A Distributed Power Management Policy for Wireless Ad Hoc Networks

Transcription

1 A Distributed Poer Management Policy for ireless Ad Hoc Netorks Carla F. Chiasserini Dipartimento di Elettronica Politecnico di Torino Torino, Italy Ramesh R. Rao Center for ireless Communications University of California, an Diego La Jolla, CA, UA Abstract This paper presents a poer management scheme that maximizes energy saving in ireless ad hoc netorks hile still meeting the required quality of service (Qo). e assume that battery-poered devices can be remotely activated by a aking-up signal using a simple circuit based on RF tag technology. In this ay, devices that are not currently active may enter a sleep state and poer up only hen they have pending traffic. Radio devices select different time-out values, so called sleep pattern, to enter various sleep states depending on their battery status and quality of service. Performances of the proposed policy are derived by simulation for a simple ad hoc netork scenario. Results sho the achieved tradeoff beteen poer saving and traffic delay. I. INTRDUCTIN In ireless communication netorks, energy consumption is a major performance metric: loer the energy consumption, longer the run-time of communication devices. Thus, there is an increasing interest in both lo poer RF devices and energy efficient communication protocols. ne of the most common techniques to reduce poer consumption in ireless communication is discontinuous reception, i.e., users may poer don and turn on their receiver again at a particular time instant in the future [], [2], [3]. In IEEE 802. [4] users in poer saving mode ake up in correspondence ith a broadcast transmission from the base station that notifies hich terminals have pending data. The same concept is used in the MAC protocol presented in [5]. In HIPERLAN [6], the ireless LAN standard specified by ETI (European Telecommunications tandards Institute), radio nodes that need to save poer, so-called p-savers, communicate their on sleep-aake schedule to the so-called p- supporter node. The p-supporter queues all the packets destined to the p-savers and transmits these packets during the p-savers active time. In this paper e introduce a completely distributed poer management technique for ad hoc netorks, hich aim to maximize energy saving in battery-poered devices hile satisfying the required traffic quality of service. e consider that a node can be in different states depending on ho many parts of the device circuitry are poered off. e assume that are sleep states, hile corre- This ork as supported by NF under grant CCR and by the Italian National Research Council. sponds to the state in hich the node is active, i.e., it is able to transmit/receive. Each sleep state is characterized by a certain poer consumption and a delay overhead; deeper the sleep state, less the poer consumption and longer the time to ake up. It is clear that longer the time spent by the nodes in sleep state, greater the energy saving is. Thus, in our scheme user devices do not ake up unless they are remotely activated by other nodes. To make this feasible e need a method to remotely activate a node through a RF (Radio Frequency) signal and such an operation must have a negligible energy cost. A possible scheme of the sitch that can be used to ake up a device hile being in sleep state, is presented in ect. II. e also consider a poer management policy implemented at the netork nodes, hich determines the state of operation of the various components in the node device during idle time. everal policies have been proposed in the literature [7], [8], [9] ith the aim to minimize both performance degradation of electronic devices and poer consumption; hoever, no policy has been proposed such that takes into account netork protocol performances and traffic Qo. The poer management technique presented in this paper is described in detail in ect. III, hile performances are presented in ect. IV. Conclusions and directions for further research are dran in ect. V. II. REMTE ACTIVATIN F THE NETRK NDE RF tags technology offers good examples of lo poer or totally passive devices that use the received RF poer to supply the logic and the transmission part of the circuit [0], [], [2]. RF tags have been used as transmitter/receiver devices (transponders) for remote localization and identification of animals, cars and other kinds of items typically over short-ranges [2], [3]. Based on RF tags technology, e can develop a sitch that can be used to remotely activate a radio device hile being in sleep state. In this ay, nodes are oken up hen necessary rather than getting active periodically to verify hether there is pending traffic. The schematic representation of the sitch, so-called RA

