Triggering ai / ML features with UE-side model inference
The coordination of AI/ML features in UE, RAN, and CN through standardized mechanisms and data collection functions addresses the lack of trigger mechanisms in existing 3GPP specifications, enabling efficient data collection and model inferencing for improved communication systems.
Patent Information
- Application Number
- PCT/US2026/027668
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-10-23
- Filing Date
- 2026-05-11
- Publication Date
- 2026-11-12
AI Technical Summary
Existing 3GPP specifications lack a standardized mechanism to trigger AI/ML enabled features for the core network, radio access network, and user equipment, and unclear communication regarding the activity of such features.
Implement methods and apparatuses in network functions of the core network, radio access network, and user equipment to coordinate AI/ML feature operations, including data collection, model inferencing, and capability negotiation using feature identifiers, with a data collection function operating within or as a standalone component to support user-plane data collection.
Enables efficient coordination and activation of AI/ML features across UE, RAN, and CN, facilitating data collection and model inferencing, enhancing features like positioning accuracy and resource management.
Smart Images

Figure US2026027668_12112026_PF_FP_ABST
Abstract
Description
PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 TRIGGERING AI / ML FEATURES WITH UE-SIDE MODEL INFERENCECROSS-REFERENCE TO RELATED APPLICATION
[0001] This application claims priority to U.S. Provisional Application No. 63 / 803,533, filed May 9, 2025, titled "Method of Triggering AI / ML for NR Air Interface with UE-side Model Inference” and U.S. Provisional Application No. 63 / 904,487, filed October 23, 2025, titled "Method of Triggering AI / ML for NR Air Interface with UE-side Model Inference,” the disclosures of which are incorporated herein by reference in their entirety.FIELD OF THE DISCLOSURE
[0002] This disclosure relates generally to methods, devices, and articles in wireless communication systems, such as 3GPP communication systems, and in particular to operating UE-side artificial intelligence / machine learning (AI / ML) models.BACKGROUND
[0003] This background description is provided for the purpose of generally presenting the context of the disclosure. Work of the presently named inventors, to the extent it is described in this background section, as well as aspects of the description that may not otherwise qualify as prior art at the time of filing, are neither expressly nor impliedly admitted as prior art against the present disclosure.
[0004] The 3rd Generation Partnership Project (3GPP) contemplates convergence between communication system functions and Artificial Intelligence (Al) technology, to improve the intelligence at the fifth-generation core (5GC) and the air interface and support network automation. More particularly, 3 GPP has begun to address such topics as data collection (DC), machine learning (ML) model training, analytics inference, etc. These technical areas require collaborative AI / ML mechanisms for coordinating functionality across such domains as user equipment (UE), radio access network (RAN), 5GC, operations and maintenance (0AM), and application functions (AF).PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0005] To support collection of training data for a UE-side model that generates complete inferences at the UE, based on the specifications related to AI / ML for New Radio (NR) air interface documented in 3GPP TR 38.843, 3GPP has agreed to study the potential support of UE data collection to meet the requirements for RAN Al support for air interface operation. Today, the remaining issues include enhancement of UE data collection in a 5G system.
[0006] Generally speaking, there are several options for data collection for UE-side model training: (i) a UE can collect and directly transfer training data to an Over-the-Top (OTT) server, in a transparent or non-transparent manner; (ii) a UE can collect training data and transfer the training data to the core network, which then transfers the training data to the OTT server; or (iii) a UE can collect training data and transferring the training data to an 0AM server, which then transfers the relevant data to the OTT server.
[0007] Communication systems can apply training data collection to different 3GPP system features to enhance resource management in terms of radio air interface, radio resource management, RAN mobility, and core network automation, etc. Examples of such features include artificial intelligence (AI) / ML based channel strength indicator (CSI) compression, AI / ML based CSI-reference signal (RS) overhead reduction, AI / ML based CSI prediction, AI / ML based beam management with downlink beam prediction, direct AI / ML positioning, AI / ML assisted positioning for positioning accuracy enhancement, etc.
[0008] However, the existing 3GPP specifications lack a standardized mechanism to trigger an AI / ML enabled feature for the core network, the RAN and the UE. Further, it remains unclear how the UE, the RAN, and the CN should communicate notifications regarding the activity of an AI / ML enabled feature.SUMMARY
[0009] An example embodiment of the techniques of this disclosure is a method implemented in one or more network functions (NFs) of a core network (CN) in a cellular communication system. The method comprises receiving, from an application function (AF) associated with an artificial intelligence and / or machine learning (AI / ML) feature, a request (i) related to collecting, from a user equipment (UE) via a radio access network (RAN), data related to the AI / MLPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 enabled feature and (ii) including a feature identifier (ID) of the AI / ML enabled feature; and configuring the collecting of the data related to the AI / ML enabled feature, via a user plane function (UPF).
[0010] Another example embodiment of these techniques is a method implemented in an NF associated with a cellular communication system. The method comprises determining a capability of a UE with respect to an AI / ML enabled feature; and receiving, from the UE and in accordance with a trigger event, data related to the AI / ML enabled feature.
[0011] Another example embodiment of these techniques is a method implemented in a UE, the method comprising: detecting a trigger event related to transferring, to a CN, data related to an AI / ML enabled feature; and in response to the trigger event, transmitting the data to the CN.
[0012] Still another example embodiment of these techniques is an apparatus comprising processing hardware and configured to implement one of the methods above.BRIEF DESCRIPTION OF THE DRAWINGS
[0013] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate one or more embodiments and, together with the detailed description, explain these embodiments.
[0014] Fig. 1 is a block diagram of an example wireless communication system in which techniques for triggering AI / ML enabled features can be implemented;
[0015] Fig. 2 is a block diagram of an example protocol stack according to which the UE of Fig. 1 can communicate with the RAN of Fig. 1;
[0016] Fig. 3 is a service-based representation of the CN architecture, which the system of Fig. 1 can implement, and in which the core network implements a privacy enforcement function (PEF);
[0017] Fig. 4 is a reference-point based representation of the CN architecture of Fig. 3, which the system of Fig. 1 can implement;
[0018] Fig. 5A is a block diagram of a known functional framework for RAN intelligence;PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0019] Fig. 5B is a block diagram of a known functional framework for AI / ML for NR air interface;
[0020] Fig. 6 is a messaging diagram of an example scenario in which a UE and a network negotiate support of AI / ML enabled feature(s) using a non-access stratum (NAS) registration request procedure;
[0021] Fig. 7A is a messaging diagram of an example scenario in which a UE and a network negotiate support of AI / ML enabled feature using a NAS registration request procedure and a radio resource control (RRC) procedure;
[0022] Fig. 7B is a messaging diagram of an example scenario similar to that of Fig. 7A, but in which the NAS registration request procedure involves an AI / ML enabled feature indication rather than AI / ML enabled feature IDs;
[0023] Fig. 8A is a messaging diagram of an example scenario in which an Application Function (AF) triggers data collection for AI / ML enabled feature by invoking a Network Exposure Function (NEF) or a Policy Control Function (PCF) service;
[0024] Fig. 8B is a messaging diagram of an example scenario in which an application function (AF) triggers data collection for AI / ML enabled feature by invoking a data collection function (DCF) service;
[0025] Fig. 9 is a messaging diagram of an example scenario that includes UE policy provisioning related to an AI / ML enabled feature;
[0026] Fig. 10 is a flow diagram of an example method for communicating a UE Data Transfer Control Message (UEDT-CM), which can be implemented in the DCF of Fig. 1; and
[0027] Fig. 11 is a flow diagram of an example method for configuring an AI / ML enabled feature, which can be implemented in one or more NFs of the CN of Fig. 1.DETAILED DESCRIPTION OF THE DRAWINGS
[0028] Using the techniques of this disclosure, a core network (CN), a radio access network (RAN), and a user equipment (UE) can coordinate operation of an AI / ML enabled feature that involves data collection. For example, an Application Function (AF) performing UE-side modelPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 training can request data collection from a cellular communication system (e.g., a 5G system or a 6G system) for an AI / ML enabled feature , with the UE performing AI / ML model inferencing using UE side AI / ML model. For such an AI / ML-enabled feature, the UE can start AI / ML model inferencing based on a corresponding generalized AI / ML model (e.g. obtained from the application server (AS) supporting federated learning) or the available AI / ML model at the UE. To support an AI / ML enabled feature requiring UE-side AI / ML model inferencing, both the RAN and the UE need to know when the AF requests activation of the AI / ML enabled feature.
[0029] The UE, the RAN, and / or the CN support negotiation of UE capabilities with respect to the AI / ML enabled feature as well as indication of CN and RAN node support of the AI / ML enabled feature, and the CN also can determine whether the AI / ML enabled feature is allowed for the UE. Further, the CN can enable the AI / ML enabled feature at the RAN node and the UE, trigger UE data collection necessary for the UE-side AI / ML inferencing process (e.g., inferencing input data from the RAN node), triggering the AI / ML inferencing process at the UE, triggering UE data collection relevant to AI / ML model training (e.g., inferencing input, output, results, and related situational information), when the AI / ML feature requires training, and trigger reporting of collected data, inferencing results, or the status
[0030] Some of the techniques involve a data collection function (DCF), which can be implemented as a component of another NF or as standalone, special-purpose NF dedicated specifically to user-plane (UP) data collection. More specifically, in one deployment, the DCF operates within an existing (legacy) NF such as for example a Location Management Function (LMF), a Policy Control Function (PCF), an Access and Mobility Management Function (AMF), or a Session Management Function (SMF) to support new service operations for UP data collection directly. As a more specific example, an AI / ML-enabled feature can be related to positioning accuracy enhancement, and the DCF can operate within the LMF to provide the UP data collection service for that feature.
