CHEATS: A Cluster-Head Election Algorithm for WSN Using a Takagi-Sugeno Fuzzy System Adonias Pires∗ , Claudio Silva∗ , Eduardo Cerqueira∗ Dionne Monteiro∗ and Raimundo Viegas Jr.∗ ∗ Federal
University of Para (UFPA) Belem, Brazil {adonias, cerqueira, dionne, rviegas}@ufpa.br,
[email protected] Abstract—Energy conservation is very important task for wireless sensor networks (WSN), since external power sources are typically unavailable and the replacement of batteries is clearly impractical for large networks. A classical protocol widely used in WSN, called LEACH (low-energy adaptive clustering hierarchy), is a cluster-based protocol which aims at reducing energy consumption in the network. However, it has some drawbacks. This paper proposes the CHEATS (Cluster-Head Election Algorithm Using a Takagi Sugeno Fuzzy System) protocol, an improvement on the LEACH protocol whose goal is to increase energy effici ency in terms of network lifetime using fuzzy logic through a Takagi-Sugeno system to calculate the probability of election of candidates for cluster-head. Simulation results have shown that the protocol proposed could prolong the lifetime and increase the throughput of the network.
I. INTRODUCTION Nowadays, it is well known that the advances in radio technology have achieved a low power consumption level that allowed the creation of very small battery-powered communication devices. When joined with low-powered microprocessors connected to any kind of tiny sensors (e.g., light intensity, temperature, 3D movement, heartbeat), a small battery-powered sensor, capable of processing data and communicating with other wireless devices is created [1]. When establishing communication between these devices starts the design of a network called wireless sensor network (WSN). A WSN can be deployed in almost all types of environment, and this is justified, among other reasons, by the small size of the devices, ease of communication and low cost, consequently there are several applications for WSNs such as military, industrial, transportation, agriculture, medical monitoring of areas of difficult access, and structural integrity. Among the several applications of WSN, this paper consider its utilization in the microclimate monitoring in forest areas, where sensor nodes, used for monitoring temperature and humidity, are scattered in a given area and sent the data collected to a central node, called base station (BS). Replace or recharge the battery of sensor nodes is expensive and impractical in some situations. The efficient utilization of sensors energy resources was and still is the main design consideration for the most proposed protocols and algorithms for sensor networks, and has dominated most of the research in WSNs [2]. It is necessary an energy aware communication protocol to ensure reduced power consumption.
In order to cope with this problem, several communications protocols have been developed in order to find ways to prolong the lifetime of the WSN, among these, clustering protocols can reduce energy consumption and prolong lifetime of network. In clustering protocols the nodes are partitioned into groups, called clusters, where each cluster has a leader, called clusterhead, which coordinates the sending of information from non-cluster-head nodes to itself, processing, aggregating and sending the data collected to BS. One of the most referenced communication protocols by routing protocols proposed in the literature is LEACH (lowenergy adaptive clustering hierarchy) [3], and several extensions of LEACH protocol have been reported in the literature [4], [5], [6], with aim of improving and solving some drawback found in LEACH, such as: the number of nodes in every cluster is not uniform, cluster member nodes deplete energy after cluster head was dead and unreasonable cluster head selection. In order to get a better balanced and reduced energy consumption, this paper considers the combination of the remaining energy of node and the distance of nodes to BS, two important variables, using fuzzy logic to get a heuristic characterization of the energy consumption in network. The CHEATS protocol, an improvement on LEACH protocol, combines multiple metrics on order to resolve the drawback about election on cluster-head. In CHEATS protocol, each node determines a probability to become a cluster-head candidate based on the remaining energy of node and the distance on node to BS. This probability is computed through a multidimensional reasoning using Takagi-Sugeno fuzzy system (TSFS). The next sections of the paper are organized as follows. Section II describes the related work. Network model is described in Section III. Proposed of CHEATS protocol is described in Section IV. Section V presents the simulation environment and analysis of the proposed work. Lastly conclude the paper in Section VI. II. RELATED WORK In recent years, many protocols and algorithms based on clusters have been proposed for wireless sensor networks, many them based on LEACH protocol try solve the problems found in algorithm. The LEACH protocol is divided in rounds, in which each round has two phases: set-up phase and steady-state phase. In
