An Efficient Power Saving MAC Protocol for IEEE Ad Hoc Wireless Networks *

Transcription

1 JOURNAL OF INFORMATION SCIENCE AND ENGINEERING 23, (2007) An Efficient Power Saving MAC Protocol for IEEE Ad Hoc Wireless Networks * Department of Computer Science and Information Engineering Chang Gung University Taoyuan, 300 Taiwan To support mobility, mobile devices must be powered by batteries with limited energy. Thus, energy-efficient management becomes one of the most important design issues in mobile wireless networks. A well-designed energy-efficient MAC protocol for mobile wireless networks should be developed for both minimum energy consumption as well as maximum data throughput. IEEE MAC protocol is based on the carrier sense multiple access with collision avoidance (CSMA/CA) medium access procedure to transmit/receive both control and data frames. We know that CSMA/CA wastes the scarce energy and bandwidth due to frame collisions and lengthens the transmission delay due to waiting backoff time, especially in heavy traffic load. In addition, the IEEE power management scheme does not specify how to determine the window size in a beacon interval. The fixed window size can not always accommodate the dynamically changing traffic conditions. To conquer these problems, we propose a novel power saving MAC protocol for IEEE ad hoc networks. The proposed protocol possesses two properties: (1) based on overhearing frames in an window, power saving (PS) stations can be scheduled to transmit data frames in order after the window, and (2) depending on the fluctuant traffic conditions, PS stations adjust the window size dynamically to save energy and achieve the superior network throughput. Through extensive simulations, we demonstrate the advantages of our new MAC protocol. Keywords: power management, medium access control (MAC), IEEE , wireless LAN (WLAN), ad hoc wireless networks, wireless networks 1. INTRODUCTION To support mobility, most portable and wearable devices are powered by batteries that have only limited and non-renewable energy. Due to technology limitations, the battery capacity will not be dramatically increased in the near future. Hence, one critical design issue is to investigate energy efficient techniques to prolong battery life for mobile wireless networks. In recent years, numerous papers have discussed various methods to reduce energy consumption. These methods can be classified into medium access control (MAC) [1, 4, 5, 8, 9, 14, 17, 19], routing [13, 15], and transport layer protocols [2, 10, 11, 21]. Woesner et al. [18] pointed out that MAC protocols could significantly reduce power consumption in mobile devices. Therefore, this paper focuses on MAC protocol energy conservation for wireless LANs (WLANs). Several power management protocols have been proposed for wireless networks. Received September 15, 2006; accepted February 6, Communicated by Ten H. Lai, Chung-Ta King and Jehn-Ruey Jiang. * This paper was supported in part by the Ministry of Economic Affairs, R.O.C., under grand No. 95-EC-17- A-19-S1-055 and National Science Council, R.O.C., under grand No. NSC E

2 1172 Chiasserini et al. [5] proposed several sleep patterns and allows mobile nodes to pick their own patterns depending on their battery condition and required quality of service. However, this protocol needs a special hardware device, called remote activated switch (RAS), which can receive wakeup signals when the mobile node has fallen to sleep. Another work proposed by Chen et al. [4], called SPAN, employs the notion of connected dominated set to save power consumption. In this protocol, some nodes must act as coordinators which are chosen according to their remaining battery power and number of neighbors. Only coordinators are needed to keep awake and other nodes can enter the sleep mode. These coordinators undertake the jobs of relaying frames for their neighboring nodes. Using an idea similar to SPAN, reference [20] proposed by Xu et al. provides a grid-based energy-saving routing protocol. With the help of the localization capability using GPS, a large area is partitioned into small areas called grids, in each of which only one node remains in the active state to relay frames for other nodes located in the same grid. Recently, Wu et al. [19] proposed three power management protocols to overcome the asynchronous problem for multi-hop mobile ad hoc networks. The proposed protocol consists of three key components: a scalable beacon transfer procedure for the beacon contention and time synchronization challenge, three randomized asynchronous power management schemes for neighbor maintenance challenges, and a data frame transfer procedure for the high energy-efficient transmissions. IEEE standard [1] specifies the MAC protocol including distributed coordination function (DCF) and point coordination function (PCF). It also provides two different power management schemes: one for infrastructure networks and the other for ad hoc networks. Our energy efficient mechanism is proposed for IEEE ad hoc WLANs. In IEEE protocol, wireless nodes have two power modes: active and power saving (PS). The power management scheme divides time into beacon intervals. At the beginning of each beacon interval, power saving nodes wake up for a short time period, called announcement traffic indication message () window. In the window, nodes exchange control frames, called frames, to inform their power saving counterparts to remain awake until the end of the beacon interval to receive data frames. After window, all nodes follow the DCF protocol to transmit their data frames. We know that DCF is based on the carrier sense multiple access with collision avoidance (CSMA/CA) which may waste scarce energy and bandwidth due to frame collisions and lengthen the transmission delay due to waiting backoff time. A lot of researches take advantage of scheduling transmissions to avoid frame collisions and to save energy consumption, which are more related to our work. However, most scheduling algorithms [12, 13, 16] are designed for centralized or master-driven systems, not for distributed systems. Reference [8] uses a distributed way to design a scheduling scheme. However, PS nodes can not completely avoid frame collisions and remove backoff time delay. We observe that PS nodes operating in IEEE power management protocol will wake up in an window so that each node can overhear all its received frames and sort the transmission order of those to-be-transmitted nodes according to their transmission durations. At the end of window, those nodes can transmit their data frames in order. Based on this idea, we propose an efficient scheduling protocol to prevent PS nodes from contending medium again after window. IEEE power management scheme does not regulate how to determine the

