A Distributed MAC Scheme Supporting Voice Services in Mobile Ad Hoc Networks 1

Transcription

1 A Distributed MAC Scheme Supporting Voice Services in Mobie Ad Hoc Networks 1 Hai Jiang, Ping Wang, H. Vincent Poor, and Weihua Zhuang Department of Eectrica & Computer Engineering, University of Aberta, Canada Schoo of Computer Engineering, Nanyang Technoogica University, Singapore Department of Eectrica Engineering, Princeton University, U.S.A. Department of Eectrica & Computer Engineering, University of Wateroo, Canada Emai: hai.jiang@ece.uaberta.ca, wangping@ntu.edu.sg, poor@princeton.edu, wzhuang@bbcr.uwateroo.ca Abstract Future mobie ad hoc networks are expected to support voice traffic. The requirement for sma deay and jitter of voice traffic poses a significant chaenge for medium access contro (MAC) in such networks. User mobiity presents particuar difficuties in this context due to the associated dynamic path attenuation. In this paper, a MAC scheme for mobie ad hoc networks supporting voice traffic is proposed. With the aid of a ow-power probe prior to DATA transmissions, resource reservation is achieved in a distributed manner, thus eading to sma deay and jitter. The proposed scheme can automaticay adapt to dynamic path attenuation in a mobie environment. Statistica mutipexing of on/off voice traffic can aso be achieved by partia resource reservation for off voice fows. Simuation resuts demonstrate the effectiveness of the proposed scheme. Index Terms medium access contro, code-division mutipe access, mobie ad hoc networks, statistica mutipexing. I. INTRODUCTION Mobie ad hoc networks are expected to support mutimedia services such as voice and video with quaity-of-service (QoS) requirements. This task is chaenging because there is no centra controer in such networks. User mobiity makes the situation even worse because of the time-varying network 1 Contact author: Dr. Hai Jiang, University of Aberta. Te: Fax: Emai: hai.jiang@ece.uaberta.ca. This paper was presented in part at IEEE GOBECOM This work was supported by the U.S. Nationa Science Foundation under Grants ANI and CNS , a research grant from the Natura Science and Engineering Research Counci (NSERC) of Canada, and a Postdoctora Feowship from NSERC of Canada. 1

2 topoogy and propagation oss between any two nodes. QoS support in mobie ad hoc networks incudes two approaches: routing at the network ayer, or medium access contro (MAC) at the ink ayer. The function of routing is to search for a network path (i.e., a route) from a traffic source node to its destination node, and to react (e.g., seect a new route) to possibe route faiures and/or network congestion. On the other hand, the function of MAC is to coordinate the channe access in an ordery manner among the mobie nodes such that efficiency can be achieved. Here, we focus primariy on the MAC in a mobie ad hoc network supporting voice traffic. In such a network, a MAC scheme shoud be: 1) distributed (because of the scaabiity requirement); 2) toerant to the hidden termina probem; 3) ensuring of sma deay/jitter (due to the rea-time nature of voice traffic); and 4) adaptive to user mobiity. In this paper, we propose an effective MAC scheme with the above desired features. The scheme is fuy distributed, thus scaing we to a arge network. The hidden termina probem does not exist in our scheme. Radio resource requirements are met by resource reservation, thus achieving sma deay and jitter. Our scheme can aso adapt to user mobiity to a certain extent. The rest of this paper is organized as foows. Reated work is discussed in Section II. The proposed MAC scheme is presented in Section III. Statistica mutipexing of voice traffic in our scheme is investigated in Section IV. Performance evauation is given in Section V, foowed by concuding remarks in Section VI. II. RELATED WORK AND DISCUSSION The most prevaent MAC schemes for ad hoc networks are carrier-sense mutipe access (CSMA) and its variants, and busy-tone based schemes. The CSMA schemes are inherenty distributed. However, it is we accepted that CSMA schemes suffer from the hidden termina probem. In addition, these schemes are based on an idea assumption, i.e., that each transmitter has an interference range so that a receiver of another ink outside the interference range does not receive any interference from that transmitter. However, in practica networks, the interference from a transmitter to any receiver depends on the transmission power, the distance between the two nodes, and the environment between 2

