A Framework to Minimize Energy Consumption for Wireless Sensor Networks

A Framework to Minimize Energy Consumption for Wireless Sensor Networks Feng Shu, Taka Sakurai,HaiL.Vu, Moshe Zukerman Department of Electrical and Electronic Engineering, The University of Melbourne, VIC 3010, Australia Email: f.shu@ee.unimelb.edu.au Centre for Advanced Internet Architectures, Faculty of I.C.T., Swinburne Univ. of Technology, P.O. Box 18, VIC 31, Australia Abstract This paper presents a framework to minimize energy consumption in the medium access control (MAC) layer for wireless sensor networks. While satisfying a range of quality of service (QoS) requirements, such as the packet transmission success rate and maximum delay constraint, we optimally choose the lengths of periods in which sensors are active and inactive, such that the energy consumption per unit time in the entire network is minimized. We first use our framework to optimize the values of the MAC attributes macbeaconorder and macsuperframeorder in an IEEE 80.15.4 beacon-enabled star network. Then we consider a much simpler protocol, which we call select-and-transmit (S&T), and the same framework is applied to find the optimal lengths of the active and inactive portions. Finally, we compare the minimal energy consumption of the IEEE 80.15.4 MAC and S&T under the same QoS requirements and show that the IEEE 80.15.4 MAC outperforms S&T in most cases. However, the S&T MAC performs better than the standard under our framework in certain scenarios, e.g., event-driven sensor networks where the packet transmission success rate is usually low. I. INTRODUCTION In recent years, there has been a surge in interest in wireless sensor networks (WSN) composed of low-cost, low-power sensor nodes. Medium access control (MAC) protocols play a critical role in meeting the stringent energy consumption requirement in a sensor network. A number of energy efficient MAC protocols for WSN have been proposed (e.g., [1] [7]). Ye et al. [1] propose sensor-mac (S-MAC) which uses a periodic listen-and-sleep strategy to reduce idle listening and thus minimize energy consumption. In B-MAC [], an adaptive preamble sampling scheme is employed to decrease duty cycle and idle listening. Tay et al. [3] propose a MAC protocol, CSMA/p, for event-driven sensor networks, in which the traffic sources are considered to be spatially correlated and so it is not necessary to collect information from all of the nodes that observe the same event. To the best of our knowledge, the problem to optimally determine the lengths of sensor active and inactive periods remains unaddressed in the literature thus far. This work was supported by the National ICT Australia (NICTA) and the Australian Research Council (ARC). F. Shu is supported by NICTA Victoria Laboratory. T. Sakurai and M. Zukerman are with the ARC Special Research Centre for Ultra-Broadband Information Networks (CUBIN). In this paper, we propose a framework to minimize energy consumption for WSNs. Given the quality of service (QoS) requirements (e.g., packet transmission success rate and maximum delay constraint), we optimally choose the lengths of periods in which sensors are active and inactive, such that the energy consumption per unit time in the entire network is minimized. We first use our framework to optimize the values of the MAC attributes macbeaconorder and macsuperframe- Order in an IEEE 80.15.4 (referred to as 80.15.4 hereinafter) [8] beacon-enabled star network. Then we consider a much simpler protocol, which we call select-and-transmit (S&T), in which a sensor node chooses a transmission slot uniformly in an equally slotted period without using carrier sense. The same framework is then applied to find the optimal lengths of the active and inactive portions. The remainder of this paper is organized as follows. In Sections II and III, the 80.15.4 and the S&T MACs are briefly described and discussed. In Section IV, we describe the energy models and our framework to minimize energy consumption for both the 80.15.4 MAC and the S&T MAC. Finally, results and discussion are presented in Section V. II. 80.15.4 MAC In an 80.15.4 star network, one node is appointed as the network coordinator. The network operates with a superframe structure, which may consist of active and inactive portions. Let time be divided into consecutive time intervals called beacon intervals (BI). The superframe duration (SD), which denotes the active portion of the superframe, may consist of a beacon frame (BF), a contention access period (CAP) and a contention free period (CFP). In this work, we set the length of CFP to 0; in other words, there is no CFP period in the superframe. The MAC attributes macbeaconorder (BO) and macsuperframeorder (SO) describes the BI and SD, respectively, where BO and SO are integers and 0 SO BO 14. More specifically, the lengths of BI and SD (measured in symbols) are given by abaseslotduration anumsuperframeslots BO and abaseslotduration anumsuperframeslots SO, respectively, where abaseslotduration is set equal to 60 symbols and anumsuperframeslots is equal to 16 in the standard. For

