Machine learning and game-theoretic approaches for dynamic spectrum evasion against intelligent adversarial jammers.
Hostile jammers dynamically adapt between barrage, sweeping, and reactive pulse modes. Deep Q-networks classify jammer policies within three pulse repetition intervals, selecting optimal counter-measures.
Formulating the spectrum competition as a Stackelberg game enables the ISAC base station to act as the leader, intentionally probing jammers to trap them into predictable, nullified frequency bands.
Live testing under 50-watt directional noise jammers proved continuous video link streaming and sub-meter target tracking, with zero lost frames during aggressive frequency hopping transitions.