Transfer Learning Augumented PPO-Based Resource Allotment for Hybrid VLC/RF Systems

Hybrid visible light communication (VLC) and radio frequency (RF) systems have gained significant attention for next-generation wireless networks owing to their ability to deliver high throughput and uninterrupted connectivity. Determining optimal resource allocation in such hybrid architectures, however, leads to a highly non-convex optimization problem that becomes increasingly complex in large and time-varying network scenarios. To overcome these shortcomings, this work introduces a two-stage mechanism that uses proximal policy optimization (PPO) assisted transfer learning for adaptive and efficient resource management in largescale hybrid VLC/RF systems. In the first stage, a PPO-based reinforcement learning agent learns a robust policy that ensures stable convergence even under rapidly changing link conditions. In the second stage, transfer learning uses previously trained policy parameters to quickly adapt to newly arriving mobile users, significantly minimizing retraining overhead. Extensive simulations confirm that the proposed method reduces convergence iterations by nearly 72 % compared with deep Q-network (DQN)-based approaches, while maintaining higher achievable sum-rate and stronger adaptability to network dynamics.