Terminal device and ai information transmission method performed by same device

The terminal device and AI information transmission method address the complexity of 6G core networks by enabling efficient AI information transmission and dynamic path selection, facilitating AI model training and improving network performance and flexibility.

WO2026240294A1PCT designated stage Publication Date: 2026-11-19SK TELECOM CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/KR2025/022498
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-14
Filing Date
2025-12-22
Publication Date
2026-11-19

AI Technical Summary

Technical Problem

The increasing complexity of 6G core networks due to the rise in Network Functions (NFs) and signaling protocols between NFs makes it difficult to support a powerful and resilient core network, especially with the evolution towards AI-based infrastructure and the need for efficient AI model training and inferencing across terminals.

Method used

A terminal device and AI information transmission method that supports the transmission of AI information from a terminal device to a core network, including initiation, modification, and termination procedures for AI information sessions, using a Multi-Plane Selector to dynamically select the optimal path for transmission and visibility verification to ensure data integrity and quality.

Benefits of technology

Enables efficient collection and utilization of AI information by the core network for AI model training, enhancing network performance and flexibility by supporting dynamic path selection and visibility verification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure KR2025022498_19112026_PF_FP_ABST
    Figure KR2025022498_19112026_PF_FP_ABST
Patent Text Reader

Abstract

In order to make it possible to enhance AI features / operations by effectively distributing and cooperating on the AI features / operations and consequently improve the performance / quality of the entire system, the present invention proposes a terminal device in which a terminal (e.g., terminal / UE / UENF) performing an on-device AI-based service implements a configuration procedure for supporting AI information for transmitting data (AI information) to a core, and a method for transmitting AI information by means of the device.
Need to check novelty before this filing date? Find Prior Art

Description

Terminal device and AI information transmission method performed on the device

[0001] The present invention relates to a technology for transmitting information to the core of a communication network while a terminal performs an On-Device AI-based service.

[0002] The present application claims priority to application No. 10-2025-0062506, filed on May 14, 2025, and the entire contents of such application are incorporated herein by reference for all purposes.

[0003] In current 5G / 6G, the number of Network Functions (NFs) based on Service Based Interface (SBI) within the core network is increasing, and signaling processed by SBI-based protocols between NFs is increasing significantly.

[0004] As such, as the number of interconnections between NFs increases and the core network becomes the center of the 5G / 6G system (5GS / 6GS), supporting a powerful and resilient core is becoming more complex and difficult.

[0005] Meanwhile, the 6G core network is expected to evolve into AI-based infrastructure.

[0006] Therefore, the 6G core network will be composed of a combination of various planes for traffic / signal / control (e.g., User Plane, Control Plane, Intelligence Plane, Management Plane, etc.), and it is expected that the quality of various application services for customers will be improved compared to previous 5G, and XaaS (X as a Service, X = AI product, rate, function, technology, etc.) for each customer terminal will be processed with ultra-low latency and ultra-high performance.

[0007] For reference, XaaS (X as a Service) can be a service form such as cloud, AI computing, or remote, and can include a vast number of products, tools, technologies, and solutions provided to customers (terminals) via on / off-line.

[0008] In addition, in 6G, NFs can be further subdivided and evolve into smaller, agile, and flexible NF forms, such as Micro Function (MF) and Nano Function (NF) units, and this will also be the case for NF-enhanced terminals (UENF).

[0009] In addition, 6G will have an Evolved SBA (Service Based Architecture) structure in which all communication interfaces and protocols are unified or mixed, and will be a network capable of increasing compatibility (migration and interworking) with existing 5G / B5G / LTE.

[0010] Most NFs (including UENFs) within these 6G networks will have AI features embedded, and within each NF, AI-based operations (model training, inferencing, etc.) can be performed to improve network quality, performance, and flexibility.

[0011] The problem to be solved by the present invention is to realize a terminal device and an AI information transmission method performed by said device that support the transmission of AI information from a terminal (e.g., Terminal / UE / UENF) performing an On-Device AI-based service to a Core.

[0012] A terminal device according to one aspect of the present invention comprises: a memory including instructions; and a processor that, by executing the instructions, performs a configuration for transmitting AI information related to an AI model within the terminal device, transmits the AI ​​information to a core network based on the configuration, and transmits the AI ​​information to a consumer who will use the AI ​​information through the core network.

[0013] Specifically, the above-mentioned setup for AI information transmission may include at least one of an initiation procedure for setting up an AI information transmission session to be used for AI information transmission, a modification procedure for modifying part or all of the set AI information transmission session, and a termination procedure for releasing the AI ​​information transmission session.

[0014] Specifically, the above initiation procedure is initiated when the terminal device determines that AI information transmission is necessary, and may include a process of establishing a session for AI information transmission by proceeding with resource allocation for AI information transmission based on a Plane path selected for AI information transmission among a plurality of Planes between the terminal and the core network.

[0015] Specifically, the modification procedure may include a process of modifying part or all of the AI ​​information transmission session according to pre-set modification conditions while the AI ​​information transmission based on the AI ​​information transmission session is activated, and a process of storing transmission status information related to the pause, hold, or buffering in the terminal device for AI information that is paused, held, or buffered during the modification process.

[0016] Specifically, the termination procedure may include a process of releasing the resources and Plane path of the AI ​​information transmission session according to a preset release condition while the AI ​​information transmission based on the AI ​​information transmission session is active, and a process of notifying the terminal device and the core network of detailed information related to the transmission interruption regarding the AI ​​information for which transmission is interrupted during the release process.

[0017] Specifically, the consumer may be a specific Network Function (NF) within the core network, or a server or platform that interacts with the AI ​​model within the terminal device.

[0018] Specifically, the processor can transmit the AI ​​information to the core network and then perform a visibility verification procedure to check the status of the AI ​​information in conjunction with the core network.

[0019] Specifically, the visibility verification procedure may include a process in which a responsible NF checks at least one item among authentication / identification, authorization, validation, verification, and acceptance regarding the AI ​​information transmitted from the terminal device to verify the status of the AI ​​information, and a process in which the responsible NF periodically transmits the status information of the verified AI information to the terminal device.

[0020] Specifically, the processor can continuously update the state of the AI ​​information transmission internally based on the state information of the AI ​​information verified through the visibility verification procedure, and perform predefined subsequent control according to the state information.

[0021] Specifically, the above visibility verification procedure may include a process in which the core network checks at least one item among authentication / identification, authorization, validation, verification, and acceptance regarding the AI ​​information transmitted from the terminal device.

[0022] Specifically, the aforementioned predefined subsequent control may be a control operation defined to be performed subsequently according to an item checked as abnormal among the Authentication / Identification, Authorization, Validation, Verification, and Acceptance of the state information.

[0023] A method for transmitting AI information performed in a terminal device according to one aspect of the present invention comprises: a configuration step for performing a configuration for transmitting AI information related to an AI model within the terminal device; and an information transmission step for transmitting the AI ​​information to a core network based on the configuration, so as to be delivered to a consumer who will use the AI ​​information through the core network.

[0024] Specifically, the setting for the AI ​​information transmission may include at least one of an initiation procedure for setting up an AI information transmission session to be used for AI information transmission, a modification procedure for modifying part or all of the set AI information transmission session, and a termination procedure for releasing the AI ​​information transmission session.

[0025] According to the terminal device and the AI ​​information transmission method performed on the device of the present invention, a specific technology is implemented to support the transmission of AI information from a terminal (e.g., Terminal / UE / UENF) performing an On-Device AI-based service to a Core.

[0026] In particular, according to embodiments of the present invention, a configuration for supporting the transmission of AI information is specifically defined and implemented.

[0027] Thus, according to the present invention, since the Core or a consumer (e.g., AF, AS, NF, etc.) communicating with the Core can collect and use AI information on the terminal side (e.g., Terminal / UE / UENF), the effect of providing a data (AI information) collection solution is derived in which the Core or the consumer (e.g., AF, AS, NF, etc.) can collect data (AI information) for training an AI model on the terminal side.

[0028] Figure 1 is an example diagram illustrating a network structure composed of current NFs.

[0029] FIG. 2 is a block diagram showing the configuration of a terminal device according to one embodiment of the present invention.

[0030] Figure 3 is a diagram illustrating the concept of AI information transmission between a terminal device and a core network proposed in the present invention.

[0031] Figure 4 is an example illustrating the registration procedure and the Capability Negotiation procedure in the AI ​​information transmission method of the present invention.

[0032] FIG. 5 is an example illustrating a Call-Flow for setting / resetting a Plane path in the AI ​​information transmission method of the present invention.

[0033] FIG. 6 is an example illustrating the Call-Flow of configuration and transmission according to the AI ​​information transmission method of the present invention.

[0034] FIG. 7 is an example illustrating the call-flow of visibility verification and subsequent control according to the AI ​​information transmission method of the present invention.

[0035] Hereinafter, various embodiments of the present invention will be described with reference to the attached drawings.

[0036] The present invention relates to a technology for transmitting information to the core of a communication network (e.g., future generations such as 5G and 6G) ​​while a terminal performs an On-Device AI-based service.

[0037] In current 5G / 6G, the number of Network Functions (NFs) based on Service Based Interface (SBI) within the core network is increasing, and signaling processed by SBI-based protocols between NFs is increasing significantly.

[0038] As such, as the number of interconnections between NFs increases and the core network becomes the center of the 5G / 6G system (5GS / 6GS), supporting a powerful and resilient core is becoming more complex and difficult.

[0039] Meanwhile, the 6G core network is expected to evolve into AI-based infrastructure.

[0040] Therefore, the 6G core network will be composed of a combination of various planes for traffic / signal / control (e.g., User Plane, Control Plane, Intelligence Plane, Management Plane, etc.), and it is expected that the quality of various application services for customers will be improved compared to previous 5G, and XaaS (X as a Service, X = AI product, rate, function, technology, etc.) for each customer terminal will be processed with ultra-low latency and ultra-high performance.

[0041] In addition, in 6G, NFs can be further subdivided and evolve into smaller, agile, and flexible NF forms, such as Micro Function (MF) and Nano Function (NF) units, and this will also be the case for NF-enhanced terminals (UENF).

[0042] In addition, 6G will have an Evolved SBA (Service Based Architecture) structure in which all communication interfaces and protocols are unified or mixed, and will be a network capable of increasing compatibility (migration and interworking) with existing 5G / B5G / LTE.

[0043] Most NFs (including UENFs) within these 6G networks will have AI features embedded, and within each NF, AI-based operations (model training, inferencing, etc.) can be performed to improve network quality, performance, and flexibility.

[0044] In addition, in order to satisfy the requirements related to AI model training on the terminal side (e.g., Terminal / UE / UENF) for operating 5G / 6G wireless interfaces, a data (AI information) collection solution is required that allows the Core or Consumer (e.g., AF, AS, NF, etc.) to collect data (AI information) for AI model training on the terminal side (e.g., Terminal / UE / UENF).

[0045] In this regard, the present invention proposes a specific technology (hereinafter referred to as the AI ​​information transmission method) that supports the transmission of AI information from a terminal (e.g., Terminal / UE / UENF) performing an On-Device AI-based service to a Core.

[0046] In the present invention, by proposing an AI information transmission method, a Core or a consumer (e.g., AF, AS, NF, etc.) communicating with the Core can collect and use AI information on the terminal side (e.g., Terminal / UE / UENF), thereby providing a data (AI information) collection solution.

[0047] First, Figure 1 conceptually illustrates a network structure consisting of 5G / B5G / 6G NFs currently in commercialization / use.

[0048] As can be seen in Fig. 1, a 5G / 6G system consists of a UE, a wireless access network ((R)AN), and various core network functions. This structure represents a reference architecture for a non-roaming environment, and Service-Based Interfaces (SBI) are used in the control plane.

[0049] As illustrated in FIG. 1, the UE communicates with the Access and Mobility Management Function (AMF) and (R)AN through the N1 and N2 interfaces, and user data is transmitted to the Data Network (DN) via the User Plane Function (UPF) through the N3 and N6 interfaces.

[0050] The core network includes various existing, improved, and new NFs such as AUSF (Authentication Server Function), SMF (Session Management Function), PCF (Policy Control Function), UDM (Unified Data Management), NEF (Network Exposure Function), and DCF (Data Collection Function), and these are interconnected through SBI.

[0051] These 5G / B5G / 6G network structures are designed to efficiently perform high-speed data processing, user authentication, policy control, and session management, and a similar architecture can be utilized as a basic framework in generations after 6G.

[0052] Meanwhile, AI features can be embedded in various NFs located in the core network (e.g., AMF, SMF, UPF, PCF, NRF, OAM, UDR, USF, AF, NWDAF, etc.).

[0053] For convenience of explanation, I will describe an example in which NWDAF (Network Data Analytics Function) is mentioned as an NF having an AI Feature (e.g., AI / ML).

[0054] NDWAF can be composed of all or part of internal or external functions for performing AI features.

