A Novel On-Policy DRL-Based Approach for Resource Allocation in Hybrid RF/VLC Systems

Abstract

Visible light communication (VLC) has emerged as a promising technology, delivering high-speed data transmission for 5G and beyond communication. Nevertheless, its susceptibility to blockages demands a co-deployment with traditional radio frequency (RF) systems to ensure uninterrupted connectivity. This co-deployment, known as a hybrid RF/VLC system, is a subset of heterogeneous networks (HetNets) and offers interoperability, energy efficiency, and optimal resource utilization. In hybrid RF/VLC, efficient resource allocation and load balancing are crucial. Existing Deep Q-Network (DQN) learning-based methods designed to address these issues, fail in large and dynamic environments. Our present study investigates alternative approaches for optimal resource allocation and load balancing in dynamic and large hybrid RF/VLC systems, to achieve maximum data rates for users. We propose two model-free on-policy deep reinforcement learning (DRL) based schemes, namely advantage actor-critic (A2C) and proximal policy optimization (PPO), for efficient resource allocation in hybrid RF/VLC. Simulation results show that the A2C and PPO based schemes outperform the DQN learning scheme by 31.3% and 32.5%, respectively, in terms of data rates. The proposed schemes also outperform the deep deterministic policy gradient (DDPG) in data rate maximization by up to 8.1% and 9.7%, respectively.