Channel estimation based on divergence minimization for OFDM systems with cochannel

Transcription

1 Aalborg Universitet Channel estimation based on divergence minimization for OFDM systems with cochannel interference Manchón, Carles Navarro; Fleury, Bernard Henri; Kirkelund, Gunvor Elisabeth; Mogensen, Preben Elgaard; Deneire, Luc; Sørensen, Troels Bundgaard; Rom, Christian Published in: Proceedings of the IEEE International Conference on Communications ICC 2009 DOI link to ublication from Publisher: /ICC Publication date: 2009 Document Version Publisher's PDF, also known as Version of record Link to ublication from Aalborg University Citation for ublished version APA: Manchón, C. N., Fleury, B. H., Kirkelund, G. E., Mogensen, P., Deneire, L., Sørensen, T. B., & Rom, C Channel estimation based on divergence minimization for OFDM systems with co-channel interference. In Proceedings of the IEEE International Conference on Communications ICC IEEE. DOI: /ICC General rights Coyright and moral rights for the ublications made accessible in the ublic ortal are retained by the authors and/or other coyright owners and it is a condition of accessing ublications that users recognise and abide by the legal requirements associated with these rights.? Users may download and rint one coy of any ublication from the ublic ortal for the urose of rivate study or research.? You may not further distribute the material or use it for any rofit-making activity or commercial gain? You may freely distribute the URL identifying the ublication in the ublic ortal? Take down olicy If you believe that this document breaches coyright lease contact us at vbn@aub.aau.dk roviding details, and we will remove access to the work immediately and investigate your claim. Downloaded from vbn.aau.dk on: Setember 09, 2018

2 This full text aer was eer reviewed at the direction of IEEE Communications Society subject matter exerts for ublication in the IEEE ICC 2009 roceedings Channel Estimation Based on Divergence Minimization for OFDM Sytems with Co-Channel Interference Carles Navarro Manchón, Bernard Fleury, Gunvor E. Kirkelund, Preben Mogensen, Luc Deneire, Troels B. Sørensen and Christian Rom Deartment of Electronic Systems, Aalborg University Niels Jernes Vej 12, 9220 Aalborg East Forschungszentrum Telekommunikation Wien FTW, Vienna, Austria Université de Nice, Sohia Antiolis Centre National de la Recherche Scientifique I3S, UMR 6070, France Infineon Technologies Denmark A/S Alfred Nobels Vej 25, DK-9220 Aalborg, Denmark Abstract In this aer, we resent a novel aroach for ilotaided channel estimation in OFDM systems with synchronous co-channel interference. The estimator is derived based on the Kullback-Leibler divergence minimization framework. The obtained solution iteratively udates both the desired user s and the interferer s channels, using a combination of linear minimum mean squared-error LMMSE filtering and interference cancellation, avoiding the comlex matrix inversions involved in the full LMMSE channel estimation aroach. Estimation of the noise variance is also included in the iterative algorithm, accounting for Gaussian noise and residual interference after each iteration. The estimates of both channels are used at the equalizer to reject the interfering signal, thus mitigating the degradation due to co-channel interference. Simulation results show that the receiver using the roosed estimator erforms as good as the one emloying the full LMMSE estimator and very closely to a receiver having erfect knowledge of the channel coefficients. I. INTRODUCTION Orthogonal Frequency Division Multilexing OFDM has become the selected transmission technique for several recent wireless standards, such as the IEEE standard for local and metroolitan area networks better known as WiMAX [1], or the 3GPP UTRA Long Term Evolution LTE [2]. Its ability to coe with time-disersive channels while allowing for receivers with low comlexity, its ability to easily integrate multile antenna techniques and its flexibility in terms of bandwidth usage and resource allocation are some of the advantages that have motivated its selection. In OFDM, the transmission bandwidth is divided into multile narrowband subcarriers. By the addition of a roer cyclic refix CP, these subcarriers become fully orthogonal and exerience frequency flat fading conditions