Rate-Splitting Multiple Access with a SIC-Free Receiver: An Experimental Study

IEEE COMMUNICA TIONS LETTERS 1 Rate-Split ting Multiple Access with a SIC-Free Recei v er: A n Ex perimental Study Guoqian Sun, Xinze L yu, Bruno Clerk Abstract —Most Rate-Splitting Multiple Access (RSMA) im- plementations rely on successiv e interference cancellation (SIC) at the recei ve r , whose performa nce is inherently limited by error propagation during common-stream decodin g. This paper addresses this issue by dev eloping a S IC-free RSMA recei ver based on joint demappin g (JD), which directly evaluates bit vector s ov er a co mposite constellation. Using a two-user Mult i ple- Input Sin gle-Output (MIS O) prototype, we conduct ov er -the- air measur ements to systematically compar e SIC and JD-based recei vers. The results show that the proposed SIC-free receiv er prov ides stro nger r eliability and better practicality ov er a wid er operating range, with all obser va tions being consistent with theoretical expectations. Index T erms —Rate-Splitting M ultiple Access, RSMA proto- type, Software-Defined Radio. I . I N T R O D U C T I O N F UTURE wireless communicatio n systems are requ ired to support high spectral efficiency and massi ve con nectivity under heterog eneous user condition s. In the d ownlink multi- user mu ltiple-inpu t m ultiple-ou tput ( MU-MIMO) setup, the design of multiple access schemes is crucial for mitigat- ing inter-user interfere n ce and improving system robustness. Con ventional Sp a tial Division Multiple Acc e ss (SDMA) suf- fers from per forman ce degradation under spatially cor related channels [1]. Mean while, No n-Orthog onal Multiple Access (NOMA) is less ef fectiv e when users hav e similar ch annel strengths [2], as insufficient power disp a rity weakens the effecti ven e ss of successive interferen c e cancellation ( SIC). Rate-Splitting Multiple Access (RSMA) ha s emerged as a promising multiple access fram ework to overcome these lim- itations. Rate-splitting was in itially studied fo r single-anten na interferen ce channels [3] and was later extended to multi- antenna b roadcast chan nels, where it was shown to outperform conv entional s chemes under partial Channel State In formation (CSI) at th e T ran sm itter [ 4]. By splitting each u ser’ s message into a common stream decoded by all users and priv a te stream s decoded individually , RSMA bridge s the gap between fully de- coding interferen c e and treating interferen ce as noise [5]. This flexible interference manageme nt strategy has been shown to enlarge the achiev able r a te region compar e d to SDMA and NOMA, improve spectral ef ficien cy and f airness, and enhan ce robustness against cha n nel variations and imperf e ct CSI [4], [6], [7]. As a result, RSMA has b een extensively in vestigated in terms of tran scei ver design and resource optimization u n der various CSI assumptions [8]. G. Sun, X. L yu, and B. Cle rckx are with the Department of Electrical and Electroni c Engine ering, Imperial College London, London SW7 2AZ, U.K. (e-mail: { g.sun24, x .lyu21, b .clerckx } @impe rial.ac .uk). Despite these theoretical advances, research on p ractical RSMA receiver de signs re mains relatively limited. Most ex- isting studies adop t SIC-based re ceiv ers and assum e Gau ssian codebo oks, while practical systems operate with finite c o nstel- lations and Bit-Interleaved Coded Mod u lation [9]. SIC is sen- siti ve to residu al interference an d pro ne to e r ror propagation , where erro rs in the earlier decoding stage can se verely d egrade subsequen t deco ding. Motiv ated by this gap, recent theo ret- ical work has pro posed SIC-free RSMA receiver designs, among which joint de mapping ( JD) achieves near-optimal perfor mance by directly evaluating bit-level Log-Likelihoo d Ratios ( L LRs) over a compo site constellation and d ecodes the comm o n an d