hash stringlengths 32 32 | doc_id stringlengths 5 12 | section stringlengths 5 1.47k | content stringlengths 0 6.67M |
|---|---|---|---|
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.29.4 Corresponding APIs | This subclause provides a summary on the corresponding API for solution #29.
- AIMLE Assistance of Edge Computing API (request-response or subscribe-notify model; API provider: AIMLE Server; known consumers: CAS, EAS). |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.29.5 Solution evaluation | This solution addresses Key Issue #1 and introduces procedures to assist the opreations of edge compting process for AI/ML task. This solution reuses existing mechanisms of 5GC (e.g. Member UE selection) and ADAE analytics.
NOTE: The possible impact of this solution to other solutions will be discussed and evaluated du... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.30 Solution #30: Support Transfer of Intermediate AIML Operation Information | |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.30.1 General | The following clauses specify procedures, information flows and APIs for KI #6 and #7 to support transfer of intermediate AIML operation information.
Pre-conditions:
- Due to various reasons (e.g., changes of available resource, changes of available time), an AI/ML member (e.g., AI/ML Enablement Client, VAL Client) fin... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.30.2 Procedures for intermediate AI/ML information transfer | |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.30.2.1 AI/ML Enablement Server assist intermediate AI/ML information transfer | Pre-conditions:
- The information of target AI/ML member (e.g. another AI/ML Enablement Client or VAL Client different from the source AI/ML member) is unknown at the source AI/ML member. The source AI/ML member decides that assist from the AI/ML Enablement Server is needed.
Figure 8.30.2.1-1: Procedure for AI/ML Enabl... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.30.2.2 Direct intermediate AI/ML information transfer | Pre-conditions:
- The information of target AI/ML member (e.g., another AI/ML Enablement Client or VAL Client different from the source AI/ML member) is assumed to be known at the source AI/ML member, or the source AI/ML member may get the target AI/ML member information via ML repository. The source AI/ML member decid... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.30.3 Architecture Impacts | The application enabler layer architecture impacts are the following:
- AI/ML Enablement Server is introduced to support discovery of target AI/ML member directly or via ML repository.
- AI/ML Enablement Server is introduced to support assistance of transfer intermediate AI/ML information from source AI/ML member to ta... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.30.4 Corresponding APIs | This subclause provides a summary on the corresponding API for solution #30.- AIMLE Intermediate AI/ML Information Transfer Assist API (request-response; API provider: AIMLE Server; known consumers: AIMLE Client).
- AIMLE Intermediate AI/ML Information Transfer API (request-response; API provider: AIMLE Client; known c... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.30.5 Solution evaluation | This solution addresses Key Issues #6 and #7 and introduces procedures on supporting the maintenance of an AI/ML process and supporting transfer learning at application enablement layers, assistance the transfer of intermediate AI/ML operaton status and results. This solution is feasible and doesn't introduce any depen... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.31 Solution #31: Supporting AIML inference service in edge | |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.31.1 Solution description | The solution addresses Key Issue #1 to enhance the architecture and related functions to support application layer AI/ML services in edge computing scenarios corresponding to hierarchical AIMLE deployment scenario as in clause 9.4.
At present, the vast majority of model inference services are concentrated in online int... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.31.2 Architecture Impacts | The application enabler layer architecture impacts are the following:
- Edge AIML Enablement server is introduced to provide support the inference for the edge scenario.
- Central AIML Enablement server has the capability of interacting with Edge AIML enablement server for providing the trained model for the edge scena... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.31.3 Corresponding APIs | This subclause provides a summary on the corresponding API for solution #31.
- AIML inference service API (request-response model: API provider: Edge AIMLE Server, known consumer: EAS).
