Single-User MIMO ML Successive Interference Canceling Receiver with HARQ-IR Protocol

Transcription

1 Single-Use MIMO ML Successive Intefeence Canceling Receive with HARQ-IR Potocol Elena Lukashova, Floian Kaltenbege, Raymond Knopp EURECOM Campus SophiaTech, 40 Route des Chappes, 0640 Biot, Fance Abstact Incemental Redundancy Hybid Automatic Repeat Request (HARQ-IR) etansmission potocol was developed to educe the tansmission eos ove fading channels though multiple etansmissions. In this pape, we implement HARQ-IR fo Single Use MIMO systems with Successive Intefeence Canceling (SIC) eceive. We pefom the thoughput and eliability analysis fo the MIMO system with Closed Loop Spatial Multiplexing (TM4) with two codewods (CW) and compae the esults fo SIC and Paallel Intefeence Awae (PIA) eceive. Ou SIC eceive benefits fom the log-likelihood atios combining fo the second tanspot block, accessed though the multi-ound SIC pocedue a posteioi to all the pevious ounds once the fist CW has been decoded, while PIA eceive teats both CWs in the same manne. The SIC eceive achieves highe thoughput in fequency selective envionment, eceiving most of the gain in the fist two etansmission ounds. Fo both eceives, the fist tanspot block enjoys huge pefomance impovement in the low SNR egime thanks to HARQ at the pice of spectal efficiency degadation due to euse of the space-fequency esouces. We also investigate the optimal etansmission scheme fo TM4 in the scenaios with one successfully decoded CW, while anothe one is in eo. I. INTRODUCTION LTE has been designed to allow maximum flexibility in exploiting the benefits of MIMO channels. The so called tansmission modes (TM) ange fom tansmit divesity, ove beamfoming to spatial multiplexing. At the same time, Hybid Automatic Repeat Request (HARQ) etansmission potocols wee developed to educe the tansmission eos ove fading channels though multiple etansmissions. If the CW is successfully decoded, the UE sends an acknowledgment message (ACK) on the one of the uplink channels and the enodeb poceeds to the tansmission of the next data packet. In the opposite case, the UE sends a non-acknowledgment message (NACK) and the enodeb etansmits the package. At the UE side, the log-likelihood atios (LLR) of the coesponding bits will be added up and the new decoding attempt will be pefomed. HARQ potocol has two options: Chase Combining (CC) and Incemental Redundancy (IR). HARQ-CC can be thought of as a epetition code: upon eception of the NACK message, the enodeb esends exactly the same copy of peviously tansmitted message. Thus, thee is no coding gain povided by this method. On contay, using moe complex IR type of the potocol, the enodeb sends a diffeent edundancy vesion (RV) of the message at each etansmission ound. The RV is geneated fom the diffeent factions of the systematic and paity check bits, deliveing coding gain with evey new etansmission. HARQ-IR, that is in focus of this pape, is shown to outpefom the HARQ-CC in the majoity of the scenaios. The fundamental analysis of the HARQ potocols in Gaussian collision channel was done by Caie and Tuninetti 2, whee the authos deived the closed-fom expession fo the thoughput metics. Thei wok was extended to MIMO V-BLAST systems by Dekosy 3 based on the conditional cut-off ate of MIMO tansmission. One of the featues of the LTE technology is an ability of the enodeb to adjust the Modulation and Coding Scheme (MCS) to the instantaneous channel conditions based on the Channel State Infomation (CSI). Szczecinski investigated the benefits of adaptive HARQ-IR, when the Modulation and Coding Scheme (MCS) changes ove the etansmission ounds 4. His simulations show