[0055] For example, NDWAF can be implemented with the functions of AnLF (Analytics Logical Function), MTLF (Model Training Logical Function), DCCF (Data Collection Coordination Function), ADRF (Analytics Data Repository Function), MFAF (Messaging Framework Adaptor Function) and related CF (Compute Function), AI / SF (AI / Service Function).

[0056] Meanwhile, terminals (e.g., Terminal / UE / UENF) can also have AI features (e.g., AI / ML) themselves, following the trend of embedding AI features in most NFs.

[0057] Hereinafter, an AI Feature (e.g., AI / ML) embedded within a terminal (e.g., Terminal / UE / UENF) will be referred to as an AI model (e.g., AI / ML) or On Device-AI.

[0058] AI models can be composed of various types, e.g., Large Language Models (LM), Deep / Machine Learning Models (D / ML), and can also consist of various algorithms (e.g., CNNs, RNNs, LSTM, GANs, Transformers, Linear / Logistic Regression, Decision Trees, Random Forest, SVM, k-NN, K-means clustering, Autoencoders, DBNs, Reinforcement Learning).

[0059] For convenience of explanation, the UE will be referred to and described below as a terminal type of one embodiment among the terminals (e.g., Terminal / UE / UENF) to which the present invention is applied.

[0060] FIG. 2 is a block diagram showing the configuration of a terminal device according to one embodiment of the present invention.

[0061] As illustrated in FIG. 2, the present invention is configured with a UE (100) that performs an On-Device AI-based service. To this end, the UE (100) has an AI model (e.g., AI / ML) that can perform AI Feature / operation on its own.

[0062] For example, within the UE (100), an AI model (e.g., AI / ML) may be loaded for each application (UE Application(s) (101)) that uses an AI Feature / action, or for applications (App) grouped according to specific conditions.

[0063] Accordingly, the UE Application(s) (101) illustrated in FIG. 2 can be considered to be equipped with the AI ​​model (e.g., AI / ML) that it uses.

[0064] These UE Application(s) (101) may consist of software (e.g., OS kernels / drivers, UI, Data, Security Connectivity module, etc.) and hardware (e.g., SoC, CPU, GPU, NPU, B5G / 6G Modem, RF Antenna, Li-Fi, WiFi / NFC / UWB / BT module, etc.) that operate the UE (100).

[0065] Also, as illustrated in FIG. 2, a UE (100) according to one embodiment of the present invention includes a UE AI / ML Training and Inferencing Processor (102) that performs training and inferencing of an internal AI model (e.g., AI / ML).

[0066] Here, training may involve using local data to train an AI model (e.g., AI / ML) (e.g., federated learning).

[0067] Inferencing can be the process of generating prediction results (e.g., photo / image classification, voice analysis, etc.) from a trained AI model (e.g., AI / ML).

[0068] In addition, a UE (100) according to one embodiment of the present invention has the function of performing a configuration for transmitting AI information related to an internal AI model (e.g., AI / ML) and transmitting the AI ​​information to a core network based on the configuration, and may include a Multi-Plane Selector (103) as a functional part related thereto.

[0069] In addition, the UE (100) includes a Tx / Rx Wireless Medium (104) as a wireless transmission and reception interface, through which it can perform various messages related to the configuration for transmitting AI information, transmission of AI information, etc.

[0070] Additionally, the UE (100) can perform network setting, startup, environment variable, user preference, logging / tracing, security policy, etc. related to AI information processing / transmission through configuration.

[0071] In addition, the configuration for AI information processing / transmission of the UE (100) in the present invention may be executed or interrupted depending on various internal and external environments such as the UE (100)’s Mobility status, H / O status, Area of ​​Interest, S-NSSAI, Cell, UE application usages, UE energy consumption status, UE temperature, UE sensor status, UE capability, scheduled, dynamic time-of-day, etc.

[0072] In addition, within the configuration for AI information processing / transmission of the UE (100) in the present invention, a cause and reason for the execution or interruption of the configuration may be included.

[0073] The reason the UE (100, equipped with AI / ML) transmits (or exchanges, communicates) its AI information to the core network is to effectively distribute and cooperate AI features / operations to enhance them, thereby improving the performance / quality of the entire system.

[0074] In the present invention, a specific / standardized data (AI information) transmission support technology (i.e., AI information transmission method) is implemented in which a UE (100, equipped with AI / ML) transmits (or exchanges, communicates) data (AI information) to a core network.

[0075] In particular, the present invention aims to specifically define a configuration for supporting data (AI information) transmission.

[0076] Before providing a detailed description of the present invention, I will briefly explain an App / use-case to aid understanding, in which a UE (100, equipped with AI / ML) performs an AI Feature / operation based on an internally embedded AI model (e.g., AI / ML).

[0077] To describe a use-case of one embodiment, it is for enhanced photo and image processing, in which a related app can automatically detect and classify scenes and subjects through AI / ML algorithms, and perform optimal photo shooting and video recording by adjusting brightness, contrast, saturation, etc. in real time.

[0078] To describe a use-case of another embodiment, it is for virtual assistant and voice recognition processing. In this case, the relevant app learns the user's habits and preferences to provide personalized AI / ML, and through natural language processing based on this AI / ML algorithm, it can more accurately recognize voice commands and execute subsequent actions such as converting images into text.

[0079] To describe a use-case of another embodiment, it relates to user authentication. In this case, the relevant app enhances AI / ML based on biometric authentication technologies such as facial recognition and fingerprint recognition, and continuously improves the recognition accuracy of AI / ML by adapting to changes in the user's appearance, thereby enabling accurate recognition and user authentication even with changes in the user's hairstyle, such as wearing glasses.

[0080] To describe a use-case of another embodiment, for battery life optimization, the relevant app learns and provides AI / ML to predict future usage by analyzing the usage patterns of other applications / apps, thereby managing and optimizing battery consumption, and furthermore, can maximize battery efficiency by closing unnecessary apps and adjusting power allocation.

[0081] In addition to the embodiments of the tactics, there are various App / use-cases, i.e., On-Device AI-based services, in which the UE (100, equipped with AI / ML) performs AI Features / operations based on On-Device AI (e.g., AI / ML).

[0082] The present invention will be applicable to various on-device AI-based services provided by the UE (100, equipped with AI / ML) as described above.

[0083] In particular, in the present invention, when a UE (100, equipped with AI / ML) performs an On-Device AI-based service, AI information from the UE (100, equipped with AI / ML) is transmitted to a core network at a free time (e.g., before, during, or after performing an On-Device AI-based AI Feature / operation), thereby enabling a core or a consumer (e.g., AF, AS, NF, etc.) communicating with the core to collect and use the AI ​​information of the UE (100, equipped with AI / ML).

[0084] Hereinafter, with reference to FIG. 3, the concept of AI information transmission between a terminal device and a core network proposed in the present invention will be explained.

[0085] First, I will describe the terminal device (100, Terminal / UE / UENF) of the present invention.

[0086] A terminal device (100, Terminal / UE / UENF) according to one embodiment of the present invention may have a configuration including a memory (not shown) containing instructions, and a processor (hereinafter referred to as an AI agent) that performs a configuration for transmitting AI information related to an AI model (e.g., AI / ML) within the terminal device (100, Terminal / UE / UENF) by executing the instructions, and transmits the AI ​​information to a core network based on the configuration so that it is delivered to a consumer who will use the AI ​​information through the core network.

[0087] In the following description, for the convenience of explanation, a UE will be referred to as a terminal type of one embodiment among the terminals (e.g., Terminal / UE / UENF) to which the present invention is applied.

[0088] In addition, an NF that manages / controls the transmission of AI information within the Core Network is defined, and is illustrated as a DCF (200, Data Collection Function) in Fig. 3.

[0089] This DCF (200) may be a newly defined NF, or it may be defined by adding functions to an existing NF, and as an example, an NWDAF may additionally perform the role of the DCF (200).

[0090] Accordingly, according to the present invention, the UE (100) can create an environment (e.g., session, QoS, network resources, policy, etc.) for transmitting its AI information to the Core Network by performing a configuration for transmitting its AI information with the DCF (200) within the Core Network.

[0091] And according to the present invention, when performing an On-Device AI-based service, the UE (100) transmits (or exchanges, communicates) its AI information to the DCF (200) within the Core Network at any time (e.g., before, during, or after performing the On-Device AI-based AI Feature / operation).

[0092] According to the present invention, in a structure in which a UE (100) transmits (or alters, communicates) its AI information to a DCF (200) within a Core Network at any time, a Core (e.g., DCF (200)) or a Consumer (300, e.g., AF, AS, NF, etc.) communicating with the Core can collect and use the AI ​​information of the UE (100).

[0093] More specifically, the configuration for AI information transmission may include at least one of an initiation procedure for setting up an AI information transmission session to be used for AI information transmission, a modification procedure for modifying part or all of the set AI information transmission session, and a termination procedure for releasing the AI ​​information transmission session.

[0094] As previously mentioned, in the present invention, the UE (100) is based on having an AI model (e.g., AI / ML) capable of performing AI Feature / operation on its own.

[0095] For example, for each UE Application(s) (101) using an AI Feature / action, the UE (100) may have one or more AI models (e.g., AI / ML) installed.

[0096] However, for convenience of explanation, the following description will be described as an example in which one UE Application (101) uses one AI model (e.g., AI / ML).

[0097] App (101) in UE (100) can provide an On-Device AI-based service by using an AI model (e.g., AI / ML) inside the UE when necessary to perform an AI Feature / action based on the AI ​​model (e.g., AI / ML) and utilizing the result of the performance in its own service.

[0098] Below, I will explain the operation process of the new technology proposed in the present invention, namely the AI ​​information transmission method.

[0099] According to one embodiment of the present invention, an AI agent for the AI ​​information transmission method (operation) of the present invention may be installed in a terminal device (100) and a DCF (200, where the core is 5G / B5G / 6G) or a specific NE (not shown, Network Equipment having the function of a DCF, where the core is LTE).

[0100] Accordingly, the terminal device (100) can be described as Terminal / UE / UENF with AI Agent, and can be described as an AI Agent as shown in FIG. 3.

[0101] Meanwhile, for convenience of explanation, in the following description, specific embodiments will be described by mentioning DCF (200) as a case where the core network is 5G / B5G / 6G.

[0102] In FIG. 4 below, the role of DCF (200) is illustrated and explained as an example in which NWDAF performs the role, and NWDAF is described as an NF having an AI Feature (e.g., AI / ML).

[0103] First, with reference to FIG. 4, I will specifically explain the process of carrying out the registration procedure and the Capability Negotiation procedure in the AI ​​information transmission method of the present invention.

[0104] According to the embodiment illustrated in FIG. 4, the UE (100) can transmit and receive messages with various NFs within the core network and perform necessary call processing as before (S10).

[0105] In addition, DCF (200, e.g., NWDAF) can perform analysis based on an internal AI model (e.g., AI / ML), i.e., AI Feature / action, using the information (Data) it possesses and requires for analysis, and can continuously train the AI ​​model (e.g., AI / ML).

[0106] As the App (101) providing On-Device AI-based services is executed, the UE (100) can perform AI Feature / action based on an AI model (e.g., AI / ML) as needed and utilize the result of the execution to provide On-Device AI-based services (S20).

[0107] In the present invention, in a basic structure in which a UE (100) and a DCF (200, e.g., NWDAF) within a core network perform AI Feature / operation using an AI model (e.g., AI / ML), a registration procedure and a Capability Negotiation procedure for applying the AI ​​information transmission method of the present invention are defined.

[0108] First, I will briefly explain the registration procedure for the AI ​​information transmission method of the present invention.

[0109] The registration procedure for applying the AI ​​information transmission method of the present invention means that the UE (100) negotiates with the DCF (200, e.g., NWDAF) of the target to which data (AI information) of an AI model (e.g., AI / ML) is to be transmitted, and establishes a negotiation connection for information transmission.

[0110] To explain in detail, the UE (100, AI agent) sends an AI / ML-related request message (also known as a connection request message) to the DCF (200, e.g., NWDAF) within the core network, and this can basically be conducted based on a standalone or distributed NAS (Non-Access Stratum, meaning communication between the UE and the core network) using a specific Plane.

[0111] At this time, the UE (100, AI agent) can basically proceed with a registration procedure (Negotiation connection setup) for applying the AI ​​information transmission method at the time when the UE (100) proceeds with connecting to the core / registering the location and processing the bearer / session.

[0112] A DCF (200, e.g., NWDAF) within the core network that receives a connection request message from a UE (100) may accept the connection request message based on its internal judgment and sends back to the UE (100) an AI / ML-related response message (also known as a connection response message) containing whether to accept it.

[0113] At this time, the fact that the DCF (200, e.g., NWDAF) accepts the connection request message essentially means that AI information can be transmitted between the UE (100) and the DCF (200, e.g., NWDAF) for the corresponding NAS.

[0114] When the Negotiation connection setup between the UE (100, AI agent) and the DCF (200, e.g., NWDAF) is completed through the registration procedure for applying the AI ​​information transmission method, the Capability Negotiation procedure for applying the AI ​​information transmission method can be carried out (S30).

[0115] In Fig. 4, for the sake of simplification of the drawing, the completion of the Negotiation connection setup in step S30, where the Capability Negotiation procedure is performed, is described as an "initial connection."

