NF device, and operating method of NF device

Abstract

The present invention presents a new technological solution, deviating from conventional NF unit implementations / structures capable of performing only call processing of a given function through design, for enabling necessary call processing in each NF (for example, UENF, RANF, and Core NF) on the basis of a generative language model (for example, LLM / GenAI) that generates / transfers a call processing procedure in each NF (for example, UENF, RANF, and Core NF).

Description

NF device and method of operating the NF device

[0001] The present invention relates to a technique for performing call processing in each NF (e.g., UENF, RANF, Core NF).

[0002] The present application claims priority to application No. 10-2024-0190072 filed on December 18, 2024, and the entire contents of such application are incorporated herein by reference for all purposes.

[0003] In 5G / 6G networks, the introduction of new features and signaling processed by SBI-based protocols between NFs is increasing significantly.

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

[0005] Meanwhile, NFs currently in commercial use are implemented to perform 3GPP-based call processing designed according to 3GPP standards, so it is possible to perform call processing of a given function through design.

[0006] Therefore, currently commercialized / used NFs are unable to handle standard-defined operations or undefined, errors, and parts not specified in the design / implementation (e.g., receiving unprocessable messages / signals / data).

[0007] Therefore, in current networks composed of NFs, it is difficult to avoid commercial issues and failures, and on the other hand, since excessive verification and time are required to improve operation, there are limitations in terms of inefficiency from the perspective of TCO (Total cost of ownership).

[0008] Meanwhile, current standards propose technologies for predicting, mitigating, and preventing network abnormalities based on the Network Data Analytics Function (NWDAF), which provides network analysis.

[0009] However, NWDAF-based technology also has inefficient limitations from a TCO perspective, such as requiring significant time for verification, as it is based on NF implemented to perform only the call processing of functions given through design.

[0010] The problem to be solved by the present invention is to move away from the existing NF unit implementation / structure, which can only perform call processing of a given function through design, and to realize a new technical method that performs complex NF operations to enable optimal call processing in each NF (e.g., UENF, RANF, Core NF).

[0011] A Network Function (NF) device according to one aspect of the present invention comprises: a memory containing instructions; and a processor that receives a call processing procedure to be performed by the NF device as Procedure Information (PI) from a specific NF that determines a call processing procedure between a User Equipment (UE) and a Core network by executing the instructions.

[0012] Specifically, the specific NF can determine a call processing procedure that can be performed by the NF device based on a request or subscription of the NF device and transmit a PI to the NF device for performing the determined call processing procedure.

[0013] Specifically, the specific NF can determine the call processing procedure to be transmitted to the NF device based on a generative language model (LM).

[0014] Specifically, the processor performs the following operations according to the received PI, and the following operations may include the operation of exchanging and processing signaling messages between NFs in the NF device, the operation of updating (e.g., installing, creating, modifying, deleting) the Software, Firmware, Version, and Code of the NF device, and the operation for a specific Management Procedure.

[0015] Specifically, when the processor determines that a call processing procedure to be performed for a specific event is required, it configures a Prompt used as an input to the generative language model within the specific NF based on the specific Event, and transmits a Prompt-based PI request to the specific NF, and the PI request may be a request for a call processing procedure to be performed by the NF device.

[0016] Specifically, the Prompt is configured to include information about the specific event, status information of the NF device, and information about the call processing procedure to be received, or may be configured to further include information about the time or period associated with the Prompt configuration.

[0017] Specifically, the processor may understand or reinterpret the text of the received PI to determine whether the next operation according to the text of the PI can be performed, and if it is determined that it cannot be performed, it may reject execution and then reconstruct or reorganize the Prompt to send the PI request again to the specific NF, perform only some operations in the next operation according to the text of the PI (Partial Execution), or re-select the function to request the PI among multiple functions related to the generative language model of the specific NF and then configure the Prompt to send the PI request.

[0018] A method of operation of a Network Function (NF) device according to one aspect of the present invention comprises: receiving a call processing procedure to be performed by the NF device as Procedure Information (PI) from a specific NF that determines a call processing procedure between a User Equipment (UE) and a Core network; and performing the next operation according to the received PI.

[0019] Specifically, the specific NF determines a call processing procedure to be transmitted to the NF device based on a generative language model (LM); if the NF device determines that a call processing procedure to be performed for a specific event is required, it further includes the step of configuring a Prompt used as an input to the generative language model within the specific NF based on the specific event and transmitting a Prompt-based PI request to the specific NF, wherein the PI request may be a request for a call processing procedure to be performed by the NF device.

[0020] Specifically, the following operations may include the operation of exchanging and processing signaling messages between NFs in the NF device, the operation of updating (e.g., installing, creating, modifying, deleting) the Software, Firmware, Version, and Code of the NF device, and the operation for a specific Management Procedure.

[0021] A computer program according to one aspect of the present invention may be stored in a medium to execute the steps of: receiving a call processing procedure to be performed by the NF device as a Procedure Information (PI) from a specific NF that is combined with hardware of a Network Function (NF) device and determines a call processing procedure between a User Equipment (UE) and a Core network based on a Generative Language Model (LM); and performing a subsequent operation according to the received PI.

[0022] According to embodiments of the present invention, moving away from existing NF unit implementations / structures that can only perform call processing of a given function through design, a technology configuration is realized that performs complex NF operations to enable the necessary call processing in each NF (e.g., UENF, RANF, Core NF) based on a generative language model (e.g., LLM / GenAI) that generates / transmits call processing procedures to each NF (e.g., UENF, RANF, Core NF).

[0023] As a result, the present invention supports optimal call processing generated / transmitted in each NF (e.g., UENF, RANF, Core NF) and automatically supports improvements in operation, handling of anomalies and abnormalities in each NF, thereby enabling the effect of significantly improving the inefficiency limits from a TCO perspective compared to existing methods.

[0024] In addition, since the call processing that must be carried out during design / development in the NF implementation can be minimized in the present invention, the effect of efficiently processing the initial state of On-device and On-NF development and distribution can also be expected.

[0025] Figure 1 is an example diagram illustrating a network composed of current NFs.

[0026] Figure 2 is an example diagram illustrating a network / network to which automatic call processing technology based on a generative language model (e.g., LLM / GenAI) proposed in the present invention is applied.

[0027] FIGS. 3 to 8 are drawings of a method for processing a call processing procedure based on a generative language model (e.g., LLM / GenAI) in the proposed technology of the present invention.

[0028] FIGS. 9 to 16 are drawings of a specific information (e.g., snapshot) processing method for configuring and utilizing specific information (e.g., snapshot) of NF for the optimization of the proposed technology of the present invention.

[0029] Hereinafter, various embodiments of the present invention will be described with reference to the attached drawings.

[0030] The present invention relates to a technique for performing call processing in each NF (e.g., UENF, RANF, Core NF).

[0031] In 5G / 6G networks, the introduction of new features and signaling processed by SBI-based protocols between NFs is increasing significantly.

[0032] As such, as the number of interconnections between NFs increases and the Core network becomes the center of the 5G / 6G system (5GS / 6GS), supporting a powerful and resilient Core is becoming more complex and difficult.

[0033] In this regard, Figure 1 conceptually illustrates a network composed of NFs currently in commercial use.

[0034] As can be seen in Figure 1, in the case of a network composed of UEs and NFs, the logical structure appears simple, but in reality, a large amount of signaling occurs during communication between the UEs and NFs, and in addition, there are a large number of call procedures both internally and externally within the NFs.

[0035] Furthermore, in actual commercial environments such as national office redundancy / distribution, each type of equipment / NF has a larger number of sub-nodes and communicates with one another in a full mesh structure; therefore, it can be viewed as a very complex structure where it is difficult to assess mutual influence.

[0036] Meanwhile, NFs currently in commercial use are implemented to perform 3GPP-based call processing designed according to 3GPP standards, so it is possible to perform call processing of a given function through design.

[0037] In other words, the current UE and NF can only perform standard call processing (or exceptionally, separately defined Mandatory / Optional standards) loaded during design / development, and for example, NF 1 can only perform call processing with a defined NF according to the execution rules of the loaded standard call processing.

