ICREMETAL REDUDACY (IR) SCHEMES FOR W-CDMA HS-DSCH Amitava Ghosh 1, Kenneth Stewat, Rapeepat Ratasuk 1, Eoin Buckley, and Raa Bachu 1 Advanced Radio Technology, GTSS, Motoola, Alington Heights, IL, USA 60004 Pesonal Communications Secto, Motoola, Libetyville, IL, USA 60048 email: {qa0047, qa191, atasu1, w50105, fb04c}@email.mot.com Abstact This pape compaes two Hybid Automatic Repeat Request (HARQ) schemes using Incemental Redundancy (IR) which have been poposed fo the UMTS (W-CDMA) High Speed Downlink Shaed Channel (HS- DSCH). Thei elative thoughput pefomance is epoted fo vaious channel conditions. It is shown that the twostage ate-matching scheme has maginally wose pefomance compaed to the altenative scheme fo cetain coding ates and unde some channel conditions. I. ITRODUCTIO The two fundamental foms of HARQ ae Chase combining and incemental edundancy (IR). In Chase combining [1], each etansmission epeats the fist tansmission o pat of it. In IR, each etansmission povides new code fom the mothe code to build a lowe ate code []. While Chase combining is sufficient to make Adaptive Modulation and Coding (AMC) obus IR offes the potential fo bette pefomance with high initial and successive code ates, at highe SR estimation eo and FER opeating points (i.e., a geate pobability that a tansmission beyond the fist will be needed), albeit at the cost of additional memoy and decoding complexity. The consensus in the 3GPP UMTS standads bodies is to explicitly define and allow IR, while etaining Chase-like opeation as a subset of IR since the memoy equiements at the Use Equipment (UE) fo the highest ate is deived based on Chase combining. In this pape, two types of IR schemes fo W-CDMA HS-DSCH ae discussed: a) IR based on two-stage ate matching and b) IR based on block inteleaving. The IR based on two-stage ate matching [3][4][9] is the ageed upon IR scheme fo HS-DSCH. It uses two stages of ate matching whee the fist stage is used to match the amount of coded to the UE buffeing capability and the second stage is used to geneate the diffeent edundancy vesions. It may be noted that the fist stage will be bypassed in case the UE has full buffeing capability. IR based on block inteleaving was poposed in [5][6]. The appoach uses a block inteleave in the channel coding chain followed by a edundancy vesion selecto and a vitual bit pioity mappe. The block inteleave ows and umns ae pemuted accoding to UMTS-Release 99 intenal tubo inteleave and 1 st inteleave algoithms. This technique simplifies the Release- 99 channel coding chain by emoving the ate-matching block. The block inteleave acts both as a ate-matche and channel inteleave followed by a vitual bit pioity mappe. In this scheme, othogonal e-tansmissions (i.e., each tansmission contains unique paity ) ae guaanteed and as such the pefomance of this scheme is optimal. Section II descibes the W-CDMA HS-DSCH IR scheme based on two-stage ate matching. Section III povides a bief desciption of the IR scheme based on block inteleaving. Simulation esults compaing these two methods ae then pesented in Section IV. Finally, conclusions ae dawn in Section V. II. HARQ BASED O TWO-STAGE RATE- MATCHIG The 3GPP HARQ scheme fo HS-DSCH is shown in Figue 1. This scheme is based on the ate-matching algoithm defined in Section 4..7 of [7] and can suppot both self-decodable and non self-decodable tansmissions. The Fist Rate Matching block is used to adust the numbe of available coded at the ode-b to the vitual UE s buffe size. The imum numbe of soft available in the vitual IR buffe is signaled fom highe layes fo each HARQ pocess. The Second Rate Matching block is then used to ate-match and select the set of coded fo tansmission, given the edundancy vesion (RV) selected. C TTI bit sepaation Paity 1 Paity Fist Rate Matching RM_P1_1 RM_P_1 Vitual IR Buffe p1 p Second Rate Matching RM_P1_ RM_P_ Two paametes contol the geneation of the edundancy vesions, a self-decodability paamete s and e ini vaiation paamete. The goal of the e ini vaiation paamete is to enable obust etansmissions by othogonalizing available edundancy vesions that contain the same self-decodability paamete s. The numbe of RV s allowed fo HS-DSCH is limited to 8. HARQ second stage ate matching is done with the geneal method descibed in 4..7.5 of [7]. The initial (e ini ), incement (e plus ) and decement (e minus ) value of vaiable e in the ate matching patten detemination algoithm ae based on the RV paametes s and. The paamete s can take the value RM_S p1 p bit lection Figue 1. IR Scheme fo HS-DSCH based on twostage ate-matching. W 0-7803-7589-0/0/$17.00 00 IEEE PIMRC 00
