Generative AI Agents with Large Language Model for Satellite Networks via a Mixture of Experts Transmission

TL;DR

This work addresses two key obstacles in upcoming 6G satellite networks: complex system modeling and interference-driven resource allocation. It introduces a generative AI agent framework that uses large language models and retrieval-augmented generation to customize problem formulation across four satellite-modeling aspects, aided by a two-layer semantic router and RAG. To solve the formulated problems, the authors propose MoE-PPO, a proximal policy optimization approach that leverages a mixture of specialized experts with a gating network to jointly optimize beamforming and rate allocations. Simulation results demonstrate accurate, adaptive problem formulation by the generative agent and a consistent performance boost (vs baselines) from MoE-PPO, including robustness to protocol choices (RSMA vs SDMA) and optimization goals (EE vs power minimization). The integrated framework offers a scalable path for tailoring complex satellite network configurations and efficiently solving their associated optimization tasks in dynamic 6G environments.

Abstract

In response to the needs of 6G global communications, satellite communication networks have emerged as a key solution. However, the large-scale development of satellite communication networks is constrained by the complex system models, whose modeling is challenging for massive users. Moreover, transmission interference between satellites and users seriously affects communication performance. To solve these problems, this paper develops generative artificial intelligence (AI) agents for model formulation and then applies a mixture of experts (MoE) approach to design transmission strategies. Specifically, we leverage large language models (LLMs) to build an interactive modeling paradigm and utilize retrieval-augmented generation (RAG) to extract satellite expert knowledge that supports mathematical modeling. Afterward, by integrating the expertise of multiple specialized components, we propose an MoE-proximal policy optimization (PPO) approach to solve the formulated problem. Each expert can optimize the optimization variables at which it excels through specialized training through its own network and then aggregates them through the gating network to perform joint optimization. The simulation results validate the accuracy and effectiveness of employing a generative agent for problem formulation. Furthermore, the superiority of the proposed MoE-ppo approach over other benchmarks is confirmed in solving the formulated problem. The adaptability of MoE-PPO to various customized modeling problems has also been demonstrated.

Figures (10)

• Figure 1: The outline of the paper. Specifically, to address the challenges of complex modeling and difficulty in solving formulated problems within satellite communication networks, we introduce a generative AI agent and an MoE approach. The generative AI agent is tasked with formulating the optimization problems, while the MoE framework is employed to effectively solve the optimization problems formulated by the generative AI agent.
• Figure 2: The design of generative AI agent for satellite communication modeling. A: The two-layer semantic router. The expertise knowledge is organized into 4 blocks, each of which contains 2 sub-blocks. B: The answer generation for satellite communications modeling.
• Figure 3: The diagram of the proposed MoE-PPO approach.
• Figure 4: The process of satellite communication modeling using proposed generative AI agent. (A): The user description and answers of generative AI agent. The words with key semantics are highlighted in yellow. (B): The retrieval from expertise knowledge.
• Figure 5: The impact of chunk size and number on generation accuracy.

Theorems & Definitions (8)

• Remark 1
• Remark 2
• Remark 3
• Proposition 1
• proof
• Definition 1: Back propagation maximization
• Remark 4
• Remark 5