private streams withou t serial interference cancellation [ 10], [11]. These studies demonstrate that SIC- free receivers, such as JD-based r eceiv ers can achie ve con- sistent perfo rmance gains over conventional SIC u nder harsh channel co nditions; howev er , their validation has been limited to theo retical an alysis a n d link-level simulations. On the other h and, RSMA pr ototypin g researc h using Software-Defined Radio (SDR) p latforms h av e experimen- tally demo n strated the adv antages of RSMA over SDMA and NOMA in u nicast, multicast, an d integrated sensing and commun ication scenario s [1 2]–[14]. Ne vertheless, all existing RSMA SDR prototyp es adopt SIC-based receivers, leaving the p ractical feasibility and pe r forman ce of SIC-free RSMA receivers un explored. T o the best of the author s’ knowledge, no experimen tal prototype has y et impleme n ted and evaluated a SIC-free RSMA receiv er . In this letter, we ad d ress this op en pr oblem by re a lizing a SIC-free RSMA receiver based on JD on a wideband multi- user MISO SDR pro to type. Using over-the-air measur ements, we comp are SIC and JD-b ased receivers in terms o f sum throug hput and coded b it er ror r ate (BER) under d ifferent channel co n ditions and spatial cor relations. The main con tr i- butions are as follows: 1) W e design and implemen t a SIC-fr e e RSMA recei ver based o n join t demapp in g on a practical wide b and SDR platform with finite constellations. 2) W e c onduct systematic over -the-air measurem ents to compare JD and SI C-b ased receivers under various Mod- ulation and Coding Scheme (MCS) settings and chan n el condition s. 3) Experimental r esults d emonstrate that the proposed JD- based receiv er ach iev es smooth er an d more reliable sum through put, preser ves priv ate-message de codability when th e common stream is under pre ssure, and reduces the d e c oding Sign al-to-Noise Ratio (SNR) th reshold. IEEE COMMUNICA TIONS LE TTERS 2 I I . J D - BA S E D R S M A S Y S T E M M O D E L W e fo llow the first SDR-based RSMA proto type ar c hitec- ture in [1 2]. W e co nsider a two-user MISO d ownlink, where the transmitter sends messages W 1 and W 2 to user-1 and user-2 respectively . At the transmitter, each message W k with k ∈ { 1 , 2 } is fir st divided by th e m essage splitter into a common componen t W c,k and a pri vate compo nent W p,k . The two commo n compone n ts a r e then comb ined into a single common message W c . After channel coding and mod u lation, W c produ ces a common stream. In parallel, each p riv a te compon ent W p,k is encod ed and modulated on its own, which forms the private stream s p,k . The three streams, s c , s p, 1 , and s p, 2 , are linea rly precoded to gen erate the OFDM frequen cy- domain tr ansmit sign al: x [ n ] = p c s c [ n ] + p 1 s p, 1 [ n ] + p 2 s p, 2 [ n ] , (1) where [ n ] indicates the symbol carried by the n -th sub-c a r rier , n ∈ { 0 , ..., N c − 1 } , p c , p 1 , and p 2 denote the linear precoder s assigned to the commo n stream and to the tw o pri vate streams. On th e receiver side, let X c denote the modu lation set of the common stream and X p,k the modulatio n set of u ser k ’ s priv ate stream, k ∈ { 1 , 2 } . After Cyclic Prefix (CP) rem oval and Discr ete Fourier Transform processing, the recei ved sign al at user - k is modeled as: y k [ n ] = h H k [ n ] x [ n ] + n k [ n ] = h H k [ n ] p c s c [ n ] + h H k [ n ] p 1 s p, 1 [ n ] + h H k [ n ] p 2 s p, 2 [ n ] + n k [ n ] , (2) where n k [ n ] re presents the ther mal n oise, whose variance is denoted by σ 2 and in practice it is estimated from pilots and residual error statistics. Before joint demapp ing, we define the scalar effecti ve channel co e fficient g on su bcarrier n as th e proje c tio n of the estimated MISO channel o n to a tran smit precoder, which captures the c o mbined amp litude scaling and p hase rotation