c 2011 IEEE 978-1-4673-0279-1/11/$26.00
set-up phase, a node is elected as cluster-head by choosing a random number between 0 and 1, if the number is less than a threshold T(n), given in equation 1, the node becomes a cluster-head for the current round. Other non-cluster-head nodes determine their cluster by choosing the cluster-head that requires the minimum communication energy, and the clusterhead create a schedule of transmission for the nodes in its cluster. ( P if n ∈ G 1−P ×(r mod P1 ) T (n) = (1) 0 otherwise Thein et al. [4] proposed an extension to the LEACH’s stochastic cluster-head selection algorithm by modifying the probability of each node to become cluster-head based on remaining energy level of sensor nodes for transmission. This algorithm tries to adapt clusters and rotating cluster head positions to evenly distribute the energy load among all the nodes. The Far-Zone LEACH protocol (FZ-LEACH) [5] is an improvement to the LEACH’s to solve the problem of existence of the large clusters in the sensor networks, by forming FarZone that is a group of sensor nodes, which is placed at locations where their energies are less than a threshold. This improvement is well suited for scenarios where there exist large size clusters. The aim of the proposal is to reduce energy consumption in the network and extend its lifetime. Ren et al. [6] proposed an Improved LEACH protocol, which has more reasonable set-up phase. In the cluster heads election phase, the protocol considers the residual energy and distance to sink node and also adopts round-robin in clusterhead election, and rotated time slot are predefined. Barolli et al. [7] proposed a power reduction algorithm for sensor networks based on fuzzy logic and number of neighbor nodes called F3N. The algorithm uses four parameters for CH selection: Distance of Cluster Centroid, Remaining Battery Power of Sensor and Degree of Number of Neighbor Nodes. Fuzzy unequal clustering algorithm (EAUCF) [8] is a distributed competitive algorithm, which adjusts the cluster-head radius considering the residual energy and the distance to the base station parameters of the sensor nodes, focusing on wisely assigning competition ranges to the tentative cluster-heads and using fuzzy logic for handling the uncertainties in cluster-head radius estimation. Energy-Efficient Cluster Head Selection (NECHS) [9] is a mechanism based on fuzzy logic in clustering routing, where the appropriate cluster-head is selected based on the neighbor nodes and remaining energy, which is the input variables of fuzzy system. The output is probability of each node to becoming a cluster-head. In fuzzy modeling is used the compositional inference, with center of gravity as defuzzification method. LEACH presents some drawbacks, such as the election of cluster-head that not consider the state of sensor nodes in terms of energy, being a decision purely probabilistic and not aware of the energy levels. Thus, nodes can be selected and die quickly, leading to a possible disruption of communication
with the sink, decreasing the lifetime of the network and unbalanced energy consumption. Many of proposals consider the improvement on the process election of cluster-head through the use of fuzzy logic that allows a better estimate of the imprecise knowledge by means of multiple criteria. However, the extension proposed use the so-called compositional rule of inference to infer the output of the system. This paper uses a multivariable TSFS where fuzzy implications and reasoning are reduced, quite simple and with greater computational efficiency, being well suited to sensor nodes which are very limited computing resources. Experiments were conducted with different membership functions in order to elect functions that enable a better modeling on the imprecise knowledge by considering the nonlinearity measurements. Moreover, this proposal considers the distance of cluster-head to base station, which enables to reduce the power dissipation and interference by controlling the signal strength, reducing energy consumption and packet loss. III. ASSUMPTIONS A. Network Model This paper considers a WSN used in environmental monitoring for collect meteorological data, therefore meteorological data need to be collected periodically. It is assumed a sensor network consisting of N sensor nodes randomly distributed in a square area, a BS and the following properties: • All sensor nodes are energy constrained and the BS doesn’t has energy restriction and is located inside the sensing field; • Nodes are left unattended after deployment. Thus, battery re-charge is not possible; • All sensor nodes have equal capabilities; • All