3 them. This means the interference eve is continuous. So the principe behind the preceding idea assumption is to quantize the continuous interference eve to two discrete eves: fu interference or no interference, thus making network anaysis much easier. However, it may not refect the reaity in an actua network. For instance, consider the situation shown in Fig. 1. We have a transmission ink from node D to node C. The receiver, i.e., node C is beyond the interference range of two interferers, nodes A and B. According to the idea assumption, node C receives no interference from either A or B. However, in reaity, athough the interference from either node A or B is not significant enough to affect node C s reception, the summation of interference received from A and B may make node C unabe to receive its desired signa successfuy. On the other hand, one motivation of busy-tone based schemes [1] is to better sove the hidden termina probem. Each target receiver sends a receive busy-tone to notify the nearby hidden terminas not to transmit. However, the hidden termina probem is not eiminated. For instance, coisions of request-to-send (RTS) frames can sti occur due to the hidden termina probem [2]. It is aso chaenging to determine the busy-tone coverage. Traditionay the busy-tone coverage is set to be the same as the interference range. However, simiar to the case in CSMA, if there are mutipe interferers beyond a target receiver s busy-tone coverage, the aggregate interference from them can be strong enough to corrupt the target receiver s desired reception, since the interferers cannot hear the busy-tone and thus start their transmissions. In a mobie environment, the mutipath fading can cause the busy-tone coverage to vary with time. This can severey degrade the system performance because a strong interferer may not be notified by the busy-tone due to instantaneous deep fading of the busy-tone channe. The contention-based nature of CSMA schemes and busy-tone based schemes makes it very difficut to bound deay and jitter to sma vaues for voice traffic. This observation impies that we shoud instead resort to reservation of channe resources. Resource reservation can avoid contentions, and voice traffic can be transmitted continuousy on the reserved channe, thus bounding the transmission deay and jitter. Reservation based MAC schemes have been extensivey studied in the iterature [3]- [8]. In these schemes, each node first contends in a contention period to send a reservation request. 3

4 Once the reservation is granted, the node can transmit in the reserved time sot without contention. However, these schemes sti suffer from the aforementioned idea assumption probem. Thus in this research we consider a code-division mutipe access (CDMA)-based mobie ad hoc network, in which the air interface supports mutipe simutaneous transmissions in a neighborhood. In our previous work [9], we have proposed a MAC scheme for a CDMA-based wireess mesh backbone, referred to as the MESH scheme here. Before DATA 2 transmissions, each potentia sender first transmits a owpower probe to the network. Upon reception of the probe, each active receiver estimates a potentia interference increase due to the potentia sender s DATA transmission. The active receiver transmits a busy-tone signa in a separate busy-tone channe to notify the potentia sender if the potentia interference increase is intoerabe. If the potentia sender does not detect a busy-tone, it successfuy reserves a code channe. However, targeted at a wireess mesh backbone with fixed topoogy and arge traffic intensity, the MESH scheme is not effective or efficient in a mobie ad hoc network with timevarying network topoogy and time-varying channe gain between any two nodes. Accordingy, in the foowing we propose a MAC scheme for a mobie ad hoc network supporting voice traffic, based on a simiar probing mechanism. III. THE PROPOSED MAC SCHEME A. System Mode Consider a mobie ad hoc network having N mobie nodes. At the MAC ayer, a source node can communicate with one or more of its neighbors, with a separate queue for each destination node. Each node is assigned a unique transmitting code and a unique receiving code [10]. The network topoogy changes as the mobie nodes move. We use a time frame structure as shown in Fig. 2, in which time is partitioned into fixed-ength frames. In each frame, there are M information sots (ISs), each foowed by a short bocking sot 2 A DATA packet means an information packet from the traffic source. 4

5 (BS). The ISs are used for information (i.e., DATA and ACK) and probe transmissions, whie the BSs are used for transmitting bocking messages (to be discussed). Ony constant rate voice traffic is considered 3. Each voice ink requires a minima transmission rate (e.g., voice packet generation rate from the codec) so that its deay and jitter can be kept at a very ow eve. To keep the transmission accuracy of the voice ink, a signa to interference pus noise ratio (SINR) threshod is required for each ink, i.e., for the ink from sender i to receiver j, the foowing inequaity shoud hod: G i P t ijg ij I j + η j Γ d (1) where G i is the spreading gain, P t ij is the transmit power, g ij is the path gain from node i to node j, I j is the interference (received at node j) from other inks, η j is the background noise power at node j, and Γ d is the required SINR threshod for accurate information transmission. Unike the situation for the MESH scheme, a the transmissions are within a frequency band in our proposed scheme, using CDMA technoogy. Since RAKE reception can coect signa energy from different paths, we wi assume that there is no fast (mutipath) fading 4, and the transmit power is attenuated ony due to path oss with exponent β and shadowing. A node cannot transmit and receive at the same time, since each node s transmission and reception are within the same frequency band. B. Power Aocation for DATA and ACKs Random access wireess networks usuay use two power aocation strategies: common power aocation and inear power aocation [11]. A common transmit power is used by a the nodes in the common power aocation strategy [12]. In the inear power aocation strategy [13][14], the transmit power is proportiona to the path attenuation eve from the sender to the receiver. In a mobie ad hoc 3 Note that our scheme can sti work in the case with other traffic types such as video and data. Each time frame can be (dynamicay) divided into three portions, for voice, data, and video traffic, respectivey. Then our scheme can be appied in the voice portion of each frame. 4 For a non-idea case where fast fading is not fuy addressed by a RAKE receiver, the impact of the fuctuation due to fast fading can be represented by extra packet oss rate. The extra packet oss rate depends on the mobiity environment, the RAKE receiver structure, the error-correction technoogy, and so on. 5

