Method and apparatus for congestion control in wireless communication system

Abstract

The present disclosure relates to an operation method of a network function in a wireless communication system, and the operation method comprises the steps of: acquiring AI / ML-related information from an AF; determining whether to perform congestion control for the first terminal based on the acquired AI / ML related information; receiving an NAS message from the first terminal; and performing congestion control based on the NAS message, in which whether to perform congestion control for the first terminal may be determined based on whether the first terminal participates in the AI / ML operation.

Description

Technical Field

[0001] The following description relates to a wireless communication system, and to a method and apparatus for performing congestion control. Specifically, the present invention relates to a method for determining whether to perform congestion control based on whether a terminal participates in AI / ML (artificial intelligence / machine learning) operation. Background Art

[0002] Wireless communication systems have been widely deployed to provide various types of communication services, such as voice or data. Generally speaking, wireless communication systems are multiple access systems that support communication for multiple users by sharing available system resources (bandwidth, transmit power, etc.). Examples of multiple access systems include code division multiple access (CDMA) systems, frequency division multiple access (FDMA) systems, time division multiple access (TDMA) systems, orthogonal frequency division multiple access (OFDMA) systems, and single carrier frequency division multiple access (SC-FDMA) systems.

[0003] Specifically, because a large number of communication devices require a large communication capacity, an enhanced mobile broadband (eMBB) communication technology is being proposed compared to a conventional radio access technology (RAT). In addition, not only a large-scale machine type communication (massive MTC) that provides various services anytime and anywhere by connecting multiple devices and objects is being proposed, but also a communication system that takes into account services / user terminal equipment (UE) that are sensitive to reliability and latency is being proposed. Various technical configurations are proposed for this purpose. Summary of the invention

[0004] Technical issues

[0005] The present disclosure relates to a method and apparatus for performing congestion control in a wireless communication system.

[0006] The present disclosure relates to a method and apparatus for obtaining AI / ML operation-related information provided from an AF (Application Function) in a wireless communication system and performing congestion control based on whether a terminal participates in the AI / ML operation.

[0007] The present disclosure relates to a method and apparatus for performing congestion control based on whether multiple terminals participate in an AI / ML operation in which the terminals participate in a wireless communication system.

[0008] The present disclosure relates to a method and apparatus for performing congestion control based on whether a network function serves all multiple terminals participating in an AI / ML operation in which the terminals participate in a wireless communication system.

[0009] The technical objectives to be achieved by the present disclosure are not limited to the contents mentioned above, and those skilled in the art who apply the technical configuration of the present disclosure can consider other technical objectives not mentioned herein from the embodiments of the present disclosure to be described below.

[0010] Technical Solution

[0011] As an example of the present disclosure, a method for operating a network function in a wireless communication system may include: obtaining artificial intelligence (AI) / machine learning (ML) related information from an application function (AF); determining whether to perform congestion control on a first terminal based on the obtained AI / ML related information; receiving a non-access stratum (NAS) message from the first terminal; and performing congestion control based on the NAS message. Whether to perform congestion control on the first terminal may be determined based on whether the first terminal participates in AI / ML operation.

[0012] In addition, as an example of the present disclosure, a network function operating in a wireless communication system may include: at least one transceiver; at least one processor and at least one memory, the at least one memory being operatively connected to the at least one processor and configured to store instructions, the instructions causing the at least one processor to perform specific operations when executed. The specific operation obtains artificial intelligence (AI) / machine learning (ML) related information from an application function (AF), determines whether to perform congestion control on a first terminal based on the obtained AI / ML related information, controls at least one transceiver to receive a non-access stratum (NAS) message from the first terminal, and performs congestion control based on the NAS message. Whether to perform congestion control on the first terminal may be determined based on whether the first terminal participates in AI / ML operations.

[0013] In addition, as an example of the present disclosure, a method for operating a terminal in a wireless communication system may include: sending a NAS message to a network function; and receiving a response message from the network function based on whether congestion control is performed on the terminal. The network function may obtain AI / ML related information from the AF and determine whether to perform congestion control on the terminal based on whether the terminal participates in AI / ML operation.

[0014] In addition, as an example of the present disclosure, a terminal operating in a wireless communication system may include: at least one transceiver; at least one processor; and at least one memory, the at least one memory being operatively connected to the at least one processor and configured to store instructions, the instructions causing the at least one processor to perform specific operations when executed. The specific operations may control at least one transceiver so that the terminal sends a NAS message to a network function; and control at least one transceiver to receive a response message from the network function based on whether congestion control is performed on the terminal. The network function may obtain AI / ML related information from the AF, and determine whether to perform congestion control on the terminal based on whether the terminal participates in AI / ML operations.

[0015] In addition, as an example of the present disclosure, a device may include at least one memory and at least one processor functionally connected to the at least one memory. The at least one processor may control the device to: obtain artificial intelligence (AI) / machine learning (ML) related information from an application function (AF); determine whether to perform congestion control on another device based on the obtained AI / ML related information; receive a non-access stratum (NAS) message from another device; and perform congestion control based on the NAS message. Whether to perform congestion control on another device may be determined based on whether the other device is involved in AI / ML operations.

[0016] In addition, as an example of the present disclosure, a non-transitory computer-readable medium storing at least one instruction may include at least one instruction that can be executed by a processor. The at least one instruction may control the device to: obtain artificial intelligence (AI) / machine learning (ML) related information from an application function (AF); determine whether to perform congestion control on another device based on the obtained AI / ML related information; receive a non-access stratum (NAS) message from another device; and perform congestion control based on the NAS message. Whether to perform congestion control on another device may be determined based on whether the other device participates in AI / ML operations.

[0017] In addition, the following contents can be applied together.

[0018] In addition, as an example of the present disclosure, based on the first terminal being a terminal that does not participate in AI / ML operations, the network function may determine whether to perform congestion control on the first terminal by considering the network load, and based on the first terminal being a terminal that participates in AI / ML operations, the network function may determine whether to perform congestion control on the first terminal by considering whether there are multiple terminals participating in AI / ML operations and whether the network function serves multiple terminals.

[0019] Furthermore, as an example of the present disclosure, based on the fact that the terminal participating in the AI / ML operation is only the first terminal, it may be determined whether to perform congestion control on the first terminal regardless of the AI / ML operation-related information.

[0020] In addition, as an example of the present disclosure, based on the presence of multiple terminals participating in AI / ML operations and the network function serving all multiple terminals, the network function can determine whether to perform congestion control on the first terminal by reflecting AI / ML-related information and status information of other terminals.

[0021] In addition, as an example of the present disclosure, the network function may be an access and mobility management function (AMF), and based on the status information of other terminals and the fact that the other terminals are in either user plane deactivation and CM-IDLE of AI / ML operation-related protocol data units (PDUs), the AMF may determine whether to perform congestion control on multiple terminals based on the possibility of sending NAS messages.

[0022] In addition, as an example of the present disclosure, based on performing congestion control on multiple terminals, AMF can set a back-off timer value for each of the multiple terminals so that the transmission delay termination time points of the multiple terminals are the same, and send the back-off timer value to each of the multiple terminals.

[0023] In addition, as an example of the present disclosure, based on the fact that other terminals are in either user plane activation and CM-CONNECTED of AI / ML operation-related protocol data units (PDUs), the AMF may determine not to perform congestion control on the first terminal based on the possibility of sending a NAS message.

[0024] In addition, as an example of the present disclosure, the network function can be a session management function (SMF), and based on whether the NAS message sent by the first terminal is associated with the PDU session of the AI / ML operation, the SMF can determine whether the first terminal participates in the AI / ML operation.

[0025] In addition, as an example of the present disclosure, the SMF may determine the possibility of sending a NAS message based on a quality of service (QoS) flow included in a PDU session used by other terminals for AI / ML operations as status information of other terminals, and the SMF may determine whether to perform congestion control on multiple terminals based on the possibility of sending the NAS message.

[0026] In addition, as an example of the present disclosure, based on performing congestion control on multiple terminals, the SMF can set a backoff timer value for each of the multiple terminals so that the transmission delay termination time points of the multiple terminals are the same, and send the backoff timer value to each of the multiple terminals.

[0027] In addition, as an example of the present disclosure, based on multiple terminals participating in AI / ML operations and the network function only serving some of the multiple terminals, the network function may determine not to perform congestion control on the first terminal based on AI / ML related information.

[0028] In addition, as an example of the present disclosure, based on the fact that multiple terminals participate in AI / ML operations and the network function only serves some of the multiple terminals, the network function can determine whether to perform congestion control on the first terminal based on AI / ML related information and AI / ML operation importance information.

[0029] In addition, as an example of the present disclosure, the AI / ML information may include at least one of identification information of a terminal participating in the AI / ML operation, AI / ML operation start time information, AI / ML operation required time information, AI / ML operation end time information, or AI / ML operation importance information.

[0030] In addition, as an example of the present disclosure, at least one information included in the AI / ML related information may be simultaneously transmitted to the core network.

[0031] In addition, as an example of the present disclosure, some of at least one information included in the AI / ML related information may be sent to the core network at a first time point, and other information may be sent to the core network at a second time point.

[0032] Beneficial Effects

[0033] The present disclosure may provide a method for performing congestion control in a wireless communication system.

[0034] The present disclosure may provide a method of obtaining AI / ML operation-related information provided from an AF in a wireless communication system and performing congestion control based on whether a terminal participates in the AI / ML operation.

[0035] The present disclosure may provide a method for performing congestion control based on whether a plurality of terminals participate in an AI / ML operation in which the terminal participates in a wireless communication system.

[0036] The present disclosure may provide a method for performing congestion control based on whether a network function serves all multiple terminals participating in an AI / ML operation in which the terminal participates in a wireless communication system.

[0037] The present disclosure has the effect of preventing AI / ML operations from being delayed or adversely affected based on congestion control operations in a wireless communication system.

[0038] The technical objectives to be achieved in the present disclosure are not limited to the above contents, and those skilled in the art who apply the technical configuration of the present disclosure may consider other unmentioned technical objectives from the embodiments of the present disclosure described below.