[0038] Therefore, currently commercialized / used NFs cannot immediately handle standard-defined operations, or undefined, errors, and parts not specified in the design / implementation (e.g., receiving unprocessable messages / signals / data).

[0039] Therefore, in current networks composed of NFs, it is difficult to avoid commercial issues and failures, and on the other hand, since excessive verification and time are required to improve operation, there are limitations in terms of inefficiency from the perspective of TCO (Total cost of ownership).

[0040] Meanwhile, in order to improve the operation of current NFs, it is essential to enhance the detailed understanding of various call processing (signaling, data) via End-to-End (E2E). However, currently, there is a lack of analysis technology for E2E, and it is very difficult to identify the cause of abnormalities in specific messages, and it takes a considerable amount of time.

[0041] In this regard, current standards present technologies for predicting, mitigating, and preventing network abnormalities based on the Network Data Analytics Function (NWDAF) that provides network analysis.

[0042] However, NWDAF-based technology also has inefficient limitations from a TCO perspective, such as requiring significant time for verification, as it is based on NF implemented to perform only the call processing of functions given through design.

[0043] Accordingly, the present invention proposes a new technical method for performing complex NF operations to enable optimal call processing in each NF (e.g., UENF, RANF, Core NF), moving away from existing NF unit implementations / structures that can only perform call processing of a given function through design.

[0044] To explain in detail, current Telco Networks are enhancing the functional efficiency of the system for call processing by learning and analyzing various data in line with the evolution of AI technology, and then judging and predicting phenomena.

[0045] In AI technology, the foundation of analysis lies in collecting training data, and it is crucial to collect appropriate data using appropriate methods and analyze / learn from it as training data.

[0046] In particular, in a network composed of numerous NFs, call processing is performed simultaneously in each distributed NF (e.g., UENF, RANF, Core NF) within a very short time (moment). Therefore, for data in such a communication environment to be utilized as training data, real-time performance and synchronization between each NF / system are essential depending on the purpose of collection.

[0047] In this regard, FIG. 2 is an example diagram illustrating a network / network to which automatic call processing technology based on a generative language model (e.g., LLM / GenAI) proposed in the present invention is applied.

[0048] That is, the present invention is a core technology of Telco Edge AI and AI-assisted Infra, and terminals (UE, UENF), base stations (RAN, RANF), and cores (Core NF) can automatically perform their own roles, functions, and operations by linking with a generative language model (e.g., LLM / GenAI) that generates / transmits call processing procedures based on 3GPP standards.

[0049] As a result, in the present invention, since each NF (e.g., UENF, RANF, Core NF) can automatically generate / transmit optimal call processing, it is possible to automatically identify and address operational improvements, anomalies, and abnormalities of each NF, and furthermore, the call processing that must be implemented during design / development in NF implementation can be minimized.

[0050] In summary, the core content of the automatic call processing technology based on a generative language model (e.g., LLM / GenAI) proposed in the present invention can be divided into a method for operation / processing between each NF and PF (Procedure Function, a function using a generative language model (e.g., LLM / GenAI) / newly defined as an NF), a method for each NF to receive its own call processing procedure, rule, etc. from the PF (related to Feature 1 described below), and a method for each NF to share the current call processing state with the PF, etc. (related to Feature 2 described below).

[0051] As illustrated in FIG. 2, the present invention proposes feature 2 (Figs. 9 to 16 described below) which collects and manages various data about the network (snapshot described below) on a whole basis and utilizes the data for AI processing, so that complex inference can be supported in individual systems of the network or in general application servers and terminals outside the Telco Network, and feature 1 (Figs. 3 to 8 described below) which generates and transmits future call processing procedures to be performed in the NF.

[0052] In particular, the present invention is centered on feature 1 (also known as an LLM / GenAI-based PI processing method) which generates and transmits a future call processing procedure to be performed in NF.

[0053] Accordingly, the present invention enables the execution of necessary call processing in each NF (e.g., UENF, RANF, Core NF) by utilizing a generative language model (e.g., LLM / GenAI) that generates and delivers call processing procedures to each NF (e.g., UENF, RANF, Core NF) through an LLM / GenAI-based PI processing method, moving away from the existing NF unit implementation / structure that can only perform call processing of a given function through design.

[0054] Meanwhile, the NF to which the present invention can be applied refers to the terminal (UE), RAN, and Core NF.

[0055] At this time, Core NF will include nodes of the Control Plane (CP) in 5G / B5G / 6G, namely ANF, SMF, PCF, NEF, UDM / AUSF, etc., and nodes of the User Plane (UP), UPF.

[0056] It can be defined as a User Plane Function.

[0057] Meanwhile, recently, research has been conducted to convert terminals (UE) and RAN (Radio Access Network) into NFs by evolving them into B5G / 6G Architectures and implementing them as NFs, thereby enabling communication not only between the NFs of the Control Plane and User Plane but also between the UE (hereinafter UENF) and RAN (hereinafter RANF) using an improved SBA structure and SBI.

[0058] Accordingly, the NF device proposed in the present invention below may be an NF of UP or an NF of CP, and furthermore, may be a UENF or a RANF. In addition, the NF device proposed in the present invention may be composed of a CNF (Cloud-native Network Function; Pod unit) or a VNF (Virtual Network Function; VM unit).

[0059] As previously mentioned, the present invention proposes an automatic call processing technology based on a generative language model (e.g., LLM / GenAI) that enables call processing to be performed in each NF (e.g., UENF, RANF, Core NF) based on a generative language model (e.g., LLM / GenAI) that generates / transmits a call processing procedure to each NF (e.g., UENF, RANF, Core NF).

[0060] Specifically, the automatic call processing technology based on a generative language model (e.g., LLM / GenAI) proposed in the present invention can be divided into feature 1, which generates and transmits a future call processing procedure to be performed in NF based on a generative language model (e.g., LLM / GenAI), and feature 2, which is snapshot processing.

[0061] Below, I will specifically explain Feature 1, which processes (generates / transmits) Procedure Information (PI) based on a generative language model (e.g., LLM / GenAI) proposed in the present invention.

[0062] In the present invention, a procedure function (PF) can be newly defined as a function / NF using a generative language model (e.g., LLM / GenAI). Of course, the name of the PF is merely an example and could be defined with a different name.

[0063] Referring to FIG. 3, in the present invention, each NF (100, e.g., UENF, RANF, Core NF) can receive and perform its own call processing procedure (Procedure, in particular, information for performing the Procedure (hereinafter, PI (Procedure Information))) from PF (200).

[0064] That is, each NF (100, e.g., UENF, RANF, Core NF) can receive appropriate guidance on the "next action" considering the current network as a PI, such as which call processing procedure to process and in what way or with which NF to send and receive data / messages.

[0065] Here, since the procedure can operate differently depending on the pre-configured information or dynamic decision information of various NFs, the state information (state-machine) of the current processing target (Context), and the form and content (parameter) of the received or transmitted message, it is practically impossible to perform operations considering complex logic at the NF level.

[0066] This is because NFs operating based on existing designs may perform unintended actions, including fail, error, and abnormal behavior, depending on abnormal behavior not considered in advance, and also because even among functionally identical NFs, they may operate differently depending on various standard specifications, release, software, library, and code versions.

[0067] There are limitations to specific designers, developers, or operators considering all these various variables in advance or identifying and responding to them within a short period of time via E2E.

[0068] Accordingly, the present invention aims to improve all the difficulties, problems, and limitations of the aforementioned method by proposing an automatic call processing technology based on a generative language model (e.g., LLM / GenAI), defined as the evolution of Telco Edge AI and AI-assisted Infra.

[0069] To briefly explain, the automatic call processing technology based on a generative language model (e.g., LLM / GenAI) proposed in the present invention stores / loads a procedure through a snapshot, and through communication with a PF (200, generative language model (e.g., LLM / GenAI)), each NF (100) can perform an appropriate / optimal operation according to the procedure given by the PF (200) without complex computational logic.

