Detection of the number of signals and estimation of their directions of arrival (DOAs) are fundamental problems in signal array processing. In this paper we present three main contributions to this problem.First, we show that there is an explicit relation between model selection and the breakdown phenomena of the Maximum Likelihood estimator (MLE). Second, we provide a simple formula for the location of the breakdown of the MLE, using tools from extreme value theory. This gives an explicit formula for the signal strength required for reliable detection. Third, we apply these results and propose a new joint detection-estimation algorithm with state-of-the-art performance. We present computer simulations that demonstrate the performance of our algorithm, compared to other popular methods.
Index Termsjoint detection-estimation, statistical hypothesis tests, extreme value theory, MLE breakdown phenomena.