[0116] The Capability Negotiation of Step S30 refers to a procedure for setting various information / data, conditions, etc., necessary for performing the AI ​​information transmission method of the present invention, such as the target AI model to perform the AI ​​information transmission method of the present invention and the setting of a Plane path to be used when transmitting data (AI information).

[0117] Specifically, the UE (100, AI agent) sends a list of AI models (e.g., AI / ML) that it supports (used by the App (101)) to the DCF (200, e.g., NWDAF) within the core network. This list of AI models may include the types of AI models supported by the UE (100), AI algorithms, performance, processing time / latency, resource usage, accuracy, etc.

[0118] The DCF (200, e.g., NWDAF) selects common or compatible AI models (functions) by comparing the list of AIs of the UE (100) with the list of AI models it has (e.g., AI / ML). The DCF (200, e.g., NWDAF) can add, delete, or update AI models as needed through communication with other NFs or external nodes (e.g., NEF (Network Exposure Function), AF (Application Function), and AS (Application Server)).

[0119] DCF (200, e.g., NWDAF) transmits a list of selected AI models (functions) to UE (100), and UE (100, AI agent) selects a target AI model from the list of AI models (functions) of DCF (200, e.g., NWDAF) to finally perform the AI ​​information transmission method of the present invention, and replies this to DCF (200, e.g., NWDAF).

[0120] Thus, according to the present invention, among the AI ​​models (functions) that are common or compatible in the UE (100) and DCF (200, e.g., NWDAF), an AI model (e.g., AI / ML) to be processed for AI information can be set by the final selection by the UE (100, AI agent) (S40).

[0121] Of course, the UE (100, AI agent) may not make a final selection from the list of AI models (features) of the DCF (200, e.g., NWDAF).

[0122] For example, if the UE (100, AI agent) does not make a final selection and substitutes the final selection with any, the DCF (200, e.g., NWDAF) can make a final selection of the target AI model to perform the AI ​​information transmission method of the present invention with the UE (100) from its list of AI models (functions).

[0123] Alternatively, if the UE (100, AI agent) does not make a final selection and substitutes it with no, the DCF (200, e.g., NWDAF) may compare the UE (100)’s list of AI models with its own list of AI models (e.g., AI / ML) to reconstruct a list of common or compatible AI models (functions) and send it to the UE (100).

[0124] Meanwhile, in the Capability Negotiation procedure of step S30, the transmission / reply for setting the AI ​​model (e.g., AI / ML) of the AI ​​information processing target can be processed using a default Plane path if this Capability Negotiation procedure is the first between the UE (100) and the DCF (200, e.g., NWDAF), and if there has been a previously conducted Capability Negotiation procedure between the UE (100) and the DCF (200, e.g., NWDAF), it can be processed using a Plane path set in that procedure.

[0125] Of course, according to the present invention, it is also possible to negotiate the setting of the Plane path in the Capability Negotiation procedure of step S30.

[0126] For example, if the default Plane path of the UE (100, AI agent) is the User Plane, it is possible to set it to another Plane, such as the Control Plane or Management Plane, through negotiation.

[0127] What is important here is that the UE (100, AI agent) or the DCF (200) in the core network can take the initiative to update or change the default or previously set Plane path (e.g., UP) to another Plane at any time (e.g., before, during, or after transmitting AI information, or when the setting condition is satisfied, etc.).

[0128] The reason for updating or changing the Plane path settings in this way is to dynamically utilize the optimal Plane path by comprehensively considering the AI ​​performance required for the AI ​​Feature / operation of the UE (100, AI agent) / DCF (200) within the core network and the quality / performance of AI information transmission, which may vary depending on various factors such as the wireless environment and location in the UE (100), pre-configured conditions, and the wireless quality and performance of the Plane path, and as a result, the AI ​​Feature / operation can be enhanced.

[0129] In the present invention, if an attempt is made to update or change (hereinafter collectively referred to as "change") a Plane path setting from a Plane currently in use (e.g., UP) to another Plane (e.g., CP, MP, etc.) but is impossible or fails (hereinafter collectively referred to as "failure"), an alternative operation (e.g., AI model reduction, AI model size reduction, AI performance requirement reduction, etc.) may be performed according to the cause of the attempt at this change (e.g., wireless environment and location, pre-set conditions, wireless quality and performance of the Plane path, etc.).

[0130] As an example, if the current change is attempted due to a cause that fails to meet the wireless quality and performance of the currently used Plane path, the wireless quality and performance of the (maintained) Plane path may be indirectly improved by performing an action to reduce the AI ​​performance required for the AI ​​Feature / action of the UE (100, AI agent) / DCF (200) within the core network as a compromise, as a secondary action due to the cause.

[0131] In addition, in the present invention, when changing the Plane path setting (existing Plane → new Plane), a migration (Copy, Move, Delete, Backup) procedure is performed to move all or part of the AI ​​information contained in the existing Plane to the new Plane. At this time, the success or failure of the current Plane path change can be determined comprehensively by also considering the resources (performance, latency) required for execution.

[0132] In addition, in the present invention, the execution cycle / frequency of the Capability Negotiation procedure, the confirmation and status of the Plane setting, can be mutually verified between the UE (100) and the DCF (200) through mutual exchange / communication.

[0133] Below, a specific embodiment regarding the method of establishing a Plane path between the UE (100, AI agent) and the DCF (200) within the core network will be described.

[0134] Figure 5 illustrates a Call-Flow for setting / resetting a Plane path in the AI ​​information transmission method of the present invention.

[0135] In the AI ​​information transmission method of the present invention, a Plane that can be set as a Plane path includes at least one of a Control Plane, a User Plane, a Management Plane, a Compute Plane, a Message Plane, an AI Plane, a Training Plane, an Inference Plane, an Intelligence Plane, a Service Plane, and an Agent Plane.

[0136] Here, each Plane can be composed of various channels, protocols, and interfaces.

[0137] That is, a Plane can be composed of various channels and application protocols (e.g., header, message, information elements, payload), such as TCP / IP, UDP / IP, HTTP / 2, HTTP / 3, QUIC, and SBI (Service based interface), and the operation of the Plane can be composed of various communication methods (distributed, centralized, peer-to-peer, decentralized), such as Request / Response, Subscription / Notification, Message Queues, Polling, and Multi / Broadcasting.

[0138] In addition, each Plane can be configured with various virtual interfaces (vi) rather than just one (e.g., User Plane_vi#1, User Plane_vi#2, AI Plane_vi#2, Service Plane_vi#5). This can be configured differently depending on the Slice, virtual IP / Port address, and protocol parameter settings supported by each tunnel or end-to-end NF.

[0139] In addition, for each Plane, wireless quality (e.g., QoS, etc.) and performance (e.g., minimum / maximum speed, latency / jitter, etc.) for monitoring can be configured when set as a Plane path.

[0140] Referring to FIG. 5, a specific embodiment is described as follows: when an AI model (e.g., AI / ML) for AI information processing between the UE (100) and the DCF (200) is set as described above, the DCF (200) configures a list of Planes that can be used when transmitting data (AI information) related to the set AI model (e.g., AI / ML) and transmits it to the UE (100) (S110).

[0141] The UE (100, AI agent) checks the received (CN Side) Plane list and transmits the list of Planes it can use to the DCF (200) (S120).

[0142] In this case, the DCF (200) checks the received (UE Side) Plane list and, based on its internal judgment, sends a Recommended list to the UE (100) suggesting one or more Planes to be set with the UE (100) (S130).

[0143] In the embodiment of FIG. 5, a Recommended list presenting one Plane (e.g., CP) is transmitted to the UE (100). That is, this list can be considered as a Suggestion, Command, or Selected list that takes into account UE Side and Core Side information.

[0144] In addition, information related to multiple histories associated with Planes and AI models within the Recommended list, other terminals, Slices, types, RATs, and ratings and scoring (e.g., priority, accuracy) for regions / locations may also be included.

[0145] The UE (100, AI agent) takes into account the AI ​​performance required for the AI ​​model (e.g., AI / ML) of the AI ​​information processing target set earlier, internal judgment conditions, etc., and finally selects whether to accept the Plane (e.g., CP) of the Recommended list (S140).

[0146] When the UE (100, AI agent) accepts / selects a Plane (e.g., CP) from the Recommended list (S140 Yes), it can set the Plane path with the DCF (200) to CP by transmitting the acceptance details to the DCF (200).

[0147] Meanwhile, when the UE (100, AI agent) rejects or selects a Plane (e.g., CP) from the Recommended list (S140 No), it sends a message requesting reselection (Reselection request) to the DCF (200) (S150), and at this time, the UE Side may also suggest a desired Plane (e.g., UP).

[0148] DCF (200)) can check the message (Reselection request) and, based on the re-evaluation, present another Plane or send a Recommended list back to the UE (100) to accept (e.g., ok UP) the Plane presented on the UE Side (S160).

[0149] In this way, when the UE (100, AI agent) accepts / selects a Plane (e.g., UP) from the Recommended list, it can set the Plane path with the DCF (200) to UP by sending the acceptance details to the DCF (200).

[0150] Thus, according to the present invention, in the UE (100) and the DCF (200) within the core network, a Plane path can be established by the final acceptance / selection by the UE (100, AI agent) from the list of available Planes mutually transmitted between the UE (100) and the DCF (200).

[0151] Accordingly, in the present invention, AI information required can be transmitted / delivered between the UE (100, equipped with AI / ML) and the DCF (200) within the core network at any time (e.g., before, during, or after performing an On-Device AI-based AI Feature / operation) (S170).

[0152] Meanwhile, FIG. 5 illustrates an example in which the UE (100) sets the Plane path with the DCF (200) to UP by transmitting acceptance content to the DCF (200), and then, in the case where the Plane path is changed (e.g., UP->AP) according to the acceptance (S140 'Yes') of the UE (100) described later, AI information is transmitted to the same DCF (200) that was communicating in the previous step.

[0153] Alternatively, according to the present invention, in the step S170 where a Plane path is finally selected and set through the process of exchanging a DCF (200) and a Plane / Recommended list, and in the case where the Plane path is changed (e.g., UP->AP), NF and AI information other than the DCF (200) that was communicated in the previous step may be transmitted.

[0154] For example, each NF within the core network (e.g., DCF (200) in the case of FIG. 5) may have a Plane responsible for it depending on the settings of the purpose, conditions, etc. of AI information transmission.

[0155] In this case, if the NF responsible for the Plane path set or changed this time is not DCF (200) but another NF (not shown) after step S170 of setting the Plane path and the Plane path change (e.g., UP->AP) described later, the UE (100, AI agent) can transmit AI information using the other NF (not shown) and the Plane path.

[0156] If, after step S170 where the UE (100, AI agent) sets the Plane path and the Plane path change (e.g., UP->AP) described later, the NF responsible for the Plane path that was set or changed this time is still DCF (200), then the same DCF (200) and AI information may be transmitted.

[0157] At this time, the Plane / Recommended list exchanged between the UE (100, AI agent) and the DCF (200) may include information about the NF responsible for the Plane included in the list (e.g., address information, etc.).

[0158] Accordingly, if the NF responsible for the set or changed Plane path is not DCF (200) but another NF (not shown), the UE (100, AI agent) can perform procedures for transmitting AI information with the other NF (not shown) responsible for the set or changed Plane path based on the information in the list exchanged previously (e.g., necessary procedures among connection, registration procedure, and Capability Negotiation procedure).

[0159] Figure 5 illustrates a method in which a UE (100, AI agent) transmits / transmits AI information to a DCF (200) as an example of transmitting AI information.

[0160] Meanwhile, the UE (100, AI agent) and DCF (200) can monitor the wireless quality (e.g., QoS, etc.) and performance (e.g., minimum / maximum speed, delay / jitter, etc.) set on the Plane path while transmitting AI information using a mutually configured Plane path.

[0161] In the embodiment of FIG. 5, the DCF (200) receiving AI information can monitor the wireless quality (e.g., QoS, etc.) and performance (e.g., minimum / maximum speed, delay / jitter, etc.) set in the Plane path.

[0162] Accordingly, DCF (200) can determine whether the Plane path is in a state / environment that requires change based on at least one of the wireless environment and location in the UE (100) and pre-set conditions, as a result of monitoring the wireless quality and performance set in the Plane path (S180).

[0163] Of course, the UE(100, AI agent) can also determine whether the Plane path is in a state / environment that requires a change.

[0164] When DCF (200) determines that a change in the Plane path is required in a certain state / environment, it checks the (UE Side) Plane list received in step S120 of the tactic and, based on an internal judgment, sends a Recommended list to the UE (100) suggesting another Plane (e.g., AP (AI Plane)) to be set with the UE (100) (S190).

[0165] In this way, when the UE (100, AI agent) accepts / selects a Plane (e.g., AP (AI Plane)) from the Recommended list (S140'Yes), it can update or change the Plane path with the DCF (200) to the AP (AI Plane) by transmitting the acceptance details to the DCF (200).