more detailed description of the standard, readers are referred to [8], [9]. III. S&T MAC In [10], we provide a detailed analysis of a simple S&T MAC scheme. In the following, we briefly review the S&T MAC and its analysis. We again consider a beacon-enabled star network in which one node is designated as the coordinator. We also borrow the same notation of BF, SD, BI, CAP for the S&T MAC. In the S&T superframe structure, we divide the active portion, SD, into coarse slots such that each slot is sufficient to transmit one packet. Let T bn, T and T bi denote the lengths (all measured in number of slots) of BF, CAP and BI, respectively. In the S&T MAC scheme, a backlogged node wakes up at the beginning of a BI, listens to the BF, and then chooses a slot for transmission in CAP according to a uniform distribution. After waiting for a node idle period (NIP), T nip, which equals TS 1 slots, it transmits its next packet and then goes back to sleep for a node sleep period, T nsp, until the end of the current BI. If two or more nodes select the same slot, a collision results. Because a node chooses a transmission slot uniformly in T, the mean of T nip in the S&T MAC is given by, E[T nip ]= T 1. (1) We can also compute the mean of the node sleep period T nsp as follows: E[T nsp ] = T bi T bn E[T nip ] 1 = T bi T bn T +1. () The probability Q k, that at least k out of n nodes successfully transmit their packets (i.e., without collisions) in a given T using the S&T MAC scheme, is given by: k 1 Q k =1 P j, (3) j=0 where P k denotes the probability that exactly k out of n nodes are successful, and is given as follows. If n T, T! P n = T n (T n)! If 0 k n and k<t, P k = 1 T n min(n,t ) ( )( )( ) n T w w!( 1) w k (T w) n w. w w k w=k Readers are referred to [10] for the derivation of P k.fig.1 demonstrates that when k is relatively small compared to n,the probability Q k can be made to quickly converge to 1 through a modest increase in T, while a much larger T is required for the same effect if k is large. These results suggest that the S&T MAC is most likely to be a competitive MAC solution for scenarios where only a small proportion of the packets are needed to report some event (e.g., event-driven workload [3]). Probability Q k 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0. k = 0.1 k = 7 k = 1 0 0 50 100 150 00 50 Length of CAP, T (in slot time) Fig. 1. Probability Q k versus length of active period T. n =15. IV. ENERGY MODELS AND ENERGY CONSUMPTION MINIMIZATION STRATEGY In this section, we first describe the energy models and then examine our framework to minimize the energy consumption for both the 80.15.4 MAC and the S&T MAC. We consider N to be a random variable representing the number of backlogged nodes at the beginning of a BI, and let n denote an outcome of N. We also allow the requirement for the number of successful packets k to vary as a function of n, and we capture this dependence through the notation k(n). In addition, we introduce the following notation. N all : number of nodes (excluding the coordinator) in the network. p N : probability mass function of N. D max : maximum delay constraint, [s]. W k(n) : probability of at least k successes out of n nodes for 80.15.4 MAC. Q k(n) : probability of at least k successes out of n nodes for S&T MAC. η k(n) : a user-specific QoS requirement for W k(n) and Q k(n). S bn : slot duration for the BF, for 80.15.4 MAC, [s]. S sd : length of SD in 80.15.4 MAC, [s]. S bi : length of BI in 80.15.4 MAC, [s]. S nip : average node idle time spent in backoff state, for 80.15.4 MAC, [s]. S nsp : average node sleep time, for 80.15.4 MAC, [s]. m cca : average number of clear channel assessment (CCA) operations performed in one BI, for 80.15.4 MAC. m tx : average number of packet transmissions in one BI, for 80.15.4 MAC. t slot : slot size, for S&T MAC, [s]. P active : average coordinator power consumption during SD, [mw]. P idle : node power consumption in idle period, [mw]. P sleep : node power consumption in sleep period, [mw]. : overall energy consumption in one BI, for 80.15.4 MAC, [mj]. E std sum

