DRL-based Power Allocation in LiDAL-Assisted RLNC-NOMA OWC Systems

TL;DR

This work tackles the challenge of CSI imperfection and residual SIC errors in dense indoor optical wireless networks implementing NOMA. It introduces a LiDAL-assisted RLNC-NOMA system and leverages a Normalised Advantage Function (NAF) based DRL to learn near-optimal continuous power allocations, aiming to maximize the average sum rate under localization-derived CSI errors. The proposed approach achieves performance close to exhaustive search, outperforms DDPG and GRPA in convergence and sum-rate, and demonstrates robustness to location errors and environmental noise. The results indicate practical potential for dynamic, high-density OWC deployments where precise CSI is hard to obtain.

Abstract

Non-orthogonal multiple access (NOMA) is a promising technique for optical wireless communication (OWC), enabling multiple users to share the optical spectrum simultaneously through the power domain. However, the imperfection of channel state information (CSI) and residual errors in decoding process deteriorate the performance of NOMA, especially when multi-parameteric and realistic dense-user indoor scenarios are considered. In this work, we model a LiDAL-assisted RLNC-NOMA OWC system, where the light detection and localization (LiDAL) technique exploits spatio-temporal information to improve user CSI, while random linear network coding (RLNC) enhances data resilience in the NOMA successive decoding process. Power allocation (PA) is a crucial issue in communication systems, particularly in the modeled system, due to the complex interactions between multiple users and the coding and detection processes. However, optimizing continuous PA dynamically requires advanced techniques to avoid excessive computational complexity. Therefore, we adopt a deep reinforcement learning (DRL) framework to efficiently learn near-optimal power allocation strategies, enabling enhanced system performance. In particular, a DRL-based normalized advantage function (NAF) algorithm is proposed to maximize the average sum rate of the system, and its performance is analyzed and compared to other widely used DRL-based and conventional PA schemes, such as deep deterministic policy gradient (DDPG), gain ratio PA (GRPA), and exhaustive search.

Figures (8)

• Figure 1: LiDAL-based RLNC-NOMA system model.
• Figure 2: Schematic architecture of RL framework.
• Figure 3: DRL-based NAF neural networks architecture.
• Figure 4: DRL-based NAF vs DDPG learning convergence.
• Figure 5: DRL-based NAF vs DDPG loss function evaluation.