7
Published in IET Communications Received on 5th September 2013 Revised on 7th December 2013 Accepted on 29th December 2013 doi: 10.1049/iet-com.2013.0793 ISSN 1751-8628 Advanced metering infrastructure performance using European low-voltage power line communication networks Javier Matanza 1 , Sadot Alexandres 1 , Carlos Rodríguez-Morcillo 2 1 Automatic and Electrical Department, ICAI School of Engineering, Madrid, Spain 2 Institute for Research in Technology (IIT), Madrid, Spain E-mail: [email protected] Abstract: Power line communication has recently attracted the attention of energy companies as a useful and natural technology for building the advanced metering infrastructure. In this context, device language message specication/companion specication for energy metering (DLMS/COSEM) is an increasingly popular standardised application protocol for communication between utilities and their customers. This study analyses the communication performance that can be expected when using the power line communication technology, powerline intelligent metering evolution (PRIME), to send DLMS/COSEM messages. Physical phenomena such as background and impulsive noise sources, channel attenuation and multi-path effect are taken into account during the rst step in the evaluation of this technologys communication performance in the physical layer. This metric is then used in upper layers to compute the packet error rate. An analysis is carried out at the application layer in terms of expected latency in different communication environments. Several simulations are performed in a European low-voltage topology to compute the number of metres that can be read within 15 min. These simulations were carried out using MATLAB and OMNeT++ software. 1 Introduction Recent years have seen an increasing interest in narrow-band power line communication (PLC) on the part of utility companies. Although this technology is not new, there are several characteristics that make it suitable for smart grid applications, such as the automatic meter-reading [1, 2]. As a result, several transceiver designs have appeared in the past few years as designers have attempted to offer solutions for communication via power lines. Early models (X-10 or KNX) made use of a single-carrier technology to modulate data. These solutions were targeted for in-home scenarios using domotic applications. One of the rst solutions to implement multi-carrier modulation while working in the low-voltage (LV) network is powerline intelligent metering evolution (PRIME), which was introduced in 2007. Since then, a number of designs have been released such as G3-PLC (from the French utility EDF); G.hnem (recommended by the International Telecommunications Union (ITU-T) and IEEE1901.2 (from the IEEE). There are several works in the literature that provide an analysis of the performance of these solutions [35]. These studies include several power line scenarios. However, although accurate, they only focus on the physical layer of the communication, and their main output consists of bit error rate (BER) curves. Nevertheless, these kinds of studies have been found to be very helpful from the industrys perspective, as mentioned in [6]. This work improves the previous performance analyses by including an implementation of the network level. This is achieved via a simulation framework in which both networking and physical effects are taken into account. Owing to the characteristics of the PLC channel, aspects such as the channel access play an important role in the behaviour of the network, as described in this paper. To the best of the authorsknowledge, this approach has barely been addressed in the literature so far. In [7], an interesting analysis of the available throughput in multi-hop power lines is conducted. Nevertheless, no channel noise is added to the transmission, so no errors are modelled in the communication. Additionally, studies like that presented in [8] attempt to analyse the effective data rate achieved in a PLC network. However, the BER is xed for all nodes in the simulation, independent of their position, which is not a realistic situation. In [9], a method is proposed to abstract the physical (PHY) layer from the simulations by means of packet error rate against signal-to-noise ratio (SNR) curves. However, when computing those curves, a xed packet length is set, leading to a not realistic situation either. A similar study to the one presented in this paper is given by Zaballos et al. [1]. They analyse the time required to read 100 m over a 64 kbps channel. The packet loss probabilities are taken into account in the simulations. However, no noise sources or channel attenuation are considered in the study. www.ietdl.org IET Commun., 2014, Vol. 8, Iss. 7, pp. 10411047 doi: 10.1049/iet-com.2013.0793 1041 & The Institution of Engineering and Technology 2014

Advanced metering infrastructure performance using European low-voltage power line communication networks

  • Upload
    carlos

  • View
    213

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Advanced metering infrastructure performance using European low-voltage power line communication networks

www.ietdl.org

IE

d

Published in IET CommunicationsReceived on 5th September 2013Revised on 7th December 2013Accepted on 29th December 2013doi: 10.1049/iet-com.2013.0793