3 AN EFFICIENT POWER SAVING MAC PROTOCOL 1173 window size in a beacon interval. However, Woesner et al. [18] has pointed out that the fixed window size can not reach the optimal throughput. Therefore, another design issue of this paper is to adjust the window size for saving more power consumption and acquiring the better throughput. Jung et al. [9] designed a scheme to dynamically choose an window size to improve network throughput. However, the scheme allows a node to change window size only one level at each beacon interval and does not support scheduling scheme for frame transmissions. To conquer the problems, we propose a novel strategy that can adjust the window size dynamically according to the measured traffic load. In a nutshell, this paper makes two contributions to the energy efficiency in WLANs. First, an efficient scheduling transmission protocol is proposed to avoid PS nodes contending for the medium again after the window without any extra overhead. Second, we propose a novel strategy to dynamically adjust the window size to accommodate to variable traffic conditions. Thus our protocol can reduce the power consumption of PS nodes and improve the network throughput. The rest of this paper is organized as follows. Section 2 gives an overview of the power management scheme of IEEE ad hoc networks. Our proposed protocol, including a scheduling transmission mechanism and an adjusting window size scheme, is presented in section 3. Extensive simulation results are given in section 4. Finally, we conclude this paper and discuss future works in section IEEE POWER MANAGEMENT SCHEME FOR AD HOC NETWORKS In this section, we briefly review the main power management scheme in an IEEE ad hoc WLAN. The standard regulates that all nodes are fully connected and, at each beacon interval, can start at about the same time, called target beacon transmission time (TBTT). At each TBTT, each node contends to send a beacon for time synchronization with all other nodes. To avoid collisions of transmitted beacon frames, a node should wait a random number of slots between 0 and 2 CWmin 1 before transmitting its beacon frame. If a node with buffering unicast frames to a PS node, it will send an frame to the PS node within the window period. On receiving the frame, the PS node responses an ACK and both the sender and receiver will remain awake for the remaining beacon interval. After the end of window, all nodes follow the normal DCF access procedure to transmit/receive their data frames. TBTT AW data frame TBTT next AW A Beacon B ACK ACK Power Saving State C Beacon Interval (i.e. BI) Fig. 1. An Example of IEEE power management scheme in ad hoc network.

4 1174 Fig. 1 illustrates an example of the IEEE power management scheme. At the beginning of a beacon interval, nodes A, B, and C wake up. If node C receives no frame for it during the window period, it will go back into the doze mode. On node A having data frames for PS node B, it first transmits an frame to B during the window. Node B replies an ACK to A. After the window, node A and B exchange their data frames and ACKs using the DCF access procedure. All dozing nodes wake up again at the beginning of the next window. Since DCF is based on contention-based CSMA/CA mechanism, it may waste scarce battery resources and bandwidth due to frame collisions and lengthen the frame delay due to waiting for a period of backoff time. 3. ENERGY-EFFICIENT MAC PROTOCOL A well-designed energy-efficient MAC protocols should be concern with higher data throughput and better energy savings. To realize the considerations as well as to have the compatibility of IEEE power management protocol, we propose a novel scheme to schedule those to-be-transmitted data frames for PS nodes and an intelligent scheme to adjust the size of the window dynamically. Our scheme follows most of IEEE regulations such as the structure of a beacon interval and the time synchronization mechanism among all nodes. We describe the detail operations of the two mechanisms in the following two subsections separately. 3.1 The Scheduling Transmission Mechanism It is clear that reducing nodes backoff idle time and avoiding frame collisions can improve energy efficiency and network throughput. We ve found that it can be achieved easily by the overhearing technique using the broadcast nature of wireless medium. Thus just little modification of IEEE power management protocol, we have an efficient scheduling mechanism for PS node transmissions. In our protocol, the duration of a to-be-transmitted data frame, named working duration, is piggybacked in an frame. The total working duration of a node is the sum of the working durations for all frames related to the node. To minimize the total power consumption, we have to minimize the total waiting time because the more wait time a node has, the more battery energy it wastes. To minimize total waiting time, we follow shortest job first policy basically, so the node with the shortest total working duration has the highest priority to transmit its buffering frames after the window. Therefore, each node can easily determine the first PS node transmission scheduled locally by sorting through the total number of working durations for all PS nodes. The scheduling transmission mechanism is designed for PS nodes to extend their battery life. After completing all scheduled transmissions, other nodes obey the normal DCF procedure to contend in the medium to send data frames. Moreover, we employ a mechanism similar to that in Ref. [9] which allows a PS node to go back to sleep when it completes all of its data transmissions. This is different from the IEEE standard in which a PS node remains awake during the entire beacon interval even if it has received its data frames before the end of this beacon interval. In the following presentation, we first introduce the basic scheduling scheme and then give our final scheme.

