Beacon-driven Leader Based Protocol over a GE Channel for MAC Layer Multicast Error Control

Transcription

1 I. J. Counications, Network and Syste Sciences, 2008, 2, Published Online May 2008 in Scies ( Beacon-driven Leader Based Protocol over a GE Channel for MAC Layer Multicast Error Control Zhao LI, Thorsten HEFET 2 Student Meber IEEE, USTC CS China & Saarland University, Gerany 2 Senior Meber IEEE, Saarland University, Gerany E-ail: {li, herfet}@nt.uni-saarland.de Abstract In wireless networks current standard MAC layer protocols don t provide any error correction schee for broadcast/ulticast. In this paper, we enhance a Leader Based Protocol (LBP and propose a Beacon-driven Leader Based Protocol (BLBP for the MAC layer ulticast error control. To guarantee a very low Packet Loss atio (PL under strict delay constraints for video ulticast over a Gilbert-Elliott (GE channel, we analyze BLBP and copare it with LBP and different application layer ulticast error control schees via siulation experients. Both the theoretical analysis and siulation results show that BLBP can correct nearly all the errors for all receivers in the MAC layer and is ore efficient than LBP. BLBP is also ore efficient than the application layer Autoatic epeat request (AQ schee and the total ulticast delay is uch shorter. BLBP is very good for real-tie ulticast applications with strict delay constraints. Keywords: BLBP, Multicast Error Control, Gilbert-elliott Channel. Introduction With the rapid developent of wireless networking technologies, it is becoing possible to supply wireless terinal users not only with data connections, but also with real-tie counication services. The eerging real-tie ulticast applications in wireless networks include the local distribution of High Definition TV (HDTV and Digital Video Broadcasting (DVB [], video-on deand, video conferencing, gaing, local VoIP, IPTV, Internet-adio distribution, P2P broadcasting, etc. Most of these applications require a very low Packet Loss atio (PL under strict delay constraints. The characteristics of wireless network can be suarized as a bandwidth variation and terinal heterogeneity plus a high degree of packet losses. It is known that the Gilbert-Elliot (GE channel [2,4] with a 2-state Markov odel is a good approxiation for the packet loss odel in wireless channels. However, the current standard MAC layer protocols don t provide any error correction schee for broadcast/ulticast. Hence the ulticast error is controlled in the application layer. The existing application layer ulticast error control schees include autoatic repeat request (AQ, forward error correction (FEC and hybrid error correction (HEC [5 0]. Unfortunately the total ulticast delays of the application layer schees are always high and soeties do not satisfy the strict application delay constraints, or these schees are not efficient when the delay constraints are short. Copared with application layer schees, MAC layer ulticast error control schees recover the ulticast loss locally and lead to uch shorter delays. Currently, very few reliable MAC layer ulticast schees, such as Leader Based Protocol (LBP [], have been proposed for IEEE 802. based wireless networks. LBP elects one of the ulticast group receivers as the leader. On erroneous reception of a data frae, the leader does not send an acknowledgeent (ACK, propting a retransission. On erroneous reception of the data frae at the non-leader receivers, LBP allows negative acknowledgeents (NACKs fro these receivers to collide with the ACK fro the leader, thus destroying the ACK and propting the sender to retransit the data frae. We refer to this ACK/NACK ja as JACK. However, LBP suffers fro two ain probles: First, when the entire data frae is lost, the non-leader Copyright 2008 Scies. I. J. Counications, Network and Syste Sciences, 2008, 2,

2 BEACON-DIVEN LEADE BASED POTOCOL OVE A GE CHANNEL 45 FO MAC LAYE MULTICAST EO CONTOL receivers can not reply NACKs because they don t know when or how to send the, as the destination is unknown especially when equest-to-send (TS and Clear-To-Send (CTS is not used for sall data fraes. As a result, LBP is not reliable for the non-leader receivers. Second, LBP has poor perforance when the channel error rates are high. The non-leader receivers send NACKs whenever the received frae is in error, regardless of whether this erroneous frae has been received correctly before or not. This is because the receivers in LBP can not access the data frae sequence nuber before the frae is received, as there is no such field in the structure of TS/CTS fraes for ulticast. So the sender has to retransit until all receivers