A Wireless MAC Protocol with Efficient Packet Recovery

Transcription

1 A Wreless MAC Protocol wth Effcent Packet Recovery Muhammad Naveed Aman and Bplab Skdar Department of Electcal, Computer and Systems Engneeng, Rensselaer Polytechnc Insttute, Troy NY 8 USA Abstract Exstng wreless medum access control (MAC protocols provde relablty aganst corrupted packets by provdng mechansms for packet error detecton and retransmsson. The effcency of exstng mechansms for provdng relablty s usually low snce they requre the entre packet to be retransmtted even though only parts of t may have been corrupted. To address ths ssue, ths paper presents a MAC protocol wth an effcent packet recovery mechansm for packets corrupted due to both channel errors and collsons. The proposed MAC protocol frst determnes the cause of the errors n a packet and then uses the acknowledgment (ACK packets to provde feedback on the sectons of the packets that have errors. To mnmze the packet recovery tme, the proposed MAC protocol allows the sender to retransmt the corrupted sectons of the packet mmedately, wthout requng a new channel access. Usng smulatons, t s shown that the proposed MAC protocol has hgher effcency and ncreases the acheved throughput. I. INTRODUCTION The causes of errors n wreless networks may be broadly classfed as ether ( caused by poor channel condtons resultng from factors such as fadng, path loss and nose and ( those caused by collsons resultng from smultaneous transmssons from more than one node. The MAC layer of most wreless networks provde relablty aganst these errors by ncludng mechansms to detect the presence of errors and retransmttng the corrupted packets. A fundamental lmtaton wth exstng protocols that lmts the acheved throughput s that the retransmsson mechansms usually resend the entre packet, even though only parts of t may have been corrupted. Snce most errors are confned to parts of a packet, retransmttng the entre packet leads to waste of channel bandwdth and s an overhead that should be avoded. To address ths ssue, ths paper presents a wreless MAC protocol capable of effcent packet recover n the presence of errors that may be caused by the channel or collsons. The presence of errors n a receved packet s usually detected through the use of error detecton codes, such as the cyclc redundancy check (CRC used n IEEE 8. [. Mechansms such as automatc repeat request (ARQ are then used to retransmt the corrupted packet [. Snce CRCs cannot solate the bts wth errors, most ARQ schemes retransmt the entre packet. A methodology for markng the bts that are lkely n error s presented n [3 and t s shown that the throughput may be ncreased by retransmttng only these bts. However, ths approach requres specalzed /4/$3. c 4 IEEE. hardware, and has a hgh overhead feedback strategy. A packet recovery mechansm that works by combnng multple copes of a packet from dfferent access ponts s proposed n [4. The overhead and delay assocated wth combnng agents lmts the applcaton of ths scheme. Moreover, only AWGN channels are consdered n [4 and ts effectveness n more practcal channels has not been establshed. In [5, a softwareonly soluton to packet recovery based on harnessng partal packets s presented. Ths scheme reles on ncremental redundancy and codng schemes, whch can become a bottleneck at hgh data rates. In contrast, the proposed packet recovery mechansm does not requre any customzed hardware, has lower overhead, and provdes real tme operaton. The man contbuton of ths paper s a MAC protocol that effcently recovers packets that are corrupted dung transmsson. The proposed protocol s based on the use of our mechansm to solate the cause and locaton of errors n transmtted packet. The proposed methodology to localze the errors s based on calculatng the Error Vector Magntude (EVM of the receved symbol. The recever n the proposed protocol frst detects the symbols that were corrupted dung transmsson. It then conveys the locatons of the corrupted portons of the packet to the transmtter usng the ACK packet and the transmtter can then uses precse retransmssons of only the corrupted symbols. The performance of the proposed MAC protocol