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

Juhyung Lee, Yanqing Lu, Klaus Doppler

Keywords: On-device large language model (LLM), Device-to-device (D2D) communication, Cross-layer wireless optimization, Parameter-efficient fine-tuning (PEFT), Context-aware reward design, Wi-Fi Aware (WFA), Edge intelligence

TL;DR: PEARL is the first on-device LLM for cooperative D2D optimization, using peer context, context-aware rewards, and head-based fine-tuning to adapt Wi-Fi Aware parameters with sub-20 ms inference and up to 16% energy savings.

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 fine-tuning. 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. Code, real-world demo, and dataset are available at https://github.com/abman23/pearl.

Submission Number: 69