Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
1
Integrating Wi-Fi Deployment with BlockChain Technology
for Accessibility and Security Optimization
John Martin Ladrido a) Lorwin Felimar Torrizo, and Lawrence Materum
Electronics and Communications Engineering Department
De La Salle University, 2401 Taft Ave., Malate, Manila Philippines
a) Corresponding author: [email protected]
Abstract. There are currently 4 billion people around the world who are using the Internet while two-thirds of the world’s
population has a mobile phone. By 2021, it is projected that 63% of total Internet traffic will come from mobile and wireless
devices. Currently when it comes to area network coverage, cellular data or LTE is the technology that is being utilized when it
comes to wireless traffic. However, LTE is over utilized such that its current data rates are not enough for consumers. On the other
hand, Wi-Fi is limited to the access point area coverage, thus it is not really a feasible replacement for cellular technology when it
comes to network coverage. This paper proposes a new type of deployment of Wi-Fi using BlockChain Technology. This method
will increase the accessibility of Wi-Fi users using BlockChain Technology and at the same time, addressing the security and
authentication problems encountered in traditional Wi-Fi Technology. The current Wi-Fi Technology is composed of 2 major
components which are; an access point and authentication mechanism. Authentication mechanism has many variants, but the
Authentication Server could only be deployed either on-premise or the cloud-based which are both centralized. This paper proposes
to utilize BlockChain Technology which is a decentralized system for authentication, thus increasing security as there is no single
entity or monopoly in authority. Everyone that is part of the BlockChain will be checking for integrity. For accessibility, the
proposal will be similar to offloading, data roaming or 802.11u, wherein the users will be able to connect to any access points that
is part or connected to the BlockChain network.
Keywords— Access Point, Authentication, BlockChain, Hotspot, Security, Smart Contracts, Wi-Fi, WiFi
INTRODUCTION
Currently wireless devices are steadily outgrowing wired devices in terms of Internet traffic. It is projected that
63% of total Internet traffic will come from mobile and wireless devices by 2021 [1]. Cellular Technology or LTE
(Long-Term Evolution) is one of the most utilized technologies in accessing the Internet due to its accessibility and
wide area coverage. However, the fact that it is being over utilized makes its throughput low and performance delayed
[2]–[4]. Thus, it is not uncommon to offload the traffic using 802.11u, where LTE and 3G traffic are being handed
over to Wi-Fi for better performance and cost cutting. Wi-Fi, compared to LTE, is more superior in LTE in terms of
throughput, delay performance, traffic received and data losses [5]. Having stated that, this paper assumes that Wi-Fi
is superior to LTE with only one major disadvantage: its accessibility or network coverage, being limited to the range
of its access point.
This paper will focus on how to improve traditional Wi-Fi deployments and address current hindrances and
accessibility issues; the process of Wi-Fi authentication, for example, requires the user to connect via SSID, and by
connecting, users must enter a password or need to authenticate by registering using an e-mail address, cellular-phone
number, or even using social media accounts such as Facebook, Twitter, etc. These traditional Wi-Fi authentications
are causing inconvenience for the users, and at the same time resulting security and privacy risk. In terms of
vulnerability - in this type of authentication, access points or hotspots are vulnerable for spoofing, making the users
2
connect to an access point without the capability to verify whether the SSID that the user is trying to connect to is
malicious or not. Since majority of these users are non-technical people who just simply connect to whatever SSIDs
are available, they are likely to provide credentials and giving access to sensitive information to malicious attackers.
This is a one-way authentication which users are being authenticated by a central authority, but the users have no
means of authenticating the connection they are trying to access whether it is malicious or not. BlockChain technology
can address this integrity issue, for the BlockChain application will only let the users connect to its own BlockChain
network [6]. As the BlockChain architecture was designed as immutable, it is nearly impossible for a BlockChain
node to compromise the BlockChain system/network.
The new Wi-Fi deployment will utilize BlockChain Technology to address two main problems: accessibility and
security or authentication. Although using BlockChain does not increase the coverage of an access point, by utilizing
BlockChain, user can connect to any access point without the need for a password or registration of credentials every
time it wants to connect. This increases accessibility in Wi-Fi technology as users could just automatically connect
without asking for permission from a central authority, and at the same time, this eliminates security and privacy risk
as users are no longer required to register every time they want to connect. This also eliminates users to be spoofed,
as users will only connect to access points that are part of the BlockChain.
In comparison to traditional authentication providers such as, Windows Authentication, Remote Authentication
Dial-In User Server (RADIUS), Diameter and others, the proposed Wi-Fi deployment will utilize the BlockChain as
its authentication provider. These traditional authentication protocols are integrated with the traditional way where
users provide their credentials, thus the Wi-Fi authentication processing time heavily depends on connection speed,
the capability of the device or server to process the authentication request of the user, and speed of the user to provide
its credentials required by the authentication Server. On the other hand, the proposed Wi-Fi authentication processing
time will depend entirely on the BlockChain architecture both in something that the user should provide for
authentication and the authentication mechanism/protocol. In terms of scalability and availability, the traditional
authentication is limited to the central authority’s resources or where it had deployed its infrastructure. While on the
proposed architecture, the BlockChain propagates, or distributed to anyone who wants to participate on the network
making it more robust, resilient, available and accessible.
The biggest difference between the two is how data is stored; in traditional authentication, data is stored in a
database which is vulnerable to hacking, eavesdropping or the database administrator/authentication provider itself
could read/decrypt and write/alter the private data for personal gain. In comparison to BlockChain, the data is
encrypted in a block and these blocks are continuously being chained creating the so-called BlockChain. Although it
is virtually possible to decrypt this chain of blocks and gain access to personal or private information, feasibility
becomes an issue because in order to read, write, and modify a BlockChain, it is needed to decrypt the entire
BlockChain before a server node in the BlockChain network updates the BlockChain. In short, a malicious user would
require administrative rights in order to alter or modify data as per traditional authentication; while it would require
computing power that surpasses every server node in the BlockChain network to alter or modify the BlockChain.
Examples of widely used BlockChain architecture are Bitcoin which takes about 15 minutes or Smart Contracts
by Ethereum which takes about 15 seconds processing/block time. The cryptochain mechanism is better to be on a
semi private-public access for shorter latency, especially with Bitcoin.
BLOCKCHAIN
This section provides an overview on the history of BlockChain, starting from its Bitcoin roots up until the recent
trends of proposed applications.
Bitcoin: The First Implementation of a BlockChain
Introduced by Satoshi Nakamoto through the paper titled Bitcoin: A Peer-to-Peer Electronic Cash System [7],
Bitcoin is an implementation of one of the many digital currencies using BlockChain technology/solution. As a
3
currency with an estimated current market cap around 110 Billion US Dollars [8], Bitcoin is currently being used by
tens of thousands of people due to its characteristics:
• Privacy: transactions can be done without divulging one’s identity.
• Open: anyone who has access to the Internet can make a Bitcoin transaction.
• Decentralized; no central entity that controls it, P2P.
• Secure and Immutable: digital signatures and validation protocols ensure that every node and user behaves
correctly, without needing intermediaries.
• Transparent: Implements a shared repository that is maintained by peers—everyone can access data and view
transactions.
Construction-wise, a ‘Bitcoin’ is basically a chain of digital signatures that points the coin’s path/history
throughout the Bitcoin ecosystem, which can also be described as a public ledger system distributed over a network
that records transactions executed among network participants (known as the Bitcoin’s BlockChain). Due to this,
Bitcoin solved conceptual issues that digital currencies have struggled to overcome:
• It introduced scarcity of a digital token (not duplicable) and its use as a decentralized DRM (digital rights
management) system as well as its disintermediation of banks for financial purposes. More than 50% of the
people in the world do not use formal or semiformal financial services and Bitcoin can potentially enable
their capacity to manage their own assets over a digital framework.
• It solved the problem of transfer of value between Internet users without having to rely on a third party; true
peer-to-peer networking of transferring value and being validated by BlockChain itself.
• Its internal mechanisms serve as a form of non-repudiation, and has the no-single-point-of-failure character
similar to the Internet, making it failure-resistant. Additionally, while a BlockChain user will depend on many
third parties, they are not reliant on a specific third party.
BlockChain: Expanding Outside Cryptocurrencies
While Bitcoin was one of the first systems to implement it, several other applications are considering BlockChain
as a mechanism in overcoming issues that utilizing BlockChain might be able to solve:
• necessity for non-traditional database technology
• participant’s capability to update the data
• privatization of data
• establishing trust between users
• reducing / discouraging possibility of attacks
• needing redundancy of data
• removing a need for a third party for whatever purposes
Based on the current trend of proposed BlockChain applications, the perceived timeline of BlockChain utilization
can be described in three stages. [9]
The first stage is strongly related to Bitcoin and cryptocurrencies. Since the birth of Bitcoin, more than 600
cryptocurrencies have been created. The most famous ones are Ethereum, Monero, and Ripple. In general,
cryptocurrencies use technology starting from Merkle’s Tree up to Adam Bank’s HashCash. However, Bitcoin was
the first application to provide P2P without intermediation of a centralized authority. [10]
The second stage is about registering, confirming, and transferring contracts or properties. During this stage, as
well as the third stages, the application of BlockChain comes from its capability of being a decentralized copy of
records to more sophisticated applications. [10]
For the third stage, the application field is no longer restricted to finance and goods transactions, but embraces
sectors like government, health, science, education, and more futuristic applications [10] One conceptual proposal that
falls to this stage involves file sharing of digital mindfiles (digital copies of human mind files). [11]
4
Smart Contracts and Ethereum
BlockChain expanded into the realm of smart contracts, which operates as a self-executing code once business
agreements have been met. As an analogy, if BlockChains are considered distributed ledgers, smart contracts turn it
into a distributed computer. BlockChain makes it possible since due to decentralization, it can be assumed that two
remote records are identical, thus we can also trust that code executed in two remote sites that base their operation
from the records will function in an identical way, all other things being equal. Smart contracts combine traditional
operation of systems and the strengths of BlockChain, which enables decentralized autonomous organizations (DAOs)
implementing the agreements and transactions.
The Ethereum BlockChain, which introduced the concept of smart contracts in 2014, allows user-created smart
contracts executed under decentralized control, executing code through the Ethereum Virtual Machine (EVM), which
conducts based on data received externally. In Ethereum, a cryptocurrency called Ether is implemented, which serves
as ‘fuel’ to be able to run smart contracts through the Ethereum system, with the amount needed depending on the
complexity of computations involving the contract [12]. Smart contracts see potential for particular business
applications, such as logistics [13], education [14], monitoring [15], and many more.
RELATED LITERATURE
There are a couple of studies that propose Wi-Fi and/or authentication processes which utilize BlockChain for the
advantages it provides.
A proposal done by Kotobi and Bilen (2017) centered on enabling spectrum access in cognitive radio networks.
In essence, their study attempts to address the ‘auction system’ that exists when license holders of full spectrum
allocations attempt to ‘share’ the spectrum during the times when the spectrum is idle, and the BlockChain provides
the facilitation of use of a virtual currency, with a secure method of validating the process. [16]
Hammi et al. (2018) proposed an authentication mechanism that is BlockChain-based. Called BCTrust, their aim
was to provide a “robust, transparent, flexible and energy efficient BlockChain-based authentication mechanism” that
can be utilized by wireless sensor networks (WSN) for Internet-of-Things (IoT). Their study included an
implementation using the Ethereum BlockChain, which, when compared to traditional methods, can potentially obtain
up to 89.82% of energy savings. [17]
Sanda and Inaba (2016) utilized BlockChain as a method in logging user access in and out of a Wi-Fi network.
Their system utilized ColoredCoins, a method in pushing/adding metadata on top of the Bitcoin BlockChain such that
it can categorize specific coins as representations of specific real-world assets or usage. Within the Wi-Fi network,
the derived ColoredCoin-based tokens on the BlockChain will be exchanged during the process of authentication to
the network, and the exchange provides a one-is-to-one correspondence in identifying a single user. [18]
Niu et al. (2017) did a follow-up work from Sanda and Inaba, stating that the lack of focus in privacy in the
previous work led to their proposed solution of implementing an anonymous yet accountable authentication scheme
for Wi-Fi hotspot access. Their study focused on having the connecting credentials involve the user’s Bitcoin address,
and establishing a blacklist system for users misbehaving in the network. [19]
PROPOSED WI-FI AUTHENTICATION PROCESS
Tokenized Authentication Process
This study proposes a network of Wi-Fi access points, enabled by a number of servers that are connected to the
same BlockChain network in order to provide a decentralized record of access credential records as well as network
5
traffic. The system enables any user to be able to connect seamlessly to any of the access points AP(1) to AP(n) while
maintaining the authentication process uniform between all access points.
The proposed system has a similar functionality as Sanda and Inaba’s work. In their proposed system, the Wi-Fi
authentication process is brought about by utilizing ColoredCoins-based tokens provided by the server to the user as
a sort of identification tag, wherein every time that a user connects to the access point, the server requests the user to
lend the token in order to verify the users public address coupled with the signatures incorporated within the coin [18].
One of the biggest limitations of Sanda and Inaba’s work is the ColoredCoins dependency to the Bitcoin network,
making each token transfer involves a transaction that entails transaction fees, as well as having long transaction times.
In order to combat this limitation, the proposed system will utilize Ethereum’s capability of creating custom tokens
on its own smart contracts. These custom tokens, which are also able to contain metadata, will serve as an identification
tag similar to how ColoredCoins were used on the reference work.
The platform involves multiple parties that are assigned their own respective address: the user (U) which translates
to a device capable of connecting through Wi-Fi, the access points coupled with their own local server (S) that manages
the local area network, and an administrator (A) which serves as a registration related node for setting up additional
servers. Each party will need a wallet which establishes their address as well as to store the tokens (T) that will be
distributed for validation purposes. The network diagram is shown in Figure 1.
FIGURE 1. Proposed Blockchain Network Diagram
Prior to users being able to connect to the access points, there is first a need for the servers to register their access
points to the administrator. The process is as follows:
6
FIGURE 2. Server Registration to Administrator
• Server S connects to administrator A.
• Through an API dedicated for server registration, S sends its public key as well as additional information to
A, including the amount of intended registered users the access point of S is intended to have.
• A sends a particular amount of token T to the wallet of S, containing encrypted metadata related to the
registration details of S, and broadcasts this transaction to the BlockChain network (i.e. to the other servers).
It should be noted that the only time a server interacts with the administrator is during this registration process.
Once the registration has been confirmed, each server operates as independent nodes with no dependencies related to
the administrator.
Meanwhile, a user U should be able to register on any server S though the following process:
FIGURE 3. User Registration to Server
• User U connects to Server S through its access point
• Through an API dedicated to user registration, U sends its public key as well as required information to S.
• S sends a token to the wallet of U, which has metadata appended to it related to the registration details of U,
and broadcasts this transaction to the BlockChain network (i.e. to the other servers)
7
The received token contains encrypted metadata relating to the origins of the token, and thus will be the primary
authentication tool that represents the ‘rights’ of a user to connect to any access point connected to the BlockChain
network.
When a user U attempts to authenticate his device in order to utilize the Wi-Fi access point, the following process
occurs:
FIGURE 4. User Authentication to Server
• User U connects to Server S through its access point
• Through an API dedicated to authentication for Wi-Fi access, U sends its token to S.
• Server S confirms U’s registration details through verifying the metadata content of the coin in conjunction
with U’s address / public key.
• Upon confirmation, S returns the token to U, and allows network access.
• The token transactions are broadcasted by S to the BlockChain network.
Expanded Use Case Example: Private Wi-Fi Sharing Networks
The proposed system enables users to be able to connect to access points connected to the BlockChain using a
token-based authentication method. In addition, with the flexibility of smart contracts, it becomes fully possible to
implement functions to the network that can provide network-related operational features.
One potential application is a bandwidth-sharing ecosystem wherein the deployment of access points is done by
multiple individuals or private entities, in the hope that their deployment of an access point enables them to be able to
connect to the access point of another person located in a different location.
The concept is inspired by BitTorrent [20] trackers, particularly the private and invite-only ones. BitTorrent is a
well-known P2P file-sharing protocol, enabled by trackers that list the file available for transfer. While publicly
accessible trackers generally provide unlimited access for other people to download the files listed, private trackers
generally implement a “give as much as you take” system, tallying the ratio of amount of uploads compared to
downloaded data. In this system, those with low sharing rates are penalized (such as restricted access, throttled
download speed).
A private Wi-Fi sharing network can be deployed in the same manner as a BitTorrent private tracker, which
similarly operates on a “give and take” principle, but instead of sharing files, Internet bandwidth is the one being
8
shared. The amount of usage of users to each access point is kept track of, which is analogous to how the
upload/download ratio is also tracked in private trackers.
In this system, while there may or may not be restrictions on who can register a server to the administrator, it is
imperative that only those that have deployed nodes of access points will be able to connect to the access points of
others as users. In such a case, the BlockChain keeps track of which addresses have already deployed nodes.
The token system can be implemented in such a way that the administrator’s distributed tokens operate as a
representation of the ratio between the amount of bandwidth his or her deployed access point has provided to other
users versus the amount of bandwidth he has used from others. Every time a user utilizes another’s deployed access
point, part of his tokens will serve as payment to the access point provider depending on the amount of usage. If a
user’s amount of tokens drops below a certain threshold, he will be not able to use any other’s access point resources
until he has accumulated enough tokens (through other users providing tokens for usage of his/her deployed access
point) to remove the restriction.
CONCLUSION
In this paper, we proposed a new method in utilizing Wi-Fi technology in order to address security and accessibility
concerns. This proposed method will enable us to further safely exploit and fully utilized the bandwidth that Wi-Fi
technology could offer us.
By incorporating BlockChain technology with Wi-Fi we have further optimized Wi-Fi capabilities in terms of
security and accessibility. Users only need to register in an administrator once, to obtain a Token which will act as
ticket for the user to connect to any access point that is part of the BlockChain. Compare to existing Wi-Fi
authentication method, this eliminates the need for user to provide user information like: registering every time user
connects to a new access point, asking for a new voucher if the hotspot is using voucher-based authentication, asking
for a new username and password if the administrator decides to change the credential in a specific access point,
asking for a new pre-shared key, and other types of traditional authentication method. Scalability and availability are
introduced as the BlockChain propagates or distributed to anyone who wants to participate in the network, unlike the
traditional authentication which is limited to the central authority’s resources. By eliminating this traditional
authentication method, we further increased the accessibility of Wi-Fi technology. Although traditional Wi-Fi
authentication schemes seems faster processing-wise due to the inherent processing requirements of BlockChain
construction, the proposed scheme eliminates the need of the user to ask or request for permission from a central
authority every time it wants to join or connect to the Wi-Fi, leading to faster times for the user to connect to the
network. The accessibility is further widened as the proposed authentication is distributed and not bounded by the
central authority’s presence. In terms of security, privacy risks are no longer an issue as users uses a token for
authentication every time it connects to a hotspot or access point which eliminates credential registration in traditional
Wi-Fi authentication. Data alteration/modification is also addressed, as malicious users who want to commit such
actions for personal gains must surpass the computing power of every participating server node in the network. User
can no longer be spoofed or vulnerable to malicious access points as their devices will only communicate with hotspots
or access points that are part of the BlockChain.
Anyone who wants to deploy an access point only needs to register to an administrator once. In deployment, access
point must have its local server that will be used for registering and authenticating users. Before being part of the
BlockChain, this server must register to an administrator and the administrator will be the one to add this server in the
BlockChain network. The administrator and server only interact during registration process and communications end
when the registration is completed. Thus, making the server an independent node.
Although the registration process seems centralized, every transaction is broadcasted to the network, validated and
recorded in every node that is part of the BlockChain network, making it decentralized by nature.
9
REFERENCES
[1] Cisco, “The Zettabyte Era: Trends and Analysis - Cisco.” [Online]. Available:
https://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-index-vni/vni-
hyperconnectivity-wp.html. [Accessed: 17-Aug-2018].
[2] IEEE, IEEE Standard 802.11s-2011, Specific requirements Part 11 : Wireless LAN Medium Access Control (
MAC ) and Physical Layer ( PHY ) specifications Amendment 10 : Mesh Networking IEEE Computer Society.
2011.
[3] Cisco, “Architecture for Mobile Data Offload over Wi-Fi Access Networks,” vol. 8, no. 2008, pp. 1–23, 2012.
[4] S. Martin, “802.11u and Hotspot 2.0 promise Wi-Fi users a cellular-like experience | Network World,” 2011.
[Online]. Available: https://www.networkworld.com/article/2180880/tech-primers/802-11u-and-hotspot-2-0-
promise-wi-fi-users-a-cellular-like-experience.html. [Accessed: 16-Dec-2018].
[5] C. C. Kevin, “Simulation Comparison between LTE and Wi-Fi in Networks,” pp. 0–19.
[6] Blockchain Semantics, “How does blockchain achieve immutability and permanence of its records? |
Blockchain Semantics Blog.” [Online]. Available: https://www.blockchainsemantics.com/blog/immutable-
blockchain/. [Accessed: 16-Dec-2018].
[7] S. Nakamoto, “Bitcoin: A Peer-to-Peer Electronic Cash System,” Www.Bitcoin.Org, p. 9, 2008.
[8] “Bitcoin (BTC) - USD - Live Bitcoin price and market cap.” [Online]. Available:
https://www.cryptocompare.com/coins/btc/overview/USD. [Accessed: 17-Aug-2018].
[9] M. Swan, Blockchain: Blueprint for a New Economy, 1st ed. 2015.
[10] V. Gatteschi, F. Lamberti, C. Demartini, C. Pranteda, and V. Santamaria, “To Blockchain or Not to
Blockchain: That Is the Question,” IT Prof., vol. 20, no. 2, pp. 62–74, 2018.
[11] M. Swan, “Blockchain Thinking,” IEEE Technol. Soc. Mag., no. December, pp. 41–52, 2015.
[12] “Ethereum Project.” [Online]. Available: https://www.ethereum.org/. [Accessed: 11-Aug-2018].
[13] Y. Yuan, F. Y. Wang, and Ieee, “Towards Blockchain-based Intelligent Transportation Systems,” 19th Int.
Conf. Intell. Transp. Syst., pp. 2663–2668, 2016.
[14] M. Hericko, “EduCTX : A Blockchain-Based Higher Education Credit Platform,” vol. 6, no. October 2017,
2018.
[15] J. Gao et al., “GridMonitoring: Secured Sovereign Blockchain Based Monitoring on Smart Grid,” IEEE
Access, vol. 6, pp. 9917–9925, 2018.
[16] K. Kotobi and S. G. Bilen, “Blockchain-enabled spectrum access in cognitive radio networks,” 2017 Wirel.
Telecommun. Symp., pp. 1–6, 2017.
[17] M. T. Hammi, P. Bellot, and A. Serhrouchni, “BCTrust: A decentralized authentication blockchain-based
mechanism,” IEEE Wirel. Commun. Netw. Conf. WCNC, vol. 2018–April, pp. 1–6, 2018.
[18] T. Sanda and H. Inaba, “Proposal of new authentication method in Wi-Fi access using Bitcoin 2.0,” 2016
IEEE 5th Glob. Conf. Consum. Electron. GCCE 2016, pp. 0–4, 2016.
[19] Y. Niu, L. Wei, C. Zhang, J. Liu, and Y. Fang, “An anonymous and accountable authentication scheme for
Wi-Fi hotspot access with the Bitcoin blockchain,” 2017 IEEE/CIC Int. Conf. Commun. China, ICCC 2017,
vol. 2018–Janua, no. Iccc, pp. 4–6, 2018.
[20] “BitTorrent.” [Online]. Available: https://www.bittorrent.com/. [Accessed: 17-Dec-2018].
10
PAPR Improvements by Using Clipping and Filtering in
Direct Spectrum Division Transmission
Sumika Omata1, a), Motoi Shirai1, Takatoshi Sugiyama1,
Izumi Urata2, Fumihiro Yamashita2
1 Electrical Engineering and Electronics, Kogakuin University, Tokyo, Japan 2 NTT Access Service System Laboratories, Yokosuka, Japan
Abstract. Currently, demands for broadband communication are increasing in wireless communications. However, there
is a problem that effective use of frequency bands cannot be realized due to scattered unused bands caused by DAMA
(Demand Assigned Multiple Access). As a method to solve this problem, DSDT (Direct spectrum division transmission)
has been studied and developed. In this scheme the single carrier modulated signal is finely divided into sub-spectra in
the frequency domain on the transmission side and they are allocated to the unused bands over the dedicated bandwidth.
Then they are combined on the reception side, and finally the original single carrier modulated signal is recovered.
However, there is a problem that the PAPR (Peak-to-average power ratio) increases in direct spectrum division
transmission as the number of sub-spectrum increases. In order to improve the PAPR degradations, we propose a new
direct spectrum division transmission applied by clipping and filtering method, which can be easily realized without
transmitting the PAPR control information to the receiving side. The performances of the proposed scheme are evaluated
by computer simulations. It is confirmed that the PAPR performances of the DSDT with clipping and filtering are
improved without BER (Bit error rate) degradation by setting the adequate clipping thresholds and adjacent channel
interference due to clipping can be reduced by filtering.
Keywords— Wireless Communications; Direct Spectrum Division Transmission; PAPR; Clipping and Filtering;
Adjacent channel interference
INTRODUCTION
Presently, due to the development of radio access network, there is a problem that the frequency is insufficient. For example, in satellite communications, the base station allocates a frequency band to each earth station every time a data transmission request from the earth station, and the earth station releases the allocated frequency band after the communication terminated. (by using Demand Assigned Multiple Access: DAMA [1].)
By repeating allocation and release of various bands by this scheme, unused bands are scattered over the satellite transponder, and even if the sum of the unused bands is wider than the required bandwidth, there is no consecutive unused bands. As a result, a new broadband user cannot be accommodated, and as a result the frequency use efficiency deteriorates.
11
FIGURE.1 Outline of Direct Spectrum Division Transmission.
As a method to solve this problem, there is a direct spectrum division transmission (DSDT) [2]. As shown in FIGURE.1, the single carrier modulated signal is divided into a plurality of sub-spectra on the transmission side, they are allocated the unused bands, and transmitted. At the receiving side, the sub-spectra are combined and the single-carrier modulation signal is regenerated.
This makes it possible to effectively use unused bandwidth over a satellite transponder. Moreover, since the DSDT can be realized by inserting spectrum division adapters, existing satellite communication facilities. It is possible to improve the frequency use efficiency while effectively utilizing the existing satellite communication system. It is expected to realize high speed satellite communication and increased number of accommodated users.
FIGURE.2 Configuration of DSDT adapter.
12
The concept of the DSDT system is shown in FIGURE.2. On the transmission side (Tx side), the DSDT adapter convents the output signal of the existing modem into a frequency-domain signal by applying a fast Fourier transform (FFT) and divides it into multiple sub-spectra by using a spectrum dividing filter bank circuit. The divided sub-spectra are allocated by frequency shifter and converts them into a time-domain signal by applying an inverse FFT (IFFT).
On the receiving side (Rx side), the DSDT adapter converts the received sub-spectra into the frequency domain by an FFT circuit. The they are frequency re-shifted and combined by using a spectrum combining filter bank circuit. Finally, they are converted into the time domain by an IFFT. The output signal of the DSDT adapter is the original modulated signal and it is demodulated by the existing modem.
PROBLEMS OF DIRECT SPECRUM DIVISION TRANSMISSION
In DSDT, there is a problem that the PAPR(Peak to Average Power Ratio[3]) of the transmission signal increases by DSDT spectrum divisions. FIGURE.3 shows relationship between division number and PAPR in direct spectrum division transmission. The modulation method was QPSK, and the roll-off factor α = 0.2 was simulated. The PAPR degradations are 0.6dB and 1.8dB in 2 and 4 spectrum divisions, respectively.
FIGURE.3 Relationship between division number and PAPR.
Before now, in the PAPR reduction method, an SLM method[4] for controlling the subcarrier phase, a phase sequence blind estimation method[5][6], and the like are studied. In the research on the PAPR reduction method in the DSDT, a scheme is being studied by adding different phase sequences for each sub-spectrum on the transmission side[7], This scheme achieves no degradation of the transmission quality, but it takes longer time to determine the parameters depending on the number of divisions and the phase resolution.
Therefore, in this paper, we propose a method of applying clipping and filtering[8][9] without processing delay due to reduce PAPR in direct spectrum division transmission and it does not need to transmit PAPR control information to the receiving side.
0
1
2
3
4
5
6
7
8
No Division 2 4
PAP
R (
dB
)
13
OUTLINE OF PROPOSED METHOD
FIGURE.4 Simulation block diagram of the proposed DSDT.
In the clipping and filtering method, the peak power is limited to a clipping threshold, and the side lobes regenerated by clipping are suppressed with filtering. PAPR which is increased by spectrum dividing can be relatively easily reduced. In addition, it is not necessary to notify the PAPR control information to the receiving side, and simple realization is possible. However, clipping and filtering may cause deterioration of transmission quality due to the influence of nonlinear distortion.
FIGURE.4 shows Simulation block diagram of the proposed direct spectrum division transmission. On the transmission side of the proposed scheme, the peak signal is suppressed by setting the clipping threshold after dividing the single carrier modulated signal into sub-spectra with a spectrum division filter. Then, the regenerated side lobes are suppressed by filtering. On the reception side, the sub-spectrum is combined through the spectrum combination filter, and the original single-carrier modulation signal is restored and demodulated.
The clipping threshold is set as β times of the transmission signal average power. FIGURE.5 shows the method of determining the threshold β. I is the real part. Q is the imaginary part. The length of the arrow indicates its power. When the power of the transmission signal at a certain time is smaller than the threshold β, the transmission is performed with the power as it is, and when the power is larger than the threshold β, the transmission power is limited to the threshold β.
The relationship between clipping threshold β and PAPR performances and BER performances with clipping was evaluated by simulation.
14
FIGURE.5 How to determine the threshold β.
SIMULATION RESULT
TABLE 1. Major simulation parameters.
Modulation scheme QPSK, 10Msymbol/s
Roll-off factor (α) 0.2
FFT size 4096 points
Number of spectrum divisions No division, 2
Propagation path AWGN (Additive white Gaussian noise)
Clipping & Filtering Clipping threshold: β = 1.0 ~ 2.5
Filter: Raised cosine characteristic: α = 0.2
ACI QPSK, 10Msymbol/s
No guard band, DSDT signal to QPSK signal power ratio = 0dB
The simulation parameters are shown in TABLE 1. As an initial study, the division number is 2. Clipping threshold β is 1.0 to 2.5. As an adjacent channel interference (ACI) environment, another QPSK spectrum is allocated between the DSDT sub-spectra without guard band. DSDT signal to the QPSK signal power ratio is set to be 0dB.
Power Spectra of the Direct Spectrum Division Transmission Signals
FIGURE.6 shows the single carrier modulation spectrum, and FIGURE.7 shows the spectrum after division. It can be seen that the spectrum can be divided by using the division filter.
15
FIGURE.6 No division spectrum (original QPSK). FIGURE.7 After division sub-spectra of DSDT.
FIGURE.8 to 10 show spectra of clipping threshold β of 2.0, 1.5 and 1.0. It can be confirmed that the side lobe of the sub-spectrum is reproduced by the nonlinear distortion caused by setting the clipping threshold too small. In the case of FIGURE.8, the clipping threshold is as large as 2.0, and since there is almost no nonlinear distortion, there is not much change from the spectrum of FIGURE.7. FIGURE.11 shows the spectrum after filtering. It can be seen that sidelobes reproduced by using filtering can be reduced.
FIGURE.8 DSDT sub-spectra with clipping threshold β=2.0. FIGURE.9 DSDT sub-spectra with clipping threshold β=1.5.
16
FIGURE.10 DSDT sub-spectra with clipping threshold β=1.0. FIGURE.11 DSDT sub-spectra with after filtering.
PAPR Performances of DSDT Signal by the Proposed Scheme
Next, what we actually evaluated quantitatively is shown. FIGURE.12 shows the PAPR performances when the number of divisions is 2. It can be seen that PAPR increases by spectrum divisions. The horizontal axis represents PAPR, and the vertical axis represents the CCDF (Complementary Cumulative Distribution Function) value of PAPR. PAPR can be reduced from this figure by lowering the clipping threshold. For examples, PAPRs with CCDF = 10-2 are about 5.8dB and 4.7dB for clipping threshold β of 2.0 and 1.5 respectively.
FIGURE.13 shows the relationship between clipping threshold β and PAPR. The horizontal axis is the clipping threshold β, and the vertical axis is the PAPR at CCDF = 10-2 From FIGURE.5, for example, PAPR by setting β = 1.8 is equivalent to a single carrier spectrum without division. In this case, the PAPR improvement amount by clipping and filtering is 0.6dB.
FIGURE.12 PAPR performances in 2 divisions. FIGURE.13 Relation between clipping threshold β and PAPR.
1.E-03
1.E-02
1.E-01
1.E+00
3 4 5 6 7
CC
DF
PAPR (dB)
No Division
After Division
β=2.0
β=1.5
β=1.0
0
1
2
3
4
5
6
7
1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4
PAP
Rat
CC
DF=
10-2
(dB
)
Clipping threshold β
No Division
After Division
Proposal
0.6dB
17
Bit Error Rate Performances of DSDT Signal by Proposed Scheme
Next, the BER (Bit Error Rate) performances was evaluated. FIGURE.14 shows the BER performances of the proposed scheme with the clipping threshold β as a parameter. The horizontal axis represents Eb/No, and the vertical axis represents BER. At the clipping threshold β = 1.5, the BER performance is equivalent to that of single carrier transmission without division, but at β = 1.0, it is found that the required Eb/No at BER = 10-4 points is deteriorated by about 1.9dB. This is due to the fact that deterioration due to nonlinear distortion caused by the clipping and filtering.
FIGURE.14 BER performances of DSDT signal. FIGURE.15 Relationship between clipping threshold and
required Eb/No.
Next, FIGURE.15 shows the relationship between the clipping threshold β and the required Eb/No at BER = 10-4
points. When the threshold β is made too small, deterioration of the BER performances due to nonlinear distortion is observed. However, there is no deterioration of the BER performances from the time of single-carrier transmission without division until the clipping threshold β = 1.5. Therefore, it has been clarified that PAPR can be reduced without deteriorating transmission quality by setting an appropriate clipping threshold in the proposed scheme.
1.E-05
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
0 2 4 6 8 10
BER
Eb/No (dB)
β=1.5
β=1.0
No Division
8.0
9.0
10.0
11.0
1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4R
equ
ired
Eb
/No
at
BER
=10
-4(d
B)
Clipping threshold β
No Division
Proposal
18
Bit Error Rate Performances in an Adjacent Channel Interference Environment
FIGURE.16 Spectrum only clipping (β = 1.0). FIGURE.17. Spectrum applied both clipping and filtering
(β = 1.0).
Next, BER performance influence by applying the proposed scheme in an adjacent channel interference environment is evaluated.FIGURE.16 shows spectra in an adjacent channel interference environment applied only clipping to the DSDT spectra. On the other hand, FIGURE.17 shows spectra applied both clipping and filtering to the DSDT spectra. In the both figures the clipping threshold β is set to be 1.0.
From these figures, when clipping is only applied, adjacent interference is occurred by nonlinear distortion due to the clipping,. However, it can be seen that the influence of nonlinear distortion can be reduced by applying filtering.
Next, FIGURE.18 shows the BER performances of an adjacent channel QPSK signal. In this figure, the DSDT uses only clipping or both clipping and filtering. For, example, clipping threshold β is set to be 1.0. By using the DSDT with both clipping and filtering, required Eb/No at BER of 10-4 improvement of 0.5dB can be achieved compared to that of with only clipping. On the other hand, the BER performances of the DSDT signal in an adjacent channel interference environment are almost similar to those without adjacent channel interference in FIGURE.14 and FIGURE.15.
FIGURE.18 BER performance of the adjacent channel QPSK signal.
1.E-05
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
0 2 4 6 8 10
BER
Eb/No [dB]
without DSDT spectra
with DSDT spectra (only
clipping, β=1.5)
with DSDT spectra (both clipping & filtering, β=1.5)
19
CONCLUSION
In this paper, we propose a new direct spectrum division transmission (DSDT) using clipping and filtering method to reduce PAPR which increases according to the spectrum divisions.
The clipping and filtering is a method in which the peak power which is increased by dividing the spectrum can be
limited within the threshold value and the peak power can be greatly reduced and it is not necessary to notify the control information to the receiving side and can be easily realized System. We evaluated the PAPR performances and the BER performances by varying the threshold value using this clipping and filtering method.
For example, in the case of 2 divisions of the single carrier spectrum, PAPR before division was 5.4dB at CCDF =
10-2, 6.0dB after division, and it was confirmed that PAPR increased by division. In order to obtain the same PAPR as before splitting, it was necessary to set the clipping threshold β = 1.8, and the PAPR improvement at that time was 0.6dB. Moreover, in the evaluation of the BER performances, deterioration of the BER performances due to the nonlinear distortion is observed when the threshold value is made too small, but it was revealed that there was no degradation of the BER performances from the time of single carrier transmission until the clipping threshold value = 1.8.
In the evaluation in the adjacent channel, it has been clarified that better BER performances of the adjacent channel QPSK signal can be obtained in an adjacent interference environment. For example, by using DSDT with clipping and filtering, required Eb/No at BER of 10-4 improvement of 0.5dB can be achieved compared to that by using DSDT with only clipping. Furthermore, the BER performance of the DSDT signal is almost similar to no adjacent channel environment.
