Deceiving Reactive Jamming in Dynamic Wireless Sensor Networks: A Deep Reinforcement Learning Based Approach

A reactive jamming attack, which performs spectrum jamming only during legal signal transmission based on the knowledge of sensor behaviors, poses a significant threat to wireless sensing networks (WSNs). In this paper, a novel deceiving approach is proposed for defending reactive jamming in dynamic WSNs. Specifically, when the maximum transmission power is given, we first formulate the anti-jamming process as an optimization problem to maximize the average received power while eliminating the effects of the jamming attack. Then the interaction between reactive jamming and legitimate sensors is modeled with the Markov decision process (MDP). Finally, a deep Q network (DQN) based jamming deceiving method is proposed to solve the formulated optimization problem. Simulation results show that the proposed anti-jamming scheme can converge quickly and is superior to the classical counterparts in terms of the mean of received signal power.