Flexible Channel Access Against Unknown Dynamic Jamming Attack: A Reinforcement Learning Approach.

This paper investigates multi-domain anti-jamming communication in a malicious jamming and channel fading environment using deep reinforcement learning (DRL). Unlike existing research, which mainly focuses on single-domain anti-jamming methods such as frequency selection and power control, multi-domain anti-jamming is considered to cope with dynamic jamming and channel fading in the cognitive radio communication scene. We consider whether to transmit at the current time slot in the time domain and joint channel-bandwidth selection in the frequency domain. To represent the environment state, A two-dimensional time-frequency spatial-spectral matrix is designed. The channel switching and bandwidth adjustment overhead are considered in the instant reward to avoid frequent channel switching and bandwidth adjustment. We propose an intelligent anti-jamming algorithm based on DRL and long short term memory network (LSTM) layer is introduced into the DRL network. Simulation results demonstrate that our proposed algorithm is adaptive in dynamic jamming environments and outperforms other anti-jamming algorithms.