Separation of radio-frequency interference from SAR signals via dictionary learning

Abstract

This paper investigates the concept of employing dictionary learning to adaptively capture the principal components of radio-frequency interference (RFI) in order to improve the efficiency as well as robustness of RFI extraction/ separation from ultra-wideband (UWB) synthetic aperture radar (SAR) signals. Interference sources here pose critical challenges for UWB systems since RFI might have significant bandwidth overlap with the spectrum of our SAR signals of interest. Moreover, RFI components often exhibit nonstationary, time-varying characteristics; hence, they are difficult to predict and model accurately. In this paper, we propose a dictionary learning framework to adaptively capture RFI components within a local spatiotemporal neighborhood. We then exploit these learned RFI dictionaries to improve the noise-source separation process via sparse recovery: interferences are represented as a sparse subspace spanned by a few dictionary atoms while SAR signals are modeled as a sparse linear combination of time-shifted transmitted pulses. We validate the effectiveness and illustrate the robustness of our proposed framework via extensive RFI-SAR separation results, using both simulated data and real-world data collected by the U.S. Army Research Laboratory (ARL) UWB BoomSAR system.