Reinforcement Learning for Cognitive Radar Task Scheduling

Cognitive Radars (CRs) have the capability to adapt to their environment and accumulate knowledge from their interactions with the environment. This paper deals with the Radar Resource Management (RRM) problem where the radar assigns limited time resources to a set of tasks. The problem is modeled as an optimization problem where the aim is to minimize the number of delayed and dropped tasks which is an NP-hard problem. We propose a modified Monte Carlo Tree Search (MCTS) approach to find an effective solution. We further develop a Reinforcement Learning (RL) solution that uses a Neural Network (NN) to guide the modified MCTS. This produces a stable RL algorithm that learns on its own, requires no external training data, and can adapt to a varying environment. The results show the proposed RL algorithm outperforms other techniques including commonly used heuristics and produces close to optimal results.