Cancellation of Multiuser Interference due to Carrier Frequency Offsets in Uplink OFDMA

Transcription

1 Cacellatio of Multiuser Iterferece due to Carrier Frequecy Offsets i Upli OFDMA S. Maohar, V. Tiiya, D. Sreedhar, ad A. Chocaligam Departmet of ECE, Idia Istitute of Sciece, Bagalore 56001, INDIA Abstract I upli orthogoal frequecy divisio multiple access (OFDMA) systems, multiuser iterferece (MUI) occurs due to differet carrier frequecy offsets (CFO) of differet users at the receiver. I this paper, we preset a multistage liear parallel iterferece cacellatio (LPIC) approach to mitigate the effect of this MUI i upli OFDMA. The proposed scheme first performs CFO compesatio (i time domai) followed by DFT operatios (where is the umber of users) ad multistage LPIC o these DFT outputs. We preset a detailed performace ad complexity compariso of the proposed scheme with aother scheme proposed recetly by Huag ad Letaief which performs CFO compesatio ad iterferece cacellatio i frequecy domai usig Circular Covolutio (we refer to this scheme as HLCC scheme). We show that the HLCC scheme performs better tha our scheme whe the idividual CFO values are small, whereas our scheme performs better tha the HLCC scheme whe the CFO differeces are small (eve if the idividual CFO values are large). Also, our scheme has lesser complexity tha HLCC scheme whe the umber of subcarriers is large, which is typical i OFDMA systems. eywords Upli OFDMA, carrier frequecy offset, multiuser iterferece, iterferece cacellatio. I. INTRODUCTION Recetly, there has bee icreased research focus o orthogoal frequecy multiple access (OFDMA) o the upli 1]- 11]. I OFDM/OFDMA systems, carrier frequecy offset (CFO) betwee trasmit ad receive carrier frequecies results i loss of orthogoality amog differet subcarriers at the receiver. I upli OFDMA, correctio to oe user s CFO would misalig other users. Thus, other user CFO will result i multiuser iterferece (MUI) i upli OFDMA. There have bee few recet attempts that address the issue of MUI due to other user CFO i upli OFDMA 7]-11]. Iterferece cacellatio (IC) techiques ca be employed at the base statio receiver to mitigate the MUI effects 8]-11]. Recetly, i 9], Huag ad Letaief preseted a IC approach which performs CFO compesatio ad MUI cacellatio i frequecy domai usig circular covolutio. We refer to this scheme i 9] as Huag-Letaief Circular Covolutio (HLCC) scheme. The circular covolutio approach was proposed earlier by Choi et al i 6] as a alterative to the direct time-domai method of CFO compesatio. Huag ad Letaief refer the scheme i 6] as scheme ( stads for the first letters of the ames of the four authors of 6]). The scheme does ot perform MUI cacellatio. The HLCC scheme, o the other had, uses circular covolutio for both CFO compesatio (as i 6]) as well as MUI cacellatio. More recetly, i ], we proposed a miimum mea square error (MMSE) receiver for MUI cacellatio i upli This wor was supported i part by the Swarajayati Fellowship, Departmet of Sciece ad Techology, New Delhi, Govermet of Idia, uder Project Ref: No. 6/3/00-S.F. OFDMA. We derived a recursio to approach the MMSE solutio ad showed that this recursive MMSE solutio ecompasses the ad HLCC schemes as special cases. Structure-wise, a commo feature i 6], HLCC 9], ad MMSE ] schemes is that all these detectors/cacellers first perform a sigle DFT operatio o the received samples ad the resultig DFT output vector is further processed to achieve CFO compesatio ad MUI cacellatio usig circular covolutio. Here, we propose ad aalyze a alterate MUI cacellatio receiver structure which first