Therefore, it has been clarified that the proposed scheme is one of the effective schemes to reduce PAPR and adjacent channel interference in the direct spectrum division transmission.
REFERENCES
[1] K.Nakahira, K.Kobayashi and K.Ohata, “Channel Allocation Algorithm for Novel Polarization Tracking free Ku-band Mobile Satellite Communication Systems”, IWSSC 2008, pp.75-79, Oct 2008.
[2] J.Abe, K.Nakahira and K.Kobayashi, “A Blind Phase Compensation Method for Direct Spectrum Division Transmission”, IEEE GLOBECOM 2011, Dec 2011.
[3] H. Ochiai and H. Imai, “On the distribution of the Peak to Average Power Ratio in OFDM Signals”, IEEE Trans. Commun., vol. 49, n2, pp. 282-289, Feb. 2001
[4] S.Heo, H.Noh, J.No and D.Shin, “A Modified SLM Scheme With Low Complexity for PAPR Reduction of OFDM Systems” IEEE Transactions on Broadcasting, vol.53, no4, Dec.2007
[5] O.Muta, “Phase-Sequence Blind Estimation for Phase-Rotation based PAPR Reduction in MIMO Coded-OFDM Systems”, 2011 5th International Conference on Signal Processing and Communication Systems(ICSPCS), Dec.2011
[6] O.Muta, “Effect of phase control based PAPR reduction in MIMO adaptive modulated vector coding systems”, 2011 International Conference on Wireless Communications and Signal Processing (WCSP), Nov.2011
[7] R.Miyatake, J.Abe and T.Sugiyama, “A Study on peak to average powor ratio reduction technique for direct spectrum divion transmission”, IEICE Trans. Commun, B-3-22, p.274 May.2015
[8] J.Armstrong, “Peak-to-average power reduction for OFDM by repeated clipping and frequency domain filtering”, Electronics Letters, vol.38 No.5, Feb.2002
[9] T.Fujii and M.Nakagawa, “Adaptive Clipping Level Control for OFDM Peak Power Reduction Using Clipping and Filtering”, IEICE Trans. Fundamentals, vol.E85-A, no7 Jul.2002
20
Design and Implementation of ITU-R M.1371 Data
Transfer Protocol and NMEA AIS Sentence Using
Open Source Software for Universal Shipborne
Automatic Identification System Application
Gerino Mappatao a), Jomel Lorenzo, Noriel Mallari,
Jean Clifford Espiritu, Lawrence Materum, and Alexander Abad
Electronics and Communications Engineering Department
De La Salle University, 2401 Taft Ave., Malate, Manila Philippines
a) Corresponding author: [email protected]
Abstract. The adoption and usage of Automatic Identification System (AIS) plays a major role in providing safety in marine
transportation. However, it is costly and not economical for small vessel operators. This paper aims to provide a low-cost alternative
in the market by making use of open source software (OSS) to implement the AIS data transfer protocol as defined by ITU-R
M.1371. The protocol was implemented using an Infineon XMC microcontroller to handle the processes of transmission and
reception of AIS messages. Based on the comparison of the input and output AIS messages for verifying the implementation, no
mismatch were found in the transmission and reception. The results proved the accuracy of the OSS-built realization of the protocol.
Keywords—navigation; marine navigation; radio navigation; radio frequency identification; sense and avoid; traffic
control; ships; marine vehicles; marine transportation; marine safety; marine accidents; automatic identification system;
INTRODUCTION
Vessel traffic management is important in marine transportation. The Universal Shipborne Automatic
Identification System (AIS) is an electronics communications system used by ships and vessel traffic services. It is
designed for addressing the growing demand in sea transportation and its ensuing traffic for the purpose of tracking
ships by allowing AIS-equipped ships to view marine traffic within their vicinity and to be seen by other AIS-equipped
ships, sea and shore traffic control stations, and authorities in the area [1–4]. The focus of AIS is for providing safety
and security measures for the ships. This includes collision avoidance, emergency report, identification of hazards
carried by the ship, interrogation, search and rescue, and fleet tracking among others. Class A and Class B are two
types of AIS used in shipboard AIS transceivers [5–7]. Existing commercial Class A and Class B transceivers are
expensive [8–10]. In order for AIS to be readily adoptable in small vessels in non-first-world economies, it has to be
low in cost but durable. Example of such countries with archipelagos and numerous shipping vessels include
Indonesia, Myanmar, and the Philippines. An attempt to address this need has been done by implementing AIS
transceivers on a DSP by using ADI Visual DSP Kernel [11]. The approach, however, is costly from a mass-production
viewpoint due to the required licensing feature of the development platform. Hence, there is a need for a low-cost
alternative. One of the aims of this work is to lower the cost by utilizing Open Source Software (OSS) and readily
available embedded systems platforms like the Infineon XMC microcontroller—a 32-bit, ARM-based microcontroller
for commercial and automotive applications. OSS is a type of computer software where licensing allows anyone to
distribute software for any purpose. This allows researchers and software developers to utilize existing code or libraries
21
for quick and rapid prototyping. OSS exists for a variety of applications, but as of this writing, there is no OSS available
for a complete implementation of an AIS that targets a specific embedded systems development platform. Through
OSS, licensing would not be an issue and therefore drastically reduce the cost of AIS transceivers for non-first world
sea-faring economies, especially those with large number of small shipping vessels. In a similar sense, this approach
was the one undertaken by OpenBTS [12]. This paper focuses on the implementation of the AIS data transfer protocol
defined by ITU-R M.1371 [13] as it is a critical standard of for realizing a Class A AIS. Two OSS libraries were used
in this work: dAISy [14] and the AIS Sentence Parser [15]. dAISy, is a commercial off-the-shelf and low-cost AIS
message receiver, but is not intended for implementation as a full Class A AIS. This research work takes advantage
of dAISy code for the data transfer protocol, with modifications to support full AIS implementation in the future. The
AIS Sentence Parser is utilized to format the AIS messages to NMEA, with the purpose of sending to a user interface
(UI) for implementing features on higher levels of the network stack. The paper is organized as follows. The next
section is a discussion of the ITU-R M.1371 Data Transfer Protocol and its implementation, which includes the
message format of the received data from other vessels (AIVDM), the protocol messages, and design considerations.
This is followed by data and results out of the implementation, and lastly a conclusion of the items achieved in this
work.
ITU-R M.1371-5 DATA TRANSFER PROTOCOL AND ITS IMPLEMENTATION
FIGURE 1. ITU-R M.1371 frame contents [13].
A full description of the data transfer protocol is defined in [13]. The following list outlines a summary of the
contents of the packet format based on [13]. It is to be noted that the data is line-encoded using non-return to zero
invert (NRZI).
1. Bit-stuffing – If five consecutive 1’s are found in the stream, a 0 zero should be inserted after the five consecutive
1’s.
2. Training Sequence – A twenty-four bit sequence of alternating 1’s and 0’s.
3. Start flag – A standard HDLC flag with a constant 8-bit sequence: 01111110, and not subject to bit-stuffing.
4. Data – 168-bit long payload of the transmission packet.
5. Frame Check Sequence (FCS) – Uses cyclic redundancy check (CRC), 16-bit polynomial to calculate the
checksum.
6. End flag – Identical to the Start Flag.
7. Buffer – Buffer bits to consider bit-stuffing, distance delay, and synchronization jitter
AIVDM Message Format
FIGURE 2. AIVDM packet [16].
AIVDM is one form of NMEA format that is solely used for marine AIS [16]. Figure 2 is one example of a typical
AIVDM packet and it contains seven fields. The following explains the content of the AIVDM packet.
1. Field 1 – It identifies that this is an AIVDM packet.
2. Field 2 – The count of parts for long messages since an AIVDM packet is limited to 82 characters as a maximum.
3. Field 3 – Part number of the message.
22
4. Field 4 – A sequential message ID for multi-part messages.
5. Field 5 – Radio channel code. The AIS uses the high side of the duplex from two VHF radio channels: AIS Channel
A is 161.975 MHz (87B); AIS Channel B is 162.025 MHz (88B).
6. Field 6 – Data payload and this is the AIS message to be decoded. Field 7 is number of fill bits required to pad the
data payload to a 6 bit boundary, ranging from 0 to 5.
ITU-R M.1371-5 Messages
According to [13], there are 26 types of AIS message. Table 1 shows the sample breakdown for messages 1 and 3
taken from Table 2 (sample messages to be used for testing based on the ITU-R standard).
TABLE 1. Message 1 and 3: Position Report Class A
Parameter Number of bits Encoding
Message Type 6 Constant: 1-3
Repeat Indicator 2 Message repeat count
MMSI 30 9 decimal digits
Navigation Status 4
Rate of Turn (ROT) 8
Speed Over Ground (SOG) 10
Position Accuracy 1
Longitude 28 Minutes / 10000
Latitude 27 Minutes / 10000
Course Over Ground (COG) 12 Relative to true north, to 0.1 degree precision
True Heading (HDG) 9 0 to 359 degrees, 511 = not available
Time Stamp 6 Second of UTC timestamp
Maneuver Indicator 2
Spare 3 Not used
RAIM flag 1 Receiver autonomous integrity monitoring (RAIM) flag
of electronic position fixing device
Radio Status 19
Design Considerations
Basically, the setup of the hardware (Figure 4) is the connection of the receive and transmit pin to simulate reception and transmission of the AIS. In addition to that, Bluetooth connectivity is added to provide an output for checking the accuracy of the system.
FIGURE 3. Hardware connections of XMC Microcontroller.
23
Figure 4 shows the block diagram of the code implemented in the microcontroller. First, the input AIS message will be converted into bits for transmission by the txStreamHandler. After transmitting, the ReceiveHandler will receive all the bits transmitted which will be converted into NMEA format. Lastly, the NMEA message will be transmitted to a laptop for verification of result through Bluetooth.
FIGURE 4. Block Diagram of XMC code.
Figure 5 shows the flowchart of the code in the microcontroller. First step is the initialization phase which resets
everything to prevent errors. Then, the input AIS message will be converted into bits and transfer to the streamBuffer.
After enqueue process, the txStreamHandler will do the transmission of bits while the ReceiveHandler will receive
the bits until there is none left in the streamBuffer. The process of txStreamHandler and ReceiveHandler is discussed
in the next subsection. After the de-queuing process, the packet is now complete and ready to be processed for
conversion into NMEA format. Finally, the NMEA message will now be transmitted through Bluetooth for
comparison.
The flowchart for the txStreamHandler is shown in Figure 6 and implements the NRZI format. The process of
transmission will depend on the input bit from the streamBuffer. If the input bit is equal to zero, then the output will
toggle, otherwise, the output will not change. In addition to that, the txStreamHandler checks the content of the
streamBuffer and if the bit is not available in the buffer then the output will be on idle mode.
FIGURE 5. Flowchart of XMC code.
24
FIGURE 6. Flowchart of transmit state machine.
On the other hand, the flowchart of ReceiveHandler is shown in Figure 7 and this decodes the NRZI transmitted
bits. First is the input bit from the txStreamHandler is compared with the previous bit to determine the value to be
stored in rx_bitstream. The state machine of ReceiveHandler has three states: off, reset, wait for preamble and start
flag. The off state explains the receiver is off. The reset state initializes all variables for reception of the preamble and
start flag. The wait for preamble and start flag state determines if the packet is valid by receiving the preamble and
start flag. After that, the data will now be received while receiving, the calculation of cyclic redundancy check (CRC)
will be implemented. Finally, the end flag will now be received, and the packet will be checked by computing the
CRC if there are errors.
25
FIGURE 7. Flowchart of receive state machine.
DATA AND RESULTS
The test setup as shown in Figure 8 consists of the XMC microcontroller for transmission and reception, oscilloscope for display of signals, and laptop for the output. The microcontroller will send the AIVDM packet through Bluetooth to the laptop for comparison of input and output.
26
FIGURE 8. Test Setup.
Figure 9(a) shows the oscilloscope output of the transmitted signal. The output looks like a pulse because the microcontroller is transmitting at 9600 baud and the oscilloscope cannot keep up with its speed. To ensure the transmitted signals are correct and is following the NRZI format, Figure 9(b) shows the same output from the previous figure using a slower baud rate so that the oscilloscope will be able to keep up with the speed.
(a) (b)
FIGURE 9. Transmit Oscilloscope output using (a) standard baud rate, and (b) slower baud rate
Ultimately, the test will be successful if the input is the same with the output. The microcontroller will transmit
the received signal in AIVDM format to the laptop through Bluetooth for comparison of results. Figure 10 shows the
sample AIVDM packet to be received by the laptop. Table 2 shows the comparison of input and output showing that
the boldface text in the output column is the same with the input column.
27
FIGURE 10. Sample Output.
TABLE 2. Comparison of input and output messages shows perfectly matching results.
Input Output Remarks
13Q6o4?P008aqmR8CnJ@0?v20831 !AIVDM,1,1,,A,13Q6o4?P008aqmR8CnJ@0?v20831,0*3E Matched
38JNQT10008a`@F8FtVD16l:0E5r !AIVDM,1,1,,A,38JNQT10008a`@F8FtVD16l:0E5r,0*69 Matched
57c:l841sp0TAD7?C7DDpE8MV10th58U<000001
@0H322t0000000000000000000000000
!AIVDM,2,1,1,A,57c:l841sp0TAD7?C7DDpE8MV10th58U<000001@0H322t00000
00000,0*11!AIVDM,2,2,1,A,000000000000000,2*25 Matched
35?Rbr10018a<LH8FMN6qo>B0000 !AIVDM,1,1,,A,35?Rbr10018a<LH8FMN6qo>B0000,0*15 Matched
18;?bt?P01`ah1>8F2@I7Ov@2D1e !AIVDM,1,1,,A,18;?bt?P01`ah1>8F2@I7Ov@2D1e,0*1C Matched
58JmFP02:KpoUH=GB21Tu=<608QDn22222222
216>`L<=6kV0Bk@CRC0H88888888888880
!AIVDM,2,1,2,A,58JmFP02:KpoUH=GB21Tu=<608QDn22222222216>`L<=6kV0
Bk@CRC0,0*12
!AIVDM,2,2,2,A,H88888888888880,2*56
Matched
18;0NB00008agV48EQ<`Db@J0<19 !AIVDM,1,1,,A,18;0NB00008agV48EQ<`Db@J0<19,0*2C Matched
19t6Su0P018agJD8F0P8POvJ2<1w !AIVDM,1,1,,A,19t6Su0P018agJD8F0P8POvJ2<1w,0*56 Matched
13PO0B001:8aa9b8GKMl;CTJ8D1a !AIVDM,1,1,,A,13PO0B001:8aa9b8GKMl;CTJ8D1a,0*7E Matched
18:bwc?P01`agq`8F1FL??vN289T !AIVDM,1,1,,A,18:bwc?P01`agq`8F1FL??vN289T,0*60 Matched
18;HqT0P028adMB8GB`=9?w:08E` !AIVDM,1,1,,B,18;HqT0P028adMB8GB`=9?w:08E`,0*2C Matched
38?O@J0000`adbT8G6<2J2tP201C !AIVDM,1,1,,B,38?O@J0000`adbT8G6<2J2tP201C,0*6C Matched
18;HqT0P008adM@8GBO==?vT0D1W !AIVDM,1,1,,B,18;HqT0P008adM@8GBO==?vT0D1W,0*57 Matched
405>ND1v9mR?:8aaAJ8FOS1000S: !AIVDM,1,1,,B,405>ND1v9mR?:8aaAJ8FOS1000S:,0*5D Matched
18;?bt?P00`ah1D8F2Aq7OvT2D1f !AIVDM,1,1,,B,18;?bt?P00`ah1D8F2Aq7OvT2D1f,0*4A Matched
18:u70wP00`ah0r8F1u>4?vT2<1n !AIVDM,1,1,,B,18:u70wP00`ah0r8F1u>4?vT2<1n,0*14 Matched
18;K2B0P00`afLv8FORN4?vV24A0 !AIVDM,1,1,,B,18;K2B0P00`afLv8FORN4?vV24A0,0*60 Matched
405>ND1v9mR@D8aaAP8FORQ00<1k !AIVDM,1,1,,B,405>ND1v9mR@D8aaAP8FORQ00<1k,0*18 Matched
38:i3u50008ahv@8F18U8iF`2DUJ !AIVDM,1,1,,B,38:i3u50008ahv@8F18U8iF`2DUJ,0*05 Matched
18;?W2?P02`ah9j8F2VN4?vb2@<S !AIVDM,1,1,,B,18;?W2?P02`ah9j8F2VN4?vb2@<S,0*4C Matched
17c:l80P01`agP@8F1LL4wvb2<1t !AIVDM,1,1,,A,17c:l80P01`agP@8F1LL4wvb2<1t,0*69 Matched
19t6S>hP008ah2r8F5l6Vwver0SU !AIVDM,1,1,,A,19t6S>hP008ah2r8F5l6Vwver0SU,0*2F Matched
19t6Su0P008agJD8F0P`cgvf28=W !AIVDM,1,1,,A,19t6Su0P008agJD8F0P`cgvf28=W,0*10 Matched
33PO0B0Oi48aaE28GK0TDCi08101 !AIVDM,1,1,,A,33PO0B0Oi48aaE28GK0TDCi08101,0*20 Matched
13Q6o4?P008aqmR8CnJ@0?wF00Rk !AIVDM,1,1,,A,13Q6o4?P008aqmR8CnJ@0?wF00Rk,0*78 Matched
33PO0B0Oi18aaKv8GJa4LSmJ8000 !AIVDM,1,1,,A,33PO0B0Oi18aaKv8GJa4LSmJ8000,0*39 Matched
58;0NB01aF4@ATPR221Bu>0d51BpHDhUR0t<6
21S4AA3=5a>0>3@DREQC1Dp88888888880
!AIVDM,2,1,1,A,58;0NB01aF4@ATPR221Bu>0d51BpHDhUR0t<621S4AA3=5a>0
>3@DREQ,0*53
!AIVDM,2,2,1,A,C1Dp88888888880,2*63
Matched
405>ND1v9mRA`8aaAn8FOTA00D1k !AIVDM,1,1,,A,405>ND1v9mRA`8aaAn8FOTA00D1k,0*6E Matched
19t6Su0P008agJb8F0Moe?wD2@Hg !AIVDM,1,1,,A,19t6Su0P008agJb8F0Moe?wD2@Hg,0*64 Matched
58:bwc80000000000000000000000000000000002@
L29t0000000000000000000000008
!AIVDM,2,1,2,A,58:bwc80000000000000000000000000000000002@L29t000000000
0,0*1F
!AIVDM,2,2,2,A,000000000000008,2*2E
Matched
CONCLUSION
In this paper, the implementation of the data transfer protocol of the ITU-R M. 1371 standard through OSS was achieved. The data transfer protocol is subdivided into two sides: transmit and receive. For the transmit side, the input AIS message was converted first into bits following the NRZI format by the XMC microcontroller before transmission. On the other hand, the receiver side decoded the bits in NRZI format to receive the original streamed bits. After transmission and reception, the streamed bits were then converted into NMEA format for comparison of input and output. Based on the results, perfectly matched results were obtained, which validates the implementation.
28
ACKNOWLEDGMENTS
This research work was made possible by supports from Infineon and the Office of the Vice Chancellor for
Research and Innovation of De La Salle University, as well as a grant from Philippine Council for Industry, Energy
and Emerging Technology Research and Development of the Department of Science and Technology.
REFERENCES
[1] T. Eriksen and Ø. Olsen, “Vessel tracking using automatic identification system data in the Arctic,” in
Sustainable Shipping in a Changing Arctic, Springer, 2018, pp. 115–136.
[2] Z. Fang, H. Yu, R. Ke, S.-L. Shaw, and G. Peng, “Automatic Identification System-Based Approach for
Assessing the Near-Miss Collision Risk Dynamics of Ships in Ports,” IEEE Transactions on Intelligent
Transportation Systems, no. 99, pp. 1–10, 2018.
[3] I. Harre, “AIS adding new quality to VTS systems,” The Journal of Navigation, vol. 53, no. 3, pp. 527–539,
2000.
[4] C. E. Page, “Maximizing maritime safety and environmental protection with AIS: (Automatic identification
system),” in OCEANS 2017 - Anchorage, 2017, pp. 1–4.
[5] P. Last, C. Bahlke, M. Hering-Bertram, and L. Linsen, “Comprehensive analysis of automatic identification
system (AIS) data in regard to vessel movement prediction,” The Journal of Navigation, vol. 67, no. 5, pp.
791–809, 2014.
[6] A. Norris, “Automatic Identification Systems–the Effects of Class B on the Use of Class A Systems,” The
Journal of Navigation, vol. 59, no. 2, pp. 335–347, 2006.
[7] F. Xiao, H. Ligteringen, C. Van Gulijk, and B. Ale, “Comparison study on AIS data of ship traffic behavior,”
Ocean Engineering, vol. 95, pp. 84–93, 2015.
[8] Em-Trak, “Em-Trak AIS: www.em-trak-row.com/.” 2018.
[9] Furuno, “Furuno AIS: https://www.furuno.com/en/products/ais/.” 2018.
[10] Simrad, “Simrad AIS: www.navico-commercial.com/en-US/Products/AIS/.” 2018.
[11] R. Essaadali, C. Jebali, K. Grati, and A. Kouki, “AIS data exchange protocol study and embedded software
development for maritime navigation,” in IEEE 28th Canadian Conference on Electrical and Computer
Engineering, 2015, pp. 1594–1599.
[12] D. A. Burgess, H. S. Samra, and others, “The openBTS project,” Report available at http://openbts.
sourceforge. net, http://openBTS. org, 2008.
[13] ITU-R, “ITU-R M. 1371-5-Technical characteristics for an automatic identification system using time-division
multiple access in the VHF maritime mobile band.” 2014.
[14] dAISY, “dAISY: https://github.com/bcl/aisparser/.” 2018.
[15] AIS-Sentence-Parser, “AIS-Sentence-Parser: https://github.com/bcl/aisparser/.” 2018.
[16] E. S. Raymond, “AIVDM/AIVDO protocol decoding, version 1.56.” 2016.
29
1Routing Wavelength Assignment: Novel Heuristic Based
on Bacterial Flagella for Optical Networks
A. B. Rodriguez, M. A. Ganga and L. J. Ramírez
Abstract— Optical transport networks are subject to an incessant demand that generates traffic of dynamic characteristics, and
these networks have a high probability of blocking and high use of the network. Current heuristics allow a reduction of some of
these indicators, but they do not achieve a balanced use of the resources of the network. This research proposes a new heuristic
based on the flagella of the bacteria that use these elements for their motor effect, being able to move in different directions. The
research proposes a heuristic similar to the motor effect of the flagella, managing to find very quickly a solution to the problem of
routing and wavelength assignment in optical networks. The results are significantly better than the comparison heuristics, but the
use of the network is not completely improved, the heuristics are still better in this indicator. On the other hand, a new integrating
indicator is presented that allows to compare the heuristics in an integral way, and the bacterial algorithm has better results than
the rest of the algorithms.
Keywords—RWA, heuristics, metaheuristics, optical networks, WDM
INTRODUCTION
Constant demand for services and the increase of innovations based on data transport over long distances have
triggered traffic in high speed networks, orienting the development of optical networks to unthought of levels. Optical
started with optical fibers, with each thread of the optical fiber carrying one wavelength. The achievable distances
were determined by the loss in the fiber, which should not exceed 0.2 dB/km; soon the WDM technology (Wavelength
Division Multiplexing) appeared, allowing the confinement of various wavelengths in the same optical fiberthread.
Linear phenomena determined the transmission distance, and nonlinear phenomena, and certainly and the nonlinear
phenomena reduced the transmission, which was established at 2.5 Gbps for each wavelength, achieving WDM links
of up to 10 Gbps, and up to 100 Gbps in multifiber links. Innovations in the demand, such as buffering and streaming,
used very widely by the OTT (Over The Top) services increase even more the demand on the transport networks,
choking the systems and orienting the research towards better and more efficient optical transport. Current optical
networks aim to be fully optical, with increasingly fewer electronic layers, whose main problem is the reduction of
the transmission speed caused by the optoelectronic reconversion [1][8].
In monofiber WDM networks, problems beyond the conventional physical problems arise, with the routing playing
a preponderant role because it has no buffer that allows storing the service request awaiting the processing. These
networks were initially designed to satisfy a static demand, and the conventional algorithms were sufficient to solve
this problem. However, this did not last much because the demand grew in such a way that the networks had to provide
dynamic service, giving rise to problems known as RWA (Routing Wavelength Assignment). This problem is
characterized by the fact that the universe of possible solutions is variable, and under those conditions it is not
subjected to optimization, in such a way that it is approachable by solutions the heuristic algorithmia, which does not
aim to optimize this heuristic, but only looks for functional solutions [6]. In the literature we can find diverse heuristic
and metaheuristic solutions such as Genetic Algorithms, Simulated Annealing, and Tabu Search among the most
widely used, and novel metaheuristics like Snake-One, Snake Two and Snake Three, Bat Algorithm, Firefly
Algorithm, among others [7][10][11]. All the mentioned algorithms are compared under two known indicators: The
first one is the blocking probability, which indicates the probability of a request being blocked, i.e., not being attended,
30
and the second is the use of the network, which indicates the utilized capacity under dynamic traffic conditions.
Heuristic algorithms attempt to reduce these indicators without the need for optimizing the routing.
The traffic demand is differentiated by the existing relation between the arrival time and the requested connection
time or Holding Time. Two types of traffic are derived from this, the static one, which is called SLE – Static Lightpath
Establishment, and the dynamic one, which is called DLE – Dynamic Lightpath Establishment. The devices in the
optical networks, like Add/Drop Multiplexers, which are multiplexers with capacity for addition and extraction of
wavelength, MUX, which are only wavelength multiplexers, attenuators, amplifiers, reconverters, etc. The latter allow
distinguishing two types of transport networks, such as networks with wavelength conversion (NWC) or network
without wavelength conversion (NWOC), i.e., that given a lightpath, it can have the same wavelength (without
conversion) or different wavelengths (with conversion).
In the networks that support dynamic traffic, the universe of solutions (possible routes) is variable, and this does
not allow the optimization, and the heuristic algorithms present their greatest contribution; which is manifested by
finding many solutions rapidly and from a random origi [12].
The present study shows a heuristic based on the motion of bacterial flagella, which has been called Bacterial
Algorithm, and it is compared using the conventional indicators with heuristics under the same conditions.
.
ROUTING IN OPTICAL NETWORKS
The heuristic solution is reduced to finding a set of links (route) and their associated wavelength, which is called
Lightpath (LP) (see Equation 1).
𝐿𝑃𝑖𝑗𝑘 = (𝑒𝑖𝑚 → 𝑒𝑛𝑗 , 𝜆𝑖𝑗𝑘) (1)
Where:
𝑒𝑖𝑚 → 𝑒𝑛𝑗 : Set of links between nodes 𝑖 and 𝑗, for the service node 𝑘, 𝑚 as node adjacent to the origin and 𝑛 as node
adjacent to the route’s destination.
𝜆𝑖𝑗𝑘 : Wavelength between nodes 𝑖 and 𝑗, for service node 𝑘.
In RSC networks, the wavelength is the same for each link, and the condition of using the same wavelength is called
Constraint Continuity Wavelength restriction, or CCW, while for RCC networks the wavelength can be different at
each link, and it is called Wavelength Reuse, or WR. For example, Figure 1 shows a route from node 2 to node 5 for
a system of eight nodes labelled from 0 to 7 with four wavelengths per link (𝜆0, 𝜆1, 𝜆2, 𝜆3).
Figura 1. Red optical con LP implementado.
31
The heuristics start their algorithmic process from a population obtained randomly, and then the processes
corresponding to each algorithm. In the case of the bacterial algorithm it is represented by 𝐵0 and it is defined by:
𝐵0 = 𝑏𝑖𝑗𝑘/𝑏𝑖𝑗𝑘 ∀ 𝑖 ∈ [0, 𝑁 − 1] ∧ 𝑗 ∈ [1, 𝑁 + 1] ∧ 𝑘
∈ [0, 𝑁 − 1] (2)
Equation 2 shows the lightpath for the CCW case. When using the wavelength continuity, the WDM-CCW networks
have a low latency when establishing the connection, but the universe of available solutions is exhausted rapidly when
the traffic increases, generating a high congestion in the network, having as an immediate consequence an elevation
of the blocking probability. On the other hand, the WDM-WR networks have an efficient utilization of the network
resources, with a high latency in establishing the connection.
Two strategies can be obtained from the literature: the first one solves the route and the wavelength assignment,
dividing the problem into two parts (SP –Subdivision of the Problem), and the second solves the problem in an integral
manner (IS – Integral Solution). The SP strategy decreases the network’s temporal complexity, but in WDM-CCW
networks the wavelength assignment becomes an obstacle, in general improving the use of the network, but increasing
the blocking probability.
The strategy of the IS, using a single process for getting the LP increases substantially the time for getting the
solution, but improves the use of the network’s resources, competing better in the blocking probability. Furthermore,
various aptitude functions based on the delay per link, the ASE noise, the number of channels, the congestion per link,
the congestion of the most used link, etc., which are maximized or minimized according to the searched criterion.
HEURISTICS IN OPTICAL NETWORKS
Traffic dynamics determine a universe of stochastic and variable over time solutions, under these conditions the
requests arriving at a node must be processed immediately, otherwise they must be blocked, and that is why a fast and
effective solution is required, but not necessarily the optimum one. Heuristic algorithms are more effective in problems
like the DLE, because they use a random initial population, leading to a fast solution.
In what follows we will describe those heuristics, such as: Genetic Algorithms (GA), Simulated Annealing (SIAN)
(Rodriguez et al., 2014), Tabu Search (TASE) (Rodriguez et al., 2014; Li et al., 2018), and the Snake family, among
others. Among the evolutionary heuristics we find one of the first and most important, like the genetic algorithms,
which are based on processes called Reproduction, Ordering, Mortality, Mutation, and Population Control. The initial
population is a table that is obtained randomly, and the elements of a row are mixed with the elements of the adjacent
row, then the rows are arranged in decreasing order, based on an aptitude function that allows convergence toward a
minimum cost. Then the last rows are eliminated by filling randomly, guaranteeing the constant population at each
iteration, or decreasing it according to a population control polícy. This process succeeds in obtaining rapidly low cost
routes that allow determining a set of lightpaths capable of satisfying the request for service [2][5][17]. The Simulated
Annealing algorithm is based on the molecular cooling principle, very similar to those used in martensitic processes
for obtaining alloys with hardening characteristics.
The process depends on the cooling rate. Taken to abstraction, the heating process is the rotation of the peripheral
elements by layers of the rando populational matrix. Arranging in this way in increasing order, at the end of the matrix
we would have the lowest cost routes, and the rotating speed of the uncooled layers should continue decreasing until
a route with a lower cost than a given threshold is achieved. More information is available in [13][15] on the algorithm
and its implementation.
The Tabu Search algorithm moves the population matrix over the columns using a sequence of changes between
columns that rewards those which approach them toward the solution and punishes those that move them far from the
solution. Based on this simple protocol of actions, it is guaranteed that the routes will be formed rapidly and that the
process reaches the solution fast. The results are impressive in other areas which are found in the following papers
[7][9][13][15], which provide more information with respect to the algorithm and its implementation. The Snake
32
family of heuristics was first presented by the research team in 2015 under the name Snake One (SNK1); it then
evolved into Snake Two (SNK2), and finally ended in Snake Three (SNK3). This metaheuristic allows finding the
solution very rapidly, and it is based on horizontal and vertical movements in the network’s costs matrix, like the
motion of a snake. A first modification called Snake Two (SNK2) uses this algorithm monitoring the available capacity
of the links, forcing the use of links to their maximum capacity, in this way using the network’s capacity more
efficiently than its predecessor nodes, forcing the use of the nodal capacity at its congestion threshold, so the traffic is
concentrated in some areas, leaving availability for future demand. These algorithms achieve a sensible decrease of
the blocking probability and administer better the use of the network [13][14].
BACTERIAL ALGORITHM
Minute microorganisms (MOO), which can only be observed under a microscope or forming colonies visible to
the naked eye. This enormous group of living beings includes the viruses, the bacteria, yeasts, and molds that live all
over the world. Their biological structure is incredibly elementary because they are single-celled, but of vital
importancie for the global biological system. Some MOOs are responsible for the degradation of some foods, and they
can even cause serious diseases if they are eaten. Nevertheless, there are other MOOs that are beneficial, and used
properly they can improve the conservation or change some properties of foods. Bacteria in general have a motion
element called flagellum, and bacteria are classified according to their flagella into monotrichous when they have a
single flagellum; lophotricous, cuando when they have several flagella on the same side of the bacteria (pole);
amphitricous when they have flagella on the bacteria’s poles; and perithricous when they have flagella all over their
[3][4]. These flagella are the elements that provide motion to bacteria; in general, they are undulated and even
helicoidal, and by rotating them they succeed in moving in some direction. The heuristic used is based on the
undulation and rotation of the flagella, increasing the monotrichous flagella up to the lophotricous, to improve the
searching power (see figure 2).
Figura 2. Monotrichous and polytrichous bacterial flagella.
Taking as reference the topology shown in figure 3, and assuming that each linkage initially has 16 channels (cost),
elements 𝑏𝑖06 and 𝑏𝑖(13)6 of the bacterial populational matrix correspond to the node of origin and the node of
destination, where 𝑖 and 𝑗 ∈ [0,13]. Equation 2 shows the initial population, where node 8 receives the request for
service toward node 6, and the intermediate elements of the matrix were filled randomly.
33
Figure 3. Example topology with 14 nodes.
Links that are nonexistent in figure 3 are assumed to have a very high cost (999), and under these conditions an
available route will be the one that has a cost lower than the maximum allowed cost (C_mp), so it can be a viable
route.
Figure 4. Bacterial matrix for the 14-node example.
𝐶𝑚𝑝 = (𝑁 − 1)𝐶𝑚𝑒 (3)
Where:
𝑁 : Number of nodes of the topology.
𝐶𝑚𝑒 : Maximum cost of a link, given by the channel’s capacity.
If 𝑁 = 14 and 𝐶𝑚𝑒 = 16, the maximum allowed cost 𝐶𝑚𝑝 = 208. Looking at the routes (rows) of the initial
population (Figure 4), they are more than 208. For example, row 0, or the first row (8 – 2 – 7 – 11 – 9 – 10 – 0 – 6 –
3 – 6 – 0 – 2 – 3 – 6) has the cost of 11,021 (11x999+2x16) because it has 11 links that do not exist. Then the processes
of ordering, definition of the number of flagella, elimination of high costs, recalculation of costs according to the
bacterial algorithm that improves rapidly the aptitude function are carried out; it must be mentioned that this
methodology only finds good routes that are not necessarily the optimum route.
Four variables are used to define the flagella: 𝑛𝐹 indicates the number of bacterial flagella; 𝑑𝐹 indicates the diameter
of the flagella; 𝑣𝐹 indicates the rotating speed of the bacterial flagella; and 𝑝𝐹 indicates the rotational power of the
34
flagella. Figure 5 shows the rotational process of the flagella, where the four descriptive variables are subjected to
optimization before the simulation.
Figure 5. Flowchart of the bacterial algorithm.
Figure 6. Flagellar rotation process.
After each rotation the costs are recalculated by row, arranging the rows in increasing order with respect to the costs
column, then a number of high cost rows are deleted (𝑒𝐹) and again they are refilled with random values. The rows
of the bacterial population do not change, following a constant growth policy. The costs are calculated again and they
are arranged in ascending order, and the cycle is repeated until the previously indicated C_me is fulfilled. Figure 6
shows the flow diagram of the bacterial heuristic, where matrix E is found, which has the network’s topology, matrix
C which monitors the costs associated to each wavelength and link, the lambda matrix that monitors the wavelength
in operation and allows the determination of the availability, the time T matrix, which monitors the holding time of
the link, i.e., the operating time of the link, and the bacterial matrix that allows the calculation of the lightpath.
35
SIMULATION SCENARIO
The simulation was carried out using the topology of the NSFNET (Network Science Foundation NETwork) shown
in Figure 7, the same that uses 14 nodes (OXC – Optical Cross Connect), eight wavelengths, and 21 WDM optical
links; this network can generate up to 20 Gbps under the conditions mentioned above.
Figure 7. NSFNET optical network, with 14 nodes, 21 links, and eight wavelengths.
Figure 7 shows the OXC optical switches distributed along the territory of the United States. The low concentration
of links per node is noteworthy, because it allows two improvements: the first one is the decreased blocking probability
by decreasing the number of node requests, and the second one improves the implementation of the network by using
the smallest number of links, at the same time that the operating and maintenance costs decrease drastically, this
configuration is used to guarantee equal conditions to compare the effectiveness of the heuristic algorithms in
operation. For the simulation it was considered that the traffic is distributed uniformly among all the pairs of nodes,
the holding time is distributed exponentially, with a mean of 100 ms; the time to configure and test a connection is
500 µs., the time for processing messages at a node is 10 µs; the uniform traffic and the average propagation between
two nodes is 14.7 m.; the signaling is inside, i.e., it uses the same lightpath; and the time for transmitting or changing
a network control packet (signaling) is 0. The simulation was carried out with 108 requests, reaching loads of up to
180 erlangs. This load was calculated with the average rate of incoming requests by the average connection time,
allowing a scan every 5 erlangs.
RESULTS AND GRAPHS
To make the algorithmic comparison use will be made of two indicators like the Blocking Probability (BP), which
measures the probability of attending an incoming request; the network use (NU), which measures the percentage of
the capacity used to improve the heuristic, including another indicator (see Equation 5) called the blocking rate (BR),
which relates the mean value of the BP and the mean value of the NU. The maximum value is 10,000 when BP is 1
(there is no service in the nodes) and 100% of use of the network (saturated network).
