A Robust Variable Forgetting Factor QS-decomposition Algorithm for Subspace Tracking
Jianqiang Lin, S. C. Chan · 2018
This paper proposed a robust variable forgetting factor (VFF) QS-decomposition algorithm for subspace tracking in impulse noise environment. The QS-decomposition algorithm was originally proposed to estimate recursively the principal orthonormal subspace of covariance matrix of a vector time series. Motivated by the close relationship between the projection approximation subspace tracking (PAST) algorithm and the recursive least squares (RLS) with multiple outputs, the local optimal forgetting factor (LOFF) algorithm recently proposed is extended to multiple outputs and is incorporated into the QS algorithm to improve its tracking and steady state mean squares error performances in nonstationary and stationary environment respectively. Furthermore, the M-estimation function is employed to improve the robustness of the proposed VFF QS-tracker against possible impulses or outliers which may be encountered in practice. Experimental results show that the proposed robust VFF-QS algorithm is able to achieve better performance in stationary and nonstationary environments than the conventional QS algorithm, especially in the presence of impulsive noise.