receive the data frae correctly at the sae tie. There are a lot of unnecessary transissions; hence LBP is not efficient particularly for lossy channels. In this paper, we enhance LBP and propose a Beacon-driven Leader Based Protocol (BLBP. The sender sends a beacon frae before the data frae to lead the non-leader receivers to set tiers and to announce the sequence nuber of the following data frae. BLBP solves the probles of LBP well. Each of the non-leader receivers can send feedback when the tier ties out. Both the leader receiver and non-leader receivers can send ACK and NACK respectively based on sequence check, hence avoids the unnecessary transissions in LBP. To guarantee a very low PL under strict delay constraints for video ulticast over a GE channel, we analyze BLBP and copare it with LBP and two application layer ulticast error control schees via siulation experients. One is an AQ based schee fro [8], called HEC-P, which cobines a NACK based AQ schee with a packet repetition (P technique. The other one is a nearly optial application layer ulticast error control schee called Hybrid AQ (HAQ Type I [9,0], which cobines FEC and NACK based AQ schee together. The reainder of this paper is organized as follows. Section 2 presents the related work. In section 3, we exaine why it is necessary to correct ulticast errors in the MAC layer. We describe BLBP in section 4 and analyze its perforance over a GE channel in section 5. And in section 6, we evaluate the perforance of BLBP and copare it with LBP, HEC-P and HAQ Type I via siulation experients. We conclude in section elated Work For the application layer ulticast error control, any authors [5,8] studied the AQ based schees and concluded that when cobined with feedback suppression and other accessorial techniques, AQ is effective to repair ulticast packet losses for sall groups with low error rates. However, the application layer AQ always take a long tie and they are not efficient at high error rates and with large nubers of receivers due to feedback iplosion and the liitation to scale. Another technique coonly used to handle losses for ulticast in the application layer is FEC, whereby redundant inforation in the data strea enables the receiver to correct losses without contacting the sender. izzo [6] studied the feasibility of software encoding/decoding for packet-level FEC. A (n, k block erasure code converts k source packets into a group of n coded packets, such that any k of the encoded packets can be used to reconstruct the k source packets. Usually, the first k packets in each group are identical to the original k data packets; the reaining n-k packets are referred to as parity packets. The advantage of using block erasure codes for wireless ulticasting is that a single parity packet can be used to correct independent single-packet losses aong different receivers. The integrated FEC/AQ schees or any other kinds of cobination of ore than one error control schees are referred to as HEC schees in this paper. Previous works [7 0] indicate that HEC schees are uch ore efficient for recovering data packets than the schees with either FEC or AQ alone. We consider the HEC- P schee fro [8], which cobines a NACK based AQ schee and a packet repetition technique. The nuber of feedback/retransission rounds and the nuber of packet repetitions in each round are adapted to the network condition. HEC-P is actually an AQ based schee without FEC coding. We also consider HAQ Type I fro [9,0], in which the sender sends a certain aount of parity packets using FEC following the original k data transissions. If the loss rate obtained after reconstruction at the receiver is still too high, AQ is used to retransit ore parity packets. Tan [0] developed forulas to optiize the perforance of HAQ Type I while guaranteeing the required PL under strict delay constraints. HAQ Type I is a nearly optial application layer ulticast error control schee. However, these application layer ulticast schees always take long ulticast delays or they are not efficient when the delay constraints are short. So we study MAC layer ulticast error control schees to support real-tie ulticast with strict delay constraints. For the reliable MAC layer ulticast, besides LBP [], Tourrihes [2] proposed a robust broadcast using a collision detector to infor the sender whether the broadcast packet is successful or not. However, this schee can not guarantee the reliability of ulticast transissions because the feedbacks are only fro the detector instead of all receivers theselves. Gupta et al. [3] proposed a tone-based solution for ulticast in both infrastructure and ad-hoc 802. networks. They use dual busy tones to siulate NACKs or Negative CTS (NCTS. Although this schee is good to detect and correct the ulticast errors, it requires an additional Copyright 2008 Scies. I. J. Counications, Network and Syste Sciences, 2008, 2,

