Causal-Aware Channel-Adaptive Semantic Communication With Emergent Encoding

Abstract

Semantic communications improve efficiency by transmitting semantics instead of raw data, but under severely constrained resources and noisy channels, it remains challenging to identify and robustly transmit informative semantic units. In this letter, we propose a Causal-Aware Channel-Adaptive Semantic Communication framework (CA2-SemCom) that jointly addresses semantic selection and encoding. Specifically, scene-graph semantics are organized into a causal graph, where causal Shapley values quantify the importance of semantic units for channel-adaptive semantic selection under limited bit budgets. Furthermore, an emergent semantic encoding mechanism induces adaptive symbol reuse structures according to semantic similarity and channel conditions, improving semantic alignment robustness under channel noise. Simulation results over AWGN and Rayleigh fading channels demonstrate that the proposed approach outperforms conventional bit-level and learning-based schemes under low bit budgets.