[0039] Those skilled in the art will understand that the effects that can be achieved through the embodiments of the present disclosure are not limited to the above effects, and other advantageous effects of the present disclosure will be more clearly understood through the following detailed description. That is, those skilled in the art can derive unexpected effects according to the implementation of the present disclosure from the embodiments of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings are provided to help understand the present disclosure, and embodiments of the present disclosure may be provided together with the detailed description. However, the technical features of the present disclosure are not limited to specific drawings, and the features disclosed in each drawing may be combined with each other to form new embodiments. The reference numerals in each drawing may refer to structural elements.

[0041] Figure 1 is a diagram showing an example of a communication system applicable to the present disclosure.

[0042] Figure 2 is a view showing an example of a UE to which an embodiment of the present disclosure is applied.

[0043] Figure 3 is a view showing an example of functional separation of a general NG-RAN and a fifth generation core (5GC).

[0044] Figure 4 is a view showing an example of a general architecture of a fifth generation (5G) system.

[0045] Figure 5 is a diagram illustrating an AI / ML-based endpoint splitting operation applicable to the present disclosure.

[0046] Figure 6 is a diagram illustrating an AI / ML model distribution method applicable to the present disclosure.

[0047] Figure 7 is a diagram illustrating a method of performing federated learning-based learning applicable to the present disclosure.

[0048] Figure 8 is a diagram illustrating a method of performing congestion control in a network function applicable to the present disclosure.

[0049] Fig. 9 is a diagram illustrating a method of performing congestion control in a network function applicable to the present disclosure.

[0050] Fig.10 is a diagram illustrating a method of performing congestion control in a network function applicable to the present disclosure. DETAILED DESCRIPTION

[0051] The following embodiments are achieved by combining the structural elements and features of the present disclosure in a predetermined manner. Unless specified separately, each structural element or feature should be selectively considered. Each structural element or feature can be performed without being combined with other structural elements or features. In addition, some structural elements and / or features can be combined with each other to form embodiments of the present disclosure. The order of the operations described in the embodiments of the present disclosure can be changed. Some structural elements or features of an embodiment may be included in another embodiment, or may be replaced by corresponding structural elements or features of another embodiment.

[0052] In the description of the drawings, processes or steps that render the scope of the present disclosure unnecessarily ambiguous will be omitted, and processes or steps that can be understood by those skilled in the art will be omitted.

[0053] Throughout the specification, when a part "comprises" or "includes" a component, this indicates that other components are not excluded, but may be further included, unless otherwise specified. The terms "unit", "-or / er" and "module" described in this specification indicate a unit for processing at least one function or operation, which can be implemented by hardware, software and a combination thereof. In addition, "a" or "an", "one", "the" and similar related words can be used to include both singular and plural expressions, unless in the context of describing this specification (especially in the context of the following claims) it indicates something different from this specification or is obviously contradictory to the context.

[0054] In this specification, the embodiments of the present disclosure are described focusing on the relationship between data reception and transmission between a base station and a mobile station. Here, a base station means a terminal node of a network that performs direct communication with a mobile station. In this document, a specific operation described as being performed by a base station may be performed by an upper node of the base station in some cases.

[0055] That is, in a network consisting of a plurality of network nodes including a base station, various operations for communicating with a mobile station may be performed by a base station or a network node other than a base station. Here, "base station" may be replaced by terms such as "fixed station", "Node (Node) B", "eNode B (eNB)", "gNode B (gNB)", "ng-eNB", "advanced base station (ABS)" or "access point".

[0056] In addition, in an embodiment of the present disclosure, “terminal” may be replaced by terms such as “user terminal equipment (UE)”, “mobile station (MS)”, “subscriber station (SS)”, “mobile subscriber station (MSS)”, “mobile terminal” or “advanced mobile station (AMS)”.

[0057] In addition, the transmitting end refers to a fixed and / or mobile node that provides data services or voice services, and the receiving end means a fixed and / or mobile node that receives data services or voice services. Therefore, in the case of an uplink, the mobile station can be the transmitting end, and the base station can be the receiving end. Similarly, in the case of a downlink, the mobile station can be the receiving end, and the base station can be the transmitting end.

[0058] The embodiments of the present disclosure may be supported by standard documents disclosed in at least one of the following radio access systems: IEEE 802xx system, 3rd Generation Partnership Project (3GPP) system, 3GPP Long Term Evolution (LTE) system, 3GPP Fifth Generation (5G) New Radio (NR) system and 3GPP2 system, and specifically, the embodiments of the present disclosure may be supported by the following documents: 3GPP TS (Technical Specification) 38.211, 3GPP TS 38.212, 3GPP TS 38.213, 3GPP TS 38.321 and 3GPP TS 38.331.

[0059] In addition, the embodiments of the present disclosure are applicable to other radio access systems, but are not limited to the above-mentioned systems. As an example, they are applicable to systems applied after the 3GPP 5G NR system, and are not limited to a specific system.

[0060] That is to say, obvious steps and parts not described in the embodiments of the present disclosure can be described with reference to the above documents. In addition, all terms disclosed in this document can be interpreted by standard documents.

[0061] Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Figure 1 The detailed description disclosed above is intended to describe example embodiments of the present disclosure, and is not intended to illustrate any only embodiment of the technical configuration capable of implementing the present disclosure.

[0062] In addition, specific terms used in the embodiments of the present disclosure are provided to help understanding of the present disclosure, and these specific terms can be used in any other modified forms without departing from the technical idea of ​​the present disclosure.

[0063] The following technology can be applied to various radio access systems such as code division multiple access (CDMA), frequency division multiple access (FDMA), time division multiple access (TDMA), orthogonal frequency division multiple access (OFDMA), single carrier frequency division multiple access (SC-FDMA), etc.

[0064] For the purpose of explanation, the following description is based on a 3GPP communication system (e.g., LTE, NR, etc.), but the technical idea of ​​the present disclosure is not limited thereto. LTE may refer to technology after 3GPP TS 36.xxx version 8. Specifically, LTE technology after 3GPP TS 36.xxx version 10 may be referred to as LTE-A, and technology after 3GPP TS 36.xxx version 13 may be referred to as pre-LTE-A. 3GPP NR may refer to technology after TS 38.xxx version 15. 3GPP 6G may refer to technology after TS version 17 and / or version 18. "xxx" means the specific number of a standard document. LTE / NR / 6G may be collectively referred to as a 3GPP system.

[0065] Contents described in standard documents issued earlier than the present disclosure may be referenced for background technology, terminology and abbreviations used in the present disclosure. As an example, 36.xxx and 38.xxx standard documents may be referenced.

[0066] For terms, abbreviations and other backgrounds that may be used in this document, reference may be made to the following standard document descriptions published prior to this document. Specifically, terms, abbreviations and other background technologies related to LTE / EPS (Evolved Packet System) may refer to the 36.xxx series, 23.xxx series and 24.xxx series, and NR (New Radio) / 5GS related terms and abbreviations and other backgrounds may refer to the 38.xxx series, 23.xxx series and 24.xxx series.

[0067] Hereinafter, the present disclosure is described based on the terms defined above.

[0068] The three main demand areas of 5G include (1) enhanced mobile broadband (eMBB), (2) massive machine type communication (mMTC), and (3) ultra-reliable low latency communication (URLLC).

[0069] Some use cases may require multiple areas to be optimized, and others may focus on just one key performance indicator (KPI). 5G supports these different use cases in a flexible and reliable way.

[0070] Communication system applicable to the present disclosure

[0071] Although not limited thereto, the various descriptions, functions, processes, proposals, methods and / or operational flowcharts of the present disclosure disclosed herein may be applied to various fields requiring wireless communication / connection (e.g., 5G) between devices.

[0072] In the following, a more detailed description will be given with reference to the accompanying drawings. In the following drawings / descriptions, unless otherwise mentioned, the same reference numerals may illustrate the same or corresponding hardware blocks, software blocks or functional blocks.

[0073] Figure 1 An example communication system suitable for use with the present disclosure is illustrated.

[0074] refer to Figure 1 , the communication system applied to the present disclosure includes a wireless device, a base station and a network. Here, a wireless device refers to a device for performing communication using a wireless connection technology (e.g., 5G NR, LTE), and may be referred to as a communication / wireless / 5G device. Without being limited thereto, the wireless device may include a robot 100a, a vehicle 100b-1 and 100b-2, an extended reality (XR) device 100c, a handheld device 100d, a home appliance 100e, an Internet of Things (IoT) device 100f, and an artificial intelligence (AI) device / server 100g. For example, a vehicle may include a vehicle equipped with a wireless communication function, an autonomous driving vehicle, a vehicle capable of performing vehicle-to-vehicle communication, and the like. Here, vehicles 100b-1 and 100b-2 may include an unmanned aerial vehicle (UAV)) (e.g., a drone). The XR device 100c may include an augmented reality (AR) / virtual reality (VR) / mixed reality (MR) device and may be implemented in the form of a head-up display (HUD) mounted on a vehicle, a television, a smartphone, a computer, a wearable device, a home appliance, a digital signage, a vehicle, a robot, etc. The handheld device 100d may include a smartphone, a smart tablet, a wearable device (e.g., a smart watch, smart glasses), a computer (e.g., a laptop, etc.), etc. The home appliance 100e may include a television, a refrigerator, a washing machine, etc. The IoT device 100f may include a sensor, a smart meter, etc. For example, the base station 120 and the network 130 may be implemented as wireless devices, and a specific wireless device 120a may operate as a base station / network node for another wireless device.

[0075] The wireless devices 100a to 100f may be connected to the network 130 through the base station 120. AI technology may be applied to the wireless devices 100a to 100f, and the wireless devices 100a to 100f may be connected to the AI ​​server 100g through the network 130. The network 130 may be configured using a 3G network, a 4G (e.g., LTE) network, or a 5G (e.g., NR) network. The wireless devices 100a to 100f may communicate with each other through the base station 120 / network 130, but perform direct communication (e.g., side link communication) that does not pass through the base station 120 / network 130. For example, the vehicles 100b-1 and 100b-2 may perform direct communication (e.g., vehicle-to-vehicle (V2V) / vehicle-to-everything (V2X) communication). In addition, the IoT device 100f (e.g., a sensor) may perform direct communication with another IoT device (e.g., a sensor) or other wireless devices 100a to 100f.