6 network, it is difficut to estimate the path attenuation in advance from a sender to a receiver due to user mobiity. Here we consider a common power aocation strategy, in which each traffic source node uses a constant power P to transmit its DATA frames to its destination, and the traffic destination uses the same power eve to feed back ACKs. C. Mutipe Access Procedure The mutipe access capabiity of the network is based on resource reservation. Consider the case of a new ca arriva at source node i in time frame 0 to destination node j 5. The mutipe access procedure is iustrated in Fig. 3, whie the finite state machine is demonstrated in Fig. 4. C.1 Seection of DATA IS Before the ca arriva, node i is in state Ide. Upon the new ca arriva, node i first seects an IS for DATA transmission, referred to as its DATA IS, i.e., node i transits to state DATA IS Se. The ISs at which the destination node j is sending traffic shoud be skipped, because node j cannot send and receive at the same time at the same band. In frame 1, node i measures its experienced interference, and monitors the transmitting code of node j. Among the ISs over which node j s transmitting code is not detected, node i seects one with the minimum interference eve, say IS k, as its DATA IS. If no IS is avaiabe, the ca is dropped, i.e., node i proceeds to state Ca dropping. C.2 Transmissions of Probe and Bocking Messages Prior to DATA transmission in the seected DATA IS, node i shoud make sure its DATA transmission does not corrupt any existing reception at any active receiver 6 in the IS. To achieve this goa, node i 5 Note that our MAC scheme ony deas with unidirectiona traffic (from node i to j) rather than bidirectiona traffic (from node i to j and vice versa) for the foowing reason. In a mobie ad hoc network, it is ikey that there is no direct ink between a pair of users (say user A and user B). In this case, it is not necessary that traffic from user A to user B foows the same path as that in the reverse direction. Therefore, our scheme ony reserves resources for unidirectiona communications, and the routing protoco in the network ayer wi ensure that reservation is performed in both directions. 6 Here an active receiver can be an existing traffic destination node for DATA reception, or an existing traffic source node for ACK reception. 6

7 first sends a ow-power test signa into the network, i.e., in state DATA Probe Tx. Specificay, at IS k of frame 2, node i sends a probe spread by a common probe code with a arge spreading gain, and with transmit power eve α P (α 1). No data is carried in the probe. With its ow power, the probe is very unikey to corrupt any existing reception at the IS. Athough the probe is sent with ow power, we assume that it can be received by active receivers at the IS because of the arge spreading gain. At IS k, an active receiver, say node, first measures the received probe power eve denoted by P r,p and its own interference eve due to other existing transmissions. It estimates the potentia interference increase due to DATA transmission of the probe sender as P r,p /α, and determines whether the potentia interference increase can corrupt its own reception. If so, it sends a bocking message at BS k. The bocking message is spread by a common bocking code 7. No data is carried by the bocking message. The power aocation for the bocking message wi be discussed in Section III-D. At BS k, node i monitors the bocking code. If the detected bocking message power is above a threshod P r,b th, it gives up on IS k for DATA transmission and tries another IS unti no bocking message is detected. C.3 Seection of ACK IS If no bocking message is detected, node i sends a request message to destination node j, at IS k in frame 3 spread by node j s receiving code, and sends its first DATA frame at IS k in frame 4 spread by node i s transmitting code, both with power P. That is, the node proceeds to state RQT. Upon receiving the request message, node j seects an IS for its ACK, referred to as ACK IS, i.e., node j proceeds to ACK IS Se. Generay it is desired that, at the ACK IS, the interference experienced by node i is sma so that the ACK can be received correcty. So in the request message, node i indicates (among the ISs except IS k) the IS with minima experienced interference, say IS m. After node j has seected IS m as the ACK IS, it proceeds to state ACK Probe Tx. That is, it sends an 7 Unike [9] where a busy-tone is used over a separate contro channe, the bocking message in this research is sent in the data channe, thus avoiding the extra hardware cost to impement the busy-tone channe. 7