BR = 106BP
NU (4)
Table 1
Distribution of the blocking rate per heuristic.
AGEN SIAN TASE SNK1 SNK2 SNK3 BAC
9.615 9.571 6.233 4.468 3.915 4.383 3.984
36
Table 1 shows that the bacterial heuristic (BAC) succeeds in decreasing the blocking rate with respect to the rest of
the heuristics, assuming a process that improves the use and service of the optical network. This important
characteristic is due mainly to the speed with which the algorithm finds the solution, but in contrast with the rest of
the heuristics, it finds only one solution, while others like AGEN, SIAN, and TASE, which in the same algorithmic
process find more than one solution, giving robustness to the routing system in case of faults in the communication.
Figure 8. Distribution of the blocking probability.
Figure 8 shows the blocking probability distribution of the heuristics, where the BAC heuristic succeeds in reducing
sensibly the blocking probability, which remains very low up to 130 erlangs; once it crosses this threshold it behaves
like the rest of the heuristics, but it continues being smaller.
Fig. 9. Distribution of the network’s utilization.
Figure 9 shows the better performance of the NU with respect to the SNK heuristics, but it does not improve on the
rest of the heuristics, meaning that the cost of obtaining low blocking probabilities is an increased use of the network.
From Figures 6 and 7 it is found that there are two important load landmarks: the first one at 30 erlangs, where the
heuristic changes the NU slope; below this threshold the slope is much greater than above it; the second load landmark
is found at 120 erlangs, where the blocking probability is quite low with respect to the blocking probabilities beyond
this threshold. This implies that there are two stress landmarks in the networks, one due to blocking probability that
is too soon, i.e., at 30 erlangs, and the other is from the use of the network at 120 erlangs.
37
CONCLUSION
The WDM optical networks facing the high data transfer demand are suffocated, and now more than ever require
all their processes to be optimized or improved. Routing and wavelength assignment is one of the processes that are
constantly studied and evolving. The heuristics studied in this research, denoted by AGEN, SIAN, TASE, SNK1,
SNK2, and SNK3, show their performance through the BP, UN, and BR indicators. The proposed heuristic, called
Bacterial Algorithm, improves sensibly the BP, but at a high cost of the UN. However, when the blocking rate is
analyzed as a transverse comparison element, the BAC heuristic improves the performance of the rest of the compared
algorithms. Nowadays optical networks are evolving toward multifiber optical networks, so these algorithms will be
tested in new scenarios that incorporate the multifiber element in the links of the optical networks, expecting that
performance to help improve the capacity of these new networks.
ACKNOWLEDGEMENTS
Special thanks are due to Project Code 081872RG, Dirección de Investigación, Científica y Tecnológica (DICYT)
of the Universidad de Santiago de Chile; to Fortalecimiento USACH USA1799_GM181622 Project; to Project IMP-
ING-2660 of the Vicerrectoría de Investigaciones of the Universidad Militar Nueva Granada, Colombia, and to the
Grupo de Investigación en Nuevas Tecnologías (GINT-DTI-USACH) for their importante support for the
development of the research.
REFERENCES
[1] Assis K. D. R., Ferreira dos Santos A. and Giozza W. F., Hybrid Algorithms for Routing and Assignment Wavelengths in Optical Networks.
Latin America Transactions, vol 8(1), pp. 214-220. 2010 [2] Barpanda R. S., Sahoo B., Turuk A. K. and Majhi B., Genetic Algorithm techniques to solve Routing and Wavelength Assignment problem in
Wavelength Division Multiplexing all-optical networks. Communication Systems and Networks (COMSNETS), vol 1, pp. 1-8. 2011
[3] Belas Roberts, Biofilms, flagella, and mechanosensing or surfaces by bacteria, Cell Press, TIMI-1096, Trends in Microbiology, pp. 1–11. 2014 [4] Bonnie C. and Hughes M, The flagellum in bacterial pathogens: for motility and a whole lot more, PII: S1084-9521(15)00230-X DOI:
http://dx.doi.org/doi:10.1016/j.semcdb.2015.10.032 Reference: YSCDB 1863. 2015
[5] Brandao J., Noronha T. and Ribeiro C., A biased random-key genetic algorithm to maximize the number of accepted lightpaths in WDM optical networks, Journal Of Global Optimization, vol 65(4), pp. 813-835. 2016
[6] Brigitte J. and Daryalal M., Efficient Spectrum Utilization in Large Scale RWA Problems, IEEE-ACM Transactions On Networking, vol 25(2), pp. 1263-1278. 2017
[7] Duhamel C., Mahey P., Martins A., Saldanha R. and de Souza M., Model-hierarchical column generation and heuristic for the routing and
wavelength assignment problem, 4OR, vol 14(2), pp. 201-220. 2016 [8] Koganti R. and Sidhu D., Analysis of routing and wavelength assignment in large WDM networks, Procedia Computer Science, vol 34, pp.
71-78. 2014
[9] Li X., Zhu L. and Baki F., A. Chaouch, Tabu search and iterated local search for the cyclic bottleneck assignment problem, Computers and Operations Research, vol 96, pp. 120-130. 2018
[10] Marković G., Artificial bee colony approach for routing and wavelength assignment in optical WDM networks, Intelligent Systems for Optical
Networks Design: Advancing Techniques, pp. 174-193. 2013 [11] Montes-Castañeda B., Patiño-Garzón J. and Puerto-Leguizamón G., A comparative study of multiobjective computational intelligence
algorithms to find the solution to the RWA problem in WDM networks, DYNA, vol 82(194), pp. 221-229. 2015
[12] Pereverzev A. and Ageyev D., Heuristic method of adaptive routing for solving dynamic RWA, CriMiCo 2013 International Crimean Conference Microwave and Telecommunication Technology, Conference Proceedings, vol 6652921, pp. 496-497. 2013
[13] Rodríguez, A., L. Ramirez and J. Martinez, Heuristic algorithms: Novel solution for RWA problem in WDM optical networks, Journal of
Telecommunication, Electronic and Computer Engineering, vol 10(1-4), pp. 1-7. 2018 [14] Rodriguez A., Ramirez L. and Basile F., Metaheuristic solution for the routing and wavelength assignment (RWA) problem in wavelength-
division multiplexing (WDM) optical networks, Informacion Tecnologica, vol 28(6), pp. 133-146. 2017
[15] Rodriguez A., Gutierrez A., Rivera L. and Ramirez L., Ruteo y Asignation de Longitud de Onda: Comparation de Algoritmos Genéticos y Simulated Annealing. Inf. tecnol., vol 4, pp. 13-18. 2014
[16] Rodriguez A., Ramirez L., Rivera L. y Gutierrez A., Routing wavelength assignement: A solution with tabu search in dynamic traffic. Ingeniare.
Rev. chil. ing. 4, 495-503(2014). [17] Zakouni A., Luo J. and Kharroubi F., Genetic algorithm and tabu search algorithm for solving the static manycast RWA problem in optical
networks, Journal of Combinatorial Optimization, vol 33(2), pp. 726-741. 2017
38
fNIRS-Measured Impact of Expertise Development on
Severity of Mental Stress: A Neuroergonomic Approach
Emad A. Alyan1, a), Naufal M. Saad1, b), Nidal S. Kamel1, c) and Fares Al-Shargie2, d)
1Centre for Intelligent Signal and Imaging Research (CISIR), Department of Electrical and Electronic Engineering,
Universiti Teknologi PETRONAS (UTP), Bandar Seri Iskandar, Perak Malaysia. 2Biosciences and Bioengineering Research Institute Department of Electrical Engineering, American University of
Sharjah (AUS), United Arab Emirates.
b)Corresponding author: [email protected] a)[email protected], c)[email protected], d)[email protected]
Abstract. This study aimed to investigate the impact of expertise development in workplace context by measuring
hemodynamic response on the frontotemporal regions. Chronic disorders and productivity loses are two of the major
mental stress factors that influence workers in the workplace. Evaluating mental stress based on neuroergonomics
approach becomes increasingly relevant for improving operator’s performance and efficacy. In this study, functional
near-infrared spectroscopy (fNIRS) measurement was performed on twelve healthy participants of two equal groups
(stress-trained and control) to investigate stress coping mechanisms during stressful complex cognitive tasks. The results
revealed a higher cortical activation level of oxygenated hemoglobin in stress-trained group as compared to control group
who showed highly stress effects on left anterior-superior temporal region, dorsomedial PFC and right ventrolateral PFC
areas. Using support vector machine (SVM), the accuracy of discrimination of stress-trained group is 96.25%. The study
demonstrated the feasibility of neuroergonomics in enhancing human capability and reducing stress level.
Keywords— Neuroergonomics; Mental Stress; fNIRS; PFC
INTRODUCTION
Human work-related mental stress plays a huge role in the mental health of the workers as major factor
contributing to chronic disorders, productivity loses and cost burden [1]. This occurs when workers believe the
demands of the work override their resources and skills, negatively affecting the physiological, psychological,
behavioral and social domains [2]. Stress causes a rapid activation of the sympathetic nervous system (SNS)
accompany with the release of norepinephrine from widely distributed brain network to the PFC [3]. Stress also
involves the activation of HPA, which regulates the release of glucocorticoids. The SNS system, via catecholamines
signaling and interacting with glucocorticoids (HPA system), plays a key role in both normal homeostasis and in
sympathetically mediated responses to stress [4]. Animal and human studies have demonstrated detrimental effects
of glucocorticoids on PFC functioning [5]. In particular, in the top-down control process, PFC is thought to be the
primary center regulating deeper brain structures such as amygdala. Therefore, PFC cognitive process could be
biased under psychological stress. A variety of external stress treatments have shown to alter the PFC functioning
acutely in animals and human subjects [6, 7].
In order to reduce mental stress and reach a high degree of performance and efficiency, neuroergonomics
approach was introduced. Neuroergonomics is an emerging field in ergonomics (human factors) and neuroscience
that understands human brain and behavior in relation to performance at work and everyday settings [8, 9]. It deals
with understanding real-life contexts, and tasks performed in natural real-world environments. It offers a reliable
assessment of human mental stress through measurement of brain hemodynamic or electromagnetic activity [8, 9].
Given the importance of evaluating and measuring worker’s mental stress, an accurate assessment could help to
prevent stress-related health problems in the workplace [10-12].
39
Nowadays, advanced functional modalities are gaining more attention for their potential to identify cortical
activity associated with cognitive mental tasks, other neurological disorders [13], and to assess brain behavior in the
work environment. Among these modalities, functional near-infrared spectroscopy (fNIRS) has been widely utilized
to investigate the hemodynamic changes over local points on the head, i.e. blood oxygenation and volume, during
select cognitive tasks [14]. These characteristics demonstrate fNIRS as a good applicant for monitoring cognitive
activity-related hemodynamic changes in real world environments [15]. Several studies have used fNIRS to
objectively evaluate cortical hemodynamic activity. For instance, Tsunashima et al. reported that PFC activity
increases while increasing task complexity. This direct correlation was also corresponded to fMRI findings [16].
Similarly, PFC and occipital behavior with and without an automated driving system have been assessed by a train
simulator. Such study demonstrated lower PFC activity during automated system [17]. On the other hand, negative
emotion can be induced by stressful task associated with time pressure and feedback. That may impair the
performance of working memory which leads to reduction in the firing rate of neurons [18]. By way of illustration,
Tian et al. found reduction in the oxygenated hemoglobin at the dorsolateral PFC (DLPFC) in subjects with post-
traumatic stress disorder [19, 20]. Likewise, cortical activities reduction was also observed by [21] during inhibiting
fear response.
To date, researchers have worked on the development of expertise through training and practicing, to facilitate
optimal performance in stressful and complex situations that is clearly predicted by performance improvement [15,
22, 23]. As experienced people can maintain their performance under stressful situation by adapting their decision
strategies, setting up priorities and adjusting their communication patterns [24]. Because experience familiarizes
people with relevant stressors such as task-load, time constraints, noise, and other stressors, and thereby minimize
feelings of ambiguity and uncertainty [25-27]. Therefore, practicing the tasks under similar operational
circumstances to those in the real-world setting should reduce the degree of performance decrement encountered in
the actual operational setting [28]. In that, brain activity measurement using neuroimaging technologies (e.g. fNIRS)
while having stressful situation with advanced practice of stressors can provide a practical evidence for the existing
literature that used subjective methods [27, 29].
Given that time pressure and negative feedback are two of the most laboriously stressors in the workplace.
Where time pressure is time restriction necessary to perform a task that might lead to an impaired performance due
to the increased cognitive load [30]. Negative feedback is also used to increase the level of stress on the participants.
Both stressors have been combined into a single task, called Montreal Imaging Stress Task (MIST). MIST has been
used in the laboratory as a psychosocial stress task that increases cortisol levels and evokes differential brain activity
when compared to control status, an identical task without the time pressure and negative feedback [31]. As
reported, the MIST (versus control condition) has been identified in the stress literature as an effective source of
mental stress that can be induced on the PFC [32-34]. Under the MIST, a significant deactivation was observed,
during fMRI and PET scanning, in the brain limbic system including the medio-orbitofrontal cortex, hippocampus,
hypothalamus and amygdala [35]. Dagher and colleagues also used the MIST to examine the consequences of stress
on brain activity for nicotine-dependent individuals [36]. Revealing related dysfunctions in the PFC of healthy
individuals to adapt to psychological stressors may provide the insight necessary to identify individuals’ capability
to the assigned task. The current paper investigates the effects of expertise development through task practice on
stressed participants by means of measuring brain hemodynamic response using fNIRS.
MATERIALS AND METHODS
A. Participants
A total of twelve males, right-handed adults (age between 20- 30) participated in this study. All participants were
informed and gave written consents prior to the experiment. They had no history of psychiatric, neurological illness
or psychotropic drug use.
40
B. Experiment Protocol
The experiment was developed based on mental arithmetic task. The arithmetic task involved 3 integers between
the range of 0 to 99 with at least 2 two-digit integers. The operators used in the arithmetic calculations were
presented randomly using the operands of +, –, and x (i.e. 18x2-30). Participants were randomly assigned to two
groups of six people, either a control group or stress-trained group. Control group practiced the arithmetic task
without stressors for duration of three-hours before the actual experiment. Stress-practiced group received identical
practice but in the presence of time pressure and negative feedback as stressors. The average time recorded during
the practice of control group was reduced by 10%, and set as time pressure. Feedbacks, such as “correct”,
“incorrect”, and “timeout”, appeared to the participants based on their performance of solving the arithmetic tasks.
Both groups then performed a stressful mental arithmetic task with identical feedbacks as in the practice phase of the
stress-practiced group, and a fixed-time pressure of 90% of the average-time taken during the practice phase.
The experiment was conducted in a room with a good lighting and temperature of 20 Celsius. The design of the
workstation was good to meet the comfortability of real-workplace environment. In this study, the arithmetic task
was presented in block design to ensure the activation of hemodynamic functions (ten blocks were used) [37]. In
each block, the arithmetic task was produced for 30s followed by 20s rest. During the 20s rest condition, all
participants were instructed to keep relax and focus on fixation cross on the monitor display. Fig.1 shows the task
sequence in block design.
FIGURE 1. Experiment design of mental stress under time pressure and negative feedback. A total of ten active blocks and
each block contains arithmetic tasks for 30 s followed by 20 s rest.
C. fNIRS Measurement and Preprocessing
In this study, we utilized multi-channel fNIRS system (OT-R40, Hitachi Medical Corporation, Japan) to measure
the hemodynamic responses over the frontotemporal regions. This system can determine relative changes of
absorbance oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) by using two wavelengths (695
and 830 nm), based on a modified Beer-Lambert law [38]. The used fNIRS probe was fixed to 3×11 thermoplastic
shells that consist 17 emitters to transmit the near-infrared lights to 16 detectors through 52 channels at sampling
rate of 10 Hz. The distance of the channel between the source and detector was 3cm. The probe was arranged on the
skull according to the international EEG 10–20 system and adjusted to the electrode position Fpz.
We mapped the fNIRS channels on ten different regions of interest over the bilateral frontotemporal regions [39]
as shown in Fig.2. These regions of interest included the dmPFC (Ch5, Ch6, Ch15, Ch16, Ch17, Ch26, Ch27),
vmPFC (Ch36, Ch37, Ch38, CH47, Ch48), right dlPFC (Ch3, Ch4, Ch13, Ch14, Ch24, Ch25), left dlPFC (Ch7,
Ch8, Ch18, Ch19, Ch28, Ch29), right vlPFC (Ch34, Ch35, Ch45, Ch46), left vlPFC ( Ch39, Ch40, Ch49, Ch50),
right psPF (Ch1, Ch2, Ch11, Ch12), left psPF (Ch9, Ch10, Ch20, Ch21), right asTC (Ch22, Ch23, Ch32, Ch33,
Ch43, Ch44) and left asTC (Ch30, Ch31, Ch41, Ch42, Ch51, Ch52).
41
The hemodynamic signals were pre-processed to remove low-frequency physiological change and high-
frequency systemic noise using Butterworth bandpass filter between 0.01 to 0.5 Hz. The motion artefacts were
removed using independent component analysis method available in the developed Platform for Optical Topography
Analysis Tool (POTATo) [40]. The data were decomposed into components (in this case 52 component) and one
component that showed artefact was removed and the other components were reconstructed to form the clean
signals. Ten analysis blocks were defined from the onset of the task condition to the end of the rest condition (period
of 30 seconds of mental arithmetic task and 20 seconds of rest). Each block was baseline-corrected by subtracting
the mean value of the 20s rest condition from each of the data point in the task condition.
FIGURE 2. (A) The schematic diagram of fNIRS probes. The probes arrays are according to the international
10/20 system. The Red and blue circles indicate NIRS emitters and detectors, respectively. (B) NIRS probe set
with 52 channels were placed over bilateral frontotemporal regions.
D. Feature Extraction and Classification
In this study, the peak values and mean of the ΔHbO concentration, and their combination were extracted over
the analysis block using moving time-window of 500 ms [41]. As in the literature, several studies reported that the
peak values of ΔHbO signal performed best in the fNIRS [42, 43]. Also, previous studies found that combining two
or more features could improve the classification accuracies, in particular the combination of mean and peak values
of HbO [44, 45]. According to Naseer et al., the combinations containing signal mean and signal peak gave the
highest classification accuracy than other feature combinations [46] . The signal peak is estimated using the max
function in MATLAB, and the signal mean is calculated as follows:
(1)
42
where M is the length of HbO signal, and ΔHbO represents the segmented HbO signal.
These features were then fed into support vector machine (SVM) classifier using radial basis function kernel
[47], as per the following equation:
(2)
Where x and xi are the two feature vectors, and represents the width of RBF.
The k-fold method has been used to perform the cross-validation testing. The obtained features were randomly divided into ten groups, (10- fold cross validation) one group used for testing and the other nine groups used for training the SVM classifier. The process repeated ten times until each group had the chance of being in the test and training space.
RESULT AND DISCUSSION
fNIRS Data Analysis
This study examined the impact of stress-training on the neurophysiological measures of the hemodynamic
response on frontotemporal cortices while the subjects (twelve healthy subjects) were performing MIST, comparing
stress-trained group with control group. The results of the HbO demonstrated significantly differences in brain
activities between the stress-trained group and control group. Also, a significant improvement in the activation of
HbO, related to stress-trained group over control group, was observed in most of the channels located on the left
hemispheric and few channels on the right hemisphere. The observed activation in the left hemisphere could be
explained due to the arithmetic task which has been proven to highly activate the left hemisphere of the PFC. Fig. 3
and 4 show the topographic maps of prefrontal oxygenated hemoglobin (HbO) responses during control group and
stress-trained group, respectively. The labels on the topographic images with red color shows high activation and
blue color shows less activation. According the topographic image of Fig.3, specific regions in the left and right
hemispheres were deactivated in the case of control group. These regions are at the left psFC (Ch20, Ch21), left
asTC (Ch30, Ch31, Ch42), left vlPFC (Ch40), dmPFC (Ch6, Ch16, Ch26), right psFC (Ch1), asTC (Ch23) and right
vlPFC (Ch35, Ch45). On the other hand, in the case of stress-trained group, most of the channels measured were
significantly activated with increased HbO concentration as shown in Fig. 4. These activated channels were located
on the left psFC (Ch9, Ch10, Ch20, Ch21), left asTC (Ch30, Ch31, Ch42, Ch52), left dlPFC (Ch8, Ch18, Ch19,
Ch29), left vlPFC (Ch39, Ch40, Ch49, CH50), vmPFC (Ch36, Ch47) and right vlPFC (Ch46). The study found a
positive correlation between the HbO concentration with stress-trained group, unlike the negative correlation with
the control group which is consistent with previous stress studies [18, 35, 41]. This demonstrated that, stress training
during work training could enhance the brain activities, and thus improves the performance in a stressful task.
Fig. 3 and 4 show also the block-averaged time courses of HbO (red line) and HbR (blue line) for stress-trained
group and control group, respectively, across selected channels of the measured brain region. These channels
represent the highest activation (left dlPFC; Ch19), average activation, (right dlPFC; Ch13 & left asTC; Ch41), and
less activation (right asTC; Ch33) in the stress-trained group. Additionally, a dramatic increase in HbO
concentration level associated with a slight decrease in the HbR, can be clearly observed, in relative to the baseline
over the selected channels. In contrast, Fig. 3 shows significant decline of HbO concentration in all selected
channels during the control group. After the task ended, the HbO concentration returned to the baseline again at the
rest condition. The experimental results in Fig. 3 and 4 further demonstrated that the HbO in the left PFC area is
highly increased in stress-trained group.
43
FIGURE 3. Pre-frontal cortical distribution of Oxy-Hb for control group during mental arithmetic task.
Oxy-Hb (red) and Deoxy-Hb (blue) hemodynamic responses.
FIGURE 4. Pre-frontal cortical distribution of Oxy-Hb for stress-trained group during mental arithmetic
task. Oxy-Hb (red) and Deoxy-Hb (blue) hemodynamic responses.
Statistical Analysis
The statistical analysis of t-test showed that, 40 channels showed significant differences in brain activations
between stress-trained group and control group with mean p-value of less than 0.05. This shows that, control group
was highly sensitive to stress exposure and reports stress-training as a good index to maintain human capability.
Only few channels [Ch2, Ch32, Ch33, Ch43, Ch28, Ch29, Ch38, Ch39, Ch49, Ch50-Ch52] did not show significant
differences with most of them located on the left ventral PFC area. Using SVM classifier, brain activity, for
44
discriminating stress-trained group from that of control group, was evaluated based on accuracy, sensitivity,
specificity and ROC. The results are shown in Table 1, and the typical ROC curves of the individual and combined
features are shown in Fig.5. Given the mean as feature, accuracy, sensitivity, specificity and AROC are equal to
88.4%, 88.8%, 88.3% and 95.4%, respectively. In contrast, peak value offers higher performance with accuracy,
sensitivity, specificity and AROC of 90.8%, 91.2%, 90.4% and 96.8%, respectively. Combing both features (mean
and peak value) gave the highest performance of accuracy, sensitivity, specificity and ROC with 96.25%, 96.0%,
96.4% and 99.4%, respectively (Fig. 5). The inclusion of features combination significantly increased the accuracy,
sensitivity, specificity and AROC by 5.7%, 5.0%, 6.2% and 2.6%, respectively, more than with the individual
features (peak value) classifier. It is noticeable that the designed task gave the optimum results by achieving higher
classification of fNIRS-based arithmetic task as compared with recent studies [48-55]. This confirmed the
hypothesis that combing features such as mean and peak value of HbO could improve the classification accuracies.
TABLE 1. Evaluation parameters of the SVM classifier.
Feature Accuracy% Sensitivity% Specificity% AROC%
Mean 88.4 88.8 88.3 95.4
Peak 90.8 91.2 90.4 96.8
Mean+Peak 96.25 96.0 96.4 99.4
FIGURE 5. ROC curves of SVM classifier. Red line represents the ROC of combined features,
black line represents the ROC of peak values and blue line represent the ROC of mean values.
CONCLUSION
The study investigated the effects of expertise development through task practice as a human factor on mental
stress responses. The results demonstrated significant improvement in human capability as indicated by increasing
the activation level of oxygenated hemoglobin in stress-trained group as compared to control group. The results also
demonstrated that, people without stress training are highly prone to stress exposure. The effects of stress on control
group was highly localized to the left anterior-superior temporal region, dorsomedial PFC, and right ventrolateral
45
PFC areas. The results of SVM classifier showed that, stress-trained group can be discriminated with an average
accuracy of 96.25%. The overall study reported the experience as a good factor to maintain human capability and
reduce stress level.
ACKNOWLEDGMENTS
The authors would like to take this opportunity to thank Universiti Teknologi Petronas (UTP) for the financial
support.
REFERENCES
[1] J. Hassard, K. Teoh, T. Cox, P. Dewe, M. Cosmar, R. Gründler, et al., "Calculating the cost of work-related
stress and psychosocial risks," 2014.
[2] J. E. Driskell and E. Salas, Stress and human performance: Psychology Press, 2013.
[3] S. Qin, E. J. Hermans, H. J. van Marle, J. Luo, and G. Fernández, "Acute psychological stress reduces
working memory-related activity in the dorsolateral prefrontal cortex," Biological psychiatry, vol. 66, pp.
25-32, 2009.
[4] V. Engert, S. Vogel, S. I. Efanov, A. Duchesne, V. Corbo, N. Ali, et al., "Investigation into the cross-
correlation of salivary cortisol and alpha-amylase responses to psychological stress,"
Psychoneuroendocrinology, vol. 36, pp. 1294-1302, 2011.
[5] B. S. McEwen, "The neurobiology of stress: from serendipity to clinical relevance1," Brain research, vol.
886, pp. 172-189, 2000.
[6] A. F. Arnsten, "Catecholamine modulation of prefrontal cortical cognitive function," Trends in cognitive
sciences, vol. 2, pp. 436-447, 1998.
[7] A. F. Arnsten and P. S. Goldman-Rakic, "Noise stress impairs prefrontal cortical cognitive function in
monkeys: evidence for a hyperdopaminergic mechanism," Archives of general psychiatry, vol. 55, pp. 362-
368, 1998.
[8] R. Parasuraman and M. Rizzo, Neuroergonomics: The brain at work: Oxford University Press, 2008.
[9] R. Parasuraman, "Neuroergonomics: Research and practice," Theoretical issues in ergonomics science, vol.
4, pp. 5-20, 2003.
[10] T. G. Pickering, "Mental stress as a causal factor in the development of hypertension and cardiovascular
disease," Current hypertension reports, vol. 3, pp. 249-254, 2001.
[11] J. Wijsman, B. Grundlehner, H. Liu, H. Hermens, and J. Penders, "Towards mental stress detection using
wearable physiological sensors," in Engineering in Medicine and Biology Society, EMBC, 2011 Annual
International Conference of the IEEE, 2011, pp. 1798-1801.
[12] S. B. Scott, J. E. Graham-Engeland, C. G. Engeland, J. M. Smyth, D. M. Almeida, M. J. Katz, et al., "The
effects of stress on cognitive aging, physiology and emotion (ESCAPE) project," BMC psychiatry, vol. 15,
p. 146, 2015.
[13] H. Bleton, S. Perera, and E. Sejdić, "Cognitive tasks and cerebral blood flow through anterior cerebral
arteries: a study via functional transcranial Doppler ultrasound recordings," BMC medical imaging, vol. 16,
p. 22, 2016.
[14] A. Villringer and B. Chance, "Non-invasive optical spectroscopy and imaging of human brain function,"
Trends in neurosciences, vol. 20, pp. 435-442, 1997.
[15] H. Ayaz, M. P. Çakir, K. İzzetoğlu, A. Curtin, P. A. Shewokis, S. C. Bunce, et al., "Monitoring expertise
development during simulated UAV piloting tasks using optical brain imaging," in Aerospace Conference,
2012 IEEE, 2012, pp. 1-11.
[16] H. Tsunashima and K. Yanagisawa, "Measurement of brain function of car driver using functional near-
infrared spectroscopy (fNIRS)," Computational intelligence and neuroscience, vol. 2009, 2009.
[17] T. Kojima, H. Tsunashima, T. Shiozawa, H. Takada, and T. Sakai, "Measurement of train driver’s brain
activity by functional near-infrared spectroscopy (fNIRS)," Optical and Quantum Electronics, vol. 37, pp.
1319-1338, 2005.
46
[18] F. Al-Shargie, T. B. Tang, and M. Kiguchi, "Assessment of mental stress effects on prefrontal cortical
activities using canonical correlation analysis: an fNIRS-EEG study," Biomedical optics express, vol. 8, pp.
2583-2598, 2017.
[19] A. F. Arnsten, M. A. Raskind, F. B. Taylor, and D. F. Connor, "The effects of stress exposure on prefrontal
cortex: Translating basic research into successful treatments for post-traumatic stress disorder,"
Neurobiology of stress, vol. 1, pp. 89-99, 2015.
[20] F. Tian, A. Yennu, A. Smith-Osborne, F. Gonzalez-Lima, C. S. North, and H. Liu, "Prefrontal responses to
digit span memory phases in patients with post-traumatic stress disorder (PTSD): a functional near infrared
spectroscopy study," NeuroImage: Clinical, vol. 4, pp. 808-819, 2014.
[21] T. Jovanovic, T. Ely, N. Fani, E. M. Glover, D. Gutman, E. B. Tone, et al., "Reduced neural activation
during an inhibition task is associated with impaired fear inhibition in a traumatized civilian sample,"
Cortex, vol. 49, pp. 1884-1891, 2013.
[22] H. Ayaz, P. A. Shewokis, S. Bunce, K. Izzetoglu, B. Willems, and B. Onaral, "Optical brain monitoring for
operator training and mental workload assessment," Neuroimage, vol. 59, pp. 36-47, 2012.
[23] R. McKendrick, H. Ayaz, R. Olmstead, and R. Parasuraman, "Enhancing dual-task performance with
verbal and spatial working memory training: continuous monitoring of cerebral hemodynamics with
NIRS," Neuroimage, vol. 85, pp. 1014-1026, 2014.
[24] T. Kontogiannis and S. Malakis, Cognitive engineering and safety organization in air traffic management:
CRC Press, 2017.
[25] G. Keinan, N. Friedland, and V. Sarig‐Naor, "Training for task performance under stress: The effectiveness
of phased training methods," Journal of Applied Social Psychology, vol. 20, pp. 1514-1529, 1990.
[26] T. Kontogiannis, "Training effective human performance in the management of stressful emergencies,"
Cognition, Technology & Work, vol. 1, pp. 7-24, 1999.
[27] C. K. McClernon, M. E. McCauley, P. E. O’Connor, and J. S. Warm, "Stress training improves
performance during a stressful flight," Human factors, vol. 53, pp. 207-218, 2011.
[28] J. E. Driskell and J. H. Johnston, "Stress exposure training," Making decisions under stress: Implications
for individual and team training, pp. 191-217, 1998.
[29] J. E. Driskell, J. H. Johnston, and E. Salas, "Does stress training generalize to novel settings?," Human
factors, vol. 43, pp. 99-110, 2001.
[30] M. A. DeDonno and H. A. Demaree, "Perceived time pressure and the Iowa Gambling Task," Judgment
and Decision making, vol. 3, p. 636, 2008.
[31] K. Dedovic, R. Renwick, N. K. Mahani, V. Engert, S. J. Lupien, and J. C. Pruessner, "The Montreal
Imaging Stress Task: using functional imaging to investigate the effects of perceiving and processing
psychosocial stress in the human brain," Journal of Psychiatry and Neuroscience, vol. 30, p. 319, 2005.
[32] F. Al-Shargie, T. B. Tang, and M. Kiguchi, "Stress Assessment Based on Decision Fusion of EEG and
fNIRS Signals," IEEE Access, vol. 5, pp. 19889-19896, 2017.
[33] D. Rosenbaum, P. Hilsendegen, M. Thomas, F. B. Haeussinger, F. G. Metzger, H.-C. Nuerk, et al.,
"cortical hemodynamic changes during the Trier Social Stress Test: An fnirs study," NeuroImage, vol. 171,
pp. 107-115, 2018.
[34] F. Al-shargie, T. B. Tang, and M. Kiguchi, "Mental stress grading based on fNIRS signals," in Engineering
in Medicine and Biology Society (EMBC), 2016 IEEE 38th Annual International Conference of the, 2016,
pp. 5140-5143.
[35] J. C. Pruessner, K. Dedovic, N. Khalili-Mahani, V. Engert, M. Pruessner, C. Buss, et al., "Deactivation of
the limbic system during acute psychosocial stress: evidence from positron emission tomography and
functional magnetic resonance imaging studies," Biological psychiatry, vol. 63, pp. 234-240, 2008.
[36] A. Dagher, B. Tannenbaum, T. Hayashi, J. C. Pruessner, and D. McBride, "An acute psychosocial stress
enhances the neural response to smoking cues," Brain research, vol. 1293, pp. 40-48, 2009.
[37] Z. Y. Shan, M. J. Wright, P. M. Thompson, K. L. McMahon, G. G. Blokland, G. I. De Zubicaray, et al.,
"Modeling of the hemodynamic responses in block design fMRI studies," Journal of Cerebral Blood Flow
& Metabolism, vol. 34, pp. 316-324, 2014.
[38] A. Sassaroli and S. Fantini, "Comment on the modified Beer–Lambert law for scattering media," Physics in
Medicine & Biology, vol. 49, p. N255, 2004.
[39] Y. Gu, S. Miao, J. Han, K. Zeng, G. Ouyang, J. Yang, et al., "Complexity analysis of fNIRS signals in
ADHD children during working memory task," Scientific reports, vol. 7, p. 829, 2017.
47
[40] S. Sutoko, H. Sato, A. Maki, M. Kiguchi, Y. Hirabayashi, H. Atsumori, et al., "Tutorial on platform for
optical topography analysis tools," Neurophotonics, vol. 3, p. 010801, 2016.
[41] F. Al-Shargie, M. Kiguchi, N. Badruddin, S. C. Dass, A. F. M. Hani, and T. B. Tang, "Mental stress
assessment using simultaneous measurement of EEG and fNIRS," Biomedical optics express, vol. 7, pp.
3882-3898, 2016.
[42] J. Shin and J. Jeong, "Multiclass classification of hemodynamic responses for performance improvement of
functional near-infrared spectroscopy-based brain–computer interface," Journal of Biomedical Optics, vol.
19, p. 067009, 2014.
[43] M. Stangl, G. Bauernfeind, J. Kurzmann, R. Scherer, and C. Neuper, "A haemodynamic brain–computer
interface based on real-time classification of near infrared spectroscopy signals during motor imagery and
mental arithmetic," Journal of near infrared spectroscopy, vol. 21, pp. 157-171, 2013.
[44] K.-S. Hong, M. J. Khan, and M. J. Hong, "Feature extraction and classification methods for hybrid fNIRS-
EEG brain-computer interfaces," Frontiers in human neuroscience, vol. 12, 2018.
[45] A. Zafar and K.-S. Hong, "Detection and classification of three-class initial dips from prefrontal cortex,"
Biomedical optics express, vol. 8, pp. 367-383, 2017.
[46] N. Naseer, F. M. Noori, N. K. Qureshi, and K.-S. Hong, "Determining optimal feature-combination for
LDA classification of functional near-infrared spectroscopy signals in brain-computer interface
application," Frontiers in Human Neuroscience, vol. 10, 2016.
[47] C.-C. Chang and C.-J. Lin, "LIBSVM: a library for support vector machines," ACM transactions on
intelligent systems and technology (TIST), vol. 2, p. 27, 2011.
[48] J. Shin, K.-R. Müller, and H.-J. Hwang, "Near-infrared spectroscopy (NIRS)-based eyes-closed brain-
computer interface (BCI) using prefrontal cortex activation due to mental arithmetic," Scientific Reports,
vol. 6, 2016.
[49] N. Naseer, N. K. Qureshi, F. M. Noori, and K.-S. Hong, "Analysis of Different Classification Techniques
for Two-Class Functional Near-Infrared Spectroscopy-Based Brain-Computer Interface," Computational
Intelligence and Neuroscience, vol. 2016, 2016.
[50] K.-S. Hong, N. Naseer, and Y.-H. Kim, "Classification of prefrontal and motor cortex signals for three-
class fNIRS–BCI," Neuroscience letters, vol. 587, pp. 87-92, 2015.
[51] N. Naseer, M. J. Hong, and K.-S. Hong, "Online binary decision decoding using functional near-infrared
spectroscopy for the development of brain–computer interface," Experimental brain research, vol. 232, pp.
555-564, 2014.
[52] M. J. Khan, M. J. Hong, and K.-S. Hong, "Decoding of four movement directions using hybrid NIRS-EEG
brain-computer interface," Frontiers in human neuroscience, vol. 8, p. 244, 2014.
[53] H.-J. Hwang, J.-H. Lim, D.-W. Kim, and C.-H. Im, "Evaluation of various mental task combinations for
near-infrared spectroscopy-based brain-computer interfaces," Journal of biomedical optics, vol. 19, pp.
077005-077005, 2014.
[54] S. D. Power and T. Chau, "Automatic single-trial classification of prefrontal hemodynamic activity in an
individual with Duchenne muscular dystrophy," Developmental neurorehabilitation, vol. 16, pp. 67-72,
2013.
