RAN1 / Meeting 119 / NR_AIML_air

AI/ML for NR Air Interface

  NR_AIML_air · 213 contributions

Meeting outcomes

  From the official RAN1#119 chairman's report

Additional study on AI/ML for NR air interface

conclusion

For Direction A 4-1 and 3a-1, send LS to RAN2 for the study of feasibility of parameter / dataset exchange Wednesday further clarification regarding above LS, additionally CC SA2, SA3, SA5. Note: Samsung denies CC.

agreement

For Direction A Option 3a-1 and Direction C, study the feasibility of scalable model structure specification over numbers of Tx ports, CSI feedback payload sizes, and bandwidths, number of slots. Feasibility of Direction A Option 3a-1 is contingent on the feasibility of the scalable model structure specification. Feasibility of Direction C is contingent on the feasibility of the scalable model structure specification and model parameters specification. Note: Angular/Delay/Doppler domain conversion and quantization are considered to be part of feasibility study

agreement

For studying the standardized model structure, For temporal domain Case 0, In case of spatial-frequency domain input, adopt Transformer as the backbone structure. Further study angular-frequency and angular-delay domain input For temporal domain Case 2, In case of spatial-frequency domain input, take into account the model structure of Case 0 with an adaptation. Study the following options for the adaptation: Option 1: Input/output adaptation with additional layers Conv-LSTM LSTM Option 2: Latent adaptation on top of Case 0 model structure Reusing Case 0 structure and changing quantization operation and/or feedback payload over different slots Note: other options are not precluded Further study angular-frequency and angular-delay domain input For temporal domain Case 3, In case of spatial-frequency domain input, take into account the model structure of Case 0 with an adaptation. Study the following options for the adaptation: Option 1: Input/output adaptation Option 2: Latent adaptation Further study angular-frequency, angular-delay, and angular-delay-Doppler domain input. Take precoding matrix in the spatial-frequency domain in a subband granularity as the input, and the reconstructed precoding matrix in the spatial-frequency as the output, for the study of standardized model structure. Note: Processing steps at the UE to derive angular, delay, and/or Doppler domain basis, and corresponding reconstruction steps at the gNB are considered. Note: scalar/vector quantization and dequantization are considered. Quantization-aware training with jointly updated quantization method/parameters (Case 2-2) is used. Presented in Wednesday session.

agreement

For NW to collect data for training, study following spec impacts Data format: codebook-based Rel-16 eType2 or Rel-18 eType2 for PMI prediction. FFS if enhancement is needed FFS number of samples in the report. FFS whether channel or precoder is needed for temporal Cases 3 Configuration of rank/layer, number of subbands Mechanism for ground-truth reporting FFS: Report additional information regarding the samples, e.g., data quality, FFS the definition of data quality and corresponding parameters. FFS if enhancements in CSI-RS and SRS configuration is needed. FFS: Report associated information that captures UE side additional condition FFS: Configuration / reporting of temporal aspects for temporal Case 2 and Case 3, e.g., association between input and output CSI FFS: details of CSI measurement For UE to collect data for training, study following spec impacts NW configuration or UE request, e.g., RS configuration/transmission for data collection Whether enhancements in CSI-RS configuration is needed. Configuration of temporal aspects for temporal case 2/3, e.g., association between input and output CSI FFS: Need of configuration of ID, and configuration of ID.

agreement

To capture all observations in section 2 of R1-2410725 to TR38.843.

agreement

Draft LS R1-2410915 with the following updates: Revising “The size becomes 600K * (2000 bits) / (8bits/Byte) = 150 MB+ 11.6 MB = 161.6MB if we assume N1=N2.” To “The size becomes N2 * (2000 bits) / (8bits/Byte) + 11.6 MB” , and Revising Action: RAN1 respectfully asks RAN2’s feedback on the feasibility of standardized signaling (over-the-air and/or other approaches) for Dataset sharing consisting of {(Target CSI, CSI feedback)} Encoder parameter sharing Encoder parameter sharing + dataset sharing consisting of {target CSI} to Action: RAN1 respectfully asks RAN2’s feedback on the feasibility of standardized signaling (over-the-air and/or other approaches) for NW-side sharing model parameters and/or dataset to the UE or UE-side for the following options Dataset sharing consisting of {(Target CSI, CSI feedback)} Encoder parameter sharing Encoder parameter sharing + dataset sharing consisting of {target CSI} Deleting “RAN1 respectfully requests RAN2’s feedback on the feasibility of standardized signaling, considering the size of dataset / encoder parameters provided above.” Deleting “The size is based on Case 0 of two-sided CSI compression; the size may or may not be larger for Case 2 and Case 3.” from “For Option 4-1: sharing {target CSI, CSI feedback} dataset:” Final LS is approved in

agreement

For temporal domain aspects Case 3, study LCM aspects and specification impacts, consider the following options for training data collection Option 1: The target CSI for training is derived based on the predicted CSI of the future slot(s). Option 2: The target CSI for training is derived based on the measured CSI of the future slot(s). Note: During inference, the input to the CSI generation part is derived based on the predicted CSI. consider following options for the monitoring labels Option 1: The monitoring label is derived based on the predicted CSI of the future slot(s). CSI prediction output is used as input to CSI generation part. Note: This corresponds to monitoring of CSI compression only. CSI prediction may be monitored separately. Option 2: The monitoring label is derived based on the measured CSI of the future slot(s) Option 2a: CSI prediction output is used as input to CSI generation part. Note: This corresponds to end-to-end monitoring of CSI prediction and compression. Option 2b: Measured CSI of the future slot(s) is used as input to CSI generation part for monitoring purpose. Note: This corresponds to monitoring of CSI compression only. CSI prediction may be monitored separately. Study how the functionality/model control (activation, deactivation, switching, and fallback) for CSI prediction and CSI compression interacts. Final summary in R1-2410724.

agreement

For study of MI-Option2 (i.e., model identification with dataset transfer) for the two-sided model, ID-X can be used for pairing the UE-part and the NW-part of a two-sided model FFS: other information needed for pairing

agreement

Regarding the relationship of model ID, first indication, and second indication for model transfer/delivery Case z4, further study the following options: Opt.1: model ID consists of the information of the first indication and the second indication E.g., model ID is a combination of the first and second indications Opt.2: The second indication is assumed as the model ID Opt.3: Model ID is assigned by network and is separated from the first indication and the second indication

conclusion

Regarding MI-Option2 (i.e., model identification with dataset transfer) for the two-sided model, from RAN1 perspective, how to construct the dataset, including whether a dataset constructed from one cell or from multiple cells is up to network implementation.

conclusion

For the study of model delivery/transfer Case z4, if the model delivery/transfer is directly used for inference, the following options are identified as the candidate solutions to determine the readiness of AI model with the transferred parameters for inference (either or combination of the following options) Option 1: UE sends signaling to network to notify that the AI model with the transferred parameters is ready for activation Option 2: The AI model with the transferred parameters can be assumed ready for activation from a minimum applicable time after the completion of model delivery/transfer Final summary in R1-2410778.

Specification support for beam management

agreement

For UE-sided model, at least for BM-Case 1, the beam information in inference result report is CRI/SSBRI of resource in Set A.

        Company alignment · AI-analyzed
      

Samsung partially adopted Samsung proposed specific data collection contents including L1-RSRPs for Set A, which aligns with the focus on Set A, but the agreement restricts the report content to CRI/SSBRI identifiers rather than the full RSRP data Samsung advocated for.

China Telecom partially adopted China Telecom advocated for full Set A measurement for accurate monitoring, which is consistent with the agreement's reliance on Set A resources, though the agreement limits the immediate reporting to identifiers.

conclusion

For BM-Case 2 of UE-side model, only fixed Set B across different time instance is supported for single CSI report.

        Company alignment · AI-analyzed
      

China Telecom contradicted China Telecom explicitly 'Opposes using only single resource set configurations for Set B in UE-sided models,' which directly contradicts the agreement that 'only fixed Set B... is supported'.

Samsung partially adopted Samsung proposed configurability between alternatives for CSI-ReportConfig; while the agreement fixes Set B for Case 2, it does not necessarily preclude the configurability Samsung sought for other aspects or cases, but it narrows the scope for Case 2 specifically.

agreement

For both BM-Case 1 and BM-Case 2, for UE-sided model for inference, when Set A and Set B are configured within CSI report configuration, two CSI-ResourceConfigId s are configured for Set A and Set B separately. Presented in Wednesday session.

        Company alignment · AI-analyzed
      

CAICT adopted CAICT explicitly 'Proposes... separate CSI-ResourceConfigId configuration (Alt 3) for Set A/B,' which matches the agreement that 'two CSI-ResourceConfigIds are configured for Set A and Set B separately'.

Samsung partially adopted Samsung supports configurability between Alt 1 and Alt 3; the agreement adopts the separation (Alt 3) but does not explicitly confirm the flexibility to choose Alt 1, thus partially adopting the structural proposal.

agreement

In Step 3, following configurations are provided from NW to UE: UE is allowed to do UAI reporting via OtherConfig, The applicability report is based on A) and/or B) It is up to RAN 2 to design the container A) one or more of CSI-ReportConfig for inference configuration (wherein the associated ID may be configured in CSI framework as working assumption applied) Note: CSI report configuration for UE-side model inference can’t be activated immediately upon receiving Step 3 B) One set or multiple sets of inference related parameters for applicability report only (not for inference) It is up to RAN2 to design the container. The set of inference related parameters selected from the IEs in/or the IEs referred by CSI-ReportConfig as a starting point, e.g., the associated ID Note: this doesn’t imply the associated ID is mandatory Set A related information Set B related information Report content related information For BM-Case 2, Time instances related information for measurements Time instances related information for prediction In Step 4, UE reports applicability for all the above A) one or more CSI-ReportConfig and/or B) set(s) of inference related parameters FFS on whether/what other information along with the applicability is needed If A) is configured in Step 3, Applicable aperiodic CSI Report and semi-persistent CSI report can be activated/triggered by NW after the applicability reported. Applicable periodic CSI Report is considered as activated only if the applicability of the corresponding CSI-ReportConfig is reported in RRCReconfigurationComplete. In Step 5, NW can optionally configure CSI-ReportConfig for inference configuration in RRCReconfiguration, where the associated ID may be configured in CSI framework as working assumption applied. Note: Step 5 may be optional if UE has already been configured with CSI-ReportConfig in Step 3

        Company alignment · AI-analyzed
      

Fujitsu contradicted Fujitsu 'Advocates for Option-3 in applicable functionality reporting, favoring UE-driven parameter reporting rather than network-driven configuration' and 'Opposes Option-1 and Option-2,' whereas the agreement establishes a framework where the NW provides configurations (A) and/or parameters (B) in Step 3 for UE to report applicability, which aligns more with network-driven options (1/2) that Fujitsu opposed.

ETRI adopted ETRI 'Strongly supports Options 1 and 2 over Option 3 for applicability procedures,' and the agreement describes a process involving NW-provided configurations (Option 1/2 style) for applicability reporting, aligning with ETRI's preference.

Google adopted Google 'Strongly supports Option 1 for inference configuration management,' and the agreement includes 'one or more of CSI-ReportConfig for inference configuration' (Option 1) as a valid method for applicability reporting in Step 3.

agreement

At least for the monitoring Type 1 Option 2 of UE-side model monitoring (when applicable), support to reuse CSI framework for the configuration for monitoring result report in L1 signaling: Dedicated resource set(s) for monitoring and report configuration for monitoring are configured in a dedicated CSI report configuration used for monitoring The ID of an inference report configuration is configured in the configuration for monitoring to link the inference report configuration and monitoring report configuration FFS how to identify the connection between RSs in the resource set(s) for monitoring and Set A beams FFS on whether to support all the combination on time domain behavior of the reportConfigType for infernece report and the reportConfigType for monitoring report FFS on the timing related issues UE measures the dedicated resource set(s) for monitoring.

        Company alignment · AI-analyzed
      

CMCC adopted CMCC 'Supports dedicated resource sets and report configurations for UE-side model monitoring (Type 1 Option 2), linking the monitoring configuration to an inference report configuration via CSI-ReportConfig ID,' which matches the agreement's requirement for dedicated resources and linking IDs.

Samsung adopted Samsung 'prefers dedicated CSI report configurations for monitoring resources,' which aligns with the agreement's support for 'Dedicated resource set(s) for monitoring... configured in a dedicated CSI report configuration'.

Fujitsu adopted Fujitsu 'prefers separate configurations for monitoring versus inference,' which is consistent with the agreement's use of a 'dedicated CSI report configuration used for monitoring' linked to the inference configuration.

Fraunhofer partially adopted Fraunhofer 'Advocates for 2-phase monitoring with adaptive frequency,' which aligns with the monitoring focus, but the agreement defers timing issues ('FFS on the timing related issues'), meaning the specific adaptive frequency mechanism is not yet decided.

conclusion

For the CSI-ReportConfig for inference configuration provided in Step 5, aperiodic CSI Report and semi-persistent CSI report can be activated/triggered by NW after RRCReconfigurationComplete. periodic CSI Report is considered as activated after RRCReconfigurationComplete. Note: UE is not expected to be configured with a CSI-ReportConfig for inference configuration for a non-applicable set of inference parameters or a non-applicable CSI-ReportConfig Any specification impact is a separate discussion

        Company alignment · AI-analyzed
      

Google adopted Google 'Strongly supports Option 1 for inference configuration management to avoid multi-stage RRC configuration latency issues,' and the agreement clarifies activation timing relative to RRCReconfigurationComplete and applicability reports, providing the clarity Google sought to avoid latency.

Samsung partially adopted Samsung proposed 'scaling the legacy Z timeline based on UE capability' for CPU handling; the agreement defines activation times but does not explicitly adopt Samsung's specific proposal for scaling the Z timeline, leaving CPU handling details potentially open or separate.

agreement

Send LS to RAN2 with below information. RAN1 thanks RAN2 for the LS on applicable functionality reporting for beam management UE-sided model. In RAN1’s discussion of RAN 2 terminologies on beam management, The concept/terminology “functionality” of Supported functionalities may refer to UE-capability information/parameters i.e., Rel-19 AI/ML-enabled Features/FGs The concept/terminology “ functionality” of Applicable functionalities may refer to CSI-ReportConfig for inference configuration or a set of inference related parameters The Activated functionalities may be enabled based on CSI framework. Therefore, the meaning and the granularity of “functionality” for Applicable functionalities, Activated functionalities and Supported functionalities may or may not be the same. RAN 1 made the following agreements related to the Questions from RAN 2: RAN1 would like to provide replies on the following questions from RAN2 in R2-2407848: Q1: In Step 2, what is the granularity of functionality? For example, whether it is a use case (e.g. beam management), whether it is a sub-use case (e.g. beam management Case 1), or others? Answer to Q1: In Step 2, RAN1 expects that UE reports its UE-capability information/parameters, i.e., Rel-19 AI/ML-enabled Features/FGs (including components and corresponding value ranges). These UE capability information/parameters will depend on how FGs are defined including the granularity, that will be discussed in RAN1 later in the WI. Q2: What is the content of NW-side additional condition, i.e. is it correct the RAN2 assumption of a NW-side additional condition assumed as associated ID? Answer to Q2: RAN 1 did not have agreement on the content of NW-side additional condition. RAN1 agreed to support associated ID and it can be used to ensure the consistency of NW-side additional condition across training and inference for UE-sided model for BM-Case 1 and BM Case 2. UE may assume the similar properties of a DL Tx beam or beam set/list associated with the same associated ID, while FFS whether/how to define similar properties of a DL Tx beam or beam set/list. Q3: Is NW-side additional condition functionality specific? Answer to Q3: Please also refer to the answer to Q2 to understand the ongoing discussion about the associated ID for NW-side additional condition. And please refer to the agreements related to the Questions from RAN 2. Q4: RAN2 wonders what information is needed in Step 3 for UE to decide whether a functionality is applicable before Step 4. More specifically, RAN2 would like to ask the following questions (Q4-1 to Q4-5): Answer to Q4: And please refer to the agreements related to the Questions from RAN 2. Q4-1: In RAN2, it is FFS whether NW-side additional condition is mandatory or optional. In order to discuss further, RAN2 would like to understand whether it is feasible for UE to decide the applicable functionalities without NW-side additional condition? Answer to Q4-1: There is no consensus yet on whether it is mandatory or optional. There is no conclusion yet on whether it is feasible or not for UE to decide the applicability without NW-side additional condition, and RAN 1 is discussing the related issues. Q4-2: In RAN2, it is FFS whether configuration (e.g. inference configuration) other than NW-side additional condition can be included in Step 3. RAN2 would like to understand whether it is feasible and required for gNB to provide configuration (e.g. inference configuration) other than NW-side additional condition in Step 3 for UE to determine applicable functionalities? Answer to Q4-2: Please refer to the agreements related to the Questions from RAN 2. Q4-3: For UE evaluating applicable functionality reporting, if the answer to Q4-2 is Yes, what is the relationship between NW-side additional condition and configuration (e.g. inference configuration)? For example, is NW-side additional condition part of inference configuration, or is inference configuration part of NW-side additional condition, or is NW-side additional condition separate from inference configuration, etc? Answer to Q4-3: Please refer to the agreements related to the Questions from RAN 2. Q4-4: If the answer to Q4-2 is Yes, what is the content of configuration (e.g. inference configuration) for UE to determine applicable functionalities? Answer to Q4-4: Please refer to the agreements related to the Questions from RAN 2. Q5: What is the content of applicable functionality reporting in Step 4? Answer to Q5: Please refer to the agreements related to the Questions from RAN 2. Q6: What is the content of inference configuration in Step 5? Answer to Q6: Please refer to the agreements related to the Questions from RAN 2. The content of inference configuration as CSI-ReportConfig is to be designed later in RAN1. Q7: If inference configuration is provided in Step 3, does it activate the functionality immediately upon receiving Step 3? Answer to Q7: Please refer to the agreements related to the Questions from RAN 2. Q8: If inference configuration is not provided in Step 3, does configuration in Step 5 activate the functionality immediately upon receiving Step 5? Answer to Q8: Please refer to the agreements/conclusion related to the Questions from RAN 2. Q9: If more than one functionality are configured in Step 3 or Step 5, whether multiple/all applicable functionalities can be activated? Answer to Q9: Please refer to the agreements related to the Questions from RAN 2. Q10: Is L1/L2 signaling for functionality activation/deactivation needed? Answer to Q10: Please refer to the agreements related to the Questions from RAN 2. With that, RAN1 understands that L1 and MAC signalling can be used for aperiodic CSI Report and semi-persistent CSI report. Decision: The draft LS is endorsed. Final reply LS is approved in R1-2410898. Final summary in R1-2410892.

        Company alignment · AI-analyzed
      

CMCC adopted CMCC 'Proposes that the granularity of UE capability reporting for AI/ML be at the sub-use case level,' and the agreement states that 'functionality' of Supported functionalities refers to 'UE-capability information/parameters... Rel-19 AI/ML-enabled Features/FGs,' aligning with CMCC's desire for detailed capability reporting.

Apple partially adopted Apple 'Proposes separating UE capabilities for data collection for training versus inference/monitoring,' and the agreement distinguishes between 'Supported functionalities' (UE-capability) and 'Applicable functionalities' (CSI-ReportConfig/parameters), partially adopting the separation of concerns.

Specification support for positioning accuracy enhancement

conclusion

For measurement report of AI/ML assisted positioning Case 3a, regarding the report of LOS/NLOS indicator, LOS/NLOS indicator can’t be reported independently from other measurements

        Company alignment · AI-analyzed
      

CATT ADOPTED CATT pushes against 'unnecessary LOS/NLOS reporting when AI/ML derives timing information,' which aligns with the conclusion that the indicator cannot be reported independently from other measurements.

ETRI CONTRADICTED ETRI emphasizes the 'need for separate handling of timing information and LOS/NLOS indicators,' which is directly contradicted by the agreement that the indicator 'can’t be reported independently'.

agreement

For the definition of sample-based measurement, for gNB/TRP measurement of an estimated channel response between a pair of UE and TRP, the starting time of the list of Nt consecutive samples is determined as follows. starting time = first detected path rounded down with timing granularity T. Note: UE-side measurement is a separate discussion.

        Company alignment · AI-analyzed
      

ETRI CONTRADICTED ETRI supports 'Option B for measurement window timing (earliest detectable sample),' whereas the agreement defines starting time as 'first detected path rounded down,' which is a different definition.

Indian Institute of Tech (M) CONTRADICTED IIT(M) advocates for 'Option B (starting time based on first detectable path power),' which conflicts with the agreed definition of 'first detected path rounded down with timing granularity T'.

agreement

For model performance monitoring of AI/ML positioning Case 1, support at least: Option A. The target UE side performs monitoring metric calculation. The target UE may signal the monitoring outcome to the LMF. FFS: content of monitoring outcome FFS: Option B Presented in Thursday session.

        Company alignment · AI-analyzed
      

Google ADOPTED Google supports 'UE-side monitoring (Option A) over network-side monitoring,' which matches the agreement to 'support at least: Option A. The target UE side performs monitoring metric calculation.'

Ericsson PARTIALLY ADOPTED Ericsson supports 'label-free self-monitoring by the model inference entity,' aligning with UE-side monitoring, but the agreement defers 'Option B' (network-side) via FFS, whereas Ericsson's position implies a broader support for flexible monitoring entities.

Fraunhofer PARTIALLY ADOPTED Fraunhofer advocates for 'network-centric control (LMF-based functionality management),' which is not fully adopted as the agreement prioritizes UE-side (Option A) and defers LMF-side (Option B) to FFS.

CEWiT PARTIALLY ADOPTED CEWiT supports 'distributed model monitoring responsibilities (gNB for Case 3a, LMF for Cases 2b/3b),' but the agreement only mandates UE-side monitoring for Case 1 and defers other options, partially aligning with the distributed concept but not the specific entity mapping.

agreement

For Rel-19 AI/ML based positioning, for Case 3b, in addition to path-based measurement that is referring to the measurement in the existing specifications (up to Rel-18), additionally support the following enhancement to the measurement, The measurement is composed of Nt' values of the estimated channel response in time domain. The Nt’ values are selected from a list of Nt consecutive channel response values, which have timing granularity T. The timing information for the Nt' values are reported with a timing granularity T, where T=2kxTc. k represents the timing reporting granularity factor. Tc is the basic time unit for NR. The associated measurement (e.g., power if reported) corresponds to the measurement for the reported Nt' values. The timing information is defined relative to a reference time, same as the path-based measurement. The Nt’ selected time domain channel measurement values are expected to be those with the highest power. The starting time of the list of Nt consecutive values is determined as: starting time = first detected path rounded down with timing granularity T. LMF can signal parameter values of Nt, Nt', k to gNB via NRPPa. Candidate set values: Nt'<=24. FFS: Nt' values. Nt = {32, 64, 128} FFS: k The gNB/TRP may use different Nt', Nt and/or k values other than the signalled parameter for measurement reporting. In this case, it’s up to LMF implementation to process the reported measurement FFS: whether transmit offset from gNB to LMF Note: measurement by UE is a separate discussion. Note: the purpose of the time domain channel measurements, such as for Rel-19 AI/ML based positioning, is not specified

        Company alignment · AI-analyzed
      

Apple PARTIALLY ADOPTED Apple supports sample-based measurements and increasing Nt to 128, which matches the agreement's Nt set including 128; however, Apple proposes CIR with phase, while the agreement focuses on time domain channel response values without explicitly mandating phase reporting in this text.

