Geographic Routing in Clustered Wireless Sensor ... - Semantic Scholar

1

Geographic Routing in Clustered Wireless Sensor Networks among Obstacles Hari Prabhat Gupta∗ ,

S. V. Rao† ,

Abstract—An important issue of research in wireless sensor networks (WSNs) is to dynamically organize the sensors into a wireless network and route the sensory data from sensors to a sink. Clustering in WSNs is an effective technique for prolonging the network lifetime. In most of the traditional routing in clustered WSNs assumes that there is no obstacle in a field of interest. Although it is not a realistic assumption, it eliminates the effects of obstacles in routing the sensory data. In this paper, we first propose a clustering technique in WSNs named Energyefficient Homogeneous Clustering that periodically selects cluster heads according to a hybrid of their residual energy and a secondary parameter, such as the utility of the sensor to its neighbors. In this way, the selected cluster heads have equal number of neighbors and residual energy. We then present a route optimization technique in clustered WSNs among obstacles using Dijkstra’s shortest path algorithm. We demonstrate that our work reduces the average hop count, packet delay, and energy-consumption of WSNs. Index Terms—Clustering, energy-efficient, obstacles, routing.

I. I NTRODUCTION A typical wireless sensor network (WSN) consists of several tiny and low-power sensors which use radio frequencies to perform distributed sensing tasks. WSNs find their applications in many areas that include plant monitoring, battlefield surveillance, fire detection, and leakage of toxic chemicals, radiations, and gas detection [1]–[5]. In such WSNs, a large number of sensors are deployed in a field of interest (FoI) in stochastic manner. In stochastic deployment, sensors are usually dropped randomly in large numbers to guarantee reliability [1], [4], [6], [7]. Minimising the energy consumed while ensuring the connectivity of a network is an important issue to be addressed in WSNs because the batteries powering the sensors may not be accessible for recharging often. Cluster-based routing in WSNs has been investigated by researchers to achieve the network scalability and management, which maximizes the lifetime of the network by using local collaboration among sensors [2]– [5], [8]–[14]. In a clustered WSN, every cluster has a cluster head (CH). CHs periodically collect, aggregate, and forward data to the sink. In any application of WSNs, connectivity is considered to be an important metric to measure the quality of service of WSNs. A network is said to be connected if all sensors in the The authors are with the Department of Computer Science and Engineering, Indian Institute of Technology Guwahati, India (e-mail: ∗ {hprabhatgupta, † [email protected]). amit.yadav0788, dutta.tanima}@gmail.com Copyright (c) 2013 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending a request to [email protected].

Amit Kumar Yadav∗ ,

Tanima Dutta∗

FoI can reach to the sink. Geographic routing [15]–[23] has been considered as an attractive approach in large scale WSNs because it does not require the global topology of a WSN. A sensor can make routing decisions based on the geographic position of itself and its neighbors. The sensor forwards the sensory data to a neighbor, which is closest to the sink. This reduces the average hop count. However, geographic routing cannot optimize the number of hops when a sensor has no neighbor closer to the sink. This problem is known as local minimum problem in the literature [15]. The occurrence of the problem can be caused by many factors, such as sparse deployment of sensors, physical obstacles, and sensor failures. • Major contributions: In this paper, we propose an energyefficient homogeneous clustering technique in WSNs and a route optimization technique in clustered WSNs among obstacles. The major contributions of our work in this area are as follows: 1) We propose an Energy-efficient Homogeneous Clustering (EHC) technique in WSNs, that selects the CHs to create a connected backbone network. EHC is a distributed technique, where sensors make local decisions on whether to join a backbone network as a CH or to a member of a cluster. The decision of each sensor is based on their residual energy and an estimate of how many of its neighboring CHs will benefit from it being a CH. We give a distributed technique where CHs rotate with time, demonstrating how localized sensor decisions lead to a homogeneous connected global topology. 2) We propose a Route Optimization Technique (ROT) in clustered WSNs among obstacles. ROT forms an energy-efficient path between the CHs selected by EHC technique and the sink. ROT uses Dijkstra’s shortest path algorithm [24]. What attracts us is that we do not change the underlying forwarding strategy of existing geographic routing [16]. ROT works under the routing layer and above the MAC and physical layers in WSNs. 3) We analysis message and time complexities of our work which are nearly optimal. We derive an expression to estimate the energy consumption of the network considering EHC and ROT techniques. The rest of the paper is organised as follows: In the next section, we briefly discuss the literature to address the clustering and the local minimum problem in WSNs. We propose EHC and ROT techniques in Section II. The complexity analysis and the energy consumption calculation of EHC and ROT are presented in Section III. In Section IV, we present the simulation results conducted to evaluate the preference of EHC

