Wireless communication methods and communication equipment

Abstract

A wireless communication method and communication device are provided. The method includes: a first device sending a first request to one or more second devices, the first request being used to request the one or more second devices to train a model associated with a first service. In this embodiment, multiple second devices are introduced to participate in the model training process. Since different second devices are associated with data features of different dimensions, this embodiment can train the model based on multi-dimensional data features, thereby helping to improve the accuracy of the model training results. Furthermore, analyzing the performance of the communication device based on the trained model helps to improve the comprehensiveness and accuracy of the performance analysis results.

Description

[0001] This invention is a divisional application of Chinese Patent Application No. 202380099907.5, entitled "Method and Communication Device for Wireless Communication", which entered the Chinese national phase of PCT international patent application PCT / CN2023 / 130747 filed on November 9, 2023. Technical Field

[0002] This application relates to the field of communication technology, and more specifically, to a method and device for wireless communication. Background Technology

[0003] In communication systems, relevant technologies can analyze terminal devices based on single data features. For example, based on a horizontal federated learning architecture, the connection establishment performance of a terminal device can be analyzed according to connection establishment information. Therefore, how to perform performance analysis based on different dimensions of user data features to improve the comprehensiveness and accuracy of the analysis results is a problem that needs to be solved. Summary of the Invention

[0004] This application provides a method and apparatus for wireless communication. The various aspects covered in this application are described below.

[0005] In a first aspect, a wireless communication method is provided, comprising: a first device sending a first request to one or more second devices, the first request being used to request the one or more second devices to train a model associated with a first service.

[0006] In a second aspect, a wireless communication method is provided, comprising: a second device receiving a first request sent by a first device, the first request being used to request the second device to train a model associated with a first service.

[0007] Thirdly, a wireless communication method is provided, comprising: a first device sending a third request to one or more second devices, the third request being used to request the one or more second devices to perform model inference of a first service association.

[0008] Fourthly, a wireless communication method is provided, comprising: a second device receiving a third request sent by a first device, the third request being used to request the second device to perform model inference of a first service association.

[0009] Fifthly, a communication device is provided, the device being a first device, the device comprising: a first sending unit, configured to send a first request to one or more second devices, the first request being configured to request the one or more second devices to train a model associated with a first service.

[0010] In a sixth aspect, a communication device is provided, the device being a second device, the device comprising: a first receiving unit, configured to receive a first request sent by a first device, the first request being configured to request the second device to train a model associated with a first service.

[0011] In a seventh aspect, a communication device is provided, the device being a first device, the device comprising: a first sending unit, configured to send a third request to one or more second devices, the third request being configured to request the one or more second devices to perform model inference of a first service association.

[0012] Eighthly, a communication device is provided, the device being a second device, the device comprising: a receiving unit, configured to receive a third request sent by a first device, the third request being configured to request the second device to perform model inference of a first service association.

[0013] A ninth aspect provides a communication device, including a processor, a memory, and a communication interface, wherein the memory is used to store one or more computer programs, and the processor is used to invoke the computer programs in the memory to cause the terminal device to perform some or all of the steps in the method of the first or third aspect.

[0014] In a tenth aspect, a communication device is provided, including a processor, a memory, and a communication interface, wherein the memory is used to store one or more computer programs, and the processor is used to invoke the computer programs in the memory to cause the terminal device to perform some or all of the steps in the method of the second or fourth aspect.

[0015] Eleventhly, embodiments of this application provide a communication system including the aforementioned communication device. In another possible design, the system may further include other devices that interact with the communication device as described in the embodiments of this application.

[0016] In a twelfth aspect, embodiments of this application provide a computer-readable storage medium storing a computer program that causes a communication device to perform some or all of the steps in the methods described above.

[0017] In a thirteenth aspect, embodiments of this application provide a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program operable to cause a communication device to perform some or all of the steps of the methods described in the foregoing aspects. In some implementations, the computer program product may be a software installation package.

[0018] In a fourteenth aspect, embodiments of this application provide a chip including a memory and a processor, the processor being able to call and run a computer program from the memory to implement some or all of the steps described in the methods of the foregoing aspects.

