Augmenting Radar Doppler Resolution with Neural Networks

Abstract

In order to generate high-resolution Doppler profiles a radar needs to emit a large number of pulses within a coherent processing interval (CPI). For a radar operating in demanding scenarios it can be difficult to sustain a long CPI across search directions. In this work, an application of small neural networks is proposed to augment the Doppler resolution beyond the one detailed by basic radar parameters. A specific neural network structure is proposed which can be trained to operate on complex valued time-domain data and yield a frequency transformed output with an increased Doppler bin resolution. It is shown that by making use of these techniques a radar can improve its ability to detect targets and to distinguish closely spaced targets. A limited increase in the Doppler bin resolution can be sustained with little to no negative impact on the false alarm rate.