Broadband source localization in a slowly time-varying environment by multichannel Kalman filtering
An adaptive multichannel (AMC) strategy is generalized to operate in a nonstationary environment. A recursive means of generating cross spectral density matrix (CSDM) estimates is presented in terms of a Kalman filter formulation. The Kalman filter is derived from a multichannel (MC) extension of the scalar dynamic autoregressive (AR) model assuming a slowly time varying environment. This permits the use of a random walk model for the process dynamics. By exploiting the structural relationship between the state vector of the MC Kalman filter and its output mean squared error, a computationally efficient structure is formed. Simulation results are presented which show, under a randomly time-varying source location scenario, that the MC Kalman filter generates CSDM estimates which yield high-resolution bearing estimation performance.