scieee Open visual document viewer

Performance analysis of a hybrid ARQ system in half duplex transmission at 2400 BPS

Dalmau Royo, Jordi,Serrat Fernández, Juan

Abstract

Hybrid ARQ/FEC protocols have been proposed to provide high data link integrities whilst keeping at the same time a high mean throughput rate. Nevertheless, hybrid ARQ strategies offer a lot of choices and none of them can be considered the optimum in any case. Three alternative protocol strategies using BCH codes are evaluated and the HF channel models used for the tests are discussed.

Full text

248 PERFORMANCE ANALYSIS OF A HYBRID ARQ SYSTEM IN HALF DUPLEX TRANSMISSION AT 2400 BPS J Dalxau-Roy0 and J Se a -Fe ndndez Uni e si a Poli ecnica de Ca alunya, Spain INTRODUCTION Today HF communica ion sys ems ely on he concep o adap a ion. In ac , he bene i s o exploi ing he ime/ equency a ian capaci y o he HF channel ha e been epo - ed elsewhe e. Adap i e sys ems can play an impo an ole in any one o he i s h ee laye s o a communica ion sys em. This pape is ocused on he desc ip ion o he bases o a le el-2 p o ocol aimed o inco po a e such adap i e capaci y. Hyb id MQ/FEC p o ocols ha e been p oposed o p o ide high da a link in eg i ies whils keeping a he same ime a high mean h oughpu a e. Ne e heless, hyb id ARQ s a egies O e a lo o choices and none o hem can be conside ed he op imum in any case. Among he di e en aspec s o consi- de in o de o choose o o design one o such p o ocols we mus ake in o accoun he sys em applica ion cons ain s. In his iew we speci y he ollowing: The p o ocol is in ended o p o ide le el-2 se ices in a hal duplex sys em exploi ed wi h wo di e en wo king equencies. Long o wa d messages and sho eplies mus be allowed in o de o minimize he e ec s o he le el-1 o e heads. A esidual bi e o a e o abou can be ole a ed a ough bi a es o 2400 bps. Al hough we will go insigh h oughou he pape i is enough o jus i y some o he app oaches ha we ha e conside ed. In he ollowing sec ion we p esen he al e na i es which we e conside ed. Then we p oceed desc ibing he h ee which appa- en ly we e he mos p omising candida es. Finally we discuss he labo a o y es s esul s wi h emphasis on he channel models used o he es s. PROTOCOL AND CODING STRATEGIES The HF channel in oduces a mix u e o andom and bu s e o s, he so called di use e o s as desc ibed by Da nell and Tech [l]. As a consequence he coding s a- egies should be sui able o comba his ype o e o s. The basic di use e o co ec ing echni- ques a e he ollowing ones: -- - E o de ec ion and e ansmission - Simple in e lea ing - Fi e codes - Reed Solomon codes - Th eshold decoding echniques - Code combining E o de ec ion and e ansmission is gene- ally easy o implemen . Mo eo e , i equi- es less edundancy han o he s o co ec he same numbe o e o s and i is indepen- den o he bu s leng h. Howe e , his echnique equi es a eliable eedback link and enough bu e s o age and logic a bo h he ansmi e and he ecei e . In addi- ion i p esen s low h oughpu a es in noisy o wa d channels. Simple in e lea ing comba s he e ec s o e o bu s s by dis ibu ing hem o e sepa a e coded da a s eams. I is also easy o implemen and many good andom e o co ec ing codes a e known. I s main d aw- back is ha he e o s co ela ion is no ully exploi ed in decoding. Mo eo e , in e lea ing may s ill be inadequa ed wi h e y long bu s s. Fi e codes p esen a e y high e iciency and a e o simple implemen a ion. Ye , long Fi e codes a e equi ed o a mode a ed bu s co ec ion capabili y. Al hough Reed Solomon codes show a code e iciency