E2HRC: An Energy-Efﬁcient Heterogeneous Ring Clustering … 2017 - ns2 -base... · · 2017-11-15(IPv6 Routing Protocol for Low Power and Lossy Networks). A new clustering algorithm

2169-3536 (c) 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. Seehttp://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2666818, IEEE Access

JOURNAL OF LATEX CLASS FILES, VOL. X, NO. X, JANUARY XXXX 1

E2HRC: An Energy-Efficient Heterogeneous Ring ClusteringRouting Protocol for Wireless Sensor Networks

Wenbo Zhang, Ling Li, Guangjie Han, Member, IEEE, Lincong Zhang

Abstract-A heterogeneous ring domain communication topol-ogy with equal area in each ring is presented in this study inan effort to solve the energy balance problem in original RPL(IPv6 Routing Protocol for Low Power and Lossy Networks). Anew clustering algorithm and event-driven cluster head rotationmechanism are also proposed based on this topology. Theclustering information announcement message and clusteringacknowledgment message were designed according to RFC andoriginal RPL message structure. An Energy-Efficient Heteroge-neous Ring Clustering (E2HRC) routing protocol for wirelesssensor networks is then proposed and corresponding routingalgorithms and maintenance methods are established. Relatedmessages are analyzed in detail. Experimental results show thatin comparison against the original RPL, the E2HRC routingprotocol more effectively balances wireless sensor network energyconsumption, thus decreasing both node energy consumption andthe number of control messages.

Index Terms—Wireless sensor network, clustering algorithm,routing algorithm, E2HRC.

I. INTRODUCTION

A wireless sensor network is composed of wireless sensornodes and a sink node. Nodes are wirelessly interconnected toone another and to the sink. These networks are characterizedas Low-power and Lossy Networks (LLNs), as individualnodes possess limited power and operate in harsh environ-ments. If a node is not in direct communication range with thesink, the data it captures is reported in a multi-hop manner.In this way, nodes located closer to the sink end up relayingdata for nodes that are farther away, thus creating hotspotsnear the sink. These hotspot nodes tend to deplete energyfaster, thus reducing the wireless sensor networks lifetime.Wireless sensor network energy consumption is a popularresearch topic [1][2]. Considering of LLN characteristics andpossible applications, the Internet Engineering Task Force(IETF) Routing Over Low-power and Lossy networks (ROLL)group has standardized a low-power and lossy network routingarchitecture called RPL [3][4]. This protocol is an open

This work is sponsored by the New Century Program for Excellent Talentsof the Ministry of Education of China, Liaoning Province Innovation GroupProject (LT2011005), the Shenyang Ligong University Computer Science andTechnology Key Discipline Open Foundation(2012,2013), Liaoning FourthBatch of Distinguished Professor Project(2014), and Liaoning BaiQianWanTalents Program.

Wenbo Zhang is now with the school of Information Science and En-gineering, Shenyang Ligong University, Shenyang, 110159, China (Email:[email protected]).

G. Han is with Department of Information and CommunicationSystems, Hohai University, Changzhou, 213022, China (e-mail:[email protected]).

L. Li and L. Zhang are with the school of Information Science andEngineering, Shenyang Ligong University, Shenyang, 110159, China (e-mail:[email protected]; [email protected]).

and accepted technical standard in regards to wireless sensornetwork IP-based development. The salient design feature ofRPL is a routing framework that allows the use of differentrouting metrics and objective functions (OFs) to manageLLNs, including limitations and heterogeneous applicationrequirements.

Many other RPL-based routing protocols have been pro-posed for different optimization objects. Hakeem [5], for ex-ample, investigated packet forwarding and other factors to findthat when nodes number is 150, the network is stable enoughto satisfy practical application general requirements (basedon the cortex M3-nodes for original RPL routing protocolin networks [6]). Li [7] utilized an original RPL protocolimplemented by TinyRPL and contikiRPL operating systemsto test real-world performance per factors including routingfairness and packet delivery rate; the results showed that theprotocol merits further improvement in regards to networkpacket and routing control packet overhead. Gaddour [8] testedenergy consumption, packet delay, and packet loss rates ofthe original RPL protocol to find that energy consumptionand packet loss rates significantly increase as the number ofnetwork nodes increases. The original RPL routing protocol isplanar and the objective function is singular, rendering it un-able to adapt to the excessive packet numbers that accompanyan excessive number of nodes. Iova [9] performed detailedtest and performance analyses of the original RPL routingprotocol to explore end-to-end delay, DIO packet transmission,and other parameters. The results showed that the networkis flooded with control packets as soon as route maintenancebecomes necessary. These control packets substantially impactoverall energy efficiency and network stability.

If the inter-cluster transmission mode in the wireless sensornetwork clustering routing algorithm is single-hop transmis-sion, the outer cluster head nodes will be located far awayfrom the base station. Inner cluster head node energy con-sumption is greater than that of other nodes due to lengthytransmission distance [10][11]. If the inter-cluster transmissionmode is multi-hop, the inner nodes will be close to thesink and consume more energy due to the large amountof data transmitted to them from outer cluster head nodes[12]. In a ring domain multi-sector cluster network, the inter-cluster transmission mode is generally multi-hop, meaning theouter cluster head nodes send data layer-by-layer through thecluster head node of the adjacent ring to the sink [13]. Theoptimal cluster number of each ring makes the size of theouter cluster larger than that of the inner cluster, thus intra-cluster transmission energy consumption carried out by theouter cluster head nodes is higher than that of intra-clustertransmission carried by the inner cluster head nodes [14]. In a




uniform split ring, the larger area of the outer cluster producesadditional transmission energy consumption [15]. Inner andouter cluster region shape is thus made more uniform byimposing a non-uniform split ring structure [16]. However,an uneven split ring can also balance inner and outer clusterhead energy consumption owing to an increase in outer ringspacing.