0 o 1 to distinguish between pioitizing tematic and paity. The paamete (ange 0 to ) changes the initial eo vaiable e ini in the case of punctuing. In case of epetition both paametes and s change the initial eo vaiable e ini. The paametes X i, e plus and e minus ae calculated as pe Table 1 shown below. RM S Paity 1 RM P1_ p1 Paity RM P_ p X i e plus e minus t, a a p 1 t, p1 p1 a a p t, p p Table 1. Paametes fo the second-stage ate matching. Denote the numbe of befoe second ate matching as fo the tematic, p1 fo the paity 1, and p fo the paity, espectively. Denote the numbe of physical channels used fo the Coded Composite Tanspot Channel (CCTCH) by P. is the numbe of available to the CCTCH in one adio fame (defined as =P 3 1, whee 1 is given in [7]). The ate matching paametes ae detemined as follows. Fo + p1 + p, punctuing is pefomed in the second ate matching stage. The numbe of tansmitted tematic in a etansmission is = min{, } fo a tansmission of self-decodable type and = { ( p1 + p),0} in the non selfdecodable case. Fo > + p1 + p, epetition is pefomed in the second ate matching stage. A simila epetition ate in all bit steams is achieved by setting the numbe of tansmitted tematic as follows (Eq. 1) + p The numbe of paity in a tansmission is then and p1 (Eq. ) p (Eq. 3) fo the fist and second paity steam, espectively. The paamete a in Table 1 is chosen using a = fo the fist paity steam and a = 1 fo the second paity steam. The ate matching paamete e ini is calculated fo each bit steam accoding to the edundancy vesion paametes and s using e plus e ini( ) = Xi 1 mod eplus + 1 (Eq. 4) in case of punctuing, and e plus e ini( ) = Xi ( ) s + 1 mod + 1 eplus (Eq. 5) in case of epetition, whee {0,1,, -1} and is the total numbe of edundancy vesions allowed. ote that vaies depending on the modulation mode (fo 16- QAM, = and fo QPSK, = 4). Subsequently, the out of the nd ate-matching stage ae mapped based on pioity via the HARQ bit lection uni and inteleaved via one o moe Release 99 inteleaves based on the modulation level (QPSK o 16- QAM). The HARQ bit lection is achieved using a ectangula inteleave of size ow. The numbe of ows and umns ae detemined fom: ow = log ( M ) (Eq. 6) = F / ow whee M is the modulation size and F is the numbe of coded and ate-matched to be tansmitted. Data is witten into the inteleave umn by umn, and ead out of the inteleave umn by umn. The paamete is the numbe of tansmitted tematic. Intemediate values and c ae calculated using: and c = (Eq. 7) (Eq. 8) If c =0, the tematic ae witten into ows 1. Othewise tematic ae witten into ows 1 +1 in the fist c umns and ows 1 in the emaining c umns. The emaining space is filled with paity. The paity ae witten umn wise into the emaining ows of the espective umns. Paity 1 and ae witten in altenating ode. In the case of 16-QAM fo each umn the ae ead out of the inteleave in the ode ow 1,
ow, ow 3, ow 4. In the case of QPSK fo each umn the ae ead out of the inteleave in the ode ow1, ow, ow 3, ow 4. The inteleaving afte the bit lection unit is done using two Release 99 nd inteleave with fixed size (3x30). Fo QPSK, only one inteleave is used wheeas fo 16-QAM, thee ae two identical inteleaves of the same fixed size. Fo 16-QAM, the fist two out of the bit lection unit ae ead into the fist inteleave and the next two ae ead into the nd inteleave and so on. III. HARQ BASED O BLOCK-ITERLEAVIG This scheme was poposed in [5][6]. A block diagam illustating the functionality of this simple scheme is shown in Figue. It consists of a slightly modified Release 99 inteleave, a edundancy vesion selecto, and a Bit Pioity Mappe. The HARQ functionality is implemented by the block-inteleave and the edundancy vesion (RV) selecto. Tubo Code Paity RV selecto R 99 inteleave + Row eaangement R 99 inteleave + Row eaangement Bit Pioity Mappe Data is ead ow-wise into each inteleave, with dummy padded if < ow. This is identical to the pocedue used fo the Release 99 second inteleave. To facilitate flexibility in suppoting vaiable coding ates, both umns and ows ae pemuted pio to eading out of the block matix contents. The same umn pemutation is