An Adaptable Energy-Efficient Medium Access Control Protocol for Wireless Sensor Networks

Transcription

1 An Adaptable Energy-Efficient ium Access Control Protocol for Wireless Sensor Networks Justin T. Kautz 23 rd Information Operations Squadron, Lackland AFB TX Barry E. Mullins, Rusty O. Baldwin, Scott R. Graham Air Force Institute of Technology, Wright-Patterson AFB OH {Barry.Mullins, Rusty.Baldwin, Abstract We propose a medium access control protocol for wireless sensor networks (WSN) called Adaptive sensor ium Access Control (AMAC), which is based on the Sensor ium Access Control (S-MAC) protocol [1]. Since WSNs are energy constrained, the lifetime of the network must be increased by making it as energy efficient as possible. Whereas S-MAC uses a fixed duty cycle for sleeping, AMAC adapts to traffic conditions by incorporating multiple duty cycles. Under a high traffic load, AMAC has a short duty cycle and awakes more often. Under a low traffic load, AMAC has a longer duty cycle and awakes infrequently. The AMAC protocol is simulated in OPNET Modeler. Analysis indicates that AMAC uses 15% less power and 22% less energy cost per byte than S-MAC with a tradeoff in twice the latency. For an application insensitive to latency, the AMAC protocol offers an extended lifetime. 1. Introduction Wireless Sensor Networks (WSN) offer a powerful tool in information collection which can be deployed anywhere remotely with little to no maintenance required. A WSN collects and relays information until its energy is exhausted. Due to the inability to replenish the WSN with power after deployment, it becomes necessary to extend the lifetime of the network as much as possible to increase its effectiveness. Creating an energy-efficient, scalable medium access control (MAC) protocol is a vital part of this. There have been many approaches to creating energy-efficient wireless protocols; some are briefly discussed in Section 2. Creating an energy-efficient MAC is one technique to save energy. This paper describes how we modified an existing MAC protocol to make it more energy-efficient. The description of AMAC and the experiment to test it is provided in Section 3. Experimental results are discussed in Section 4. Section 5 provides the final conclusions and recommendations for future work. 2. Background WSNs are a logical extension of wireless networks, though with different priorities on performance such as throughput, latency, bandwidth, and energy consumption. One of the key differences between regular wireless networks and WSNs is the limited lifetime of a WSN. Normally, nodes in a WSN are powered by batteries and deployed to remote locations where it is not possible to change the battery. Such networks are deployed ad hoc with a limited communication range implying multi-hop routing is required to transfer data across the network [2]. Since the energy supply is limited, energy consumption is one of the primary metrics of interest when designing a WSN. Nodes can benefit from lower-level MAC innovations aimed more towards maximizing efficiency in point-to-point communications. Some MAC innovations take advantage of the fact that a node in a WSN spends the majority of its time idly listening to the medium instead of transmitting or receiving data. To save energy, the transceiver can be turned of for a time, called sleep cycles. The S-MAC protocol uses such an approach [1][3]. U.S. Government Work Not Protected by U.S. Copyright 1