ypically g ea e han 90%, one o i s main d awbacks is conce ned wi h hei igid co ec ion capabili y, as i has been s a ed by Reed and Hopkinson 121. Th eshold decoding echniques a e able o handle la ge bu s s and hey a e also com- me cially a ailable. Howe e , he code design is s ill an a and de ices exis o only a limi ed ange o code a es. Code combining p esen s an a bi a ily low ou pu e o a e e en unde he mos se e e channel condi ions. The s a egy is he code a e adap a ion o he channel condi ions. Howe e , i p esen s a e y low h oughpu unde se e e condi ions. Once he basic di use e o co ec ing echniques ha e been examined, i can be concluded ha a sui able blend o hem should be he bes solu ion o cope wi h such e o pa e ns. In his way, a s a egy based on he ollowing p emises was pu sued: - E o de ec ion and co ec ion by e ansmission - Use o simple in e lea ing - A ce ain deg ee o code a e adap a- ion o he p e ailing channel condi- ions These conside a ions lend o a ype-I1 hyb id ARQ, as in e e ence [l]. The design o such coding s a egy mus conside he ollowing aspec s: 249 - The selec ion o codes o b ing ou an op imal p ocess o e o de ec ion and co ec ion. - The in e lea ing pa ame e s. - The way o adap he code a e. - The e o p o ec ion o he eedback channel. Because o i s hal duplex ope a ion, he selec ion o he e ansmission s a egy (Go Back-N, Selec i e Repea , e c.) is clea ly o e come. In addi ion, he in e lea ing dep h is cons ained and limi ed by he ame leng h. BCH and RS a e known as ones o he mos e icien codes in e ms o ade o be - ween e o p o ec ion capabili y and a e. Because o his, we choosed BCH codes since he begining al hough wi hou disca ing RS codes. In o de o p o ide a easonably high e o p o ec ion whils keeping a he same ime a sho code block leng h, we picked up code leng hs o 15 and 31 bi s. Speci ically we selec ed he BCH(15,10), BCH(15,5) and BCH(31,ll) in o de o be used as explained in he ollowing sec ion. As a as he eedback channel is conce ned we a e limi ed by he sho leng h o he eedback ame. Then, codes wi h di use e o co ec ion capabili y mus be adop ed he e. Fo ins ance, di use con olu ional codes a e a alid al e na i e. This is because we selec ed his amily o h eshold decodable codes o cope wi h he e o pa e ns o his channel and speci ically a con olu ional code o leng h 14. PROPOSED APPROACHES In he ollowing pa ag aphs we speci y h ee p o ocol s a egies which a e om now iden i ied as Sys em 1, Sys em 2 and Sys em 3. The h ee sys ems exchange o wa d a- mes, con eying he in o ma ion om he ansmi e o he ecei e , and backwa d ames only o acknowledgemen pu poses. The in o ma ion bi s o be ansmi ed a e segmen ed in blocks, encoded and placed, o ally o pa ially, in o he ansmission bu e wi h o he blocks in o de o cons i- u e he o wa d ame. Each block is iden- i ied by i s posi ion in he ame. I he ecei e is unable o decode o i i de- ec s e o s in o a gi en block a NACK is sen back in he posi ion associa ed o his block in he backwa d ame. A e he eceip o a NACK he ansmi e may encode again he block wi h a di e en code (Sys ems 1 and 2) o i may send he emai- ning bi s o he i s encoding p ocess (Sys em 3) in o de o allow he ecei e o do a eliable e o co ec ion p ocess. The e o e, he h ee sys ems may use only one o mo e o wa d ames o insu e ha each segmen o in o ma ion bi s is app o- p ia ely deli e ed a he dis an end. Sys em 1 Sys em 1 makes use o segmen s o 10 in o - ma ion bi s. The i s ime ha a segmen is picked up i is encoded wi h he BCH(15,lO). In his case he ecei e ex- ploi s he p