that the adaptive HARQ bings significant gains compaed to non-adaptive scheme fo high SNR in channels with outdated CSI at the tansmitte. The inteesting esults of the combination of HARQ and adaptive modulation and coding wee obtained in, whee with only one bit of feedback, the authos achieve pefomance close to the egodic capacity. In this pape we stictly follow the 3GPP LTE standads 6, 7, 8, which define downlink HARQ potocol as non-adaptive and asynchonous. The advanced MIMO eceive achitectue takes coss-laye intefeence into account. This may put specific constains on the design of the etansmission scheme 9. Symbol-level combining scheme fo HARQ with Intefeence-Awae (IA) successive decoding was poposed in 0. The detailed analysis of the optimal combining schemes fo MIMO systems with HARQ was done in. In this pape we study the Single- Use (SU) MIMO system with ou Maximum-Likelihood (ML) IA Successive Intefeence Canceling (SIC) eceive 2 with HARQ potocol. The eceive passes the full eceived signal though the Matched Filte (MF) detecto, computes IA soft LLR metic 3 fo the fist steam and decodes the fist CW. Afte the fist CW is successfully decoded, the eceive econstucts it with the coesponding RV fo a cuent and pevious (if available) HARQ ounds, multiplies the signal with the coesponding compensated effective chan-

2 nel coefficients and subtacts it fom the MF outputs so that the detection of the second steam is intefeence-fee, thus benefiting fom multi-ound SIC LLRs combining fo the second tanspot block (TB). The majoity of the state-of-at liteatue focuses on the methods to handle HARQ potocol in MIMO systems at the UE side. At the same time, thee ae vey few souces descibing the etansmission options of the enodeb. In June 6 we pefomed a dive test campaign in Sophia Antipolis, Fance, duing which the full message flow between the enodeb and the UE was captued. This allowed us to study the HARQ implementation in the pactical MIMO LTE system with Eicsson enodeb, configued in Cyclic Delay Divesity (TM3) using the DCI fomat 2A with 2 antenna pots. When the UE awaits to eceive two TBs pe subfame, they ae associated with the same HARQ pocess. If only one CW is in eo, the enodeb pefoms a etansmission only of the eoneous CW, deactivating the CW that was successfully decoded. Howeve, thee is no pecoding infomation field fo the 2 antenna pot configuation in the DCI fomat 2A. The tansmit divesity (Alamouti pecoding) is the only option available fo the enodeb fo a etansmission of a single CW in TM3. This inspied us to study the etansmission stategies fo Closed Loop Spatial Multiplexing (TM4) based on OpenAiInteface (OAI) 4 downlink (DL) simulato. TM4 uses the DCI fomat 2 with an active pecoding infomation field fo 2 antenna pot tansmission. Fo the single CW etansmission, the enodeb uses Tempoay Pecode Matix Indicato (TPMI) in the DCI fomat 2 to signal the etansmission scheme. Ou main contibutions in this pape ae the following: We implement HARQ potocol suppot fo the R-ML IA SIC eceive in OAI DL simulato. We pefom the thoughput and eliability analysis of TM4 with HARQ potocol with R-ML IA SIC and R-ML We analyze diffeent etansmission options applying the thoughput metic in the situations with actual and outdated CSI in fequency-selective fading channels in the scenaios, when one of the CW is decoded, and anothe one is in eo. II. SYSTEM MODEL In this pape we often efe to the tems CW, TB and DCI. The MAC laye handles the TBs. The PHY laye maps TBs on the CWs. TB size defines a maximum numbe of bits that can be sent in ms. The DCI can be seen as an inteface between the MAC and PHY layes and caies the infomation about the MCS, New