2

and ROT. We conclude the paper in Section V. • Clustering in WSNs: The limited battery power and the difficulty in recharging the batteries in a hostile environment require that the sensors be deployed with a high density for a long lifetime of WSNs. Distributed clustering techniques are more useful in WSNs. Low Energy Adaptive Clustering Hierarchy (LEACH) [11] selects CHs based on a predetermined probability in order to rotate the CH role among the sensors to balance of the residual energy of the sensors. Following the idea of LEACH, a number of protocols have been presented in the literature [9], [10], [25]. Hybrid Energy-Efficient Distributed (HEED) [13] clustering selects the CHs based on the residual energy of sensors and a secondary parameter, such as proximity to its neighbors. SPAN selects CHs based on the residual energy and number of neighbors [14]. The CHs form a connected network that is used to forward the data. An Energy Efficient Clustering Scheme [26] allocates a fewer number of sensors to clusters with longer distances to the sink. A Fuzzy-logic based clustering approach is proposed in [2]. • Local minimum problem: Several geographical routing protocols have been proposed in recent years to address the local minimum problem in WSNs. Most of the existing solutions for the local minimum problem use perimeter routing technique (PRT) [16], [19], [20], [27]–[29]. By the PRT, when greedy forwarding fails at a local minimum, i.e., no neighbors closer to the sink, packets tend to be routed along the hole boundaries. The Greedy Perimeter Stateless Routing (GPSR) uses greedy forwarding and switches to perimeter routing mode when a local minimum problem is reached [16]. The right-hand rule is used in the perimeter routing mode, where packets are forwarded along the edge counterclockwise on the face of a planar graph. Jia et al. [19] presented the idea of Hole Avoiding In advance Routing protocol (HAIR) to bypass holes in advance. In WSNs, packets are typically routed from sensors to a sink. When a sensor recognizes itself as a local minimum, it asks its neighbor sensors to mark itself as a hole sensor. Data packets are sent to non-hole sensors when possible. Li et al. [20] proposed a new geographic Hole Bypassing Forwarding (HBF) protocol to address the hole diffusion problem in WSNs. The HBF protocol models a hole using a virtual circle whose radius is adjustable within a certain range and is calculated on a per-packet basis. The information associated with the virtual circle is used for selecting an anchor point to bypass the hole in order for a packet to reach a particular sink sensor. Nguyen et al. [29] presented a novel routing protocol named Greedy Forwarding with Virtual Destination (GFVD) strategy. The basic idea is that during the transmission of a packet, a new destination called virtual destination is put in place when the packet is forwarded to a stuck sensor. The abstracted holes protocol in [15] uses a distributed convex hull algorithm to achieve a constant path stretch with lower communication and storage overhead. Motivation: The work in this paper is motivated by the following limitations observed in the literature. The work [2], [9]–[11], [25], [26] requires all CHs to perform direct transmissions to the sink, thus it suffers from the cost of longdistance transmissions. As a result, the sensors that are far away from the sink draining their energy much earlier than

others. The work [10], [13], [14] requires multiple rounds to form a clustered WSN. The multiple rounds introduce communication and processing overhead and taxing the energy as a result. Moreover, HEED [13] has a worst case processing time complexity of O(N ) per sensor, where N is the number of sensors in the WSN. In [14], a sensor associates with a CH in a step. A cluster with a higher CH degree may become highly loaded. Another drawback of existing clustering techniques [5], [12], [13] is that they require more than one transmission power levels for routing the data. Such techniques are not suitable for low-cost sensors which have usually single power level. The work [16], [19], [20], [29] makes a detour path along the perimeter of the hole or obstacles. A new detour is required again and again, which causes partial breakdown of the WSN. In [16] due to the right-hand rule, GPSR does not achieve shorter routing paths. Similarly, GFVD [29] does not construct a shorter routing path and may burden the energy consumption of sensors. The anchor may be far from the hole in [20], and thus detouring the hole in a longer path. In summary, there is no work in the literature on routing in clustered WSNs among obstacles. The existing work focuses on individual aspects, i.e., on clustering, routing, or routing amount obstacles in WSNs. Considering these limitations in the literature, we propose an EHC technique in WSNs that periodically selects CHs according to their residual energy and the utility of the sensor to its neighbors. The main difference between existing clustering techniques and EHC technique is the utility of the CHs in WSNs. In the EHC technique, a sensor becomes a CH if the utility of the sensor is higher than its neighbors. Different from the existing work, a CH in EHC technique has maximum eleven neighboring CHs and does not make any assumptions about the density of sensors. The worst case processing time and space complexities of EHC technique is O(1) per sensor. We present a route optimization technique in clustered WSNs among obstacles using Dijkstra’s shortest path algorithm. II. ROUTING IN A C LUSTERED WSN Network model: A network consists of N sensors, deployed at random uniformly in a FoI Ψ among obstacles. The sensors are stationary and powered by the batteries. We assume the binary disc communication model in which a sensor, denoted by s, can communicate with other sensors within the disc of radius C centered at s, denoted by A(s, C), where kA(s, C)k = πC 2 . Thus, C denotes the communication range of s. Two sensors i and j can communicate with each other directly and are known as neighbors if the Euclidean distance between them is not more than C. The number of neighboring CHs of a CH is said to be the CH degree. In this paper, the lifetime of WSNs is the time from the start of the network operation to the death of the first sensor in the network. The lifetime of WSNs is divided into rounds to balance the energy consumption among sensors. Each round consists of two phases: decision phase and working phase. At the beginning of a round, all sensors participate in the decision phase to form a clustered WSN using the EHC technique. In the working phase, the sensory data from the sensors in

3

a cluster are transmitted directly to their CH which then aggregates and forwards data to other CHs, which en-route to the sink using ROT.