poposed fo both the tematic and paity inteleaves and is defined in [7]. The poposed ow pemutations ae deived diectly fom the inta-ow pemutations of the Release 99 Tubo code intenal inteleave. B. Redundancy Vesion Selecto The edundancy vesion selecto detemines the subset of tematic and paity to be tansmitted ove the ai. A salient featue of this method is that the selecto does not suggest a paticula ode of tansmission. Using this simple appoach the selected edundancy vesion sequence may be chosen to suppot Chase, patial and full IR schemes. Each edundancy vesion contains a stating umn α in eithe the tematic o paity inteleave (i.e. α { i, k} fo {0, 7}, i { S, P} and k { 1,, i, _ } ). The pocedue fo computing α is as follows. Fis compute then v = S, _ + P, _ (Eq. 9) RV selecto Figue. Simple IR/BPM Scheme. fo mod(,4) VR v mod(,4) 8 (Eq.10) {0, 7}. Finally, calculate α via A. Block Inteleave The encode used hee is the ate-1/3 Tubo encode specified in [7]. The un-punctued codewod ae sepaated into a tematic and a paity steam denoted by x S, k, x P, k espectively whee x S, k {0,1} and x P, k { p 1, k, p, k } (i.e., x P, k {{0,0}, {0,1}, {1,0}, {1,1}}). Each steam is ead into an ow block matix inteleave while the tail ae buffeed sepaately and ae late appended onto the un-punctued instantaneous codewod (whee the instantaneous codewod is the codewod tansmitted in a specific tansmission time inteval). Thus, the 1 buffeed tail ae always tansmitted. As an example, if the fist tansmission is at R=3/4 and the numbe of infomation is 600, then 600 tematic, 188 paity ae ead fom the and Paity inteleave espectively into the Vitual Bit Pioity mappe. Finally the 1 tail ae ead. The numbe of umns in each inteleave is always fixed at 30 while the numbe of ows is vaiable (dependent on the numbe of infomation ) and is detemined in exactly the same manne as the tubo code intenal inteleave defined fo Release 99 in step of section 4..3..3.1 of 5.1. if α = S, VR } (Eq. 11) VR S, _ { < and VR S, ˆŸ _ α = P, (Eq. 1) othewise. This gives fou unique stating umns evenly spaced thoughout the lowest code ate suppoted by the mobile. Coded ae ead fom the selected stating umns umn-wise eithe in inceasing ode ( y i ow +1), y i ow +), ) fo {0,1,,3}) o in deceasing ode ( y i ow 1), y i ow ), ) fo {4,5,6,7}. Coded continue to be ead until the end ( y i ˆŸ Š ow) fo {0,1,,3}) o stat ( y i, 1 fo {4,5,6,7}) of the inteleave is eached. Reading then continues umn wise fom the othe inteleave î, ( iˆ { S, P}, iˆ i ) eithe fom the fist umn o the last umn.
Column 1 Column 0 Column 5 Column 15 RV0 RV1 RV RV3 ex the two IR schemes ae compaed using a symbol-level simulato fo HS-DSCH. The esults pesented hee assume ideal channel estimation, QPSK/16-QAM modulation and non self-decodable e-tansmissions. Table lists additional elevant simulation paametes. ow=4 RV5 RV6 RV7 RV4 As an example, conside the case shown in Figue 3 whee the encoded message consists of 70 with the UE able to suppot the full ate 1/3 code. Assuming the edundancy vesions chosen fo tansmission ae RV0 and RV and each tansmission consisting of 960, then the fist tansmission will consist of codewod followed by the 1 tail and the second tansmission will consist of the codewod followed by the 1 tail. C. Vitual Bit Pioity Mappe Pioity bit mapping is based on utilizing the diffeing bit eliability offeed by highe ode constellations (16- QAM o highe). Since the tematic potions of a codewod ae of geate impotance to decode pefomance than the paity potions, they should be placed in positions of high eliability when highe ode constellation is used. Since channel inteleaving has aleady implicitly been pefomed, codewod can be ead diectly fom the and Paity punctuing block inteleaves, to fom the desied 16-QAM symbol in a simple and staightfowad manne. IV. SIMULATIO RESULTS The effective code ate is defined as n i /n i +n p whee n i is the numbe of infomation and n p is the numbe of unique paity, afte the second tansmission is plotted as a function of the initial code ate fo the above two schemes (assuming full buffeing) in Figue 4. The two schemes have identical esults fo initial code ates lowe than 1/. At highe code ates, howeve, the ate-matching scheme cannot select only unique paity in the second tansmission, thus esulting in a highe effective code ate. The block inteleaving appoach, howeve, is able to select