Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On-Device LLM for Context-Aware Wi-Fi Roaming

Ju-Hyung Lee, Yanqing Lu, Klaus Doppler

TL;DR

The paper tackles the challenge of maintaining seamless Wi-Fi roaming in dynamic mobile environments where fixed threshold rules and traditional DRL approaches struggle. It introduces an on-device LLM that reasons over structured, real-time context to perform context-aware AP selection and adaptive thresholding, using chain-of-thought prompting, LoRA fine-tuning, Direct Preference Optimization, and quantization to meet edge latency budgets of $10$--$100$ ms. Experimental results on indoor and outdoor data show the LLM-based approach can reduce unnecessary handovers while preserving strong signal quality, surpassing legacy heuristics and DRL baselines, and the work provides practical deployment guidelines and a real-world demonstration. Overall, this work demonstrates a viable path for applying application-layer AI to govern lower-layer wireless control at the edge, enabling robust, context-aware roaming in mobile scenarios.

Abstract

Roaming in Wireless LAN (Wi-Fi) is a critical yet challenging task for maintaining seamless connectivity in dynamic mobile environments. Conventional threshold-based or heuristic schemes often fail, leading to either sticky or excessive handovers. We introduce the first cross-layer use of an on-device large language model (LLM): high-level reasoning in the application layer that issues real-time actions executed in the PHY/MAC stack. The LLM addresses two tasks: (i) context-aware AP selection, where structured prompts fuse environmental cues (e.g., location, time) to choose the best BSSID; and (ii) dynamic threshold adjustment, where the model adaptively decides when to roam. To satisfy the tight latency and resource budgets of edge hardware, we apply a suite of optimizations-chain-of-thought prompting, parameter-efficient fine-tuning, and quantization. Experiments on indoor and outdoor datasets show that our approach surpasses legacy heuristics and DRL baselines, achieving a strong balance between roaming stability and signal quality. These findings underscore the promise of application-layer LLM reasoning for lower-layer wireless control in future edge systems.

On-Device LLM for Context-Aware Wi-Fi Roaming

TL;DR

The paper tackles the challenge of maintaining seamless Wi-Fi roaming in dynamic mobile environments where fixed threshold rules and traditional DRL approaches struggle. It introduces an on-device LLM that reasons over structured, real-time context to perform context-aware AP selection and adaptive thresholding, using chain-of-thought prompting, LoRA fine-tuning, Direct Preference Optimization, and quantization to meet edge latency budgets of -- ms. Experimental results on indoor and outdoor data show the LLM-based approach can reduce unnecessary handovers while preserving strong signal quality, surpassing legacy heuristics and DRL baselines, and the work provides practical deployment guidelines and a real-world demonstration. Overall, this work demonstrates a viable path for applying application-layer AI to govern lower-layer wireless control at the edge, enabling robust, context-aware roaming in mobile scenarios.

Abstract

Roaming in Wireless LAN (Wi-Fi) is a critical yet challenging task for maintaining seamless connectivity in dynamic mobile environments. Conventional threshold-based or heuristic schemes often fail, leading to either sticky or excessive handovers. We introduce the first cross-layer use of an on-device large language model (LLM): high-level reasoning in the application layer that issues real-time actions executed in the PHY/MAC stack. The LLM addresses two tasks: (i) context-aware AP selection, where structured prompts fuse environmental cues (e.g., location, time) to choose the best BSSID; and (ii) dynamic threshold adjustment, where the model adaptively decides when to roam. To satisfy the tight latency and resource budgets of edge hardware, we apply a suite of optimizations-chain-of-thought prompting, parameter-efficient fine-tuning, and quantization. Experiments on indoor and outdoor datasets show that our approach surpasses legacy heuristics and DRL baselines, achieving a strong balance between roaming stability and signal quality. These findings underscore the promise of application-layer LLM reasoning for lower-layer wireless control in future edge systems.
Paper Structure (19 sections, 1 equation, 10 figures, 7 tables)

This paper contains 19 sections, 1 equation, 10 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Wi-Fi Roaming and Mobility Management
  4. LLMs as Decision-Making Agents
  5. Task (1): Context-Aware AP Choice by LLM
  6. Problem Statement: Best BSSID Selection
  7. Approach: LLM-based Decision with Contextual Information
  8. Evaluation
  9. Experiments
  10. Limitations
  11. Task (2): Online Roaming Optimization by On-Device LLM
  12. Problem Statement: Dynamic Threshold Selection
  13. Approach: Adaptive Threshold via On-Device LLM
  14. Experiments
  15. Discussion
  16. ...and 4 more sections

Figures (10)

  • Figure 1: Cross-layer control via on-device LLM: Rule‑based handover (legacy) vs. on‑device LLM context‑aware handover.
  • Figure 2: Common failure modes of rule-based Wi-Fi roaming.
  • Figure 3: Impact of prompt engineering.
  • Figure 4: Comparison of # HO (left) and AvgRSSI (right) for each method for best BSSID selection (Task 1).
  • Figure 5: Average LLM inference time (A100 GPU).
  • ...and 5 more figures