This paper proposes a novel time}frequency maximum likelihood (t}f ML) method for direction-of-arrival (DOA) estimation for nonstationary signals impinging on a multi-sensor array receiver, and compares this method with conventional maximum likelihood DOA estimation techniques. Time}frequency distributions localize the signal power in the time}frequency domain, and as such enhance the e!ective SNR, leading to improved DOA estimation. The localization of signals with di!erent time}frequency signatures permits the division of the time}frequency domain into smaller regions, each containing fewer signals than those incident on the array. The reduction of the number of signals within di!erent time}frequency regions not only reduces the required number of sensors, but also decreases the computational load in multidimensional optimizations. Compared to the recently proposed time}frequency MUSIC (t}f MUSIC), the proposed t}f ML method can be applied to coherent environments, without the need to perform any type of preprocessing that is subject to both array geometry and array aperture. # 2000 The Franklin Institute. Published by Elsevier Science Ltd. All rights reserved. Keywords: Time}frequency distribution; Direction "nding; Maximum likelihood; Spatial time}frequency distribution; Array processing 1.

This paper proposes a novel time-frequency maximum likelihood #t-f ML# method for direction-of-arrival #DOA# estimation for non-stationary signals, and compares this method with conventional maximum likelihood DOA estimation techniques. Time-frequency distributions localize the signal power in the time-frequency domain, and as such enhance the e#ective SNR, leading to improved DOA estimation. The localization of signals with di#erent t-f signatures permits the division of the time-frequency domain into smaller regions, each contains fewer signals than those incident on the array. The reduction of the number of signals within di#erent time-frequency regions not only reduces the required number of sensors, but also decreases the computational load in multi-dimensional optimizations. Compared to the recently proposed time-frequency MUSIC #t-f MUSIC#, the proposed t-f ML method can be applied in coherentenvironments, without the need to perform anytype of preprocessing that is subject to both array geometry and array aperture.

We address the problem of blind source separation of non-stationary signals of which only instantaneous linear mixtures are observed. A blind source separation approach exploiting both auto-terms and cross-terms of the time-frequency (TF) distributions of the sources is considered. The approach is based on the simultaneous diagonalization and anti-diagonalization of spatial TF distribution matrices made up of, respectively, auto-terms and cross-terms. Numerical simulations are provided to demonstrate the effectiveness of the proposed approach and compare its performances with existing TFbased methods.