Assured Wireless Operations Through Dynamic Data-Driven Open Radio Access Systems
Abstract
The overarching goal of this project is to lay the theoretical foundations of Dynamic Data Driven Open Radio Access Network Systems, in short, 3D-O-RANs. Through 3D-O-RANs, the Air Force will achieve strategic mission-critical objectives by guaranteeing assured wireless communications in congested, contested and contaminated environments. First, we will investigate effective, efficient and reliable distributed spectrum sensing in 3D-O-RAN through a novel methodology named split deep reinforcement learning with multiple exits, which combines distributed control and classification into the same data-driven model. Next, we will mathematically optimize 3D-O-RAN operations through dynamic slicing of network, computation, and memory resources, taking into account the semantics of the sensed data into the problem formulation. Our core intuition is that different application classes can tolerate different levels of image compression and still deliver minimum classification accuracy. Therefore, the sensed image data can be dynamically compressed according to the application semantics, so as to minimize network load while still guaranteeing key application performance indicators. Next, we will investigate novel techniques to certify the performance of the neural networks used by 3D-O-RAN by reshaping them according to the ongoing optimization objectives. To this end, we propose mission-driven dynamic knowledge distillation with student pruning, which certifies execution latency and hardware occupation while dynamically guaranteeing desired accuracy levels. Our theoretical findings will be extensively validated through a system-level prototype of 3D-O-RAN, which will be obtained through a custom-built System-on-Chip platform, as well as with experiments on Colosseum, the world s largest network emulator, and Arena, an over-the-air 64-antenna wireless testbed at Northeastern University.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 29, 2024
- Source ID
- FA95502310261
Entities
People
- Francesco Restuccia
Organizations
- Air Force Office of Scientific Research
- Northeastern University
- United States Air Force