Modified Split-Rendering Architecture to enable AI-assisted Application-aware MAC for XR Slice

Supporting Extended Reality (XR) over traditional networks is challenging due to its bursty nature, low latency, and high-reliability requirements. We demonstrate that DiffServ in EPON results in low network utilization, around 10%. Additionally, XR frames are highly correlated due to changes in the application-layer data caused by human activity. By exploiting this correlation, in this letter, we introduce a cross-layer MAC scheduling termed "Application-aware MAC," which provides a manifold increase in network utilization by introducing a small prediction error in the application layer. For enabling this, in this letter, we also suggest a modified split rendering architecture with additional NFVs.