2 6 P Remote Activated itch (RA) Receiver tandard Receiver/ Transmitter Logic 3 2 Poer tatus Battery Fig.. cheme of the communication device circuit. 4 5 Device Electronics (Remote Activated itch), is shon in Figure. henever a node becomes idle, it enters a sleep state, i.e., the standard receiver/transmitter is turned off as ell as part of the device electronics. aking-up signals are received and demodulated by the RA, then the signal information is passed to the logic circuit that detects the sequence. If the received sequence matches the device s sequence, it turns on the standard receiver. Notice that the RA receiver may be either totally passive (e.g., an amplitude demodulator) or supplied by the battery source through connection. III. THE PER MANAGEMENT CHEME Poer management policies are used to make decisions on the state of operation of electronic systems such as portable computers and radio communication devices. In the field of computer systems, several poer management techniques have been proposed [7], [8], [9]. The objective of our scheme is to implement a simple poer management policy at the netork nodes, hich guarantees a poer saving alays greater than zero. The adopted policy can queue service requests and adapt to different needs and characteristics of the nodes. Recall that a device can be in different states here,.., are sleep states and corresponds to the state in hich the node is active, i.e., it is transmitting/receiving. Each sleep state is characterized by a certain poer consumption, denoted by ( =,.., ), and a delay overhead, denoted by ( =,.., ). e have that and "!#. Let us define $ as the poer cost of the transition from ( state (&%' ) to state, and ) as the minimum time that a node has to spend in sleep state to obtain a positive energy gain. e derive ) from the folloing formula, ) % $ $ 7 E +*-, /.0 2 *3, /. 54 /.0 *-, 0/.0 /.0 () For the sake of simplicity the cost associated ith transitions from state to state 7-8:9<;=7 >?A@CBDBDBD@FE#G is included in the cost of the back transitions to state. here the left term is the energy gain obtained from being in state rather than in state 4 and the right term is the additional cost due to transition from to ith respect to 4 ) transition from N to. Thus e have, %IHKJML *3, $ /. 54 /..0 *,.0 $.0 (2) e assume that nodes select different sleep patterns QR TCU<% V R TCU R TXU Y!#XZ R TCU ([\%] 3 _^ ) depending on their battery status and required quality of service. The values R TCU (\% ) are the time-out values b that determine hen nodes enter sleep i state (`%a ). For [c%d e^ and %gfh, e impose the folloing constraint =!5&j R TCU R TCU 4lk+ (3) here k is a system parameter such that k+ j ) n. Thus, henever a node enters a sleep state ith %mfh, it can move to state only after a time period equal to k. If a node has little battery capacity left, it ill use a vector QR TXU ith small values for R TXU to quickly enter deep sleep states. In this case, the node ill save energy, hoever, the delay penalty to pay may be significant. Instead, if a node is concerned mainly about traffic quality of service, it ill use a vector Q R TCU ith higher and higher values for R TCU as decreases such that it ill enter deep sleep states less rapidly. No, let us consider the generic node o. e define p_q as the time spent by o in sleep state, i.e., the time elapsed from the time instant hen node o passed from state to state. e assume that o is assigned a aking-up signal (also called aking-up sequence), denoted by rhq. henever a node in the ad hoc netork, let s say node s, has packets to forard to node o, it regularly transmits the akingup sequence r q until node o poers on. (In the folloing the inter-paging time is denoted by IP.) Upon detecting the signal rtq, node o, currently in state, akes up if and only if p q j R TCU 4lk (4) i.e., node o alays stays in a sleep state long enough to compensate for the transition poer cost. If the condition expressed in (4) is satisfied, node o becomes active after a time period equal to. As soon as o is active, it sends back to node s an acknoledgment and remains aake until it receives all the pending packets. A diagram of the proposed scheme is shon in Figure 2. IV. REULT Results are obtained for the netork scenario shon in Figure 3 ith four source nodes that generate traffic, and to destination nodes that receive only. The traffic load, denoted by

