TONIC: Token-Centric Semantic Communication for Task-Oriented Wireless Systems

2026-05-23 · Source: cs.LG updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Wireless Communication Systems · Depth: Expert, extended

Summary

TONIC is a novel token-centric semantic communication framework designed for task-oriented wireless systems, addressing the fundamental mismatch between traditional bit-level communication and the token-based information processing of modern foundation models. It operates by converting source samples into token sequences, estimating token-level task relevance, and applying utility-aware unequal error protection at the transmitter under a fixed channel-use budget. At the receiver, TONIC employs confidence-aware gating to transform unreliable token decisions into recoverable erasures, which are then restored by a Transformer-based completion model for final task inference. This modular and interpretable architecture consistently outperforms separation-based schemes, pixel-domain DeepJSCC, and other token-domain baselines in image classification tasks across AWGN, Rayleigh, and Rician channels, demonstrating superior performance under matched communication budgets.

Key takeaway

For Machine Learning Engineers or AI Scientists designing wireless systems for foundation models, you should adopt a token-centric communication paradigm, prioritizing task-relevant tokens over uniform bit fidelity. Implement utility-aware unequal error protection at the transmitter and confidence-aware gating with generative completion at the receiver. This approach significantly improves task accuracy, especially in bandwidth-limited or noisy environments, by converting residual uncertainty into recoverable erasures rather than harmful substitutions, leading to more robust and efficient semantic communication.

Key insights

Align wireless communication with token-level semantics for foundation models, optimizing task performance over bit fidelity.

Principles

• Token-level task relevance should guide communication resource allocation.
• Erasures can be less harmful than incorrect substitutions with strong completion priors.
• Modular design can combine semantic awareness with practical digital communication.

Method

Transmitter converts samples to tokens, estimates utility, and applies UEP. Receiver gates low-confidence tokens to erasures, then a Transformer model completes them for task inference.

In practice

• Use gradient-based utility for practical token importance estimation.
• Calibrate group-wise confidence thresholds offline for robust receiver policy.
• Prioritize protection and conservative acceptance for high-utility token groups.

Topics

• Semantic Communication
• Token-Centric Systems
• Unequal Error Protection
• Generative Completion
• Wireless Image Transmission
• Transformer Models

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Wireless TokenCom: RL-Based Tokenizer Agreement for Multi-User Wireless Token Communications — Hybrid RL optimizes tokenizer agreement, resource allocation, and beamforming for efficient multi-user wireless video TokenCom.
• Structured Intent as a Protocol-Like Communication Layer: Cross-Model Robustness, Framework Comparison, and the Weak-Model Compensation Effect — Structured intent representations significantly enhance goal alignment and reduce variance across diverse AI models and languages.
• Semantic Cache Distillation: Efficient State Transfer via Reuse and Selective Patching — Semantic Cache Distillation efficiently transfers LLM KV caches using compact semantic codes, significantly speeding up time-to-first-token while preserving quality.
• To See the Unseen: on the Generalization Ability of Transformers in Symbolic Reasoning — Transformer models exhibit representational collapse of unseen token unembeddings, hindering symbolic reasoning generalization.
• Mean-Field Parallel Decoding for Discrete Diffusion Language Models — A training-free framework coordinates parallel token updates in discrete diffusion models, improving generation quality and latency.
• Scaling Down Is the New Scaling Up — Future agentic AI prioritizes hardware-aware design for efficient on-device operation over raw model size.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.