DRL-based power allocation in LiDAL-assisted RLNC-NOMA OWC systems

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, imperfect channel state information (CSI) and residual decoding errors deteriorate NOMA performance, especially in realistic dense-user indoor scenarios. In this work, we model an OWC system that integrates light detection and localization (LiDAL) and random linear network coding (RLNC) within a NOMA framework. LiDAL exploits spatio-temporal information to improve user CSI, while RLNC enhances data resilience in the successive decoding process, resulting in a LiDAL-assisted RLNC-NOMA OWC system. Power allocation (PA) is crucial in this system due to complex interactions between multiple users and the coding and detection processes, but optimizing continuous PA dynamically can be computationally prohibitive. To address this, we adopt a deep reinforcement learning (DRL) framework to efficiently learn near-optimal PA strategies. In particular, a DRL-based normalized advantage function (NAF) algorithm is proposed to maximize the average sum rate, and its performance is compared to deep deterministic policy gradient (DDPG), gain ratio PA (GRPA), and exhaustive search. The results indicate that NAF closely matches exhaustive search, is 39% faster than DDPG, and improves the average sum rate by 4.6% over GRPA, while accounting for user location estimation errors.