[0166] Of course, if DCF (200) determines that a change in the Plane path is necessary in a certain state / environment, it may re-execute the tactical step S110 to re-set the Plane path.

[0167] Alternatively, if the UE (100, AI agent) determines on its own that a change to the Plane path is required in a certain state / environment, it may perform the operation starting from step S150 of the tactic to update or change the Plane path with the DCF (200) to another Plane (e.g., AP (AI Plane)).

[0168] Meanwhile, when the UE (100, AI agent) rejects or selects a Plane (e.g., AP (AI Plane)) from the Recommended list (S140'No), it can send a message requesting reselection (Reselection request) to the DCF (200) (S200).

[0169] Alternatively, when the UE (100, AI agent) rejects or selects a Plane (e.g., AP (AI Plane)) from the Recommended list (S140'No), if it determines that it is impossible to update or change the settings to the Plane (e.g., AP (AI Plane)) from the Recommended list, it may perform an alternative action (e.g., AI performance reduction, etc.) based on the cause of the attempt to update or change this time (e.g., wireless environment and location, pre-configured conditions, wireless quality and performance of the Plane path, etc.) (S200).

[0170] As described above, according to the present invention, a UE (100, AI agent) or a core network (e.g., DCF (200)) may take the initiative to update or change a default or previously set Plane path (e.g., UP) to another Plane at any time (e.g., before, during, or after transmission of AI information, or when a set condition is satisfied, etc.).

[0171] At this time, the setting / resetting function of the Plane path illustrated in Fig. 5 can be performed by the AI ​​agent of the UE (100), in particular the Multi-Plane Selector (103).

[0172] Meanwhile, although the description in the tactics suggests presenting one Plane related to one AI model (e.g., AI / ML), it is not limited to this, and the Recommended list may include multiple Planes, and M Planes may be delivered to and selected by the UE (100). That is, the UE (100) may select specific N AI models (e.g., AI / ML) according to conditions so that they can be separated (Separate / Split) or aggregated (Aggregate / Merge) into M Planes.

[0173] Additionally, the UE (100) can select to combine specific N AI models into other AI models by expanding or reducing them according to conditions. This allows for the optimization of the AI ​​models themselves, and the state of multiple M planes can be considered when performing the optimization operation.

[0174] Through this, the UE (100) can transmit and receive data (AI information) of multiple AI models through multiple planes to and from a core network (e.g., DCF (200)) by considering multiple environments.

[0175] Meanwhile, the process of setting / resetting the Plane path described with reference to FIG. 5 may be performed while the Capability Negotiation procedure of step S30 shown in FIG. 4 is in progress, or it may be performed at any time separately from the Capability Negotiation procedure (e.g., before, during, or after transmitting AI information, or when the setting condition is satisfied, etc.).

[0176] That is, the UE (100, AI agent) sets a Plane path to be used when transmitting AI information and a target AI model to perform the AI ​​information transmission method through the Capability Negotiation process (see S40 in FIG. 4), and during the Capability Negotiation process, monitors the quality and performance set in the Plane path and, based on the result and at least one of the wireless environment and location in the UE (100) and the pre-set conditions, can re-conduct Capability Negotiation to update or change the pre-set (at least one of the Plane path and target AI model).

[0177] Additionally, according to the present invention, the Capability Negotiation procedure and / or the process of setting / resetting the Plane path can be attempted / re-executed at any time by the UE (100, AI agent), and this process can be attempted / re-executed for each App (101) providing an On-Device AI-based service within the UE (100).

[0178] Meanwhile, according to the present invention, the UE (100, AI agent) may select a generation of the core network (e.g., LTE, 5G NSA, 5G SA, 6G Core Network / RAT, etc.) to perform the AI ​​information transmission method of the present invention depending on the environment (e.g., wireless environment and location, preset conditions, etc.).

[0179] For example, the UE (100, AI agent) can select / reselect a generation of the core network (e.g., LTE, 5G NSA, 5G SA, 6G Core Network / RAT, etc.) to perform the AI ​​information transmission method before (e.g., Negotiation connection setup), during the procedure, or after the procedure (including UE, RAN H / O procedure, UE / Network Triggering), and this selection / reselection can be performed for each App (101) that provides an On-Device AI-based service within the UE (100).

[0180] Furthermore, according to the present invention, the UE (100, AI agent) may adjust the throughput (resources, performance, etc.) of an internal AI model (e.g., AI / ML), training, and AI operations (e.g., inference) according to the environment (e.g., wireless environment and location, preset conditions, etc.).

[0181] For example, the UE (100, AI agent) may adjust the target AI model (e.g., AI / ML) to perform the AI ​​information transmission method at a time when it is determined that adjustment is necessary according to the environment (e.g., wireless environment and location, pre-set conditions, etc.) before the Capability Negotiation procedure (e.g., Negotiation connection setup), during the procedure, or after the procedure, and may also adjust the Train and Inference throughput (resources, performance, etc.) for it, and such adjustment may be performed for each App (101) that provides On-Device AI-based services within the UE (100).

[0182] Referring again to FIG. 4, the UE (100, AI agent) performs an AI information transmission method (S50~S80) to transmit data (AI information) according to the Capability Negotiation previously conducted through steps S30~S50 with the DCF (200, e.g., NWDAF) within the core network described above.

[0183] Meanwhile, prior to specifically describing the operation of transmitting AI information, according to the present invention, the AI ​​information transmitted between a terminal (e.g., Terminal / UE / UENF) and a DCF (200) within a core network may include sensitive content (e.g., personal information, model information, sensitivity to data, privacy).

[0184] Accordingly, in the present invention, for an AI information transmission method for transmitting AI information, authentication and security may be provided as necessary, and settings for this may be supported.

[0185] For example, in one embodiment, the AI ​​information to be transmitted (content related to AI / ML) may be sensitive under specific conditions (e.g., personal information, model information, sensitivity to data, privacy).

[0186] Therefore, through consultation between the terminal (e.g., Terminal / UE / UENF) and the DCF (200) within the core network, authentication / encryption can be used differently for each condition regarding AI information (content related to AI / ML).

[0187] To briefly explain one example of the process, the DCF (200) in the core network requests an authentication method for transmitting AI information from the terminal (e.g., Terminal / UE / UENF).

[0188] A terminal (e.g., Terminal / UE / UENF) transmits an authentication method of an AI model (e.g., AI / ML) that it supports to a DCF (200) within the core network, and the DCF (200) selects an appropriate authentication method among the transmitted authentication methods and notifies the terminal (e.g., Terminal / UE / UENF).

[0189] The terminal (e.g., Terminal / UE / UENF) and DCF (200) perform an authentication procedure with each other using an authentication method based on selection / notification, and depending on the result of this, they can perform an AI information transmission method based on the current Capability Negotiation setting (S50~S80).

[0190] Hereinafter, with reference to FIG. 6, the configuration and transmission call-flow according to the AI ​​information transmission method of the present invention will be explained.

[0191] As explained with reference to Figure 4 of the invention, when the registration procedure and Capability Negotiation procedure between the UE (100) and the DCF (200) in the core network are completed, the UE (100) creates an environment (e.g., session, QoS, network resources, policy, etc.) for transmitting (or exchanging, communicating) its AI information to the DCF (200) in the core network at a time when it is free to perform On-Device AI-based services.

[0192] Specifically, according to the present invention, the UE (100) can create an environment (e.g., session, QoS, network resources, policy, etc.) for transmitting its AI information to the Core Network by performing a configuration for transmitting its AI information with the DCF (200) within the Core Network.

[0193] As a specific embodiment, the configuration for AI information transmission may include at least one of an initiation procedure for setting up an AI information transmission session to be used for AI information transmission, a modification procedure for modifying part or all of the set AI information transmission session, and a termination procedure for releasing the AI ​​information transmission session.

[0194] In the present invention, in addition to the three aforementioned procedures of configuration (initiation, modification, and termination), detailed procedures within each major procedure may be included.

[0195] For example, detailed procedures within each major procedure may include discovery and registration procedures related to AI information within and between the UE (100) and the UE (100), negotiation procedures, activation procedures, monitoring reporting procedures, scaling and cloning procedures, suspension and pausing, resumption and reconnection, and deregistration and cleanup procedures.

[0196] In addition, the execution of the procedures described above (Discovery and Registration procedures, ... Deregistration and Cleanup procedures) may be performed within the Initiation, Modification, and Termination procedures, or before and after the procedures.

[0197] According to the embodiment illustrated in FIG. 6, the UE (100) can transmit and receive messages with various NFs within the core network and perform necessary call processing as before (S210).

[0198] In addition, DCF (200, e.g., NWDAF) can perform analysis based on an internal AI model (e.g., AI / ML), i.e., AI Feature / action, using the information (Data) it possesses and requires for analysis, and can continuously train the AI ​​model (e.g., AI / ML).

[0199] In addition, AI models can be defined as protocols (e.g., MCP, Model Communication Protocol) that enable interaction such as message exchange between AI models (e.g., LLM → Vision model request), tool invocation and result processing (e.g., calculator API call), multimodal model collaboration (e.g., text ↔ image ↔ audio), and traceable message delivery (including context, ID, and user permissions).

[0200] For example, the UE (100) can make requests to an LLM model in natural language through the MCP (“weather, AI-drawn animal photos, interpretation and summary of specific documents”) and organize them into a specific Script / JSON structure to specify who sent it / to whom it is sent / what work is requested / what data is passed.

[0201] In this regard, in the external network and core network, detailed query and LLM results can be generated through various tools, AI models and API servers, and the generated results can be responded to in the form of script / Json, and finally, the UE (100) can receive the result / response in natural language.

[0202] UE (100) can provide an On-Device AI-based service by utilizing the results of the execution of an App (101) that provides an On-Device AI-based service when necessary, and by performing an AI Feature / action based on an AI model (e.g., AI / ML) (S220).

[0203] According to the present invention, a UE (100, AI agent) determines whether it is necessary to transmit related AI information to a core network while performing part or all of an AI Feature / operation based on an AI model (e.g., AI / ML) (S230).

[0204] At this time, the determination of whether transmission of AI information is necessary can be made based on various criteria, such as by determining based on configuration information received from the core network in the past, determining based on the history with Consumers (300, e.g., AF, AS, NF, etc.) that the UE (100, AI agent) previously communicated with, or determining based on the conditions set in the Capability Negotiation procedure described above.

[0205] Here, Consumer (300) may be a specific NF (e.g., NWDAF) within the core network, or a server or platform that interacts with an AI model (e.g., AI / ML) within the UE (100) (e.g., AF, AS, etc.).

[0206] According to the present invention, if the UE (100, AI agent) determines that it is necessary to transmit AI information to the core network, it can perform configuration with the core network.

[0207] According to the present invention, when performing a configuration for transmitting AI information, the UE (100) can classify the operation into a first configuration and a second configuration.

[0208] For example, the first configuration can be defined as a step of exchanging and setting initial information (e.g., clean state) between the UE (100) and the core network, and the operation can be stored in memory and cache.

[0209] The second configuration can be defined as a step of exchanging and configuring minimal information between the UE (100) and the core network based on the temporal memory and cache included in the first configuration, and this can be reconfigured according to the temporal memory and cache included in the first configuration.

[0210] Thus, according to the present invention, the UE (100) can classify and perform configurations for transmitting AI information into multiple operations (first, second,...) depending on the type of multiple base stations, the multi-SIM and non-3gpp network of the UE (100), and can also perform multiple configurations for transmitting AI information in parallel.

[0211] Referring again to FIG. 6, according to the present invention, if a UE (100, AI agent) determines that it needs to transmit AI information to a core network, it can perform a configuration with the core network, in particular an initiation procedure (S240).

[0212] Specifically, the initiation procedure of step S240 is initiated when the UE (100, AI agent) determines that AI information transmission is necessary, and can be performed as a process of establishing a session for AI information transmission by allocating resources for AI information transmission based on a Plane path selected for AI information transmission among a plurality of Planes between the UE (100, AI agent) and the core network.

[0213] To describe a specific embodiment, when the UE (100, AI agent) determines that it needs to transmit AI information to the core network, it performs an initiation procedure with the DCF (200, e.g., NWDAF) in the core network, that is, a session establishment process for transmitting AI information (S240).

[0214] At this time, the UE (100, AI agent) performs the process of establishing a session for transmitting AI information by allocating resources for transmitting AI information based on the Plane path selected / configured in the Capability Negotiation procedure with the DCF (200, e.g., NWDAF) described earlier among multiple Planes.

[0215] The session configured in this manner for AI information transmission refers to a dedicated session for transmitting data / information regarding AI models and inference, distinct from general PDU sessions (IMS sessions, data sessions, etc.). Accordingly, the session for AI information transmission may have different characteristics and features compared to PDU sessions (IMS sessions, data sessions, etc.) in terms of transmission plane, interface, protocol, service conditions, etc.

[0216] In addition, according to the present invention, during the initiation procedure between a UE (100, AI agent) and a DCF (200, e.g., NWDAF) within a core network, network resources such as QoS, bandwidth, rate, and slice for the session for transmitting the AI ​​information, as well as AI / ML-related policies (allow, disallow, threshold setting, filter conditions, etc.) may be set together.