Baicells PARTIALLY ADOPTED Baicells strongly advocates for sample-based measurements, which is adopted, but pushes for phase information support which is not explicitly included in the agreed measurement composition of 'Nt' values of the estimated channel response in time domain'.

CATT PARTIALLY ADOPTED CATT advocates for sample-based measurements, which is adopted, but proposes 'unified starting time across multiple TRPs using configured or predefined offsets,' whereas the agreement defines starting time relative to the 'first detected path' and defers transmit offset.

Huawei CONTRADICTED Huawei proposes capping 'Nt at 64' and 'Nt' at 16,' which contradicts the agreement allowing 'Nt = {32, 64, 128}' and 'Nt'<=24' with FFS on values, exceeding Huawei's proposed caps.

Fujitsu PARTIALLY ADOPTED Fujitsu advocates for sample-based measurement, which is adopted, but deprioritizes phase information and opposes explicit indicators, while the agreement leaves phase and specific parameter values (k, Nt') as FFS.

agreement

For AI/ML based positioning Case 1, all assistance information from legacy UE-based DL-TDOA, other than info #7, can be provided from LMF to UE. For info #7, RAN1 study, if necessary, choose one alternative from the following: Alternative 1. Info #7 is provided implicitly via associated ID. Associated ID is signaled by LMF to indicate whether info #7 is consistent between training and inference. Alternative 2. Info #7 can be provided either implicitly or explicitly by LMF. Note: no UE capability is introduced on whether info #7 is provided implicitly or explicitly, and the UE can request info #7 to be provided explicitly or implicitly. If provided implicitly, associated ID is signaled by LMF to indicate whether info #7 is consistent between training and inference. Alternative 3. Info #7 is not be provided from LMF to UE. If info #7 is not provided, UE may assume info #7 is consistent between training and inference. Alternative 4. Info #7 is provided explicitly from LMF to UE. Final summary in R1-2410921.

        Company alignment · AI-analyzed
      

Ericsson PARTIALLY ADOPTED Ericsson proposes using 'associated IDs or explicit assistance data,' which aligns with the agreement to reuse legacy info and study alternatives for Info #7, but the specific alternative is deferred to FFS.

Huawei PARTIALLY ADOPTED Huawei insists on 'reusing legacy DL-TDOA assistance data explicitly,' which matches the agreement to provide legacy info, but opposes 'implicit signaling,' while the agreement defers the method for Info #7 (including implicit options) to further study.

Fujitsu CONTRADICTED Fujitsu pushes against 'associated ID mechanisms,' but the agreement includes 'Associated ID' in Alternative 1 and Alternative 2 for handling Info #7 consistency.

Position changes since RAN1#118bis

  AI-synthesized from contributions · all text is paraphrased

9.1.1 Specification support for beam management 2 evolved · 4 new

      Evolved positions
    

Spreadtrum shifted verify sources →

Spreadtrum shifted from emphasizing UE-initiated control to favoring network-provided configurations while maintaining opposition to complex new metrics.
Huawei strengthened verify sources →

Huawei shifted from a conservative approach emphasizing reuse of existing frameworks to a more aggressive stance pushing for significant enhancements and expanded capabilities.

      New contributors this meeting
    

CMCC verify sources →

Advocates for comprehensive AI/ML beam management support with flexible configuration options for both network-side and UE-side models, pushing for enhanced UE capabilities beyond current 64 RS limits and Top-K beam sweeping procedures.
InterDigital verify sources →

Strongly advocates for Option 2 applicability approach for UE-side models to avoid excessive configuration overhead, supporting single CSI-ResourceConfigId configuration and UE-assisted performance monitoring over network-only approaches.
Quectel verify sources →

Advocates for UE autonomous control in AI/ML beam management, supporting UE-initiated data collection and model management while extending existing CSI frameworks rather than creating entirely new signaling structures.
Samsung verify sources →

Advocates for a comprehensive framework supporting both network-side and UE-side models with flexible configuration options, separate CPU counting for AI/ML vs legacy CSI reports, and enhanced signaling mechanisms.

9.1.2 Specification support for positioning accuracy enhancement 2 evolved · 3 new

      Evolved positions
    

Huawei strengthened verify sources →

Huawei strengthened their position by expanding opposition from general complex signaling to specifically opposing tight specification of sample-based measurements and mandatory phase information.
ZTE strengthened verify sources →

ZTE strengthened their position by providing specific quantitative evidence (1.2-2.2x performance improvement) to support their advocacy for CIR with phase information.

      New contributors this meeting
    

CMCC verify sources →

Advocates for sample-based measurements over path-based measurements and pushes for reusing existing legacy mechanisms where possible to minimize specification impact.
New H3C Technologies Co. verify sources →

Strongly advocates for using CIR as AI/ML model input despite higher overhead, arguing it preserves more channel information, and specifically supports relative phase methods over double differential methods.
Samsung verify sources →

Advocates for a comprehensive AI/ML positioning framework that balances performance with overhead reduction, supporting flexible data collection approaches and UE autonomy in data provider decisions.

9.1.3 Specification support for CSI prediction 1 evolved · 3 new

      Evolved positions
    

ZTE strengthened verify sources →

ZTE evolved from a systematic approach with conditional support to a stronger opposition stance, now emphasizing model generalization capabilities as sufficient without additional consistency mechanisms.

      New contributors this meeting
    

CMCC verify sources →

Advocates for maximum reuse of existing AI/ML beam management mechanisms for CSI prediction and supports hybrid approaches combining associated IDs with performance monitoring, favoring intermediate KPIs over eventual KPIs.
Ericsson verify sources →

Argues against extensive specification enhancements for training/inference consistency, claiming UE-sided CSI prediction models demonstrate sufficient generalization capability across various network conditions.
Samsung verify sources →

Strongly advocates for network-assisted AI/ML model training consistency with TRP-related signaling, emphasizing critical need to address severe performance degradation from antenna configuration mismatches.

9.1.4.1 CSI compression 1 evolved · 3 new

      Evolved positions
    

Huawei strengthened verify sources →

Huawei maintained their support for Direction A but strengthened their position by specifically favoring Option 4-1 and adding strong opposition to Direction C based on practical implementation concerns.

      New contributors this meeting
    

CMCC verify sources →

Advocates for prioritizing Direction A and B over Direction C for inter-vendor collaboration, supporting temporal domain CSI compression with demonstrated performance gains and promoting existing R16 eType-II and Rel-18 Doppler codebooks as starting points.
Ericsson verify sources →

Strongly advocates for Direction C (fully standardized reference models) as the most feasible inter-vendor solution, supporting 3GPP channel model-based synthetic data while opposing over-the-air delivery for Direction A due to complexity concerns.
Samsung verify sources →

Strongly advocates for angle-delay (W2) domain compression over spatial-frequency (W) domain compression due to superior generalizability, and pushes for temporal aspects in CSI compression (Cases 2 and 3) with specific focus on basis vector refresh.

9.1.4.2 Other aspects of AI/ML model and data 2 evolved · 2 new

      Evolved positions
    

Huawei strengthened verify sources →

Huawei expanded their opposition to cross-vendor collaboration from just Case z1 to Cases z1, z2, z3, and z5, taking a stronger stance against complexity.
ZTE strengthened verify sources →

ZTE provided more specific technical justification for opposing MI-Option 2 with concrete resource overhead figures (1-10GB transfers) and added explicit support for Case z4.

      New contributors this meeting
    

CMCC verify sources →

Advocates for prioritizing MI-Option 2 and MI-Option 3 for model identification while opposing MI-Option 4 due to limited performance gains and high specification efforts. Supports studying model transfer Case z4 with standardized reference model structures.
Samsung verify sources →

Advocates for simplified model identification approaches that avoid complex cross-vendor collaboration, supporting network-side additional condition indicators, standardized reference models (MI-Option 4), and restricting data collection to 3GPP network entities.

Sub-topics

  5 agenda items · grouped by 3GPP agenda numbering

9.1.1 Specification support for beam management 52 contributions

Contributions address specification support for AI/ML-based beam management in NR, focusing on data collection for training and inference, configuration of Set A and Set B resources, and performance monitoring mechanisms. Companies debate the separation of monitoring and inference configurations, the definition of Beam Accuracy Indicators (BAI), and the reuse of existing CSI framework mechanisms versus introducing new signaling. Key technical discussions include overhead reduction strategies, CPU handling for UE-side inference, and consistency mechanisms via DL Tx IDs and Associated IDs.

      Company positions
    

Samsung ×9 — Proposes specific data collection contents for NW-side training, including L1-RSRPs for Set A and Set B with timestamps conveyed via high-layer signaling. Supports configurability between Alt 1 and Alt 3 for CSI-ReportConfig and introduces DL Tx IDs to ensure consistent spatial domain transmission filters between Set A and Set B. Proposes introducing a Beam Accuracy Indicator (BAI) for Type 1 Option 2 performance monitoring, calculated over X CSI reports, and prefers dedicated CSI report configurations for monitoring resources. Regarding CPU handling, proposes separate CPU counting for AI/ML-based CSI reports and scaling the legacy Z timeline based on UE capability. Advocates for a flexible framework combining multiple options for UE-side model applicability, opposing overly restrictive single-option approaches.
Apple — Proposes separating UE capabilities for data collection for training versus inference/monitoring to reflect that UEs capable of inference may not support training data collection. Requires the associated ID in assisted information to be PLMN unique and managed by the core network or O&M to ensure consistency between training and inference, embedding this ID in reference signal configuration. Proposes leveraging the MDT framework for NW-side model training data collection and supporting L1 beam reporting for performance monitoring. For overhead control, proposes two-part beam reporting using bitmaps to omit weak beams' RSRPs and differential RSRPs for un-omitted beams, with the strongest beam index indicated across measurement occasions for BM Case-2. Argues that the effective time for beam reporting should reference the CSI measurement source rather than the beam report time to handle fixed time gaps in AI/ML models.
CAICT — Advocates for comprehensive UE-sided model monitoring with flexible network configuration, supporting both periodic and event-triggered reporting. Proposes per-cell based Associated ID design and separate CSI-ResourceConfigId configuration (Alt 3) for Set A/B. Pushes for finer quantization steps in L1-RSRP reporting, specifically less than 3dB accuracy. Supports flexible UE implementation of monitoring algorithms, positioning against overly rigid standardization of UE-side monitoring procedures.
CATT — Advocates for flexible AI/ML beam management solutions that reuse existing CSI framework mechanisms while introducing minimal specification impact. Supports both UE-sided and network-sided models with comprehensive performance monitoring. Opposes overly complex new signaling mechanisms, favoring practical approaches like reusing legacy TCI delay requirements and CPU mechanisms rather than creating entirely new frameworks.
China Telecom — Advocates for comprehensive support of multiple data collection options for flexibility across different Life Cycle Management purposes, including training, monitoring, and inference. Proposes beam prediction accuracy KPIs as primary performance monitoring metrics and full Set A measurement for accurate UE-assisted performance monitoring. Opposes using only single resource set configurations for Set B in UE-sided models. Deprioritizes link quality metrics for AI/ML model performance monitoring, arguing these do not directly reflect prediction accuracy.
CMCC — Proposes that L1 signaling be supported for NW-sided training data collection and that Top-K beam sweeping be supported for NW-side inference to enhance prediction accuracy. Requires that for UE-side models, the overall CPU be separately counted between legacy CSI reporting and AI/ML-based CSI reporting, while being shared among AI/ML features. Prefers Option 1 for BM-Case 2 differential RSRP reporting, which includes CRI of top-K predicted beams per instance. Supports dedicated resource sets and report configurations for UE-side model monitoring (Type 1 Option 2), linking the monitoring configuration to an inference report configuration via CSI-ReportConfig ID. Proposes that the granularity of UE capability reporting for AI/ML be at the sub-use case level, including details on Set A/B size, RS type, and model outputs.
Ericsson — Advocates for a comprehensive AI/ML framework that leverages existing CSI mechanisms with minimal specification impact while maximizing functionality. Strongly supports expanding aperiodic CSI-RS resources from 16 to 64 beams to enable practical AI/ML deployment. Proposes uncertainty quantification in UE predictions to enable trustworthy AI/ML and adaptive Top-K beam selection based on model confidence. Supports extensive overhead reduction mechanisms for NW-sided models. Opposes complex new signaling frameworks, favoring reuse of existing CSI infrastructure with targeted enhancements.
ETRI — Advocates for a comprehensive framework supporting both UE-sided and NW-sided AI/ML models with emphasis on overhead reduction and consistency mechanisms. Strongly supports Options 1 and 2 over Option 3 for applicability procedures. Proposes L1-RSRP difference as a key performance metric and flexible Associated ID configurations across multiple locations. Supports differential RSRP reporting methods. Opposes overly complex signaling by preferring direct activation methods and supporting omission of RSRP values when differences are minimal to reduce overhead.
Fraunhofer — Strongly advocates for UE-sided AI/ML models over network-sided models, emphasizing their advantages in reduced signaling overhead, shorter measurement-to-inference delay, and ability to use more measurements. Supports bitmap-based indexing as the most efficient beam reporting method. Advocates for 2-phase monitoring with adaptive frequency. Opposes beam indication for future time instances, arguing it offers little advantage while incurring high specification workload and considerable UE behavior modifications.
Fujitsu — Advocates for Option-3 in applicable functionality reporting, favoring UE-driven parameter reporting rather than network-driven configuration. Supports aperiodic CSI-RS for temporal beam prediction and prefers separate configurations for monitoring versus inference. Pushes for high-resolution quantization in training data. Opposes Option-1 and Option-2 for functionality reporting due to feasibility concerns. Advocates for L1-RSRP difference as the preferred performance metric for network-side models over beam prediction accuracy.
FUTUREWEI — Advocates for maximizing reuse of existing CSI framework to minimize specification effort. Supports simpler approaches that avoid complex new metrics like probability and confidence information. Opposes supporting multiple alternatives that would increase complexity, specifically opposing Opt 3/4 for inference reporting and other performance monitoring alternatives beyond Alt 1. Favors practical solutions like using RS ID as implicit beam ID rather than defining new beam identifiers.
Google — Advocates for comprehensive AI/ML beam management support including both network-side and UE-side models. Supports flexible reference signal configurations, including AP-CSI-RS and SSB, and enhanced reporting mechanisms with confidence indicators. Proposes separate TCI state pools for AI/ML predictions. Supports practical implementation considerations like CPU resource management and performance monitoring relaxation. Strongly supports Option 1 for inference configuration management to avoid multi-stage RRC configuration latency issues.

      Open issues
    

Selection of specific options for UE-side model applicability reporting (Option 1 vs 2 vs 3) and configuration methods (Alt 1 vs Alt 3).
Definition and calculation methodology for Beam Accuracy Indicator (BAI) and other performance monitoring metrics.
Separation of CPU counting and timeline handling for AI/ML-based CSI reports versus legacy CSI reports.
Overhead reduction mechanisms for beam reporting, including differential RSRP and bitmap usage.
Consistency mechanisms between training and inference, specifically the role and scope of Associated IDs and DL Tx IDs.

9.1.2 Specification support for positioning accuracy enhancement 39 contributions

Contributions focus on defining model inputs for AI/ML-based positioning, specifically debating sample-based versus path-based measurements and the inclusion of phase information (CIR). Companies discuss training data collection procedures, consistency mechanisms between training and inference (e.g., associated IDs vs. assistance data), and model monitoring frameworks, including label-free methods and entity responsibilities (UE vs. LMF/gNB).

      Company positions
    

Ericsson ×7 — Presents a technical case against path-based measurements and CIR/phase inputs due to signaling overhead and alignment difficulties, initially favoring sample-based measurements with total-power PDP inputs. Later documents advocate supporting both sample-based and path-based measurements as a compromise to avoid blocking progress, while continuing to push against CIR support. Proposes using associated IDs or explicit assistance data from the LMF to ensure consistency between training and inference, and supports label-free self-monitoring by the model inference entity.
Apple — Proposes supporting both path-based and sample-based measurement inputs, arguing that sample-based is a special case of path-based with equi-spaced timing. Requires increasing the number of additional paths (Nt) to up to 128 to ensure performance in low-complexity scenarios and supports using Channel Impulse Response (CIR) including phase information, proposing training data compensation for phase mismatch. Defines specific quality indicators for channel measurements and ground truth labels, and proposes that the default monitoring entity is the one hosting the model.
Baicells — Strongly advocates for sample-based measurements over path-based measurements, citing superior positioning accuracy and avoidance of algorithm inconsistencies between vendors. Pushes for phase information support in Case 3b despite overhead concerns. Advocates for minimizing specification impact by reusing existing IEs and procedures wherever possible rather than defining new mechanisms.
CATT — Strongly advocates for sample-based channel measurements over path-based measurements due to superior AI/ML performance and reduced ambiguity. Proposes unified starting time across multiple TRPs using configured or predefined offsets and treats AI/ML positioning as an independent method requiring specialized assistance information. Pushes against separate quality indicators for different measurement components and unnecessary LOS/NLOS reporting when AI/ML derives timing information.
CEWiT — Advocates for sample-based over path-based channel measurement reporting due to better positioning accuracy, despite higher overhead. Supports distributed model monitoring responsibilities (gNB for Case 3a, LMF for Cases 2b/3b) and pushes for semi-supervised learning to leverage both PRU and non-PRU UEs for training data collection. Implicitly argues against centralized model management approaches, promoting UE autonomy in parameter selection while maintaining network assistance.
CMCC — Slightly prefers sample-based measurements to avoid potential errors from intermediate path-based processing at the UE, proposing that the entity deriving the AI model should provide recommended sample numbers. Argues that offline training with marginal specification impact is feasible but questions the feasibility of obtaining ground-truth labels via PRUs due to large dataset sizes. Proposes reinterpreting existing 'Timing Measurement Quality' IEs for gNB-side models and reusing legacy LOS/NLOS reporting formats, while deprioritizing certain consistency options in favor of further study.
ETRI — Strongly advocates for vendor flexibility in AI/ML positioning implementation, pushing for UE vendors and gNB manufacturers to determine their own model input formats rather than enforcing standardized formats. Supports Option B for measurement window timing (earliest detectable sample) and emphasizes the need for separate handling of timing information and LOS/NLOS indicators in AI/ML-assisted positioning. Focuses on practical implementation considerations in NLoS environments where traditional positioning methods fail.
Fraunhofer — Advocates for comprehensive AI/ML positioning frameworks that maintain network-centric control (LMF-based functionality management) while supporting flexible measurement approaches. Pushes for complex-valued CIR reporting to preserve information richness, event-based training data collection optimization, and two-stage monitoring processes. Is positioned against UE-autonomous functionality management without network oversight, advocating for balanced approaches that consider both performance and overhead costs.
Fujitsu — Advocates for sample-based measurement with configured offset (Option D) and deprioritizes phase information study due to negligible gains versus overhead. Proposes using LOS/NLOS indicator as a confidence level rather than LOS likelihood and supports multiple timing information reporting. Pushes against associated ID mechanisms for multi-TRP scenarios due to impractical training burden and opposes explicit AI/ML indicators in positioning reports.
Google — Advocates for transparent model performance monitoring that does not require UEs to disclose location information to the network, supporting UE-side monitoring (Option A) over network-side monitoring. Pushes for flexible channel measurement options including phase information and configurable sample-based versus path-based reporting. Supports reuse of existing measurement frameworks rather than creating new mechanisms.
Huawei — Argues that ambiguity in sample-based measurements can be avoided by implementation rather than strict specification, proposing flexible determination of selection window parameters by gNBs while capping Nt' at 16 and Nt at 64 to protect proprietary channel estimation. Supports enhancing legacy path-based reporting by increasing the number of reported paths to 16 and proposes reusing legacy 'LoS/NLoS Information' IEs for Case 3a. Opposes introducing new positioning methods or implicit signaling for Case 1, insisting on reusing legacy DL-TDOA assistance data explicitly.
Indian Institute of Tech (M) — Advocates for Option B (starting time based on first detectable path power) for sample-based measurements, claiming superior positioning accuracy compared to path-based measurements. Pushes against increasing the number of reported paths beyond 16, arguing it adds unnecessary overhead without significant measurement quality enhancement.

      Open issues
    

Whether to support sample-based, path-based, or both measurement types for model inputs.
Inclusion of phase information (CIR) in model inputs versus deprioritizing it due to overhead and alignment complexity.
Mechanisms for ensuring consistency between training and inference (e.g., associated IDs vs. explicit assistance data).
Entity responsible for model monitoring (UE-side vs. LMF/gNB-side) and monitoring methods (label-free vs. label-based).
Handling of LOS/NLOS indicators in AI/ML-assisted positioning outputs.

9.1.3 Specification support for CSI prediction 34 contributions

Contributions focus on ensuring consistency between training and inference for UE-sided AI/ML CSI prediction models, specifically debating the necessity of network-side additional conditions and associated IDs. Companies are divided on whether factors like TXRU mapping and antenna tilt angles require explicit signaling or if generalized models with UE-side performance monitoring suffice. The discussion also covers specification impacts on CSI-RS configuration for data collection and the reuse of mechanisms from AI/ML beam management.