at the receiver . Accor d ingly , the effecti ve gains seen by the common and pr ivate stream s at user k on subcarrier n are: g c,k,n = h H k [ n ] p c , g p,k,n = h H k [ n ] p k . (3) For each subcarrier n , the r eceiv er u tilizes these effectiv e gain s to form the composite co nstellation: S k,n = n g c,k,n x c + g p,k,n x p,k , x c ∈ X c , x p,k ∈ X p,k o . (4) User- k subseque n tly uses a join t demap per on each subcar- rier n to select a composite symbol pair ( s c [ n ] , s p,k [ n ]) ∈ X c × X p,k , which enables receiver k to d e c ode common and priv ate streams simu ltaneously . W ithin the SIC-free r e c ei ver , each user does not pe r form successive subtractio n of the com mon stream before decoding its priv ate stream. Instead, the bit vectors of both the common and private stream s ar e ob tained directly from the compo site co nstellation in a single joint d emappin g stage, where the contr ibutions of different streams are jo intly considered rather than successively ca n celed. This eliminates the hie r archical decod in g dep endency inher e nt to SIC-based receivers. At user - k , the joint demapper ou tp uts two sets o f bit vectors that are sent to: 1) the chann el decod er for W c , p roducin g an estimate of the co mmon m e ssage, c W c,k ; 2) the chan nel deco der f or W p,k , prod u cing an estimate of the p r iv ate message c W p,k . In the en d, c W c,k and c W p,k are comb ined to form user k ’ s estimate of its desired message , c W k . T o keep the analy sis par allel to the SIC-based prototyp e, we ad opt the same wideba n d coded- OFDM co n vention with polar code as in [1 2]. Specifically , to gen erate ea c h stream s ∈ { s c , s p, 1 , s p, 2 } , a single p olar cod ew ord is interleav ed, modulated and mapp ed across all N c OFDM subcar r iers within o n e fra m e. Under this setting, deep fades on some da ta- bearing subcarrier s may dom inate the d ecoding reliability . As a theoretical b aseline fo r tran smitter-side benchm arking, we consider a simple bottlenec k pro xy to provide a conservati ve wideband reference for MCS b e n chmark ing. W e d efine th e per-stream wideband spectral-efficiency prox y as: R s , min n ∈{ 0 ,...,N c − 1 } log 2 (1 + SINR s [ n ]) , (5) where SINR s [ n ] is th e per-subcarrier Signal-to-I n terferenc e- plus-Noise Ratio (SINR). For th e c o mmon stream, we further use SINR s c [ n ] = min k SINR c,k [ n ] to reflect that the commo n message m ust b e dec o ded by both users. Impor tan tly , this pr oxy serves pr im arily as a theoretical baseline. In practice, our SDR proto type supports a finite set of M CSs, and we evaluate system perform ance in bits per second (bps) to align with hard ware capab ilities and realistic deployment metr ics. T o bridge the discre te physical lay er parameters with the co ntinuou s system data rate, we co nsider the system ef fecti ve ban dwidth B ef f and a n MCS group M = { ( m c , r c ) , ( m 1 , r 1 ) , ( m 2 , r 2 ) } . Let R ( m, r ) , B ef f · m · r denote the nomina l wideban d data rate of a selected stream, where m = log 2 ( M ) is the numbe r of b its p er symbol an d r is the co de r ate. The expected JD sum throu ghput (in bps) is therefor e f ormulated as: T JD mcs ( P , M ) = R ( m c , r c ) Pr( ˆ W c, 1 = W c , ˆ W c, 2 = W c ) + 2 X k =1 R ( m k , r k ) Pr( ˆ W p,k = W p,k ) , (6) where P , { p c , p 1 , p 2 } . The comm on stream con tributes to the th rough put o n ly when bo th u sers de c ode it successfu lly , while each pr iv ate stream co n tributes accor ding to the decod - ing su c cess o f its intended user . Based on the expected sum thro ughp u t defined in ( 6), we ev alu ate the system p erforma n ce by exh a u sti vely scan ning a standard ized MCS sub set. For eac h MCS selection, the decodin g success prob abilities are obtain ed emp irically fr om measuremen ts. T his p rocedur e ensures a