[0070] The call processing procedure mentioned in the present invention refers to a procedure based on signaling messages exchanged through mutual interfaces between NFs.

[0071] The signaling messages constituting this may be transport or application layer messages such as SCTP, GTP-C, HTTP / 2, HTTP / 3, QUIC, etc.

[0072] In addition, this signaling message is exchanged between 3GPP-based terminals / base stations / cores, or among themselves, and may be related to execution procedures including call processing based on requests from terminals or each network unit (e.g., Connection, Registration, Mobility, Session, Radio Resource, Bearer, Identity, AMF, Location, Policy, Security Management).

[0073] And each message can be various unit information defined by 3GPP, such as UE context, NF context, mobility management context, session management context, security context, access stratum context, non-access stratum context, bearer context, radio resource control context, UE context management context, handover context, paging context, and QoS flow context information. For example, UE context can be configuration information combining the terminal's idle / active state, battery level, and power level, while mobility can be configuration information combining the terminal's mobility.

[0074] Ultimately, in the present invention, the call processing procedure, i.e., Procedure, that each NF (100) receives from the PF (200) can be described as information for performing the next operation / signaling processing for each Procedure required by each NF.

[0075] As an example of the call processing procedure mentioned in the present invention, a procedure can be described in which three NFs (NF1, 2, 3) transmit and receive specific signaling messages to and from each other.

[0076] For example, NF1 to NF2 can be Connection Management Procedures.

[0077] And NF2 ~ NF3 can be Session Management Procedures.

[0078] And NF3 ~ NF1 can be Mobility Management Procedures.

[0079] In other words, the call processing procedure, or Procedure, proceeds through signaling messages between each NF and can be divided into various types of Procedures.

[0080] A Procedure is not limited to a 1:1 relationship between NFs, but can be defined as involving multiple NFs depending on the type of Procedure. Communication between NFs can be unidirectional (e.g., send_only) or bidirectional (send / receive). Additionally, from a specific perspective, a Procedure can be a request and response, or a subscription and notification for temporary or continuous information delivery requests.

[0081] The PI that each NF (100, e.g., UENF, RANF, Core NF) receives from the PF (200), that is, the procedure information for performing a procedure determined / generated by the PF (200) through LLM / GenAI-based processing (Inferencing), may be an operation / method for exchanging and processing signaling messages between NFs as an appropriate guide for the next operation considering the current network, or it may be an operation of changing the version, software, etc. of the NF (100).

[0082] For example, in the case of a Mobility Management Procedure, the corresponding AMF PI examples are as follows.

[0083] 1. UE Configuration Update Procedure: This procedure is initiated when the AMF wants to update the UE's configuration.

[0084] PI (Procedure Information)

[0085] 1) Transferred from AMF to UE: UE Configuration Update Command

[0086] - Included information: Configuration update indication, 5G-GUTI, TAI list, Allowed NSSAI, etc.

[0087] 2) Transfer from UE to AMF: UE Configuration Update Complete

[0088] 2. Registration Procedure: The basic procedure for a UE to register with a network

[0089] PI (Procedure Information)

[0090] 1) Transfer from UE to AMF: Registration Request

[0091] 2) Passed from AMF to UDM: Nudm_UEContextManagement_Get Request

[0092] 3) Passed from UDM to AMF: Nudm_UEContextManagement_Get Response

[0093] 4) Transfer from AMF to UE: Registration Accept

[0094] 5) Transfer from UE to AMF: Registration Complete

[0095] 3. Handover Procedure: The procedure that occurs when a UE moves to another base station.

[0096] PI (Procedure Information)

[0097] 1) Transfer from existing gNB to AMF: Handover Required

[0098] 2) Transfer from AMF to new (to be transferred) gNB: Handover Request

[0099] 3) Transfer from the new (to be moved) gNB to the AMF: Handover Request Acknowledge

[0100] 4) Transfer from AMF to existing gNB: Handover Command

[0101] 5) Transfer from existing gNB to UE: RRC Reconfiguration

[0102] 6) Transfer from UE to new (to move) gNB: RRC Reconfiguration Complete

[0103] 7) Transfer from the new (to be moved) gNB to the AMF: Handover Notify

[0104] 4. Service Request Procedure: Used when the UE attempts to transition from the CM-IDLE state to the CM-CONNECTED state.

[0105] PI (Procedure Information)

[0106] 1) Passed from UE to RAN / AMF: Service Request

[0107] 2) Passed from AMF to gNB: Initial Context Setup Request

[0108] 3) Passed from gNB to AMF: Initial Context Setup Response

[0109] 4) Pass from AMF to UE: Service Accept

[0110] Ultimately, the key feature of the present invention is that it enables these procedures (especially PIs) to be automatically generated using AI techniques in 6G.

[0111] In other words, the present invention is a technology of a new concept in which, through the automation of the determination / generation and transmission of a Procedure (particularly PI), it is possible to determine what a terminal will request and what processing will be done in the network using AI / ML.

[0112] And, as previously explained, the PI mentioned in the present invention may be configured in relation to one of various Management Procedures, may be configured in relation to a combination of multiple Management Procedures, or may be configured in relation to / in relation to a part of a specific Management Procedure (e.g., NF Context Information).

[0113] Additionally, PI can be configured to include state-machine operations (e.g., snapshot, i.e., exact / wildcard dynamic match / action rule based on specific conditions). Accordingly, NF (100) that receives PI can execute the next operation according to PI, but can also execute the snapshot processing operation (related to feature 2).

[0114] In addition, PI can update / change the functions / operations (e.g., Software, Firmware, Version, Code, etc.) of NF (100, e.g., UENF, RANF, Core NF). Through this, NF (100) can evolve from general (common) NF functions / operations into an adaptive and advanced NF that takes into account the current network conditions and operations.

[0115] To continue the explanation, as illustrated in FIG. 3, each NF (100, e.g., UENF, RANF, Core NF) has a structure that includes an LLM / GenAI Client capable of communicating with a PF (200, LLM / GenAI), and can request a Procedure (specifically a PI) to be performed by the NF itself from the PF (200, LLM / GenAI).

[0116] PF (200) may have a structure of a server including a generative language model (hereinafter LLM / GenAI), or may have a structure using a separate LLM / GenAI.

[0117] At this time, the LLM / GenAI used or included by the PF (200) can be trained to utilize the standard specifications of the network to which the present invention applies, for example, 3GPP standard specifications, as training data to perform inferencing on an input (Input, Prompt described later) meaning instructions or questions, and to output a Procedure (particularly PI) to be performed as a result of the processing.

[0118] In this invention, no limitations or features are placed on the basic learning process of LLM / GenAI, so a detailed description will be omitted.

[0119] In the present invention, PF (200, LLM / GenAI) is responsible for determining / generating the call processing procedure between the UE (e.g., UENF) and the Core network based on LLM / GenAI.

[0120] That is, PF (200, LLM / GenAI) can determine the call processing procedure to be transmitted to NF (100) based on LLM / GenAI.

[0121] Specifically, the PF (200, LLM / GenAI) can determine a procedure that can be performed on the NF (100) based on a request or subscription of the NF (100), and can send a PI to the NF (100) for performing the determined procedure.

[0122] In one embodiment, in the case of a request, the PF (200, LLM / GenAI) can, upon receiving a PI request from the NF (100), inferencing it based on LLM / GenAI to determine / generate a procedure to be performed in the NF (100), and as a result, that is, a procedure according to the inferencing, that is, a procedure to be performed in the NF (100) (especially a PI), can be transmitted to the NF (100) as a response to the PI request.

[0123] Here, each NF (100) including an LLM / GenAI Client can be implemented as a 3GPP NF and can typically be configured as an NF-generic, Common NF, or a specific NF-specific or RAN-specific or UE-specific NF such as RAN, UE, AMF, SMF, UPF, etc.

[0124] When NF (100) communicates with PF (200), it can be connected to an existing or new Interface (Plane).

