Research On Key Technologies Of Single Channel Blind Separation For Co-frequency Mixed Signals

Single Channel Blind Separation is one of effective ways to process co-frequency mixed signals, this thesis mainly talks about the single channel blind separation of co-frequency mixed signals, some significant problems are researched in depth, which include the time delay estimation of co-frequency mixed signals, efficient Per-Survivor Processing single channel blind separation algorithms and single channel blind separation algorithm in time varying-channel. The main work and innovative points completed in the thesis can be summarized as follows:1. A maximum likelihood time delay estimation algorithm of mixed signals based on trust region algorithm is proposed aiming at the time delay estimation of co-frequency mixed signals, and also giving the global convergence proof of this algorithm. This algorithm make up the shortage of the grid search method in maximum likelihood time delay estimation algorithm take advantage of the quick convergence and accuracy of trust region algorithm, thus accomplished the time delay’s quickly and accurately estimation of mixed signals. The last, proving the effectiveness of this algorithm by simulation experiments.2. A decision feedback-based Per-Survivor Processing(DF-PSP) single channel blind separation algorithm is proposed aiming at the problem of performance degradation of the reduced-complexity truncation Per-Survivor Processing(T-PSP) single channel blind separation algorithm, and slso give the complexity analysis of this algorithm. This algorithm combines the ideas of decision feedback, utilizes the designed feedback filter and the decision sequences which are obtained by the most probable survivor path from the Viterbi detector, to eliminating the interference caused by the tail of Channel impulse response, thus to make up the performance degradation resulted from the Channel impulse response’s truncating processing of T-PSP single channel blind separation algorithm. The simulation results show that this algorithm is superior to the T-PSP single channel blind separation algorithm at the same complexity in processing mixed QPSK signals can obtain larger performance improvement, and also improve performance of the algorithm by oversampling.3. A decision feedback and feedfoward-based Per-Survivor Processing(DFF-PSP) single channel blind separation algorithm is proposed aiming at the problem of DF-PSP single channel blind separation algorithm is difficult to eliminate the interference caused by the forward of Channel impulse response, and slso analyses the complexity of this algorithm. This algorithm designs a delayed feedforward filter based on DF-PSP single channel blind separation algorithm, and also according to the decision sequences which are obtained from the most probable survivor path, to eliminating the interference caused by the forward of Channel impulse response, thus to make up the disadvantage of DF-PSP single channel blind separation algorithm. The simulation results show that this algorithm is superior to the DF-PSP single channel blind separation algorithm at about double complexity in processing mixed QPSK signals can obtain a certain level of performance improvement, and also compare the performance and complexity of T-PSP、DF-PSP、DFF-PSP single channel blind separation algorithm and full state maximum likelihood sequence estimation(MLSE) algorithm, proving the effectiveness of this algorithm. The last, improve performance of the algorithm by oversampling.4. A new single channel blind separation algorithm based on Basis Expansion Model Per-Survivor Processing(BEM-PSP) algorithm is proposed aiming at the poor performance of least mean square Per-Survivor Processing(LMS-PSP) single channel blind separation algorithm in time varying-channel, and also give the complexity analysis of this algorithm. This algorithm divides the full signal processing stage into preprocessing stage and back-end processing stage. In the stage of preprocessing, obtained a portion of accurate CIR by processing a portion of received mixed signals use LMS-PSP single channel blind separation algorithm. In the stage of back-end processing, accomplished estimations of the CIR during time period according to a portion of accurate CIR obtained from preprocessing stage and also combining the channel modeling ideas of BEM, the last estimated the mixed signals with Viterbi algorithm, thus accomplished single channel blind separation of the mixed signals. The simulation results show that this algorithm is superior to the LMS-PSP single channel blind separation algorithm at about fifty percent of complexity and has greater performance in processing mixed QPSK signals, and also this algorithm can obtained better performance increase in same oversampling ratio.