[0019] In this embodiment, multiple second devices are introduced to participate in the model training process. Since different second devices are associated with data features of different dimensions, this embodiment can train the model based on multi-dimensional data features, thereby improving the accuracy of the model training results. Furthermore, analyzing the performance of the communication device based on the trained model helps improve the comprehensiveness and accuracy of the performance analysis results. Attached Figure Description

[0020] Figure 1 This is the wireless communication system used in the embodiments of this application.

[0021] Figure 2 This is a schematic diagram of the training process of vertical federated learning.

[0022] Figure 3 This is a schematic diagram of the reasoning process in vertical federated learning.

[0023] Figure 4 This is a diagram illustrating an application of a horizontal federated learning architecture in a communication system.

[0024] Figure 5 This is a schematic flowchart of a wireless communication method according to an embodiment of this application.

[0025] Figure 6 This is an example diagram illustrating the application of the vertical federated learning architecture of this application in a communication system.

[0026] Figure 7 This is a schematic flowchart of a wireless communication method according to another embodiment of this application.

[0027] Figure 8 This is a flowchart illustrating the model training method based on a vertical federated learning architecture according to an embodiment of this application.

[0028] Figure 9 This is a flowchart illustrating the model inference method based on a vertical federated learning architecture according to an embodiment of this application.

[0029] Figure 10 This is a schematic diagram of a communication device provided in an embodiment of this application.

[0030] Figure 11 This is a schematic diagram of a communication device provided in another embodiment of this application.

[0031] Figure 12This is a schematic diagram of a communication device provided in another embodiment of this application.

[0032] Figure 13 This is a schematic diagram of a communication device provided in another embodiment of this application.

[0033] Figure 14 This is a schematic structural diagram of a communication device according to an embodiment of this application. Detailed Implementation

[0034] The technical solutions in this application will now be described with reference to the accompanying drawings.

[0035] Communication system architecture The technical solutions of this application embodiment can be applied to various communication systems, such as: Global System for Mobile Communication (GSM) system, Code Division Multiple Access (CDMA) system, Wideband Code Division Multiple Access (WCDMA) system, General Packet Radio Service (GPRS), Long Term Evolution (LTE) system, Advanced Long Term Evolution (LTE-A) system, LTE Frequency Division Duplex (FDD) system, LTE Time Division Duplex (TDD) system, New Radio (NR) system, evolution system of NR system, LTE-based access to unlicensed spectrum (LTE-U) system, NR-based access to unlicensed spectrum (NR-U) system, non-terrestrial networks (NTN) system, terrestrial networks (TN) system, and Universal Mobile Telecommunications (UMT) system. The technologies provided in this application include UMTS (Underground Universe Telecommunications System), WLAN (Wireless Local Area Networks), WIFI (Wireless Fidelity), and 5G (Fifth Generation) systems. The technical solutions provided in this application can also be applied to other communication systems, such as future communication systems like sixth-generation mobile communication systems and satellite communication systems.

[0036] Traditional communication systems typically support a limited number of connections and are easy to implement. However, with the development of communication technology, mobile communication systems will not only support traditional communication but also, for example, device-to-device (D2D) communication, machine-to-machine (M2M) communication, machine-type communication (MTC), vehicle-to-vehicle (V2V) communication, or vehicle-to-everything (V2X) communication. The embodiments of this application can also be applied to these communication systems.

[0037] The communication system in this application embodiment can be applied to carrier aggregation (CA) scenarios, dual connectivity (DC) scenarios, and standalone (SA) network deployment scenarios.

[0038] The communication system in this application embodiment can be applied to unlicensed spectrum, which can also be considered as shared spectrum; or, the communication system in this application embodiment can also be applied to licensed spectrum, which can also be considered as dedicated spectrum.

[0039] A key feature of communication system architecture (such as 5G system architecture) is that it can be a service-oriented architecture, meaning that network elements (service providers) in the core network can provide specific services and make them available to other network elements (consumers) through well-defined application programming interfaces (APIs).

