Key Technical Points of RAN3 in R18 for 5G Radio Group

In the 3GPP Technical Radio Access Network (TSG RAN) specification group, RAN3 is responsible for the overall architecture of UTRAN, E-UTRAN, and G-RAN, as well as the protocol specifications of related network interfaces. Specific details in R18 are as follows:

I. AI/ML and IAB Mobile Architecture for RAN3

1.1 AI/ML for NG-RAN (Model Deployment, F1/Xn-based Inference)

• Working Principle: CU/DU exchange AI model parameters (tensor shape, quantization) via F1AP/XnAP. gNB-DU runs inference locally (beam/CSI prediction) and sends the results to CU. The model is updated with incremental parameters (without requiring complete retraining).
• Progress: Lack of standardized AI integration; vendors use proprietary silos.
• Implementation Results: Interoperable AI across multi-vendor RANs has been achieved (verified by Ericsson and Nokia).

1.2 Mobile IAB (Node Migration, RACH-less Handover, NCGI Reconfiguration) 

• Operating Principle: IAB-MT performs L1/L2 handover to the target parent node; the serving user equipment (UE) performs handover via NCGI (NR cell global ID) reallocation.
• Work Progress: The target gNB allocates UL timing via XnAP before migration. The topology is advertised in the SIB (mobileIAB-Cell).
• Implementation Results: Static IAB fails during vehicle movement (events cover vehicles, trains); throughput drops by 60% during topology changes. Seamless backhaul migration maintains 5% UE throughput during 60 mph movement.

1.3 SON/MDT Enhancements (RACH Optimization, NPN Logging).

• Operating Principle: MDT logs RACH failures and L1/L2 movement events for specific slices. The SON algorithm automatically adjusts the number of RACH operations based on slice load. NPN (Non-Public Network) logging includes enterprise identifiers and coverage maps.
• Work Progress: Rel-17 SON cannot recognize slice interactions; enterprise NPN lacks diagnostic data.
• Implementation Results: RAC optimization improved by 40%, NPN deployment verification was automated.

1.4 QoE Framework (AR/MR/Cloud Gaming, RAN-visible QoE based on data center).

• Working Principle: gNB collects XR attitude data, rendering latency, and packet loss rate through QoE measurements (MAC CE/RRC). It reports to OAM/NWDAF via XnAP/NGAP. Dynamic QoS adjustment is performed based on video stuttering events and motion sickness indicators.
• Progress: RAN is unaware of application QoE; operators are unaware of XR performance degradation.
• Implementation Results: Video stuttering was reduced by 30% through predictive scheduling.

1.5 Network Slicing (S-NSSAI Alternative, Partially Allowing NSSAI).

• Working Principle: Partial NSSAI allows the use of a subset during congestion; S-NSSAI is dynamically replaced by NGAP. Timing Synchronization Status (TSS) is reported every 10 seconds during GNSS outages to achieve gNB clock correction.
• Progress: NSSAI mismatch caused 20% of slice handover failures; GNSS outages caused 15% timing drift in the FR2 band.
• Implementation Results: NSSAI consistency reached 99%, and timing accuracy during outages was less than 1μs.

1.6 Timing Resilience (NGAP/XnAP TSS Reporting).

• Working Principle: The NGAP and XnA protocols were enhanced with the addition of a Timing Synchronization Status (TSS) reporting mechanism between network nodes to detect and compensate for timing drift or GNSS outages. This ensures that gNBs can dynamically adjust their clocks based on TSS messages to maintain synchronization.
• Progress: Timing alignment is critical for NR, especially in high-frequency bands and NTN. GNSS outages or network failures can cause timing drift, impacting throughput and mobility. The TSS mechanism improves network resilience by enabling rapid correction, reducing link failures and service degradation caused by timing errors.

II. RAN3 Technology Applications

• Vehicle-mounted Relays (VMR for event coverage).
• Enterprise-grade NPN Phase 2 (SNPN Reselection/Handover).
• Automation (AI/ML SON automatically adjusts coverage).

III. RAN3 Practical Applications

• CU/DU: F1AP extension for AI model parameters (e.g., input/output tensors); Mobile IAB MT migration is achieved through Xn handover.
• Application Examples: Mobile IAB-DU reselection broadcasts the mobile IAB-Cell indicator; UEs use SIB-assisted priority ranking, thereby reducing topology change latency by 40%.