PEARL: Peer-Enhanced Adaptive Radio via On-Device LLM

2025-09-28

AI Summary

This paper addresses the cross-layer optimization challenge of Wi-Fi Aware (WA) in device-to-device (D2D) communication by proposing PEARL—the first collaborative optimization framework tailored for on-device large language models (LLMs). PEARL jointly models publisher and subscriber states via a lightweight decoupled architecture, peer-context-aware mechanisms, and a LoRA+-enhanced head fine-tuning strategy with KL-regularized reward alignment to enable real-time, adaptive WA parameter configuration. Innovatively, it introduces a normalized reward function integrating application latency tolerance and battery state. Evaluated under real-world measurement-driven scenarios, PEARL significantly improves energy efficiency and latency: it achieves up to 16% energy savings in collaborative low-battery settings compared to heuristic and compact-model baselines, while the PEARL-Lite variant maintains inference latency below 20 ms.

Technology Category

• Machine Learning: Hardware-aware ML
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Search and Optimization: Learning to Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Energy management for devices in mobile Web and WoT environments
• Search and Retrieval-Augmented AI: Personalized, context-aware and across-device search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

We present PEARL (Peer-Enhanced Adaptive Radio via On-Device LLM), a framework for cooperative cross-layer optimization in device-to-device (D2D) communication. Building on our previous work on single-device on-device LLMs, PEARL extends the paradigm by leveraging both publisher and subscriber states to guide Wi-Fi Aware (WA) parameter selection. A context-aware reward, which normalizes latency by application tolerances and modulates energy by device battery states, provides richer supervision for KL-based finetuning. We study two lightweight variants: PEARL (Head + Low-Rank Adaptation (LoRA)) achieves the best overall performance, while PEARL-Lite (Head-only) delivers sub-20 ms inference at near-identical objective scores. Across synthetic scenarios grounded in real measurements, PEARL improves objective scores over heuristic and compact model baselines and reduces energy by up to 16% in cooperative low-battery cases. These results demonstrate that peer-aware context, reward-aligned training, and head-based efficiency make LLMs practical for always-on, on-device cross-layer control.

Problem

• Optimizing device-to-device communication through cooperative cross-layer control
• Enhancing Wi-Fi Aware parameter selection using publisher and subscriber states
• Reducing energy consumption while maintaining performance in low-battery scenarios

Innovation

• Peer-aware context guides Wi-Fi parameter selection
• Reward-aligned training uses latency and energy supervision
• Head-based efficiency enables fast on-device inference