Data Indicato (NDI), RV, TPMI and esouce allocation. We conside 2 2 MIMO system, whee the connection between the enodeb and the UE is established in TM4 with DCI fomat 2 and 4 ounds HARQ-IR etansmission potocol is configued. The enodeb sends TB 0 and TB mapped onto spatially multiplexed CW 0 and CW at the initial tansmission Table I: TPMI bit field intepetation fo two CW tansmission Bit field TPMI intepetation j j 2 last PMI on PUSCH ound = 0. The eceived signal vecto y l C 2 fo the l-th subcaie seen by the UE is given by = 0 : y l = H l P l x l + n l, l =, 2..., L, () whee x l Q M0,M is the vecto of two complex symbols x 0 and x with vaiance of σ0 2 and σ, 2 Q M0,M := Q M0 Q M is a Catesian poduct of two modulation alphabets Q M0 and Q M, M 0, M {2, 4, 6} ae the modulation odes of the QAM constellations. The vecto n l is Zeo Mean Ciculaly Symmetic Complex Gaussian (ZMCSCG) white noise of double-sided powe spectal density N 0 /2 at the 2 eceive antennas of UE. The matix H l is a 2 2 channel matix built with espect to the applied in simulation section channel model and P l is pecoding matix employed by the enodeb at the l-th RE. Fo the sake of simplicity, we dop the subcaie index and eplace the multiplication of H and P with the effective channel H eff : y = H eff x + n, H eff = h eff0 h eff. (2) Thee ae thee pecoding options signaled though the TPMI field fo the DCI fomat 2 with 2 active CWs. The bit field to PMI intepetation is pesented in Table I 8. In ou simulations, the enodeb is configued to use TPMI2 the PMI eceived with latest CSI epot on Physical Uplink Shaed Channel (PUSCH). Since with ou SIC eceive decoding of CW is only accessible if CW 0 is decoded coectly, the pecode is selected in the way to minimize the Block Eo Rate () of the fist steam. Specifically, the UE selects the pecode matix P, which ensues that the effective channel of the fist steam is stonge than the one of the second steam by evaluating the coelation coefficient ρ 0 = h H eff h eff0. Compaing the eal and imaginay pats of ρ 0, the UE picks up one P fom the two options:, 2 fo R(ρ 0 ) I(ρ 0 ) P = (3) 2 j j, fo R(ρ 0 ) < I(ρ 0 ) III. R-ML IA SIC RECEIVER IMPLEMENTATION Fig. pesents the basic scheme of R-ML IA The eceived signal y (2) goes though the linea MF ĤH eff. The eceived signal is tansfomed into ymf0 = y ĤH MF effh x0 + x n 0, (4) n

3 Table II: TPMI bit field intepetation fo a single CW etansmission Bit field TPMI intepetation 0 Alamouti Figue : R-ML IA SIC eceive scheme whee Ĥeff is the estimated channel matix. The compensated signal belonging to the TB is ( y MF = ρ ) ) x 0 + (ĥeff0 heff0 + ĥ eff heff x + n, () whee ρ = ĥ eff0 heff00 + ĥ eff heff0 is the coelation coefficient. Afte the MF, the eceive computes the IA soft bit LLR metic fo lowe-ate TB 0, teating TB as intefeence. The bit metic fo TB 0 is then passed to the tubo-decode. If the Cyclic Redundancy Check (CRC) test confims the coect decoding of TB 0, the SIC pocedue is tiggeed. If the CRC fails, the LLR values of undecoded CWs ae stoed at the eceive and ae updated duing the next HARQ ounds. The SIC block aims to econstuct the decoded signal fom the bits, belonging to TB 0. It pefoms e-encoding and emodulation pocesses, multiplication of the symbols with the channel estimates, subtaction of the ecoveed signal fom the MF output and LLR computation fo TB. Afte the stipping unit, TB enjoys intefeence-fee detection. IV. RETRANSMISSION PROTOCOLS Suppose TB 0 is decoded on the ound = dec. Fo all the ounds dec, the enodeb sends a new RV of TB 0 and TB that ae mapped into CW 0 and CW and stoes the compensated eceived signal as well as the coelation coefficients. dec : TB 0 CW 