T Commun., 2014, Vol. 8, Iss. 7, pp. 1041–1047oi: 10.1049/iet-com.2013.0793

ISSN 1751-8628

Advanced metering infrastructure performanceusing European low-voltage power linecommunication networksJavier Matanza1, Sadot Alexandres1, Carlos Rodríguez-Morcillo2

1Automatic and Electrical Department, ICAI School of Engineering, Madrid, Spain2Institute for Research in Technology (IIT), Madrid, Spain

E-mail: [email protected]

Abstract: Power line communication has recently attracted the attention of energy companies as a useful and natural technologyfor building the advanced metering infrastructure. In this context, device language message specification/companion specificationfor energy metering (DLMS/COSEM) is an increasingly popular standardised application protocol for communication betweenutilities and their customers. This study analyses the communication performance that can be expected when using the power linecommunication technology, powerline intelligent metering evolution (PRIME), to send DLMS/COSEM messages. Physicalphenomena – such as background and impulsive noise sources, channel attenuation and multi-path effect – are taken intoaccount during the first step in the evaluation of this technology’s communication performance in the physical layer. Thismetric is then used in upper layers to compute the packet error rate. An analysis is carried out at the application layer in termsof expected latency in different communication environments. Several simulations are performed in a European low-voltagetopology to compute the number of metres that can be read within 15 min. These simulations were carried out usingMATLAB and OMNeT++ software.

1 Introduction

Recent years have seen an increasing interest in narrow-bandpower line communication (PLC) on the part of utilitycompanies. Although this technology is not new, there areseveral characteristics that make it suitable for smart gridapplications, such as the automatic meter-reading [1, 2]. As aresult, several transceiver designs have appeared in the pastfew years as designers have attempted to offer solutions forcommunication via power lines. Early models (X-10 orKNX) made use of a single-carrier technology to modulatedata. These solutions were targeted for in-home scenariosusing domotic applications. One of the first solutions toimplement multi-carrier modulation while working in thelow-voltage (LV) network is powerline intelligent meteringevolution (PRIME), which was introduced in 2007. Sincethen, a number of designs have been released such asG3-PLC (from the French utility EDF); G.hnem(recommended by the International TelecommunicationsUnion (ITU-T) and IEEE1901.2 (from the IEEE).There are several works in the literature that provide an

analysis of the performance of these solutions [3–5]. Thesestudies include several power line scenarios. However,although accurate, they only focus on the physical layer ofthe communication, and their main output consists of biterror rate (BER) curves. Nevertheless, these kinds of studieshave been found to be very helpful from the industry’sperspective, as mentioned in [6].

This work improves the previous performance analyses byincluding an implementation of the network level. This isachieved via a simulation framework in which bothnetworking and physical effects are taken into account.Owing to the characteristics of the PLC channel, aspectssuch as the channel access play an important role in thebehaviour of the network, as described in this paper.To the best of the authors’ knowledge, this approach has

barely been addressed in the literature so far. In [7], aninteresting analysis of the available throughput in multi-hoppower lines is conducted. Nevertheless, no channel noise isadded to the transmission, so no errors are modelled in thecommunication.Additionally, studies like that presented in [8] attempt

to analyse the effective data rate achieved in a PLCnetwork. However, the BER is fixed for all nodes in thesimulation, independent of their position, which is not arealistic situation.In [9], a method is proposed to abstract the physical (PHY)

layer from the simulations by means of packet error rateagainst signal-to-noise ratio (SNR) curves. However, whencomputing those curves, a fixed packet length is set, leadingto a not realistic situation either. A similar study to the onepresented in this paper is given by Zaballos et al. [1]. Theyanalyse the time required to read 100 m over a 64 kbpschannel. The packet loss probabilities are taken into accountin the simulations. However, no noise sources or channelattenuation are considered in the study.

1041& The Institution of Engineering and Technology 2014

Page 2: Advanced metering infrastructure performance using European low-voltage power line communication networks

www.ietdl.org

In the present study, the PHY, media access and logically

link control layers of PRIME are chosen as a PLC solution.The channel effects are analysed and included in thesimulation. To complete the communication layer stack, apopular application for remote metering has been used:device language message specification/companionspecification for energy metering (DLMS/COSEM). As itwas also done in [1], the time to read all meters in an LVnetwork is used as a metric to evaluate thecommunication’s performance. The simulation results showhow channel effects influence the performance of theapplication layer.The structure of this paper is as follows: Section 2 offers a

conceptual and mathematical description of the PLC channeleffects and how they will be modelled in the simulations. InSection 3, the same type of description is provided withrespect to the PRIME’s transceiver, with respect to both thephysical and the access and logical link layers of thedevices. A description of how the DLMS/COSEMapplication layer can be assembled in the PRIME layerstack is given in Section 4. The way channel andnetworking effects are integrated in the simulationframework is addressed in Section 5. The results obtainedfrom these simulations are shown and discussed in Section6. Finally, the paper concludes with a discussion of thesignificance and implications of the findings.