performs CFO compesatio i time domai, followed by DFT operatios (where is the umber of users) ad multistage liear parallel iterferece cacellatio (LPIC) o these DFT outputs. We scale the estimated MUI by weights before cacellatio. We obtai closed-form expressios for the optimum weights that maximize the average sigal-to-iterferece ratio (SIR) at the output of the differet LPIC stages. Comparig the proposed weighted LPIC (WLPIC) scheme with the HLCC scheme i 9], we show that i) i terms of complexity, the proposed WLPIC scheme is less complex tha the HLCC scheme, particularly whe the umber of subcarriers is large (which is typical i OFDMA systems), ad ii) i terms of performace, while the bit error performace of the HLCC scheme is affected by the idividual CFO values of all the users, the performace of the WLPIC scheme is affected by the differece betwee the desired user s ad other users CFO values. Because of this, the HLCC scheme performs better tha the WLPIC scheme whe the idividual CFO values are small, whereas the WLPIC scheme performs better tha HLCC scheme whe the CFO differeces are small (eve if the idividual CFO values are large). II. UPLIN OFDMA SYSTEM MODEL We cosider a upli OFDMA system with users, where each user commuicates with a base statio through a idepedet multipath chael as show i Fig. 1. We assume that there are N subcarriers i each OFDM symbol ad oe subcarrier ca be allocated to oly oe user. The iformatio symbol for the ith user o the th subcarrier is, S i, where S i is the set of subcarriers (i) assiged to user i ad E ] =1. The, i=1 S i = deoted by (i) {0, 1,...,N 1} ad S i Sj = φ, fori j. The legth of the guard iterval added is N g samples ad is assumed to be loger tha the maximum chael delay spread. After IDFT processig ad guard iterval isertio at the trasmitter, the time domai sequece of the ith user, x (i), is give by x (i) = 1 N /06/$0.00 (c) 006 IEEE (i) S i e jπ N, Ng N 1. (1)

2 Iput 1 Iput Fig. 1. (1) IDFT IDFT Add Guard Iterval Add Guard Iterval x (1) x D/A D/A Upli OFDMA system model. RF RF h (1) h RF A/D r Basebad Processig The ith user s sigal, after passig through the chael, is give by s (i) = x (i) h (i) () where deotes liear covolutio ad h (i) is the ith user s chael impulse respose. It is assumed that h (i) is o-zero oly for =0,...,L 1, where L is the maximum chael delay spread, ad that all users chaels are statistically idepedet. We assume that h (i) s are i.i.d. complex Gaussia with zero mea ad E ( h (i),i ) ] = E ( h (i),q ) ] =1/L, where h (i),i ad h(i),q are the real ad imagiary parts of h(i). The chael coefficiet i frequecy domai H (i) is give by L 1 H (i) = h (i) e jπ N, (3) =0 H (i) ad E ] =. The received basebad sigal after coarse carrier frequecy tracig (leavig some residual carrier frequecy offset) is give by r = s (i) e jπɛ i N + z, N g N 1, (4) i=1 where ɛ i,i =1,..., deotes the ith user s CFO ormalized by the subcarrier spacig, ad z is the AWGN with zero mea ad variace σ. We assume that all users are time sychroized ad that ɛ i,i=1,,are ow at the receiver. Figure shows the receiver basebad processig icludig i) CFO compesatio i time domai ad guard time removal, ii) DFT operatios (oe for each user), ad iii) liear parallel iterferece cacellatio (LPIC) i multiple stages. Note that the CFO compesatio is carried out i time domai by multiplyig r with e jπɛ i N, i =1,, (this method of CFO compesatio is referred to as the direct method i 6]). The received sigal after CFO compesatio ad guard time removal for the ith user is give by y (i) = r e jπɛ i N, 0 N 1, (5) which forms the iput to the ith DFT