[55] J. Shin, J. Kwon, J. Choi, and C.-H. Im, "Ternary Near-Infrared Spectroscopy Brain-Computer Interface
With Increased Information Transfer Rate Using Prefrontal Hemodynamic Changes During Mental
Arithmetic, Breath-Holding, and Idle State," IEEE Access, vol. 6, pp. 19491-19498, 2018.
48
Web Crawling Technique for Vulnerability Assessment on
Web
Fietyata Yudha1, a) Andi Muhammad Panji T1,b), Laksono Adiputro AR1,c), Erika
Ramadhani1,d)
1Department of Informatics, Universitas Islam Indonesia, Yogyakarta, Indonesia
.
a)Corresponding author: [email protected]
b)[email protected] c)[email protected]
Abstract. Internet becomes popular communication in the world. The web application also growing up along with the
increasing of internet users. Various desktop application are moving into a web-based application. Web-based application
usually accessible over the internet. The application type is open source even open data over the internet. It's interesting
for some internet user, in this case, in a security enthusiast. The developement of web application before launching to the
internet required a security test. The test manually perform by surfing the whole web page which is using a security test
application. Manually, crawling techniques are possible to perform a surf in the whole web pages. This technique also
utilized in the search engine application. The crawler will produce the list of URL that will test using several vulnerability
testing technique. This research proposed a security test application for the web application by utilizing web crawling
technique. The developed application can perform web application security test, concentrating on SQL injection, Cross-
Site Scripting (XSS) and phishing attack vector.
Keywords—Vulnerability Assessment; Web Application; Crawling
INTRODUCTION
Nowadays, the internet becomes popular communication in the worlds. Every people are connected to the internet
today. One type of services provided by the internet is the web. The web became popular time by time after invented
in 1989. By the time evolving, the web become more powerful, attractive, good-looking, and accessible to the internet
user. Several desktop application changed to the web-based application to get the additional features also
experiencing the users. Usually, web-based application is accessible over the internet. Certain web applications use
sensitive data and provide critical services, it will be open data and/or open source. It is interesting for some
enthusiastic internet user.
The increasing of the internet user over the worlds, make an issues impact in security. Computer security is a
precautionary measure that taken to protect a system from attacks carried out by irresponsible parties. A weak system
will affect the user's trust when utilized the system. Security system required to maintain and protect user privacy.
There are several events related to hacking or breaking into the systems caused by the lack of attention of service
providers to system security. The type of system that is generally targeted by computer security enthusiast, hackers,
and crackers is a web-based system. This occurred because of recently the web-based system is growing rapidly.
Hackers are someone who hacks by looking for security holes on a system, giving ideas then solutions to system
administrators if there is a security gap. While crackers are someone who hacks by looking for security holes from a
49
system, then using these loopholes for individual interests such as stealing data, deleting the data, destroying data,
then doing other things that harm the system owner.
There are certain types of attacks that commonly attacked the web application. Several kind of attack models are
SQL Injection, Phishing, and Cross-Site Scripting (XSS). Fig. 1 depicted the top 10 attack technique used by an
attacker to compromise the system. SQL injection is one of the techniques that have a higher value[1]. The attack
carried out by using a manual technique or by using application. In manual technique, the tester will manually surf all
of the web content to find the vulnerability of the web application. Otherwise, the use of application to perform testing
which is very easy. The tester only input the target URL in the web based application and the application will launch
the result of the test.
FIGURE 1. Top 10 Computer Security Attack Technique
Scanning is the second step process in the system security testing methodology. Scanning will read the detailed
information about the targeted system. The system security tester may utilize this information to make a decision for
the next step such as to gain access into the system. The crucial step in the testing of a web application is vulnerability
scanning. Since scanning may provide all vulnerability information about the target. Surfing the whole page of the
web application is a form of scanning process that refers to the manual technique of web application testing then it
utilize as a web crawling technique. This technique may read the HTML source of web page and produce as needed,
one of the outputs resulted by crawling is URL. The URL resulted from crawling used to check the vulnerability.
This research proposed software assurance tool on web application security scanner that develops a simple
application according to the implementation of web crawling technique to perform testing on the web page open
source web application by implementing crawling technique, to give information about the compromised web
application and perform simple test for the compromised web application. The application will focus on the phishing,
SQL Injection, and Cross-Site Scripting (XSS) attack vector.
LITERATURE REVIEW
Fonseca, et al.[2] conducted research on comparing web vulnerability scanning tools for SQL injection and XSS
(Cross Site Scripting). The research constructed the injection of software errors on web applications in order to
compare the efficiency of every tools in discovering the vulnerabilities as a result of the injected bugs. The results of
the assessment of three foremost web application vulnerability scanners shows that dissimilar scanners produce quite
different results.
50
Auronen[3] categorized the assessed tools into two categories, i.e., enterprise-level tools and focused tools with
the objective to consider how comprehensive the tool is in its view to the assessment process. This research also
divides the tools into additional two categories, i.e., open-source and closed-source tools. The closed-source tools
typically completed products and utilize by non-security expert personnel.
Web application scanners support the developers to decrease the number of exposures in web applications. Web
application scanners are crawling over a web application pages, examine the application vulnerabilities then
simulating the attacks. Different types of tools and best practices would have applied throughout the development
lifecycle. [4] defined web application scanner, as a part of the NIST SAMATE project.
Kie, et al.[5] conducted research by developing an application naming Ardilla. Its purpose to test web application
by using SQL Injection and Cross-Site Scripting (XSS) attack technique. Ardilla wrote in the form of the PHP
(Hypertext Preprocessor) programming language. To use this application, user must specify types of attack that will
perform. Until today the Ardilla has not published publicly because this application is still in the development stage.
In other side, Kals, et al.[6] offered SecuBat, a generic and modular web vulnerability scanner that examines
websites for vulnerable to SQL and XSS. SecuBat functioned to identify a huge amount of possibly vulnerable
websites. Yudha & Muryadi[7] conducted preliminary design of web application vulnerability assessment. The
application design provided three attack vector of web application namely 1.SQL injection, 2.Cross-site Scripting
(XSS), and 3.Phising. This study suggested a web security assessment tool developed by using Python. The reason to
use Phyton because it has several advantages such as cross-platform capabilities.
Furthermore, Mirtaheri, et al.[8] wrote a brief history of web crawlers. There are several motivation for a system
to the use of web crawler. Its content indexing for search engines, automated testing for a web application, and
automated security testing and vulnerability assessment. Ahuja, et al.[9] said that crawler is the important thing to get
information from the webpage such a search engine. Developing an effective web crawler application is a not a
difficult task. Using the right strategies also an effective architecture will deliver an intelligent web crawler
application. Implementation of a good algorithm will increase the performance of the application likewise the result.
There are minimum requirements for any large-scale crawling system, according to the search strategies and
algorithms. The requirement is enabling the flexibility, high performance, error acceptance, maintainability and
configurability[10]. Due to the evolution of the web and its information structure, the algorithms of web crawling are
changing to provide more accurate search results. Pellegrino, et al.[11] presented jAk, a web application scanner which
implements the crawler. The crawler produces a navigation graph which use to choose the next act. The tool assessed
and compared with four supplementary web-application scanners and tested using 13 web applications as a target. The
trial results demonstrate that jAk be able to discover web applications vulnerability about 86% larger than the other
tools.
Bhojak, et al.[12] present a new technique to find vulnerability of different web pages. This research implements
web crawling technique to perform web vulnerability scanner. The result of this research is a tool that could perform
SQL injection test and Cross-Site Scripting test on a web page. The tool claimed that has a better result than two other
scanners.
METHODOLOGY
Web page vulnerability performed by opening the webpage and looking for vulnerable URL or script that run in the webpage. The process by surfing page to page to find the vulnerability of the website. The process will take a long time and a big effort for the tester. Applicating the crawling technique will simplify the process when developing a testing application.
51
FIGURE 2. Application Flow
Fig. 2 shows the process of developed application. The application has three main functions such as crawling
function, URL parsing function, and URL testing function. Crawling function would use a plugin to simplify the
developing process. Crawler plugins would give the application URL data that contained on the index page of the
target site. The URL list result would parse by the application to find vulnerable URL/Page. In the practice of injection
(SQL Injection and XSS) the vulnerable URL usually used "?id=" phrase. After the URL has found, the application
would test the URL and catch some error from the test. According to the error standard of the web page the application
would define the result. This practice would implement the URL testing method in the application.
From the method as explained above, the developed application will perform two test to the target based on the
injection method and one test for phishing test. In the injection test, the developed application will perform SQL
injection test to find SQL vulnerability in web apps then test for Cross-Site Scripting (XSS) attack possibilities. To
perform the test of a web app using the developed application, the user enters the URL. The application will scan the
target URL by crawling the index page of the web apps to get information about other pages are available. There are
three step to test the target web apps. There are phishing, SQL Injection, and cross-site scripting(XSS). These steps
carried out sequentially from phishing to cross-site scripting.
To find phishing vulnerability on the page, the developed application would find "login" phrase on the URL list
result. If the phrase found, the application will report that the target vulnerable to phishing. SQL injection and cross-
site scripting the application will find the "id=" in the URL list. On SQL injection step the application would add a
single quote (‘) at the end of founded URL and visit the page to catch the error. If the application found the standard
SQL error on the visited page, it would report as SQL injection vulnerability. For the Cross-site scripting, the apps
would insert the cross-site code into the URL then catch the error.
RESULT AND DISCUSSION
According to the literature review and the methodology, this research would implement web crawling to create
web vulnerability scanner based on Python. Beautifulsoup implemented for web crawling on the developed
application. Beautifulsoup is a Python library to pull data out of HTML and XML files. It works with chosen parser
to deliver idiomatic ways of navigating, searching, and modifying the parse tree[13].The result of beautifulsoup would
process by developed application to find the vulnerability in the web page and provide the result to the user. The
developed application will capture URL from the crawling process as depict in Fig. 3. The URL result would test
using the rules of Phishing, SQL Injection, and Cross-Site Scripting provided in methodology. For capturing the error
52
of the testing, this application developed using the urllib plugin in Python. Urllib would open the tested URL then
crawl the result with beautifulsoup to check reported error.
(a) (b)
FIGURE 3. BeautifulSoup Crawling Result
Fig. 4 shows the interface of the developed application. The GUI interface developed using Qt5. To perform testing
with developed application, user just enter the tested URL. The application would launch the test immediately then
provide information on the bottom section of the application. This application may provide a different experience for
users with a simple action by the users. This application provides an additional feature for SQL injection. Its feature
is SQL database dump.
(a) (b)
FIGURE 4. Application Interface
Testing the developed application performed by testing various web application. It scans 50 webs applications to
get the result from different web vulnerability. The following TABLE 1 shows the result of application test. From
the result, we can conclude that the developed application has the same result with a manual test.
53
TABLE 1. Application Test Result Compared with Manual Test
No Number
Of
Sample
Phishing SQL
Injection
Cross-
Site
Scripting
No Number
Of
Sample
Phishing SQL
Injection
Cross-
Site
Scripting
1 Sample 1 - - 26 Sample 26 -
2 Sample 2 - - 27 Sample 27 -
3 Sample 3 - - 28 Sample 28 - -
4 Sample 4 - - 29 Sample 29 - -
5 Sample 5 - - 30 Sample 30 - - -
6 Sample 6 - - 31 Sample 31 - - -
7 Sample 7 - - 32 Sample 32 - - -
8 Sample 8 - - 33 Sample 33 - - -
9 Sample 9 - - 34 Sample 34 - - -
10 Sample 10 - 35 Sample 35 - - -
11 Sample 11 - - - 36 Sample 36 - - -
12 Sample 12 - - - 37 Sample 37 - - -
13 Sample 13 - 38 Sample 38 - - -
14 Sample 14 - - - 39 Sample 39
15 Sample 15 - - - 40 Sample 40 - - -
16 Sample 16 - - 41 Sample 41 - - -
17 Sample 17 - - - 42 Sample 42 - - -
18 Sample 18 - - - 43 Sample 43 - - -
19 Sample 19 - - - 44 Sample 44 - - -
20 Sample 20 - - - 45 Sample 45 - - -
21 Sample 21 - - - 46 Sample 46 - -
22 Sample 22 - - 47 Sample 47 -
23 Sample 23 - - - 48 Sample 48 - - -
24 Sample 24 - - 49 Sample 49 - - -
25 Sample 25 - - - 50 Sample 50 - - -
CONCLUSION
The primary contribution of this research is to develop a simple application based on an implementation of web
crawling technique to perform testing on the web page for phishing, SQL injection, and Cross-Site Scripting (XSS).
The result shows that the application has the same result with the manual process of web page test. It means that the
crawling technique may be utilize to develop web application vulnerability assessment tools.
For future development, the application expanded to cover A1 – A10 from OWASP security risk list to meet the
NIST SP-500-269 about software assurance tool on web application security scanner. Also, develope the lightweight
application, in order the application can run on a single board based PCs, however this study would begin with the
performance analysis of the application.
ACKNOWLEDGMENTS
We are grateful to Informatics Department, Universitas Islam Indonesia for their support for the research project.
We would like to express our gratitude to Mr. Yudi Prayudi for providing guidance.
54
REFERENCES
[1] P. Passeri, “February 2018 Cyber Attacks Statistics – HACKMAGEDDON,” hackmageddon Cyber Attacks
Statistics, 2018. [Online]. Available: https://www.hackmageddon.com/2018/04/06/february-2018-cyber-
attacks-statistics/. [Accessed: 24-Apr-2018].
[2] J. Fonseca, M. Vieira, and H. Madeira, “Testing and comparing web vulnerability scanning tools for SQL
injection and XSS attacks,” Proc. - 13th Pacific Rim Int. Symp. Dependable Comput. PRDC 2007, pp. 365–
372, 2007.
[3] L. Auronen, “Tool-Based Approach to Assessing Web Application Security,” Netw. Secur., pp. 1–20, 2002.
[4] E. Fong and V. Okun, “Web application scanners: Definitions and functions,” Proc. Annu. Hawaii Int. Conf.
Syst. Sci., pp. 1–7, 2007.
[5] A. Kie, P. J. Guo, and M. D. Ernst, “Automatic Creation of SQL Injection and Cross-Site Scripting Attacks,”
in Proceeding of International Conference Software Engineering, 2009, pp. 199–209.
[6] S. Kals, E. Kirda, C. Kruegel, and N. Jovanovic, “SecuBat : A Web Vulnerability Scanner,” WWW ’06 Proc.
15th Int. Conf. World Wide Web, pp. 247–256, 2006.
[7] F. Yudha and A. Muhammad Panji Muryadi, “PERANCANGAN APLIKASI PENGUJIAN CELAH
KEAMANAN PADA APLIKASI BERBASIS WEB,” CyberSecurity dan Forensik Digit., vol. 1, no. 1, pp.
1–6, 2018.
[8] S. M. Mirtaheri, M. E. Dinçktürk, S. Hooshmand, G. V. Bochmann, G.-V. Jourdan, and I. V. Onut, “A Brief
History of Web Crawlers,” in CASCON, 2013.
[9] M. S. Ahuja, J. S. Bal, and Varnica, “Web Crawler : Extracting The Web Data,” Int. Eng. Res. J., vol. 1, no.
8, pp. 629–632, 2015.
[10] B. M. Jasani and C. K. Kumbharana, “Analyzing Different Web Crawling Methods,” Int. J. Comput. Appl.,
vol. 107, no. 5, pp. 975–8887, 2014.
[11] G. Pellegrino, C. Tschürtz, E. Bodden, and C. Rossow, “jAk: Using Dynamic Analysis to Crawl and Test
Modern Web Applications,” in Proceedings of the 18th International Symposium on Research in Attacks,
Intrusions, and Defenses, 2015, pp. 295–316.
[12] P. Bhojak, V. Shah, K. Patel, and D. Gol, “Automated Web Application Vulnerability Detection With
Peneteration Testing,” Kalpa Publ. Comput., vol. 2, pp. 177–187, 2017.
[13] L. Richardson, “Beautiful Soup Documentation — Beautiful Soup 4.4.0 documentation,” 2015. [Online].
Available: https://www.crummy.com/software/BeautifulSoup/bs4/doc/. [Accessed: 27-Nov-2018].
55
Investigation of SEGR Effects on Power VDMOSFET for
Various Heavy Ion Radiation
Saranya Krishnamurthy a), Ramani Kannan b), Fawnizu Azmadi Hussin,
Erman Azwan Yahya
Department of Electrical and Electronic Engineering, Universiti Teknologi PETRONAS, 32610, Perak, Malaysia
a)Corresponding author: [email protected]
Abstract. Power metal-oxide semiconductor field-effect transistors (MOSFETs) are the most important devices for power
converters in harsh environment applications. However, the power MOSFETs suffer significantly from Single Event Gate
Rupture (SEGR) and Single Event Burnout (SEB) during heavy ion exposure and heavily damages the device. This paper
investigated the influence of SEGR effects on Vertical Double-Diffused power MOSFET (VDMOSFET) for various heavy
ion radiation in order to define the sensitive criteria of the power MOSFET. In addition, the SEGR performance for different
gate oxide thickness was performed. Investigations relied on two-dimensional SILVACO TCAD simulation tools:
ATHENA for process and ATLAS for device simulation. The examination of SEGR mechanism for different incident ions,
various gate oxide thickness and a comparison of behaviour illustrated the understanding of device robustness. It also
discovered that the device is much sensitive for the heavy ion with higher radiation value and improved SEGR performance
for the larger gate oxide thickness.
Keywords—Power MOSFET; Single Event Gate Rupture; Silvaco TCAD simulation.
INTRODUCTION The outer space applications include national security, earth observation, Global Navigation Satellite System
(GNSS), communication satellites, weather forecasting satellites, satellite television and data broadcasting has become
a significant part of everyday life. The switching power converters and electronic devices mounted in artificial
satellites are required to be highly reliable and efficient. Power MOSFETs are widely used as switching devices in
satellites for its high switching speed over than other power semiconductor switches, such as power diodes, power
bipolar junction transistors (BJTs) and insulated-gate bipolar transistors (IGBTs). Predominantly, these devices are
highly sensitive to the radiation source in space includes galactic cosmic rays, solar flares, trapped radiation belts and
solar particle events. Furthermore, these radiation sources are divided into particle and photon. The particle radiations
are electrons, protons, neutrons, alpha particles and heavy ions. Gamma rays and X-rays are photon radiations. Power
MOSFETs are extensively used in many power electronic applications especially in space and harsh environment.
However, utilizing power semiconductor devices under continuous radiation and high temperature reduces their
lifespan by affecting its electrical properties and performance [1]. Generally, two major radiation effects create
permanent damage to power MOSFET namely Total Ionizing Dose (TID) and Single Event Effect (SEE). The
cumulative effect of ionizing radiation is referred as TID. Dose is defined as the quantitative measure of accumulated
energy absorbed from ionizing radiation per unit mass. TID effects are associated with gate oxide damage, it will
change the threshold voltage and also increase the leakage current of device [2]. Single Event Effects are transient
phenomena because it is induced by a single heavy energetic particle that permanently damage the power MOSFET
within few microseconds. SEE effects are further classified into two catastrophic effects SEB and SEGR. SEB effect
increases the drain current due to the heavy ion passes through the device result in source to drain shortage. In the
case of SEGR effect, it damages the gate oxide dielectric [3]. Thus, there is a necessity to study the vulnerability of
SEGR effects on power MOSFETs in space applications.
Power MOSFETs have conquered the power switching semiconductor device market subsequently from 1980s
and introduce in the most of power switching converters. The power MOSFETs are an essential part of space-borne
electronic systems due to their benefits of fast switching speed, low operating voltage and positive temperature
coefficient. The early 1980s, the silicon-based Vertical Double-Diffused MOSFET structure (VDMOSFET or
DMOSFET) emerge as a leading power MOSFET introduced by International Rectifier [3-7]. Some literature papers
56
discussed about the hardening techniques that are used to mitigate the SEGR effects in Silicon VDMOSFET. The
techniques are addition of buffer layer, localized carrier lifetime control and elimination of P-well region below the
N+ source [8-11].
In this work, two dimensional ANTHENA simulator used to create a power VDMOSFET structure followed by
the processing steps includes doping, deposition, etching, annealing and contact formation. In addition, ATLAS
simulator was performed to observe the SEGR performance for a different heavy ion radiation. The investigation was
also used to analyze the SEGR reliance on gate oxide thickness, threshold voltage and drain current for the LET of
various incident ions. The examination of SEGR mechanism for different incident ions, various gate oxide thickness
and a comparison of behaviour illustrated the understanding of device robustness. It also discovered that the device is
much sensitive for the heavy ion with higher radiation value and improved SEGR performance for the larger gate
oxide thickness. The remaining part of the paper will further discuss the radiation effects on power MOSFET with
SEB and SEGR mechanisms, process and device simulation and illustrates the results and discussion.
RADIATION EFFECTS ON POWER MOSFET The ionizing radiation in the device degrades the performance and efficiency of the device by changing the electrical
characteristics which tends to ionization mechanism . Figure.1 shows the power MOSFET energy band gap diagram, the structure is positively biased via gate terminal.
During the radiation exposure, the highly charged particle such as electrons, protons and gamma rays passes
through the oxide layer creates more number of electron-hole pair inside the oxide layer and as well as the interface
layer of MOSFET structure [4]. Compared to the holes, the generated electrons were significantly more versatile to
move out of the oxide layer in a quick occasion. However, some immobile electrons and holes that escaped from initial
recombination will remain behind in oxide. This trapped charges at the SiO2/Si interface induces an inversion layer
during the off-state, which causes increase in drain current, leakage current and threshold voltage shifts [5]. The
ionizing radiation of the space environment causes both TID and SEEs. The physical mechanisms of these effects are
distinctive to each other, explained briefly as follows:
Total Ionizing Dose (TID) The cumulative effect of ionizing radiation is referred as TID. Dose is defined as the quantitative measure [12] of
accumulated energy absorbed from ionizing radiation per unit mass as given in equations (1).
dx
dEd
1= (1)
where, d is the dose, ϕ is the flux of incident particles, E is the energy and ρ is the density of the target.
The radiation dose is quantified either using the radiation absorbed dose (rad) or SI unit which is Gray [Gy] i.e.1
Gy = 100 rad. The ionizing radiation rates are material-specific, it depends on the charges accumulation over period
of time in different materials such as silicon or insulating oxide (SiO2).
57
FIGURE 1. The Ionization process in power MOSFETs
Single Event Effect (SEE) SEEs are transient phenomena because it is induced by a single heavy energetic particles that permanently damage
the power MOSFET within few microseconds. Further classified into two catastrophic effects SEB and SEGR. The SEB mechanism shown in Fig. 2. SEB is a destructive SEE effect in power devices, mainly due to the inherent parasitic bipolar junction transistor present inside the MOSFET structure. When heavy ion passes through the device structure, charges accumulate along its ions track based on the angle. The deposited charge leads to high electric field density inside the JFET region as well as in epitaxial layer because of parasitic BJT on. The ionization rates of electron and holes determined by the shape, intensity and electric field time evolution inside the epitaxial layer. Due to the regenerative phenomenon the very high electric field density always ON the parasitic BJT and pull down all the holes to drain. Thus, SEB does not damage the insulator but increasing in high source to drain current because of effectively shorts in source to the drain. The high level of currents causes the device to burnout and lead to meltdown the substrate. The charge deposition in the oxide layer measured in terms of Linear Energy Transfer (LET), units pc/µm or MeVcm2mg-1. LET is defined as energy exchange between interaction particle and target material along certain linear track as given in equation (2).
dx
dELET
1= (2)
where, E is the energy of particle , ρ is the density of the target and x is the linear track. The relation between two units is calculated as 1 pC/μm =100 MeVcm2mg-1 [6].
58
FIGURE 2. Single Event Burnout (SEB) in power MOSFET
SEGR effect is also a destructive of SEE highly associated with power MOSFETs due to heavy ion radiation. The
SEGR mechanism is shown in Fig. 3. SEGR occurs when the ion track is nearly perpendicular to the vertical axis of the device but does not depends on temperature. When heavy ion strikes the neck region of the power MOSFET during OFF state, electron-hole pairs are formed according to the ion LET provided that a high electric field in the order of MV/cm. The generation rate of electric field-dependent inside the dielectric material also plays a vital role in the formation of electron-hole pairs. When the transitory conductivity of the gate oxide run into the high electric stress across the oxide layer, it gives rise to increase in gate-to-source leakage current and permanent damage in the gate insulator (SiO2) layer.
FIGURE 3. SEGR effects in power MOSFET
59
Furthermore, high electric field inside the power MOSFET also causes threshold voltage shifts and increase in drain-source current, which depends on changes in drain-source resistanceSEB is a destructive SEE effect in power devices, dependent on the inherent parasitic BJT in the MOSFET structure. A heavy ion transverses power MOSFET it deposits charge along its ions track. The deposited charge leads to turning ON the parasitic BJT and creates a flux of electrons flowing down to the drain. The intensity, shape and time evolution of electric field inside the epitaxial layer, determine ionization rates of electron and holes. When electric field peak moves from p-n junction to epi-sub junction, it causes holes generation. If electric field intensity is high enough, the holes flux towards the BJT structure keeps it turned ON, consequently creating a regenerative phenomenon. Thus, SEB does not damage the insulator but effectively shorts the source to the drain then resulting in the high source to drain current. The high level of currents causes the device to burnout and lead to meltdown the substrate. Charge deposition is given quantitatively by the Linear Energy Transfer (LET) measured in units of pc/µm or MeVcm2mg-1 [6].
PROCESS AND DEVICE SIMULATION
In this work, the SEGR simulation was performed to study the change in electrical characteristics when heavy ion
passes through the neck region of device VDMOSFET. In this case two simulator in silvaco software was used,
ATHENA simulator for designing the device with fabrication steps and ATLAS simulator for analyzing the electrical
characteristics.
FIGURE 4. Device used for SEGR case study with net doping and materials
Silvaco Simulator The overview of device simulator with physical models is given in Fig. 5. Silvaco TCAD simulation software
consists of different modules such as ATHENA, ATLAS, DeckBuild, DevEdit and Tonyplot.
60
FIGURE 5. Silvaco device simulator overview [13]
The fabrication process flow modeling of the device is termed by process TCAD (ATHENA) and the operation of
the device modeling is termed by device TCAD (ATLAS). It has built in library for different semiconductor materials
for various analysis process. It also allows to simulate different MOSFET Devices with variation of different
parameters. The ANTENA module consists of the following features: design and analysis simulation of all critical
fabrication steps, dopant distributions, prediction of multi-layer topology, import mask files, stresses, run-time
extraction of process and device parameters, automatic user-defined mesh generation and control, optimization of
process flow and calibration of model parameters.
The ATLAS module used for the operation of device and circuit simulation using SPICE netlists, SmartSpice
models and fully integrated with Athena & DevEdit. It will allow analyzing transient time, DC, AC transfer
characteristics and time evaluation responses in 2D/3D for all semiconductor technology devices. The special module
inside atlas is GIGA-Electro-Thermal that used for analyze the lattice temperature of the material and the dependence
in the interface of transport parameters. Silvaco ATLAS TCAD software is used to design the geometrical construction
and Silvaco Victory Process feature is use to simulate the design with radiation module. Then, the electrical
characteristics and behavior of design will be analyzed based on static and dynamic topology by using TonyPlot and
Monte-Carlo features in Silvaco Software. The DevEdit module used for re-mesh or edit an existing device associated
with DeckBuild or Virtual Wafer Fab (VWF). TonyPlot is a parasitic extraction tools that transforming simulation
data into a high quality plot. It provides visualization and flexible graphical analysis features such as rulers, probes,
zoom, labels, views and overlay for multiple plot support [13].
Process Simulation
In order to evaluate the SEGR performance for various heavy ion, a typical power VDMOSFET structure was
created using ATHENA simulator. The device parameter used in the simulations followed the default VDMOSFET
structure as shown in Fig. 4. A process flow of power MOSFET followed few steps, at first highly doped n+ silicon
substrate was formed using the device parameter as shown in Table 1, above that lightly phosphorous doped n-
epitaxial layer for the 7.5µm-thickness was deposited. Next step to form the P-base region, two boron well ion-
implanted and diffused into the epitaxial layer with a surface concentration at 1e19. Followed by highly doped
phosphorous for the formation of n+ contact with a surface concentration of 5e14 was diffused. The gate oxide
thickness 50nm was deposited in the gate formation process. Finally passivation, annealing and source contacts made
to form the drain, source and gate terminals. During simulation of SEGR performance, the gate oxide thickness
changed to 100nm, 150nm and 200nm. The changing in oxide thickness is limited based on the maximum transient
oxide field and critical gate to source bias [12].
61
TABLE 1. Device parameter for simulation
Device Parameter Value
Gate oxide thickness 0.05
N-Drift layer thickness 7.5
P-body Doping 1e19
n+ source doping 5e14
Device Simulation
For device simulation, the structure exported to the ATLAS device simulator. In order to perform SEGR
simulation, few models are required to obtain more precise device simulation results. To compare the electrical
characteristics for various heavy ion radiation, the ion with chosen LET values of 26.6 (Ni), 37.2 (Br), 59.7 (I) and
82.3 (Au) strikes the centre (X=0) of the device. To creating the ion strike, the syntax must include “singleeventupset”
statement, LET value, entry and exit point parameters. The statements “radius” and “radialgauss” will define the ion
strike track radius and the charge density respectively. The ion strike initial time and duration defined by the t0 and
tc. To enable the numerical integration of ion strike track, need to specify the seu.integrate in method statement.
Fowler-Nordheim model used to monitor the gate current. The “probe” statement included to calculate the DC
simulation that specifies the physical quantities located at specific co-ordinates in the device structure. The statement
mentioned in the atlas as below:
set let=26.6 /37.2/59.7/82.3
set density=$let*0.011
method seu.integrate
models fn.cur probe name=Source_Field field dir=270 x=3.4 y=-0.02
A fine mesh is generally required to form electron-hole pairs. The biasing conditions should maintain VGS and
VDS constant in order to monitor the leakage current during SEGR simulation. The required gate rupture ion strike
will occur during the application of gate voltage =-13.9 volts and drain voltage = 30 volts. Finally the Tony plot used
to extract the line graph of drain current, threshold voltage and breakdown voltage [12].
TABLE 2. Heavy Ion with Energy Levels
Ion Beam Energy,
MeV
LET,
MeVcm2mg-1
Au-197 350 82.3
I-127 320 59.7
Br-79 286 37.2
Ni-58 278 26.6
RESULTS AND DISCUSSION In this section, numerical simulations using silvaco ATLAS were performed for four different incident ions include
Nickel, Bromine, Iodine and Gold with LET energy levels shown in Table 2 [13]. Figure. 6 shows the snap-shot of hole concentration time evolution during the heavy ion (Br-37.2) strike at time intervals of 5ps, 50ps, 100ps and 150 pico seconds after the ion strike. It also displays the effects of electric fields present before the heavy ion strike and subsequent transportation after the time evaluation.
62
FIGURE 6. Hole concentration at 5, 50, 100 and 150ps after the ion strike
The drain current as a function of transient time for different ion strike was compared as shown in Fig. 7. The drain
current reached its peak value within a few picoseconds for gold compared to the nickel as incident ion during the input parameters: Vgs=-13.92V, Tox=50nm and Vds=30V. In addition, to observe the SEGR performance for various gate oxide thickness with respect to time was performed. Changing the gate oxide thicknesses for four different values ranging from 50 to 200 in nanometre were simulated and calculated the drain current as function of time. For every change in oxide thickness, the structure underneath the Si/SiO2-interface was left unchanged and only the thickness of the gate-oxide was adjusted to the desired value. It observed that the Vds drain current reduced for higher gate oxide thickness as shown in Fig. 8. Unfortunately, the threshold voltage value increases when increasing the gate oxide thickness. It clearly shown that the threshold voltage value is directly proportional to gate oxide thickness. The comparison of threshold voltage value form the Vgs to Vds curve was observed and shown in Fig. 9.
FIGURE 7. Drain current after various ion strike
63
FIGURE 8. Drain current after ion strike for different gate oxide thickness
FIGURE 9. Threshold voltage Vs Drain current for different gate oxide thickness
CONCLUSION In this paper, study of two dimensional SEGR simulation for power VDMOSFET was presented. The sensitive
volume of VDMOSFET to SEGR was investigated by four different ion strike at neck region of device. It also investigated the SEGR dependence on gate oxide thickness, threshold voltage and drain current characteristics as function of time. The study confirms that the SEGR effect increases with increase in LET value of heavy ion radiation. Furthermore, it can also be noted that the effect was less sensitive to higher gate oxide thickness. The simulation results also shows that the threshold voltage (Vth) is directly proportional to oxide thickness (Tox). Hence, there is a trade off between Vth and Tox that has to be considered during SEGR device simulation. Consequently, the Tox need to be optimised in such a way that both Vth and SEGR performance are set to desired levels.
64
ACKNOWLEDGMENTS The authors would like to acknowledge the support of Universiti Teknologi PETRONAS (UTP) through the Award
of Yayasan Universiti Teknologi PETRONAS Fundamental Research Grant 015CL0-069.
REFERENCES
[1] T. Zhaohuan, H. Gangyi, C. Guangbing, T. Kaizhou, L. Yong, L. Jun,and X. Xueliang, “A novel structure for
improving the SEGR of a VDMOS,”Journal of Semiconductors, vol. 33, no. 4, p. 044002, 2012.
[2] J. L. Titus, “An updated perspective of single event gate rupture and single event burnout in power mosfets,”
IEEE Transactions on nuclear science, vol. 60, no. 3, pp. 1912–1928, 2013.
[3] Z. Tang, X. Fu, F. Yang, K. Tan, K. Ma, X. Wu, and J. Lin, “SEGR and SEB-hardened structure with DSPSOI
in power MOSFETs,” Journal of Semiconductors, vol. 38, no. 12, p. 124006, 2017.
[4] M. Zerarka, P. Austin, G. Toulon, F. Morancho, H. Arbess, and J. Tasselli, “Behavioral study of single-event
burnout in power devices for natural radiation environment applications,” IEEE Transactions on Electron
Devices, vol. 59, no. 12, pp. 3482–3488, 2012.
[5] D. Li-Li, C. Wei, G. Hong-Xia, Y. Yi-Hua, G. Xiao-Qiang, and F. RuYu, “Scaling effects of single-event gate
rupture in thin oxides,” Chinese Physics B, vol. 22, no. 11, p. 118501, 2013.
[6] J.-M. Lauenstein, N. Goldsman, S. Liu, J. L. Titus, R. L. Ladbury, H. S. Kim, A. M. Phan, K. A. LaBel, M.
Zafrani, and P. Sherman, “Effects of ion atomic number on single-event gate rupture (SEGR) susceptibility of
power mosfets,” IEEE Transactions on Nuclear Science, vol. 58, no. 6, pp. 2628–2636, 2011.
[7] T. Zhaohuan, L. Rongkan, T. Kaizhou, L. Jun, H. Gangyi, L. Ruzhang, R. Huaping, and W. Bin, “A novel terminal
structure for total dose irradiation hardened of a p-VDMOS,” Journal of semiconductors, vol. 35, no. 5, p. 054005,
2014.
[8] C.-H. Yu, Y. Wang, F. Cao, L.-L. Huang, and Y.-Y. Wang, “Research of single-event burnout in power planar
vdmosfets by localized carrier lifetime control,” IEEE Transactions on Electron Devices, vol. 62, no. 1, pp. 143–
148, 2015.
[9] J. Lu, H. Liu, X. Cai, J. Luo, B. Li, B. Li, L. Wang, and Z. Han, “Single event burnout hardening of planar power
MOSFET with partially widened trench source,” Journal of Semiconductors, vol. 39, no. 3, p. 034003, 2018.
[10] X. Wan, W. S. Zhou, S. Ren, D. G. Liu, J. Xu, H. L. Bo, E. X. Zhang, R. D. Schrimpf, D. M. Fleetwood, and T.
Ma, “SEB hardened power mosfets with high-k dielectrics,” IEEE Transactions on Nuclear Science, vol. 62, no.
6, pp. 2830–2836, 2015.
[11] J. Mo, H. Chen, L. Wang, and F. Yu, “Total ionizing dose effect and single event burnout of vdmos with different
inter layer dielectric and passivation,” Journal of Electronic Testing, vol. 33, no. 2, pp. 255–259, 2017.
[12] S. Inc, ATLAS & ATHENA User’s Manuals, Silvaco.
[13] L. Scheick, “Testing guideline for single event gate rupture (SEGR) of power mosfets,” Pasadena, CA: Jet
Propulsion Laboratory, National Aeronautics and Space Administration, 2008., Tech. Rep., 2008.
65
Resource-Optimized FPGA Implementation of Mersenne
Twister using SDP BRAM
Gabriel Mallari1,a), Jan Cedrick Marco1,b), Gideon Clarence Ong1,c),
Kien Aira Roldan1,d) and Dr. Ann Dulay1,e)
1ECE Department, Gokongwei College of Engineering, De La Salle University
b)[email protected] c)[email protected] d)[email protected]
Corresponding author: e)[email protected]
Abstract. Pseudorandom number generators (PRNG) are key components in diverse computational methods. The Mersenne
Twister algorithm is a popular choice of Pseudorandom Number Generator because of its long period and high quality of
randomness. In this paper, the researchers present a resource-efficient hardware implementation of the MT algorithm. This design
is centered on the simple dual port mode (SDP) of the Block RAM. Using this mode, the RNG can have a throughput of 1 sample-
per-clock cycle while only consuming 1 BRAM. It is implemented on a Kintex 7 board with a maximum frequency of 515.331
MHz. It requires only 42 slices and 1 BRAM, an improvement over previous works. The correctness of the designs is verified
through a comparison with the output of a software MT, as well as through statistical tests for randomness.