- ML models retrieval API (request-response model: API provider: Central AIMLE server, known consumer: Edge AIMLE Server). |
a968fed150b6156e01512b97375c462a | 23.700-82 | 8.31.4 Solution evaluation | This solution addresses Key Issue #1 and in particular the architecture enhancements for edge scenarios (in hierarchical deployments). This solution is feasible and provides value by enabling an edge deployed AIMLE server to fetch a trained model from the central AIMLE / ML repository and support ML model inference loc... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 9 Deployment scenarios | |
a968fed150b6156e01512b97375c462a | 23.700-82 | 9.1 General | This clause provides the different deployment models for AIML enablement (AIMLE) services. There could be three deployment options:
- AIMLE server can be deployed at a centralized cloud platform and collects data from multiple EDNs.
- AIMLE server can be deployed at the edge platform.
- Hierarchical AIMLE server deploy... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 9.2 Deployment model #1: Cloud-deployed AIMLE server | In this deployment, as shown in Figure 9.2-1, the AIMLE server is centrally located and can provide support for AIML operations to the application and edge services (EAS/EES, VAL server, other SEAL services).
Figure 9.2-1 cloud deployed AIML enabler |
a968fed150b6156e01512b97375c462a | 23.700-82 | 9.3 Deployment model #2 Edge-deployed AIMLE server | In this deployment, as shown in Figure 9.3-1, the AIML enabler server is located at the EDN as EAS and provides AIML enablement services to the other EAS(s) or other edge native applications at the edge platform. AIMLE services can be deployed by the ECSP or the MNO to provide value-add services related to AI/ML operat... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 9.4 Deployment model #3: Hierarchical AIMLE server deployment | In this deployment, multiple AIMLE servers can be located at different EDNs/DNs and can be deployed by the same provider. Such hierarchical deployments allow the local – global ML operations (e.g., federated learning across domains).
The ML support services that the edge deployed AIML enabler correspond to the AIML ena... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 10 Business Relationships | Figure 5-1 shows the business relationships that exist for the AIMLE functionality and that are needed to support a single VAL user.
Figure 5-1: Business relationships for VAL services
The VAL user belongs to a VAL service provider based on a VAL service agreement between the VAL user and the VAL service provider. The ... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 11 Overall evaluation | |
a968fed150b6156e01512b97375c462a | 23.700-82 | 11.1 Summary of enablement capabilities and APIs | Various solutions proposed in the present document aim 1) defining new analytics to be supported by ADAES or introducing or 2) introducing a new SEAL service (AIMLE, ML Repository) to support AI/ML services for vertical use cases. This clause provides a summary of the capabilities as well as the corresponding potential... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 11.1.1 Summary of AIMLE services | This clause provides a summary of the new AIMLE and ML repository capabilities based on the Solutions.
Table 11.1.1-1 is listing solutions proposing new capabilities related to AIMLE:
Table 11.1.1-1: Solutions proposing new AIMLE capabilities
Solution
Proposed Capabilities
#2
Support ML client information fetching
#3
S... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 11.2 Key issue #1: Support of Architecture Enhancement and Functions for Application Layer AI/ML Services | The open issues studied in the Key Issue #1 are as follows:
1. Whether and how to enhance the architecture and related functions to support application layer AI/ML services.
2. Whether and how the above architecture enhancement and related functions supporting the management/execution of AI/ML lifecycle operations.
3. ... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 11.3 Key issue #2: AI/ML-enhanced ADAES | The open issues studied in the Key Issue #2 are as follows:
1. Whether and how to enable the ADAE layer (including A-DCCF, A-ADRF) to derive analytics or provide analytics services based on AI/ML methods? This includes the study of necessary enhancements to the ADAE layer architecture, if any.
2. Whether and how the AD... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 11.4 Key issue #3: Support for federated learning | The open issues studied in the Key Issue #3 are as follows:
1. How to support federated learning at application enablement layers?
2. Identify procedures for supporting FL at the application enablement layer, including FL entity discovery, registration, communication, reporting.