0, TB CW. The eceived signal vecto is y MF = ĤH eff H x + n, = 0,..., dec. (6) A. Multi-ound SIC pocedue As soon as TB 0 is decoded, the multi-ound SIC pocedue is tiggeed. Since the UE now knows x 0, it may go though all the pevious ounds = 0,..., dec to econstuct ρ x 0 with the coesponding RV, pass it though the stipping unit to obtain ) ỹ MF = (ĥeff0 heff0 + ĥ eff heff x + n (7) and compute the coesponding multi-ound intefeence-fee LLR metics. Afte ate matching, that combines the multiound LLRs, the eceive will make an attempt to decode TB. If the decoding is successful, the UE sends {ACK 0, ACK } message to the enodeb and moves to the next package. Othewise, {ACK 0, NACK } is sent, and if the limit of etansmission ounds has not been eached yet, the enodeb will etansmit a new RV of TB j 4 2 j st column of the last PMI on PUSCH 6 2nd column of the last PMI on PUSCH B. Retansmission of TB If TB is not decoded afte dec ounds of SIC pocedue, it is not feasible to etansmit TB 0 again, thus, the enodeb will deactivate it in the etansmission. The deactivation is pefomed though the DCI: the MCS fo the deactivated TB is set to zeo, while the coesponding RV is set to. If only one TB is active, it must be mapped on CW 0 8: dec < ( max ) : TB CW 0, TB 0 is disabled. Afte disabling TB 0, the enodeb keeps DCI fomat 2 and faces a choice of the pecoding options, which ae signaled to the UE in TPMI. It includes Alamouti pecoding and single laye pecoding using eithe a pedefined pecode o the one fom latest CSI epot on PUSCH. A TPMI bit field intepetation fo the DCI fomat 2 fo 2 antenna pots with one active CW is shown in Table II 8. In this pape we focus on 3 options: TPMI {0,, 6}, since Alamouti pecoding is designed to incease the eliability in low SNR egime, and TPMI and TPMI6 may impove system pefomance at high SNR. Apat fom the diffeent TPMI options, we also conside two possibilities: if thee is an actual CSI fo each etansmission ound, o if the last CSI was eceived on dec and thus is outdated fo the dec < ( max ) ounds when single TB is etansmitted. Ou pecode is optimal fo CW 0, thus when TB is mapped to CW 0, TPMI is expected to pefom bette than TPMI6. This makes the compaison between TPMI with actual CSI (assuming only PMI feedback, Channel Quality Indicato (CQI) is not taken into account) and TPMI6 with outdated CSI inteesting. a) TPMI0: The enodeb configues the etansmission in Alamouti mode. In this case the eceived signal fo ounds dec < ( max ) can be seen as: x0 x whee X = x x. 0 y = 2 Xh + n, (8) b) TPMI and TPMI6: the enodeb configues the single laye etansmission based on the latest PMI epot on PUSCH,

4 using pecode p fom the fist (TPMI) o second (TPMI6) column multiplied by 2 of latest epoted PMI. In this case the eceived signal fo ounds dec < ( max ) can be seen as: y = Hpx + n. (9) The potocol implementation is descibed in Algoithm - Algoithm 3. Algoithm Retansmission algoithm fo TM4 SIC detection : 0, TPMI 2, max 4, sic 0 2: TB 0 flag active, TB 0 CW 0, TB CW 3: TB 0 dec false, TB dec false 4: while ( ( max )) & (TB 0 dec = false) do : enodeb: encoding, ate matching, modulation fo both TBs. 6: UE:measuements, feedback epoting on PUSCH. 7: UE:MF and stoe the compensated channels. 8: UE:LLR 0 computation and TB 0 decoding 9: if TB 0 is decoded then 0: = dec : 2: TB 0 dec = tue UE: call pocedue SIC 3: else 4: UE: {NACK 0, NACK } : + 6: end if 7: end while Algoithm 2 Multi-ound SIC Pocedue : pocedue SIC( dec, MF outputs fom pevious ounds) 2: fo 0 sic dec do 3: econstuct ρ x 0, obtain y MF,sic 4: LLR sic computation : end fo 6: Combine LLR and decode TB 7: if TB is decoded then 8: 9: TB dec = tue, UE: ACK 0, ACK beak Move to the next package 0: else : UE: {ACK 0, NACK } 2: + 3: UE: call pocedue Retansmit TB 4: end if : end pocedue Algoithm 3 Retansmit TB : pocedue Retansmit TB () 2: 3: TB 0 flag deactivated, TB CW 0 enodeb: choose TPMI {0,..6} 4: fo ( max ) do : enodeb: encoding, ate matching, modulation fo TB. 6: UE:MF and stoe the compensated channels. 