2 A node will often receive transmissions intended for another node. To save energy, the transceiver could be turned off while nearby nodes are transmitting. Both the Power Aware Multi-Access protocol with Signaling (PAMAS) protocol [4] and S- MAC exhibit this behavior. Other approaches use a schedule-based as opposed to contention-based algorithm to maximize communication efficiency, e.g., the Traffic Adaptive ium Access (TRAMA) Protocol [5]. 3. Methodology S-MAC is energy efficient compared to standard wireless protocols like [1][6], yet it was not designed to be adaptable or flexible. The duty cycle of S-MAC, the duration of time it is awake compared to the entire cycle, is set before the network is deployed. The application drives the selection of the appropriate duty cycle; once set, the duty cycle is not varied. This static configuration limits the flexibility of S-MAC. The goal of our research is to make S-MAC more energy efficient by allowing it to dynamically adjust to various network conditions. We show that the average energy savings is, on average, better for this dynamic protocol, since it adjusts to traffic conditions and sleeps more during periods of inactivity. This behavior increases latency during these periods of low activity AMAC Protocol Design AMAC is similar in functionality to S-MAC, and therefore few changes are needed to the original protocol. AMAC changes its sleep duration (duty cycle) based upon recent traffic trends while maintaining communication. Duty cycles of inverse powers of 2 (e.g., 2 -n ) maintain common periods of activity as shown in Figure 1. An AMAC node informs neighboring nodes of changes which allows neighboring nodes to properly schedule communication with the node. Convergence of node schedules is on the order of minutes. This process continues throughout the operation of the network. Bounds are placed on the duty cycle to prevent it from waking up too often or sleeping too long between periods of activity. The first negates the effect of having an energy-efficient protocol, and the second decreases the response to network traffic increase after periods of inactivity. Moreover, synchronization must be maintained, and therefore an upper bound of 1/4 and lower bound of 1/64 were chosen for the duty cycle. Duty Cycle 1/2 1/4 1/4 1/16 1/32 = Listen Period = Sleep Period Figure 1. Duty cycle comparisons. Common periods of activity are maintained between neighbors. AMAC adapts to varying traffic conditions by maintaining a variable called currentusage which uses exponential forgetting. currentusage is affected by whether or not the listen period/slot of the duty cycle is used. A used slot means that data was either sent or received. The current usage value is Xn Xn-1 (1- ) ( ) 1 if slot was used, (1) 0 if slot was not used where is the sensitivity level. The sensitivity level can be varied between zero and one. The closer to zero, the more sensitive AMAC is to change; the closer to one, the less sensitive AMAC is to change. When the sensitivity is one, it behaves exactly like S-MAC. The different sensitivity levels of AMAC are discussed further in Section 3. Duty cycle transition points were determined via pilot studies and provide overlap with adjacent duty cycle speeds to minimize rapid fluctuation of duty cycles. The duty cycles are derived as 1 1 T DC min, max, DC if X n DC T 1if Xn 3DC, (2) 0 otherwise where DC is the duty cycle. With the duty cycle information from neighboring nodes, three optimizations can be made. The first involves the SYNC period; a node only wakes up for the SYNC period during the slot of its slowest neighbor. This prevents a node from sending a SYNC when all of its neighbors may not be listening. The second optimization is when a node transmits to a 2

3 node on a slower duty cycle. If the slower neighbor is not currently awake, the node does not send an RTS, thereby alleviating unproductive network traffic and the energy loss therein. The third optimization involves a slower neighbor transmitting to a faster neighbor. If the slower node fails to attain the medium, meaning it was not successful in sending an RTS or receiving the corresponding CTS, it reschedules to wakeup during the faster neighbor s next wakeup period. congestion and dependence on successful transmissions for currentusage can lead to network deadlock where each node eventually transmits at the slowest duty cycle. To avoid congestion, if a node s packet buffer exceeds an upper boundary, it enters a packet dump mode. When the packet buffer goes below a lower boundary it exits that mode. The boundaries are set to be 20% and 75% of the buffer size based upon initial tests. In packet dump mode, a node only acts as a receiver during adjacent slots. These are slots not synchronized with the next slowest duty cycle. Figure 2 shows an example with the 1/2 duty cycle. Additionally, that node counts every slot it hears an RTS or CTS as a used slot increasing currentusage which increases the duty cycle. This increases the transmission opportunity of that node allowing it to transmit faster and empty its buffer. Duty Cycle 1/2 1/4 = Common Slot = Adjacent Slot Figure 2. Adjacent slots illustrated 3.2. Experimental Design The evaluation technique is simulation using OPNET 11.0.A. A detailed model of AMAC was created using the S-MAC model as a baseline. A full factorial experiment with 20 repetitions was conducted. The first factor is traffic sensitivity,, with sensitivity levels of none (1.00), low (0.99), medium (0.98), and high (0.96). The high level, 0.96, was chosen so that at least two successful transmissions (used slots) will cause a transition to a faster duty cycle regardless of X n current value. For example, assuming X n-1 is zero, (1) can be used to derive a reasonable high traffic sensitivity level by equating the upper boundary for the slowest duty cycle (3/64) to two used slots (with 15 intermediate unused cycles): 3 st 01 used slot 15 unused slots 64 nd 2 used slot (3) The exponential forget factor is multiplied by the first success fifteen times in (3) since X n is based on the fastest duty cycle, there are fifteen slots of the fastest cycle between each slot of the slowest cycle. Therefore, the maximum for two successful sequential transmissions to cause a transition from the slowest duty cycle works out to be approximately The maximum for one successful transmission to cause a transition from the slowest duty cycle is the upper boundary for the slowest duty cycle, which is approximately The high level for the experiment, 0.96, is the midpoint between these two values, 0.97 and Therefore, at least two successful sequential transmissions will cause a transition from the slowest duty cycle. The medium and low levels were chosen so that it would require two and four times the amount of time as the high level to transition between duty cycles respectively. This means that the medium and low levels are two and four times less sensitive to change in traffic. The second factor is the packet interarrival period which is exponentially distributed and varied between low, medium and high, defined as 5, 10, and 20 packets per second respectively. The third factor is the duty cycle for burstiness which is exponentially distributed and varied between low, medium, and high, defined as 33%, 66%, and 100%. The period for the burstiness duty cycle is 9 minutes. This value is based upon the minimum time it takes to go from the fastest to the slowest duty cycle with the low sensitivity, 66% burstiness duty cycle and no usage. The node topology is the fourth factor and is varied between Topology 1 and Topology 2 to simulate two different variations of topologies. Topology 1 (Figure 3), the two source - two sink topology, has the highest expected power for an individual node at the intermediary node (node C). Topology 2 (Figure 4), a line network, has the longest expected latency. The 3

