Avoiding Interference in Multi-Emitter Environments: A Reinforcement Learning Approach

摘要

In this paper, we investigate the interference avoidance performance of a single phased-array radar in a multi-emitter environment under a partially observable Markov decision process (POMDP) model. Considering deep Q-network (DQN) and long short term memory (LSTM) networks utilizing deep reinforcement learning algorithm, in the proposed solution, the radar learns how to predict that a certain beam position will incur interference in the next time slot according to its past observations of the estimated direction of arrivals (DoAs). This allows the radar to proactively choose a highly probable vacant beam position. Simulation results show that the proposed neural networks successfully predict vacant beam positions. This significantly reduces the average probability of interference up to 0.1%, which ensures a high reliability in radar systems.