[0031] In other implementations, the DCF is a dedicated NF configured to support operations related to UP data collection for, or on behalf of, the existing (legacy) NFs such as an LFM, a PCF, an AMF, an SMF, a User Plane Function (UPF), a Network Data Analytics Function (NWDAF), or a Sensing Service Function (SSF) for example. According to this approach, NFs can request a UP data collection service from the dedicated DCNF. Examples of thisPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 deployment include an AI / ML enabled feature for an NWDAF analytics enhancement, where the DCF is a UP data collection service provider configured to provide UP data collection service for the NWDAF operating as a service consumer, or an AI / ML enabled feature for positioning accuracy enhancement, where the DCF is a UP data collection service provider configured to provide UP data collection service for the LMF operating as a service consumer.
[0032] In yet other implementations, a DCF operates in an NWDAF with coordination functionality, i.e., with a Data Collection Coordination Function (DCCF). According to this approach, NFs operate as consumers and request a data collection service from the NWDAF with the DCCF. Accordingly, the NWDAF with DCCF can enhance operations ofNnwdaj ' DataManagement services, to collect training data from a UE.
[0033] The AI / ML-enabled features can have respective feature identifiers (IDs). A UE in general supports one or more AI / ML-enabled features and uses a feature ID to associate particular training data with a particular AI / ML-enabled feature. More generally, a UE and / or the network (e.g., the CN) can use feature IDs when handling UP data collection and policies that control UP data connections for UP data collection (UPDC). In some cases, for a specific AI / ML-enabled feature, one or more AI / ML models may apply. When only one AI / ML model applies to an ALM- enabled feature, the Feature ID can identify an AI / ML model as a Model ID (and, conversely, the Model ID can identify the feature in this case). A UE can support one or more AI / ML-enabled features, and each AI / ML-enabled feature can have one or more AI / ML models applicable in different situations.
[0034] As discussed in more detail below, an AF performing AI / ML model training can request data collection from a CN for an AI / ML enabled feature, with a UE performing AI / ML model inferencing using a UE-side AI / ML model. For this AEML-enabled feature, the UE can start AI / ML model inferencing based on a corresponding generalized AI / ML model (e.g. obtained from an OTT server supporting federated learning) or an AI / ML model available at the UE. To provide a customized AI / ML model for the UE, the AS or the OTT server performing AI / ML model training can request that the DCF operating in the core network report data related to the AI / ML-enabled feature. The UE can receive, from the DCF, a request for data collection. In particular, the DCF can request that the UE perform data collection and report related inferencing input data, inferencing output data, inferencing results, and corresponding situationalPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 information (e.g. location, time, etc.), to the DCF and the AF. The UE then can obtain a customized UE-side AI / ML model for the AI / ML enabled feature. UE data collection can include all the types of data mentioned above and required for AI / ML model training.
[0035] The following techniques related to triggering mechanisms for an AI / ML enabled feature, UE data collection, and reporting for UE-sided model training are discussed below: (i) negotiation support of an AI / ML enabled feature (see Figs. 6, 7A, and 7B), (ii) AF triggering data collection for AI / ML enabled feature using (see Figs. 8A and 8B), (iii) UE policy enhancements to support the approaches of Figs. 6-8B, (iv) triggering the AI / ML enabled feature (see Figs. 6-8B).
[0036] Referring first to Fig. 1, an example wireless communication system 100 can implement one or more of the techniques of this disclosure for triggering AI / ML enables features. The example wireless communication system 100 includes UE 102, a base station (BS) 104, a base station 106, and a core network (CN) 110, such as a fifth generation (5G) core (5GC). The base stations 104 and 106 can operate in a RAN 105 connected to the CN 110. The CN 110 can also be implemented as a sixth generation (6G) core or another suitable core network.
[0037] The base station 104 covers a cell 124, and the base station 106 covers a cell 126. If the base station 104 is a gNB, the cell 124 is an NR cell. If the base station 124 is an ng-eNB, the cell 124 is an evolved universal terrestrial radio access (E-UTRA) cell. Similarly, if the base station 106 is a gNB, the cell 126 is an NR cell, and if the base station 126 is an ng-eNB, the cell 126 is an E-UTRA cell. The cells 124 and 126 can be in the same Radio Access Network Notification Areas (RNA) or different RNAs. The cells 124 and 126 can partially overlap, so that the UE 102 can select, reselect, or hand over from one of the cells 124 and 126 to the other. In general, the RAN 105 can include any number of base stations, and each of the base stations can cover one, two, three, or any other suitable number of cells. The UE 102 can support at least a 5GNR (or simply, “NR”) air interface to communicate with the base stations 104 and 106. Each of the base stations 104, 106 can connect to the CN 110 via an interface (e.g., SI or NG interface). The base stations 104 and 106 also can be interconnected via an interface (e.g., X2 or Xn interface) for interconnecting NG RAN nodes.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0038] The CN 110 include an AMF 164, an SMF 166, a UPF 170, a PCF 160, a Network Exposure Function (NEF) 154, and an AF 158. The CN 110 also includes a DCF 182 which, as discussed above, can operate as a standalone NF or as a component of another NF. Additional NFs that can operate in the CN 110 or a similar CN are discussed with reference to Figs. 3 and 4. While not shown in Fig. 1 to avoid clutter, the CN 110 may include processing hardware, which may include one or more general-purpose processors (e.g., CPUs) and a non-transitory computer-readable memory storing instructions that the one or more general-purpose processors execute. Additionally or alternatively, the processing hardware can include special-purpose processing units.
[0039] The base station 106 is equipped with processing hardware 142 that can include one or more general -purpose processors such as CPUs and non-transitory computer-readable memory storing machine-readable instructions executable on the one or more general-purpose processors, and / or special-purpose processing units. The base station 106 also includes a transceiver 142 to communicate with the UE 102 over a radio interface. The base station 106 further includes a memory 144 storing a data collection controller 146 configured to implement one or more of the techniques for AI / ML feature triggering and data collection or transfer. The base station 104 may be configured in a similar manner as the base station 106.
[0040] The UE 102 is equipped with processing hardware 130 that can include one or more general-purpose processors such as CPUs and non-transitory computer-readable memory storing machine-readable instructions executable on the one or more general-purpose processors, and / or special-purpose processing units. The UE 102 also includes a transceiver 132 to communicate with the RAN 105 over a radio interface. Further, the UE 102 includes a memory 134 storing a AI / ML feature controller 131 configured to implement one or more of the techniques for managing feature capability signaling and data transfer discussed in this disclosure. The memory 134 in some cases further stores a UE-side ML model (not shown).
[0041] Fig. 2 illustrates, in a simplified manner, an example protocol stack 200 according to which the UE 102 can communicate with an eNB / ng-eNB 230 or a gNB 232 (e.g., one or more of the base stations 104, 106). In the example stack 200, a physical layer (PHY) 202A of EUTRA provides transport channels to the EUTRA MAC sublayer 204A, which in turn provides logical channels to the EUTRA RLC sublayer 206A. The EUTRA RLC sublayer 206A in turnPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 provides RLC channels to a EUTRA PDCP sublayer 208 and, in some cases, to an NR PDCP sublayer 210. Similarly, the NR PHY 202B provides transport channels to the NR MAC sublayer 204B, which in turn provides logical channels to the NR RLC sublayer 206B. The NR RLC sublayer 206B in turn provides data transfer services to the NR PDCP sublayer 210. The NR PDCP sublayer 210 in turn can provide data transfer services to Service Data Adaptation Protocol (SDAP) 212 or a radio resource control (RRC) sublayer (not shown in Fig. 2). The UE 102, in some implementations, supports both the EUTRA and the NR stack as shown in Fig. 2, to support handover between EUTRA and NR base stations and / or to support DC over EUTRA and NR interfaces. Further, as illustrated in Fig. 2, the UE 102 can support layering of NR PDCP 210 over EUTRA RLC 206A, and SDAP sublayer 212 over the NR PDCP sublayer 210.
[0042] The EUTRA PDCP sublayer 208 and the NR PDCP sublayer 210 receive packets (e.g, from an Internet Protocol (IP) layer, layered directly or indirectly over the PDCP layer 208 or 210) that can be referred to as service data units (SDUs), and output packets (e.g., to the RLC layer 206A or 206B) that can be referred to as protocol data units (PDUs). Except where the difference between SDUs and PDUs is relevant, this disclosure for simplicity refers to both SDUs and PDUs as “packets.”
[0043] On a control plane, the EUTRA PDCP sublayer 208 and the NR PDCP sublayer 210 can provide signaling radio bearers (SRBs) or RRC sublayer (not shown in Fig. 2) to exchange RRC messages or non-access-stratum (NAS) messages, for example. On a user plane, the EUTRA PDCP sublayer 208 and the NR PDCP sublayer 210 can provide data radio bearers (DRBs) to support data exchange. Data exchanged on the NR PDCP sublayer 210 can be SDAP PDUs, Internet Protocol (IP) packets, or Ethernet packets.