7: UE:LLR computation and combining 8: UE:TB decoding 9: if TB is decoded then 0: : TB dec = tue, UE: {ACK 0, ACK } beak Move to the next package 2: else 3: UE: {ACK 0, NACK } 4: + : end if 6: end fo 7: end pocedue V. NUMERICAL RESULTS AND DISCUSSION A. Simulation paametes The link-level simulations (LLS) wee caied out fo the MHz 8-tap Rayleigh fading channel and Extended Pedestian A (EPA) channel with zeo Dopple fequency. Fo the EPA channel modeling we applied low and modeate coelation matix, efeing to them as to EPAL and EPAM espectively. The enodeb sends 00 packets with Physical Downlink Contol Channel symbol ove the wide ange of noise vaiances. Evey etansmission ound is dawn fom a new channel ealization. Since TM4 is designed fo high data ates tansmission, we chose MCS 0 and MCS, such that 0 MCS 0 28, MCS 0 MCS 28. Fo each fading envionment we conside thee etansmission options: TPMI0, TPMI with actual CSI, and TPMI6 with outdated CSI. B. MCS Optimization and Thoughput Analysis fo the Multiple Rounds The total system thoughput of the MIMO system with 2 TBs and max HARQ ounds can be seen as T tot = (T 0 + T ), (0) max =0 T 0 = + R0 ( 0 ), T = + R ( ), whee T 0 and T ae the thoughput values fo TB 0 and TB on the ound, R 0 and R ae the ates coesponding to MCS 0 and MCS, 0 and ae the coesponding Block Eo Rates. The fact that CW is attempted fo a decoding only if CW 0 is decoded is taken into account in. SIC eceives ae vey sensitive to the choice of MCS: if the instantaneous channel does not suppot the ate of MCS 0, CW 0 is not decoded and the SIC pocedue is thus not tiggeed. On the othe hand, if the fist steam is decoded, the second steam becomes intefeence-fee and can potentially cay highe infomation ates. Thee exist an optimal combination of MCS 0 and MCS that maximizes aveaged long-tem thoughput (0). We apply the methodology fom 2: LLS taces ae geneated fo all possible MCS combinations fo 4 HARQ ounds. Afte that fo each SNR point we select MCS 0 and MCS, which coesponds to the maximum thoughput. Spectal efficiency declines with evey etansmission ound due to euse of space-fequency esouces by the same TB. Thus, the maximum contibution to the aveaged long-tem thoughput is done duing the fist ound. In the pevious publication 2, that was ecently submitted, we showed that ou SIC eceive achieves up to.8 Mbps thoughput gain in MHz bandwidth flat Rayleigh fading compaed to ou R-ML IA eceive 6 in scenaios without HARQ potocol. In this pape we extend the compaison to the multiple HARQ ounds and fequency-selective Rayleigh 8-tap and EPA channel models. Fig. 2 illustates the total system thoughput T tot afte 4 HARQ ounds and thoughput T0 0 +T0

5 Thoughput, Mbps 3 0 Thoughput, Mbps 3 0 Thoughput, Mbps 3 0 T tot TPMI0 T tot TPMI UP T tot TPMI6 OUT T T (a) SIC eceive 8-tap Rayleigh fading channel. (b) SIC eceive EPAL channel. (c) SIC eceive EPAM channel. Thoughput, Mbps 3 0 Thoughput, Mbps 3 0 Thoughput, Mbps 3 0 T tot TPMI0 T tot TPMI UP T tot TPMI6 OUT T T (d) PIA eceive 8-tap Rayleigh fading channel. (e) PIA eceive EPAL channel. (f) PIA eceive EPAM channel. Figue 2: Compaison of SIC and PIA eceive s total thoughput T tot afte 4 HARQ ounds and thoughput T T 0 afte the fist ound = 0 applying optimized MCS 0 and MCS. We conside TPMI={0,, 6} duing the etansmissions of single TB. TPMI is studied