3GPP TSG RAN WG1 #120bis			R1- 2501829
Wuhan, China, April 7th – 11th, 2025

Source: 	vivo
Title:	Discussion on UE features for AIML Air interface
Agenda Item:	9.15.1
Document for:	Discussion and Decision

Conclusion
In this contribution we provide updates on UE features for AI/ML based beam management, positioning and CSI prediction.
Introduce the following Rel. 19 UE FGs for AI/ML based beam management in section 2.
Introduce the following Rel. 19 UE FGs for AI/ML based positioning in section 2.
Introduce the following Rel. 19 UE FGs for AI/ML based CSI prediction in section 2.

3GPP TSG RAN WG1 #120bis	R1-2501920
Wuhan, China, April 7th – 11th, 2025

Title: 	Discussion on other aspects of AI/ML model and data
Source: 	ZTE Corporation, Sanechips
Agenda item:	9.15.1
Document for:	Discussion/Decision

Conclusion
In this contribution, we provide our analysis and proposals for the UE feature design for Rel-19 AI/ML use cases for air interface.
General principle
Proposal 1: Consider the following two principles for Rel-19 UE feature design for AI/ML use cases for physical layer
Principle#1: Balance between UE capability report and applicability report
Principle#2: Balance between UE implementation flexibility and UE implementation fragmentation

AI beam prediction
Proposal 2: For Rel-19 AI beam prediction, three basic UE feature groups are sufficient
One basic UE feature group for more than 4 beam related information in L1 signalling
One basic UE feature group for beam prediction case 1 with UE side model
One basic UE feature group for beam prediction case 2 with UE side model
Proposal 3: Consider the following UE features for AI beam prediction in Rel-19.

AI POS enhancement
Proposal 4: New UE feature for AI POS enhancement case 1 is needed, while no need to define any new UE feature for AI POS enhancement case 3a and case 3b.
Proposal 5: Consider the following UE features for AI POS enhancement in Rel-19.

Proposal 6: RAN1 further discusses whether new UE features are required for the following aspects specifically for the AI POS enhancement case 1.
Common DL PRS Processing Capability, e.g., FG 13-1
DL PRS Resources, e.g., FG 13-3, 13-3a, 13-3b
Support of PRS measurement in RRC_INACTIVE state, e.g., FG 27-18a

AI CSI prediction
Proposal 7: RAN1 prioritizes UE feature discussion for other AI use cases first and come back to the UE feature discussion for AI CSI prediction in May or August meeting.

3GPP TSG RAN WG1 #120bis		                                                                                                                                                        R1-2501979
Wuhan, China, April 7th – 11th, 2025

Source:	CATT, CICTCI
Title:	Discussion on UE features for AI/ML for NR Air Interface
Agenda Item:	9.15.1
Document for:	Discussion and Decision

Conclusion
In this contribution, we provided our views regarding RAN1 UE features for AI/ML for NR air interface. The following UE capabilities are proposed:
Beam management

Positioning accuracy enhancement

CSI prediction

3GPP TSG RAN WG1 #120bis	  																		R1-2502047
Wuhan, China, April 7th – 11th, 2025

Agenda item:	9.15.1
Source: 	Ofinno
Title: 	Views on UE features for AI/ML for NR Air Interface
Document for:	Discussion/Decision

Views on UE features for AI/ML for NR Air Interface
In this contribution, we present our views on UE features for AI/ML for NR air interface. We propose:

Proposal 1: For FG XX-1-1, support BM-Case1 as a separate FG
Proposal 1-1: Support the following component in FG XX-1-1 based on the agreement below:
Number of CSI reports for inference that can be configured/activated/triggered.




Proposal 2: For FG XX-1-2, support BM-Case2 as a separate FG
Proposal 2-1: Support the following components in FG XX-1-2 based on the agreement below:
Maximum supported N [1, 2, 4, 8]
Maximum time gap [10ms, 20ms, 40ms, 80ms, 160ms]
Minimum instances of measured Set B of RSs for prediction








AI/ML based beam management
 

TDoc file conclusion not found

3GPP TSG-RAN WG1 Meeting #120bis																										R1-2502102
Wuhan, China, April 7th – 11th, 2025

Agenda Item:	9.15.1
Source:	LG Electronics
Title:	Discussions on UE features for AI/ML for NR Air Interface
Document for:	Discussion and Decision

Conclusion
In this contribution, we provided our views on Rel-19 UE features for AI/ML for NR Air Interface.