bloc. The output of the DFT bloc for the ith user o the th subcarrier is give by Y (i) where = H (i) (i) + l=1 l =i q q S l H q (l) q (l) + Z (i), (6) 4649 r j π ε1 e j π ε e /N /N Remove Guard Iterval Remove Guard Iterval DFT DFT Select Subcarriers for user 1 Select Subcarriers for user Output for user 1 Output for user MUI Estimatio ad Cacellatio (Stage ) MUI Estimatio ad Cacellatio (Stage 3) Decisio Decisio Fig.. Receiver basebad processig CFO compesatio (i time domai) ad iterferece cacellatio. q = si π( q + δ ( li) N si π ( q + δ N li) exp j(1 1 ) N )π( q+δ li), ad δ li is the differece betwee the ith user ad lth user CFO values, give by δ li = ɛ l ɛ i. (8) The chael coefficiet H (i) compoet Z (i) is give by v v (1) (7) is give by (3) ad the oise N 1 Z (i) = z (i) e jπ(+ɛ i) N. (9) =0 Note that the d term i (6) represets the MUI preset at the DFT output. I the case of sigle user detectio (), the DFT outputs, Y (i) s, ca be directly used to mae the symbol decisio. Additioal processig (e.g., iterferece cacellatio) may be performed o Y (i) s to mitigate the MUI effects. III. PROPOSED WEIGHTED LINEAR PIC SCHEME The proposed multistage weighted liear PIC (WLPIC) scheme is explaied as follows. Let m deote the stage idex. We tae the DFT outputs, Y (i) s, i (6) as the first stage (m =1) outputs of the receiver, i.e., Y (i),(1) = Y (i). I the case of, the symbol decisios are made directly from Y (i),(1) s. Parallel iterferece cacellatio is performed i the subsequet stages. I a give PIC stage m, m > 1, a estimate of the MUI is made based o the soft values of the previous stage outputs. These MUI estimates are scaled by weights ad cacelled from the DFT outputs, Y (i),(1). The iterferece cacelled output of the ith user o the th subcarrier i the mth stage, Y (i),(m), m>1, ca be writte as Y (i),(m) = Y (i),(1) w(i),(m) Y (l) q,(m 1) () where Y (i),(1) l=1 l =i give by (7). It is oted that l=1 l =i the MUI estimate, ad w (i) q q S l } {{ } MUI estimate is the 1st stage output give by (6) ad ρ(i),(l) q is q S l q Y (l) q,(m 1) is,(m) is the weight with which this MUI estimate is scaled ad cacelled. It is oted that the becomes a special case of the proposed WLPIC scheme for

3 w (i),(m) =0, i,, m. Also, we call the WLPIC scheme with uity weights o all subcarriers (i.e., w (i),(m) =1, i,, m)as covetioal LPIC (CLPIC) scheme. I the CLPIC scheme, the operatios eeded for choice of optimum weights ad scalig of MUI estimates with these weights are avoided (because of uity weights). However, performace better tha that of the CLPIC ca be achieved by usig optimum weights. We propose to obtai the optimum weights w (i),opt,(m) by maximizig the average SIR at the mth stage output. I a ucoded system, the symbol decisio for the ith user o the th subcarrier at the output of the mth stage ca be made based o the output Y (i),(m). For example, the symbol decisio at the mth stage output for the case of BPS modulatio ca be obtaied as ( )) (i),(m) (Re = sg H (i) Y (i),(m), (11) where H (i) deotes the cojugate of H (i). For the case of M-QAM/M-PS modulatio, symbol decisio ca be made usig the miimum Euclidea distace rule. I a coded system, the Y (i),(m) s are fed to the decoder. IV. AVERAGE SIR AT ND AND 3RD STAGE OUTPUTS I 11], we have derived expressios for the average SIR at the output of the d ad 3rd stages of the proposed WLPIC scheme. Also, we used these average SIR expressios to obtai closed-form expressios for the optimum weights w (i),opt,(m). Refer 11] for closed-form expressios for average SIR at d ad 3rd stage outputs ad the optimum weights. I Fig. 3, we plot the