Multi-Modal Deep Reinforcement Learning Framework for Interference-Limited RIS-Assisted UAV Wireless Network

In this work, we investigate minimization of power consumption in interference limited reconfigurable intelligent surface (RIS)-assisted unmanned aerial vehicle (UAV) networks. We formulate a power minimization optimization problem with interference mitigation and quality-of-service (QoS) constraints. The problem is nonconvex; therefore, we consider deep reinforcement learning (DRL) framework to handle it. To build the agents in the proposed multi-modal DRL, multi-modal neural networks are used where two types of features are extracted: deep modality features and shallow modality features. These features are then concatenated to perform the prediction. The proposed multi-modal DRL leverages the concept of multi-modality feature design to improve the prediction accuracy while maintaining the inference speed. We show via simulation that our proposed MM-DRL achieves better performance compared to the existing algorithm.