3GPP TSG RAN WG1 #120bis	R1- 2502133
Wuhan, China, 7 – 11 April 2025

Agenda item:		9.15.1
Source:	Nokia
Title:	Views on UE features for AI/ML for NR Air Interface
Document for:		Discussion and Decision

Conclusion
In this contribution, we presented our views, suggestions, and proposals on UE features for AI/ML for NR Air Interface, and we have made the following proposals
Proposal 1: Adopt the proposed FGs and components in Table 1 to Rel-19 UE features for AI/ML Beam Management for NR Air Interface 
Proposal 2: Adopt the proposed FGs and components in Table 2 to Rel-19 UE features for AI/ML Positioning for NR Air Interface 

3GPP TSG RAN WG1 #120bis								R1-2502178
Wuhan, China, April 7th – 11th, 2025
Source: 	CMCC
Title:	Discussion on UE features for AI/ML for NR air Interface
Agenda item:	9.15.1
Document for:	Discussion & Decision

1 

TDoc file conclusion not found

3GPP TSG-RAN WG1 Meeting #120bis	 R1-2502246
Wuhan, China, April 7 – 11, 2025

Agenda Item:	9.15.1
Source:	Huawei, HiSilicon
Title:	UE features for AI/ML for NR air interface
Document for:	Discussion and Decision

Conclusions
In this contribution, we discussed the basic FGs for Rel-19 NR_AIML_air. The following proposals are provided.
Proposal 1: Regarding the general principle for UE feature of UE-side models:
Each basic FG should include complete LCM components of data collection, inference, monitoring, and fallback.
Adopt the notation of ‘functionality for use case x’ as the description of the FG.
For BM and CSI prediction, dedicated CSI processing unit pool (XPU) is introduced as UE capability.
Proposal 2: Introduce the following basic FG for functionality for beam management:
FG 58-1: Increased number of CSI-RS for beam measurement and beam reporting.
Proposal 3: The basic capability of FG 58-1 should include the components listed in Table 1.
Proposal 4: Introduce the following 4 basic FGs for functionality for beam management of UE-side model:
FG 58-2: Functionality for beam ID prediction.
FG 58-3: Functionality for beam RSRP prediction.
FG 58-4: Functionality for temporal-domain beam ID prediction.
FG 58-5: Functionality for temporal-domain beam RSRP prediction.
Proposal 5: The basic capability of FG 58-2 to FG 58-5 should include the components listed in Table 2.
Proposal 6: Introduce the following basic FG for functionality for positioning accuracy enhancement:
FG 58-6: Functionality for fingerprint-based positioning.
Proposal 7: The basic capability of FG 58-6 should include the components listed in Table 3.
Proposal 8: Introduce the following basic FG for functionality for CSI prediction:
FG 58-7: Functionality for CSI prediction.
Proposal 9: The basic capability of FG 58-7 should include the components listed in Table 4.
Other enhanced FGs may borrow the corresponding enhanced FGs for Rel-18 CSI prediction.

3GPP TSG RAN WG1 #120bis									 R1-2502292
Wuhan, China, April 7th – 11th, 2025
	 							    		

Source:	OPPO
Title:	UE features for AIML for air interface
Agenda Item:	9.15.1
Document for:	Discussion and Decision

Conclusion
In this contribution, we discuss the potential UE features for AIML for air interface with the following proposals:
Proposal 1: The UE features shall be considered and supported for AIML for beam management.
Proposal 2: Support information related to DL PRS and UL SRS as in the capability signaling of legacy methods, detailed capability value can be further discussed if needed.
Proposal 3: Introduce the Rel. 19 UE FGs for AI/ML based CSI prediction.