3 46 Z. LI ET AL. channel for the tone, which is not always feasible in practice. Our BLBP enhances LBP with a beacon frae to lead the non-leader receivers to set tiers and to announce the sequence nuber of the following data frae. BLBP avoids the probles of LBP and is ore efficient. 3. Motivation The eerging real-tie ulticast applications in wireless networks (such as wireless HDTV, DVB, gae, video conference require strict delay constraints. Error recovery based on application layer AQ is suboptial because the end-to-end application layer feedback and retransission take a too long tie due to application layer protocol waiting, MAC layer queuing, hardware handling, etc. Moreover, the application layer AQ based schees are not efficient when the error rates are high and the nubers of receivers are large due to feedback iplosion and the liitation to scale. The FEC coding based application layer schees can satisfy the strict delay constraints but they are not efficient when the delay constraints are very strict particularly for sall ulticast groups. And the FEC based ones are not adaptive to the heterogeneity of receivers because the code has to be set based on the receiver with the worst channel condition. Copared with application layer schees, MAC layer ulticast error control schees take a uch shorter tie due to the faster feedback and retransission. Due to the JACK schee, BLBP and LBP even achieve coplete feedback suppression. So the MAC layer ulticast error control schees are very good for real-tie ulticast applications. For non-realtie ulticast applications such as file disseination and shared whiteboards, reliable MAC layer ulticast saves tie as well as both network and end-syste resources. Moreover, for ulti-hop ulticast with wired network and wireless LAN as the last hop, the need for additional transissions due to errors in the wireless LANs puts unnecessary processing burden on the original reote sender. These additional transissions go over the entire wired ulticast tree and also the wireless links, taking a long tie, wasting bandwidth and also leading to processing of unwanted redundant retransissions at those receivers which ight have already received the packet. The siilar thing also happens in ulti-hop wireless networks such as wireless esh networks and wireless ad hoc networks. If the access points (AP (or base stations were to take the responsibility of supplying retransissions rather than the original sender, then the load of supplying retransission gets distributed across access points and takes a shorter tie. The total error correction cost will be uch shorter and it is easier to guarantee the final PL under strict delay constraints in the application layer. 4. Main Schee of BLBP The MAC layer reliable ulticast BLBP requires a slight odification to the IEEE 802. MAC layer protocols. As entioned earlier, 802. DCF (Distributed Coordination Function unicast assued TS/CTS is switched on to solve the hidden terinal proble is ore reliable than broadcast/ulticast, because unicast uses TS/CTS signaling and ACK/retransission schee in the MAC layer and broadcast/ulticast does not. BLBP enhances LBP with a MAC control frae called beacon shown in Figure. Besides the sae fields in TS/CTS fraes, such as frae control header, transission duration, receiver address (A, transitter address (TA and frae check sequence (FCS, the beacon frae also includes the sequence nuber of the following data frae. The use of the beacon frae is to lead the non-leader receivers to set tiers and to announce the sequence nuber of the following data frae. Figure. The forat of the beacon frae The ain schee of BLBP is shown in Figure 2. A receiver is selected as the leader for the ulticast group. The AP first sends a TS frae to all receivers, and only the leader receiver transits a CTS frae in reply to the AP. The AP is then assured that the channel is granted and starts the transission of the beacon frae with the sequence nuber of the following data frae. On receipt of the beacon frae, each of the non-leader receivers sets a tier according to the beacon frae. The AP then transits the data frae following the beacon frae. The leader receiver replies an ACK frae if the data is correct or it has already got the data based on sequence check, or does nothing otherwise. When the tier ties out, each non-leader receiver replies a NACK if the data is error and it has not received it correctly yet based on sequence check, or does nothing otherwise. Then if the AP receives an ACK, this transission is done. Otherwise, the AP repeats the whole procedure and retransits again until the nuber of ties is beyond the retransission liit. For exaple, in the retransission phase in Figure 2, although this tie the data frae is lost, the leader receiver still replies an ACK because it knows this data frae has been received correctly already in the first transission, thanks to the beacon frae. For the deterination of a leader, we use a schee Copyright 2008 Scies. I. J. Counications, Network and Syste Sciences, 2008, 2,