sensor nodes are not mobile; • All sensor nodes can transmit with enough power to reach the BS, if needed; • Links are symmetric. Thus, two nodes can communicate using the same transmission power; • All sensor nodes are location-unaware, i.e. not equipped with GPS-capable antennas; • Each sensor node can change the power level dynamically and has a fixed number of transmission power levels. B. Considerations on the use of fuzzy logic During the lifetime of a WSN cluster-based, the conditions to cluster formation is usually a function of inaccurate measurements, such as the energy of a sensor node and the distance, therefore can be conveniently expressed in linguistic terms, e.g., the remaining energy can be low, average or high. In this case, the fuzzy logic can manipulate variables with imprecise and linguistic values, and is a multivalued logic that allows intermediate values to be defined between conventional threshold values, allowing judicious cluster-head election across multiple criteria. Fuzzy logic is very well suited for implementing routing and clustering heuristics and optimizations like link or cluster
head quality classification. However, other techniques of computational intelligence are not yet adequate, such as Neural Networks and Evolutionary Algorithms, because they have very high processing demands and are usually centralized solutions and the use of swarm intelligence requires high communication overhead for sending ants separately for managing the routes [10]. These techniques are inappropriate for this proposal, because the clustering of nodes is distributed, nodes are limited in processing capacity, and should ensure an appropriate reduction of energy consumption on the network without major increases in the overload control. IV. CHEATS P ROTOCOL This section presents the CHEATS protocol and the design of fuzzy system proposed. The protocol considers the use of remaining energy and the distance of node to BS for the election of cluster-head by determining the nodes choice probability. This proposal changes the setup phase and maintains the steady-phase of LEACH. A. Algorithm Cluster-head Election During initialization of the network the BS broadcasts a “hello” message to all nodes, each node can compute the approximate distance to the BS on the received signal strength and it adjust the transmission power to the distance. In the next step, the nodes compute the probability to become cluster-head candidate using the TSFS. To this each node generates a random number µ ∈ [0,1], then makes a decision whether it can be a cluster head according to its judgment, that compare µ with the probability and a sampling rate α, determined by the equation 2. If µ is less than both, the node becomes a cluster-head candidate for the current round, and this broadcasts an advertisement message containing the information of probability, then each node sort the received probabilities and performs a sampling based on α. If the node ID is contained between the α first, then it is going to be a cluster-head. α = 2 × nCh/100,
(2)
where, nCh is the desired amount of cluster-head for the network. After all the nodes perform this operation, the cluster-head nodes waiting the JOIN-MESSAGE of non-candidates-nodes and candidates who did not become cluster-head, to formation of cluster, such as LEACH, in the following operations are the same as LEACH protocol. The chosen cluster-head node perform this operation only once every α1 rounds. The Algorithm 1 and Table I describe the new setup phase. B. Fuzzy System Model In this paper is used a general TSFS, which maps crisp inputs into crisp outputs constituted of four modules: a ruler base, an inference engine, fuzzifier and defuzzifier, where the T-norm and propagation operation is the algebraic product, the aggregation operator is sum, and the defuzzification algorithm
Algorithm I: Cluster-Head Election if Sink then Broadcast advMessage at time 0 end if if Receive advMessage then estimateDistance end if µ ← rand(0,1) probability ← fuzzySystem(energy, distance) if µ < probability & µ < α then beCandidateHead ← true end if if beCandidateHead = true then Broadcast candidateHeadMessage(id,probability) setTimer(delay) while timerExpired() = false do c.probability ← msgReceveid.probability c.id ← msgReceveid.id Add c to candidateClusterHead set S’ end while CH’ = sort(S’, probability, α) if (search(id,CH’)) then setTimer(delay) else CH’ = sort(CH’, RSSI) Broadcast joinHeadMessage(CH’[0].ID,id) end if else setTimer(delay) while timerExpired() = false do c.probability ← msgReceveid.probability c.id ← msgReceveid.id Add c to candidateClusterHead set S’ end while CH’ = sort(S’, probability, α) CH’ = sort(CH’, RSSI) Broadcast joinHeadMessage(CH’[0].ID,id) end if
TABLE I D ESCRIPTION OF FUNCTIONS AND VARIABLES OF THE ALGORITHM
Functions and Variables fuzzySystem(energy,distance) SetTimer(delay) sort(set,criteria,sample) c CH’ S’ search(key,Set)