2 PLC channel description

2.1 Power line transfer function

Channel transfer function provides a relation between theinput and output signal powers in the channel. Thisrelationship is, typically, frequency-dependent, meaning thatsome frequencies are attenuated more than others.A common approach in the evaluation of the transfer

function in power line channels is to use transmission linetheory. This methodology was proposed for the first timeby Banwell and Galli in 2001 [10]. One major advantage of

Fig. 1 General scenario for the European and the US cases

a US distribution networkb European distribution network. Both adapted from [12]

1042& The Institution of Engineering and Technology 2014

this approach is that it is computationally more efficientthan a multi-path time analysis [10].Equation (1) represents the equivalent circuit of the power

line channel as a two-port network, where the ZL stands forthe receiver input impedance. The input and output voltagesand currents are related by the ABCD parameters as theequation shows

VS

IS

[ ]= A B

C D

[ ]VL

IL

[ ](1)

Given the above relation, the frequency response of thechannel can be computed by

H( f ) = VL

VS

= ZLAZL + B

(2)

The ABCD parameters depend both on the cables’ length andtheir electrical properties. Reference [11] shows in more detailhow transfer functions can be computed using transmissionlines theory and cables’ properties.With respect to the physical topology of the LV networks,

Fig. 1 shows a general scenario for the European and the UScases. European LV distribution networks typically consist ofa transformer which serves one cell. As stated in [12], this cellgathers a number of distribution lines or branches where usersare connected. In urban environments, these branches can bearound 400 m long [13] and may have around 30 users perbranch. Typically, there are 10 branches, meaning that eachtransformer serves around 300 subscribers.In the case of the US, the LV distribution network has fewer

branches, as Fig. 1a shows. The use of a greater number oftransformer produces smaller cells with up to 20 users [12].Independent of the topology, the total path can be computed

by modelling small segments (short transmission line, shuntloads and so on) using transmission matrices theory. Theoverall transmission matrix is the multiplication of thetransmission matrices of all those small segments. To

IET Commun., 2014, Vol. 8, Iss. 7, pp. 1041–1047doi: 10.1049/iet-com.2013.0793

Page 3: Advanced metering infrastructure performance using European low-voltage power line communication networks

www.ietdl.org

reproduce a realistic scenario, the procedure described in [14]has been reproduced in the European scenario describedbefore. The NAYY150SE and NAYY50SE cables have beenassumed for the main distribution and local loop,respectively, as done in [14]. The physical parameters hasbeen set to r = 15.6 mm, rinsu = 3.6 mm, r = 9.4 mm and rinsu= 2.8 mm, for the NAYY150SE and NAYY50SE, where rrepresents the conductor radius and rinsu represents theinsulation thickness. The relative permittivity, the specificresistance of the conductor and the loss angle are common toboth cables and have been set to er = 4, ρ = 2.8 × 10−8 Ωmand tan(δc) = 0.01. Additionally, the houses’ input impedancehas been generated randomly according to a uniformdistribution of [0, 5] xj [−5, 5]Ω, as reported in [14].

2.2 Channel noise

In this study, two types of PLC disturbances have beenconsidered [12]: background noise and impulsive noise.The combined noise is seen by the receiver as a stationarymixture process of two random processes. It can berepresented as

nk = wk + ik (3)

where wk and ik stand for the background and impulsivenoises and k represents a time index.As stated in [15], the background noise is mainly because

of several low-power noise sources. The power spectrumdensity of this kind of noise was derived frommeasurements reported in [15]. The expression is