has been evaluated usng both analyss and smulatons. The rest of the paper s organzed as follows. Secton II presents the proposed MAC protocol. Sectons III and IV present mechansms for detectng the cause of packet errors and localzng the errors, respectvely. Secton V presents smulaton results, and Secton VI concludes the paper. II. A MAC PROTOCOL WITH EFFICIENT PACKET RECOVERY The proposed MAC protocols uses carer sense multple access wth collson avodance (CSMA/CA for channel arbtraton and ts basc operaton s based on IEEE 8. [. The proposed protocol may be consdered to an extenson for IEEE 8. for effcent packet recovery. The operaton of the protocol conssts of two steps: ( when a packet s deemed to be corrupt (as determned by a CRC check, the frst step s to detect the sectons of the packet that are corrupted, and ( facltate the effcent and targeted retransmsson of only the

2 DIFS data BO ACK SIFS SIFS SIFS SIFS one packet transmsson Backoff slot SIFS DIFS Fg.. Proposed MAC Protocol for Effcent Packet Recovery corrupted sectons of the packet. A. Error Localzaton tme A key part of the packet recovery mechansm s the mechansm that localzes the sectons of the packet that are corrupted. For the purposes of error localzaton, we consder each packet to be dvded nto a number of equal szed blocks. For each corrupted packet, we frst establsh the cause (collson or channel error usng the mechansm descbed n Secton III. If the packet s corrupted by a collson we use a thresholdng mechansm to dentfy the blocks n error, whle K-means clusteng s used f the packet s corrupted by channel errors, as descbed n Secton IV. B. MAC Protocol The operaton of the proposed MAC protocol s shown n Fgure. Under CSMA/CA based channel access, each node wth a packet to transmt frst senses the state of the channel. If the channel s sensed to be dle for an nterval greater than the Dstbuted Inter-Frame Space (DIFS, the node proceeds wth ts transmsson. If the channel s busy, the node defers ts transmsson tll the channel becomes dle. The node then ntalzes ts backoff tmer wth a randomly selected backoff nterval and decrements ths tmer every tme t senses the channel to be dle. The node transmts ts data when the backoff tmer reaches zero. An exponental backoff mechansm such as that n IEEE 8. s used n the presence of collsons. After recevng a packet, the recever wats for a Short Inter-Frame Space (SIFS, before t transmts an ACK. However, unlke IEEE 8. that only uses postve ACKs, the proposed MAC protocol uses the ACK to convey nformaton about the sectons of the packet that are corrupted. Once the blocks n error are dentfed, a btmap ndcatng these blocks s sent to the transmtter usng the ACK packet. On recevng the block-level error btmap, the sender proceeds to retransmt the packet but only ncludes the blocks wth errors. A key aspect of the proposed protocol s that the retransmssons do not requre a new channel access. Once the ACK s receved, the sender retransmts the corrupted blocks mmedately after a SIFS peod. Snce all other nodes n the vcnty of the sender have to wat for a DIFS peod before they can begn ther transmsson or backoff, ths prevents any collsons. The recever reples to the retransmtted blocks wth an ACK that agan uses a btmap to ndcate the blocks that may stll be n error. The sender agan retransmts the corrupted blocks after a SIFS peod and ths contnues tll all the blocks of the packet are receved correctly. The MAC layer at the recever assembles all the blocks of the packet before sendng t to the upper layer. An upper lmt may be placed on the number of retransmsson attempts that are allowed for a block before the current transmsson s aborted. Such a lmt may be requred for farness and to prevent a sngle node from monopolzng the channel for too long. The entre packet needs to be retransmtted f the current transmsson attempt exceeds the number of retransmssons attempts for ts blocks n a sngle channel access. The advantage of the proposed protocol stems from three factors. Frst, only the sectons of the packet that are corrupted are retransmtted, thereby elmnatng the waste of channel resources. Secondly and more sgnfcantly, the retransmsson