3GPP TSG RAN WG1 #120bis					           	                                                         R1-2502391 
Wuhan, China, April 7th –11st, 2025
Agenda item:	9.15.1
Source:	Samsung
Title:	UE features for AI/ML for NR Air Interface
Document for:	Discussion and Decision

Conclusion
The proposals made in this contribution are summarized below.
Proposal 1: Introduce the following Rel. 19 UE FGs for AI/ML based beam management.

Proposal 2: Introduce the following Rel. 19 UE FGs for AI/ML based positioning.

Proposal 3: Introduce the following Rel. 19 UE FGs for AI/ML based CSI prediction.

3GPP TSG RAN WG1 #120bis                                                    R1-2502459
Wuhan, China, April 7th – 11th, 2025

Agenda Item:	9.15.1
Source:	Xiaomi
Title:	Discussion on UE features for AI/ML for NR Air Interface
Document for:	Discussion

Conclusion
Based on the discussion, we propose the FGs need to be supported.
AI/ML for beam management
Proposal 2-1: Support the following FGs/rows for AI/ML based beam management. 


AI/ML for positioning

Proposal 3-1: Consider “Support of Inference operation for AI/ML direct positioning with UE side model” as one UE feature for Case 1 
Set Existing FG 13-1 as Prerequisite feature groups 
Proposal 3-2: For performance monitoring, wait for sufficient progress on whether set it as one separate new UE feature 
Proposal 3-3: For training data collection:
For Case of part A and part B generated by UE, there is no need to define related UE feature
For Case of part A generated by UE and part B generated by LMF, whether new UE feature is necessary is up to other WG
Proposal 3-4: Consider the following table as baseline for UE feature discussion in Case 1 AI/ML positioning

AI/ML for CSI prediction
Proposal 4-1: Support to introduce the feature group on AI model based CSI prediction as one UE feature group and the prerequisite feature group of the new introduced feature group is support of Rel-16-based doppler CSI.
Proposal 4-2: Support to introduce a new feature group on aperiodic CSI report timing relaxation at least for AI/ML model switches/activates.
Proposal 4-3: Support to introduce new feature on maximum number of aperiodic CSI-RS resources that can be configured in the same CSI report setting for AI/ML model based CSI prediction.
Proposal 4-4: Support to introduce feature group on performance monitoring for CSI prediction model.
Proposal 4-5: Support to introduce feature group on data collection for CSI prediction model training if CSI-RS resource enhancement is considered.
Proposal 4-6: The following feature groups on AI/ML-based CSI prediction in the Table could be supported.


References

3GPP TSG RAN WG1 #120bis																                             R1- 2502639
Wuhan, China, April 7th – 11st, 2025

Agenda item:	9.15.1
Source: 	Apple Inc.
Title: 	Views on UE features for AI/ML for NR Air interface  
Document for:	Discussion/Decision

General structure for AI based UE feature discussion 
In this contribution, we present our high-level views on UE features for Rel-19 AI/ML for NR air interface. At a high level, a set of proposals on UE features related to LCM across different use cases are proposed. Detailed FGs for each use caseare discussed separately. 
For the one sided model, depending on the location of AI/ML inferencing, either UE-side model or NW-side model are specified. Supporting UE side model or NW side model requires separate UE capabilities. 
Proposal 1: Per use case, define separate UE feature groups for NW side model and UE side model.  

For NW side model, the main specification effort is on data collection for training, inferencing and performance monitoring. 
On data collection for training, the current MDT framework is being enhanced in RAN2 with logging capability to enable efficient and large delay tolerant data transfer.  
On data collection for inferencing, typically L1 reports are used to enable NW inference operation which is delay sensitive. 
On data collection for performance monitoring, it is still under discussion whether L1 report is used, or L3 report is used for AI based beam management. UE support highly depends on the final design of the detailed performance monitoring methods.  
For positioning, all signaling is based on LPP. 
Due to the large variance  of the requirement for data collection, separate UE feature groups are required for data collection for the NW side model. 

