Optimal Beacon Interval for TDMA-based MAC in Wireless Sensor Networks

Optimal Beacon Interval for TDMA-based MAC in Wireless Sensor Networks Abstract An energy-efficient Medium Access Control (MAC) protocol can significantly elongate the lifetime of wireless sensor networks (WSNs), by reducing the duty-cycle of sensor nodes to an ultra-low level. TDMA-based MACs can greatly reduce idle listening, collision, overhearing and therefore are very energy efficient in data transfer at the cost of requiring tight time synchronization through periodically beacon dissemination. The length of the beacon interval may greatly affect the energy efficiency of a TDMA-based MAC. A shorter beacon interval leads to higher synchronization cost, while a longer beacon interval will lead to a larger guard time due to clock drift. Therefore, there is a tradeoff between these two parts of energy consumption. In this paper, we investigate the optimal beacon interval for TDMAbased MAC in low duty-cycle sensor networks, and then present a strategy that adaptively utilizing the optimal beacon interval in a TDMA-based MAC protocol (called Opt-TDMA). Experimental results demonstrate that Opt-TDMA is more energy-efficient than non-optimal TDMA and SCP-MAC. Index Terms Beacon Interval, TDMA-based MAC, Energy Efficiency, Wireless Sensor Networks I. INTRODUCTION Wireless Sensor Network has become a mainstream solution in many important applications such as target detection and tracking, environmental monitoring, industrial automation, and tactical systems. Energy efficiency is one of the primary considerations in the design of WSN protocols. According to [1], the major power consumption of a sensor node is on the radio. Therefore, a MAC protocol which implements an efficient duty-cycle mechanism is indispensable in prolonging the lifetime of sensor networks. Radio energy consumption comes from sending/receiving data and other control overhead. Beside, there are other unnecessary sources of energy waste, including idle listening, collision and overhearing. Among them, idle listening is considered to be one of the most significant sources of energy consumption in sensor networks. The central way to solve the problem of idle listening is to lower the radio duty-cycle [2], which is defined as the ratio between active time and a full active/sleep period. In the active state a node is able to transmit or receive data, while in the sleep state the node completely turns off its radio to save energy. A study in [3][4] classifies duty-cycle MAC protocols into two categories: asynchronous and synchronous strategies. Asynchronous approaches such as B-MAC [5], X-MAC [6], A-MAC [7] and PW-MAC [8], allow each node chooses its own duty-cycle independently, so the sender and the receiver are completely decoupled in their own duty cycles. In order to satisfy the low-power requirements of the sensor network, asynchronous MAC protocols employ Low Power Listening (LPL) strategy. In this way, a sender transmits a preamble lasting at least as long as the sleep period of the receiver. The receiver periodically probes the channel and stays awake to receive data once it detects the preamble. These protocols remove the synchronization control overhead and achieve high energy efficiency. However, due to the long preamble, they become less efficient in latency and packet delivery ratio when the network is crowded. Also, these protocols suffer from overhearing problem because the long preamble may wake up the nodes which are not the destination of packets. Synchronous MACs can be divided into two classes: contention-based and TDMA-based. Existing contention-based synchronous approaches such as S-MAC [9], T-MAC [1], and DW-MAC [11] synchronize the nodes sleep schedules such that neighboring nodes are awake at almost the same time. The schedule determines the moments when a node should be in active or sleep states. Neighbor nodes start exchanging packets only within the common active time and the CSMA/CA is used for channel access in the active time. The key advantage of this approach is that the sender does not need long preamble and the receiver reduces idle listening time. However, these protocols introduce extra synchronization overhead, and the throughput of the network declined due to channel contention. In contrast to the aforementioned duty-cycle protocols, TDMA-based solutions [12][13] establish a contention-free transmission