3. Whether and how to support the data c... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 11.5 Key issue #4: Supporting Vertical FL at enablement layer | The open issues studied in the Key Issue #4 are as follows:
1. How we can ensure that all training functions have an aligned sample range, e.g., the same users, to support VFL?
2. How can we discover what features are available between domains (for the same sample range) in order to support VFL?
This clause provides an... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 11.6 Key Issue #5: Support for of AI/ML operation splitting between AI/ML endpoints and in-time transfer of AI/ML models | The open issues studied in key issue 5 are:
1. Whether and how to enhance the architecture and related functions to support management and/or configuration for split AI/ML operation, and in-time transfer of AI/ML models. The management and configuration aspects including discovery of requried nodes for split AI/ML oper... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 11.6.1 Split AI/ML operation | Solution #19, #23 and #28 have studied how to address split AI/ML operation in Key Issue #5. A first aspect cited in Key Issue #5 is how to "support management and/or configuration for split AI/ML operation including discovery of requried nodes for split AI/ML operation and support of different models of AI/ML operatio... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 11.6.2 Transfer of AI/ML models | Solution #24 and #25 have studied how to address in-time transfer of AI/ML models in Key Issue #5.
A first aspect cited in Key Issue #5 is how to "support management and/or configuration for in-time transfer of AI/ML models”.
- Solution #24 proposes that AIML model information is managed as a resource in the model repo... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 11.7 Key issue #6: Support for transfer learning | The open issues studied in the Key Issue #6 are as follows:
1. How to support transfer learning at application enablement layers?
This clause provides an overall evaluation of the key issue #6. The solutions #12, #20, and #30 cover different aspects for the open issue in the KI #6.
Solutions #12, #20, and #30 address d... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 11.8 Key issue #7: Discovery or Support of Member Selection and Maintenance for Application Layer AIML Service | The open issues studied in the Key Issue #7 are as follows:
1. How to support the AI/ML for member selection and re-selection (e.g., policies).
2. How to support the AI/ML member participation configurations.
3. How to support the AI/ML maintaining the AI/ML process.
4. How to utilize the AI/ML policies and configurati... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 12 Conclusions | |
a968fed150b6156e01512b97375c462a | 23.700-82 | 12.0 General conclusions and recommendations | The present technical report described the AI/ML enablement capabilities for supporting vertical use cases. This technical report fulfills the objectives of the study on application architecture for enabling AIML services. The report includes the following:
1. Definition of terms and abbreviations used in the study (cl... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 12.1 Conclusions of key issue #1 | Following solution considerations will be considered for the normative work for KI #1:
- Solution #11 can be considered for normative work for AIML service lifecycle management, and #12 can be considered as candidate for normative work for AI/ML model lifecycle management.
- Solution #13 can be considered for normative... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 12.2 Conclusions of key issue #2 | Following solution considerations will be considered for the normative work for KI #2:
- Solution #4 can be considered for normative work to support ML-enabled ADAE analytics.
- Solution #5 can be considered can be considered for normative work for ML model storage and discovery.
- The common parts of Solution #4 and S... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 12.3 Conclusions of key issue #3 | Following solution considerations will be considered for the normative work for KI #3:
- Solutions #13 and #22 can be considered for normative work.
- Solutions #2, #6, and #7 can be merged for FL member discovery and selection, the solution after merging can be considered for normative work.
- Solution #8 can be consi... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 12.4 Conclusions of key issue #4 | Following solution considerations will be considered for the normative work for KI #4:
- Solution #18 can be considered for normative phase. |
a968fed150b6156e01512b97375c462a | 23.700-82 | 12.5 Conclusions of Key Issue #5 | The following principles will be considered in the normative phase for KI #5.
Split AI/ML operation:
- Solution #19 will be considered for AIML split operation pipeline discovery (request/response), events subscription/notification and split an operation pipeline creation/registration (request/response). AIML split ope... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 12.6 Conclusions of key issue #6 | Following solution considerations will be considered for the normative work for KI #6:
- Solution #12 can be considered for normative work for ML model lifecycle management,
- Solution #20 can be considered for normative work for supporting the discovery and selection of models.
