A HYBRID ACCESS POINT CENTERED POLICY FOR OPTIMAL AGE OF INFORMATION IN WIRELESS POWERED SENSOR NETWORK

This thesis investigates the optimization of the Age of Information (AoI) in Wireless Powered Communication Networks (WPCNs), focusing on a Hybrid Access Point (HAP)-centered policy. AoI is a critical metric for ensuring information freshness in Internet of Things (IoT) applications, such as smart cities, healthcare, and industrial automation, where timely data delivery is essential.To address the challenge of minimizing AoI while managing energy-constrained devices, this study employs the Irregular Repetition Slotted Aloha (IRSA) protocol, enhanced with Q-Learning, a reinforcement learning technique. IRSA improves upon traditional Aloha-based protocols by allowing devices to transmit multiple replicas of their packets across different time slots, increasing the likelihood of successful packet decoding through Successive Interference Cancellation (SIC). However, determining the optimal number of replicas for each device is complex, especially in dynamic environments with fluctuating energy levels and channel conditions. To tackle this challenge, the research models the system using a Markov Decision Process (MDP) framework, where devices learn optimal transmission strategies through reinforcement learning. The Q-Learning algorithm, combined with an epsilon-greedy policy, balances exploration and exploitation to minimize AoI efficiently. A Python-based simulation evaluates system performance under various conditions, including different numbers of users and slots. The study examines three distinct scenarios, each with different user and slot configurations. The results demonstrate that the Q-Learning algorithm effectively reduces AoI over time, with significant improvements observed after the training phase. The findings highlight the trade-off between system gain (user-to-slot ratio) and AoI, emphasizing the importance of balancing network load to optimize performance. The research concludes that adaptive transmission policies, guided by reinforcement learning, can significantly enhance network efficiency by dynamically adjusting transmission actions based on real-time conditions. Despite promising results, challenges remain, particularly regarding the algorithm’s sensitivity to hyperparameters and its adaptability to highly dynamic environments. Future research could explore hybrid models that combine reinforcement learning with other optimization techniques to improve robustness and scalability. This study contributes to the advancement of intelligent communication strategies in WPCNs, paving the way for more efficient and adaptive IoT networks.