[0044] Fig. 3 is a service-based representation 300 of an example CN architecture, which the system of Fig. 1 can implement as the CN 110, and in which an FTCF is implemented in a AMF 364 or as a dedicated NF. Generally speaking, similar operations, devices and functions are labeled with similar reference numbers that share two least significant digits, with differences discussed below where appropriate. With the exception of the differences shown in the figures and discussed below, any of the other implementations discussed with respect to a particular event (e.g., for messaging and processing) may apply to events labeled with similar reference numbers in other figures.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0045] In the representation 300, the overall non-roaming reference architecture of the Policy and Charging Control (PCC) framework for the 5GS includes components illustrated using solid lines, and the other components are illustrated using dashed lines. According to this representation, network functions enable other authorized network functions to access their services. The components that are outside the PCC framework include a Network Slicing Selection Function (NSSF) 302, a Network Repository Function (NRF) 306, a Unified Data Management (UDM) 308, an Edge Application Server Discovery Function (EASDF) 310, a Network Slice Specific Authentication and Authorization Function (NSSAAF) 312, an Authentication Server Function (AUSF) 314, a Service Communication Proxy (SCP) 316, and a Network Slice Admission Control Function (NSACF) 318. The non-PCC architecture further includes the UE 102 and the (R)AN 105.
[0046] The PCC framework in the architecture 300 includes a UDR 352, an NEF 354, an NWDAF 356, an AF 358, aPCF 360, a Charging Function (CHF) 362, an AMF 364, an SMF 366, and a UPF 370. The UPF 370 can access a data network (DN) 330, in which an application server (AS) 331 can operate. The CN 300 can further include an LMF 345.
[0047] In some implementations, the NWDAF 356 can support data collection for network automation and interact with a data collection application function (DCAF) 342 operating in the trusted domain of the operator network, to collect data from a UE application as an input for analytics generation and ML model training. This framework can apply to the UE collecting and directly transferring training data to an OTT server as discussed above, where the DCAF 342 operates as a data collection network function (DCNF) that connects to the UPF 370 via an N6 interface.
[0048] A DCF 382 is illustrated in Fig. 3 as a standalone component. However, as discussed above, the DCF 382 in other deployments can operate as a component of another NF such as the LMF 345, the AMF 364, the PCF 360, etc. In some deployments, the DCF operates as a data consumer NF. In other deployments, the CN includes a Data Consumer NF 383 as an additional node.
[0049] The AMF 364 is generally configured to manage registration, connection, and mobility of a UE (such as the UE 102A or 102B) and provide transport for session management (SM) messages between the UE 102A or 102B and the SMF 366. The SMF 366 is generallyPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 configured to manage sessions, allocate IP addresses for UEs, and provides downlink (DL) notifications. In some implementations, the SMF 366 also includes following functionalities for PIN service: providing per-QoS flow non-3GPP QoS assistance information to the UE (e.g., PEGC), and supporting IP address allocation to UE and PDR configuration with packet filter set for PIN to UPF for framed routing based on PIN group information from the UDM 308.
[0050] The UDM 308 is generally configured to handle user identification, access authorization based on subscription data, and subscription management. In some implementations, the UDM 308 supports the functionality of PIN group management handling.
[0051] The UDR 352 is generally configured to store subscription-related information, such as subscription data, policy data, structured data for exposure, and application data. The UPF 370 is generally configured to handle packet routing and forwarding. The NEF 354 is generally configured to expose a network’s capabilities and services to authorized third-party applications.
[0052] The AF 358 in some deployment operates in a trusted domain or outside the trusted domain, i.e., in a non-trusted domain. The trusted domain is generally internal to the CN 110 and includes such components as the UDM 308, the UDR 352, the PCF 360, the AMF 364, the SMF 366, and the UPF 370. Generally speaking, an AF operating outside the trusted domain (which may be operated by an authorized third-party entity) can access the network functions of the CN 110 only via the NEF 354, whereas an AF operating within the trusted domain can access at least some of the network functions of the CN 110 directly, or may access these functions via the NEF 354 in some deployments.
[0053] Fig. 4 is a reference-point based representation 400 of the example 5GS architecture discussed with reference to Fig. 3. In Fig. 4, the non-roaming reference architecture of the PCC framework for the 5GS is illustrated as blocks and connections with solid lines, and components and connections outside the PCC framework are illustrated using dashed lines. The interfaces are labeled with the corresponding protocols, e.g., N3 between the RAN 105 and the UPF 370, N1 between the AMF 364 and the UE 102, etc.
[0054] As illustrated in Fig. 4, the RAN in at least some implementations can communicate with the Data Consumer NF 383 directly rather than via the AMF 364. Thus, a UE can access thePATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 Data Consumer NF 383 without using the NAS protocol. Further, he CN 400 can include a dedicated interface between the DCF 382 and the AS 331.
[0055] The communication system shown Figs. 1-4 in may include additional, fewer, and / or alternative devices or functionalities, and may be configured to perform additional, fewer, or alternate actions, including functionalities / actions described herein.
[0056] Referring to Fig. 5A, a known functional framework a RAN includes a Data Collection entity that provides training data to a Model Training entity and inference data to a Model Inference entity. The Model Training entity generates model deployments or updates for the Model Inference entity, which outputs inference results to an Actor while supporting model performance feedback. Fig. 5B illustrates another known functional framework specifically for the New Radio (NR) air interface, detailing interactions where data collection provides training data, monitoring data, and inference data to respective model training, management, and inference nodes, while updated models are transferred from model storage.
[0057] Generally speaking, similar events in Figs. 6-11 are labeled with similar reference numbers that share two least significant digits, with differences discussed below where appropriate. For example, event 610 is similar to event 710, and event 612 is similar to event 712.
[0058] Referring first to Fig. 6, a scenario 600 involves the UE 102 and the network negotiating support of AI / ML enabled feature IDs using a non-access stratum (NAS) Registration Request procedure. The scenario of Fig. 6 is consistent with triggering the AI / ML enabled feature via an N2 message originated from an O&M or an AF request, received from the AMF 364 or from an UDM / PCF / DCF via the AMF. The CN 110 can use one or more AI / ML enabled feature ID(s), and the CN can determine the triggering of each AI / ML enabled feature based on the negotiation of UE capability and network support, which can include CN support and RAN support.
[0059] During a non-UE specific RAN configuration update procedure, the RAN 105 indicates support of specific AI / ML enabled feature IDs. In particular, the RAN 105 sends 602, to the AMF 364, an uplink non-UE associated AI / ML feature transport message including the feature IDs of RAN-supported AI / ML enabled features.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0060] The UE 102 sends 610, to the AMF 364, a Registration Request message that includes feature IDs of supported AI / ML enabled features, e.g., as part of the 5GMM capability of the UE 102. A UE capability relevant to an AI / ML enabled feature can include support of the corresponding AI / ML NR air interface functionalities or related measurements between the RAN node and the UE. The AMF 364 stores 610 the UE capabilities in the AMF UE context. The AMF 364 sends 612, to the UDM / UDR 308 / 352, subscription retrieval or update data. In particular, the AMF 364 can send, to the UDM 308, an Nudm_UECM_Registration message to register the UE at the UDM 308. The AMF 364 further can send, to the UDM 308, an Nudm_SDM_Get message to retrieve new AI / ML service subscription data containing the list of feature IDs and explicit indications of whether each feature is allowed or not allowed. The AMF 364 can sends an Nudm_SDM_Info or Nudm_SDM_Notifi cation message to store the UE capability with respect to AI / ML enabled features, as part of the AI / ML service subscription data or application data at the UDM / UDR 308 / 3 2.
[0061] If the UE supports a specific feature ID and subscription is allowed, the AMF 364 discovers and selects 614 a PCF 360 or DCF 382 to serve the UE by invoking an Nnrf_NFDiscoveiy service operation. The Nnrf_NFDiscovery operation can be implemented similar to clause 5.2.7.3.2 of TS 23.502 of TS 23.50. The AMF 364 can create a PCF context and DCF context for the UE association. The AMF 364 sends 614, to the PCF 360, an Npcf UEPolicyControl Create Request message including an indication of the AI / ML enabled feature. Alternatively, the AMF 364 sends 614, to the DCF 382, a special-purpose, dedicated request such as an Ndcf AIML featureRegister Request message including the feature ID(s). In response, the PCF 360 or DCF 382 sends 616, to the AMF 364, an Npcf_UEPolicyControl Create Response or Ndcf A I ML featureRegister Response message.
[0062] The AMF 364 then determines 617 the network support of the AI / ML enabled feature(s) based on the following factors: (i) UE capability as indicated in the registration request procedure, (ii) indication of the allowed AI / ML enabled feature(s) in the UE subscription, (iii) core network support of the AI / ML enabled feature(s), and (iv) the RAN support of the AI / ML enabled feature(s). Finally, the AMF 364 sends 630, to the UE 102, a Registration Accept message including an allowed AI / ML enabled feature ID(s) indication (e.g., as part of a 5GS Network Feature Support IE). Otherwise, the AMF 364 includes 630 a suitable cause code indicating a lack of subscription, RAN node support, or core network support. The UE 102PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 receives 630 the allowed indications and stores the information in the UE context, on per-feature-ID basis. In particular, the UE 102 stores, in the UE context for each AI / ML enabled feature, an indication of whether this AI / ML enabled feature is allowed or not allowed.
[0063] Thus, as illustrated in Fig. 6, the AMF 364 selects, for the UE 102, a PCF and a DCF and creates a PCF context and a DCF context, respectively, for the UE association. The AMF 364 and the UE 102 communicate regarding UE capability with respect to AI / ML enabled feature ID(s) via NAS messaging.
[0064] Fig. 7A illustrates a scenario 700 that involves combining the NAS registration request procedure with an RRC procedure to obtain UE capabilities for a list of feature IDs. Events 710, 712, 714, 716, and 717 are similar to events 610, 612, 614, 616, and 617 discussed with reference to Fig. 6.
