Deep Q-Network Based Resilient Drone Communication:Neutralizing First-Order Markov Jammers

TL;DR

This paper presents a deep reinforcement learning approach using Deep Q-Networks to develop resilient drone communication systems that can neutralize reactive jammers employing first-order Markov strategies, enhancing robustness in electronic warfare.

Contribution

The study introduces a novel DRL-based frequency hopping method that learns to counter reactive jamming using transition statistics, demonstrating effective neutralization in complex radio environments.

Findings

The DQN-based transmitter learns a near-random frequency hopping policy.

Moderate error correction codes significantly reduce packet loss.

The approach effectively neutralizes first-order reactive jammers.

Abstract

Deep Reinforcement Learning based solution for jamming communications using Frequency Hopping Spread Spectrum technology in a 16 channel radio environment is presented. Deep Q Network based transmitter continuously selects the next frequency hopping channel while facing first order reactive jamming, which uses observed transition statistics to predict and interrupt transmissions. Through self training, the proposed agent learns a uniform random frequency hopping policy that effectively neutralizes the predictive advantage of the jamming. In the presence of Rayleigh fading and additive noise, the impact of forward error correction Bose Chaudhuri Hocquenghem type codes is systematically evaluated, demonstrating that even moderate redundancy significantly reduces packet loss. Extensive visualization of the learning dynamics, channel utilization distribution, epsilon greedy decay, cumulative reward, BER and SNR evolution, and detailed packet loss tables confirms convergence to a near optimal jamming strategy. The results provide a practical framework for autonomous resilient communications in modern electronic warfare scenarios.