Keywords—FPGA, Mersenne Twister, Optimized, PRNG, Dieharder, NIST
INTRODUCTION
Random Number Generator (RNG) is an integral part to a wide range of applications, such as simulations,
sampling, numerical analysis, computer programming, decision making, aesthetics, and recreation [1]. Specific
applications well-suited to FPGA technology are modeling and optimization as in the Monte Carlo method,
cryptography, numerous stochastic simulation methods, and as noise simulators. It involves an intensive
computational operation that is used for high quality random number generation. Due to its widespread use, many
techniques have been developed to efficiently generate these random numbers.
There are several well-known Pseudorandom Number Generators (PRNG) such as the Mersenne Twister (MT),
Linear-Feedback Shift Register (LFSR), Blum Blum Shub, and Xorshift. These all vary in complexity, quality, period,
and speed. The Mersenne Twister PRNG is one of the most widely used, owing to its good properties and especially
long period.
Many of the applications requiring RNGs are well-suited to FPGA technology. The parallelism in hardware makes
calculations much faster than when performed in software [2]. Thus, much research has been done on the MT PRNG
as shown in Table 1. The resource efficiency can be further improved from the current approaches. The design in [3]
has a throughput of 1 sample/cycle at the cost of 2 BRAM. The mixed RAM used in [4] is very efficient but does not
strictly adhere to the MT algorithm. Most works on the FPGA implementation of Mersenne Twister deal with
efficiently handling the storage of its very large state while still maintaining a good throughput.
This paper presents a hardware architecture for the FPGA implementation of MT19937 that achieve a throughput
of 1 sample per cycle with minimal resources used.
66
Section II presents the Mersenne Algorithm, specifically MT19937. Section III describes our proposed hardware
architecture for the MT-19937 method. Section IV discusses the implementation of the proposed architecture and
evaluation of the results including resources utilization and statistical testing. Section V contains the conclusion of the
paper.
TABLE 1. Prior Studies
Author PRNG Frequency Slices BRAM FPGA
S. Chandrasekaran and A. Amira [5] MT 24.234
MHz 330 2 Virtex-E
Y. Li, J. Jiang, H. Cheng, M. Zhang, and S. Wei [3] MT 450 MHz 41 2 Virtex-6
P. Echeverra and M. Lopez [6] MT 418.6 MHz 272 2 Virtex-5
N. Saraf and K. Bazargan, [4] MT 505.31 MHz 53 1 Virtex-7
S. Banks, P. Beadling, and A. Ferencz [7] MT 300 MHz 56 4 Virtex-5
An V. Sriram and D. Kearney [8] MT 319 MHz 87 - Virtex II-
Pro
X. Tian and K. Benkrid [9] MT 265 MHz 128 2 Virtex-4
MERSENNE TWISTER ALGORITHM
One of the commonly used types of Pseudorandom number generator is Mersenne Twister. It is known for having
an incredibly long period of 219937 − 1 when using the parameter set for MT-19937. A sequence of -bit word vectors
are generated and are uniformly distributed between 0 and 2𝑤 − 1 [10]. The following recurrence in (1) is used:
𝑥𝑘+𝑛 = 𝑥𝑘+𝑚 ⊕ (𝑥𝑘𝑢 | 𝑥𝑘+1
𝑙 )𝐴 (𝑘 = 0, 1, . . . ) (1)
Where,
𝑛 is the degree of recurrence
𝑚 is the middle word, and1 ≤ 𝑚 ≤ 𝑛 𝑟 is a parameter used in (𝑥𝑘
𝑢 | 𝑥𝑘+1𝑙 ), and 0 ≤ 𝑟 ≤ 𝑤 − 1
(𝑥𝑘𝑢 | 𝑥𝑙+1
𝑙 ) is the concatenation of the upper 𝑤 − 1 bits of 𝑥𝑘 and the lower 𝑟 bits of 𝑥𝑘+1
𝐴 is a constant 𝑤 × 𝑤 matrix of the form presented in (2)
𝐴 = (0 𝐼𝑤−1
𝑎𝑤−1 (𝑎𝑤−2 , … , 𝑎0) (2)
The parameters, 𝑤, 𝑟, and 𝑛 are chosen so that 2𝑤𝑛−𝑟 is a Mersenne prime. Matrix is in this form so that the
matrix multiplication can be reduced to (3). Thus, the entire recurrence can be calculated using only bitwise operations.
𝑥𝐴 = 𝑥 ≫ 1 𝑖𝑓 𝑥0 = 0
(𝑥 ≫ 1) ⊕ 𝑎 𝑖𝑓 𝑥0 = 1 (3)
To improve the equidistribution of the sequence, a tempering matrix is multiplied to 𝑥𝑘+𝑛. This is done by the
transformations in (4). The result of these transforms is the final output of the MT RNG.
𝑦 ← 𝑥𝑘+𝑛 𝑦 ← 𝑦 ⊕ (𝑦 ≫ 𝑢)
𝑦 ← 𝑦 ⊕ ((𝑦 ≪ 𝑠) AND 𝑏))
𝑦 ← 𝑦 ⊕ ((𝑦 ≪ 𝑡) AND 𝑐))
67
𝑦𝑘 ← 𝑦 ⊕ (𝑦 ≫ 𝑙) (4)
The generation process of Mersenne Twister is seen in Fig. 1. In short, 3 words in the state vector are used in the
transform step to generate a new word, which is then fed to the tempering block to get the final output.
FIGURE 1. Generation Process of Mersenne Twister
The Mersenne Twister Algorithm has different sets of parameters as mentioned in [10]. But the parameters as
shown in Table 2 is referred as MT-19937 which has a period of 219937 − 1. Other parameters exist for MT-11213A
and MT-11213, but MT-19937 has the longest period and is most commonly used.
PROPOSED ARCHITECTURE
The straight forward implementation of the MT algorithm based on Fig. 1 needs a 32x624 dual port BRAM to
store the entire seed. From the recurrence (1), 3 RD (read) and 1 WR (write) operation are needed for each state update
step. Since the BRAM can only perform 2 operations at once, two cycles would be needed to generate one random
number.
If a throughput of one sample per cycle is desired, a basic implementation for this would be to use 3 BRAM, all
containing the entire state. Such a configuration allows 3 RD operations per cycle, while updating each of the BRAM
with the new state; however, this is not an efficient design as 3 BRAM are required to realize it.
TABLE 2. MT19937 Period and Tempering Parameters
Period Parameters Tempering Parameters
w = 32 u = 11
n = 624 s = 7
m = 397 b = 0x9D2C5680
68
r = 31 t = 15
a = 0x9908B0DF c = 0xEFC60000
l = 18
These basic implementations reveal the biggest obstacle: the BRAM only has two ports, while the recurrence
requires four operations.
The proposed design makes use of the 64-bit width simple dual port (SDP) BRAM. The 624-word state vector is
stored by concatenating the even and odd states (see Fig. 2) so that we end up with a 64x312 BRAM. Port A is
essentially performing two 32-bit RD operations in one cycle.
FIGURE 2. State Organization Scheme for SDP RAM
FIGURE 3. Timing Diagram of the Proposed Architecture
Fig. 3 is the timing diagram of the proposed MT RNG architecture. Because the RD port takes two states at once,
there will be times when data is read in advance. These are temporarily stored in three registers. Register 1, provides
a one cycle delay to the MSB of register 2. The input of register 2 alternates between register 3 and the upper half of
port A. Register 3 provides a one cycle delay for the lower 32 bits of port A.
69
FIGURE 4. Circuit Diagram of the Proposed Architecture
Fig. 4 shows the resulting circuit diagram. The control unit and MUX set to realize the above timing diagram. The
Write_Enable signal selects which half of the address is overwritten with xk+624.
TABLE 3. Address Generation Algorithm of the Proposed Architecture
Input(s):
counter0, counter1 : read address counters
counter2 : write address counter
p : number of pipeline stages
Output(s):
addressA : read port address
addressB : write port address
write enable : 8-bit enable 1: counter0 = 0 2: counter1 = 396 3: counter2 = -p initialize counters 4: for each cycle do 5: addressB = counter2 6: if counter1 is even then 7: addressA = counter0 /2 take the upper 9 bits 8: write enable = ffff000016 9: else 10: addressA = counter1 /2 11: write enable = 0000ffff16 12: end if 13: for each counter do 14: if counter == 623 then 15: counter = 0 16: else 17: counter = counter+1 18: end if 19: end for 20: end for
The control unit provides the addressing for the BRAM. Three counters are maintained; counter 1 and 2 for port
A and counter 3 for port B. The generation algorithm for these addresses are shown in Table 3. From the diagram it is
clear that the addresses count up at half the clock frequency. Assuming the number of pipeline stages is 1, the address
generation of the WR port lags the RD port by one cycle.
70
RESULTS AND DISCUSSION
The proposed architecture is implemented on Kintex 7 FPGA. The numbers generated by the hardware are passed
to the PC via UART. The numbers are formatted so that it will be read properly by the test software. There are two
softwares used for the randomness test, the Dieharder and the NIST. The utilized resources for the implementation is
also obtained and compared with previous studies. The results are presented in the next subsections.
Randomness Tests
The sequence produced by the proposed architecture was tested for quality of randomness using Dieharder and
NIST. The tests were run 10 times each and a summary is presented in Table 4. The sequence was determined to pass
both tests and can be said to be highly random.
TABLE 4. Results of NIST and Dieharder
Test Suite Passed Weak Failed %Passed
NIST 1880 - 0 100%
Dieharder 1087 52 1 95.35%
TABLE 5. Comparison Table of Resources Used
Platform Slices LUTs FFs BRAM Frequency
MHz
Throughput
Msps
Period Passed
Both tests
Virtex 7
Proposed Architecture 42 128 124 1 515.331 515.331 219937 − 1 Yes
[4] [Saraf and Bazargan,
2017]
53 149 118 1 505.31 505.31 219937 − 1 Yes
Virtex 6
Proposed Architecture 53 135 124 1 383.289 383.289 219937 − 1 Yes
[3] [Li et al., 2012] 41 152 143 2 450 450 219937 − 1 Yes
Kintex 7
Proposed Architecture 42 128 124 1 515.331 515.331 219937 − 1 Yes
Resource Utilization
Table 5 shows the resource utilization of the proposed architecture as compared to previous implementations.
Comparing the proposed architecture with [3], the slices were higher by 12 but the LUTs and FFs were reduced to
11.18% and 13.29% respectively. More significantly, the number of BRAMs used was reduced by 50%. On the other
hand, comparing it with [4], the FFs of the proposed architecture were higher by 5.08% but the slices and LUTs were
71
reduced to 20.75% and 14.09% respectively.
The past design that closely matches the amount of resources used for the proposed architecture is [3]. The
advantage of the design proposed in this work is that the output sequence strictly follows MT-19937, unlike the out-
of-order operation implemented in [4].
CONCLUSION
This research implemented the Mersenne Twister algorithm, specifically MT-19937, on an FPGA. The proposed
architecture used specific features of the BRAM to efficiently use resources. The resulting implementation was of
high throughput while using less resources than the current literature. The architecture was verified to output a random
sequence that exactly matches software MT when given the same seed. Statistical tests of Dieharder and NIST were
also performed to verify the quality of randomness of the output sequences generated with software and hardware.
ACKNOWLEDGMENT
We would like to acknowledge Dr. Lawrence Materum for suggesting the area of hardware RNGs, Engr. Roderick
Yap for his invaluable feedback.
REFERENCES
[1] D. E. Knuth, The Art of Computer Programming, Volume 2 (3rd Ed.): Seminumerical Algorithms. Boston, MA,
USA: Addison-Wesley Longman Publishing Co., Inc., 1997.
[2] E. T. D. Pellerin and M. Xu, “Fpga-based hardware acceleration of c/c++ based applications.,” Impulse
Accelerated Technologies, Aug. 2007.
[3] Y. Li, J. Jiang, H. Cheng, M. Zhang, and S. Wei, “An efficient hardware random number generator based on the
mt method,” in 2012 IEEE 12th International Conference on Computer and Information Technology, pp. 1011–
1015, Oct 2012.
[4] N. Saraf and K. Bazargan, “A memory optimized mersenne-twister random number generator,” in 2017 IEEE 60th
International Midwest Symposium on Circuits and Systems (MWSCAS), pp. 639–642, Aug 2017.
[5] S. Chandrasekaran and A. Amira, “High performance fpga implementation of the mersenne twister,” in 4th IEEE
International Symposium on Electronic Design, Test and Applications (delta 2008), pp. 482–485, Jan 2008.
[6] P. Echeverr´ıa and M. L´opez-Vallejo, “High performance fpga-oriented mersenne twister uniform random number
generator,” J. Signal Process. Syst., vol. 71, pp. 105–109, May 2013.
[7] S. Banks, P. Beadling, and A. Ferencz, “FPGA Implementation of Pseudo Random Number Generators for Monte
Carlo Methods in Quantitative Finance,” in 2008 International Conference on Reconfigurable Computing and
FPGAs, pp. 271–276, Dec 2008.
[8] V. Sriram and D. Kearney, “An fpga implementation of a parallelized mt19937 uniform random number
generator,” EURASIP J. Embedded Syst., vol. 2009, pp. 7:1–7:6, Jan. 2009.
[9] X. Tian and K. Benkrid, “Mersenne twister random number generation on fpga, cpu and gpu,” in 2009 NASA/ESA
Conference on Adaptive Hardware and Systems, pp. 460–464, July 2009.
[10] M. Matsumoto and T. Nishimura, “Mersenne twister: a 623-dimensionally equidistributed uniform pseudo-
random number generator,” vol. 8, pp. 3–30, 01 1998.
72
Implementation of Seat Identification and Attendance
System using Beacons
Jung-In Choi1, b) and Hwan-Seung Yong2, a)
1 Software Education Center, Pusan National University, Busan, Korea 2Dept. of Computer Science & Engineering, Ewha Womans University, Seoul, Korea
a)Corresponding author: [email protected]
Abstract. Systems that provide services to users through wearable sensors or smart devices are increasing. Thus, research
on the use of Bluetooth low-energy beacons or near-field communication is expanding for detecting location in indoors.
This paper proposes a seat identification system that uses beacons to detect the micro-location and smart device to recognize
user's movement. Using the proposed equation, the smart device can calculate the distance between user(device) and
desks(beacons). With calculated distance, the smart device can predict a user’s seat. Further, we automatically check the
user’s attendance with user's movement using the distance and the acceleration sensor in a smart device. The accuracy is
greater than 90%.
Keywords— seat identification; beacon; attendance system; indoor positioning
1. INTRODUCTION
An increasing number of systems use various sensors to recognize user activities in real time and to provide
services. To minimize user inconvenience, the sensors are usually not attached to the body of the user, and activities
are recognized through a variety of smart-device sensors. However, GPS sensors in smart devices provide only limited
positioning capability in indoor environments; [1] thus, when users are indoors, these devices cannot detect precise
positions. Further, because image, video, and voice sensors record user activity and situation, it may compromise user
privacy. Therefore, this study uses Bluetooth low energy beacon for indoor positioning. The beacon devices are small,
inexpensive, and designed to run on batteries for many months or years [2].
In the present paper, we first propose a seat identification system that uses Bluetooth low energy beacons to identify
a specific indoor location of a user. We implement the proposed indoor positioning system in a typical classroom
environment. Using this system and acceleration sensor, we propose an automatic attendance checking system for the
users inside a classroom. To recognize a specific location, we determine that the most suitable interval of beacon
installation is in the range of 75 to 300 cm. This paper is organized as follows. We explore related works in Section
2, and we explain the overall structure of the seat identification and attendance system, and the environment and
algorithm used for identification in Section 3. In addition, we propose an equation to calculate the distance between
the user and the desk. In Sections 4 and 5, we experimentally test the system and describe the results.
2. RELATED WORK
In this section, we explore related works in the field of indoor positioning and sensor based attendance systems.
73
2.1 Indoor Positioning
According to the existing National Human Activity Patten Survey research, people spend an average of 87% of
their time indoors and 6% of their time in vehicles [3]. Therefore, people spend more time indoors than outdoors and
in vehicles. To identify the user location in an indoor environment, many researchers use various sensors. Ongoing
research studies that the use of WiFi signals [4, 5], radio-frequency identification (RFID) [6, 7], ultrasonic sensors [8,
9], near-field communication (NFC)-type sensors [10, 11], ultra-wideband (UWB) [12, 13], and Bluetooth low-energy
beacons [14-16] has been conducted for indoor location identification.
Rich Site Summary (RSS)-based wireless local area network (WLAN) indoor positioning system is popular for
localization and navigation. Ma et al. suggested a RSS-based indoor positioning system and proposed a green WLAN
to reduce power consumption. They used the fingerprint algorithm and singular value thresholding theory and built a
radio map [17]. Chen et al. used WiFi, pedestrian dead reckoning, landmarks, and Kalman filter to create a sensor
fusion framework and developed an Android app for real-time indoor localization and navigation. Their system
provides an average localization accuracy of 1 m, which indicates good performance; however, it is not suitable for
recognizing a specific location [5]. Yang et al. suggested an RFID-based real-time indoor positioning system using an
online sequential extreme learning machine (OS-ELM). Their system captured wireless signals from tags, which are
used to train past data. OS-ELM has a high learning speed and high generalization performance [6]. RFID can contain
the domain information and other data; it exhibits good performance and also has high accuracy. However, an RFID-
based system needs to include tags in every object and train its signal data, which is complex. In addition, some
environments require a much simpler processing system. Sim et al. proposed an indoor navigation wearable system
based on an ultrasonic sensor and visual markers for blind users. They use an ultrasonic sensor for obstacle detection.
In addition, blind users are required to wear glasses containing several sensors [9]. Ultrasonic sensors are suitable for
indoor navigation; however, obstacles cannot be detected in all directions, and blind spots may occur. These problems
can be solved using many sensors, but the navigation system can only learn a limited direction and not all directions
at once; thus, whether the aforementioned problems occur or not are irrelevant. Montañés et al. proposed a smart
indoor positioning system based on Quick Response code [10]. This technology uses small NFC chips, which can
cover indoor and outdoor environments. They do not require power usage and can contain information. Therefore,
this technology is suitable for navigation in any direction; however, it is not automatic. UWB is a wireless technology
that can transmit a large amount of data over a wide bandwidth without requiring high power usage. The transmission
distance is 10 times greater than that of Bluetooth. Khawaja et al. used UWB to detect the trajectory and speed of a
moving object [12]. Jeong et al. suggested a membership management algorithm with several tags. Their system
manages registration, positioning, and tracking [13].
All sensors for indoor positioning are suitable for every situation. Our aim is to identify specific location. In
additionally, the system is cost effective. Therefore, in the present study, a system design using Bluetooth low energy
(BLE) beacon was selected. BLE is an emerging wireless technology for short-range communication [17]. Sung et al.
proposed filter-based beacon distance measurement, and the accumulated difference between the locations was
reduced by approximately 31% [14]. Kriz et al. proposed a method to improve the positioning accuracy using WiFi
and BLE. They implemented their system in a real-life situation and showed that it improved the positional accuracy
by 23% [15]. Komai et al. proposed a system for monitoring the movement and activity details of elderly people using
BLE [16]. Table 1 presents an overview of sensors for location-based services adapted from [18].
TABLE 1. Overview of sensors for location-based services.
Suitable
Environment
Localization
Accuracy
Extra
Device on
User-Side
Power
Consumption Cost
GPS Outdoor 10m No High Moderate
RFID Indoor 1-3m Yes Low Moderate
UWB Indoor 10-50m Yes Low High
WiFi Indoor 2-5m No High Low
BLE Indoor and outdoor 1-5m No Low Low
74
Existing indoor location identification systems that use existing beacons employ only one beacon for a large area
and can only achieve recognition if the user is in a certain location or beacons are installed at fixed intervals and used
for indoor navigation.
2.2 Sensor based Attendance System
Taking attendance of students for every class is a time consuming process for teachers, especially when class size
is large [19]. Many research works have been conducted on an automatic attendance system with indoor positioning
sensors. Fransisco et al. [20] proposed an attendance registration system using RFID. Students enter a classroom and
place their cards near the reader to register that they are present in that class. The attendance logger will save this
attendance information in the RFID server. Vishal et al. [21] proposed a student attendance system based on Bluetooth
and RFID reader applications. RFID reader gets a student information, and Bluetooth helps a student record his or her
attendance. Fadi et al. [19] proposed a system that uses multi-factor authentication. QR code, captured facial data, and
an application are used to confirm the student identity. Recently, some devices have used personal biometric
information for authorization. Abhishek [22] proposed an attendance system using face recognition technique. The
system captures a student’s facial data and extracts features. Samuel et al. [23] also used the face recognition system
but it captures the whole classroom and recognizes all students. They obtained about 82% accuracy of seated students.
However, some students may feel uncomfortable when the system captures their image. Seifedine et al. [24] proposed
an attendance management system based on iris recognition. They scanned a student’s eyes using a sensor, extracts
the minutiae, and matches the result with that in their database. Using sensors, biometric information and a user’s
smart device can help to avoid students from cheating in terms of their attendance. The attendance cannot be registered
solely on the basis of sensors; students must be involved to a certain extent. The entire system cannot automatically
detect when a student leaves. Biometric information is difficult to duplicate but the information accuracy is lower than
that of the other sensors. The problem of low accuracy and the non-automated problem are solved when we use the
Bluetooth low energy beacon. In the research by Jung et al., [25] this system was used, and a beacon was installed in
a classroom to check the attendance of up to 100 people. Other research studies have also been performed using a
single beacon to check attendance [26-28]. However, in the existing research studies, although the locations of users
in a large area can be detected, specific locations cannot be recognized. In this study, we propose an implementation
of seat identification system, which can be used in classrooms to check student’s attendance, via beacons. Unlike
existing studies, our system can check each student’s seat by measuring the signal strengths from all the beacons.
3. SEAT IDENTIFICATION AND ATTENDANCE SYSTEM
In this section, we describe an overview of our seat identification and attendance system. Figure 1 shows it.
FIGURE 1. Overview of attendance system based on seat identification.
We propose a seat identification system to calculate the specific location of a user in an indoor environment. To
implement this system, we need to install beacons in the classroom. We consider that all students have their own smart
75
devices nowadays. Therefore, we set up a beacon in every desk, which transmits the data to the user’s smart device
using Bluetooth. A smart device is installed in the attendance system application programming interface (API), and it
detects the user’s seat. Furthermore, a smart device transmits a user’s seat information to the server, and it is registered
in the attendance system. After that, a smart device detects a user’s activity using an acceleration sensor. If a user
moves to another seat or outside the classroom, a device notifies the change to the server, and the smart device
recalculates the user’s seat.
FIGURE 2. Beacon-equipped desk intervals for attendance system.
To implement a seating system, beacons are installed in all classroom desks. The beacon signal transmits in the
shape of a fan, rather than a circle; thus, the beacon is attached to the front and center of each desk. Figure 2 shows
the diagram of beacon-equipped desks placed at different intervals according to the classroom properties. The four
desks at the left are at 75-cm intervals, and the four desks at the right are at 1-m intervals. The small circles in the
diagram indicate the beacon locations. The beacons are attached to the desks and the desks are arranged at regular
intervals. In this research, the regular intervals at which the beacons are installed represent the beacon matrix.
The experiments used BLE beacons by RECO Inc. and a Nexus 7 second-generation phone. The beacon transmit
power was set at -20 dBm. The transmit power was used to determine the distance between the beacon and the user.
The -20 dBm power transmits the signal at a radius of approximately 3 m. In this research, the API provided by RECO
was modified to receive the beacon signal.
3.1 Seat Identification and Attendance Algorithm
The attendance-checking system proposed in this research determines if a user enters the classroom carrying a
mobile device with Bluetooth. If the user enters the classroom, the system operates and the details are as follows:
1. Transmits signals from all beacons to the devices through Bluetooth [time, beacon ID, and distance (m)].
2. Saves the beacon signal in the device at 30-s interval 1 min after the user enters the classroom.
3. Calculates the average distance to each beacon every 30 s in the device and then deletes the signal data without
average distance result [start time, beacon ID, and average distance (m)].
4. Completes identification of the beacon attached to a particular seat: determines the closest signal to the user,
and if it remains for 1 min, transmits it to the server (beacon ID and user ID).
5. Considers the seat identified and runs the seat identification system: recognizes the user location according to
the classroom seating chart stored in the server and reflects it in the user attendance.
6. After the user’s device sends the user’s seat information to the server (attendance-checking system), the device
turns off the Bluetooth signal to reduce power consumption.
7. The acceleration sensor works instead of the beacon.
8. If the user moves his/her seat, the device turns on the Bluetooth and detects the user’s location.
If the user enters the area where beacons are installed, the beacon signal is sent by Bluetooth technology to his or
her mobile phone. At this time, all beacon signals in the classroom are transmitted. After the user sits down in a seat,
the user location is calculated; thus, starting at 1 minute after the beacon signals are received, the transmitted data are
saved every 30 seconds. The saved data are used to determine the average distance between the user and each beacon.
While this event is happening, if the beacon distance is measured to be less than 20 cm from the user, it is determined
to be the user seat, and this information is transmitted to the server. At this time, the user school number is transmitted
76
as a unique ID. If the seat transmitted to the server remains the same, it is recorded in the attendance-checking system
that the user has been in the same seat for more than 1 min. Subsequently, the beacon signal calculations are stopped,
and the device turns off the beacon signal and uses the acceleration sensor to detect user activity.
FIGURE 3. Acceleration signals for sitting, up and down, walking, jogging.
If the acceleration sensor detects the user standing, walking, running, or performing a similar activity, the user’s
device turns on the Bluetooth signal to detect the user’s location. Figure 3 shows a sample graph of the acceleration
sensor signal. In (a) and (b) activities, the signals are static because the user is in his/her seat. When he/she is not on
the seat, the graph of acceleration changes, especially z-axis.
Algorithm 1 shows a method for determining the user’s position as a seat. For verifying the user’s seat, the device
checks the z-axis of the acceleration sensor. If it exceeds the threshold, it means the user is moving. Thus, a user’s
device keeps the Bluetooth ON and recalculates the user’s location. If not, it means that the user is seated in the
recognized seat. The threshold may be different for each person. When a user’s location is confirmed as one specific
seat, the confirmed seat transfers to the attendance system.
3.2 Modified Friis Transmission Equation
The beacon transfers Timestamp, Major number, Minor number, Tx-power, RSSI(dB), and Proximity data to the
smart device. To measure the distance between the device and the beacons, the free-space Friis Model is used [29].
𝑃𝑟(𝜔) = 𝑃𝑡(𝜔)𝐺𝑡(𝜔)𝐺𝑟(𝜔)𝜆2
(4𝜋𝑟)2 ()
where Pt and Pr are the transmitted and received powers, respectively, Gt and Gr are the receiving and transmitting
antenna gains, respectively, λ is the wavelength at the operating frequency, and r is the distance between the
transmitting and receiving antennas. This result does not include the propagation losses, polarization mismatch, or
impedance mismatch at either the transmitting or receiving antennas [30]. According to the RECO Inc. beacon and
Apple Inc. iBeacon official websites, the accuracy of the distance between the devices and the beacon is 68% with the
77
offered API [31,32]. Under different Tx power settings, the estimated distance remains inaccurate [33]. In addition,
the beacon does not provide the direction. Thus, the precise location of the user cannot be known. When the distance
between the user and the beacons was 5 m in the experiment, the result of the Friis transmission equation was 1–10
m. To increase the distance accuracy rate, the current research proposes the use of the average result of the Friis
transmission equation. To choose the average time, the number of seconds required for accurate results is tested.
The experiment is performed in different environments, which are described as follows.
1. The distance between the beacons is 1 m, and the user is within 10, 20, and 50 cm of the beacon.
2. The distance between the beacons is 75 cm, and the user is within 10, 20, and 50 cm of the beacon.
FIGURE 4. Experimental results for different scenarios using the modified equation.
The experimental result of the average time is shown in Figure 4, which indicates an accuracy of more than 80%
in less than 30 seconds. Using the average values at 30 seconds, an error of 20% occurs; however, the error is less
than the 10 cm intervals. An error of approximately 10 cm indicates that variations due to the position or direction of
the user in possession of a smart device do not affect the system to a great extent. Therefore, this paper suggests using
an average of 30 seconds of the Friis transmission equation result. The modified equation is given by
r =1
𝑠∑ (
𝜆
4𝜋√
𝑃𝑡
𝑃𝑟𝐺𝑡𝐺𝑟 )𝑆
τ=0 ()
where τ is the total number of transmission data, s is number of transmission data for 30 seconds, and r represents
the distance between the beacon and the user.
4. PERFORMANCE EVALUATION RESULT
To identify the direction, the beacons were installed at narrow intervals. The beacons were installed at regular
intervals of 75 cm and 1 m, and the installation environment was configured in the server. In our previous study, we
experimented with 10 beacon matrices [34]. However, there was an error in measuring the distance. Thus, we tested
with 20 and 30 beacon matrices. The testing position of the user device was divided into the right side, left side, center
of the desk, and top of the chair; thus, each beacon test was performed at least four times. The distance between the
beacon and the smart device was within 30 cm. The time required for the experiment was 5 minute in which the
average distance from the user over 30 seconds was calculated.
78
(a) (b)
FIGURE 5. Distance measurement results with the device at Desk 22 (a) using 1 m intervals, (b) using 75 cm intervals.
Figure 5 shows the arrangement of 30 beacons and the results of experiment. Figure 5 (a), and (b) show the test
results after arranging the desks at 1 m intervals and 75 cm and placing the device at the right side of Desk 22, which
has a beacon attached. The numbers shown are the 30 seconds averages of the distances between the device and the
beacons. At Desk 22, the distance between the beacon and the device is 0.011 m (b) and 0.014 m (b), which is the
closest distance; thus, the current location is determined to be Desk 22. Table 2 lists the accuracy of each experiment.
TABLE 2. Distance recognition accuracy using beacons.
10 beacons 20 beacons 30 beacons
75cm 80% 90% 100%
1 m 90% 100% 100%
According to the results listed in Table 2, the accuracy increased as the number of beacons increased. When 10
beacons were used, 9 beacons were arranged in a square, and the beacon in the middle could not properly send its
signal, which resulted in errors. In this research, we considered this to be due to signal interference. However, 30
beacons were tested at all four positions, and the user desk was identified with 100% accuracy each time.
With 20 or more beacons installed, the user location could be identified with 100% accuracy when the desks were
at regular 1 m intervals, and with 30 or more beacons installed, 100% accuracy was attained for the 75 cm and 1 m
intervals. This system could be used in an actual classroom environment because the distance between the seats in a
university classroom is equal to or greater than 1 m. In a company meeting room, no individual desk is present;
however, if the beacons are installed according to the space between people, the intervals would be 1 m. In middle
and high school classroom, the desks are arranged at 75–95 cm intervals depending on the school. The results of
additional testing at 75 cm intervals showed 100% accuracy when 30 beacons were used. Therefore, we expect that
the system proposed in this research can also be used in middle and high school classrooms.
79
FIGURE 6. Attendance system interface.
Figure 6 shows the execution screen of the attendance system. When a user, who is a teacher, enters the ID and
password (a), the system opens and shows the calendar and time (d). After a user clicks on “Start Attendance Check”
(d), the system start checking the attendance (b). If a student is sitting in his/her seat, his/her smart phone recognizes
the seat and transfers it to the server, and it appears on the interface (e). If a teacher wants to know the information of
a seated student, he/she can click on the seat, and the system would show the required information (c). When a class
ends, the teacher clicks on “END” (e) and the system shows the resulting class attendance (f) and it saves in the server.
5. CONCLUSION
With the development of IoT devices and smart devices, research works and application of services on activity recognition are ever increasing. However, GPS sensors in smart devices provide only limited positioning capability in an indoor environment. When users are indoors, smart devices cannot detect precise positions. Therefore, many research works use different sensors for indoor positioning. All sensors for indoor positioning are suitable for different situation. Our work was intended at developing an automatic attendance system. For this, we first identified a specific location with no blind spots. We then designed a system keeping cost effectiveness in mind. For this, we designed a system using Bluetooth low energy beacon and a smart device, which has a built-in acceleration sensor.
The main contribution of our paper is the system’s ability to detect with very high accuracy the location of indoor users located at 75-cm or 1-m intervals using beacons and to detect a user’s movement using the acceleration sensor embedded in smart devices. The system presented in this research could determine the specific location of a user in a room and automatically identify the user seat by attaching beacons to the seats at 75-cm and 1-m intervals. After identifying the user seat, it transfers the data to server and the user location is recalculated when the user is not in his/her seat. A teacher and students do not have to experience the inconvenience of attendance checks. For future research, the locations of students will be analyzed to study their personal relationships and friendliness with one another. With location identification, the correlation between grades and seating location can also be understood. We plan to group the classroom seats into front, back, side, and middle groups and analyze the effect of the seat location on the grades.
80
ACKNOWLEDGMENTS
This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2018R1D1A1B07042967). This paper is part of the doctoral thesis of first author.
REFERENCES
[1] Y. Zhuang, L. Hua, L. Qi, J. Yang, P. Cao, Y. Cao, et al., “A survey of positioning systems using visible led
lights,” IEEE Communications Surveys & Tutorials, vol. 20, no. 3, pp. 1963-1988, 2018.
[2] R. Faragher, and R. Harle, “Location fingerprinting with bluetooth low energy beacons,” IEEE Journal on
Selected Areas in Communications, vol.33, pp.2418–2428, 2015.
[3] N.E. Klepeis, W. C. Nelson, W.R. Ott, et al., “The National Human Activity Pattern Survey (NHAPS): A resource
for assessing exposure to environmental pollutants,” Journal of Exposure Analysis and Environmental
Epidemiology vol.11, pp.231–252, 2001.
[4] L. Ma, and Y. Xu, “Received signal strength recovery in green WLAN indoor positioning system using singular
value thresholding,” Sensors, vol.15, pp.1292–1311, 2015.
[5] Z. Chen, H. Zou, H. Jiang, et al., “Fusion of WiFi, smartphone sensors, and landmarks using the Kalman filter
for indoor localization,” Sensors, vol.15, pp.715–732, 2015.
[6] Z. Yang, P. Zhang, and L. Chen, “RFID-enabled indoor positioning method for a real-time manufacturing
execution system using OS-ELM,” Neurocomputing, vol.17, pp.121–133, 2016.
[7] H. Xu, Y. Ding, P. Li, R. Wang, and Y. Li, “An RFID indoor positioning algorithm based on bayesian probability
and k-nearest neighbor,” Sensors, vol. 17, no. 8, pp. 1806, 2017.
[8] D. Gualda, J. Ureña, J. García, et al., “Locally-referenced ultrasonic—LPS for localization and navigation,”
Sensors, vol.14, pp.21750–21769, 2014.
[9] F. Ma, F. Liu, X. Zhang, P. Wang, H. Bai, and H. Guo, “An ultrasonic positioning algorithm based on maximum
correntropy criterion extended Kalman filter weighted centroid,” Signal, Image and Video Processing, pp. 1-9,
2018.
[10] W.C.S.S. Sim, and V.F. de Lucena, “Blind user wearable audio assistance for indoor navigation based on visual
markers and ultrasonic obstacle detection,” in Conf. 2016 IEEE International Conference on Consumer
Electronics (ICCE), Las Vegas, NV, USA, 7 January – 11 January 2016, pp.60–63.
[11] J. A. P. Montañés, A. M. Rodríguez, and I. S. Prieto, “Smart indoor positioning/location and navigation: A
lightweight approach,” IJIMAI, vol.2, pp.43–50, 2013.
[12] B. Ozdenizci, V. Coskun, and K. Ok, “NFC internal: An indoor navigation system,” Sensors, vol.15, pp.7571–
7595, 2015.
[13] W. Khawaja, K. Sasaoka and I. Guvenc, “UWB radar for indoor detection and ranging of moving objects: An
experimental study,” in Conf. 2016 International Workshop on Antenna Technology (iWAT), Cocoa Beach, FL,
USA, 29 February – 2 March 2016, pp.102–105.
[14] S. Jeong, J. B. Park, and S. H. Cho, “Membership management and positioning method of multiple objects using
UWB communication,” in Proc. International Conference on Communication, Computing and Virtualization
(ICCCV), vol.79, pp.640–648, 2016.
[15] Y. Sung, J. Kwak, Y. S. Jeong, et al., “Beacon distance measurement method in indoor ubiquitous computing
environment,” Advances in Parallel and Distributed Computing and Ubiquitous Services, vol.368, pp.125–130,
2016.
[16] P. Kriz, F. Maly, and T. Kozel, “Improving indoor localization using bluetooth low energy beacons,” Mobile
Information Systems, vol. 2016, 2016.
[17] K. Komai, M. Fujimoto, Y. Arakawa, et al., “Beacon-based multi-person activity monitoring system for day care
center,” in Conf. IEEE International Conference on Pervasive Computing and Communication Workshops
(PerCom Workshops), Sydney, NSW, Australia, 14 March – 18 March 2016, pp.1–6.
81
[18] C. Gomez, J. Oller and J. Paradells, “Overview and evaluation of bluetooth low energy: An emerging low-power
wireless technology,” Sensors, vol.12, pp.11734–11753, 2012.
[19] H. Zou, H. Jiang, Y. Luo, et al., “BlueDetect:An iBeacon-Enabled Scheme for Accurate and Energy-Efficient
Indoor-Outdoor Detection and Seamless Location-Based Service,” Sensors, vol. 16, no. 2, pp.268, 2016.
