AI2MMUM: AI-AI Oriented Multi-Modal Universal Model Leveraging Telecom Domain Large Model

2025-05-15

AI Summary

To address the demand for generalized, multimodal air-interface tasks in 6G systems—characterized by heterogeneous modalities (e.g., RF signals and text), diverse physical-layer objectives, and instruction-driven operation—this paper proposes the first AI-AI collaborative universal physical-layer model. Our method introduces an implicit, learnable prefix instruction mechanism; freezes a pre-trained RF encoder; incorporates lightweight modality adapters for cross-modal alignment; and employs task-specific lightweight heads for direct physical-layer output. Built upon a telecommunications-domain large language model backbone, the framework undergoes multimodal alignment fine-tuning. Evaluated on WAIR-D and DeepMIMO datasets, it achieves state-of-the-art performance across five critical physical-layer tasks—including channel estimation and environment recognition—demonstrating unprecedented flexibility and generalization capability for air-interface intelligence.

Technology Category

• Machine Learning: Multimodal Learning
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Computer Vision: Multi-modal Vision

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Designing a 6G-oriented universal model capable of processing multi-modal data and executing diverse air interface tasks has emerged as a common goal in future wireless systems. Building on our prior work in communication multi-modal alignment and telecom large language model (LLM), we propose a scalable, task-aware artificial intelligence-air interface multi-modal universal model (AI2MMUM), which flexibility and effectively perform various physical layer tasks according to subtle task instructions. The LLM backbone provides robust contextual comprehension and generalization capabilities, while a fine-tuning approach is adopted to incorporate domain-specific knowledge. To enhance task adaptability, task instructions consist of fixed task keywords and learnable, implicit prefix prompts. Frozen radio modality encoders extract universal representations and adapter layers subsequently bridge radio and language modalities. Moreover, lightweight task-specific heads are designed to directly output task objectives. Comprehensive evaluations demonstrate that AI2MMUM achieves SOTA performance across five representative physical environment/wireless channel-based downstream tasks using the WAIR-D and DeepMIMO datasets.

Problem

• Designing a 6G-oriented universal model for multi-modal data processing
• Enhancing task adaptability with fixed keywords and learnable prompts
• Achieving SOTA performance in wireless channel-based downstream tasks

Innovation

• AI2MMUM model for multi-modal 6G tasks
• LLM backbone with fine-tuning for telecom
• Lightweight task-specific heads for outputs