[0076] Wireless communication / connection 150a, 150b and 150c can be established between wireless devices 100a to 100f / base station 120 and base station 120 / base station 120. Here, wireless communication / connection, such as uplink / downlink communication 150a, side link communication 150b (or D2D communication) and base station to base station communication 150c (e.g., relay, integrated access backhaul (IAB)), can be established through various wireless connection technologies (e.g., 5G NR). The wireless device and the base station / wireless device, and the base station and the wireless device can send / receive radio signals through wireless communication / connection 150a, 150b and 150c. For example, wireless communication / connection 150a, 150b and 150c can send / receive signals through various physical channels. To this end, based on various proposals of the present disclosure, various configuration information setting processes for sending / receiving radio signals, various signal processing processes (e.g., channel coding / decoding, modulation / demodulation, resource mapping / demapping, etc.), and at least a part of the resource allocation process can be performed.

[0077] Figure 2 is a view showing an example of a UE to which an embodiment of the present disclosure is applied.

[0078] refer to Figure 2 UE 200 may include a processor 202, a memory 204, a transceiver 206, one or more antennas 208, a power management module 241, a battery 242, a display 243, a keyboard 244, a subscriber identity module (SIM) card 245, a speaker 246, and a microphone 247.

[0079] The processor 202 may be configured to implement the descriptions, functions, processes, proposals, methods and / or operational flow charts disclosed in this specification. The processor 202 may be configured to control one or more different components of the UE 200 to implement the descriptions, functions, processes, proposals, methods and / or operational flow charts disclosed in this specification. The layers of the wireless interface protocol may be embodied in the processor 202. The processor 202 may include an ASIC, other chipsets, logic circuits and / or data processing devices. The processor 202 may be an application processor. The processor 202 may include at least one of a DSP, a central processing unit (CPU), a graphics processing unit (GPU), and a modem (modulator and demodulator).

[0080] The memory 204 may be operably coupled to the processor 202 and store various information for operating the processor 202. The memory 204 may include ROM, RAM, flash memory, memory card, storage medium and / or other storage devices. In the case of software implementation, the technology described herein may be implemented using a module (e.g., process, function, etc.) that executes the description, function, process, proposal, method and / or operation flow chart disclosed in this specification. The module may be stored in the memory 204 and executed by the processor 202. The memory 204 may be implemented in the processor 202 or outside the processor 202, in which case the memory 204 may be communicatively coupled to the processor 202 by various methods known in the art.

[0081] The transceiver 206 can be operably coupled to the processor 202 and send and / or receive radio signals. The transceiver 206 can include a transmitter and a receiver. The transceiver 206 can include a baseband circuit for processing radio frequency signals. The transceiver 206 can send and / or receive radio signals by controlling one or more antennas 208.

[0082] The power management module 241 may manage power of the processor 202 and / or the transceiver 206. The battery 242 may supply power to the power management module 241.

[0083] The display 243 may output a result processed by the processor 202. The keyboard 244 may receive an input to be used in the processor 202. The keyboard 244 may be displayed on the display 243.

[0084] SIM card 245 is an integrated circuit for securely storing an International Mobile Subscriber Identity (IMSI) and related keys, and can be used to identify and authenticate a user in a handheld telephone device such as a mobile phone or computer. In addition, contact information can be stored in many SIM cards.

[0085] The speaker 246 may output sound-related results processed in the processor 202. The microphone 247 may receive sound-related input to be used in the processor 202.

[0086] In an embodiment of the present specification, a UE may operate as a transmitting device in an uplink and as a receiving device in a downlink. In an embodiment of the present specification, a base station may operate as a receiving device in an UL and as a transmitting device in a DL. In the present specification, a base station may be referred to as a Node B (Node B), an eNode B (eNB), and a gNB, and may not be limited to a specific form.

[0087] In addition, as an example, the UE can be embodied in various forms according to the use example / service. The UE can be configured by various components, devices / components and / or modules. For example, each UE may include a communication device, a control device, a storage device and additional components. The communication device may be a communication circuit and a transceiver. For example, the communication circuit may include one or more processors and / or one or more memories. For example, the transceiver may include one or more transceivers and / or one or more antennas. The control device may be electrically coupled to the communication device, the storage device and the additional components, and control the overall operation of each UE. For example, the control device may control the electrical / mechanical operation of each UE based on the program / code / instruction / information stored in the storage device. The control device may send the information stored in the storage device to the outside (e.g., another communication device) via the communication device through a wireless / wired interface, or store the information received from the outside (e.g., another communication device) through the wireless / wired interface in the storage device.

[0088] The additional components may be configured in various ways depending on the type of UE. For example, the additional components may include at least one of a power supply device / battery, an input / output (I / O) device (e.g., an audio I / O port, a video I / O port), a drive device, and a computing device. In addition, without limitation, the UE may be configured in a variety of ways. Figure 1 The robot 100a, Figure 1 The vehicles 100b-1 and 100b-2, Figure 1 XR device 100c, Figure 1 Handheld device 100d, Figure 1 Home appliances 100e, Figure 1 IoT devices 100f, digital broadcast terminals, hologram devices, public safety devices, MTC devices, medical devices, fintech devices (or financial devices), security devices, climate / environmental devices, Figure 1 AI server / device 100g, Figure 1 The UE may be implemented in the form of a base station 120 and a network node. The UE may be used in a mobile or fixed location depending on the use case / service.

[0089] Various components, devices / components and / or all modules of the UE may be interconnected via a wired interface, or at least a portion thereof may be wirelessly connected via a communication device. In addition, each component, device / component and / or module of the UE may further include one or more components. For example, the control device may be configured by a collection of one or more processors. As an example, the control device may be configured by a collection of a communication control processor, an application processor (AP), an electronic control unit (ECU), a GPU, and a memory control processor. As another example, the storage device may be configured by RAM, dynamic RAM (DRAM), ROM, flash memory, volatile memory, non-volatile memory and / or a combination thereof.

[0090] 5G system architecture to which the present disclosure is applicable

[0091] The 5G system is an advanced technology derived from 4G LTE mobile communication technology, and supports new radio access technologies (RATs), extended long term evolution (eLTE) as an extension technology of LTE, non-3GPP access (for example, wireless local area network (WLAN) access), etc. through the evolution of the existing mobile communication network structure or net state structure.

[0092] The 5G system is defined based on services, and the interactions between network functions (NFs) in the architecture of the 5G system can be represented in the following two ways.

[0093] - Reference point representation: indicates the interaction between NF services in NFs described by a point-to-point reference point (e.g., N11) between two NFs (e.g., AMF and SMF).

[0094] -Service-based representation: A network function (e.g. AMF) within the control plane (CP) allows other authenticated network functions to access its services. If necessary, this representation also includes point-to-point reference points.

[0095] 5GC (5G Core) can include various components, Figure 6 Some of the various components are shown, including Access and Mobility Management Function (AMF), Session Management Function (SMF), Policy Control Function (PCF), Prose User Plane Function (UPF), Application Function (AF), Unified Data Management (UDM) and Non-3GPP Interworking Function (N3IWF).

[0096] The UE is connected to the data network via the UPF via the Next Generation Radio Access Network (NG-RAN) including the gNB. The UE can even be provided with data services through untrusted non-3GPP access, such as a wireless local area network (WLAN). In order to connect non-3GPP access to the core network, the N3IWF can be deployed.

[0097] N3IWF performs the function of managing the interworking between non-3GPP access and 5G system. When the UE is connected to non-3GPP access (e.g., WiFi called IEEE 801.11), the UE can connect to the 5G system through N3IWF. N3IWF performs control signaling with AMF and connects to UPF through N3 interface for data transmission.

[0098] AMF can manage access and mobility in the 5G system. AMF can perform the function of managing (non-access layer) NAS security. AMF can perform the function of handling mobility in idle state.

[0099] The UPF performs the function of sending and receiving user data by the gateway. The UPF node 440 can perform all or part of the user plane functions of the serving gateway (S-GW) and the packet data network gateway (P-GW) of 4G mobile communication.

[0100] UPF is a component that operates as a boundary point between the Next Generation Radio Access Network (NG-RAN) and the core network and maintains the data path between the gNB and the SMF. In addition, when the UE moves over the area served by the gNB, the UPF acts as a mobility anchor. The UPF can perform the function of processing PDUs. For mobility within the NG-RAN (which is defined after 3GPP Release 15), the UPF can route packets. In addition, the UPF can also use another 3GPP network (RAN defined before 3GPP Release-15, such as the Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access Network (UTRAN), Evolved (E)-UTRAN, or Global System for Mobile Communications (GERAN) / Enhanced Data Rates for Global Evolution (EDGE) RAN) as a mobility anchor. The UPF may correspond to the termination point of the data interface toward the data network.

[0101] PCF is a node that controls the operator's policies. AF is a server for providing various services to UE. UDM is a server that manages subscriber information, such as the Home Subscriber Server (HSS) of 4G mobile communications. UDM stores subscriber information in a unified data repository (UDR) and manages the subscriber information.

[0102] SMF can perform the function of allocating the Internet Protocol (IP) address of the UE. In addition, SMF can control the packet data unit (PDU) session.

[0103] For ease of explanation, in the following, the figure marks may be omitted for AMF, SMF, PCF, UPF, AF, UDM, N3IWF, gNB or UE, which may operate with reference to the contents described in the standard documents released earlier than this document.

[0104] Figure 3 : is a diagram showing an example of expressing a structure of a wireless communication system applied to the present disclosure from a node perspective.

