Mitigation of Scheduling Violations in Time-Sensitive Networking using Deep Deterministic Policy Gradient

Time-Sensitive Networking (TSN) is designed for real-time applications, usually pertaining to a set of Time-Triggered (TT) data flows. TT traffic generally requires low packet loss and guaranteed upper bounds on end-to-end delay. To guarantee the end-to-end delay bounds, TSN uses Time-Aware Shaper (TAS) to provide deterministic service to TT flows. Each frame of TT traffic is scheduled a specific time slot at each switch for its transmission. Several factors may influence frame transmissions, which then impact the scheduling in the whole network. These factors may cause frames sent in wrong time slots, namely misbehaviors. To mitigate the occurrence of misbehaviors, we need to find proper scheduling for the whole network. In our research, we use a reinforcement-learning model, which is called Deep Deterministic Policy Gradient (DDPG), to find the suitable scheduling. DDPG is used to model the uncertainty caused by the transmission-influencing factors such as time-synchronization errors. Compared with the state of the art, our approach using DDPG significantly decreases the number of misbehaviors in TSN scenarios studied and improves the delay performance of the network.