Fusion Model Using a Neural Network and MLE for a Single Snapshot DOA Estimation with Imperfection Mitigation

In this paper, we discuss the fusion of deep neural networks with the Maximum Likelihood Estimation (MLE) algorithm for estimating both the Direction Of Arrival (DOA) of MIMO radars and the number of sources for a single-snapshot scenario using a Minimum Redundancy antenna Array (MRA) with imperfections and evaluating it with synthetic and real-world data. The combination of Deep Learning (DL) with the classical MLE seems to be a viable solution for this problem, as it is less computationally intensive than a classical MLE while not losing generalization and being better at estimating the number of sources. In both our experiments, using synthetic and real-world data, our method performs close to MLE and appears to be a deployable solution for real scenarios. Besides reducing the computational load of MLE, the novelty of our model lies in the fact that the deep learning model learns the gain and phase errors as a function of the direction of arrival instead of applying a simple calibration.