International Conference on Communications
Efficient Quantum Soft Actor-Critic Model for Dynamic Spectrum Sharing in Intelligent O-RAN
Vu-Hai Nguyen, Yared Abera Ergu, Ren‐Hung Hwang, Trung Q. Duong, Van-Linh Nguyen
Abstract
The Open Radio Access Network (O-RAN) architecture enables intelligent Dynamic Spectrum Sharing (DSS), although state-of-the-art deep reinforcement learning (DRL) methods have demonstrated remarkable practical results, their integration with quantum computing brings greater potential for enhancements in performance. In this study, we present Quantum Soft Actor-Critic (QSAC), a novel hybrid quantum-classical methodology aimed at tackling barren plateaus and demonstrating enhanced performance compared to conventional DRL benchmarks, including Soft Actor-Critic (SAC) and Twin Delayed Deep Deterministic Policy Gradient (TD3), achieving a higher demand satisfaction ratio and markedly improved ϵ-band compliance rates under demanding conditions. Additionally, we present a comprehensive architectural design for the effective integration of our framework into the O-RAN non-real-time and near-real-time RICs. Our quantum-enhanced actor significantly reduces the number of trainable actor parameters by over 99% compared to its conventional SAC equivalent (from around 73,000 to merely 377), indicating an important improvement in model efficiency. To promote repeatability and facilitate future research, we offer the open-source implementation of our study.