Heterogeneous-Modality TinyLM Scheduling at the MAC Layer

Develop a MAC-layer scheduler for 6G base stations that allocates resources among TinyLM semantic streams from text, image, speech, and industrial-control modalities despite their modality-specific and incommensurable fidelity metrics.

Background

Different semantic modalities are evaluated using different metrics, such as BLEU for text, PSNR for images, PESQ for speech, and task accuracy for industrial control. The paper notes that the Semantic Efficiency Score does not provide a common cross-modal currency because it normalizes within individual systems.

A viable scheduler must therefore either establish a universal semantic metric or optimize multiple modality-specific objectives and priority queues. The paper reports that neither approach has yet been demonstrated in a multimodal 6G MAC context.

References

Extending either approach to several token streams sharing a bottleneck raises a fairness question that a bit-level controller cannot pose, because how much rate a flow can give up depends on the generative prior at its receiver, a quantity the network does not observe. Whether protocols for this regime should be designed or, following the protocol-learning and language-oriented approaches surveyed by Park et al., learned from the task itself remains open.

— From Semantic to Token Communication: The Next Paradigm for Large-Model-Driven 6G Intelligent Connectivity  (2609.10714 - Ma et al., 9 Sep 2026) in Section 5.8, Token-Level Protocol Design

No scheduler handles the realistic case of one base station simultaneously serving TinyLMs across incommensurable modalities.

— Closing the Semantic-Edge Gap: Tiny Language Models for 6G Wireless Intelligence  (2609.03747 - Kamath et al., 3 Sep 2026) in Challenge 9, Section 8