[0065] The AMF 364 sends 720, to the UE 102, a Registration Accept message including the feature IDs of the allowed AI / ML enabled feature(s) or a suitable cause code indicating the reason the AI / ML enabled feature ID is not allowed (e.g., no subscription to the of AI / ML enabled feature, no support at the RAN node, or no support at the CN). The AMF 364 can include the feature IDs in a 5GS Network Feature Support IE. More specifically, the AMF 364 sends 720, to the RAN 105, an N2 Initial Context Setup Request message, induing the network supported AI / ML enabled feature IDs and aNAS-PDU containing the Registration Accept message. The RAN 105 stores 722 the received network-supported IDs. The RAN 105 sends 730 an RRC message including the Registration Accept message to the UE 102, which stores the allowed feature IDs.
[0066] To query for granular capabilities, the RAN node can send 732, to the UE 102, a UECapabilityEnquiry message with either a general AI / ML enabled feature indication (not referring to any particular AI / ML enabled feature by feature ID) or a specific AI / ML enabled feature ID. The UE 102 responds 724 with a UECapability Information message, indicating the capability of the UE 102 with respect to one or more specific feature IDs (when the UECapabilityEnquiry includes the general AI / ML enabled feature indication) or the capability of the UE 102 with respect to the specific AI / ML enabled feature identified by feature ID in the UECapabilityEnquiry message. In the latter case, the UECapabilitylnformation can include simply a supported / non-supported indication. The RAN 105 stores and updates 736 the networkPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 supported AI / ML enabled feature IDs. The RAN 105, an O&M, or an N2 message from the CN 110 (e g., during a PDU Session Establishment / Modifi cation request procedure) can trigger this capability update.
[0067] The scenario 700B is generally similar to that of Fig. 7A, except that here the UE 102 informs the network regarding its capabilities with respect to one or more AI / ML enabled features by including, in the Registration Request message, an AI / ML enabled feature indication (rather than the feature ID(s) as in the events 610 and 710). This allows the network to later trigger and configure AI / ML features accordingly. Similar to scenarios 600 and 700A, a UE capability relevant to an AI / ML enabled feature can include support of the corresponding AI / ML NR air interface functionalities or related measurements between the RAN node and the UE. As part of the registration procedure, and when the CN 110 indicates support of an AI / ML enabled feature in a corresponding N2 message, the RAN 105 can query the UE 102 for UE capability with respect to relevant AI / ML enabled feature ID(s), using RRC messages, and store the indication(s) at the RAN 105 as a part of the UE context.
[0068] Thus, in the scenario 700B, one AI / ML enabled feature indication represents one or more AI / ML enabled feature(s) in the CN 110. The triggering of each AI / ML enabled feature is determined between the NG-RAN and UE. The AMF 264 determines network support of an AI / ML enabled feature based on the following factors: (i) the UE capability as reflected in the AI / ML enabled feature indication included in Registration request procedure, (ii) the subscription of the UE to the AI / ML enabled feature, and (iii) support of the AI / ML enabled feature at the CN 110. In view of the support indication of an AI / ML enabled feature in an N2 message, the RAN 105 can query the UE for its capability with respect to the AI / ML enabled feature, identifiable by a feature ID as in the examples above, using RRC messages. The RAN 105 stores the UE capability in the UE context at a RAN node. The UE 102 stores, in the UE context, an indication of whether the AI / ML enabled feature is allowed or not allowed. The scenario of Fig. 7B is consistent with triggering the AI / ML enabled feature by an O&M, by an N2 message from the PCF / DCF via the AMF 364 and originated via an AF request, or by an N2 message from the SMF 366, via the AMF 364, based on UE request.
[0069] In particular, The UE 102 sends 711 a Registration Request message including an AI / ML enabled feature indication as part of its 5GMM capability, which the AMF 364 stores inPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 the AMF UE context. The AMF 364 selects a DCF to create a DCF context. The AMF 364 sends 712 an Nudm_UECM_Regi strati on message to the UDM 308 to register 712 the UE at the UDM 308. The AMF 364 further sends 712, to the UDM 308, an Nudm_SDM_Get message to request retrieval of UE subscription data. In this scenario, the UDM / UDR 308 / 352 store AI / ML service subscription data, which includes a list of AI / ML enabled feature indications, each indicating whether the respective AI / ML enabled feature is allowed or not allowed for the UE 102. Still further, the AMF 364 sends 712, to the UDM 308, an Nudm_SDM_Info message or Nudm_SDM_Noti fi cation message including AI / ML enabled feature ID(s) to store UE capability for the AI / ML enabled feature indication as part of the AI / ML service subscription data, or as part of application data (stored at the UDR 352, for example).
[0070] The AMF 364 determines 717 network support of the AI / ML enabled feature based on whether the UE 102 provided the AI / ML enabled feature indication and whether the UDM 308 indicates that the AI / ML enabled feature is allowed for the UE 102.
[0071] The AMF 364 sends 721, to the UE 102, a Registration Accept message including an allowed AI / ML enabled feature indication or a relevant cause code, included for example in an 5GS Network Feature Support IE. In particular, the AMF 364 sends 721, to the RAN 105, an N2 Initial Context Setup Request message. This message indicates network support of the AI / ML enabled feature indication and a NAS-PDU IE including a Registration Accept message. The RAN 105 stores 723 the received network supported AI / ML enabled feature indication. The RAN 105 then sends 731, to the UE 102, an RRC message including the Registration Accept message and the allowed AI / ML enabled feature indication. The UE 102 stores the allowed AI / ML enabled feature indication locally.
[0072] The RAN 105 then proceeds with granular capability negotiation. To this end, the RAN 105 sends 732, to the UE 105, a UECapabilityEnquiry message inluding either a general indication or a specific feature ID. The UE 102 responds 734 with the UEC apability Information message. The RAN node stores 736 the specific supported AI / ML enabled feature IDs, allowing future triggering by the NG-RAN, O&M, or N2 messages.
[0073] Now referring to Fig. 8A, here the AF 358 uses an Nnef_AFsessionWithQoS or Npcf PolicyAuthorization Create service operation to request UE data collection and transfer, to the AS 331, the collected data for AI / ML training for standardized datasets related to a specificPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 AI / ML enabled feature. The AF 358 and the AS 331 can identify the AI / ML feature by a feature ID. The UE 102 and the RAN 105 can trigger data collection for the AI / ML enabled feature, which these entities also can identify by the feature ID, in those where the AF 358 has not already requested UP data collection.
[0074] After the UE 102 and the CN 110 perform 840 a PDU Session Establishment procedure (which can be implemented similar to steps 1-6 in clause 4.3.2.2.1 of TS 23.502), the AF 358 sends 842, to the NEF 354 or the PCF 360 or, an Nnef_AFsessionWithQoS_Update or Npcf PolicyAuthorization Create request. This request can pertain to the UE 102 or a group of UEs. The request can include one or more of (i) an AI / ML enabled feature ID, (ii) QoS parameters for UE data transfer, (iii) indication of UP data collection, or (iv) UE data report configuration information. The UE data report configuration can include the address and port of the AS 331 and subscription for a UE data report with a dedicated, special-purpose event ID for the DCF 382. The subscription also can indicate a reporting schedule (e.g., time, frequency). Using the Nnef_AFsessionWithQoS service in this manner allows the AF 358 to request, from the CN 110, to provide a specific QoS for UE data collection and transfer.
[0075] When the request 582 indicates the AI / ML enabled feature ID, the CN 110 can use the feature ID to trigger the specific AI / ML enabled feature at the RAN 105 and / or the UE 102, when not already enabled. Further, the SMF 366 can provide the AI / ML enabled feature ID to the RAN 105 along with the QoS profile, which the RAN 105 can use to enable the AI / ML enabled feature operation at the corresponding RAN node. The RAN node can perform an RRCreconfiguration procedure to enable the AI / ML enabled feature operation at the UE 102. Still further, the SMF 366 can provide the AI / ML enabled feature ID to the UE 102 along with the QoS rule, which the UE 102 can use to to enable the AI / ML enabled feature operation at the UE 102.
[0076] When the request 582 indicates UP data collection, the SMF 366 provides this indication to the UE 102 along with the QoS rule, which the UE 102 can use to enable UP data collection at the UE 102 for the associated AI / ML enabled feature. When the request 582 incl dues UE data report configuration information, the CN 110 can use this information to trigger the DCF 382 to transfer the collected UE data based on subscription of event exposure for associated AI / ML enabled feature.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0077] The PCF 360 obtains 846, from the UDM / UDR 308 / 352, information related to UE capability with respect to the AI / ML enabled feature (corresponding to the indicated feature ID) and the DCF address information. The PCF 360 also checks user consent for data collection for the purpose of the AI / ML enabled feature for NR air interface, using a data-key such as a Subscription Permanent Identifier (SUPI) of the UE and / or the AI / ML enabled feature ID. The PCF 360 proceeds with the steps illustrated in Fig. 8A if the user consent indicates that the data collection for the AI / ML enabled feature for the NR air interface. Otherwise, the PCF 360 replies to the AF 358 with a suitable cause value, to indicate lack of user consent for the AI / ML enabled feature for the NR air interface.
[0078] The SMF 366 and the PCF 360 then initiate 848 an SM Policy Association Modification procedure for the PDU session. The PCF 360 provides a PCC rule to the SMF 366 for the impacted PDU session for the UE 102 (or multiple PCC rules for the impacted PDU sessions for a group of UEs). The PCC rule includes one or more of the following, for the AI / ML enabled feature: (i) an AI / ML enabled feature ID, (ii) QoS parameters for the UE data transfer, (iii) indication of UE data collection, (iv) DCF address information, which the PCF 360 can retrieve from the UDM 308 if available, and (v) UE data report configuration information, which can include the address and port of the AS 331 for UE data transfer from the DCF 382 to the AS 331 and subscription for a UE data report with a dedicated, special-purpose event ID for the DCF 382.