[20] F. Masalha and N. Hirzallah, “A students attendance system using QR code,” International Journal of Advanced
Computer Science and Applications, vol.5, pp.75–79, 2014.
[21] F. Silva, V. Filipe, and A. Pereira, “Automatic control of students' attendance in classrooms using RFID,” in Conf.
3rd International Conference on Systems and Networks Communications, 2008, ICSNC'08, Sliema, Malta, 26
October – 31 October 2008, pp. 384–389.
[22] V. Bhalla, T. Singla, A. Gahlot, et al., “Bluetooth based attendance management system,” International Journal
of Innovations in Engineering and Technology (IJIET), vol.3, pp.2319–1058, 2013.
[23] A. Jha, “Classroom attendance system using facial recognition system,” Int J Math Sci Technol Manag vol.2, pp.
4–7, 2007.
[24] S. Lukas, A. R. Mitra, R. I. Desanti, et al., “Student attendance system in classroom using face recognition
technique,” in Conf. 2016 International Conference on Information and Communication Technology
Convergence (ICTC), Jeju, South Korea, 19 October – 21 October, 2016, pp. 1032–1035.
[25] S. Kadry, and M. Smaili, “Wireless attendance management system based on iris recognition,” Scientific
Research and Essays, vol.5, pp.1428–1435, 2013.
[26] H. Jung, D. Shin, K. Cho, et al., “BLE-OTP authorization mechanism for iBeacon network security,” Journal of
KIISE, vol.42, pp.979–989, 2015.
[27] J. H. Lee, S. Y. Yang, and W. J. Kim, “The design of attendance management system with beacon,” in Conf. the
Korea Institute of Electronic Communication Sciences (KIECS). Geosan, Korea, vol. 9, no. 1, June, 2015, pp.
301–303.
[28] S. Noguchi, M. Niibori, E. Zhou, et al., “Student attendance management system with Bluetooth low energy
beacon and android devices,” in Conf. 18th International Conference on IEEE Network-Based Information
Systems (NBiS), 2015, Taipei, Taiwan, 2 September – 4 September 2015, pp.710–713.
[29] M. Y. Bae, and D. J. Cho, “Design and implementation of automatic attendance check system using BLE beacon,”
International Journal of Multimedia and Ubiquitous Engineering, vol.10, pp.177–186, 2015.
[30] C. A. Balanis, “Antenna Theory: Analysis and Design,” New York: John Wiley & Sons, 1997, pp. 88.
[31] D. M. Pozar, R. E. McIntosh, and S. G. Walker, “The optimum feed voltage for a dipole antenna for pulse
radiation,” IEEE Transactions on Antennas and Propagation, vol.31, pp.563–569, 1983.
[32] RECO Beacon Homepage. http://reco2.me/
[33] iBeacon for developers. http://developer.apple.com/ibeacon/
[34] P. Martin, B. J. Ho, N. Grupen, et al., “An iBeacon primer for indoor localization: demo abstract,” in Proc. the
1st ACM Conference on Embedded Systems for Energy-Efficient Buildings, vol.31, pp.190–191, 2014.
[35] J. I. Choi, and H. S. Yong, “Implementation of Closed Interval Indoor Positioning using Beacon,” in Proc. Asia
Pacific Proceedings of Applied Science and Engineering for Better Human Life, vol.12, pp.34–38, 2017.
82
Service Oriented Architecture to Integrate E-Commerce and
Social Media in Indonesia
B. Junedi Hutagaol 1,a), Dennis 1,b), Frido Oktaviandre 1,c), Matius Richard 1,d),
Valentinus 1,e) And Sfenrianto 1,f)
1)Information System Management Department, BINUS Graduate Program – Master of Information System
Management, Bina Nusantara University, Jakarta, Indonesia 11480
Corresponding author: a)[email protected] b)[email protected] c)[email protected] d)
[email protected] e)[email protected] f)[email protected]
Abstract – Customer E-commerce in Indonesia growth very fast and have three main behaviour, that not effective and
efficient . First, they read more than or equal to 10 e-commerce. Second, they looking for the promotion. Third, they become
more confidence if there is a recommendation from social media. In order to help e-commerce customer to buy a products,
we proposed service oriented architecture to integrate e-commerce and social media based on behaviour of Indonesian
people. Service oriented architecture will help customer to find a lowest price by comparing the product in all e-commerce,
searching for a promotion and also view the review from other people at social media and other platform. This system will
increase the efficiency and effectiveness people for searching the best price, find a promotion and review the product from
social media.
Keywords : Price comparison, promotion, social interaction, service oriented architecture (SOA).
INTRODUCTION
E-commerce growth very fast In Indonesia, as we can see the transaction of e-commerce in 2012 around 2,5 billion
rupiah and in 2013 it increase to 4,5 billion rupiah [1]. Customer of e-commerce in Indonesia have three main
behaviour deciding to buy products. First, read more than or equal to 10 e-commerce website to find the best price
before they decide to buy [2]. Second, Looking for the promotion, because promo is an important factor that can be
considered for buying on the e-commerce. There is an connection between promotion and product complementarity,
which leads to different Indonesian customer e-commerce purchasing intention [3]. Third, Indonesian e-commerce’s
customers have become more confidence if there is recommendation from known friend in social media, as well as
some previous customer testimonial in fan page especially in facebook, because 99% e-commerce in Indonesia have
facebook account [4].
Because of this three main behaviours, Indonesian e-commerce’s customers is wasting to much time for buying a
product. So in this paper we try to propose a new e-business concept to solve that problem and fulfill customer needs
by using Service Oriented Architecture (SOA) to aggregate all of the product detail and price from e-commerce in
Indonesia, customer social interaction combine with social interaction from social media e-commerce it self, and
media to share and find product promotion. All of customer’s needs above will be provided in one media. Customer
only visit this e-business to make decision to buy the product they need.
In our e-business we engage SOA because that can be implemented as Web service. So we can implement the e-
business in mobile, web platform or other platform [5]. SOA can connected system with reuse services for new purpose
83
then will reduce development cont. [6] In SOA model with the same capabilities of e-business terms of integration
and interoperability is faster and proposed to be applied [7].
The reset of the paper is arranged as follows: section 2 present related article for our SOA; section 3 describes the
methodology we used in our paper; section 4 describes our model implementation of serviced oriented architecture
and newly introduced website to integrate all product e-commerce and review in social media; section 5 the conclusion
of our paper. We conclude that the proposed architecture in resolving the issues we identified in user behaviour e-
commerce in Indonesia.
RELATED WORKS
To avoid lack of quality health staff in remote area especially in developing countries, health sector need to
integrate their system. in order to speed up the process of medical care at the hospital, Boukaye Boubacar Traore et
al. proposed integration between the Model Driven Architecture (MDA) approach and service oriented (SOA) for
improving telemedicine services. What needs to be considered in the development of this system is the availability of
services in either taking or updating patient data in real time. Services that are built should be modular and ensure the
functions provided are independent for each medical action to be performed. This journal takes an approach that
enables the decomposition of telemedicine applications as services that have the MDA concept. This concept has the
advantage of utilizing business processes and facilitating the integration of using SOA between different platform
applications such as mobile phones, tablets running in remote areas. [8]
Most of banking have different variety of system to support their business. This approach spend expensive cost in
operational. In this journal Li Zhang et al. propose an implementation banking system to integrate all business process
based on SOA. For succesfull purpose, it is important to integrate between business and technology closely. IT team
need to understand the business process clearly, in the other hand business team need to understand the technical
implementation ability. Li Zhang and team identify then analyze the banking business process based on user needs.
After identified and analyze the business process, they create a workflow to design, configure and execute the business
process. Business Process run in the different system. Reuse and the heterogeneous architecture are the two
fundamental problems needed to be resolved. SOA proposed consits of The presentation layer, The control layer, The
services transferring layer, The service components layer, The application deployment layer, The background systems
layer. This SOA will reduce duplicate services and create reusable services. [9]
SOA offer a reusable and flexible service which can reduce time and cost for rapid developing system. Those
things will be achieved while implemented well. In this paper Nils Joachim et al., try to define SOA governance
mechanism to develop SOA integration. They use survey as methodology. They gathered all the reason why
implement SOA and challenging they had got. Analyze and finalize the result with Partial Least Squares (PLS), The
most important SOA governance mechanisms are using standards, establishing clear service management processes,
increasing the qualification of employees, and facilitating IT/business communication. The SOA governances seen
from managerial, business analyst, and technical perspective. Managerial become easier to manage all services listed,
technical person must be easier to identify and reuse services, business analyst will validate the system business
process easier. The result will be help full for the current and next development using the same services. For conclusion
governance mechanism is the most important things to see while development SOA with integration, scalability,
modularity and reuse services. [10]
METHODOLOGY
The objective of this paper to solved the problem related to user behaviour of customer in e-commerce in Indonesia.
This analysis using systematics literature review and document literature.
84
FIGURE 1 Methodology Systematic Literature Review
We use systematic literature review as a guideline methodology, it is used to identify and review the problem of
e-commerce in Indonesia after that we search the literature that can solved the problem. And we manage and evaluate
the literature in order to acquire a good understanding of the used of the service oriented architecture and design the
architecture of the system that can aggregate the price of the product and promotion all around the e-commerce and
get the review and social interaction of the product in social media.
RESULT AND DISCUSSION
The main architecture of the proposed SOA are presented in Figure 2.
FIGURE 2 Service Oriented Architecture to integrate e-commerce and social media in one places
85
In Figure 1, the proposed SOA is divided into three layer. Each layer in SOA have different in functionality. The
three layer are: data layer, business layer and request and response presentation layer.
Request and response presentation layer is the most top layer application. It’s function is to separate user request
and response that will used to query the data. In this layer it’s represent the data retrieved from the service into the
user based on their platform. The business layer is the core of the SOA. In business layer we compose the service that
will used for the e-business in social media interaction, promotion aggregation and price comparison on the product.
It will used by presentation layer to showed to the user that access the data. Data layer is the lowest layer in our
proposed SOA. In this layer data harvested from the social media and e-commerce and stored it in database to be used
it when there is a problem in service in third party. I this paper our third party is e-commerce and social media.
The systems developed in our proposed SOA will be used to aggregate the price comparison, social interaction
and promotion data. We divided our proposed SOA into three feature that will be implement in our SOA system.
Social interaction data
FIGURE 3 Architecture for social interaction
We gathered the data from Application Programming Interface (API) from all e-commerce and social media. To
gathered the comment of the product in the e-commerce and the comment of the product in social media. We use the
service and registered in service registry and stored it in social interaction database to be used if there is a problem in
the service of social media and e-commerce API.
86
Promotion aggregation
FIGURE 4 Architecture for promotion
We gathered the promotion data from the social media and e-commerce API using our external service. The data
that already get it from e-commerce and social media stored in database using local service. Before the data stored the
data will be validated before processed it to database. The data that requested from user will be request it through
service registry to search in database or from API.
Price comparison
FIGURE 5 Architecture for price comparison
87
We gathered the data from e-commerce using external service and validate and cleans the data and transform it to
be stored in our formatted product database. The data will be accessed by user, if the user using our price comparison
system, our service registry will be connect to the API of e-commerce. If there are a trouble that makes the API offline
we will use our internal storage to provided the data.
CONCLUSION
We proposed the Service Oriented Architecture using systematic literature review to help customer in e-commerce
solve the problem finding promotion, price comparing and reviewing the products. This concept to provide users with
easy access and in-time service when needed. There are three major issues we have identified in customer behavior
which is not efficient and effective. First, customer difficult to find the best price. Second, customer difficult to find
the valid promotion. Third, customer difficult to find review product from other customer testimonial.
In this paper, we proposed a service oriented architecture to build e-business solutions, incorporating social
interaction data, promotion aggregation and price comparison, as well as dynamically external and local integration
services. Furthermore, the proposed architecture exploits added-value services like product review, valid promotion
and price comparison to provide services in real time and increased data accuracy.
The main contribution of this paper is focused on resolving issues discovered costumer needs for decision making
to buy products. Product information from many e-commerce, product promotion information combine from social
media and e-commerce, and customer social interaction in one media. All information covered in one system. This
system will reduce customer time to find product needed in effective and efficient way.
REFERENCES
[1] 1. C. L. a. T. R. Sri Astuti Pratminingsih, "Factors Influencing Customer Loyalty Toward Online,"
International Journal of Trade, Economics and Finance, vol. 4, no. 3, pp. 104-110, 2013.
[2] 2. E. S. R. O. Rully Agus Hendrawan, "Evaluation of E-Commerce Product Reviews Based on Structural,
Metadata, and Readability Characteristics," Procedia Computer Science, vol. 124, pp. 280-286, 2017.
[3] 3. Y. Zhang, J. Deng and Y. Xu, "The Effect of Different Price Promotion Ways on Consumers’ Purchasing
Intention," American Journal of Industrial and Business Management, pp. 1192-1208, 2017.
[4] 4. A. A. Syuhada and W. Gambetta, "Online Marketplace for Indonesian Micro Small and Medium,"
Procedia Technology , vol. 11, pp. 446-454, 2013.
[5] 5. S. R. Loya, K. Kawamoto, K. Kawamoto and K. Kawamoto, "Service Oriented Architecture for Clinical
Decision Support: A Systematic Review and Future Directions," Journal of Medical Systems, pp. 1-22,
2014.
[6] 6. D. W. M. J. X. S. J Clement, "Service-Oriented Reference Architecture for Smart cities," 2017 IEEE
Symposium on Service Oriented System Engineering, 2017.
[7] 7. V. G.-I. D. R.-F. Jose V. Espí-Beltrána, "Enabling distributed manufacturing resources through SOA: The
REST approach," Robotics and Computer–Integrated Manufacturing, vol. 46, pp. 156-165, 2017.
[8] 8. B. B. Traore, B. Kamsu-Foguem and F. Tangara, "Integrating MDA and SOA for improving telemedicine
services," Telematics and Informatics , vol. 33, pp. 733-741, 2016.
[9] 9. Y. W. F.-j. L. Li ZHANG, "The Banking System Reference Implementation based on SOA," Fourth
International Joint Conference on Computational Sciences and Optimization, 2011.
88
[10] 10. D. B. T. W. Nils Joachim, "The influence of SOA governance mechanisms on IT flexibility and," Journal
of Strategic Information Systems, vol. 22, pp. 86-101, 2013.
89
Power Transfer in Low Voltage Hybrid AC-DC Microgrids
Under Islanding Operation: Model and Simulation
Hasti Afianti 1, 2, a) Ontoseno Panangsang1, b) Adi Soeprijanto1, c)
1Department of Electrical Engineering, Institut Teknologi Sepuluh Nopember Surabaya, Indonesia 2Engineering Faculty, Bhayangkara University of Surabaya, Indonesia
a) Corresponding author: [email protected]
Abstract. The concept of microgrids is increasingly being used for our electricity, especially in remote areas, where the
use of Renewable Energy Sources (RES) is also developing. The microgrids can connect the power plant that uses RES,
which is known that the continuity of the power plant with RES depends on the RES itself. One of the problems that arise
in microgrids with RES is power allocation. The discussion of the transfer of active and reactive power in Hybrid AC -
DC microgrids model is the subject of this paper, the model and simulation are carried out on the islanded condition and
low voltage system using Simulink MATLAB. Simulations have been done in three different cases that prove the transfer
of active and reactive power in the system and the voltage in both sub microgrids are well maintained under various load
conditions.
Keywords— Hybrid AC- DC Microgrids, Low Voltage Distribution, Interlinking Converter, Islanding Operation
INTRODUCTION
Recently, the use of Renewable Energy Sources (RES) is increasing, especially in remote areas where there is no
electricity network. So that it will be very effective if using natural energy sources available in the area to supply
electricity with RES technology. Some popular RES technology like photovoltaic system, battery, fuel cell generates
naturally DC power.
Meanwhile, currently there are many devices that use DC sources, for example Light Emitting Diode (LED) for
lights, computers and others. This equipments use a converter if would be connected to the microgrids. So it would
be more efficient to build a DC microgrids, which connect DC sources and DC loads directly, it will reduce the use
of converters [1].
Recently DC distribution system or DC microgrids are being develops. DC microgrids can operate islanded, and
it is possible to operate connected with grid or AC microgrids.
There are many RES such as wind, micro hydro centrals, etc that produce AC power and the load on the AC
system are certainly more varied, such as lights, television, radio, air conditioning and engines. The AC system is
more developed compared to the DC system. AC to AC converter is usually used for voltage system control in
electric power distribution.
Combination between AC and DC microgrids is an effective way of using RES in remote areas, because the
natural resources generated can be either AC or DC power [1-3].
90
Hybrid AC-DC microgrids is microgrids that separates based on the power generated and its load, AC power
generated will be located with AC load, and DC power generated will be located with DC load. An interlinking
converter may connect the ac power generated with dc power generated energy.
The advantages of hybrid AC-DC microgrids in general can be summarized as follows[1-3]:
1. increase installed power capacity.
2. Reducing the use of power electronic equipments: converter, Power Factor Correction (PFC)
3. improve power quality on microgrid AC, because DC load will be directly connected to DC source so that it
doesn't cause harmonic pollution.
Active and reactive power transfer in the low voltage system becomes an interesting discussion. The nature of
the impedance line of the low voltage system is not the same as the impedance line on the medium or high voltage
system [4]. This makes some difference in the analysis of the completion of power transfer in the distribution
system.
In this paper we discuss the active and reactive power transfer in low voltage hybrid AC-DC microgrids under
islanding conditions. Simulations has been done in three different cases: isolation, connected between AC and DC
sub microgrids and maximum load condition in AC sub microgrids.
RESEARCH METHOD
Hybrid AC – DC Microgrid in Islanding Operation
In hybrid AC-DC microgrids there are 3 main parts, namely: 1. AC sub microgrids, 2 DC sub microgrids, and 3.
interlinking converter that connects AC and DC sub microgirds [3].
DG 1 DG 2 DG 3
DC 1
Source
DC 2
SourceDC Microgrids
AC Microgrids
Interlinking
Converter
AC LoadDC Load
PIC
Non Linier Load
AC
Bus 1
AC
Source
Bus 1
DC Bus 3
DC Bus 1 DC Bus 2
Bus 4
AC
Source
Bus 2
AC
Source
Bus 3
AC
Bus 2
AC
Bus 3
FIGURE 1 Hybrid AC – DC Microgrids
91
In islanding mode, the total load must be divided and managed independently by both sub microgrids, which
involve control strategies on interlinking converter to minimize microgrids dynamics. The right loading strategy is
also needed to maintain system stability.
Interlinking Converter
Hybrid AC-DC microgrids in islanded conditions have advantages in power fulfillment, because of the
possible transfer of power between the two sub microgrids, AC and DC models. For that, the interlinking converter
which is the link between AC sub microgrids and DC sub microgrids requires a control that is able to detect the real
conditions in the system.
DC
uga
ugb
ugc
PLL
iqid
+
+-
-
L
L
L
DC Voltage
Regulator
vdc
+
-
ia
ib
ic
vdc ref
id ref
iq ref
abcd-q
Current
Regulator
PWM
Modulator
FIGURE 2. Interlinking Converter
In Fig 2, the interlinking converter uses three legs with six Insulated Gate Bipolar Transistor (IGBT)s, and the
control uses a closed loop. Control input in the form of current and voltage from AC sub microgrids and voltage
from DC sub microgrids [5].To equalize the frequency and phase angle between the interlinking and control
converter systems, the voltage from the AC sub microgrids become input for the Phase Locked Loop (PLL) circuit,
before it is transformed to the d-q coordinate along with the current from the AC sub microgrids. The Voltage from
DC sub microgrids will be compared with the DC referance voltage. The results of this comparison are controlled
and become a reference which is then compared with the current from the AC sub microgrids that has been
transformed to dq coordinate. Finaly in the Pulse Width Modulation (PWM) modulator, the voltage from the d-q
coordinate is changed to abc coordinates, which then becomes input for the PWM generator and produces pulses
that move the switch on IGBTs in the interlinking converter.
Power Transfer in Low Voltage AC Sub Microgrids
In transmission or high voltage system the distance between conductors is much greater than in the
distribution or low voltage, that is because the flux linkages between the conducting phases are greater, so that the
external component of the line inductance is much higher in high voltage system than in the low voltage system, for
this causes inductive reactance high voltage line is three or more times higher than resistance, making resistance
contribution to the modular impedance value negligible.
Inductance and resistance values in the low voltage line are inversely proportional to the high voltage system,
the inductance value in the low voltage line is very small and the resistance value is large [8-9].Typical line
parameter in grid Rꞌ, Xꞌ and typical rated current for the high, medium and low voltage line is shows in Table I [8-9]
92
TABLE 1. Typical Line Parameter
Type of Line Rꞌ (Ω/km) Xꞌ (Ω/km) Rꞌ/ Xꞌ
Low Voltage 0,642 0,083 7,7
Medium Voltage 0,161 0,190 0,85
High Voltage 0,06 0,191 0,31
From Tab 1 the grid with high voltage has a high inductive value while at medium and low voltage resistive value is
higher. This condition indicates that the low voltage system is resistive.
FIGURE 3. Influence of active and reactive power for different line impedance ratios
(a) R/X = 0, (b) R/X = 1, (c) R/X = ∞
The relation of P', Q', P and Q for different ratios of inductance X against resistivity R from line impedances is
illustrate in Fig 3 [10].
In the inductive lines are mainly P '≅ P and Q' ≅ Q, while for resistive lines are mainly P '≅ -Q and Q' ≅ P. If the
ratio of R / X is unknown, a compromise might assume R and X are equal.
r
Vs ∠φsV∠φ S = P + jQ
FIGURE 4. Equivalent circuit in AC sub microgrids
The active and reactive power transfer at two connected low voltage sources are illustrated in Fig 4, assuming the
source voltage is from the generator (DG) and the bus.
The power transfer from generator to bus is described as [10]
(1)
Active and reactive power that flow accros between generator and bus are described as [8]:
(2)
(3)
(4)
Q
U
Q’
S
P P’ f
R Z
X
(a) (b) (c)
U
Q’
S
P
f
R
Z X
P’ Q
φ
φ U
Q’
S
P’ f
R
Z
X
- Q
P
φ
φ
93
Where are generator voltage amplitude, generator phase angle, bus amplitude voltage, bus phase
angle, ang ω, ωs are angular frequency of the generator and the bus, respectively, r represents the line resistance. The
formulation in 2 and 3 are for single-phase system. They multiplied by three if the three-phase real and reactive
power are interest [11].
Power Flow in Low Voltage DC Microgrid
In DC sub microgrids, the bus voltage depends on the DC voltage source and the load can be assumed pure resistive
value, as line impedance.
The active power for DC load is [13]:
PDC = VDC Ii (5)
Where Ii is the sum of the currents derived from the DC source and the current flowing into the interlinking
converter.
The active power transfered to AC sub microgrids :
Pg = PDC - Σ (PDC-AC , PLC ) (6)
Where, PDC-AC is the loss power which due to the conduction and commutation of the converter IGBT, where PLC is
the instantaneous power absorbed by the filter
Assuming, there is no power losses in the converter and the filter. Then
Pg ≈ PDC (7)
Only active power is transfered to AC microgrid:
Qg = 0 (8)
SIMULATION AND DISCUSSION
Hybrid AC – DC microgrids is modeled by 2 Distributed Generation (DG)s in the DC sub microgrids and 3
DGs in the AC sub microgrids, all of the DGs connected in 10 bus as Fig 1. Simulations have been performed with
three different cases, they are:
1. Isolated condition
2. Connected between AC and DC sub microgrids
3. Maximum load condition in AC sub microgrids
The system parameters are in Tab 2 for AC sub microgrids and Tab 3 for DC sub microgrids.
TABLE 2. AC Sub microgrids Parameter
AC Source Bus 1 AC Source Bus 2 AC Source Bus 3
Power 5 (KVA) 5 (KVA) 5(KVA)
Nominal Voltage 380 V 380 V 380 V
Nominal Frequency 50 Hz 50 Hz 50 Hz
AC Bus 1 AC Bus 2 AC Bus 3
Line impedance 0.2 + j 0.025 Ω/km 0.2 + j 0.025 Ω/km 0.2 + j 0.025 Ω/km
Load R = 2000 Ω R = 2000 Ω R = 2000 Ω
94
TABLE 3. DC Submicrogrid Parameter
DC Bus 1 DC Bus 2
Load 100 Ω 1000 Ω
Nominal Voltage 400 V 400V
In DC sub microgrids, the load given to DC Bus 1 is 1000 Ω and at 0.67 seconds is given an additional installed
parallel load of 500 Ω. While the DC Bus 2 is given an initial load of 100 Ω and an additional load of 1000 Ω which
is parallel installed as well. DC Bus 3 is a bus that connects DC sub microgrids with interlinking converter, this bus
is not given a source and load.
In AC sub microgrids there are 7 buses, namely AC Source Bus 1, AC Source Bus 2, AC Source Bus 3
connected with synchronous machines and AC Bus 1, AC Bus 2, and AC Bus 3 is a load bus, and the last one is AC
Bus 4, AC Bus 4 is a bus that connects all three AC buses with an interlinking converter and the bus is not given a
load. 2000 Ω load is given for each load bus, specifically for AC Bus 1 load is given through a converter, this is to
condition non-linear loads. At 0.67 seconds the system is given an additional load installed in parallel, on the AC
Bus 1 the additional load is given in the form of 500 ohms and an inductive load of 200 H, while the AC Bus 2 and
AC Bus 3 have an additional load of 2500 ohms.
First Case: Isolated condition
In this conditions, either AC or DC sub microgrids can fulfill their power requirements, but in this condition
there is no power transfer between the two sub microgrids. Fig 5 shows the DC power in DC bus 1 and 2. The bus
connected to the interlinking converter has no power, either on DC bus 3 on DC sub microgrids or on AC bus 4 on
AC sub microgrids.
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 105
0
200
400
600
800
1000
1200
1400
1600
1800
Time (s)
Pow
er
(Watt
)
DC Bus 1
DC Bus 2
Load Changes
(a)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 105
-50
0
50
100
150
200
250
300
350
400
450
Time (s)
Voltage (
V)
or
Curr
ent
(A)
Current (A)
Voltage (V)
(b) FIGURE 5. DC Sub microgrids in isolated condition (a) Power in DC Bus 1 and 2 (b) Voltage and Current in DC Bus 3
95
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 105
-10000
-8000
-6000
-4000
-2000
0
2000
4000
6000
8000
Time (s)
Active P
ow
er
(VA
)
(a)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 105
-5000
0
5000
10000
15000
20000
Time (s)
Reactive P
ow
er
(Var)
(b)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 105
-4000
-3000
-2000
-1000
0
1000
2000
3000
4000
5000
Time (s)
Active P
ow
er
(VA
)
(c)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 105
-10000
-8000
-6000
-4000
-2000
0
2000
Time (s)
Reactive P
ow
er
(Var)
(d)
FIGURE 6. AC sub microgrids in isolated condition (a) Active Power AC Bus 1 (b) Reactive Power AC Bus 1 (c) Active Power
AC Bus 2 (b) Reactive Power AC Bus 2
In Fig 6 it appears the difference in active and reactive power produced by AC Bus 1 and AC bus 2. The
provision of non-linear loads seems to make a difference in active and reactive power generated from AC bus 1 to
AC bus 2 and AC bus 3 which is not given a non linear load. The difference is more obvious when there is an
increase in load in parallel, where on AC Bus 1 resistive and inductive loads are added, while in AC Bus 2 and AC
Bus 3 only resistive loads.
Second Case: Connected between AC and DC sub microgrids
In this case, there is an interlinking converter that connects the DC sub microgrids and AC sub microgrids. With
the interlinking converter, the power from DC sub microgrids can flow to the AC sub microgrids, this can be seen
from the difference in power shape between after and before the two sub microgrids are connected. In FIGURE 7, it
is clear that active power and reactive power do not show fluctuations despite additional loads, and with the power
flow from DC sub microgrids active and reactive power on AC Bus 1 and AC Bus 2 and AC Bus 3 get the same
power flow.
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 105
0
200
400
600
800
1000
1200
1400
1600
1800
Time (s)
Pow
er
(Watt
)
DC Bus 1
DC Bus 2
Load Changes
(a)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 105
-100
0
100
200
300
400
500
Time (s)
Voltage (
V)
or
Curr
ent
(A)
Voltage
Current
(b)
96
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 105
-6
-4
-2
0
2
4
6x 10
5
Time (s)
Active P
ow
er
(VA
)
(c)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 105
-2
0
2
4
6
8
10
12x 10
5
Time (s)
Reactive P
ow
er
(Var)
(d)
FIGURE 7. Hybrid AC – DC microgrids in connected condition (a) Power in DC Sub microgrids (b) Voltage and
current in DC Bus 3 (c) Active Power in AC Bus 1 (d) Reactive Power in AC Bus 1
FIGURE 8. Active power of AC sub microgrids
FIGURE 9. Reactive power of AC sub microgrids
97
Third Case: Maximum load condition in AC submicrogrid
The peak load is obtained by giving the load periodically until the system reaches the limit of its ability to
maintain voltage stability. The results of the simulation show that the system can bear the burden of three times the
initial load. Differences in active and reactive power under normal load conditions and maximum loads can be seen
in the Fig 8 and Fig 9.
Differences in source and load conditions are given to test system reliability. Whether in an isolated or
connected condition with an interlinking converter, the voltage and current in the system can be maintained
properly, so that the power transfer to the system also runs well.
CONCLUSION
Hybrid AC-DC microgrids in island conditions have shown good performance, in isolated conditions or
connected conditions. The simulation results show that there is a good transfer of power from the dc sub microgrid
to the ac sub microgrid. Interlinking converters can smoothly transfer power even though load conditions change.
Simulations on Hybrid AC-DC microgrids models have been carried out in three different cases. Every simulation
gets results for power transfer performance and interlinking converters can keep system operation more stable in a
variety of load conditions.
REFERENCES
[1] Hasti Afianti, Ontoseno Panangsang, Adi Soeprijanto, “Management strategy of hybrid microgrid to reduce multiple conversion”, presented at International Conference on Electrical Engineering, Informatics and Its Education 2015 (CEIE- 2015), Malang, Indonesia, October 3, 2015
[2] Sung –Hwan Park, Jing Yon Choi, Dong-Jun Won,” Cooperative control between the distributed energy resource in AC / DC hybrid microgrid”, IEEE XPlore, presented at Innovative Smart GridTechnologies Conference, 2014
[3] Navid Eghtedarpour, Ebrahim Farjah,” Power control and management in an AC/DC hybrid microgrid”, IEEE Transaction On Smart Grid, Volume 5 Issue 3, pp: 1494-1505, 2014
[4] Xiong Liu, Peng Wang, Poh Chiang Loh, “ A hybrid AC/DC microgrid and its coordination control”, IEEE Transaction On Smart Grid, Vol. 2, No. 2, pp: 278-286, Juni, 2011
[5] Xiancheng Zheng, Fei Gao, Husan Ali, Huamei Liu, “A droop control based three phase bidirectional AC-DC converter for more electric aircraft applications”, Energies, Vulume 10, Issue 3, 2017
[6] H.M. El-Deeneb, M. I. Daoud, A. Elserougi, S Abdel-Khalik, S. Ahmed, A. M. Massoud, “Maximum power transfer of PV-fed inverter-based distributed generation with improved voltage regulation using flywheel energy storage systems, presented at IECON 2014, Dallas TX USA, 29 Okt – 1 Nov 2014
[7] Thomas Basso, “IEEE 1547 and 2030 standards for distributed energy resources interconection and interoperability with the electricity grid”, Technical Report NREL (National Renewable Energy Laboratory)/TP-5D00-63157, 2014
[8] Xiaochao Hou, Yao Sun, Wenbin Yuan, Hua Han, Chaolu Zhong, Josep M. Guerrero, “Conventional P-ω / Q-V droop control in highly resistive line of low-voltage converter-based AC microgrids”, Energies, Volume 9, Issue 11, 2016
[9] Chendan Li, Sanjay K. Chaudhary, Mehdi Savaghebi, Juan C. Vasquez, Josep M. Guerrero, “Power flow analysis for low-voltage AC and DC microgrids considering droop control and virtual impedance”, IEEE Transactions On Smart Grid, Vol. 8, No. 6, pp 2754-2764, 2017
[10] K. De Brabandere, B. Bolsens, J. Van den Keybus, A. Woyte, J. Driesen , R. Belmans, “A voltage and frequency droop control method for parallel inverters”, IEEE for Transactions on Power Electronic, volume 22, issued 4, pp 1107-1115, 2007
[11] Henry Louie, “ Off-Grid Electrical Systems in Developing Countries”, Springer, 2018
[12] N. Hamrouni, M.Jraidi, A. Cherif, “New method of current control for LCL-interfaced grid – connected three phase voltage source inverter: Revue des Energies Renouvelables”, 13(1), 2010
[13] Hasti Afianti, Mochamad Ashari, Ontoseno Panangsang, Adi Soeprijanto, Suyanto, “Power transfer enhancement in hybrid AC – DC microgrids”, Journal of Engineering and Applied Sciences Vol. 11, No. 7, pp: 1660-1664, 2016
98
Accuracy Performance of
Wireless Multipath Clustering Approaches
Antipas T. Teologo Jr.a) and Lawrence Y. Materum
Electronics and Communications Engineering Department
De La Salle University, 2401 Taft Ave., Malate, Manila, Philippines
a)Corresponding author: [email protected]
Abstract. This paper presents the four most common clustering approaches used in wireless multipaths – Gaussian
Mixture Model (GMM), K-Power Means (KPM), Ant Colony Clustering (ACC) and Kernel Power Density (KPD).
Clustering of multipath components is very important in wireless communications as most of the current channel models
are cluster-based. Clustering accurately the multipath components means an accurate channel model which will lead to a
reliable performance evaluation of a wireless network. A review of the basic concepts and procedures of the mentioned
clustering algorithms is provided including their various strengths and shortcomings. A focus on the accuracy is the main
highlight of this study showing that these four clustering approaches, considering different reference propagation
channels, claim to perform best. A common ground-truth propagation data, however, is necessary in comparing the
performance of these multipath clustering approaches.
Keywords—channel models; clustering methods; multipath channels; radiowave propagation
INTRODUCTION
In the field of mobile communications, channel modeling plays a big role most especially in terms of system
simulations and evaluations. The main objective of channel modeling is to characterize the multipath components
(MPCs) in various environments. To ensure a reliable system design and good performance evaluation in a wireless
communication system, an accurate channel model is necessary [1]. In channel modeling, two methods have been
typically used which are the clustered and non-clustered models. Over the past two decades, clustering of MPCs
gained much attention in the research community and is now the basis of various channel models nowadays such as
the European Cooperation in Science and Technology (COST) 259, COST 2100, 3GPP Spatial Channel Model, and
the European Wireless World Initiative New Radio (WINNER) due to the correspondence of the majority of the
signals apart from the line-of-sight to form as significant clusters of signal energy. Various parameters of MPCs
such as the angular spreads, delay, number and position are used in the clustered structure modeling to group MPCs
into clusters. Wider bandwidth requirement, which is increasing in demand, in some wireless communication
systems such as fourth generation (4G), fifth generation (5G) and multiple-input multiple-output (MIMO) systems
can be achieved by utilizing the cluster-based channel model.
99
FIGURE 1. Image of multipath clustering between BS and MS [2].
If cluster-based channel model is desired, parameterization of clusters’ position, number, delay and angular
spreads is important and this can be done through clustering of MPCs. Figure 1 shows a clustering scenario of
multipaths between a base station (BS) and mobile station (MS). Several multipath clustering algorithms have
already been proposed which are either manual (visual inspection of data) or automatic and some are combination of
both. Visual inspection has been widely used in the past [3][4][5][6][7][8] but because of its limitations most
especially in clustering high-dimensional data, there was the development of the different automatic clustering
algorithms for better channel modeling. However, finding an efficient and accurate clustering algorithm is still a
challenge.
In the field of machine learning, clustering analysis is very well discussed [9] but on how to use this in clustering
of MPCs in a wireless channel is still problematic since clustering results by automatic clustering approaches do not
all correspond to physical objects present in the wireless propagation channel. MPC contains attributes such as
delay, electric field, magnetic field, angle of departure, angle of arrival, polarization, Doppler frequency, and
launching/landing antenna/material-element, and incorporating the impacts of these attributes is the major concern
in MPC clustering. Some algorithms consider only the power and delay attributes while others consider all. There
are several developments already in automatic clustering but at the expense of accuracy and complexity. Despite of
these progress, automatic cluster identification still poses some limitations. Most of these existing algorithms suffer
their accuracy in clustering those MPCs because of the failure in considering that the data they are clustering varies
with the environment and movement of objects within that environment within a part or the entirety of the
transmitted signal. Attributes of real-world MPCs should be incorporated in the clustering algorithm to make sure
that it takes into account the actual set-up in the real-world environment. Another thing, prior information such as
the number of clusters is also required for most clustering techniques and this is also difficult for automatic
clustering since it is not known in. Also, most clustering algorithms still require user-specified parameters adding as
well into the list of its limitations [1].
The goal of this study is to show four clustering approaches which are recently used in wireless multipath
clustering, namely, Gaussian Mixture Model (GMM), K-Power Means (KPM), Ant Colony Clustering (ACC) and
Kernel Power Density (KPD)-based algorithm. These approaches were considered as they represent different types
of clustering algorithms in reference to the taxonomies of [9]. Each algorithm belongs to a unique category and there
is no duplication in the clustering type. The algorithms in their respective category have an improved performance in
terms of clustering wireless multipaths, however, each algorithm utilized different reference propagation data in
evaluating performance. These four techniques also utilize multiple parameters in their parameter settings, which
could impact the resulting clustering results. The development of these algorithms starting from their inception to
100
their current state are covered in the following sections including the different challenges and limitations that they
encounter in achieving accurate clustering performance. From the results, the authors emphasize directions that the
wireless propagation engineering community could do in order to attain the best multipath clustering methods.