average SIR at the output of the d stage as a fuctio of weights w (i),() obtaied through both aalysis (Eq. (1) i 11]) as well as simulatios. The followig system parameters are cosidered: N =3, =4, ɛ 1,ɛ,ɛ 3,ɛ 4 ]= 0.1, 0.3, 0.5, 0.15], ad SNR=5 db. The chael model used is a oe sample spaced two-ray equalgai Rayleigh fadig model. Two types of subcarrier allocatio, amely, a) iterleaved allocatio ad b) bloc allocatio are cosidered. I bloc allocatio, a cosecutive bloc of subcarriers are alloted to oe user, the ext bloc to aother user, ad so o. I iterleaved allocatio, the subcarriers of each user are uiformly iterleaved with the subcarriers assiged to the other users. The followig observatios ca be made from Fig. 3. For the cosidered chael model ad system parameters, bloc allocatio results i a higher output SIR tha iterleaved allocatio. Also, the maximum average output SIR occurs at a optimum weight (maximum SIR of about 15 db at w (i),() 0.7 for iterleaved allocatio ad a maximum SIR of about 1 db at w (i),() 0.6 for bloc allocatio). Closed-form expressios for these optimum weights for d stage ad 3rd stage, respectively, are give i Eqs. (3) ad (33) i 11]. V. RESULTS AND DISCUSSION I this sectio, we preset the umerical ad simulatio results of the average SIR ad BER performace of the proposed WLPIC scheme ad compare with those of other detec Average SIR at d Stage Output (db) Bloc Allocatio N = 3, = 4, SNR = 5 db cfo] = 0.1, 0.3, 0.5, 0.15] Iterleaved Allocatio Simulatio Aalysis Weight, W_{1,()}^{(1)} Fig. 3. Average SIR of the 1st user at the d stage output of the proposed WLPIC scheme as a fuctio of the weight o the 1st subcarrier, w (1) 1,(). N =3, =4, 16-QAM, ɛ 1,ɛ,ɛ 3,ɛ 4 ]= 0.1, 0.3, 0.5, 0.15], SNR = 5 db. Iterleaved ad bloc allocatio. Aalysis vs simulatio. Average SIR (db) N = 3, = 4, No oise cfo]= 0.1, 0.3, 0.5, 0.15], ray chael Iterleaved allocatio CLPIC Stage WLPIC Stage CLPIC Stage 3 WLPIC Stage Subcarrier Idex, Fig. 4. Average SIR as a fuctio of subcarrier idex,, for differet detectors. N =3, =4, ɛ 1,ɛ,ɛ 3,ɛ 4 ]= 0.1, 0.3, 0.5, 0.15]. No oise (σ =0). Iterleaved allocatio. Aalysis. tors i the recet literature. The chael model used throughout this sectio is a oe sample spaced two-ray equal-gai Rayleigh fadig model. I Fig. 4, we plot the aalytically computed average SIR as a fuctio of the subcarrier idex =1,,,N uder o oise coditio (i.e., σ =0)for a), b) d ad 3rd stages of the CLPIC scheme (where =1, i, ), ad c) d ad 3rd stages of the WLPIC scheme, for a upli OFDMA system with N =3 subcarriers, =4users, iterleaved allocatio, ad CFOs of the differet users ɛ 1,ɛ,ɛ 3,ɛ 4 ]= 0.1, 0.3, 0.5, 0.15]. w (i),() = w(i),(3) From Fig. 4, it ca be see that the gives the least SIRs i all subcarriers sice o iterferece cacellatio is performed. Whe iterferece cacellatio is performed usig CLPIC scheme (where uity weights are used), the d stage output SIR improves sigificatly compared to that of. The CLPIC 3rd stage output SIR improves further compared to the CLPIC d stage output SIR. The WLPIC scheme (where optimized weights are used) performs sigificatly better tha both as well as CLPIC. For example, the 3rd stage of the WLPIC results i a average SIR of about 3 db o all the subcarriers which is sigificatly larger tha those of