does not requre a separate channel accesses. The channel access tme s a major contbutor to the delays expeenced by the packets [6. By allowng retransmssons after only a SIFS peod as opposed to a full-fledged channel access nvolvng backoffs and deferments to transmssons from other nodes. Fnally, the thrd advantage of the proposed MAC protocols s that snce t can dentfy the cause of a corrupted packet, a node does not have to do an exponental backoff n the cases where the errors are caused by channel errors. Extenson for RTS/CTS: the descpton of the protocol above llustrated the operaton of the protocol n the absence of request-to-send (RTS and clear-to-send (CTS packets. RTS- CTS packet exchange before the data transmsson are usually recommended to address the problem of hdden termnals. Under the operaton wth RTS-CTS based reservaton, a node wth packet to transmt sends a RTS packet to the recever (after successful channel access usng CSMA/CA and the recever responds wth a CTS packet f t s wllng to accept the packet and s currently not busy. The RTS/CTS packets contan tmng nformaton about the length of the ensung transacton, and all nodes that overhear ths exchange defer ther transmssons tll the current transmsson s complete. Wth the proposed retransmsson mechansm, the tmng nformaton contaned n the RTS/CTS packets may not be vald f retransmssons are requred for any of the blocks of the packet. Consequently t s possble that hdden nodes may nterfere wth subsequent transmssons. To address ths stuaton, the proposed MAC protocol adds tmng nformaton to the ACK packets. When an ACK packet sends a btmap wth the lst of blocks wth errors, t also needs to nclude the tmng nformaton that the retransmsson wll take. Nodes overheang the ACK packet wll update ther network allocaton vector (NAV and defer ther transmsson. At the sender s sde, no modfcaton s requred snce the retransmsson begns after a SIFS nterval whle all other nodes have a wat of at least a DIFS nterval. The MAC header of the retransmtted packet can also convey the tmng nformaton requred for the nodes n the vcnty of the sender. III. DETECTING THE CAUSE OF PACKET CORRUPTION Ths secton presents a methodology for detectng the cause of packet errors. Ths nformaton s necessary for accurately solatng the locatons of the symbols wth errors. If X k

3 3 denotes the reference or transmtted sgnal and Y k denotes the receved (dstorted sgnal, then the error vector s E k = Y k X k. Let the transmtter and recever be denoted by T x and R x, respectvely. Consder an OFDM system wth T subcarers and a frequency flat multpath Raylegh fadng channel. The receved tme doman OFDM sgnal y n s y n = H n x n +η n +ζ n ( where H n s the Raylegh channel coeffcent, η n s addtve whte Gaussan nose wth zero-mean and vaance ση, and ζ n s the nterference due to collson. The reference sgnal n our model s obtaned from the receved sgnal by demodulatng the receved symbols and then modulatng them once agan. Ths re-modulated sgnal works as an approxmaton for the transmtted sgnal. x n s the n th tme doman OFDM sgnal and can be obtaned from X k, the M-QAM modulated symbol at the kth subcarer as [7, T x n = IDFT{X k } = X k e jπkn/n ( k= Let us ntroduce a random vaable Z defned as Z = E k. (3 N P k= where P s the average power of all the symbols for a gven modulaton, and N s the number of receved symbols. Let e n be the error vector n the tme doman.e., e n = y n x n. We also know that e n = IDFT{E k } and y n = IDFT{Y k }. Thus, applyng Parseval s theorem, we can rewte (3 as Z = e n. (4 N P Note that Z s the sum of N..d. random vaables. Usng the central lmt theorem (for large N, we approxmate Z as a Gaussan wth probablty densty functon (pdf f Z (z = e (zµz σ Z. (5 πσ Z To obtan the pdf f Z (z of Z n the presence and absence of collsons, we need to fnd ts mean (µ Z and vaance (σz for both cases. We frst evaluate µ Z and σz for the case when there s no collson. In ths case the error vector s gven by e n = H n x n +η n x n = x n (H n +η n. (6 We assume the path loss law l(r = α mean power of the Raylegh channel (H n. Then, µ Z = ε{z} = ( π [σx + P, and take r α +σ η as the. (7 To fnd σz, we need to