5 AN EFFICIENT POWER SAVING MAC PROTOCOL The basic scheduling transmission scheme The basic scheme has two versions: the first is called the simple version which uses the idea of shortest transmission duration and the second is the improved version that will approach the total minimum waiting time for PS nodes within a beacon interval. We present the detailed steps for the simple version first and then the improved version. The Simple Version Our simple version is divided into three steps. To ease our explanation, we use the example as shown in Fig. 2 to illustrate the three steps for the simple version. Step 1: Construct the transmission table Following IEEE regulations, all nodes are fully connected and time synchronization so that all PS nodes can wake up at nearly the same time TBTT. At the TBTT, each node wakes up for an window interval. If a node with buffering a unicast frame to a PS node, it will send an frame to the PS node within the window period. On receiving the frame, the PS node responds with an ACK to the sender of the frame and completes the reservation for the data frame transmissions. The information about the sender, receiver, and working duration of the to-be-transmitted data frame are contained in frames denoted as (Sender_ID, Receiver_ID, Working_Duration). Because of the broadcast nature of wireless medium, any other node can overhear the frame and append to its transmission table as shown in Table 1. Table 1. Transmission table. Sender_ID Receiver_ID Working_Duration A B 10 A C 20 B C 30 D E 40 A (A, B, 10) (A, C, 20) SIFS Remaining BI For active nodes to contend for data transmision. B ACK (B, C, 30) C D (D, E, 40) E Window Beacon Interval Window Fig. 2. An example of scheduling protocol. Fig. 2 is an example to illustrate our main idea. To simplify our presentation, we omit the beacon transfer procedure which is same as the IEEE protocol. There are

6 1176 five PS nodes involved in the frames transmission: A, B, C, D, and E. During the window, only four frames are announced successfully, i.e., (A, B, 10), (A, C, 20), (B, C, 30), and (D, E, 40). Therefore, at the end of the window all nodes in the network should maintain the same transmission table as shown in Table 1 if no transmission error has occurred. Step 2: Determine the transmission order (a) To determine the first transmission node, each node sums up the working durations for individual node in its transmission table. The node with minimum total working duration is the first transmission node in the current transmission table. (b) To determine the next transmission node, the related entries for first transmission node will be deleted from the transmission table. The system then goes back to step 2(a) to determine the next transmission node. The loop of the two steps 2(a) and 2(b) is performed until the transmission order of all PS nodes is determined. To illustrate this step, we also use Fig. 2 as our example. From Table 1, A has two related frames, i.e., (A, B, 10) and (A, C, 20), so its total working duration is = 30. B also has two related frames, i.e., (A, B, 10) and (B, C, 30), so its total working duration is = 40. The calculations of other nodes are same as A and B. After the calculations are complete, nodes A, B, C, D, and E have total working durations of 30, 40, 50, 40, and 40, respectively. Therefore, node A is the first transmission node because it has the minimum total working duration. To determine the next transmission node, all entries related to A in the transmission table are deleted. Table 2 is the new transmission table. From Table 2, we can easily determine the next transmission nodes are B and C. The loop of the two steps is performed until the transmission order of all PS nodes is determined. The final transmission schedule is that after A finishes its transmissions, B transmits the data frame to C and then D transmits the data frame to E. Table 2. The new transmission table after the related entries of node A has been deleted. Sender_ID Receiver_ID Working_Duration B C 30 D E 40 Step 3: Transmission data frames for PS nodes At the end of the window, each PS node exchanges its data frames according to its individual order and the available time specified in the working duration. We continue the above example to explain this operation. After window and a SIFS time, A and B first exchange frames and A waits for a SIFS time to send the buffering frames to C. While completing all of its traffic, A goes to sleep. In the next transmission node B waits for a SIFS time and then transmits its data frames to C immediately. The next transmission, from node D to E, follows the same rule as B and C to exchange frames. At the end of the scheduling transmission time, the nodes not in the transmission table begin to transmit their frames following DCF regulations.