8 probe at IS m in frame 3 8 in the same manner as node i sending a probe for DATA IS. If no bocking message is detected at BS m, node j sends an ACK spread by its transmitting code at IS m in frame 4 for the DATA frame received 9 in frame 4, i.e., node j proceeds to state Info Rcv ACK. If node i does not receive an ACK at IS m of frame 4, it goes back to state DATA IS Se, i.e., seects another DATA IS, and repeats the preceding procedure with a maxima number of attempts. If an ACK is received, it proceeds to the next state Info Tx. C.4 Continuous Transmissions at Reserved DATA and ACK ISs At subsequent time frames from frame 5, node i in state Info Tx transmits its DATA frames spread by its transmitting code at IS k, and node j in state Info Rcv ACK transmits the ACKs spread by its transmitting code at IS m, unti the competion of the ca. It can be seen that caeve resource reservation is performed. Upon the ca competion, no ink termination procedure is necessary, because our scheme is based on rea-time interference measurement, and other active receivers can measure the interference eve reduction when the target ink stops using the reserved resource. D. Power Aocation for the Bocking Message If an active receiver at an IS is to send a bocking message in response to a probe from a potentia sender 10, the transmit power of the bocking message shoud be designed carefuy so that the potentia sender can be notified. The objective of bocking message power aocation is to ensure that the potentia sender can detect a bocking message with power above threshod P r,b th. At an IS, it is possibe that one or more potentia senders may send probes for resource reservation. Since the resource reservation is performed at the ca eve, it is reasonabe to assume that at an IS 8 The probe is sent at IS m in frame 3 when m > k, or in frame 4 when m < k. For simpicity of presentation, we use the former case as an exampe in this paper. 9 Since the request message and DATA are sent in the same way from node i, DATA can aso be received correcty if the request can be deivered successfuy. 10 Here a potentia sender can be a traffic source node for DATA transmission or a traffic destination node for ACK transmission. 8

9 there are at most two potentia probe senders in a neighborhood. This assumption is reasonabe due to the fact that the duration of a frame (usuay in the order of 10 ms) is much shorter than a ca duration (usuay in the order of minutes). Since each voice user may ony need one request (and thus one probe) for its ca duration, it is reasonabe to have the above assumption. We aso assume channe reciprocity in terms of shadowing effect. Consider the case in which an active receiver, say node, detects the probe power eve P r,p, and wants to send a bocking message. Node sends a bocking message with transmit power P t,b { 2 P r,b th = max αp P r,p, P r,b th P } Λ where Λ is the interference margin [15][16] of node, defined as the maximum extra interference that can be toerated by node. From the interference margin definition, we have G P g I + η + Λ = Γ d (3) where G and g are the spreading gain and path gain for the transmission to node, respectivey. Then Λ is given by Λ = G P g Γ d I η. (4) We first consider the scenario in which ony one potentia sender, node i, sends a probe at an IS. The path gain from node i to node is From (2) we have P t,b 2 P r,b th (2) g i = P r,p αp. (5) P r,p αp Due to the reciprocity of the channe, g i = g i, and so we have P t,b Then the received bocking message power at the potentia sender i is = 2 P r,b th. (6) g i 2 P r,b th. (7) g i P t,b g i 2 P r,b th (8) 9

10 which is greater than the detection threshod P r,b th. Hence the potentia sender is notified to give up the IS. Next, we consider the scenario in which two potentia senders, node i 1 and node i 2, send probes at an IS. As a common probe code is used, and no data is carried by a probe, node can coect energy from both probes via a RAKE receiver. The tota received probe power at node is aso denoted by P r,p, and node wants to send a bocking message. Without oss of generaity, assume the path gain from node i 1 to is not ess than that from i 2 to, i.e., g i1 g i2. We have P r,p = αp g i1 + αp g i2 2αP g i1. (9) When node sends a bocking message with power given by (2), the received bocking message power at node i 1 is P t,b g i1 = P t,b g i1 r,b 2 Pth αp P r,p th αp r,b 2 P 2αP g i1 = P r,b th. As a resut, node i 1 wi be notified to give up the IS. g i1 g i1 (10) In the foowing, we prove that, if node i 2 s potentia information transmission (i.e., DATA or ACK) generates extra interference that is intoerabe at node, node s bocking message with power given by (2) can reach node i 2 with a power eve above the threshod P r,b th. Proof: If node i 2 s information transmission can corrupt node s reception, this means P g i2 Λ. (11) We have g i2 Λ P. (12) When node sends a bocking message with power given by (2), the received bocking message power 10

