Performance analysis of an adaptive subspace filter for signal enhancement

In this paper, we investigate the noise suppression performance of a novel adaptive filter based on subspace methods. In contrast to conventional subspace filters, this approach requires no eigenvalue or singular value decomposition. Therefore, our adaptive subspace filter can be implemented for real-time operation using general purpose digital signal processors. In addition, only the noisy signal, and no reference signal is needed. We compare the adaptive subspace filter (ASF) with the adaptive line enhancer (ALE) which also requires no reference signal. For stationary signals the noise suppression performance is similar to that of the ALE. However, in a nonstationary environment the ASF exhibits far better tracking behavior, and faster convergence rate