7 AN EFFICIENT POWER SAVING MAC PROTOCOL 1177 The Improvement Version The way we describe above to estimate the energy efficiency follows the principle that minimize the total waiting time for all PS nodes. The idea is based on that the less the waiting time is the less the energy consumption. This is why we use the algorithm, shortest job first, to calculate the scheduling transmission order. However, there is an exception that has less total nodes waiting time than shortest job first. We also use Fig. 2 to illustrate the exception. When the entry (D, E, 40) is changed to (D, E, 35), the total working duration for nodes A, B, C, D, and E are 30, 40, 50, 35, and 35, respectively. From Table 1, we know D and E only communicate with each other so that they can go back to doze mode immediately after completing their transmission. Thus, we can save both of D and E s waiting time. The special type of communication is called pair communication. In original principle, node A will be the first transmission node because it has the minimum total working duration. However, if there is a pair communication in the transmission table, the scheduling scheme first calculate and compare the two cases: (i) exchange the transmission pair order with the first priority order, and (ii) do not exchange orders, i.e., using the short job first order. If the total waiting time of case (i) is less than that for case (ii), this pair communication has first transmission priority. In the above exception we let the pair communication, D and E, be the first transmission. After completing the pair communication, step 2(a) of the simple version deletes the entry for D and E. At this time, node A has the shortest total working duration and it will be the second one to send its data frame. Now the transmission order is D to E, A to B and C, and the last is B to C. We find that in following this order the total energy consumption is ( ) = 325. It is less than the original order, A to B and C, B to C, and then D to E, with the total energy consumption ( ) = Adjust Window Size It has been mentioned that the window size affects network performance. To conserve more PS node power and improve the network throughput, we propose an adaptive scheme to dynamically adjust the duration of the window to a suitable size. An window starts at the beginning of a beacon interval which is same as IEEE power management protocol, but the end of the window will depend on the traffic load of the current beacon interval. We have two rules to dynamically end the window size. The detail operation is shown below. The variables/constants used in our presentation are listed in Table During an window period, all nodes are awake and via overhearing frames they have the same channel information regardless if the channel is busy or idle. If the channel is idle more than T DIFS + T CWmin, we deem that there will be no other node wanting to send frames. Accordingly, all nodes end the window and enter the scheduled transmission phrase. That is IF ChannelIdleTime T DIFS + T CWmin THEN AW = (T curr TBTT) + T DIFS + T CWmin.

8 1178 Table 3. The new transmission table after the related entries of node A has been deleted. T SIFS T DIFS T ACK T CWmin T Framei T Framemin T T curr BI TBTT TBTT next time of short inter-frame spacing time of distributed inter-frame spacing time to transmit an ACK time of min. contention window time to transmit ith data frame in scheduling table time to transmit a min. data frame time to transmit an frame last ACK transmission ending time Beacon Interval the beginning time of current BI the beginning time of next BI T curr SIFS TBTT DIFS+CWmin data frame For active nodes to contend for data transmision. TBTT next RemainingBI A B ACK ACK C AW Power Saving State Beacon Interval (i.e. BI) Next BI Fig. 3. This figure shows the first period ending rule. If nodes sense the channel is idle for T DIFS + T CWmin, nodes end window immediately. Fig. 3 is an example to illustrate this rule. During window, there is only one frame sent from A to B. After they exchange their control frames, the channel becomes idle. All nodes will not end the window to enter the scheduled transmission until the idle time lasts for T DIFS + T CWmin. According to our scheduling transmission mechanism, A starts to send the data frame to node B, and after SIFS idle time, B replies an ACK to node A. After completing their transmissions, A and B can turn back to sleep mode. Because node C has nothing to send or receive, it goes to sleep mode at the end of window. The rest beacon interval is for active nodes to use. 2. As mentioned in section 3.1, each node can obtain the duration of frame transmissions by overhearing frames in a fully connected topology and calculate the total duration of all its currently receiving frames. If the total duration of the scheduled transmissions reaches BI limitation, i.e. there is no any data frame can be transmitted in the rest BI even if the shortest frame, all nodes end the current window immediately and enter the scheduled transmission. That is RemainingBI = TBTT next T curr n i= 1 ( T + 2 T + T ), Framei SIFS ACK

