Collaborative Anti-jamming in Cognitive Radio Networks Using Minimax-Q Learning

Cognitive radio is an efficient technique for realization of dynamic spectrum access. Since in the cognitive radio network (CRN) environment, the secondary users (SUs) are susceptible to the random jammers, the security issue of the SU’s channel access becomes crucial for the CRN framework. The rapidly varying spectrum dynamics of CRN along with the jammer’s actions leads to challenging scenario. Stochastic zero-sum game and Markov decision process (MDP) are generally used to model the scenario concerned. To learn the channel dynamics and the jammer’s strategy the SUs use reinforcement learning (RL) algorithms, like Minimax-Q learning. In this paper, we have proposed the multi-agent multi-band collaborative anti-jamming among the SUs to combat single jammer using the Minimax-Q learning algorithm. The SUs collaborate via sharing the policies or episodes. Here, we have shown that the sharing of the learned policies or episodes enhances the learning probability of SUs about the jammer’s strategies but reward reduces as the cost of communication increases. Simulation results show improvement in learning probability of SU by using collaborative anti-jamming using Minimax-Q learning over single SU fighting the jammer scenario. Index Terms — Cognitive radio networks, Stochastic game theory, Collaborative games, Markov decision process, Reinforcement learning