3GPP RAN#108 AML

swjtuhong

3GPP TSG RAN Meeting #108                                                                                    RP-251860

Prague, Czech Republic, June 9-13,2025                                                     (revisionof RP-yyxxxx)

Source:                     Qualcomm

Title:                          NewWI: Artificial Intelligence (AI)/Machine Learning (ML) for NR air interfaceenhancements

Documentfor:        Approval

AgendaItem:           15.1.1

3GPP™ WorkItem Description

Information on Work Items canbe found at http://www.3gpp.org/Work-Items
See also the 3GPPWorking Procedures, article 39 and the TSG Working Methods in 3GPP TR 21.900

Title:                        NewWI: Artificial Intelligence (AI)/Machine Learning (ML) for NR air interfaceenhancements

Acronym:                NR_AIML_air_Ph2

Unique identifier:     XXX

NOTE:      For new WIs/SIs leave the Uniqueidentifier empty and make a proposal for an Acronym.

                   For a revised WI/SI: Take Uniqueidentifier and acronym as shown in 3GPP workplan.

                   If this is a RAN WIDincluding Core and Perf. part, then Title, Acronym and Unique identifierrefer to the feature WI.

                   Please tick (X) theapplicable box(es) in the table below:

                   Either:

This WID includes a Core  part	X
This WID includes a  Performance part	X

                   or:

This WID includes a Testing  part
and it addresses the  following 3GPP work area:	Radio Access
	Core Network
	Services

Potential targetRelease:   Rel-20

NOTE: In case ofcontradiction with the target dates of clause 5, clause 5 determines the targetrelease.

1           Impacts

Affects:	UICC  apps	ME	AN	CN	Others  (specify)
Yes		X	X
No	X			X
Don't know

2           Classificationof the Work Item and linked work items
2.1          Primaryclassification

This description is either a…

Study Item

or a

Normative Work Item:    tick applicable boxes below
	Stage 1
X	Stage 2
X	Stage 3
	Other (e.g. testing)

2.2          ParentWork Item

For a brand-new topic, use “N/A” in thetable below. Otherwise indicate the parent Work Item.

Parent Work /  Study Items
Acronym	Working Group	Unique ID	Title (as in 3GPP  Work Plan)
FS_NR_AIML_air	RAN1	940084	Study on Artificial  Intelligence (AI)/Machine Learning (ML) for NR Air Interface
FS_NR_AIML_air_Ph2	RAN1	1060078	Study on Artificial  Intelligence (AI)/Machine Learning (ML) for NR air interface Phase 2

NOTE:      RAN agreedsome time ago, that it describes the feature WI + Core/Perf. part WI or Testingpart WI in one    WID. Therefore the tableabove should include the feature WI data (In case the feature covers Core andPerf.  part, please list under WorkingGroup the leading WG of the Core part).

2.3          Otherrelated Work Items and dependencies

Other related Work/Study  Items (if any)
Acronym	Unique ID	Title	Nature of  relationship
NR_AIML_air	1020093	Artificial Intelligence (AI)/Machine Learning  (ML) for NR Air Interface	Preceding work item.

NOTE:      Also related or dependent WIs/SIs in otherTSGs shall be indicated here.

Dependencyon non-3GPP (draft) specification:

None.

3           Justification

Theapplication of AI/ML techniques to NR air interface has been studied inFS_NR_AIML_air and FS_NR_AIML_air_Ph2. Normative support for the generalframework for AI/ML for air interface and some of the use cases studied in FS_NR_AIML_airwas provided through the NR_AIML_air related project.

In thiswork item, we provide the normative support for the CSI feedback enhancement (two-sidedmodel) use case as well as standards-based UE data collection.

4           Objective
4.1          Objectiveof SI or Core part WI or Testing part WI

Provide specification support for thefollowing aspects:

AI/ML frameworkfor two sided AI/ML models within the realm of what has been studied in theFS_NR_AIML_air_Ph2 and NR_AIML_air projects for CSI feedback enhancement[RAN2]:

-         Signalling and protocol aspects of Life CycleManagement (LCM) enabling functionality and model selection, activation,deactivation, switching, fallback

o    ID relatedsignalling is part of the above objective

-         Necessary signalling/mechanism(s) for LCM tofacilitate model training, inference, performance monitoring,.

CSIfeedback enhancement, encompassing (two-sided model) [RAN1, RAN2]:

-         CSI spatial/frequency compression without temporalaspects (“Case 0”),

-         Specify necessary signalling/mechanism(s) as per theidentified potential specification impacts in FS_NR_AIML_Air , including:

o    Modelpairing procedure including ID and applicability reporting

o    Inferenceaspects including target CSI type, measurement and report configuration, CQI RIdetermination, payload determination, quantization configuration codebook, UCImapping, CSI processing criteria and timeline, priority rules for CSI reports

o    NW and UEside data collection for training,

§  Target CSIformat

§  Note Theframework defined in BM and CSI prediction use cases could be reused

o    Specifyperformance monitoring

Inter-vendortraining collaboration for two-sided AI/ML models

-         Fully defined/specified reference model (“DirectionC”) with RAN1 scalability study outcome taken into account [RAN4/RAN1] – check-point in RAN#110 upon RAN4feedback

o   Specification of standardized encoder modelstructure plus parameter exchange (“Direction A, sub-option 3a-1” withouttarget CSI sharing) leveraging defined/reference model of “Direction C” andtaking RAN1 scalability study outcome into account [RAN4/RAN1/RAN2/RAN3] – check-point in RAN#110 upon SA WGfeedback

