Study On Sparse Doubly-selective Channel Estimation And Efficient Decoding In Symmetric A Stable Noise Environment

This thesis is devoted to a study on the technologies of sparse doubly-selective channel estimation and high efficiency channel decoding in SaS noise environments, which has already been a hot topic of research in signal processing recently, including channel estimation based on adaptive filter and compressed sensing, soft de-mapping in SaS noise, low complexity and low latency TPC iterative decoding and application of fountain codes. The work finished in this paper is a part of a large-scale engineering project undertaken by the Lab the author works with. The major contribution and innovative achievements obtained in this thesis are summarized as follows.1. SaS noise model and several currently existing BEM-based doubly-selective channel models are studied in depth. Based on above analysis the DFT-BEM model is chosen for use in this thesis.2. With regard to sparse doubly-selective channel estimation, following achievements were made.● Targeting at the situation that some large amplitude pulse noise with SaS distribution could severely influence the channel estimation, a method called 5σ pre-processing is provided, and based on which a new algorithm based on H∞ adaptive filter is proposed. Simulation results show that 5σ pre-processing can remarkably improve the performances. Compared with the conventional sIPNLMS, new algorithm shows obvious performance superiority in both sparse multipath channels and sparse double selective channels. The length of training sequence needed by proposed algorithm is longer than that of the algorithm based on compressed sensing, however, the proposed algorithm has no floor effect in approximately sparse channels.● Based on theoretical analysis of residual errors of the existing OMP-based sparse channel estimation algorithm with SaS distribution assumption, a novel robust stopping criterion for the algorithm was proposed, and the iteration number of the algorithm is just the wanted estimated value of the sparsity. Simulation results have proved the robustness of the new criterion, and the estimation accuracy of sparsity will exceed 90% when GSNR is higher than 5dB. In addition, the overall performances of improved algorithm based on above new stopping criterion are similar to that of the conventional OMP algorithm with known sparsity. Also, the proposed algorithm show obvious superiority over the existing IPNLMS and mCoSaMP algorithms that are without the need of sparsity parameter in advance.● An improved CoSaMP algorithm for SaS noise environments is proposed, in which the method of minimum p norm was introduced and LS estimation was replaced by LPN estimation. Also, a new kind of sparsity vector substitute was proposed in the new algorithm which does not need to be strictly conform to the RIP criterion, thus can remarkably reduce the overhead of training sequence and increase the channel capacity. Besides, a new sparsity estimation method is provided for the improved algorithm. Simulations show the sparsity estimation accuracy of the new CoSaMP algorithm can be higher than 70% if GSNR is above 5dB, And due to the use of the method of minimum p norm, the overall channel estimation performances of the new CoSaMP algorithm are a little bit better than that of existing CoSaMP algorithm without the need of sparsity.3. As regards efficient decoding, following achievements were made.● An improved soft de-mapping algorithm suitable for SaS noise environments is proposed. As the conventional soft de-mapping algorithms originally designed under Gaussian noise assumption, the new algorithm can still use the Euclidean distance as the metric parameter with the only difference that a pre-processing unit is introduced between the soft de-mapping and decoding, thus is very easy to use and of lower computational complexity. Simulation results show that, with the same bit error rate, the GSNR (geometric signal-to-noise ratio) needed by proposed algorithm is about 0.3 dB lower than that of the Huber algorithm and almost the same as that of numerical calculation algorithm based on SaS probability density function.● Targeting at the problems of large delay and high complexity of existing Chase-Pyndiah algorithm and the performance loss of Argon algorithm, an improved parallel iterative decoding structure for TPC is designed. The new algorithmic structure has the same performance as that of the serial structure, but the latency is reduced to half of the serial structure. In the design, an optimum search method of the least reliable bit positions is proposed for reducing the decoding delay. A series of improvements were implemented including simplified calculation of Euclidean distances between candidate codes and soft input vectors, the use of Gray code test patterns, and the optimized search for ML code word and competing codeword. All of these measures greatly reduced the complexity and decoding latency.● The topic of what codes are most suitable for the situations with sparse doubly-selective channel and severe SaS noise was studied in this thesis. Analysis and simulations show that fountain code is a suitable one for this purpose with the advantages of low power consumption, making full use of the time-varying channel capacity, easy to be concealed and without the need of feedback. Simulation results have proved that fountain code outperforms ARQ, and its transmission time is almost independent of frame error ratio.