schedule where each node is assigned one or multiple time-slots. Nodes are allowed to transmit only in the allocated time-slot(s). So time synchronization is a key part in the TDMA-based MAC protocol. Furthermore, a node which is not the sender or intended receiver in the current slot can sleep to conserve energy. These protocols usually can achieve an ultra-low duty-cycle (e.g. Dozer [3] achieves radio duty cycle in the magnitude of.2% ) and be more energy efficient. In low duty-cycle data-collection sensor networks, time synchronization between nodes is realized by periodically beacon dissemination from sink to leaf nodes in the data-gathering tree. The interval between two adjacent time synchronization is needed to be investigated. On one hand, the shorter the beacon interval is, the more synchronization cost comes out due to frequent beacon sending and receiving. On the other hand, increasing beacon interval leads to a larger guard time and longer idle listening due to clock drift. This is because the guard time has to be large enough to accommodate the synchronization errors, while these errors are affected by the frequency deviation (skew) between two clocks and thus accumulated over time. Therefore, a longer beacon interval will lead to larger synchronization errors and hence a larger

guard time is necessary. So we have a question: what is the optimal beacon interval that can minimize the overall energy consumption? The works on the optimization of MAC protocol parameters are quite limited. By tuning the configurations to the optimal values, the performance of a MAC protocol can be remarkably enhanced. T-MAC [1] adaptively changes the duration of the active periods to avoid idle listening according to the traffic loads, but its optimization problem is not considered. X-MAC [6] presents an approximately optimal algorithm for adapting the receivers duty-cycle, that is, both the sleep time and listening time are configured to minimize the expected energy consumption per packet depending on different traffic load. For the low duty-cycle synchronous MAC, the optimal channel polling period for both LPL and SCP is proposed in [2]. Compared to LPL, the optimized SCP can not only achieve the lower bound of energy consumption for hybrid MAC, but also adapt well to variable traffic. However, all these optimization methods are customized for their specific MAC protocols and therefore are just one part of these protocols. In this paper, we analyze the optimal beacon interval for TDMA-based MAC protocols. Different from the previous works, our approach is suitable for most of the TDMA-based MACs with ultra-low duty-cycle in data-collection applications (such as Dozer [3], Koala [14] and so on), not just optimized for one specific MAC protocol. To demonstrate the effects of the optimization, a TDMA-based MAC protocol using the optimal beacon interval called Opt-TDMA is presented. To the best of our knowledge, this is the first work which is specialized to discuss the optimization problem of beacon interval in TDMA-based MACs for low duty-cycle datacollection WSNs. II. SYSTEM MODEL 1) Energy Consumption: Fig.1 illustrates a basic link communication model in WSN. A and B are two sensor nodes and can hear each other directly. A is a master node and B is a slave node. Node B (slave node) periodically generates data packets and sends them to node A while node A (master node) periodically sends beacons to synchronize node B. These two nodes utilize the same coordinated duty-cycle. The data report interval and beacon interval are different. Node A and B operate at a low duty-cycle mode, being active and asleep periodically to save energy. In order to wake up at almost the same time, node A and B must rendezvous at a common time point. Due to the synchronization errors between two nodes which is caused by clock drift, the receiver (A or B) has to turn on its radio in advance with a short time (advanced time) to listen to the channel, and check if a beacon or packet is coming for a longer time (guard time). If a SFD (start-offrame delimiter) byte of the head of a beacon/data packet is detected, the radio will be kept on until the reception is over. Otherwise, the radio will be turned off immediately. In this paper, our analysis focuses on the energy consumption of the radio, without considering other components (which is similar to [3]), such as CPU and sensors. There are