Signal Detection with a Digital Antenna Array using Machine Learning

Abstract

Radars and radar warning receiver systems engage in a perpetual battle for superiority. This constant struggle demands the development of new methods, and recent advancements in machine learning and image recognition may serve as sources of inspiration. In this thesis, a “You Only Look Once”-style neural network is designed to detect pulsed radar signals and estimate their angle of arrival, frequency, time of arrival, and pulse width. Employing a digital multi-channel uniform linear antenna array, a short-time Fourier transform is applied to the output from each antenna element. Afterward, beamforming is performed for various angles combining the spectrograms, producing a 3D image upon which image recognition is applied. We find that using simulated data this method works for detecting two sinusoidal signals, even in noisy environments.