Description Process and return the probability using fuzzy logic Start a timer Returns a sample of the highest values aˆ aˆ of a given set according to a criterion Structure for Cluster-Head information Set of Cluster-Heads Set of candidates for Cluster-Heads Research an id in a given set
is weighted average. The architecture of the fuzzy system used is shown in Figure 1. In TSFS, the fuzzy implication is based on a fuzzy partition of input space. In each fuzzy subspace a linear input-output relation is formed. The output of fuzzy reasoning is given by the aggregation of values inferred by some implications that are applied to an input [11]. Input linguistic variables of the system are the remaining energy, expresses in percentage, and the distance of node to BS, expresses in meters. The remaining energy is represented in percentage to normalize the values of energy in different sensor nodes enabling expands the proposed algorithms to
trapezoidal and triangular functions, fairly satisfactory with the bell-shaped function, showing good results with small parameter values for the center curve, and achieved better results with the Gaussian function by better describe the nonlinearity and uncertainty about the variables. The function is given below: f (x, σ, c) = e
others sensor nodes with different radios. The specifications related to the universe of discourse of input and output functions of the system and their respective linguistic values (LV) are as follows: • Residual energy: u=[0,100] : LV = Low, Average, High • Distance: u=[0,100] LV = Small, Average, Big • Probability: u=(0,1] LV = very high, high, moderately high, little high, average, little low, moderately low, low, very low The capability to determine appropriate membership functions and meaningful fuzzy operations in the context of each particular application is crucial for making fuzzy set theory practically useful [10]. The characteristics of a fuzzy system will depend on the application that it intends to solve. Thus, different applications will require different T-norm, different membership functions, or different defuzzification algorithms, and the same problem can be solved by different designers using fuzzy systems with different characteristics [12]. Several experiments were performed to examine which membership functions that better represent the linguistic states, using the functions trapezoidal, triangular, bell-shaped, Gaussian, Z-shaped and S-shaped for both linguistic representation of the states of the input. For the representation of linguistic states low, high, small and big of input variables, the degrees of membership to these sets should remain constant from certain values of the universe of discourse. The functions in the form of S and Z, equation 3 and 4 respectively, are chosen because they demonstrate to be the most suitable for the states representation. 0, x
b
For linguistic intermediaries representing the states of the input system and the cost function of output, experiments were performed and the results were unsatisfactory with the
(5)
The membership functions designed for the system are presented in Fig. 2(a) and 2(b): Membership Degree
Fuzzy Diagram [10]
1
Low
High
Average
0.8 0.6 0.4 0.2 00
20
40
60
80
100
Energy (%)
(a) Fuzzy sets for RemainingEnergy Membership Degree
Fig. 1.
−(x−c)3 2σ2
1
Small
Big
Average
0.8 0.6 0.4 0.2 00
20
40 60 DistanceToBS (m)
80
100
(b) Fuzzy sets for DistanceToBS Fig. 2.
Membership Functions
The rules are expressed as logical implications in the form of IF-THEN statements into a mapping from fuzzy input sets to output functions. The rules are determined based on the whole network behavior analysis through extensive simulations over time. The rules that result in a class of probability high, ensures an excellent chance for the election of these nodes and differentiates into the class according to their distance to BS. However, the network behavioral analysis showed that average levels of energy cluster-head can lead them to death quickly, so despite the nodes present themselves as good candidates to chosen, they should be avoided in order to prevent their death. In this case the rule base provides a median probability. For cases in which the output of the system presents a low level of energy, this deal with situations that presents values of energy poor in large part of the network. Thus, the low class of probability tries reduces the number of cluster-head nodes in network. In all cases the distance of node to BS is a very important variable, because the nodes next to BS require a minimum power that enables communication, which may be controlled and it should have more probability of choice, decreasing the energy dissipation in network. The output of rules is zero-order functions, and the rules are expressed as:
IF f (x1 isA1 , ..., xk isAk ) T HEN y = p0
(6)
Table II shows the fuzzy inference rules used in system.