4 experimental factors and their levels are shown in Table Main Effects Plot for ETE Delay (S) 350 Mean of ETE Delay (S) Figure 3. Topology 1: Two hop network with two sources and sinks [3] 150 Figure 6. Mean end-to-end delay Figure 4. Topology 2: Ten node linear network with one source and sink Table 1 - List of factors and their levels Traffic Interarrival Period (S) Burstiness (duty cycle) Topology 5 33% Cross 10 66% Line % 4. Analysis and Results The analysis explores the fundamental tradeoffs in end-to-end latency and energy costs for sending messages using the AMAC protocol. The mean power and end-to-end delay for the different levels of sensitivity are plotted in Figure 5 and Figure 6. The overall mean for all levels is indicated by the bisecting line. The power in Figure 5 shows an overall average energy savings of 15% for AMAC (low, med, high) over S-MAC (none). This was not as much as expected; however, S-MAC used a 6.75% duty cycle which is close to optimal for the interarrival rates [1]. Furthermore, the burstiness of traffic did not allow for long periods of inactivity. As Figure 7 and Figure 8 show, as traffic decreases, the AMAC energy savings become more pronounced. Figure 6 shows that S-MAC has the least amount of end-to-end delay., medium and high sensitivity have two to three times more latency than S-MAC. The streamput response in Figure 9, which is the number of bytes delivered from source to destination per unit time, shows that the original S-MAC is better. Nodes using AMAC on faster duty cycles have to queue packets destined for slower nodes which increases the end-to-end delay and also lowers streamput. S-MAC has a constant duty cycle and therefore has no such problems and therefore has higher streamput Main Effects Plot for Power (mj/s) 2.75 Interaction Plot for Power (mj/s) Burstiness Mean of Energy (mj/s) Mean Figure 5. Mean power Figure 7. Mean power for interaction of sensitivity and burstiness 4

