Large Language Model as a Catalyst: A Paradigm Shift in Base Station Siting Optimization

TL;DR

This work tackles the challenging problem of base station siting (BSS) in urban mobile networks by introducing an LLM-empowered optimization paradigm augmented with retrieval-augmented generation (RAG). It presents three heuristic strategies—PoL (prompt-optimized LLM), LaBa (LLM-driven autonomous agent), and CLaBa (cooperative multi-agent)—to balance human input, automation, and collaboration in solving BSS tasks. Empirical evaluation on a real-world urban dataset shows that the LLM-based strategies can meet the 90% traffic-coverage target while achieving lower deployment costs than traditional baselines, with RAG further boosting accuracy and robustness. The proposed framework demonstrates a pathway toward AI-native, scalable network optimization where human expertise and AI capabilities are integrated to enable AI-as-a-service for next-generation networks, albeit with careful attention to data quality, hallucination risks, and deployment considerations. $D_{min}$, $d_h$, $d_d$, $ heta_{cp}$, and cost terms like $C_h$, $C_d$ are used to formalize the optimization, underscoring the role of precise mathematical modeling in enabling reliable LLM-guided solutions.

Abstract

Traditional base station siting (BSS) methods rely heavily on drive testing and user feedback, which are laborious and require extensive expertise in communication, networking, and optimization. As large language models (LLMs) and their associated technologies advance, particularly in the realms of prompt engineering and agent engineering, network optimization will witness a revolutionary approach. This approach entails the strategic use of well-crafted prompts to infuse human experience and knowledge into these sophisticated LLMs, and the deployment of autonomous agents as a communication bridge to seamlessly connect the machine language based LLMs with human users using natural language. Furthermore, our proposed framework incorporates retrieval-augmented generation (RAG) to enhance the system's ability to acquire domain-specific knowledge and generate solutions, thereby enabling the customization and optimization of the BSS process. This integration represents the future paradigm of artificial intelligence (AI) as a service and AI for more ease. This research first develops a novel LLM-empowered BSS optimization framework, and heuristically proposes three different potential implementations: the strategies based on Prompt-optimized LLM (PoL), LLM-empowered autonomous BSS agent (LaBa), and Cooperative multiple LLM-based autonomous BSS agents (CLaBa). Through evaluation on real-world data, the experiments demonstrate that prompt-assisted LLMs and LLM-based agents can generate more efficient and reliable network deployments, noticeably enhancing the efficiency of BSS optimization and reducing trivial manual participation.

Figures (13)

• Figure 1: The coverage and planning of base stations within a given region. The real-world map is shown on the bottom layer; existing macro and micro base stations are displayed on the middle layer; both planned and existing macro and micro base stations are marked in the top layer, along with proposed upgrades to address areas with poor coverage.
• Figure 2: Diagram of the prompt-optimized LLM-based (PoL) strategy, showcasing the iterative workflow involving information gathering, automatic modeling & optimization, as well as automatic code generation & correction (including auto-coding, auto-debugging, and also auto-correction). This workflow is guided by human-initialized prompts, enabling efficient solutions to the BSS problem.
• Figure 3: An example prompt for mathematical modeling of PoL strategy.
• Figure 4: Diagram of the LLM-empowered autonomous BSS agent (LaBa) strategy, illustrating the workflow from mathematical problem formulation and automated code generation to iterative debugging, error correction, and testing, ultimately delivering a validated solution for BSS task.
• Figure 5: Diagram of the cooperative multiple LLM-based autonomous BSS agents (CLaBa) framework, illustrating the workflow of task division and collaboration among agents: ${\rm Agent}_1$ for problem formulation, ${\rm Agent}_2$ for code generation, ${\rm Agent}_3$ for code execution and feedback, and ${\rm Agent}_4$ for testing and feedback, with iterative collaboration to refine and solve BSS task.