- The common parts of Solution #12 and S... |
a968fed150b6156e01512b97375c462a | 23.700-82 | 12.7 Conclusions of key issue #7 | Following solution considerations will be considered for the normative work for KI #7: - Solution #6 and Solution #14 are complementing each other and can be considered for the normative work. It is recommended to specify Solution #6 and Solution #14 as a part the AIMLE Client Discovery and Selection API. - Solution #2... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 1 Scope | This study will focus on the following objectives:
- AI/ML cross-domain coordination aspects on whether and how to consider 5GC enhancements to LCS to support AI/ML based Positioning considering conclusions of the RAN study in TR 38.843 [6].
NOTE 1: UE data collection, model delivery and transfer to the UE and model id... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 2 References | The following documents contain provisions which, through reference in this text, constitute provisions of the present document.
- References are either specific (identified by date of publication, edition number, version number, etc.) or non‑specific.
- For a specific reference, subsequent revisions do not apply.
- Fo... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 3 Definitions of terms and abbreviations | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 3.1 Terms | For the purposes of the present document, the terms given in TR 21.905 [1] and the following apply. A term defined in the present document takes precedence over the definition of the same term, if any, in TR 21.905 [1].
Horizontal Federated Learning (HFL): a federated learning technique without exchanging/sharing local... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 3.2 Abbreviations | For the purposes of the present document, the abbreviations given in TR 21.905 [1] and the following apply. An abbreviation defined in the present document takes precedence over the definition of the same abbreviation, if any, in TR 21.905 [1].
AP Active Participant
HFL Horizontal Federated Learning
PP Passive Particip... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 4 Architectural Assumptions and Requirements | The present study will not consider service-based interfaces with RAN and with UE.
The architecture for the present study shall comply with the existing NWDAF framework as specified in TS 23.288 [5], and 5GS framework as specified in TS 23.501 [2], TS 23.502 [3] and TS 23.503 [4].
The architecture for the present study... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5 Use Cases, Motivations and Key Issues | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.1 Use Cases | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.1.0 Guidelines | Use cases as captured in the following sub-clauses are related to Key Issue #2, Key Issue #3 and Key Issue #4. Table 5.2.0-1 shows the mapping of Key Issues to Use Cases.
NOTE 1: Capturing use cases for Key Issue #4 is optional.
NOTE 2: For KI#1, use cases are based on Positioning accuracy enhancements, case 2b and cas... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.1.1 Use Case #1: NWDAF-assisted QoS enhancement | Currently, the QoS parameters are determined by the PCF based on its knowledge, e.g. AF requirements, analytics provided by the NWDAF, etc. After applying the determined QoS parameters to the service, the PCF may determine whether or not the current QoS can fully satisfy the service requirements based on the Service Ex... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.1.2 Use Case #2: Enhancements to QoS Determination with NWDAF Assistance | A use case is provided for how the network can benefit from the NWDAF-assistance for QoS determination and setup for the purpose of optimising the overall network performance and signalling based on operator's policy.
After UE registers with the 5GS, a PDU session set up might be required. Each PDU session is associate... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.1.3 Use Case #3: NWDAF assistance in device signalling storm prevention and mitigation | In some scenarios, e.g. NB-IoT CP optimization scenario, UEs send small data over NAS signalling. In case e.g. the application on the NB-IoT UEs is not implemented correctly, e.g. report data at the same time, the NB-IoT devices in some area may be active at the same time so that a large amount of NAS signalling may be... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.1.4 Use Case #4: Motivation and Support for VFL in 5GC | It is well known in the AI/ML literature that VFL is a federated learning setting where multiple parties perform training on data sets that share the same sample space but differ in feature space. Because of this, an alignment in sample and feature spaces among participating entities is usually required before applying... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.1.5 Use Case #5: NWDAF support for observed service experience analytics based on VFL | When NWDAF provides observed service experience analytics, as in other analytics that require input data from the AF, policies in the PLMN and or the AF may prevent raw data to be exchanged directly between NWDAF and an external AF, as NWDAF is in the PLMN and the AF is outside the PLMN and the user data has high priva... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.1.6 Use Case #6: Analytics-assisted prevention of abnormal NF behaviour causing signalling storm and mitigation of its impact in the network | The presented use case is to elaborate on how analytics can assist entities in 5GC, e.g. NFs, OAM, etc. to prevent and mitigate the impact of abnormal behaviour i.e. signalling storm in the network.
