Decision Making for Communication Anti-Jamming Tasks with Knowledge-Graph-Based Q-Learning

Abstract

Due to the severe threats posed by smart jammers, anti-jamming decision making has become an essential technology for wireless communications. Most of the existing anti-jamming decision-making approaches have adopted Q-Learning to improve accuracy. However, the performances of these approaches drop dramatically in fast-varying jamming environments. Thus, an advanced Q-Learning approach utilizing domain knowledge graph as prior knowledge is proposed to select the optimal strategies with high flexibility and accuracy in different jamming environments. Specifically, by taking a knowledge graph that contains anti-jamming knowledge to initialize the Q-table, Q-Learning can avoid becoming stuck at local suboptimal solutions and obtain accurate strategies with fewer iterations. The iterations of the proposed approach are one third of those of other approaches based on Q-Learning and the average rewards of the proposed approach have improved by 2 percent. Numerical results demonstrate the optimality and excellent performance of the proposed approach over various existing benchmarks.