Learning-based One bit DoA Estimation with Single Snapshot

Direction of Arrival (DoA) estimation is one of the important and practical problems in the context of array signal processing. One-bit quantization of measurements using a simple comparator can signi cantly reduce computational complexity, cost, and power consumption. In this paper, the DoA estimation problem with a single snapshot isexamined based on both unquantized and also one-bit quantized measurements using an Arti cial Neural Network (ANN), when either raw measurements or the calculated sample covariance matrix is fed to the input layer. Scenarios with one, two, or three possible sources are considered. We show that under the conditions mentioned above, the learning-based DoA regression outperforms both the subspace-based MUSIC and the compressed sensing-based Complex Binary Iterative Hard Thresholding (CBIHT) methods. Moreover, it is preferable to simply use raw measurements as the input of the ANN, compared to calculation of the sample covariance matrix.