The network may, based on the analytics, discover abnormal NF behaviour, i.e. signalling storms, and begin an enforcement... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.2 Key Issues | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.2.0 Mapping of Key Issues to Use Cases | Table 5.2.0-1: Mapping of Key Issues to Use Cases
Key Issues
Use cases
2
4, 5
3
1, 2
4
3, 6 |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.2.1 Key Issue #1: Enhancements to LCS to support Direct AI/ML based Positioning | This key issue aims to provide solutions for whether and how to consider enhancements to support AI/ML based positioning for case 2b, 3b as defined in TR 38.843 [6], which will investigate the following aspects:
- Study whether and how an AI/ML model for direct AI/ML positioning (i.e. case 2b/3b) is handled:
- Which en... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.2.2 Key Issue #2: 5GC Support for Vertical Federated Learning | This key issue aims to provide solutions for enabling 5GC support for vertical federated learning (VFL) involving NWDAF and/or AF, where no raw data need to be exchanged but some level of coordination is still required when training and inference are performed on local models. In particular, datasets used for each loca... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.2.3 Key Issue #3: NWDAF-assisted policy control and QoS enhancement | The NWDAF can gather quite a lot of data from 5GC NFs, AF and OAM and thus may further assist the PCF in making PCC decisions (which traditionally determine QoS parameters based on its own data and knowledge as well optional statistics and predictions collected from the NWDAF).
This key issue aims to study whether and ... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 5.2.4 Key Issue #4: NWDAF enhancements to support network abnormal behaviours (i.e. signalling storm) mitigation and prevention | This key issue aims to provide solutions for prediction, detection, prevention, and mitigation of network abnormal behaviours, i.e. signalling storm, with the assistance of NWDAF. In particular, the following aspects will be addressed:
- Identify scenarios that can result in a signalling storm situation.
- Whether and ... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6 Solutions | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.0 Mapping of Solutions to Key Issues | Table 6.0-1: Mapping of Solutions to Key Issues and Use Cases
Key Issues
Use cases (optional)
Solutions
1
2
3
4
1
2
3
4
5
6
#1
X
#2
X
#3
X
#4
X
#5
X
#6
X
#7
X
#8
X
#9
X
#10
X
#11
X
#12
X
#13
X
X
#14
X
X
X
#15
X
X
X
#16
X
X
#17
X
X
#18
X
X
X
#19
X
X
#20
X
X
X
#21
X
X
X
#22
X
X
X
#23
X
X
X
#24
X
X
#25
X
X
#26
X
X
X
#27
X... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.1 Solution #1: Direct AI/ML based Positioning for case 2b/3b | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.1.1 Description | This solution is for Key Issue#1: Enhancements to LCS to support Direct AI/ML based Positioning.
For Direct AI/ML based Positioning in LMF side, it will be the LMF that provides the estimated UE location based on AI mechanism. In this solution, the LMF can be collocated with an NWDAF containing AnLF.
The ML model for i... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.1.2 Procedures | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.1.2.1 Direct AI/ML based Positioning in LMF | Figure 6.1.2.1-1: Direct AI/ML based Positioning in LMF collocated with AnLF
0-2. UE may trigger 5GC-MO-LR Procedure, or LCS client may trigger 5GC-MT-LR Procedure as defined in TS 23.273 [7].