Proposal 2: For NW-side model, define separate UE features for data collection for training, inference and performance monitoring.   
Note: positioning case 3a and 3b with NW-side model or LMF side model are not related to the UE feature discussion.  

For UE side model, one key issue is whether a UE that supports inference should also support performance monitoring. Performance monitoring was discussed in various use cases, and at a  high level, the functionality level activation/de-activation/fall-back are controlled by the NW. Various methods of performance monitoring are also discussed. UE features indicating support for  one or multiple options of performance monitoring should be defined. 
Proposal 3: For UE-side model, define separate UE features for data collection for training, inference and performance monitoring.   UE features indicating support for  one or multiple options of performance monitoring should be defined
 
Particularly for data collection for training, the NW side model and UE side model has a very different discussion. For NW side model, logged measurement for RRC_connected mode UEs are specified in R19. For UE side data collection for UE side model, different options have been studied but not yet specified.  

Proposal 4: For NW side model data collection, define separate UE capability under UE-BasedPerfMeas-Parameters:   
loggedMeasurements-dataCollection-r19
Indicates whether the UE supports logged measurements for NW-side data collection in RRC_Connected state. A UE that supports logged measurements shall support both periodical logging and event-triggered logging. The minimum memory size of logged measurements for NW-side data collection is 64KB.  

Proposal 5: For NW side model data collection, similar to immediate MDT framework, the specific measurement to log is separate capability. 
For example, L1 RSRP for set A/B for AI based beam management
AI/ML for beam management 
In this section

FGxx-1x covers the AI/ML for beam management     

The following new FGs are for AI/ML-based beam management






AI/ML based positioning 
In this section

FGxx-2-1-x covers UE-based direct positioning with UE-side model (Case 1)
FGxx-2-1-x covers AI/ML for positioning Case 3a and 3b

3.1 FGxx-2-1x: UE-based direct positioning with UE-side model (Case 1)

In RAN2 #129, the following agreement was made



As such, we make the following proposals:
Proposal 6: From the Rel-16 UE feature list for NR Positioning FG 13-1, 13-1a, 13-7, 13-7a are applicable AI/ML-based positioning case 1


Proposal 7: From the Rel-17 UE feature list for NR Positioning FG 27-3-2, 27-3-3, 27-6, 27-9, 27-10, 27-10a and 27-11 are applicable AI/ML-based positioning case 1

Proposal 8: From the Rel-18 UE feature list for NR Positioning, the applicable FGs for AI/ML-based positioning case 1 are FFS.

The following new FGs are for AI/ML-based positioning.




3.2 FGxx-2-2-x: AI/ML for positioning Case 3a and 3b

Proposal 9: From the Rel-16 UE feature list for NR Positioning all FGs on SRS transmission are applicable AI/ML-based positioning case 3a and 3b


Proposal 10: From the Rel-17 UE feature list for NR Positioning all FGs on SRS transmission are applicable AI/ML-based positioning case 3a and 3b


Proposal 11: From the Rel-18 UE feature list for NR Positioning, the applicable FGs for AI/ML-based positioning case 3a and 3b are FFS.

AI/ML based CSI prediction 
FGxx-3x covers the AI/ML for CSI prediction. There is only one meeting for AI based CSI prediction, with limited agreement. 
 






Proposal 12: From the Rel-18 UE feature list for Doppler Codebook, FG 40-3-2-1a, FG 40-3-2-1a-1, FG 40-3-2-1b,  FG 40-3-2-2, FG 40-3-2-3, FG 40-3-2-3a, FG 40-3-2-4, FG 40-3-2-4b, FG 40-3-2-5, FG 40-3-2-6, FG 40-3-2-7, FG 40-3-2-7a, FG 40-3-2-8,  FG 40-3-2-9, FG 40-3-2-10, are applicable AI/ML CSI prediction.  


Conclusion: 
 In this contribution, we discussed high level proposals on UE feature discussion for AI/ML for air interface. The proposals are: 

Proposal 1: Per use case, define separate UE feature group for NW side model and UE side model.  