in updated CSI scenaio, while TPMI6 is consideed in outdated CSI envionment. Thoughput T 0 0 +T 0 afte the fist ound is independent fom the TPMI used duing the etansmissions of single TB. afte the fist ound = 0 achieved with MCS 0 and MCS fo SIC and PIA detection. Ou SIC eceive outpefoms the PIA eceive on the fist ound = 0 in all the channel models with the gains vaying fom 2 4 Mbps in high SNR egime to 7 Mbps in low SNR egime. Multiple etansmission ounds educe this gap, but the SIC eceive still pefoms bette at high SNR (up to 2 3 Mbps). The wose the channel is, the bigge is the contibution of the etansmissions in tems of thoughput. Howeve, the TPMI duing single TB etansmission does not have a noticeable impact on the thoughput. Thee is a slight pefeence to TPMI and TPMI6 in actual and outdated CSI scenaios ove Alamouti coding in the fequency-selective Rayleigh channel, while in EPA channels thee is no visible diffeence. C. Reliability Analysis In this section we focus on the contibution of the multiple HARQ ounds to the eliability of the MIMO system with ou The MCS optimization methodology applied in the pevious section fits well to povide an initial idea about thoughput, achievable with multiple HARQ ounds. Howeve, it eflects pooly the TPMI influence on the system pefomance in the cases with a single CW etansmission. Fo the following eliability analysis we detach fom the optimized MCS 0 and MCS and conside a few paticula MCS. To cove all the modulation odes, we select thee MCS combinations: MCS 0 = 2 and MCS = 6, MCS 0 = 6 and MCS =, MCS 0 = and MCS = 26. The main contibution fom the HARQ potocol is pefomance impovement in the lowe SNR egime. Fig. 3 intoduces the of TB 0 fo 4 HARQ ounds. At the taget level of, the 2 6 MCS combination eceives 0 db and 3. db impovement in Rayleigh and EPA channel espectively. The SNR gain inceases fo the highe MCS and eaches 2 and db fo 6 22 and MCS combinations in Rayleigh and EPAM channel espectively. Howeve, these benefits ae coming at pice of spectal efficiency degadation with evey etansmission ound, as the space-fequency esouces ae eused by the same TB. Since the fist steam is teated by the SIC and PIA eceives in the same way, the R-ML PIA eceive is expected to show identical pefomance fo TB 0. We illustate the analysis of the second steam with examples fom the EPAM channel, in which thee is a significant ( Mbps) diffeence in thoughput pefomance between scenaios with 4 HARQ ounds and scenaios without HARQ, as we peviously showed on Fig. 2. Afte TB 0 is decoded on ound dec, the SIC eceive econstucts TB 0 with the coesponding RV fo each ound 0 dec, thus obtaining multi-ound LLRs fo TB. Fig. 4 shows, how many fames belonging to TB wee passed to the multi-ound SIC pocedue and how many wee suc-

6 cessfully decoded though it on the ound. In the low SNR egime, a negligible amount of attempts was done at = 0 due to the high of TB 0 in Fig. 3, while a significant amount of them was pefomed in the next ounds, since the of TB 0 is emakably lowe afte etansmissions even at low SNR. Howeve, the majoity of attempts failed, meaning that the gain fom the LLR combining though multi-ound SIC pocedue is not sufficiently high to impove the pefomance in the low SNR egime. At modeate SNR level, about 0% of the attempts ae decoded though the SIC pocedue, while in high SNR level this value eaches 00%. We conclude that ou SIC eceive implementation benefits fom multi-ound LLR combining at modeate SNR and the majoity of combining gain is coming fom the second ound =, while in the low SNR egime gains ae not that impessive. At each ound ( > 0), the total amount of etansmissions et tot is composed