[0125] In the case of an existing Interface (Plane), NF (100) can use P2P (point-to-point, e.g., GTPC-v2, SCTP protocol) and SBI (service-based interface, e.g., HTTP2 / , HTTP / 3 protocol) when communicating with PF (200). Alternatively, NF (100) can communicate based on a new Interface (Plane) (e.g., AI / ML-plane, Training-plane, LLM-plane, model-plane, etc.) when communicating with PF (200).

[0126] And each NF (100) may have a Client capability representing LLM / GenAI execution information.

[0127] Client capability representing LLM / GenAI execution information in each NF(100) can basically be composed of the following three:

[0128] ·Input Message (Input Value)

[0129] · Process

[0130] ·Output Message (Output Value)

[0131] The key point is that each NF (100) can optimize the Procedure (especially PI) that needs to be performed in the NF (100), i.e., the next action, through a Client capability that represents LLM / GenAI execution information.

[0132] At this time, the next operation performed according to the PI may be an operation / method of exchanging and processing signaling messages between NFs in NF (100), and may be an operation of updating (e.g., installing, creating, modifying, deleting) / changing the Software, Firmware, Version, Code, etc. of NF (100).

[0133] In addition, another key point is that each NF (100) can perform operations starting from an NF-generic or Common NF through a PI request / response, and then as a more sophisticated and specific NF (e.g., a specific NF-specific or RAN-specific or UE-specific).

[0134] To elaborate, each NF (100) containing a Client for LLM / GenAI execution may have the following components.

[0135] 1. API Integration Layer: This layer is responsible for communication with LLM services, such as the GPT (Generative Pre-trained Transformer) model and the BERT (Bidirectional Encoder Representations from Transformers) model. It manages API calls, authentication, and data formatting.

[0136] 2. Prompt Management: Handles the creation, storage, and optimization of prompts used to interact with the LLM. It may include a prompt library and a template system.

[0137] 3. Response Processing: It is responsible for parsing and processing the response received from the LLM. It may also include streaming functions for real-time output.

[0138] 4. Error handling and retry logic: It can manage API errors and rate limits, and perform a retry mechanism for failed requests.

[0139] 5. Caching Layer: Optionally include a caching system that stores and retrieves previous responses to reduce API calls and improve performance.

[0140] 6. Configuration Management: Manages settings such as API endpoints, model parameters, and client-specific configurations.

[0141] And, each NF (100) including an LLM / GenAI Client can perform a PI request / response as follows.

[0142] · PI Request Process

[0143] 1. Prompt Preparation: The client prepares a prompt containing the user's question or instruction. This prompt is configured in a form that the LLM can understand and respond to.

[0144] 2. API Call: Sends an API request including a prepared prompt. The API can be built using a Service-Based Interface (SBI), such as HTTP / 2 or HTTP / 3. This PI request typically includes the following information:

[0145] - Model to use (e.g., ChatGPT-4o, ChatGPT-4, ChatGPT-5, ChatGPT-Next, Claude 3, Gemini 1.5, Llama 3.1), prompt text, other parameters (maximum number of tokens, etc.)

[0146] · PI Response Process

[0147] 1. Text generation: PF(200, LLM / GenAI) processes the received prompt and generates an appropriate text response.

[0148] 2. Response Format: The response can generally be returned in JSON format (of course, other formats are also possible) and may include the following information.

[0149] - Generated text, number of tokens used, other metadata

[0150] 3. Response processing: NF(100) parses the received PI response and processes it as needed (e.g., next action, etc.).

[0151] Below, I will explain an example of how an LLM / GenAI Client in NF (100) configures a prompt.

[0152] A prompt refers to an instruction or question that an LLM / GenAI Client inputs into LLM / GenAI, and the following explains example prompt styles 1, 2, and 3.

[0153] 1. "received a request from RAN, the message is "Service Request"

[0154] "Service Request contains values (abc, def, xyz)"

[0155] "Tell me the next action including relevant procedures when received from RAN"

[0156] 2. "Procedure Management : Registration Procedure"

[0157] Received a request from UE, RAN or NF.

[0158] The message was InitialUEMessage containing Tracking Area Update request

[0159] UE ID = abc, CellID = xyz,

[0160] Tell me the next procedure action, and tell me about the UE related procedures

[0161] 3. "Procedure Management : Service Request"

[0162] Received a request AMF.

[0163] The message was PDU_session_Create_SMContext

[0164] Content has SUPI, selected DNN, request DNN, S-NASSI(s), PDU Session ID, AMF ID, Request Type, PCF ID, Priority Access, N1 SM container(PDU session Establishment Request), ..)

[0165] The content is already processed before

[0166] Update me the 3gpp-release version based on this and new procedure if needed

[0167] FIG. 4 shows an example of the operation / processing between each NF and PF and the method of receiving a Procedure (particularly PI) from PF in each NF in the proposed technology of the present invention.

[0168] Each NF (100) can register itself with PF (200) through communication with PF (200) (Registration, etc.).

[0169] At this time, the NF (100) can register its own status, NF capabilities related to the Procedure, etc. Based on the registration information of each NF (100), the PF (200) can understand the capabilities of the NF and PF from each other and, if necessary, discover the necessary NF.

[0170] For example, PF (200) may have a number of functions (e.g., ChatGPT-4o, ChatGPT-4, ChatGPT-5, ChatGPT-Next,…, performance, throughput, resource, latency, accuracy, precision, etc.) in relation to the generative language model, and NF (100) may select or re-select the function to request PI by considering the multiple functions of such PF (200).

[0171] NF(100) determines whether a procedure to be performed is required when a specific event occurs.

[0172] For example, NF(100) can recognize that a specific event has occurred when a UE Registration message is received and determine whether a Procedure is required.

[0173] NF(100) can proceed with the next action, i.e., call processing, if a procedure for a specific event (e.g., UE Registration message) exists in the internal storage, and if the procedure does not exist in the internal storage or if the message cannot be processed by the call processing logic within NF(100), it can be determined that a procedure is required (Y).

[0174] If NF (100) determines that a Procedure is required (Y), it can construct a Prompt, i.e., a Text-based Input, as described in the tactical example and send a Prompt-based PI request to PF (200). At this time, the scope of the PI request depends on the Prompt information generated by NF (100).

[0175] Since the PF (200, LLM / GenAI) has already learned the 3GPP standard specifications, when a PI request is received from the NF (100), it processes (Inferencing) it based on the LLM / GenAI to determine / generate the procedure to be performed in the NF (100), and as a result, that is, the procedure according to the processing (Inferencing), that is, the procedure to be performed in the NF (100) (especially the PI), can be sent to the NF (100) as a response to the PI request.

[0176] NF(100) can determine whether a Procedure (especially a PI) is possible when it is returned from PF(200, LLM / GenAI).

[0177] For example, NF (100) can understand the text of the PI returned from PF (200, LLM / GenAI) and, if it determines (Y) that there is no load issue (e.g., overload, processing load high) in performing the next action, it can perform the action (Perform Procedure).

[0178] NF (100) can determine that if the text of the PI is difficult to understand or if there is a load (e.g., overload, processing load high) in executing the next action according to the text, it cannot execute (N), reject execution, and reconfigure / reorganize the prompt and send the PI request back to PF (200).

[0179] In addition, if NF (100) determines that execution is impossible (N), in the next action according to the text of the current PI, 1) Partial Execution (perform only some actions), 2) re-select the function that requested the PI from among the functions related to the generative language model of PF (200) (e.g., ChatGPT-4o, ChatGPT-4, ChatGPT-5, ChatGPT-Next, Claude 3, Gemini 1.5, Llama 3.1) and then reorganize the Prompt / request PI, 3) handle the error / retry, or 4) do nothing.

[0180] In particular, in order to receive the optimal PI / next action from PF (200, LLM / GenAI), it is very important for NF (100) to optimally configure / generate the prompt, i.e., textual input, when requesting a PI.

[0181] Of course, in the present invention, NF (100) may initially configure / generate the Prompt as a long text.

[0182] In the present invention, in order to receive an optimal PI / next action, it is necessary to configure / generate a Prompt with the following information.