W ( f ) = 10K−3.95×10−5 f [Hz][W/kHz] (4)

where K follows a normal distribution with mean μ =−5.64and a standard deviation of σ = 0.5. As it can be seen from(4), this noise is slightly coloured and its power decreaseswith the frequency.Additionally, the other main disturbance in the PLC

channel is the impulsive noise. This kind of noise has beenfound to be very damaging to PLC communications [12,16] and is in the focus of many researchers [5, 17]. Theimpulsive noise is because of different transients that occurin the network. This asynchronous impulsive noise istypically described by the Bernoulli–Gauss or theMiddleton’s Class A noise models [12]. In this particularpaper, the Middleton’s model has been chosen in order toprovide some comparison with other recent studies that alsoused this method, mainly [5, 8, 18–20]. The probabilitydensity function of this type of noise (pA(n)) can be written as

pA(n) =∑1m=0

e−A Am

m!

1�������2ps2

m

√ e− n2/2s2m

( )(5)

where

s2m = s2 m/A+ G

1+ G= s2

im

A+ s2

g (6)

A is known as the impulsive index, s2m is the total noise power

(both background and impulsive noise) and G = s2G/s

2I is the

mean power ratio between background and impulsive noises,sometimes referred to as the Gaussian-to-impulsive varianceratio. A more detailed description of the implementation ofthis noise model can be found in [21].

IET Commun., 2014, Vol. 8, Iss. 7, pp. 1041–1047doi: 10.1049/iet-com.2013.0793

3 PLC transceiver

PRIME standard [22] defines the PHY, medium accesscontrol (MAC) and logical link control (LLC) layers of thetransceiver. The PHY manages all mechanisms related withthe signal modulation, whereas the other two layers takecare of the correct way of using the channel and alsoimplement some functions related with logical connections.With respect to the PHY layer, PRIME uses the spectrum

between 42 and 89 kHz to insert orthogonal frequencydivision multiplexing (OFDM) symbols. To compose thetransmitted signal, the output of a 512-length inverse fastFourier transform is sampled with a 250 kHz clock. Thisgives a sub-carrier spacing of Δf = 250 kHz/512 =488.28125 Hz. Additionally, a cyclic prefix extension of 48samples (192 μs) is added to all OFDM symbols to avoidinter-symbol interference. This produces OFDM symbols oflength 560 samples (2.24 ms). Before the data modulate thesub-carriers, information bits are passed through adifferential phase shift keying (DPSK) modulator and extraredundancy may be added to increase the robustness againstnoise. A more detailed description of PRIME’s PHY layeris provided in [23].With respect to the MAC layer, PRIME basically defines

two kinds of nodes: base node (BN) and service node (SN).The role of BN is to manage the network resources andconnections. Only one BN should exist per network and therest of the nodes must register to it in order to be able totransmit data.SNs act as leaves or branch points of the logical

tree-structured network. The initial state of the SN isdisconnected. They may change to terminal by registeringto the network’s BN and only when they are already atterminal state they can be promoted to SWITCH. Althoughthe objective of a node in terminal state is to provideconnectivity to the upper layers, SWITCH nodes are alsoresponsible for forwarding traffic to and from other nodeswhose signals are too attenuated when they reach the BN.Given the physical characteristics of the power line channel,

the medium acts as a bus channel for all devices connected toit, that is, the medium is shared and no more than one node canbe transmitting at the same time. Another mission of the BN isto add some synchronism to the network so that every devicehas the chance to transmit. To do so, time is divided intoabstract slots known as MAC frames. BNs and SNs canaccess the channel during both the shared contention period(SCP) and the contention-free period (CFP). In the beginningof the MAC frame, the BN and switches (if present in thenetwork) send broadcast beacon messages. These messagescontain information regarding the time limits of the MACframe and the data needed to request a registration to theBN. Once all the SN’s in the network know the time limitsof the SCP and CFP, they contend to access the channelusing a carrier sense multiple access/collision avoidance(CSMA/CA) procedure.Additionally, the BN also performs network management

functions. It polls all SNs in the network to obtain anup-to-date state of all the devices connected to it. To do so,keep-alive messages are periodically sent. In the event thata response is not received from a particular SN, it will beconsidered as disconnected and it will have to re-register inorder to transmit more information.The MAC layer is also responsible for performing what is

called PHY robustness management. This makes use of thedifferent communication modes available in the PHY layer(see [23]). Depending on the channel conditions, nodes may

1043& The Institution of Engineering and Technology 2014

Page 4: Advanced metering infrastructure performance using European low-voltage power line communication networks

www.ietdl.org

transmit with several communication modes in order to usethe maximum transmission rate given the channelconditions. The performance obtained by the differentcommunication mode is shown in Section 4.