four Beacon Guard time Advance time Beacon Data Sending Receiving Listening Fig. 1. Basic communication model Sink Fig. 2. Data-gathering tree stable radio states: listening, sending, receiving and sleeping. The power consumptions of these four states are P l, P s, P r and P sleep, respectively. The total energy consumption can be modelled as: E = E l + E s + E r + E sleep = P l t listen + P s t s + P r t r + P sleep t sleep (1) where t s, t r, and t sleep are the time in sending, receiving and sleeping, receptively. t listen is the time for listening to the channel before the packets arrive. A. Topology and Frame Structure The wireless sensor network we studied is based on the tree topology which is widely used in data-collection applications, called data-gathering tree (Fig.2). The tree topology can be easily formed by a tree-base or hierarchical routing protocol (e.g. CTP and LEACH). In this data-gathering structure, data are collected from nodes to the sink and beacons are disseminated from the sink to all nodes. Each node may play two independent roles, one as a parent and one as a child. A parent node receives data packets when its children nodes are allowed to upload data, and a child node receives the synchronization beacons from its parent. To simplify our analysis, we present a TDMA-based MAC for data and beacon transmission. In our TDMA-based protocol, the whole timeline is divided into sequential frames and a frame is divided into 2n slots, A B

time for sending/receiving a byte. Because the parent node will not only transmit the data packets sensed by itself, but also forward the data packets from its children nodes, so the coefficient of the time spent in sending data packets is (n + 1). Therefore, the total time of sending state satisfies the following equation: t s = L sync t B r sync + (n + 1)L data t B r data (3) Fig. 3. The frame structure where the former n slots are beacon slots (called beacon subframe) and the latter n slots are data slots (called data subframe). The last slot of beacon slots is allocated for nodes to join the network, which is called the access slot. Fig.3 illustrates the structure of the frame. Each node is assigned a unique time slot number (ID) SlotNum and its beacon and data sending time slots are determined based on SlotNum. The ID of beacon sending time slot is SlotNum and ID of the data sending time slot is SlotNum + n. The receiving time slots of beacon and data packets are determined by the slot number of its parent and child slots. When the node is in the beacon sub-frame, it receives the beacon packet at its parents beacon-sending slot and broadcasts the beacon at its own beacon-sending slot. When the node is in the data subframe, it listens to the channel at its own children nodes datasending slots for a guard time and sends data at its own datasending slot to its parent. All the nodes listen to the channel at the access time slot for network maintenance. The nodes are in sleep mode in all other time slots. III. OPTIMAL BEACON INTERVAL In this section, we calculate the expected power consumption (per node) based on the data-gathering tree topology, and then analyze the optimal beacon interval for the TDMA-based protocol (Opt-TDMA). In a tree-based topology, the parent node sends beacons to the children nodes and the children nodes send data packets to the parent node. Considering a parent node with n children nodes, the energy consumption of the parent node is the sum of energy spent in each state, expressed as the following equation (P sleep is far less than other components, thus not considering). E = E l + E s + E r = P l t listen + P s t s + P r t r (2) As a parent node, the time spent on sending state includes sending beacons (synchronization packets) and data packets. The time spent on broadcasting the beacons is L sync t B r sync and the time spent on sending data packets is (n + 1) L data t B r data, where L sync and L data are the beacon length and data packet length, respectively. r sync and r data are the beacon rate and data packet rate, respectively. t B is the The time spent on receiving state also includes receiving beacons and data packets. The parent node acts as a child node of its own parent, so it receives beacons from its own parent. The time spent on receiving the synchronization packets is L sync t B r sync, and the time spent on receiving data packets from its children nodes is nl data t B r