4 0 1 BPS, N = 3 subcarriers, = 4 users cfo] = 0.1, 0.3, 0.5, 0.15], ray chael WLPIC,S WLPIC,S3 CLPIC,S CLPIC,S3 No MUI Bit Error Rate Average SIR (db) 5 0 N = 64 subcarriers, = 4 users, No oise cfo] = 0.1, 0.1, 0.05,0.05], ray chael Fig. 5. Bit error rate performace of the proposed WLPIC scheme for BPS. N =3, =4, ɛ 1,ɛ,ɛ 3,ɛ 4 ]= 0.1, 0.3, 0.5, 0.15]. Iterleaved allocatio. Simulatio. the other detectors. Thus the performace beefit of usig the optimized weights i WLPIC istead of uity weights i CLPIC or zero weights i is clearly evidet i Fig. 4. For the same set of parameters i Fig. 4, we plot the simulated BER performace of, CLPIC (d ad 3rd stages) ad WLPIC (d ad 3rd stages) i Fig. 5 for BPS. The sigle user performace (o MUI) is also show for compariso purposes. From Fig. 5, it ca be see that the proposed WLPIC scheme results i sigificatly better BER performace tha both the as well as the CLPIC scheme. The 3rd stage of the WLPIC scheme is foud to approach the sigle user (o MUI) performace. We have observed similar SIR ad BER improvemet for the case of bloc allocatio as well as 16-QAM. A. Compariso with HLCC ad Schemes I this subsectio, we preset a compariso of the performace ad complexity of the proposed WLPIC scheme with other detectors reported i the recet literature, amely, a) the HLCC scheme i 9], b) scheme i 6], ad c). It is oted that while the proposed WLPIC scheme ad the HLCC scheme are essetially iterferece cacellers, the ad schemes are detectors without iterferece cacellatio. Aother iterestig observatio is that while the ad HLCC schemes implemet CFO compesatio i frequecy domai usig circular covolutio, the ad WLPIC schemes implemet CFO compesatio usig the direct time-domai method. Because of this, as we will see ext, a HLCC versus WLPIC compariso shows similar comparative behaviour as a versus compariso, i terms of performace ad complexity. 1) SIR ad BER Compariso: We ote that, sice the CFO compesatio is doe usig circular covolutio i HLCC scheme, the performace of the HLCC scheme is affected by the ( idividual CFO values of all the users, ) ɛ i,i=1,,, see Eqs. (), (15) ad () i 9]. Whereas i the proposed WLPIC scheme, CFO compesatio is doe i time domai ad hece the performace of the WLPIC scheme is affected by the differeces betwee the desired user s ad other users CFO values, δ il = ɛ i ɛ l,i,l=1,,,, i l ( see Eqs. (6),(7),(8) ). Because of this, the HLCC Subcarrier Idex, Fig. 6. Compariso of the SIR performace of the proposed WLPIC scheme with HLCC ad schemes. N = 64, = 4, ɛ 1,ɛ,ɛ 3,ɛ 4 ]=0.1, 0.1, 0.05, 0.05]. No oise (σ =0). Iterleaved allocatio. Aalysis. scheme performs better tha the WLPIC scheme whe the idividual CFO values are small, whereas the WLPIC scheme performs better tha the HLCC scheme whe the CFO differeces are small. We illustrate both these cases i Figs. 6 through 9, by cosiderig two cases of CFO values, amely, CFO-1 = ɛ 1,ɛ,ɛ 3,ɛ 4 ]=0.1, 0.1, 0.05, 0.05], ad CFO- = ɛ 1,ɛ,ɛ 3,ɛ 4 ]=0.15, 0.1, 0.16, 0.08]. It is oted that the CFO-1 values i the above are the same oes used for geeratig the performace plots i 9]. As ca be see, these CFO-1 values correspod to the case where the idividual CFO values are smaller tha the differece betwee the CFO values ( e.g., 0.1 < (0.1 ( 0.05)) = 0.15 ).Othe other had, CFO- values correspod to the case where the CFO differeces are smaller tha the idividual CFO values ( e.g., 0.15 > ( ) = 0.03 ). I Figs. 6 ad 7, we preset a compariso of the average SIR performace of the various detectors for N =64, =4, iterleaved allocatio, ad o oise. Figure 6 is for CFO-1 ad Fig. 7 is for CFO-. From Fig. 6, it ca be see that both the HLCC ad WLPIC schemes give sigificatly higher SIR tha the ad schemes. Also, HLCC scheme results i higher SIR tha WLPIC scheme i this case sice the idividual CFO values are smaller tha the CFO differeces. I Fig. 7, o the other had, we see that the WLPIC scheme offers sigificatly higher SIR tha