[0079] Based on the received PCC rule and / or on local policies, the SMF 366 determines 849 to enable the AI / ML enabled feature and the corresponding UP data collection by configuring each of the UE 102, the RAN 105, and the UPF 370 with the QoS information derived from the PCC rule, and additionally configures the DCF 382 with information related to UP data transfer from the UE 102, as well as a UE data report to the AS 331.
[0080] The SMF 366 performs 850 an SMF-initiated PDU Session Establishment / Modification procedure. In particular, the SMF 366 sends 851, to the AMF 364, an Namf_Communication_NlN2MessageTransfer message including an N2 container enclosing a QoS profile and the AI / ML enabled feature ID, and N1 container enclosing a QoS rule. The RAN node 105 then receives 852, from the AMF 364, the N2 PDU Session Request message and retrieves the QoS profile. The SMF 364 configures 851-853 the UE 102 with the QoS rule, thePATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 AI / ML enabled feature ID, and the indication of UP data collection included in PDU Session Establishment Accept or Modification Command message. To this end, the SMF 366 includes the QoS in the N1 container as indicated above and sends 851 the N1 container in the Namf Communication NlN2MessageTransfer message to the AMF 366. The RAN 105 receives 852 the N2 PDU Session Request message and forwards 853, to the UE 102, the PDU Session Establishment Accept or the PDU Session Modification Command message. The RAN 105 responds 854 with an N2 PDU Session Response message, and the AMF 364 and the SMF 366 complete 855 the PDU Session Establishment / Modification procedure with Nsmf PDU Session_updateSmContext Request / Response messages.
[0081] Based on the QoS profile and the AI / ML enabled feature ID, the RAN 105 may query the UE capability with respect to the AI / ML enabled feature and send, to the UE 102, an RRCReconfiguration message in response to determining to enable the AI / ML enabled feature operation (e.g. the AI / ML enabled feature needs inferencing input from a RAN node). In some implementations, the RAN 105 includes AI / ML enabled feature ID as a component of the QoS profile.
[0082] Based on the QoS rule, the AI / ML enabled feature ID, and indication of UE data collection , the UE 102 can start the corresponding AI / ML enabled feature operation and data collection of the standardized data of the corresponding AI / ML enabled feature. In some implementations, the QoS rule includes the AI / ML enabled feature ID. In some implementations, the QoS rule includes the indication of UE data collection. In some implementations, the QoS rule includes both the AI / ML enabled feature ID and the indication of UE data collection.
[0083] With continued reference to Fig. 8A, the SMF 366 configures 860 the PSA UPF 370 by sending an N4 Session Modification Request message including N4 rules with the following information for the uplink direction: (i) a packet detection rule (PDR), which the UPF 370 can use to detect data traffic flow with the destination address set to the address and port number of the DCF 382, which turn is associated with the AI / ML enabled feature ID, (ii) a forwarding action rule (FAR) that includes the address and port number of the DCF 382, and (iii) a QoS enforcement rule (QER) used to associate the data traffic flow with the proper QoS parameters for UP data transfer, for the AI / ML enabled feature. When traffic detection is based on the PDR,PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 the PSA UPF 370 enforces the FAR that defines how the detected packet is to be forwarded. The FAR includes, according to the option that enhances the existing attributes with new settings, (i) the destination interface set as at core-side and (ii) the forwarding policy that indicates steering traffic toward the target DCF 382. The FAR includes, according to another option, a new (dedicated, special-purpose) attribute to indicate in-network forwarding information for the target DCF 382.
[0084] The SMF 366 also can configure N4 rules for a UE data report to the AS 331, with the following information for the uplink direction: (i) a PDR to detect the data traffic flow with source address set as DCF address and port number for a UE data report associated with the AI / ML enabled feature ID, and (ii) a FAR that includes the AS address and port number.
[0085] The SMF 366 discovers 862 a suitable DCF (e.g., the DCF 382), if the DCF address information is not available in the PCC rules, and configures the DCF 382 using a new (dedicated, special-purpose) Ndcf service operation with the following information (i) the AI / ML feature ID, (ii) connection information for data transfer including the AS address information and the PSA UPF information, and (iii) the UE data reporting schedule information, e.g. time, frequency. Regarding the connection information, if the SMF 366 provides only the AS address information is provided, the DCF 382 can request a secure connection with the AS 331; if the SMF 366 provides both the PSA UPF information and the AS address information, the DCF 382 can request a secure connection with the AS 331 via the PSA UPF 370. In this case, the PSA UPF 370 can request a secure connection with the AS 331.
[0086] The RAN 105 and the UE 102 start or activate 868 the AI / ML enabled feature identified by the AI / ML enabled feature ID, if not already started. The UE 102 then performs 870 data collection of the AI / ML enabled feature.
[0087] The UE 102 transfers 880 the collected data to DCF 382 via the PSA UPF 370. Based on the configured N4 rules, when the PSA UPF 370 detects a data traffic from the UE 102 based on the PDR, the PSA UPF 370 forwards the data traffic to the configured DCF address indicated in the FAR. When the DCF 382 receives 890 the data traffic from the UE 102 via the PSA UPF 370, the DCF 382 may perform the following based on the information related to the configured UP data transfer schedule information such as time and frequency (see event 862): if the SMF 366 configured only the AS address information, the DCF 382 requests a secure connection withPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 the AS 331 directly over a new (dedicated) interface between the DCF 382 and the AS 331; if the SMF 366 configured both the PSA UPF information and the AS address information, the DCF 382 requests a secure connection with the AS 331 via PSA UPF 270. In the latter case, the PSA UPF 370 requests a secure connection with the AS 331. The PSA UPF 370 forwards the data traffic received from the DCF 382 to the AS 331.
[0088] Now referring to Fig. 8B, another technique for the AF 358 triggering data collection for an AI / ML enabled feature involves using a dedicated AF-triggered data collection application programming interface (API). The following is contemplated for AF-requested UP data collection: (i) a dedicated API for AF-requested UP Data collection, and (ii) support for both direct AF-to-DCF and mediated AF-to-DCF-via-PCF interaction models.
[0089] While the technique illustrated in Fig. 8A leverages existing 5GS service operations such as Nnef_AFsessionWithQoS and Npcf_PolicyAuthorizationCreate, the technique illustrated in Fig. 8B uses a new, dedicated service-based API, such as Naf_DataCollectionService, to provide a standardized and extensible entry point for all AF-triggered UP data collection for AIML-enabled features. This dedicated service provides an API for application developers and allows the network to expose data collection as a native capability. This dedicated service has sufficient flexibility to support both operator-controlled (mediated) and trusted direct-access models, to support both 5G and 6G architectures.
[0090] Aspects of technique of Fig. 8B include the following: (i) a unified API, which provides a single, purpose-built API for all AF-triggered data collection requests, whether for immediate data transfer or for provisioning UE policies (URSP) for subsequent UE-triggered transfers; (ii) a rich data model, so that the API carries a structured set of parameters including a feature ID, QoS Parameters, and a flexible Traffic Descriptor (TD). This allows the AF 358 to precisely specify its requirements; and (iii) architectural flexibility, so that different NFs can expose the Naf_DataCollectionService to support different trust models.
[0091] One trust model is a mediated model (e.g., a PCF-mediated model). The PCF 360 exposes the service and operates as a policy enforcement point. Generally speaking, this is the default model for untrusted or third-party AFs. Another trust model is a direct model: the DCF 382 directly exposes the service for trusted, intra-operator, or pre-authorized AFs, enabling a more streamlined interaction. Both models allow for a QoS framework reuse. In particular, thePATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 existing PCF-SMF-UPF chain enforces the QoS and configures the user plane, ensuring no changes are needed to the core 5G QoS mechanisms.
[0092] In an example deployment, The Naf_DataCollectionService is defined with the following key service operations: (i) create, (ii) update, (iii) delete, and (iv) subscribe or notify. An AF care involve the create operation to request the creation of a new data collection session or to provision a UE policy. The AF can invoke the update operation to modify parameters of an active data collection task (e.g., change the reporting frequency), invoke the delete operation to terminate a data collection task, and the subscribe / notijy operation to subscribe to events such as the completion of data collection or error notifications.
[0093] The create operation can involve the following parameters: of 11). which is an identifier of the requesting AF, targetUEs, which is a list of target UEs (identified SUPI or Generic Public Subscription Identifier (GP SI)), featur elD, which is a unique identifier for the AI / ML-enabled feature, used for end-to-end correlation, serviceMode, which is an enumeration indicating the requested mode such as IMMEDIATE (for immediate data collection) or PROVISION POLICY (to configure URSP rules for a later UE-triggered session), qosParameters, which is a structured object containing the required QoS characteristics for the data flow (5QI, GBR, MBR, ARP), trafficDescriptor, which is a flexible structure allowing the AF to specify how to identify the data traffic, including options for a target DCF address (FQDN / IP), a "data collection" Connection Capability, or other standard 5-tuple identifiers, and dataReportingConfig, which is information regarding where and how a device should report the collected data (e g., the AS address, reporting schedule).