4 BEACON-DIVEN LEADE BASED POTOCOL OVE A GE CHANNEL 47 FO MAC LAYE MULTICAST EO CONTOL fro LBP []: The first receiver that joins the ulticast group acts as the leader. The group is cancelled if there is no leader. The other group ebers can rebuild/rejoin the ulticast group if necessary when tie out. Please note that it is possible to reduce the aount of control traffic flow for leader election purposes when a higher layer group anageent protocol like IGMP (Internet Group Manageent Protocol is running above the link layer []. The leader does not affect the perforance because BLBP supplies fair service for all receivers. 5.. GE Channel Model The GE channel odel is a two-state Markov chain shown in Figure 3. In the Good state (G errors occur with (low probability P G while in the Bad state (B they occur with (high probability P B. The errors occur in clusters or bursts with relatively long error-free intervals (gaps between the. The state transition is suarized by its transition probability atrix in forula (. Figure 3. GE channel odel Figure 2. Main schee of BLBP (i denotes receiver i BLBP solves the probles of LBP well. All the nonleader receivers can send feedbacks when the tiers tie out. Both the leader receiver and non-leader receivers send ACK and NACK respectively based on sequence check thanks to the beacon frae, hence it avoids the unnecessary transissions in LBP. Clearly, BLBP can correct all the errors for all receivers due to the ACK/NACK feedback and retransission in the MAC layer. BLBP is even ore efficient than the application layer AQ schees because BLBP suppresses the ultiple feedbacks into just a JACK. However, the loss of beacon fraes will decrease the perforance (reception rate of the non-leader receivers. Fortunately, the beacon fraes are uch ore reliable (nearly error free than data fraes because they are uch saller and are transitted using the lowest data rate, like other control fraes in 802. (TS, CTS, ACK. Moreover, due to TS/CTS signaling, the beacon fraes also avoid collision loss. Please also note that BLBP can run without TS/CTS exchanges for sall data fraes just like 802. DCF unicast. Although our discussion is in the context of 802. DCF, BLBP is actually applicable to all ACK/retransission based MAC protocols, such as 802. PCF (Point Coordination Function etc. 5. Perforance Analysis In this section we first introduce the GE channel odel and then analyze the theoretical perforance of BLBP over the GE channel odel with both teporal error correlation and spatial error correlation. β β P = α α ( To reflect ost reasonable choices for real scenarios, it is assued that P G =0 and P B =. This odel is always referred to as the siplified GE odel. Our analysis and siulation experients in the following sections are based on the siplified GE odel. The occupancy ties for state B and G are both geoetrically distributed with respective eans ( α and ( β, which are also called as the expected error burst length and the expected error free length respectively. The steady state probabilities of being in states G and B are π G = ( α /(2 α β and π B = ( β /(2 α β respectively. So the average packet loss rate produced by the GE channel odel is PG( α + PB( β p = PGπG + PBπB = ( α + β (2 For the siplified GE channel odel, the PL will be ( β p = ( α + β (3 Following [4], the variance of the error sybol (or packet X is σ 2 = EX ( p 2 = p( p. So we get the correlation of two consecutive error sybols X and X 2 : E(( X p( X2 p τ = = α + β (4 2 σ which is also referred to as the teporal error correlation. Finally, the two paraeters of the siplified odel (α and β can be expressed in the ters of the ore eaningful quantities p and τ by solving forulas (3 and (4. These yields Copyright 2008 Scies. I. J. Counications, Network and Syste Sciences, 2008, 2,