In this study, we designed a heterogeneous ring clusteringalgorithm and E2HRC routing protocol that minimizes overallenergy consumption and balances the consumed energy. Rel-ative messages were also designed and implemented based onoriginal RPL routing protocol message structure.

The remainder of this paper is organized as the follows:Section II elaborates on related works and our contributions;Section III establishes a topology control model based on thering communication area; Section IV describes the E2HRCclustering algorithm; Section V presents the E2HRC routingalgorithm; Section VI presents the results of our experiments;and Section VII offers a conclusion.

II. RELATED WORKS

The RPL is an IPv6 distance vector routing protocol de-signed for networks with limited resources (energy, computa-tion, etc.) [17]. The RPL works by constructing a connectiontree, called Destination Oriented Acyclic Graph (DoDAG), torelay traffic from many sensors to all nodes.

To build this DoDAG, RPL calculates an objective functiondefined as a combination of metrics and constraints. Thisfunction allows the best path to the destination to be chosen.Metrics and constraints are managed as RPL introduces newICMPv6 (Internet Control Messages Protocol) messages. Eachnode periodically broadcasts a DIO (DoDAG InformationObject) message to advertise graph characteristics (objectivefunction, node rank, etc.). Each node joins a DoDAG andmaintains its path towards the sink. A node broadcasts a DIS(DoDAG Information Solicitation) when it needs to join theDoDAG and immediately receive DIO. The DAO (DoDAGDestination Advertisement Object) is transmitted by each nodeto validate the path choice and to notify the selected parent.This parent then sends a DAO acknowledgment to the sendernode when it receives the DAO.

Kim [18] addressed the downward routing reliability prob-lem in RPL and designed an asymmetric transmission power-based network where the root directly transmits downlinkpackets to destination nodes using higher transmission powerthan low power nodes possess. Ko [19] proposed the inter-operability problem of two operation modes (MOPs) definedin the RPL standard, ultimately demonstrating that there isa serious connectivity problem when two MOPs are mixedwithin a single network. The DualMOP-RPL, which allowsnodes with different MOPs to communicate gracefully in asingle network while preserving the high bi-directional datadelivery performance, was established to solve this problem.None of these studies have investigated RPL load balancingproblems over a real multi-hop LLN testbed, however [20].

Many energy consumption algorithms have been proposedto optimize and improve original RPL routing protocol.

Gaddour [21], for example, proposed the CO-RPL, whichfunctions by using multiple route selection to improve theoriginal RPL object function. CO-RPL exhibits better energyconsumptions and packet pass rates than original RPL, butpacket congestion and control packet flooding are still possiblein large-scale wireless sensor networks. Zhang [22] establisheda new RPL routing object function based on energy-efficiency,but only energy conservation (not energy balancing) is con-sidered during routing.

A multi-father node routing strategy was proposed for theoriginal RPL by Iova [23] that can increase wireless sensor net-work redundancy, increase stability, balance energy consump-tion, and extend lifespan. Even with this routing maintenanceand routing reconstruction, the network is still congested bya large number of packets; node memory becomes exhaustedwith an increase in redundant routes, necessitating more nodememory to satisfy the performance requirement.

Research indicates that hierarchical networks have unpar-alleled advantages over other networks. Yu [24] proposed anenergy-aware clustering algorithm to solve unbalanced energydissipation accompanied by a new routing algorithm based onthis clustering algorithm; the routing algorithm operates bydividing a nodes competitive radius when attempting to join acluster. The node with the most energy is then selected.

Yan [25] built a clustering topology control algorithm basedon a wireless sensor network with an uneven gradient. Othernodes calculate their own weight and send these weights totheir neighbors to participate in cluster head selection basedon global parameters transmitted by the sink. Nodes withpowerful communication capabilities are effectively selectedas cluster head nodes, however, this causes nodes to send toomany campaign control messages, resulting in a flood of con-trol messages to the wireless sensor network and essentiallywasting node energy.

Liu [26] proposed the EDUC algorithm based on theprobability model with regards to energy and distance. To bespecific, the avoiding time message that the candidate clusterheads send to campaign control is calculated according to nodeenergy and distance: The more energy a node has and thecloser it is to the sink, the shorter the avoiding time. It is thuslikely that a certain node becomes a real cluster head. Thisalgorithm may cause a large number of nodes close to thesink to become cluster head nodes, however. When clusterhead nodes are unevenly distributed in the wireless sensornetwork, the traffic is increased in the clusters; further, therouting convergence time is longer due to low network powerand low loss.

Liu [27] proposed a multiple hop clustering routing algo-rithm (RBMC) based on ring structure in which the com-munication area is divided into uniformly spaced concentriccircles, thus realizing multiple hop communication betweenclusters supposing there is equal energy consumption in eachring. The number of cluster head nodes in each ring can thenbe calculated successfully. This algorithm can balance energyconsumption in the wireless sensor network, while uniformlyspaced concentric circles create an inner ring coverage areathat is smaller than the outer ring thus preventing an energyhole in the inner nodes; the packet loss rate increases greatly




due to signal interference from crowded nodes in inner clus-ters, however.

In the scheme proposed by Krishnan [28], each node con-firms the avoiding time associated with their sending clusteringestablishment message by calculating its own threshold. Ifthe threshold is greater, the cluster establishment situation isbetter and avoiding time is shorter. This method can effectivelydistribute cluster heads very evenly, however, undesired nodeswill be elected to the cluster head if each nodes avoiding timeis too short or the communication link is not stable enough.The routing convergence time also increases if the avoidingtime of each node is too long. This method is not suitable forlow power, low loss networks.

The main contributions of this paper are as follows:1) A topology control model based on ring domain commu-

nication routing is proposed. Nodes are divided into dif-ferent levels in this model based on different positions.Various ring domains are also divided based on differentlevels. This causes the next hop node with optimaldirection angle to be selected during inter-ring domaincommunication, thus reducing the energy consumed bysending and receiving data packets.