1

A. Energy-efficient Homogeneous Clustering (EHC) In this section, we first propose EHC technique and then describe its properties. 1) EHC Description: EHC works in the following two steps to form a clustered WSN: • Initial cluster head election: The goal of this step is to elect the CHs in a distributed manner. Let P be the probability that the expected number of CH-candidates for a round is k of N sensors in Ψ. The probability that there are at least one CH-candidate in the region A(i, C) is 1 − e−kkA(i,C)k/kΨk with high probability [30]. The probability P is therefore given by kkA(i,C)k k − kΨk , where 1 − e ≥ 0.99 N 1.46 kΨk . ≥ N C2

P=

(1)

Einit − Ei + R × T, Einit

2 3 4 5 6 7 8 9 10

Input: i, N , C, Einit , Ei , T , kΨk ; Output: ICH, p, ni ; Set ni ← {φ}; /* initially i has zero degree */ Set nmin ← 100; /* set nmin with a large number */ Set p ← {φ}; /* set p with null value */ Pick random numbers R and R1 in (0, 1); if P ≤ R1 and expire of delay1i and ni = φ then Broadcast a Cadve (i, Ei , ni ) message; /* i is a ICH */ if receive Cadve (j, Ej , nj ) message from ICH j then Add ICH j in neighbor list ni ; if |nj | ≤ nmin then nmin ← nj and p ← j;

We estimate the randomized back-off delay to resolve the advertisement contention for selecting GCHs. The randomized back-off delay for a NCH i is denoted by delay2i =

The probability P is either stored in each sensor off-line or can be sent by the base station initially at the time of deployment. At the beginning of each round, sensor i picks a random number in (0, 1). If the random number is less than P, then sensor i is a CH-candidate. With this mechanism, approximately k of N sensors are elected as CH-candidates. The random number does not depend on the previous round. Note that if a sensor i elects to become a CH-candidate, i broadcasts an advertisement message CHadve (i, Ei , ni ) to inform other sensors of its availability, where Ei and ni are the residual energy and the list of neighboring CHs of i, respectively. Advertisement contention occurs when multiple CH-candidates advertise at the same time. To resolve the contention, we use a randomized back-off delay. The randomized back-off delay for a CH-candidate i is denoted by delay1i =

Procedure 1: Initial cluster head election /*Run at a sensor i in the beginning of a round */

(2)

where Einit , R, and T are the initial energy of sensors, a random number in (0, 1), and the round-trip delay for a small control packet, respectively. The randomized back-off delay assures that a CH-candidate with higher residual energy among its neighbors will have higher probability to become a CH. The choice of delay1i is a reasonable method for evenly consumption of energy of the sensors while preventing additional overhead. The CHs elected in this step are denoted by Initial CHs (ICHs). The pseudocode for initial cluster head election is given by Procedure 1. • Connected network formation: We elect more CHs to ensure that the CHs can form a connected network, since ICHs are not connected. A Non-Cluster Head sensor (NCH) is elected as a CH, denoted by Gateway CH (GCH), if two or more neighboring ICHs are not connected. In the rest of the paper, a CH represents either a GCH or ICH. Preference is given to the NCHs which have higher amount of residual energy and maximum number of neighboring CHs. Lemma 1 shows that a NCH has maximum five neighboring ICHs.

5 − |ni | Einit − Ei + + R × |ni |T. Einit 5

(3)

Thus, a NCH with the higher residual energy and a number of neighboring ICHs will be elected as a GCH with high probability among neighboring GCH-candidates. If a NCH i wants to associate with a CH j, where j has a minimum CH degree in ni , i sends an associate message (denoted by Casso (j, i, ni )) to j and receives a subsequent confirmation message (denoted by Cconf (i, j, nj )) from j. A NCH updates information of neighbor i whenever it receives Cconf (j, i, ni )) message from i. The pseudocode of this step is given by Procedure 2. 2) Properties of EHC technique: First property of a clustered WSN is that all sensors are clustered. Line 10 in Procedure 2 illustrates that an isolated sensor will become a CH. Therefore, each sensor in WSN is either a CH or a member of a cluster. All CHs are connected is the second property of connected WSNs. Line 14 in Procedure 2 shows that if a sensor has two or more neighboring ICHs, which are not connected, the sensor will become a GCH. Third property in a connected clustered WSN is that each NCH has exact one CH. Line 17 in Procedure 2 describes that each NCH allows to associate with only one CH. Lemma 1. Maximum degree of a CH in EHC is eleven. Proof: Procedure 1 shows that ICHs are not directly connected with each other. The degree of a ICH is therefore the number of neighboring GCHs. Consider a GCH M as shown in Fig. 1(a). We assume the ICHs Q and R are not connected. The Euclidean distances between M and Q, M and R, and Q and R are denoted by c, b, and a, respectively. If M is connected with Q and R, then 0 ≤ {b, c} ≤ C < a. To estimate the maximum number of ICHs surrounding M , we take the minimum value of the angle ∠RM Q =cos−1 (b2 + c2 − a2 /2bc). Substituting a, b, and c, the angle ∠RM Q > π/3, i.e., the maximum number of neighboring GCHs of a ICH is five. The maximum degree of a GCH is the number of neighboring CHs. Fig. 1(b) shows a scenario where ICHs P , Q, and R

4

a FoI [1], [4], [6], [7], [31]. In this section, we propose ROT in clustered WSNs that optimizes the path length during data transmission without any extra overhead. We consider m obstacles in Ψ, where m ≥ 0. Each sensor S knows about its location. Let view-vertices V = m i=1 Vi , Sn where Vi = j=1 vij is a set of view-vertices of an obstacle i, 1 ≤ i ≤ m and n > 0 [32]. The view-vertices of all the obstacles are stored in each sensor initially at the time of deployment or can be updated by the sink. Fig. 2 shows the view-vertices of two obstacles. 1) ROT description: In the early phase of ROT, a backbone network is constructed using the proposed EHC, where a sensor is a CH or a member of a cluster. Consider a source CH i and a sink t as shown in Fig. 2. Before i sends data to the sink t, it identifies the obstacles between t and itself. If there is no obstacle, i forwards data to t using geographic forwarding (GF) [16]. Otherwise i finds a shorter path (SP) → − to t, denoted by it , through the view-vertices of obstacles using Dijkstras shortest path (DSP) [24] algorithm. i sets the view-vertices along SP as the intermediate destinations (IDs). When data reach the nearest CH of a ID, denoted by j, ROT reruns between j and t to find a new SP. The pseudocode of ROT is described in Procedure 3.

