Adaptive Bandwidth Radar for UAV Swarm Detection Using Reinforcement Learning

The problem of detecting and resolving UAV swarms using radar systems is considered in this paper. Conventional FMCW radars operating with fixed waveform configurations are limited by the trade-off between range resolution and spatial coverage within the maximum unambiguous range. To address this, an adaptive bandwidth selection approach based on Proximal Policy Optimization (PPO) is proposed within the cognitive radar framework. The radar adjusts its transmitted bandwidth on a per-CPI basis using closed-loop feedback to improve swarm detectability. The approach is validated using a dedicated FMCW radar simulator with realistic target motion and detection modeling. Results show that the learned policy via PPO consistently outperforms fixed-bandwidth baselines and approaches optimal-level performance (i.e., that achievable by access to ground-truth information) across multiple swarm scenarios.