2) A clustering algorithm for a routing protocol based onenergy balance is proposed. A clustering probabilitymodel is used to divide the network into various sizedheterogeneous clusters based on node residual energyand relative node position in the cluster. A combinationof heterogeneous cluster and cluster head rotation mech-anisms serves to balance node energy consumption inthe network to avoid the generation of a network energyhole.

3) The E2HRC routing protocol based on energy-efficientheterogeneous ring clustering is described in detail,including those messages for clustering, clustering ro-tation, route establishment, and route maintenance.

III. TOPOLOGY CONTROL MODEL BASED ON RINGCOMMUNICATION

Sensor node antennas are generally omni-directional inpractical applications, allowing wireless signals to spreadoutward in spherical waves. However, the signal spreads ina circular wave in the two-dimensional plane. Thus, nodesin the same domain in the ring radius also receive wirelesssignals without considering channel interference. Ring do-main communication topology can effectively use 360◦ signaltransduction while also reducing message collision during thetransmission process. Here, we determined the domain gradebased on the received signal strength indicator (RSSI). Signalintensity decreases during the wireless signal transmission pro-cess as transmission distance increases. We can thus transformsignal intensity attenuation as signal transmission distance,using the function between signal strength attenuation andsignal transmission distance to approximate the distance - thisis the principle of distance measurement based on RSSI. Wethen calculated the radio signal transmission power and therelational expression to determine the distance between thesource and the sink, allowing us to determine whether or not

Fig. 1. Topology of ring communication

the node belongs to the gradient ring. This topology is shownin Fig.1.

We then determined the node ring domain levels. The nodemaximum transmit power and minimum sensing power arespecified in IEEE 802.15.4. Per the class divisions, we onlyspecified the relationship between different ring radii and notthe specific first-layer ring domain radius size. We then setthe third layer loop domain boundary as the minimum sensingpower boundary point to guarantee network coverage and rout-ing model availability. The node level ring domain commu-nication topology thus inherits original RPL rank parameters,effectively avoiding routing loopback and maintaining existingrouting protocol advantages. This structure was introduced tomake clustering boundaries clear and to effectively controlthe size of the heterogeneous cluster, providing effective inputparameters for the sensor network clustering algorithm.

IV. E2HRC CLUSTERING ALGORITHM

In a low-power, vulnerable wireless sensor network, thenode location is fixed and the link relatively unstable. Thisallows only one cluster to form during the network operationperiod. During cluster head rotation, events are defined basedon the cluster head rotation mechanism. There is no need tointroduce clustering round number parameters in the clusterprobability mode assuming that the initial node energy isisomorphic. Cluster probability threshold Tk,j(s) is definedas follows:

Tk,j(s) = {Cch(k)

NC

∗ Eleft(j)Einit(j)

∗ ω s ∈ G(k)0 otherwise

(1)

where the subscript k represents the first k layer circle, jrepresents the first j node in the first k layer circle, Cch(k)represents optimal cluster head nodes in the first k layer, k istotal number of nodes in the ring communication area, C isthe number of rings in the ring communication area, Eleft(j)is residual node energy, Einit(j) is initial node energy, ω isa cluster adjustment factor, and G(k) represents all nodes inthe k layers annular communication regions.




A. Clustering Algorithm Description

Before describing the clustering algorithm, we will firstdefine the following data structure.

Definition 1 (Global configuration information I). I ={S,N,C}, where S is the total area of network coverage; Nis the number of nodes in the network; C is the total numberof rings of the network.

Definition 2 (DIO CLUSER data packet). Packet containsglobal configuration information of the Network I belongingto the domain layer k, the ring domain clustering probabilitythreshold T (k), and the optimal cluster head number Ch(k).

Definition 3 (DIO CCH data packet). Candidate cluster headdata packet contains the address information Addr, remainingenergy E, domain k, and the distance from the cluster headnode dhead of cluster head candidate node.

Definition 4 (DIO CH data packet). Cluster head notificationdata packet contains layer domain k belonging to cluster headand address information.

The algorithm can then be described step-wise as follows:• The sink is used to calculate clustering probability thresh-

old nodes T (k) in each layer ring communication areabased on network global configuration information andcluster probability model. This global configuration in-formation and threshold T (k) are encapsulated in theDIO CLUSER data packets and multicast.

• The ring domain is determined and T (k) is analyzedin the DIO CLUSER data packet based on the distancebetween RSSI and the sink, where a random number his then generated after receiving the DIO CLUSER datapacket.

• If the random number h is less than T (k), the nodebecomes the candidate cluster head node and sends themulticast to the DIO CCH packet.

• It must first be determined whether or not the candidatecluster head belongs to the same ring domain if clusterhead A receives the DIO CCH data packet sent by clusterhead B. If it does not, the packet is discarded and distance

d among nodes is calculated. If d <

√r2k+1−r

2k

ch(k), node

A acts based on candidate cluster head election criterionαE+βdhead to calculate and compare both values. If thevalue of node A is small, it reflects the status of candidatecluster nodes and sends DAO CH data to node B, thenselects node B as the network cluster node.

• Before sending the DIO CH data packet, each candidatecluster head node waits for a certain time to becomethe cluster head node. Based on target function (ObjectFunction, OF), the common nodes select cluster headnodes and send DAO CH packets to the cluster.

• Unless the cluster head node has not received anyDIO CH packets, the node initially sends DIS CH dataout to seek cluster head nodes. The node automaticallybecomes the cluster head node and sends multicastDIO CH packets if the cluster head node has still notbeen identified.

• The entire network clustering process is complete onceall nodes are successfully integrated into the cluster.

In the clustering process, the entire wireless sensor networkcontains nodes in the ordinary state, candidate cluster headnodes, and the cluster head nodes defined by different conver-sion conditions.