[0183] · Received Events (Event Information)

[0184] · My states (state information of NF(100))

[0185] · Desired response (information on the call processing procedure you wish to receive)

[0186] Additionally, to ensure a smooth response, you may include the time / period related to the prompt configuration and the corresponding content (e.g., file, pcap, trace).

[0187] The details regarding optimizing the Prompt configuration in this way will be mentioned again in the section describing feature 2 of snapshot processing, which will be explained in detail later.

[0188] In the present invention, NF (100), which receives a response (PI) to a PI request from PF (200), can understand and reinterpret the Text of the PI as described above.

[0189] The reason for the reinterpretation is to determine whether the next action according to PI can be performed, either by NF itself or based on setting criteria such as overload during execution.

[0190] Meanwhile, Fig. 5 shows examples of Prompt configuration / generation in the present invention.

[0191] In the present invention, NF (100) can generate / configure a Prompt for a PI request. The key here is to configure the Prompt optimally.

[0192] Basically, NF(100) is based on a "received Event" that determines that a Procedure to be performed is needed, and can configure a Prompt containing the contents necessary to perform the Event.

[0193] At this time, NF (100) can configure the Prompt to include information about what information is in the received Event, “Received Events”, what is the status of NF (100) itself, “My Status”, and what answer is expected, “Request for next action”.

[0194] FIG. 6 illustrates a scenario in which, as an embodiment of the present invention, an NF (100, e.g., SMF) performs a PI request / response for an Event in which a message in SBI JSON format is received.

[0195] Referring to Fig. 6, NF (100, e.g., SMF) determines whether a procedure to be performed based on a message is required when a specific event, namely the reception of a message from AMF, occurs.

[0196] For example, NF(100) can recognize that a specific event has occurred when a UE Registration message is received and determine whether a Procedure is required.

[0197] If NF (100, e.g., SMF) determines that a Procedure is required, it can generate / configure a Text-based Input, i.e., a Prompt, to transmit to PF (200) and send a PI request to PF (200) (Request PI). At this time, the scope of the PI request depends on the Prompt information generated by NF (100).

[0198] Accordingly, in the present invention, NF (100) may apply / use various techniques to optimally configure the Prompt (also known as Prompt Engineering).

[0199] For example, you can select and use Prompt Optimization / Engineering features that enable Zero-shot (a method of instructing the LLM to perform a task without providing specific examples), Few-shot (a method of providing a few examples to the LLM to elicit a response for a given task), Chain-of-thought (a method of significantly improving performance in complex tasks by guiding the LLM to clearly express inference steps), Role (a method of obtaining output that matches the desired intent by assigning a specific role or character to the LLM), and Iterative refinement (a method of continuously improving the prompt based on the LLM's response).

[0200] When a PI request is received from an NF (100, e.g., SMF), the PF (200, LLM / GenAI) processes (Inferencing) it based on LLM / GenAI to determine / generate a Procedure to be performed at the NF (100), and can send / reply to the NF (100, e.g., SMF) the result, i.e., the Procedure according to the processing (Inferencing), i.e., the Procedure to be performed at the NF (100) (especially PI), as a response to the PI request (Response PI).

[0201] NF(100, e.g., SMF) can determine whether a Procedure (especially PI) is feasible when it is returned from PF(200, LLM / GenAI).

[0202] NF(100, e.g., SMF) determines that it is possible to execute the following actions if there is no load (e.g., overload, high processing / memory load) in understanding / reinterpreting the text of the PI returned from PF(200, LLM / GenAI) and performing the next action accordingly (Yes).

[0203] Accordingly, NF(100, e.g., SMF) can process the specific event that occurred this time by performing the next action according to PI, such as sending a specific message (e.g., N1N2 Message to UE) to AMF.

[0204] If NF (100, e.g., SMF) determines that the Text of the PI is difficult to understand or that there may be a load (e.g., overload, processing load high) in performing the next action according to the Text, it determines that execution is impossible (N), and can proceed again with the process of creating / configuring a Prompt and sending the PI request to PF (200).

[0205] In addition, if NF (100, e.g., SMF) determines that execution is impossible (N), in the next action according to the text of the current PI, 1) Partial Execution (perform only a part of the action), 2) re-select the function that requested the PI from the multiple functions related to the generative language model of PF (200) (e.g., ChatGPT-4o, ChatGPT-4, ChatGPT-5, ChatGPT-Next, Claude 3, Gemini 1.5, Llama 3.1) to configure the Prompt / request the PI, 3) handle the error / retry, or 4) handle it as Do Nothing.

[0206] Meanwhile, in the tactics section, a scenario is described in which a PI request / response is performed between one NF (100) and PF (200, LLM / GenAI), but this is merely an example for the convenience of explanation.

[0207] As illustrated in FIG. 7, in the present invention, the PF (200, LLM / GenAI) can identify the Procedure roles of the entire Network, i.e., Core NF, base station (RANF), and terminal (UENF), and can transmit PI to one or more NFs (100, e.g., UENF, RANF, Core NF) required for each Management Procedure.

[0208] Additionally, as illustrated in FIG. 8, in the present invention, an NF (100, e.g., UENF, RANF, Core NF) can transmit a PI request of the same Prompt to one or more PFs (200, e.g., different LLM versions), and can select the optimal one from the returned PI response to execute the next action according to the PI (Text).

[0209] As explained above, according to the automatic call processing technology based on a generative language model (e.g., LLM / GenAI) of the present invention (in particular, the LLM / GenAI-based PI processing method of Feature 1), a specific technical configuration is realized to enable the necessary call processing to be performed in each NF (e.g., UENF, RANF, Core NF) based on a generative language model (e.g., LLM / GenAI) that generates and delivers a call processing procedure to each NF (e.g., UENF, RANF, Core NF), moving away from the existing NF unit implementation / structure that can only perform call processing of a given function through design.

[0210] Accordingly, according to the present invention, based on a generative language model (e.g., LLM / GenAI), the basic call processing operation of each NF (e.g., UENF, RANF, Core NF) can be changed, and the call processing execution rules of the NF can be changed to achieve stability / high performance / low latency.

[0211] Next, with reference to FIGS. 9 to 16, Feature 2 of the snapshot processing method proposed in the present invention will be explained in detail below.

[0212] Referring to FIG. 9, in the present invention, each NF (100, e.g., UENF, RANF, Core NF) can transmit specific information (hereinafter, snapshot) that it has configured / generated and holds (stored) to a PF (200, LLM / GenAI) (snapshot-based logic).

[0213] Accordingly, in the present invention, the PF (200, LLM / GenAI) can collect and manage various data (snapshots) about the Network on a whole basis through the delivery of snapshots of each NF (100, e.g., UENF, RANF, Core NF), and can utilize the data for AI processing.

[0214] The purpose of the specific information, namely the snapshot, proposed in the present invention is ultimately to optimize the input (Prompt) to PF(200, LLM / GenAI) and also to train LLM / GenAI.

[0215] To this end, in the present invention, each NF (100, e.g., UENF, RANF, Core NF) may have a structure including a memory (not shown) containing instructions, and a processor (not shown, e.g., LLM / GenAI Client) that executes the instructions to construct a snapshot and transmits it to a PF (200, LLM / GenAI).

[0216] In the present invention, PF (200, LLM / GenAI) operates as a function that determines / generates a call processing procedure (Procedure, particularly PI) between the UE and the Core network through LLM / GenAI-based processing (Inferencing) in accordance with Feature 1 of the present invention described above (call processing procedure logic), and in particular, can use a transmitted snapshot of each NF (100, e.g., UENF, RANF, Core NF) for such determination / generation (e.g., LLM / GenAI additional learning).

[0217] As previously explained, in the present invention, the PF (200, LLM / GenAI) can determine a procedure that can be performed on the NF (100) based on a request or subscription of the NF (100), and can transmit a PI for performing the determined procedure to the NF (100) (call processing procedure logic).

[0218] In particular, PF(200, LLM / GenAI) can use specific information (snapshot) collected from each NF(100, e.g., UENF, RANF, Core NF) through transmission in determining / generating the Procedure.