in time-invariant channels [3]. This allows for simle equalization of the signal at the receiver, while keeing a high sectral efficiency due to the use of orthogonal overlaing subcarriers. In OFDM systems with frequency re-use, however, the signal transmitted from other cells may create co-channel interference which, if not correctly treated, can induce a severe degradation of the receiver erformance, esecially at the cell edge. Much work has been done in interference cancellation techniques for OFDM, as in [4] [6]. These methods, however, assume erfect knowledge of the channel at the receiver. In [7], a minimum mean-squared error interference rejection combiner MMSE-IRC for OFDM receivers with multile antennas is roosed. The combiner arameters are estimated using a discrete-fourier-transform-based robust MMSE instantaneous correlation estimator, which is therefore sensitive to the leakage effect [8] when the channel delays are not erfectly aligned with the receiver samling grid. In this work, we roose a ilot-aided channel estimator for OFDM systems with severe synchronous co-channel interference in both the data and ilot subcarriers. The estimator is derived by alying the Kullback-Leibler KL divergence minimization DM aroach, which was resented in [9] for multiuser detection in a code-division multile access system. Our roosed scheme is able to estimate the desired user s and the interferer s channels based on merely the signal received at ilot subcarriers. The estimates are then used in a MMSE-IRC combiner, effectively mitigating the effect of the interference. A similar roblem was studied in [10]. The solution roosed there, however, requires a reamble in which no interference is resent at the ilot subcarriers. Our estimator, on the contrary, can effectively searate and estimate both channels when the ilot signals of the desired user and the interferer overla in frequency for every OFDM symbol. The remainder of the aer is organized as follows. The signal model for our considered system is resented in Section II. In Section III, the DM framework is briefly introduced, and the channel estimator is derived. The erformance of the estimator is assessed by means of Monte-Carlo simulations in Section IV. Finally, some concluding remarks are given in /09/$ IEEE Authorized licensed use limited to: Aalborg Universitetsbibliotek. Downloaded on January 11, 2010 at 09:57 from IEEE Xlore. Restrictions aly.

3 This full text aer was eer reviewed at the direction of IEEE Communications Society subject matter exerts for ublication in the IEEE ICC 2009 roceedings b k 1 b k 2 Coding Coding c k 1 c k 2 Pilot generation x,1 k QAM Modulation x d,1k Pilot generation x,2 k QAM Modulation x d,2k Multilexing Multilexing x 1k IFFT + CP Tx 1 Insertion x 2k IFFT + CP Tx 2 Insertion Rx 1 Rx 2 r 1k FFT CP remov FFT CP remov r k 2 r,1k Channel Demux Estimation h 1,mk r k d,1 MMSE-IRC x d,1k Equalization Detection r d,2k h 2,mk Demux Channel Estimation r,2k Fig. 2. Block diagram of the receiver c 1k Decoding b 1k Fig. 1. Block diagram of the transmitters. Section V. The following notation will be used throughout the aer. Vectors are reresented by boldface lowercase letters, while matrices are denoted as boldface uercase letters; T and H denote resectively the transose and conjugate transose of a vector; tr{ } denotes the trace oeration, and diag{x} reresents a diagonal matrix with the elements of vector x; x y denotes direct roortionality, i.e., x = αy, and x e y denotes exonential roortionality, i.e., ex[x] = ex[β + y], for arbitrary constants α and β; finally, E qx {fx} reresents the exectation of the function fx with resect to the robability distribution q x x of x. II. SIGNAL MODEL We consider an OFDM system with single transmit antenna and one interferer, as deicted in Fig. 1. The first transmitter reresents the user of interest, while the second transmitter reresents a synchronized interferer transmitting in the same time-frequency resources. The information bits b m k, m = 1, 2, k =0,...,N b 1 are coded, yielding a stream of coded bits c m k, k =0,...,N c 1. These are modulated onto a set of QAM symbols x d,m k, k =0,...,N d 1 to be maed onto an OFDM block. The number of subcarriers used for data transmission in an OFDM block is N d = N c /M, N c being the number of coded bits transmitted in one OFDM block and M