Denoising of Radar Pulse Streams With Autoencoders

Abstract

There are many cases in which the noise corrupts the signals in a significant manner. To better analyze these signals, the noise must be removed from the signals for further data analysis, and the process of noise removal is referred to as denoising. In this letter, we propose a novel approach to the pulse denoising problem by extracting features from time of arrival (TOA) sequences using the autoencoders. The noise-contaminated TOA sequence is first coded into a binary vector and then fed into an autoencoder for training. Then, the trained autoencoder is capable of generating the original TOA sequence without lost and spurious pulses. Moreover, the proposed method does not require a noise-free TOA sequence as a priori as with conventional autoencoders. Simulation results show that the new technique can deal with TOA sequences with complex pulse repetition interval (PRI) modes that have not been tackled before. In addition, the proposed method has a better performance in noisy environments than conventional methods and general deep neural network structures.