On the performance of semi-blind subspace-based channel estimation

This paper is devoted to the analysis of a "semi-blind" estimation framework in which the standard least-squares estimator (based on a known training sequence) is enhanced by using the statistical structure of the observations. More specifically, we consider the case of a general time-division multiple access (TDMA) frame-based receiver equipped with multiple sensors and restrict our attention to second order based subspace methods that are suitable for most standard communication applications due to their moderate computational cost. The semi-blind channel estimator is obtained as a regularized least-squares solution where a blind subspace criterion plays the role of the regularization constraint. The main contribution of the paper consists of showing by asymptotic analysis how to optimally tune the balance between the blind criterion and the least-squares fit, depending on the design parameters of the system. Simulations show that the proposed solutions are robust and significantly improve the efficiency of the equalization.