4 DLMS/COSEM integration with PRIME

This section describes how the communication layer stack canbe adapted so that DLMS/COSEM messages can betransmitted using the power line network.The principal idea is described in Fig. 2. The so-called

‘bridge node’ would be the medium voltage (MV)/LVtransformer (as seen in Fig. 1b). It must have twocommunication interfaces, one of them implementing theInternet layer stack in order to be able to receive andtransmit DLMS/COSEM messages from and to the utility’sapplication server. The other interface should be compatiblewith PRIME specification and, as such, it should includemedium access mechanisms and logical connectivity (MACand LLC layers in Fig. 2).

Fig. 2 PRIME and DLMS/COSEM communication stack

Fig. 3 Scheme of the simulation framework

1044& The Institution of Engineering and Technology 2014

The DLMS/COSEM applications at both the utility and theend user communicate with each other in a transparent mode,independent of the physical medium.

5 Simulation framework

As mentioned in Section 1, one of the key features of thiswork is the binding of physical power line channel effectstogether with the network behaviour. To accomplish this,the simulation framework displayed in Fig. 3 wasimplemented.The framework developed consists of a co-simulation

environment between MATLAB and a popular networksimulator: OMNeT++ [24]. As can be seen in Fig. 3, all theeffects related to the PHY layer of the communicationshave been taken into account via MATLAB simulations.Both PRIME receiver and transmitter have beenimplemented following the specification [22], and thepower line channel model includes all the attenuation andnoise effects described in Section 2. The main outputobtained from this part of the simulation framework is thetransceiver’s behaviour in terms of BER against SNR.These BER values are used then by the OMNeT++simulator to compute the probability of a packet to arrive atits destinations with or without errors. The MAC and LLClayers defined in PRIME are also developed with OMNeT++together with the application layer, implementing DLMS/COSEM protocol.The BER against SNR curves obtained from the MATLAB

simulations are shown in Fig. 4. Fig. 4a shows theperformance achieved by PRIME when the channel isimpaired by background noise. The different curves standfor the different communication modes defined in thestandard. It should be noted that the performance obtainedusing a constellation with a lower number of symbols(DBPSK for instance) is more robust to noise than whencompared with one which has a higher number of symbols(D8PSK), since the same noise amplitude is less likely toproduce a symbol misinterpretation. However, DBPSK’s

IET Commun., 2014, Vol. 8, Iss. 7, pp. 1041–1047doi: 10.1049/iet-com.2013.0793

Page 5: Advanced metering infrastructure performance using European low-voltage power line communication networks

Fig. 4 PRIME PHY performance

a With background noiseb With impulsive noise

www.ietdl.org

transmission speed is three times lower when compared withwhat is achieved with D8PSK.The use of forward error correction (FEC) mechanisms also

improves the performance in terms of BER, but once againhas a lower transmission rate [23].Additionally, in those cases where communications are

also affected by impulsive noise, the performancedeteriorates dramatically. This can be clearly seen inFig. 4b, where the impulsive noise model described inSection 2.2 is included in the simulations. In this case,impulsive parameters Γ and A were taken from [5] and setto 0.01 and 0.1, respectively.To emulate the telematics effects of the PLC network,

a popular network simulator, OMNeT++, was used.OMNeT++ [24] is a modular, C++ -based, events-orientednetwork simulator that has a number of communicationlibraries. This simulator can take into account all the effectsregarding packet segmentation, medium access, bustopology, delays and re-transmissions.As mentioned before, OMNeT++ is used for modelling the

telematics of the network. Thanks to this network simulator,effects like packet segmentation, medium access, bustopology, delays and re-transmissions can easily be takeninto account.MATLAB and OMNeT++ are linked by the BER, SNR

and the communication mode values. The process isexplained in the following lines. For a given transmission,OMNeT++ obtains two parameters from the receivedpacket: (i) communication mode used in the currentdialogue and (ii) the SNR of the received signal. OMNeT++ then retrieves from MATLAB the corresponding BERagainst SNR curve and, using the calculated SNR value,obtains a BER for the current transmission. This BER isused together with the length of the packet to decidewhether there have been errors in the reception or thepacket is correct and can be processed. This procedure isschematically expressed in Fig. 3.