fom the faction of single TB etansmis- and the faction of two TBs etansmission sions et single et multipl : et single = et tot et multipl. () In Fig. we compae the amount of etansmissions of single TB et single on ound in EPAM channel fo ou SIC and PIA eceives. In the low SNR egime the amount of et single is almost identical, while in modeate and high SNR egime SIC eceive has about 0% less of et single due to the benefits of LLR combining though multi-ound TB 0 econstuction and subsequent subtaction. This suppots the idea that the main benefits of SIC eceive ae achieved duing the fist two ounds = 0 and =. In pactice, thee ae cases fo the PIA eceive, when TB is decoded, while TB 0 is in eo. In this case TB is deactivated and the UE equests the etansmission of single TB 0. Such cases ae negligibly ae due to ou pecode selection stategy, that maximizes the pobability of CW 0 to be decoded. We now investigate the of TB (Fig. 6). At the vey fist ound = 0 with all the MCS combinations, ou SIC eceive significantly outpefoms PIA detection due to the SIC pocedue, fo example, it achieves of at SNR level 0 db lowe than fo PIA eceive fo 2 6MCS combination. The SNR level fo of is slightly lowe fo the SIC eceive afte the fist ound, while the following etansmission ounds ae binging significant benefits to the PIA detection, while fo the SIC eceive gain between the thid and fouth etansmission ound is not emakable. Regading the TPMI fo the et single, both eceives show slight pefeence fo the Alamouti pecoding. This means, that the etansmission scheme, employed by the Eicsson enodeb fo TM3 is the optimal etansmission scheme fo TM4. D. A Note on Computational Effot In the peceding wok 2 we showed, that ou SIC eceive is % moe time efficient than ou PIA eceive given only one tansmission ound. Howeve, thee ae less etansmissions of the single TB in high SNR egime with HARQ suppot if SIC detection is applied, meaning that oveall pocessing time is futhe educed. On the othe side, the SIC detection clealy has high CPU consumption, since we need to stoe the MF outputs and coelation coefficients fo TB 0 fo each etansmission to be able to econstuct the signal fom peceding ounds. VI. CONCLUSION Inspied by the dive test measuement campaign, we have pesented HARQ potocol implementation fo ou R-ML IA SIC and R-ML PIA eceives fo LTE TM4 in OAI DL simulato. The SIC eceive achieves highe thoughput in all the simulated scenaios. Fo both eceives, the fist TB accesses huge pefomance impovement in the low SNR egime thanks to HARQ (up to 0 db eduction to achieve of ) at the pice of spectal efficiency degadation with evey etansmission ound, as the space-fequency esouces ae eused by the same TB. The Eicsson enodeb that was used to pefom the dive tests, was configued in TM3 and applied Alamouti pecoding as the only option allowed by the DCI fo this tansmission mode. Ou analysis fo the DCI fomat 2 showed that Alamouti pecoding is favoable fo the etansmissions of the single CW in TM4 fom the eliability point of view, while thee is no noticeable pefeence to any of the etansmission schemes fom the thoughput point of view. The futue publication will povide the feedback computation methodology fo PIA and SIC eceives and extend ou analysis to TM Ray8 EPAM Ray8 EPAM Ray8 EPAM Ray8 EPAM (a) MCS 0 = 2, MCS = 6. (b) MCS 0 = 6, MCS = 22. (c) MCS 0 =, MCS = 26. Figue 3: of TB 0 in 8-tap Rayleigh and EPAM channels fo 4 HARQ ounds (identical fo SIC and PIA eceive).

