Deep polar transformer semantic coding framework for reliable 6G autonomous vehicle communication

The development of sixth generation (6G) networks for Autonomous Vehicles (AVs) requires an unprecedented level of ultra-reliable low-latency communication (URLLC). However, existing communication paradigms such as separate source channel coding (SSCC) and deep joint source channel coding (DeepJSCC) are inherently limited in meeting these demands. These approaches either waste bandwidth by reconstructing human-readable data that is unnecessary for control tasks or provide uniform protection to all data regardless of its control relevance. To overcome these challenges, this study proposes a novel semantic polar coding framework termed POLAR-CODE. The proposed approach integrates semantic extraction and channel coding into a unified end-to-end architecture optimized for control-oriented communication. At its core, POLAR-CODE employs a task-driven semantic Transformer to extract compact, control-relevant representations, such as actuator signals and safety margins. It combines them with a semantics-aware Polar coding scheme. A key innovation is the Gradient-Guided Polarization (GGP) mechanism, which dynamically adjusts bit reliability based on the semantic importance of encoded features. Extensive simulations of cooperative vehicle platooning scenarios using vehicle-to-everything (V2X) channel models demonstrate the effectiveness of the approach. Results show that POLAR-CODE achieves approximately 97% payload reduction, improves lane-keeping stability by approximately 42%, and reduces emergency-braking collisions by approximately 81% compared to state-of-the-art DeepJSCC methods under stringent reliability and low signal-to-noise ratio (SNR) conditions.