Proposal 2: For NW-side model, define separate UE feature for data collection for training, inference and performance monitoring.   
Note: positioning case 3a and 3b with NW-side model or LMF side model are not related to UE feature discussion.  

Proposal 3: For UE-side model, define separate UE feature for data collection for training, inference and performance monitoring.   

Proposal 4: For NW side model data collection, define separate UE capability under UE-BasedPerfMeas-Parameters:   
loggedMeasurements-dataCollection-r19
Indicates whether the UE supports logged measurements for NW-side data collection in RRC_Connected state. A UE that supports logged measurements shall support both periodical logging and event-triggered logging. The minimum memory size of logged measurements for NW-side data collection is 64KB.  

Proposal 5: For NW side model data collection, similar to immediate MDT framework, the specific measurement to log is separate capability. 
For example, L1 RSRP for set A/B for AI based beam management

Proposal 6: From the Rel-16 UE feature list for NR Positioning FG 13-1, 13-1a, 13-7, 13-7a are applicable AI/ML-based positioning case 1

Proposal 7: From the Rel-17 UE feature list for NR Positioning FG 27-3-2, 27-3-3, 27-6, 27-9, 27-10, 27-10a and 27-11 are applicable AI/ML-based positioning case 1

Proposal 8: From the Rel-18 UE feature list for NR Positioning, the applicable FGs for AI/ML-based positioning case 1 are FFS.

Proposal 9: From the Rel-16 UE feature list for NR Positioning all FGs on SRS transmission are applicable AI/ML-based positioning case 3a and 3b

Proposal 10: From the Rel-17 UE feature list for NR Positioning all FGs on SRS transmission are applicable AI/ML-based positioning case 3a and 3b

Proposal 11: From the Rel-18 UE feature list for NR Positioning, the applicable FGs for AI/ML-based positioning case 3a and 3b are FFS.

Proposal 12: From the Rel-18 UE feature list for Doppler Codebook, FG 40-3-2-1a, FG 40-3-2-1a-1, FG 40-3-2-1b,  FG 40-3-2-2, FG 40-3-2-3, FG 40-3-2-3a, FG 40-3-2-4, FG 40-3-2-4b, FG 40-3-2-5, FG 40-3-2-6, FG 40-3-2-7, FG 40-3-2-7a, FG 40-3-2-8,  FG 40-3-2-9, FG 40-3-2-10, are applicable AI/ML CSI prediction.  

TDoc file conclusion not found

3GPP TSG-RAN WG1 Meeting #120bis	Tdoc R1-2502661
Wuhan, China, April 7th – April 11st, 2025

Agenda Item:	9.15.1
Source:	Ericsson
Title:	UE Features for Rel-19 AI/ML for NR Air Interface
Document for:	Discussion, Decision

Conclusion
In the previous sections we made the following observations: 
Observation 1	Depending on the alternative chosen, an FG for support Case 1 positioning assistance data  should be introduced.
Observation 2	If RAN1 agrees to support sample-based and path-based measurement for Case 1, then a UE feature may need to be introduced on the measurement type for training data collection of Part A.


