A Comprehensive Survey of Large AI Models for Future Communications: Foundations, Applications and Challenges

Authors

Feibo Jiang, Cunhua Pan, Li Dong, Kezhi Wang, Merouane Debbah, Dusit Niyato, Zhu Han

Paper

https://arxiv.org/abs/2505.03556

Code

https://github.com/jiangfeibo/ComLAM

Abstract

The 6G wireless communications aim to establish an intelligent world of ubiquitous connectivity, providing an unprecedented communication experience. Large artificial intelligence models (LAMs) are characterized by significantly larger scales (e.g.,billions or trillions of parameters) compared to typical artificial intelligence (AI) models. LAMs exhibit outstanding cognitive abilities, including strong generalization capabilities for fine-tuning to downstream tasks, and emergent capabilities to handle tasks unseen during training. Therefore, LAMs efficiently provide AI services for diverse communication applications, making them crucial tools for addressing complex challenges in future wireless communication systems. This study provides a comprehensive review of the foundations, applications, and challenges of LAMs in communication. First, we introduce the current state of AI-based communication systems, emphasizing the motivation behind integrating LAMs into communications and summarizing the key contributions. We then present an overview of the essential concepts of LAMs in communication. This includes an introduction to the main architectures of LAMs, such as transformer, diffusion models, and mamba. We also explore the classification of LAMs,including large language models (LLMs), large vision models (LVMs), large multimodal models (LMMs), and world models,and examine their potential applications in communication.Additionally, we cover the training methods and evaluation techniques for LAMs in communication systems. Lastly, we introduce optimization strategies such as chain of thought (CoT), retrieval augmented generation (RAG), and agentic systems. Following this,we discuss the research advancements of LAMs across various communication scenarios, including physical layer design, resource allocation and optimization, network design and management,edge intelligence, semantic communication, agentic systems, and emerging applications. Finally, we analyze the challenges in the current research and provide insights into potential future research directions.

Contents

• A Comprehensive Survey of Large AI Models for Communications: Foundations, Applications, and Challenges
  • Abstract
  • Contents
  • I. INTRODUCTION
    • A.Background
      • 1) Deep learning-assisted communication
      • 2) Reinforcement learning-assisted communication
      • 3) Generative AI-assisted communication
    • B. Motivation
      • 1) Definition
      • 2) Distinction between LAMs and GAIs
      • 3) Distinction between LAMs and pretrained FMs
    • C. Related survey work
      • 1) Limited model coverage
      • 2) Incomplete application landscape
    • D. Contributions
      • 1) Foundations of LAMs for communications
      • 2) Applications of LAMs for communications
      • 3) Research challenges of LAMs for communications
  • II. FOUNDATIONS OF LAMS FOR COMMUNICATIONS
    • A. Key architecture of LAMs
      • 1) Transformer model
      • 2) Diffusion model
      • 3) Mamba model
    • B. Classification of LAMs
      • 1) Large language model
      • 2) Large vision model
      • 3) Large multimodal model
      • 4) World model
    • C. Training of LAMs for communications
      • 1) Pre-training of LAMs for communications
      • 2) Fine-tuning of LAMs for communications
      • 3) Alignment of LAMs for communications
    • D. Evaluation of LAMs for communications
      • 1) Communication Q&A
      • 2) Communication tool learning
      • 3) Communication modeling
      • 4) Communication code design
    • E. Optimization of LAMs
      • 1) Chain of thought
      • 2) Retrieval-augmented generation
      • 3) Agentic system
    • F. Summary and lessons learned
      • 1) Summary
      • 2) Lessons learned
  • III. LAMS FOR PHYSICAL LAYER DESIGN
    • A. Channel and beam prediction based on LAMs
    • B. Automated physical layer design based on LAM
    • C. Summary and lessons learned
      • 1) Summary
      • 2) Lessons learned
  • IV. LAMS FOR RESOURCE ALLOCATION AND OPTIMIZATION
    • A. Computing resource allocation
    • B. Spectrum resource allocation
    • C. Energy resource optimization
    • D. Summary and lessons learned
      • 1) Summary
      • 2) Lessons learned
  • V. LAMS FOR NETWORK DESIGN AND MANAGEMENT
    • A. Network design
    • B. Network management
    • C. Summary and lessons learned
      • 1) Summary
      • 2) Lessons learned
  • VI. LAMS FOR EDGE INTELLIGENCE
    • A. Edge training and application of LAMs
    • B. Edge resource scheduling meets LAMs
    • C. Federated learning of LAMs
    • D. Summary and lessons learned
      • 1) Summary
      • 2) Lessons learned
  • VII. LAMS FOR SEMANTIC COMMUNICATION
    • A. LLM-based semantic communication systems
    • B. Other LAM-based semantic communication systems
    • C. Summary and lessons learned
      • 1) Summary
      • 2) Lessons learned
  • VIII. LAM-BASED AGENTIC SYSTEMS
    • A. Agentic systems based on LLMs
    • B. Agentic systems based on other GAI models
    • C. Summary and lessons learned
      • 1) Summary
      • 2) Lessons learned
  • IX. LAMS FOR EMERGING APPLICATIONS
    • A. Smart healthcare
    • B. Carbon emissions
    • C. Digital twins
    • D. Artificial intelligence of things
    • E. Integrated satellite, aerial, and terrestrial networks
    • F. Integration of UAVs and LLMs
    • G. Summary and lessons learned
  • X. RESEARCH CHALLENGES
    • 1). Lack of high-quality communication data
    • 2). Lack of structured communication knowledge
    • 3). Generative hallucination in communication
    • 4). Limitations of reasoning ability
    • 5). Poor explainability in LAMs
    • 6). Adaptability in dynamic environments
    • 7). Diversity of communication tasks
    • 8). Resource constraints at the edge
    • 9). High inference latency
    • 10). Security and privacy
  • Communication datasets for LAMs
  • Classification of LAMs
  • Paper With Code
  • The Team
  • Acknowledgments
  • Update log