B. Cluster Head Rotation Mechanism

Cluster head node energy consumption is much greater thanthat of a common node because it serves as the messagepacket forwarder in the present cluster and conducts messageforwarding between the backbone networks. If no cluster headrotation mechanism is provided, an energy hole readily formsin the network, leading to local and even entire networkparalysis. There is a great deal of control message transmissionand data packet transmission interference when conventionalregular replacement of the cluster head mechanism is ap-plied to a low-power lossy network, thus inducing networkthroughput. In this case, cluster head node rotation leads toglobal repair of the backbone network, consuming a significantamount of energy. We propose an event-based cluster headrotation mechanism with the following packet definitions.

Definition 5 (DIO CH POISONING). Cluster head rotationdata packet, including cluster head node address Addr be-longing to the ring domain class Rank.

Definition 6 (DIO CH POISONING ACK). Cluster headnode rotation response data packet.

Definition 7 (DIO CH APP). Cluster head node appointmentpacket, including cluster head node address Addr appoint-ment.

Definition 8 (DIO CH LOSE). Cluster head node lost con-tact data packet. When cluster nodes fail to connect a certainnumber of times, data packets are multicast to other clusternodes.

Cluster head rotation events are triggered when cluster headenergy drops to the set energy threshold or when contact withthe cluster head node is unexpectedly lost, thus launching anew cluster head selection. Cluster head rotation is triggeredunder different initiators utilizing two mechanisms: Clusterhead node initiation and non-cluster head node initiation.

During cluster head rotation initiated by the clus-ter head node, cluster head node energy consumptionreaches a threshold E0 and cluster head nodes multi-cast DIO CH POISONING data packets to initiate rota-tion and local repair in the backbone network withinthe network structure. After the cluster node receivesDIO CH POISONING, the cluster head node is sent re-sponse packet DIO CH POISONING ACK, and calculatesthe appropriate candidate cluster head node based on eachsending nodes packet parameters followed by the sending ofthe DIO CH APP packet. After the backbone network nodereceives DIO CH POISONING, the routes associated with thenode are marked as unreachable and the system waits for thenew cluster head node to send network requests. If the newcluster head address is the same as that of the node, it then




waits for other nodes in the cluster to send the DAO CH datapacket.

Conversely, if the address differs, it sends DAO CH data tothe new cluster head node. This new cluster head node thensends DIO messages to locally repair the backbone network.Cluster head rotation is thus initiated by non-cluster headnodes. The node multicasts the DIO CH LOSE packets withinthe cluster if cluster member nodes connected to the clusterhead node time out in the network architecture cluster. Afterthe cluster member node receives the DIO CH LOSE datapacket, it determines a cluster head node address is its own,thus increasing cluster head node loss count. If this nodelost count reaches threshold M , the cluster nodes initiatecluster head rotation. The first initiated cluster head nodebecomes the temporary proxy cluster head and sends theDIO CH POISONING data packet. This completes the clusterhead rotation mechanism.

The communication area features a relatively uniform dis-tribution of cluster heads and heterogeneous cluster networktopology of each ring domain containing varying cluster sizes.The distribution of the communication ring domain cluster isdense and cluster size is relatively small and attributable tothe close proximity of the sink. This network saves energyduring data packet forwarding. The system realizes distributedcluster head rotation, avoids empty holes, and effectivelyreduces cluster head rotation in the network control packetssent, thus reducing network energy consumption, network loadrate, and packet conflict, ultimately improving overall networkperformance.

C. Design and Process for Clustering Information An-nouncement Message

During the cluster establishment phase, rank value calcula-tion, cluster head competition, or nodes joining in the clustermay occur. Only the clustering information announcementmessage (DIO CLUSTER) control message is designed toreduce control message transmission volume. When a nodereceives the clustering information announcement message,it performs different processing procedures based on its ownstate. It sends the cluster confirmation message (DAO CACK)and registers its own address and other information to thenode selected cluster head. The sink first sends the clusteringinformation announcement message. Subsequently, the innernode processes the message information according to the nodestate and modifies part of their related fields in the messagebefore transferring it. The related node states during thisprocess are as follows.

Definition 9 (Dissociated). The initial state in which the ranknode value is unknown when in this state, and thus the nodecan only send a DIS request message to obtain the DIOinformation message.

Definition 10 (Waited). The node receives the clustering in-formation announcement message and becomes the nonclusterhead node by calculating the clustering probability model.The node then waits to receive the clustering informationannouncement message.

Fig. 2. DIO CLUSER option message

Definition 11 (Competing). The rank value has been con-firmed and becomes the candidate cluster head node. The nodewaits to receive the DIO message from the cluster head nodewith the same rank value to calculate the cluster head nodeand join the cluster.

Definition 12 (Announcing). The node has cluster head nodecampaigns and announces to the other nodes with the samerank value to join the cluster, telling those with different rankvalues to build a backbone network routing.

All needed information during clustering should be includedin the cluster information announcement message as a nodeprocesses different states differently after receiving the clusterinformation announcement message. The detailed messageformat is shown in Fig.2.

The descriptions of specific options in the message are asfollows:• Type: Message format type, defined as 0x0a.• Option Length: Length of optional referring to the resid-

ual byte number of the message after the length of Typeand Option Length are eliminated. (Here, it is 6 bytes.)

• RevSinkPower: Sink sending power, which is used toconveniently calculate the distance between the node andthe sink to determine the rank value.

• CoverArea: The coverage area used to calculate clusterthreshold probability in each ring region.

• NodeCount: The number of nodes.• LeftPower: The residual energy of the node used to com-

pete between candidate cluster head nodes and parameterselection in backbone network routing.

• CluserFactor: The clustering adjustment factor, a cluster-ing probability threshold parameter.

• NS: Used to identify sending node state, classified as dis-sociated, waited, competing, or announcing. The messagereceiver judges the receiving and processes different statemessages.

Nodes with different states finally become common nodesor cluster head nodes by sending the clustering informationannouncement message, thus finishing the network clusteringprocess. However, when a node wants to join a cluster, itmust obtain unicast confirmation information sent by thecluster node to ensure the joining cluster success because eachnode sends clustering information announcement messages viabroadcast.