[0217] In addition, according to the present invention, during the initial setup (Initiation) procedure between a UE (100, AI agent) and a DCF (200, e.g., NWDAF) within a core network, the UE (100, AI agent) also obtains address information of a Consumer (300, e.g., AF, AS, NF, etc.) which is the destination of the session for transmitting the AI ​​information.

[0218] Accordingly, the UE (100, AI agent) can verify whether its data (AI information) is being transmitted and received normally based on the address information of the Consumer (300, e.g., AF, AS, NF, etc.), and this is related to the visibility verification procedure between the UE (100, AI agent) and the core network described later.

[0219] Meanwhile, according to the present invention, the initiation procedure between the UE (100, AI agent) and the DCF (200, e.g., NWDAF) within the core network may fail, and such failure may occur due to reasons such as insufficient performance, network overload, etc., when there is a mismatch in conditions for the initiation procedure.

[0220] According to the present invention, if the initial initiation procedure between the UE (100, AI agent) and the DCF (200, e.g., NWDAF) within the core network fails, the UE (100, AI agent) and the DCF (200, e.g., NWDAF) may attempt to re-initiate through a re-initiation procedure, and during such retry, a different condition from the existing (in the case of previous failure) may be selected, and the retry cycle may be adjusted according to a policy such as a back-off time.

[0221] The initial setup (Initiation) procedure of step S240 described in the present invention may be similar to a kind of negotiation, but is essentially / fundamentally an operation of preparing between two peers, such as allocating and securing resources for transmitting AI information, distributing resources between UE (100) and Core, and reviewing load.

[0222] As illustrated in FIG. 6, the UE (100, AI agent) transmits its AI information to the DCF (200, e.g., NWDAF) within the core network based on the setup (session for transmitting AI information set up on the selected Plane path) after the initial setup (Initiation) procedure of step S240.

[0223] In this way, the DCF (200, e.g., NWDAF) in the core network uses the AI ​​information if it is a Consumer of the AI ​​information of the UE (100, AI agent), and transmits the AI ​​information to other Consumers (300, e.g., AF, AS, NF, etc.) of the AI ​​information of the UE (100, AI agent) so that they can use it (S250).

[0224] Below, the AI ​​information transmission operation for transmitting AI information in the present invention will be described in detail.

[0225] Before providing a detailed explanation, the following is an example of an AI information transmission scenario in which AI information transmission can be applied between a terminal (e.g., Terminal / UE / UENF) and a core network (specifically, DCF (200)).

[0226] First, according to one example, federated learning can be supported through AI information transmission.

[0227] For example, a terminal transmits parameters of an AI / ML model trained in a local / internal storage to a core network as "AI information," and the core network can update a global model by aggregating AI / ML models from multiple terminals using the "AI information."

[0228] When the AI ​​information transmission operation of the present invention is applied to such an AI information transmission scenario, the core network can improve the performance of AI / ML models by utilizing data from multiple terminals, i.e., "AI information," while protecting personal information. In this case, the terminals can participate in training without sharing sensitive original data.

[0229] According to another example, the performance of segmented AI / ML tasks can be supported through the transmission of AI information.

[0230] For example, if part of an AI / ML model runs on the terminal and "AI information" related to that execution is transmitted to the core network, the remainder, excluding the parts that do not run, can be executed at the core network's edge or in the cloud. In this case, the core network can coordinate and manage these partitioned executions.

[0231] When the AI ​​information transmission operation of the present invention is applied to such an AI information transmission scenario, complex AI / ML tasks can be performed efficiently while considering the limited computing resources of the terminal. In addition, tasks can be dynamically divided and distributed according to network conditions.

[0232] According to another example, transfer learning can be supported through AI information transmission.

[0233] For example, the core network provides a pre-trained AI / ML model to the terminal as "AI information," and the terminal can fine-tune the AI ​​model using local data from a local / internal storage based on this.

[0234] When the AI ​​information transmission operation of the present invention is applied to such AI information transmission scenarios, high-performance AI / ML models can be built more efficiently than training a model from scratch at the terminal. Additionally, personalized models can be created while saving network resources.

[0235] According to another example, network data analysis can also be supported through AI information transmission.

[0236] For example, network analysis information provided by the NWDAF within the core network can be transmitted to the terminal's AI / ML application as "AI information".

[0237] When the AI ​​information transmission operation of the present invention is applied to such an AI information transmission scenario, the performance of the terminal's AI / ML application can be optimized by utilizing network status, traffic patterns, QoS information, etc. For example, the terminal (e.g., AI / ML application) can adjust the timing of data transmission based on network congestion prediction information.

[0238] According to another example, AI information transmission can also support updates to AI / ML models and parameters.

[0239] For example, the core network provides the latest AI / ML models or parameters to the terminal, and the terminal can download them to update its local model.

[0240] When the AI ​​information transmission operation of the present invention is applied to such AI information transmission scenarios, consistent AI / ML performance can be maintained across the entire network, and new data or environmental changes can be responded to quickly. In addition, considering the limited storage space of the terminal, only necessary models can be selectively updated.

[0241] That is, the AI ​​information transmission operation proposed in the present invention refers to a procedure / method / call processing, etc., for transmitting AI information by applying the aforementioned AI information transmission scenarios between a terminal (e.g., Terminal / UE / UENF) and a core network (particularly, DCF (200)) for various App / use-cases (e.g., voice recognition, image quality improvement, XR, etc.) described above.

[0242] In the case of Fig. 4, which was explained in the tactics section, I will describe an embodiment that supports the execution of a segmented AI / ML task as an example of performing an AI information transmission operation (S50~S80).

[0243] As explained above, the UE (100) can perform necessary existing call processing through message transmission and reception with various NFs in the core network (S10), and in particular, the UE (100, AI agent) performs an AI information transmission operation to transmit AI information according to the Capability Negotiation and configuration (configuration, especially the initial configuration (Initiation) procedure) previously performed with the DCF (200, e.g., NWDAF) in the core network.

[0244] Specifically, in the UE (100), an App (101) executed to provide an On-Device AI-based service can perform some AI Feature / action for the service using an AI model (e.g., AI / ML) (S50).

[0245] In this regard, the UE (100, AI agent) can extract information to be transmitted and configure AI information based on the Capability Negotiation and configuration (configuration, particularly the initial configuration (Initiation) procedure) previously conducted with the DCF (200, e.g., NWDAF) (S50), and can transmit / transmit this AI information to the DCF (200, e.g., NWDAF) (S60).

[0246] Assuming an embodiment that supports performing partitioned AI / ML tasks, the AI ​​information transmitted to the DCF (200, e.g., NWDAF) will be configured to include data / information that allows the DCF (200, e.g., NWDAF) to request the remaining AI Features / actions for services other than some AI Features / actions performed by the App (101) in the UE (100).

[0247] At this time, the UE (100, AI agent) can transmit the AI ​​information configured in step S50 to the DCF (200, e.g., NWDAF) before, during, or after the execution of the AI ​​Feature / action (in this case, AI information configured in relation to the previous AI Feature / action) (S60).

[0248] At this time, the UE (100, AI agent) transmits / transmits AI information using the AI ​​information transmission session set up in the previously selected Plane path (e.g., UP, CP, MP, etc.).

[0249] A DCF (200, e.g., NWDAF) within a core network that receives AI information from a UE (100, AI agent) can use the received AI information to identify data / information requesting the remaining AI Feature / operation, and after verifying this, can perform the remaining AI Feature / operation based on an AI model (e.g., AI / ML) mutually configured with the UE (100, AI agent) through the Capability Negotiation process (S65).

[0250] Then, DCF (200, e.g., NWDAF) sends the result of the AI ​​Feature / operation performed in step S65 to the UE (100, AI agent) (S70), and the UE (100, AI agent) can reflect this response in the AI ​​Feature / operation of the corresponding AI model (e.g., AI / ML) (S80).

[0251] In step S80, the method of reflecting the response (result of execution of AI Feature / action) of the DCF (200, e.g., NWDAF) into the AI ​​Feature / action of the AI ​​model (e.g., AI / ML) within the UE (100) may be a method of combining the result (response) of the divisional execution of the entire AI Feature / action for the On-Device AI-based service of the App (101) in the DCF (200, e.g., NWDAF) and using it in the service, provided that the embodiment supports the execution of divided AI / ML tasks.

[0252] In addition, the method of reflecting the response (result of execution of AI Feature / action) of the DCF (200, e.g., NWDAF) in step S80 into the AI ​​Feature / action of the AI ​​model (e.g., AI / ML) within the UE (100) can be various, such as reinforcing the AI ​​model (e.g., AI / ML) in a local / internal repository based on the response or fine-tuning it.

[0253] Meanwhile, as another example of performing an AI information transmission operation in the present invention, S50 to S80 can be explained by mentioning an embodiment that supports federated learning of a core network.

[0254] In this case, the AI ​​information transmitted by the UE (100, AI agent) to the DCF (200, e.g., NWDAF) in steps SS50 and 60 can be extracted / configured to include parameters of an AI model (e.g., AI / ML) trained in a local / internal repository within the UE (100).

[0255] In step S65, the DCF (200, e.g., NWDAF) can update / enhance the large-scale / global AI model (e.g., AI / ML) of the core network by aggregating / learning the AI ​​model parameters of multiple UEs (100) based on the AI ​​information being transmitted.

[0256] In this case, at step S70, the DCF (200, e.g., NWDAF) may reply to the UE (100, AI agent) with the latest updated / enhanced large-scale / global AI model (e.g., AI / ML) or parameters, and the UE (100, AI agent) may update the AI ​​model (e.g., AI / ML) by reflecting this reply (S80, embodiment supporting AI / ML model and parameter update).

[0257] Meanwhile, as mentioned in the tactics, the UE (100, AI agent) can determine whether there is a situation where the settings according to the Capability Negotiation conducted in step S30 need to be changed, that is, whether the Capability Negotiation needs to be re-conducted (S90).

[0258] For example, the UE (100, AI agent) monitors the wireless quality (e.g., QoS, etc.) and performance (e.g., minimum / maximum speed, latency / jitter, etc.) of the Plane path transmitting AI information, and determines whether the wireless quality / performance set for this Plane path is satisfied.

[0259] When the UE (100, AI agent) determines that the wireless quality / performance set in this Plane path is unsatisfactory (e.g., low QoS, high-latency, high jitter, etc.) or that there is an abnormal state (e.g., data retransmission, error, faults, etc.), it may determine that Capability Negotiation needs to be performed again.

[0260] Accordingly, if the UE (100, AI agent) determines that the Capability Negotiation needs to be re-conducted at step S90, it can re-conduct the Capability Negotiation procedure with the DCF (200, e.g., NWDAF) by transmitting the determination details and the data / information required for the re-conducting of Capability Negotiation to the core network, i.e., the DCF (200, e.g., NWDAF) (S30').

[0261] At this point, through the re-execution of Capability Negotiation in step S30, only the process of resetting the Plane path may be performed, or if necessary, the target AI model may also be reset.

[0262] Referring again to FIG. 6, the configuration for AI information transmission according to the present invention is described as follows: After the initial configuration (Initiation) described above, the UE (100, AI agent) can perform a modification procedure (S260) while the session-based AI information transmission for AI information transmission is activated.

[0263] Specifically, the modification procedure of step S260 may include a process of modifying part or all of the session for AI information transmission according to pre-set modification conditions while the session-based AI information transmission for AI information transmission is activated, and a process of storing transmission status information related to the pause, hold, or buffering in the terminal device for AI information that is paused, held, or buffered during the modification process.

[0264] To describe a specific embodiment, the UE (100, AI agent) may perform a process of modifying part or all of the session for transmitting AI information according to a pre-set modification condition while in the operation of transmitting AI information (i.e., when the transmission of AI information is active).

[0265] At this time, the modification condition is a condition based on the settings, and may be a condition set in the form of, for example, UE(100, AI agent) / core (specifically, DCF(200)), threshold exceeded, error occurrence, event, etc.

[0266] Also, the modification procedure of step S260 can be triggered by the UE (100, AI agent) and can also be triggered by the core network (specifically, DCF (200)).

[0267] According to the present invention, a UE (100, AI agent) can perform a modification procedure with a core network (particularly, DCF (200)) to modify part or all of the session for transmitting AI information (e.g., adjusting the transmission rate, changing QoS, changing acceptable AI / ML models, etc.).

[0268] According to the present invention, in the modification procedure of step S260, part or all of the session for AI information transmission can be modified or changed, similar to the initiation procedure of step S240.

[0269] Depending on the success or failure of the modification procedure, a modification / change action is executed. At this time, the timing of the modification / change and the time required (e.g., duration, processing time) are also taken into account, and the UE (100, AI agent) / core (specifically, DCF (200)) performs the modification procedure by including this information.

[0270] In addition, during the modification process, the transmission of AI information may be paused, held, or buffered.