      Company positions
    

LG Electronics ×4 — Concludes that gNB antenna tilt angle variations have negligible impact on CSI prediction performance, arguing no additional specification support is needed for tilt angle consistency. Pushes for prioritizing Type 1 and Type 3 performance monitoring over Type 2 due to lower reporting overhead and higher accuracy, explicitly opposing Type 2 monitoring due to payload and quantization issues. As document moderator, facilitates consensus that tilt angle impacts are negligible, while acknowledging a lack of industry consensus on TXRU mapping effects, suggesting continued study rather than immediate standardization for the latter.
Huawei ×2 — Presents a technical case against introducing network-side indications, such as associated IDs, for CSI prediction consistency, citing feasibility issues and proprietary disclosure risks regarding network planning information. Argues that generalized AI/ML models achieve satisfactory performance using mixed datasets (Generalization Case 3), rendering explicit consistency indications unnecessary. Proposes ensuring consistency via UE-side performance monitoring in an implementation manner, allowing the UE to select the best-matching model without network signaling overhead.
vivo ×2 — Identifies TXRU virtualization mapping as a critical network-side additional condition, presenting simulation evidence that mismatches cause substantial SGCS loss (up to -44.4%) in high-speed and outdoor scenarios. Advocates for adopting the associated ID solution established in the Beam Management use case to link data collected under specific network-side conditions to unique identifiers. Pushes against ignoring the significant performance degradation caused by TXRU mapping mismatches between training and inference phases.
Apple — Argues that the requirement for network-side additional conditions depends on whether the UE model relies on spatial signatures, distinguishing between time-domain models and 3D CNNs. Proposes that if generalization case 2 evaluation shows performance degradation, consistency signaling is needed, specifically defining assisted information or an identifier to abstract antenna virtualization and multi-TRP details for spatial signature-dependent models. Proposes enabling the UE to signal whether this assisted information is required during the data collection request procedure.
AT&T — Strongly advocates for site/cell specific AI/ML models over generalized models for CSI prediction, arguing they provide superior performance. Pushes for network-side additional conditions and associated ID mechanisms to enable localized model training and ensure training/inference consistency. Argues against relying solely on generalized models that show significant performance degradation across diverse scenarios.
CATT — Advocates for relaxing consistency requirements for network-side antenna configurations like tilt angles and TXRU mapping, demonstrating through simulations that they have negligible impact. Pushes for performance monitoring-based methods to handle other consistency factors, such as interference distributions, which cannot be addressed through associated IDs. Positions against overly restrictive consistency requirements that would unnecessarily complicate CSI prediction implementations.
China Telecom — Advocates for reusing the associated ID mechanism from AI-based beam management for CSI prediction to ensure training/inference consistency. Recognizes that UE assumptions about the same associated ID may need different interpretations for CSI prediction compared to beam management. Supports starting with associated ID within a cell before extending to multiple cells due to practical deployment challenges in aligning network-side conditions across cells.
CMCC — Proposes studying two options for ensuring consistency: one based on associated ID and another based on performance monitoring, suggesting a combined solution where the ID preliminarily guarantees consistency without exposing proprietary information. Requires that if associated ID is supported, the UE assumes consistency of network-side additional conditions with the same ID at least within a cell, configured within the CSI framework. Proposes using intermediate KPIs for Type 3 monitoring and reusing data collection mechanisms from AI/ML beam management while adapting Rel-18 CSI parameters.
Ericsson — Argues the inapplicability of the inconsistency issue identified for UE-sided spatial beam prediction to CSI prediction, concluding no specification enhancements are needed for UE speed, deployment scenario, carrier frequency, antenna tilt, or TXRU mapping due to negligible performance degradation. Proposes studying specification impacts on CSI-RS configuration to indicate association between observation and prediction window resources for data collection. Opposes Type 2 based performance monitoring, citing large reporting overhead and significant network complexity.
ETRI — Advocates for extending the already-agreed associated ID support from AI/ML beam management to CSI prediction to reduce specification workload and ensure training/inference consistency. Pushes for leveraging existing CSI framework configurations, such as CSI report or resource configuration, to implement the associated ID. Argues this approach minimizes new specification requirements while addressing model performance degradation when network conditions change.
Fujitsu — Advocates for using associated ID as the primary mechanism to ensure training/inference consistency, similar to the approach agreed for beam management. Pushes against considering tilt angle as a key network-side condition, based on simulation results showing negligible performance impact. Supports focusing on gNB antenna configurations and scenario types (indoor/outdoor, LOS/NLOS) as the critical factors that should be associated with the same ID.
Google — Advocates for UE-side only CSI prediction models with UE-reported preferred CSI-RS configurations to avoid configuration mismatches. Proposes associated ID management per CSI report configuration to maintain training/inference consistency. Pushes for dynamic activation/deactivation capabilities to enable network energy saving while maintaining model performance.

      Open issues
    

Whether TXRU virtualization mapping requires explicit network-side indication (associated ID) or if generalized models suffice.
The necessity and scope of network-side additional conditions for ensuring training/inference consistency.
The appropriate type of performance monitoring (Type 1, 2, or 3) and associated KPIs for CSI prediction.
Specification impacts on CSI-RS configuration for data collection and the association between observation and prediction windows.

9.1.4.1 CSI compression 41 contributions

Contributions focus on standardizing AI/ML-based CSI compression for NR Release 19, specifically addressing inter-vendor training collaboration, temporal domain processing, and performance monitoring. Companies debate the feasibility of dataset sharing (Direction A) versus parameter sharing (Direction B), with significant divergence on specification complexity and proprietary risks. Key technical discussions include the use of Transformer backbones, handling UCI loss through reset mechanisms, and defining ground-truth data collection methods using enhanced codebooks.

      Company positions
    

Qualcomm ×10 — Strongly advocates for Direction A Option 4-1 (dataset sharing) over Direction B (parameter sharing) due to lower specification impact and feasibility concerns with common encoders. Supports prioritizing temporal domain Cases 2 and 3 with separate prediction and compression as a baseline, and emphasizes Transformer-based model structures for Case 0. Proposes NW-side monitoring as the primary approach using ground-truth CSI reporting, while opposing over-the-air signaling for large dataset exchange and joint prediction/compression approaches due to complexity.
Apple — Deprioritizes inter-vendor training collaboration Option 3a-1, arguing it cannot handle UE-side additional conditions as effectively as Option 4-1 with assisted nominal decoder training. Proposes enabling semi-persistent CSI reporting and DCI-based state reset for temporal cases to mitigate SGCS loss from UCI drop-induced state desynchronization. Argues for using precoded CSI-RS to implicitly transmit output CSI for low-overhead performance monitoring and proposes studying RLF/BFD-like mechanisms for UE-initiated reports.
CATT — Strongly advocates for Direction B (Option 3b) over Direction A, supporting direct NW encoder parameter sharing to the UE without offline engineering based on evaluation showing better performance and less complexity. Pushes for standardized OTA signaling with RRC as a starting point, spatial-frequency domain model input, and Transformer backbone for Case 0. Is against localized models due to implementation complexity and opposes UE-side performance monitoring based on reference or proxy models.
CEWiT — Advocates for network-side monitoring as the primary approach for AI/ML CSI compression, supporting basis components transmission methods over high-resolution codebook methods for reduced payload. Pushes for Option-1 CQI determination (not based on CSI reconstruction output) as the starting point while arguing against Option-1c. Strongly supports reusing existing legacy signaling methods and procedures rather than creating entirely new mechanisms.
CMCC — Prioritizes inter-vendor collaboration Directions A and B for further study, arguing they offer better performance and lower specification effort than Direction C. Proposes that for Direction A, the NW-side must share dataset information and performance targets to enable UE-side offline engineering, while acknowledging that over-the-air parameter exchange introduces extra overhead. Prioritizes NW-side monitoring based on ground-truth CSI reports and proposes using Rel-16 eType-II and Rel-18 Doppler codebooks as starting points for ground-truth CSI data collection.
Ericsson — Requires that reference models for inter-vendor collaboration be designed using 3GPP channel model based synthetic data rather than field data, citing excessive work and bias risks. Opposes UE-side first training for Options 3/4/5, arguing it necessitates multiple parallel models at the NW-side, and rejects over-the-air delivery for Direction A due to signaling overhead. Concludes that UE reporting of high-resolution target CSI is necessary to enable NW-side intermediate KPI monitoring, specifically proposing enhancements to the eType-II format.
ETRI — Advocates for supporting both Direction A (UE-side offline engineering) and Direction B (direct parameter sharing) inter-vendor training collaboration approaches, emphasizing the importance of maximizing reuse of existing CSI-RS specifications for temporal domain aspects. Pushes against Direction C due to severe performance degradation and advocates for UE-side performance monitoring over network-side alternatives due to complexity and overhead considerations.
Fujitsu — Strongly advocates for Direction A Option 4-1 over 3a-1 due to similar performance with less specification effort, and supports synthetic data training for reference models in Direction C. Prefers NW-side monitoring over UE-side proxy model monitoring due to reliability concerns and generalization issues. Pushes against Option 2 for CQI determination due to complexity and delay concerns, favoring codebook-based quantization approaches.
FUTUREWEI — Advocates for using enhanced Rel-16 eType II codebook with new parameters for both data collection and monitoring to achieve better performance, despite overhead concerns. Pushes for differentiated delivery methods based on latency requirements, supporting over-the-air for Direction B and upper layer signaling for Direction A. Is against rushing into proxy model adoption without thorough LCM complexity analysis and advocates for careful performance evaluation before supporting precoded RS-based monitoring methods.
Google — Advocates for a comprehensive AI/ML-based CSI compression framework prioritizing transformer-based standardized models over other architectures. Pushes for flexible hybrid AI/ML and non-AI/ML approaches based on rank indicators and separate processing units for ML inference versus channel estimation. Supports dataset sharing (target CSI + CSI feedback) for inter-vendor collaboration and opposes requiring common encoders across UEs or using SCS as a performance monitoring metric.
Huawei — Argues that for Direction A inter-vendor collaboration, sharing datasets (Option 4-1) incurs less proprietary risk than sharing model parameters, proposing that model backbone information is unnecessary for UE autonomy. Presents technical cases showing that UCI missing causes significant performance loss due to accumulated CSI misalignment, proposing reset mechanisms to mitigate this. Opposes specifying UE-side proxy models for monitoring, citing imbalanced generalization and excessive LCM burden, instead supporting NW-side ground-truth CSI monitoring.
IIT Kanpur — Advocates for enhanced temporal correlation representation in AI/ML models for CSI compression, emphasizing that joint prediction tasks (Case 3) require more sophisticated temporal modeling compared to reconstruction tasks (Case 2). Pushes for improvements in temporal diversity in datasets and better data pre-processing for training data to handle non-ideal UCI conditions effectively.

      Open issues
    

Feasibility and specification complexity of Direction A (dataset sharing) versus Direction B (parameter sharing) for inter-vendor training collaboration.
Handling of UCI loss and state desynchronization in temporal domain CSI compression (Cases 2 and 3).
Definition of performance monitoring mechanisms, specifically the trade-off between NW-side ground-truth reporting and UE-side proxy model monitoring.
Standardization of model structures, particularly the adoption of Transformer backbones and handling of proprietary model accommodation.

9.1.4.2 Other aspects of AI/ML model and data 35 contributions

Contributions focus on defining the scope and mechanisms for AI/ML model identification, lifecycle management (LCM), and data transfer for NR air interface, particularly for two-sided models. Companies debate the complexity of standardizing model structures versus relying on offline vendor coordination, with significant divergence on the feasibility of over-the-air dataset delivery and the necessity of specific model transfer cases like Case z4.

      Company positions
    

OPPO ×5 — Proposes a unified LCM framework combining functionality-based and model ID-based operations, preferring network-assigned model IDs with 1-to-1 mapping to associated IDs. Supports studying MI-Option 2 with dataset transfer and standardized model structures for Case z4, arguing for 3GPP specification of model structures rather than offline vendor collaboration. Opposes overly complex model identification schemes and partial parameter transfer due to implementation burden and feasibility concerns.
Apple — Proposes minimizing standardization scope by deprioritizing complex model transfer cases requiring cross-vendor collaboration. Supports flexible multi-cell operation and limits 3GPP specifications to essential signaling aspects, opposing detailed AI model implementation standardization to reduce UE implementation burdens.
AT&T — Proposes a unified LCM framework with functionality as the default and network-controlled model ID assignment. Supports standardized model transfer approaches focusing on known model structures while opposing fragmented LCM approaches and solutions that risk proprietary disclosure or excessive cross-vendor collaboration burdens.
CATT — Proposes simplified model identification schemes with flexible dataset-to-model mappings. Opposes overly complex standardized model transfer mechanisms, specifically arguing that Case z4 is not necessary for Rel-19 due to feasibility challenges. Advocates focusing study efforts on two-sided models only while deprioritizing UE-first training approaches.
China Telecom — Proposes prioritizing Case z4 model transfer using Alternative B to provide network flexibility in model structure management. Deprioritizes Case z1 due to cross-vendor collaboration complexity and supports standardizing UE readiness reporting mechanisms and clarifying model ID relationships in two-sided AI/ML implementations.
CMCC — Proposes prioritizing MI-Option 2 and MI-Option 3 for model identification while deprioritizing MI-Option 4 due to specification effort and RAN4 dependencies. For Case z4, proposes standardizing reference model structures and exchanging parameters, including specific procedures for UE capability reporting and parameter updates. Argues that meta information, including CSI generation/reconstruction I/O and model backbone, is necessary for dataset transfer validity.
Continental Automotive — Proposes configurable mapping relations as the primary mechanism for model identification and parameter transfer, strongly supporting the MI-Option 2 framework with ID-X transmission from network to UE. Supports leveraging model structure similarity to enable partial parameter transfer and pre-configured mapping relations to optimize efficiency across different model structures.
Ericsson — Presents technical case against over-the-air dataset delivery from network to UE, citing high complexity, signaling overhead, and questionable feasibility. Proposes postponing model identification discussions for MI-Options 2, 3, and 4 until the two-sided CSI compression use case resolves data distribution mismatch and inter-vendor interoperability issues. Prioritizes 'case y' for model transfer and opposes Case z4 unless NW-sided training collaboration is deemed infeasible, arguing that vendor-specific conformance testing would break multi-vendor interoperability.
ETRI — Proposes establishing many-to-many relationships between identifiers (Associated ID, ID-X) and Model IDs across MI-Option 1 and MI-Option 2 to enable flexible model training scenarios. Supports including inference time information in known model structures, arguing that computational complexity metrics alone are insufficient since inference time varies non-linearly with model structure and hardware configurations.
Fujitsu — Proposes comprehensive dataset-based model identification (MI-Option 2) with proper ID management and test dataset procedures. Opposes impractical model transfer cases, specifically arguing to deprioritize Case z1 due to unclear benefits over Case y and Case z4 for UE-sided models due to real-time update impracticality. Opposes partial parameter transfer studies due to complexity and lack of justification.
FUTUREWEI — Proposes network-controlled model identification with associated IDs limited to cell scope, arguing against using associated IDs as model IDs. Deprioritizes complex model transfer cases (z1, z2) that do not provide clear benefits over simpler OTT approaches. Uniquely proposes exploring two-sided models without explicit model identification to reduce complexity, while supporting Case z4 for non-3GPP-transparent scenarios.
Google — Proposes simplifying AI/ML model identification by assuming model IDs are known after UE connection (MI type A) and prioritizing MI-Option 1 to maintain understanding of model properties. Opposes the complexity of MI-Option 2 and MI-Option 3, arguing for deprioritization of dataset transfer and model transfer methods. Supports consolidating ID-X with model ID to avoid redundant identification schemes.

      Open issues
    

Feasibility and necessity of standardizing model structures (Case z4) versus relying on offline vendor coordination or simpler transfer cases.
Complexity and signaling overhead of over-the-air dataset delivery for two-sided model training.
Appropriate model identification options (MI-Option 1 vs 2 vs 3 vs 4) and the relationship between Model IDs, Associated IDs, and ID-X.
Scope of standardization for partial parameter transfer and model lifecycle management procedures.

Contributions

R1-2409786 Apple discussion not treated 5

Discussion on RAN2 LS on applicable functionality reporting for beam management UE-sided model

Apple analyzes three options for AI/ML beam management UE-side model configuration agreed at RAN1 #118bis, focusing on the signaling flow between network and UE. The document raises four key proposals regarding the investigation of these options, the dependencies of dynamic UE capability reporting, the definition of…

Apple proposes that RAN1 thoroughly investigate the three options for AI/ML beam management UE-side model configuration agreed at RAN1 #118bis, arguing against rushing into a design due to the far-reaching impact on UE-side model operation. They question whether dynamic UE…

R1-2409787 Apple discussion not treated 9.1.1

Discussion on AI/ML beam management

Apple presents a comprehensive framework for AI/ML-based beam management in Rel-19, focusing on the lifecycle of models, overhead control for beam reporting, and consistency between training and inference. The document contains 13 proposals and 4 observations addressing UE capabilities, data collection mechanisms, UCI…

Apple proposes separating UE capabilities for data collection for training versus inference/monitoring to reflect that UEs capable of inference may not support training data collection. They require the associated ID in assisted information to be PLMN unique and managed by the…

R1-2409788 Apple discussion not treated 9.1.2

Discussion on Specification Support for AI/ML-based positioning

Apple Inc. presents 45 proposals for Rel-19 AI/ML-based positioning, focusing on specification impacts for sample-based versus path-based measurements, model input types (CIR, PDP, DP), and data collection procedures. The document argues for supporting both measurement types, increasing path support to 128, and…

Apple proposes supporting both path-based and sample-based measurement inputs, arguing that sample-based input is a special case of path-based input with equi-spaced timing. They require increasing the number of additional paths (Nt) to 16, 32, 64, or 128 to ensure performance…

R1-2409789 Apple discussion not treated 9.1.3

Discussion on AI/ML-based CSI prediction

Apple discusses the necessity of network-side additional conditions for AI-based CSI prediction, arguing that the need for consistency signaling depends on whether the UE model relies on spatial signatures. The document presents three proposals regarding the evaluation of generalization cases, the definition of…

Apple argues that the requirement for NW-side additional conditions depends on the specific UE-side model design, distinguishing between models that do not depend on spatial signatures (e.g., time domain LSTM, time/frequency domain 2D CNN) and those that do (e.g., 3D CNN). They…

R1-2409790 Apple discussion not treated 9.1.4.1

Discussion on AI based CSI compression

Apple analyzes inter-vendor training collaboration options for AI/ML-based CSI compression, arguing that Option 3a-1 should be deprioritized in favor of Option 4-1 with assisted nominal decoder training to handle UE-side additional conditions. The document presents 11 proposals and 12 observations covering…

Apple deprioritizes inter-vendor training collaboration option 3a-1, arguing it cannot handle UE side additional conditions as effectively as option 4-1 with Alt 1, which utilizes a nominal decoder. They propose that option 4-1 requires additional assisted information, including…

R1-2409791 Apple discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML model and data

Apple discusses AI/ML model management aspects for 5G NR air interface, proposing 4 specific recommendations covering model transfer deprioritization, enhanced transfer procedures, multi-cell association IDs, and data collection standardization scope.

Apple advocates for minimizing standardization scope and reducing implementation burdens by deprioritizing complex model transfer cases that require cross-vendor collaboration, while supporting flexible multi-cell operation and limiting 3GPP specifications to essential signaling…

R1-2410335 AT&T discussion not treated 9.1.4.2

Other Aspects of AI/ML framework

AT&T's contribution discusses the general framework for AI/ML in NR air interface, focusing on unified lifecycle management (LCM), model identification procedures, and model transfer mechanisms. The document contains 13 detailed proposals covering framework unification, model identification options, and standardized…

AT&T advocates FOR a unified LCM framework that combines functionality-based and model-ID-based operations with functionality as default, network-controlled model ID assignment, and standardized model transfer approaches focusing on known model structures. They push AGAINST…

R1-2410336 AT&T discussion not treated 9.1.3

Discussion on AI/ML for CSI prediction

AT&T discusses AI/ML for CSI prediction in NR Air Interface, focusing on consistency between training and inference for UE-sided models. The document contains 6 observations and 3 main proposals advocating for network-side additional conditions and associated ID mechanisms to improve localized model performance.

AT&T strongly advocates FOR site/cell specific AI/ML models over generalized models for CSI prediction, arguing they provide superior performance. They push FOR network-side additional conditions and associated ID mechanisms to enable localized model training and ensure…

R1-2410414 Baicells discussion not treated 9.1.2

Discussion on specification support for AI-ML based positioning accuracy enhancement

Baicells presents their views on AI/ML-based positioning accuracy enhancement for R19, covering model input/output, training data collection, quality indicators, and model monitoring across positioning sub-use cases. The document contains 11 proposals and 9 observations addressing technical aspects from sample-based…

Baicells strongly advocates FOR sample-based measurements over path-based measurements, arguing that sample-based provides superior positioning accuracy (1.06-1.62x better performance) and avoids algorithm inconsistencies between vendors. They push FOR phase information support…

R1-2410367 CAICT discussion not treated 9.1.1

Discussions on AI/ML for beam management

This CAICT document proposes 10 technical proposals for AI/ML beam management in 5G NR, covering UE-sided model performance monitoring, reference signal configuration for Set A/B, and measurement reporting for network-sided models.

CAICT advocates FOR comprehensive UE-sided model monitoring with flexible network configuration (supporting both periodic and event-triggered reporting), per-cell based Associated ID design, and separate CSI-ResourceConfigId configuration (Alt 3) for Set A/B. They push FOR finer…

R1-2409911 CATT discussion not treated 5

Discussion on reply LS on applicable functionality reporting for beam management

CATT provides a comprehensive response to RAN2's liaison statement on AI/ML beam management for UE-side models, presenting 12 detailed proposals covering associated ID configuration, inference parameter handling, and CSI reporting mechanisms. The document advocates for Option 2 among three possible approaches for…

CATT advocates FOR Option 2 approach for applicability reporting based on efficiency analysis, supporting optional associated ID usage and reusing legacy CSI reporting frameworks. They push AGAINST mandatory associated ID requirements and immediate activation upon Step 3…

R1-2409925 CATT discussion not treated 9.1.1

Specification support for AI/ML-based beam management

CATT presents a comprehensive technical contribution on AI/ML-based beam management for 5G NR Rel-19, covering both UE-sided and network-sided models with 39 detailed proposals addressing configuration, inference, reporting, and performance monitoring aspects.

CATT advocates FOR flexible AI/ML beam management solutions that reuse existing CSI framework mechanisms while introducing minimal specification impact. They push FOR supporting both UE-sided and network-sided models with comprehensive performance monitoring, AGAINST overly…

R1-2409926 CATT discussion not treated 9.1.2

Specification support for AI/ML-based positioning

CATT's comprehensive technical document presents 35 proposals and 7 observations for AI/ML-based positioning across the NR air interface, covering data collection, model inference, performance monitoring, and consistency issues for all positioning cases (1, 2a, 2b, 3a, 3b).

