Reduced-Complexity Adaptive Filtering Techniques for Communications Applications

Adaptive filtering algorithms are powerful signal processing tools with widespread use in numerous engineering applications. Computational complexity is a key factor in determining the optimal implementation as well as real-time performance of the adaptive signal processors. To minimize the required hardware and/or software resources for implementing an adaptive filtering algorithm, it is desirable to mitigate its computational complexity as much as possible without imposing any significant sacrifice of performance. This thesis comprises a collection of thirteen peer-reviewed published works as well as an integrating material. The works are along the lines of a common unifying theme that is to devise new low-complexity adaptive filtering algorithms for communications and, more generally, signal processing applications. The main contributions are the new adaptive filtering algorithms, channel equalization techniques, and theoretical analyses listed below under four categories: 1) adaptive system identification ? affine projection algorithm with selective projections ? proportionate affine projection algorithm with selective projections ? modified RLS algorithm with enhanced tracking capability 2) adaptive inverse system identification ? low-complexity adaptive decision-feedback equalization of MIMO channels ? partial-update adaptive decision-feedback equalization ? MIMO-DFE coupled with V-BLAST for adaptive equalization of wideband MIMO channels 3) adaptive set-membership system identification ? modified quasi-OBE algorithm with improved numerical properties ? low-complexity implementation of the quasi-OBE algorithm ? steady-state mean squared error and tracking performance analysis of the quasi-OBE algorithm ? tracking performance analysis of the set-membership NLMS algorithm 4) adaptive linearly-constrained system identification ? reduced-complexity constrained recursive least-squares algorithm ? linearly-constrained recursive total least-squares algorithm ? linearly-constrained line-search algorithm for adaptive filtering. The main techniques utilized to alleviate the computational complexity in the proposed algorithms are ? adoption of a time-varying projection order for the affine projection algorithm ? the concept of partial updates ? the dichotomous coordinate-descent iterations for solving the associated systems of linear equations ? the method of weighting for linearly-constrained filters. The proposed algorithms yield significant savings in terms of computational complexity compared with the existing algorithms and techniques. Numerical computer simulations show that, in most cases, this is realized without incurring any noticeable performance degradation. Establishing sensible trade-offs between complexity and performance is another substantial benefit offered by the proposed algorithms. Moreover, there is a good match between simulation results and theoretically predicted values of the performance metrics for which the analyses have been carried out. The detailed account of each work together with the appropriate literature survey, theoretical and numerical analysis, discussions, and concluding remarks has been reported in the corresponding papers that are embodied in the thesis.