[0271] According to the present invention, in the process of modifying part or all of the session for transmitting AI information during a modification procedure, if there is AI information that is paused, held, or buffered during transmission, transmission status information related to the pause, hold, or buffering of said AI information can be temporarily stored in a cache within the UE (100).

[0272] The transmission status information temporarily stored here may be in a form that includes how far the AI ​​information has been transmitted, whether the transmission was successful, sequence, chunk information, etc.

[0273] And the storage time and duration for the temporary storage of transmission status information can be discussed / agreed upon between the UE (100, AI agent) and the core (specifically, the DCF (200)) when performing the modification procedure.

[0274] The difference between the modification procedure of step S260 and the initiation procedure of step S240 in this invention is that the modification procedure is performed when AI information transmission is already activated, and therefore can occur while AI information is actually being transmitted.

[0275] Accordingly, during the modification procedure, transmission status information (e.g., how far the AI ​​information has been transmitted, whether the transmission was successful, sequence, chunk information, etc.) for AI information that is paused, held, or buffered can also be modified in accordance with the modification of the session for transmitting AI information, and this can also be discussed / consulted between the UE (100, AI agent) and the core (especially, DCF (200)).

[0276] In this case, the UE (100, AI agent) transmits its AI information to the DCF (200, e.g., NWDAF) within the core network based on the configuration (session for transmitting updated AI information to the selected Plane path) after the modification procedure of step S260.

[0277] In this way, the DCF (200, e.g., NWDAF) in the core network uses the AI ​​information if it is a Consumer of the AI ​​information of the UE (100, AI agent), and transmits the AI ​​information to other Consumers (300, e.g., AF, AS, NF, etc.) of the AI ​​information of the UE (100, AI agent) so that they can use it (S250).

[0278] Additionally, regarding the configuration for AI information transmission according to the present invention, the UE (100, AI agent) can perform a termination procedure after the initial configuration (Initiation) described above, while the session-based AI information transmission for AI information transmission is activated (S270).

[0279] Specifically, the termination procedure of step S270 may include a process of releasing the resources and Plane path of the AI ​​information transmission session according to a pre-set release condition while the AI ​​information transmission based on the AI ​​information transmission session is active, and a process of notifying the terminal device and the core network of detailed information related to the transmission interruption regarding the AI ​​information for which transmission interruption occurs during the release process.

[0280] To describe a specific embodiment, the UE (100, AI agent) can perform a process of releasing the session for transmitting AI information according to a preset release condition while in the operation of transmitting AI information (i.e., in a state where AI information transmission is enabled).

[0281] In this case, the release condition is a condition based on the settings, and may be a condition set in the form of, for example, transmission speed, QoS, etc.

[0282] Also, the termination procedure of step S270 can be triggered by the UE (100, AI agent) and can also be triggered by the core network (specifically, DCF (200)).

[0283] According to the present invention, the UE (100, AI agent) can perform a termination procedure with the core network (particularly, DCF (200)) to release the resources and Plane path of the session for transmitting AI information.

[0284] For example, according to the present invention, during the termination procedure, the resources of the session for transmitting AI information, the compute for transmitting AI information, and the internal cache of the UE (100) may be released according to different release conditions, and accordingly, the related Plane path may also be released.

[0285] In addition, according to the present invention, regarding AI information for which transmission is interrupted during the process of releasing resources and Plane paths of a session for transmitting AI information during a termination procedure, detailed information related to the transmission interruption is notified to the UE (100, AI agent) and the core network (particularly, DCF (200)), and furthermore, to the Consumer (300).

[0286] The details notified here may be in the form of including where the AI ​​information was interrupted—such as the location of the transmission interruption and the cause of the transmission interruption.

[0287] According to the present invention, the UE (100, AI agent) can stop the transmission of AI information that was using the session for transmitting AI information released after the termination procedure (S280).

[0288] As described above, according to embodiments of the present invention, in a basic structure in which a terminal (e.g., Terminal / UE / UENF) performs an AI Feature / operation using an AI model (e.g., AI / ML), a method (AI information transmission method) for transmitting data (AI information) related to the terminal's On-Device AI to a specific NF (e.g., DCF) within the core is defined / implemented, and a Capability Negotiation procedure for applying AI information transmission is defined / implemented.

[0289] In addition, according to embodiments of the present invention, configuration, namely initiation, modification, and termination procedures, are specifically defined to support the transmission of AI information from a terminal (e.g., Terminal / UE / UENF) performing an on-device AI-based service to a core.

[0290] Thus, according to the present invention, since the Core or a consumer (e.g., AF, AS, NF, etc.) communicating with the Core can collect and use AI information on the terminal side (e.g., Terminal / UE / UENF), the effect of providing a data (AI information) collection solution is derived in which the Core or the consumer (e.g., AF, AS, NF, etc.) can collect data (AI information) for training an AI model on the terminal side.

[0291] In addition, according to the present invention, the performance and quality of the entire system can be improved, such as by improving AI performance performed at terminals (e.g., Terminal / UE / UENF) and core networks (AI Feature) and by generally enhancing the quality of services provided through AI-based infrastructure.

[0292] Furthermore, according to the present invention, a terminal device (100, Terminal / UE / UENF) can transmit and link AI information to a core network (particularly, DCF (200)) and perform a visibility verification procedure to check the status of the AI ​​information.

[0293] The reason a visibility verification procedure is required is that when a terminal device (100, Terminal / UE / UENF) transmits AI information to an external source (e.g., Consumer (300)) via the core network, both the terminal device (100, Terminal / UE / UENF) and the core network can verify whether the AI ​​information has been transmitted correctly, whether there are any errors during transmission, and whether the AI ​​information is being processed in the correct form.

[0294] Ultimately, according to the present invention, a visibility verification procedure is defined / proposed to observe and track how transmitted AI information of a terminal device (100, Terminal / UE / UENF) moves and is processed within the entire transmission system.

[0295] Hereinafter, with reference to FIG. 7, the call flow of visibility verification and subsequent control according to the AI ​​information transmission method of the present invention will be explained.

[0296] The UE (100, AI agent) transmits / transmits AI information using the AI ​​information transmission session set up in the previously selected Plane path (e.g., UP, CP, MP, etc.).

[0297] A DCF (200, e.g., NWDAF) within a core network that receives AI information from a UE (100, AI agent) can perform a visibility verification procedure by checking items regarding the AI ​​information to verify the status of the AI ​​information (S300, S350) before delivering the AI ​​information to a Consumer (300, e.g., AF, AS, NF, etc.) and delivering the verified status information of the AI ​​information to the UE (100, AI agent) (S310, S360).

[0298] Meanwhile, in FIG. 7, DCF (200) is shown performing the visibility verification procedure, but this is only one embodiment and a separate other NF may be responsible for the visibility verification procedure.

[0299] In the present invention, the visibility verification procedure may proceed as a process of checking at least one item among authentication / identification, authorization, validation, verification, and acceptance.

[0300] In this context, the Authentication / Identification item refers to a series of authentication processes, such as verifying who the user of the AI ​​model (e.g., AI / ML) of the relevant AI information is, or whether the terminal / system is trustworthy; examples include login and fingerprint / face recognition.

[0301] Furthermore, authorization is a series of permission granting / authorization processes, such as determining whether an authenticated user has the authority to perform specific tasks on an AI model (e.g., AI / ML) or granting access rights to systems or data (AI information); examples of this include access control and resource restriction.

[0302] Furthermore, validation is a series of processes that checks whether input data (AI information) or requests of an AI model (e.g., AI / ML) conform to predefined rules or formats; examples include AIML Input Check, Sanity Check, and format checking.

[0303] Furthermore, verification is a series of processes to confirm whether data (AI information) is accurate or genuine, or whether it operates or functions normally. For example, in AI / ML testing, it is related to checking, confirming, and proving whether functions operate according to requirements.

[0304] Furthermore, Acceptance is a series of processes for determining or approving whether input data (AI information), requests, or results of a transmitted AI model (e.g., AI / ML) are ultimately acceptable; for example, it is related to Approval, Consent, and Admission regarding AI / ML.

[0305] Accordingly, the DCF (200, e.g., NWDAF) within the core network, which receives AI information from the UE (100, AI agent), checks the items of the tactical method regarding the AI ​​information to verify the status of the AI ​​information (S300, S350), and this verification can be performed sequentially for each item or in parallel.

[0306] And the DCF (200, e.g., NWDAF) within the core network can periodically transmit status information of the AI ​​information verified in step S300 or S350 to the UE (100, AI agent) (S310, S360).

[0307] In this way, the UE (100, AI agent) can continuously update the status of its AI information transmission internally based on the status information of the AI ​​information verified through a visibility verification process.

[0308] At this time, the status information of the AI ​​information confirmed (transmitted) through the visibility verification procedure may differ depending on whether the visibility verification procedure is performed by the DCF (200, e.g., NWDAF) within the core network before transmitting the AI ​​information to the Consumer (300, e.g., AF, AS, NF, etc.) or after transmitting it to the Consumer (300, e.g., AF, AS, NF, etc.).

[0309] For example, when a DCF (200, e.g., NWDAF) within the core network performs a visibility verification procedure after delivering AI information to a Consumer (300, e.g., AF, AS, NF, etc.), the "state information of AI information" verified (delivered) may be in the form of adding the result processed up to the Consumer (300, e.g., AF, AS, NF, etc.) to the "state information of AI information" verified (delivered) when a visibility verification procedure is performed prior to delivery to the Consumer (300, e.g., AF, AS, NF, etc.).

[0310] And the UE (100, AI agent) may also perform predefined subsequent control according to the state information of the AI ​​information.

[0311] At this time, the predefined subsequent control may be a control action defined to be performed subsequently according to the status information of the AI ​​information confirmed through the visibility verification procedure (an item checked as abnormal among Authentication / Identification, Authorization, Validation, Verification, and Acceptance).

[0312] Specifically, the UE (100, AI agent) can transmit the AI ​​information using a session for transmitting AI information, and in the case where approval success (OK) is received as the status information of the AI ​​information according to the status information of the AI ​​information confirmed through the visibility verification procedure (S310), the AI ​​information can be delivered to the Consumer (300) who will use the AI ​​information through the core network (specifically, DCF (200)).

[0313] According to one embodiment, the UE (100, AI agent) may, when an approval failure (Not OK) is received as status information of AI information based on status information of AI information confirmed through a visibility verification procedure (S360), hold the use of the session for transmitting AI information and stop the transmission of AI information (S370).

[0314] Furthermore, the UE (100, AI agent) can perform subsequent control actions (e.g., initial setup retry (Re-initiation), Re-establishment, or re-authentication, etc.) based on the abnormal check item that caused the current approval failure (Not OK) (S375).

[0315] According to another embodiment, if a DCF (200, e.g., NWDAF) within a core network confirms that an acknowledgment failure (Not OK) is confirmed as status information of AI information (S360), then even if it receives the corresponding AI information transmitted from the UE (100, AI agent), it may not use it or perform the action of transmitting it to the Consumer (300) (S380, transmission not performed).

[0316] As described above, according to embodiments of the present invention, in a basic structure in which a terminal (e.g., Terminal / UE / UENF) performs an AI Feature / operation using an AI model (e.g., AI / ML), a configuration, namely an initiation, modification, and termination procedure, is specifically defined to support the transmission of AI information from a terminal (e.g., Terminal / UE / UENF) performing an On-Device AI-based service to a core.

[0317] In addition, according to embodiments of the present invention, when a terminal device (100, Terminal / UE / UENF) transmits AI information to an external source (e.g., Consumer (300)) via a core or through a core, a visibility verification procedure is defined to enable the terminal device (100, Terminal / UE / UENF) and the core network to verify whether the AI ​​information has been transmitted correctly, whether there are any errors during transmission, and whether the AI ​​information is being processed in the correct form.

[0318] Meanwhile, the present invention defines and implements an information transmission method, i.e., an AI information transmission method, capable of transmitting information between a terminal (e.g., Terminal / UE / UENF) and a core network as described above. Based on this, it is also possible to transmit various data / information of a terminal (e.g., Terminal / UE / UENF) to the core network through a method of transmitting by including it in the AI ​​information or through a method of transmitting the AI ​​information.

[0319] To describe one embodiment, the UE (100, AI agent) can perform the operation of periodically transmitting (delivering) its state information according to the Capability Negotiation already conducted with a specific NF (e.g., NWDAF, AMF, SMF, PCF, etc.) within the core network.

[0320] In this way, a specific NF (e.g., NWDAF, AMF, SMF, PCF, etc.) within the core network can check the status information of the UE (100) and transmit (delivery) the necessary update information at the necessary time.

[0321] Here, the data / information of the terminal (e.g., Terminal / UE / UENF) may consist of log and trace types, and the terminal (e.g., Terminal / UE / UENF) can configure / learn / perform On-Device AI (e.g., AI / ML) based on the said content.

[0322] That is, an AI model (e.g., AI / ML) within a terminal (e.g., Terminal / UE / UENF) may be an AI / ML model configured based on the relevant data / information, or an AI / ML model that is continuously trained based on the relevant data / information.

[0323] A terminal (e.g., Terminal / UE / UENF) can perform AI operations (Inferencing) based on an AI model (e.g., AI / ML) configured / trained in this way.

