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

Area of Science:

• Optical Wireless Communication (OWC)
• Wireless Networks
• Signal Processing

Background:

• Non-orthogonal multiple access (NOMA) enhances optical wireless communication (OWC) by enabling simultaneous user access via power domain sharing.
• Imperfect channel state information (CSI) and decoding errors degrade NOMA performance, particularly in dense indoor user scenarios.

Purpose of the Study:

• To develop a LiDAL-assisted RLNC-NOMA OWC system for improved performance.
• To address the computational challenges of dynamic power allocation (PA) optimization in complex NOMA systems.
• To leverage deep reinforcement learning (DRL) for efficient near-optimal PA strategy learning.

Main Methods:

• Integration of light detection and localization (LiDAL) for enhanced user CSI.
• Implementation of random linear network coding (RLNC) to improve data resilience.
• Application of a DRL-based normalized advantage function (NAF) algorithm for PA optimization.

Main Results:

• The proposed NAF algorithm closely approximates exhaustive search for PA optimization.
• NAF demonstrates a 39% speed improvement over deep deterministic policy gradient (DDPG).
• NAF achieves a 4.6% increase in average sum rate compared to gain ratio PA (GRPA), while considering location errors.

Conclusions:

• The LiDAL-assisted RLNC-NOMA OWC system effectively enhances communication performance.
• DRL, specifically the NAF algorithm, provides an efficient and effective solution for dynamic PA optimization in NOMA-OWC systems.
• The proposed approach robustly handles user location estimation errors, improving overall system reliability.