[0219] To describe a specific embodiment, PF (200, LLM / GenAI) can utilize snapshots collected from each NF (100, e.g., UENF, RANF, Core NF) in determining / generating a Procedure (particularly PI) by further training LLM / GenAI with snapshots from each NF (100, e.g., UENF, RANF, Core NF).

[0220] Here, the specific information (snapshot) defined in the present invention may be composed of a Context defined to represent the call processing status of each NF (100, e.g., UENF, RANF, Core NF).

[0221] At this time, the Context may be defined to include time or period information in which specific information (snapshot) is configured, and detailed state information including detailed Context and implementation logic of the UE and NF related to the specific information (snapshot).

[0222] In the present invention, the generative language model (e.g., LLM / GenAI) used by PF (200) when determining / generating a Procedure (particularly PI) is a model that has learned 3GPP standard specifications.

[0223] However, in actual commercial network situations, messages defined by the carrier itself that are not defined in the 3GPP standard specifications may be used, and various situations not anticipated by the standard may occur.

[0224] As mentioned earlier, the purpose of the snapshot in the present invention is to optimize the input (Prompt) to PF(200, LLM / GenAI).

[0225] A general LLM / GenAI that has learned only 3GPP standard specifications will not be able to receive the details of a specific call process and / or non-standard messages and the correct answer (Inferencing result, PI) according to each Context state.

[0226] Accordingly, in the present invention, by using a snapshot of each NF (100) to further train the PF (200, LLM / GenAI) and including or reflecting the snapshot in the Prompt requesting the Procedure (PI) to deliver an optimized input (Prompt) to the PF (200, LLM / GenAI), accurate and precise answers (Inferencing results, PI) in actual commercial network situations can be quickly received from the PF (200, LLM / GenAI).

[0227] The snapshot processing method (feature 2) of the present invention presupposes the LLM / GenAI-based PI processing method of feature 1 described above, and enables the creation of new PIs or the updating of existing PIs by participating in the process of determining / generating a Procedure (particularly PI) in PF (200, LLM / GenAI) (e.g., additional LLM / GenAI learning).

[0228] Thus, the snapshot processing method of the present invention (feature 2) can, as a result, optimize the configuration of the Prompt (PI request of feature 1) used as an input to LLM / GenAI for commercial network conditions.

[0229] FIGS. 10 and 11 illustrate, as an example, the concept that each NF (100, e.g., UENF, RANF, Core NF) constitutes its own specific information, i.e., snapshot, according to the snapshot-based logic of the present invention.

[0230] As can be seen in Fig. 10, each NF (100, e.g., UENF, RANF, Core NF) can be a Service Consumer requesting a Service Request or a Service Producer.

[0231] However, as an example, if explained from the perspective of a Service Producer as shown in FIG. 10, in the present invention, information generated from various internal and / or external processing logic (communication between NFs) during the process of processing a received Service Request in an NF (100, e.g., UENF, RANF, Core NF) as a Service Producer can be defined as "state-machine information" representing the call processing status of the NF (100).

[0232] Figure 11 illustrates the information, or "state-machine information," generated from an NF (100, e.g., UENF, RANF, Core NF) from a Service Producer perspective, as in Figure 10, in a different way.

[0233] At this time, the snapshot of the present invention may be configured / created by capturing the call processing state ("Information", or "state-machine information") of NF (100, e.g., UENF, RANF, Core NF).

[0234] In addition, snapshots containing (capturing) this state can be generated in formats such as raw text, log / trace data, code, AI model, image / thumbnail, and video / preview.

[0235] Accordingly, a snapshot of an NF (100, e.g., UENF, RANF, Core NF) can be configured as a Context defined to represent the call processing status of an NF (100, e.g., UENF, RANF, Core NF), which includes time or period information in which the snapshot was configured (e.g., Capture), and detailed status information including detailed Context and implementation logic (e.g., Code, Programming Language, Script) of the UE and NF associated with the snapshot.

[0236] Ultimately, in the present invention, the snapshot serves as background information for determining the call processing status / execution logic of NF (100, e.g., UENF, RANF, Core NF), and may have generalized background information (e.g., network status, network condition, terminal characteristics, etc.) by considering detailed context / status information and differences after primary analysis with existing information of the message.

[0237] In this case, the network status / situation can represent the following information.

[0238] · NF external NF integration status and message statistics, request-response success rates, and related statistics by interface

[0239] · Load information of the NF itself and linked NF

[0240] · Request-response time / whether retransmission occurs during the integration with external NF, etc.

[0241] · Information related to the NF / terminal, etc., used for services provided by the NF (including UE / NF Context information used for call processing - timer, state, subscriber information, ...)

[0242] · Power consumption information during the NF operation process

[0243] And, terminal characteristics can represent the following information.

[0244] · Information on service request / processing target terminal(group)

[0245] · Location, Area, Model, Slice, Signaling Pattern, Capability, Requested Service, Usage (Statistics, etc.), Common Set Information of Requested Terminals (Groups)

[0246] In the present invention, NF (100, e.g., UENF, RANF, Core NF) can extract (e.g., Capture) information generated from internal and / or external processing logic to configure and store a snapshot, i.e., internal and external information, and can store each for each call processing.

[0247] To explain a specific embodiment, in the present invention, each NF (100, e.g., UENF, RANF, Core NF) can configure / store information for each step of the call processing process (method of exchanging and processing signaling messages between NFs) and feature procedures (especially PIs) during the call processing process when configuring / saving a snapshot, by determining the target information and storage method for each type of PI, and can transfer it to another NF (e.g., PF (200)) if necessary.

[0248] In this case, the present invention can classify and organize snapshots by Procedure (PI), and organize the steps of the Procedure, what processing is performed on what information, etc., in the form of a table or tree.

[0249] In addition, in the present invention, when saving a snapshot, it can be compressed and stored, and can be configured, created, or stored in various forms such as image, binary, and structure.

[0250] In addition, in the present invention, when storing a snapshot, information regarding the time or period during which the snapshot was configured (e.g., Capture, gathering, etc.) may be included and stored.

[0251] FIG. 12 illustrates the basic procedure of snapshot-based logic according to the snapshot processing method proposed in the present invention.

[0252] As illustrated in FIG. 12, according to the present invention, each NF (e.g., UENF, RANF, Core NF) can extract (e.g., Capture) information generated from internal and / or external processing logic, i.e., internal and external information, and store a snapshot representing its call processing status by configuring / storing these, and can utilize its snapshot when necessary.

[0253] As an example, NFs (e.g., UENF, RANF, Core NF) enable additional training of LLM / GenAI using snapshots in PF(200, LLM / GenAI) by utilizing their snapshots in snapshot-based logic and passing them to PF(200, LLM / GenAI).

[0254] Of course, as illustrated in FIG. 12, according to the present invention, each NF (e.g., UENF, RANF, Core NF) can, when a specific event is received, generate a Prompt requesting a Procedure (PI) to be executed based on the message of the specific event, transmit the PI request including the Prompt to the PF (200), and receive a response thereto (PI request / response).

[0255] Accordingly, an NF that receives a PI response (e.g., UENF, RANF, Core NF) can process the specific event that occurred this time by performing the next action according to the text of the PI being responded to.

[0256] As an example, the following operations may be possible for NFs (e.g., UENF, RANF, Core NF) according to the PI being responded to.

[0257] - Service Producer NF selection optimization

[0258] (Producer NF's functional capability, power consumption information, Load)

[0259] - Functional execution information of features requiring support

[0260] - Processing logic for parsing non-standard message formats

[0261] - Assigning / Changing NF's role (service)

[0262] - Judgment on common logic

[0263] In this case, according to the present invention, each NF involved in processing a specific event (e.g., UENF, RANF, Core NF) can extract information generated from internal and / or external processing logic, i.e., internal and external information (e.g., Capture), and save a snapshot (e.g., save new, update existing, etc.).

[0264] Meanwhile, Figure 13 illustrates a procedure through Snapshot Transfer between NFs of snapshot-based logic according to the snapshot processing method proposed in the present invention.