-         Specification of standardized dataset format/contentplus dataset exchange (“Direction A, sub-option 4-1”) [RAN1/RAN2/RAN3/RAN4] – check-point in RAN#110 upon SA WGfeedback

Datacollection [RAN2]:

-         Further study and specify if deemed necessary,standards-based UE data collection at least for UE-side model training in closecollaboration with SA WGs and taking into account the agreed principles duringthe Rel-19 study –check-point in RAN#110 based on SA WG feedback and SA2 study on data collection

Interoperabilityand RRM requirement [RAN4]:

-         To define the requirement for the encoder, specifythe standardized test decoder to guarantee interoperability

o    The fullyspecified test decoder, data set, and corresponding reference encoder in RAN4to be used for inter-vendor collaboration

-         Necessary RRM requirement for specified LCMprocedures.

-         NOTE: offline training is assumed for the purpose ofthis project. Strive to leverage on RAN1’s study on scalable model structure inFS_NR_AIML_Air.

4.2          Objectiveof Performance part WI

NOTE:      Leave empty if the WI proposal does not contain a RANperformance part.

Performancerequirements [RAN 4]:

-         Demodulation requirement for CSI compression Case 0

-         RRM performance test cases for the corresponding RRMrequirement for LCM procedures.

4.3          RANtime budget request (not applicable to RAN5 WIs/SIs)

NOTE:      For all new RAN related WIs/SIs which are not led byRAN WG5 the WI/SI rapporteur has to fill out the attached Excel table torequest time budgets for corresponding RAN WG meetings.
The Excel table has to be filled out for all affected RAN WGs and up to thetarget date of the WI/SI.
One time unit (TU) corresponds to ~ 2 hours in the meeting.
If no TU is needed, then leave the field empty otherwise enter a number >0in the field.

                   For revisions of already approved WI/SIdescriptions: Please remove the Excel table from the WID/SID's zip file.The time budgets are already recorded. If you want to modify them, then thishas to be done via the status report and not via a revised WID/SID.

                   If this WID is covering Core and Performance part,then please fill out one line for each part in the attached Excel table.

additional comments to the time budget request in theattached Excel table:

5           ExpectedOutput and Time scale

New  specifications {One line per specification. Create/delete  lines as needed}
Type	TS/TR number	Title	For info   at TSG#	For approval at TSG#	Remarks

NOTE:      If this is a RAN WI including Core andPerf. part, then all new Core part specs have to be listed first and then allnew Perf. part specs. Indicate "Core part" or "Perf. part"under Remarks for each spec.
By default a new specs can only be new for one of both parts.

Impacted  existing TS/TR {One line per specification. Create/delete  lines as needed}
TS/TR No.	Description of change	Target completion plenary#	Remarks
38.300	Feature introduction.	TSG#115	Core part
38.214	Feature introduction.	TSG#115	Core part
38.331	Feature introduction.	TSG#115	Core part
38.133	Feature introduction.	TSG#115	Core part
38.401	Feature introduction.	TSG#115	Core part
38.133	Feature introduction.	TSG#117	Perf. Part

NOTE:      If this is a RAN WI including Core andPerf. part, then all new Core part specs have to be listed first and then allnew Perf. part specs. Indicate "Core part" or "Perf. part"under Remarks for each spec.
If an existing spec is affected by both (Core part and Perf. part), then it hasto be listed twice with appropriate approval dates.

6           Workitem Rapporteur(s)

Juan Montojo, Qualcomm Inc., juanm@qti.qualcomm.com[RAN1]

Xiaofeng Liu, CAICT, liuxiaofeng1@caict.ac.cn [RAN4]

Marco Belleschi,Ericsson, marco.belleschi@ericsson.com, [RAN2]

Fahad SyedMuhammad, Nokia, fahad.syed_muhammad@nokia.com[RAN3]

NOTE:      The first listed Rapporteur has theoverall responsibility for this WI (incl all secondary tasks).

7           Workitem leadership

Primary: RAN1

Secondary: RAN2, RAN3, RAN4

8           Aspectsthat involve other WGs

For a Stage 2 WID requiring Stage 3 to be done by another group: on abest-effort basis, indicate which potential WG is expected to specify the Stage3: possible values: "Not applicable"

NOTE:      For RAN WIs: Section 8 applies only to WGs outside ofTSG RAN because all RAN WG aspects have to be covered in section 4.

9           SupportingIndividual Members

Supporting IM name

Qualcomm

CAICT

Ericsson

Nokia

CATT

CEWiT

China Telecom

China Unicom

ETRI

FirstNet

Futurewei

III

IITM

Intel

InterDigital

ITRI

Kyocera Corporation

LG Electronics

MediaTek

NEC

Nvidia

Ofinno

Panasonic

Samsung

Sanechips

SK Telecom

Spreadtrum

Telefonica

UNISOC

vivo

ZTE