Based on the discussion in the previous sections we propose the following:
Proposal 1	Introduce a UE feature for BM-Case 1, including the support of:
a.	CRI-only prediction
b.	CRI+RSRP prediction
c.	The maximum values of resources in set A
Proposal 2	Introduce a UE feature for BM-Case 2, including the support of:
a.	CRI-only prediction
b.	CRI+RSRP prediction
c.	The maximum values of resources in set A
d.	The time gap values between two consecutive future time instances for prediction.
e.	The maximum values on the number of future time instances to predict (e.g. 1,2,3)
Proposal 3	For AI/ML BM, discuss if a separate UE feature for UE-assisted performance monitoring of a UE-sided model is needed, or if it should be part of the BM-Case 1 and/or 2 feature
Proposal 4	For AI/ML BM, introduce a UE feature for the data collection, including the support of a maximum value of resources in set A
Proposal 5	For the BM feature, discuss UE capabilities in support for the maximum number of
a.	configured or active associated IDs
b.	CSI reports for inference that can be configured/activated/triggered
Proposal 6	For the NW-sided beam prediction, introduce a UE feature for reporting more than 4 beams in L1-signalling
Proposal 7	Regarding Rel-19 AI/ML positioning use case, UE features only need to be considered for Case 1.
Proposal 8	Introduce a single FG for support Case 1 positioning method and related LCM procedures for the functionality, including at least model inference, model performance monitoring, functionality activation/deactivation.
a.	Note: the FG may not be a L1 FG.
Proposal 9	No additional FG are needed for case 1 monitoring if option B is not specified.
Proposal 10	Discuss whether UE features need to be introduced for model performance monitoring of Case 1, after RAN1 finalize the discussion of Option A-1/2/3 and Option B.
Proposal 11	The basic feature for Rel-19 CSI prediction using UE-sided model for the case of  is defined as follows:
a.	Include components 1-5 and 9 of feature group 40-3-2-1,
b.	Additional components related to CSI processing criteria and timeline with P/SP/A-CSI-RS, performance monitoring, data collection, etc., if needed, can be added further pending further RAN1 agreements.
Proposal 12	For Rel-19 CSI prediction using UE-side model, include the Rel-16 eType2 codebook as prerequisite to the basic feature for Rel-19 CSI prediction using UE-side model.
Proposal 13	Support the following additional feature for Rel-19 CSI prediction using UE-sided model for the case of :
a.	Include components 1-4 of feature group 40-3-2-1a
b.	Make the above basic feature in Proposal 11 as pre-requisite of this additional feature
c.	Additional components related to CSI processing criteria and timeline, performance monitoring, data collection, etc., if needed, can be added further pending further RAN1 agreements
 

3GPP TSG RAN WG1 #120bis	                                   					       												                           R1-2502738
Wuhan, China, April 7th – 11th, 2025

Agenda Item:	9.15.1
Source:	Moderator (AT&T)
Title:	Summary of UE features for AI/ML for NR Air Interface
Document for:	Discussion/Decision

Conclusion
Agreements reached during RAN1 #120bis as part of this agenda item are summarized in [ ].  

3GPP TSG RAN WG1#114	 	 	R1-2502785
Wuhan, CN, Apr. 7th – 11st, 2025
Source:	NTT DOCOMO, INC.
Title:	Discussion on UE features for AI/ML for NR Air Interface
Agenda Item:	9.15.1
Document for: 	Discussion and Decision

Conclusion
In this contribution, we discussed on UE features for AI/ML for NR Air Interface. Based on the discussion, we made following proposals.
FGs for AI/ML in beam management
Proposal 1:  Introduce the following FG for the enhancements for the data collection in NW side beam prediction.

Proposal 2:  Introduce the following FG for the enhancements on the UE side data collection.

Proposal 3:  Introduce the following FG for inference result reporting for BM-Case1.

Proposal 4:  Introduce the following FG for inference result reporting for BM-Case2.

Proposal 5:  Introduce the following FG for UE assisted performance monitoring.
FGs for AI/ML for positioning accuracy enhancement
Proposal 6: For the new capability of AI/ML based positioning for case 1, two alternatives can be considered.
Alternative 1: For case 1, regarding DL PRS Resources, the legacy capability for DL-TDOA can be reused for AI/ML positioning case 1. New capability is not necessary for AI/ML positioning case 1. 
Alternative 2: Similarly to the legacy positioning methods, the following capabilities for AI/ML based positioning can be considered based on RAN2 agreement:
DL PRS Resources for AI/ML based positioning
DL PRS Resources for AI/ML based positioning on a band
DL PRS Resources for AI/ML based positioning on a band combination

TDoc file reading error

3GPP TSG RAN WG1 #120-bis                                                                                                                                                                                                                                                                   					                    R1-2502864 Wuhan, China, April 7th – 11th , 2025 

Agenda item:		9.15.1
Source: 		Qualcomm Incorporated
Title: 			Discussion on UE features for AIML use cases
Document for:		Discussion and Decision