data. Therefore, the total time of receiving state satisfies the following equation: t r = L sync t B r sync + nl data t B r data (4) When the node is going to receive the message, it has to turn on the radio in advance due to clock drift. So the total time of listening to the channel is: t listen = t guard r sync + nt guard r data (5) where t guard is a guard time with a length that is large enough to tolerate a given maximal clock drift. According to [2], the guard time is affected by the beacon interval T sync and the clock drift rate θ, which is decided by the frequency derivation of two clocks and also affected by environmental factors such as temperature and voltage. The maximum clock difference between two nodes is: t diff = 2T sync θ (6) where the coefficient of two reflects the worst case when each nodes clock drifts in the opposite direction. Because the sender does not know which direction its clock drifts with regard to the receiver, it needs to put the guard time in both directions, making the total guard time to be twice the maximum clock difference [2]. The guard time can be given as follows: t guard = 2t diff = 4T sync θ (7) Substituting Eq.3,4,5,7 into Eq.2, we can obtain the energy consumption as: E = r sync [t guard P l + L sync t B (P s + P r )]+ r data [nt guard P l + L data t B ((n + 1)P s + np r )] = 1 T sync L sync t B (P s + P r ) + T sync r data n4θp l + r data L data t B ((n + 1)P s + np r ) + 4θP l (8)

TABLE I SYMBOLS AND VALUES USED IN RADIO ENERGY ANALYSIS Symbol Meaning CC242 T data Data packet period varying r data Data packet rate (1/T data ) Varying T sync SYNC packet period Varying r sync SYNC packet rate (1/T sync) Varying θ Clock drift 5 ppm L data Data packet length 5 B L sync SYNC packet length 36 B t B Time to Tx/Rx a byte 32 µs P l Power in listening 56.4 mw P s Power in sending 52.2 mw P r Power in receiving 56.4 mw Optimal sync peroid Tsync (s) 6 5 4 3 2 1 n=1 n=2 n=3 n=4 n=5 n=6 Assuming that the lengths of a beacon and a data packet are fixed, we can see from Eq.8 that the expected power consumption of a parent node varies with its children size n, data rate r data, clock drift rate θ and the synchronization period T sync. The trade-off of power consumption considering T sync is: the increasing of T sync reduces the energy cost of sending/receiving beacons, but increases the length of guard time and hence the cost on idle listening. Therefore, the optimal value of beacon interval is the one to provide the best tradeoff of these two parts of overhead and minimize the overall energy consumption. We can derive the optimal value of beacon interval by solving the following equation. de dt sync = (9) Substituting Eq.8 into Eq.9, we have the optimal beacon interval as: T sync = L sync t B (P s + P r ) n4θp l r data (1) IV. PERFORMANCE EVALUATION In this section, we provide the performance evaluation of the proposed optimal beacon interval (Opt-TDMA) scheme, in comparison with SCP and a TDMA-MAC without optimized beacon interval (non-opt TDMA). SCP is a synchronous MAC based on scheduled channel polling and contention-based data transmission, which has an average duty-cycle typically less than 1%. The beacon interval of non-opt TDMA is configured as 5 s. The energy consumption of the network is measured by calculating the radios operating time in different modes as sending, receiving and listening on data packets and beacons. The default parameters of evaluation for all those MACs are shown in Table I. A. Optimal Values of Beacon Interval Fig.4 depicts the optimal beacon interval of the Opt-TDMA. As Fig.4 demonstrates, the optimal beacon interval is positively correlated to data interval but negatively correlated to the network size. The reasons are as follows. There is 5 1 15 2 25 3 Fig. 4. Optimal beacon interval for different network size n a tradeoff between two parts of energy consumption: the energy consumed in sending/receiving beacons, and the energy consumed in idle listening during guard time. When the data generation interval increases, the frequency of idle listening for data packets will decrease. Therefore, too frequent synchronization is not necessary since we can simultaneously reduce beacon sending/receiving frequency at the cost of only slightly increasing the power consumption on idle listening during guard time. From the figure we also notice that the optimal beacon interval increases as the number of children