Adaptive resource allocation for semantic-aware vehicular wireless networks with hierarchical codebooks

Vehicular wireless networks must exchange task-relevant information, such as perception data, cooperative sensing, and driving intention, under constrained freshness constraints. Semantic communication can reduce the bandwidth by transmitting task-level representations rather than raw bits of information. However, the joint decision of radio allocation and semantic granularity remains open in high-mobility vehicle-to-everything (V2X) communication networks. This paper proposes an adaptive framework that couples hierarchical codebooks with deep reinforcement learning (DRL) for joint resource scheduling and codebook-level selection in V2X networks. Each codebook level offers a different distortion trade-off, and the DRL-based controller can adaptively select a codebook level and allocate resource blocks and power, so that the proposed method can ensure the age of information (AoI) for each vehicle within its maximum AoI tolerance. We propose a system model that captures OFDMA scheduling, power control, and semantic-level decisions, and formulate an AoI-constrained optimization that minimizes long-term weighted semantic distortion; and design a DRL-based online policy with feasibility projection to respect AoI limits in every timeslot. The approach decomposes into lightweight per-slot decisions and scales to multi-RSU deployments. Simulations with cooperative perception demonstrate significant reductions in semantic distortion at spectrum usage, while satisfying the stringent AoI constraints.