being the modulation order. The data symbols are then multilexed with a sequence of ilot symbols x,m k, k =0,...,N 1, N being the number of ilot subcarriers er block. We assume that ilot symbols are allocated to the same subcarriers at both transmitters. The resulting sequence of symbols x m k, k =0,...,N u 1 is then maed to the N u = N d + N active subcarriers of the OFDM system and transmitted through the wireless channel after insertion of a cyclic refix CP. We assume in this work that the CP is long enough to coe with the time disersion in both the desired and interfering channels. The structure of the receiver is shown in Fig. 2. We assume a receiver with two antennas. The extension to a higher number of antennas is straightforward. After FFT and CP removal, the received signal at the k th subcarrier of the n th antenna ort is given by r n k =x 1 kh n1 k+x 2 kh n2 k+w n k, 1 where w n k is additive white Gaussian noise AWGN with variance σw 2 and h nm k reresents the frequency-domain channel gain from transmitter m to receive antenna n at the k th subcarrier. In 1, we assume that the channel resonse is static during one OFDM block. Hence, full orthogonality between subcarriers is achieved. The received signal at antenna ort n for all subcarriers can be re-written in matrix-vector notation as r n = X 1 h n1 + X 2 h n2 + w n 2 with r n = [r n 0 r n N u 1] T, h n,m = [h nm 0 h nm N u 1] T, w n =[w n 0 w n N u 1] T and X m = diag{[x m 0 x m N u 1]} being a diagonal matrix containing the transmitted symbols. The demultilexer following the FFT and CP removal block searates the signal received at ilot and data subcarriers. The ilot signals r,n = [r,n 0 r,n N 1] T are fed to the resective channel estimator blocks, while the data signals r d,n =[r d,n 0 r d,n N d 1] T are sent to the equalizer. Based on the signal received on the ilot subcarriers, the channel estimation block which will be exlained in detail in Section III rovides the equalizer with estimates h n,m of the channel frequency resonses of both the desired and interfering channels. Using these estimates and the signal received at data subcarriers, the equalizer erforms MMSE- IRC filtering to recover the desired transmitted symbols as x d,1 k = h H d,1k H H d k H d k+σ 2 wi 1 rd k. 3 In the above equation, r d k =[r d,1 kr d,2 k] T, H d k = [h d,1 kh d,2 k] and h d,m k =[h d,1m kh d,2m k] T,with h d,nm k denoting the gain of the time-frequency resonse of the nm channel at the k th data subcarrier, and I denotes the 2x2 identity matrix. Finally, the coded bits of the user of interest c 1 k are recovered from the equalized symbols in the QAM detector and are fed to the channel decoder which yields the estimates of the information bits b 1 k. III. CHANNEL ESTIMATOR In this section, two channel estimation aroaches are resented. The first one is the LMMSE estimator, which will be used as a benchmark for the erformance evaluation of our estimator. Next, our roosed channel estimator based on the DM framework is introduced. The DM rincile is briefly exlained, along with the alication to our secific scenario. Authorized licensed use limited to: Aalborg Universitetsbibliotek. Downloaded on January 11, 2010 at 09:57 from IEEE Xlore. Restrictions aly.

4 This full text aer was eer reviewed at the direction of IEEE Communications Society subject matter exerts for ublication in the IEEE ICC 2009 roceedings More details about the DM framework and its relation to other known algorithms can be found in [9]. A. LMMSE Channel Estimator The linear LMMSE channel estimator aims at minimizing the mean-squared error of the estimate. For the signal model under consideration, the LMMSE transfer function estimate of the n1 channel reads: h 1 = arg min E{h n1 h n1 H h n1 h n1 } h n1 = Σ hn1h,n1 X H,1X,1 Σ h,n1 X H,1 + X,2 Σ h,n2 X H,2 + Σ 1 r,n 4 where Σ = E{ H } = σwi, 2 Σ hn1h,n1 = E{h n1 h H,n1} and Σ h,nm = E{h,nm h H,nm}. Note that we assume no correlation between the channels n1 and n2. The estimator requires the inversion of an N N matrix every OFDM symbol, which is normally too comlex to comute in a mobile receiver for a system with a large number of subcarriers. In the rest of the section, we resent an iterative aroach which avoids this matrix inversion. B. Divergence minimization Let Φ denote a vector including as comonents all the unknown arameters