Single-Snapshot Direction-of-Arrival Estimation of Multiple Targets using a Multi-Layer Perceptron

Abstract

An alternative approach to high-resolution direction-of-arrival estimation in the context of automotive FMCW signal processing is shown by training a neural network with simulation as well as experimental data to estimate the mean and distance of the azimuth angles from two targets. Testing results are post-processed to obtain the estimated azimuth angles which can be validated afterwards. The performance of the proposed neural network is then compared with a reference implementation of a maximum likelihood estimator. Final evaluations show super-resolution like performance with significantly reduced computation time, which is expected to have an impact on future multi-dimensional high-resolution DoA estimation.