[0105] refer to Figure 3 , the UE is connected to the data network (DN) through the next generation RAN. The control plane function (CPF) node performs all or part of the functions of the mobility management entity (MME) of 4G mobile communication and all or part of the serving gateway (S-GW) and PDN gateway (P-GW) functions. CPF nodes include AMF and SMF.

[0106] The UPF node performs the function of a gateway for sending and receiving user data.

[0107] The Authentication Server Function (AUSF) authenticates and manages the UE. The Network Slice Selection Function NSSF is a node of the following network slices.

[0108] The Network Open Function (NEF) provides a mechanism to securely open services and functions of the 5G core.

[0109] Figure 3 The reference points in are described as follows. N1 represents the reference point between UE and AMF. N2 represents the reference point between (R)AN and AMF. N3 represents the reference point between (R)AN and UPF. N4 represents the reference point between SMF and UPF. N5 represents the reference point between PCF and AF. N6 represents the reference point between UPF and DN. N7 represents the reference point between SMF and PCF. N8 represents the reference point between UDM and AMF. N9 represents the reference point between UPFs. N10 represents the reference point between UDM and SMF. N11 represents the reference point between AMF and SMF. N12 represents the reference point between AMF and AUSF. N13 represents the reference point between UDM and AUSF. N14 represents the reference point between AMFs. N15 represents the reference point between PCF and AMF in non-roaming scenarios and between AMF and PCF of the visited network in roaming scenarios. N16 represents the reference point between SMFs. N22 represents the reference point between AMF and NSSF. N30 represents the reference point between PCF and NEF. N33 may denote a reference point between AF and NEF, and the above-mentioned entities and interfaces may be configured with reference to contents described in a standard document released earlier than this document.

[0110] The radio interface protocol is based on the 3GPP radio access network standard. The radio interface protocol is horizontally divided into a physical layer, a data link layer, and a network layer, and vertically divided into a user plane for transmission of data information and a control plane (signaling) for transmission of control signals.

[0111] The protocol layer may be divided into L1 (layer 1), L2 (layer 2), and L3 (layer 3) based on three lower layers of an open system interconnection (OSI) reference model widely known in communication systems.

[0112] Hereinafter, the present disclosure describes each layer of the wireless protocol. Figure 4 is a view showing an example of the structure of a radio interface protocol between a UE and a gBN.

[0113] refer to Figure 4 The access stratum (AS) layer may include a physical (PHY) layer, a medium access control layer, a radio link control (RLC) layer, a packet data convergence protocol (PDCP) layer, and a radio resource control (RRC) layer, and operations based on each layer may be performed with reference to the contents described in a standard document released earlier than the current document.

[0114] AI / ML technology can be applied to the application layer and the network layer. As an AI / ML technology, deep learning technology can be applied to the field of wireless resource management and allocation, but is not limited to this. In addition, AI / ML technology can also be applied to the MAC layer and the physical layer. In particular, attempts are being made to combine deep learning with wireless transmission at the physical layer.

[0115] For example, an AI / ML algorithm based on deep learning in a machine learning manner may require a large amount of training data to optimize training parameters. However, there may be limitations in obtaining data as training data in a specific channel environment, and a method for applying AI / ML technology to a network while taking this into consideration will be described below.

[0116] Machine learning refers to a series of operations to train a machine to create a machine that can perform tasks that are difficult or impossible for humans to perform. Machine learning requires data and a learning model. In machine learning, data learning methods can be mainly divided into three types: supervised learning, unsupervised learning, and reinforcement learning. Neural network learning is to minimize output errors. Neural network learning is a process of repeatedly inputting training data into a neural network, calculating the error between the target and the output of the neural network for the training data, and backpropagating the error of the neural network from the output layer of the neural network to the input layer in the direction of reducing the error, thereby updating the weight of each node of the neural network. Supervised learning uses training data marked with the correct answer, and unsupervised learning may use training data that is not marked with the correct answer. That is, for example, in the case of supervised learning for data classification, the training data may be data in which each training data is marked with a category. The labeled training data is input to the neural network, and the error can be calculated by comparing the output (category) of the neural network with the label of the training data. The calculated error is backpropagated from the neural network in the opposite direction (i.e., from the output layer to the input layer), and the connection weights of each node in each layer of the neural network can be updated according to backpropagation. The amount of change in the connection weight of each node to be updated can be determined according to the learning rate. The calculation of the input data by the neural network and the back propagation of the error can constitute a learning cycle (epoch). The learning rate can be applied differently according to the number of repetitions of the learning cycle of the neural network. For example, in the early stages of learning of the neural network, a high learning rate can be used so that the neural network can quickly acquire a certain level of performance, thereby improving efficiency, and in the later stages of learning, a low learning rate can be used to increase accuracy. In addition, as an example, the learning method can vary according to the characteristics of the data. For example, if the purpose is to accurately predict the data sent from the transmitter in the communication system at the receiver, it is preferred to use supervised learning rather than unsupervised learning or reinforcement learning to perform learning. The learning model corresponds to the human brain, and the most basic linear model can be considered, but the machine learning paradigm using a highly complex neural network structure (such as an artificial neural network) as a learning model can be referred to as deep learning. The neural network core used as a learning method can mainly include a deep neural network (DNN), a convolutional deep neural network (CNN) and a recursive Boltzmann machine (RNN), and these learning models can be applied.

[0117] Hereinafter, a method for performing AI / ML services and application support and AI / ML traffic transmission by supporting 5GS considering the above-mentioned AI / ML technology will be described. For example, learning can be performed on a learning model based on AI / ML technology, and reasoning can be performed by a learning model on which learning has been performed. Here, the network can support the above-mentioned learning and reasoning operations. As a more specific example, the AI / ML operation of the application layer can support the operations of Table 1 below. With reference to Table 1 below, the AI / ML task division point of the AI / ML endpoint can be determined by considering the relationship between entities. As another example, AI / ML models and data can be distributed and shared through 5GS. For another example, 5GS can support federated learning (federated learning, FL), and the model parameters of FL can be applied, but are not limited to this.

[0118] [Table 1]

[0119]

[0120] In addition, the network can support network resource utilization monitoring operations to support AI / ML functions at the application layer. In addition, the network can perform support by determining whether to expose the acquired information to support AI / ML functions (e.g., location, QoS, load, congestion, etc.) and to which entity to expose the information. In addition, the network can acquire and manage information to support AI / ML functions and is not limited to a specific implementation.

[0121] As another example, the network may support federated learning (FL) and apply the model parameters of FL. That is, the model parameters of FL may adapt to the communication environment and may perform signal transmission. The network may perform at least one of the functions of protecting personal privacy, reducing the load of the base station through distributed processing, or reducing the traffic between the base station and the terminal through FL, but may not be limited thereto. For the above operations, the network may set the FL model parameters and send the set FL model parameters to multiple terminals. Then, each of the multiple terminals may update the acquired model parameters based on the acquired training data and send the updated model parameter information to the network. The network may eventually update the model parameters based on the model parameter information acquired from each of the multiple terminals. However, in the above process, since the transmission and reception of local model parameters are required between the network and the terminal, the traffic of the local model parameters may cause a large burden in the wireless communication environment, and a method for solving this problem will be described.

[0122] Figure 5is a diagram showing an AI / ML-based endpoint split operation applicable to the present disclosure. AI / ML can be used in a range of application areas and can replace existing algorithms (e.g., speech recognition, image recognition, video processing, etc.) of devices (e.g., smart phones, vehicles, robots, etc.) within a communication system. For example, the communication system may support at least one of the AI / ML operation splitting between the AI / ML endpoints of Table 1 above, the distribution / sharing of AI / ML models / data via the communication system, or the FL support via the communication system.

[0123] Here, refer to Figure 5 , the AI / ML operation splitting for AI / ML reasoning can be considered. Considering the current task and environment, the AI / ML operation / model can be split into multiple parts. Specifically, the endpoint (e.g., terminal) can transfer computationally intensive operations and energy-intensive operations to the network part, and directly handle privacy-sensitive operations and delay-sensitive operations. More specifically, part of the AI / ML operation / model can be located in the terminal device 510 and the network AI / ML endpoint 520, respectively. That is, the entire AI / ML model is not configured in one entity, but can be split and located in multiple entities. Here, when reasoning is performed based on the AI / ML model, input can be provided as input to the AI / ML operation / model part in the terminal device 510. Thereafter, intermediate data can be derived from the corresponding AI / ML operation / model part by the terminal device 510 as output. The intermediate data can be sent to the network AI / ML endpoint (520) and provided as input to the AI / ML operation / model part within the network. The AI / ML operation / model part within the network can derive the final reasoning output data based on the provided input. That is, AI / ML operations / models can be split into each entity and appropriate operations are performed considering each entity’s tasks and environment.

[0124] Figure 6 is a diagram illustrating an AI / ML model distribution method applicable to the present disclosure. For example, considering the task and environment, a device with various functions (e.g., a smart phone, a vehicle, a robot, etc.) needs to use multiple AI / ML models. In other words, considering the task and environment, the device needs to switch to a suitable AI / ML model and apply it. Here, since the storage space of the device may be limited, it may not be desirable to configure all candidate AI / ML models in the device. In view of the above, the candidate AI / ML models of the multiple AI / ML models may be configured as an AI / ML model pool. For example, an AI / ML model pool may be provided in a network and may be distributed to the devices from network endpoints. To this end, the device 610 may continuously monitor the performance of the AI / ML model. As a more specific example, refer to Figure 6(a), the first device 610 may receive an AI / ML model update request from the second device 620, and acquire a suitable AI / ML model through a network endpoint based on the AI / ML model update request. Here, the second device 620 may continuously monitor the AI / ML model, and if the AI / ML model is not suitable in view of the new task and environment, the second device 620 may request an AI / ML model update from the first device 610. The first device 610 may acquire a suitable AI / ML model from the network and send information about the AI / ML model to the second device 620.