D. Design and Process for Clustering Confirmation Infor-mation Message

The clustering confirmation information message is mainlyused when the waited node unicasts DAO CLUSER ACK to




Fig. 3. Clustering confirmation information message

the cluster head node after receiving the clustering informationannouncement message. The three-way handshake mechanism(similar to TCP protocol) is adopted to ensure the correctreception of cluster joining confirmation information and toavoid the interference of any malicious message during themessage sending process. The clustering structure confirma-tion information message is illustrated in Fig.3.• Type: Extensible message, defined as 0x0b.• Option Length: Length of extensible message type, 6

bytes.• SYN/ACK: Request/acknowledgement mark. The node

sending the cluster joining request is SYN; the clusterhead node confirms as ACK and value is computed asSYN+ACK.

• ADDR: Node address, or sending node address informa-tion.

The node receives clustering information announcementmessage DIO CLUSER sent by the cluster head node from theabove clustering confirmation information message sendingprocess and thus decides whether or not to join the clusterbased on the objective function. If the node chooses to joinin the cluster, it sends a clustering confirmation informationmessage including SYN and node address and applies to thecluster head node, where the computer expects SYN+ACK.When the cluster head node receives the application, it judgeswhether to accept the node request or not based on theclusters current state. If the cluster head node agrees, itsends a DAO CLUSER ACK message including SYN+ACK.The waiting node then compares the received SYN+ACKwith its own expectation after receiving this message, andsuccessfully joins the cluster and sends data packets if theyare equal. If these expectations are unequal, the node discardsthe message and continues to wait. The node quits joininga cluster during a timeout period and receives the clusteringinformation announcement message from other cluster headnodes.

V. E2HRC ROUTING ALGORITHM

A. Algorithm Description

We established backbone network type routing mechanismbased on optimal direction angle (DA), node residual energy(EN), and hop difference (HC). When selecting relay nodes,the nodes in the backbone network consider the optimaldirection angle, the residual energy, and the minimum numberof hops to reduce network energy consumption and balancenetwork energy.

Definition 13 (Optimal direction angle, DA). We joined thesink to the relay node in a straight line and joined the relaynode to source node in another straight line. If the included

Fig. 4. Diagram of communication area

angle between those two lines is less than or equal to 180◦

and greater than or equal ◦, this included angle is the optimaldirection angle.

Theorem 1. Communication between node m and sink nodeby relay node n with optimal direction angle can reduce energyconsumption (compared with direct communication with sinknode .) as shown in Fig.4.

Proof: Connect points m, n, , to 4mnδ, using the trigono-metric function theorem:

d2(m, δ) = d2(m,n) + d2(n, δ)− 2d(m,n)d(n, δ) cosα (2)

Given that mn ⊥ L, α =∠mnδ > 90◦, cosα <0, which is−2d(m,n)d(n, δ) cosα > 0. Then, from per Eq.(2)

d2(m, δ) > d2(m,n) + d2(n, δ) (3)

The following shows that node energy loss is proportionalto the square of the communication distance[29].

ETx(u, v) =

{qEelec + qεfsd(u, v)

2 d(u, v) < d0qEelec + qεmpd(u, v)

4 d(u, v) ≥ d0(4)

d0 =

√εfsεmp

(5)

where Eelec is consumed energy for circuit transmissionunit data, the threshold value d0 is a constant, ε is a constant,and when d(u, v) < d0, ε is εfs, power amplifiers use thefree-space transmission mode. When d(u, v) ≥ d0 , ε is εmpand power amplifiers use the multi-path fading channel transfermode. α is the path loss exponent in outdoor radio propagationmodels, where usually α ≥ 2 (in this study, α is either 2 or4), thus: P (m, δ) > P (m,n) + P (n, δ) and

P (m, δ) > P (m,n, δ) (6)




where relay node n is located on line segment mδ, d(m, δ) =d(m,n) + d(n, δ), on both sides of Eq.(4) defines the square,so, P (m, δ) > P (m,n, δ).

The proposed routing algorithm is a stepwise process.• The sink sends a DIO message received by cluster head

nodes on the first communication ring domain, whichthen unconditionally selects the sink as a parent nodeand sends the DAO message downward. The nodes inthe other communication loop domains do not touch theDIO message. If the node in the first communication loopdomain does not receive the DIO message sent by thesink, the node sends a DIS message to the sink.

• The DIO router messages containing the distance betweenthe node and the sink, node rank value, the residualenergy of nodes, and other parameters are sent whenselecting the optimal parent node to provide referenceselection for the nodes in the lower loop domain.

• The energy optimization network model is first appliedto calculate the DA among current inter-nodes, relaynodes, and the sink when the cluster head nodes of theother layer domains receive a DIO message. The node isdiscarded when DA is less than 90◦.

• The formula θ = α ∗ DAπ + β ∗ ENEinit

+ γ ∗ HC10 is usedto identify the node with maximum routing metric valueof θ as the optimal parent node, and is then sent in theDAO message to form a downward route.

• The node sends a DIS message to judge whether rankvalue is greater than the receiving nodes own rank valueif no optimal parent node has been identified when thebackbone network times out. If the rank value is greater,the node sends a DIO route notification message andyields the optimal angle of the judgment according tothe formula θ′ = β ∗ EN

Einit+ γ ∗ HC10 , thus selecting the

optimal parent node to receive the DAO message.The backbone network in the communication loop domain is

formed through this routing network algorithm. The backboneis similarly routed to the original RPL routing protocol, whichsupports point-to-multipoint routing, point-to-point routing,and multipoint-to-point routing. The algorithm is improvedby optimal parent node selection after introducing two pa-rameters, optimal direction angle and energy, to reduce andbalance overall network energy consumption. Communicationwithin the cluster includes point-to-multipoint communicationbetween ordinary nodes and cluster head nodes, point-to-pointcommunication between cluster head nodes and ordinary oramong ordinary nodes, and multipoint-to-point communicationbetween ordinary nodes and cluster head nodes.