nodes decreases. This is because fewer children nodes lead to less data packets to be received in condition of a given data report interval. According to the analysis above, a slower data reception rate will render the optimal beacon interval to be longer. Therefore, the optimal beacon interval increases with fewer children nodes. B. Comparison with non-opt TDMA and SCP Fig.5 and Fig.6 demonstrate the average power consumption of sensor node for non-opt TDMA, Opt-TDMA and SCP depending on the data generation interval with different clock drift rate. From these two figures, we can see that the average power consumption increases as the data generation interval decreases. This is because a faster data packet rate leads to heavy traffic load, and therefore more power is consumed on data transfer. we can also see that the power consumption is getting larger when the clock drift rate is higher. The reason is that a more serious clock drift rate introduces larger synchronization errors, thus a longer guard time is required to accommodate these synchronization errors, then will enlarge the overhead of idle listening and the overall power consumption. As shown in Fig.5, the difference of power consumption between non-opt TDMA and Opt-TDMA increases as the clock drift rate or data rate increases. As analysis above, a more serious clock drift rate will enlarge the idle listening overhead. Then more energy can be saved adopting Opt-TDMA when the clock drift rate becomes higher. On the other hand, the optimal beacon interval increases

Energy consumption per second (mw).12.1.8.6.4.2 θ=5 ppm, Opt TDMA θ=5 ppm, non Opt TDMA θ=1 ppm,opt TDMA θ=1 ppm, non Opt TDMA θ=15 ppm,opt TDMA θ=15 ppm, non Opt TDMA Energy consumption per second (mw).25.2.15.1.5 n=1,opt TDMA n=1, non Opt TDMA n=2, Opt TDMA n=2, non Opt TDMA n=3, Opt TDMA n=3, non Opt TDMA 5 1 15 2 25 3 5 1 15 2 25 3 Fig. 5. Power consumption for Opt-TDMA and TDMA with different clock drift Fig. 7. Power consumption for Opt-TDMA and TDMA with different network size Energy consumption per second (mw).12.1.8.6.4.2 θ=5 ppm, Opt TDMA θ=5 ppm, SCP θ=1 ppm,opt TDMA θ=1 ppm,scp θ=15 ppm,opt TDMA θ=15 ppm,scp Energy consumption per second (mw).25.2.15.1.5 n=1,opt TDMA n=1, SCP n=2, Opt TDMA n=2, SCP n=3, Opt TDMA n=3, SCP 5 1 15 2 25 3 5 1 15 2 25 3 Fig. 6. drift Power consumption for Opt-TDMA and SCP with different clock Fig. 8. Power consumption for Opt-TDMA and SCP with different network size with the decreasing of the data rate (as shown in Fig.4). Therefore, the synchronization rate reduced significantly with the decreasing of the data rate, reducing the gap of energy consumption between Opt-TDMA and non-opt TDMA. In Fig.5 and Fig.6, we also find out that the power consumption of Opt-TDMA is lower than non-opt TDMA and SCP. The reasons are as follows. (1) Compared with non-opt TDMA (Fig.5), Opt-TDMA uses the optimal beacon interval and minimizes the total power consumption of synchronization and idle listening. Although non-opt beacon intervals can minimize either synchronization cost or idle listening overhead, they can not minimize the overall power consumption. (2) Compared with SCP (Fig.6), although SCP also optimizes the channel polling period, the channel access is based on contention and not able to completely avoid collisions, thus wasting more energy. Fig.7 and Fig.8 display the power consumption of these three protocols depending on data report interval with different children number. As shown in Fig.7, the power consumption of Opt-TDMA is lower than non-opt TDMA in any data generation interval, especially in the high data rate. So the Opt- TDMA is more suitable in massive data flow network. From Fig.8 we can see that the difference of power consumption between Opt-TDMA and SCP increases when the number of children nodes increases. For example, the difference between two blue lines (n is 3) is shown obviously larger than two red lines (n is 1). The reason is that collisions occur more frequently in SCP when more children nodes have packets to be transmitted, while in Opt-TDMA, it is collision-free and no contention occurs due to the scheduled data transmission. Therefore, the power consumption of Opt-TDMA is smaller than SCP, especially when the network size is large. V. CONCLUSION AND FUTURE WORK In this paper, we investigate the effect of beacon interval length on the energy efficiency of TDMA-based MAC in low duty-cycle sensor networks. By providing the best tradeoff of the energy consumption between synchronization cost and idle listening overhead, the optimal length of beacon interval is analyzed. To implement this optimal solution, a

