Anti-Jamming Communications Using Spectrum Waterfall: A Deep Reinforcement Learning Approach

• Published in: IEEE communications letters Vol. 22; no. 5; pp. 998 - 1001
• Main Authors: Xin Liu, Yuhua Xu, Luliang Jia, Qihui Wu, Anpalagan, Alagan
• Format: Journal Article
• Language: English
• Published: IEEE 01.05.2018
• Subjects: Aerodynamics; Anti-jamming; Convolutional neural networks; deep Q-network; deep reinforcement learning; Heuristic algorithms; Jamming; Machine learning; Signal to noise ratio; Time-frequency analysis
• ISSN: 1089-7798; 1558-2558
• DOI: 10.1109/LCOMM.2018.2815018

This letter investigates the problem of anti-jamming communications in a dynamic and intelligent jamming environment through machine learning. Different from existing studies which need to know (estimate) the jamming patterns and parameters, we use the temporal and spectral information, i.e., the spectrum waterfall, directly. First, to cope with the challenge of infinite state of spectrum waterfall, a recursive convolutional neural network is designed. Then, an anti-jamming deep reinforcement learning algorithm is proposed to obtain the optimal anti-jamming strategies. Finally, simulation results validate the proposed approach. The proposed algorithm does not need to model the jamming patterns, and naturally has the ability to explore the unknown environment, which implies that it can be widely used for combating dynamic and intelligent jamming.