High-Resolution Neural Network Processing of LFM Radar Pulses

Abstract

In radar applications, the bandwidth of a transmitted pulse determines the range resolution and the ability to disclose densely spaced targets. The processing of radar signals is often carried out through matched filtering (MF) which aims to maximize the signal to noise ratio. This work presents an alternative processing scheme for oversampled radar signals based on small-sized neural networks. The networks are trained with an objective to return MF outcomes corresponding to a higher bandwidth pulse. The article demonstrates how such a neural network design can be constructed and compares against traditional processing and detection.