[0040] Figure 1An exemplary system architecture diagram of a wireless communication system applicable to embodiments of this application is shown. Taking a 5G system architecture as an example, the wireless communication system may include multiple network elements, nodes, or devices, such as terminal devices, access network (AN) devices, user plane function (UPF) network elements, access and mobility management function (AMF) network elements, session management function (SMF) network elements, policy control function (PCF) network elements, and application function (AF) network elements. The wireless communication system may also include a data network (DN), etc.

[0041] The functions of each part or network element involved in the wireless communication system in the 5G network are illustrated below.

[0042] Terminal equipment: Terminal equipment can also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station (MS), mobile terminal (MT), remote station, remote terminal, mobile device, user terminal, terminal, wireless communication equipment, user agent, or user device. In the embodiments of this application, the terminal equipment can be a device that provides voice and / or data connectivity to a user, and can be used to connect people, objects, and machines, such as handheld devices with wireless connectivity, vehicle-mounted devices, etc. The terminal devices in the embodiments of this application may be mobile phones, tablets, laptops, handheld computers, mobile internet devices (MIDs), wearable devices, vehicle devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical surgery, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, etc.

[0043] Access network equipment: Access network equipment provides network access functionality for authorized terminal devices in a specific area and can use transmission channels of different quality according to the terminal device's level and service requirements. Access network equipment manages radio resources, provides access services to terminal devices, and thus completes the forwarding of control signals and data between the terminal devices and the core network.

[0044] Access network equipment can be a device in a wireless network. Access network equipment can also be called radio access network (RAN) equipment or network equipment; for example, access network equipment can be a base station. In the embodiments of this application, access network equipment can refer to a radio access network (RAN) node (or device) that connects a terminal device to a wireless network. A base station can broadly encompass, or be replaced by, various names including: NodeB, evolved NodeB (eNB), next-generation NodeB (gNB), relay station, access point, transmitting and receiving point (TRP), transmitting point (TP), master MeNB, auxiliary SeNB, multi-mode radio (MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed unit (DU), positioning node, etc. A base station can be a macro base station, micro base station, relay node, donor node, or similar entities, or combinations thereof. A base station can also refer to a communication module, modem, or chip installed within the aforementioned equipment or apparatus. Base stations can also be mobile switching centers, devices that perform base station functions in device-to-device (D2D), vehicle-to-everything (V2X), and machine-to-machine (M2M) communications, network-side devices in 6G networks, and devices that perform base station functions in future communication systems. Base stations can support networks using the same or different access technologies. The embodiments of this application do not limit the specific technologies or device forms used in the access network equipment.

[0045] Base stations can be fixed or mobile. For example, a helicopter or drone can be configured to act as a mobile base station, and one or more cells can move depending on the location of the mobile base station. In other examples, a helicopter or drone can be configured as a device to communicate with another base station.

[0046] In some deployments, the access network device in this application embodiment may refer to a CU or a DU, or the access network device may include both a CU and a DU. The gNB may also include an AAU.

[0047] Access network equipment and terminal equipment can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on water; and they can also be deployed in the air on airplanes, balloons, and satellites. This application embodiment does not limit the scenario in which the access network equipment and terminal equipment are located.

[0048] UPF (User Plane Function) element: The UPF is a user plane function in the core network, responsible for forwarding and receiving user data (such as service data streams) from terminal devices. The UPF can connect to access network devices (such as base stations) and external data networks for data transmission. For example, the UPF can receive user data from the DN (Digital Network Node) and transmit it to the terminal device through the access network device; alternatively, the UPF can also receive user data from the terminal device through the access network device and then forward it to the DN. The transmission resources and scheduling functions providing services to the terminal device in the UPF are managed and controlled by the SMF (Service Provider Function). In some embodiments, the UPF can be divided into intermediate-UPF (I-UPF) and anchor-UPF (A-UPF). The I-UPF is connected to the access network, and the A-UPF is a session anchor UPF, also known as a PDU session anchor (PSA).