B. Design and Implementation for Routing Scheme in theNetwork

The node is in the announcing state during the clusteringstage. The node saves the candidate neighbor node informationwhen receiving DIO CLUSER messages from other nodes indifferent ring region ranks in the same state. Collecting theneighbor node information during the clustering stage enablesthe neighbor node to collect as much information as possibleby taking advantage of the broadcasting of the DIO CLUSER

Fig. 5. Route message notification message (DIO RAI)

message, thus improving message utilization. However, thenode only needs to unicast the targeted routing announce-ment information to candidate neighbor nodes to establishthe corresponding routing information and avoid broadcastingthe messages in this stage in the routing establishment stage,relieving network pressure and reducing network total energyconsumption.

Here, we propose a message notification message for routeestablishment that functions via the message sending andreceiving process. Combined with the objective function, wethus establish the upward routing, downward routing, andpoint-to-point routing processes. The routing message notifi-cation message DIO RAI is sent along with the DIO messagein terms of the extension head as shown in Fig.5.

Each field in the message is defined as follows.• Type: Extension head type, denoted as 0x0c.• Option Length: Extension head length, denoted as 10

bytes.• Prefix Length: The prefix in the network always repre-

sents a co-owned node part in a cluster, using the sameprefix for the nodes in the same cluster. When the nodejoins the cluster, the cluster head allocates an address forthe node according to its prefix.

• Source Address: Node address sending data packets.• Target Address: Node address either in the higher ring

region or lower ring region.• RSSILoss: Strength loss of received signal.• Upward routing: The routing that the data packets are sent

in the sink. To form this routing, the node with a lowerrank value sends the upper level node routing messageto the lower level node with higher rank value. Afterreceiving the routing notification message, the lower levelnode calls the objective function to judge if it is theoptimal forwarding node.

• Downward routing: The opposite direction of upperrouting, where the node sends a notification messageincluding the routing information of lower level nodeto upper node. After receiving the message, the upperlevel node calls the objective function to judge if it is theoptimal forwarding node.

• Point to point routing: The forwarding route of nodedata packets in the same ring region. This data packetforwarding is realized by first choosing upper routingto find the common ancestor node, and then downwardrouting to forward. (We do not specifically deal with thistype of routing in the routing establishment stage.)

The routing information directed by the optimal directionangle objective function based on energy optimization isformed in a backbone network through the above sending




Fig. 6. Cluster head rotation message

Fig. 7. Cluster head rotation message notification confirmation

and receiving routing information message processes. Afterrouting is established, the node dynamically adjusts to thetransmitting power by the next hop node in the routinginformation when forwarding the data packets, thus reducingnode energy consumption when sending data packets.

C. Design and Implementation for Network Repair Message

The network is inevitably plagued by node unreachabilitydue to the characteristics of the inner areas (such as the energyof nodes or software problems) or external factors (such aschanges of external environment) as network running time isprolonged. When this occurs, corresponding repair measuresare started up to repair the network.

In the cluster head rotation scheme, the packets involvedinclude the cluster head rotation message notification packet(DIO CH POISONING), cluster head rotation message no-tification confirmation packet (DIO CH POISONIG ACK),cluster head appointment packet (DIO CH APP), andcluster head lost contact message notification packet(DIO CH LOSE).

The cluster head rotation message notification packet is akind of information notification packet that is actively sentby the cluster head node to the cluster members. The goal ofpacket sending is to enable cluster members to send a clusterhead rotation message notification confirmation packet afterreceiving this message, thus enabling the cluster head node toselect a new cluster head node . The packet structure is shownin Fig.6.• Type: Extensible head type, the value is 0x0d.• Option Length: Extensible head length, the value is 2

bytes.• Reserved: The reserved field.The node sends the parameters (e.g., remaining energy) to

the cluster head node through the cluster head rotation messagenotification confirmation packet. The node then becomes awaiting node. The packet structure is shown in Fig.7.• Type: Extensible head type, the value is 0x0e.• Option Length: Extensible head length, the value is 2

bytes.• Left Power: The remaining energy of the node.• Reserved: The reserved field.The new cluster head node has been determined when the

old cluster head node sends the cluster head appointmentpacket. The cluster head appointment packet is sent to the

Fig. 8. Cluster head data message packet

Fig. 9. Cluster head lost contact message

nodes in the original cluster and the cluster head nodes inother ring regions among the old cluster head nodes. Afterreceiving this packet, the cluster members judge if the nodebecomes the cluster head node or joins a new cluster. Thecluster head nodes in other ring regions will send the routingestablishment related packets to the new cluster head nodeand repair the routing in the backbone network. The packetstructure is shown in Fig.8.• Type: Extensible head type, the value is 0x0f.• Option Length: Extensible head length, the value is 34

bytes.• Prefix Length: The prefix length of the new cluster head

node address.• Reserved: The reserved field.• Old Address: The original address information. The node

will judge by itself whether the new cluster uses it or not.• New Address: The address information of the new cluster

node.The cluster head lost contact message notification packet

(DIO CH LOSE) is used to announce to the cluster head thata node has been lost. The packet structure is shown in Fig.9.• Type: Extensible head type, the value is 0x10.• Option Length: Extensible head length, the value is 18

bytes.• Prefix Length: The prefix length of the cluster head node

address.• Reserved: The reserved field.• CLUSTER Address: The address information of cluster

head node.There are many various packet types among nodes during

the rotating cluster head, though the procedural process ofthese packets is relatively simple. This range is mainly used to




TABLE ISIMULATION PARAMETERS

Attribute Parameter Value

Number of the Nodes 120Simulation time(s) 3600Simulation area 500*500 m2

Simulator CoojaMote Type Sky moteOperating System Contiki 2.7Radio medium UDGRM(Unit Disk Graph

Medium):Distance LossMote startup delay (ms) 1000Transmission range (m) 50

realize communication negotiation among nodes through mes-sages. Here, we display various packets processing proceduresbased on flowcharts between various nodes. When the clusterhead node rotates, the routing information of the backbonenetwork also changes because the backbone network nodesare composed of cluster head nodes. Cluster head rotation isexecuted via local autonomy to minimize energy consumptionand network message congestion. Thus, the repair and reestab-lishment of routing information updates local information aswell. Repair routing and reestablishment are triggered by twodifferent events: The routing repair scheme actively triggeredby routing nodes and the routing repair scheme triggered byproxy nodes. The difference between these two schemes isthe trigger event; the process of each node is essentially thesame. The packets in need of routing repair are the same asthe clustering confirmation information messages.