[0265] As can be seen in FIG. 13, according to the present invention, each NF (e.g., UENF, RANF, Core NF), as in one embodiment, NF (e.g., UENF), can save a snapshot while performing its call processing.

[0266] The snapshot stored here can be combined with the state of NF (e.g., UENF), call processing trace, log, etc., and can be compressed for efficiency and security purposes.

[0267] This Snapshot information must be transferred to PF(200, LLM / GenAI), but if an NF (e.g., UENF) does not have a direct connection with PF(200, LLM / GenAI), it may be transferred to PF(200, LLM / GenAI) via another NF (e.g., RANF, Core NF) (Snapshot_Transfer).

[0268] For example, another NF (e.g., RANF, Core NF) can review the snapshot of the NF being delivered (e.g., UENF), merge the snapshot of the NF (e.g., UENF) into its own snapshot, and then deliver it to PF(200, LLM / GenAI) using snapshot-based logic.

[0269] At this time, other NFs (e.g., RANF, Core NF) may exclude duplicate content (e.g., same time, period, same context information, etc.) when merging into their own Snapshot based on the review.

[0270] With reference to FIGS. 12 and 13, when the PF (200, LLM / GenAI) updates an existing PI as it proceeds with further training of the LLM / GenAI, it can send the updated PI to each NF (e.g., UENF, RANF, Core NF) that subscribes to the updated PI (using a notification message, etc.).

[0271] At this time, when the PF (200, LLM / GenAI) transmits the PI updated this time to each NF (e.g., UENF, RANF, Core NF), it may transmit a PI selected or processed by taking into account the capability of the corresponding NF (in particular, LLM / GenAI Client).

[0272] In this way, each NF (e.g., UENF, RANF, Core NF) immediately receives the PI updated by additional training using a snapshot in PF (200, LLM / GenAI) and stores it in the internal storage, so that when a specific related event occurs, the operation of the Procedure (PI) already existing in the internal storage can be performed without a request for the PI (Perform Procedure).

[0273] FIG. 14 illustrates an example of the process / logic for storing a snapshot in the present invention.

[0274] In each NF (100, e.g., UENF, RANF, Core NF), a processor (not shown, e.g., LLM / GenAI Client) can determine the target information to be configured as a snapshot when the configuration of specific information, i.e., a snapshot, is initiated, determine whether a comparison with a previous snapshot is necessary and how to store the determined target information, and store the determined target information according to the determined comparison necessity and how to store it, thereby configuring and storing the snapshot.

[0275] At this time, in each NF (100, e.g., UENF, RANF, Core NF), the processor (not shown, e.g., LLM / GenAI Client) may initiate the configuration of a snapshot when a message that cannot be processed by the call processing logic within the NF (100, e.g., UENF, RANF, Core NF) is received or when a preset period is reached.

[0276] In addition, in each NF (100, e.g., UENF, RANF, Core NF), the processor (not shown, e.g., LLM / GenAI Client) may directly or indirectly transmit the snapshot configured this time to the PF (200, LLM / GenAI) when a significant change occurs compared to the previous snapshot transmitted within a set time range or when a set period is reached.

[0277] Referring to FIG. 14, according to the present invention, NF (100) can initiate the configuration of a snapshot at the time of a specific event, such as when an external request is received, when a message that cannot be processed internally is received, or when a preset period is reached (S10).

[0278] According to the present invention, when NF (100) initiates the configuration of a snapshot, it can determine target information to be configured as a snapshot (S20).

[0279] For example, the target information to be configured as a snapshot may be a range of information at the system / terminal level, or a range determined through a decision procedure that selects some of the system / terminal-related information.

[0280] In this regard, FIG. 16 shows examples of target information that can be configured as a snapshot in the present invention.

[0281] That is, in the present invention, any information that can be converted into data in a network can be the target information that can be configured as a snapshot.

[0282] As described in FIG. 16, in the present invention, a snapshot may be composed of detailed status information (e.g., power consumption, software patch, software version, etc.) and time information (or period information), etc. of a UE (target terminal) and NF (target system) associated with the event, depending on the event that initiates the configuration of the snapshot.

[0283] In addition, in the present invention, when storing / transmitting a snapshot to PF(200, LLM / GenAI), it may be stored / transmitted by changing it into a form (generalized) that supports general call processing.

[0284] According to the present invention, NF (100) can determine whether a comparison with a previous snapshot is necessary and how to store it after determining the target information to be configured as a snapshot (S20).

[0285] The process of comparing the target information to be configured as a snapshot with the previous snapshot can be complex in some cases and is a procedure that consumes various resources such as time and computing power; therefore, whether or not to perform the comparison process can have a significant impact on the overall process.

[0286] Accordingly, for example, NF (100) may determine that, based on the snapshot target information determined in step S20, a comparison with the previous snapshot is unnecessary when the current snapshot needs to be saved in a clean state (e.g., initial, explicit request, etc.) (S30→S50).

[0287] NF (100) determines that if the current snapshot is saved in accordance with a request for execution such as periodic backup, it is necessary to compare the target information of the snapshot determined in step S20 with the previous snapshot (e.g., (ID:xxx)) (S30→S40), and through comparison, only the changed information can be identified (S40).

[0288] According to the present invention, NF (100) can store information (e.g., the entire target information, or the part of the information that has changed from the previous one based on the comparison, or the part of the selected information) based on the comparison determination of the comparison necessity and storage method among the target information of the snapshot determined in step S20, and can configure and store a snapshot according to the event of the start of the current snapshot configuration (S50).

[0289] At this time, according to the present invention, NF (100) may store the snapshot only in internal storage, or may store it together in external storage (e.g., PF (200, LLM / GenAI)).

[0290] At this time, according to the present invention, NF (100) can perform an operation of storing the snapshot according to the storage method when storing the snapshot, and then storing the snapshot in an optimal way so that it can be delivered to the system / terminal.

[0291] To explain a specific embodiment, the present invention aims to manage and optimize internal / external storage and is characterized by storage / management that takes into account the characteristics of the stored snapshot information and the specificity of the method for requesting and providing subsequent snapshot information.

[0292] Snapshots stored in internal / external storage may be stored continuously in the form of a common structure or a substructure of a specific structure after the first time, and may have a common tendency of the information being stored.

[0293] In addition, snapshots stored in internal / external storage are storage requests for the entire network, and detailed information can be categorized by specific subscriber, terminal, or system units; the categorized information may share some similarities with one another or may have completely different forms.

[0294] Accordingly, in the present invention, when configuring and storing a snapshot in each NF (100), a judgment can be made to store it in an optimal manner that considers the form of the snapshot information having such diversity, and storage / management can be performed according to the judgment.

[0295] And, according to the present invention, NF (100) can utilize its own snapshot stored in internal storage when needed.

[0296] As an example, NF (100) enables additional training of LLM / GenAI using snapshots in PF (200, LLM / GenAI) by utilizing its snapshot in snapshot-based logic and passing it to PF (200, LLM / GenAI).

[0297] As another example, NF (100) may create a Prompt requesting a Procedure (PI) by including or reflecting its own snapshot when creating the Prompt and sending the PI request including it to PF (200), thereby utilizing it in the call processing procedure logic.

[0298] In addition, in the present invention, a method of requesting / providing snapshot information stored in each NF (100) may also be possible.

[0299] For example, one can assume a Consumer who requests additional analysis of the network from an individual system, NWDAF, AF, or terminal based on snapshot information collected by storing or receiving.

[0300] In this case, since the Consumer does not know which snapshot information is stored in the repository, it can send a snapshot request that includes conditions regarding the expected snapshot information.

[0301] For example, a snapshot request can be delivered in a form that specifies the following items.

[0302] · Delivery frequency: One-time (or periodic or upon reaching a specific threshold, etc.)

[0303] · Delivery Method: Direct notification (http) - Control Plane, AI Plane (dedicated path for transmitting and receiving source data for AI processing, new), etc.

[0304] · Target: Specific systems; SystemXYZ, SystemABC (or specific terminals and associated systems; XYZ, ABC, ...)