Procedure 2: Connected network formation /*Run at a sensor i (NCH or CH) after Procedure 1*/

1 2 3

Input: i, p, ni , N , C, Einit , Ei ; Output: GCH; if i is a CH then if receive Casso (i, j, nj ) message from a NCH j then Send Cconf (j, i, ni ) message to j; if receive Cadve (j, Ej , nj ) message from a GCH j then Add j in neighbor list ni ; Send Cconf (j, i, ni ) message to j;

4 5 6 7

else

17

if expire of delay2i then if |ni | = 0 then Broadcast Cadve (i, Ei , ni ); /* i is a GCH */ else if |ni | = 1 then Send Casso (j, i, ni ) to a CH j ∈ ni ; else if CHs in ni are not connected then Broadcast Cadve (i, Ei , ni ); /* i is a GCH */ else Sends Casso (p, i, ni ) message to p, p ∈ ni ;

18

else

8 9 10 11 12 13 14 15 16

19 20 21 22

if receive Cadve (j, Ej , nj ) or Cconf (l, j, nj ) then Add CH j in neighbor list ni ; if |nj | ≤ nmin then p ← j and nmin ← nj ;

Procedure 3: Route Optimization Technique /*Run at a sensor i */ are not connected. We assume the Euclidean distances between P and Q, P and R, and Q and R are denoted by c, b, and a, respectively. If GCHs M and N are connected with P , Q and P , R, respectively, then 0 ≤ {b, c} ≤ 2C < a. To estimate the maximum number of GCHs surrounding P , we take the minimum value of the angle ∠RP Q = cos−1 (b2 + c2 − a2 /2bc). For 0 ≤ {b, c} ≤ 2C < a, the angle ∠RP Q > π/6, i.e. The maximum number of neighboring CHs of a GCHs is therefore eleven. The maximum degree of a CH is therefore eleven.

1 2 3 4 5 6 7 8 9 10 11

Input: clustered WSN, vij , V ; Output: SP ; if i is a source then Set |ID| ← {φ} and Send data to the associate CH; else if i is the nearest CH to ID or |ID| = φ then − → if there exist obstacles for it then Apply DSP algorithm on the vertices V ; Set view-vertices along the SP as IDs; Send data to the next ID using GF ; else Send data to t using GF ; else

Send data to t using GF ;

R R C

N

2C M

Q

C P

M

v14

Q

i (a)

v24

v13

v23

v25

v15

(b)

Fig. 1: Illustration of maximum degree of a CH. The ICHs and GCHs are denoted by “•” and “◦”, respectively. Part(a) illustrates an ICH M can communicate with non-connected GCHs Q and R. Part(b) shows the ICHs M and N can communicate with non-connected GCHs.

B. Route Optimization Technique (ROT) The goal of a route optimization technique is to achieve a path from the source to the sink but we also want to achieve the goal at a minimum cost, i.e. shortest path in terms of hop counts among obstacles. Most of the literature on routing in WSNs does not have any special treatment for the obstacles in

v11

v12

v21

t

v22

Fig. 2: Illustration of ROT. View-vertices of two obstacles are v1p and v2q , where 1 ≤ {p, q} ≤ 5. CHs are denoted by “◦”. Line (− − −) shows the shorter path.

2) Property of ROT: Due to the obstacles in the FoI, the path generated in the literature [15], [16], [20] can deviate far from the shortest path. Fig 3 illustrates an example of the path formation in ROT and GPSR. The Euclidean distance between source i and sink t is d + b. A b × l rectangularshaped obstacle separates i and t such that t is behind the obstacle. The length of the path formed by ROT and GPSR

5

p are (l/2)2 + d2 + (l/2 + b) and d + l + b, respectively. If d = l and d >> b, the path shrink in ROT is given by 2d −

q (d2 + d2 )2 + d2

(4)

× 100 = 19%.

2d

The path shrink reduces the energy consumption during routing the sensory data and therefore prolongs the lifetime of WSNs, stability, and delay. d

b

t

s

l

GPSR Geographical routing with ROT

assumed that the view-vertices are stored in each sensor offline, no additional message overhead is required for ROT. Total message complexity is therefore O(N ). Message complexity is O(1) per sensor and is near optimal. B. Time complexity The delay in Procedure 1 is the sum of the delay1i and T (last advertisement message). Similarly, the delay in Procedure 2 is the sum of delay2i and T . The total time complexity of EHC is therefore O(ni T ). Since, N ni , the time complexity of the EHC is O(1) per sensor. Assuming that there are p viewvertices and the DSP algorithm does not use any min-priority queue. The time complexity of ROT is O(|IDs|p2 ). Total time complexity is therefore the sum of the time complexity of EHC and ROT, i.e., O(|IDs|p2 ).

Fig. 3: Illustration of geographic routing with ROT and GPSR.

C. Energy consumption calculation We conduct the following experiments with ns 2.34 simulator to illustrate the motivation of ROT. The description of the simulator is given in section IV. An obstacle of size 300m × 300m is located at (500, 500) as shown in Fig. 4(a). The average path length using GPSR on EHC is 21 hops and GPSR on ROT and EHC is 17 hops. The path shrink is therefore 19.04%. Similarly, the path shrink is 21% when two obstacles are located in the FoI as shown in Fig. 4(b). Backbone network

ET x =

8

6

l(el + ef d2 ) l(el + em d4 ) =ler ,

Y Coordinate (×102 m)

8

ERx

if d < dth if d ≥ dth (5)

6

4

4

2 0

Backbone network

10

Y Coordinate (×102 m)

10

The energy consumption per round is the sum of energy consumed per round in EHC and ROT, given by Eround = EEHC + EROT . In terms of energy consumption, transmitting a single bit of data is equivalent to execution of 800−1000 instructions [33]. We therefore consider the energy consumption as the energy dissipated in transmitting and receiving packets. • Energy consumption model: We consider the following energy consumption model [34]:

2

0

2

4

6

8

X Coordinate (×102 m)

(a) One obstacle in the FoI.