to be estimated and Φ r be the osterior robability density function df of Φ given an observation r. The DM framework aroximates Φ r by an auxiliary df qφ minimizing the KL divergence [11] D qφ Φ r dφqφ log qφ 5 Φ r In our alication, the arameters to estimate are the channel resonses of the desired and interfering channels, as well as the inverse of the noise covariance matrix, i.e., Φ = {h,n1, h,n2,,n }, where Σ,n = E{,n,n}. H The index indicates that only ilot subcarriers are taken into account. In order to get a solution that can be comuted with tractable comlexity, the auxiliary df is assumed to factorize according to qφ =qh,n1, h,n2,,n = q h,n1 h,n1 q h,n2 h,n2 q Σ 1 w,n,n. 6 The observation is the received signal at the ilot subcarriers, i.e r = r,n = X,1 h,n1 + X,2 h,n2 +,n. 7 The algorithm iteratively minimizes the KL divergence with resect to one of the factors in 7, while the other factors are ket fixed, resulting in an iterative scheme. Note that the channel estimation rocess is done indeendently for each of the receive antennas. In the remainder of the section we therefore dro the receive antenna subindex n in order to simlify the notation e.g. denotes h,n1. The algorithm is started with initial distributions q [0], q [0] h,2 h,2 and q [0], and these distributions are successively udated according to the udating stes detailed in the following two subsections. C. Udate of the channel vectors In this subsection, the derivation of the udating ste for q h,1 is detailed. Due to the symmetry of the roblem, the udate for q h,2 h,2 is analogous. To udate q h,1, the distributions q [i] h,2 h,2 and q [i] are ket fixed, and q h,1 is udated by solving the following roblem: minimize D q h,1 q [i] h,2 h,2 q [i], h,2, r subject to 8 q h,1 d =1 q h,1 0. The distribution q [i+1] solving 8 is found to be q [i+1] [ { { }}] ex q h q log r, h,2, Σw 1,2 where is the rior df of. The log-likelihood function in 9 reads log r, h,2, { e log tr r X,1 X,2 h,2 9 r X,1 X,2 h,2 H}. 10 The marginalization of 10 with resect to q [i] h,2 h,2 and q [i] yields { { }} q h q log r, h,2,,2 { e tr Ω 1 w, [i] A [i]}, 11 { } where Ω [i] w, 1 q and A [i] = r X,1 X,2 h [i],2 r X,1 X,2 h [i],2 H +X,2 Σ [i] h,2 X H,2.12 Details on Ω [i] w, 1 and Σ [i] h,2 are given later on in this section. Note that terms indeendent of have been neglected in the derivation as they do not affect q [i+1]. For Rayleigh fading channels, the rior distribution of is{ Gaussian} with zero mean and covariance matrix Σ h,1 = E h H,1. Using this rior distribution and 11 in 9, we obtain an udated distribution, which is also Gaussian, with df: q [i+1] ex [ h [i+1],1 H Σ [i+1] 1 ] h [i+1],1 13 Authorized licensed use limited to: Aalborg Universitetsbibliotek. Downloaded on January 11, 2010 at 09:57 from IEEE Xlore. Restrictions aly.

5 This full text aer was eer reviewed at the direction of IEEE Communications Society subject matter exerts for ublication in the IEEE ICC 2009 roceedings with mean vector h [i+1],1 = Σ h,1 Σ h,1 X H,1X,1 + Ω [i] r X,2 h [i],2 and covariance matrix Σ [i+1] = w, 1X H,1 14 +Ω [i] w, 1 X H,1X, By insecting 14, it is seen that the channel resonse udating ste has the form of a LMMSE or Wiener filter [12], alied to the interference-cancelled received signal at ilot ositions. As the interference cancellation is not ideal, the takes into account both the noise and the residual interference ower, in order to correctly smooth the channel resonse, as it is shown in the next subsection. Note that the udate of the channel coefficients in 14 does only rovide estimates of the channel resonse at ilot subcarriers. Estimates of the full frequency resonse at all active subcarriers are obtained by using estimate of the noise covariance matrix Ω [i] w, h 1 = Σ h1 Σ h,1 X H,1X,1 + Ω [i] r X,2 h [i],2 w, 1X H,1 16 instead of 14 { in the } last iteration of the algorithm with Σ h1 = E h 1 h H,1. D. Udate of the noise covariance matrix When udating q Σ 1 w, the distributions q [i] and q [i] h,2 h,2 are ket fixed, and the otimization roblem to solve is the following: minimize D q [i] q [i] h,2 h,2 q Σ 1 w, h,2, r subject to 17 q Σ 1 d =1 The solution reads q [i+1] ex q Σ 1 0. [ { { q h q log r, h,2,,1 h,2 }}]. 18 The marginalization of 10 is taken with resect to h [i] h [i],2, resulting in { { }} q q h,2 log r, h,2, where e log w tr { B [i] =r X,1 h [i],1 X,2h [i],2 r X,1 h [i],1 X,2h [i],2 H,1 and w,b [i]}, 19 + X,1 Σ [i] X H,1 + X,2 Σ [i] h,2 X H,2. 