ope ies o he BCH(15,lO) o co ec one e o o o de ec wo e o s in he block. The e o e only when wo e o s a e de ec ed, a NACK will be sen back and he ecei e will be wai ing o he e- ansmission o his segmen . On he o he cases, he 10 in o ma ion bi s will be deli e ed o he use a e he app op ia e e o co ec ion p ocess i necessa y. When a NACK is ecei ed, he ansmi e spli s he segmen in wo hal es, encodes independen ly each one wi h he BCH(15,5) and places bo h blocks in o he nex o wa d ame. Now he ecei e will exploi he capaci y o his code o co ec up o h ee e o s o o de ec i he decoded block is eally a codewo d. In he e en ha he decoded block is no a codewo d, a new NACK would be sen back in o de o o ce he ansmi e o o wa d again he same code block. Summa izing, since he i s e ansmission each hal segmen is encoded wi h a BCH(15,5), ansmi ed as many imes as necessa y and p ocessed independen ly o he o he hal . The numbe o e ansmissions is limi ed o a easonable amoun and app o- p ia e con ol p ocedu es a e ca ied ou in o de o always iden i y he posi ion o a gi en hal segmen o in o ma ion bi s inside he o wa d ame. The ACK/NACK bi s o each ecei ed block a e encoded wi h a hal a e con olu ional di use code o cons i u e he backwa d ame. Sys em 2 Sys em 2 exhibi s only a slig h di e ence wi h espec o Sys em 1 conce ning he way in which i exploi s he BCH(15,lO) code. This is, he BCH(15.10) is now used only o de ec bu no o co ec - e o s in ne i s ansmission o a segmen . The e o e a NACK will be sen back o he ansmi e i 1,2,3,5,6, .. e o s appea inside he 15 bi block. Besides his, Sys em 2 p oceeds in he same way as Sys em 1. Sys em 3 Sys em 3 wo ks wi h segmen s o 11 in o ma- ion bi s encoded wi h he BCH(31,ll). Among he 31 bi s a he encode ou pu only 16 a e selec ed and s o ed in he ansmission bu e o be included in he nex o wa d ame. These a e he 11 bi s o he segmen and 5 edundancy bi s specially selec ed in o de o maximize he minimum dis ance be ween he 16 bi codewo ds. By compu e sea ch we ealized ha he minimum ee dis ance is 3. This means ha he ecei e will be able o de ec up o 2 e o s in each 16 bi block. When some e o is de ec ed a NACK will be sen back. Then, a e he eceip o a NACK he ansmi e will load he emaining 15 edundancy bi s plus a s u ing bi in he ansmission bu e in o de o be o wa ded in he nex ame. The e o e he ecei e will be enabled o decode he 31 bi block o he BCH(31.11) ansmi ed in wo conse- cu i e o wa d ames. The amewo k o his p ocedu e is explained by Kallel [31 and i can also be seen as a modi ied memo y ARQ as desc ibed by Se inken [4]. In case ha he i s ansmi ed block we e conside ed e o ee, he ecei e would deli e 11 in o ma ion bi s o he use and an ACK would be issued. Then he ansmi e would dele e he abo e men ioned -- 250 15 edundancy bi s and i would use his posi ion in he bu e o send a new encoded segmen . The backwa d ame is buil up om he ACK/NACK bi s o each ecei ed block as in Sys ems 1 and 2. LABORATORY TEST BED In o de o assess he pe o mance o he h ee sys ems se e al es s we e ca ied ou based on he a angemen shown in Figu e 1. In his igu e PC1 emula es a ansmi ing s a ion and PC2 he ecei ing s a ion. The link be ween bo h s a ions is a h ee wi e line plugged on he asynch onous se ial po o each PC. The p ocessing in ol ed in each o he h ee sys ems is ca ied ou by an ad-hoc so wa e de eloped o each PC. On he o he hand, as no modem o channel simula o a e used, he e ec s o he HF channel need o be simula- ed a one o bo h PCs. In o de o do i a o al o nine channel models comming om he same