E std sum(n) =P active S sd + P sleep (S bi S sd ) + }{{} N all E }{{ 0 } for coordinator for all nodes + m cca E cca + m tx E t + n(p idle S nip + P sleep S nsp ) +(N all n)p sleep (S bi S bn ). }{{}}{{} (4) for backlogged nodes for non backlogged nodes E s&t sum(n) =(P active (T bn + T )+P sleep (T bi T bn T )) t slot + }{{} N all E }{{} 0 for coordinator for all nodes + n(p idle E[T nip ]t slot + E t + P sleep E[T nsp ]t slot ) +(N all n)p sleep (T bi T bn )t slot. }{{}}{{} (5) for backlogged nodes for non backlogged nodes Esum s&t : overall energy consumption in one BI, for S&T MAC, [mj]. E 0 : node energy consumption to wake up and listen to the BF, [mj]. E t : node energy consumption to transmit a packet, [mj]. E cca : energy consumption to perform a CCA operation, [mj]. A. Energy Models A typical sensor node is comprised of several sub-systems such as a microcontroller unit, a radio and a sensing device, each of which consumes energy when a node works [11]. In our energy models, we only investigate the energy consumption of the radio unit, which is closely related to the design of MAC protocols. To simplify the energy models, we do not use the acknowledgment mode for both of the two MAC schemes. In the following we demonstrate, using two examples, how different MAC protocols give rise to different energy models. 1) 80.15.4 MAC: In the 80.15.4 MAC, the overall energy consumption in a network during one BI consists of several components. Firstly, the network coordinator is active for S sd and inactive for (S bi S sd ). Secondly, every node spends energy E 0 in waking up and listening to the BF. When a node wakes up, it switches from the sleep mode to receive mode and turns on its radio. Thirdly, for the backlogged nodes, they perform m cca clear channel assessment (CCA) operations in one BI on average and m tx of the nodes eventually transmit their packets using the slotted CSMA-CA mechanism. The CCA mode we choose for our energy model is carrier sense only, in which a busy medium is reported only when a signal with the modulation and spreading characteristics of 80.15.4 is detected [8]. Moreover, a backlogged node on average spends S nip in the backoff state and goes back to sleep for S nsp, i.e., the rest of the current BI. Finally, non-backlogged nodes immediately go to sleep after the BF and expend P sleep (S bi S bn ) amount of energy. In summary, given that there are N all nodes in a network and n of them are backlogged, the overall energy consumption inabi,esum, std is given by (4). ) S&T MAC: Similarly, the overall energy consumption in one BI for the S&T MAC, Esum, s&t is given by (5). Input: System Parameters Fig.. Energy Optimization MAC and its energy model Output: Energy minimization framework. Optimum MAC Parameters Note that there are two major differences between the energy consumption models of the two MAC schemes. Firstly, a S&T node does not perform carrier sense before transmission. Secondly, in the standard, a backlogged node may not have a chance to transmit its packet before the end of the current CAP, while in the S&T MAC, every backlogged node will transmit its packet. B. Energy Consumption Minimization Strategy Fig. shows the energy consumption minimization framework. Our aim is to minimize the expectation of the overall energy consumption per unit time in one BI while satisfying a range of QoS requirements. To this end, we formulate an optimization problem for the 80.15.4 MAC as follows: min : s.t. 1 N all S bi n=0 p N (n) E std sum(n) W k(n) η k(n) S bi + S sd S bn D max 0 SO BO 14 (6) where n =0, 1,..., N all. The decision variables are SO and BO. Note that S sd and S bi are one-to-one functions of SO and BO, respectively. The first constraint is to satisfy that at least k out of n nodes are successful with probability no less than η k(n), and the second constraint is to meet the maximum delay requirement. The worst case packet delay we consider is equal to S bi + S sd S bn. The third constraint 0 SO BO 14 is defined in the standard. Similarly, taking the expectation of the overall energy consumption per unit time in one BI, the energy consumption