Fig. 5 DLMS/COSEM data transmission diagram

6 Performance results

This section contains simulation results obtained with theframework described previously. These results are presentedin the following two subsections.

IET Commun., 2014, Vol. 8, Iss. 7, pp. 1041–1047doi: 10.1049/iet-com.2013.0793

6.1 Time to read 1 meter

The process required to read 1 meter, in accordance with thecommunication layer stack described in Section 4, is shown inFig. 5. Since PRIME is a connection-oriented protocol, beforesending the application request, a connection has to beestablished. The ESTABLISH.req message is triggered bythe GET.req received from the DLMS/COSEM layer. Therequested entity answers the ESTABLISH.req with thecorresponding ESTABLISH.res.When the connection is established, application data are

sent encapsulated in LLC-layer frames (DATA(GET.req)message in Fig. 5). The requested data values are includedin a GET.res message by the SN. After the reception of adata frame at the BN, an acknowledgment (ACK) messageis sent back to the transmitter to confirm the reception.Given this message exchange, the time to read 1 meter

(indicated in Fig. 5) can be computed with

Tread1Meter =ESTABLISH.req+ ESTABLISH.res

VTXCONTROL

+ GET.req+ GET.res+ 2 · LLCHEADER

VTXDATA

(7)

where numerators stand for the bit length of each one of themessages and the denominators are the corresponding

1045& The Institution of Engineering and Technology 2014

Page 6: Advanced metering infrastructure performance using European low-voltage power line communication networks

www.ietdl.org

transmission speeds. It is worth mentioning that not allmessages may be transmitted at the same speed. As definedin PRIME standard, all control messages (both the establishrequest and response in this case) must be transmitted usingthe most robust communication mode (DBPSK-FEC ON),which means 21.4 kbps. Data frames, however, can be sentusing the most suitable communication mode given thechannel conditions. Transmission speeds in those cases are21.4, 42.9, 42.9, 85.7, 64.3 and 128.6 kbps forcommunication modes DBPSK-FEC ON, DBPSK-FECOFF, DQPSK-FEC ON, DQPSK-FEC OFF, D8PSK-FECON and D8PSK-FEC OFF, respectively.

Fig. 7 Proposed topology in order to allow all meters in thenetwork to be polled

6.2 Time to read several meters

It is obvious that the situation analysed above does notrepresent a real-life scenario. Typically, utilities would liketo read all meters in the network. However, results cannotbe extrapolated from a 1 meter to a several-meter scenariofor several reasons. One of these is that the medium acts asa bus where all devices contend with each other fortransmitting. As mentioned, collisions are handled by aCSMA/CA algorithm; however, the analytical study of theperformance in such scenario is complex.This section presents a study of the time required to read all

meters in a network, as it was also proposed in [1]. Theframework described in Section 5 has been used to emulatea European topology, as was described in Section 2. Theresults show the required time as a function of the numberof branches present in the network. In addition, thisperformance is compared with a 15 min reference, assuggested in [16].Two polling strategies are also studied: simultaneous and

sequential polling. When using sequential polling, the BNwaits for the response of the SN currently being polledbefore sending a new request. In this simultaneous case, theBN massively sends GET.req messages to all meters. Thelengths of GET.req and GET.res messages are set to 254and 144 bytes, respectively. These numbers would allow fora request for 22 measurements in a single fragment, whichmight be sufficient for any conventional application, asstated in [25].Fig. 6 shows the time required to read all meters using

different types of strategies in different scenarios as afunction of the number of branches existing in the network.In concrete, the line with stars represents this performance

Fig. 6 Comparison of the time required to read all meters as afunction of the number of branches; averaged values