Fig. 1: The development history of LAMs.

Fig. 2: The role of LAMs in AI.

Fig. 3: Overall organization of the survey.

Fig. 4: Applications of LAMs in Communication. LAMs can be applied across various domains in communication, including physical layer design, resource allocation and optimization, network design and management, edge intelligence, semantic communication, agentic systems, and emerging applications.

Communication datasets for LAMs

Category	datasets	Release Time	Link	Download
General datasets	Common Crawl	2020		Code
	Pile	2023		Code
	Dolma	2024		Code
	RedPajama-data	2024		Code
Communication content filtering	Common Crawl	2024		Code
	RedPajama	2024		Code
Communication pre-training datasets	TSpec-LLM	2023	Paper	Code
	OpenTelecom dataset	2024	Paper	Code
	CommData-PT dataset	2025	Paper
	TeleQnA dataset	2024	Paper	Code
	Tele-Data dataset	2024	Paper	Code
Communication fine-tuning datasets	TelecomInstruct dataset	2024	Paper
	CSI dataset compliant with 3GPP standards	2024	Paper
	CommData-FT dataset	2025	Paper
Communication alignment datasets	TelecomAlign dataset	2024	Paper
	Dataset for multi-server multi-user offloading problem dataset	2024		Code

Classification of LAMs

LAM Category	Specific Models	model	Release Time	Link	Download
Large Language Model	GPT series	GPT-1	2020	Paper	Code
		GPT-2	2023	Paper	Code
		GPT-3	2023	Paper
		GPT-4	2023	Paper
		OpenAI o1	2024	Paper	Code
	Gemma series	Gemma 1	2024	Paper
		Gemma 2	2024	Paper
	LLaMA series	LLaMA-1	2023	Paper	Code
		LLaMA-2	2023	Paper	Code
		LLaMA-3	2024	Paper	Code
Large Vision Model	SAM series	SAM-1	2023	Paper	Code
		SAM-2	2024	Paper	Code
	DINO series	DINO V1	2021	Paper	Code
		DINO V2	2023	Paper	Code
	Stable Diffusion series	Stable Diffusion V1	2022	Paper	Code
		Stable Diffusion V2	2022	Paper	Code
		Stable Diffusion V3	2024	Paper
Vision Language Model	LLaVA	LLaVA	2024	Paper	Code
	Qwen-VL	Qwen-VL	2023	Paper	Code
		Qwen-VL-Chat	2023	Paper	Code
	Mini-GPT4	Mini-GPT4	2023	Paper	Code
Large Multimodal Model	CoDi series	CoDi-1	2024	Paper	Code
		CoDi-2	2024	Paper	Code
	Meta-Transformer	Meta-Transformer	2023	Paper	Code
	ImageBind	ImageBind	2023	Paper	Code
World Model	Sora	Sora	2024	Paper
	JEPA	JEPA	2022	Paper
	Vista	Vista	2024	Paper	Code
Lightweight Large AI Model	TinyLlama	TinyLlama	2024	Paper	Code
	MobileVLM	MobileVLM	2024	Paper	Code
	Mini-Gemini	Mini-Gemini	2024	Paper	Code
Large Reasoning Model
	OpenAI o3-mini	OpenAI o3-mini	2025	Paper
	DeepSeek	DeepSeek-R1	2025	Paper	Code

