3GPP TSG-RAN WG2 Meeting #130	 R2-2503778
St. Julians, Malta, May 19th – 23rd, 2025

Agenda Item:	8.3.1    
Source:	Mediatek Inc.
Title:	Report of [POST129bis][020][AI Mob] Sim. Results Figures (Mediatek)
Document for:	Discussion, Decision

Conclusion
Proposal 1: Endorse the performance evaluation spreadsheets as attachment to the TR.
Proposal 2: Endorse the CDF figures plotted based on the collected simulation results from companies in each individual spreadsheet.
Proposal 3: Select the CDF curves that align with the current observation trends to be included in the TR. The rapporteurs will choose the appropriate CDF curves as needed during the drafting of the TR. 

3GPP TSG-RAN WG2 Meeting #130	                    R2-2503780
St. Julians, Malta, May 19th – 23rd, 2025

Agenda Item:	8.3.1 
Source: 	MediaTek Inc., vivo, CATT, Qualcomm Incorporated, Nokia, Huawei, HiSilicon, Ericsson, Samsung
Title:	Support of Spatial Domain Prediction for AI Mobility
Document for:	Discussion, Decision

Conclusion
Observation on the performance on spatial domain prediction in beam dimension
Observation 1: As MRRS increases, both the average L1 beam-level RSRP difference and average L3 cell-level RSRP difference increase. 
Observation 2: As UE speed increases, the prediction accuracy decreases. 
Observation 3: Under the same MRRS setting, the selection of measured beam pattern can significantly affect the prediction accuracy (i.e. L3 cell-level RSRP difference) of the spatial domain prediction in beam dimension.
Observation 4: Under the same MRRS setting, sub-use case 1 demonstrates higher prediction accuracy than RRM sub-use case 3.
Observation 5: Under the same MRRS setting, cluster-based approach demonstrates higher prediction accuracy than the single-cell approach.
Observation 6: By spatial domain prediction in beam dimension, 50% reduction in reference signal and UE measurement effort can be achieved, without any degradation in mobility performance compared with legacy HO in the metrics such as HOF, Ping-pong, data interruption time, and average TOS. 

Proposal 
Spatial domain prediction across beams
Proposal 1a: Confirm the definition of ‘intra-frequency intra-cell spatial domain prediction (in beam dimenstion)’ in the TR, i.e. Intra-frequency intra-cell spatial domain prediction is evaluated for the 1st study goal by measuring a sub set of configured SSB as input to the model to derive L3 filtered cell-level measurements for every time instance of the same cell. 
Proposal 1b: Intra-frequency spatial domain prediction (in beam dimension) can utilize the ‘cluster approach’, i.e., in addition to the measurement of subset of configured SSB of the same cell, measurement results from other cell(s) are used as input to the model.
Spatial domain prediction across cells
Proposal 2: Intra-frequency spatial domain prediction (in cell dimension) involves measuring one or more cells as input to the model to derive L3 filtered cell-level measurements for every time instance of another cell(s).
Proposal 3: The scenario of intra-frequency spatial domain prediction (in cell dimension) is not considered by the UE-side model in 5G. RAN2 to discuss whether to consider it by the network-sided model. RAN2 assumes that a network-sided model requires no specification impacts. 
 
Observation based on current evaluation results
Proposal 4: Capture the following observations for intra-frequency spatial domain prediction (in bean dimension) in the TR:
Increasing MRRS correlates with decreased prediction accuracy. 
Increasing UE speed correlates with decreased prediction accuracy. 
Under the same MRRS setting, different measurement skipping patterns can result in different prediction accuracy.
Under the same MRRS setting, sub-use case 1 demonstrates higher prediction accuracy than RRM sub-use case 3.
Under the same MRRS setting, cluster-based approach demonstrates higher prediction accuracy than the single-cell approach.
For intra-frequency spatial domain prediction (in beam dimension) with MRRS=50%, the AI algorithm can achieve comparable system performance as the legacy mechanism.
Proposal 5: The specification impact of intra-frequency spatial domain prediction in beam dimension will be studied by RAN2 for both UE-side and network-side model. 

Appendix-2: Simulation Assumption