CATT strongly advocates FOR sample-based channel measurements over path-based measurements due to superior AI/ML performance and reduced ambiguity, unified starting time across multiple TRPs using configured/predefined offsets rather than implementation-dependent methods, and…

R1-2409927 CATT discussion not treated 9.1.3

Specification support for AI/ML-based CSI prediction

CATT's document analyzes consistency requirements between training and inference phases for AI/ML-based CSI prediction in 5G NR networks. The document presents 3 observations and 3 proposals, demonstrating through simulations that network-side conditions like antenna tilt and TXRU mapping have negligible impact, while…

CATT advocates FOR relaxing consistency requirements for network-side antenna configurations (tilt angles, TXRU mapping) since they show negligible performance impact, while pushing FOR performance monitoring-based methods to handle other consistency factors like interference…

R1-2409928 CATT discussion not treated 9.1.4.1

Additional study on AI/ML-based CSI compression

This CATT document provides extensive analysis and evaluation results for AI/ML-based CSI compression inter-vendor training collaboration, containing 27 proposals and 25 observations covering various approaches including Direction A (parameter/dataset sharing with UE offline engineering), Direction B (NW encoder…

CATT strongly advocates FOR Direction B (Option 3b) over Direction A, supporting direct NW encoder parameter sharing to UE without offline engineering based on their evaluation showing better performance and less complexity. They push FOR standardized OTA signaling with RRC as…

R1-2409929 CATT discussion not treated 9.1.4.2

Additional study on AI/ML for other aspects

This 3GPP RAN1 technical document from CATT addresses various aspects of AI/ML for NR air interface, focusing on model identification, model transfer/delivery, and data collection issues. The document contains 15 proposals and 3 observations covering model identification options, transfer procedures, and feasibility…

CATT advocates for flexible and practical approaches to AI/ML model management, pushing FOR simplified model identification schemes with flexible dataset-to-model mappings while pushing AGAINST overly complex standardized model transfer mechanisms. They specifically argue that…

R1-2410571 CEWiT discussion not treated 9.1.2

Discussion on specification support for AI/ML Positioning Accuracy enhancement

CEWiT presents 18 proposals for AI/ML-based positioning accuracy enhancement, covering sample-based measurement reporting, training data collection procedures, and model monitoring frameworks across different positioning use cases. The document addresses time domain channel measurements, quality indicators, timing…

CEWiT advocates FOR sample-based over path-based channel measurement reporting due to better positioning accuracy despite higher overhead, supports distributed model monitoring responsibilities (gNB for Case 3a, LMF for Cases 2b/3b), and pushes for semi-supervised learning to…

R1-2410572 CEWiT discussion not treated 9.1.4.1

Discussion on AI/ML for CSI compression

CEWiT contributes technical analysis on AI/ML-based CSI compression for NR air interface, addressing data collection, model monitoring, and inference aspects. The document contains 14 proposals and 4 observations covering spatial-temporal-frequency CSI compression implementations.

CEWiT advocates FOR network-side monitoring as the primary approach for AI/ML CSI compression, supporting basis components transmission method over high-resolution codebook methods for reduced payload. They push FOR Option-1 CQI determination (not based on CSI reconstruction…

R1-2409994 China Telecom discussion not treated 9.1.1

Discussion on AI/ML for beam management

China Telecom's contribution discusses AI/ML for beam management in NR, focusing on Life Cycle Management (LCM) aspects including data collection, model inference, and performance monitoring for both network-sided and UE-sided models. The document contains 16 technical proposals covering BM-Case 1 (spatial beam…

China Telecom advocates FOR comprehensive support of multiple data collection options for flexibility across different LCM purposes (training, monitoring, inference), beam prediction accuracy KPIs as primary performance monitoring metrics, and full Set A measurement for accurate…

R1-2409995 China Telecom discussion not treated 9.1.3

Discussion on AI/ML for CSI prediction

China Telecom discusses ensuring consistency between training and inference phases for AI/ML-based CSI prediction in 5G NR, focusing on the reuse of associated ID mechanisms from beam management. The document contains 2 proposals addressing network-side additional condition consistency and UE assumptions for CSI…

China Telecom advocates FOR reusing the associated ID mechanism from AI-based beam management (AI 9.1.1) for CSI prediction to ensure training/inference consistency, but recognizes that UE assumptions about the same associated ID may need different interpretations for CSI…

R1-2409996 China Telecom discussion not treated 9.1.4.2

Discussion on other aspects of AI ML model and data

China Telecom's contribution discusses AI/ML model transfer/delivery and identification for NR air interface, presenting 4 proposals focused on deprioritizing certain model transfer cases and prioritizing specific alternatives for Case z4 implementation.

China Telecom advocates FOR prioritizing Case z4 model transfer with Alternative B approach (providing network flexibility in model structure management) and deprioritizing Case z1 due to cross-vendor collaboration complexity. They push FOR standardizing UE readiness reporting…

R1-2409494 CMCC discussion not treated 5

Discussion on LS on applicable functionality reporting for beam management UE-sided model

CMCC analyzes the signaling procedures for applicable functionality reporting in NR AI/ML beam management, specifically addressing Steps 3-5 of the RAN2-defined lifecycle. The document presents 15 proposals and 2 observations, arguing that NW-side additional conditions should be optional and functionality-specific,…

CMCC proposes that NW-side additional conditions be functionality specific and optional, arguing that UE can determine applicable functionalities based on inference configuration or parameters provided in Step 3 without mandatory NW-side conditions. They support both Option 1…

R1-2409499 CMCC discussion not treated 9.1.1

Discussion on specification support for beam management

CMCC presents 49 proposals and 2 observations regarding the specification impacts of AI/ML-based beam management for NR, covering data collection, inference, and monitoring for both NW-side and UE-side models. The document addresses configuration of Set A and Set B, reporting content for training and inference, Top-K…

CMCC proposes that L1 signaling be supported for NW-sided training data collection and that Top-K beam sweeping be supported for NW-side inference to enhance prediction accuracy. They require that for UE-side models, the overall CPU be separately counted between legacy CSI…

R1-2409500 CMCC discussion not treated 9.1.2

Discussion on specification support for positioning accuracy enhancement

CMCC analyzes specification impacts for AI/ML-based positioning in NR, presenting 17 proposals and 12 observations across methodology, measurement definitions, data collection, and model monitoring. The document argues for sample-based measurements over path-based ones to reduce processing ambiguity and emphasizes the…

CMCC slightly prefers sample-based measurements for AI/ML positioning to avoid the potential errors introduced by intermediate path-based processing at the UE. They propose that the entity deriving the AI model should provide the recommended number of samples (Nt') and that…

R1-2409501 CMCC discussion not treated 9.1.3

Discussion on AI/ML for CSI prediction

CMCC discusses specification impacts for AI/ML-based CSI prediction in Rel-19, focusing on ensuring consistency between training and inference phases. The document presents 8 proposals covering consistency mechanisms via associated IDs and performance monitoring, data collection reuse from beam management, and…

CMCC proposes studying two options for ensuring consistency of NW-side additional conditions across training and inference for AI-based CSI prediction: one based on associated ID and another based on performance monitoring. They propose combining these solutions, where the…

R1-2409502 CMCC discussion not treated 9.1.4.1

Discussion on AI/ML for CSI compression

CMCC presents evaluation results for AI/ML-based temporal domain CSI compression (Cases 2 and 3), demonstrating significant SGCS gains over Rel-16 and Rel-18 benchmarks. The document proposes prioritizing inter-vendor collaboration Directions A and B, specifies requirements for dataset/parameter exchange, and…

CMCC prioritizes inter-vendor collaboration Directions A and B for further study, arguing they offer better performance and lower specification effort than Direction C. They propose that for Direction A, the NW-side must share dataset information and performance targets to…

R1-2409503 CMCC discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML model and data

CMCC discusses model identification options (MI-Option 2, 3, and 4) and model transfer/delivery cases for NR AI/ML, proposing to prioritize MI-Option 2 and 3 while deprioritizing MI-Option 4 due to specification effort and RAN4 dependencies. The document outlines specific procedures for dataset transfer, model…

CMCC proposes prioritizing MI-Option 2 and MI-Option 3 for model identification while deprioritizing MI-Option 4 due to specification efforts and RAN4 dependencies. For MI-Option 2, they propose studying whether the NW or UE assigns the model ID and suggest that dataset IDs can…

R1-2410844 CMCC other endorsed 9.1

Session notes for 9.1 (AI/ML for NR Air Interface)

This is a session notes document from CMCC serving as Ad-hoc Chair for RAN1 #119 meeting on AI/ML for NR Air Interface. The document contains agreements and conclusions across 4 main areas: beam management, positioning accuracy enhancement, CSI prediction, and additional AI/ML studies including CSI compression and…

CMCC as Ad-hoc Chair is facilitating consensus building across all AI/ML use cases for NR air interface. They are advocating FOR systematic specification support across beam management, positioning, and CSI applications while promoting standardized approaches for model…

R1-2410192 Continental Automotive discussion not treated 9.1.4.2

Other aspects of AI/ML model and data

Continental Automotive presents their position on AI/ML model identification and transfer mechanisms for NR air interface, proposing 4 specific enhancements. The document focuses on configurable mapping relations for model-related information exchange and efficient parameter transfer across different model structures.

Continental advocates FOR configurable mapping relations as the primary mechanism for model identification and parameter transfer, strongly supporting the MI-Option2 framework with ID-X transmission from network to UE. They push FOR leveraging model structure similarity to…

R1-2409443 Ericsson discussion not treated 9.1.2

AI/ML for Positioning Accuracy Enhancement

Ericsson presents a comprehensive technical case for Rel-19 AI/ML-based positioning, strongly favoring sample-based measurements over legacy path-based reporting due to lower complexity and better generalization across different channel estimators. The document contains 73 proposals and 58 observations, arguing…

Ericsson argues inapplicability of Rel-18 carrier phase positioning for AI/ML inputs, proposing to down-prioritize CIR model inputs due to high signaling overhead and difficulty aligning phase measurements between training and inference. They present a technical case against…

R1-2409449 Ericsson discussion not treated 9.1.3

AI/ML for CSI prediction

Ericsson presents evaluation results for UE-sided AI/ML CSI prediction, concluding that no specification enhancements are needed to ensure consistency between training and inference regarding UE speed, deployment scenario, carrier frequency, NW antenna tilt, or TXRU mapping, as generalization performance degradation…

Ericsson argues inapplicability of the inconsistency issue identified for UE-sided spatial beam prediction to the UE-sided CSI prediction use case, which relies on temporal domain correlation rather than spatial beam sets. They conclude that no specification enhancement is…

R1-2409450 Ericsson discussion not treated 9.1.4.1

AI/ML for CSI compression

Ericsson presents a comprehensive analysis of inter-vendor training collaboration options for AI/ML-based CSI compression, arguing for the use of 3GPP synthetic data and standardized phase normalization to ensure interoperability. The document evaluates three main directions (A, B, and C), concludes that UE-side first…

Ericsson requires that reference models for inter-vendor collaboration (Options 1, 3a, 3b) be designed using 3GPP channel model based synthetic data rather than field data, citing excessive work and bias risks. They oppose UE-side first training for Options 3/4/5, arguing it…

R1-2409731 Ericsson discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML

Ericsson analyzes model identification options for two-sided AI/ML models in NR, specifically focusing on CSI compression use cases. The document presents 7 proposals and 6 observations, arguing that over-the-air dataset delivery is infeasible and recommending that model identification discussions be postponed until…

Ericsson opposes the over-the-air delivery of datasets from the network to the UE for two-sided model training, citing high complexity, signaling overhead, and questionable feasibility. They propose that model identification discussions for MI-Options 2, 3, and 4 be postponed…

R1-2410354 Ericsson discussion not treated 9.1.1

AI/ML for beam management

Ericsson presents a comprehensive technical document on AI/ML for beam management in NR air interface, containing 20 proposals and 3 observations. The document addresses both UE-sided and NW-sided AI/ML models for beam management enhancement, covering data collection, inference reporting, performance monitoring, and…

Ericsson advocates for a comprehensive AI/ML framework that leverages existing CSI mechanisms with minimal specification impact while maximizing functionality. They strongly push FOR: (1) expanding aperiodic CSI-RS resources from 16 to 64 beams to enable practical AI/ML…

R1-2410714 Ericsson discussion noted 9.1.2

Summary #1 of specification support for positioning accuracy enhancement

This 3GPP RAN1 meeting summary document (R1-2410714) from Ericsson covers AI/ML for NR Air Interface positioning accuracy enhancement, containing over 90 proposals and conclusions across 8 major sections covering model input, output, training data collection, inference, and monitoring.

Ericsson advocates FOR: (1) Supporting both sample-based and path-based measurements as compromise solution to enable progress, (2) Reusing existing legacy IEs and frameworks wherever possible to minimize specification impact, (3) LMF-centric functionality management with…

R1-2410715 Ericsson discussion noted 9.1.2

Summary #2 of specification support for positioning accuracy enhancement

This document contains approximately 15 proposals and 2 conclusions from Ericsson covering AI/ML positioning enhancements including sample-based vs path-based measurements, LOS/NLOS indicators for model output, training data collection procedures, and model inference consistency requirements.

Ericsson advocates FOR: (1) supporting both sample-based and path-based measurements as a compromise solution to avoid blocking progress, (2) reusing existing legacy IEs and signaling frameworks wherever possible to minimize specification impact, (3) label-free monitoring…

R1-2410716 Ericsson discussion noted 9.1.2

Summary #3 of specification support for positioning accuracy enhancement

This document is RAN1#119 Summary #3 from Ericsson covering AI/ML positioning accuracy enhancement with approximately 120+ proposals/conclusions across model input, output, training data collection, inference, and monitoring. The document extensively discusses sample-based vs path-based measurements, LOS/NLOS…

Ericsson as document moderator presents a balanced compromise approach, advocating FOR: (1) supporting both sample-based and path-based measurements to avoid blocking progress despite company divisions, (2) reusing existing signaling frameworks and IEs where possible to minimize…

R1-2410717 Ericsson discussion noted 9.1.2

Summary #4 of specification support for positioning accuracy enhancement

This 3GPP RAN1 document (Tdoc R1-2410717) from Ericsson summarizes discussions on AI/ML-based positioning accuracy enhancements from RAN1#119, containing over 80 proposals across model input definitions, output specifications, training data collection, and model inference procedures. The document focuses on technical…

Ericsson advocates FOR sample-based measurements over path-based measurements, supporting flexible parameter configurations (Nt, Nt', k) while maintaining backward compatibility with legacy positioning methods. They push FOR reusing existing IEs and frameworks to minimize…

R1-2410718 Ericsson discussion noted 9.1.2

Summary #5 of specification support for positioning accuracy enhancement

This RAN1 document from Ericsson presents 95 proposals across 6 major technical areas for AI/ML-based positioning enhancement in NR, covering model input definitions, model output specifications, training data collection, inference procedures, and performance monitoring frameworks.

Ericsson advocates FOR sample-based measurements as the primary approach for AI/ML positioning input (supporting majority view over compromise approaches), FOR reusing existing legacy signaling frameworks and IEs where possible to minimize specification impact, and FOR…

R1-2410921 Ericsson discussion noted 9.1.2

Final summary of specification support for positioning accuracy enhancement

This 3GPP RAN1 document (R1-2410921) from Ericsson presents a final summary of discussions on AI/ML for NR Air Interface positioning accuracy enhancement from RAN1#119 meeting. The document contains over 160 proposals and conclusions across model input/output, training data collection, inference consistency, and model…

Ericsson advocates FOR sample-based measurements as an enhancement to legacy path-based reporting, supporting both alternatives to avoid blocking progress while preferring sample-based for better AI/ML positioning performance. They push AGAINST requiring phase information (CIR)…

R1-2410257 ETRI discussion not treated 9.1.1

Discussion on specification support for beam management

ETRI's contribution presents 20 proposals and 3 observations on AI/ML-based beam management for NR air interface, covering configuration procedures, performance monitoring, measurement reporting, and consistency mechanisms for both UE-sided and NW-sided models across spatial and temporal domain beam prediction.

ETRI advocates for a comprehensive framework supporting both UE-sided and NW-sided AI/ML models with emphasis on overhead reduction and consistency mechanisms. They strongly push FOR Options 1 and 2 over Option 3 for applicability procedures, L1-RSRP difference as a key…

R1-2410258 ETRI discussion not treated 9.1.2

Discussion on specification support for positioning accuracy enhancement

ETRI presents a comprehensive technical contribution on AI/ML for NR air interface positioning accuracy enhancement, covering model input/output formats, training data collection, and performance monitoring across different positioning cases. The document contains 21 detailed proposals addressing various aspects of…

ETRI strongly advocates for vendor flexibility in AI/ML positioning implementation, pushing for UE vendors and gNB manufacturers to determine their own model input formats rather than enforcing standardized formats. They support Option B for measurement window timing (earliest…

R1-2410259 ETRI discussion not treated 9.1.3

Discussion on specification support for CSI prediction

ETRI discusses specification support for CSI prediction using AI/ML models, focusing on ensuring consistency between training and inference phases to prevent performance degradation. The document contains 3 proposals and 1 observation, advocating for associated ID support similar to beam management solutions.

ETRI advocates FOR extending the already-agreed associated ID support from AI/ML beam management to CSI prediction to reduce specification workload and ensure training/inference consistency. They push FOR leveraging existing CSI framework configurations (CSI report configuration…

R1-2410260 ETRI discussion not treated 9.1.4.1

Discussion on AI/ML for CSI compression

ETRI presents comprehensive views on AI/ML-based CSI compression for NR air interface Release-19 further study, addressing temporal domain aspects and inter-vendor training collaboration. The document contains 8 proposals and 13 observations covering specification impacts, training collaboration directions, and…

ETRI advocates FOR supporting both Direction A (UE-side offline engineering) and Direction B (direct parameter sharing) inter-vendor training collaboration approaches, emphasizing the importance of maximizing reuse of existing CSI-RS specifications for temporal domain aspects.…

R1-2410261 ETRI discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML model and data

ETRI presents their technical views on AI/ML model identification and data management for NR air interface, making 4 specific proposals and 1 observation focused on establishing many-to-many relationships between IDs and models, and incorporating inference time information into model structures.

ETRI advocates FOR establishing many-to-many relationships between various identifiers (Associated ID, ID-X) and Model IDs across both MI-Option1 and MI-Option2, enabling flexible model training scenarios where single datasets can train multiple models and single models can…

R1-2410205 Fraunhofer discussion not treated 9.1.2

AI/ML positioning accuracy enhancement

This Fraunhofer technical document presents 19 proposals and 1 observation for AI/ML positioning accuracy enhancements in NR air interface, covering measurement enhancements, training data collection optimization, and comprehensive model lifecycle management frameworks for Cases 1 and 3a.

Fraunhofer advocates FOR comprehensive AI/ML positioning frameworks that maintain network-centric control (LMF-based functionality management) while supporting flexible measurement approaches and intelligent model lifecycle management. They push FOR complex-valued CIR reporting…

R1-2410255 Fraunhofer discussion not treated 9.1.1

Specification support for beam management

This Fraunhofer document presents 17 proposals for AI/ML-based beam management in 5G NR, covering both UE-sided and network-sided models with focus on performance monitoring, configuration optimization, and overhead reduction. The proposals address two main use cases (BM-Case1 and BM-Case2) with comprehensive…

Fraunhofer strongly advocates FOR UE-sided AI/ML models over network-sided models, emphasizing their advantages in reduced signaling overhead, shorter measurement-to-inference delay, and ability to use more measurements. They push FOR bitmap-based indexing as the most efficient…

R1-2410048 Fujitsu discussion not treated 9.1.1

Discussion on specification support on AI/ML for beam management

Fujitsu presents a comprehensive technical document with 35 proposals for AI/ML-based beam management in 5G NR Release 19, covering RAN2 liaison responses, training data collection, UE-side and network-side model inference, and performance monitoring across both spatial (BM Case-1) and temporal (BM Case-2) beam…

Fujitsu advocates FOR Option-3 in applicable functionality reporting (UE-driven parameter reporting rather than network-driven configuration), supports aperiodic CSI-RS for temporal beam prediction, prefers separate configurations for monitoring vs inference, and pushes for…

R1-2410049 Fujitsu discussion not treated 9.1.2

Discussion on specification support for AIML-based positioning accuracy enhancement

This Fujitsu contribution discusses AI/ML-based positioning accuracy enhancement for NR air interface, providing 12 proposals and 7 observations across model input/output, training-inference consistency, and performance monitoring aspects for various positioning cases.

Fujitsu advocates FOR: sample-based measurement with configured offset (Option D), deprioritizing phase information study due to negligible gains vs overhead, using LOS/NLOS indicator as confidence level rather than LOS likelihood, supporting multiple timing information…

R1-2410050 Fujitsu discussion not treated 9.1.3

Discussion on specification support for CSI prediction

Fujitsu presents evaluation results and proposals for ensuring consistency between training and inference for AI/ML-based CSI prediction with UE-side models in 5G NR. The document contains 1 observation and 2 proposals focused on identifying key network-side conditions that impact model generalization performance.

Fujitsu advocates FOR using associated ID as the primary mechanism to ensure training/inference consistency for CSI prediction, similar to the approach already agreed for beam management. They are pushing AGAINST considering tilt angle as a key network-side condition, based on…

R1-2410051 Fujitsu discussion not treated 9.1.4.1

Discussion on CSI compression with AI/ML

Fujitsu's comprehensive technical document analyzes CSI compression with AI/ML for NR air interface, presenting evaluation results for non-ideal UCI feedback and detailed discussions on inter-vendor training collaboration across three directions (A, B, C). The document contains 24 proposals and 21 observations…

Fujitsu strongly advocates FOR Direction A option 4-1 over 3a-1 due to similar performance with less specification effort, synthetic data training for reference models in Direction C, NW-side monitoring over UE-side proxy model monitoring due to reliability concerns, and…

R1-2410052 Fujitsu discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML model and data

Fujitsu's contribution addresses unresolved aspects of AI/ML model identification and model transfer/delivery for NR air interface, presenting 12 proposals and 1 observation across model identification procedures (MI-Option2), model transfer cases, and unified procedures for two-sided models.

Fujitsu advocates FOR comprehensive dataset-based model identification (MI-Option2) with proper ID management and test dataset procedures, while pushing AGAINST impractical model transfer cases - specifically arguing to deprioritize Case z1 (due to unclear benefits over Case y)…

R1-2410029 FUTUREWEI discussion not treated 9.1.1

Discussion on specification support for AI/ML-based beam management

FUTUREWEI presents 11 proposals for AI/ML-based beam management specification support in NR Release 19, covering performance monitoring, model inference, data collection, and assistance information. The document focuses on reusing existing CSI framework while avoiding overly complex new metrics and signaling…

FUTUREWEI advocates FOR maximizing reuse of existing CSI framework to minimize specification effort, supporting simpler approaches that avoid complex new metrics like probability and confidence information. They push AGAINST supporting multiple alternatives that would increase…

R1-2410030 FUTUREWEI discussion not treated 9.1.4.1

Discussion of CSI compression on AI/ML for NR air interface

Futurewei presents a comprehensive analysis of AI/ML-based CSI compression for NR air interface, covering inter-vendor training collaboration options, specification impacts, and performance evaluation results. The document contains 10 formal proposals and 6 observations addressing parameter/model exchange methods,…

Futurewei advocates FOR using enhanced Rel-16 eType II codebook with new parameters for both data collection and monitoring to achieve better performance, despite overhead concerns. They push FOR differentiated delivery methods based on latency requirements (over-the-air for…

R1-2410031 FUTUREWEI discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML model and data on AI/ML for NR air-interface

This Futurewei document discusses AI/ML model identification and transfer/delivery for NR air interface, presenting 14 proposals and 4 observations covering four model identification options and various model transfer cases. The document advocates for network-assigned model IDs, deprioritization of certain transfer…

Futurewei advocates for network-controlled model identification with associated IDs limited to cell scope, argues against using associated IDs as model IDs, and pushes to deprioritize complex model transfer cases (z1, z2) that don't provide clear benefits over simpler OTT…

R1-2410149 Google discussion not treated 9.1.1

AI/ML based Beam Management

This 3GPP RAN1 technical document from Google presents 27 proposals for AI/ML-based beam management enhancements in 5G NR, covering beam measurement, reporting, indication, failure recovery, and performance monitoring across both network-side and UE-side AI/ML models.