[0094] In an example deployment based on the techniques of Fig. 8A, the PCF 360 exposes the Naf DataCollectionService operation. The AF 358 sends, to the PCF 360, a create request. The PCF 360 performs the necessary policy checks, user consent verification (e.g., via the UDM 308), and service authorization. If authorized, the PCF 360 triggers the SMF 366 to set up a PDU session and the QoS, and instructs the DCF 372 (e.g., via an Ndcf_DataCollectionSetup service) to prepare for the data transfer. This technique ensures maximum operator control and may be the preferred approach for third-party AFs and untrusted AFs.
[0095] For a trusted environment (e.g., when the AF 358 is a part of the internal operations of the mobile network operator (MNO) that controls the CN 110), the DCF 382 can directly exposePATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 the Naf_DataCollectionService. The AF 358 in this case sends a create request directly to the DCF 382, which is responsible for performing a simplified authorization check (based on a preconfigured trust relationship) and user consent verification (e.g., via the UDM 308). To set up the QoS path, the DCF 382 can interact with the PCF 360 using an Npcf ' DataCol lection Pol icy service, effectively delegating the QoS setup task. The PCF 360 receives this request, creates the PCC rules, and triggers the SMF 366 as in the scenarios discussed above. This technique streamlines the initial signaling path for trusted applications while still leveraging the standard QoS framework.
[0096] Next, Fig 8B illustrates a procedure consistent with the direct (trusted) model, which assumes a trusted relationship between the AF 358 and the DCF 382. The AF 358 can discover the DCF 382 via the NRF 306 or local configuration.
[0097] As illustrated in Fig. 8B, a scenario 800B begins with a procedure 840, similar to the scenario 800A. The AF 358, which in this scenario is a trusted AF, initiates 841 a DCF service operation, Ndcf_DataCollectionService_Create request, directly with the DCF 382. This is a dedicated, purpose-built service operation designed for data collection. The request contains all necessary service requirements such as the feature ID, the one or more target UEs, the desired QoS Parameters (5Q1, GBR, etc.), and a flexible TD (e g., target DCF FQDN) to identify the data flow.
[0098] The DCF 382 performs 843 authorization and user consent check. To this end, the DCF 382 sends 844, to the UDM 308, a Nudm_SDM_Get request to retrieve the user’s subscription data and explicitly verify that consent has been granted for this data collection service, using the feature ID as a key. This user consent check is similar to the user consent check the PCF 360 performs 846 in the scenario 800A, but here it is the DCF 382 that performs the check. In response, the UDM 308 sends 844, to the DCF 382, a Nudm_SDM_Get response to confirm that the user consent is granted. If consent is not granted, the DCF 382 rejects the request 841. This is also similar to the response the PCF 360 receives 846 in the scenario 800A, but here the UDM 308 sends the response to the DCF 382.
[0099] After confirming user consent, the DCF 382 delegates the task of policy and QoS enforcement by sending 845, to the PCF 360, a request such as an Npcf DataCollectionPolicy Create request. This special -purpose service operation provides thePATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 PCF 360 with the necessary policy context, including the feature ID, the TD, and the QoS Parameters received 841 from the AF 358. Unlike the scenario of Fig. 8A, this operation involves a DCF-to-PCF interaction path. This operation can trigger a backend policy mechanism similar to that discussed with reference to Fig. 8A.
[0100] The PCF 360 receives 845 the request and performs 846 policy decision-making. The PCF 360 can combine the request parameters with the subscription profile of the UE 102 and the local policies of the MNO associated with the CN 110 to generate the PCC rules needed for this service, similar to the scenario 800A.
[0101] Next, after performing several operations discussed with reference to Fig. 8A, the SMF 366 sends 861, to the PCF 360, a successful response for the Npcf_SMPolicyControl_UpdateNotify service operation received 848 earlier. This explicit confirmation completes the policy enforcement transaction and signals to the PCF 360 that the PCC rules have been successfully installed and the user plane path is ready. The PCF 360, after receiving 361 the confirmation that the data path is ready, sends 863 a successfulNpcf DataCollectionPolicy Create Response back to the DCF 382. This transmission completes the service request that the DCF 382 initiated 845 earlier. This step can be important for proper a SBA request-response handling between the DCF 382 and the PCF 360.
[0102] The DCF 382 at this point has confirmed that the end-to-end policy and data path are established. The DCF 382 sends 865, to the AF 358, the finalNaf DataCollectionService Create Response, to confirm the successful setup of the data collection service. The transmission 865 provides the final confirmation to the original requester and is a feature of the direct model discussed above.
[0103] As indicated above, when an AI / ML enabled feature is enabled, the UE 102 needs to request UP data collection for an AI / ML model training in the network / OTT server in order to obtain a customized AI / ML model for the UE 102. In some deployments, the CN 110 enhances the UE policy as discussed below. These enhancements can include one or more of following changes to URSP rule provisioning by the PCF 360 or the DCF 382 via the PCF 360, or a new DCRP rule provisioning by the DCF 382.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0104] One option is based on a TD enhancement. The TD can include the existing IP descriptors indicating the destination DCF address, port number, and a protocol ID for a secure UP connection. This technique assumes that the UE 102 is configured with the address and port number of the DCF 382 via a DL NAS Transport procedure, or that the UE 102 received application layer information from the AF / AS 358 / 331 supporting AI / ML model training. When the UE 102 requests communication of UP data traffic, the UE 102 establishes the PDU Session Establishment / Modification request procedure based on the Route Selection Descriptor (RSD).
[0105] Another option is also based on a TD enhancement. In this case, the TD includes Connection Capabilities (CC) indicating a dedicated standardized component value as UP data collection. When the UE 102 communication of UP data traffic, the UE 102 performs the PDU Session Establishment / Modification request procedure based on the RSD for the matched TD indicating the CC for UP data collection.
[0106] Yet another option is based on a TD enhancement. The TD includes a dedicated component of AI / ML enabled feature ID(s) for data traffic related to UP data collection and transfer. This technique assumes that the UE 102 is configured with allowed AI / ML enabled feature ID(s) during a registration procedure or during a DL NAS Transport procedure, or that the UE 102 received application layer information from the AF / AS 358 / 331 supporting AI / ML model training. When the UE requests for transferring UP data traffic, the UE establishes the PDU Session Establishment / Modification request procedure based on RSD for the matched TD indicating the AI / ML enabled feature ID.
[0107] Another option is based on an RSD enhancement. The RSD indicates the following information associated with PDU session related information: a dedicated Data Network Name (DNN) and a dedicated Single Network Slice Selection Assistance Information (S-NSSAI) for UP data collection.
[0108] Referring to Fig. 9, the UE 102 performs 910 a registration procedure to negotiate UE capability with respect to an AI / ML enabled feature over the NR air interface and network support. This procedure can include the operations described in TS 23.502 clause 4.2.2.2 as well as the steps discussed above with reference to Figs. 6-7B.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0109] If the network supports UE policy handling, the PCF 360 or the DCF 382 performs 914 UE Configuration Update procedure to provision UE policies for the UE 102. The DCF 382 can perform the UE Configuration Update procedure directly or via PCF 360 by sending, to the UE 102 via AMF 364, an Namf Communication NlN2MessageTransfer message to provision session management UE policies, e.g. URSP rules enhanced for UP data collection or a DCRP rule modified as discussed above), when the PCF / DCF 360 / 382 decides to update UE policy based on triggering conditions. The triggering condition can include one or both: (i) the PCF / DCF 360 / 382 determines to update the UE policy, e.g. based on operator policies or O&M, or (ii) the PCF / DCF 360 / 382 receives an AF request for UP data collection and transfer at the UE for a specific AI / ML enabled feature that uses AI / ML model training service at the network / OTT server.
[0110] Based on the triggering condition for UE policy evaluation, e.g. when the UE receives 970 an upper layer (application) or lower-layer (e.g. RRC / MAC / SDAP) request or a NAS message from the network, for enabling an AI / ML enabled feature and the corresponding UP data collection of the UE side AI / ML model inferencing, the UE 102 evaluates UE policies for UP data collection (as discussed above. If the UE 102 finds a matched UE policy, the UE 102 determines whether to perform a PDU Session Establishment / Modification request procedure for UP data collection related to one or more AI / ML enabled features, and UP data transfer to the network, e.g., for AI / ML model training service at an OTT server.
[0111] Based on the RSD of the matched UE policy, the UE 102 enforces the UE policy enhanced for UP data collection and performs 940 a PDU Session Establishment / Modification request procedure (e.g., similar to the procedure described in TS 23.502 clause 4.3.2. A part of the PDU Session Establishment / Modification request procedure can include receiving an AF request for a required QoS update for UP data collection related to a specific AI / ML enabled feature, identified by a feature ID. Based on the PCC rules the PCF 360 provided, the SMF 366 performs QoS flow binding, configures the UE 102 with a QoS rule, configures the RAN 105 with a QoS profile, configured the PSA UPF 370 with N4 rules, and configured the DCF 382 with the AI / ML enabled feature ID and AS address information. The UE 102 receives a PDU Session Establishment Accept message or a PDU Session Modification Command message from the network, which confirms the enabling of the AI / ML enabled feature, UP data collection, andPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 UP data transfer using the PDU Session to the DCF 382 via PSA UPF 370. The DCF 382 reports stored UE data to the AS directly or via PSA UPF 360, based on the subscribed event for UE data transfer.
[0112] Further, the following techniques for triggering AI / ML enabled feature ID(s) is contemplated. According to one approach based on the technique of Figs. 6 and 7A (which involves exchanging allowed AI / ML enabled feature ID(s) for the NR air-interface), the RAN 105 can trigger an AI / ML enabled feature based on (i) an N2 message from the AMF 364, originated at an O&M server and initially sent to the AMF, (ii) an N2 message from the UDM / PCF / DCF 308 / 360 / 382, received via the AMF 364, which originated from an AF request or from an O&M server to the UDM / PCF / DCF 308 / 360 / 382, or (iii) an N2 message received from the SMF 366 via the AMF 364 in, which is originated from an AF request (see Fig. 8A) or from an O&M server to the UDM / PCF / DCF 308 / 360 / 382.
