DEEP-MLE: FUSION BETWEEN A NEURAL NETWORK AND MLE FOR A SINGLE SNAPSHOT DOA ESTIMATION

In this paper, we propose a novel framework called Deep-MLE, which gives a solution to the single-snapshot Direction Of Arrival (DOA) estimation problem, up to 4 distinct targets, using a radar equipped with a Minimum Redundancy antenna Array (MRA). This framework works by fusing a Deep Learning (DL) technique - 1D Residual Neural Network (1D ResNet) - with a classical DOA algorithm - Maximum Likelihood Estimation (MLE). By combining two very different approaches, we can address some of their limitations, such as the computational complexity of MLE. On the other hand, our proposed Deep-MLE uses MLE to correct, to some degree, the estimations made by the Neural Network (NN). The results from our framework are promising as it seems to be a viable solution to the DOA estimation problem, having a better performance than models using pure MLE or NN.