A low power algorithm for sparse system identification using cross correlation

We present a novel algorithm and architecture for adaptive sparse system identification. The algorithm uses a cross correlation to identify active tap weights and uses the scaled version of the cross correlation estimate to seed a reduced complexity adaptive filter. We call the algorithm the sparse cross correlation (SCC) algorithm. Simulations for the finite precision case are presented. Comparisons of area, critical path, power and algorithmic convergence between the normalized least mean squares (NLMS) algorithm and the SCC algorithm are presented. The SCC algorithm is shown to be lower power in both the steady state (trained) and transient (training) operation. Results for a test implementation show that approximately 20% smaller circuit area and approximately 40% lower power consumption than the standard NLMS algorithm can be achieved.