Cooperative Proactive Eavesdropping over Two-Hop Suspicious Communication Based on Reinforcement Learning

Legitimate surveillance has attracted more and more concern, and effective proactive intervention can eavesdrop the illegitimate information. In this paper, we propose legitimate eavesdropping over a two-hop suspicious communication link by two full-duplex legitimate monitors (LMs) based on multi-agent deep deterministic policy gradient (MADDPG) algorithm in two phases. In phase 1, the suspicious transmitter sends information to the suspicious assistant relay, and the assistant relay decodes and forwards the received message to the suspicious receiver in phase 2. Meanwhile, two LMs cooperatively emit jamming to suspicious relay and receiver during each phase. Particularly, each LM is considered to be an energy-limited device, and eavesdropping is a long-term process, so we adopt expected eavesdropping energy efficiency (EEE) over a period of time to evaluate eavesdropping performance. However, for two LMs, how to cooperatively make jamming power decision at each hop in a dynamic environment is a huge challenge. Therefore, MADDPG algorithm, as a multi-agent reinforcement learning approach with the advantage of dynamic decision-making, is utilized to solve the issue of jamming power decision for each LM. In the simulation, the results show that our proposed cooperative jamming scheme can obtain higher expected EEE.