Google advocates FOR comprehensive AI/ML beam management support including both network-side and UE-side models, pushing for flexible reference signal configurations (AP-CSI-RS, SSB), enhanced reporting mechanisms with confidence indicators, separate TCI state pools for AI/ML…

R1-2410150 Google discussion not treated 9.1.2

AI/ML based Positioning

Google's contribution discusses ML-based positioning for NR, presenting 4 key proposals covering training data collection, model monitoring, and measurement configurations. The document builds on previous RAN1 agreements and focuses on Case 1 UE-based positioning with emphasis on transparent model performance…

Google advocates FOR transparent model performance monitoring that does not require UEs to disclose location information to the network, supporting UE-side monitoring (Option A) over network-side monitoring (Option B). They push FOR flexible channel measurement options including…

R1-2410151 Google discussion not treated 9.1.3

AI/ML based CSI Prediction

Google proposes enhancements for AI/ML based CSI prediction in NR air interface, focusing on UE-side model consistency between training and inference. The document contains 2 main proposals addressing CSI-RS configuration reporting and associated ID management for CSI report configurations.

Google advocates FOR UE-side only CSI prediction models with UE-reported preferred CSI-RS configurations to avoid configuration mismatches, and FOR associated ID management per CSI report configuration to maintain training/inference consistency. They push FOR dynamic…

R1-2410152 Google discussion not treated 9.1.4.1

AI/ML based CSI Compression

Google's technical document on ML-based CSI compression for 5G NR presents 15 proposals covering CSI report content, processing units, model monitoring, data collection, and inter-vendor collaboration. The document addresses key aspects of AI/ML integration into NR air interface including compressed W2 reporting, dual…

Google advocates for a comprehensive AI/ML-based CSI compression framework that prioritizes practical deployment considerations. They push FOR: (1) transformer-based standardized models over other architectures, (2) flexible hybrid AI/ML and non-AI/ML approaches based on rank…

R1-2410153 Google discussion not treated 9.1.4.2

AI/ML Model and Data

Google presents a technical document on AI/ML model identification and data collection for NR air interface, proposing 8 specific proposals covering model identification types, UE data collection procedures, and ID configuration schemes. The document advocates for simplifying model identification approaches by…

Google advocates FOR simplifying AI/ML model identification by assuming model IDs are known after UE connection (MI type A) and prioritizing MI-Option 1 for maintaining understanding of model properties. They push AGAINST the complexity of MI-Option 2 and MI-Option 3, arguing…

R1-2410174 HONOR discussion not treated 9.1.1

Discussion on AI/ML for beam management

HONOR's technical document presents proposals for AI/ML-based beam management in NR, addressing performance monitoring metrics, inference reporting for both UE-sided and network-sided models, and measurement report optimization. The document contains 13 specific proposals across performance monitoring, model…

HONOR advocates FOR Alt 2 definition of Top-K beam prediction accuracy (measuring if top measured beam is within top-K predicted beams) as more appropriate for scenarios with second-round beam sweeping, supports UE pre-evaluation mechanisms to reduce signaling overhead, and…

R1-2409395 Huawei discussion not treated 9.1.1

Discussion on AI/ML for beam management

This Huawei Tdoc (R1-2409395) addresses AI/ML for NR Air Interface beam management, presenting 40 proposals and 9 observations across data collection, NW-side and UE-side models, inference, monitoring, and LCM. Key technical stances include supporting larger beam sets (up to 256), reusing legacy TCI timelines for…

Huawei proposes studying mechanisms to support beam sets exceeding 64 resources, either through multiple legacy sets or a single set up to 256 resources. They require the associated ID for UE-side models to be limited to a cell-specific manner to avoid proprietary disclosure and…

R1-2409396 Huawei discussion not treated 9.1.2

Discussion on AI/ML for positioning accuracy enhancement

This Huawei contribution discusses AI/ML for positioning accuracy enhancement in NR, covering model input, output, training, consistency, monitoring, and lifecycle management. It contains 30 proposals and 17 observations, arguing for implementation flexibility, reuse of legacy mechanisms, and protection of proprietary…

Huawei argues that ambiguity in sample-based measurements for Case 3b can be avoided by implementation rather than strict specification, proposing that gNBs flexibly determine selection window parameters (Nt, Nt', k) while capping Nt' at 16 and Nt at 64 to protect proprietary…

R1-2409397 Huawei discussion revised 9.1.3

Discussion on AI/ML for CSI prediction

Huawei argues against introducing network-side indications, such as associated IDs, to ensure consistency between training and inference for UE-side AI/ML models in CSI prediction, citing feasibility issues and proprietary disclosure risks. The document presents 5 observations and 2 proposals, asserting that…

Huawei presents a technical case against the necessity and feasibility of introducing associated IDs or other NW-side indications for CSI prediction consistency. They argue that the massive number of impacting factors, such as antenna layout and down tilt angles, makes…

R1-2409398 Huawei discussion not treated 9.1.4.1

Discussion on AI/ML for CSI compression

This Huawei contribution discusses inter-vendor training collaboration for AI/ML-based CSI compression in NR Rel-19, focusing on Direction A (dataset/parameter exchange) and temporal domain cases. It presents 17 proposals and 11 observations covering training methods, overhead alleviation, specification impacts for…

Huawei argues that for Direction A inter-vendor collaboration, sharing datasets (Option 4-1) incurs less proprietary risk than sharing model parameters (Option 3a-1/3b). They propose that model backbone information is unnecessary for Option 4-1, allowing UE autonomy in structure…

R1-2409399 Huawei discussion not treated 9.1.4.2

Discussion on other aspects of the additional study for AI/ML

Huawei analyzes model identification options for two-sided AI/ML models in NR, proposing to exclude MI-Option 1 as it applies only to one-sided cases. The document details information elements for dataset and model transfers (MI-Options 2 and 3), argues for Case y as the baseline for UE-trained models to avoid…

Huawei argues that MI-Option 1 is inapplicable to the revised WID scope for two-sided models and proposes excluding it from further discussion. For MI-Option 2 and 3, Huawei specifies that model identification relies on the delivery of Dataset IDs or Model IDs alongside their…

R1-2410326 Huawei discussion not treated 5

Discussion on the LS reply to RAN2 on functionality in AI/ML

This Huawei document discusses the framework for functionality-based Life Cycle Management (LCM) for AI/ML in NR air interface, specifically addressing RAN2 questions on beam management UE-side model functionality reporting. The document contains 16 proposals and 3 observations covering functionality granularity,…

Huawei advocates FOR finer granularity of applicable functionalities (Option 2) rather than coarse use case/sub-use case granularity, enabling better network control and more efficient UE memory usage. They push FOR reusing legacy CSI framework signaling rather than introducing…

R1-2410654 Huawei discussion not treated 9.1.3

Discussion on AI/ML for CSI prediction

Huawei argues against introducing associated IDs for ensuring training/inference consistency in CSI prediction for AI/ML-enhanced NR air interface, presenting 5 observations and 2 proposals. The document demonstrates through simulation results that generalized AI/ML models can achieve satisfactory performance using…

Huawei advocates AGAINST introducing network-side associated IDs or explicit indications for CSI prediction consistency, arguing that such mechanisms are both infeasible (due to massive impacting factors and network burden) and unnecessary (since generalized models work well…

R1-2410589 IIT Kanpur discussion not treated 9.1.4.1

Discussion an AI/ML based CSI Compression

IIT Kanpur presents evaluation results for AI/ML-based CSI compression using temporal-spatial-frequency domain approaches, focusing on Case 2 (reconstruction with temporal correlation) and Case 3 (joint prediction). The document contains 2 key observations about the performance differences between reconstructive and…

IIT Kanpur advocates for enhanced temporal correlation representation in AI/ML models for CSI compression, particularly emphasizing that joint prediction tasks (Case 3) require more sophisticated temporal modeling compared to reconstruction tasks (Case 2). They push for…

R1-2410373 Indian Institute of Tech (M) discussion not treated 9.1.1

Discussion on Specification Support of AI/ML for Beam Management

This document from Indian Institute of Technology addresses AI/ML for beam management in NR air interface, providing responses to RAN2 questions and proposing specification support mechanisms. It contains 8 proposals and 5 observations covering signaling procedures, performance monitoring, and resource configuration.

IIT advocates FOR mandatory NW-side additional conditions in beam management, supporting Option 1 for UE-side model inference with CSI-ReportConfig framework. They push FOR implicit L1/L2 signaling through performance thresholds rather than explicit signaling, and advocate FOR…

R1-2410424 Indian Institute of Tech (M) discussion not treated 9.1.2

Specification Support of AI/ML for Positioning Accuracy Enhancement

This document from Indian Institute of Technology proposes recommendations for AI/ML-based positioning accuracy enhancement in NR, comparing sample-based and path-based measurements for model training. It contains 1 proposal and 1 observation focusing on optimal measurement approaches and path requirements.

IIT advocates FOR Option B (starting time based on first detectable path power) for sample-based measurements over Option A, claiming superior positioning accuracy with sample-based approaches compared to path-based measurements. They push AGAINST increasing the number of…

R1-2410425 Indian Institute of Tech (M) discussion not treated 9.1.4.1

Discussion and evaluation results on AI/ML for CSI Compression

This document from Indian Institute of Technology presents detailed implementation approaches for AI/ML-based CSI compression inter-vendor compatibility options, specifically evaluating parameter exchange (Option 3a-1) versus dataset exchange (Option 4-1) methods. The document contains 11 observations across…

IITM advocates FOR detailed implementation of both Option 3a-1 (parameter exchange) and Option 4-1 (dataset exchange) with specific multi-step training procedures, supporting dataset sharing for lower payload sizes while showing both methods are viable for higher payloads. They…

R1-2409741 Intel discussion not treated 9.1.1

Specification support for beam management

Intel presents a comprehensive set of proposals for AI/ML-aided beam management in NR Rel-19, covering configuration, data collection, inference, and performance monitoring for both network-side and UE-side models. The document contains approximately 35 distinct proposals aimed at leveraging existing CSI frameworks…

Intel proposes that NW-sided models utilize implicit Set A/B configuration via legacy CSI frameworks to minimize signaling overhead, whereas UE-sided models require explicit configuration of Sets A and B, potentially using a new IE for Set A resources. They require the…

R1-2409742 Intel discussion not treated 9.1.2

Specification support for positioning accuracy enhancement

Intel presents 26 proposals and 2 observations regarding specification support for AI/ML-based positioning accuracy enhancements in Rel-19, focusing on data collection, model input/output characterization, and consistency between training and inference. The document argues that sample-based measurements are a specific…

Intel argues that the sample-based measurement approach is a specific implementation of the path-based approach already supported in existing specifications, proposing to support path-based measurement with potential enhancements to the number of reported paths. They propose…

R1-2409743 Intel discussion not treated 9.1.4.1

AI/ML for CSI compression

Intel analyzes the impact of data distribution mismatch on AI/ML-based CSI compression performance, highlighting asymmetric performance losses between different subarray configurations and model complexities. The document presents three proposals addressing the careful selection of synthetic data parameters for…

Intel presents technical evidence that performance loss due to data distribution mismatch is asymmetric and dependent on model complexity, specifically noting that training on 1x1 subarray data and testing on 4x1 subarray data leads to >10% SGCS loss, whereas the reverse causes…

R1-2409744 Intel discussion not treated 9.1.4.2

Other aspects of AI/ML model and data

Intel presents 15 proposals and 7 observations regarding AI/ML model identification and transfer mechanisms for NR Rel-19, specifically focusing on CSI compression use cases. The document argues for model-ID-based identification to enable granular Life Cycle Management (LCM) and proposes prioritizing model…

Intel proposes supporting model-ID-based identification for two-sided CSI compression models to enable finer granularity in Life Cycle Management (LCM) compared to functionality-level identification. They argue that MI-Option 1 (data collection configuration), MI-Option 2…

R1-2409455 InterDigital discussion not treated 9.1.1

Discussion on AI/ML for beam management

InterDigital presents 29 proposals and 24 observations regarding AI/ML for beam management in NR, focusing on configuration frameworks, reporting overhead reduction, and lifecycle management. The document argues for Option 2 for UE-side model applicability to minimize signaling overhead and supports a unified…

InterDigital supports Option 2 for UE-side model applicability identification, arguing that Option 1 requires excessive configuration overhead for candidate CSI report configurations. They prefer Alt 2 for beam configuration, utilizing a single CSI-ResourceConfigId for both Set…

R1-2409845 InterDigital discussion not treated 9.1.2

Discussion on support for AIML positioning

InterDigital's comprehensive technical document presents 38 proposals and 24 observations for AIML positioning in NR, focusing on Case 1 (UE-based positioning with UE-side model), Case 3a (gNB-side model), and Case 3b (LMF-side model). The document prioritizes consistency mechanisms, ground truth quality indicators,…

InterDigital advocates FOR leveraging existing positioning methods (UE-based DL-TDOA) as the foundation for AIML positioning rather than creating new methods, FOR path-based measurements over sample-based alternatives, FOR LMF-controlled ground truth quality indicators, and FOR…

R1-2410042 InterDigital discussion not treated 9.1.3

On AI/ML-based CSI prediction

InterDigital presents evaluation results on AI/ML-based CSI prediction for NR air interface, demonstrating that UE-sided models can generalize well across different network conditions without requiring complex associated ID mechanisms. The document contains 1 formal proposal and 3 observations focused on simplifying…

InterDigital advocates FOR simplifying CSI prediction by eliminating the need for associated ID mechanisms, arguing that UE-sided models can generalize well across different network conditions without complex consistency checking. They are pushing AGAINST the adoption of…

R1-2410043 InterDigital discussion not treated 9.1.4.1

On AI/ML-based CSI compression

InterDigital presents comprehensive analysis of AI/ML-based CSI compression for NR air interface, covering beam domain processing, temporal-spatial-frequency compression, model monitoring, and inter-vendor collaboration. The document contains 25 observations and 5 proposals addressing performance improvements,…

InterDigital advocates FOR beam domain processing as a superior approach to spatial domain processing, offering better generalization across antenna configurations and reduced complexity. They strongly push FOR temporal-spatial-frequency (TSF) compression Case 2 over…

R1-2410044 InterDigital discussion not treated 9.1.4.2

On other aspects of AI/ML model and data

InterDigital's contribution addresses AI/ML model identification for two-sided models, data collection for training, and model transfer/delivery aspects for NR air interface. The document contains 14 proposals and 9 observations across model identification challenges, CSI enhancement requirements, positioning use…

InterDigital advocates for deferring complex model identification discussions until RAN4 makes progress, emphasizing that dataset-only exchange is insufficient for two-sided model compatibility. They push for LMF-centralized ground truth quality control in positioning, enhanced…

R1-2410547 ITL discussion not treated 9.1.1

Specification support for AI/ML beam management

ITL's comprehensive technical document presents 25 detailed proposals for AI/ML-based beam management in NR air interface, covering both network-side and UE-side AI/ML models for spatial and temporal beam prediction (BM-Case1 and BM-Case2), along with performance monitoring mechanisms.

ITL advocates FOR flexible beam reporting with up to 256 beams per report, larger quantization steps to reduce overhead, and supporting both network-side and UE-side AI/ML models with unified TCI framework. They push FOR reusing existing CSI reporting frameworks and…

R1-2410588 ITL discussion not treated 9.1.2

Discussions on specification support for positioning accuracy enhancement for AI/ML

ITL presents technical specifications for AI/ML-based positioning enhancements in NR, addressing timing reference methods and training data generation entities across multiple positioning cases. The document contains 3 main proposals covering reference time definitions for multi-RTT support and standardizing training…

ITL advocates FOR unified reference timing approaches across different AI/ML positioning cases, supporting multi-RTT capability by using different timing references for UL and DL parts. They push FOR including multiple entities (PRU, Non-PRU UE, LMF) as valid sources for…

R1-2410201 KAIST discussion not treated 9.1.4.1

Discussion on AI/ML for CSI compression

KAIST proposes enhancements to AI/ML-based CSI compression for temporal domain aspects, specifically addressing non-ideal UCI feedback scenarios. The document contains 2 main proposals focusing on incorporating additional information beyond UCI loss probability for better CSI reconstruction and historical information…

KAIST advocates FOR incorporating multiple error indicators (ACK/NACK probability, data error probability) beyond just UCI loss probability when managing historical CSI information in temporal domain AI/ML compression. They push FOR more sophisticated error evaluation mechanisms…

R1-2410504 KDDI Corporation discussion not treated 9.1.1

Specification support for beam management

KDDI Corporation proposes 7 specifications for AI/ML beam management in 5G NR, focusing on overhead reduction mechanisms for network-sided models and performance monitoring approaches for UE-sided models. The document addresses both temporal correlation exploitation in BM-Case2 and efficient monitoring alternatives…

KDDI advocates FOR overhead reduction mechanisms that exploit temporal correlation in RSRP values and flexible threshold-based beam reporting, while pushing FOR efficient Alt 3 performance monitoring over more resource-intensive alternatives. They advocate AGAINST approaches…

R1-2410344 KT discussion not treated 9.1.1

Discussion on AI/ML based beam management

KT's technical document proposes 11 specific enhancements for AI/ML-based beam management in NR air interface, addressing UE-sided model consistency, resource set associations, and performance monitoring. The document focuses on supporting both associated ID-based and performance monitoring approaches for ensuring…

KT advocates FOR multi-cell/cell-group level associated IDs (instead of cell-level only) to reduce UE complexity, FOR dual approach combining both associated ID-based and performance monitoring methods for consistency assurance, FOR Alt 2 using single CSI-ResourceConfigId for…

R1-2409569 Kyocera discussion not treated 9.1.1

Specification Support for AI/ML for Beam Management

Kyocera presents a comprehensive framework for AI/ML-assisted beam management in NR Rel-19, covering configuration, inference reporting, consistency, and performance monitoring for both NW-sided and UE-sided models. The document contains 35 proposals and 4 observations, focusing on defining Set A/B configurations,…

Kyocera proposes that for UE-side AI/ML models, Set A be virtually configured for reference mapping while Set B is explicitly configured for measurements, requiring a new IE to associate beams with these sets. They require the introduction of an 'associated ID' within the CSI…

R1-2410018 Lenovo discussion not treated 9.1.1

AI/ML specification support for beam management

This Lenovo contribution presents 26 proposals for AI/ML specification support in NR beam management, covering data collection, model inference for both UE-side and NW-side implementations, performance monitoring, and UE capability reporting across spatial-domain (BM-Case1) and temporal (BM-Case2) beam prediction use…

Lenovo advocates for a comprehensive AI/ML beam management framework that supports both UE-side and NW-side inference with strong emphasis on: 1) UE autonomy in data collection and model training initiation, 2) unified configuration approaches supporting both BM-Case1 and…

R1-2410019 Lenovo discussion not treated 9.1.2

Specification impacts for AI/ML positioning

Lenovo's comprehensive technical document on AI/ML positioning for 3GPP RAN1, presenting 30 proposals and 4 observations covering specification impacts for enhanced positioning accuracy. The document addresses measurement definitions, model inputs/outputs, training data collection, and implementation consistency…

Lenovo advocates FOR a hybrid measurement approach supporting both sample-based and path-based timing representations depending on the use case, with strong emphasis on reusing legacy frameworks where possible (LOS/NLOS indicators, assistance data signaling). They push FOR…

R1-2410020 Lenovo discussion not treated 9.1.3

On AI/ML for CSI prediction

Lenovo's document analyzes training/inference consistency challenges for UE-sided AI/ML-based CSI prediction in 5G NR, presenting 7 key proposals and 8 observations. The contribution evaluates four approaches to maintain consistency and recommends focusing on monitoring-based techniques while deprioritizing model…

Lenovo advocates FOR monitoring-based approaches (Type 1 and Type 3) and limited NW-side additional condition indication for maintaining training/inference consistency in UE-sided CSI prediction. They push AGAINST model transfer from network to UE due to significant overhead…

R1-2410021 Lenovo discussion not treated 9.1.4.1

On AI/ML for CSI compression

Lenovo's technical contribution on AI/ML for CSI compression addresses data collection, model monitoring, inter-vendor collaboration, and quantization schemes. The document contains 19 specific proposals across 6 major technical areas, with emphasis on two-sided model training and deployment challenges.

Lenovo advocates FOR Direction A inter-vendor collaboration (UE-side offline engineering) over Direction B (direct parameter use), strongly pushing AGAINST Direction C (fully standardized models) due to performance limitations and specification complexity. They prioritize…

R1-2410022 Lenovo discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML model and data

Lenovo's document discusses model identification procedures for two-sided AI/ML models in CSI compression, focusing on inter-vendor training collaboration options. The document contains 9 key proposals addressing model structure sharing, dataset identification, and standardized model formats for collaborative AI/ML…

Lenovo advocates FOR standardized model identification procedures that enable inter-vendor collaboration while maintaining vendor-specific optimization capabilities. They support using dataset IDs (ID-X) for model-dataset association and favor exploring multiple open format…

R1-2410193 LG Electronics discussion not treated 9.1.1

Discussions on AI/ML for beam management

LG Electronics' comprehensive contribution on AI/ML for NR beam management covers data collection, inference, and performance monitoring for both network-sided and UE-sided models, presenting 25 detailed proposals and 4 observations across spatial and temporal beam prediction scenarios.

LG Electronics advocates for practical, overhead-conscious AI/ML beam management solutions, strongly pushing for Alt 4 configuration approach and Set B-only configurations to reduce signaling overhead, while opposing AP CSI-RS support for temporal prediction due to resource…

R1-2410194 LG Electronics discussion not treated 9.1.3

Discussions on CSI prediction

This LG Electronics contribution discusses AI/ML-based CSI prediction for NR air interface, focusing on consistency between training and inference, and presents simulation results showing marginal performance loss for gNB antenna tilt angle variations. The document contains 4 proposals and 1 observation addressing…

LG Electronics advocates FOR simplifying specification requirements by concluding that no additional spec support is needed for gNB antenna tilt angle consistency, while pushing FOR prioritizing Type 1 and Type 3 performance monitoring over Type 2 due to lower reporting overhead…

R1-2410195 LG Electronics discussion not treated 9.1.4.1

Study on CSI compression

LG Electronics proposes 9 technical proposals for improving AI/ML CSI compression performance in NR, focusing on temporal/spatial/frequency domain compression, inter-vendor training collaboration, and addressing practical implementation challenges like UCI feedback issues and rank adaptation.

