Recursive `ML´ bearing estimation: initialization and sources number update

Maximum-likelihood (ML) and approximate ML may be considered as the upper state of the art in high-resolution methods, but they suffer from initialization of the sources´ number and position. Starting from a crude initialization with a low-resolution method, the authors propose a time recursive method for simultaneous update of the sources´ number and location. For the current estimate of the sources´ number the algorithm computes the best ML position estimate over the past observations. The corresponding signal is subtracted from the observations, and the residue is tested for the noise-only hypothesis. If the test fails, the sources´ number is incremented, a new initialization is provided, and ML estimation proceeds. Emphasis is on the stationary case