A new method for source detection, power estimation, and localization in large sensor networks under noise with unknown statistics

Most statistical inference methods for array processing assume an array of size N fixed and a number of snapshots T large. In addition, many works are based on the assumption of a white noise model. These two assumptions are increasingly less realistic in modern systems where N and T are usually both large, and where the noise data can be correlated either across successive observations or across the sensor antennas. In this paper an approach to handle this kind of scenario is presented. New algorithms for source number estimation, power estimation, and localization by a sensor array under noise with unknown correlation model are proposed. The results fundamentally rely on recent advances in small rank perturbations of large dimensional random matrices.