4.0 [14], a discrete event network simulator, and the Castalia Simulator [15]. The basic parameters used are listed in Table III. 50
ENERGY High Hight Hight Average Average Average Low Low Low
DISTANCE Small Average Big Small Average Big High Average Low
Distance (m)
TABLE II F UZZY I NFERENCE RULES PROBABILITY very high y=1 high y=0.9 moderately high y=0.8 little high y=0.6 average y=0.5 little low y=0.2 moderately low y=0.1 low y=0.07 very low y=0.02
(8)
where p is the order of output function. Then the truth value of proposition is calculated by the equation: A(xi ),
10
20 30 Distance (m)
Fig. 3.
where {x0 ,...,xk } is the set of system input, {A1 ,...,Ak } the set of membership functions that defines the rule, being y the output of rule, f is logical function connecting the propositions in the premise, and g is the function that implies the value of y. For each implication Ri , yi is calculated by the function gi in the consequence, thus:
n Y
20
0 0
IF f (x1 isA1 , ..., xk isAk ) T HEN y = g(x1 , ..., xk ), (7)
wi =
30
10
The inference engine was designed according to the approaches data-driven. In this case, available data are supplied to expert system, which uses them to evaluate relevant production rules and draw all possible conclusions [13]. The fuzzy implication Ri is defined by a set of rules in which each rule is of the format:
yi = p0 + p1 x1 , ..., +pk xk ,
40
(9)
i=1
where A(xi ) is the grade of the membership of x0 and wi is the degree of truth of rule Ri . The output y inferred from n implications is given by the weighted averages of {y1 ,...,yn }: Pn wi yi wi = Pi=1 (10) n i=1 wi V. SIMULATION RESULTS
A. Evaluation Setup It is considered a network with 100 nodes randomly distribution in a size field of 50mx50m, where the base station is the triangle in black and red stars are the sensor nodes, as shown in Fig 3. Simulation is performed using Omnet++
40
50
Topology
TABLE III PARAMETERS Parameter Field Size Location of Base Station Number of Nodes Number of Cluster-head Initial Energy of sensor node The data packet size Time Round Radio Model Channel Model
Value 50m X 50m (50m,50m) 100 5 20J 26bytes 20s CC2420 ZigBee-ready Free Space
B. Simulation Results The CHEATS was compared with LEACH, in terms of network lifetime, energy consumption and throughput, with LEACH using a probability of 0.05, which corresponds to an average of 5 clusters to the scenario proposed. In the data analysis of the simulations were carried out 30 simulations with a confidence interval of 99%. Fig. 4 shows the total number of nodes alive over time, each node initially given 20 J of energy. This graphic show that nodes in LEACH protocol die more quickly compared to the proposal. The results show an increase in the network lifetime by 10% compared to LEACH protocol, this occurs because the selection the cluster-head, by determining the probability of choice through the TSFS, for the formation cluster consider the remaining energy of nodes, allowing nodes with low energy levels have a lower probability of choice in relation to others with higher levels. Simulations were performed on the same scenario to evaluate the first node death time, half of the network nodes and the total death. Fig. 5 shows that the proposal could be higher than the LEACH throughout the operation of network. Fig. 6 illustrates the energy consumption, showing the accumulated energy dissipation on the time. The energy consumption of the CHEATS is less than LEACH protocol. This occurs because of power control performed by cluster-head and the probability for election of cluster-head that consider
80 60 40 20 0
1400
1500
Fig. 4.
1600
1700 1800 Time (s)
1900
2000
Packets Received at BS
Number of Alive Nodes
LEACH CHEATS
100
1200
LEACH CHEATS
1000 800 600 400 200 0
500
1000
2100
Fig. 7.
Number of nodes alive with time
1500 2000 Time (s)
2500
3000
Packets Received at BS
mines its probability to become cluster-head candidate based on energy remaining and the distance of nodes to BS, using a Takagi-Sugeno fuzzy system to deal with the uncertainties in the estimates. Simulation results show an improvement in the performance on the LEACH protocol in terms of energy dissipation, network lifetime of 10%, and more than 50% in throughput. R EFERENCES
Fig. 5.
Network Lifetime
Energy Dissipation (J)
the distance to BS, allowing nodes closer has more probability to be cluster-head, reducing the energy dissipation in the network. 16 14 12 10 8 6 4 2 0
LEACH CHEATS
200
400
600
Fig. 6.
