Radio Frequency Interference (RFI) Detection in Instrumentation Radar Systems: a Deep Learning Approach

Transient broadband Radio Frequency Interference (RFI) is a pervasive problem for measurement radars instrumenting outdoor test ranges. These noise signals manifest themselves unpredictably in measurement data and must be detected and removed to preserve overall data quality. Large data volumes make the manual identification of RFI intractable over the long term, but trained human experts can successfully identify RFI when the data is represented visually. Convolutional Neural Networks (CNNs) have been used to some success in mimicking human visual inspection across a wide range of disciplines, and are recently applied to RFI detection in the closely related field of radio astronomy. We present a CNN architecture which is resistant to overfitting, trains quickly, and exhibits excellent performance in flagging RFI-contaminated data in a range instrumentation radar system (Receiver Operating Characteristics Area Under Curve (AUC) = 0.998, Precision-Recall AUC = 0.998).