1046& The Institution of Engineering and Technology 2014

when using sequential polling. It can be seen that the 15minutes threshold is only exceeded in scenarios with up toseven branches. In addition, the line with circles stands forthe results obtained when using a simultaneous polling. Inthis case, the time required to read all meters increases upquickly with the number of branches. The 15 min time limitis now clearly exceeded when the network has five branches.As mentioned in Section 2.2, the PLC channel may also be

impaired by impulsive noise interference. The consequencesin terms of physical communication performance wereshown in Section 5. To understand how impulsive noiseaffects the network performance, OMNeT++ simulationswere fed with the BER/SNR curves shown in Fig. 4b. Ascan be seen, a direct consequence of the presence ofimpulsive noise is that nodes need a higher SNR toguarantee communications without errors. It is translated ina higher number of SWITCH nodes in the network, sincesignals need to be re-transmitted more often. This highernumber of SWITCH nodes produce extra traffic and,therefore, the time required to read all meters is increased.The results are represented by the line with squares inFig. 6. When the network has six or more branches, theamount of traffic is so great that the network cannot be set up.Simulations carried out reveal that a scenario with a LV

network of ten branches cannot be managed. A simplesolution to enable PRIME technology in Europeanenvironments with such a high number of branches wouldhave to use several BN’s per network. This could beachieved by creating several subnetworks. Each of whichwould need to be isolated from the rest in order to preventinterference from PRIME’s signals. A simple low-pass filterwould allow the transmission of power while blocking thehigh-frequency messages. A scheme for this idea is shownin Fig. 7.

7 Conclusions

This paper describes the implemented simulation frameworkfor PLC. In contrast to other studies, simulations carried outtake into account all the processes in the communicationlayer stack.Physical effects perturbing the communication, such as the

different types of noise sources and the attenuation sufferedby the signal, were modelled by MATLAB simulations.The output of these simulations is then applied, in the form

IET Commun., 2014, Vol. 8, Iss. 7, pp. 1041–1047doi: 10.1049/iet-com.2013.0793

Page 7: Advanced metering infrastructure performance using European low-voltage power line communication networks

www.ietdl.org

of BER against SNR curves, to a network simulationframework implemented with OMNeT++. There, processesrelated to the network topology were modelled.To complete the communication layer stack, a popular

interface model of electrical power communications(DLSM/COSEM) was used. The simulation frameworkshows how physical phenomena taking place in the channelis translated into the application layer in terms of timerequired to read a given number of meters.The benefit of the approach presented is that, because of the

decoupling of the physical effects on the networking, changesin the channel conditions (different types of noise sources) orin the PHY layer (new transceiver designs) can be easilyincluded in the simulation framework. Moreover, thisapproach also allows for the testing of other applicationsthat could use PLC as a communication medium (as theIEC60870-5-104, IEEE1901.2, G.hnem or G3-PLC).The authors also believe that this simulation framework

could result in an interesting tool for utilities whendeploying their smart grid solutions using PLCs. With theuse of this tool, aspects such as the maximum number ofsubscribers per area or the expected communication latencyor the polling frequency can be approximated.The simulation results show that a typical European LV

network cannot be completely polled using a single BN.Time required to read all meters increases considerablywhen the PLC network is affected by impulsive noise. Aproposed solution would be the isolation of a group ofbranches to create smaller subnetworks. Each subnetworkshould be operated by a different BN.

8 References

1 Zaballos, A., Vallejo, A., Majoral, M., Selga, J.M.: ‘Survey andperformance comparison of AMR over PLC standards’, IEEE Trans.Power Deliv., 2009, 24, (2), pp. 604–613

2 Sivaneasan, B., Gunawan, E.: ‘Modeling and performance analysis ofautomatic meter-reading systems using PLC under impulsive noiseinterference’, IEEE Trans. Power Deliv., 2010, 25, (3), pp. 1465–1475

3 Razazian, K., Umari, M., Kamalizad, A.: ‘Error correction mechanism inthe new G3-PLC specification for powerline communication’. Int.Symp. Power Line Communications and its Applications (ISPLC),2010, pp. 50–55, ISPLC2010

4 Kim, I.H., Varadarajan, B., Dabak, A.: ‘Performance analysis andenhancements of narrowband OFDM powerline communicationsystems’. IEEE Int. Conf. Smart Grid Communications(SmartGridComm), 2010, pp. 362–367

5 Mengi, A., Han Vinck, A.J.: ‘Successive impulsive noise suppression inOFDM’. Int. Symp. Power Line Communications and its Applications(ISPLC), 2010, pp. 33–37

6 Sanz, A., Pinero, P.J., Montoro, D., Garcia, J.I.: ‘High-accuracydistributed simulation environment for PRIME networks analysis andimprovement’. IEEE Int. Symp. Power Line Communications and itsApplications (ISPLC), Beijing, P.R. China, 2012, pp. 108–113

7 Kim, M.S., Son, D.M., Ko, Y.B., Kim, Y.H.: ‘A simulation study of thePLC-MAC performance using network simulator-2’. IEEE Int. Symp.

