App–RAN mutual awareness for MEC live video analytics

Technology
In development
University
Scouted & summarized by Halo AI

ARMA is an end-to-end live video analytics system developed by Seoul National University that ensures consistent low latency in 5G MEC environments. By implementing a mutually-aware, decoupled scheduling framework on the Open-RAN Intelligent Controller, ARMA coordinates application and network schedulers to optimize bitrate and DNN complexity. This approach enables high latency SLO satisfaction, achieving a 97% success rate in prototype testing.

Overview

ARMA (App-RAN Mutual Awareness) is an end-to-end live video analytics system developed by researchers at Seoul National University to address the challenge of maintaining consistent low latency in Mobile Edge Computing (MEC) environments. In standard 5G MEC platforms, latency Service Level Agreement (SLO) satisfaction can drop significantly due to fluctuations in wireless channels and scene complexity. ARMA solves this by fostering cooperative interaction between application-layer and Radio Access Network (RAN) schedulers, which are typically operated independently. By enabling mutual awareness, the system ensures that scheduling decisions are coordinated, leading to high SLO satisfaction rates.

Technical specifications
  • Mutually-aware decoupled scheduling: Built upon the real-time RAN Intelligent Controller (RIC) in an Open-RAN architecture, this framework allows app and RAN schedulers to exchange minimal information for resource monitoring without disclosing proprietary algorithms.
  • Bitrate-DNN interplay: The system jointly controls video bitrate and Deep Neural Network (DNN) complexity. It leverages the ability of heavier DNNs to compensate for accuracy losses from lower bitrates, and vice versa, to optimize latency under varying resource conditions.
  • Mutually-aware resource monitoring: ARMA tracks network and compute resource availability by using physical-layer resource allocation information from the RAN to improve bandwidth estimation at the application layer.
  • SLO splitting: The system independently splits the overall end-to-end latency SLO into feasible network and compute-stage deadlines for each user.
  • Preemption mechanism: To handle sudden resource fluctuations, the system includes a preemption mechanism that accelerates the inference execution of users falling behind schedule by temporarily preempting others.
Technology readiness level

The technology is at a prototype stage (TRL 4-6). It has been implemented and evaluated...

Generated Sep 27, 2026. Please verify key information.

About Seoul National University

Seoul National University is a comprehensive public research university serving a large student and faculty community in Korea’s capital, recognized for cross‑disciplinary depth and scale. Industry partners engage through an industry–academia cooperation office that streamlines sponsored research, NDAs and master agreements, plus shared core facilities and prototyping spaces across its Gwanak and medical campuses. Integration with a major teaching hospital system enables clinical translation, while the Seoul location places teams close to corporate headquarters, government R&D agencies and a vibrant startup ecosystem for rapid collaboration and scale‑up. Research is supported by competitive national funding from the National Research Foundation of Korea and ministries such as Science and ICT, Trade, Industry and Energy, and Health and Welfare; a dedicated technology transfer office manages IP, licensing and startup formation.

Halo home
Partner smarter. Move faster.
Get new partnering requests
delivered to your inbox.