LMS estimation of sparsely parametrized channels via structural detection

Considers the LMS estimation of “long” channels which have time domain impulse responses consisting of sparsely spaced nonzero taps or groups of taps. Standard LMS estimation of such long sparsely parametrized channels, in which all taps are estimated, suffers from poor transient and/or asymptotic performance. Analyses carried out by the authors indicate that performance improvements can be achieved by estimating only those taps which are nonzero or “active”. The authors develop a simple procedure, based on the least squares method, which for sufficiently large N (the number of sample intervals) detects the correct number and position of active taps. Using this structure detection procedure the authors propose an LMS based estimation algorithm. Simulations indicate that this “LMS-structure detection” algorithm provides considerable performance improvement over the standard LMS algorithm