evaluate ε{z } gven as follows: ( Z = e n = NP N P e n e n. (8 n =n = We now assume block fadng wth a block length of m symbols. Then, ε{z } s gven by ε{z } = [ NP mε{e 4 n}+ N m ( ε{e m n } ( where ε{e n} s gven by { ε{e n} = σx } π + α +ση ( and ε{e 4 n} s gven by (9. Combnng (7 and (, we can obtan the vaance of Z as σ Z = ε{z }(µ Z. Now consder the case when a packet s corrupted by collsons. We assume a network wth J nterferers at dstance r j from R x, that transmt wth probablty p ndependent of each other. The startng locaton of a collson wthn a packet s assumed to be unformly dstbuted between and N. If a collson starts at symbol n, the error vector s gven by { e en n < n n = (3 e n +ζ J n n where ζ J s gven as follows ζ J = J B H W. (4 = Here B s are..d. Bernoull random vaables wth parameter p, H s Raylegh dstbuted wth mean power / α, and W s the OFDM symbol transmtted by nterferer, whch s approxmately..d. Gaussan dstbuted wth zero-mean and vaance σx. We can then rewte (4 as follows: Z = NP ( n e n + Takng the expectaton of (5, we get ε{z } = σx π + P n=n ( e n +ζ J + p(n+ N J r α j= j. (5 +σ η and smlarly, we have ε{z } = [( ( N N P mε{e 4 (N 3 n}+ 4 m 3 (N + (ε{e (N m n} N+ +( 3 ε{e n}ε{e n }+ (N+ [ mε{e n4 }+ 4m N( N +5N3 ( 5N N+ + 3 m(n++ ( ε{e n } (6 4m whereε{e n} andε{e 4 n} are gven by ( and (9. respectvely. ε{e n } s gven by { } ε{e n π J } = σx α + α +ση+pσ x rj α (7 j= and ε{e n 4 } s gven by (. Thus, we can now obtan the

4 4 ε{e 4 n} = (3σ 4 x ε{e n4 } = [ ( (3σ 4 x 8 α ( [ { 6 σx α 8 r α 4 + π r 3 α α α 4 + r 3 α r α π + ( + +3ση +6σxσ η ( π + +3ση +6σxσ η } +ση pσx J r α j= j +6pσx 4 J j= r α j π + π α + + (9 ( vaance of Z as σ Z = ε{z }(µ Z. A. Classfyng the Cause of Packet Errors To classfy the cause of each packet loss, we frst calculate the EVM of each receved packet and then compare t aganst a threshold value. If the calculated EVM s greater than the threshold, the packet s classfed as a collson, and vce versa. The optmum threshold value s determned by choosng the threshold that leads to equal false postve and false negatve rates (.e., the threshold that leads to the crossover error rate. A false postve s defned as an event where the cause of a packet loss s attbuted to a collson, when the actual cause was a weak sgnal (not a collson. If we denote the threshold by γ, then the probablty of a false postve s gven by P Z [Z > γ = P e [ f Z (zdz (8 where P e s the symbol error rate. Smlarly a false negatve s defned as an event where the cause of a packet loss s attbuted to a weak sgnal, when the actual cause was a collson. The probablty of a false negatve s gven by [ P Z [Z γ = P e f Z (z dz. (9 To obtan the threshold that leads to the crossover error rate, we can fnd γ by equatng (8 and (9. Thus, to get the threshold we need to solve the followng equaton numecally: f Z (zdz + f Z (z dz =. ( IV. IDENTIFYING THE LOCATION OF ERRONEOUS BLOCKS The error dentfcaton mechansm detects blocks of symbols wth errors nstead of ndvdual symbols wth errors. To dentfy the symbols wth errors, we calculate the EVM over blocks of m symbols (m. Ths block based approach serves two purposes: ( justfyng the use of (5 and ( loweng the overhead of the feedback sent to the transmtter. A. Error Locatons n Packets Corrupted by Collsons Once a packet s classfed as a collson, the recever compares the EVM of each block wth a threshold. The threshold s determned usng the mechansm descbed n Secton III. Note that the threshold for dentfyng blocks wth errors s dfferent from the threshold for dentfyng the cause of a packet loss. The dfference ases because the cause of a packet loss s classfed usng all the symbols n the packet whle blocks wth errors are detected based only on the symbols wthn a block. The blocks wth EVM greater than the threshold are tagged as erroneous blocks. B. Error Locatons n Packets Corrupted by Channel Nose To detect the error locatons n case of corrupton due to channel nose, we perform K-means clusteng [8 on the EVM values of the blocks n the packet. Clusteng s used nstead of thresholdng because the dfference n the vaances of the pdfs of the EVM of blocks affected by channel nose, and the