9 AN EFFICIENT POWER SAVING MAC PROTOCOL 1179 IF RemainingBI 3T SIFS + T Framemin + T + 2T ACK + T DIFS THEN AW = T curr TBTT. We use Fig. 4 to explain the rule. After the announcements in period, A, B, and C, have the transmission durations of Frame A B, Frame B C, and Frame C B. When they find the remaining beacon interval (i.e. RemainingBI) is less than 3T SIFS + T Framemin + T + 2T ACK + T DIFS, it means there is no more time for exchanging an and an ACK frame and transmitting the shortest data frame. They then end window immediately and start to transmit their scheduled data frames after SIFS. TBTT T curr SIFS TBTT next RemainingBI A A->B ACK B->C B ACK ACK B->A B->C C AW ACK C->B Beacon Interval (i.e. BI) C->B Next BI Fig. 4. This figure shows the second period ending rule. After calculating, if the remaining beacon interval is less than 3T SIFS + T Framemin + T + 2T ACK + T DIFS, nodes end window immediately. 4. PERFORMANCE EVALUATION We have developed an event-driven simulator to evaluate the performance of the proposed protocol and compare it with IEEE protocol. Our proposed protocol contains two distinct mechanisms, i.e. scheduling transmission and adjusting window size. To further investigate the individual effect of the two mechanisms, the proposed protocol are evaluated with three different protocols in the simulations: the protocol with the window size adjustment mechanism without scheduling transmission, a protocol with the mechanism scheduling without adjusting the window size, and a protocol with a combination of the two individual mechanisms denoted as, Shortest, and + Shortest, respectively, for the following text and figures. For the power consumption parameters used in the simulations, we refer from [6, 7] based on real experiments on Lucent WaveLAN cards. A cost to transmit/receive a data frame which has the form P const + P byte Frame length (μjoule), where P const is the fixed energy consumption independent of packet length, P byte is the power consumption per byte, and Frame length is the frame length in bytes. The value P const and P byte to transmit a data frame are set to 420 and 1.9, and to receive a data frame are set to 330 and A node has no frame to transmit/receive is called Idle in the active mode and Doze in the PS mode. The power consumption of Idle/Doze is set to 808 mw and 27 mw, respectively. For the network parameters, the channel bit rate is 2 Mbps, beacon interval is 0.1

10 1180 sec, beacon window is 0.01 sec, window is 0.02 sec, data frame is 2048 bytes, beacon frame is 61 bytes, frame is 28 bytes, and data/ ACK frame is 14 bytes. The window size in and the Shortest protocol is fixed at 0.02 sec while in the + Shortest and protocol is variable depended on the network traffic load. When a data frame arrival from higher-layer to MAC sublayer at a node, the destination of each frame is selected randomly from all of the nodes. Each result is the average of 50 times and each time runs 10 seconds. The performance is evaluated using 3 metrics: energy efficiency (bytes/joule), throughput (bytes/sec), and average delay of transmitted data frames (sec). Since our MAC protocol is primarily designed for PS nodes, we first give a simulation with the environment of pure PS nodes and compare ours with protocols on various metrics. Note that the arrivals of frames at a PS node follow the Poisson distribution with mean rate 1 frames/sec. The simulation will increase PS nodes with the number of 20 per step. Fig. 5, Fig. 6, and Fig. 7 show the energy efficiency, throughput, and average delay for the transmitted data frames, respectively. Energy efficiency: Fig. 5 demonstrates the advantageous position of the proposed protocols in the metric of energy efficiency. By means of dynamically adjusting window size, the proposed protocols permit suitable number of data frames to be transmitted by PS nodes after the window. After that, other PS nodes can enter the sleep mode for saving their variably battery. We know using DCF, data frame collisions and backoff waiting time will be more and longer than using the scheduling transmission mechanism when the total PS nodes number arises. This is the reason that the energy efficient of the protocol is little lower than of the Shortest one as the number of PS nodes is high. The Shortest protocol gives PS nodes the first order to transmit data frames so saves their waiting time. However, under the heavy traffic load, the Shortest protocol will waste a little energy because it uses the fixed window that cannot be ended early even if everybody ensures no more frame to be transmitted. By possessing all the merits in both and Shortest ones, the protocol + Shortest performs the best energy efficiency among all protocols. Throughput: The throughput results are drawn in Fig. 6. Dissecting the results, all of proposed protocols are more outstanding throughput than The reason is the same as the previous analysis. From the figure we also find a phenomenon that the throughput of + Shortest protocol is the best one and the protocol is the worst one. And the gap between the Shortest and protocol is more visible as the number of PS nodes becomes larger. The reason about the phenomenon is that the Shortest protocol offers the tremendous throughput for PS nodes by transmitting the scheduled data frames without any collisions and backoff delay time. However, the PS nodes in the protocol still need to wait for at least the DIFS and backoff time, and may incur collisions with other nodes during data transmission time. This makes the protocol limited in throughput performance. This means that the benefit of using the Shortest protocol will be better than of using the one in the environment with the large number of pure PS nodes. The + Shortest protocol has the highest throughput because it has the advantages of both adjusting window size and scheduling transmissions.