3. The LMF determines whether to use legacy UE positioning methods as defined in TS 23.273 [7] or AI/ML based positioning metho... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.1.2.2 Training procedure for Direct AI/ML positioning model | The NWDAF containing MTLF can provide trained LMF-side model to consumer as follows:
Figure 6.1.2.2-1: Procedure for Training of the Direct AI/ML positioning model
If the NWDAF consumer (i.e. the LMF) requests a trained model in step 1, the NWDAF provides the trained ML model to the consumer with the inference input da... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.1.2.3 Procedure for ML model training for Direct AI/ML based positioning and data collection | Figure 6.1.2.3-1: Direct AI/ML based positioning and data collection
1. When the LMF determines to use AI/ML based positioning method to obtain UE location, the LMF may get a trained ML model from NWDAF containing MTLF. The LMF decides to get a trained ML model from NWDAF may be based on LMF implementation or based on ... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.1.3 Impacts on services, entities and interfaces | UE:
- Depending on RAN WG decision, perform and report new measurements.
RAN:
- Depending on RAN WG decision, perform and report new measurements.
LMF:
- Determine to use AI/ML based positioning to obtain UE location, based on LMF's AI/ML based positioning related capability and measurement data types reported from UE ... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.2 Solution #2: Support for AI/ML Direct Positioning Training, Inference and Data Collection with LMF-side models | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.2.1 Description | This solution addresses KI#1. The solution presents three different aspects regarding 5GC support for AI/ML Direct Positioning with LMF-side models, namely training, inference and data collection procedures. The solution addresses Case 2b and 3b as defined in TR 38.843 [6].
The main features of the solution are as foll... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.2.2 Procedures | Clauses 6.2.2.1, 6.2.2.2 and 6.2.2.3 show step-by-step procedures for training, inference and data collection aspects of AI/ML Direct Positioning with LMF-side models, respectively.
NOTE: In all procedures of this solution, LMF may be a standalone NF or co-located with NWDAF containing MTLF/AnLF for model training/mode... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.2.2.1 Training procedure for AI/ML Direct Positioning with LMF-side models | Figure 6.2.2.1-1: Training procedure for AI/ML Direct Positioning with LMF-side models
The training procedure for AI/ML Direct Positioning with LMF-side models in Figure 6.2.2.1-1 is described step by step below.
1. LMF subscribes to training data from NWDAF for an AI/ML model to be used for direct positioning. The tra... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.2.2.2 Inference procedure for AI/ML Direct Positioning with LMF-side models | Figure 6.2.2.2-1: Inference procedure for AI/ML Direct Positioning with LMF-side models
This procedure assumes that a trained AI/ML model is available in LMF for Direct Positioning, i.e. the procedure in Figure 6.2.2.1-1 has already taken place. The inference procedure for AI/ML Direct Positioning with LMF-side models ... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.2.3 Impacts on services, entities and interfaces | NWDAF:
- Enhance support for model training services to enable AI/ML positioning training and inference at LMF.
LMF:
- Support for training/inference for AI/ML Direct Positioning and data collection. |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.3 Solution #3: Training of the AI/ML positioning model | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.3.1 Functional Description | A request of LMF-side model that outputs location of a specific UE is provided by the service consumer. After training of the ML model, the NWDAF provides the trained ML model to the service consumer (e.g. LMF) and the service consumer can perform inference afterwards.
Although the service consumer can obtain UE locati... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.3.1.1 Input of model training | Table 6.3.1.1-1: Data Collected for LMF-side model training
Information
Source
Description
Measurements
>Measurement data
The measurement data collected by LMF.
>Positioning method
TBD
The positioning method corresponding to the measurement data.
>Time stamp
Time stamp of measurement data.
Ground truth data
TBD
The Gro... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.3.2 Procedures | The NWDAF can provide trained LMF-side model to consumer as follows:
Figure 6.3.2-1: Procedure for Training of the AI/ML positioning model
If the consumer request a trained model in step 1, the NWDAF provides the trained model to the consumer with the training input data information to help the consumer perform inferen... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.3.3 Impacts on existing services, entities and interfaces | To implement the proposed solution, the NWDAF should support the proposed model training with input data(listed in table 6.3.1.1-1) collection capability. The trained model shall have the ability to calculate UE location. Besides, the NWDAF shall be able to provide the trained model to consumer. |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.4 Solution 4: Data Collection Framework for Direct AI/ML positioning | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.4.1 Description | Editor's note: This clause will describe the solution principles and architecture assumptions for corresponding key issue(s). Sub-clause(s) may be added to capture details.
