Dynamic Residual Bandwidth Distribution Based on Adaptive Chaotic Q-Learning for Low-Latency Communications in IoT Gateways

Abstract Efficiently managing networks is crucial for handling the increasing data flow in IoT devices. IoT gateways act as intermediaries between devices and cloud servers, managing data transfer. Effective bandwidth management is key to ensuring smooth data movement through these gateways. However, current methods struggle to adapt to different needs, especially during peak usage periods. To address these challenges, our article introduces a new method that combines smart bandwidth management with a reinforcement learning algorithm. Our idea is to utilize unused bandwidth to enhance network performance without incurring extra costs. We propose specific strategies for allocating bandwidth to different types of connections. Additionally, our proposed Chaotic Q-learning algorithm integrates the-Chaotic Greedy Selection Strategy to dynamically optimize band-width allocation. Experimental analysis demonstrates the superior performance of our proposed algorithm. The-Chaos-Greedy strategy requires fewer iterations (15 vs. 25 for-Greedy) and achieves higher rewards. Moreover, Chaotic Q-learning (CQL) achieves higher throughput (80 Mbps vs. 70 Mbps for Q-learning) with reduced packet loss (2.5% vs. 5.5%), lower latency (35 ms vs. 45 ms), and higher fairness (0.90 vs. 0.78 fairness index). Furthermore, the bifurcation diagram generated by CQL sheds light on network resource allocation dynamics and aids in studying system stability, crucial for optimizing performance.