LG Electronics strongly advocates FOR extending CSI compression to temporal/spatial/frequency (TSF) domain to leverage past channel information correlation, supporting dataset exchange method (Option 4-1) over model parameter exchange (Option 3a-1) for inter-vendor collaboration…

R1-2410196 LG Electronics discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML model and data

LG Electronics presents 8 proposals addressing AI/ML model identification, lifecycle management, and transfer mechanisms for NR air interface. The document focuses on clarifying functionality granularity, model identification options, and addressing practical challenges in model delivery scenarios.

LG Electronics advocates for a functionality-based approach to AI/ML model lifecycle management with flexible sub-use-case specific configurations, while pushing against overly rigid model identification schemes. They emphasize practical implementation challenges like…

R1-2410817 LG Electronics discussion noted 9.1.3

Summary #1 of CSI prediction

This 3GPP RAN1 document from LG Electronics summarizes CSI prediction evaluation results and consistency issues between training and inference for UE-sided AI/ML models. The document contains numerous observations from multiple companies regarding generalization performance across different network conditions, with…

LG Electronics, as the moderator, takes a consensus-building approach advocating for concluding that tilt angle and TXRU mapping have negligible impact on CSI prediction generalization performance based on majority company results. They push for drawing high-level conclusions…

R1-2410818 LG Electronics discussion noted 9.1.3

Summary #2 of CSI prediction

This 3GPP RAN1 technical document from LG Electronics presents a comprehensive summary of CSI prediction discussions covering consistency between training and inference, with over 100 observations and proposals from multiple companies. The document focuses on evaluating whether network-side additional conditions like…

As the document moderator, LG Electronics takes a balanced approach by summarizing industry consensus rather than advocating for specific technical solutions. They facilitate the discussion toward concluding that tilt angle has negligible impact on CSI prediction performance,…

R1-2410899 LG Electronics discussion noted 9.1.3

Summary #3 of CSI prediction

This 3GPP RAN1 document (R1-2410899) from LG Electronics summarizes evaluation results for AI/ML based CSI prediction, focusing on consistency between training and inference regarding network-side conditions like antenna tilt angles and TXRU mapping. The document contains numerous observations and conclusions across…

LG Electronics, as the document moderator, presents a comprehensive summary showing mixed industry consensus on CSI prediction consistency issues. They advocate for concluding that antenna tilt angles have negligible impact and don't require additional signaling, while…

R1-2410564 Mavenir discussion not treated 9.1.3

AI/ML for CSI prediction

Mavenir presents evaluation results comparing AI/ML-based CSI prediction using LSTM networks against Kalman filter baselines, demonstrating superior performance across different UE speeds. The document contains 6 proposals and 3 observations focusing on CSI prediction implementation, performance monitoring, and…

Mavenir advocates FOR AI/ML-based CSI prediction using LSTM networks with UE-side implementation, emphasizing the need for enhanced CSI-RS configurations and proper fallback mechanisms. They push FOR leveraging uplink measurement reference signals in TDD mode to reduce CSI-RS…

R1-2410508 MediaTek discussion not treated 9.1.4.1

Additional study on AI/ML for NR air interface - CSI compression

MediaTek's contribution addresses additional study aspects for AI/ML-based CSI compression in NR Release 19, presenting 10 technical proposals covering temporal-domain compression, error tolerance, inter-vendor collaboration approaches, model monitoring techniques, and data collection methods for training AI/ML models.

MediaTek advocates FOR: (1) using temporal-domain CSI compression with separate AI/ML models for prediction and compression rather than unified models, (2) prioritizing AI/ML model transfer over dataset transfer for inter-vendor collaboration, (3) leveraging uplink CSI from SRS…

R1-2410519 MediaTek discussion not treated 9.1.1

Discussion on specification support for AIML-based beam management

MediaTek's technical contribution for RAN1 #119 presents 48 proposals addressing specification support for AI/ML-based beam management in NR, covering associated ID consistency, performance monitoring, Set A/B configuration, training data collection, and responses to RAN2 liaison statements.

MediaTek advocates FOR: (1) flexible associated ID configuration that is not one-to-one mapped with inference configurations to reduce overhead, (2) separate CPU counting for AI/ML vs legacy CSI processing, (3) NW-determined Set B to achieve RS overhead savings, (4) support for…

R1-2410531 MediaTek discussion not treated 9.1.2

Design for AI/ML based positioning

MediaTek's technical document presents 13 proposals across 5 main sections addressing AI/ML-based positioning design for NR air interface, covering sample-based measurement reporting, receiver diversity, LOS/NLOS indicators, training consistency, performance monitoring, and assistance data aspects.

MediaTek advocates FOR implementation flexibility in sample-based measurements by avoiding strict specification of intermediate steps like Nt consecutive samples, and FOR protecting sensitive network deployment information by using implicit signaling through associated IDs…

R1-2410537 MediaTek discussion not treated 9.1.3

AI/ML - Specification support for CSI Prediction

MediaTek's document evaluates AI/ML model generalization for CSI prediction across different network configurations, showing that models can generalize well across tilt angles and TXRU mappings with less than 2% performance loss. The document contains 8 proposals and 4 observations focused on ensuring consistency…

MediaTek advocates FOR demonstrating that AI/ML models for CSI prediction can generalize well across different network configurations (tilt angles and TXRU mappings), supporting the use of associated ID as a baseline approach for training-inference consistency while proposing…

R1-2410347 Meta discussion not treated 9.1.1

AI/ML for Beam Management

Meta presents a comprehensive technical document on AI/ML for beam management in 5G NR, covering configuration aspects, performance monitoring, and beam indication enhancements. The document contains 14 formal proposals addressing both UE-sided and network-sided AI/ML models for beam management use cases.

Meta advocates for flexible AI/ML beam management implementation with simplified configuration approaches (avoiding explicit CSI framework requirements for set A), comprehensive performance monitoring including both Type 1 and Type 2 options with event-driven mechanisms, and…

R1-2409852 NEC discussion not treated 9.1.2

Discussion on specification support for AIML based positioning accuracy enhancement

This NEC contribution provides a comprehensive discussion on AI/ML for NR positioning accuracy enhancement, covering model input/output, training data collection, lifecycle management, and consistency between training and inference. The document contains 42 detailed proposals and 6 observations addressing various…

NEC strongly advocates FOR sample-based measurements over path-based measurements, arguing that sample-based approach provides better standardization control and avoids vendor-specific path detection algorithms. They push FOR supporting both UE-side and network-side AI/ML model…

R1-2409853 NEC discussion not treated 9.1.3

Discussion on specification support for CSI prediction

NEC's contribution discusses specification support for AI/ML-based CSI prediction with UE-sided models, focusing on consistency between training and inference phases and normative work preparation. The document contains 14 proposals and 1 observation covering consistency mechanisms, performance monitoring, model…

NEC advocates for reusing existing associated ID mechanisms from beam management cases for training-inference consistency, emphasizing UE-side performance monitoring (Type 1 and 3) over network-side monitoring to minimize overhead while maintaining control. They push for…

R1-2409854 NEC discussion not treated 9.1.4.1

Discussion on CSI compression

NEC presents their views on AI/ML-based CSI compression for NR air interface, focusing on inter-vendor training collaboration solutions and specification impacts. The document contains 10 proposals and 3 observations addressing training collaboration methods, performance monitoring approaches, and model inference…

NEC strongly advocates FOR Direction A (parameter/model/dataset sharing enabling UE-side offline engineering) as the priority approach for inter-vendor collaboration, arguing it provides better performance and reduces NW complexity compared to other directions. They push AGAINST…

R1-2409855 NEC discussion not treated 9.1.1

Discussion on specification support for beam management

NEC's technical document on AI/ML enhancements for NR beam management presents comprehensive specifications for both UE-sided and NW-sided models across spatial and temporal beam prediction cases. The document contains 39 proposals and 14 observations covering beam management lifecycle management, performance…

NEC advocates FOR UE-centric beam prediction with flexible UE-determined reporting (K-value selection, confidence reporting), comprehensive performance monitoring using RSRP difference metrics, and enhanced TCI framework extensions. They push AGAINST rigid network-only control…

R1-2409856 NEC discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML model and data

NEC's contribution discusses AI/ML model and data aspects for NR air interface, focusing on model identification procedures, model transfer methodologies, and UE capability reporting. The document contains 11 proposals and 10 observations covering model identification for one-sided models, inter-vendor collaboration…

NEC advocates FOR supporting model identification for one-sided models using associated IDs, Alternative B for model transfer methodology z4 (network-initiated signaling approach), RRC-based model parameter transfer with UE readiness indication, and UE failure reporting with…

R1-2409859 NEC discussion not treated 5

Discussion on RAN2 LS on applicable functionality reporting for beam management UE-sided model

This NEC document responds to RAN2 questions about applicable functionality reporting for AI/ML beam management in NR, presenting 13 proposals and 2 observations addressing signaling procedures, network-side conditions, and functionality activation mechanisms.

NEC advocates FOR Option 3 approach where UE initiates parameter alignment reporting rather than network-side configuration in Step 3, supports sub-use case granularity for functionality definitions, and pushes AGAINST providing inference configuration in Step 3. They emphasize…

R1-2409543 New H3C Technologies Co. discussion not treated 9.1.2

Discussion on AI/ML for positioning accuracy enhancement

This document from H3C proposes specific algorithms for eliminating initial phase mismatch in Channel Impulse Response (CIR) measurements used as AI/ML model inputs for NR positioning. It defines reference sample selection criteria (strongest path, first satisfied sample) and reference phase calculation methods (mean,…

H3C supports employing CIR as AI/ML model input for both direct and assistant positioning, arguing it preserves more channel information than PDP or DP. They require the elimination of initial phase mismatch in CIR measurements before AI/ML model input to prevent performance…

R1-2409985 Nokia discussion not treated 9.1.1

AI/ML for Beam Management

Nokia presents a comprehensive technical document with 22 proposals and 7 observations addressing AI/ML for beam management in 5G NR, covering both UE-sided and NW-sided models for spatial (BM-Case1) and temporal (BM-Case2) beam prediction. The document systematically addresses inference operations, performance…

Nokia strongly advocates FOR reusing existing CSI reporting frameworks as the foundation for AI/ML beam management while proposing systematic enhancements for beam prediction capabilities. They push FOR practical parameter values (K=1,2,4 for beams, N=1,2,4 for time instances),…

R1-2409986 Nokia discussion not treated 9.1.2

AI/ML for Positioning Accuracy Enhancement

Nokia's comprehensive technical document on AI/ML for positioning accuracy enhancement presents 51 detailed proposals and 11 observations covering performance monitoring, training-inference consistency, data collection, and inference operations for various AI/ML positioning use cases.

Nokia advocates FOR comprehensive standardization of both label-free and label-based monitoring approaches with strong LMF control over functionality decisions, supporting both path-based and sample-based representations while pushing AGAINST CIR support for inference input due…

R1-2409987 Nokia discussion not treated 9.1.3

AI/ML for CSI Prediction

Nokia presents simulation results demonstrating robustness of their AI/ML CSI predictor against training/inference mismatches and proposes leveraging Release 18 non-AI/ML CSI prediction framework for AI/ML-based CSI prediction. The document contains 5 observations and 1 proposal across sections on generalization…

Nokia advocates FOR reusing existing Release 18 non-AI/ML CSI prediction framework and UE-sided beam management agreements to minimize standardization effort for AI/ML CSI prediction. They demonstrate that their AI/ML architecture design (operating per tap/port without spatial…

R1-2409988 Nokia discussion not treated 9.1.4.1

AI/ML for CSI Compression

Nokia's contribution analyzes AI/ML for CSI compression, focusing on inter-vendor training collaboration across three directions (A, B, C) and proposes deprioritizing Directions B and C while advancing Direction A. The document presents 6 proposals and 6 observations covering dataset transfer optimization, phase…

Nokia strongly advocates FOR Direction A (parameter/dataset sharing for UE-side offline engineering) while pushing AGAINST Directions B and C. They argue Direction B creates unsolvable proprietary information disclosure issues and implementation complexity, while Direction C…

R1-2409989 Nokia discussion not treated 9.1.4.2

Other aspects of AI/ML for two-sided model

Nokia's contribution analyzes model identification and transfer/delivery aspects for two-sided AI/ML CSI compression use cases in 5G NR, presenting 5 observations and 2 proposals on how different inter-vendor collaboration options map to specific model identification procedures.

Nokia advocates FOR prioritizing MI-Option 2 with dataset transfer for inter-vendor collaboration Option 4, supporting flexible model transfer mechanisms that can handle partial parameter updates and non-compiled models requiring UE-side offline engineering. They are pushing FOR…

R1-2410376 NTT DOCOMO discussion not treated 9.1.1

Discussion on AI/ML for beam management

NTT DOCOMO's technical document presents 20 comprehensive proposals for AI/ML-based beam management in 5G NR, covering both UE-sided and network-sided model aspects including configuration, inference, performance monitoring, and data collection optimization.

DOCOMO advocates for a pragmatic, network-centric approach that prioritizes operational efficiency and overhead reduction. They strongly push FOR: reusing existing CSI frameworks, network-controlled decision making (type 1 monitoring only), statistical performance reporting, and…

R1-2410377 NTT DOCOMO discussion not treated 9.1.2

Discussion on AI/ML for positioning accuracy enhancement

NTT DOCOMO's document provides a comprehensive technical analysis of AI/ML for NR positioning accuracy enhancements, covering data collection, model inference, performance monitoring, and lifecycle management aspects. The document presents 16 detailed proposals and 2 observations addressing specification impacts for…

NTT DOCOMO advocates FOR maximum reuse of existing legacy positioning mechanisms and IEs to minimize specification overhead while enabling AI/ML positioning enhancements. They push FOR sample-based measurements as baseline for better consistency between training and inference,…

R1-2410378 NTT DOCOMO discussion not treated 9.1.3

Discussion on AI/ML for CSI prediction

NTT DOCOMO proposes criteria for deciding whether to standardize training/inference consistency mechanisms for AI/ML CSI prediction, based on comparing performance losses to the marginal 0-7.8% UPT gains observed in Release 19 studies. The document contains 2 proposals focused on reusing beam management mechanisms as…

NTT DOCOMO advocates FOR a pragmatic approach to AI/ML CSI prediction standardization, arguing that consistency mechanisms should only be standardized if performance losses are significant relative to the already marginal gains. They push FOR reusing existing beam management…

R1-2410379 NTT DOCOMO discussion not treated 9.1.4.1

Discussion on AI/ML for CSI compression

NTT DOCOMO proposes a combined approach using both Direction A and Direction C for AI/ML-based CSI compression in 5G NR, advocating for eT2-like CSI monitoring schemes and performance optimization methods. The document contains 2 formal proposals and 3 observations addressing inter-vendor collaboration and performance…

NTT DOCOMO advocates FOR a hybrid approach combining Direction A and Direction C for AI/ML CSI compression, emphasizing that Direction C provides essential baseline performance and interoperability while Direction A enables advanced field adaptation. They push FOR eT2-like CSI…

R1-2410380 NTT DOCOMO discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML model and data

NTT DOCOMO presents a comprehensive discussion on AI/ML model identification, data collection, and transfer mechanisms for 5G NR air interface, making 9 proposals and 6 observations covering scenario-specific models, model identification procedures, and standardization approaches.

NTT DOCOMO strongly advocates FOR scenario/site-specific AI/ML models over generalized models, arguing they provide superior performance by learning specific environment tendencies that are difficult to mathematically model. They push FOR hybrid approaches combining multiple…

R1-2410587 NTU discussion not treated 9.1.1

On Associated ID for Beam Management Use Case

NTU proposes a collaborative approach for AI/ML-based beam management to address codebook mismatch issues between UE prediction models and network implementations. The document contains 2 main proposals focusing on relaxing 'similar properties' requirements for associated IDs and enabling dual reporting of predicted…

NTU advocates FOR a collaborative UE-network approach using standardized codebooks (like RAN4 testing codebooks) to reduce UE implementation complexity, opposing the current associated ID approach which requires UEs to support multiple AI/ML models for different vendor…

R1-2409780 NVIDIA discussion not treated 9.1.1

Specification support for AI-enabled beam management

NVIDIA presents a comprehensive framework for AI/ML-enabled beam management in 5G-Advanced, focusing on spatial (BM-Case 1) and temporal (BM-Case 2) downlink beam prediction. The document contains 11 proposals and 1 observation, advocating for specification support for beam set associations, model lifecycle…

NVIDIA proposes specification support for associating Set A of beams with Set B of beams for both spatial (BM-Case 1) and temporal (BM-Case 2) DL beam prediction, establishing the foundational mapping for AI/ML inputs. They require specification support for using L1-RSRP…

R1-2409781 NVIDIA discussion not treated 9.1.2

Specification support for AI-enabled positioning

NVIDIA presents a comprehensive framework for AI/ML-enabled positioning in 5G-Advanced, focusing on specification support for measurements, model lifecycle management, and data consistency. The document contains 1 observation and 9 proposals covering channel measurement reporting, training data generation, and model…

NVIDIA proposes supporting both sample-based and path-based time domain channel measurements, alongside the inclusion of phase information, to enhance AI/ML model inputs for positioning. They require that assistance data for UE-based positioning (Case 1) ensures consistency…

R1-2409782 NVIDIA discussion not treated 9.1.3

Specification support for AI-enabled CSI prediction

NVIDIA presents a technical contribution on specification support for AI-enabled CSI prediction, highlighting four key proposals and two observations regarding inference location, training/inference consistency, and post-deployment monitoring. The document argues for evaluating gNB-side inference alongside UE-side…

NVIDIA proposes that inference for one-sided AI/ML CSI prediction models be evaluated at both the gNB and UE sides to assess comparative gains. They argue that inconsistency between training and inference arises when using stochastic channel models, and therefore propose…

R1-2409783 NVIDIA discussion not treated 9.1.4.1

Additional study on AI-enabled CSI compression

NVIDIA argues that stochastic channel models are insufficient for demonstrating AI/ML CSI compression gains, proposing instead the use of site-specific models trained on data generated via ray tracing in defined reference scenarios. The document outlines six proposals covering performance study methodologies,…

NVIDIA argues that stochastic channel models are insufficient for demonstrating AI/ML CSI compression gains, proposing the consideration of site-specific AI/ML models. They propose defining a common reference scenario with site specificity, selecting between real-scenario maps…

R1-2409784 NVIDIA discussion not treated 9.1.4.2

Additional study on other aspects of AI model and data

NVIDIA argues for the necessity of deterministic, physics-based propagation modeling (specifically ray tracing) for accurate AI/ML data generation in 5G-Advanced and 6G. The document proposes concluding the need for model identification in Lifecycle Management (LCM) for two-sided models, collecting UE-sided training…

NVIDIA argues that deterministic, physics-based modeling, specifically ray tracing, is indispensable for generating accurate training data, presenting a technical case against relying solely on stochastic channel models which lead to poor test accuracy in site-specific…

R1-2410101 OPPO discussion not treated 9.1.1

On specification for AI/ML-based beam management

OPPO's technical document on AI/ML-based beam management for 3GPP RAN1 contains comprehensive responses to RAN2 liaison questions and presents 50+ technical proposals covering NW-side and UE-side model inference, training, monitoring, and consistency mechanisms. The document addresses both BM-Case1 (spatial-domain)…

OPPO strongly advocates for leveraging existing CSI framework for AI/ML beam management configuration and supports UE-side model autonomy through Type 2 performance monitoring where UE makes LCM decisions. They push against requiring UE-side additional conditions for NW-side…

R1-2410102 OPPO discussion not treated 9.1.2

On specification for AI/ML-based positioning accuracy enhancements

OPPO's technical document presents a comprehensive analysis of AI/ML-based positioning accuracy enhancements for NR Release 19, covering five positioning cases (Case 1, 2a, 2b, 3a, 3b) with 26 detailed proposals and 3 observations addressing measurement enhancement, training/inference consistency, data collection,…

OPPO advocates for minimal specification impact by leveraging existing protocols (LPP, NRPPa) and leaving implementation details to vendors for flexibility, while strongly opposing phase information reporting due to limited performance gains versus overhead. They push for…

R1-2410103 OPPO discussion not treated 9.1.3

On specification for AI/ML-based CSI prediction

OPPO proposes solutions for ensuring consistency between training and inference phases in AI/ML-based CSI prediction for NR air interface, addressing both intra-cell and inter-cell scenarios. The document contains 5 proposals and 3 observations focusing on associated ID frameworks and signaling mechanisms.

OPPO advocates FOR adopting the associated ID framework from BM-Case 2 to solve consistency issues in AI/ML CSI prediction, supporting flexible model-to-ID relationships where one model can work with multiple associated IDs. They push FOR further standardization work on…

R1-2410104 OPPO discussion not treated 9.1.4.1

Additional study on AI/ML-based CSI compression

OPPO presents analysis on inter-vendor collaboration approaches for AI/ML-based CSI compression, evaluating multiple directions and options for standardization. The document contains 17 proposals and 5 observations covering reference model standardization, parameter/dataset exchange methods, and data collection…

OPPO advocates FOR deprioritizing Direction C (fully standardized reference models) in Rel-19 due to high standardization effort and compatibility concerns, while pushing FOR prioritizing Case 0 model structure work and encoder-only specification in Direction C. They strongly…

R1-2410105 OPPO discussion not treated 9.1.4.2

Additional study on other aspects of AI/ML model and data

OPPO's study document on AI/ML model identification and data management proposes 9 key technical proposals covering model identification procedures, lifecycle management approaches, and model transfer mechanisms for 3GPP Release 19 NR air interface AI/ML work item.