5 48 Z. LI ET AL. α = p+ τ ( p (5 β = ( p + τ p (6 The transition probability atrix then becoes p( τ p( τ P = ( p( τ ( p( τ And the I-step transition probability atrix is: I I I p( τ p( τ P = I I ( p( τ ( p( τ (7 (8 Now we copute P[a,b], the probability of a errors in a sequence of b sybols following [4]. Let P G [a,b] be the probability of a errors in b transissions with the channel ending in state G. Siilarly, let P B [a,b] be the probability of a errors in b transissions with the channel ending in state B. Then P[ ab, ] = P[ ab, ] + P[ ab, ] (9 G For b =, 2, 3 and a =0,, 2 b, assuing the siplified GE channel, then P [ a, b] = P [ a, b ] β + P [ a, b ]( α (0 G G B P [ a, b] = P [ a, b ] α + P [ a, b ]( β ( B B G The initial conditions for the recursion are G B P [0,0] = ( α /(2 α β (2 P [0,0] = ( β /(2 α β (3 B and PG[ a,0] = PB[ a,0] = 0 for a 0. Note that with these initial conditions, all nuerical values coputed will be steady state results BLBP over the GE Channel Model with Teporal Error Correlation Now we analyze BLBP over the GE channel odel. As in ost referenced papers, it is assued that the MAC layer control fraes (TS/CTS/Beacon/ACK are error free fro the error odel. This also akes sense in practice because the control fraes are very sall and are sent in the lowest data rate, and hence they are ore reliable than data fraes. We first consider only the teporal error correlation. In other words, it is assued that the error events at different receivers are independent. To be clear, we show the used sybols here. P: The original packet error rate for all receivers; : The nuber of receivers; : The retransission liit; N: The total nuber of transissions required to transit a ulticast packet correctly to all the receivers; N r : The nuber of transissions required for receiver r to receive a packet correctly; PL target_ac : The PL target in the MAC layer; D target_ac : The Delay target in the MAC layer. First we consider the final PL for receiver r, shown in forula (4. PL( r = P[ +, + ] = pα (4 About the deterination of the retransission liit, there are two constraints, the PL constraints and the delay constraints which are shown as follows: p PL t arg et _ ac α (5 T * BLBP D t arg et _ ac (6 where T BLBP =T CC + T TS + T CTS + T BEACON + T DATA + T ACK + DIFS + 4SIFS is the tie of one transission in BLBP. T CC denotes the channel contention tie which can be calculated theoretically following [5] or by easureents in practice. T TS, T CTS, T BEACON, T DATA, and T ACK are the transission ties of fraes TS, CTS, BEACON, DATA and ACK respectively. DIFS denotes the Distributed Inter Frae Space while SIFS is the Short Inter Frae Space. Note that the PL target and the delay target ay not be satisfied at the sae tie soeties, especially when the delay constraint is too strict. We will explore this further in siulation experients. Now we consider the expected nuber of transissions for one ulticast data packet. The probabilities that N r n, Nr = n and Nr > n are shown in forulas (7, (8 and (9 respectively. P[ N n] = P[ n, n] r = + = 0, n = 0 n pα, n,2... PN [ = n] = PN [ n] PN [ n ] r r r n 2 n pα pα, n = 2, = p, n = n pα,,2... PN [ r > n] = Pnn [, ] =, n = 0 (7 (8 (9 So we get the expected nuber of transissions for one ulticast data packet required for receiver r: p( α EN ( r = PN [ r > n] = + ( α 0 (20 Next the probabilities that N n, N = n and N > n are shown in forulas (2, (22 and (23 respectively. Copyright 2008 Scies. I. J. Counications, Network and Syste Sciences, 2008, 2,

6 BEACON-DIVEN LEADE BASED POTOCOL OVE A GE CHANNEL 49 FO MAC LAYE MULTICAST EO CONTOL PN [ n] = PN [ n] n ( α r = p, n =, = 0, n = 0 PN [ = n] = PN [ n] PN [ n ] n n 2 ( α ( α p p, n = 2, = ( p, n = PN [ > n] = PN [ n] n ( α p, n =,2... =, n = 0 r (2 (22 (23 Finally we get the expected nuber of transissions for one ulticast data packet for all receivers, shown in forula (24. 0 ( ( n pα EN [ ] = PN [ > n] = + (24 Siilarly, we get the expected nuber of transissions for one ulticast data packet for all receivers in LBP, shown in forula (25. ( ( E [ N] = P[,] LBP 0 n ( ( p (25 = 0 And the redundant inforation (I of BLBP is: I = E[ N] (26 Siilar as in [6], aong pure AQ based ulticast error control schees, BLBP needs the iniu nuber of transissions (shown in forula (24 to let all receivers receive the packet correctly. There are no unnecessary transissions due to the sequence check schee and the coplete feedback suppression based on the JACK schee. We will explore this further in siulation experients by coparing BLBP with LBP and an application layer pure AQ schee BLBP over the GE Channel Model With Both Teporal and Spatial Error Correlation Now we consider BLBP over a GE channel odel with both the teporal error correlation and the spatial error correlation. The error events at different receivers are in a certain correlation. We assue two kinds of error events, the