PPO based Real-Time Beamforming and Power Splitting in Near-Field SWIPT Systems

Abstract

Near-field wireless communication in millimeter-wave and terahertz systems unlocks high-capacity and energy-efficient connectivity but also introduces new challenges in beamforming and power control due to spherical wavefronts and spatial non-stationarity. This paper presents a near-field constraint-aware proximal policy optimization (NF-CA-PPO) framework for simultaneous wireless information and power transfer (SWIPT) in a multi-transmitter environment. The proposed learning framework formulates joint beamforming and power-splitting optimization as a Markov Decision Process (MDP) and integrates constraint satisfaction directly into the policy learning stage through projection-based and penalty-aware updates. Unlike conventional convex optimization or standard deep reinforcement learning (DRL) methods, NF-CA-PPO guarantees physical feasibility, improves sample efficiency, and achieves real-time adaptability under near-field channel variations. Simulation results show that NF-CA-PPO reduces total transmit power and attains faster, more stable convergence than baseline (deep deterministic policy gradient), while consistently satisfying signal-to-interference-plus-noise ratio (SINR) and energy-harvesting constraints. © 2025 IEEE.