OPPO advocates FOR a unified lifecycle management (LCM) framework that combines both functionality-based and model ID-based operations, supporting network-assigned model IDs with 1-to-1 mapping between model IDs and associated IDs. They push FOR Alternative B in model transfer…

R1-2410106 OPPO discussion not treated 5

Discussion on LS on applicable functionality reporting for beam management UE-sided model

This OPPO document addresses RAN2's liaison statement on UE-side model inference for beam management, proposing 10 specific answers to technical questions about the 6-step functionality-based lifecycle management procedure. The document contains 11 key proposals focusing on deprioritizing Option 1 for applicability…

OPPO advocates FOR deprioritizing Option 1 in favor of Options 2 and 3 for UE-side model inference applicability, arguing that Option 1's early inference configuration in Step 3 creates unstable RRC behavior requiring withdrawal of configurations after UAI reporting. They push…

R1-2410107 OPPO LS out not treated 5

Draft reply on LS on applicable functionality reporting for beam management UE-sided model

This is a liaison statement from RAN1 (OPPO) responding to RAN2's questions about AI/ML beam management UE-sided model functionality reporting for Release 19. The document provides detailed replies to 10 questions covering UE capability reporting, network-side conditions, inference configuration, and functionality…

OPPO advocates for flexible UE capability reporting with optional associated IDs for network-side conditions, supporting reuse of legacy CSI reporting mechanisms for AI/ML beam management while maintaining UE autonomy in determining applicable functionalities through model…

R1-2410775 OPPO discussion noted 9.1.4.2

Summary #1 for other aspects of AI/ML model and data

This 3GPP RAN1 document (R1-2410775) from OPPO serves as a moderator summary for AI/ML model and data aspects in Rel-19, containing approximately 20+ proposals across model identification, training data collection, and model transfer/delivery. The document consolidates company contributions and proposes agreements for…

OPPO, as the moderator, is advocating FOR a systematic study of model identification mechanisms for two-sided models, particularly supporting MI-Option2 with dataset transfer and standardized model structures for Case z4. They push FOR 3GPP specification of model structures…

R1-2410776 OPPO discussion noted 9.1.4.2

Summary #2 for other aspects of AI/ML model and data

This OPPO-moderated document presents a comprehensive summary of AI/ML model identification, training data collection, and model transfer/delivery discussions for RAN1 #119, containing over 40 proposals across multiple technical areas. The document addresses three main topics: model identification procedures for…

OPPO, as document moderator, advocates for a balanced approach supporting both functionality-based and model ID-based LCM operations while prioritizing standardized model structures over offline vendor collaboration. They push FOR unified LCM frameworks, 3GPP-specified model…

R1-2410777 OPPO discussion revised 9.1.4.2

Summary #3 for other aspects of AI/ML model and data

This OPPO-moderated document (R1-2410777) presents a comprehensive summary for RAN1 agenda item 9.1.4.2 on AI/ML model and data aspects, containing 31 proposals and 15 observations covering model identification procedures, training data collection, and model transfer/delivery mechanisms for Release 19.

OPPO, as the moderator, advocates FOR a unified LCM framework supporting both functionality-based and model ID-based operations, with network assignment of model IDs preferred for consistency. They push FOR studying model transfer Case z4 with standardized structures while being…

R1-2410778 OPPO discussion noted 9.1.4.2

Summary #4 for other aspects of AI/ML model and data

This 3GPP RAN1 technical document from OPPO summarizes discussions on AI/ML model and data aspects for NR air interface, containing approximately 30 proposals across model identification, training data collection, and model transfer/delivery topics. The document focuses on two-sided model scenarios and establishes…

OPPO, as the moderator, is advocating FOR a unified approach to AI/ML model identification that supports both functionality-based and model-ID-based operations, with network-assigned model IDs preferred for consistency. They push FOR standardized reference models (MI-Option4)…

R1-2409957 Panasonic discussion not treated 9.1.1

Discussion on specification support for beam management

Panasonic's technical document for 3GPP RAN1 discusses AI/ML-based beam management for NR air interface, presenting 12 detailed proposals covering UE-side and NW-side model inference, consistency mechanisms, and performance monitoring. The document addresses technical specifications for both spatial-domain (BM-Case1)…

Panasonic strongly advocates FOR Option 2 approach for UE-side model inference applicability (against Options 1 and 3), arguing it provides better optimization and efficiency by avoiding unnecessary CSI-ReportConfig IEs. They push FOR cross-cell applicability of 'associated IDs'…

R1-2410173 Panasonic discussion not treated 9.1.4.2

Discussion on other aspects for AI/ML for air interface

Panasonic proposes 3 key concepts for AI/ML air interface framework: MNO responsibility for associated ID management, consistency scope within MNO boundaries, and dataset association representation through ID-X. The document addresses model identification, data collection, and transfer mechanisms for two-sided AI/ML…

Panasonic advocates FOR MNO-centric management of associated IDs to preserve vendor proprietary information while enabling multi-cell AI/ML model consistency. They push AGAINST vendor-only management of associated IDs and support removing brackets around network proprietary…

R1-2410248 Panasonic discussion not treated 9.1.3

Discussion on consistency of training / inference for AI/ML-based CSI prediction

Panasonic presents their view on ensuring consistency between training and inference for AI/ML-based CSI prediction in NR, focusing on UE-side model development and proposing solutions for handling network-side additional conditions. The document contains 1 observation and 6 proposals addressing data collection,…

Panasonic advocates FOR UE-side only training/development for CSI prediction models and proposes using configuration IDs instead of actual network configurations to preserve proprietary information. They push FOR reusing beam prediction conclusions and implementing…

R1-2410249 Panasonic discussion not treated 9.1.4.1

Discussion on AI/ML for CSI compression

Panasonic's technical document presents their comprehensive view on AI/ML-based CSI compression for 5G NR, containing 5 proposals and 29 observations covering inter-vendor collaboration approaches, temporal domain aspects, and specification impacts. The document advocates for combining Direction A (parameter/dataset…

Panasonic strongly advocates FOR the combination of Direction A (parameter/dataset sharing enabling UE-side offline engineering) with Direction C (fully standardized reference models) as the optimal approach for Release 19, positioning this as more practical than Direction B…

R1-2410463 Qualcomm discussion not treated 5

Discussion for LS reply on applicable functionality reporting

Qualcomm's discussion document addresses RAN2's liaison statement questions about applicable functionality reporting for beam prediction in UE-sided AI/ML models. The document contains no formal proposals but provides detailed technical responses to support the development of beam management procedures.

Qualcomm strongly advocates FOR mandatory associated ID signaling and inference configuration in Step 3, arguing this is the only feasible solution for UE applicability determination. They push AGAINST performance monitoring approaches and immediate activation upon Step 3…

R1-2410466 Qualcomm discussion not treated 9.1.1

Specification support for AI-ML-based beam management

This Qualcomm document presents 19 proposals and 3 observations for AI/ML-based beam management in 5G NR, focusing on ensuring consistency between training and inference phases for UE-side models, performance monitoring mechanisms, and signaling configurations within the CSI framework.

Qualcomm strongly advocates FOR associated ID-based consistency mechanisms over performance monitoring approaches, pushing for joint Set A/Set B configuration within existing CSI framework to minimize specification impact. They argue AGAINST performance monitoring-based…

R1-2410467 Qualcomm discussion not treated 9.1.2

Specification support for AI-ML-based positioning accuracy enhancement

Qualcomm presents a comprehensive technical document for AI/ML-based positioning accuracy enhancement in 5G NR, containing 31 proposals and 34 observations spanning positioning integration, model training/inference consistency, data collection, model input/output aspects, and lifecycle management. The document…

Qualcomm strongly advocates FOR treating AI/ML positioning as enhancements to existing positioning methods (reusing mature procedures and IEs) and FOR explicit provision of assistance data rather than implicit indication approaches. They push AGAINST network-side training for…

R1-2410468 Qualcomm discussion not treated 9.1.3

Specification support for CSI prediction

This Qualcomm document proposes introducing associated IDs to ensure consistency between training and inference phases for AI/ML CSI prediction with UE-side models. The document contains 2 formal proposals and 4 observations focused on addressing how network-side additional conditions affect channel distribution and…

Qualcomm advocates FOR introducing associated IDs to categorize network-side additional conditions for CSI prediction, arguing that cell ID alone is insufficient due to multiple TRPs/RRUs within cells and potential field deployment variations not captured in RAN1 models. They…

R1-2410469 Qualcomm discussion not treated 9.1.4.1

Additional study on CSI compression

Qualcomm's comprehensive technical document on AI/ML-based CSI compression for two-sided models presents 25 proposals and 22 observations covering inter-vendor collaboration, model performance monitoring, inference aspects, and complexity reduction. The document advocates for dataset sharing (option 4-1) over…

Qualcomm strongly advocates FOR dataset sharing (option 4-1) over encoder parameter sharing (3a-1), UE-side performance monitoring using SGCS estimators, and non-over-the-air solutions for model/dataset exchange. They are AGAINST over-the-air delivery of training data as…

R1-2410470 Qualcomm discussion not treated 9.1.4.2

Other aspects of AI/ML model and data

Qualcomm's contribution to RAN1 meeting #119 discusses AI/ML model identification and transfer/delivery aspects for NR air interface. The document contains 2 main proposals focused on concluding current model identification work and deprioritizing certain model transfer cases.

Qualcomm advocates for concluding the current model identification work as sufficient and moving forward, while pushing against pursuing certain model transfer/delivery cases (specifically Case z4 for UE-sided models) due to engineering challenges and UE capability requirements.…

R1-2410719 Qualcomm discussion noted 9.1.4.1

Summary#1 of Additional study on AI/ML for NR air interface: CSI compression

This technical document from Qualcomm serves as a moderator summary for additional study on AI/ML-based CSI compression for NR air interface, containing over 120 proposals from various companies across temporal domain aspects, localized models, inter-vendor training collaboration, and monitoring approaches. The…

Qualcomm, as the moderator, advocates FOR prioritizing temporal domain Cases 2 and 3 with separate prediction and compression as baseline, supporting Direction A (dataset sharing via option 4-1) combined with Direction C for inter-vendor collaboration, and emphasizing…

R1-2410720 Qualcomm discussion noted 9.1.4.1

Summary#2 of Additional study on AI/ML for NR air interface: CSI compression

This 3GPP RAN1 technical document (Tdoc R1-2410720) presents a draft summary of AI/ML-based CSI compression studies, containing over 300 proposals from various companies covering temporal domain aspects, inter-vendor training collaboration, monitoring, and inference aspects. The document focuses on addressing UCI loss…

Qualcomm, as the document moderator, advocates for prioritizing Direction A Option 4-1 (dataset exchange) for inter-vendor collaboration while supporting Direction C for minimum performance assurance. They push for dataset sharing over parameter sharing due to lower…

R1-2410721 Qualcomm discussion noted 9.1.4.1

Summary#3 of Additional study on AI/ML for NR air interface: CSI compression

This 3GPP RAN1 technical document (R1-2410721) from Qualcomm serves as the draft summary for AI/ML-based CSI compression study, containing over 100 company proposals across temporal domain aspects, inter-vendor collaboration, monitoring, and inference aspects. The document focuses on addressing UCI loss mitigation,…

Qualcomm, as the moderator, advocates FOR: (1) Prioritizing Direction A Option 4-1 (dataset exchange) over Option 3a-1 due to lower specification complexity while maintaining performance, (2) Supporting Direction C for minimum performance guarantee using RAN4-compatible models,…

R1-2410722 Qualcomm discussion noted 9.1.4.1

Summary#4 of Additional study on AI/ML for NR air interface: CSI compression

This 3GPP RAN1 technical document (Tdoc R1-2410722) from Qualcomm presents a comprehensive draft summary on AI/ML for NR air interface CSI compression, containing approximately 120+ proposals across four major sections covering temporal domain aspects, localized models, inter-vendor training collaboration, and…

Qualcomm, as the moderator, advocates FOR prioritizing Case 2 and Case 3 temporal domain aspects with separate prediction and compression as baseline, supporting Direction A Option 4-1 (dataset exchange) over Option 3a-1 (parameter exchange) due to lower specification impact,…

R1-2410723 Qualcomm discussion noted 9.1.4.1

Summary#5 of Additional study on AI/ML for NR air interface: CSI compression

This 3GPP RAN1 document from Qualcomm presents a comprehensive summary of AI/ML-based CSI compression study results with over 100 proposals from multiple companies covering temporal domain aspects, inter-vendor training collaboration, monitoring, and data collection. The document captures extensive evaluation results…

Qualcomm, as the moderator, advocates for a balanced approach prioritizing Direction A Option 4-1 (dataset exchange) for inter-vendor collaboration while maintaining Direction C (fully specified models) as a minimum performance baseline. They push against premature…

R1-2410724 Qualcomm discussion noted 9.1.4.1

Final summary of Additional study on AI/ML for NR air interface: CSI compression

This 3GPP RAN1 document (R1-2410724) from Qualcomm serves as the final meeting summary for AI/ML-based CSI compression studies in Release 19, containing over 200 proposals from multiple companies across temporal domain aspects, inter-vendor collaboration, monitoring, and data collection. The document concludes the…

Qualcomm, as the document moderator, advocates for a pragmatic approach prioritizing Direction A Option 4-1 (dataset sharing) over Option 3a-1 due to lower specification complexity, while supporting Direction C as a minimum performance baseline. They push for NW-side target CSI…

R1-2410725 Qualcomm discussion noted 9.1.4.1

Updated summary of Evaluation Results for AI/ML CSI compression

This document from Qualcomm presents comprehensive evaluation results for AI/ML based CSI compression in NR air interface, containing 16 main observations across different test cases covering SGCS performance, FTP traffic, full buffer scenarios, CSI feedback reduction, and localized models with varying complexity…

Qualcomm, as the moderator, is advocating for comprehensive standardization of AI/ML based CSI compression techniques across multiple scenarios and configurations. They are pushing FOR detailed performance characterization across different payload sizes, traffic patterns, and…

R1-2410915 Qualcomm LS out revised 9.1.4.1

[Draft] LS on signalling feasibility of dataset and parameter sharing

This liaison statement from Qualcomm/RAN1 to RAN2 requests feedback on standardized signaling feasibility for AI/ML-based CSI compression inter-vendor collaboration, specifically for sharing datasets and model parameters between network-side and UE-side entities. The document contains multiple company comments but no…

Qualcomm/RAN1 is advocating FOR standardized signaling mechanisms to enable inter-vendor collaboration in AI/ML-based CSI compression through dataset and parameter sharing between network-side and UE-side entities. They are pushing FOR feasibility assessment of three specific…

R1-2410922 Qualcomm LS out approved 9.1.4.1

LS on signalling feasibility of dataset and parameter sharing

This is a liaison statement from RAN1 to RAN2 requesting feedback on the feasibility of standardized signaling for sharing AI/ML model parameters and datasets between network and UE sides for two-sided CSI compression. The document contains only one explicit observation and no formal proposals, focusing instead on…

Qualcomm advocates FOR standardized signaling approaches to enable inter-vendor collaboration in AI/ML-based CSI compression, specifically supporting Options 3a-1 and 4-1 for parameter and dataset sharing with offline UE-side engineering. They are pushing FOR feasibility…

R1-2409447 Quectel discussion not treated 9.1.1

Discussion on AI/ML for Beam Management

Quectel presents 15 proposals regarding AI/ML-based beam management in NR, focusing on data collection frameworks for network and UE-side models, inference reporting configurations, and performance monitoring mechanisms. The document addresses specification impacts for Beam Management Case 1 (BM-Case1) and Case 2…

Quectel proposes extending UE capability for RS measurement beyond 64 RS per resource set to support larger Set A configurations, utilizing bitmaps for Set B pattern selection. They support configuring two resource sets for Set A and Set B separately within a single…

R1-2409840 Ruijie Networks Co. Ltd discussion not treated 9.1.1

Discussion on specification support for beam management

This 3GPP RAN1 document from Ruijie Networks presents 5 technical proposals for AI/ML-based beam management in NR air interface, focusing on UE-side model configuration, timing references, associated ID management, performance monitoring, and applicability reporting procedures.

Ruijie Networks advocates FOR flexible timing reference mechanisms based on Set B measurements rather than uplink slots, separate associated ID configuration for Set A/B to handle different beam scenarios, dedicated monitoring configurations to avoid mixing inference and…

R1-2409841 Ruijie Networks Co. Ltd discussion not treated 9.1.2

Discussion on specification support for positioning accuracy enhancement

Ruijie Networks presents 4 proposals for AI/ML positioning accuracy enhancement in NR air interface, focusing on sample-based measurements, model output reporting, training data collection, and model monitoring for various AI/ML positioning cases.

Ruijie Networks advocates for maximizing reuse of existing information elements and frameworks rather than creating new ones, supporting flexible options for sample-based measurements that preserve potentially useful information before first detected path, and establishing…

R1-2409581 Samsung discussion not treated 9.1.1

Discussion for supporting AI/ML based beam management

Samsung presents 30 proposals and 1 observation regarding AI/ML-based beam management for NR, covering both NW-side and UE-side models. The document addresses data collection for training and inference, spatial and temporal enhancements for beam reporting, consistency mechanisms via DL Tx IDs and associated IDs,…

Samsung proposes specific data collection contents for NW-side training, including L1-RSRPs for Set A and Set B and timestamps, conveyed via high-layer signaling. For UE-side inference, they support configurability between Alt 1 and Alt 3 for CSI-ReportConfig and introduce DL Tx…

R1-2409582 Samsung discussion not treated 9.1.2

Discussion for supporting AI/ML based positioning accuracy enhancement

Samsung presents a comprehensive discussion on AI/ML-based positioning accuracy enhancement, outlining 29 observations across triggering, model selection, data collection, inference, monitoring, and consistency checks. The document emphasizes the need for processed channel measurements rather than raw data, defines…

Samsung argues that full-size raw channel measurements are unsuitable for data collection due to prohibitive overhead and storage costs, proposing instead that truncated or feature-extracted measurements be used. They support explicit signaling mechanisms where UEs can notify…

R1-2409583 Samsung discussion not treated 9.1.3

Views on AI/ML based CSI prediction

Samsung presents observations and proposals for AI/ML-based CSI prediction in NR, highlighting the performance degradation when models are trained on mismatched TRP antenna settings and the need for network assistance to ensure training/inference consistency. The document contains 2 key observations regarding…

Samsung argues that site-specific AI/ML models significantly outperform generic models, particularly at large prediction horizons, and demonstrates that data distribution mismatch due to different TRP antenna settings (e.g., [2,8,2] vs [4,4,2] TXRU mapping) causes up to 77%…

R1-2409584 Samsung discussion not treated 9.1.4.1

Views on additional study for AI/ML based CSI compression

Samsung presents views on further studies for AI/ML-based CSI compression in Rel-19, focusing on temporal aspects (Case 2 and Case 3), performance-complexity trade-offs, and inter-vendor training collaboration. The document contains 18 proposals and 16 observations, arguing that angle-delay (W2) domain compression…

Samsung proposes that angle-delay (W2) domain compression is significantly more robust to data distribution mismatches than spatial-frequency (W) domain compression, citing up to 37.9% degradation for W-domain versus only 0.7% for W2-domain when mixing deployment scenarios. They…

R1-2409585 Samsung discussion not treated 9.1.4.2

Views on additional study for other aspects of AI/ML model and data

Samsung analyzes model identification and data handling for AI/ML in NR, presenting 13 proposals and 5 observations across model-level management, two-sided model consistency, and data privacy. The document argues that explicit model identification is unnecessary for ensuring training-inference consistency, proposing…

Samsung argues that explicit model identification is not required to ensure consistency between model training and inference, proposing instead that the indication of associated IDs for network-side additional conditions is sufficient. They propose studying MI-Option1 for…

R1-2410733 Samsung discussion noted 9.1.1

FL summary #0 for AI/ML in beam management

This Samsung-moderated 3GPP RAN1 document (Tdoc R1-2410733) presents a comprehensive summary of AI/ML beam management contributions from meeting #118, containing over 100 proposals across multiple technical areas. The document covers UE-side and network-side models for both spatial (BM-Case1) and temporal (BM-Case2)…

Samsung advocates for a flexible framework combining multiple options for UE-side model applicability (merging Options 1 and 2), supports dedicated monitoring configurations separate from inference configurations, and pushes for practical beam accuracy indicators with multiple…

R1-2410734 Samsung discussion noted 9.1.1

FL summary #1 for AI/ML in beam management

This Samsung-moderated FL summary document for RAN1#119 contains over 250 proposals and observations across 9 main sections covering AI/ML beam management, including RAN2 LS handling, performance monitoring, configuration aspects, and inference reporting. The document addresses both UE-sided and NW-sided AI/ML models…

Samsung advocates for a flexible approach supporting multiple options for UE-sided model applicability reporting, favoring dedicated monitoring configurations separate from inference configurations, and supporting enhanced quantization steps for overhead reduction. They push for…

R1-2410735 Samsung discussion noted 9.1.1

FL summary #2 for AI/ML in beam management

This 3GPP RAN1 technical document (Tdoc R1-2410735) from Samsung serves as FL summary #2 for AI/ML in beam management, containing over 200 proposals and observations across multiple technical areas including performance monitoring, configuration methods, inference reporting, and beam indication mechanisms.

Samsung advocates for a comprehensive framework supporting both UE-side and NW-side AI/ML models with emphasis on: 1) Flexible applicability reporting combining multiple options rather than forcing single approach, 2) Dedicated monitoring configurations separate from inference…

R1-2410736 Samsung discussion noted 9.1.1

FL summary #3 for AI/ML in beam management

This document is Samsung's FL summary #3 for AI/ML in beam management from RAN1 #119, containing over 40 proposals and observations covering UE-side and NW-side model configurations, performance monitoring, data collection, inference reporting, and beam indication frameworks.

Samsung, as the document moderator, advocates for: (1) merging applicability options to give RAN2 flexibility in container design while supporting both CSI-ReportConfig and inference parameter approaches, (2) comprehensive performance monitoring with beam accuracy indicators and…

R1-2410737 Samsung discussion noted 9.1.1

FL summary #4 for AI/ML in beam management

This is Samsung's summary document (R1-2410737) for AI/ML in beam management from RAN1 #119 meeting, containing over 200 proposals and observations from multiple companies covering UE-side and NW-side model configurations, performance monitoring, data collection, and beam indication frameworks.

Samsung advocates for practical AI/ML beam management implementations that reuse existing CSI frameworks while introducing minimal spec impact. They push FOR: reusing legacy CSI-ReportConfig structures, supporting both UE-side and NW-side models with clear associated ID…

R1-2410892 Samsung discussion noted 9.1.1

FL summary #5 for AI/ML in beam management

This is Samsung's FL summary #5 for AI/ML in beam management (Tdoc R1-2410892), containing over 100 proposals addressing UE-sided and NW-sided models, performance monitoring, configuration frameworks, and beam indication mechanisms across multiple technical issues.

Samsung, as the moderator, advocates for a comprehensive AI/ML beam management framework supporting both UE-sided and NW-sided models with practical implementation considerations. They push for flexible configuration options (supporting multiple alternatives rather than…

R1-2410893 Samsung LS out revised 9.1.1

[DRAFT] Reply LS on applicable functionality reporting for beam management UE-sided model

This is a liaison statement from RAN1 to RAN2 responding to questions about beam management UE-sided AI/ML model functionality reporting. The document provides answers to 9 detailed questions and includes 2 agreements and 1 conclusion about the configuration and activation procedures for AI/ML-enabled beam management…

Samsung/RAN1 advocates for a flexible AI/ML functionality reporting framework that supports both CSI-ReportConfig-based inference configuration and separate inference parameter sets for applicability reporting. They push for optional network-side additional conditions and…

R1-2410898 Samsung LS out approved 9.1.1

Reply LS on applicable functionality reporting for beam management UE-sided model

Samsung's response to RAN2's liaison statement regarding applicable functionality reporting for beam management UE-sided AI/ML models, containing 4 key observations about terminology definitions and 3 agreements/conclusions on configuration procedures and CSI reporting mechanisms.