7 Attempts Attempts Attempts (a) MCS 0 = 2, MCS = 6. (b) MCS 0 = 6, MCS = 22. (c) MCS 0 =, MCS = 26. Decoded fames Decoded fames Decoded fames (d) MCS 0 = 2, MCS = 6. (e) MCS 0 = 6, MCS = 22. (f) MCS 0 =, MCS = 26. Figue 4: Amount of TB fames attempted to be decoded (a,b,c) and successfully decoded (d,e,f) though SIC pocedue fo the ound in EPAM channel. Fames 2 3 Fames (a) MCS 0 = 2, MCS = 6, (d) MCS 0 = 2, MCS = 6, (b) MCS 0 = 6, MCS = 22, (e) MCS 0 = 6, MCS = 22, (c) MCS 0 =, MCS = 26, (f) MCS 0 =, MCS = 26, Figue : Amount of etansmissions of single TB et single eceives. on ound in EPAM channel fo SIC (a,b,c) and PIA (d,e,f)

8 TPMI0 0 TPMIUP 0 TPMI6 TPMI0 TPMIUP TPMI6 2 TPMI0 2 TPMIUP 2 TPMI6 3 TPMI0 3 TPMIUP 3 TPMI6 (a) MCS 0 = 2, MCS = 6, (b) MCS 0 = 6, MCS = 22 (c) MCS 0 =, MCS = 26, TPMI0 0 TPMIUP 0 TPMI6 TPMI0 TPMIUP TPMI6 2 TPMI0 2 TPMIUP 2 TPMI6 3 TPMI0 3 TPMIUP 3 TPMI6 (d) MCS 0 = 2, MCS = 6, (e) MCS 0 = 6, MCS = 22, (f) MCS 0 =, MCS = 26, Figue 6: of TB in EPAM channel fo 4 HARQ ounds fo SIC (a,b,c) and PIA (d,e,f) eceive fo TPMI={0,, 6}. TPMI is studied in updated CSI scenaio, while TPMI6 is consideed in outdated CSI envionment. REFERENCES E. N. Onggosanusi, A. G. Dabak, Y. Hui, and G. Jeong, Hybid aq tansmission and combining fo mimo systems, in Communications, 03. ICC 03. IEEE Intenational Confeence on, vol., May 03, pp vol.. 2 G. Caie and D. Tuninetti, The thoughput of hybid-aq potocols fo the gaussian collision channel, IEEE Tansactions on Infomation Theoy, vol. 47, no., pp , 0. 3 A. Dekosy, A cutoff ate based coss-laye metic fo mimo-haq tansmission, in 0 IEEE 6th Intenational Symposium on Pesonal, Indoo and Mobile Radio Communications, vol. 4, Sept 0, pp Vol L. Szczecinski, C. Coea, and L. Ahumada, Vaiable-ate tansmission fo incemental edundancy hybid aq, in Global Telecommunications Confeence (GLOBECOM 0), 0 IEEE, Dec 0, pp.. T. Villa, R. Mez, R. Knopp, and U. Takya, Adaptive modulation and coding with hybid-aq fo latency-constained netwoks, in Euopean Wieless, 2. EW. 8th Euopean Wieless Confeence, Apil 2, pp GPP, Evolved univesal teestial adio access (e-uta); physical laye pocedues, 3GPP, Tech. Rep. Technical Specification V2.3, Oct 4. 7, Evolved univesal teestial adio access (e-uta); physical channels and modulation, 3GPP, Tech. Rep. Technical Specification 36.2 V2.6.0, July. 8, Evolved univesal teestial adio access (e-uta); multiplexing and channel coding, 3GPP, Tech. Rep. Technical Specification V2.7.0, July. 9 D. Toumpakais, J. Lee, A. Matache, and H. L. Lou, Pefomance of mimo haq unde eceive complexity constaints, in IEEE GLOBE- COM IEEE Global Telecommunications Confeence, Nov 08, pp.. 0 H. Kwon, J. Lee, and I. Kang, Symbol-level combining fo hybid aq on intefeence-awae successive decoding, in 3 IEEE Global Communications Confeence (GLOBECOM), Dec 3, pp E. W. Jang, J. Lee, H. L. Lou, and J. M. Cioffi, Optimal combining schemes fo mimo systems with hybid aq, in 07 IEEE Intenational Symposium on Infomation Theoy, June 07, pp E. Lukashova, F. Kaltenbege, and R. Knopp, ML successive intefeence cancelling vs paallel intefeence awae detection in LTE SU-MIMO systems, in Submitted to 7 Euopean Confeence on Netwoks and Communications (EuCNC): Physical Laye and Fundamentals (PHY) (EuCNC7 - PHY), Oulu, Finland, Jun R. Ghaffa and R. Knopp, Low complexity metics fo bicm siso and mimo systems, in Vehicula Technology Confeence (VTC 0- Sping), 0 IEEE 7st, May 0, pp OpenAiIntefacePlatfom, Euecom. Online. Available: 3GPP, Evolved univesal teestial adio access (e-uta); Use Equipment (UE) adio tansmission and eception, 3GPP, Tech. Rep. Technical Specification 36.0 V2.9.0, Oct. 6 E. Lukashova, F. Kaltenbege, R. Knopp, and C. Bonnet, PHY laye abstaction fo SU-MIMO LTE system employing paallel Intefeence- Awae detection, in Intenational Wokshop on Link and System Level Simulations 6 (IWSLS2 6), Vienna, Austia, June 6.