[0113] According to another approach based on the technique of Fig. 7B (which involves exchanging of an allowed AI / ML enabled feature for the NR air-interface), the RAN 105 can send, to the UE 102, an RRC request message to obtain UE capability with respect to AI / ML enabled feature(s) and received, from the UE 102, an RRC response message including the supported / unsupported information corresponding to the UE capability with respect to the AI / ML enabled feature ID(s). The RAN 105 can use this mechanism when the capability is not already available and stored at the RAN as part of the UE context.
[0114] In any case, the RAN 105 stores the indication of of UE capability with respect to the AI / ML enabled feature ID(s) in the UE context. The trigger for this action at the RAN can be based on (i) a request from an O&M server, (ii) an N2 message from the UDM / PCF / DCF 308 / 360 / 382 transmitted via the AMF 364 and originated in response to a request from the AF 358, or (iii) an N2 message from the SMF 364 transmitted via the AMF 364 and originated in response to a request from the AF 358 or from an O&M server transmitted to the SMF 366.
[0115] Referring back to Figs. 8 A and 8B, the UE 102 and the DCF 382 can exchange UE Data Transfer Control Messages (UEDT-CMs) to initiate, terminate and manage UE data transfer over the UP connection, for an AI / ML enabled feature. Each control message may contain UE ID, e.g. SUPI, AI / ML enabled feature ID, Validity time period, Validity areas.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0116] Referring to Fig. 10, a method 1000 can be implemented in a DCF such as the DCF 382. At block 1081, the DCF generates UE Data Transfer Control Message (UEDT-CM) including one or more of (i) a UE ID, (ii) feature ID of the AI / ML enabled feature, (iii) validity time period, or (iv) validity area. At block 1082, the DCF transmits, to the UE via the RAN, the UEDT-CM. At block 1083, the DCF received, from the UE via the RAN, a second UEDT-CM.
[0117] A UEDT-CM can be one of the following: (i) UEDT Connection Establishment request, used to request for establishment of a new UP connection for UE data transfer, (ii) UEDT Connection Establishment Complete, used to acknowledge for the establishment completion of a UP connection for UE data transfer, (iii) UEDT Connection Release request, used to request for releasing UP connection for UE data transfer, (iv) UEDT Connection Release Confirm, used to acknowledge for confirming the release of UP Collection for UE data transfer, (v) UEDT Start request, used to request for starting UE data transfer, (vi) UEDT Start Confirm, used to acknowledge for starting a UE data transfer, (vii) UEDT Suspend request, used to request for suspending UE data transfer, (viii) UEDT Suspend Confirm, used to acknowledge for the confirmation of suspending UE data transfer, (ix) UEDT Resume request, used to request for resuming UE data transfer, (x) UEDT Resume Confirm, used to acknowledge for the confirmation of resuming UP data transfer, (xi) UE Data Collection Status Update indication, used to request status update for the UP Data Collection for UE data transfer (this UEDT-CM can additionally indicate what status information is requested for the update, e.g. data volume of data collected, transferred, time window of the collected data to be transferred, earliest timestamp of the collected data to be transferred, latest timestamp of the collected data to be transferred, last timestamp of the data being transferred), or (xii) UP Data Collection Status Update response indication, used to respond status update for the UP Data Collection (this UEDT-CM can respond the status information requested for the update, e.g. data volume of data collected, transferred, time window of the collected data to be transferred, earliest timestamp of the collected data to be transferred, latest timestamp of the collected data to be transferred, last timestamp of the data being transferred, etc. for UP data collection of AIML enabled features).
[0118] Finally, Fig. 11 illustrates a method 1100 for configuring an AI / ML enabled feature, which can be implemented in one or more NFs of the CN 110. At block 1102, the one or more NFs receive, from an AF associated with an AI / ML feature, a request related to collecting dataPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 related to the AI / ML feature and the feature ID of the AI / ML enabled feature (see, e g., Fig. 8A event 842, Fig. 8B event 841). At block 1106, the one or more NFs configure collection of data related to the AI / ML feature via a UPF (see, e.g., Fig. 8A or Fig. 8B event 848).
[0119] The following list of examples reflects a variety of the embodiments explicitly contemplated by the present disclosure.
[0120] Example 1. A method implemented in a network function (NF) associated with a cellular communication system, the method comprising: determining a capability of a user equipment (UE) with respect to an artificial intelligence and / or machine learning (AI / ML) enabled feature; and receiving, from the UE and in accordance with a trigger event, data related to the AI / ML enabled feature.
[0121] Example 2. The method of example 1, wherein the data related to the AL / ML enabled feature is inference data related to an AI / ML model with which the AI / ML enabled feature is associated.
[0122] Example 3. The method of example 1 or 2, wherein the AI / ML enabled feature is associated with channel state information (CSI) compression.
[0123] Example 4. The method of example 1 or 2, wherein the AI / ML enabled feature is associated with CSI reference signal (CSLRS) overhead reduction.
[0124] Example 5. The method of example 1 or 2, wherein the AI / ML enabled feature is associated with CSI prediction.
[0125] Example 6. The method of example 1 or 2, wherein the AI / ML enabled feature is associated with beam management with downlink beam prediction.
[0126] Example 7. The method of example 1 or 2, wherein the AI / ML enabled feature is associated with direct AI / ML positioning.
[0127] Example 8. The method of example 1 or 2, wherein the AI / ML enabled feature is associated with direct AI / ML assisted positioning for positioning accuracy enhancement.
[0128] Example 9. The method of any of the preceding examples, wherein: the receiving of the data related to the AI / ML enabled feature includes receiving a feature identifier (ID) identifying the AI / ML enabled feature.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0129] Example 10. The method of any of the preceding examples, wherein: the receiving of the data related to the AI / ML enabled feature includes receiving a model ID identifying the AI / ML model with which the AI / ML enabled feature is associated.
[0130] Example 11. The method of any of the preceding examples, further comprising: transmitting, to the UE, a trigger command, wherein the trigger event includes the trigger command.
[0131] Example 12. The method of example 11, wherein the transmitting of the trigger command includes: transmitting, from an Access and Mobility Management Function (AMF) to a radio access network (RAN), an N2 message.
[0132] Example 13. The method of example 11, wherein the transmitting of the trigger command includes: transmitting, from a subsystem including one or more of a Unified Data Management Function (UDM), a Policy Control Function (PCF), or a Distributed Coordination Function (DCF), to a radio access network (RAN), an N2 message.
[0133] Example 14. The method of example 12 or 13, wherein the trigger command originates at an Operations and Maintenance (O&M) node.
[0134] Example 15. The method of any of examples 1-10, wherein: the trigger event is a preconfigured condition associated with the AI / ML enabled feature.
[0135] Example 16. The method of any of example 15, wherein the preconfigured condition is default collection of data for the AI / ML enabled feature.
[0136] Example 17. The method of any of the preceding examples, the method further comprising: transmitting, to the UE, a data collection command to trigger, at the UE, data collection related to the AI / ML enabled feature.
[0137] Example 18. The method of any of the preceding examples, further comprising: negotiating, with the UE, capability of the UE with respect to the AI / ML enabled feature.
[0138] Example 19. The method of example 18, wherein the negotiating includes: receiving, from the UE, a registration request including an indication of whether the UE supports the AI / ML enabled feature.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0139] Example 20. The method of example 19, further comprising: transmitting, to the UE, a registration accept message including an indication of whether the CN supports the AI / ML enabled feature.
[0140] Example 21. The method of example 18, wherein: the negotiating includes using a non-access stratum (NAS) registration request procedure.
[0141] Example 22. The method of example 2, further comprising: transmitting, to the UE, a UE instance of the AI / ML model.
[0142] Example 23. The method of example 2 or 22, further comprising: training a network instance of the AI / ML model using the data related to the AI / ML enabled feature.
[0143] Example 24. The method of any of the preceding examples, wherein the receiving of the data related to the AI / ML enabled feature includes using a format defined specifically for the AI / ML enabled feature.
[0144] Example 25. The method of example 1, further comprising: triggering data collection for the AI / ML enabled feature using an Nnef_AFsessionWithQoS service operation.
[0145] Example 26. The method of any of the preceding examples, further comprising: triggering data collection for the AI / ML enabled feature using an Npcf_PolicyAuthorization_Create service operation.
[0146] Example 27. The method of example 1, further comprising: transmitting, to a different NF, a request to create a data collection session or to provision a UE policy.
[0147] Example 28. The method of example 11, wherein the request to create the data collection session includes one or more of an identifier of the NF, a list of target UEs; an AI / ML enabled feature identifier, a service mode identifier, quality of service (QoS) characteristics, traffic descriptors, and information on where to report collected data.
[0148] Example 29. The method of example 11, wherein the different NF is a data collection function (DCF).
[0149] Example 30. The method of 29, wherein the DCF delegates policy and QoS enforcement to a policy control function (PCF).PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0150] Example 31. The method of example 30, wherien the DCF delegates policy and QoS enforcement using a Npcf DataCollectionPolicy Create request message.
[0151] Example 32. A method implemented in a user equipment (UE), the method comprising: detecting a trigger event related to transferring, to a core network (CN), data related to an artificial intelligence and / or machine learning (AI / ML) enabled feature; and in response to the trigger event, transmitting the data to the CN.