VI. SIMULATION AND ANALYSIS

We set region size to 500m∗500m with 120 nodes (one sinknode and 119 route notes) randomly distributed throughout.The network topology was formed based on the most commonsink topology (radial distribution around the network). A listof specific simulation environment parameters is provided inTable 1.

Fig.10 displays where nodes in the network vary over time.Original RPL and E2HRC routing protocol energy consump-tion is also varied. The data acquisition start time duringsimulation was open for all nodes, and the end time for eachnon-sink node was 1 min to successfully send 10 packets.The x-axis in the figure is number of different nodes in thenetwork and the y-axis is average energy consumption of thenetwork, as defined by Eq.(2), which yields average networkpower consumption Paverage. Psum is the total network energyconsumption, and Cnode is the number of network nodes. Weconducted replicate experiments to ensure accurate simulationresults. The lower end of each nodes data in the y-axisdirection is the minimum value among multiple simulationruns.

Paverage =PsumCnode

(7)

Energy balance is very important for the wireless sensorwhen this network is working. The experimental results for

Fig. 10. Comparison of average network energy consumption

Fig. 11. Comparison of average energy consumption of nodes

energy balance tests are shown in Fig. 11, where the x-axismarks nodes that participated in the experiment and the y-axisis the average energy consumption of nodes.

As shown in Fig.11, the fluctuating range of the nodeenergy consumption curve in the wireless sensor networkwhen original RPL was used was higher than that when theE2HRC routing protocol proposed in this paper was used.Per our analysis, some energy-balanced steps followed theE2HRC routing protocol such as by establishing ring domaincommunication, calculating cluster head ratio, and determiningthe probability factor. Accordingly, cluster head election andthe probability factor adjustment mechanism were adopted toguarantee uniformity in the cluster head distribution in therings. The energy consumption of wireless sensor network waseffectively balanced and the lifespan was extended.

Packet loss ratio is an important index to measure networkperformance. If the wireless sensor networks packet loss ratiois too high, the transmission performance of the network willbe too low to use. As shown in Figs.12 and 13, the bandwidthis wider in E2HRC compared to RPL resulting in a higherpacket loss ratio in certain wireless sensor networks and inthe case of the same number of nodes.

Further analysis showed that as the number of nodes in-creases in the wireless sensor network where the originalRPL is used, packet loss ratio gradually increases as well.However, the packet loss ratio gradually decreases as number




Fig. 12. Packet loss ratio of original RPL

Fig. 13. Packet loss ratio of E2HRC routing protocol

of nodes increases in wireless sensor networks where theE2HRC routing protocol is used. Further, the greater thenumber of nodes, the better the wireless sensor network withthe E2HRC routing protocol compared to the original RPL interms of packet loss ratio. These results were obtained beforethe best father node was selected, and considering optimaldirection angle and residual energy. In effect, the proposedmethod could effectively reduce signal attenuation for long-distance communication.

Control packets are very important for topology establish-ment and maintenance as well as wireless sensor network routerepair. Data packets are not forwarded to their destinationnodes if control packets are more congested than the wire-less sensor network. Additionally, excessive control packetsconsume more node energy when they are sent or received,truncating the lifespan of the wireless sensor network. Thecontrol packet monitoring data numbers in the wireless sensornetwork from 1:00 to 20:00 is shown in Fig.14. Compared tothe original RPL, the E2HRC routing protocol effectively de-creased the number of control packets as time progressed. Theproposed clustering algorithm and cluster rotation mechanismalso achieved distributed cluster rotation to prevent an energyhole, and effectively decreased the number of control packetsduring cluster rotation, thus balancing the network load andimproving network performance.

Fig. 14. Control packet ratio of RPL and E2HRC routing protocol

VII. CONCLUSION

We established a heterogeneous ring communication topol-ogy, proposed a related clustering algorithm for this topology,and built the E2HRC routing protocol to improve originalRPL performance in this study. The proposed method yieldsbetter average energy consumption and overall performancethan RPL while balancing the energy consumption of thewhole wireless sensor network. We also designed a messagingstructure for clustering and routing and verified that bothprotocols are efficient and effective.

REFERENCES

[1] Guangjie Han, Jinfang Jiang, et al. Green Routing Protocols for WirelessMultimedia Sensor Networks. IEEE Wireless Communications Magazine.2016, DOI: 10.1109/MWC.2016.1400052WC.

[2] Hayes T, Ali F H. Location aware sensor routing protocol for mobilewireless sensor networks. Iet Wireless Sensor Systems, vol. 6, no. 2, pp.49-57, 2016.

[3] Thubert P, Winter T, Brandt A, et al. RPL: IPv6 Routing Protocolfor Low power and Lossy Networks, Internet Requests for Comment, rfc6550, pp. 853-861, 2012.

[4] Tripathi J, De Oliveira J C, Vasseur J P. A performance evaluationstudy of RPL: Routing Protocol for Low power and Lossy Networks,In Proceedings of Information Sciences and Systems. IEEE, pages 1-6,2010.

[5] Hakeem S A A, Barakat T M, Seoud R A A. New Real EvaluationStudy of Rpl Routing Protocol Based on Cortex M3 Nodes of Iot-LabTest Bed, Middle-East Journal of Scientific Research, vol. 23, no. 8, pp.1639-1651, 2015.