The current LCS framework allows the LMF to collect positioning measurement related data from UE and the RAN via control plane by providing positio... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.4.2 Procedures | Editor's note: This clause describes high-level procedures and information flows for the solution.
The procedure for configuring the UE and RAN node is as follows. For simplicity only the control plane procedure for reporting data is shown.
Figure 6.4.2-1: Data collection via LMF
1. A consumer requires data to train an... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.4.3 Impacts on services, entities and interfaces | Editor's note: This clause captures impacts on existing services, entities and interfaces.
LMF:
- Needs to handle requests from consumer (e.g. NWDAF MTLF) to collect positioning related measurement data from RAN/UE
NOTE: There may be potential impacts on LPP and/or NRPPa for collecting data for AI based positioning, wh... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.5 Solution #5: LMF selection to support the LMF-sided direct AI/ML positioning | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.5.1 Description | LMF selection functionality is supported by the AMF to determine an LMF for location estimation of the target UE. The LMF selection functionality is also supported by the LMF if it determines that it is unsuitable or unable to support location for the current UE based on the accurate requirement.
LMF reselection is a f... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.5.2 Procedures | The procedure will be built based on the procedures defined by the clause 5.1 of TS 23.273 [7] for LMF discovery and selection with no flow modification. |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.5.3 Impacts on services, entities and interfaces | AMF:
- Enhanced to support the LMF discovery and selection/reselection based on the LMF AI based positioning capability.
LMF:
- Enhanced to support the AI based positioning capability.
UDM:
- Enhanced to store the AI based positioning subscription data to indicate whether the UE is authorized to use AI based positionin... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.6 Solution #6: LMF based ML model training and Inference | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.6.1 Description | This is a solution proposed for KI#1: Enhancements to LCS to support Direct AI/ML based Positioning.
In order to support the Direct AI/ML based positioning for case 2b, 3b, the LMF is required to be enhanced to support the AI/ML- based positioning. How LMF performs ML model training and inference, is based on implement... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.6.2 Procedures | This is the data collection procedure to support ML model training and inference for AI/ML based positioning in LMF.
Figure 6.6.2-1: Data collection for LMF based ML model training and inference
1. LMF requests from the UE to report the positioning measurement parameters as described in TS 23.273 [7]. UE reports positi... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.6.3 Impacts on services, entities and interfaces | LMF, NG-RAN, UE:
NOTE: The collected measurement parameters from UE in step 1 and NG-RAN in step 2 is determined by RAN WGs. If there are new measurement parameters that provided by UE or NG-RAN to LMF in Rel-19 depends on RAN's progress in Rel-19.
Editor's note: How does LMF selects UEs to collect data for ML model tr... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.7 Solution #7 Training of LMF-side Model to determine location | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.7.1 Description | Models will closely relate to topologies and antenna in an area or a radio cell and will thus be specific to an area or cell or at least consider the cell as input. For models related to UE measurement (for Case 2b: UE-assisted/LMF-based positioning), details of the UE such as number of reception points may also matter... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.7.2 Procedures | |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.7.2.1 Training an LMF-side model for NG-RAN assisted positioning | Figure 6.7.2.1-1: Training an LMF-side model for NG-RAN assisted positioning
1. The NG RAN nodes inform the AMF about their capability to support NG-RAN assisted positioning. The NG RAN nodes may also provide periodic reports about their load and/or send indications when being in overload.
Editor's note: How the NG_RAN... |
c8f50ee24a14933c33290076ea3625ad | 23.700-84 | 6.7.2.2 Training an LMF-side model for UE assisted positioning | Figure 6.7.2.2-1: Training an LMF-side model for UE assisted positioning
1. The NG RAN nodes inform the AMF about their capability to support UE assisted positioning. The NG RAN nodes may also provide periodic reports about their load and/or send indications when being in overload.
Editor's note: How the NG_RAN .indica... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.