10

0

0

2

4

6

8

10

X Coordinate (×102 m)

(b) Two obstacles in the FoI.

Fig. 4: Illustration of a clustered WSN among obstacles. The sensors marked “”, “•” and “N” are ICHs, CHs, and NCHs, respectively.

III. C OMPLEXITY A NALYSIS AND E NERGY C ONSUMPTION C ALCULATION OF EHC AND ROT In this section, we present the analysis of message complexity, time complexity, and estimate the energy consumption per round of EHC and ROT. A. Message complexity In a WSN of N sensors, let k1 , k2 , N − k1 − k2 are the number of ICHs, GCHs, and NCHs, respectively. The CHs broadcast advertisement messages to inform others of their availability. A NCH sends an association message to a CH and receives a subsequent conformation message from the CH. Therefore, the proposed EHC requires an additional cost of 2(N − k1 − k2 ) in Procedure 2. The total message complexity of EHC is given by 2N −k1 −k2 =O(N ). Since, we

where ET x is the energy dissipated to transmit l bits over a distance d and dth is a distance threshold. ERx is the energy dissipated to receive l bits, el =er =50nJ/bit, ef = 10pJ/bit/m2 , and em = 0.0013pJ/bit/m4 . The energy cost in sleep mode is very low, that we set equal to 1/1000 of the energy spent for receiving. The energy consumption when transmitting, receiving, and in sleep mode follows the first order energy model [17]. 1) Energy consumption in EHC: Let the length of a control message is l bits and Ψ is the FoI where sensors are deployed. The total energy consumption during a round is given by EEHC =

1.46 kΨk l πC 2 N 2 e + e C + − 1 er + l f C2 kΨk 1.46 kΨk l πC 2 N 2 e + e C + − 1 e + r l f C2 kΨk 5 × 1.46 kΨk l πC 2 N 2 e + e C + − 1 e + r l f C2 kΨk πC 2 N N l el + ef C 2 + − 1 er . (6) kΨk

Eq. 6 can be briefly explained as follows: k of N sensors are ICH-candidates in Procedure 1. The ICH-candidates compete with each other to become a CH and announce their candidacy within their communication region. The ICH-candidates announcements are received by the peer sensors, which is approximately πC 2 kN/ kΨk-1 sensors per ICH-candidate. The

6

first term in Eq. 6 is for ICH-candidates to announce their candidacy and the reception of these announcements by peer sensors. Let k2 be the number of GCHs. Lemma 1 shows that a GCH has maximum five ICHs. GCHs announcements are received by the NCHs and ICHs. The neighboring CHs reply back with a confirmation message. The second and third terms stand for these events. Finally, each NCH sends a message for association to a CH and the CH replies back with a confirmation message. The last term stands for these events. Since N 7 ∗ 1.46 kΨk/C 2 ,the energy consumption ofa sensor in a round is given by l el + ef C 2 +

πC 2 N kΨk

− 1 er .

2) Energy consumption in ROT: Let d be the length of the SP using the proposed ROT. The total number of hops in the SP, denoted by h, is given by h ≥ d/C. Let the source generates no number of packets of lo bits in a round. Let N 0 is the number of CHs in the network. Energy consumption during a round in ROT is therefore given by 2 0 ! EROT = (no lo + no lo h) ×

el + ef C 2 +

πC N − 1 er kΨk

. (7)

Equation 7 can be briefly explained as follows: The first term shows that the source sends the sensory data to its CH. All sensors in the communication region of the source receive the sensory data. The second term illustrates routing the sensory data from CH to the sink using h hops. 3) Energy consumption per round: Total energy consumption per round, denoted by Eround , is expressed as Eround =EEHC + EROT , 7 × 1.46 kΨk l = + N l × C2 2 πC N el + ef C 2 + − 1 er + kΨk πC 2 N 0 (no lo + no lo h) × el + ef C 2 + − 1 er . kΨk (8)

IV. P ERFORMANCE E VALUATION In this section, we discuss the results from a simulation study of the proposed EHC and ROT implemented in ns 2.34 simulator [35] using CMU wireless extension [36]. We run our simulation on top of the 802.11.4 MAC layer and most of our simulation parameters are similar to those in [17]. There are 1000 static sensors, randomly deployed (using setdest in ns-2.34) in a 1000m × 1000m square-shaped FoI. The sink is located at the position (990, 990). The initial energy of a sensor is assumed to be 500J and simulation time is 500 seconds. Assuming TwoRayGround propagation model, drop tail priqueue with 50 packets queue length, and the traffic rate λ varies from 0.25 to 6 packets per second. The data packets are all 250B long. The length of control packets is 25B. The channel data rate is 38.4 kbps. Data traffic is generated according to a Poisson process of intensity λ packets per second. We repeat each experiment 50 times with different seeds and reported the average value, each time on a