[6] Pannuto P, Lee Y, Foo Z, et al. M3: a mm-scale wireless energyharvesting sensor platform, International Workshop on Energy NeutralSensing Systems, pages 1-2, 2013.

[7] Shujun L. Performance Evaluating of RPL Routing Protocol under RealEnvironment, Journal of Changchun University of Science & Technology,vol. 37, no. 2, pp. 151-154, 2014.

[8] Gaddour O, Koubaa A. RPL in a nutshell: A survey, ComputerNetworks, vol. 56, no. 14, pp.3163-3178, 2012.

[9] Iova O, Theoleyre F, Noel T. Stability and efficiency of RPL underrealistic conditions in wireless sensor networks // Personal Indoor andMobile Radio Communications (PIMRC), 2013 IEEE 24th InternationalSymposium on. IEEE, pages, 2098-2102, 2013.

[10] Guangjie Han, Li Liu, Jinfang Jiang, Lei Shu, Gerhard Hancke. Anal-ysis of Energy-Efficient Connected Target Coverage Algorithms forIndustrial Wireless Sensor Networks, IEEE Transactions on IndustrialInformatics, 2015. DOI: 10.1109/TII.2015.2513767.

[11] Tekkalmaz M, Korpeoglu I. Distributed power-source-aware routingin wireless sensor networks, Wireless Networks, vol.22, no.4, 1381-1399,2016.




[12] Guangjie Han, Jiawei Shen, Li Liu, Lei Shu. BRTCO: A NovelBorder Line Recognition and Tracking Algorithm for Continuous Ob-jects in Wireless Sensor Networks, IEEE Systems Journal, 2016, DOI:10.1109/JSYST.2016.2593949.

[13] Li S, Zeng W, Zhou D, et al. Compact Conformal Map for GreedyRouting in Wireless Mobile Sensor Networks, IEEE INFOCOM, pp. 12,no. 11, 1632-1646, 2016.

[14] Guangjie Han, Yuhui Dong, Hui Guo, Lei Shu, Dapang Wu. Cross-layer Optimized Routing in Wireless Sensor Networks with Duty-cycleand Energy Harvesting, Wireless Communications and Mobile Comput-ing, Vol.15, No.16, pp. 1957-1981, 2015.

[15] Ahvar E, Lee G M, Crespi N, et al. RER: a real time energy efficientrouting protocol for query-based applications in wireless sensor networks,Telecommunication Systems, vol.61, no.1, pp. 1-15, 2016.

[16] Guangjie Han, Aihua Qian, et al. A Grid-Based Joint Routingand Charging Algorithm for Industrial Wireless Rechargeable SensorNetworks, Computer Networks, Vol.101, No.6, pp. 19-28, 2016.

[17] O. Gaddour, A.Kouba, RPL in a nutshell: A survey , 56th ComputerNetworks, 2012.

[18] H. S. Kim, H. Cho, et al., MarketNet: An Asymmetric TransmissionPower-based Wireless System for Managing e-Price Tags in Markets, inProceedings of the 13th ACM International Conference on EmbeddedNetworked Sensor Systems (SenSys’15), pp. 281-294, Nov. 2015.

[19] J. Ko, J. Jeong, et al., DualMOPRPL: Supporting Multiple Modesof Downward Routing in a Single RPL Network, ACM Transactions onSensor Networks, vol. 11, no. 2, pp. 39:1-39:20, Mar. 2015.

[20] Hyung-sin Kim, Hongchan Kim,et al., Load Balancing underHeavy Traffic in RPL Routing Protocol for Low Power and LossyNetworks, IEEE TRANSACTIONS MOBILE COMPUTING, 2016, DOI10.1109/TMC.2016.2585107.

[21] Gaddour O, Koubaa A, Rangarajan R, et al. Co-RPL: RPL routingfor mobile low power wireless sensor networks using Corona mecha-nism[C]//Industrial Embedded Systems (SIES), 2014 9th IEEE Interna-tional Symposium on. IEEE, pages 200-209, 2014.

[22] Zhang Zongjie, Liu Tong, Ma Xiaoyua et al. Wireless Sensor NetworkRPL Routing Protocol Optimization, Highlights of Sciencepaper Online,vol.7, no.8, 715-721, 2014.

[23] Iova O, Theoleyre F, Noel T. Using multiparent routing in RPL toincrease the stability and the lifetime of the network, Ad Hoc Networks,no.45-62, 2015.

[24] Yu J, Qi Y, Wang G, et al. A cluster-based routing protocol for wirelesssensor networks with nonuniform node distribution, AEU-InternationalJournal of Electronics and Communications, vol.66, no.1, 54-61, 2012.

[25] Yan X F, Zhang Y K, Li TWang X X. A Clustering TopologyAlgorithm Based on Uneven Gradient in WSN, Journal of ZhengzhouUniversity (Engineering Science), vol.35, no.6, 47-51, 2014.

[26] Liu F A, Zhang C H, Nan W U. Adaptive hierarchical routingalgorithm for WSN based on energy and distance, Application Researchof Computers, vol.31, no.11, pp.3434-3437, 2014.

[27] Liu Z, Qiu Z D. Ring based multi-hop clustering routing algorithmfor wireless sensor networks, Journal on Communications, vol.29, no.3,104-113, 2008.

[28] Krishnan P, Krishnakumar A S, Ju W H, et al. A system for LEASE:Location estimation assisted by stationary emitters for indoor RF wirelessnetworks, Twenty-third AnnualJoint Conference of the IEEE Computerand Communications Societies. IEEE, pages 1001-1011, 2004.

[29] Heinzelman W R, Chandrakasan A P, Balakrishnan H. et al. AnApplication-Specific Protocol Architecture for Wireless Microsensor Net-works, IEEE Trans. on Wireless Communication, 2002, 1(4):660-670.

Documents

E2HRC: An Energy-Efﬁcient Heterogeneous Ring Clustering … 2017 - ns2 -base... · · 2017-11-15(IPv6 Routing Protocol for Low Power and Lossy Networks). A new clustering algorithm