IET Commun., 2014, Vol. 8, Iss. 7, pp. 1041–1047doi: 10.1049/iet-com.2013.0793

Power Line Communications and its Applications (ISPLC), April2008, vol. 2, pp. 99–104

8 Korki, M., Hosseinzadeh, N., Moazzeni, T.: ‘Performance evaluation ofa narrowband power line communication for smart grid with noisereduction technique’, IEEE Trans. Consum. Electron., 2011, 57,pp. 1598–1606

9 Kim, K.-H., Lee, H.-B., Lee, Y.-H., Kim, S.-C.: ‘PHY abstractionmethodology for the performance evaluation of PLC channels’. IEEEInt. Symp. Power Line Communications and its Applications (ISPLC),2010, pp. 28–32

10 Banwell, T.C., Galli, S.: ‘A new approach to the modeling of the transferfunction of the power line channel’. IEEE Int. Symp. Power LineCommunications and its Applications (ISPLC), 2001

11 Matanza, J., Alexandres, S., Rodriguez-Morcillo, C.: ‘PRIMEperformance in power line communication channel’. IEEE Int. Symp.Power Line Communications and its Applications (ISPLC), Udine,Italy, 2011, pp. 159–164

12 Ferreira, H.C., Lampe, L., Newbury, J.: ‘Power line communications:theory and applications for narrowband and broadbandcommunications over power lines’ (Wiley, 2010)

13 Deconinck, G.: ‘An evaluation of two-way communication means foradvanced metering in Flanders (Belgium)’. IEEE Instrumentation andMeasurement Technology Conf. Proc. (IMTC), 2008, pp. 900–905

14 Lampe, L., Han Vinck, A.J.: ‘On cooperative coding for narrow bandPLC networks’, AEU – Int. J. Electron. Commun., 2011, 65, (8),pp. 681–687

15 Hooijen, O.G.: ‘A channel model for the residential power circuit usedas a digital communications medium’, IEEE Trans. Electromagn.Compat., 1998, 40, (4), pp. 331–336

16 Nassar, M., Lin, J., Mortazavi, Y., Dabak, A., Kim, I.H., Evans, B.L.:‘Local utility power line communications in the 3–500 kHz band:channel impairments, noise, and standards’. IEEE Signal ProcessingMagazine, 2012, pp. 116–127

17 Lampe, L.: ‘Bursty impulse noise detection by compressed sensing’.IEEE Int. Symp. Power Line Communications and its Applications(ISPLC), Udine, Italy, April 2011, pp. 29–34

18 Papilaya, V.N., Han Vinck, A.J.: ‘Investigation on a new combinedimpulsive noise mitigation scheme for OFDM transmission’. 2013IEEE 17th Int. Symp. Power Line Communications and itsApplications, 2013, no. 1, pp. 86–91

19 Mengi, A.: ‘On combined coding and modulation’. PhD thesis,Universitat Duisburg-Essen, 2010

20 Mengi, A., Han Vinck, A.J.: ‘Impulsive noise error correction in16-OFDM for narrowband power line communication’. Int. Symp.Power Line Communications and its Applications (ISPLC), Dresden,Germany, 2009, pp. 31–35

21 Matanza, J.: ‘Improvements in the PLC Systems for Smart GridsEnvironments’. Pontifical University of Comillas, Madrid PhD thesis,2013

22 ITU-T. G.9904: ‘Narrowband orthogonal frequency divisionmultiplexing power line communication transceivers for PRIMEnetworks’. Technical Report, 2012

23 Matanza, J., Alexandres, S., Rodriguez-Morcillo, C.: ‘PRIMEperformance under impulsive noise environments’. IEEE Int. Symp.Power Line Communications and its Applications (ISPLC), Beijing, P.R. China, 2012, pp. 380–385

24 OMNeT++ Community. OMNeT++. Available at http://www.omnetpp.org/

25 Feuerhahn, S., Zillgith, M., Wittwer, C., Wietfeld, C.: ‘Comparison ofthe communication protocols DLMS/COSEM, SML and IEC 61850for smart metering applications’. IEEE Int. Conf. Smart GridCommunications, October 2011, pp. 410–415

1047& The Institution of Engineering and Technology 2014