11 at node i 2 is P t,b g i2 = P t,b P r,b th P Λ P r,b th P Λ = P r,b th. g i2 g i2 As node i 2 detects bocking message power above the threshod, it gives up the IS. Λ P (13) In summary, when there are two probe senders in an IS, and the aggregate potentia interference from their information transmissions is intoerabe by an active receiver, the potentia sender with stronger interference wi be bocked. The potentia sender with weaker interference wi be bocked if its potentia interference to the target active receiver is arger than the receiver s interference margin. E. Impact of User Mobiity and Soutions The impact of user mobiity on each active ink at the MAC ayer is two-fod. One is due to the time-varying path attenuation of the desired signa, and the other is due to the varying interference eve. The path attenuation consists of three components: path oss, shadowing, and fast (mutipath) fading. Fast fading can be addressed by the RAKE receiver that coects signa energy from different paths. The time-varying path oss, shadowing, and interference eve make the SINR of each ink fuctuate with time. Generay our scheme can automaticay adapt to the varying SINR of each ink, because each receiver determines whether or not to admit new inks into the network based on rea-time measurement and estimation of its desired signa and interference eves. However, it is possibe that the SINR of some ongoing inks at an IS cannot remain above the threshod Γ d (a condition referred to as a ink faiure) because of the fuctuations. Athough a new route searched by the routing scheme can hep to sove the probem, it is desirabe to dea with the probem at the MAC ayer first before resorting to the routing scheme, considering the overhead and time needed for a new route search, and the duration needed to set up a new route. To address the ink faiure probem, for each active ink, the traffic destination node indicates in its ACK two candidate DATA ISs at which the destination node experiences the minimum interference 11

12 and does not transmit. The destination node aso indicates a detected ink faiure (if any) in the ACK. If the traffic source node (say node i) does not receive correcty the ACK from its destination (say node j), this means the ACK is corrupted 11. As node i does not have the information of whether the DATA transmission is corrupted or not, we et the ink be re-estabished in a new DATA IS and a new ACK IS. The main difference from the estabishment procedure given in Section III-C is as foows. Node i sends two probes in the two candidate DATA ISs indicated in the previous ACK. At each candidate IS, a request is sent if no bocking message is detected. The request aso indicates two candidate ACK ISs in which node i experiences the minimum interference and does not transmit. If at east one request is received at node j (which means a new DATA IS is estabished), node j sends two probes at the candidate ACK ISs. If at east one probe at a candidate ACK IS is not bocked by active receivers at that IS, a new ACK IS is estabished. If the new DATA IS and ACK IS cannot be estabished simutaneousy, the ca is dropped. On the other hand, if node i successfuy receives an ACK which indicates that a faiure of DATA transmission happens, a simiar procedure is executed, except that ony the estabishment of a new DATA IS is needed. Generay, if the ink faiure is due to arge interference, it is ikey that the ca can be estabished successfuy in other DATA and/or ACK ISs at which nodes generating arge interference are not active. However, if the ink faiure is due to a arge path oss of the desired signa (e.g., the traffic source and destination are separated by a arge distance) or due to shadowing of the desired signa (e.g., the path between the traffic source and destination is bocked by a ta buiding), it is very ikey that the ink quaity of the target ca at other ISs is not good enough either, and thus the ca wi be dropped. In this case, a new route shoud be seected by the routing scheme. 11 If an expected DATA frame is not received, the traffic destination node responds with a NACK. For presentation simpicity, we use ACK to represent both ACK and NACK in this paper. 12

13 F. Summary The proposed MAC scheme has the desired features discussed in Section I. The scheme is performed in a distributed manner at the mobie nodes, and is thus scaabe to arge networks. The hidden termina probem does not exist, since each active receiver makes an admission decision on a new ca based on rea-time measurement of its own desired signa and interference eves. The reservation nature of our scheme can guarantee sma deay and jitter for voice traffic as ong as the minimum required rate can be met. Our scheme can adapt automaticay to user mobiity to a certain extent, aso benefiting from the fact that each active receiver makes rea-time measurements of desired signa and interference eves. In addition, the idea assumption required by CSMA and busy-tone based schemes (as discussed in Section II) is unnecessary for our scheme. Furthermore, athough the bocking message has simiar functionaity to the busy-tone in busy-tone based schemes, our scheme does not have the probem of different channe gains in the separate busy-tone and DATA frequency bands, because the bocking message and DATA are sent in the same band. IV. STATISTICAL MULTIPLEXING OF VOICE TRAFFIC The resource reservation procedure in the preceding section is suitabe for constant rate voice traffic, i.e., the sender generates packets at a constant rate during the ca ifetime. However, voice traffic is normay characterized by an on/off nature. No information packets are generated in an off state, thus posing some chaenges on the resource reservation procedure. When an on voice fow turns off, other active receivers measure ess interference. If subsequenty one or more new cas are admitted at the serving ISs of the off voice fow, and ater the off voice fow switches to the on state, it is ikey that some active receivers at the voice fow s serving ISs experience more interference than they can toerate, thus eading to ink corruption. This probem is due to the inaccurate interference estimation when a voice fow turns off. Furthermore, it is more desirabe to achieve mutipexing gain of on/off voice traffic. To address these probems, the foowing soutions are proposed. 13