[0125] As another example, refer to Figure 6 (b), the DNN (deep neural network) of FL may exist in the cloud. Here, the model information of the DNN may be sent to the first device 610. The first device 610 may perform some training on the DNN using a subset of the training data based on the model information, and send the new gradient information to the cloud. The above cycle may be repeated, so that the DNN of FL in the cloud may be updated. In addition, the model information of the DNN may also be sent to the second device 620. The second device 620 may perform some training on the DNN using a subset of the training data based on the model information, and send the new gradient information to the cloud. The above cycle may be repeated, so that the DNN of FL in the cloud may be updated. That is, the cloud may obtain updated learning model information based on the training data owned by each of the multiple devices 610 and 620, and integrate the information to update the overall model information. For example, the device may apply the training data to a local model constructed based on the sent model information and send the gradient to the cloud, and the cloud may update the global model based on the gradient received from the device.

[0126] Figure 7 is a diagram showing a method for performing learning based on federated learning applicable to the present disclosure. Figure 7 , the FL training server 710 may determine at least one terminal participating in FL learning (or operation). For example, the FL training server 710 may be configured in a network, but may not be limited thereto. Figure 7 , the learning of the FL training server 710 may be iterated at regular intervals. For example, in the Nth learning iteration at a specific interval, the FL training server 710 may receive a training resource report from at least one terminal 721, 722, 723, 724, and 725. The training resource report may include information related to FL training in each of the at least one terminal 721, 722, 723, 724, and 725, and may not be limited to a specific form. The FL training server 710 may select a terminal to perform learning in a given cycle based on the training resource report information obtained from each of the at least one terminal 721, 722, 723, 724, and 725. For example, in Figure 7 Terminal A 721, terminal C 723, and terminal D 724 are selected in the Nth learning iteration, but this is merely an example for ease of explanation and may not be limited thereto. For example, the FL training server 710 may send a global model and learning configuration information to the terminals 721, 723, and 724 selected in the model distribution and training configuration. The global model is a learning model constructed based on FL, and the selected terminals 721, 723, and 724 may construct a local model based on the global model information. Thereafter, each of the selected terminals 721, 723, and 724 may perform learning by applying the acquired training data to the constructed local model, and report the training result to the FL training server 710. For example, the training result may be the gradient of the DNN, but may not be limited thereto. The FL training server 710 may receive a training result report from each of the selected terminals 721, 723, and 724 in the training result reporting step, and update the global model. For example, the terminal participating in the FL operation (or learning) provides a training result report to the FL training server 710 at a preset time for each learning iteration, and the next learning iteration can be initiated based on this. The above operation can be iterated for each learning iteration. Here, for example, if the learning environment or conditions (e.g., computing resources of the device, wireless channel conditions, other services of the learning device, etc.) do not change, the learning terminal reselection and training reconfiguration may not be required for each learning iteration in which learning is performed. That is, the learning terminals of the same group can participate in learning with the same configuration through multiple iterations. However, for example, independent and identical sampling implementations from all terminals may be required, and considering the above points, the learning terminals may be configured differently according to the iteration, but may not be limited thereto.

[0127] Here, as an example, a method of applying NAS (Non Access Stratum) level congestion control to a terminal participating in FL operation (or learning) (or a terminal selected for FL operation (or learning)) is described below.

[0128] As a specific example, a terminal participating in FL operation (or learning) needs to provide training results to the server for each learning iteration, but may not be able to provide training results to the server at a preset timing based on network congestion. In the above case, the FL training server may not be able to terminate the learning iteration, resulting in a delay or exceeding the preset time. Here, if the FL training server does not obtain the training results from the terminal, the FL training server may perform global model updates or FL model aggregation without results, which may adversely affect the FL model convergence. As an example, the terminal may obtain a backoff timer value based on NAS-level congestion control to postpone signal reception, and the above delay may occur based on this. Here, the backoff timer value may be several (tens) minutes or several (tens) hours, but may not be limited thereto. That is, since the network conditions of the communication system (e.g., network congestion conditions, etc.) may affect the operation of the AI / ML service / application, a method for effectively supporting the AI / ML service by considering the characteristics of the AI / ML service may be required, which will be described below.

[0129] For example, based on NAS level congestion control, the network may provide a back-off timer value to the terminal, and the terminal may postpone signal transmission based on the back-off timer value. The network may determine the back-off timer differently for each terminal, so that each terminal may transmit a signal at a different timing after the back-off timer. That is, a terminal that has received a back-off timer may not transmit a NAS signal unless the back-off timer expires, a request is received from the network, or a service with a high priority should be performed.

[0130] For example, in the case of network overload, AMF may reject NAS messages from terminals using 5G to access the network. If the NAS request is rejected, AMF may send a mobility management back-off timer to the terminal. Except for the deregistration process or the process not affected by congestion control (e.g., high priority access, emergency services) and mobile terminal services, the terminal may not initiate any other NAS requests until the back-off time expires.

[0131] Here, as an example, if the terminal performs a deregistration procedure while the backoff timer is running, the backoff timer may continue to run even after the deregistration procedure. In addition, as an example, if the terminal is already in the CM-CONNECTED state while the backoff timer is running, the terminal may perform a mobility registration update. In addition, if the terminal receives a paging request or a NAS notification message from the AMF while the backoff timer is running, the terminal may stop the backoff timer and start a service request procedure or a mobility registration update procedure through the corresponding access.

[0132] In addition, as an example, if the terminal is in the CM-IDLE state and has not left the registration area, the terminal may send a service request message indicating that the message is excluded from NAS congestion control. As another example, if the terminal is in the CM-IDLE state and has left the registration area, the terminal may send a mobility registration update request message including a request indicating that the message is excluded from NAS congestion control.

[0133] As another example, if the terminal is in the CM-CONNECTED state, the terminal may send a PDU session modification request with a PS data off state change via a UL NAS transport message while the backoff timer continues. Here, the message may be sent with an indication that it is excluded from NAS congestion control.

[0134] In addition, when NAS MM (mobility management) congestion control is activated in the AMF, if the terminal sends a NAS MM message to the AMF indicating that the NAS MM message is excluded from NAS congestion control, the AMF may not reject the NAS MM message, but may forward the NAS SM (session management) message to the corresponding SMF. Here, the forwarded message may be forwarded together with an indication that it is excluded from NAS congestion control. The SMF may confirm that the NAS SM message is not subject to congestion control, and if it is subject to congestion control, it may reject the message. In addition, as an example, even if a cell / RAT / access type and PLMN change is performed, the mobility management back-off timer may not be stopped. That is, the cell / RAT / TA / access type change may not stop the mobility management back-off timer.

[0135] As another example, if a 5GSM core network function change or an always-on PDU session request indication needs to be reported while the NAS MM congestion control timer of the terminal is running and MM signaling cannot be initiated, the terminal can postpone initiating the relevant MM signaling until the MM congestion control timer expires. In addition, if there is a reserved communication mode in the terminal, the AMF can select the fallback timer value considering the communication mode of the terminal so that the terminal does not miss the reserved mode.

[0136] In addition, if the terminal is in the CM-CONNECTED state, the AMF may not reject the registration request message for mobility registration update. On the other hand, the AMF may reject the service request message sent by the terminal already in the CM-CONNECTED state and the ULNAS transmission message containing the mobility management backoff timer.

[0137] As another example, if the terminal receives a DLNAS transport message from the AMF while the backoff timer is still running, the backoff timer may be stopped. In addition, considering CM-IDLE mobility, the AMF may reject the registration request for mobility registration update by sending a registration reject message including the mobility management backoff timer to the terminal.

[0138] For example, as described above, the back-off timer value is provided to the terminal, and the terminal may perform NAS-level congestion control by deferring reception of a NAS message based on the back-off timer value.

[0139] As another example, DNN (Data Network Name)-based congestion control may be performed. For example, the SMF may reject a PDU session establishment request message or a PDU session modification request message, except for a PS data off state change for reporting a specific DNN for which a backoff timer is running. Thus, DNN-based congestion control may be applied to the terminal. In addition, as an example, the SMF may send a PDU session release command message with a DNN backoff timer to the terminal to release a PDU session belonging to a congested DNN. For example, if the DNN backoff timer is included in the PDU session release command message, the reactivation request may not be set as the cause value.

[0140] As another example, when the network data analysis function (NWDAF) is configured, the SMF may determine the back-off timer to be provided to the terminal based on the session management congestion control information provided from the NWDAF. For example, when the DNN back-off timer is activated, the terminal may not initiate a session management process for a congested DNN. On the other hand, the terminal may perform a session management process for another DNN. For example, when there is no DNN associated with the back-off timer, the terminal may initiate a session management request for all PDU session types for a specific DNN. In addition, as an example, in the event of a cell / TA / PLMN / RAT change, an unreliable non-3GPP access network change, or an access type change, the terminal may not stop the back-off timer. That is, the DNN back-off timer may work regardless of the above process.

[0141] As another example, the terminal can initiate a session management process for high priority access and emergency services even when the DNN backoff timer is activated. In addition, the terminal can initiate a session management process to report the data closing state change to the network. In addition, as an example, if the terminal receives a network-initiated session management message for a congested DNN associated with the backoff timer instead of a PDU session release command, the terminal can stop the backoff timer and respond to the network. In addition, if the terminal receives a PDU session release command message for a congested DNN, the terminal can stop the backoff timer unless it receives a new backoff timer from the SMF. The terminal can initiate a PDU session release process and may not stop the backoff timer when the associated PDU session is released.

[0142] In addition, as an example, congestion control based on S-NSSAI (single network slice selection assistance information) may be performed. Congestion control based on S-NSSAI may be applied to handle NAS signal congestion of a terminal with a backoff timer associated with S-NSSAI, regardless of the presence of DNN. For example, if it is determined that the S-NSSAI is in a congested state, the SMF may apply congestion control based on S-NSSAI to the terminal for an SM request that does not include a PS data off state change report, and may provide a backoff timer and a home public land mobile network (HPLMN) congestion indication. If the terminal receives an S-NSSAI-based backoff timer without an HPLMN congestion indication, the terminal may apply the S-NSSAI backoff timer only to the PLMN from which the backoff time is received. On the other hand, if the terminal receives an S-NSSAI-based backoff timer together with an HPLMN congestion indication, the terminal may apply the S-NSSAI-based backoff timer to both the PLMN from which the backoff time is received and other PLMNs. As another example, the SMF may release a PDU session belonging to a congested S-NSSAI. Here, the SMF may send a PDU session release request message to the terminal, the PDU session release request message including a backoff timer associated with the S-NSSAI or a combination of the S-NSSAI and a specific DNN. For example, if NWDAF is deployed, the SMF may utilize the session management congestion message provided by the NWDAF to determine the backoff timer to be provided to the terminal.