20 By choosing the rior df to be flat, 18 becomes [ { q [i+1] w ex tr w,b [i]}], 21 which has the form of a Wishart distribution [13] as w, W N N +2, B [i] 1. The mean of is therefore Ω [i+1] w, 1 E q [i+1] { } B [i] 1 =. 22 N +2 In order to simlify the algorithm, it is assumed that Σ reresents the covariance matrix of a white Gaussian noise vector with = diag{σw 2,...,σw 2 }. In this case, the udated df is given by [ { q σ 2σw 2 σw 2 N ex σw 2 tr B [i]}] 23 which is a chi-square distribution [13]. Secifically σw 2 χ 2 N +2, and the exectation of σ 2 is { } { } σw 2 [i+1] = E [i+1] q σw 2 = tr B [i] σ 2 N +2 E. Imlementation Issues 1 Matrix inverse in the udate of the channel vectors: As it can be observed in 14, the inversion of a matrix of dimensions N N is still required for the udate of the channel coefficients vector. To avoid the matrix inversion, 14 can be rewritten as: 1U h [i+1],1 = US S +σw 2 [i] I H N X H,1 r X,2 h [i],2 25 where Σ h,1 = USU H is the singular value decomosition SVD of the channel covariance matrix. We have also made use of the fact that X H,1X,1 = I N for constant unit-ower ilots, and the simlification of the noise covariance matrix introduced in 23 and 24. Note that the matrix to invert is now a diagonal matrix, which can be inverted with just N comlex oerations. Also, in a wide-sense stationary channel, the rior covariance matrices of the channels will not change over time, and therefore the SVDs need to be comuted only once for each channel. 2 Initialization: Details on how to udate each of the dfs have been given reviously in this section. An initialization of them for the first iteration of the algorithm, however, is needed. In our roosed imlementation, the channel resonses are initialized to null vectors, i.e., h [0],m = [0,...,0] T, and their covariance matrices are initialized to the rior covariance matrices of the channel, Σ [0] h,m = Σ h,m.asforthe noise variance, it is initialized to the AWGN variance, i.e., σw 2 [0] = σw 2. In subsequent iterations, this initialization is udated with the residual interference after the interference cancellation erformed in the udates of the channel resonse vectors. Authorized licensed use limited to: Aalborg Universitetsbibliotek. Downloaded on January 11, 2010 at 09:57 from IEEE Xlore. Restrictions aly.

6 This full text aer was eer reviewed at the direction of IEEE Communications Society subject matter exerts for ublication in the IEEE ICC 2009 roceedings 3 Udating schedule: Another imortant asect having an imact on the erformance of the algorithm is the order in which the dfs are udated. So far, no analytical way of determining the otimal udating order has been found. In this article, we evaluate the following udating order: q h,1, q h,2, q σ 2. Intuitively, the desired user channel should be as strong or stronger than the interfering channel, thus it is selected to be estimated first. Once a first estimate of the desired channel is available, the interfering channel can be estimated more accurately. Finally an estimate of the residual noise lus interference is obtained to imrove the channel estimates in subsequent iterations. Simulation results which have not been included here due to lack of sace showed no relevant gain by udating the recirocal of the noise variance between the estimates of the desired and interfering channel. Therefore, this ste is not included in the algorithm, yielding a less comlex scheme with no areciable loss in erformance. IV. PERFORMANCE EVALUATION In this section, we evaluate the erformance of the roosed channel estimator by means of Monte-Carlo simulations. In order to do so, we define an OFDM system with arameters insired by the 3GPP Long Term Evolution LTE 5 MHz downlink hysical layer arameters [2]. The system oerates with an FFT size of 512, with 300 active subcarriers, and a frequency sacing of 15 KHz between them. Pilot subcarriers are transmitted in every OFDM symbol, with a frequency sacing of 6 subcarriers i.e. 300 KHz between them. Both the desired and interfering signals have their ilots in the same subcarriers, and erfect synchronization between the transmitters is assumed. Hence, ilots of both transmitted signals overla in frequency. The ilot sequences are made of