TDMA-based MAC (Opt-TDMA) is presented in the context of data-collection sensor networks. By configuring the beacon interval to its optimal value according to the data packets rate and network size, Opt-TDMA can reduce the overall power consumption of both sending/receiving beacons and data packets. The experiment results show that Opt-TDMA is more energy-efficient than non-opt TDMA and SCP, by using optimal beacon interval and contention-free transmission. Opt- TDMA can save more energy especially when the network size is large and the clock drift rate is high. That is, Opt-TDMA is more suitable for those large-scale networks and those networks in worse environments. Our analysis on the optimal beacon interval can be also extended to all synchronous MAC protocols. Therefore, the future work includes applying this strategy into other synchronous MAC protocols and investigating their performances. REFERENCES [1] Yingchao Han, Hongmei Li, and Jinghui Qiu. The analysis and summary about energy saving technologies of wireless sensor network. In Electronic and Mechanical Engineering and Information Technology (EMEIT), 211 International Conference on, volume 2, pages 883 885. IEEE, 211. [2] Wei Ye, Fabio Silva, and John Heidemann. Ultralow duty cycle mac with scheduled channel polling. In Proceedings of the 4th international conference on Embedded networked sensor systems, pages 321 334. ACM, 26. [3] Nicolas Burri, Pascal Von Rickenbach, and Roger Wattenhofer. Dozer: ultra-low power data gathering in sensor networks. In Information Processing in Sensor Networks, 27. IPSN 27. 6th International Symposium on, pages 45 459. IEEE, 27. [4] Pei Huang, Li Xiao, Soroor Soltani, Matt W Mutka, and Ning Xi. The evolution of mac protocols in wireless sensor networks: A survey. Communications Surveys & Tutorials, IEEE, 15(1):11 12, 213. [5] Joseph Polastre, Jason Hill, and David Culler. Versatile low power media access for wireless sensor networks. In Proceedings of the 2nd international conference on Embedded networked sensor systems, pages 95 17. ACM, 24. [6] Michael Buettner, Gary V Yee, Eric Anderson, and Richard Han. X-mac: a short preamble mac protocol for duty-cycled wireless sensor networks. In Proceedings of the 4th international conference on Embedded networked sensor systems, pages 37 32. ACM, 26. [7] Prabal Dutta, Stephen Dawson-Haggerty, Yin Chen, Chieh-Jan Mike Liang, and Andreas Terzis. Design and evaluation of a versatile and efficient receiver-initiated link layer for low-power wireless. In Proceedings of the 8th ACM Conference on Embedded Networked Sensor Systems, pages 1 14. ACM, 21. [8] Lei Tang, Yanjun Sun, Omer Gurewitz, and David B Johnson. Pw-mac: An energy-efficient predictive-wakeup mac protocol for wireless sensor networks. In INFO- COM, 211 Proceedings IEEE, pages 135 1313. IEEE, 211. [9] Wei Ye, John Heidemann, and Deborah Estrin. Medium access control with coordinated adaptive sleeping for wireless sensor networks. Networking, IEEE/ACM Transactions on, 12(3):493 56, 24. [1] Tijs Van Dam and Koen Langendoen. An adaptive energy-efficient mac protocol for wireless sensor networks. In Proceedings of the 1st international conference on Embedded networked sensor systems, pages 171 18. ACM, 23. [11] Yanjun Sun, Shu Du, Omer Gurewitz, and David B Johnson. Dw-mac: a low latency, energy efficient demandwakeup mac protocol for wireless sensor networks. In Proceedings of the 9th ACM international symposium on Mobile ad hoc networking and computing, pages 53 62. ACM, 28. [12] Sameh Gobriel, Daniel Mosse, and Robert Cleric. Tdmaasap: Sensor network tdma scheduling with adaptive slot-stealing and parallelism. In Distributed Computing Systems, 29. ICDCS 9. 29th IEEE International Conference on, pages 458 465. IEEE, 29. [13] Wen-Zhan Song, Renjie Huang, Behrooz Shirazi, and Richard LaHusen. Treemac: Localized tdma mac protocol for real-time high-data-rate sensor networks. Pervasive and Mobile Computing, 5(6):75 765, 29. [14] Razvan Musaloiu-E, C-JM Liang, and Andreas Terzis. Koala: Ultra-low power data retrieval in wireless sensor networks. In Information Processing in Sensor Networks, 28. IPSN 8. International Conference on, pages 421 432. IEEE, 28.