Wireless Communication Enables Dense-User Indoor Scenarios Via RLNC-NOMA OWC and DRL Power Allocation

Non-orthogonal multiple access (NOMA) offers a pathway to increased spectral efficiency in wireless communication, yet its performance is often hampered by imperfect channel information and decoding errors, particularly in realistic indoor environments. Researchers Ahmed A. Hassan, Ahmad Adnan Qidan, Taisir Elgorashi, and Jaafar Elmirghani, all from IEEE, address these challenges by modelling a novel system combining light detection and localization (LiDAL) with random linear network coding (RLNC) within a NOMA framework. Their work demonstrates how LiDAL can refine channel state information, while RLNC bolsters data reliability during decoding. Crucially, the team proposes a deep reinforcement learning (DRL) approach to optimise power allocation, a critical aspect of NOMA systems, and demonstrates its superiority over existing techniques in maximising the average sum rate of the system. This research offers a significant step towards robust and efficient wireless communication in dense user scenarios.

This work pioneered the integration of light detection and localization (LiDAL) with random linear network coding (RLNC) within a NOMA framework, aiming to enhance both CSI accuracy and data resilience.

The LiDAL technique harnessed spatio-temporal information to refine user CSI estimates, while RLNC bolstered the successive decoding process inherent to NOMA by creating random linear combinations of data packets. Central to the study was the development of a deep reinforcement learning (DRL) framework for dynamic power allocation (PA). Researchers implemented a DRL-based normalized advantage function (NAF) algorithm specifically designed to maximize the average sum rate of the system.

Experiments employed a simulated OWC environment to train and evaluate the NAF algorithm, meticulously comparing its performance against established PA schemes including deep deterministic policy gradient (DDPG), gain ratio PA (GRPA), and exhaustive search. The system delivered performance metrics based on the average sum rate achieved under varying levels of CSI imperfection and user density. To quantify CSI errors, the study utilized the Cramer, Rao lower bound (CRLB) as an unbiased error estimator, providing a benchmark for assessing the effectiveness of the LiDAL technique.

The DRL agent was trained using a reward function directly linked to the average sum rate, encouraging the algorithm to discover PA strategies that optimize system throughput. Rigorous testing demonstrated the NAF algorithm converged 33% faster than the DDPG algorithm and improved the average sum rate by 4.6% compared to the GRPA method. Notably, the optimized power allocation achieved by the proposed NAF algorithm closely matched the performance of exhaustive search, a computationally intensive method considered the gold standard.

This result highlights the efficiency and effectiveness of the DRL approach in navigating the complex interactions between multiple users, coding processes, and detection mechanisms within the OWC system. The methodology enabled a breakthrough in achieving near-optimal PA in dynamic, realistic indoor scenarios, paving the way for more robust and efficient OWC deployments.

LiDAL-RLNC-NOMA Achieves Optimal Power Allocation

Scientists have demonstrated a significant advancement in optical wireless communication (OWC) through the development of a LiDAL-assisted RLNC-NOMA system. This work models a system where light detection and localization (LiDAL) refines user channel state information (CSI), while random linear network coding (RLNC) bolsters data resilience within a non-orthogonal multiple access (NOMA) framework.

The research addresses the critical challenge of power allocation (PA) in complex, multi-user indoor environments, where interactions between users and coding processes demand sophisticated solutions. Experiments revealed that a deep reinforcement learning (DRL) framework, utilizing a normalized advantage function (NAF) algorithm, effectively learns near-optimal PA strategies to maximize the average sum rate of the system.

Results demonstrate the optimized power allocation achieved by the DRL-based NAF algorithm closely aligns with that of an exhaustive search method, while simultaneously accounting for inaccuracies in user location estimates. The team measured a convergence rate 33% faster for the NAF algorithm when compared to the deep deterministic policy gradient (DDPG) algorithm, showcasing improved efficiency in learning optimal strategies. Further tests prove the NAF algorithm improves the system’s average sum rate by 4.6% compared to the gain ratio PA (GRPA) method, highlighting a substantial gain in performance.

The study meticulously quantified the impact of imperfect CSI on NOMA-based OWC systems, employing the Cramer, Rao lower bound (CRLB) as an unbiased error estimator to characterize CSI imperfections. Integrating RLNC with NOMA further enhances data reliability by enabling reconstruction of original information from received combinations of superimposed NOMA signals, mitigating error propagation. The research successfully addresses the computational complexity inherent in optimizing PA, particularly in dense indoor networks with fluctuating user CSI.

By leveraging reinforcement learning, scientists circumvent the limitations of traditional methods like genetic algorithms and simulated annealing, which can become trapped in local optima. This breakthrough delivers a practical, efficient solution for achieving optimal performance in realistic, multi-parameter indoor OWC scenarios, paving the way for high-speed connectivity for data-intensive applications.

LiDAL-RLNC-NOMA Power Allocation via Deep Learning

This work presents a novel approach to power allocation in LiDAL-assisted RLNC-NOMA optical wireless communication systems, addressing the challenges posed by imperfect channel state information and decoding errors in dense indoor environments. Researchers developed a deep reinforcement learning framework, specifically a normalized advantage function algorithm, to dynamically optimise power distribution amongst users.

This allows the system to maximise the average sum rate, improving overall communication efficiency. The proposed system integrates light detection and localization with random linear network coding and non-orthogonal multiple access, leveraging spatio-temporal information to refine channel estimation and enhance data resilience. Simulation results demonstrate the effectiveness of the DRL-based power allocation strategy when compared to conventional methods like deep deterministic policy gradient and gain ratio power allocation.

The authors acknowledge limitations stemming from the modelling of the indoor environment and the assumptions made regarding user mobility, suggesting future work could explore more complex scenarios and refine the localization process. Further research may also investigate the scalability of the proposed algorithm to systems with a significantly larger number of users.

Non-orthogonal multiple access (NOMA) offers a pathway to increased spectral efficiency in wireless communication, yet its performance is often hampered by imperfect channel information and decoding errors, particularly in realistic indoor environments. Their work demonstrates how LiDAL can refine channel state information, while RLNC bolsters data reliability during decoding. This research offers a significant step towards robust and efficient wireless communication in dense user scenarios.