numbe o eal HF links we e consi- de ed. Because o he in ol emen o he esul s and hei signi icance wi h he used models we de o a pa o he p esen sec- ion o deal wi h his opic. Figu e 2 shows he unc ional block diag am o he so wa e unning on he ansmi e (PC1). On op o i s laye ed s uc u e we ha e he module de o ed o he use in e a- ce which is in ended o allow him he p o- g amming and ini ializa ion be o e i s execu ion. Among he ini ializa ion pa ame- e s he use mus speci y he name o he ile o be ansmi ed, he ype o p o ocol and he HF simula ion model. A he bo om we ha e he communica ion module which o hese es s was designed o ansmi / ecei e cha ac e s h ough he asynch onous po o he PC. The in e ace be ween he communica- ion module and he uppe laye s has been ca e ully speci ied in o de o allow o he communica ion links in he u u e. The o he laye s cons i u e he co e o he ARQ p o ocols desc ibed in he p e ious sec ion in o de o p ocess he backwa d ames and o gene a e he o wa d ames. Conce ning he so wa e o PC2 i looks like Figu e 2 bu now he in e medium le els a e in ended o p ocess he o wa d ame and o gene a e he backwa d ame. An special ea u e o he use in e ace a PC2 is ha i also gene a es a epo ile con aining he mos ele an da a o he ansmission/ ecep ion p ocess like he numbe o esidual e o s, mean h oughpu a e, numbe o blocks ansmi ed one ime wo imes, e c. All he esul s p esen ed in he nex sec ion ha e been ob ained om he epo iles w i en a he end o each un. HF channel models The HF channel is based on a disc e e model de i ed om wo se s o expe imen s: - HF Ionosphe ic communica ions a 2400 bps wi h PSK-2 modems epo ed by Spun icchia and Ca oggio 151. - HF g ound wa e mobile communica ions a 600 bps wi h FSK-2 modems epo ed by Dalmau 161. The second se o expe imen s co esponds o da a ob ained in he me opoli an a ea o Ba celona. Due o he sho leng h o he links (be ween 5 and 30 Km) he main p opa- ga ion mode is g ound wa e. The egis e ed e o pa e ns exhibi di use e o s like hose encoun e ed in sky wa e p opaga ion. The e o e, he esul s go om es s using hese channels may be also ex ensi e o any HF channel. A F i chman's 171 pa i ioned h ee s a e model was selec ed because i gi es he bes ade o be ween accu acy and complexi y. TEST RESULTS In Figu e 3 o 5 we show a summa y o he esul s ob ained a e se e al uns. Each un was in ended o ansmi a ile o 60 Kbi s om PC1 o PC2 in o de o e alua e he use 's bi e o a e and he mean h oughpu a e. The use 's BER is compu ed di iding he e idual e o s in he ecei ed ile by 6 loB. I can be a gued ha his p ocedu e does no gi e a eliable es ima e mainly i he e o coun is low. This is because we w i e in he app op ia e igu es he absolu- e e o coun . As a as he mean h oughpu is conce ned i mus be said ha i is compu ed di iding he 6 lo4 in o ma ion bi s by he o al numbe o bi s ansmi ed h ough he channel in bo h di ec ions bu excluding he s a and s op bi s o each cha ac e in he asynch onous link. Then i gi es an es ima e o he maximum a ainable mean h oughpu a e o a hypo he ical synch onous sys em using he same ough bi a e o bo h modems. Figu e 3 plo s use 's BER e sus channel BER o he nine channels and he h ee sys ems conside ed. The numbe o blocks pe ame is always 160 and he backwa d ame is encoded by he 1/2 a e di use con olu io- nal code. As i can be seen he h ee sys ems show he same beh iou up o chan- nel BERs o abou 4 IO-'. A his poin Sys em 1 begins o lose pe o mance wi h espec o he o he s. On he o he hand, Sys em 3 alls be ween 1 and 2 excep o he wo s channel whe e i deli