[0324] In addition, the terminal (e.g., Terminal / UE / UENF) can transmit (transfer) some or all of the data / information to the core network, and the data / information of the terminal (e.g., Terminal / UE / UENF) can be classified into approximately 11 types as follows, and various log and trace information can be extracted for each type.

[0325] ·UE's Network, Protocol / Network / Telephony, System / UI, Application / Storage, Security, Location, Sensor, Media, Synch, Accessibility / System, diagnostic

[0326] Below, I will list and explain examples of log and trace information of each type regarding data / information of terminals (e.g., Terminal / UE / UENF).

[0327] ·UE Network-related: Signal strength (RSRP, RSRQ, SINR), Cell ID and sector information, Handover event, RRC connection establishment / release, RACH procedure, Radio bearer setup / modification, Radio link failure, Frequency band information, Modulation and Coding Scheme (MCS) information, MIMO config information, Power control parameters, Uplink / downlink throughput, Packet loss rate, Latency measurement results, Network slicing information, VoLTE / VoNR usage, eSIM provisioning, Carrier aggregation, Beam management, NSSAI (Network Slice Selection Assistance Information)

[0328] ·UE Protocol / Network / Telephony-related(단말 프로토콜 / 네트워크 / 통신 기능 관련): RRC message, NAS message, MAC layer, PDCP message, RLC message, IP packet data, TCP / UDP session, SCTP, HTTP / 2, HTTP / 3, QUIC protocol related, Wi-Fi connection / disconnection, Cellular / Mobile data connection / disconnection, Bluetooth connection, NFC usage, Hotspot / tethering, Airplane mode toggle, Roaming status, Network type switching, Wi-Fi calling, Network selection, Call history, SMS / MMS, Contact access, Voicemail access, SIM card change, Dual SIM usage, Emergency call, Call recording, RCS (Rich Communication Services),

[0329] ·System / UI-related(시스템 / UI 관련): UE Kernel, System boot, Process creation / termination, Memory usage, CPU usage, Battery consumption, Temperature sensor, Disk I / O, System call, Interrupt, Screen state change, Touch event, Keyboard input, Gesture recognition, Screen orientation change, Accessibility service usage, UI rendering performance, Custom view inflation, Dia / popup display, Navigation event

[0330] ·Application / Storage-related(애플리케이션 / 저장소 관련): App start / stop events, App crash, App performance metrics, User interaction, API call, Database query, Application network request / response, Push notification, Background task, In-app purchase, App installation, App uninstallation, App update, In-app purchase, App store browsing, App rating / review, App download, App update check, App compatibility check, App permission change

[0331] ·Security-related(보안 관련): Authentication attempt, Encryption key exchange, Security policy application, Malware detection, Firewall, VPN connection, Biometric authentication, Certificate validation, Secure element access, Trusted execution environment

[0332] ·Location-related(위치 관련): GPS coordinate, Cell-based location tracking, Wi-Fi-based location tracking, Geofencing event, Location permission request, A-GPS assistance data, GNSS raw measurement, Fusion location provider, Indoor positioning, Location spoofing detection

[0333] ·Sensor-related(센서 관련): Accelerometer data, Gyroscope data, Proximity sensor data, Ambient light sensor data, Pressure sensor data, Heart rate sensor data, Magnetometer data, Barometer data, Step counter data, Gesture recognition

[0334] ·Media-related(미디어 관련): Audio codec usage, Video codec usage, Microphone usage, Speaker usage, Camera usage, Audio routing, Video encoding / decoding, Audio / video synchronization, Noise cancellation, Voice recognition, Media playback, Media encoding / decoding, DRM-related, Streaming quality, Media scanner, Playlist modification, Album art retrieval, Subtitle processing, Media metadata extraction, Media casting

[0335] ·Synch-related(동기화 관련): File system access, Storage space usage, Data backup / restore, File encryption / decryption, Cache management, External storage mount / unmount, File integrity check, Storage performance, File transfer, Storage permission, Cloud sync, Local backup creation, Data restoration, Account sync, App data sync, Contacts sync, Calendar sync, Photo / video backup, Document sync, Bookmark sync

[0336] ·Accessibility / System-related(접근성 / 시스템 관련): Screen reader usage, Accessibility settings change, Switch access, Magnification usage, Color inversion, Captions usage, Text-to-speech engine, Accessibility shortcut usage, Accessibility timeout, Accessibility button usage, OS update check, system update download, System update installation, Patch level information, A / B system update, Recovery mode, Bootloader, OTA package verification, Update rollback, Firmware over-the-air (FOTA)

[0337] ·Diagnostic-related(진단 관련): debugging activation, Developer option change, command execution, Overlay drawing, GPU rendering, Strict mode violation, Memory leak detection, Layout bounds visualization, Show taps, Pointer location, Battery health status, Storage device health status, RAM diagnostic, CPU temperature, Display panel, Touchscreen calibration, Fingerprint sensor, Camera sensor, Microphone test, Speaker test

[0338] Furthermore, since the present invention defines and implements an AI information transmission method capable of transmitting information between a terminal (e.g., Terminal / UE / UENF) and a core network, it may be possible to transmit information (such as Snapshot, PI request / response, etc. described later) to the core network to receive the procedure information (PI) that the terminal (e.g., Terminal / UE / UENF) itself must perform, through a method of transmitting the AI ​​information or a method of transmitting the AI ​​information.

[0339] To explain in detail, in the present invention, a Procedure Function (PF) may be defined as a specific function / NF that uses a generative language model (e.g., LLM / GenAI) within a core network. Of course, the name of the PF is merely an example and may be defined by a different name.

[0340] However, for the sake of convenience of explanation, the following examples will be described as cases where the PF as a specific function / NF using a generative language model (e.g., LLM / GenAI) within the core network is NWDAF.

[0341] In this case, NWDAF(LLM / GenAI) is responsible for determining / generating the call processing procedure between the terminal (e.g., Terminal / UE / UENF) and the core network based on LLM / GenAI.

[0342] The call processing procedure mentioned here refers to a procedure based on signaling messages transmitted between terminals (e.g., Terminal / UE / UENF) and NFs, and between NFs, through mutual interfaces.

[0343] The signaling messages constituting this may be transport or application layer messages such as SCTP, GTP-C, HTTP / 2, HTTP / 3, QUIC, etc.

[0344] In addition, this signaling message is exchanged between 3GPP-based terminals / base stations / cores, or among themselves, and may be related to execution procedures including call processing based on requests from terminals or each network unit (e.g., Connection, Registration, Mobility, Session, Radio Resource, Bearer, Identity, AMF, Location, Policy, Security Management).

[0345] And each message can be various unit information defined by 3GPP, such as UE context, NF context, mobility management context, session management context, security context, access stratum context, non-access stratum context, bearer context, radio resource control context, UE context management context, handover context, paging context, and QoS flow context information. For example, UE context can be configuration information combining the terminal's idle / active state, battery level, and power level, while mobility can be configuration information combining the terminal's mobility.

[0346] The call processing procedure, i.e., Procedure, received by a terminal (e.g., Terminal / UE / UENF) or NF from NWDAF (LLM / GenAI) by applying the AI ​​information transmission method of the present invention can be described as information for performing the next operation / signaling processing for each Procedure required by the terminal (e.g., Terminal / UE / UENF) or NF.

[0347] For example, as an example of the call processing procedure mentioned in the present invention, a procedure can be described in which each of the three NFs (NF1, 2, 3) transmit and receive specific signaling messages to and from each other.

[0348] For example, NF1 to NF2 can be Connection Management Procedures.

[0349] And NF2 ~ NF3 can be Session Management Procedures.

[0350] And NF3 ~ NF1 can be Mobility Management Procedures.

[0351] In other words, the call processing procedure, or Procedure, proceeds through signaling messages between each NF and can be divided into various types of Procedures.

[0352] A Procedure is not limited to a 1:1 relationship between NFs, but can be defined as involving multiple NFs depending on the type of Procedure, and communication between NFs can be unidirectional (one-way communication, e.g., send_only) or bidirectional (two-way communication, send / receive). Additionally, from a specific perspective, a Procedure can be a request and response, or a subscription and notification for temporary or continuous information delivery requests.

[0353] The PI received by the terminal (e.g., Terminal / UE / UENF) or NF from NWDAF(LLM / GenAI), that is, the procedure information for executing a procedure determined / generated by NWDAF(LLM / GenAI) through LLM / GenAI-based processing (Inferencing), may be a method of transmitting and processing signaling messages between the terminal (e.g., Terminal / UE / UENF) and NF, or between NFs, as an appropriate guide considering the current network for the next operation, and may be an operation of changing the version, software, etc. of the terminal (e.g., Terminal / UE / UENF) or NF itself.

[0354] Terminals (e.g., Terminal / UE / UENF) or NWDAF(LLM / GenAI) can select the type of LLM / GenAI they can utilize by negotiating before, during, or after PI execution (e.g., ChatGPT type / version, GPT-4o, GPT-4o Mini, GPT-4.5, GPT 5.x).

[0355] Specifically, when the AI ​​information transmission method of the present invention is applied to an example of PI request / PI response exchange, the UE (100) performs existing call processing through message transmission and reception with various NFs in the core network, and for each call processing, extracts (e.g., captures) information generated from internal and / or external processing logic, i.e., internal and external information, and configures and stores it as a snapshot.

[0356] In addition, in the present invention, when storing a snapshot, information regarding the time or period during which the snapshot was configured (e.g., capture, gathering, etc.) may be included and stored.

[0357] Accordingly, in the present invention, the UE (100) can extract (e.g., capture) information generated from internal and / or external processing logic, i.e., internal and external information, and store a snapshot representing its call processing status by configuring / storing these, and can utilize its own snapshot when necessary.

[0358] In this case, as a method utilizing snapshots, the UE (100, AI agent) can transmit its snapshot to NWDAF (LLM / GenAI).

[0359] More specifically, UE (100, AI agent) can send its Snapshot to NWDAF (LLM / GenAI) in accordance with the Capability Negotiation previously conducted with NWDAF (LLM / GenAI).

[0360] Accordingly, NWDAF(LLM / GenAI) can perform additional training of LLM / GenAI using snapshots transmitted / exchanged from multiple UEs (100).

[0361] Of course, when the UE (100, AI agent) receives a specific event, it can generate a Prompt to request a Procedure (PI) to be executed based on the message of the specific event, send the PI request including this Prompt to the NWDAF (LLM / GenAI), and receive a PI response.

[0362] For example, if the UE (100, AI agent) has a Procedure for this specific event in its internal storage, it can proceed with the next action, i.e., call processing, accordingly. However, if the Procedure does not exist in its internal storage or if the message cannot be processed by the call processing logic within the UE (100), it determines that a Procedure is required, generates a Prompt, i.e., a Text Input, as described above, and transmits a PI request containing this to the NWDAF (LLM / GenAI).

[0363] More specifically, the UE (100, AI agent) can exchange PI requests / responses by sending a PI request to the NWDAF (LLM / GenAI) and receiving a PI response from the NWDAF (LLM / GenAI) in accordance with the Capability Negotiation conducted with the NWDAF (LLM / GenAI).

[0364] And, the UE (100, AI agent) that has exchanged the current PI request / response with NWDAF (LLM / GenAI) can process the specific event that occurred this time by performing / executing the next action according to the PI response (e.g., action according to the text of the PI being responded to).

[0365] More specifically, when the UE (100, AI agent) determines that it needs to request a call processing procedure (e.g., when receiving an unprocessable message / signal / data, etc.), it can generate / configure a textualized input, i.e., a Prompt, to be transmitted to the NWDAF (LLM / GenAI) via an On Device-AI (e.g., including an LLM / GenAI Client), and transmit a PI request containing the Prompt to the NWDAF (LLM / GenAI). At this time, the scope of the PI request depends on the Prompt information generated by the UE (100).

[0366] In this way, NWDAF(LLM / GenAI) creates / determines a Procedure that can be performed by the UE (100) using LLM / GenAI based on the PI request and transmits it to the UE (100) as a PI, and the UE (100, AI agent) can handle the situation that caused the current PI request (e.g., receiving an unprocessable message / signal / data, etc.) by performing / executing the next action according to the PI transmitted / responded from NWDAF(LLM / GenAI).

[0367] In this case, according to the present invention, a terminal (e.g., Terminal / UE / UENF) involved in processing a specific event and each NF can extract information, i.e., internal and external information, generated from internal and / or external processing logic (e.g., Capture) and save a snapshot (e.g., new save, existing update, etc.).

[0368] Meanwhile, in addition to the terminal (e.g., Terminal / UE / UENF) of the embodiment illustrated in FIG. 6, each NF (e.g., RANF, AMF, SMF, UPF, PCF, NRF, OAM, UDR, UDSF, AF,...) may transmit a PI request to NWDAF(LLM / GenAI) when deemed necessary, and NWDAF(LLM / GenAI) may generate / determine a procedure that can be executed at the corresponding NF based on the PI request and transmit it to the corresponding NF as a PI response (exchange of information (Snapshot, PI request / response) between NFs).