TABLE I ELECTRICAL SPECIFICATIONS. Parameter Value Voltage (V) 3 Transmit current (ma) 0 Receive current (ma) 15 Idle current (ma) 10 Avg. current (coord. active) (ma) 16 Sleep current (ma) 0.03 Initialize radio and time(ma, s) 6, 3.5 10 4 Turn on radio and time (ma, s) 1, 1.3 10 3 Sleep to RX and time (ma, s) 15,.5 10 4 TABLE II SYSTEM PARAMETERS. Parameter Value Data packet size (PHY) (bytes) 5 Beacon frame size (bytes) 5 Maximum packet delay (s) 5 Packet arrival rate, λ (pkts/s) 0.5 QoS requirement η k(n) 0.90 Number of slots for BF in S&T MAC, T bn 1 Slot duration for S&T MAC (s) 0.01 minimization problem for the S&T MAC is given by min : N 1 all p N (n) E T bi t sum(n) s&t slot n=0 s.t. Q k(n) η k(n) (T bi + T )t slot D max T k(n) T bi T + T bn (7) where n =0, 1,..., N all. Here the decision variables are T and T bi. Again, the first and the second constraints are the proportion of successful packet transmissions and maximum delay, respectively. The third constraint sets a lower bound for T, which must be greater than or equal to the number of required successful nodes k(n). The fourth constraint ensures that the length of an SD is not greater than that of a BI. V. NUMERICAL RESULTS In this section, we describe the traffic model we used and define three different scenarios to evaluate the performance of the two schemes under our proposed framework. A. Traffic Model In the scenario we consider, a node becomes backlogged if during the previous BI one or more packets have generated for transmission. The probability mass function of the random variable N, which represents the number of backlogged nodes at the beginning of a BI, is denoted by p N (n), where n denotes an outcome of N. We make the observation that if more than one packet is in the transmission buffer, it is possible for users to make choices regarding the order of transmission and whether redundant packets can be dropped. In particular, if packets awaiting transmission contain information on the same event, only the latest is relevant, and it is sensible to discard the old ones. In our setup, we assume that when a node generates a new packet, it will be accommodated in the transmission buffer if the buffer is empty. When there is already a packet in the node s buffer, the action taken depends on the status of the node. If the node is active, it means the buffered packet is attempting to transmit, so the new packet will wait to transmit in the next BI. If the node is sleeping, the new packet will replace the old one, which will be discarded. B. Scenario Definition As mentioned in Section IV, k is defined as a function of n. To evaluate the performance of energy consumption for the two MAC protocols, we define a generic function k(n) = min(α 1 n, α N all ), where α 1,α (0, 1) and α 1 α.this function typically allows k to increase proportionally with n until it hits the upper bound α N all when n is close to N all. We now define three scenarios for evaluation: 1) Scenario A: α 1 =0.,α =0.1. ) Scenario B: α 1 =0.5,α =0.4. 3) Scenario C: α 1 =0.8,α =0.7. In Scenario A, we are interested in the successful transmission of a very small proportion of the packets. Applications consistent with these scenarios could be event-driven sensor networks (refer to [3] for detailed examples). Conversely, in Scenario C most of the packets need to be transmitted successfully. Scenario B corresponds to a case where a moderate number of transmissions need to be successful. C. Results and Discussion We adopted the electrical specifications used in [] (reproduced in Table I), which are based on the CC1000 transceiver. Some other system parameters are shown in Table II. We chose Poisson processes for packet generation of the nodes, which means that for each node, the probability of being backlogged at the beginning of a BI is given by p nd =1 e λt, where λ is the packet arrival rate and t = S bi for the 80.15.4 MAC and t = T bi t slot for the S&T MAC. Given a network with N all sensor nodes, the probability mass function of N is given by p N (n) = ( N all ) n p n nd (1 p nd ) Nall n. To compute the energy consumption based on (4), we obtained values of W k (n), m cca and m tx through simulation of the 80.15.4 MAC for each set of parameters (i.e. SO, BO, n and N all ). We have repeated the process a sufficient number of times such that the radius of the 95% confidence interval based on student-t distribution is within 5% of the energy consumption value. For the S&T MAC, given a set of values of T, T bi, n and N all, we have analytical results of (1) and () for the energy consumption computation, and therefore no simulations were required. Since SO, BO, T and T bi are all integer-valued and their ranges are constrained, we were able to use exhaustive search to solve the optimization problems. We used the analytical result for Q k(n) (3) to exclude cases that violate the constraint Q k(n) η k(n). The optimal solutions for the two MAC schemes in the.4 GHz frequency band defined in the IEEE 80.15.4 standard