the HLCC scheme ( > 45 db SIR for WLPIC vs 5 db SIR for HLCC ). As poited out earlier, this is because, though the idividual CFO values are large, the CFO differeces are smaller for CFO-. A similar performace behaviour i terms of simulated BER for 16-QAM ca be observed i Fig. 8 (for CFO-1) ad Fig. 9 (for CFO-). For similar reasos of CFO implemetatio i frequecy domai versus time domai, it ca also be observed that performs better tha for CFO-1 whereas performs better tha for CFO-. Such scearios (lie CFO-, where performs better tha ) are ot discussed i 6] ad 9]. Coded FER Performace: We also carried out a compariso study of the various detectors i terms of coded frame error rate (FER) performace. We cosidered a rate-1/ covolutioal code with costrait legth 5. The system parame-

5 QAM, N = 64 subcarriers, = 4 users cfo] = 0.1, 0.1, 0.05, 0.05], ray chael 45 1 Average SIR (db) N = 64 subcarriers, = 4 users, No oise cfo] = 0.15,0.1,0.16,0.08], ray chael Bit Error Rate 3 No MUI Subcarrier Idex, Fig. 7. Compariso of the SIR performace of the proposed WLPIC scheme with HLCC ad schemes. N = 64, = 4, ɛ 1,ɛ,ɛ 3,ɛ 4 ]=0.15, 0.1, 0.16, 0.08]. No oise (σ = 0). Iterleaved allocatio. Aalysis. ters cosidered iclude =4users, N =64subcarriers, ɛ 1,ɛ,ɛ 3,ɛ 4 ]=0., 0., 0.18, 0.15], iterleaved allocatio, 4-QAM, ad -ray Rayleigh fadig chael. As i 9], each frame cosists of OFDM symbols, ad it is assumed that the chaels do ot vary withi oe frame but vary from frame to frame. I each frame, a 8 40 bloc bit iterleaver is employed. Figure shows the simulated coded FER performace for various detectors. For this system sceario, WLPIC scheme performs better tha HLCC scheme. Agai, this is because, i this case, the CFO differeces are much smaller tha the idividual CFO values. Liewise, performs better tha scheme i this sceario. ) Complexity Compariso: I additio to the above SIR ad BER/FER performace compariso, we carried out a complexity compariso amog the differet detectors as well. The complexities of various detectors i terms of umber of complex multiplicatios required are listed i Table I. The complexities of ad schemes are same as those give i 6]. Compared to scheme, HLCC has a additioal complexity of N + N / per cacellatio stage ( as per Eqs. (18),(19) i 9] ). Liewise, compared to scheme, WLPIC has a additioal complexity of N N / per cacellatio stage ( as per Eq. () ). The complexity compariso betwee HLCC ad WLPIC schemes as a fuctio of umber of subcarriers, N, for =16users ad m =, 3 (d, 3rd stages) is show i Fig. 11. It ca be see that for a give, HLCC is less complex for small N, whereas WLPIC has lesser complexity tha HLCC for large N, which is typical i OFDMA systems. For example, for N = 4, =16ad m =, HLCC has a complexity of 11,84,768, whereas WLPIC has a lesser complexity of,55,3. It is further oted that complexity reductio techiques similar to those give i 9] for HLCC scheme (e.g., by way of igorig wea subcarriers or other user subcarriers far-off from desired user s subcarriers) ca be doe for the WLPIC scheme as well. I additio to the performace ad complexity compariso preseted i the above, effect of imperfect CFO estimatio ad received power imbalace, as studied for HLCC Fig. 8. Bit error rate as a fuctio of average SNR for differet detectors for 16-QAM. =4, N =64, ɛ 1,ɛ,ɛ 3,ɛ 4 ]=0.1, 0.1, 0.05, 0.05]. Iterleaved allocatio. Simulatio. Bit Error Rate 0 1 No MUI 16 QAM, N = 64 subcarriers, = 4 users cfo] = 0.15, 0.1, 0.16, 0.08], ray chael Fig. 9. Bit error rate as a fuctio of average SNR for differet detectors for 16-QAM. =4, N =3, ɛ 1,ɛ,ɛ 3,ɛ 4 ]=0.15, 0.1, 