11 AN EFFICIENT POWER SAVING MAC PROTOCOL 1181 Fig. 5. Energy efficiency under pure PS nodes. Fig. 6. Throughput under pure PS nodes. Fig. 7. The average delay of a transmitted frame under pure PS nodes. (a) (b) Fig. 8. Energy efficiency with various active nodes under the fixed number of PS nodes at 10 in (a) and at 40 in (b). Delay: Fig. 7 illustrates the average delay for the transmitted data frames. The figure shows that our protocols present better outcome than In addition, the gap of average delays between ours and is larger when we increase the number of PS nodes into the network. This is because that, as PS nodes increase, two characters, the fixed window and the DCF rule used in data frame transmission make the average delay longer. Because the flexible protocol avoid too many nodes competing the wireless medium in order to transmit data frames and the ingenious Shortest protocol to reduce the total waiting time of PS nodes, our proposed protocol deservedly minimize the average delay of all nodes. Lower the transmitted frame delay represents the lower energy consumption of PS nodes in the environment with pure PS nodes. Combined the above analysis for energy efficiency, throughput, and the average frame transmission delay, we have shown the superiority of our proposed protocols in each aspect of interest to users. To make the simulation environment correspond to a real network, we designed four simulations with both active nodes and PS nodes by setting the mean arrival rate of frames 4 frames/sec and 1 frames/sec respectively. In the first two simulations, we fixed the number of PS nodes at 10 in simulation 1 and 40 in simulation 2, and increase the active nodes with the degree of 10. Simulation 1 produces Fig. 8 (a), Fig. 10 (a), and Fig. 2 (a) while simulation 2 generates Fig. 8 (b), Fig. 10 (b), and Fig. 12 (b). In the last two simulations, we fixed the number of active nodes at 10 in simulation 3 and 40 in simulation 4,

12 1182 (a) (b) Fig. 9. Energy efficiency with various PS nodes under the fixed number of active nodes at 10 in (a) and at 40 in (b). (a) (b) Fig. 10. Throughput with various active nodes under a fixed number of PS nodes at 10 in (a) and at 40 in (b). (a) (b) Fig. 11. Throughput with various PS nodes under a fixed number of active nodes at 10 in (a) and at 40 in (b). and increased the number of PS nodes with the degree of 10. Simulation 3 produces Fig. 9 (a), Fig. 11 (a), and Fig. 13 (a) while simulation 4 generates Fig. 9 (b), Fig. 11 (b), and Fig. 13 (b).

13 AN EFFICIENT POWER SAVING MAC PROTOCOL 1183 Energy efficiency: The energy efficiency is shown in Figs. 8 and 9. The + Shortest protocol, which combines the scheduling with adjusting the window size dynamically, has the highest performance. That means, under consuming the same amount of energy, our method transmits more data frames. The reasons for this improvement are that we employ the scheduling transmission mechanism to decrease the possibility of frame collision and the energy consumption in waiting for DIFS and backoff time. The dynamically adjusted window avoids PS nodes remaining unnecessarily awake. Fig. 8 (b) shows when the power saving node number is high (e.g. 40) and the number of active nodes is low (e.g. 10), our protocol outperforms the by nearly two times. This is because that the serious energy waste of the PS nodes effects the waiting backoff time and keeps the node unnecessarily awake. Similarly, our protocol gains higher energy efficiency than in Fig. 9. In Fig. 9 (a), by increasing the number of PS nodes, the energy efficiency lines of our protocols increase while decreases. Throughput: To prove that our energy efficient MAC protocols don t sacrifice the throughput, it deserves to probe into the network throughput of these protocols. The simulation results in Figs. 10 and 11 point out that the + Shortest protocol still has the highest throughput. This proves that the + Shortest protocol has an excellent performance in both energy efficiency and throughput. Lengthening window size on the heavy load allows suitable number of nodes to exchange the frame and the ACK during the period. After that, they don t need to contend channel and they can surely send their buffering data frames. By shortening the window on the light load, we decreases the period because that no any other node tends to transmit data to the PS node. This increases the remaining time for active nodes to transmit more data frames in each beacon interval. Thus, the + Shortest protocol has lower collisions, less idle time and more the throughput. From the simulation results we could infer that when the number of PS nodes is greater than the active nodes and the difference between them becomes larger, our protocol attains good performance than the protocol in both energy efficiency and throughput. Typically, increasing the total nodes will make the throughput arise unless the total offer load of nodes is greater than the network saturation point. In Fig. 11 (b), the trend for all lines descends as the number of nodes increases. To answer the phenomenon we must find out the cue for the network overload. We observe that the network is saturated at about 40 active nodes and 10 PS nodes. Thus, the more nodes join the network, the more collisions of the transmitted frames will be occurred. This is why the output lines descend. Considering the throughput of the Shortest protocol in Figs. 10 and 11, we can know the benefit of scheduling frame transmissions. Because of no contention, PS nodes transmit their frames in order. Thus the more PS nodes the more throughput. The protocol, dynamically adjusts the node window size, directly extends the remaining time for the data transmissions after ends. When the demand from PS nodes is low, window will be terminated beforehand because that ChannelIdleTime is greater than or equal to T DIFS + T CWmin. In the other situation, when the PS requests are heavier, window will still end at the earlier time point because the total PS node

