Online Antenna Selection For Enhanced Doa Estimation

The performance of direction of arrival (DOA)estimation using antenna arrays is fundamentally limited by the Cramer-Rao bound (CRB), which is intimately tied to the array configuration. In systems where a subset of antennas can be selected from a larger array, the array configuration can be recruited to enhance the DOA estimation performance by choosing the optimal subarray that minimizes the CRB. This strategy has been demonstrated to provide performance improvements, albeit at a substantial computational cost. Here, we leverage the power of unsupervised learning to reduce the computational cost of antenna selection for enhanced DOA estimation resulting in a practical online implementation of the array reconfiguration. We formulate the changing DOA estimation problem as a game in the context of online convex optimization, and employ a gradient-based technique that makes a move at each step in order to minimize the total loss after T steps. We compare the performance of the proposed method with the exhaustive search and a Dinkelbach-type algorithm that provides an approximate solution. We show that the proposed method is able to provide a solution that is close to the exhaustive search at a fraction of the computational time, thus permitting online implementation of the selection strategy.