0.16, 0.08]. Iterleaved allocatio. Simulatio. scheme i 9], ca be ivestigated for the WLPIC scheme also. These effects o WLPIC are expected to be o the same lies as observed for HLCC i 9]. VI. CONCLUSIONS We proposed a iterferece cacellatio scheme for MUI mitigatio i upli OFDMA. The proposed scheme performed CFO compesatio i time domai, followed by DFT operatios (o a per-user basis) ad multistage liear parallel iterferece cacellatio o these DFT outputs. Estimates of the MUI for cacellatio were obtaied usig soft values of the outputs from the previous stages ad the MUI estimate were scaled by weights before cacellatio. The proposed scheme was show to effectively cacel the MUI caused by the other user CFOs. We showed that the scheme proposed by Huag ad Letaief (HLCC scheme) performs better tha our scheme whe the idividual CFO values are small, whereas our scheme performs better tha the HLCC scheme whe the CFO differeces are small (eve if the idividual CFO values are large). Also, our scheme has lesser complexity tha HLCC scheme whe the umber of subcarriers is large, which is typical i OFDMA systems. 465

6 REFERENCES Coded Frame Error Rate 0 1 N = 64 subcarriers, = 4 users, cfo] = 0., 0., 0.18, 0.15] Iterleaved allocatio, 4 QAM, ray chael Rate 1/ Cov. code, Costrait legth = Fig.. Coded FER performace compariso amog differet detectors. =4, N =64, ɛ 1,ɛ,ɛ 3,ɛ 4 ]=0., 0., 0.18, 0.15]. Iterleaved allocatio, 4-QAM, rate-1/ covolutioal code, costrait legth = 5. Simulatio. WLPIC performs better tha HLCC. Detector HLCC WLPIC Complexity N log N + N N log N + N +(m 1) N log N N log N +(m 1) N N ] N + N log 3 ( 1)N ] ] log 3 ( 1)N ] N N 1]. im, Y. Ha, ad S-L. im, Joit subcarrier ad power allocatio i upli OFDMA systems, IEEE Comm. Letters, vol. 9, o. 6, pp , Jue 005. ] Z. R. Cao, U. Tureli, ad Y-D. Yao, Determiistic multiuser carrier frequecy offset estimatio for iterleaved OFDMA upli, IEEE Tras. o Commu., vol. 5, o. 9, pp , ] H. Wag ad B. Che, Asymptotic distributios ad pea power aalysis for upli OFDMA, Proc. IEEE ICASSP 004, May ] M. O. Pu, C.-C. J. Juo, ad M. Morelli, Joit sychroizatio ad chael estimatio i upli OFDMA systems, Proc. IEEE ICASSP 005, March ] A. M. Toello, N. Laureti, ad S. Pupoli, Aalysis of the upli of a asychroous multiuser DMT OFDMA system impaired by time offsets, frequecy offsets, ad multipath fadig, Proc. IEEE VTC 000 (Fall), vol. 3, pp , October ] J. Choi, C. Lee, H. W. Jug, ad Y. H. Lee, Carrier frequecy offset compesatio for upli of OFDM-FDMA systems, IEEE Commu. Letters, vol. 4, o. 1, pp , December ] Z. Cao, U. Tureh, ad Y. D. Yao, Aalysis of two receiver schemes for iterleaved OFDMA upli sigals, 36th Asilomar Cof. o Sigals, Systems ad Computers, vol., pp , November 00. 8] R. Fatacci, D. Marabissi, ad S. Papii, Multiuser iterferece cacellatio receivers for OFDMA upli commuicatios with carrier frequecy offset, Proc. IEEE GLOBECOM 04, pp , ] D. Huag ad. B. Letaief, A iterferece cacellatio scheme for carrier frequecy offsets correctio i OFDMA systems, IEEE Tras. o Commu., vol. 53, o. 7, pp , July 005. ] D. Sreedhar ad A. Chocaligam, MMSE receiver for multiuser iterferece cacellatio i upli OFDMA, Proc. IEEE VTC 006 (Sprig), Melboure, May ] S. Maohar, V. Tiiya, D. Sreedhar, ad A. Chocaligam, A multiuser iterferece cacellatio scheme for upli OFDMA, Proc. IEEE WCNC 006, April 006. TABLE I Complexity compariso amog differet detectors. 7 Complexity (i # complex multiplicatios) = 16 HLCC, m = WLPIC, m = HLCC, m = 3 WLPIC, m = Number of subcarriers, N Fig. 11. Complexity compariso of the proposed WLPIC scheme with HLCC scheme. =16, m =,