[0143] As described above, NAS-level congestion control can be considered, and DNN-based congestion control and S-NSSAI congestion control can be considered. Here, as an example, the above-mentioned NAS-level congestion control can be applied to a terminal participating in FL operation (or learning) (or a terminal selected for FL operation (or learning)). As an example, a terminal participating in FL operation (learning) can switch to the CM-IDLE state while performing learning. The terminal needs to switch to the CM-CONNECTED state to complete the learning and send the training results to the server. Here, the terminal sends a service request message to the AMF, but the AMF can apply NAS mobility management congestion control. Therefore, the AMF can send a service rejection message including mobility management backoff time information to the terminal, and the terminal can switch to the CM-CONNECTED state after the backoff timer expires.

[0144] As another example, while the terminal participating in FL operation (learning) performs learning, the user plane of the PDU session used to send the training results to the FL training server may be deactivated. When the learning is completed, the terminal needs to activate the user plane of the PDU session to send the training results to the server. Therefore, the terminal can send a service request message to the AMF. However, the AMF may apply NAS mobility management congestion control and send a service rejection message including mobility management backoff timer information to the terminal. Here, the PDU session may be a PDU session for AI / ML services / operations / businesses / applications or a PDU session for FL operation (learning), and may not be limited to a specific form.

[0145] As another example, a terminal participating in FL operation (learning) may need to change the PDU session to send the training results to the server. For example, PDU session modification may be required to modify the QoS (quality of service) of the PDU session or add a QoS flow. Here, the terminal sends a PDU session modification request message to the SMF, but the SMF may send a PDU session modification rejection message including backoff timer information by applying DNN-based congestion control or S-NSSAI-based congestion control. Here, the PDU session can be a PDU session for AI / ML services / operations / businesses / applications or a PDU session for FL operation (learning), and may not be limited to a specific form.

[0146] As described above, if the terminal postpones the sending of the training result based on the backoff timer based on NAS congestion control, the FL training server may not be able to obtain the training result from the corresponding learning iteration. Therefore, the FL training server may terminate the corresponding learning iteration and perform a global model update without receiving the training result. In other words, the FL training server will skip the corresponding training result and perform a global model update, which may adversely affect the model update convergence. As another example, the FL training server may wait to receive the training result while the terminal postpones the sending and terminate the learning iteration after receiving the training result, but this may cause a delay. As an example, the backoff timer value may be several (tens) minutes or several (tens) hours, but may not be limited thereto.

[0147] That is, in the above situation, since network congestion may affect the operation of AI / ML services / applications, a method for effectively supporting AI / ML services by considering the characteristics of AI / ML services may be required, which will be described below.

[0148] For example, considering that the congestion control related event support method of AI / ML service may be composed of a combination of one or more of the following operations / configurations / steps, and is not limited to a specific form.

[0149] For example, AI / ML services may be used interchangeably with the following AI / ML-based services, AI / ML businesses, AI / ML applications, AI / ML transmissions, AI / ML operations, and AI / ML operation businesses, and are not limited to a specific form.

[0150] In addition, as an example, it can be used interchangeably with the following FL (Federated Learning) operation, FL training, FL training operation, and distributed learning operation, and is not limited to a specific form.

[0151] In addition, UE (user equipment) and terminal may be used interchangeably below, but are not limited to this. In addition, as an example, AF (application function), AS (application server) and server may be used interchangeably and are not limited to a specific form. As an example, AF and AS may be AI / MLAF and AI / MLAS, but may not be limited to this. As an example, although a congestion control-related event support method based on AI / ML services is described below, it is not limited to AI / ML services and may be applicable to other services and is not limited to a specific form.

[0152] As an example, the content described below may refer to but is not limited to the operation of a 5G system (e.g., TS23.501, TS23.502, TS23.503).

[0153] AF provides AI / ML operation (or learning) related information to the core network

[0154] The AF may provide the core network with information related to the AI / ML operation (or learning). For example, the AI / ML operation (or learning) may be the FL operation (or learning), but may not be limited thereto.

[0155] Here, the AI / ML operation (or learning) related information may include at least one of the following Table 2. For example, the information in the following Table 2 may be included (or provided) based on any one of a combination form, an explicit form, an implicit form, a compressed form, and other forms.

[0156] As a specific example, the AI / ML operation (or learning) related information may include information for identifying a terminal participating in the AI / ML operation (or learning). Here, when multiple terminals participate in the AI / ML operation (or learning), the information about the terminal participating in the AI / ML operation (or learning) may be information in the form of a terminal list. The information used to identify the terminal may be at least one of a general public subscription identifier (GPSI), a subscription permanent identifier (SUPI), an IP address information, an external group ID, or other forms of identification information, but may not be limited thereto. In addition, as an example, the AI / ML operation (or learning) related information may include at least one of the start time, required time, or end time information of the AI / ML operation (or learning). As another example, the AI / ML operation (or learning) related information may include at least one of the expected start time, expected required time, or expected end time information of the AI / ML operation (or learning), but may not be limited thereto. In addition, as an example, the AI / ML operation (or learning) related information may include the importance information of the AI / ML operation (or learning). As another example, the AI / ML operation (or learning) related information may include the urgency information of the AI / ML operation (or learning). As an example, the importance information and the urgency information may be service requirements and may be provided as level information such as “high / medium / low.” As another example, the importance information and the urgency information may be instruction information such as “important / urgent” and are not limited to a specific form.

[0157] For example, the AF may provide AI / ML operation (or learning) related information to the core network at the same time. As another example, the AF may provide AI / ML operation (or learning) related information at different times. As a specific example, the AF may first provide information 1), 2) and 5) of the following Table 2, and then provide information 4) when the AI / ML operation (or learning) is actually completed. However, this is only an example for ease of explanation and may not be limited to this.

[0158] [Table 2]

[0159]

[0160] In addition, as an example, the AF can provide AI / ML operation (or learning) related information to the core network by itself. As another example, the AF can provide AI / ML operation (or learning) information to the core network based on a core network request.

[0161] Specifically, the core network may obtain AI / ML operation (or learning) related information from the AF as input. As another example, the core network may obtain AI / ML operation (or learning) related information from the AF by request. As another example, the core network may obtain AI / ML operation (or learning) related information from the AF based on an event method.

[0162] As another example, the core network may obtain AI / ML operation (or learning) related information through a new process different from the existing process, and may not be limited to a specific form.

[0163] AI / ML operation (or learning) related information can be used in the network function (NF) within the core network. In the following, the AMF and SMF that perform NAS-level congestion control will be described, but it may not be limited to this. For example, based on the information in Table 2 for identifying the terminal participating in the AI / ML operation (or learning), the AI / ML operation (or learning) related information can be provided to at least one of the AMF or SMF serving the terminal. Here, at least one of the AMF or SMF can identify whether there are terminals participating in the same AI / ML operation (or learning) in addition to the terminals it serves based on the information used to identify the terminals participating in the AI / ML operation (or learning). In addition, if there are terminals participating in the same AI / ML operation (or learning) in addition to the terminals it serves, at least one of the AMF or SMF can identify which terminals it does not serve.

[0164] Apply congestion control of AMF based on AI / ML operation (or learning) related information provided by AF

[0165] AMF may receive (or obtain) AI / ML operation (or learning) related information provided by AF. AMF may then determine the need for congestion control based on the network congestion status and initiate congestion control. Here, AMF may receive a NAS message sent by the terminal. AMF may then check whether the terminal is a terminal participating in AI / ML operation (or learning). For example, if the terminal is a terminal that does not participate in AI / ML operation (or learning), AMF may apply the existing congestion control operation to the terminal. That is, AMF may determine whether to apply the congestion control operation to the terminal by considering the network load.

[0166] On the other hand, if the terminal is a terminal participating in AI / ML operation (or learning), the AMF may check whether there are multiple terminals participating in AI / ML operation (or learning). If there are multiple terminals participating in AI / ML operation (or learning), the AMF may check whether it serves all terminals. If there is only one terminal participating in AI / ML operation (or learning), the AMF may determine whether to apply congestion control based on AI / ML operation (or learning) related information. For example, the AMF may determine to apply congestion control in the same manner as before, without considering the AI / ML operation (or learning) related information, and is not limited to a specific form.

[0167] On the other hand, if there are multiple terminals participating in the AI / ML operation (or learning) and the AMF serves all terminals, the AMF may determine whether to apply congestion control based on the AI / ML operation (or learning) related information. For example, the AMF may determine whether to apply congestion control by reflecting the status of other terminals other than the terminal to which the above-mentioned NAS message has been sent. The status of other terminals can be used to check whether there is a possibility of sending NAS messages from other terminals to the AMF. For example, the terminal state may be in a user plane deactivated state or CM-IDLE for a PDU session for AI / ML operation (or learning). If the terminal state is a user plane deactivated state or CM-IDLE for a PDU session for AI / ML operation (or learning), a NAS message may be sent.

[0168] For a more specific example, if the AMF determines based on the status information of other terminals that other terminals other than the terminal that has sent the NAS message may send a NAS message, the AMF may determine to apply congestion control to all terminals. The AMF provides the back-off timer value to the terminal, and the time value may be set to be the same. For example, since the AMF may receive NAS messages from each terminal at different times, the AMF may set the back-off timer value so that based on the back-off timer value, the time when the back-off timer expires is the same at all terminals, and the back-off timer value may be provided to the terminal, but may not be limited thereto.