ANT COLONY CLUSTERING
Ant colony clustering algorithm is a kind of swarm-intelligent method that is basically inspired by the behavior
of the ant colonies. This relatively new ant-based clustering algorithm came from the observation on how ants
perform their clustering of corpses and sorting of their larval.
Ant-based clustering model
Ants are modeled as the agents that randomly move in their given environment. They can pick up, transport and
drop objects scattered in the environment. The sorting and clustering of the elements are done in a way that ants tend
to pick up objects in the vicinity wherein they have the similar ones [10]. There is already a gradual shift in using
biologically-inspired algorithms in performing complex data processing by most of the recent researchers [11] and
as being inspired by the real ant’s behaviors, ant colony optimization is now making its way in data-clustering [12].
One of the problems, however, is to discover how blind animals such as the ants are able to determine the shortest
path to take in moving from their colony to feeding sources and back again. It was then discovered that pheromone
trails are used by the ants to communicate with each other to decide in which path to take. The pheromone substance
is being left on the ground by a moving ant to mark its taken path. A moving ant can detect the previously laid trail
and can decide to follow it leaving its own pheromone and thus reinforcing the trail [12]. The more the path is being
followed by the ants, the more the trail becomes attractive to be followed. Figure 2 shows the behavior of ants in
moving from points A to E. Figure 2a shows the path without any obstacle along the way. When an obstacle is
present along the way as shown in Figure 2b, ants at point B travelling from point A to E will have to decide if either
to turn left or right (or at point D if travelling from point E to A). Big part of the ant’s decision is based from the
intensity of the pheromone trails produced by the preceding ants. The higher the pheromone level, the higher the
probability that the ant will turn on that direction. For the first ant to reach point B or D, since there is no previous
pheromone yet, the decision of turning to left or right has the same probability. The first ant following path BCD
will reach point D before the first ant following path BHD. As a result, any returning ant from point E and
approaching point D will most probably take the path of DCB going back to point A because more ants are also
taking this path. Since more ants are taking the shorter path BCD, pheromone quantity increases in this path and will
therefore give a higher probability that any single ant to follow will decide to follow this path.
101
In the classic ACC method, it starts with the projection of a data object onto a virtual workspace. Ant-agents are
being designated to perform common tasks such as to randomly pick up objects and drop these based on the local
environment similarity. However, in this kind of set-up, a large computation resource is needed as a result of huge
amount of aimless or random movements of the ants which can also lead in slow convergence. With this problem
and by exploiting the population self-similarity, a more efficient ACC was introduced in the study of [13]. The first
step involves the modeling of each MPC as a virtual ant-agent iO . Next, these agents are then projected onto a
virtual 2-D workspace, which is an amplitude-time plane (x,y). Correspondingly,
)()( , k
i
k
i
k
i yxO = (1)
denotes the i -th agent’s position during the k -th iteration. An environment similarity ),( )()( k
i
k
ir OO serves as a
guide in the agent’s movement. Given by ),( )( rOL k
i , the local environment of the i -th agent )(k
iO is being defined
by the range r and the neighbor agents )(k
jO giving,
rOO k
j
k
i )()( , (2)
Hence, r indicates the maximum radius distance that can be found between the current position of agent )(k
iO
and its neighbors of interest. Local environment similarity will be evaluated first by the i -th agent during each
iteration to decide on which region to move into. Once the local similarity is very weak, the i -th agent will find
other feasible regions but if the similarity is high, the agent will move closer to this region and update its position.
In this promising algorithm, time of arrivals and the decaying amplitude are the two large scale characteristics of
each MPC that were being explored. IEEE 802.15.4a ultra-wideband (UWB) channel model is used as the
propagation channel data and it is being validated using the probability of missing clusters, probability of wrong
clusters, relative error of mean deviations and probability of error clusters. ACC showed an attractive identification
performance in most scenarios as compared to wavelet-based method and linear regression. It provides compelling
processing paradigm on realistic UWB propagations and this method can be used for other time-series analysis.
Among the major factors in the success of this new automatic clustering algorithm are on the good design of its
population similarity and on the efficient rule for the ant movement.
FIGURE 2. Experimental set-up using ACC. (a) Ants following a path from points A to E. (b) Points A to E with obstacle and
ants choosing the two paths with equal probability. (c) More pheromone is laid on the shorter path [12].
102
K-POWER MEANS FRAMEWORK
KPM algorithm is just based from the K-Means algorithm with MPCs power being incorporated. As an
unsupervised learning type of algorithm, KPM clustering requires a priori knowledge for the initialization of
clusters. KPM differs from the standard K-Means due to the power weighting in the determination of MPC distance
[1]. K-Means algorithm is a kind of partitional approach which directly divides data objects into some pre-specified
number of clusters [9]. Typically, K-Means is utilized with Euclidean metric in order to determine the distance
between points and cluster centers which makes it easy to determine spherical or ball-shaped clusters in data.
Although it is a popular method, K-Means must be initialized with the number of clusters present in the radio
channel. This information is a priori unknown and improvement to solve this problem gave rise to the introduction
of K-Power Means in the different studies.
KPM algorithm incorporates the power of MPCs which makes it different from K-Means. KPM is a type of
automatic clustering algorithm that is parameter-based [14], [15] which minimizes the distance of angle-of-arrival
(AoA), angle-of-departure (AoD) and propagation delay time (DT) of the MPCs that belong to the same cluster.
KPM iteratively locates the centroids of the cluster in order to minimize the total sum of distances of each MPC to
its respective centroid and this centroid is said to be as the center of mass of the given cluster. Figure 3 presents the
idea of KPM. KPM algorithm begins with the initialization of K different cluster centroids k ,...,, 21
randomly. Then, assign each MPC sample x to the reasonable cluster centroid j and for each set x , set,
( ) )()( ,.minarg: k
jMPCxj
k xdc = (3)
where superscript )(k represents the k -th iteration and c represents the store indices of MPC clustering in the
k th iteration. Next step is to update the cluster centroids for each j , setting,
+
=
==
x x
k
xx
k
k
jjc
xjc
)(
)(
)1(
1
.1: (4)
then, just repeat the process involving (3) and (4) until convergence. KPM differs from the standard K-Means due to
the power weighting in the determination of MPC distance, MPCd
KPM utilization
In the study of [16], it utilized KPM with different initialization procedure to cluster synthetic data generated
from Saleh-Valenzuela (SV) model. By considering the different parameters such as delay, azimuth and power and
utilizing Calinski-Harabasz (CH), Dunn’s, Xie-Beni (XB) and Pakhira, Bandyopadhyay and Maulik (PBM) indices,
it was found out that none of the indices is always able to predict correctly the desired number of clusters but the XB
and D53 (Dunn’s) presented the best results with almost equal performances. In the study of [17], KPM algorithm
was used by taking the multipath component distance as a distance metric in parameter space as being applied in the
60 GHz channel model. In this study, Space Alternating Generalized Expectation-Maximization (SAGE) is used to
estimate the MPCs. It was discovered that the cluster peak power variation around the mean could appropriately
modeled using a log-normal distribution. Another study [18] used the KMP framework to cluster data from the
MIMO channel indoor environment at 11 GHz. Geometry of the scattering points (SPs) measured from the ray
tracer were exploited and used by KPM framework for clustering. Results indicate that the proposed method has
higher performance as compared with the conventional KPM in terms of characterization of channel and has less
complexity. Recently, study of [2] has also employed KPM for MPCs clustering together with SAGE algorithm to
estimate MPCs and multipath component distance-based algorithm for tracking.
103
This new method is called hybrid clustering approach. Using the MIMO channel model for subway station
scenario, this novel approach showed an effective way of clustering MPCs and was able to capture all the
characteristics of the clusters.
FIGURE 3. Main idea illustration of the KPM algorithm [1].
KERNEL POWER DENSITY-BASED ALGORITHM
KPD is also a density-based algorithm just like the density-based spatial clustering for applications with noise
(DBSCAN) but with some notable differences. KPD utilizes the kernel density instead of samples and considers
only the neighboring points for the density computation. Aside from the addition of power in the clustering, the
relative density and a threshold are used for the determination of the connectivity of two clusters [1]. First step in
KPD algorithm is the calculation of the density, x , of each MPC sample using K nearest MPCs as indicated in (5).
( )
−−
−−
−−=
x
xyRxyTxy
KyKy
yRxR
Ky
yTxT
Ky
yx
yx
,
,,
,
,,
2
, ,,
exp.exp.exp).exp(
(5)
where y represents an MPC which is arbitrary and such that xy . For MPC x , xK gives the set of K nearest
MPCs and xKy(.), serves as the standard deviation in the domain of (⋅). Next, for each sample of MPC, is the
determination of the relative density in order to make sure that the density over various regions using (6) is
normalized. This will make sure that various clusters will have the same level of density.
max )( yxKy
xx
x
= (6)
After getting the relative density, searching for the key MPCs is next. Finding the key MPC will serve as the basis
for the initial cluster centroids. Then, for each MPC, neighboring MPC as in (7) with high-density will be
searched and connect into it
),(minarg:,
~
yxdxxyy
=
(7)
104
FIGURE 4. Main concept of the KPD clustering algorithm [1].
There is a possibility of connecting two MPCs to each other over multiple paths but those MPCs that connect to
an identical key MPC belong to the same group and considered as one cluster. Last step is the cluster merging
wherein each MPC is connected to its K nearest MPCs. In case that any two key MPCs are reachable and any MPC
in any path connecting the two key MPCs has relative density greater than its density threshold, the two key MPCs’
clusters will just be merged as one cluster [19]. Figure 4 shows the main illustration of the KPD algorithm.
KPD is another novel clustering framework that uses the kernel density and the power of MPCs. It uses the
relative density and only considers K nearest MPCs. In the study of [1], KPD-based algorithm obtained 100 percent
identification and showed the best performance as compared with KPM and DBSCAN. It also achieved fairly low
computational complexity which is ideal for cluster-based channel modeling and enables applications in MIMO
channels with no prior knowledge of the distribution of propagation MPC estimates. In [19], KPD is used and
validated using a simulated channel based on the 3GPP Spatial Channel Model Extended (SCME) MIMO channel
model where F measure is utilized to evaluate the algorithm’s performance. Different number of clusters were used
to test the algorithm and KPD performance is compared with KPM.
VARIATIONAL GAUSSIAN MIXTURE MODEL
As a generative model, GMM takes into consideration that all MPC dimensions exhibit in various proportions
the Gaussian distribution. With high accuracy, GMM can be used to approximate any given data set and can also be
regarded as a soft clustering community [2]. Figure 5 shows the graphical model for GMM. Random variables are
represented by nodes in circles while the model’s parameters are represented by nodes in squares. Random variables
are those in double-circled nodes while hidden random variables are in single-circled nodes. Variable Z is used to
represent the component needed to produce an observed sample x . Using the graphical model, nodes distribution
are,
kkZ == )Pr( (8)
And,
);;()()|( kkx xNxpkZxXp ==== (9)
Where represents the mixing coefficients, );;( kkxN gives the density function of the Gaussian probability
and k is the mean and k denotes the covariance matrix. The Gaussian probability density function is given as:
105
)det()2(
)()(2
1exp
)(
1
k
D
k
k
T
k xx
xN
−−−
=
−
(10)
In GMM, channel multipaths’ statistical characteristics are the basis for clustering. The K basic Gaussian
distributions are linearly combined with each distribution has its own mean and covariance. Each Gaussian
distribution is called a component. GMM can give close approximation of any continuous density function by
adjusting the means, covariances and coefficients, and by utilizing adequate number of Gaussian distributions. It has
a lot of advantages and because of this, it is commonly used in different areas such as object detection [20], feature
selection [21], and classification [22], [23], [24].
FIGURE 5. Graphical model for GMM [2].
GMM in multipath clustering
GMM is another novel clustering algorithm introduced in the paper of [2] where the channel multipaths
covariance structure and mean information were incorporated to reveal the channel multipaths’ similarity. A
combination of Expectation-Maximization (EM) algorithm and Variational Bayesian (VB) algorithm for the search
of GMM parameters’ posterior estimation and for the optimization of GMM parameters, respectively, was done. EM
can be used in estimating the parameters of the GMM but it has the tendency to yield a covariance matrix that is
singular which will make its variance along the principal axis becomes zero.
EM algorithm also needs some cross-validation methods in order to determine the optimal number of
components, making it not ideal for training the GMM. Because of these shortcomings, Bayesian theory is being
applied for the training of GMM, enhancing the searching ability of EM and for the faster determination of the
optimal number of components. Results showed that Variational GMM model has better performance than K-Means
and it was also discovered that GMM needs a greater number of multipaths than the number of model parameters.
106
PERFORMANCE COMPARISON OF THE CLUSTERING ALGORITHMS
TABLE 1. Comparison of the four clustering algorithms.
Clustering
Algorithm
Algorithm
Category
Propagation
Channel Data
Parameters Used Results/Remarks
Ant Colony
Clustering
(ACC) [13]
Miscellaneous IEEE 802.15.4a ultra-
wideband channel
Amplitude and time
of arrivals of
multipath
components
Attractive identification
performance of ACC was
obtained in most scenarios
K-Power
Means (KPM)
Framework
[18]
Partitional MIMO channel
indoor environment
at 11 GHz
Delay time; angle-
of-arrival (AoA)
and angle-of-
departure (AoD)
Proposed method has higher
performance in terms of
characterization of channel
and has less complexity than
the conventional KPM
Kernel-Power
Density
(KPD)-Based
Algorithm [1]
Kernel-based Synthetic and
measured 3GPP 3D
MIMO channel
Model
Azimuth of arrival
(AoA) and
elevation of arrival
(EoA)
KPD obtains 100 percent
correct identification; KPD
shows best performance using
F measure and Silhouette
coefficient as compared to
KPM and DBSCAN
Variational
Gaussian
Mixture
Model
(GMM) [2]
Mixed model Synthetic Data Sets
and Outdoor-to-
indoor data (O2I)
delay; elevation-of-
arrival (EoA);
elevation-of -
departure (EoD);
azimuth-of-arrival
(AoA) and azimuth
of departure (AoD)
For synthetic data sets, GMM
obtained better clustering than
K-means; For O2I Channel
Data, CI values for GMM was
better than K-means
Table 1 gives the comparison of the four clustering algorithms. It can be seen that the four techniques use
different propagation channel data in evaluating their performances. It can also be observed that each algorithm has
considered different parameters although common are the angles (or direction) of the multipath components. For the
accuracy results, mostly are not giving a numerical value of its accuracy. Moreover, each multipath clustering
approach utilizes different method in evaluating its accuracy performance. The lack of common accuracy measure
and reference propagation data makes it difficult for the wireless propagation engineering community to compare
which approach is really the appropriate one to use.
CONCLUSIONS AND RECOMMENDATIONS
Continuous improvement on how to accurately cluster MPCs have been very evident in the considered clustering
techniques. From KPM to ACC to the novel methods of KPD and GMM, there have been a lot of efforts already to
optimize the different parameters of MPCs to arrive in an improved algorithm.
Choosing which among of these algorithms will perform best is still a challenge since their accuracy in terms of
clustering could not really be validated. Most common, each algorithm has been improved based on a certain
scenario or just for a certain parameter only and its performance comparison is limited only to one or two other
107
clustering algorithms. None of these studies has proved already that it could be the best choice, in whatever type of
environment, for wireless multipath clustering. A common base line reference (i.e. a ground-truth) as well as
accuracy metrics should be utilized in order to compare these multipath clustering approaches and a standard
evaluation of the accuracy of these techniques is highly recommended.
ACKNOWLEDGMENTS
The authors would like to thank the Commission on Higher Education (CHED) of the Philippines, Office of the
Vice Chancellor for Research and Innovation (OVCRI) and the University Research Coordination Office of De La
Salle University for all the support in making this study a possible one.
REFERENCES
[1] R. He, Q. Li, B. Ai, Y. L.-A. Geng, A. F. Molisch, V. Kristem, Z. Zhong, and J. Yu, “A kernel-power-
density-based algorithm for channel multipath components clustering,” IEEE Transactions on Wireless
Communications, vol. 16, no. 11, pp. 7138–7151, 2017.
[2] Y. Li, J. Zhang, Z. Ma, and Y. Zhang, “Clustering Analysis in the Wireless Propagation Channel with a
variational Gaussian Mixture Model,” IEEE Transactions on Big Data, 2018.
[3] M. Toeltsch, J. Laurila, K. Kalliola, A. F. Molisch, P. Vainikainen, and E. Bonek, “Statistical characterization
of urban spatial radio channels,” IEEE Journal on selected areas in Communications, vol. 20, no. 3, pp. 539–
549, 2002.
[4] J. Laurila, K. Kalliola, M. Toeltsch, K. Hugl, P. Vainikainen, and E. Bonek, “Wideband 3D characterization
of mobile radio channels in urban environment,” IEEE Transactions on antennas and propagation, vol. 50,
no. 2, pp. 233–243, 2002.
[5] K. Yu, Q. Li, and M. Ho, “Measurement investigation of tap and cluster angular spreads at 5.2 GHz,” IEEE
Transactions on Antennas and Propagation, vol. 53, no. 7, pp. 2156–2160, 2005.
[6] L. Vuokko, P. Vainikainen, and J. Takada, “Clusters extracted from measured propagation channels in
macrocellular environments,” IEEE Trans. Antennas Propag., vol. 53, no. 12, pp. 4089–4098, Dec. 2005.
[7] C. Oestges and B. Clerckx, “Modeling outdoor macrocellular clusters based on 1.9-GHz experimental data,”
IEEE Transactions on Vehicular Technology, vol. 56, no. 5, pp. 2821–2830, 2007.
[8] Q. Pei, B. Kang, L. Zhang, K. R. Choo, Y. Zhang, and Y. Sun, “Secure and privacy-preserving 3D vehicle
positioning schemes for vehicular ad hoc network,” EURASIP Journal on Wireless Communications and
Networking, 2018.
[9] R. Xu and D. Wunsch, “Survey of clustering algorithms,” IEEE Trans. Neural Netw., vol. 16, pp. 645–678,
May 2005.
[10] W. Gao, “Improved ant colony clustering algorithm and its performance study,” Computational intelligence
and neuroscience, vol. 2016, p. 19, 2016.
[11] P. Espejo, S. Ventura, and F. Herrera, “A survey on the application of genetic programming to
classification,” IEEE Transactions on Systems, Man, and Cybernetics - Part C: Applications and Reviews,
vol. 40, pp. 121–144, 2010.
[12] M. Dorigo, V. Maniezzo, and A. Colorni, “Ant system: optimization by a colony of cooperating agents,”
IEEE Transactions on Systems, Man, and Cybernetics, vol. 26, pp. 29–41, 1996.
[13] B. Li, C. Zhao, H. Zhang, Z. Zhou, and A. Nallanathan, “Efficient and robust cluster identification for ultra-
wideband propagations inspired by biological ant colony clustering,” IEEE Transactions on
Communications, vol. 63, no. 1, pp. 286–300, 2015.
[14] N. Czink, P. Cera, J. Salo, E. Bonek, J. p. Nuutinen, and J. Ylitalo, “A Framework for Automatic Clustering
of Parametric MIMO Channel Data Including Path Powers,” in IEEE Veh. Technol. Conf., 2006, pp. 1–5.
[15] N. Czink, R. Tian, S. Wyne, F. Tufvesson, J. Nuutinen, J. Ylitalo, E. Bonek, and A. Molisch, “Tracking time-
variant cluster parameters in MIMO channel measurements,” in 2007 Second International Conference on
Communications and Networking in China, 2007.
108
[16] S. Mota, F. Perez-Fontan, and A. Rocha, “Estimation of the number of clusters in multipath radio channel
data sets,” IEEE Transactions on Antennas and Propagation, vol. 61, no. 5, pp. 2879–2883, 2013.
[17] C. Gustafson, K. Haneda, S. Wyne, and F. Tufvesson, “On mm-wave multipath clustering and channel
modeling,” IEEE Transactions on Antennas and Propagation, vol. 62, no. 3, pp. 1445–1455, 2014.
[18] P. Hanpinitsak, K. Saito, J. Takada, M. Kim, and L. Materum, “Multipath Clustering and Cluster Tracking
for Geometry-Based Stochastic Channel Modeling,” IEEE Transactions on Antennas and Propagation, vol.
65, no. 11, pp. 6015–6028, 2017.
[19] R. He, Q. Li, B. Ai, Y. L. Geng, A. Molisch, V. Kristem, Z. Zhong, and J. Yu, “An automatic clustering
algorithm for multipath components based on kernel-power-density,” in 2017 IEEE Wireless
Communications and Networking Conference, 2017.
[20] C. Ari and S. Aksoy, “Detection of compound stuctures using a gaussian mixture model with spectral and
spatial constraints,” IEEE Transactions on Geoscience and Remote Sensing, vol. 52, pp. 6627–6638, 2014.
[21] M. H. C. Law, M. A. T. Figueiredo, and A. K. Jain, “Simultaneous feature selection and clustering using
mixture models,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 26, pp. 1154–1166,
2004.
[22] M. M. Dundar and D. Landgrebe, “A model-based mixture-mixture supervised classification approach in
hyperspectral data analysis,” IEEE Transactions on Geoscience and Remote Sensing, vol. 40, pp. 2692–2699,
2002.
[23] S. Tadjudin and D. A. Landgrebe, “Robust parameter estimation for mixture model,” in IEEE International
Geoscience and Remote Sensing Symposium Proceedings, 1998.
[24] S. Prasad, M. Cui, W. Li, and J. Fowler, “Segmented mixture-of-gaussian classification for hyperspectral
image analysis,” IEEE Geoscience and Remote Sensing Letters, vol. 11, pp. 138–142, 2014.
109
Dual-Channel Implementation of ITU-R M.1371
using XMC Microcontroller for a
Class A Automatic Identification System
Jomel Lorenzo, Gerino Mappatao a), Noriel Mallari,
Lawrence Materum, and Alexander Abad
Electronics and Communications Engineering Department
De La Salle University, 2401 Taft Ave., Malate, Manila Philippines
a) Corresponding author: [email protected]
Abstract. Increasing demand for maritime transport will increase the traffic in seaways. Implementing Automatic
Identification System (AIS) will help manage traffic by tracking the marine vessels and at the same time, provide security
and safety measures of the ship. This paper focuses on the low-cost dual-channel implementation of Class A AIS using
Infineon XMC microcontroller to accurately handle the messages on both channels. Class A is a type of AIS used for
large vessels. Under the ITU-R M.1371 standard, Class A AIS is required to have two dedicated receivers to cover all the
messages transmitted by other ships. This work proposes a low-cost firmware alternative with dAISy 2+ as a dual-
channel AIS receiver as a basis for implementation of the dual-channel in XMC. Intensive bench tests show that there
were no mismatched cases based on the comparison of the outputs from both receivers. The proposed dual-channel
Class A AIS implementation is more effective than dAISy since it outperformed dAISy per se based on dAISy’s missed
messages.
Keywords—navigation; marine navigation; radio navigation; radio frequency identification; sense and avoid; traffic
control; ships; marine vehicles; marine transportation; marine safety; marine accidents; automatic identification system
INTRODUCTION
Marine transportation has been here in the world for a long time dealing with transport of raw materials, people,
and others. Around 90% of the world is covered by maritime transport because of its cost effectiveness [1].
However, the demand is still increasing, and it is expected that it will increase by 60% after 30 years [2]. In order to
address the growing demand and traffic in the ocean, automatic identification system (AIS) [3–8] is required for
marine vessels. AIS is an electronic communications system that tracks AIS-equipped ships within range. In
addition to that, the messages of AIS includes other information about the ship such as identification, speed,
hazardous goods, and others in order to provide security and safety measures for the ships [9–12]. AIS transceivers
are subdivided into classes: Class A, and Class B. Class A AIS is for large vessels that are prioritize due to the larger
risks that such vessels bring at sea and shore, whereas Class B AIS is for small vessels [13]. In terms of exchanged
messages between AIS transceivers, Class B transmits reports at a lesser rate than Class A AIS. Such reporting is
due to the nature of vessels. This paper focuses on the dual-channel implementation for Class A AIS as per
requirement for ITU-R M.1371, where the authors’ aim is on the economical side as emphasized in [14]. AIS
operates in two channels: 87B (161.975 MHz) and 88B (162.025 MHz) [15]. These two channels use time division
multiple access (TDMA) method, which is a shared wireless channelization scheme that allows multiple users (for
AIS, i.e. vessels) to enter the channel through different time slots in order to utilize the whole bandwidth of one
channel. This paper aims to reduce the costs in implementing dual-channel. In order to achieve that, it makes use of
the data transfer protocol that uses open-source software which significantly reduces the development cost in
implementing Class A AIS [14]. The data transfer protocol is based on ITU-R M.1371 [15] for both receive
110
channels, and is implemented through the Infineon XMC microcontroller embedded system platform. This proposed
Class A AIS is different from the non-hardware-based AIS simulator [16] and relatively high-priced AIS
implementation like in [17] and [18] due to is additional sensors. The next section will discuss the design
consideration for this implementation, which includes the dual-channel receiver and its block diagram for integration
in the Class A AIS system. The section is followed by evaluation, which discusses the testing procedure, and
afterwards trailed by the data and results of the tests. The paper ends with conclusion of the achieved items.
DESIGN CONSIDERATIONS
Class A AIS has four channels that includes two TDMA channel for receiving and transmitting, one digital
selective calling (DSC) channel for receiving and one DSC channel for transmitting as shown in Figure 1, which is
the same as indicated in ITU-R M.1371 [15]. This paper will only implement two TDMA receive channel because
this is where the AIS messages resides and could be parsed for implementation of the transmit feature. According to
ITU-R M.1371, one frame is divided into 2250 time slots where the AIS could transmit in one channel. For
receiving two channels, the expected number of messages to be received must be less than 4500 per frame.
FIGURE 1. Schematic of Class A AIS [19].
Dual-Channel Receiver
In this paper, dAISy 2+ will be used as a basis for implementation of dual-channel in XMC microcontroller.
There two things that are important consideration in order to receive AIS message properly: bitstream and clock.
Bitstream is the message that decoded by the data transfer protocol while the clock is for the synchronization in
receiving the bitstream. Inside dAISy, the received signal is converted into a bitstream. This bitstream serves as an
input for the XMC microcontroller and then converted to NMEA message once it is decoded [14]. The NMEA
message will serve as the output for dAISy and XMC. The clock from dAISy functions as an input in the
microcontroller for synchronization to determine the timing in receiving the bits. Figure 2 shows the dAISy 2+
receiver with labels where the bitstream, clock, and NMEA message are taken.
111
FIGURE 2. dAISy 2+ dual-channel receiver
Block Diagram
Figure 3 shows the block diagram of the dual-channel implementation. As stated in the design consideration,
dAISy receives the signal and then produces a bitstream that will be decoded by the microcontroller. Afterwards it
converted to NMEA format that will be transmitted via UART to be displayed in the output terminal.
FIGURE 3. Block Diagram for Dual-Channel Implementation
METHODOLOGY
The testing procedure for validating the dual-channel capability of implementation on the XMC microcontroller
has three steps:
1. Setup of the antenna – The location is near the sea port to create an ideal testing ground for receiving.
2. Receive duration – In order to test the dual-channel capability, both dAISy and XMC are to receive within a
minimum time of three minutes, which is equivalent to three frames in TDMA in which it is less than or
equal to 6750 time slots for each channel.
3. Comparison of the outputs – The bitstream received by dAISy is the same as the input for the XMC.
Therefore, both output from dAISy and XMC must be the same to verify the receiving capabilities of the
implementation on XMC. A sample comparison of the outputs is shown in Figure 4.
112
DATA AND RESULT
Figure 4 shows the test setup and the output terminal that receives the messages from dAISy and XMC.
FIGURE 4. Test Setup
Table 1 shows only a part of the results from dAISy and XMC for comparison since the total number of
messages received is 927. Based on the comparison, the possible outcome could be matched, missed, and
mismatched—matched means that the messages from both receivers are identical; missed means that the other
receiver was not able to receive the message; mismatched means that the message was received by both receivers.
As can be seen in the results, dAISy2+ had some errors. The summary of all testing results is shown in Table 2.
Based on Table 2, there are no mismatched cases. This means that all messages are decoded properly and no
errors occurred for both receivers. Another finding is that the XMC received more messages than dAISy which
makes the XMC as more effective receiver than dAISy in terms of rapid processing. This is expected because XMC
is a microcontroller that is more powerful than the microcontroller used in dAISy. For a total of 927 messages, there
are 48 messages that was missed receive by dAISy.
113
TABLE 1. Comparison of Output NMEA message from dAISy 2+ and XMC
Output (dAISy 2+) Output (XMC) Remarks
!AIVDM,1,1,,B,18:rlu0P00`agoh8F0q`nOwT2D21,0*54 !AIVDM,1,1,,B,18:rlu0P00`agoh8F0q`nOwT2D21,0*54 Matched
!AIVDM,1,1,,A,19t6S>hP008agvT8F5v00?wUrhMl,0*3E !AIVDM,1,1,,A,19t6S>hP008agvT8F5v00?wUrhMl,0*3E Matched
!AIVDM,1,1,,A,18;NI800188aTTR8G:@1m1S`28Nm,0*5B !AIVDM,1,1,,A,18;NI800188aTTR8G:@1m1S`28Nm,0*5B Matched
!AIVDM,1,1,,A,16Sg1w80008adBJ8EkQP08;`24qH,0*15 !AIVDM,1,1,,A,16Sg1w80008adBJ8EkQP08;`24qH,0*15 Matched
!AIVDM,1,1,,B,18;:Wq0P028aRd08G>ppoOwb0D1O,0*48 !AIVDM,1,1,,B,18;:Wq0P028aRd08G>ppoOwb0D1O,0*48 Matched
!AIVDM,1,1,,A,18Lj838000`aTQt8Ehj<VA7d04qH,0*40 Missed
!AIVDM,1,1,,A,38;OdE8000`ai0b8F1LDfQMb02fQ,0*20 !AIVDM,1,1,,A,38;OdE8000`ai0b8F1LDfQMb02fQ,0*20 Matched
!AIVDM,2,1,1,B,58SJAoP2A0tg=LTgV21Hpj19D9V2222222222217
D0DA:7510JD283CS,0*10
!AIVDM,2,1,5,B,58SJAoP2A0tg=LTgV21Hpj19D9V2222222222217
D0DA:7510JD283CS,0*14
Matched
!AIVDM,2,2,1,B,888888888888880,2*26 !AIVDM,2,2,5,B,888888888888880,2*22 Matched
!AIVDM,1,1,,A,177cJw000g8aQrd8E33coqbP00T8,0*5E !AIVDM,1,1,,A,177cJw000g8aQrd8E33coqbP00T8,0*5E Matched
!AIVDM,1,1,,A,18M1KF000O`a>=h8DMJDQCVN2@9@,0*75 !AIVDM,1,1,,A,18M1KF000O`a>=h8DMJDQCVN2@9@,0*75 Matched
!AIVDM,1,1,,A,1856KIPP008ahCn8F4E00?vP2<21,0*29 Missed
!AIVDM,1,1,,B,10V9cEhP00`aai@8EwT6hOvP2<2=,0*74 !AIVDM,1,1,,B,10V9cEhP00`aai@8EwT6hOvP2<2=,0*74 Matched
!AIVDM,1,1,,B,38:tA91P01`a7ft8GE2MJgvP2000,0*10 !AIVDM,1,1,,B,38:tA91P01`a7ft8GE2MJgvP2000,0*10 Matched
!AIVDM,1,1,,A,18:tb9500u8aUrH8GIe96PRR04qL,0*79 !AIVDM,1,1,,A,18:tb9500u8aUrH8GIe96PRR04qL,0*79 Matched
!AIVDM,2,1,9,A,58;CeR427wlhAD7;?K@mB1<j0l4iDpLL5V22221
@7h:66w9=083@CRC0,0*41
!AIVDM,2,1,9,A,58;CeR427wlhAD7;?K@mB1<j0l4iDpLL5V22221
@7h:66w9=083@CRC0,0*41
Matched
!AIVDM,2,2,9,A,H20DPSmD`0m4`80,2*58 !AIVDM,2,2,9,A,H20DPSmD`0m4`80,2*58 Matched
!AIVDM,1,1,,B,19t6Su0P008agr@8F63P0?vR20S>,0*4A !AIVDM,1,1,,B,19t6Su0P008agr@8F63P0?vR20S>,0*4A Matched
!AIVDM,1,1,,B,18:uo60P01`aWar8Gkg7EOvR2<2L,0*04 !AIVDM,1,1,,B,18:uo60P01`aWar8Gkg7EOvR2<2L,0*04 Matched
!AIVDM,1,1,,B,34bABn0P008ah208F5i@0?vT:Rui,0*45 !AIVDM,1,1,,B,34bABn0P008ah208F5i@0?vT:Rui,0*45 Matched
!AIVDM,1,1,,A,177cJw000g8aQrd8E33coqbP00T8,0*5E !AIVDM,1,1,,A,177cJw000g8aQrd8E33coqbP00T8,0*5E Matched
!AIVDM,1,1,,A,18;3Uq00018adLV8GEOe901H0@Ir,0*0E !AIVDM,1,1,,A,18;3Uq00018adLV8GEOe901H0@Ir,0*0E Matched
!AIVDM,1,1,,A,18;CeR0P01`aWBR8GgHhT?wF2<2?,0*26 Missed
!AIVDM,1,1,,A,38Hs4f5000`afbl8F629`QI<0lMJ,0*74 !AIVDM,1,1,,A,38Hs4f5000`afbl8F629`QI<0lMJ,0*74 Matched
!AIVDM,1,1,,B,18;9370P018aeD28Fn>W4gwKr0Sc,0*65 Missed
!AIVDM,1,1,,B,18Lj838000`aTR@8Ehj<EQ7J0<2O,0*15 !AIVDM,1,1,,B,18Lj838000`aTR@8Ehj<EQ7J0<2O,0*15 Matched
!AIVDM,1,1,,B,177cJw000i8aQhP8E42t79eL04qL,0*32 !AIVDM,1,1,,B,177cJw000i8aQhP8E42t79eL04qL,0*32 Matched
!AIVDM,1,1,,A,10V9cEhP00`aai@8EwTFhOwH2<2>,0*1D !AIVDM,1,1,,A,10V9cEhP00`aai@8EwTFhOwH2<2>,0*1D Matched
!AIVDM,1,1,,B,18M1KF000O`a>EP8DLvlMCUJ2D3V,0*20 !AIVDM,1,1,,B,18M1KF000O`a>EP8DLvlMCUJ2D3V,0*20 Matched
!AIVDM,1,1,,A,18;0NB00008ahSD8EaDRA0IH08Jr,0*44 !AIVDM,1,1,,A,18;0NB00008ahSD8EaDRA0IH08Jr,0*44 Matched
!AIVDM,1,1,,A,19NWpM@0038aOqB8DfTABaeN00Rg,0*6E !AIVDM,1,1,,A,19NWpM@0038aOqB8DfTABaeN00Rg,0*6E Matched
!AIVDM,1,1,,A,19t6Su0P008agr88F64h0?wN2<28,0*78 !AIVDM,1,1,,A,19t6Su0P008agr88F64h0?wN2<28,0*78 Matched
!AIVDM,1,1,,B,1856KIPP008ahCl8F4Dh0?wN24qL,0*58 !AIVDM,1,1,,B,1856KIPP008ahCl8F4Dh0?wN24qL,0*58 Matched
!AIVDM,1,1,,B,18:tb9500v8aUdt8GHM96PSN0@L?,0*5D !AIVDM,1,1,,B,18:tb9500v8aUdt8GHM96PSN0@L?,0*5D Matched
!AIVDM,1,1,,B,177<jL0017`a1528ETF25AgN0D14,0*4E !AIVDM,1,1,,B,177<jL0017`a1528ETF25AgN0D14,0*4E Matched
!AIVDM,1,1,,A,18:kT78P00`af3D8FFlpvOwP20SA,0*4F !AIVDM,1,1,,A,18:kT78P00`af3D8FFlpvOwP20SA,0*4F Matched
!AIVDM,1,1,,A,14bABn0P008ah1R8F5hh0?wR:hM>,0*5E Missed
TABLE 2. Comparison Summary
dAISy 2+ XMC
Matched 879 879
Missed Messages 48 0
Total Number of Received Message 879 927
114
CONCLUSION
A low-cost dual-channel Class A AIS receiver using XMC microcontroller was implemented in this paper. The
implementation is compliant to the data transfer protocol since the messages were successfully decoded without
failure for both channels. The resulting messages were converted to NMEA message output format, which were also
without error. An off-the-shelf dual-channel receiver, dAISy 2+, was used in order to verify the dual-channel ability
of the implementation. The intensive validation of receiving sufficient messages from both receivers shows that the
implementation is reliable. Based on the evaluation results, the XMC implementation received more messages than
dAISy making it as a more rapid-processing receiver on top of the no mismatch runs. This proves that the
implementation of the Class A AIS is effective in receiving AIS messages. Overall, dual-channel feature in Class-A
AIS using XMC microcontroller was successfully implemented and achieved correct results.