error event at the sender which leads to the n correlated packet loss aong all receivers and the error events at different receivers which lead to independent packet losses. Soe new sybols are shown as follows. P : The error rate caused at receivers (called input error, independent of each other; P out : The error rate caused at the sender (called output error, which leads to correlated loss covering all receivers; p = pout + ( pout p ; in βin βin Pin = αin α : The transition probability atrix in of the input errors at all receivers; βout βout Pout = αout α : The transition probability out atrix of the output error at the sender; λ : The spatial error correlation aong different receivers, λ = p p. out Cobining the output error and the input error, we can copute the total error odel, which is a 4-states Markov chain, shown as follows. βout βin βout ( βin ( βout βin ( βout ( βin βout ( αin βoutαin ( βout ( αin ( βout α in P = ( αout βin ( αout ( βin αoutβin αout ( βin ( αout ( αin ( αout αin αout ( αin αoutαin (27 For the convenience of analyzing, we use a GE channel odel to approxiate the total error odel. This is also confired by the siulation experients. The total GE channel odel is β β P = α α (28 where α = ( β ( p/ p and β = βoutβ. in So we get the final PL for all receivers, shown in forula (29. PL( r = P[ +, + ] = pα (29 Now we consider the expected nuber of transissions for one ulticast data packet for all receivers. The probability that N > n is shown as follows. PN> [ 0] = (30 PN [ > ] = p + p p (3 ( ( ( out out in > out + ( out ( ( in ( Pout[ n i, n i] ( αout ( ( ( pin / p P[ i, i] PN [ n] P [ nn, ] p ( p / ppnn [, ] n + i= n n ( α ( ( ( α n i i ( pout ( αout ( αout ( ( pinα = p + p p n i= out out out in + (32 Copyright 2008 Scies. I. J. Counications, Network and Syste Sciences, 2008, 2,

7 50 Z. LI ET AL. Finally the expected nuber of transissions for one ulticast data packet for all receivers can be calculated as in forula (33. EN [ ] = PN [ > n] 0 ( ( ( n n pout ( αout + ( pout ( pinα (33 n n i i + pout ( αout ( αout ( pinα + i= Note that it is direct and siple to calculate the average channel holding tie of BLBP fro the expected nuber of transissions in practice. We will evaluate the perforance of BLBP and copare BLBP with LBP and different application layer ulticast error control schees by siulation experients in the following section. 6. Perforance Evaluation In this section, we first evaluate the perforance of BLBP and confir the theoretical results by siulation experients. Then we copare BLBP with the application layer ulticast error control schees HEC- P and HAQ Type I. The etrics used for evaluation include the average nuber of transissions, the axiu ulticast delay and the total I. Consistent with any references [8,0,4], we also consider the redundant transission only as I. Table. Application targets and paraeters PL equireent e-6 Delay Constraint 20-00s TP Payload Length 36Bytes Multicast load interval 2.5s TT 3.5s Original Error ate 0% Packet sent 40-00e6 Table 2. The retransission liit and the teporal error correlation (PL constraint e-6 Error The teporal error correlations ates We conduct our siulation study using ns-2 and ipleent BLBP based on the 802.e siulation odel fro [7]. All client nodes are one hop to the AP and at ost two hops to each other. We use IEEE 802.a paraeters to odel the physical layer. The data rate we choose is 24Mbps. The first receiver that joins the ulticast group acts as the leader. The application targets and paraeters are presented in Table. Note that the PL target (e-6 is very strict. We use this final PL target as the MAC layer PL target. The total payload length in the MAC layer is 356 bytes, and there is no fragentation in the MAC layer or the network layer. The retransission liit of BLBP is deterined following only the PL constraint shown in forula (5. Soe exaples are shown in Table 2. We guarantee the PL target first and explore the total ulticast delays of BLBP in different network scenarios and channel conditions. The application layer ulticast error control schees HEC-P and HAQ Type I are ipleented based on the real-tie transport protocol (TP [8]. We use unicast for the feedback in HEC-P and HAQ Type I instead of broadcast because unicast is ore reliable. The siplified GE channel odel is ipleented in the physical layer, but it is used only for data fraes. The MAC control fraes (TS/CTS/Beacon/ACK are error free fro the error odel. (The control fraes also ay be lost because they ight collide with the background traffic. This also akes sense in practice because the control fraes are very sall and are sent in the lowest data rate, and