Samsung advocates FOR a flexible CSI framework-based approach to AI/ML functionality management with clear separation between supported, applicable, and activated functionalities. They push FOR allowing multiple CSI reports for inference based on UE capability and support…

R1-2410359 Sharp discussion not treated 9.1.1

Discussions on specification support for beam management

Sharp's contribution discusses comprehensive specification support for AI/ML beam management in NR Release 19, covering both UE-side and network-side models for spatial (BM-Case 1) and temporal (BM-Case 2) beam prediction. The document contains 23 technical proposals and 1 observation across performance monitoring,…

Sharp advocates for comprehensive AI/ML beam management support using existing CSI frameworks as baseline while proposing specific enhancements: they strongly support Alternative 3 (two separate CSI-ResourceConfigIds for Set A and Set B) over other alternatives, favor beam…

R1-2410360 Sharp discussion not treated 9.1.2

Discussion on specification support for AI/ML based positioning accuracy enhancements

Sharp's technical document discusses specification support for AI/ML based positioning accuracy enhancements, covering sample-based measurements, model input/output, consistency between training and inference, and performance monitoring. The document contains 9 proposals and 10 observations addressing various aspects…

Sharp advocates for flexible, implementation-friendly approaches to AI/ML positioning. They push FOR supporting multiple options simultaneously rather than down-selecting (Options A, C, D can co-exist), continuous phase information over single-phase, and Option 1-1…

R1-2410426 Sharp discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML model and data

Sharp discusses model identification procedures for AI/ML in NR air interface, focusing on two-sided models with 3 observations and 2 proposals. The document analyzes MI-Option2 (model identification with data transfer) and MI-Option3 (model identification with model transfer).

Sharp advocates FOR flexible use of ID-X as a unified model identifier that can represent both network and UE parts of two-sided models in MI-Option2, and FOR mandatory model IDs in MI-Option3. They push FOR simplifying the model identification procedure by potentially…

R1-2410216 Sony discussion not treated 9.1.1

Discussion on specification support for beam management

Sony's document presents 18 proposals addressing AI/ML-based beam management for 5G NR, focusing on model inference procedures, time synchronization requirements, and performance monitoring mechanisms for both UE-side and network-side AI models.

Sony advocates FOR explicit time-related information reporting in BM-Case 2, dynamic adjustment of time windows for both data collection and prediction instances, probability-based reporting for Top K beam predictions, and event-triggered monitoring with measurement overhead…

R1-2410217 Sony discussion not treated 9.1.2

Support for AI/ML for positioning accuracy enhancement

Sony's contribution presents a comprehensive framework for AI/ML-enhanced positioning in NR, covering the entire AI/ML lifecycle from data collection to model deployment and monitoring. The document contains 15 detailed proposals addressing key aspects including CIR-based data collection, model transfer mechanisms,…

Sony strongly advocates for sample-based time domain channel measurements over path-based measurements for AI/ML model input, emphasizing information richness for positioning accuracy. They push for comprehensive CIR reporting mechanisms and advocate for robust consistency…

R1-2410218 Sony discussion not treated 9.1.3

Further views on consistency issues in CSI prediction

Sony's contribution addresses consistency issues between AI/ML model training and inference for CSI prediction in NR systems, presenting 2 specific proposals for RAN1 study. The document identifies how differences in UE capabilities, network configurations, and model functionality configurations can cause performance…

Sony advocates FOR comprehensive study of training-inference consistency issues in AI/ML CSI prediction models, pushing for systematic characterization and resolution of inconsistencies caused by UE capability differences, network configuration mismatches, and functionality…

R1-2409625 Spreadtrum discussion not treated 9.1.1

Discussion on AIML for beam management

Spreadtrum presents 15 proposals and 6 observations regarding AI/ML for NR Beam Management, focusing on UE-side and NW-side model configurations, inference reporting, and performance monitoring. The document argues for specific signaling frameworks to ensure consistency between training and inference, particularly…

Spreadtrum supports Option 1 or Option 2 for UE functionality determination, requiring the associated ID to be configured for both training and inference to guarantee consistency. They oppose configuring only Set B resources for UE-side inference and argue that larger…

R1-2409626 Spreadtrum discussion not treated 9.1.3

Discussion on AIML for CSI prediction

Spreadtrum discusses the consistency of training and inference for UE-sided CSI prediction models, proposing the reuse of the 'associated ID' mechanism from Beam Management to ensure network-side conditions remain consistent. The document contains two main proposals: using the associated ID to guarantee consistency…

Spreadtrum argues that the 'associated ID' mechanism, previously introduced for Beam Management (AI-BM), should be reused for CSI prediction to ensure consistency of network-side additional conditions across training and inference. They present a technical case against…

R1-2409627 Spreadtrum discussion not treated 9.1.4.1

Discussion on AIML for CSI compression

Spreadtrum presents evaluation results for AI-based CSI Spatial-Temporal-Frequency (S-T-F) compression, demonstrating superior SGCS and UPT performance over Rel-16 and Rel-18 baselines. The document contains 8 proposals and 7 observations addressing inter-vendor training collaboration directions, CQI determination,…

Spreadtrum supports extending CSI compression to Spatial-Temporal-Frequency (S-T-F) domains, presenting technical evidence that S-T-F compression yields higher SGCS and UPT gains than S-F compression or Rel-16/18 codebooks. They propose using SGCS as the primary performance…

R1-2409628 Spreadtrum discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML model and data

Spreadtrum presents three proposals and four observations regarding AI/ML for the NR air interface in Rel-19, focusing on data collection, model transfer, and identification. The document argues that RAN1 should deprioritize certain model transfer cases and exclude specific model identification options for two-sided…

Spreadtrum proposes that RAN1 should not discuss data collection for UE-side model training, preferring mechanism 1a or waiting for RAN2 progress due to privacy concerns regarding network exposure. They require that model transfer/delivery Case z1 be deprioritized in Rel-19,…

R1-2410202 TCL discussion not treated 9.1.4.1

Discussion on AIML CSI compression

TCL presents a comprehensive technical contribution on AI/ML CSI compression for NR air interface, addressing resource configuration, priority rules, UE capabilities, collaborative training, and overhead reduction. The document contains 7 proposals and 5 observations covering specification impacts for CSI compression…

TCL advocates FOR practical AI/ML CSI compression solutions that maintain compatibility with legacy CSI frameworks while introducing necessary AI/ML-specific enhancements. They push AGAINST overly complex collaborative training options (Options 1-2) that require full model…

R1-2410203 TCL discussion not treated 9.1.4.2

Discussions on other aspects of AlML In NR air interface

TCL presents a discussion on AI/ML framework aspects for NR air interface, focusing on model identification and additional conditions management. The document contains 3 observations and 3 proposals addressing ID hierarchical relationships, functional overlaps, and overhead reduction strategies.

TCL advocates FOR a hierarchical ID structure where model IDs represent generic model types rather than specific trained models to reduce overhead, and FOR eliminating duplicate ID definitions between associated ID and dataset ID. They push AGAINST maintaining large numbers of…

R1-2410204 TCL discussion not treated 9.1.1

Discussion on AIML beam management

TCL proposes comprehensive enhancements to 3GPP NR beam management by integrating AI/ML techniques to simplify conventional P1/P2/P3 beam pairing processes and improve beam failure detection/recovery procedures. The document contains 10 proposals and 4 observations covering beam prediction, reference signal…

TCL advocates for a comprehensive AI/ML integration into NR beam management that goes beyond basic beam prediction to include unified BFD/BFR frameworks, enhanced TCI signaling, and sophisticated reporting mechanisms. They push for significant protocol changes including new TCI…

R1-2410215 TCL discussion not treated 9.1.2

Discussion on specification support for positioning accuracy enhancement

TCL presents their position on AI/ML based positioning for NR air interface, covering performance monitoring, training data collection, and consistency between training and inference. The document contains 11 proposals and 2 observations addressing various aspects of AI/ML positioning implementation.

TCL advocates FOR Option A-2 for label-based model monitoring to reduce data transfer overhead, supports AI-specific reference signal configurations with hierarchical resource type structures, and promotes explicit indication of critical assistance data IEs for consistency. They…

R1-2409749 Tejas Networks Limited discussion not treated 9.1.1

Specification support for beam management

Tejas Networks Limited submits 36 proposals and 6 observations to address open issues for AI/ML-based beam management in NR, covering both UE-sided and NW-sided models. The document focuses on ensuring consistency between training and inference via Associated IDs, optimizing reporting overhead through differential…

Tejas Networks proposes that the Associated ID be configured within the CSI-Report Config to ensure consistency between training and inference phases for UE-sided models, mapping to both Set A and Set B resources. They support using L1 signaling for Type 1 NW-side performance…

R1-2409750 Tejas Networks Limited discussion not treated 9.1.2

Specification support for positioning accuracy enhancement

Tejas Networks Limited submits a contribution discussing AI/ML for positioning accuracy enhancement, focusing on model input definitions, training data collection, and model performance monitoring. The document contains 23 proposals and 15 observations addressing sample-based and path-based measurements, LoS/NLoS…

Tejas Networks proposes redefining model outputs for AI/ML positioning by reporting the timing of arrival of the LoS path whether real or virtually inferred, and introducing a new LoS/NLoS indication that characterizes the reported measurement itself rather than the link…

R1-2409751 Tejas Networks Limited discussion not treated 9.1.3

Discussion on study for AI/ML CSI prediction

Tejas Networks discusses AI/ML for CSI prediction in Rel-19, focusing on ensuring consistency between training and inference, Life Cycle Management (LCM) modes, data collection mechanisms, and performance monitoring strategies. The document presents 14 proposals and 3 observations addressing issues such as…

Tejas Networks proposes using an associated ID to align training and inference conditions, mitigating performance degradation from interference variations and TxRU mapping differences. They support AI/ML model identification in LCM mode, where the Network identifies models by…

R1-2409752 Tejas Networks Limited discussion not treated 9.1.4.1

Discussion on AI/ML for CSI Compression

Tejas Networks Limited presents 11 proposals and 5 observations regarding AI/ML-based CSI compression for NR Release 19, focusing on evaluation methodologies, inter-vendor collaboration, and monitoring frameworks. The document proposes specific assumptions for CSI-Net model architecture, defines scenarios for handling…

Tejas Networks proposes specific architectural assumptions for CSI-Net, including a convolutional encoder with a 33 kernel size and RefineNet-based decoder, to standardize evaluation metrics for temporal domain CSI compression. They require the use of Model IDs to identify…

R1-2409753 Tejas Networks Limited discussion not treated 9.1.4.2

Discussion on Other aspects of AI/ML model and data

Tejas Networks discusses model identification and data handling for AI/ML in NR, focusing on the consistency of NW-side additional conditions via 'associated IDs' and the mapping between these IDs, datasets, and model IDs. The document contains 7 observations and 5 proposals covering training/inference consistency, ID…

Tejas Networks proposes that use-case specific data collection configurations and site-specific information be mapped to associated IDs to ensure consistency between training and inference. They claim the UE should assign model IDs and report them to the NW, rather than the NW…

R1-2409960 Transsion Holdings discussion not treated 9.1.1

Discussion on specification support for AI/ML beam management

Transsion Holdings presents 13 proposals for AI/ML beam management enhancements in 5G NR, covering data collection, model inference, and performance monitoring for both network-side and UE-side AI/ML models. The document addresses specification impacts across training data collection, inference procedures, and…

Transsion advocates for UE-centric approaches including UE-initiated data collection rather than network-controlled collection, supports separate resource set configurations (Set A and Set B) for better flexibility, and pushes for reduced signaling overhead through larger…

R1-2409659 vivo LS out not treated 5

Draft reply LS on applicable functionality reporting for beam management UE-sided model

This document is a Liaison Statement from RAN1 to RAN2 regarding the applicable functionality reporting for UE-sided AI/ML models in beam management for Release 19. It provides technical answers to ten specific questions from RAN2 concerning the role of network-side additional conditions, associated IDs, and inference…

vivo argues that it is not feasible for the UE to determine applicable functionalities without NW-side providing associated IDs first, emphasizing that ensuring consistency of NW-side additional conditions across training and inference is crucial to avoid unpredictable…

R1-2409660 vivo discussion not treated 5

Discussion on LS on applicable functionality reporting for beam management UE-sided model

vivo analyzes three options for applicable functionality reporting for AI/ML beam management UE-sided models, arguing that Option 1 is inefficient due to signaling waste and CSI framework conflicts. The document prioritizes Option 2 and Option 3, which decouple associated ID interaction from inference configuration to…

vivo presents a technical case against Option 1, arguing it leads to significant wastage of NW signaling, communication resources, and UE storage resources due to invalid inference configurations. They oppose Option 1 because it contradicts the existing CSI framework, where…

R1-2409668 vivo discussion not treated 9.1.1

Specification support for beam management

This document from vivo addresses specification support for AI/ML-based beam management in NR, focusing on consistency issues, performance monitoring, and reporting overhead reduction. It contains 53 proposals and 8 observations covering UE-side and NW-side models, associated ID interactions, Set A/B configurations,…

Vivo argues that ensuring consistency of Set A and Set B properties (beam width, pointing angle, indexing) via an associated ID is critical to prevent severe performance degradation, prioritizing local associated IDs to protect NW proprietary information. They oppose Option 1…

R1-2409669 vivo discussion not treated 9.1.2

Specification support for positioning accuracy enhancement

This document from vivo analyzes specification impacts for AI/ML-based positioning in NR, focusing on data collection, model inference, and consistency between training and inference. It presents simulation results demonstrating the superiority of sample-based measurements over path-based measurements and proposes…

vivo argues that sample-based measurements significantly outperform path-based measurements for AI/ML positioning, particularly in scenarios with limited TRPs or bandwidth, and proposes specifying sample-based reporting with defined parameters (Nt, Nt’, k). They prefer reusing…

R1-2409670 vivo discussion revised 9.1.3

Study on consistency issue for CSI prediction

This document analyzes the impact of TXRU virtualization mapping mismatches on CSI prediction generalization performance, identifying significant SGCS losses under high-speed and outdoor user conditions. It contains 3 observations regarding performance degradation and 2 proposals, one identifying TXRU mapping as a…

vivo identifies TXRU virtualization mapping as a critical NW-side additional condition that significantly impacts the generalization performance of UE-sided CSI prediction models. They present simulation evidence showing that mismatches in TXRU mapping between training and…

R1-2409671 vivo discussion not treated 9.1.4.1

Discussion on CSI compression

This document from vivo analyzes inter-vendor training collaboration for AI/ML-based CSI compression in NR, covering Directions A (UE-side offline engineering), B (NW-side encoder sharing), and C (standardized reference models). It presents 25 proposals and 17 observations, arguing that Direction A requires careful…

vivo argues that for Direction A, exchanging target CSI from NW to UE is unnecessary for option 3a-1 if the NW includes mixed UE data in training, but proprietary concerns regarding decoder structure disclosure for Alt 1 training must be studied. They propose addressing data…

R1-2409672 vivo discussion not treated 9.1.4.2

Other aspects of AI/ML model and data

This document from vivo analyzes model identification and transfer mechanisms for NR AI/ML, specifically focusing on the feasibility of Case z4 (known model structure transfer). It presents 22 proposals and 7 observations covering associated ID scopes, reference model standardization, hyper-parameter specifications…

vivo proposes supporting local associated IDs for multiple cells to balance training consistency with network privacy, rather than using global IDs that expose deployment choices. They conclude that Case z4 (known model structure transfer) is feasible from a device…

R1-2410673 vivo discussion not treated 9.1.3

Study on consistency issue for CSI prediction

Vivo presents a study on training-inference consistency issues for AI/ML-based CSI prediction in NR, analyzing how TXRU virtualization mapping mismatches cause significant performance degradation (up to 44.4% loss). The document contains 2 proposals and 5 observations addressing generalization performance impacts.

Vivo advocates FOR addressing training-inference consistency issues in AI/ML CSI prediction by adopting associated ID solutions from beam management use cases, and FOR recognizing TXRU mapping as a critical network-side condition. They push AGAINST ignoring the significant…

R1-2409877 Xiaomi discussion not treated 9.1.1

Discussion on AI/ML for beam management

This Xiaomi contribution discusses specification impacts of AI/ML for beam management in NR, covering both spatial (BM Case 1) and temporal (BM Case 2) beam prediction for UE-side and NW-side models. The document contains 47 proposals and 1 observation across functionality identification, data collection, signaling…

Xiaomi advocates FOR comprehensive AI/ML beam management support with both spatial and temporal prediction capabilities, flexible functionality identification based on beam set relationships, optional NW-side additional conditions, and dual benchmark approaches for performance…

R1-2409878 Xiaomi discussion not treated 9.1.2

Discussion on AI/ML-based positioning accuracy enhancement

Xiaomi presents a comprehensive discussion on AI/ML-based positioning accuracy enhancement for NR, covering input types, data collection, functionality identification, inference, and performance monitoring across different positioning cases. The document contains 28 proposals and 2 observations addressing key aspects…

Xiaomi strongly advocates FOR sample-based input over path-based input for Cases 2b and 3b due to better positioning accuracy and reduced ambiguity issues, while pushing AGAINST CIR support due to high overhead with marginal performance gains. They favor UE-side performance…

R1-2409879 Xiaomi discussion not treated 9.1.3

Discussion on AI/ML model based CSI prediction

Xiaomi proposes methods to ensure consistency between AI/ML model training and inference for UE-side CSI prediction, presenting simulation results showing that models trained with consistent TXRU mapping configurations achieve significantly better performance. The document contains 2 proposals and 1 observation…

Xiaomi advocates FOR explicit network indication of additional conditions (like TXRU mapping) to UEs to maintain AI/ML model performance consistency, and FOR leveraging beam prediction consistency methods as a foundation for CSI prediction. They push FOR high-priority study of…

R1-2409880 Xiaomi discussion not treated 9.1.4.1

Views on AI/ML model based CSI compression

Xiaomi presents comprehensive views on AI/ML-based CSI compression feedback for NR, covering performance evaluation of temporal domain compression (Case 2), inter-vendor collaboration options, performance monitoring, and specification impacts. The document contains 19 proposals and 14 observations addressing key…

Xiaomi strongly advocates for AI-TSF (temporal-spatial-frequency) compression over traditional AI-SF compression, demonstrating significant performance gains (8.9-13.71% over eType II codebook). They push for practical inter-vendor collaboration solutions, favoring over-the-air…

R1-2409881 Xiaomi discussion not treated 9.1.4.2

Further study on AI/ML model and data

Xiaomi proposes technical solutions for AI/ML model delivery/transfer and identification in NR air interfaces, arguing for standardized solutions particularly for network-side trained models. The document contains 13 proposals and 10 observations focused on Case z4 model transfer and various model identification…

Xiaomi advocates FOR network-side AI model training and standardized model transfer/delivery solutions, particularly Case z4 with specified known model structures. They push AGAINST over-the-air dataset delivery and argue for cell-group unique IDs as a compromise between UE…

R1-2409478 ZTE discussion not treated 5

Discussion and reply LS on applicable functionality reporting for beam management UE-sided model

ZTE analyzes the applicable functionality reporting procedures for UE-sided AI/ML beam management models, specifically addressing questions from a RAN2 Letter of Agreement. The document presents 15 distinct proposals across four sections, arguing against Option 1 for inference applicability due to resource waste and…

ZTE opposes Option 1 for UE-side model inference applicability, arguing it causes substantial waste of air interface resources and complicates NW scheduling. They support Option 3, emphasizing its compatibility with legacy CSI-ReportConfig design. ZTE proposes that AI/ML beam…

R1-2409479 ZTE discussion not treated 9.1.1

Discussion on AI/ML-based beam management

ZTE proposes functionality-based LCM without model ID signaling for AI/ML beam management, emphasizing overhead reduction through bitmap-based beam reporting and threshold-based data omission. The document outlines specific enhancements for NW-side data collection, UE-side inference reporting, and performance…

ZTE proposes utilizing functionality-based LCM without model ID based signaling for AI/ML beam management, arguing that model transfer challenges and existing associated ID support diminish the need for model-ID-based approaches. They support bitmap-based methods for beam…

R1-2409480 ZTE discussion not treated 9.1.2

Discussion on AI/ML-based positioning enhancement

ZTE presents a comprehensive contribution on AI/ML-based positioning enhancements for Rel-19, containing 30 proposals and 11 observations across model input, output, training, and monitoring. The document strongly favors sample-based measurements over path-based ones due to reduced implementation ambiguity and argues…

ZTE proposes supporting sample-based measurements for Rel-19 AI/ML positioning, arguing that implementation ambiguities in path-based measurements cannot be removed, whereas sample-based ambiguities can be resolved via LMF configuration. They require the starting point of Nt…

R1-2409481 ZTE discussion not treated 9.1.3

Discussion on specification support for AI CSI prediction

ZTE presents simulation results evaluating the generalization capability of AI-based CSI prediction models across different down tilt angles and TXRU mappings. The document contains two observations regarding model performance and one proposal concluding that neither down tilt angle nor TXRU mapping should be treated…

ZTE argues that down tilt angle and TXRU mapping should not be defined as network-side additional conditions requiring an associated ID for AI CSI prediction. They present technical evidence showing that AI models generalize well across different down tilt angles, with only a…

R1-2409482 ZTE discussion not treated 9.1.4.1

Discussion on study for AI/ML CSI compression

ZTE analyzes inter-vendor training collaboration options for AI/ML-based CSI compression in NR Release 19, focusing on Directions A (UE-side offline engineering), B (on-device operation), and C (fully standardized reference model). The document presents 32 proposals and 10 observations, arguing for the down-selection…

ZTE proposes conducting comparisons between Case 2 and Case 3 for potential down selection to reduce specification impact analysis efforts, and deferring specification impact analysis for inter-vendor training collaboration until feasibility studies conclude. For Direction A,…

R1-2409483 ZTE discussion not treated 9.1.4.2

Discussion on other aspects of AI/ML model and data

ZTE analyzes model identification options for two-sided AI/ML models in NR, arguing against dataset transfer (MI-Option 2) due to high overhead and latency, while favoring model parameter transfer (MI-Option 3) and standardization of reference models (MI-Option 4). The document contains 14 proposals and 8…

ZTE presents a technical case against MI-Option 2 (dataset transfer), citing huge resource overhead, large latency, and potential performance degradation due to backbone misalignment. They prefer MI-Option 4 (standardization of reference UE-part model) and MI-Option 3 (model…

AI/ML for NR Air Interface

Meeting outcomes

Position changes since RAN1#118bis

Sub-topics

Contributions

Your notes