Cooperative Multi-UAV Dynamic Anti-Jamming Scheme with Deep Reinforcement Learning

Key challenges for manned/unmanned aerial vehicles(MAV/UAV) cooperative operation with distributed command and control (C2) structure network face are the assignment of spectrum and the resilience against interference. In response, we propose a cooperative multi-UAV dynamic anti-jamming (CMDA) approach that, in contrast to previous work, can adapt to scenario with fast channel variations due to malicious sweep jamming in highly dynamic environment. In this work, the multi-UAV anti-jamming problem is translated into a decentralized partially observable Markov decision process (Dec-POMDP), which is solved through multi-agent reinforcement learning (MARL) method. Specifically, every autonomous UAV acts based on different partial observations about the environment and updates local Q-network according to the common reward individually. The objective is to maximize the global reward and approximate the optimal power control and spectrum allocation policy without prior knowledge of the time-varying wireless channel state in jamming environments. Finally, the simulation results demonstrate that proposed scheme provides lower average packet loss rate (PLR) and better jamming resistance than previous approaches.