[0152] Example 33. The method of example 32, wherein the data is inference data, the method further comprising: generating, at the UE, the inference data for the AI / ML enabled feature using an AI / ML model.
[0153] Example 34. The method of example 32 or 33, wherein the data related to the AL / ML enabled feature is inference data related to an AI / ML model with which the AI / ML enabled feature is associated.
[0154] Example 35. The method of any of examples 32-34, wherein the AI / ML enabled feature is associated with channel state information (CSI) compression.
[0155] Example 36. The method of any of examples 32-35, wherein the AI / ML enabled feature is associated with CSI reference signal (CSLRS) overhead reduction.
[0156] Example 37. The method of any of examples 32-35, wherein the AI / ML enabled feature is associated with CSI prediction.
[0157] Example 38. The method of any of examples 32-35, wherein the AI / ML enabled feature is associated with beam management with downlink beam prediction.
[0158] Example 39. The method of any of examples 32-35, wherein the AI / ML enabled feature is associated with direct AI / ML positioning.
[0159] Example 40. The method of any of examples 32-35, wherein the AI / ML enabled feature is associated with direct AI / ML assisted positioning for positioning accuracy enhancement.
[0160] Example 41. The method of any of examples 32-40, wherein: the transmitting of the data includes transmitting a feature identifier (ID) identifying the AI / ML enabled feature.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0161] Example 42. The method of any of examples 32-41, wherein: the transmitting of the data includes transmitting a model ID identifying the AI / ML model with which the AI / ML enabled feature is associated.
[0162] Example 43. The method of any of examples 32-42, further comprising: receiving, from the CN, a trigger command, wherein the trigger event includes the trigger command.
[0163] Example 44. The method of example 43, wherein the receiving of the trigger command includes: receiving, from an Access and Mobility Management Function (AMF) to a radio access network (RAN), an N2 message.
[0164] Example 45. The method of example 43, wherein the receiving of the trigger command includes: receiving, from a subsystem including one or more of a Unified Data Management Function (UDM), a Policy Control Function (PCF), or a Distributed Coordination Function (DCF), via a radio access network (RAN), an N2 message.
[0165] Example 46. The method of example 44 or 45, wherein the trigger command originates at an Operations and Maintenance (O&M) node.
[0166] Example 47. The method of any of examples 32-42, wherein: the trigger event is a preconfigured condition associated with the AI / ML enabled feature.
[0167] Example 48. The method of any of example 47, wherein the preconfigured condition is default collection of data for the AI / ML enabled feature.
[0168] Example 49. The method of any of examples 32-48, further comprising: receiving, from the CN, a data collection command to trigger, at the UE, data collection related to the AI / ML enabled feature.
[0169] Example 50. The method of any of examples 32-48, further comprising: detecting a data collection event corresponding to a certain amount of data collected; and initiating data collection for the AI / ML enabled feature in response to detecting the data collection event.
[0170] Example 51. The method of any of examples 32-48, further comprising: detecting a data collection event corresponding to a certain amount of time elapsed relative to a reference time; and initiating data collection for the AI / ML enabled feature in response to detecting the data collection event.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0171] Example 52. The method of any of examples 32-51, further comprising: negotiating, with the CN, capability of the UE with respect to the AI / ML enabled feature.
[0172] Example 53. The method of example 52, wherein the negotiating includes: transmitting, to the CN, a registration request including an indication of whether the UE supports the AI / ML enabled feature.
[0173] Example 54. The method of example 53, further comprising: receiving, from the CN, a registration accept message including an indication of whether the CN supports the AI / ML enabled feature.
[0174] Example 55. The method of any of examples 32-54, wherein: the transmitting of the data related to the AI / ML enabled feature includes using a format defined specifically for the AI / ML enabled feature.
[0175] Example 56. An apparatus comprising processing hardware and configured to implement a method of any of the preceding examples.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01
[0176] Generally speaking, description for one of the above figures can apply to another of the above figures. Examples, implementations and methods described above can be combined, if there is no conflict. An event or block described above can be optional or omitted. For example, an event or block with dashed lines in the figures can be optional. In some implementations, “message” is used and can be replaced by “information element (IE)”, and vice versa. In some implementations, “IE” is used and can be replaced by “field,” and vice versa. In some implementations, “configuration” can be replaced by “configurations” or “configuration parameters”, and vice versa. The “attach” can be replaced with “registration.”
[0177] In some implementations, the “PDN connectivity procedure” can be replace by “PDU session establishment procedure.” In such cases, the “PDN Connectivity Request” and “Activate Default EPS Bearer Context Request” can be replaced by “PDU Session Establishment Request” and “PDU Session Establishment Accept”, respectively and the “Activate Default EPS Bearer Context Accept” is omitted.
[0178] A user device in which the techniques of this disclosure can be implemented (e.g., the UE 102) can be any suitable device capable of wireless communications such as a smartphone, a tablet computer, a laptop computer, a mobile gaming console, a point-of-sale (POS) terminal, a health monitoring device, a drone, a camera, a media-streaming dongle or another personal media device, a wearable device such as a smartwatch, a wireless hotspot, a femtocell, or a broadband router. Further, the user device in some cases may be embedded in an electronic system such as the head unit of a vehicle or an advanced driver assistance system (ADAS). Still further, the user device can operate as an internet-of-things (loT) device or a mobile-internet device (MID). Depending on the type, the user device can include one or more general-purpose processors, a computer-readable memory, a user interface, one or more network interfaces, one or more sensors, etc.
[0179] Certain embodiments are described in this disclosure as including logic or a number of components or modules. Modules may be software modules (e.g., code, or machine-readable instructions stored on non-transitory machine-readable medium) or hardware modules. A hardware module is a tangible unit capable of performing certain operations and may be configured or arranged in a certain manner. A hardware module can comprise dedicated circuitry or logic that is permanently configured (e.g., as a special-purpose processor, such as aPATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 field programmable gate array (FPGA) or an application-specific integrated circuit (ASIC), a digital signal processor (DSP), etc.) to perform certain operations. A hardware module may also comprise programmable logic or circuitry (e.g., as encompassed within a general-purpose processor or other programmable processor) that is temporarily configured by software to perform certain operations. The decision to implement a hardware module in dedicated and permanently configured circuitry, or in temporarily configured circuitry (e.g., configured by software) may be driven by cost and time considerations.
[0180] When implemented in software, the techniques can be provided as part of the operating system, a library used by multiple applications, a particular software application, etc. The software may be executed by one or more general-purpose processors or one or more specialpurpose processors.
Claims
PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 The claims;1. A method implemented in one or more network functions (NFs) of a core network (CN) in a cellular communication system, the method comprising:receiving, from an application function (AF) associated with an artificial intelligence and / or machine learning (AI / ML) feature, a request (i) related to collecting, from a user equipment (UE) via a radio access network (RAN), data related to the AI / ML enabled feature and (ii) including a feature identifier (ID) of the AI / ML enabled feature; andconfiguring the collecting of the data related to the AI / ML enabled feature, via a user plane function (UPF).
2. The method of claim 1, wherein the receiving of the request includes receiving, at a data collection function (DCF) a data collection service request dedicated to directly requesting AI / ML data collection from the DCF.
3. The method of claim 2, wherein the request includes a traffic descriptor to identify a data flow to which the data related to the AI / ML enabled feature belongs.
4. The method of claim 1, wherein the receiving of the request includes receiving one of:(i) at a policy control function (PCF), an Npcf PolicyAuthorization Create request, or (ii) at a network exposure function (NEF), an Nnef_AFsessionWithQoS request.
5. The method of claim 5, wherein the Npcf_PolicyAuthorization_Create request or the Nnef_AFsessionWithQoS request includes an indication of a server address and a port of an application server (AS), to which a DCF is to send the data related to the AI / ML enabled feature.
6. The method of any of the preceding claims, wherein the request includes an indication of desired quality-of-service (QoS) parameters for the collecting of the data related to the AI / ML enabled feature.PATENT APPLICATION Attorney Docket No.: 31730 / 309358-01 7. The method of claim 2, 3, or 5, further comprising:communicating, between the DCF and the UE and for the AI / ML enabled feature, a UE Data Transfer Control Message (UEDT-CM) indicating one of (i) a start of the collecting, (ii) a suspension of the collecting, or (iii) resumption of the collecting.
8. The method of any of the preceding claims, further comprising: transmitting, to the RAN, a QoS profile including the feature ID.
9. The method of any of the preceding claims, further comprising: transmitting, to the RAN, an N2 initial context setup request including an indication that the CN supports the AI / ML enabled feature.
10. The method of claim 9, wherein the indication that the CN supports the AI / ML enabled feature includes the feature ID.
11. The method of any of the preceding claims, further comprising:transmitting, to the UE, a configuration including (i) a QoS rule, (ii) the feature ID, and (iii) an indication of user-plane (UP) data collection.
12. The method of claim 11, wherein the configuration is transmitted in a protocol data unit (PDU) session establishment accept command or a PDU session modification command.
13. The method of any of claims claim 2, 3, 5, or 7, further comprising: providing, to the UE, a User Equipment Route Selection Policy (URSP) rule with a traffic descriptor (TD) that indicates (i) an address of the DCF, (ii) a port number at the DCF, and (iii) a protocol ID for a secure UP connection.
14. The method of any of claims claim 2, 3, 5, or 7, further comprising: providing, to the UE, a URSP rule with a TD that includes the feature ID, in a field dedicated to UP data collection .PATENT APPLICATION Attorney Docket No.: 31730 / 309358-0115. A device comprising processing hardware and configured to implement a method of any of the preceding claims.
