Coordinating Communication and Computing for Wireless VR in Open Radio Access Networks

Abstract

Driven by diverse applications, radio access networks (RAN) are expected to embrace built-in computing and intelligence, forming a versatile wireless computing platform that closely integrates communication and computing. To fully unleash the potential of such a synergistic system, it is essential to coordinate communication and computing with intelligence unlocked by the radio intelligent controllers (RICs) in O-RAN. Building on the groundwork established by existing theoretical studies and simulations, we develop a platform that can emulate the events in the real-world system in more detail, bringing theoretical works closer to practical implementation. In this paper, we first introduce ns-GP-O-RAN, a software simulation platform developed over ns-3, enabling communication, computation task processing, large-scale data collection, and testing of system-level orchestration policies through user-level control. Taking virtual reality (VR) as an example, we formulate the computation offloading problem and develop a prediction-based computation offloading xAPP, which contains a prediction phase to predict users' end-to-end (E2E) performance with the deep neural network and a system-level decision-making phase for global orchestration with the differential evolution algorithm. We evaluate the system capacity and E2E latency over the developed ns-GP-O-RAN, which is more effective than existing approaches.