3 u ƒ, Elapsed time= IP? no Node i Packet for user j in state l Transmit ait ack x j Node j leep state l Receive x j Is τ j < no T (q) + Y l l ake up Keep sleeping TABLE I TIME-UT VALUE FR THE CNIDERED LEEP PATTERN (EXPREED IN TIME UNIT). k 4, 4 [\% [%cf [% ~ [\% R TXU R TXU 3 R TXU R TCU i ) ) ) ) y T () no Ack arrival? Transmit packets end ack Receive packets poer gain Fig. 2. Diagram of the poer management scheme. 0. D D 0.03 Fig. 4. Mean poer gain per node as a function of the traffic load for different sleep patterns. user node radio link Fig. 3. imulation netork scenario., is uniformly distributed among the source nodes. ource nodes generate packets according u"vm to a Poisson distribution ith mean value equal to, and each packet has a geometrically distributed length. The time scale is discretized into time intervals ith duration equal to the mean packet transmission time. e assume % and that the values of poer consumed in the various sleep states are normalized ith respect to the poer consumed by a node in receiving mode. Thus, e have: 0x%m, y% Dz, 0%, {%, and $ % 3 f * for }% X~. The values of the delay overhead are set equal to %b * for 5% X~. Also, e set the duration of the inter- paging time, IP, equal to 5 time units, the acknoledgment duration equal to 0.02 time units, and the transmission poer equal to 0 times the poer spent in receiving mode. Protocol performances are derived by simulation in terms of average poer gain and of average delay from the time instant hen a packet is generated to the time instant hen the packet is delivered to the destination node. Figures 4 and 5 present the performance metrics as functions of the average traffic load. Plots are derived for different sleep patterns, QR TXU ith [%, such that passing from QR U to QR x U, the time periods k ( % X~ ) increase and the time spent by the nodes in each sleep state becomes longer. Figures 4 and 5 sho that for sleep patterns QR U and QR U the packet delay is roughly constant hile the average poer gain decreases as the traffic load increases since nodes characterized by a loer traffic rate enter deep sleep states ith higher probability. For sleep patterns QR y U and QR x U, hich are

4 ! T () Fig. 5. Mean packet delay as a function of the traffic load for different sleep patterns. characterized by higher values of k ( %' X~ ), the packet delay gros significantly for high values of the traffic load alloing for a good poer gain. The presented plots can be used for system design: given a certain traffic delay, e can easily determine the poer gain that can be achieved or given the poer gain that nodes need, e have the mean delay that the traffic ill experience. Figure 6 presents the average total poer consumed by the nodes to transmit paging signals and acknoledgments, and hile being in receiving mode. The average poer consumption is derived as a function of the offered traffic load and for different sleep patterns. e notice that hen e consider patterns ith increasing values of k+ ( %ˆ 3 e~ ), the traffic becomes more bursty and the number of paging signals and acknoledgments decrease; thus, the average poer consumption decreases too. A similar effect takes place as the average traffic load gros. Hoever, poer consumption due to paging, acknoledgment transmission and time spent in receiving mode, is alays much less than the poer gain obtained through the presented scheme as e can see by comparing Figure 4 to Figure 6. Finally, e consider N FF traffic sources ith N and FF times that are random variables taking values according to a Pareto distribution:, r %cšœ tcr!5 ŠŽ e ` ; r j (5) An aggregation of such traffic sources results in a self similar traffic if the distribution of the N and FF time periods is heavy-tailed, i.e., Š{ f ; e take =0.75 and Š =.06 [4]. average poer consumption T () Fig. 6. Average poer consumption due to packet reception, paging, and acknoledgment transmissions poer gain T () 0.03 sleep patterns poer gain Fig. 7. Mean packet delay and poer gain obtained hen the packet generation process is assumed to be Pareto distributed. Figure 7 shos the obtained and poer gain as the sleep pattern varies. e have a behavior similar to the previous results presented in Figures 5 and 6: as e pass from QR U to QKR x U, and hence the time intervals k+ (#% X~ ) become longer, both traffic delay and poer gain increase. e notice that even in the presence of traffic sources ith Pareto distributed N and FF time periods, the proposed scheme allos for a significant poer gain at the expense of a limited additional delay.