TABLE III OPTIMAL SOLUTIONS. NOTE THAT T AND T bi ARE MEASURED IN SLOTS. SEE SECTION III.A. Scenario A Scenario B Scenario C N all 80.15.4 S&T 80.15.4 S&T 80.15.4 S&T SO BO T T bi SO BO T T bi SO BO T T bi 5 0 7 6 499 0 7 30 4968 0 7 97 4901 10 0 7 6 499 0 7 30 4968 0 7 0 4796 15 0 7 10 4998 0 7 34 4964 1 7 0 4796 0 0 7 11 4987 0 7 41 4957 1 7 30 4768 5 0 7 15 4983 0 7 47 4951 1 7 30 4768 30 0 7 16 498 1 7 54 4944 1 7 57 4741 35 0 7 0 4978 1 7 60 4938 7 79 4719 Energy consumption (mj/s) 0 18 16 14 1 10 8 6 4 S&T A S&T B S&T C 80.15.4 A 80.15.4 B 80.15.4 C 0 5 10 15 0 5 30 35 0 Number of nodes Fig. 3. Comparisons of optimal energy consumption. are listed in Table III. Fig. 3 illustrates the performance comparisons of the optimal energy consumption for the 80.15.4 MAC and the S&T MAC under various scenarios. For relatively small k in Scenarios A and B, the energy consumption of the two MAC schemes are comparable. In fact, the S&T MAC has lower energy consumption than the standard in Scenario A. However, in Scenario C where k is relatively large, the standard can still maintain a low level of energy consumption, while the energy of the S&T MAC increases dramatically. This is because when k is relatively large, the length of CAP, T, needs to be large (and thus the node idle period T nip is greatly increased). 40 30 0 10 [] J. Polastre, J. Hill, and D. Culler, Versatile low power media access for wireless sensor networks, in Proc. nd ACM Conference on Embedded Networked Sensor Systems (SenSys), Nov. 004, pp. 95 107. [3] Y. C. Tay, K. Jamieson, and H. Balakrishnan, Collision-minimizing CSMA and its applications to wireless sensor networks, IEEE J. Select. Areas Commun., vol., no. 6, pp. 1048 1057, Aug. 004. [4] A. El-Hoiydi and J.-D. Decotignie, WiseMAC: an ultra low power MAC protocol for the downlink of infrastructure wireless sensor networks, in Proc. 9th International Symposium on Computers and Communications, June 004, vol. 1, pp. 44 51. [5] G. Lu, B. Krishnamachari, and C.S. Raghavendra, An adaptive energy-efficient and low-latency MAC for data gathering in wireless sensor networks, in Proc. 18th International Parallel and Distributed Processing Symposium, Apr. 004, pp. 4 31. [6] T. van Dam and K. Langendoen, An adaptive energy-efficient MAC protocol for wireless sensor networks, in Proc. 1st ACM Conference on Embedded Networked Sensor Systems, Nov. 003, pp. 171 180. [7] P. Lin, C. Qiao, and X. Wang, Medium access control with a dynamic duty cycle for sensor networks, in Proc. IEEE Wireless Communications and Networking Conference, Mar. 004, vol. 3, pp. 1534 1539. [8] IEEE 80.15.4, Wireless medium access control (MAC) and physical layer (PHY) specifications for low rate wireless personal area networks (LR-WPANS), Standard, IEEE, 003. [9] F. Shu, T. Sakurai, H. L. Vu, and M. Zukerman, Optimizing the IEEE 80.15.4 MAC, in IEEE Region 10 International Conference (TENCON), 006. [10] F. Shu, T. Sakurai, H. L. Vu, and M. Zukerman, Does a WSN MAC based on uniform access without re-attempts have merits?, in IEEE Region 10 International Conference (TENCON), 006. [11] C. S. Raghavendra, Krishna Sivalingam, and Taieb Znati, Wireless Sensor Networks, Kluwer Academic Publishers, 004. VI. CONCLUSIONS We have presented an energy consumption minimization framework for wireless sensor networks, in which we optimally determined the lengths of sensor active and inactive periods, such that the energy consumption per unit time in the entire network is minimized. The IEEE 80.15.4 MAC and select-and-transmit (S&T) MAC were examined using our framework and results showed that the standard is superior in most cases but in cases where the packet transmission success rate is low (e.g., event-driven workload), the S&T MAC performed better than the standard. REFERENCES [1] W. Ye, J. Heidemann, and D. Estrin, Medium access control with coordinated adaptive sleeping for wireless sensor networks, IEEE/ACM Trans. Networking, vol. 1, no. 3, pp. 493 506, June 004.