Non-stationary Bayesian direction of arrival estimation with drifting sensor locations

We describe a new tracking algorithm for the direction of arrival estimation problem where both the locations of the sensors in the array and the directions of arrival are nonstationary. The approach taken is Bayesian. The algorithm assumes that the filtering distribution is approximately Gaussian and maintains the mean and covariance of this approximation by fitting a quadratic surface to the log posterior around the location where the log posterior is maximized. In the case where the sensor locations are stationary, the algorithm is shown to have similar performance to particle filter based algorithms but at a reduced computational cost. In the case where the sensor locations are non-stationary particle filtering is unsuccessful and the new algorithm performs significantly better than currently existing algorithms.