[0169] As another example, the AMF may determine that some or all terminals other than the terminals that have sent NAS messages are unlikely to send NAS messages. For example, if the user plane of the PDU session for AI / ML operation (or learning) is active, it may be determined that there is no possibility of sending NAS messages. In this case, the AMF may not apply congestion control to the terminals that have sent NAS messages.

[0170] As another example, consider the case where multiple terminals participate in AI / ML operation (or learning) but are not served by the same AMF. That is, the AMF may serve only some of the terminals participating in the AI / ML operation (or learning). The AMF may determine whether to apply congestion control based on information related to the AI / ML operation (or learning). As a specific example, the AMF may determine not to apply congestion control to a terminal that has sent a NAS message. As another example, if the AI / ML operation (or learning) is considered to be unimportant or not urgent based on information related to the AI / ML operation (or learning), the AMF may determine to apply congestion control to a terminal that has sent a NAS message.

[0171] Congestion control using SMF based on AI / ML operational (or learning) related information provided by AF

[0172] The SMF may receive (or obtain) information related to the AI / ML operation (or learning) provided by the AF. The SMF may then determine that congestion control is required based on the network congestion status and initiate congestion control. Here, the SMF may receive a NAS message sent by the terminal. The SMF may then check whether the terminal is a terminal participating in the AI / ML operation (or learning). For example, the SMF may check whether the received NAS message is associated with a PDU session for the AI / ML operation (or learning) and whether it corresponds to performing congestion control to check whether the terminal is participating in the AI / ML operation (or learning).

[0173] If the terminal is a terminal that does not participate in AI / ML operations (or learning), the SMF can apply the existing congestion control operation to the terminal. On the other hand, if the terminal is a terminal that participates in AI / ML operations (or learning), the SMF can check whether there are multiple terminals in the AI / ML operations (or learning) in which the terminal participates. Here, if there are multiple terminals participating in AI / ML operations (or learning), the SMF can check whether it serves all terminals. For example, the SMF can check whether the SMF serves all terminals for PDU sessions for AI / ML operations (or learning). In other words, the SMF serving the terminal can serve the PDU session, but is not limited to this.

[0174] If there is only one terminal participating in the AI / ML operation (or learning), the SMF may determine whether to apply congestion control based on the AI / ML operation (or learning) related information. Here, the SMF may determine whether to apply congestion control in the same manner as before, without considering the AI / ML operation (or learning) related information, but is not limited thereto.

[0175] On the other hand, if there are multiple terminals participating in the AI / ML operation (or learning) and the SMF serves all the terminals, the SMF may determine whether to apply congestion control based on the AI / ML operation (or learning) related information. Here, the SMF may determine whether to apply congestion control by reflecting the status of other terminals other than the terminal to which the above-mentioned NAS message has been sent. For example, the status of other terminals may be a QoS flow included in a PDU session for AI / ML operation (or learning), but may not be limited thereto.

[0176] For a more specific example, if the SMF determines that other terminals other than the terminal that has sent the NAS message have the possibility of sending a NAS message based on the status information of the other terminals, the SMF may determine to apply congestion control to all terminals. Here, the SMF provides the back-off timer value to the terminal, and the time value may be set to the same value. For example, since the SMF may receive NAS messages from each terminal at different times, the SMF may set the back-off timer value so that based on the back-off timer value, the time when the back-off timer expires is the same at all terminals, and the back-off timer value may be provided to the terminal, but is not limited thereto.

[0177] As another example, the SMF may determine that some or all terminals other than the terminals that have already sent NAS messages are unlikely to send NAS messages. For example, if the PDU session for AI / ML operation (or learning) already includes an appropriate QoS flow, it may be determined that there is no possibility of sending NAS messages. In this case, the SMF may not apply congestion control to the terminals that have already sent NAS messages.

[0178] As another example, consider a situation where multiple terminals participate in AI / ML operation (or learning) but are not served by the same SMF. That is, the SMF may serve only some of the terminals participating in the AI / ML operation (or learning). Here, as an example, the SMF may determine whether to apply congestion control based on information related to the AI / ML operation (or learning). As an example, the SMF may determine not to apply congestion control to a terminal that has sent a NAS message.

[0179] As another example, if the AI / ML operation (or learning) is considered unimportant or unurgent based on the AI / ML operation (or learning) related information, the SMF may determine to apply congestion control to the terminal that has sent the NAS message.

[0180] For example, the SMF applying congestion control to a terminal may mean that it applies congestion control to the terminal's PDU session.

[0181] Based on the above, AI / ML services can be effectively supported by determining whether to apply congestion control to a terminal participating in AI / ML operation (or learning) based on AI / ML operation (or learning) related information provided from the AF.

[0182] Figure 8 is a diagram illustrating a method of performing congestion control in a network function applicable to the present disclosure.

[0183] Reference Figure 8 , the network function may obtain AI / ML related information from the AF. (S810) For example, the network function may be the above-mentioned AMF or SMF, but is not limited thereto. In addition, the network function may obtain AI / ML related information directly from the AF or through another network function, and is not limited to a specific form.

[0184] The network function may confirm that congestion control needs to be performed on the terminal. (S820) For example, the network function may consider whether the terminal participates in the AI / ML operation. (S830) If the terminal does not participate in the AI / ML operation, the network function may perform congestion control by considering the network load in the same manner as the existing congestion control. (S840) On the other hand, if the terminal participates in the AI / ML operation, the network function may check whether the terminal participating in the AI / ML operation is the only terminal among the above-mentioned terminals. (S850) Here, if only the corresponding terminal participates in the AI / ML operation, the network function may perform congestion control by considering the network load in the same manner as the existing congestion control. (S840) On the other hand, if multiple terminals participate in the AI / ML operation, the network function may check whether it serves all terminals participating in the AI / ML operation based on the AI / ML related information, and perform congestion control accordingly. (S860)

[0185] When there are multiple terminals participating in AI / ML operation and the network function serves all multiple terminals, the network function can determine whether to perform congestion control on the terminal by reflecting AI / ML related information and status information of other terminals. Here, as an example, the network function can be AMF. When other terminals are in user plane deactivation or CM-IDLE of AI / ML operation related PDUs based on the status information of other terminals, AMF can determine to perform congestion control on multiple terminals based on the possibility of sending NAS messages.

[0186] In addition, when congestion control is performed on multiple terminals, the AMF may set a backoff timer value for each of the multiple terminals so that the transmission delay termination time points of the multiple terminals are the same, and send the backoff timer value to each of the multiple terminals. On the other hand, when the other terminal is in one of the user plane activation and CM-CONNECTED of the AI / ML operation-related PDU based on the status information of the other terminal, the AMF may determine not to perform congestion control on the terminal based on the possibility of sending a NAS message.

[0187] As another example, the network function may be an SMF. The SMF may determine whether the terminal participates in the AI / ML operation based on whether the NAS message sent by the terminal is associated with the PDU session of the AI / ML operation. In addition, the SMF may determine the possibility of sending a NAS message based on the QoS flow included in the PDU session used by other terminals for AI / ML operations as the status information of other terminals. The SMF may determine to perform congestion control on multiple terminals based on the possibility of sending NAS messages.

[0188] Here, when congestion control is performed on multiple terminals, the SMF can set a backoff timer value for each of the multiple terminals so that the transmission delay termination time points of the multiple terminals are the same, and send the backoff timer value to each of the multiple terminals.

[0189] Fig. 9 is a diagram showing a method for performing congestion control in a network function applicable to the present disclosure. Fig. 9 , the network function may determine whether to perform congestion control on a first terminal among a plurality of terminals participating in the AI / ML operation. (S910) For example, in the case Figure 8 In the case where congestion control is performed on the terminal as shown, if there are multiple terminals in the AI / ML operation in which the terminal participates, the network function can determine whether to perform congestion control. Here, the network function can check whether it serves all multiple terminals. (S920) Here, if the network function serves all multiple terminals, the network function can determine whether to perform congestion control on all terminals participating in the AI / ML operation based on the AI / ML related information and the status information of other terminals. (S930) On the other hand, if the network function only serves some of the multiple terminals, the network function can determine whether to perform congestion control on the first terminal based on the AI / ML related information and the AI / ML operation importance information. (S940) As another example, if the network function only serves some of the multiple terminals, the network function may not apply congestion control to the first terminal, as described above.

[0190] For example, the AI / ML related information may include at least one of identification information of a terminal participating in the AI / ML operation, AI / ML operation start time information, AI / ML operation required time information, AI / ML operation end time information, or AI / ML operation importance information, as described in Table 2 above. In addition, at least one information included in the AI / ML related information may be simultaneously sent to the core network. As another example, as described above, some of the at least one information included in the AI / ML related information may be sent to the core network at a first time point, and other information may be sent to the core network at a second time point.

[0191] Fig.10 is a diagram illustrating a method of performing congestion control in a network function applicable to the present disclosure.

[0192] Reference Fig.10 , the network function may obtain AI / ML related information from the AF. (S1010) The network function may determine whether to perform congestion control on the terminal based on the obtained AI / ML related information. (S1020) The network function may receive a NAS message from the terminal (S1030), and perform congestion control based on the received NAS message. (S1040) For example, when performing congestion control on the terminal, the network function may send a rejection message including a backoff timer value to the terminal.

[0193] Here, as an example, if the terminal is a terminal that does not participate in the AI / ML operation, the network function may determine whether to perform congestion control on the terminal by considering the network load. That is, if the terminal does not participate in the AI / ML operation, the network function may perform congestion control in the same manner as before. On the other hand, if the terminal is a terminal that participates in the AI / ML operation, the network function may determine whether to perform congestion control on the terminal by considering whether there are multiple terminals that participate in the AI / ML operation and whether the network function serves multiple terminals.

[0194] Here, if only one terminal participates in the AI / ML operation, the network function can determine whether to perform congestion control on the terminal regardless of the AI / ML operation related information. In other words, the network function can perform congestion control in the same way as before. On the other hand, if there are multiple terminals participating in the AI / ML operation and the network function serves all multiple terminals, the network function can determine whether to perform congestion control on the terminal by reflecting the AI / ML related information and the status information of other terminals.

