Deep Learning Based Source Direction Estimation with Magnitude-only Array Measurements

Most DOA estimation techniques require phase information of the received array signals to accurately estimate the direction of arrival (DOA). Nevertheless, in some scenarios, the phase information may not be easily accessible or reliable due to various reasons such as hardware limitations or calibration issues. One way to tackle this challenge is to discard the phase information or only measure the magnitude of the received signals. In this work, a deep-learning (DL) based DOA estimation method is proposed for effective DOA estimation with magnitude-only measurements. To improve the generalization ability and robustness of the proposed solution, an attention mechanism is employed, and to avoid the implicit assumption that the number of signals is known in the training process, labels of the data are converted into the one-hot form. Simulation results show that the proposed solution has superior performance in terms of computational complexity, accuracy, and robustness compared to traditional DOA estimation algorithms.