Detection guided NLMS estimation of sparsely parametrized channels

We consider the normalized least mean square (NLMS) estimation of a channel, which may be well approximated by a finite impulse response model with sparsely separated active or nonzero taps. Previously reported analyses imply that the convergence rate of the NLMS estimator should be greatly enhanced if only the active taps are estimated. We propose an NLMS estimator, which incorporates a least squares based active tap detection method. Simulations demonstrate that the NLMS estimator has significantly faster convergence than the standard NLMS estimator for colored as well as white input signals, Furthermore, for sparse channels, this improved convergence speed is accompanied by a lower computational cost.