all unique paity in subsequent tansmissions. As a esul the optimality of the block inteleaving based appoach is evident fom the figue since the blockinteleaving appoach could guaantee othogonal tansmissions as opposed to the two-stage ate-matching scheme. Paity S,_ =30 P,_ =30 Figue 3. Example of Redundancy Vesion Selecto and Codewod Selection. Paamete Caie Fequency Channel conditions HSDPA fame Length Io/Ioc Fast fading model Channel Coding Value GHz AWG, Flat Fading msec (3 slots) Vaiable Jakes spectum R=1/3 Tubo Max no. of Decode Ite 8 Metic fo Tubo Code Tubo Inteleave Max-Log-Map Random Table. Simulation Paametes. Figue 5 compaes the spectal efficiency of the two schemes unde static AWG channel fo =70, R=3/4 and QPSK modulation. In this case, the block-inteleaving based scheme has a maginal advantage ove the atematching scheme at I o /I oc geate than 1. db. Figue 6 compaes the spectal efficiency vs. I o /I oc of the two schemes using QPSK modulation, =70, R=3/4 code unde Rayleigh fading channel at 3 km/h. Fo the atematching appoach, we assume s=1, =0 on the fist tansmission, s=0, =0 on the second tansmission, s=0, =1 on the thid tansmission, and s=1, =1 on the fouth tansmission fo the ate matching appoach. A imum of 4 e-tansmissions wee allowed fo the packet fo both the above cases. Figue 7 compaes the pefomance using 16-QAM modulation at 10 kmph using R=3/4 code fo a message size of 1,440. Fo the ate-matching appoach, we again select s and as outlined above. It may be obseved fom both the figues that the block inteleaving appoach outpefoms the two-stage ate-matching scheme by appoximately 0.5 db unde fading channel conditions. V. COCLUSIOS The block-inteleaving scheme has the following advantages ove the 3GPP two-stage ate-matching scheme. Fis the tematic and paity inteleaves in Figue ae analogous to the second ate-matching block combined with the Release 99 inteleaves in that they inteleave as well as ate match. As such, a single set of backwads-compatible inteleaves seve as a simple appoach fo RV selection,
symbol mapping etc. Second, the pefomance of the blockinteleaving appoach is supeio to IR based on two-stage ate-matching by appoximately 0.5dB. Thid, fo the twostage ate matching scheme, the ate-matching pattens must be detemined fo evey edundancy vesion selected, wheeas in the block-inteleaving appoach only the index of the stating location is needed. Lastly, not all etansmissions ae guaanteed to have unique paity due to the ate-matching algoithm, which degades incemental edundancy pefomance. ACKOWLEDGEMETS The authos would like to acknowledge Bob Love and Bian Classon fo thei invaluable comments. REFERECES [1] D. Chase, Code combining A imum-likelihood de-coding appoach fo combining an abitay numbe of noisy packets, IEEE Tans. Comm., Vol. COM-33, o. 5, pp 385-393, May 1985. [] S. Lin and D. J. Costello, Eo Contol Coding: Fundamentals and Applications. ew Yok: Pentice Hall, 1983, pp. 477-494. [3] R1-01-1045, Eicsson, Physical-laye Hybid-ARQ functionality, 3GPP RA1#, Koea. [4] R1-01-1196, Siemens, Rate Matching and Incemental Redundancy fo HSDPA, 3GPP RA1#, Koea. [5] R1-01-144, Motoola, Revised:HARQ Scheme Using Incemental Redundancy, 3GPP RA1#, Koea. [6] R1-0-085, Motoola, Enhancement fo IR fo HSDPA, 3GPP RA1#4, USA. [7] 3GPP HSDPA-Physical Laye Aspects, 3GPP TS5.1, V5.0.0. [8] R1-0-034, Motoola, Enhancement of IR fo HSDPA, 3GPP RA1#4, USA. [9] R1-01-106, TI, Two Stage Rate Matching fo 3GPP HSDPA, 3GPP RA1 Ad-Hoc, Sophia Antipolis. 0.5 Spectal Efficiency 3.8.6.4. 1.8 1.6 Spectal Efficiency fo =70, QPSK, R=3/4, AWG Channel 1.4 1.8 1.6 1.4 1. 1 0.8 0.6 0.4 0. 0 Figue 5. Spectal Efficiency fo =70, QPSK, R=3/4, AWG Channel. Spectal Efficiency 3.5 1.5 1 0.5 Spectal Efficiency fo =70, QPSK, R=3/4, Flat Fading (3 km/h) Channel 0 10 5 0 5 10 15 Figue 6. Spectal Efficiency fo =70, QPSK, R=3/4, Flat Fading channel (3 km/h). 0.48 1.8 Spectal Efficiency fo =1440, 16 QAM, R=3/4, Fading (10 km/h) Channel 0.46 1.6 Effective Code Rate 0.44 0.4 0.4 0.38 0.36 Two Stage Rate Matching Scheme Spectal Efficiency 1.4 1. 1 0.8 0.6 0.34 0.4 0.3 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Initial Code Rate 0. Figue 4. Effective Code Rate Compaison of IR based on two-stage ate-matching and block inteleaving. 0 1 10 8 6 4 0 Figue 7. Spectal Efficiency fo =1440, R=3/4, 16- QAM, Flat Fading channel (10 km/h).