Channel Estimation Techniques For OFDM Systems

Orthogonal frequency division multiplexing (OFDM) is a promising candidate for next-generation high-speed mobile multimedia communication systems due to its high spectral efficiency, robustness to multipath-fading channel and simple hardware implementation. In OFDM systems, a real-time and accurate channel estimation is not only crucial for coherent detection, but also indispensable for adaptive link and multiple-input and multiple-output (MIMO) transmission. This thesis focuses on channel estimation for OFDM systems, including pilots symbol assisted methods and blind methods.A low-rank adaptive (LRA) channel estimator is proposed in order to simultaneously reduce the computation complexity of minimum mean square error (MMSE) channel estimator and minimize its performance loss due to mismatch of the estimator-to-channel statistics. This algorithm adaptively tracks the noise power change and the principal eigenspace of time-average channel correlation matrix, and models the MMSE channel estimator by a low-rank adaptive filter. Therefore, the LRA has much lower computation complexity than the MMSE. Theoretical analysis indicates that without prior information of channel statistics, the LRA converges to the solution of statistics-matched MMSE estimation asymptotically. Moreover, LRA not only improves the one-dimensional (1-D) channel estimation, but also can be generalized to 2-D pilot-symbols assisted channel estimation. Simulation results show the effectiveness of LRA, as well as its improvement over other techniques when channel statistics information is unknown.A subspace based blind channel estimation algorithm is proposed for an Alamouti coded STC-OFDM system with 2 transmit antennas. This algorithm makes use of the redundancy introduced by zero-padding and virtual carriers,therefore channel identifiability is guaranteed regardless of channel zero locations and underlying symbol constellations. Identifiability analysis shows that multiple channels can be identified simultaneously up to one scalar ambiguity. With a small quantity of pilots, a semi-blind estimator can be established to resolve this scalar ambiguity. The subspace blind estimate's mean square error (MSE) is obtained based on the first order perturbation theory. Simulation results testify the theoretical MSE and indicate that the semi-blind estimator is more accurate and more efficient than the training-based one. For OFDM systems, the necessary and sufficient conditions are proposed and proved for blind channel identifiability based on the finite alphabet property of information symbols. The existing sufficient condition is too strict while the proposed conditions are looser which can be applied in all cases of channel identifiability. Based on these new conditions, a novel minimum distance algorithm implementing in frequency domain (FMD) is proposed for blind channel estimation. By polyphase decomposition, the channel frequency response sequence is divided into several subsequences, and phase ambiguities of each subcarrier can be resolved by exhaustive searching only L elements in one subsequence, where L is the channel order. The applicability of FMD algorithm is more general than the existing ones because FMD algorithm can be utilized as long as parameters of the channel and the OFDM system satisfy the proposed conditions. Simulation results show that the FMD algorithm is accurate for channel estimation and has a low computation complexity.A novel blind channel estimator based on computing roots of matrices (RM) is proposed for OFDM systems. This algorithm exploits the finite alphabet property of information symbols and implements channel deconvolution by computing the Jth principle root of a low-triangular Toeplitz matrix. Therefore, RM algorithm has a computation complexity proportional to the square order of channel impulse response, which is much lower than searching algorithms in previous works, and RM is able to function in the case of large channel order that is intractable by searching algorithms. Moreover, an adaptive RM (ARM) algorithm is proposed to adjust RM estimator by steepest descent method. Simulation results indicate that RM algorithm has great accuracy comparable to the optimal exhaustive search, and ARM improves the estimation performance of RM considerably in low SNR condition.