blocks that are unaffected, s small. Ideally, blocks wth errors should form one cluster whle blocks wthout error form another cluster. The process starts wth two ponts randomly chosen as cluster centers. The EVM of the blocks n the corrupted packet are then assgned to ther closest cluster center based on the Eucldean dstance. The centrod of the two resultng clusters s then calculated and used as the new cluster centers. The EVM value assgnment process s then repeated, and ths process contnues untl the cluster centers stablze. V. RESULTS In ths secton we present the smulaton results to evaluate the performance of our proposed methodology. The smulatons were done usng a mx of MATLAB/Smulnk and NS. The transmtter and recever models were desgned accordng to the IEEE 8.a specfcatons [9. The network smulaton was done usng NS, whch n turn used the output of the channel model created usng MATLAB/Smulnk. The wreless nodes n the smulaton are connected through a frequency flat multpath Raylegh fadng channel. The channel s realzed through the Jake s model [. Results were generated for BPSK, QPSK, 6QAM and 64QAM. We only show the results for 64QAM, for whch two data rates were consdered: 48Mbps and 54Mbps. For both cases, there were 6 coded bts per subcarer and 88 coded bts per OFDM symbol. For the 48Mbps case the codng rate was /3 and data bts per OFDM symbol was 9. The correspondng numbers for the 54Mbps case was 3/4 and 6. Fgures and 3 show the throughput acheved by the proposed MAC protocol and IEEE 8. for channel data rates of 48 and 54 Mbps respectvely, for a channel sgnal to nose rato (SNR of 4 db. For these smulatons, saturated traffc condtons were assumed at each node. All nodes exchanged data wth a sngle base staton and the nodes were located on a crcle of radus 5 meters around the base staton, at regular angular ntervals. We observe that the proposed mechansm

5 5 Throughput (bts/sec 3e+6.5e+6 IEEE 8. Proposed e+6.5e+6 e Number of nodes [4 G. Woo, P. Kheradpour, D. Shen and D. Katab, Beyond the bts: cooperatve packet recovery usng physcal layer nformaton, Proceedngs of ACM MOBICOM, pp , Montreal, Canada, September 7. [5 K. Ln, N. Kushman and D. Katab, ZpTx: harnessng partal packets n 8. networks, Proceedngs of ACM MOBICOM, pp , New York, NY, September 8. [6 O. Tckoo abd B. Skdar, A Queueng Model for Fnte Load IEEE 8. Random Access MAC, Proceedngs of IEEE ICC, pp , Pas, France, June 4. [7 Y. Choo, J. Km, W. Yang and C. Kang, MIMO-OFDM Wreless Communcatons wth MATLAB, Wley-IEEE press, Sngapore,. [8 I. Wtten and E. Frank, Data Mnng: Practcal Machne Learnng Tools and Technques, Morgan Kaufman, San Francsco, CA, 5. [9 Hgh-Speed Physcal Layer n the 5 GHz Band, IEEE Standard 8.a- 999, 999. [ W. Jakes, Mcrowave Moble Communcatons, New York: Wley, 974. Fg.. Compason of throughput of the proposed packet recovery mechansm and IEEE 8. at 48 Mbps and SNR of 4dB. 3.5e+6 3e+6 IEEE 8. Proposed Throughput (bts/sec.5e+6 e+6.5e+6 e Number of nodes Fg. 3. Compason of throughput of the proposed packet recovery mechansm and IEEE 8. at 54 Mbps and SNR of 4dB. provdes more than % mprovements n the throughput at all loads. Ths mprovement ncreases as the dstance between the source and the destnaton nodes ncreases or the channel becomes poorer. VI. CONCLUSIONS Ths paper presented a MAC protocol for the fast and effcent recovery of corrupted packets. The proposed MAC protocol s based on frst determnng the cause of the errors n a packet and then selectvely retransmttng the sectons of the packet wth losses. Addtonal effcency s acheved by the proposed MAC protocol by allowng the sender to retransmt the corrupted sectons of the packet mmedately, wthout requng a new channel access. Our smulaton results show that the proposed MAC protocol performs sgnfcantly better than the conventonal IEEE 8. protocol. REFERENCES [ Wreless LAN Medum Access Control (MAC and Physcal Layer (PHY Specfcatons, IEEE Standard 8., 997. [ D. Bertsekas and R. Gallager, Data Networks, Englewood Clffs, NJ: Prentce-Hall, 987. [3 K. Jameson and H. Balakshnan, PPR: partal packet recovery for wreless networks, Proceedngs of ACM SIGCOMM, pp. 49-4, Kyoto, Japan, August 7.