ACKNOWLEDGMENTS
This research work was made possible by supports from Infineon and the Office of the Vice Chancellor for
Research and Innovation of De La Salle University, as well as a grant from Philippine Council for Industry, Energy
and Emerging Technology Research and Development of the Department of Science and Technology. Any opinions,
findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not
necessarily reflect the views of the acknowledged entities.
REFERENCES
[1] M. Fiorini, A. Capata, and D. D. Bloisi, “AIS data visualization for maritime spatial planning (MSP),”
International Journal of e-Navigation and Maritime Economy, vol. 5, pp. 45–60, 2016.
[2] Norweigian Shipowners Association, “Maritime Outlook Report,” 2018.
[3] S.-J. Chang, “Development and analysis of AIS applications as an efficient tool for vessel traffic service,” in
MTTS/IEEE TECHNO-OCEAN’04, 2004, vol. 4, pp. 2249–2253.
[4] International Electrotechnical Commission, “IEC 60945-Maritime navigation and radio communication
equipment and systems-General requirements-Methods of testing and required test results.” IEC, 2002.
[5] T. Eriksen and Ø. Olsen, “Vessel tracking using automatic identification system data in the Arctic,” in
Sustainable Shipping in a Changing Arctic, Springer, 2018, pp. 115–136.
[6] I. Harre, “AIS adding new quality to VTS systems,” The Journal of Navigation, vol. 53, no. 3, pp. 527–539,
2000.
[7] L. Heilig and S. Voß, “Information systems in seaports: a categorization and overview,” Information
Technology and Management, vol. 18, no. 3, pp. 179–201, 2017.
[8] B. Lin and C.-H. Huang, “Comparison between ARPA radar and AIS characteristics for vessel traffic
services,” Journal of Marine Science and Technology, vol. 14, no. 3, pp. 182–189, 2006.
[9] J. A. Creech and J. F. Ryan, “AIS The Cornerstone of National Security?,” The Journal of Navigation, vol.
56, no. 1, pp. 31–44, 2003.
[10] Z. Fang, H. Yu, R. Ke, S.-L. Shaw, and G. Peng, “Automatic Identification System-Based Approach for
Assessing the Near-Miss Collision Risk Dynamics of Ships in Ports,” IEEE Transactions on Intelligent
Transportation Systems, no. 99, pp. 1–10, 2018.
[11] C. E. Page, “Maximizing maritime safety and environmental protection with AIS: (Automatic identification
system),” in OCEANS 2017 - Anchorage, 2017, pp. 1–4.
[12] M.-C. Tsou, “Discovering knowledge from AIS database for application in VTS,” The Journal of Navigation,
vol. 63, no. 3, pp. 449–469, 2010.
[13] J. Pike, “Class A And Class B Automatic Identification System (AIS): www.oceantimemarine.com/class-a-
and-class-b-automatic-identification-system-ais/.” 2018.
[14] G. Mappatao, J. Lorenzo, N. Mallari, J. C. Espiritu, L. Materum, and A. Abad, “Design and Implementation
of ITU-R M.1371 Data Transfer Protocol and NMEA AIS Sentence Using Open Source Software for
Universal Shipborne Automatic Identification System Application,” (accepted for publication).
115
[15] ITU-R, “ITU-R M. 1371-5-Technical characteristics for an automatic identification system using time-
division multiple access in the VHF maritime mobile band.” 2014.
[16] K. Hata, K. Hasegawa, K. Niwa, and H. Fukuda, “AIS simulator and ITS applications,” in 48th International
Symposium on Multimedia Signal Processing and Communications, 2006, pp. 223–226.
[17] R. Essaadali, C. Jebali, K. Grati, and A. Kouki, “AIS data exchange protocol study and embedded software
development for maritime navigation,” in IEEE 28th Canadian Conference on Electrical and Computer
Engineering, 2015, pp. 1594–1599.
[18] C.-C. Lee and S.-H. Park, “Standalone Maritime Aids-To-Navigation AIS Mobile Station,” Journal of
Information and Communication Convergence Engineering, vol. 7, no. 3, pp. 297–303, 2009.
[19] U.S. Coast Guard Navigation Center, “Schematic Diagram of Class A Ship-borne AIS Station:
https://navcen.uscg.gov/pdf/AIS/IALA_AIS_ClassA_Schematic.pdf.” .
116
E-Government System Design for Indonesia using SOA
Natanael Alamas, Kristianto, Marcelina Anggraeni, Reagen Sayoga, and
Sfenrianto
Information Systems Management Department, BINUS Graduate Program - Master of Information Systems
Management, Bina Nusantara University, Jakarta, Indonesia 11480
Abstract. Non-transparent process and non-standardized procedure of public services become major issues faced by Indonesian
government which cause people to become discouraged and lose their trust of public service organizations. E-Government
system become preferred solution by Indonesian government to answer those issues, but currently they only reach regional level
with limited features. This paper proposes conceptual design of national E-Government system by using Service-oriented
Architecture (SOA). The ability of SOA to connect services between systems can help to integrate related public services
organizations to provide single and centralized E-Government system.
Keywords-E-Government, G2C, Service-oriented Architecture
INTRODUCTION
As a developing country, Indonesia has a lot of objectives to improve the performance of its government. One of
the critical objectives is to improve public service. People give poor rating for public service in Indonesia. One of
the reason is they feel the process is not transparent. Sometimes they do not know what they need to do, what is the
current progress, or who to ask for more information. Information regarding public service is available from internet,
but it could be different when people actually visit the organization. This situation makes people discourage and lose
trust when dealing with public service. Another problem is each region in Indonesia could have different procedure
of public service. Regional government has regional autonomy which allows them to make rule that only apply in
their region. This may cause some confusion when people move from one region to another since they do not know
about the procedure on new region.
Indonesian government has already realized these problems and starts to improve public service. Some regional
government start to launch mobile application or website which allows people to use their mobile phone or computer
to search for public service information, or online form for public service registration, such as for creating birth
certificate, family card, ID card, etc. This approach is already quite helpful but still does not fully answer all the
problems.
This study proposes Service Oriented Architecture (SOA) design to create E-Government system so it can help
Indonesian people, not only in some regions, but the whole nation. The system can help people easily request for
public service, get notified about the service’s progress, and be ensured about service result. The design will not
only help Indonesian government to have centralized system but also give Indonesian government the national
standard procedure of providing public services. Furthermore, Indonesian government will also have integration
which allows data exchange between public services organizations.
117
Service-oriented Architecture (SOA) is software architecture consists of connected services to deliver business
functionality. Service is an independent component that performs certain function. Service can communicate with
other services or applications through a network [1]. SOA is widely applicable in many domains. One example is
public health management system (PHMS). PHMS provide eight business functions based on people’s need. Then
PHMS links to hospital’s system to perform their business function [2]. In banking industry, SOA can be used to
create web based home banking system. They try to provide bank’s functionalities, such as opening new account,
transfer money, and perform payment to multiple channels. People can access this system from anywhere as long as
they connected to the internet [3]. There is also tourism e-commerce system. This system use SOA to integrate with
travel enterprises. People can see information about tourism, and perform order based on their travel plan [4]. E-
Commerce ecosystem started from monolithic systems where all software components that provide services are
bundled into single application. But using this approach makes it difficult to change or add services. In order to have
flexibility, E-Commerce are moving towards SOA [5]. All of examples above show us that SOA is useful to create
integrated, flexible system and provide valuable services to the people.
E-Government is defined as the use of combination between Information Technology, organizational changes,
and new skills in public administration to improve the quality of public services [6]. Study [7] defines E-
Government as delivery of government services automatically using computer systems. Study [8] mentions some
advantages of E-Government like to increase efficiency, time saving, online access of services, and improve
transparency. E-Government implementation might face challenges. Some employees can resist to the changes, this
might have negative impact to the organization [9].
RELATED WORKS
Many countries around the world have started to implement E-Government. France has Legifrance system which
already been used for more than 15 years. It is a government service portal, where the French can have various
services, contact with certain government institution, receive information, and other things. [10]. Rwanda has
IREMBO project, an initiative to digitalize all public services in single platform. They create this initiative in order
to eliminate manual processes, delays and bottleneck in service delivery [11]. Study [12] proposes framework of
SOA for E-Government in Jordan. SOA can improve the quality of E-Government because it allows interoperability
between different government administrations and data sharing in controlled way. Thailand has progresses through
many E-Government frameworks since 1990 and currently they have version ‘4.0’ plan. Their target is to improve
performance of public sector and citizen wellbeing. During their journey, they face some challenges such as lack of
ICT experienced Thai public servants, lack of system integration between public agencies, and also ambiguity in the
regulation of E-Government [13].
Japan has big target to transform from bureaucracy into an open E-Government. They want to take down vertical
organizational barriers and provide convenient E-Government services. Japan has been implementing “My Number”
system in 2016. “My Number” system is similar like social security number in United States, which act as a key to
access administrative functions. Using this system, people can access their own information and check how
government agencies use their information. Japan also has “My Portal”, one stop service system that allow people to
submit for services [14]. Malaysia established mega project named Multimedia Super Corridor in 1996. One of its
flagships is Electronic Government. E-Government’s objective is to make government sector shifting toward use of
electronic and multimedia networks. The vision of E-Government initiative is to improve service delivery from
government to people of Malaysia [15].
Some regional governments in Indonesia already start to implement E-Government in their region. They create
mobile application or website to provide online services for people in their region. This study compiles Table 1 to
show example of existing applications and its current features made by regional governments. Only people in
respective region can use the application to request for services. Available services in those applications are varied.
It depends on the readiness of application, capability and procedures in regional government.
118
TABLE 1. Examples of existing service portal and application in Indonesia
RESEARCH METHODOLOGY
This study use systematic literature review as an approach. It is performed by identifying, evaluating, and
combining information from several other studies which relevant with this study’s research questions. This study
starts by reviewing the benefit of SOA implementation and examples of SOA implementation in several industries.
Afterwards, this study looks into the concept of E-Government, examples of E-Government implementation in
several countries, benefits expected from implementing SOA, and challenges during E-Government implementation.
After reviewing all of that information, we conclude to propose the creation of E-Government system based on SOA
to help current problem faced by Indonesian government. Technical and implementation design is out of scope from
this study.
RESULT AND DISCUSSIONS
Ecosystem
Learning from existing applications made by regional government, we want to propose a design of single E-
Government system that can be used by all Indonesian people from all regions in Indonesia. We start the design by
determining possible services that will be available in the application. Available services must be services that
frequently requested by Indonesian people. By providing frequently used services from the beginning, we can attract
more people to use the application and shows how useful this application is. Based on the selected available
services, we can determine related organizations that must be connected to this application.
1. Disdukcapil (Population and Civil Registration Agency): This organization is important because it controls the
civil and population information. Civil information is very important for E-Government system’s ecosystem,
especially ID number, because it will be used as primary key that will be linked with all of other services.
2. Hospital: This organization is selected because it provides the information of the beginning and the end of
people’s life. By having this information, we can provide services to people as soon as they were born.
3. Police: This organization is selected because the application can help government to monitor and control
people’s obedience to law and regulations. One of the examples is this application will prevent underage person
from having driver’s license and make them comply with rules.
4. Ditjen Pajak (Directorate General of Taxes): This organization is selected because all of productive people in
Indonesia are regulated to report their tax payment regularly. By providing tax related services, we will not only
help a lot of people to obey the regulation, but also government to monitor and control tax information.
5. Immigration: This organization is selected because the application can help government to monitor and control
people that come to and go from Indonesia. It also help people that frequently going abroad.
Features Satu Layanan Dukcapil Dalam
Genggaman
Tangerang Live Smart Dukcapil
Birth Certificate Only information Online Online Online
Death Certificate Only information Online N/A Online
ID Card Only information Online Online Online
Children ID Card N/A Online N/A N/A
Family Card Only information Online N/A N/A
Data integration N/A N/A N/A N/A
Region Only information Only Solo Only Tangerang Only Serang
119
All of the organizations within ecosystem are integrated. It means that data or information from one organization
which are required by another organization to provide their services can be immediately retrieved without people
need to manually search or input.
Business Services
After ecosystem is identified, Next step is to identify the business services. Business services is a feature which
available in the application. For E-Government system, business services should be based on services provided by
all organizations inside ecosystem. Business services’ name must simple and self-explanatory, so user can
immediately know the function of the business services just by looking at its name. Business services also need to be
independent. If there is problem on one business service then other business services should not be impacted. They
must be still functioning normally and can be used by user. Based on five organizations inside ecosystem, we
identify following business services:
1. Create birth certificate
2. Create death certificate
3. Create family card
4. Create ID card
5. Update ID card
6. Create Children ID card
7. Update Children ID card
8. Create birth notification
letter
9. Create death notification
letter
10. Create driver’s license
11. Extend driver’s license
12. Create vehicle registration
certificate
13. Extend vehicle registration
certificate
14. Create passport
15. Extend passport
16. Create tax identification
number
17. Update tax identification
number
18. Submit tax report
19. Submit tax payment
This list can be further expanded in the future as more organizations are added into ecosystem.
List of Services
The core of SOA is in the availability of services. Services can be derived from business services process flow.
Process flow is list of sequential processes that is performed from start until end to complete a business service. As
an example, Figure 1 displays process flow of create birth certificate business service.
FIGURE 1. Create Birth Certificate process flow
120
By identifying all business services process, we can identify common processes that have similar function used
by them. These similar functions can be made into reusable services. Service can be used again when there is new
business service in the future. Below is the list of reusable services for E-Government system:
1. Verify Registration: Service to verifying the creation of new user.
2. Verify Login: Service to verifying user login to application.
3. Save User Profile: Service to create and update user profile.
4. Get User Profile: Service to retrieve user profile data and display it to user interface.
5. Get User Transaction: Service to retrieve user’s transaction information and display it to user interface.
6. Entry Data: Service to store data inputted by user. Input data can be varied based on business service.
7. Upload Document: Service to store document uploaded by user. Upload document can be varied based on
business service.
8. Entry Appointment: Service to scheduling user’s visit to selected organization.
9. Validate Data: Service to validate user’s input data. Validation rule can be varied based on business service.
10. Get External Data: Service to retrieve data from external organization. Source of data can be varied based on
business service.
11. Send Email to User: Service to send email to user. Content and attachment can be varied based on business
service.
12. Send Data: Service to send required data to external organization to fulfill business service.
13. Send Document: Service to send required document to external organization to fulfill business service.
14. Get Verification Result: Service to request data and document verification result from external organization.
15. Send Appointment: Service to send user’s visit schedule to external organization.
16. Get Appointment Result: Service to request confirmation about user’s visit schedule.
17. Get Transaction Result: Service to request final result of business service from external organization.
All of these services will be listed and maintained in service registry.
SOA Layer
Using ecosystem and list of services, we can map SOA implementation into several layers. Each layer served
different purposes and they communicate each other. This study creates Figure 2 to visualize SOA layer for E-
Government system.
1. Presentation layer: Presentation layer is the User Interface. This is a front-end application which user can
interact with. There are two presentation layers for E-Government system: mobile application and website.
2. Business service layer: Business services are the features inside application. User can view and select the
feature to fulfill certain function. There are several business services layers for E-Government system which
mentioned in section 4.2.
3. Service layer: Reusable services are maintained in this layer. Services are connected with other services
following specific flow to fulfill business service purposes. List of services for E-Government system are
mentioned in section 4.3.
4. Resource layer: Resource layer are external resource needed to fulfill business service. It is the application or
system inside external organization. Technically, interaction with external resource is usually through API to
keep internal application from being exposed. Resource layer for E-Government system are all organizations
inside ecosystem.
5. Data layer: Data layer store and maintain the data that are used for business service. It is usually accessed by
service layer and can’t be accessed directly without highest authorization. Some examples of data for E-
Government system are user profile, user transactions, and all related data from external organizations.
121
FIGURE 2. E-Government system SOA layer
CONCLUSION
To have single and centralized E-Government is one of many objectives of Indonesian government since long
time ago. E-Government system can make all regions in Indonesia to have a standardized procedure for service
request so people won’t be confused anymore if they are moving from one region to another. People also won’t be
discouraged anymore when dealing with public service because all services are fully transparent now. People can
easily access and get updated about their service request. Indonesian government also can regulate and protect
people easily because all of their data is integrated between organizations. This study’s design also allow for further
expansion in the future such as creating technical and implementation design of national E-Government system, or
adding more organizations into the ecosystem to provide more services based on urgent needs.
REFERENCES
[1] Hatitye Chindove, Lisa F. Seymour, and Francois I. van der Merwe, F, “Service-oriented Architecture:
Describing Benefits from an Organizational and Enterprise Architecture Perspective”, 2017.
[2] Zhang Yimin, Song Ping, Sun Qi, “Design of Public Health Management System Based on SOA
Architecture”, 2017.
[3] A. Appandairaj, Dr. S.Murugappan, “Service Oriented Architecture Design for Web Based Home Banking
Systems with Cloud Based Service”, 2013.
[4] Ding Wang, “Research on Design and Realization of SOA-based Tourism E-Commerce System”, 2017.
[5] Fabian Aulkemeier, Milan Schramm., Maria-Eugenia Iacob, Jos van Hillegersberg, “A Service-Oriented E-
Commerce Reference Architecture”, 2013.
122
[6] Puneet Kumar, Dharminder Kumar, Narendra Kumar, “E-Governance in India: Definitions, Challenges and
Solutions”, 2014.
[7] Zuhoor Al-Khanjari, Nasser Al-Hosni, Naoufel Kraiem, “Developing a Service Oriented E-Government
Architecture towards Achieving E-Government Interoperability”, 2014.
[8] Sethunya Rosie Joseph, “Advantages and Disadvantages of E-Government Implementation: Literature
Review”, 2015.
[9] Ali Tarhini, Ra’ed (Moh’d Taisir) Masa’deh, Ali Abdallah Alalwan, Nabeel Al-Qirim, “Factors Affecting the
Adoption of E-Government in Kuwait: A Qualitative Study”, 2017.
[10] Oksana A. Mukhoryanova, Irina V. Novikova, Slavko B. Rudich, Elena V. Bogushevich, “E-Government in
the Western European Countries, Asia, and in the USA”, 2016.
[11] Jean D. Twizeyimana, Hannu Larsson, Ake Gronlund, “E-Government in Rwanda: Implementation,
Challenges and Reflections”, 2018.
[12] Zakaria I. Saleh, Rand A. Obeidat, Yaser Khamayseh, “A Framework for an E-government Based on Service
Oriented Architecture for Jordan”, 2013.
[13] Danuvas Sagarik, Pananda Chansukree, Wonhyuk Cho, Evan Berman, “E-Government 4.0 in Thailand: The
role of central agencies”, 2018.
[14] Noriko Asato, “Development of Japan’s e-Government: My Government as a Step towards a Ubiquitous G2C
Networked Society”, 2017.
[15] Muhd Rosydi Muhammad “Managing the Implementation of E-Government in Malaysia: A Case of E-
Syariah”, 2013.
123
New Multi Objective Approach for Optimal Network
Reconfiguration in Electrical Distribution Systems Using
Modified Ant Colony Algorithm
A. OLOULADE1, a), A. MOUKENGUE IMANO2, b), A. VIANOU1, c),
H. TAMADAHO1, d), R. BADAROU1, e)
1Electrotechnic, Telecommunications and Informatics Laboratory (LETIA),
University of Abomey-Calavi, Benin. 2Electronic, Electrotechnic, Automatic, Telecommunications Laboratory (LEEAT),
University of Douala, Cameroon.
a)[email protected] b)[email protected] c)[email protected] d)[email protected]
Abstract. The losses in networks of Beninese Electrical Energy Company (SBEE) are very high and therefore constitute a
concern for the operators. This work consisted in finding an optimal topology of a 41 nodes real network of SBEE by Modified
Ant Colony Algorithms (MACA) in order to reduce the losses and ensure a continuous power supply to the customers in case
of occurrence disturbances on any branch of this network. With technological breakthrough of Automation and Supervision
Systems (SCADA), the operation of distribution networks can be ensured remotely in real time with the aim of minimizing
losses, eliminating equipment overload and improving reliability. The criteria of technical performance improvement
formulated under operating constraints are solved by Modified Ant Colony Algorithm (MACA) which is association of ant
system and fuzzy logic on the Matlab platform. The best results obtained show the effectiveness and efficiency of this method.
The SBEE's HVA networks can then be reconfigured automatically to significantly improve their continuity of supply and
reliability. The improved results obtained after tests on a standard 33-nodes and a real 41 nodes networks show the robustness
and accuracy of this MACA algorithm.
Keywords—reconfiguration; ant colony algorithm; power loss reduction; radial distribution network
INTRODUCTION
Distribution networks is very important in energy flow chain and take an important part of the electrical
industry's facilities, to the point where 30% to 40% of dedicated investments in the energy sector are devoted to
it but do not receive necessary technological impact [1]. These distribution systems, although constructed in a
loop structure, are operated in a radial configuration in order to ensure the coordination, efficiency and
effectiveness of the protection systems. Distribution engineers in normal operating conditions or sometimes in
the disturbed system periodically reconfigure their network outputs by opening or closing switches in order to
increase the reliability of the network and reduce losses or to isolate the faulty part network and ensure the
continuous power supply to customers. The aim is made of reducing operating costs by optimizing the topology
of the distribution networks under operating constraints (respect of tolerated voltage drops, respect of the balance
of charges on departures and loss rates). There are several network optimization techniques, among which
reconfiguration remains one of the most used practices [2]. The reconfiguration consists in sequentially defining
the topological status of the breaking devices in the distribution networks in order to obtain an optimal
configuration that minimizes losses and improves the node voltage profiles [3]. Network reconfiguration aims
improvement of the voltage profiles, the stability, the reliability of networks and contributes to the reduction of
technical losses by optimizing its driving.
The distribution networks of SBEE are not very automated. Indeed, the switches installed there are, for the
most part, under manual control, the few existing voltage interrupt switches are not functional, associated with
124
the mechanical switches which are not motorized and are therefore not remote controllable. These shortcomings
in SBEE's distribution networks are the operating difficulties. Thus, excessive losses are generated, interruption
times which are not in the optimal proportions and which induce important undistributed energies (END) [4].
The reconfiguration of distribution networks for loss minimization has been proposed for the first time by
Merlin and Back in [5] where a heuristic optimization approach has been used to determine an optimal
configuration minimizing losses in a tree network with a specific charge. Since then, several other methods have
been developed and oriented towards programming based on the C-language, genetic algorithms [6], evolutionary
programming [7] which proposed a reconfiguration of distribution networks by the heuristic permutation method
compared to the TABOU method. They found that the method of branch switching makes satisfactory results
over the other. In recent years, several researchers have used genetic algorithms (AGs) and ant colony to solve
optimization problems and whose matched results are relevant. In this paper, Modified Ant Colony Algorithms
(MACA) are developed and tested on a standard 33 nodes network and then on 41 real nodes of Benin Electrical
Energy Company (SBEE).
Issues in operation of SBEE distribution networks
The beninese electrical system is mainly supplied by the Benin Electric Community (CEB) imports, direct
energy purchases from independent producers (AGGREKO and MRI) and PARAS Energy in Nigeria.
This electrical system is made of 2678 HVA/LVA stations, 8 dispatch stations and 121 HVA departures. The
conduct of the SBEE networks is very little automated. In fact, overhead switches such as mechanical switches
are not equipped with adequate technology that can allow their remote operation. The search for defects on the
lines is done step by step, thus generating significant undistributed energies as illustrated in Figure 1 which shows
the proportion of undistributed energies by region of the SBEE due to bad facilities and whose cumulative energy
in 2017 is estimated at 4 286 846 kWh [4]. In fact, the operators of the SBEE networks are confronted with some
difficulties in manual operating the switches installed on the networks and are sometimes obliged to resort by
sectioning of electrical branch in default. This situation unnecessarily lengthens the duration of interventions and
consequently increases the undistributed energies. It is also observed that the level of losses in the system is
globally high as illustrated in Figure 2 and until then, constitutes a source of constant concern for the operator
and supervisors who seek to master it and rationalize the electrical service and thus, limit the resulting expenses.
This figure represents losses evolution on the SBEE networks since 2009 until 2017. The losses in a distribution
system is made of two components, technical and non-technical or commercial losses. Technical losses are due
in SBEE's distribution networks to the inadequacy of branch sections, excessively long lines exceeding the limits
tolerated by standards, overloads on some HV / LV distribution transformers and the inefficiency of some voltage
control devices. Most of the commercial losses are due to fraud on the electricity meters, errors in the reports of
the measuring reducers, the existence of illegal connections, the drift often observed in the computer systems and
the direct connection of public lighting lamps to LV networks without counting box.
Technical losses are inherent in the operation of networks and therefore can only be reduced while
commercial losses can be avoided. In the SBEE distribution networks, the non-technical losses are greater than
the other component, hence the need for operators to develop effective strategies for eradicating these losses, such
as, the tracking of these from the source stations to the connections via HVA / HVA distribution stations and MV
/ LV distribution stations and the constitution of very strong anti-fraud systems.
Figure 1 shows the evolution of loss rates on the SBEE distribution network over the years.
125
Fig. 1. Profile of loss rates on the SBEE networks
The energy losses of the network in the different regions are significant as shown in Figure 2.
Fig. 2. Rate of undistributed energies by department for the year 2017 [4]
With :
DRL1 : Regional Direction of Littoral 1
DRL2 : Regional Direction of Littoral 2
DRA : Regional Direction of Atlantique
DROP : Regional Direction of Ouémé – Plateau
DRZC : Regional Direction of Zou – Collines
DRMC : Regional Direction of Mono - Couffo
DRBA : Regional Direction of Borgou – Alibori
DRATAD : Regional Direction of Atacora – Donga
The minimization of losses is therefore one of the improvement indicators of energy supply systems and operator
performance. In fact, a 1% energy loss in a distribution network increases the company's operating cost by 2% in
order to cover the expenses incurred by these losses [8].
DRL125%
DRL213%
DRA27%
DROP8%
DRMC4%
DRZC11%
DRBA9%
DRATAD3%
END 2017
126
Problem formulation of the distribution network topology problem
The distribution networks are built in loops with opening points but operated in a radial topology to allow
the effective functioning of the protection systems. Therefore, they are made of adjacent meshes but a few times
independent. Indeed, when the meshes are independent, the number of possibilities of combinations of the opening
switches is limited and with a simple exploration of all the possible diagrams, the operators manage to find easily
optimal tie-switches which can reduce losses and improve protection plans. But the cases of the most real
operation schemes are the adjacency of several meshes, and from which flow very complex network operation
schemes. In this case, the number of possibilities is exponential, the search for optimal topologies becomes tedious
and managers often use optimizers. As a result, optimizing a distribution network mathematically returns to
finding the optimal objectives under constraints.
Optimizing the topology of a network then amounts to minimizing the objective function F (Y, Z, U) under the
constraints of equalities and inequalities expressed by G (Y, Z, U) ≤0.
Y : Set of currents in the lines
Z : Set of node voltages
U : Set of tie-switches
Objective functions and constraints
a) Power losses reduction and voltage deviation
The reduction of power losses and voltage deviation on the busbars represents an economic and technical contests
for the electricity companies. Thus, reducing the technical losses amounts to:
min f1 = ∑ Rl (Pl
2 + Ql2
Vi2 )
NL
l=1
(1)
min f2 = ∑ (Vi − Vi
sp
Vimax − Vi
min)
2N
2
(2)
with
Rl : resistance of branch l,
Pl : active power losses of branch l,
Ql : reactive power losses of branch l,
Vi : voltage at node i,
NL : Set of branches,
N : Set of nodes,
𝑉𝑖𝑚𝑎𝑥 𝑒𝑡 𝑉𝑖
𝑚𝑖𝑛 limits of node voltages,
𝑉𝑖𝑠𝑝
, specific voltage at node i.
b) Constraints
• 0,95 p. u < Vi < 1.05 p. u
• Pl2 + Ql
2 ≤ Slmax2
• For any node of the network, there must be a single path through which any other node of network can be
reached where it represents tree constraint
Radial distribution network load flow
Distribution networks are characterized by high R/X ratios and the traditional methods of power flow
calculating widely known such as Newton Raphson, Gauss Seidel and Fast decoupled methods are not suitable
for effectively solving the problems power flow problems in radial distribution networks.
127
The backward / forward sweep is one of load flow computation method which is developed and converges very
well for radial distribution networks. It is based on the laws of ohm and Kirchhoff and can be considered from
the family of iterative methods of power flow. This method is implemented on the Matlab platform and integrated
into the optimal network reconfiguration algorithm [9].
Radial network reconfiguration using ant colony and fuzzy logic
Ants are social insects whose physical or behavioral characteristics have long fascinated researchers. Ant
colony algorithms typically use the behavior of real ants [8] to solve combinatorial optimization problems. It can
be reduced in search of the shortest path through graphs based on the behavior of the ants that in their displacement
mark the way with glands contained in their abdomen called pheromones [10]. The amount of pheromone depends
on the length of the path and the amount of food found. The pheromone evaporates over time if others are
deposited there. The path of pheromones leading to food sources will be more frequented by ants. The
implementation process of this algorithm consists of the initialization steps, the implementation of the local update
rule, the evaluation of the objective function and the application of the global update.
Implementation Process of Modified Ant Colony Algorithm (MACA)
MACA is an adaptation of ant algorithm and fuzzy logic in search of optimal radial topology from a basic
configuration. This algorithm is based on the following principles :
- tie-switches are considered as path branches to avoid ;
- a new radial configuration of the network is considered as a set of unique paths between all cities (any
node) for a reference city (source node).
The program developed for the reconfiguration is made of the following essential steps :
- generation of tie-switches (branches as path to avoid through two given cities) and therefore a set of
paths composed of electrical network lines
- radiality criterion assessment of the new topology
- power flow analysis and technical performance evaluation of new power grid topology
The new approach of optimal reconfiguration seeking elaborated in this MACA paper is described by a main
program "MACA"
Table 1 : Modified Ant Colony Algorithm (MACA)
Algorithm
Step 1 : Initialize first iteration 𝑡 = 1,
Initialize pheromones : 𝜏𝑖𝑗 ← 𝜏0∀𝑖, 𝑗 = 1, … . , 𝑛
Step 2 : For each ant 𝑘 = 1 à 𝑚
Build a cycle 𝑇𝑘(𝑡) of tie-switches
Update new tie-switches
Step 3 : Search a unique path from any node to source node
and avoid tie-switches
Step 4 : Reorganize loads according to the end-of-branch
nodes and saving of the new configuration
Step 5 : For each ant 𝑘 = 1 à 𝑚
Evaluate the radiality criterion of the new configuration
Penalize radiality criterion by a high value if current
configuration is not radial
Step 6 : Compute glabal cost function 𝐿𝑘(𝑡) of 𝑇𝑘(𝑡)
Update the best global solution
Step 7 : For each arc (𝑖, 𝑗)
Update 𝜏𝑖𝑗 while confining them in [𝜏𝑚𝑖𝑛 , 𝜏𝑚𝑎𝑥]
𝑡 = 𝑡 + 1
if 𝑡 < 𝑁𝐶𝑚𝑎𝑥
Goto step 2
else
STOP
Return the best solution founded
128
Simulation parameters
Table 2 presents the main parameters tested with MACA for simulation.
Table 2 : Main parameters
Parameters Values
Size of the colony Number of
network branches
Number of iterations 100
Α 1
Β 2
Q 100
pMin 0,0005
pMax 0,7
Ρ 0,05
𝝀𝟏 0 ,7
𝝀𝟐 0,3
Γ
100 000 (high
value)
SIMULATION AND RESULTS
The Modified Ant Colony Algorithm (MACA) is tested on standard IEEE 33-bus system and is applied on SBEE
41-bus real system.
Application of MACA on IEEE 33-bus network
The 33-bus system is made of 32 closed switches and 5 tie-witches. Active and reactive loads are respectively
evaluated at 3802.19kW and 2694.66 kVar with a voltage level of 12.66 kV.
The results of MACA applied to this network are presented in table 3 :
Table 3 : Comparison of configuration for 33-bus network
Base
config.
[11]
[12] [13] Proposed method
(MACA)
Tie-switches 33 – 35 – 34 – 36 –
37
7 – 10 – 14 – 36 –
37
33 – 14 – 8 – 32 – 38
13 – 10 – 5 – 27 –
18
Active losses
(kW) 214,72 135,47 139,5 121,56
𝑽𝒎𝒊𝒏
(pu) 0,9128 0,9342 0,9437 0,9845
Losses rate (%) -- 36,9 35,03 43,38
It is observed that according to references [12] and [13], which found respective losses of 135.47 kW and 139.5
kW as a result of the simulation, MACA decreased these losses to 121.56 kW, a performance of 43.38%. These
highly improving results prove the performance and efficiency of MACA, and also testify that it can improve
significantly the continuity of service and reliability of distribution networks.
129
Application of MACA on SBEE HVA 41-bus system of Togba
The togba HVA network, whose basic topology is shown in figure 3, is made of a source station and 41 nodes
with 4 tie-switches.
Fig 3. Topology of HVA network of Togba before reconfiguration
Figure 4 shows the voltage profiles on the busbars before and after reconfiguration and table 4 shows losses and
voltage levels after reconfiguration.
Fig. 4. Voltage profiles before and after reconfiguration.
From these results, it is observed that the losses are significantly improved and reduced by 63.15% and the lowest
voltage is found at node 23 with a value of 0.9724 pu. No node voltage has been exceeded the ranges
recommended by the standards for HVA voltages on busbars. The reconfiguration of networks contributes to
qualitatively improving the performance of power systems without the need for important investments. Indeed, it
is urgent to ensure the reliability of opening points of this network to facilitate the optimization of its conduct.
130
Table 4 : Reconfiguration results of real 41-bus network of Togba
Before reconfiguration After reconfiguration
(MACA)
Tie-switches 42 – 43 – 44 – 45 9 – 24 – 31 – 36
Active losses
(kW) 403,56 148,97
Lowest voltage (bus) 0,8498 pu (37) 0,9724 pu (23)
Rate of improvement of
network state settings
Losses 63,15 %
Lowest voltages 14,42 %
MACA performance
computation duration 18,331 s
Number of iterations 71
Figure 5 shows the optimal topology of the SBEE 41-bus distribution network from MACA.
Fig. 5. Topology of 41-bus system after reconfiguration
131
CONCLUSION
This paper is about of distribution networks reconfiguration to improve the quality of power supply and
decrease losses under operating constraints. The proposed MACA method for solving this complex combinatorial
problem was tested on the standard 33-bus network and then on the actual SBEE 41-bus system. The technical
performances, the computation duration and the speed of convergence obtained, clearly show the efficiency of
this method, designed to solve problems of radial topology optimization. According to the results obtained, it
appears that the reconfiguration of a distribution network is one of the efficient and effective resource to balance
the loads on the HVA stations and decrease the operating costs. The optimization approach based on metaheuristic
methods is very efficient and effective in increase of reliability of the electricity supply systems that will
contribute to a best quality of life in a sustainable ecological context.
REFERENCES
[1] H. K. Hazaa, H. H. Abdullah, M. H. Yasen, et S. S. Mustafa, « Evaluation of electric energy losses in kirkuk distribution electric system area », Iraqi J. Electr. Electron. Eng., vol. 7, no 2, p. 144–150, 2011.
[2] B. Gotzig, « Recherche du schéma optimal d’exploitation d’un réseau de distribution électrique », PhD Thesis, Institut National Polytechnique de Grenoble-INPG, 1997.
[3] A. Jain, A. R. Gupta, et A. Kumar, « An efficient method for D-STATCOM placement in radial distribution system », in 2014 IEEE 6th India International Conference on Power Electronics (IICPE), 2014, p. 1‑6.
[4] DD/SBEE, « Rapport d’analyse des incidents sur le reseau HTA (Période : Juillet à septembre 2017) », oct. 2017.
[5] A. Merlin, « Search for a minimal-loss operating spanning tree configuration for an urban power distribution system », Proc 5th PSCC 1975, vol. 1, p. 1–18, 1975.
[6] J. Li, « Reconfiguration of power networks based on graph-theoretic algorithms », 2010.
[7] V. Tudor, « Optimal loss reduction of distribution networks using a refined genetic algorithm », Univ. Politeh. Buchar. Sci. Bull. Ser. C Electr. Eng., vol. 72, no 3, p. 29–38, 2010.
[8] D. Shirmohammadi et H. W. Hong, « Reconfiguration of electric distribution networks for resistive line losses reduction », IEEE Trans. Power Deliv., vol. 4, no 2, p. 1492–1498, 1989.
[9] A. Oloulade, A. Moukengue Imano, A. Vianou, et H. Tamadaho, « Optimization of the number, size and placement of D-STATCOM in radial distribution network using Ant Colony Algorithm », 2018.
[10] H. B. Tolabi, M. H. Ali, et M. Rizwan, « Simultaneous reconfiguration, optimal placement of DSTATCOM, and photovoltaic array in a distribution system based on fuzzy-ACO approach », IEEE Trans. Sustain. Energy, vol. 6, no 1, p. 210–218, 2015.
[11] A. Oloulade, A. Moukengue Imano ,A. Vianou , et H. Tamadaho, « Optimisation multi-critère du placement d’un D-STATCOM dans un réseau de Distribution par les Colonies de Fourmis », 2018.
[12] R. K. Jacob et V. Malathi, « Optimal Reconfiguration of Power Distribution Systems », 2015.
[13] R. Srinivasa Rao, Optimization of Distribution Network Configuration for Loss Reduction Using Artificial Bee Colony Algorithm , International Journal of Electrical and Computer Engineering, Vol:2, No:9, 2008 .