hence they are ore reliable than data fraes. First we copare the average nubers of transissions for BLBP and LBP in different channel conditions. The results are shown in Figure 4. The siulation result and the theoretical result of BLBP atch very well. As expected, BLBP is ore efficient than LBP particularly when the error rates are high and the nubers of receivers are large. This is because BLBP allows all receivers to send feedback based on sequence check thanks to the beacon frae and LBP can not. Then we explore the effect of the teporal error correlation. The average nubers of transissions and the axiu ulticast delays of BLBP with different teporal error correlations are shown in Figure 5 and Figure 6 respectively. About the expected nuber of transissions, we can see that the theoretical analysis and the siulation results atch very well. The teporal error correlation affects the average nuber of transissions very uch. However, the ulticast delays are still very low even when the teporal error correlations are high, thanks to the fast ACK and retransission in the MAC layer. So BLBP is very good for real-tie ulticast applications with strict delay constraints. Figure 7 shows the average nubers of transissions for BLBP with different spatial error correlations. The theoretical analysis and the siulation result atch very well. BLBP can take full advantage of the spatial error correlation because of the coplete feedback suppression thanks to the JACK schee. Then we copare BLBP with the application layer ulticast error control schees. The teporal error correlation and the spatial error correlation are set to 0.0 and 0.20 respectively to siulate a near realistic channel condition. First we copare BLBP with HEC- P in different channel conditions. The total Is and axiu ulticast delays are shown in Figure 8 and Copyright 2008 Scies. I. J. Counications, Network and Syste Sciences, 2008, 2,

8 BEACON-DIVEN LEADE BASED POTOCOL OVE A GE CHANNEL 5 FO MAC LAYE MULTICAST EO CONTOL Figure 9 respectively. When the error rates are high, we can see BLBP is uch ore efficient than HEC-P which is a pure AQ based schee (without FEC coding. This is because BLBP suppresses the ultiple feedbacks into just a JACK and it is ore efficient than the application layer feedback and retransission suppression in HEC-P. The results also show that the ulticast delays of BLBP are uch shorter than the delays in HEC-P. This is due to the fast MAC layer ACK/retransission in the MAC layer. Moreover, because of the uch longer delays aong each transission or retransissions, HEC-P isn t uch affected by the teporal error correlation. Figure 0 shows the total Is of BLBP, HEC-P and HAQ Type I under different delay constraints. As shown in Figure 9, HEC-P takes a long tie and hence only satisfies long delay constraints. Both BLBP and HAQ Type I can satisfy all the delay constraints fro 20s to 00s. We can see that BLBP is ore efficient than HAQ Type I when the delay constraints are short and the nubers of receivers are sall. This is because that HAQ Type I has to switch to pure FEC schee (no AQ when the delay constraint are very short, hence it is not efficient. Finally we copare the total Is of BLBP, HEC-P and HAQ Type I with the heterogeneity of receivers. The error rate for receiver is variable and all the other receivers have a fixed error rate 0.0. Figure shows the result. We can see that both BLBP and HEC-P are uch ore efficient than HAQ Type I because HAQ Type I has to set FEC code and other paraeters according to the receiver with the worst channel condition but BLBP and HEC-P are ore adaptive. Moreover, due to the effect of the teporal error correlation (shown in Figure 5, here BLBP is a little less efficient than HEC-P. 7. Conclusions In this work, we enhance LBP and propose BLBP for the MAC layer ulticast error control in wireless networks. The use of the beacon frae is to lead the non-leader receivers to set tiers and to announce the data frae sequence. On erroneous reception of a data frae that has not been correctly received before, the leader does not send an ACK, propting a retransission. On erroneous reception (The tier ties out of the data frae that has not been correctly received before, the non-leader receivers send NACKs to collide with the potential ACK fro the leader, thus propting the AP to retransit the packet. To guarantee a very low PL under strict delay constraints for video ulticast over a GE channel with both the teporal error correlation and the spatial error correlation, we analyze BLBP and evaluate its perforance via siulation experients. Both the theoretical analysis and siulation results show that BLBP