14 A. Soution to Inaccurate Interference Estimation When a voice fow is in an off state, it uses a probe (referred to as off probe) with a different code from that in Section III-C to inform other inks of the interference that wi be generated when the off voice fow turns on again. Specificay, when a voice fow from node i to node j turns off, its source and destination nodes send an off probe on the DATA and ACK IS, respectivey, with power αp and a arge spreading gain. Each active receiver keeps monitoring the off probe. At an active receiver at the DATA or ACK IS, et P r,p,off denote the received off probe power. Then when a new ca sender sends a probe (for admission) that arrives at node with power eve P r,p, node wi send a bocking message if P r,p α + P r,p,off α > Λ. (14) This means the potentia interference of off voice fow from node i to node j is counted in node s current interference eve, and thus the inaccurate interference estimation probem can be addressed. When the voice fow from node i to node j turns on ater, nodes i and j can send the DATA and ACK frame at the DATA IS and ACK IS, respectivey, without any re-admission procedure. B. Soution for Statistica Mutipexing In (14), each active receiver at an IS counts 100% of the potentia interference from off voice fows, and thus there is no statistica mutipexing gain. However, if an active receiver counts ony a portion ξ (< 1) of the potentia interference from off voice fows, ony part of the reserved capacity for off voice fows is maintained. When a new ca sender sends a probe (for admission) that arrives at an active receiver node with power eve P r,p, node wi send a bocking message if P r,p α + ξ P r,p,off α > Λ (15) where ξ is referred to as reservation factor. By this means, ony a portion ξ of the capacity for the off nodes is reserved. In other words, new cas can use a portion 1 ξ of the capacity for the off nodes. 14

15 On the other hand, when an off voice fow turns on and its source and destination nodes start to transmit in its previous DATA and ACK ISs, respectivey, it is possibe that the transmission accuracy of other inks at the ISs may be corrupted. So re-admission of off-to-on voice fows is necessary as foows. When an off voice fow turns on, the voice source node sends a probe in its previous DATA IS and in another IS at which the voice destination node experiences the east interference (as indicated in the ast ACK). At either IS, if no bocking message is detected, the voice source node sends a request which aso indicates a candidate ACK IS (different from its previous ACK IS) at which it experiences the east interference and does not transmit. If the voice destination node receives at east one request, it sends a probe in its previous ACK IS and the candidate ACK IS. A new ACK IS is estabished if at east one probe is not bocked. The ca wi be dropped if a DATA IS and an ACK IS cannot be estabished simutaneousy. Note that the admission of an off-to-on voice request is different from that of a new ca request. For a new ca request, a certain portion of the capacity shoud be reserved for off voice fows, as shown in (15). On the other hand, for an off-to-on voice request, such reservation is not necessary, and thus we have the foowing criterion for bocking a request P r,p α > Λ. (16) Therefore, when an off voice fow turns on, the source or destination node shoud use a different probe from that used by the source or destination of a new voice ca (i.e., different from that used in Section III-C). By this means, other active receivers can distinguish off-to-on voice requests from new voice ca requests. V. PERFORMANCE EVALUATION We use computer simuation to evauate the performance of the proposed scheme. Generay it is desired to compare our scheme with other MAC schemes. However, to the best of our knowedge, so far no MAC scheme is designed to address user mobiity issue, and thus it is difficut to achieve a fair 15