5 u V. CNCLUIN AND FUTURE RK This paper proposed a ne poer saving scheme that allos nodes to ake up only hen necessary. Also, a simple poer management policy is implemented: nodes select a sleep pattern according to the desired Qo and their battery status, and enter a sleep state only after they have been idle for a time period equal to or greater than the corresponding time-out value. Results shoed the poer gain that can be achieved in the case of a simple netork scenario as the average packet delay varies. It as shon that even at high traffic load (namely, =0.9) a poer gain as high as 24% can be obtained. Future ork ill deal ith an ad hoc netork scenario here the proposed poer management scheme is applied. Nodes could be aake or asleep, and for each transmitting user the problem to select the best relay among the neighboring nodes arises. The choice of the best relay node should take into account both the required transmission poer and the delay introduced hile aiting for the relay node getting active. REFERENCE [] Y.-B. Lin, Paging systems: Netork architecture and interfaces, IEEE Netork, vol., no. 4, July-Aug. 997, pp [2] A.K. alkintzis, C. Chamzas, Mobile packet data technology: An insight into MBITEX architecture, IEEE Personal Communications, vol. 4, no., Feb. 997, pp [3] A.K. alkintzis, C. Chamzas, An in-band poer-saving protocol for mobile data netorks, IEEE Trans. on Communications, vol. 46, no. 9, ept. 998, pp [4] H. oesner, J.-P. Ebert, M. chlager, A. olisz, Poer-aving Mechanisms in Emerging tandards for ireless LANs: The MAC Level Perspective, IEEE Personal Communications, pp , June 998. [5] J.-C. Chen, K.M. ivalingam, P. Agraal, R. Acharya, cheduling multimedia services in a lo-poer MAC for ireless and mobile ATM netorks, IEEE Trans. on Multimedia, vol., no. 2, June 999, pp [6] HIgh PErformance Radio Local Area Netork (HIPERLAN) Type, Functional pecification, ET , ctober 996. [7] C.-H. Hang, A.C.-H. u, A predictive system shutdon method for energy saving of event-driven computation, IEEE/ACM International Conference on Computer-Aided Design, an Jose, CA, pp , Nov [8] L. Benini, A. Bogliolo, G.A. Paleologo, G. De Micheli, Policy optimization for dynamic poer management, IEEE Trans. on Computer- Aided Design of Integrated Cir. and ys., vol. 8, no. 6, pp , June 999. [9] E.Y. Chung, L. Benini, A. Bogliolo, G. De Micheli, Dynamic poer management for non-stationary service requests, Proc. of Design, Automation and Test in Europe, Munich, Germany, pp. 77-8, March 999. [0] U. Kaiser,. teinhagen, A lo-poer transponder IC for highperformance identification systems, IEEE Journal of olid-tate Circuits, vol. 30, no. 3, March 995, pp [] D. Friedman, H. Heinrich, D.-. Duan, A lo-poer CM integrated circuit for filed-poered radio frequency identification tags, IEEE International olid-tate Circuits Conference (ICC), an Francisco, CA, UA, Feb. 997, pp [2] H.K. Heinrich, et al., Method of transporting radio frequency poer to energize radio frequency identification transponders, U.. Pat. No. 5,850,8, Dec. 5, 998. [3] H.K. Heinrich, et al., Radio frequency identification transponder ith electronic circuit enabling/disabling capability, U.. Pat. No. 5,874,902, Feb. 23, 999. [4] M. Crovella, A. Bestavros, elf imilarity in orld ide eb Traffic Evidence and Possible Causes, Proc. of the 996 ACM IGMETRIC Conference, Philadelphia, PA, May 996.