Intent Interpretation at RIC Timescales: Jev Decision Models versus Large Language Models in 6G Open RAN
Jev-1.13.0 stayed inside the 1-second RIC budget on 99.8% of intent calls; the hosted LLMs tested did not.
The paper compares decision models and LLMs for turning Open RAN intents into A1 policies. Median interpretation time ranged from 0.286 to 2.35 seconds, while A1 and E2 transfer stayed under 25 ms. Two hosted LLMs met the 1-second near-real-time budget on 17.9% and 0% of calls. In the network tests, the authors found no resolved radio-performance gain from hosted LLMs over Jev-1.13.0 at the base point. HF Daily Papers' note
The paper compares decision models and LLMs for turning Open RAN intents into A1 policies. Median interpretation time ranged from 0.286 to 2.35 seconds, while A1 and E2 transfer stayed under 25 ms. Two hosted LLMs met the 1-second near-real-time budget on 17.9% and 0% of calls. In the network tests, the authors found no resolved radio-performance gain from hosted LLMs over Jev-1.13.0 at the base point. HF Daily Papers' note
score 4