fair compa rison be- tween SIC and JD, with all resu lts grou n ded in measured decodin g o u tcomes r ather than theoretical capacity limits. I I I . J D - BA S E D R S M A P ROT OT Y P E D E S I G N A N D I M P L E M E N TA T I O N Our proto type is built u pon an SDR-ba sed M U-MISO platform a s shown in Fig. 1 . The tra nsmitter u ses one National Instrumen t ( N I ) Universal Software Radio Peripheral (USRP) IEEE COMMUNICA TIONS LE TTERS 3 Fig. 1. SIC-free RSMA pr ototype block dia gram. T ABLE I M C S G R I D I M P L E M E N T E D I N T H E P RO T O T Y P E . MCS Index Modulation m Code Rate r Data Rate 0 BPSK 1 / 2 6 Mbps 1 BPSK 3 / 4 9 Mbps 2 QPSK 1 / 2 12 Mbps 3 QPSK 3 / 4 18 Mbps 4 16QAM 1 / 2 24 Mbps 5 16QAM 3 / 4 36 Mbps driving two antennas to transmit the RSMA com mon and priv ate streams to ward the two recei ve antenn as connected to the other NI USRP . In the timing an d I /O inf rastructure part, a NI CD A-2 990 clock distributor distributes a comm on 10 MHz referen ce signal to all USRPs for time synchronization . Then , a NI CPS- 8910 PCIe switch bo x is applied to aggregate multiple PCIe ×4 link s f rom the USRPs into PCIe ×8 to the workstation. In the workstation part, the TX design fo llows the SIC-based RSMA proto type in [12]. It g e n erates rando m b it streams, applies p o lar en c o ding to fo rm th e comm on and p riv ate code- words, perfo rms p ower alloc a tion, and designs the precode r based on CSI feedback for u ser-1 and user-2. The r esulting symbols are th en go th r ough OFDM modulation with Inv erse Fast Fourier T ransform and CP insertion. Meanwhile, pilots and a pr eamble are adde d for sy n chron iz a tio n and c h annel estimation. For tran sm ission and th rough put evaluation, the system is configur ed with an ef fective bandwidth B ef f of 1 2 MHz a nd a subcarrier spacing o f 31 2.5 kHz. Based on the data rate definition R ( m, r ) from o ur system model, the correspon ding operation al data rates f o r the gen erated co mmon and private codewords are summarized in the p er-stream MCS grid in T ABLE I . Moving to the RX par t from Fig. 1 , the receiver first perfor ms frame syn c hronizatio n to remove timing offsets and then applies a Fast Fourier Transform to obtain per-subcarrier symbols in the freq uency do main. Th e u pper p art o f the receiver illustrates the ch annel estimation procedure , where pilot sym bols are fed into th e chan nel estimator to obtain the estimated CSI, wh ich is furthe r u sed in the feedback loop for precod e r de sig n . Utilizing the estimated subcarr ier-dependent CSI, th e re- Fig. 2. RSMA measure campai gn. User 1 is fixed, whil e user 2 i s sho wn at Case 4 a nd can mo ve from Case 1 to Case 6. ceiv er calculates the effective gains and con structs the com- posite constellation S k,n as defined in th e system mo d el. Each b aseband symb ol on sub carrier n is dema pped as a symbol from this sing le c o mposite alphab e t. The demapper outputs a b it vector of length b c + b p (with b c = log 2 |X c | and b p = log 2 |X p,k | ). A fixed b it ord er is used: the first b c bits are assigned to the comm on stream and th e last b p bits are assign ed to the private- k stream . The resulting bits are separated into common and priv a te bits. By collec tin g b its over all subca r riers, the bit-le vel code words fr om the common and p r iv ate streams are obtained an d for warded to the polar decoder . As for the measurem ents using our RSMA prototype, we implement two RSMA receiver ch ains under the same fram e structure. T o alig n the theoretical expected throu ghput (6) with our hardware measurem ents, we substitute the theoretical decodin g success probabilities Pr( · ) with the empirical packet delivery ratios o bserved in the SDR . W e e v aluate both SIC and JD chains using this un ified, measur ement-align ed empirical sum-thro ughpu t metric: T JD mcs , emp  P , M  = D JD s,c T  B ef f m c r c  + 2 X k =1 D JD s,k T  B ef f m k r k  . (7) where D JD s,c denotes the n u mber of run s in which the com mon stream is successfully decoded by both users, D JD s,k denotes the n umber o f runs in which the priv ate stream of user-k is successfully d ecoded, and T indicates the total number o f r uns. This equation tran slates the hardware d ecoding outc o mes into the continuo us time d omain (b ps), enab ling a dir e ct and fair compariso n o f the SIC-based and SIC-free RSMA receivers. I V . M E A S U R E M E N T C A M PA I G N A N D R E S U LT S The experimental tests we r e settled on the la b bench as shown in Fig. 2. A fu ll list of all parame te r s app lied to ou r measuremen ts is shown in T ABLE II . W e the n measur e the average chann el strength disparity α and sp a tial corre latio n ρ fro m Case 1 to Case 6 shown in Fig. 2. The relativ e ch annel strength disparity (in dB) between the two u sers is d efined as α = 10 log 10 ( k h 2 k 2 / k h 1 k 2 ) , wher e h 1 and h 2 denote the estimated chann el vectors for u ser 1 and user 2 . T o e valuate spatial correlation, we define ρ = 1 − | h H 1 h 2 | / ( k h 1 kk h 2 k ) . Unde r th is definitio n, the channels become more aligned as ρ ap proach es 0 and more orthogo nal as it approaches 1. The empirical results averaged over all r u ns for each case are listed in T ABLE III. IEEE COMMUNICA TIONS LE TTERS 4 T ABLE II P A R A M E T E R S U S E D I N T H E E X P E R I M E N T S Parame ter Notation V alue Center freq uenc y f c 2.484 GHz Tra nsmit po wer P t 16 dBm TX ante nna length 0.13 m Fraunhofer dist ance 0.28 m T otal ban dwidth 20 MHz Subcarrie rs T otal ( N c ) 64 Data 48 Pilot (FPS) 4 Guard band 12 CP le ngth 16 Effe cti ve bandwidth B ef f 12 MHz OFDM symbols pa yload 40 Experimenta l ru ns per case 75 T ABLE III E M P I R I C A L A V E R AG E S O F C H A N N E L S T R E N G T H D I S PA R I T Y α A N D S PA T I A L C O R R E L AT I O N ρ F O R T H E C A S E S I N F I G . 2 . Relati ve pathloss Spatial correla tion Cases α [dB] Ca ses ρ lo w 1 − 1 . 34 lo w 1 0.28 2 − 1 . 63 4 0.19 3 − 1 . 19 mid 2 0.52 high 4 − 11 . 74 5 0.49 5 − 11 . 87 high 3 0.81 6 − 11 . 52 6 0.79 BPSK 3/4 QPSK 1/2 QPSK 3/4 16QAM 1/2 16QAM 3/4 Private MCS 0 20 40 60 80 100 120 Sum Throughput (Mbps) Sum Throughput for Different MCS Combinations Common QPSK 1/2 - SIC Common QPSK 1/2 - Joint Common QPSK 3/4 - SIC Common QPSK 3/4 - Joint Common 16QAM 1/2 - SIC Common 16QAM 1/2 - Joint Common 16QAM 3/4 - SIC Common 16QAM 3/4 - Joint Fig. 3. Measured sum throughput versus priv ate-strea m MCS for Case 1 under SIC a nd JD. W e exhaustively search the MCS grid shown in T ABLE I. For each candidate MCS selection, decoding suc c ess prob a- bilities ar e obtain e d empirically fr om measurem ents, and the resulting su m thr oughp ut is comp uted acco rdingly . The sum throug hput per f ormance for Case 1 as a function of the M CS lev els is plo tted in Fig. 3, and th rough p ut de c o mposition into common an d two private contr ibutions for two receivers are plotted in Fig. 4. Fig. 3 shows that SIC achieves the high est peak thr oughp ut only when the co mmon stream uses a mod erate MCS, namely QPSK 3 /4, wh ereas JD remains more robust as the co mmon- stream MCS becom e s more agg ressi ve. Whe n the common stream u ses QPSK, the sum throug hput generally in c reases with the priv ate-stream MCS. T he b est measure d operating point is ob tain ed at common QPSK 3/4 and priv ate 16QAM 3/4, wh ere SIC achieves 78.69 Mbp s and JD ach ieves 65.31 Mbps. As the c ommon - stream MCS becom es mo re a ggressive, BPSK 3/4 QPSK 1/2 QPSK 3/4 16QAM 1/2 16QAM 3/4 Private MCS 0 10 20 30 40 50 60 70 80 90 Throughput (Mbps) Joint Decoding Breakdown Common QPSK 1/2 Common QPSK 3/4 Common 16QAM 1/2 Common 16QAM 3/4 Priv. UE1 Priv. UE2 BPSK 3/4 QPSK 1/2 QPSK 3/4 16QAM 1/2 16QAM 3/4 Private