can correct nearly all the errors for all receivers in the MAC layer. BLBP needs the iniu nuber of redundancy transissions aong all pure AQ based schees. BLBP is ore efficient than the application layer AQ schees and the total ulticast delay is uch shorter. BLBP is even ore efficient than the best application layer ulticast error control schee when the delay constraints are short or with the heterogeneity of receivers. BLBP is very good for the real-tie ulticast applications with strict delay constraints, especially for sall groups. In the future, we plan to extend BLBP with FEC coding and further iprove the QoS of ulticast cross the application layer and MAC layer in wireless networks. Figure 4. The expected nuber of transissions with different error rates ( λ =0 τ =0 for all receivers Figure 5. The expected nuber of transissions with different teporal error correlations (error rate 0.0 λ =0 for all receivers Copyright 2008 Scies. I. J. Counications, Network and Syste Sciences, 2008, 2,

9 52 Z. LI ET AL. Figure 6. The axiu ulticast delay with different teporal error correlations (error rate 0.0 λ =0 for all receivers Figure 9. The axiu ulticast delay with different error rates ( τ =0.0 λ =0.20 for all receivers Figure 7. The expected nuber of transissions with different spatial error correlations (error rate 0.0 τ =0.0 for all receivers Figure 0. The total I with different delay constraints (error rate 0.0 τ =0.0 λ =0.20 for all receivers Figure 8. The total I with different error rates ( τ =0.0 λ =0.20 for all receivers Figure. The total I with a bad receiver (error rate 0.0 τ =0.0 λ =0.20 for other receivers, delay constraint 00s Copyright 2008 Scies. I. J. Counications, Network and Syste Sciences, 2008, 2,

10 BEACON-DIVEN LEADE BASED POTOCOL OVE A GE CHANNEL 53 FO MAC LAYE MULTICAST EO CONTOL 8. eferences [] U.H. eiers, DVB-The Faily of International standards for Digital Video Broadcasting, in Proceedings of IEEE, vol.94, no., pp , January [2] E.N. Gilbert, Capacity of a burst-noise channel, Bell Syst. Tech. J., vol.39, pp , Septeber 960. [3] E.O. Elliott, Estiates of error rate for codes on burstnoise channels, Bell Syst. Tech. J., vol.42, pp , Septeber 963. [4] M. Mushkin and I. Bar-David, Capacity and coding for the Gilbert-Elliott channels, IEEE Trans. on Infor. Theory, vol. 35, pp , Noveber 989. [5] J. Nonnenacher and E.W. Biersack, Optial ulticast feedback, in IEEE Infoco, (San Francisco, California, pp. 964, March/April 998. [6] L. izzo, Effective erasure codes for reliable coputer counication protocols, ACM Coputer Counication eview, April 997. [7] D. Qiao and K.G. Shin, A two-step adaptive error recovery schee for video transission over wireless networks, in Proceedings of IEEE Infoco, 2000, March [8] G.P. Tan and T. Herfet, The Optiization of an TP Level Hybrid Error Correction Schee for DVB Systes in Wireless Hoe Networks under Strict Delay Constraint, IEEE Trans. on Broadcasting, Noveber [9] G. Carle and E.W. Biersack, Survey of error recovery techniques for IP-based audio-visual ulticast applications, IEEE Network, vol., no 6, pp , Noveber Deceber 997. [0] G. Tan and Th. Herfet, Application Layer Hybrid Error Correction with eed-soloon Code for DVB Services over Wireless LANs, the 3rd IEEE International Conference on Wireless Counications, Networking and Mobile Coputing (WiCOM, Shanghai, China, Septeber [] J. Kuri, S.K. Kasera. eliable ulticast in ulti-access wireless LANs, in Proceedings of Infoco 99. [2] J.Tourrilhes, obust broadcast: iproving the reliability of broadcast transission on CSMA/CA, in Proceedings of the Ninth IEEE International Syposiu on. Personal, Indoor and Mobile adio Counications, 998. [3] S.K.S. Gupta, V.Shankar, and S. Lalwani, eliable Multicast MAC Protocol for Wireless LANs, IEEE International Conference on Counications, vol., pp , [4] J.. Yee and E.J. Weldon, Jr., Evaluation of the perforance of error-correcting codes on a Gilbert channel, IEEE Tran. on Counications, vol. 43, no. 8, August 995. [5] G. Bianchi, Perforance analysis of the IEEE 802. distributed coordination function, IEEE Journal on Selected Areas in Counications, vol. 8, no. 3, March [6] D.F. Towsley, J. Kurose, and S. Pingali, A Coparison of Sender-Initiated and eceiver-initiated eliable Multicast Protocols, IEEE Journal on Selected Areas in Counications, vol. 5, no.3, April 997. [7] S. Wiethölter and C. Hoene, An IEEE 802.e EDCA and CFB Siulation Model for ns-2, [8] J. Ott, S. Wenger, N. Sato, C. Bureister, and J. ey, Extended TP profile for TCP-based feedback, draftietf-avt-rtcp-feedback-.txt, August Copyright 2008 Scies. I. J. Counications, Network and Syste Sciences, 2008, 2,