Paper with code

Category	Title	Link	Download
Variational autoencoder	Joint coding-modulation for digital semantic communications via variational autoencoder	Paper	Code
Diffusion models	Beyond deep reinforcement learning: A tutorial on generative diffusion models in network optimization	Paper	Code
Large language model	Large language model enhanced multi-agent systems for 6g communications	Paper	Code
Large vision model	Large ai model-based semantic communications	Paper	Code
In-context learning	In-context learning for MIMO equalization using transformer-based sequence models	Paper	Code
Retrieval-augmented generation	Telco-rag: Navigating the challenges of retrieval-augmented language models for telecommunications	Paper	Code
Multi-agent system	Large language model enhanced multi-agent systems for 6g communications	Paper	Code
LLM-assisted physical layer design	Llm4cp: Adapting large language models for channel prediction	Paper	Code
LLM-assisted physical layer design	Generative ai agent for next-generation mimo design: Fundamentals, challenges, and vision	Paper	Code
GAI model-assisted physical layer design	Mimo channel estimation using score-based generative models	Paper	Code
Computing resource allocation	Diffusion-based reinforcement learning for edge-enabled ai-generated content services	Paper	Code
Edge training and application of LAMs	Edge-llm: Enabling efficient large language model adaptation on edge devices via layerwise unified compression and adaptive layer tuning and voting	Paper	Code
Edge training and application of LAMs	Federated fine-tuning of billion-sized language models across mobile devices	Paper	Code
Federated fine-tuning for LAMs	Fwdllm: Efficient fedllm using forward gradient	Paper	Code
Agent systems based on LLMs	Large language model enhanced multi-agent systems for 6g communications	Paper	Code
Agent systems based on LLMs	Wirelessagent: Large language model agents for intelligent wireless networks	Paper	Code
Agent systems based on LLMs	Generative ai agent for next-generation mimo design: Fundamentals, challenges, and vision	Paper	Code
LAMs for digital twin	Towards autonomous system: flexible modular production system enhanced with large language model agents	Paper	Code
Smart healthcare	Conversational health agents: A personalized llm-powered agent framework	Paper	Code
Carbon emissions	Generative ai for low-carbon artificial intelligence of things	Paper	Code

The Team

Here is the list of our student contributors in each section.

Section	Student Contributors
The whole paper	Zhengyu Du , Yuhan Zhang
Literature Search	Jian Zou , Dandan Qi
Project Maintenance	Xitao Pan

Contact Information for Source Code Submission or Update

If you intend to add or update the source code in the repository, please contact the following email addresses: jiangfb@hunnu.edu.cn, Dlj2017@hunnu.edu.cn, 240620854087@stu.hutb.edu.cn and 240620854065@stu.hutb.edu.cn.

Update Log

Version	Time	Update Content
v1	2024/12/09	The initial version.
v2	2024/12/18	Improve the writing. Correct some minor errors.
v3	2025/05/07	Improve the writing. Correct some minor errors.

Citation

 @ARTICLE{2025arXiv250503556J,
      title = {A Comprehensive Survey of Large AI Models for Future Communications: Foundations, Applications and Challenges},
      author = {Feibo Jiang, Cunhua Pan, Li Dong, Kezhi Wang, Merouane Debbah, Dusit Niyato, Zhu Han},
      journal = {arXiv preprint arXiv:2505.03556v1},
      year = {2025}
}