Radar Signal Denoising for ISAR Imaging Using Complex-valued Neural Network

Neural networks are increasingly being applied in radar imaging. However, the application for imaging or denoising typically considers only two-dimensional data pairs. This paper examines the employment of complex-valued neural networks (CVNN) on one dimensional time series data and considers the reconstruction of ISAR images while obtaining noise reduction. For this purpose, a complex-valued convolutional neural network (CV-CNN) is proposed where the use of a phase stability loss is demonstrated. Additionally, this paper provides a deeper look into complex-valued activation functions. Numerical evaluation results show the performance of the models in terms of the Mean Squared Error (MSE) and the Unwrapped Instantaneous Phase Mean Absolute Error (UIPMAE), while graphically it can be shown that CV-CNNs can successfully denoise time signals for the reconstruction of ISAR images. Also, denoising was observable, leading to clearer ISAR images in which targets could be better identified.