Compressive Sensing Based Signal Estimation And Detection For Large MIMO-OFDM Systems

A large MIMO-OFDM (multi-input multi-output orthogonal frequency division multiplexing) system with tens or hundreds antennas at transmitter and receivers, which can achieve high multiplexing gain, high diversity gain and high spectral efficiency, is a promising way to meet the challenges brought by the increasing data rates and communication reliability in the future wireless networks. Signal detection and channel estimation are two long standing studies in signal processing area. And the accuracy and efficiency of their processes is a key to the large MIMO-OFDM systems. In this thesis, we consider the signal estimation and detection in large MIMO-OFDM systems. To be concrete, we aim to improve the accuracy and efficiency of their process by fully exploiting the sparsity in wireless communications and using the sparse signal processing framework, i.e., compressive sensing.Firstly, we consider the sparse channel estimation in OFDM systems. This problem is investigated from two aspects:pilot arrangement and spars channel impulse response recovery. Equi-spaced pilot arrangement is the most popular scheme for pilot-aided transmission in OFDM systems. In this thesis, we argue that a non-equispaced pilot pattern may outperform its equi-spaced counterpart, if we fully take into account the sparsity of the channel impulse response (CIR) which is inherent in wireless channels. More specifically, a sparsity-aware pilot arrangement scheme based on the coherence criterion is investigated in this paper. To address the resulting NP-hard combinatorial optimization problem, we propose an efficient local search algorithm. For channel estimation, we convert it to a sparse recovery problem. To enhance the applicability of our scheme, i.e., when there is no prior knowledge about the channel order, we propose to employ the Bayesian information criterion (BIC) to estimate the channel order first and then recover the sparse channel vector via existing low-complexity methods, e.g., orthogonal matching pursuit (OMP). By combining the above pilot arrangement scheme with channel estimation, our scheme exhibits substantially better performance in comparison with the conventional equispaced schemes with linear (or spline) interpolation, in terms of total number of pilot symbols and bit error rate.Secondly, we consider the signal detection in large MIMO systems. Two different modulation schemes, i.e., quadrature amplitude modulation (QAM) and spatial modulation (SM), are investigated. For QAM (such as BPSK and4-QAM), we propose a novel low-complexity detector for large MIMO systems, which is capable of achieving near-ML (maximum likelihood) performance. The main idea of our algorithm is to successively boost the detection by leveraging the hidden sparsity in the residual error of received signal. Specifically, since the SER (symbol error rate) of the MMSE (minimum mean squared error) detector is usually not high (say, less than10%), the residual error, which is the difference between the original transmitted signal and the recovered one, would exhibit significant sparsity. Therefore, by locating the non-zero entries (i.e. the incorrectly detected symbols) via compressive sensing algorithms, we can reduce the original MIMO system to a new one, whose input dimension is much less than the output dimension. This implies that a linear detector will suffice for achieving near-optimal performance, otherwise we can repeat the above procedures to iteratively boost the detection till satisfaction. Overall, our proposed algorithm can achieve performance close to the optimal ML detector, while its complexity is just on the order of the linear detectors (say, MMSE detector).Finally, we consider the detection of generalized space shift keying (GSSK) in large-scale MIMO systems, which is simplified variation of SM. In GSSK scheme, there are only a few fraction antennas activated at transmitters and the information is coded in the indices of the activated antennas. Therefore, the GSSK signal is naturally a sparse zero-one vector. Inspired by the property of the GSSK signal, we proposed a sparse K-Best detector based on the breadth-first category of sphere detector (referred to K-best sphere decoding). Different with the conventional K-Best detector searching all the transmit antennas, our proposed SK detector investigates only a few promising candidates which are activated antennas at transmitter. Overall, our proposed SK detector exploits not only the sparsity of the GSSK signal but also the constraint on its nonzero values. Therefore, the restricted isometry property (RIP) based performance analysis shows that the exact recover condition of our proposed detector is looser than that of greedy algorithms. Moreover, the empirical results show that our detector performs much better than the sparse algorithms based normalized compressive sensing (NCS) detectors while exhibits only slightly higher complexity than the latter (the low-complexity orthogonal matching pursuit (OMP) based NCS detector).In a word, the above results show that not only the system overhead can be reduced but also the performance of signal estimation and detection can be enhanced substantially if we can fully exploit the sparsity of the signal in the large MIMO-OFDM systems.