ADAPTIVE HIGH-RESOLUTION ALGORITHMS FOR TRACKING NONSTATIONARY SOURCES WITHOUT THE ESTIMATION OF SOURCE NUMBER

In this paper, a generalized inflation method which can adaptively and robustly converge to the noise-subspace is proposed to improve the pe rformances of subspace algorithms used for tracking nonstationary sour ces. This generalized inflation method, which includes an inflation fa ctor developed in the view point of orthogonal projection, preserves t he parallel structure for realizations and achieves better performance s of convergence and initialization behavior than the inflation method , adaptive PHR algorithms, and other adaptive eigensubspace algorithms when the number of sources is not known. A bound of the inflation fac tor is also suggested to secure the noise-subspace-only adaptation. Th e general inflation method in use of weighted-subspace can further imp rove the tracking performances. Simulations for analyzing the tracking performances of the algorithms are also included.