A robust subspace tracking algorithm for subspace-based blind multiuser detection in impulsive noise

Abstract

Subspace tracking is an efficient method to reduce the complexity in estimating the signal subspace required in subspace-based multiuser detection algorithm. Recursive least square (RLS)-based subspace tracking algorithms such as the PAST algorithm can be used to estimate the signal subspace adaptively with relatively low computational complexity. However, it is shown in this paper that subspace estimation using conventional autocorrelation matrix is very sensitive to impulse noise. A new robust correlation matrix, based on robust statistics, is proposed to overcome this problem. Moreover, a new robust PAST algorithm is developed, again using robust statistics, for robust subspace tracking. A new restoring mechanism is also proposed to handle long bursts of impulses, which sporadically occur in communications systems. Simulation results show that the proposed robust subspace tracking-based blind multiuser detector performs better than the conventional approach, especially under consecutive impulses. The adaptation of the proposed scheme in a dynamic multiple access channel, where users may enter and exit the shared mobile channel, is also found to be satisfactory.