DOA estimation of quasi-stationary signals via Khatri-Rao subspace

This paper addresses the problem of direction-of-arrival (DOA) estimation of quasi-stationary signals, which finds applications in array processing of speech and audio. By studying the subspace structures of the local second-order statistics (SOSs) of quasi-stationary signals, we develop a Khatri-Rao (KR) subspace approach that has two notable advantages. First, the approach can operate in underdetermined cases. It is proven that if N is the number of sensors in the array, then the proposed approach can identify up to 2N - 2 source DOAs in an unambiguous fashion. Second, the approach can handle the problem of unknown noise covariance. Essentially, the KR subspace formulation is found to provide a simple and effective way of annihilating the (unknown) noise covariance from the observed signal SOSs. Simulation results, with an emphasis on underdetermined and colored-noise cases, illustrate that the KR subspace approach provides promising mean square estimation error performance.