Restless Bandits for Metacognitive Spectrum Sharing With Software Defined Radar

• Published in: IEEE transactions on radar systems 2023, Vol.1, p.401-412
• Main Authors: Kovarskiy, Jacob A., Martone, Anthony F., Narayanan, Ram M., Sherbondy, Kelly D.
• Format: Article
• Language: English
• Subjects: Cognitive radar; dynamic spectrum access; Long Term Evolution; LTE; Metacognition; Radio frequency; Real-time systems; Reinforcement learning; Software radio
• ISSN: 2832-7357; 2832-7357
• DOI: 10.1109/TRS.2023.3306309

Recent advances in cognitive radar technologies have proposed solutions to the problem of radio frequency (RF) spectral congestion. Different strategies for radar spectrum sharing have been demonstrated to perform more efficiently in particular RF interference (RFI) conditions. These RFI conditions can change abruptly based on rapidly varying demands for spectral resources, thus degrading spectrum sharing efficiency when using one single strategy. This paper presents a "metacognition" framework to enable a cognitive radar to adapt strategies for spectrum sharing as RFI scenarios change. Here, our metacognition framework employs a restless bandit model to select from three radar spectrum sharing strategies. In prior work, these three strategies have been individually implemented on software defined radios (SDRs) and characterized in terms of spectrum sharing efficiency. Our manuscript compares the spectrum sharing efficiency of restless bandit algorithms to select from three known cognitive radar implementations to optimize performance in changing RFI conditions. Experiments consist of a real-time SDR implementation of this metacognition framework with emulated RFI signals that change over time. Using SW UCB1-tuned shows up to 8 dB improvement in average signal-to-interference plus noise ratio (SINR) and up to 13 MHz improvement in average bandwidth utilization compared to worst-case from the considered restless bandit algorithms.