Deep-Learning based Spectrum Prediction for Cognitive Automotive Radar Interference Mitigation

Interference affects the robustness of driver assistance systems that rely on automotive radars. Today's radar sensors use algorithms for suppression of interference in the received radar signal. In contrast, interference-aware cognitive radar (IACR) avoids interference based on spectral awareness, as opposed to suppressing it in post-processing. In this paper, we propose a machine learning (ML) based spectrum prediction model for IACR that infers the spectral occupation in the upcoming measurement cycle. Subsequently, interference is avoided by adapting the waveform to use the spectral resources with the least interference potential. By closing the cognitive loop, we show that IACR is capable of achieving an order of magnitude less average interference power compared to radars without a cognitive interference avoidance strategy.