14 1184 (a) (b) Fig. 12. The average ZA delay of a transmitted frame with various active nodes under a fixed number of PS nodes at 10 in (a) and at 40 in (b). (a) (b) Fig. 13. The average delay of a transmitted frame with various PS nodes under a fixed number of active nodes at 10 in (a) and at 40 in (b). transmission duration has exceeded the RemainingBI. Allowing more PS nodes transmitting frames will cause redundant competition for nodes in wireless medium during the RemainingBI. In this situation, the protocol could prevent nodes from redundantly competing. Our proposed protocol integrates these two schemes into the exceptional one brings the highest throughput in the simulation results. Delay: Since the average delay for transmitted data frames is the most important factor in power consumption as well as QoS degree, we have a series simulation results from Figs. 12 to 15 to evaluate. Figs. 12 and 13 show that the protocol + Shortest is stably superior to the protocol. The average delay of our protocols in Fig. 12 (a) forms a concave shape. To interpret the special situation, we need to divide the average delay of all nodes into two individual delays, i.e. PS nodes and active nodes, for further narrow down the reflecting causes. The average delay of the PS and active nodes are shown in Figs. 14 and 15 respectively. From the two figures, we can make some comments as follows. First, the Shortest protocol makes the PS nodes have the lowest average delay because it schedules the transmissions to PS nodes prior to active nodes after the window.

15 AN EFFICIENT POWER SAVING MAC PROTOCOL 1185 delay time (sec) Shortest 0.16 Shortest _10 20_10 30_10 40_10 active node number_ps node number delay time (sec) Shortest 0.07 Shortest _10 20_10 30_10 40_10 active node number_ps node number (a) (b) Shortest Shortest Shortest Shortest 0.35 delay time (sec) delay time (sec) _40 20_40 30_40 40_40 active node number_ps node number 0 10_40 20_40 30_40 40_40 active node number_ps node number (c) Fig. 14. The average delay for a transmitted frame from a PS node with various active nodes under the fixed number of PS nodes at 10 in (a) and at 40 in (b) as well as with various PS nodes under the fixed number of active nodes at 10 in (c) and at 40 in (d). (d) However, the protocol incurs increasing active node delay and loses some chance to contend for the medium. Because the Shortest protocol schedules the transmission time of data frames disregarding active nodes, it makes that the longer delay of active nodes as shown in Figs. 15 (a)-(d). The effect on the average delay of all nodes is also worse, especially when the number of active nodes reaches 40, as shown in Fig. 12 (a). In other words, when we focus on the topic of transmission delay, Shortest is beneficial to PS nodes but sacrifices the profit of active nodes. Second, the protocol, which allows proper nodes to send frames and ACKs during the interval, reduces the number of nodes contending for the wireless medium during the remaining time. The benefit of the protocol is that it can indeed descend the waiting time for active nodes because less nodes want to contend for the wireless medium to transmit the data frame to PS nodes. Without scheduling the data frame transmissions, PS nodes still have to contend with others for the chance to transmit data frames. Hence, we would not surprise the higher delay of PS nodes. Finally, the + Shortest protocol, which combines the and the Shortest protocols, gets the ideal delay threshold in between and the Shortest schemes. The protocol makes a suitable number of nodes which want to send data frames to PS nodes to transmit them after the window, and the Shortest protocol ensures their data transmission having the shortest waiting time. These properties give the reason for why the + Shortest has the excellent performance in respect of the average node delay.

16 Shortest Shortest Shortest Shortest delay time (sec) delay time (sec) _10 10_20 10_30 10_40 active node number_ps node number 0 10_10 10_20 10_30 10_40 active node number_ps node number (a) (b) Shortest Shortest Shortest Shortest 0.35 delay time (sec) delay time (sec) _10 40_20 40_30 40_40 active node number_ps node number 0 40_10 40_20 40_30 40_40 active node number_ps node number (c) Fig. 15. The average delay for a transmitted frame from an active node with various active nodes under the fixed number of PS nodes at 10 in (a) and at 40 in (b) as well as with various PS nodes under the fixed number of active nodes at 10 in (c) and at 40 in (d). (d) 5. CONCLUSION We developed a new energy-efficient MAC protocol for WLANs. The proposed protocol has two main contributions. First, to avoid unnecessary frame collisions and backoff waiting time during data frame transmissions, we designed a novel idea but simple scheduling mechanism for nodes to transmit their frames to power saving nodes in order. Second, to better conserve the power of PS nodes and improve to the channel utilization, we developed an intelligent strategy to dynamically change the window size. The simulation results prove that our energy-efficient MAC protocols have a better performance than IEEE power management protocol not only in energy efficiency but also in aggregate throughput. REFERENCES 1. IEEE Std IEEE Computer Society LAN MAN Standards Committee, Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, S. Agrawal and S. Singh, An experimental study of TCP s energy consumption over wireless link, in Proceedings of the 4th European Personal Mobile Communi-