[0305] ·Time: YYYY.MM.DD HH:MM:SS:msec, time-window: 1 hour

[0306] ·Payload data: Signalling size, frequency, distributions, trend

[0307] An NF (100) that receives a snapshot request checks for a snapshot that meets the conditions in an internal / external storage to deliver the requested snapshot information according to the snapshot request, and if necessary, may obtain the necessary snapshot through a query method via another NF. The NF (100) can verify / obtain a snapshot that meets the snapshot request and deliver it to the Consumer.

[0308] Meanwhile, Fig. 15 illustrates the delivery and subsequent procedures of snapshot-based logic according to the snapshot processing method proposed in the present invention.

[0309] In the present invention, each NF (100) may undergo a registration process with PF (200) at initialization.

[0310] For example, NF (100) and PF (200) can transmit mutual system configuration information (e.g., HW configuration, Capa, interconnection I / F information, etc.), and PF (200) can transmit (assign) basic information (PI) regarding the supported call processing procedure (Procedure) of each NF (100) by considering existing network status information.

[0311] Afterwards, each NF (100) can perform its own call processing based on the received information or its own internal judgment.

[0312] At this time, NF (100) can determine whether a message / signal / data that cannot be processed by the call processing logic inside NF is received, whether there is a significant change in the state of NF (e.g., overload, change in external connection state, etc.), or whether there is a need to optimize the call processing logic inside NF (e.g., reaching a cycle).

[0313] NF (100) can configure necessary snapshot information according to the next Case based on the judgment of tactics and additionally transmit it to PF (200).

[0314] For example, NF (100) can determine Case 1 when it receives a received message / signal / data that cannot be parsed or a data type that has not been processed by the existing call processing logic, and can construct a snapshot including the related unprocessable message / signal / data, the processing target, and the processing logic / range of the current NF, and transmit it to PF (200).

[0315] Alternatively, NF (100) may determine Case 2 and deliver the latest snapshot to PF (200) if there is a significant change in the latest snapshot currently configured / created compared to the snapshot most recently / last delivered to PF (200).

[0316] For example, NF (100) can determine Case 2 based on changes in the resource usage rate of NF (100) (e.g., above threshold), changes in the status of existing linked NFs (e.g., failure), and trends in the call processing operations of NF (100) (e.g., statistics on processing results including success / failure).

[0317] Alternatively, NF (100) may determine Case 3 when a period set for the purpose of synchronization to minimize discontinuous changes in the network state is reached, and may construct a snapshot containing information according to Case 3 and transmit it to PF (200). In this case, the transmitted snapshot may be used for additional training of LLM / GenAI in PF (200, LLM / GenAI).

[0318] As explained above, according to the automatic call processing technology based on the generative language model (e.g., LLM / GenAI) of the present invention (in particular, the snapshot processing method of feature 2), a specific technical configuration is realized to enable more sophisticated and optimized call processing in each NF (e.g., UENF, RANF, Core NF) by optimizing the generative language model (e.g., LLM / GenAI) for actual commercial network conditions and optimizing the input (Prompt) through the configuration and transmission / storage of snapshots, thereby enabling the performance of necessary call processing in each NF (e.g., UENF, RANF, Core NF) based on the generative language model (e.g., LLM / GenAI).

[0319] Accordingly, according to the present invention, call processing information can be shared more efficiently between terminals and networks through snapshots, network speed and signaling delay can be improved, and the performance and latency of generative language models (e.g., LLM / GenAI) can be enhanced.

[0320] According to the various embodiments of the above tactics, moving away from the existing NF unit implementation / structure that can only perform call processing of a given function through design, a specific technical configuration is realized that enables the necessary call processing to be performed in each NF (e.g., UENF, RANF, Core NF) based on a generative language model (e.g., LLM / GenAI) that generates / transmits call processing procedures to each NF (e.g., UENF, RANF, Core NF).

[0321] As a result, the present invention supports optimal call processing generated / transmitted in each NF (e.g., UENF, RANF, Core NF) and automatically supports improvements in operation, handling of anomalies and abnormalities in each NF, thereby enabling the effect of significantly improving the inefficiency limits from a TCO perspective compared to existing methods.

[0322] In addition, since the call processing that must be carried out during design / development in the NF implementation can be minimized in the present invention, the effect of efficiently processing the initial state of On-device and On-NF development and distribution can also be expected.

[0323] A method of operation of an NF device according to one embodiment of the present invention, and an automatic call processing technique based on a generative language model (e.g., LLM / GenAI), may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either individually or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the present invention, or they may be those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The above-described hardware device may be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.

[0324] Although the present invention has been described in detail with reference to various embodiments, the present invention is not limited to the embodiments described above, and the technical concept of the present invention extends to the scope in which various modifications or alterations are possible by anyone with ordinary knowledge in the technical field to which the present invention belongs, without departing from the gist of the present invention as claimed in the following claims.

Claims

1. In an NF (Network Function) device, Memory containing instructions; and An NF device characterized by including: a processor that receives, as Procedure Information (PI), a call processing procedure to be performed by the NF device from a specific NF that determines the call processing procedure between a UE (User Equipment) and a Core network by executing the above command.

2. In Paragraph 1, The above specific NF is, An NF device characterized by determining a call processing procedure that can be performed by the NF device based on a request or subscription of the NF device, and transmitting the PI to the NF device for performing the determined call processing procedure.

3. In Paragraph 1, The above specific NF is, An NF device characterized by determining a call processing procedure to be transmitted to the NF device based on a generative language model (LM).

4. In Paragraph 1, The above processor performs the next operation according to the received PI, and The following operation above is, Operation of exchanging and processing signaling messages between NFs in the above NF device, The operation of updating the Software, Firmware, Version, and Code of the above NF device, An NF device characterized by including an operation for a specific Management Procedure.

5. In Paragraph 3, The above processor is, If it is determined that a call processing procedure to be performed for a specific event is required, a prompt used as the input of the generative language model within the specific NF is configured based on the specific event, and a PI request based on the prompt is transmitted to the specific NF. An NF device characterized in that the above PI request requests a call processing procedure to be performed by the above NF device.

6. In Paragraph 5, The above Prompt is, It is configured to include information on the specific event mentioned above, status information of the NF device, and information on the call processing procedure to be received, or An NF device characterized by being configured to further include information on time or period related to the above Prompt configuration.

7. In Paragraph 5, The above processor is, Understanding or reinterpreting the text of the received PI to determine whether the next action according to the text of the PI can be performed, and An NF device characterized by, when it is determined that execution is impossible, reconstructing or reorganizing the Prompt after Reject Execution and sending a PI request to the specific NF again, performing only a part of the operation in the next operation according to the Text of the PI (Partial Execution), or re-selecting the function to request a PI from among multiple functions related to the generative language model of the specific NF and then configuring the Prompt to send a PI request.

8. In the method of operation of an NF (Network Function) device, A step of receiving a call processing procedure to be performed by the NF device as PI (Procedure Information) from a specific NF that determines the call processing procedure between the UE (User Equipment) and the Core network; A method of operating an NF device characterized by including the step of performing the next operation according to the received PI.

9. In Paragraph 8, The above specific NF determines a call processing procedure to be transmitted to the NF device based on a generative language model (LM); If the above NF device determines that a call processing procedure to be performed for a specific event is required, it further includes the step of configuring a Prompt used as an input to the generative language model within the specific NF based on the specific event, and transmitting a Prompt-based PI request to the specific NF; A method of operation of an NF device characterized in that the above PI request requests a call processing procedure to be performed by the NF device.

10. In Paragraph 8, The following operation above is, Operation of exchanging and processing signaling messages between NFs in the above NF device, The operation of updating the Software, Firmware, Version, and Code of the above NF device, A method of operation of an NF device characterized by including an operation for a specific Management Procedure.

11. A step of receiving a call processing procedure to be performed by the NF device as PI (Procedure Information) from a specific NF that is combined with the hardware of the NF (Network Function) device and determines the call processing procedure between the UE (User Equipment) and the Core network based on a Generative Language Model (LM); and A computer program characterized by being stored in a medium to execute a step of performing the next action according to the received PI.

Images