Feature groups for AIML-based beam management

The following are proposed for BM-Case1 and BM-Case2 to be considered for defining Rel. 19 UE FGs for AI/ML-based beam management.
The following cases for temporal beam prediction have been identified in TR 38.843:
Case A: based on number of measurements/RSs and prediction time. 
where T2 is the time duration for beam prediction
where Mt is the number of time instances for measurement as AI/ML inputs with a periodicity of Tper 
where Pt is the number of time instance(s) for prediction with a periodicity of Tper in T2
Case B+: based on Y times of a given minimal periodicity Tper of the reference signals for measurements.
For non-AI baseline (Option 1), UE measures all the reference signals of Set A every Tper 
For AI, UE measures the reference signals of Set B every Y times of Tper
In this case, prediction time is defined as the time from each measurement instance to the latest prediction instance before the next measurement instance.
The above cases are depicted in Figure 1 and Figure 2:

Figure 1 Example for Case A

Figure 2 Example for Case B+

Proposal 1: While designing the FGs, both flavors of temporal beam prediction in TR 38.843 should be supported (both Case A and Case B+):
Case A: based on number of measurements/RSs and prediction time. 
where T2 is the time duration for beam prediction
where Mt is the number of time instances for measurement as AI/ML inputs with a periodicity of Tper 
where Pt is the number of time instance(s) for prediction with a periodicity of Tper in T2
Case B+: based on Y times of a given minimal periodicity Tper of the reference signals for measurements.
For non-AI baseline (Option 1), UE measures all the reference signals of Set A every Tper 
For AI, UE measures the reference signals of Set B every Y times of Tper
In this case, prediction time is defined as the time from each measurement instance to the latest prediction instance before the next measurement instance.
Proposal 2: For BM-Case 1 and BM-Case 2, define separate FGs for “Set B is a subset of Set A” and “wide-to-narrow beam prediction” (Set B is composed of wide beams and Set A is composed of narrow beams).

As elaborated in our companion paper under section 9.1.1 and 9.1.4.1, for some UEs, the AIML operation may share a dedicated hardware which is independent to the legacy features, and for some other UEs, they may use the same hardware as legacy. For both of them, RAN1 needs to discuss whether a dedicated AI processing units pool is needed or the AIML processing and its pre-processing/post-processing can share the same CPU pool as legacy. Since the processing complexity is highly related to the model development choice of UEs, it would be beneficial for the UE to report their concurrent processing capabilities, in terms of either CPU criteria or other ways. The detailed UE capability reporting can be determined after the criteria is agreed.
Proposal 3: Support UE capability signaling of AIML-enabled processing units. The detailed signaling can be determined after discussion on the AI/ML processing in 9.1.1/9.1.4.1 agenda.

Proposal 4: Introduce the following Rel. 19 UE FGs for AI/ML-based beam management. 
















Feature groups for AIML-based positioning

RAN2 agreed to introduce AI/ML positioning Case 1 as a new positioning method, which means it needs to have dedicated listing of UE features that relate to capabilities of PRS, reporting, and receiving of assistance data (AD) (which is equivalent to other positioning methods). It was also agreed that UE in Case 1 can obtain most of ADs of UE-based DL-TdoA method.

Observation 1: The AI/ML positioning Case 1 is to be introduced as a new positioning method, requiring specific features related to capabilities of PRS, reporting, and receiving AD. 
Observation 2: The AI/ML positioning Case 1 is agreed to share most of ADs of UE-based DL-TDoA. 

From RAN1 perspective, UE features for AI/ML positioning Case 1 can include those equivalent features of UE-based DL-TdoA method, including features related to PRS resource capabilities, PRS processing capabilities, PRS QCL processing capabilities, on-demand PRS capabilities, capabilities related to reporting LocationEstimate, and capabilities for receiving assistance data (AD), as follow:
RS configuration related features
Supported PRS resource configurations
Supported PRS QCL processing
Supported PRS processing 
Supported on-demand PRS
Supported on-demand PRS with bandwidth aggregation (BWA)
Reporting related features:
Support for Periodical reporting
Support for Ten milliseconds response time
Support for Periodic reporting interval in milliseconds
Support for Scheduled location request
Support for Multiple location estimates in same measurement report
Supported reporting shapes
AD related features:
Support for receiving AD information (e.g., agreed ADs from AD#1 to AD#18 of previous agreements)
We propose, as a starting point, to consider UE features of UE-based DL-TdoA for introducing those of AI/ML positioning Case 1.
Proposal 1: In AI/ML positioning Case 1, for UE features, as starting point, consider UE features of UE-based DL-TdoA capabilities (e.g., PRS resource, PRS processing, PRS QCL processing, on-demand PRS, supported reporting, supported ADs, etc.), as in the following table:





