Semi-Blind Sparse Channel Estimation Using Reduced Rank Filtering

Abstract

This paper develops a novel method for semi-blind estimation of sparse channels in single-input, multiple-output and multiple-input, multiple-output single-carrier transmission systems, using single/multiple transmit and multiple receive antennas. Very little work exists on semi-blind sparse channel estimation. By a novel formulation of the received data matrix (called reduced data matrix) for sparse channels, semi-blind channel/data estimation problem has been cast into a reduced-rank filtering problem. The estimation operates on selected, non-uniformly spaced equalizer taps. Then two classes of semi-blind channel/data estimation are developed, the first being a pseudo-inverse based linear prediction method and the second based on reduced rank multi-stage Wiener filtering. The second class has the advantage of not requiring the rank of the reduced data matrix, but estimating it adaptively via stage-by-stage updates. The computational complexity of the novel reduced channel estimators are compared with the full channel estimators, which use all available equalizer taps. Simulation results show that the reduced semi-blind estimators perform much better than the full estimators and existing sparse channel estimators, with limited number of training symbols, at much reduced computational complexity.