Researches On Optimization Of Channel Estimation And Signal Detection In MIMO-OFDM Systems

MIMO-OFDM, which integrates the multiple input multiple output (MIMO) technique into the orthogonal frequency division multiplexing (OFDM) technique, is one of the key techniques of the next generation mobile communications system. The low peak to average power ratio (PAPR), high transmission quality and rapid mobility have put forward strict requirements to transmission and signal detection in MIMO-OFDM systems. In this dissertation, we focus on the optimization design of PAPR reductions, channel estimation and signal detection for MIMO-OFDM systems.The improved algorithms:C-G-SLM algorithm and C-A-PTS algorithm are advanced which use the spatial diversity of STBC MIMO systems to ameliorate the PAPR performance. Comparing with the conventional SLM and PTS method, the proposed methods reduce the PAPR while decreasing the computational complexity.The channel estimation model and estimation method of slow fading MIMO-OFDM channels based on training sequence are analyzed. According to Least-Square criterion, we have derived the optimal frequency domain and optimal time domain training sequence design. On this basis, an optimal frequency division, phase shift joint training sequence scheme is put forward which can solve the transmit antenna number limitation problem of the traditional training sequence. The proposed training sequence design method can achieve the optimal MSE and PAPR performance.The signal detection methods in MIMO-OFDM systems with low complexity are studied. Then, a sorted decision feedback aided lattice reduction detection scheme is advanced. Considering the border of transformed constellation and the correlations between different element, the element is selected to be quantized corresponding to the value of the quantization errors. The proposed scheme can approach the optimum detection (maximum likelihood detection) performance in polynomial level computational complexity.The Markov Chain Monte Carlo detection scheme with low complexity, optimal or almost optimal performance is discussed. It uses a transmit bits list by Gibbs sampling to estimate each bit's posteriori probability. But in the case of high SNR or iteration, the conventional MCMC method is likely to "trap" in a certain sample state, which leads to a biased posteriori probability estimation. An improved MCMC algorithm is proposed, which disperses the trapped sample sequence randomly in a certain range and finds initial sample value using the MMSE principle simultaneously. Simulation results evidence the improvement of BER performance and computational complexity of the proposed algorithm.