800
1000 1200 Time (s)
1400
1600
1800
2000
Energy Dissipation
Fig. 7 shows the number of data message received at BS on the time, the results show an increase of 50% in amount of received packets in BS. The cluster-head is selected based on the remaining energy diminishing the chances of selected nodes have exhausted its energy sources, that creates a higher expectation of the transmission network. The proposed algorithm also allows a better distribution of network nodes on the clusters. Furthermore the power control conducted by cluster-head reduces the level of interference in the network avoiding packet losses. The network energy depletion is fast in LEACH. As shown in figures 4, 5, 6 and 7, we can conclude that our proposed model provides better characteristics regarding the extension of lifetime of the network and throughput comparing to the LEACH protocol. VI. CONCLUSIONS In this paper is proposed an improvement of the LEACH protocol that adjusts the setup phase, where each node deter-
[1] L. Mendes and J. Rodrigues, “A survey on cross-layer solutions for wireless sensor networks,” Journal of Network and Computer Applications, vol. 34, pp. 523–534, mar 2011. [2] Y. Huang and C. Huang, “Cluster algorithm for electing cluster heads based on threshold energy,” in Proc. IEEE International Conference on Electrical and Control Engineering (ICECE’2010), jul 2010, pp. 2692– 2695. [3] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “Energyefficient communication protocol for wireless microsensor networks,” in Proc. IEEE 33rd Annual Hawaii International Conference on System Sciences, jan 2000, pp. 1–10. [4] M. C. M. Thein and T. Thein, “An energy efficient cluster-head selection for wireless sensor networks,” in Proc. IEEE International Conference on Intelligent Systems, Modelling and Simulation (ISMS’2010), jan 2010, pp. 287–291. [5] V. Katiyar, N. Chand, G. C. Gautam, and A. Kumar, “Improvement in leach protocol for large-scale wireless sensor networks,” in Proc. IEEE International Conference on Emerging Trends in Electrical and Computer Technology (ICETECT’2011), mar 2011, pp. 1070–1075. [6] P. Ren, J. Qian, L. Li, Z. Zhao, and X. Li, “Unequal clustering scheme based leach for wireless sensor networks,” in Proc. IEEE Fourth International Conference on Genetic and Evolutionary Computing (ICGEC’2010), dec 2010, pp. 90–93. [7] L. Barolli et al., “Evaluation of an intelligent fuzzy-based cluster head selection system for wsns using different parameters,” in Proc. IEEE Workshops of International Conference on Advanced Information Networking and Applications (WAINA’2010), mar 2010, pp. 388–395. [8] H. Bagci and A. Yazici, “An energy aware fuzzy unequal clustering algorithm for wireless sensor networks,” in Proc. IEEE International Conference on Fuzzy Systems (FUZZ’2010), jul 2010, pp. 1–8. [9] Y. Hu, X. Shen, and Z. Kang, “Energy-efficient cluster head selection in clustering routing for wireless sensor networks,” in Proc. IEEE 5th International Conference on Wireless Communications, Networking and Mobile Computing (WiCom’2009), jan 2009, pp. 1–4. [10] V. Kulkarni, A. Forster, and G. Venayagamoorthy, “Computational intelligence in wireless sensor networks: A survey,” IEEE Communications Society Surveys & Tutorials, vol. 13, no. 2, pp. 1–29, 2011. [11] T. Takagi and M. Sugeno, “Fuzzy identification of systems and its applications to modeling and control,” IEEE Transactions on Systems, Man, and Cybernetics, vol. 15, no. 1, pp. 116–132, 1985. [12] E. Frias-Martinez, “Universal fuzzy system to takagi-sugeno fuzzy system compiler,” in Proc. IEEE International Conference on Fuzzy Systems (FUZZ’2002), may 2002, pp. 305–309. [13] G. J. Klir and B. Yuan, Fuzzy sets and fuzzy logic: theory and applications. PTR Upper Saddle River, NJ, USA: Prentice Hall, 1995. [14] (2011) Omnet++ descrete event simulation system. [Online]. Available: http://www.omnetpp.org/ [15] NICTA. (2011) Castalia a simulator for wsns. [Online]. Available: http://castalia.npc.nicta.com.au/ 