Presenter: Piumika Karunanayake, Kotelawala Defence University, Sri Lanka

Inviting Professor: Prof. Dr. Anna Förster

Abstract: 

Wireless communication protocols are governed by a set of critical parameters that have a direct impact on overall network performance. Consequently, these parameters must be carefully aligned with the application requirements, operating environment, and the number of nodes in the network. When a single, fixed set of protocol parameters is applied across diverse applications and network conditions, optimal performance cannot be achieved. Therefore, dynamic protocol parameter optimization is essential to ensure efficient network operation. Recent research has explored several approaches for dynamic parameter optimization, including mathematical modeling, fuzzy logic based techniques, game theory, and reinforcement learning based models. In this work, we focus on the selection of parameters and the dynamic optimization of protocol parameters to enhance network performance without relying on any prior knowledge of the network. Furthermore, the proposed approach applies to both centrally operated and decentralized network architectures. Since Q-learning is a model-free reinforcement learning technique well-suited for adaptive optimization, we propose a Q-learning based parameter optimization mechanism to improve the performance of wireless networks.