Robust Deep Reinforcement Learning for Interference Avoidance in Wideband Spectrum

This paper presents a design of a cognitive engine for interference and jamming resilience based on deep reinforcement learning (DRL). The proposed scheme is aimed at finding the spectrum opportunities in a heterogeneous wideband spectrum. In this paper we discuss a specific DRL mechanism based on double deep Q-learning (DDQN) with a convolutional neural network (CNN) to successfully learn such interference avoidance operation over a wideband partially observable environment. It is shown, through simulations, that the proposed technique has a low computational complexity and significantly outperforms other techniques in the literature, including other DRL-based approaches.