MCS 0 10 20 30 40 50 60 70 80 90 Throughput (Mbps) SIC Decoding Breakdown Common QPSK 1/2 Common QPSK 3/4 Common 16QAM 1/2 Common 16QAM 3/4 Priv. UE1 Priv. UE2 Fig. 4. Case 1 throughput decomposition: top, JD; bo ttom, SIC. howe ver , the ga p between SIC an d JD is increasingly driven by reliability , i.e., successful commo n-message decodin g, rath er than peak thro ughpu t. W ith com mon 16 QAM 1/2 , SIC peaks around pr i vate QPSK 1 /2 and then dro ps to 17.49 M bps at priv ate 16QAM 3/4. With common 16QAM 3/4, SIC co llapses to near ly zero th r oughp ut acr oss the priv ate-MCS sweep. JD remains n on-zero un d er co mmon 16 QAM 3/4, alth ough it also degrad es at hig h p r iv ate o rders beca u se the composite constellation becomes very dense. The detailed d ecompo sition in Fig. 4 explain this tren d. Unde r SIC, priv ate-stream per- forman ce depen ds stro ngly on successfu l common dec o ding and acc urate inter ference can cellation; he nce, fr agility in the common stream quickly rem oves both commo n and p riv ate contributions. In con trast, JD extracts d ecoding inf ormation directly from the compo site constellatio n , allowing priv ate- stream co n tributions to persist even when the comm on stream is near thr eshold. T o fur ther demonstrate the super iority of JD over SIC in resisting er ror prop agation, we present the four- quadra n t analysis in Fig. 5 for all six cases in T ABLE III. E ach quadra n t summarizes decoding success/fail outco mes for each user u nder a fixed MCS (common , priv ate = 1 6QAM 1 /2). For the SIC-based re c e i ver , th e distributions move rap id ly tow ard joint failure as ch annel difficulty in creases. In the hardest case, the m a ss con centrates in the qu adrant where both common and priv ate fail for both users, often above 9 0%. JD shows a different transition . In the e a sier cases, a large fraction r emains in the q uadran ts with pr ivate success, and user 1 exhibits more than 30% priv ate - only success in Case 1 to 3. Even un d er worse ch annel condition s, JD avoids th e near- total collapse seen with SIC and maintain s priv ate-message decodab ility w h en the common stream is stressed, e ffecti vely shifting outcomes from ’both-fail’ to ’priv ate- only success’. IEEE COMMUNICA TIONS LE TTERS 5 Fig. 5. Four -quadrant decod ing outcomes for SIC a nd JD a cross Cases 1– 6. Left: Case s 1–3. Right : Cases 4–6. 0 5 10 15 20 SNR (dB) 10 -5 10 -4 10 -3 10 -2 10 -1 10 0 Coded BER Coded BER vs SNR (Common: QPSK 3/4, Private: QPSK 1/2) SIC - Total (Common+Private) Joint - Total (Common+Private) SIC - Common only Joint - Common only SIC - Private only Joint - Private only Fig. 6. Coded BER v ersus demap per-i nput SNR for SIC and JD. Finally , we introd uce coded BER versus SNR. The BER considered here is cod ed BER, as it dire c tly d etermines the subsequen t p acket su ccess r a te and overall sum thro u ghpu t. The objective is to show the SNR advantage of JD over SIC under the same MCS g roup (Co m mon QPSK 3/4, priv ate QPSK 1 /2 ). Fig. 6 compares cod ed BER versus the demap per input SNR γ , and we sum marise the imp r ovement using th e threshold gain ∆ γ ( β ) = γ SIC ( β ) − γ JD ( β ) , which is the SNR reduction required b y JD to reach the same coded BER target β . For the com mon stream, JD pr ovides a modest gain, abou t 1.2 dB at 10 − 3 and about 1 dB at 10 − 5 . For the p riv ate streams, the gain is much larger, abou t 5. 