different connected topology. We use the energy consumption model described in Section III-C. We consider three scenarios with no obstacle, one obstacle, and two obstacles (in Fig. 4). The simulations are considered by varying the communication range, simulation time, and number of obstacles. All the results are with 95% confidence level though the error bars are not visible in the plots. GPSR is a prominent and widely-cited geographic routing protocol. We have implemented ROT on GPSR and ROT on GPSR with EHC. The enhanced versions of GPSR are referred to as ROT-GPSR and EHC-ROT-GPSR. In Section IV-A, we define the average hop count as the average number of hops traveled by any packet to reach the sink. The simulation for analysis of average hop count is conducted by varying the communication range of the sensors. Next, we study the impact of the proposed work on the energy consumption and packet delivery ratio in WSNs, as describe in Sections IV-B and IV-C, respectively. Finally, we study the impact of the obstacles on the lifetime of network, consumed energy per packet, average hop count, and average packet delay in Section IV-D. All results in Section IV-D are compared with GPSR in percentage. A. Average Hop Count Analysis Fig. 5 shows that the average hop count decreases as the communication range increases from 25m to 100m. We compare our work with GPSR [16], GOAL [15], and HBF [20]. It illustrates that for all the protocols, the average hop count increases with obstacles since a geographic routing protocol does not guarantee shortest routing path among obstacles. Fig. 5 shows that ROT-GPSR requires less number of average hop count for the routing the packets. As expected, without ROT the routing is highly affected by the obstacles in the FoI. We noted that the EHC-ROT-GPSR does not achieve a good gain in the average hop count for the routing the packets as ROT-GPSR. This is because in EHC-ROT-GPSR, only CHs participate in routing. B. Energy Consumption in WSNs Fig. 6 shows the energy consumption in WSNs for the entire duration of the simulation for GPSR, HBF, GOAL, ROTGPSR, and EHC-ROT-GPSR. It can be concluded that when EHC is not used, all the sensors remain active to provide the routing for a short duration. Fig. 6 also illustrates that ROTGPSR is consumed less energy than GPSR, HBF, and GOAL. This is because a geographical routing without ROT requires more number of the average hop count for routing the packets. C. Packet Delivery Ratio The packet delivery ratio of a flow is the ratio of the number of packets that are received by the sink over packets submitted to the network by the source. Fig. 7 shows the overall delivery ratio for the entire duration of the simulation for GPSR, HBF, GOAL, ROT-GPSR, and EHC-ROT-GPSR. We conclude that the packet delivery ratio reduces as the average hop count increases. Both ROT-GPSR and EHC-ROT-GPSR achieve a stable performance for the entire duration of the simulation.

60

GPSR HBF GOAL ROT-GPSR EHC-ROT-GPSR

50 40 30 20 10

25

50

75 Communication range (in m)

100

Energy consu. (in KJ).

Average hop count

7

16 14 12 10 8 6 4 2 0

100

70 60 50 40 30 20 10

25

50


100

15 10 5 0

50


100

Average hop count

(c) Two obstacles in the FoI Ψ as shown in Fig. 4(b). Fig. 5: Relationship between the average hop count and the communication range of the sensors.

D. Impact of Obstacles on lifetime, hop count, and delay In this section, we assume 300m × 300m square-shaped obstacles are randomly deployed in the FoI. The impact of the number of obstacles on the average hop count (AHC) is shown in Fig. 8(a). The y-axis in Fig. 8(a) denotes the ratio of the difference of AHC in GPSR and a given protocol to the AHC in GPSR. As the number of obstacles increases, the number of view vertices increases which in turn increases AHC. Fig. 8(a) illustrates that the AHC decreases when the proposed ROT is used. Fig. 8(b) shows the impact of the number of obstacles on the average packet delay in WSNs. Similar to the previous result, the average packet delay is less when ROT is used since ROT requires less AHC for routing the sensory data to the sink. The y-axis in Fig. 9(a) denotes the ratio of the difference of the lifetime of given protocols and GPSR to the lifetime of WSNs in GPSR. We have implemented the Fuzzy-logic based clustering on GPSR and the enhanced version of GPSR are referred to as Fuzzy-GPSR. It shows that when the proposed EHC and ROT are used, only CHs remain active to provide the routing and therefore prolongs the lifetime of WSNs. The impact of the number of obstacles on the consumed energy per packet (CEP) is shown in Fig. 9(b). It illustrates that the CEP is less when the proposed ROT is used. The CEP is significantly reduced when EHC is used with ROT. In EHC, only CHs are in active state and participate in routing the sensory data.

100

200

300 400 Simulation time (in sec).

500

(b) One obstacle in the FoI Ψ as shown in Fig. 4(a). Energy consu. (in KJ).

25

500

20

(b) One obstacle in the FoI Ψ as shown in Fig. 4(a). 70 60 50 40 30 20 10


(a) No obstacle in the FoI Ψ. Energy consu. (in KJ).

Average hop count

(a) No obstacle in the FoI Ψ.

200

25 20 15 10 5 0

100

200


500

(c) Two obstacles in the FoI Ψ as shown in Fig. 4(b). Fig. 6: Demonstration of the energy consumption in WSNs. Communication range of the sensors is 100m.