[0369] As described above, if the AI ​​information transmission method of the present invention is applied to an example of PI request / PI response exchange, through the exchange of information (Snapshot, PI request / response) between a terminal (e.g., Terminal / UE / UENF) and a core network (e.g., LLM / GenAI), it is possible to implement interoperability with a core generative language model (e.g., LLM / GenAI) that generates / transmits a 3GPP standard-based call processing procedure, thereby enabling the terminal (e.g., Terminal / UE / UENF) to automatically perform various roles, functions, and operations.

[0370] As a result, in the present invention, in addition to the call processing of functions given during design in the terminal (e.g., Terminal / UE / UENF) and related NF, optimal call processing generated / transmitted can be performed, thereby enabling automatic detection and response to operational improvements, abnormalities, and abnormalities of each NF involved in the terminal and call processing, and furthermore, the call processing that must be implemented during design / development in the implementation of the terminal (e.g., Terminal / UE / UENF) / NF can be minimized.

[0371] Furthermore, since the present invention defines and implements an AI information transmission method capable of exchanging information between a terminal (e.g., Terminal / UE / UENF) and a core network, it may be possible to exchange information (state-machine information described below) with the core network to support analysis (e.g., terminal status, abnormal operation, network abnormal situation (e.g., signaling storm), etc.) in the core network, particularly in NWDAF (e.g., AI / ML), through a transmission method such as including and transmitting the AI ​​information or transmitting the AI ​​information.

[0372] Specifically, the present invention considers a basic structure in which information to be used for analysis in NWDAF is collected in the form of input from terminals (e.g., Terminal / UE / UENF), various / multiple NFs (e.g., RANF, AMF, SMF, UPF, PCF, NRF, OAM, UDR, UDSF, AF,...), and SCP (Service Communication Proxy), and NWDAF transmits the analysis results (e.g., statistics and / or predictions regarding network abnormal situations) through various analysis engines (e.g., AI / ML engine) to the relevant terminals (e.g., Terminal / UE / UENF) and various / multiple NFs in the form of output.

[0373] Ultimately, when the AI ​​information transmission method of the present invention is applied to an embodiment that exchanges for analysis support in NWDAF (e.g., AI / ML), the information transmitted to the NWDAF in the form of input refers to specific information (collected data) collected for analysis from the perspective of the NWDAF, and for the convenience of explanation, this information is described by the name "state-machine information."

[0374] Accordingly, the UE (100, AI agent) of the present invention can communicate with a core network, particularly an NWDAF (e.g., AI / ML), by transmitting specific information, i.e., state-machine information, at a necessary time through On Device-AI-based analysis, by including it in the AI ​​information and transmitting it, or by transmitting the AI ​​information.

[0375] The point in time required at this time refers to the point in time when an exceptional state deviating from a predefined normal operation is predicted through On Device-AI-based analysis operating in the UE (100).

[0376] The state-machine information transmitted to / exchanged with NWDAF (e.g., AI / ML) for the use of such collected data may be defined to include information indicating the predicted exceptional status (Unexpected Status) at a time when it is predicted through On Device-AI-based analysis operating on the UE (100).

[0377] In this case, examples of exceptional states can be of Operational and Functional types and may represent quality / performance / behavior (e.g., Operational fault, error, failure, misbehaving, abnormality, instability, etc.).

[0378] That is, the information collected in the form of input to the NWDAF (e.g., AI / ML) for analysis includes specific information transmitted / exchanged from a UE (100) capable of performing AI / ML independently (On Device-AI based analysis), namely state-machine information.

[0379] More specifically, the UE (100, AI agent) can transmit state-machine information to the NWDAF (e.g., AI / ML) based on the Capability Negotiation previously conducted with the NWDAF (e.g., AI / ML).

[0380] Specific information, namely state-machine information, may also include the location of the terminal (e.g., Terminal / UE / UENF) (GPS and positioning / location related information), radio resources of the terminal (e.g., Terminal / UE / UENF) (UE RRC, RB usage, UL / DL SNR, RSSI / RSSP, Radio Frequency), hardware resource information of the terminal (e.g., Terminal / UE / UENF) (CPU / DPU / NPU, Memory, Modem), and software resource information (OS, Application).

[0381] In addition, state-machine information may also include the following information.

[0382] · External NF integration status of terminals and NFs, and message statistics, request-response success rates, and related statistics by terminal interface

[0383] · Load information for terminals and interconnected multiple terminals

[0384] · Request-response time / whether retransmission occurs during the interoperability process with external base stations / NFs, etc.

[0385] · Related information such as NF / UE used for services provided by the terminal (including UE / NF Context information used for call processing - timer, state, subscriber information, ...)

[0386] · Power consumption information during terminal and NF operation execution

[0387] And, terminal (e.g., Terminal / UE / UENF) features can represent the following information.

[0388] · Information on service request / processing target terminal(group)

[0389] · Location, Area, Model, Slice, Signaling Pattern, Capability, Requested Service, Usage (Statistics, etc.), Common Set Information of Requested Terminals (Groups)

[0390] And the operations performed by a terminal (e.g., Terminal / UE / UENF) to support analysis in NWDAF (e.g., AI / ML) (e.g., prediction through On Device-AI-based analysis, transmission / exchange of state-machine information) can be executed during the call processing execution procedure (Management Procedure) at the terminal (e.g., Terminal / UE / UENF) (e.g., Connection, Registration, Mobility, Session, Radio Resource, Bearer, Identity, AMF, Location, Policy, Security Management).

[0391] And the status / message during the Management Procedure can be UE context, NF context, mobility management context, session management context, security context, access stratum context, non-access stratum context, bearer context, radio resource control context, UE context management context, handover context, paging context, and QoS flow context information. For example, UE context can be configuration information combining the terminal's idle / active state, battery level, and power level, and mobility can be configuration information combining the terminal's mobility.

[0392] Meanwhile, information collected in NWDAF (e.g., AI / ML) as input for analysis may also include various NFs (e.g., UENF, RANF, AMF, SMF, UPF, PCF, NRF, OAM, UDR, UDSF, AF,...) and information transmitted from SCP (state-machine information).

[0393] Of course, if each NF (e.g., UENF, RANF, AMF, SMF, UPF, PCF, NRF, OAM, UDR, UDSF, AF,...) also supports On Device-AI based analysis, it can communicate with the core network, particularly NWDAF (e.g., AI / ML), by transmitting specific information, namely state-machine information, to the AI ​​information at the necessary time through On Device-AI based analysis, or by transmitting it in a manner similar to transmitting AI information.

[0394] As described in the tactical example, if the AI ​​information transmission method of the present invention is applied to an example of information exchange (also known as state-machine information) for analysis support in NWDAF (e.g., AI / ML), the NWDAF (e.g., AI / ML) can analyze and predict abnormal situations (e.g., signaling storm, etc.) in the terminal (e.g., Terminal / UE / UENF) and core more quickly and accurately.

[0395] As described above, according to embodiments of the present invention, in a basic structure in which a terminal (e.g., Terminal / UE / UENF) performs an AI Feature / operation using an AI model (e.g., AI / ML), a method (AI information transmission method) for transmitting data (AI information) related to the terminal's On-Device AI to a specific NF (e.g., DCF) within the core is defined / implemented, and a Capability Negotiation procedure for applying AI information transmission is defined / implemented.

[0396] In addition, according to embodiments of the present invention, configuration, namely initiation, modification, and termination procedures, are specifically defined to support the transmission of AI information from a terminal (e.g., Terminal / UE / UENF) performing an on-device AI-based service to a core.

[0397] In addition, according to embodiments of the present invention, when a terminal device (100, Terminal / UE / UENF) transmits AI information to an external source (e.g., Consumer (300)) via a core or through a core, a visibility verification procedure is defined to enable the terminal device (100, Terminal / UE / UENF) and the core network to verify whether the AI ​​information has been transmitted correctly, whether there are any errors during transmission, and whether the AI ​​information is being processed in the correct form.

[0398] Thus, according to the present invention, since the Core or a consumer (e.g., AF, AS, NF, etc.) communicating with the Core can collect and use AI information on the terminal side (e.g., Terminal / UE / UENF), the effect of providing a data (AI information) collection solution is derived in which the Core or the consumer (e.g., AF, AS, NF, etc.) can collect data (AI information) for training an AI model on the terminal side.

[0399] In addition, according to the present invention, the performance and quality of the entire system can be improved, such as by improving AI performance performed at terminals (e.g., Terminal / UE / UENF) and core networks (AI Feature) and by generally enhancing the quality of services provided through AI-based infrastructure.

[0400] In addition, the AI ​​information transmission method of the present invention can be applied to / utilized in various embodiments of information transmission between cores, such as utilizing it to exchange information with a core network to receive a call processing procedure (PI) to be performed by a terminal (e.g., Terminal / UE / UENF), or utilizing it to exchange information to support analysis in NWDAF (e.g., AI / ML).

[0401] The AI ​​information transmission method according to embodiments of the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either alone or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the present invention, or those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.

[0402] Although the present invention has been described in detail with reference to preferred embodiments, the present invention is not limited to the above-mentioned embodiments, and the technical concept of the present invention extends to the scope in which various modifications or alterations are possible by anyone with ordinary knowledge in the technical field to which the present invention belongs, without departing from the gist of the present invention as claimed in the following claims.

Claims

1. In a terminal device, Memory containing instructions; and A terminal device characterized by including a processor that, by executing the above command, performs a configuration for transmitting AI information related to an AI model within the terminal device, transmits the AI ​​information to a core network based on the above configuration, and delivers the AI ​​information to a consumer who will use the AI ​​information through the core network.

2. In Paragraph 1, The settings for transmitting the above AI information are, Initial setup (Initiation) procedure for establishing an AI information transmission session to be used for AI information transmission, A modification procedure for modifying part or all of the AI ​​information transmission session configured above, A terminal device characterized by including a termination procedure for releasing the above-mentioned AI information transmission session.

3. In Paragraph 2, The above initial setup (Initiation) procedure is, It is initiated when the above terminal device determines that AI information transmission is necessary, and A terminal device characterized by including a process of establishing a session for transmitting AI information by performing resource allocation for transmitting AI information based on a Plane path selected for transmitting AI information among a plurality of Planes between the terminal and the core network.

4. In Paragraph 2, The above modification procedure is, A process of modifying part or all of an AI information transmission session according to pre-set modification conditions while the AI ​​information transmission based on the above AI information transmission session is activated, A terminal device characterized by including a process of storing transmission status information related to the pause, hold, or buffering in the terminal device for AI information that is paused, held, or buffered during the above modification process.

5. In Paragraph 2, The above termination procedure is, A process of releasing resources and Plane paths of the AI ​​information transmission session according to pre-set release conditions while the AI ​​information transmission based on the AI ​​information transmission session is enabled, A terminal device characterized by including a process of notifying the terminal device and the core network of detailed information related to transmission interruption regarding AI information where transmission interruption occurs during the above-mentioned release process.

6. In Paragraph 1, The above consumer, A terminal device characterized by being a specific Network Function (NF) within the core network, or a server or platform that interacts with the AI ​​model within the terminal device.

7. In Paragraph 1, The above processor is, A terminal device characterized by transmitting the above AI information to the above core network and then performing a visibility verification procedure to check the status of the above AI information in conjunction with the above core network.

8. In Paragraph 7, The above visibility verification procedure is, A process in which a responsible NF checks at least one item among authentication / identification, authorization, validation, verification, and acceptance regarding the AI ​​information transmitted from the terminal device to check the status of the AI ​​information, A terminal device characterized by including a process in which the above-mentioned responsible NF periodically transmits status information of the above-mentioned confirmed AI information to the terminal device.

9. In Paragraph 7, The above processor is, Based on the status information of the AI ​​information confirmed through the above visibility verification procedure, the status of the AI ​​information transmission is continuously updated internally, and A terminal device characterized by performing a predefined subsequent control according to the above state information.

10. In Paragraph 9, The above visibility verification procedure includes, A terminal device characterized by including a process in which the core network checks at least one item among authentication / identification, authorization, validation, verification, and acceptance with respect to the AI ​​information transmitted from the terminal device.

11. In Paragraph 10, The subsequent control defined above is, A terminal device characterized by control operations defined to be subsequently performed according to an item checked as abnormal among the authentication / identification, authorization, validation, verification, and acceptance of the above-mentioned status information.

12. In a method for transmitting AI information performed on a terminal device, A configuration step for performing a configuration for transmitting AI information related to an AI model within the terminal device; A method for transmitting AI information characterized by including an information transmission step of transmitting the AI ​​information to a core network based on the above settings, so that the AI ​​information is delivered to a consumer who will use the AI ​​information through the core network.

13. In Paragraph 12, The settings for transmitting the above AI information are, Initial setup (Initiation) procedure for establishing an AI information transmission session to be used for AI information transmission, A modification procedure for modifying part or all of the AI ​​information transmission session configured above, A method for transmitting AI information characterized by including a termination procedure for releasing the session for transmitting the AI ​​information.