random indeendent and uniformly distributed QPSK symbols. The ower of the interfering signal is equal to that of the desired signal, and 16QAM modulation is emloyed for the data symbols. A convolutional code is used for channel coding, with BCJR [14] decoding at the receiver. Two different channel models are considered, namely the ITU Indoor Office A channel [15] and the COST 259 Tyical Urban channel [16]. The former channel exhibits a low frequency selectivity, with a coherence bandwidth of about 3.2 MHz, while the latter has much less coherence bandwidth of around 467 KHz. Block fading is assumed, with a static channel resonse over the duration of an OFDM symbol and indeendent realizations between consecutive OFDM symbols. The same channel rofile is assumed for all wireless links desired and interfering. In Fig. 3, the mean-squared error MSE of the channel estimates of the desired and interfering channel versus the number of iterations of the estimator are shown for the two considered channels. The MSE of the LMMSE estimator is also deicted for comarison s sake. The signal-to-noise ratio SNR, which is calculated as the ratio between the desired signal ower and the noise ower for each antenna branch, is fixed to 25 db. It is observed that the iterative rocess MSE db MMSE Estimator Desired Channel TU Interfering Channel TU Desired Channel IndA Interfering Channel IndA Iteration Fig. 3. MSE of the channel estimates versus the number of iterations of the channel estimator at a fixed SNR of 25 db. BER Known Channel MMSE DM 5 iter DM 10 iter SNR db Fig. 4. BER erformance for an Indoor A Channel. imroves greatly the quality of the estimates, due to the effectiveness of the interference cancellation and the udating of the noise covariance matrix, which accounts for both the AWGN and the residual interference. A lower MSE about a 7 db difference is achieved in the Indoor A channel. This is a consequence of the lower frequency selectivity, a well-known result from LMMSE channel estimation. It is also noted that the convergence rate of the algorithm deends on the frequency selectivity of the channel as well: while 5 iterations are enough to achieve convergence in the Indoor Channel, around 10 iterations are needed in the Tyical Urban Channel. As the results show, the DM channel estimator erformance converges to the LMMSE estimator with sufficient number of iterations, and the number of iterations required for convergence deends on the frequency selectivity of the channel. The receiver s erformance is evaluated in terms of biterror-rate BER in Fig. 4 for the Indoor Office A Channel and in Fig. 5 for the Tyical Urban Channel. Results are shown for the case where the estimators use 5 and 10 iterations. As a Authorized licensed use limited to: Aalborg Universitetsbibliotek. Downloaded on January 11, 2010 at 09:57 from IEEE Xlore. Restrictions aly.

7 This full text aer was eer reviewed at the direction of IEEE Communications Society subject matter exerts for ublication in the IEEE ICC 2009 roceedings BER Known Channel MMSE DM 5 iter DM 10 iter SNR db Fig. 5. BER erformance for a Tyical Urban Channel. reference, the BER of the receiver with erfect knowledge of the channel is also deicted, as well as the BER of a receiver using the LMMSE estimator. In the Indoor Channel, the DM and LMMSE estimators exhibit the same erformance. When comared to a receiver with erfect knowledge of the channel, a very small degradation in the range of 1 db is observed in the high SNR range. Furthermore, as commented above, the erformance of the estimator does not significantly imrove after 5 iterations, with only a very marginal gain after 10 iterations. In the Tyical Urban Channel, a larger erformance deviation from the erfect channel knowledge results is observed. The degradation ranges from 1.7 db to 2.4 db at BER of 10% and 0.1% resectively. However, the degradation in the high SNR range is relatively small considering a scenario with such a severe interference. Again, the erformance of the DM estimator is very close to that of the LMMSE estimator, and only a very slight gain is observed when the number of iterations of the algorithm is increased from 5 to 10. V. CONCLUSION In this aer, we have resented a novel aroach for channel estimation in OFDM systems with synchronized cochannel interferers and overlaed ilot symbols. Based