Robust recursive bi-iteration singular value decomposition (SVD) for subspace tracking and adaptive filtering

The recursive bi-iteration singular value decomposition (Bi-SVD), proposed by Strobach (1997), is an efficient and well-structured algorithm for performing subspace tracking. Unfortunately, its performance under an impulse noise environment degrades substantially. In this paper, a new robust-statistics-based bi-iteration SVD algorithm (robust Bi-SVD) is proposed. Simulation results show that the proposed algorithm offers significantly improved robustness against impulse noise than the conventional algorithm with a slight increase in arithmetic complexity. For nominal Gaussian noise, the two algorithms have similar performance.