17 AN EFFICIENT POWER SAVING MAC PROTOCOL 1187 cations Conference, 2001, pp I. Chakraborty, A. Kashyap, A. Kumar, A. Rastogi, H. Saran, and R. Shorey, MAC scheduling policies with reduced power consumption and bounded packet delay for centrally controlled TDD wireless networks, in Proceedings of IEEE International Conference on Communications, Vol. 7, 2001, pp B. Chen, K. Jamieson, H. Balakrishnan, and R. Morris, Span: an energy-efficient coordination algorithm for topology maintenance in ad hoc wireless networks, in Proceedings of the International Conference on Mobile Computing and Networking, 2001, pp C. F. Chiasserini and R. R. Rao, A distributed power management policy for wireless ad hoc networks, in Proceedings of IEEE Wireless Communication and Networking Conference, 2000, pp L. M. Feeney and M. Nilsson, Investigating the energy consumption of a wireless network interface in an ad hoc networking environment, in Proceedings of IEEE INFOCOM, Vol. 3, 2001, pp L. M. Feeney, An energy consumption model for performance analysis of routing protocols for mobile ad hoc networks, ACM/Kluwer Mobile Networks and Applications, Vol. 6, 2001, pp C. S. Hsu, J. P. Sheu, and Y. C. Tseng, Minimize waiting time and conserve energy by scheduling transmissions in IEEE based ad hoc networks, in Proceedings of International Conference on Telecommunications, 2003, pp C. S. Hsu and Y. C. Tseng, Cluster-based semi-asynchronous power-saving protocols for multi-hop ad hoc networks, in Proceedings of International Chamber of Commerce, 2005, pp E. S. Jung and N. H. Vaidya, An efficient MAC protocol for wireless LANs, in Proceedings of IEEE INFOCOM, Vol. 3, 2002, pp R. Krashinsky and H. Balakrishnan, Minimizing energy for wireless web access with bounded slowdown, in Proceedings of the International Conference on Mobile Computing and Networking, 2002, pp R. Kravets and P. Krishnan, Application-driven power management for mobile communication, ACM/Baltzer Wireless Networks, Vol. 6, 2000, pp S. C. Lo, G. Lee, and W. T. Chen, An efficient multipolling mechanism for IEEE wireless LANs, IEEE Transactions on Computers, Vol. 52, 2003, pp J. H. Ryu and D. H. Cho, A power-saving multicast routing scheme in 2-tir hierarchical mobile ad-hoc networks, in Proceedings of IEEE Vehicular Technology Conference, Vol. 4, 2000, pp S. Singh and C. S. Raghavendra, Power efficient MAC protocol for multihop radio networks, in Proceedings of IEEE International Personal, Indoor and Mobile Radio Communication Conference, 1998, pp S. Singh, M. Woo, and C. S. Ragavendra, Power-aware routing in mobile ad hoc networks, in Proceedings of the International Conference on Mobile Computing and Networking, 1998, pp J. A. Stine and G. de Veciana, Improving energy efficiency of centrally controlled wireless data networks, ACM/Baltzer Wireless Networks, Vol. 8, 2002, pp Y. C. Tseng, C. S Hsu, and T. Y. Hsieh, Power-saving protocol for IEEE

18 1188 based multi-hop ad hoc networks, in Proceedings of IEEE INFOCOM, Vol. 1, 2002, pp H. Woesner, J. P. Ebert, M. Schlager, and A. Wolisz, Power-saving mechanism in emerging standards for wireless LANs: the MAC level perspective, IEEE Personal Communications, Vol. 5, 1998, pp S. L. Wu, P. C. Tseng, and Z. T. Chou, Distributed power management protocols for multi-hop mobile ad hoc networks, Computer Networks, Vol. 47, 2005, pp Y. Xu, J. Heidemann, and D. Estrin, Geography-informed energy conservation for ad hoc routing, in Proceedings of the International Conference on Mobile Computing and Networking, 2001, pp M. Zorzi and R. R. Ramesh, Is TCP energy efficient? in Proceedings of IEEE MoMuC, 1999, pp Shih-Lin Wu ( ) received the B.S. degree in Computer Science from Tamkang University, Taiwan, in June 1987 and the Ph.D. degree in Computer Science and Information Engineering from National Central University, Taiwan, in May Since Aug. 2001, he has been with the department of Computer Science and Information Engineering, Chang Gung University. Dr. Wu served as a Program Chair in the Mobile Computing Workshop, 2005, as a Guest Editor for Journal of Pervasive Computing and Communication special issue on Key Technologies and Applications of Wireless Sensor and Ad Hoc Networks. His current research interests include mobile computing, wireless networks, distributed robotics, and network security. Pao-Chu Tseng ( ) received the B.S. degree in Electronic Engineering, from Chung Yuan Christian University, Taiwan, R.O.C., in 1998, and the M.S. degree in Chang Gung University, Taiwan, R.O.C., in When she was in graduate school, her research focuses on power saving and reliable broadcast in wireless environment. Jhen-Yu Yang ( ) received the B.S. degree in Computer Science from the Tamkang University, Taiwan, R.O.C., in 2004, and the M.S. degree in Computer Science from the Chang Gung University (CGU), Taiwan, R.O.C. in From 2004 to 2006, he joined the High Speed Network Lab of CGU. His research interests include mobile computing and performance analysis.