Feature group for AIML-based CSI prediction
In our view, R19 AIML based CSI prediction is a new CSI codebook / reporting though the codebook type is same as legacy feature. In other words, it can be considered as an AIML version of the legacy feature where the prediction part is performed via AIML, but the same codebook is used. Thus, R19 AIML based CSI prediction should not be prerequisite on R18 doppler codebook features. Instead, it should be dependent on FG 2-35, and individually signal the supported CSI-RS triplets and supported codebook configurations.
Rel-19 AIML-based CSI prediction is a separate CSI codebook / reporting compared to legacy R19 AIML doppler codebook though same codebook is used.
Rel-19 AIML-based CSI prediction should NOT have Rel-18 predicted CSI as prerequisite feature group.
For R19 AIML-based CSI prediction UE feature, copy FG 40-3-2-1 family (R16 eType II based prediction N4=1), FG 40-3-2-1a family (R16 eType II based prediction N4>1), FG 40-3-2-1b (max number of A-CSI-RS resource), FG 40-3-2-2 family (R16 eType II based prediction R=2), FG 40-3-2-3 family (CQI), FG 40-3-2-7 (starting location), FG 40-3-2-8, FG 40-3-2-9 and FG 40-3-2-10 by changing the feature group name to AIML-based CSI prediction.
Additionally, similar to legacy UE capability signalling for concurrent codebooks FG 16-8 and FG 23-9-5, concurrent processing of AIML based CSI prediction and legacy non-AI CSI codebooks should be signaled so as to avoid UE underreporting its capability.
Support UE capability signalling for CSI-RS triplets concurrent processing of AIML-based CSI predication and other legacy CSI reports. The signalling granularity should be per-band and also per band-combination.
Another aspect of UE capability is regarding the processing criteria and timeline. As elaborated in our companion paper under section 9.1.3 and 9.1.4.1, for some UEs, the AIML operation may share a dedicated hardware which is independent to the legacy features, and for some other UEs, they may use the same hardware as legacy. For both of them, RAN1 needs to discuss whether a dedicated AI processing units pool is needed or the AIML processing and its pre-processing / post-processing can share the same CPU pool as legacy. Since the processing complexity is highly related to the model development choice of UEs, it would be beneficial for the UE to report their concurrent processing capabilities, in terms of either CPU criteria or other ways. The detailed UE capability reporting can be determined after the criteria is agreed. 
RAN1 will need to discuss additional timeline is needed for AIML rather than the existing timeline for the non-AIML prediction and whether additional time is needed for loading models to the modem. Similar to concurrent processing criteria, timeline is also related the model design and hardware architecture. Thus, it might be premature to determine a hardcoded timeline in the spec, UE capability reporting for the timeline can be a more flexible solution.
Support UE capability signaling of AIML-enabled processing units and timeline. The detailed signalling can be determined after discussion on the processing / timeline criteria in 9.1.3 / 9.1.4.1 agenda.
In summary, based on proposals above, the feature groups for AIML-based CSI prediction should contain the followings.

TDoc file conclusion not found

TDoc file reading error

3GPP TSG RAN WG1 #120bis	                                   					       												                           R1-2503131
Wuhan, China, April 7th – 11th, 2025

Agenda Item:	9.15.1
Source:	Moderator (AT&T)
Title:	Final summary of UE features for AI/ML for NR Air Interface
Document for:	Discussion/Decision

Conclusion
Agreements reached during RAN1 #120bis as part of this agenda item are summarized in [17].