V. C ONCLUSIONS In this paper, we proposed a distributed approach to determine if a sensor in WSNs is a CH to meet the desired connectivity requirements. We mainly focused on energyefficient clustered WSNs to prolong the lifetime of WSNs. We also proposed a technique to optimize the routing path among obstacles in clustered WSNs. We simulated the performance of the proposed EHC and ROT for different network scenarios and demonstrated that the energy consumption and average hop count in WSNs are reduced due to the clustering of sensors and optimization of routing path, hence the lifetime of WSNs is increased. The results demonstrated that the geometry and location of the obstacles should be considered to compute an optimized routing path. R EFERENCES [1] J. H. Lee, T. Kwon, and J. Song, “Group Connectivity Model for Industrial Wireless Sensor Networks,” IEEE Trans. Ind. Informat., vol. 57, no. 5, pp. 1835–1844, 2010. [2] J.-S. Lee and W.-L. Cheng, “Fuzzy-Logic-Based Clustering Approach for Wireless Sensor Networks Using Energy Predication,” IEEE Sensors Journal, vol. 12, no. 9, pp. 2891–2897, 2012. [3] Z. Ha, J. Wu, J. Zhang, L. Liu, and K. Tian, “A General SelfOrganized Tree-Based Energy-Balance Routing Protocol for Wireless Sensor Network,” IEEE Trans. Nuclear Science, vol. 61, no. 2, pp. 732– 740, 2014. [4] D. C. Hoang, P. Yadav, R. Kumar, and S. Panda, “Real-Time Implementation of a Harmony Search Algorithm-Based Clustering Protocol for Energy-Efficient Wireless Sensor Networks,” IEEE Trans. Ind. Informat., vol. 10, no. 1, pp. 774–783, 2014.

8

More lifetime with GPSR

Packet delivery ratio.

1

EEC-ROT-GPSR HBF GOAL ROT-GPSR Fuzzy-GPSR

80

.975

60 40

.971

20

.965

100

150 200 Communication range (in m).

250

(a) No obstacle in the FoI Ψ.

2

3

4 5 Number of obstacles

80

Less CEP with GPSR


1

(a) More lifetime of WSNs compares with GPSR in %.

1

.975

60

.971

40

.967

20

.964

100


250

(b) One obstacle in the FoI Ψ as shown in Fig. 4(a).

0

1

2


5

(b) Less CEP compares with GPSR in %. Fig. 9: Impact of the number of obstacles on lifetime of WSNs and consumed energy per packet. All results are compared with GPSR in percentage.

1


0

.976 .971 .967 .963 .960

100


250

(c) Two obstacles in the FoI Ψ as shown in Fig. 4(b).

Less AHC with GPSR

Fig. 7: Relationship between the packet delivery ratio and the communication range of the sensors. HBF GOAL ROT-GPSR EHC-ROT-GPSR

25 15 5

1

2


5

Less delay with GPSR

(a) Less average hop count (AHC) compares with GPSR in %. 25 20 15 10 5

1

2


5

(b) Less average packet delay compares with GPSR in %. Fig. 8: Impact of the number of obstacles on average hop count and average packet delay. All results are compared with GPSR in percentage.

[5] M. Tarhani, Y. Kavian, and S. Siavoshi, “SEECH: Scalable Energy Efficient Clustering Hierarchy Protocol in Wireless Sensor Networks,” IEEE Sensors Journal, vol. 14, no. 11, pp. 3944–3954, Nov 2014. [6] P. T. A. Quang and D.-S. Kim, “Enhancing Real-Time Delivery of Gradient Routing for Industrial Wireless Sensor Networks,” IEEE Trans. Ind. Informat., vol. 8, no. 1, pp. 61–68, 2012. [7] J. Niu, L. Cheng, Y. Gu, L. Shu, and S. Das, “R3E: Reliable Reactive Routing Enhancement for Wireless Sensor Networks,” IEEE Trans. Ind. Informat., vol. 10, no. 1, pp. 784–794, 2014.

[8] R. Xie and X. Jia, “Transmission-Efficient Clustering Method for Wireless Sensor Networks Using Compressive Sensing,” IEEE Trans. on Parallel and Distributed Systems, vol. 25, no. 3, pp. 806–815, March 2014. [9] H. Lu, J. Li, and M. Guizani, “Secure and Efficient Data Transmission for Cluster-Based Wireless Sensor Networks,” IEEE Trans. Parallel Distrib. Syst., vol. 25, no. 3, pp. 750–761, 2014. [10] D. Wei, Y. Jin, S. Vural, K. Moessner, and R. Tafazolli, “An EnergyEfficient Clustering Solution for Wireless Sensor Networks,” IEEE Trans. Wireless Communications, vol. 10, no. 11, pp. 3973–3983, 2011. [11] M. Handy, M. Haase, and D. Timmermann, “Low energy adaptive clustering hierarchy with deterministic cluster-head selection,” in MWCN, 2002, pp. 368–372. [12] X. Zhu, L. Shen, and T. Yum, “Hausdorff Clustering and Minimum Energy Routing for Wireless Sensor Networks,” IEEE Trans. Vehicular Technology, vol. 58, no. 2, pp. 990–997, 2009. [13] O. Younis and S. Fahmy, “HEED: A Hybrid, Energy-Efficient, Distributed Clustering Approach for Ad Hoc Sensor Networks,” IEEE Trans. Mobile Computing, vol. 3, no. 4, pp. 366–379, 2004. [14] B. Chen, K. Jamieson, H. Balakrishnan, and R. Morris, “Span: An Energy-Efficient Coordination Algorithm for Topology Maintenance in Ad Hoc Wireless Networks,” Wireless Networks, vol. 8, no. 5, pp. 481– 494, 2002. [15] M. Won, W. Zhang, and R. Stoleru, “GOAL: A parsimonious geographic routing protocol for large scale sensor networks,” Ad Hoc Networks, vol. 11, no. 1, pp. 453–472, 2013. [16] B. Karp and H. T. Kung, “GPSR: greedy perimeter stateless routing for wireless networks,” in MOBICOM, 2000, pp. 243–254. [17] C. Petrioli, M. Nati, P. Casari, M. Zorzi, and S. Basagni, “ALBAR: Load-Balancing Geographic Routing Around Connectivity Holes in Wireless Sensor Networks,” IEEE Trans. Parallel Distrib. Syst., vol. 25, no. 3, pp. 529–539, 2014. [18] Y. Cao, Z. Sun, N. Wang, H. Cruickshank, and N. Ahmad, “A Reliable and Efficient Geographic Routing Scheme for Delay/Disruption Tolerant Networks,” IEEE Wireless Communications Letters, vol. 2, no. 6, pp. 603–606, 2013. [19] T. Wang, W. Jia, G. Wang, M. Guo, and J. Li, “Hole Avoiding in Advance Routing with Hole Recovery Mechanism in Wireless Sensor Networks,” Ad Hoc & Sensor Wireless Networks, vol. 16, no. 1-3, pp. 191–138, 2012. [20] F. Li, B. Zhang, and J. Zheng, “Geographic hole-bypassing forwarding protocol for wireless sensor networks,” IET Communications, vol. 5, no. 6, pp. 737–744, 2011. [21] P. Spachos, P. Chatzimisios, and D. Hatzinakos, “Energy Aware Opportunistic Routing in Wireless Sensor Networks,” in Globecom Workshops (GC Wkshps), 2012, pp. 405–409. [22] K. Zeng, J. Yang, and W. Lou, “On Energy Efficiency of Geographic