16 comparison. Therefore, in the foowing we ony demonstrate the effectiveness of our proposed scheme to support voice traffic with QoS provisioning. Traditiona QoS metrics for voice traffic incude deay, jitter, and packet oss rate. From Section III, it can be seen that, as ong as a ca is admitted into the network, the transmitter can send voice packets continuousy in the reserved sot. So the voice deay is fixed, and is up to the duration of a time frame, and there is no jitter. Therefore, in the foowing, packet oss rate is used as the major QoS metric for voice in the simuation. The simuation is done using Matab. Consider an ad hoc network consisting of N nodes, with N/2 voice source nodes and N/2 voice destination nodes. The source nodes are randomy distributed in a 1000 meter 1000 meter square, each associated with a destination node. A the source nodes are fixed, and a the destination nodes move with a speed v in a randomy seected direction. A destination node is assumed to move within a radius of 150 m of the associated source node 12. Each time frame has a fixed duration of 20 miiseconds (ms), and is further divided into 8 ISs and 8 BSs. The spreading gain is 32 for DATA and ACKs, and 3200 for the probe. The path oss attenuation exponent β is 2.4, and the shadowing is modeed by the first-order autoregressive process given in [17]. Voice ca inter-arriva time at each source node is exponentiay distributed with mean vaue 48 seconds, and the duration of a ca is exponentiay distributed with mean vaue 30 seconds. When a ca is active, one DATA frame is generated over each time frame. The SINR requirement Γ d is 10 db. In the foowing, the simuation for each set of parameters ends unti on average each source node finishes 100 cas. In the simuation, the number N of nodes varies from 20 to 100, and the destination node veocity v varies from 0 to 20 km/hour (kph). We first verify our assumption in Section III-D that the probabiity of more than two probe senders at an IS is negigibe. Among the ISs with one or more probe senders, Tabe I ists the percentage of ISs that have one probe sender, two probe senders, and more than two probe senders, respectivey. No more than two probes are observed at an IS, thus vaidating our assumption. 12 Routing may be invoved when a destination node moves beyond this distance, a situation which is beyond the scope of this anaysis. 16

17 First we compare our scheme with the IEEE medium access contro scheme. In the simuation, N=100, no node is moving, and an active source node generates constant rate voice fows. It is observed that the 95% confidence interva of the mean packet deay in the IEEE medium access contro scheme is [1.302, 1.303] second. It is aso observed that the jitter in the IEEE medium access contro scheme is 0.7 second. On the other hand, our scheme is reservation based. Thus, the packet deay is a frame duration, i.e., 20ms, and the jitter is zero. Apparenty our scheme performs better in terms of timey deivery of voice packets. Next consider constant rate voice fows in our scheme. For an ongoing voice ca, if its SINR requirement is not satisfied, it tries to switch from one IS to another. Some voice frames wi be ost during the switching due to the switching processing time, and if the switching is not successfu, ca dropping occurs. Fig. 5 shows the voice frame oss rate with different N and v vaues. A arger vaue of N and/or v tends to increase the frequency of switching among the ISs, thus increasing the frame oss rate. It can aso be seen that a the frame oss rates are bounded by 0.1%, which is we within acceptabe range for voice service. It is aso observed that the ongoing ca dropping probabiity is bounded by 1.2%, which happens when N = 100 and v = 20 kph. So our scheme can maintain the frame oss rate and ca dropping rate at ow eves. Then we consider on/off voice fows with statistica mutipexing in our scheme. The durations of on and off states are independent and exponentiay distributed with mean vaues 352 ms and 650 ms, respectivey. For N = 100 case, Tabe II shows the ongoing ca dropping probabiity and new ca bocking probabiity. When the reservation factor increases, more capacity is reserved for off voice fows, and thus the ongoing ca dropping probabiity decreases. However, the new ca bocking probabiity increases due to the reduced avaiabe capacity for new cas. 17

18 VI. CONCLUSIONS It is chaenging to support voice traffic over a mobie ad hoc network due to the deay-sensitive nature of such traffic and due to user mobiity. In this paper we have proposed a MAC scheme to support voice traffic by the resource reservation mechanism. User mobiity is addressed by rea-time signa and interference measurement and estimation at the receivers. Statistica mutipexing is achieved by partia resource reservation for off voice fows. The resuts of this study provide some usefu insight into the deveopment and depoyment of mobie ad hoc networks supporting mutimedia services. REFERENCES [1] Z. J. Haas and J. Deng, Dua busy tone mutipe access (DBTMA) a mutipe access contro scheme for ad hoc networks, IEEE Trans. Commun., vo. 50, no. 6, pp , June [2] P. Wang, H. Jiang, and W. Zhuang, A dua busy-tone MAC scheme supporting voice/data traffic in wireess ad hoc networks, in Proc. IEEE GLOBECOM 06, San Francisco, Caifornia, USA, Nov. Dec [3] C. R. Lin and M. Geria, Asynchronous mutimedia mutihop wireess network, in Proc. IEEE INFOCOM 97, pp , Kobe, Japan, Apr [4] A. Muir and J. J. Garcia-Luna-Aceves, Group aocation mutipe access with coision detection, in Proc. IEEE INFOCOM 97, pp , Kobe, Japan, Apr [5] S. Jiang, J. Rao, D. He, X. Ling, and C. C. Ko, A simpe distributed PRMA for MANETs, IEEE Trans. Veh. Techno., vo. 51, no. 2, pp , Mar [6] Z. Ying, A. L. Ananda, and L. Jacob, A QoS enabed MAC protoco for muti-hop ad hoc wireess networks, in Proc IEEE Internationa Performance, Computing, and Communications Conference (IPCCC 03), pp , Phoenix, AZ, USA, Apr [7] S.-T. Sheu and T.-F. Sheu, DBASE: A distributed bandwidth aocation/sharing/extension protoco for mutimedia over IEEE ad hoc wireess LAN, in Proc. IEEE INFOCOM 01, pp , Anchorage, Aaska, USA, Apr

