Reinforcement Learning-based UL/DL Splitter for Latency Reduction in Wireless TSN Networks

Reducing latency in Time-Sensitive Networking (TSN) networks is critical to fulfil real-time communicationrequirements, ensuring timely data delivery, and maintaining system responsiveness. Minimizing latency enhances the reliability of industrial automation, multimedia streaming, and other time-critical applications, ultimately optimizing overall network performance and user experience. One of the most critical points lies in the wireless segments, which cannot be considered as deterministic. Specially, when the traffic load between uplink and downlink is unbalanced, it is critical to allocate resources based on the volume of such traffic and on the channel state, both significantly impacting on packet latency. In this paper we present and compare a set of approaches to scheduling time slots within a wireless frame for communication between the uplink (UL)and downlink (DL) in a TSN network. The primary objective is to reduce latency in wireless transmissions, particularly in scenarios with stringent timing requirements. By optimizing the allocation of time slots between UL and DL, our proposed scheduling algorithm aims to minimize queueing delays while ensuring efficient utilization of network resources. The results highlight the significant reduction achieved in terms of queueinglatency and packet loss through our scheduling strategy, thereby enhancing the reliability and timeliness of wireless links in TSN networks.