Recurrent-Neural-Network-Based Anti-Jamming Framework for Defense Against Multiple Jamming Policies

Conventional anti-jamming methods mainly focus on preventing single jammer attacks with an invariant jamming policy or from multiple jammers similar policies. These are ineffective against a following several different policies distinct Therefore, this article proposes method that can adapt its to the current attack. Moreover, for scenario, estimates future occupied channels using jammers’ in previous time slots is proposed. In both and scenarios, interaction between users modeled recurrent neural networks (RNNs). The performance of proposed evaluated by calculating users’ successful transmission rate (STR) ergodic (ER), compared baseline based deep $Q$ -learning (DQL). Simulation results show all considered perfectly detected high STR ER maintained. when 70% spectrum under jammers, achieves greater than 75% 80%, respectively. values reach 90% 30% attacks. addition, significantly outperform DQL scenarios.