19 [8] C. W. Ahn, C. G. Kang, and Y. Z. Cho, Soft reservation mutipe access with priority assignment (SRMA/PA): a nove MAC protoco for QoS-guaranteed integrated services in mobie ad-hoc networks, in Proc. IEEE Vehicuar Technoogy Conference (VCT 00), pp , Boston, MA, USA, Sept [9] H. Jiang, P. Wang, W. Zhuang, and X. Shen, An interference aware distributed MAC scheme for CDMA-based wireess mesh backbone, in Proc. IEEE Consumer Communications and Networking Conference (CCNC) 2007, pp , Las Vegas, Nevada, USA, Jan [10] E. S. Sousa and J. A. Sivester, Spreading code protocos for distributed spread-spectrum packet radio networks, IEEE Trans. Commun., vo. 36, no. 3, pp , Mar [11] T. Moscibroda and R. Wattenhofer, The compexity of connectivity in wireess networks, in Proc. IEEE INFOCOM 06, Barceona, Spain, Apr [12] S. Narayanaswamy, V. Kawadia, R. S. Sreenivas, and P.R. Kumar, Power contro in ad hoc networks: theory, architecture, agorithm, and impementation of the COMPOW protoco, in Proc. European Wireess 02, pp , Forence, Itay, Feb [13] S. Banerjee and A. Misra, Minimum energy paths for reiabe communication in muti-hop wireess networks, in Proc. 3rd ACM MOBIHOC, pp , Lausanne, Switzerand, June [14] J. E. Wiesethier, G. D. Nguyen, and A. Ephremides, Energy-efficient broadcast and muticast trees in wireess networks, Mobie Networks and Appications, vo. 7, no. 6 pp , Dec [15] F. Cuomo, C. Marteo, A. Baiocchi, and F. Capriotti, Radio resource sharing for ad hoc networking with UWB, IEEE J. Seect. Areas Commun., vo. 20, no. 9, pp , Dec [16] A. Muqattash and M. Krunz, Power controed dua channe (PCDC) medium access protoco for wireess ad hoc networks, in Proc. IEEE INFOCOM 03, pp , San Francisco, Caifornia, USA, Mar. Apr [17] G. L. Stuber, Principes of Mobie Communication. Kuwer Academic Pubishers, Norwe, MA, USA, 1996, pp

20 TABLE I THE PERCENTAGE OF THE ISS WITH DIFFERENT NUMBER OF PROBES. Number (N) of nodes v = 10 1 probe kph 2 probes probes v = 20 1 probe kph 2 probes probes TABLE II ONGOING CALL DROPPING PROBABILITY AND NEW CALL BLOCKING PROBABILITY FOR THE CASES WITH on/off VOICE FLOWS. ξ v = 0 kph Dropping prob. (%) Bocking prob. (%) v = 10 kph Dropping prob. (%) Bocking prob. (%) v = 20 kph Dropping prob. (%) Bocking prob. (%)

21 Interference range A C D B Desired signa Interference range Interference Fig. 1. An exampe in which the idea assumption is not vaid. IS 1 BS 1. IS 2 BS 2.. IS M BS M Frame IS: information sot BS: bocking sot Fig. 2. Time frame structure. 21

22 Frame 0 Frame 1 Frame 2 Frame 3 Frame 4, 5,... k k k m m k m Ca arriva at i for j Interference measurement by i Request by i DATA by i Probe by i Bocking message Probe by j ACK by j IS BS Fig. 3. The mutipe access procedure. Bocking For receiver For transmitter Ide RQT received No ACK ACK_IS_ Se ACK_Prob e_tx No bocking Info. competed Info_Rcv_ ACK Ca arriva DATA_IS_ Se Success Bocking DATA_Pro be_tx No bocking RQT ACK received Info_Tx Faiure Ca dropping Frame 1 Frame 2 Frame 3 Frames 4, 5,... Fig. 4. The finite state machine of the medium access protoco. 22

23 6 x veocity = 20 kph veocity = 10 kph veocity = 0 kph Voice frame oss rate Number (N) of voice nodes Fig. 5. The voice frame oss rate of a voice ca. 23