e s up o 11 e o s. The same beha iou was also obse ed in an ideal backwa d channel. This esul s a e consis en . In ac , Sys em 1 loses e o de ec ion capabili y compa ed wi h Sys em 2 because he BCH(15,lO) is also used o co ec e o s. Also Sys em 3 is less powe ul as a as e o de ec ion is conce ned because he minimum dis ance o i s e o de ec ing code is 3 ins ead o 4. Figu e 4 shows he mean h oughpu a e o he h ee sys ems e sus channel BER. Again he numbe o blocks in he o wa d ame is 160 and he backwa d ame is encoded by he same con olu ional code. Sys em 3 clea ly achie es he highes h oughpu because o i s pa icula phyloso y as a modi ied memo y ARQ sys em, hus ma ching be e he code o he channel. No signi ica i e di e- ences we e obse ed conside ing an ideal backwa d channel. Finally, igu e 5 is a plo o he use 's e o s e sus he ame leng h o Sys em 3 wo king in he wo s channel (BER = The numbe o e o s depends on he ame leng h as i could be expec ed because his 25 1 one is a ec ing he e o pa e n dis ibu- ion. Excluding he sho es ame case, he mean numbe o e o s is 3 o e he 60000 in o ma ion bi s, a good enough igu e in espec o he design objec i es. No di e- ences we e obse ed as a as h oughpu no conside ing an ideal backwa d channel. Sys em 2 clea ly ou pe o ms Sys em 1 consi- de ing use 's e o s and i shows a low h oughpu loss wi h espec o i . Sys em 3 achie es he highes h oughpu o all he channels conside ed. In ac , he di e ence wi h espec o Sys ems 1 and 2 is conside able. As as as use 's e o s is conce ned, al hough i is ou pe o med by Sys em 2 o channel BERs highe han 6.5 10- , i seems o be he bes choice assu- ming a long enough o wa d ame. On he o he hand, i s implemen a ion complexi y is also lowe han he o he sys ems. 1. 2. 3. 4. 5. 6. 7. The p o ec ion o he backwa d channel is PC 1 PC 2 enough o cope wi h he e o s in oduced by all he channels conside ed. Ac ually, he ac o conside ing he back channel ideal o no does no a ec he esul s. REFERENCES Da nell. M. 1985, "P oblems o Mobile HF Comunica ion and Techniques o Pe o mance Imp o emen ". ~~~ IEE P oceedings, 5 ,- Augus . Reed, A.P.C. and Hopkinson, J.N., 1990, "Adap i e Da a Link P o ocols: Design and Pe o mance o e Au oma ed HF Skywa e Links". Plessey, U.K. Kallel, S., 1990, "Analysis o Type I1 Hyb id ARQ Scheme wi h Code Combining". IEEE T ans. on Comms., 38, 0, Augus . Se inken, N.M., 1988, "HF Da a Message and Facsimile Te minal Sys em". 3 h IEE -, London. Spun icchia and Ca oggio, 1986, "Un P ocedimen o di Misu a e di Analisi degli E o i pe Sis emi di T ansmisio- ni Da i". XVII Conc esso In . pe 1'Ele onica". Dalmau, J.L., 1991, "A Con ibu ion o he Analysis o Coding Schemes o Mobile Da a Communica ions o e HF Channels". Doc o a e Thesis, Uni e si- a Poli ecnica de Ca alunya, Spain. F i chman, B.D., 1967, "A Bina y Cha- ac e iza ion using Pa i ioned Ma ko Chains". IEEE T ans. on In . Theo , IT-13, 2, Ap il. FOR I FIGURE 1: LABORATORY TEST OED I I USER AND DISK INTERFACE I ENCODER 1 TRANSYISION INTERLEYER HF CHANNEL BACK DECODER COYYUNICATIONS FIGURE 2: LAYERED STRUCTURE OF THE OOFTIURE RUNNINQ IN PC 1 8- e- 4- __ -~ 2- 0 - c ". .. .. c 0.1 channel be (x1E-2) - Se ies 1 + Se iem 2 * Se ies 9 sys em 1 sys em 2 sys em 3 igu e 3: use 's e o s s channel be h oughpu 0.8, 0.56 L:-- . 0.4 ' I 0.1 channel be (XI€-2) 1 sys em 1 sys em 2 sys em 3 Se ies 1 + Se ies 2 -+- Se ies 3 igu e 4: h oughpu s channel be 252 use 's e o s I' I 0' I Mo I80 200 220 240 280 280 300 320 numbe o blocks pe ame So i08 1 igu e 5: use 's e o s s numbe o blocks pe ame I -