[0195] Here, as an example, the network function may be AMF. Based on the status information of other terminals, if the other terminals are in user plane deactivation or CM-IDLE of AI / ML operation related PDUs, the AMF may determine to perform congestion control on multiple terminals based on the possibility of sending NAS messages.

[0196] In addition, when congestion control is performed on multiple terminals, the AMF may set a backoff timer value for each of the multiple terminals so that the transmission delay termination time points of the multiple terminals are the same, and send the backoff timer value to each of the multiple terminals. On the other hand, when the other terminal is in one of the user plane activation and CM-CONNECTED of the AI / ML operation related PDU based on the status information of the other terminal, the AMF may determine not to perform congestion control on the terminal based on the possibility of sending a NAS message.

[0197] As another example, the network function may be an SMF. The SMF may determine whether the terminal participates in the AI / ML operation based on whether the NAS message sent by the terminal is associated with the PDU session of the AI / ML operation. In addition, the SMF may determine the possibility of sending a NAS message based on the QoS flow included in the PDU session used by other terminals for AI / ML operations as the status information of other terminals. The SMF may determine to perform congestion control on multiple terminals based on the possibility of sending NAS messages.

[0198] Here, when congestion control is performed on multiple terminals, the SMF can set a backoff timer value for each of the multiple terminals so that the transmission delay end time points of the multiple terminals are the same and send the backoff timer value to each of the multiple terminals.

[0199] As another example, if multiple terminals participate in AI / ML operations and the network function only serves some of the multiple terminals, the network function may determine not to perform congestion control on the terminal based on the AI / ML related information. As another example, if multiple terminals participate in AI / ML operations and the network function only serves some of the multiple terminals, the network function may determine whether to perform congestion control on the terminal based on the AI / ML related information and the AI / ML operation importance information.

[0200] For example, the AI / ML related information may include at least one of identification information of a terminal participating in the AI / ML operation, AI / ML operation start time information, AI / ML operation required time information, AI / ML operation end time information, or AI / ML operation importance information, as described in Table 2 above. In addition, at least one information included in the AI / ML related information may be simultaneously sent to the core network. As another example, some of the at least one information included in the AI / ML related information may be sent to the core network at a first time point, and the other information may be sent to the core network at a second time point, as described above.

[0201] Since the examples of the above-mentioned proposed methods may also be included in one of the implementation methods of the present disclosure, it is obvious that they may be considered as a proposed method. In addition, the above-mentioned proposed methods may be implemented individually or in combination (or merge) of some of them. A rule may be defined so that information on whether to apply the proposed method (or information on the rule of the proposed method) is notified from the base station to the terminal through a predefined signal (e.g., a physical layer signal or an upper layer signal).

[0202] Without departing from the technical ideas and basic features described in the present disclosure, the present disclosure may be embodied in other specific forms. Therefore, the above detailed description should not be interpreted as being restrictive in all aspects and should be considered as illustrative. The scope of the present disclosure should be determined by a reasonable interpretation of the attached claims, and all changes within the equivalent scope of the present disclosure are included in the scope of the present disclosure. In addition, claims that do not have a clear reference relationship in the claims may be combined to form an embodiment or included as a new claim by modification after submission.

[0203] Industrial Applicability

[0204] Embodiments of the present disclosure are applicable to various radio access systems. Examples of various radio access systems include the 3rd Generation Partnership Project (3GPP) or 3GPP2 system.

[0205] The embodiments of the present disclosure are applicable not only to various radio access systems but also to all technical fields to which various radio access systems are applied. In addition, the proposed method is applicable to millimeter wave and terahertz wave communication systems using ultra-high frequency bands.

[0206] Additionally, embodiments of the present disclosure are applicable to a variety of applications, such as autonomous vehicles, drones, and the like.

Claims

1. A method for operating a network function in a wireless communication system, the method comprising: Obtain AI / ML related information from the application function AF; Determining whether to perform congestion control on the first terminal based on the obtained AI / ML related information; receiving a non-access stratum NAS message from the first terminal; as well as performing congestion control based on the NAS message, Whether to perform congestion control on the first terminal is determined based on whether the first terminal participates in AI / ML operation.

2. The method according to claim 1, in, Based on the first terminal being a terminal not participating in AI / ML operation, the network function determines whether to perform congestion control on the first terminal by considering a network load, and Wherein, based on the fact that the first terminal is a terminal participating in the AI / ML operation, the network function determines whether to perform congestion control on the first terminal by considering whether there are multiple terminals participating in the AI / ML operation and whether the network function serves multiple terminals.

3. The method according to claim 2, wherein: Based on the fact that the terminal participating in the AI / ML operation is only the first terminal, it is determined whether to perform congestion control on the first terminal regardless of the AI / ML operation related information.

4. The method according to claim 2, wherein: Based on the presence of multiple terminals participating in the AI / ML operation and the network function serving all of the multiple terminals, the network function determines whether to perform congestion control on the first terminal by reflecting the AI / ML related information and status information of other terminals.

5. The method according to claim 4, in, The network function is an access and mobility management function (AMF), and Wherein, based on the status information of the other terminals and the fact that the other terminals are in any one of user plane deactivation and CM-IDLE of the AI / ML operation related protocol data unit PDU, the AMF determines whether to perform congestion control on multiple terminals based on the possibility of sending NAS messages.

6. The method according to claim 5, wherein: Based on performing the congestion control on the multiple terminals, the AMF sets a backoff timer value for each of the multiple terminals so that the transmission delay termination time points of the multiple terminals are the same, and sends the backoff timer value to each of the multiple terminals.

7. The method according to claim 5, wherein: The AMF determines not to perform congestion control on the first terminal based on the possibility of sending the NAS message based on the other terminal being in any one of a user plane activation and a CM-CONNECTED of an AI / ML operation related protocol data unit PDU.

8. The method according to claim 4, in, The network function is a session management function SMF, and Wherein, based on whether the NAS message sent by the first terminal is associated with a PDU session of the AI / ML operation, the SMF determines whether the first terminal participates in the AI / ML operation.

9. The method according to claim 8, in, The SMF determines the possibility of sending a NAS message based on a quality of service QoS flow included in a PDU session used by the other terminal for the AI / ML operation as status information of the other terminal, and The SMF determines to perform congestion control on a plurality of terminals based on the possibility of sending the NAS message.

10. The method according to claim 9, wherein: Based on performing the congestion control on the multiple terminals, the SMF sets a backoff timer value for each of the multiple terminals so that the transmission delay termination time points of the multiple terminals are the same, and sends the backoff timer value to each of the multiple terminals.

11. The method according to claim 1, wherein: Based on the fact that a plurality of terminals participate in the AI / ML operation and the network function serves only some of the plurality of terminals, the network function determines not to perform congestion control on the first terminal based on the AI / ML related information.

12. The method according to claim 1, wherein: Based on the fact that a plurality of terminals participate in the AI / ML operation and the network function only serves some of the plurality of terminals, the network function determines whether to perform congestion control on the first terminal based on the AI / ML related information and the AI / ML operation importance information.

13. The method according to claim 1, wherein: The AI / ML information includes at least one of identification information of a terminal participating in the AI / ML operation, AI / ML operation start time information, AI / ML operation required time information, AI / ML operation end time information, or AI / ML operation importance information.

14. The method according to claim 13, wherein: At least one information included in the AI / ML related information is simultaneously transmitted to a core network.

15. The method according to claim 14, wherein: Some of at least one information included in the AI / ML related information is transmitted to the core network at a first time point, and the other information is transmitted to the core network at a second time point.

16. A network function operating in a wireless communication system, the network function comprising: at least one transceiver; at least one processor; as well as at least one memory capable of being operatively connected to the at least one processor and configured to store instructions that, when executed, cause the at least one processor to perform specific operations, The specific operations include: Obtain AI / ML related information from the application function AF; Determining whether to perform congestion control on the first terminal based on the obtained AI / ML related information; Controlling the at least one transceiver to receive a non-access stratum (NAS) message from the first terminal; and performing congestion control based on the NAS message, Whether to perform congestion control on the first terminal is determined based on whether the first terminal participates in AI / ML operation.

17. A method for operating a terminal in a wireless communication system, the method comprising: The terminal sends a NAS message to the network function; as well as receiving a response message from the network function based on whether congestion control is performed on the terminal, The network function obtains AI / ML related information from the AF, and determines whether to perform congestion control on the terminal based on whether the terminal participates in the AI / ML operation.

18. A terminal operating in a wireless communication system, the terminal comprising: at least one transceiver; at least one processor; as well as at least one memory operatively connectable to the at least one processor and configured to store instructions that, when executed, cause the at least one processor to perform specific operations, The specific operations include: controlling the at least one transceiver so that the terminal sends a NAS message to a network function; and controlling the at least one transceiver to receive a response message from the network function based on whether congestion control is performed on the terminal, The network function obtains AI / ML related information from the AF, and determines whether to perform congestion control on the terminal based on whether the terminal participates in the AI / ML operation.

19. An apparatus comprising at least one memory and at least one processor functionally connected to the at least one memory, wherein: The at least one processor controls the device to: Obtain AI / ML related information from the application function AF; determining whether to perform congestion control on another device based on the obtained AI / ML related information; receiving a Non-Access Stratum (NAS) message from the other device; and performing congestion control based on the NAS message, Whether to perform congestion control on another device is determined based on whether the other device participates in AI / ML operations.

20. A non-transitory computer-readable medium storing at least one instruction, the non-transitory computer-readable medium including the at least one instruction executable by a processor, in, The at least one instruction control device: Obtain AI / ML related information from the application function AF; determining whether to perform congestion control on another device based on the obtained AI / ML related information; receiving a Non-Access Stratum (NAS) message from the other device; and performing congestion control based on the NAS message, Whether to perform congestion control on another device is determined based on whether the other device participates in AI / ML operations.

Images