5 Interaction Plot for Power (mj/s) Main Effects Plot for Throughput (Bytes/S) Mean Interarrival Mean of Throughput (Bytes/S) Figure 8. Mean power for interaction of sensitivity and interarrival Figure 10. Mean throughput Main Effects Plot for Energy/Stream (µj/byte) Mean of Streamput (Bytes/S) Main Effects Plot for Streamput (Bytes/S) Mean of Energy/Stream (µj/byte) Figure 9. Mean streamput Throughput is defined as the amount of traffic delivered between adjacent nodes only. Figure 10 demonstrates that AMAC exhibits better throughput for all sensitivity levels. This is due to AMAC s ability to adapt its duty cycle; each node is awakened and transmits based on the characteristics of recent traffic. AMAC provides more throughput due to traffic adaptation, but provides less streamput due to temporary differences in duty cycle levels. Figure 7 and Figure 8 show that a low sensitivity provides the best energy savings. This assumption is valid as long as raw power is the most important variable. However, streamput and throughput, as shown in Figure 9 and Figure 10, show that the best level is medium despite the fact that the medium level exhibits the worst energy value for AMAC. The average energy cost of transmitting a byte is calculated by taking the ratio of power to streamput and power to throughput give energy/stream (Figure 11) and energy/link (Figure 12) measured in J/Byte. Figure 11. Mean energy/stream cost AMAC provides an energy cost per byte savings of 22% over S-MAC. Figure 11 shows that the low sensitivity level yields the best energy cost for delivering a packet from source to destination. However, Figure 12 shows the medium sensitivity level is better than low sensitivity in energy cost per byte in communication between neighbors. ium and low sensitivity are the two best candidates for use in the AMAC protocol since high sensitivity is noticeably worse in energy/stream and energy/link. Based on an analysis of the computational effects of each sensitivity level, the medium sensitivity provides the best protocol performance when considering all factors. 5. Conclusion The goal of this research is to modify an existing protocol to make it more adaptive and therefore more energy efficient. AMAC is the latest generation energy-efficient MAC by being able to adapt to traffic which enables higher energy savings. 5

6 The results are promising. AMAC performs better with a 15% decrease in overall network power, a 22% decrease in energy/byte and 4% more throughput. These came at the cost of 7% less streamput and double the latency. For applications which are not time sensitive and can accommodate some delay, AMAC offers a solution for extending the overall lifetime of a network. Mean of Energy/Link (µj/byte) Main Effects Plot for Energy/Link (µj/byte) Figure 12. Mean energy/link cost An example where AMAC will excel would be the deployment of sensors whose application or location precludes a small number of sensor events. For instance, ground motion sensors with the purpose of detecting vehicle or troop movements. Such sensors in certain locations would experience large periods of inactivity and AMAC has shown to have a faster growth in energy savings as traffic decreases (Figure 7, 8). Some issues to be addressed with AMAC are increase of latency and decrease of streamput. In addition, the energy savings and cost can be increased with further optimizations and testing various configurations. AMAC can be adapted to work with other protocol s optimizations, such as newer versions of S-MAC to ascertain whether energy-efficiency can be increased further. Also larger network configurations need to be tested with AMAC to determine how it performs under higher node and link density. Finally, how AMAC affects energy efficiency with mobility needs to be determined. 6. References [1] W. Ye, J. Heidemann, and D. Estrin, An energyefficient MAC protocol for wireless sensor networks, In 21st Conference of the IEEE Computer and Communications Societies (INFOCOM), Vol. 3, pp , June [2] I. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, "Wireless sensor networks: A survey," Computer Networks (Elsevier) Journal, vol. 38, no. 4, pp , Mar [3] W. Ye, J, Heidemann, and D. Estrin, "ium Access Control With Coordinated Adaptive Sleeping for Wireless Sensor Networks", IEEE/ACM Transactions on Networking, Vol. 12, No. 3, June 2004, pp [4] S. Singh and C.S. Raghavendra, "PAMAS: Power aware multi-access protocol with signaling for ad hoc networks", ACM Comput. Commun. Rev., vol. 28, no. 3, pp. 5-26, July [5] V. Rajendran, K. Obraczka, J.J. Garcia-Luna-Aceves, "Energy-Efficient, Collision-Free ium Access Control for Wireless Sensor Networks", in Proc. SENSYS03, [6] IEEE, Wireless LAN medium access control (MAC) and physical layer specifications, ANSI/IEEE standard , 1999 Edition, Acknowledgements We thank the reviewers for their constructive and insightful comments. The views expressed in this paper are those of the authors and do not reflect the official policy or position of the United States Air Force, Department of Defense, or the U.S. Government. 6