Retrofitting In-Vehicle Networking for Time-Sensitive Streams: Spatiotemporal Slot Scheduling and Kernel Stack Simplification

Abstract

The perception and planning functions of intelligent vehicles rely on a wide array of sensors, controllers, and actuators, but in turn they generate high-volume (Gbps-level) and low-latency-required (microsecond- to millisecond-scale) in-vehicle streams. Although the time-sensitive networking (TSN) technology has shown potential for such traffic, prior studies have only focused on simulations or isolated onboard tests, leaving the real-world performance largely unexplored. In this work, we develop a hardware-in-the-loop platform to assess in-vehicle networking performance from an end-to-end perspective. Our investigation reveals some critical issues such as prolonged transmission delays and drastic jitter across the application- and physical-layer paths. We observe that there is a structural incompatibility between in-vehicle traffic characteristic and certain TSN mechanisms: TSN’s time-slot scheduling fails to adequately account for bursty traffic patterns; TSN deployment within the operating system (OS) kernel of controllers and actuators introduces non-deterministic transmission delays. To solve them, we propose CarTSN, a delay-deterministic framework for in-vehicle networking. Specifically, we design a spatiotemporal scheduler that flexibly adapts to fragmented time slots to better accommodate bursty traffic. We also streamline the kernel networking stack in end devices to reduce packet delivery delays and jitters. Experimental results demonstrate that CarTSN reduces average transmission delay by 64.08% compared to conventional TSN-supported approaches, while maintaining packet delivery determinism at 99.54%.