2 d B at 10 − 3 and about 3 dB at 10 − 5 , which is consistent with SIC suffering error propaga tio n fro m the common stage. When both co mmon and priv ate streams must meet the target simu ltaneously , the overall gain is abo ut 1.7 dB at 1 0 − 3 and about 1 dB at 10 − 5 , mainly limited by the com m on-stream thresh old. Overall, the ∆ γ results demon stra te tha t JD strengthen s pr iv ate-stre a m robustness and p revents the sharp failures that o ccur un der serial SIC when the common stream is stressed. V . C O N C L U S I O N In this paper, we imp lemented an SDR-based SIC-free RSMA receiver pro to type based on jo in t demappin g. Altho ugh JD incurs h ig her co mputation a l complexity than serial SIC du e to joint bit-vector ev aluation, o ur measurem e n ts demon strate its p r actical feasib ility and im proved reliability acr oss a wide range of MCS selection s. By mitigatin g the e r ror prop a gation inherent in SI C, the JD- based recei ver provides a receiver-side enhancem ent that deliv ers smoother sum-th rough put behavior and reduced d ecoding thresholds without requiring tr a n smitter- side modifications. R E F E R E N C E S [1] Q. H. Spencer , A . L. Swindle hurst, and M. Haardt, “Zero-forcing methods for downlin k spatia l multiple xing in multiuser mimo channels, ” IEEE T rans. Si gnal Pr ocess. , vol. 52, no. 2 , pp. 4 61–471, 2004. [2] Y . Saito, Y . Kishiyama, A. Benjebbour , T . Nakamura , A. Li, and K. Higuchi, “No n-orthogona l multiple a ccess (noma) for cellular future radio acc ess, ” in Pro c. IE EE 77th V eh. T echnol. Conf . (VTC Spring) , 2013, pp. 1– 5. [3] T . Han and K. K obayashi , “ A ne w achie vabl e rate region for the interfe rence channel, ” IEEE T rans. Inf . Theory , vol. 27, no. 1, pp. 49–60, 1981. [4] H. Joudeh and B. Clerc kx, “Sum-rate maximiz ation for linearly precod ed do wnlink multiuser MISO systems with partial CSIT: A rate-splitt ing approac h, ” IEEE T rans. Commun. , vol. 64, no. 11, pp. 4847–4861, 2016. [5] Y . Mao, B. Clerckx, and V . O. K. Li, “Ra te-spli tting m ultiple access for down link communicati on systems: Bridging, generali zing, and out- performing SDMA and NOMA, ” EURASIP J. W ir eless Commun. Netw . , vol. 2018, no. 1, p. 133, 2018. [6] C. Hao, Y . Wu, and B. Clerckx, “Rat e analysis of two-rece i ve r MISO broadca st channe l with finite rate feedba ck: A rate-splitt ing approach, ” IEEE T rans. Co mmun. , vol. 63, no. 9, pp. 32 32–3246, 2015. [7] O. Dizdar , Y . Mao, and B. Clerckx, “Rate -splitti ng multiple access to mitigat e the curse of mobility in (massi ve) MIMO networks, ” IEEE T rans. Commun. , v ol. 69, no. 10, pp. 6765–6780, 2021. [8] Y . Mao, O. Dizdar , B. Clerckx , R. Schober , P . Popovski, and H. V . Poor , “Rate-splitt ing m ultipl e access: Fundamentals, survey , and future research trends, ” IE EE Commun. Survey s T uts. , vol. 24, no. 4, pp. 2073– 2126, 2022. [9] D. Mi, H. Chen, Z. Chu, P . Xiao, Y . W u, and C.-L. W ang, “Rate - splitti ng multipl e access with finite-alphabe t constell ations: Precoder optimiza tion and achi e v able rate performa nce, ” IEEE Tr ans. Green Commun. Netw . , vol. 8 , no. 4 , pp. 1293–1 307, 2024. [10] S. Z hang, B . Clerckx, D. V arga s, O. Haffend en, and A. Murphy , “Ra te- splitti ng multiple acce ss: Finite constell ations, recei ver design, and SIC- free impl ementat ion, ” IEEE T rans. Commun. , v ol. 72, no. 9, pp. 5319– 5333, 2024. [11] S. Zhang, B. Clerckx, and D. V arg as, “SIC-free rate-splitti ng multiple access: Const ellat ion-con strained optimizat ion and applicat ion to lar ge- scale systems, ” IEEE T rans. W irel ess Commun. , vol. 25, pp. 10 669– 10 683, 2026. [12] X. L yu, S. Aditya, J. Kim, and B. Clerckx, “Rate-splitt ing multiple access: The fi rst prototype and experimenta l valida tion of its superiority ov er SDMA and NOMA, ” IEEE T rans. W ireless Commun . , vol . 23, no. 8, pp. 9986– 10 000, 2024. [13] X. L yu, S. Aditya, and B. Clerc kx, “Rate-spli tting m ultipl e access for ov erload ed multi-g roup multic ast: A first experiment al study , ” IEEE T rans. Br oadcast. , vol. 71 , no. 1, pp . 30–41, 20 25. [14] X. L yu, S. Aditya, and B. Clerc kx, “Rate-spli tting m ultipl e access for inte grate d sensing and communicatio ns: A first experimenta l study , ” IEEE T rans. W irel ess Commun. , vol . 25, pp. 10 498–10 513, 2026.