Deep Learning Based Game-Theoretical Approach to Evade Jamming Attacks

Software-defined radios (SDRs) with substantial cognitive (computing) and networking capabilities provide an opportunity for malicious individuals to jam the communications of other legitimate users. Channel hopping is a well known anti-jamming tactic used in order to evade jamming attacks. We model the interaction between a transmitter, who uses chaotic pseudo-random patterns for channel hopping, and a sophisticated jammer, who uses advanced machine learning algorithms to predict the transmitter’s frequency hopping patterns as a non-cooperative security game. We investigate the effectiveness of adversarial distortions in such a scenario to support the anti-jamming efforts by deceiving the jammer’s learning algorithms. The optimal strategies in the formulated game indicate how adversarial distortions should be used by the players at every step of the game in order improve their outcomes. The studied jamming/anti-jamming scenario combines chaotic time series generators, game theory, and online deep learning.