9

[23] [24] [25]

[26] [27] [28] [29] [30] [31]

[32] [33] [34] [35] [36]

Opportunistic Routing in Lossy Multihop Wireless Networks,” Wirel. Netw., vol. 18, no. 8, pp. 967–983, Nov 2012. K. Zeng, K. Ren, W. Lou, and P. J. Moran, “Energy aware efficient geographic routing in lossy wireless sensor networks with environmental energy supply,” Wirel. Netw., vol. 15, no. 1, pp. 39–51, Jan. 2009. E. W. Dijkstra, “A note on two problems in connexion with graphs.” Numerische Mathematik, vol. 1, pp. 269–271, 1959. B.-C. Cheng, H.-H. Yeh, and P.-H. Hsu, “Schedulability Analysis for Hard Network Lifetime Wireless Sensor Networks With High Energy First Clustering,” IEEE Trans. Reliability, vol. 60, no. 3, pp. 675–688, 2011. M. Ye, C. Li, G. Chen, and J. Wu, “EECS: An Energy Efficient Clustering Scheme in Wireless Sensor Networks,” in IPCCC, 2005, pp. 535–540. J. You, D. Lieckfeldt, F. Reichenbach, and D. Timmermann, “ContextAware Geographic Routing for Sensor Networks with Routing Holes,” in WCNC, 2009, pp. 2589–2594. S.-F. Hwang, H.-H. Lin, and C.-R. Dow, “An energy-efficient routing protocol in wireless sensor networks with holes,” in ICUFN, 2012, pp. 17–22. N. D. Nguyen, D. T. Nguyen, M.-A. Le Gall, N. Saxena, and H. Choo, “Greedy Forwarding with Virtual Destination Strategy for Geographic Routing in Wireless Sensor Networks,” in ICCSA, 2010, pp. 217–221. C. Bettstetter, “On the minimum node degree and connectivity of a wireless multihop network,” in MobiHoc, 2002, pp. 80–91. D. Zhang, G. Li, K. Zheng, X. Ming, and Z.-H. Pan, “An EnergyBalanced Routing Method Based on Forward-Aware Factor for Wireless Sensor Networks,” IEEE Trans. Ind. Informat., vol. 10, no. 1, pp. 766– 773, 2014. S. K. Ghosh, Visibility Algorithms in the Plane. New York, NY, USA: Cambridge University Press, 2007. J. L. Hill, R. Szewczyk, A. Woo, S. Hollar, D. E. Culler, and K. S. J. Pister, “System Architecture Directions for Networked Sensors,” in ASPLOS, 2000, pp. 93–104. S. D. Muruganathan, D. Ma, R. Bhasin, and A. Fapojuwo, “A centralized energy-efficient routing protocol for wireless sensor networks,” IEEE Communications Magazine, vol. 43, no. 3, pp. S8–S13, 2005. “The Network Simulator: ns-2.34,” 2014, http://www.isi.edu/nsnam/ns. “CMU MONARCH EXTENSIONS to ns,” 2014, http://www.monarch.cs.cmu.edu.

Hari Prabhat Gupta received his Ph.D in Computer Science and Engineering from the Indian Institute of Technology Guwahati in 2014. Since then he is working in Samsung R&D Institute Bangalore, India. He has received a research fellowship from TATA Consultancy Services, India. His research interests include wireless sensor networks, ad hoc networks, and algorithms.

S. V. Rao received his Ph.D in Computer Science and Engineering from the Indian Institute of Technology Kanpur in 1999. Presently, he is a Professor in the Department of Computer Science and Engineering, Indian Institute of Technology Guwahati, India. His research interests include wireless sensor networks, ad hoc networks, data center networks, and computational geometry.

Amit Kumar Yadav received his Masters degree in Computer Science and Engineering from the Indian Institute of Technology Guwahati in 2012. Since then he is working in Dell R&D Institute, India-Bangalore. His research interests include wireless sensor networks, ad hoc networks, and algorithms.

Tanima Dutta received his Ph.D in Computer Science and Engineering from the Indian Institute of Technology Guwahati in 2014. Since then she is working in TCS Innovation Labs Bangalore, India. She has received TCS research fellowship and SAIL fellowship for persuing PhD and B.Tech degrees, respectively. Her research interests include multimedia, digital forensics, wireless sensor networks, and algorithms.