on the KL-divergence minimization rincile, an iterative algorithm for estimation of the channel gains based on the signal observed at ilot locations has been derived. The resulting algorithm combines LMMSE channel estimation with successive interference cancellation and estimation of the noise and residual interference ower. The effectiveness of the roosed estimator is assessed by Monte-Carlo simulations. The results show that our algorithm erforms as good as the full LMMSE channel estimator, with the advantage of avoiding the cumbersome matrix inversion in the latter. An overall receiver erformance very close to that of a receiver with erfect knowledge of the channel coefficients is attained, esecially in channels with low frequency selectivity. To conclude, it is worth remarking that although the estimator has been resented and evaluated for an OFDM system with synchronized co-channel interference, alication to other scenarios could be very advantageous. For instance, our estimator would allow to reduce the ilot overhead in a MIMO-OFDM system by lacing the ilot sequences of all transmit antennas in the same time-frequency locations, instead of having secific locations reserved for each of the antennas as it is common in current wireless standards, e.g. LTE. ACKNOWLEDGMENT This work has been artly funded by the FP7-ICT Network of Excellence in Wireless Communications, NEWCOM++ Contract No The authors would also like to thank Infineon Technologies Denmark A/S for the financial suort which made this work ossible. REFERENCES [1] IEEE Standard for Local and metroolitan area networks Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems Amendment 2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Oeration in Licensed Bands and Corrigendum 1, IEEE Std e-2005, [2] Evolved Universal Terrestrial Radio Access E-UTRA; LTE Physical Layer - General Descrition release 8, 3rd Generation Partnershi Project, Tech. Re. TS , V8.1.0, Nov [3] L. Hanzo, M. Munster, B. J. Choi, and T. Keller, OFDM and MC-CDMA for Broadband Multi-User Communications, WLANs and Broadcasting. West Sussex, England: John Wiley & Sons, [4] M. Munster, T. Keller, and L. Hanzo, Co-channel interference suression assisted adative OFDM in interference limited environments, Vehicular Technology Conference, VTC Fall. IEEE VTS 50th, vol. 1, vol.1, [5] J. Li, K. Ben Letaief, and Z. Cao, Co-channel interference cancellation for sace-time coded OFDM systems, Wireless Communications, IEEE Transactions on, vol. 2, no. 1, , Jan [6] C.-S. Ni and K.-C. Chen, Cochannel interference suression for coded OFDM systems over frequency-selective slowly fading channels, Vehicular Technology Conference, VTC2004-Fall IEEE 60th, vol. 1, Vol. 1, Set [7] Y. Li and N. Sollenberger, Adative antenna arrays for OFDM systems with cochannel interference, Communications, IEEE Transactions on, vol. 47, no. 2, , Feb [8] J.-J. van de Beek, O. Edfors, M. Sandell, S. Wilson, and P. Borjesson, On channel estimation in ofdm systems, Vehicular Technology Conference, 1995 IEEE 45th, vol. 2, vol.2, Jul [9] B. Hu, I. Land, L. Rasmussen, R. Piton, and B. Fleury, A divergence minimization aroach to joint multiuser decoding for coded CDMA, Selected Areas in Communications, IEEE Journal on, vol. 26, no. 3, , Aril [10] M. Raghavendra, S. Bhashyam, and K. Giridhar, Parametric channel estimation in reuse-1 OFDM systems, Communications, ICC 07. IEEE International Conference on, , June [11] T. Cover and J. Thomas, Elements of information theory. John Wiley & Sons, [12] P. Hoeher, S. Kaiser, and P. Robertson, Two-dimensional ilot-symbolaided channel estimation by Wiener filtering, Acoustics, Seech, and Signal Processing, ICASSP-97., 1997 IEEE International Conference on, vol. 3, vol.3, Ar [13] A. Guta and D. Nagar, Matrix Variate Distributions. Chaman & Hall/CRC, [14] L. Bahl, J. Cocke, F. Jelinek, and J. Raviv, Otimal decoding of linear codes for minimizing symbol error rate, Information Theory, IEEE Transactions on, vol. 20, no. 2, , Mar [15] Guidelines for evaluation of radio transmission technologies for IMT- 2000, ITU, Tech. Re. Recommendation ITU-R M.1225, [16] Deloyment asects, 3rd Generation Partnershi Project, Tech. Re. TS , Jun Authorized licensed use limited to: Aalborg Universitetsbibliotek. Downloaded on January 11, 2010 at 09:57 from IEEE Xlore. Restrictions aly.