[0049] AMF (Active Mobile Element): The AMF is a mobility management function in the core network. It can be used to implement functions other than session management in the mobility management entity (MME), such as lawful interception or access authorization (or authentication). In some embodiments, in addition to performing mobility management on terminal devices, the AMF can also be responsible for forwarding session management-related messages between the terminal device and the SMF.

[0050] SMF Network Element: SMF is the session management function in the core network. It is mainly responsible for session management, allocation and management of Internet Protocol (IP) addresses for terminal devices, selection of endpoints for manageable user plane functions, policy control, or charging function interfaces, downlink data notification, and configuration of routing information for user plane functions.

[0051] PCF Network Element: The PCF is a policy management function in the core network, responsible for formulating policies related to mobility management, session management, and charging of terminal devices. Specifically, the PCF can provide policy rule information to control plane functional network elements (such as AMF, SMF, etc.) to manage and control the mobility management and session management of terminal devices.

[0052] AF (Action Center) network element: The AF primarily supports interaction with the 3rd Generation Partnership Project (3GPP) core network to provide services, such as influencing data routing decisions, policy control functions, or providing third-party services to the network side. In other words, the AF can primarily be used to convey application-side requests to the network side. In some embodiments, the AF can be an internal operator application, such as IP Multimedia Subsystem (IMS) technology. In some embodiments, the AF can be understood as a third-party server, such as an application server in the Internet, providing relevant service information, including providing the PCF (Physical Processing Function) with service-related Quality of Service (QoS) requirement information, and sending user plane data information of the service to the A-UPF (Application-Planner Filter). In some embodiments, the AF can also be a content provider (CP). In some embodiments, if the AF is an internal operator AF and is within the same trusted domain as other network functions (NFs), it can directly interact and access other NFs; if the AF is not within a trusted domain, it needs to access other NFs through other network elements (e.g., the NEF network element mentioned below).

[0053] DN: DN refers to a network that can be used to provide data transmission. DN can be a private network, such as a local area network (LAN), an external network not controlled by an operator, such as the Internet, or a proprietary network jointly deployed by operators, such as a network providing IMS services.

[0054] Optionally, the wireless communication system may also include other network elements such as unified data management (UDM) network elements, authentication server function (AUSF) network elements, network slice selection function (NSSF) network elements, network exposure function (NEF) network elements, network data analytics function (NWDAF) network elements, and network repository function (NRF) network elements. This application embodiment does not limit this.

[0055] The UDM (User DM) network element is the subscription database in the core network. It can be used to generate and store user subscription data in the network (e.g., 5G network), manage authentication data, and perform other functions. The UDM network element can support interaction with external third-party servers. The AUSF (Authorized User Default Server) network element can be used to receive AMF (Authorized User Default Server) requests for terminal device authentication, request keys from the UDM, and then forward the issued keys to the AMF for authentication processing. The NSSF (Network Slice Default Server) network element can be used for network slice selection.

[0056] The NEF (Network Element Provider) can manage the network data exposed by 5G network elements. External untrusted applications need to access data within the core network through the NEF to ensure the security of the 3GPP network. In some embodiments, the NEF can also provide functions such as external application QoS capability opening, event subscription, and AF request distribution. The NWDAF (Network Window Assist) can collect data from various network elements and network management systems in the core network for big data statistics, analysis, or intelligent data analysis to obtain network-side analysis results or network-side prediction data. This enables various network elements to more effectively control terminal devices based on the data analysis results.

[0057] The NSSF network element is the network slice selection function in the core network. Its supported functions include: selecting the set of network slice instances to serve the UE; determining the allowed network slice selection assistance information (NSSAI), and, when necessary, determining the mapping to the subscribed single-network slice selection assistance information (S-NSSAI); determining the configured NSSAI, and, when necessary, determining the mapping to the subscribed S-NSSAI; determining the set of AMFs that may be used to query the UE, or determining a list of candidate AMFs based on the configuration.

[0058] The AUSF network element can be used to receive AMF requests for terminal authentication, request a key from UDM, and then forward the issued key to AMF for authentication processing.

[0059] NEF (Network Element Framework) elements can be used for capability exposure, meaning that network capabilities can be exported to external networks based on NEF. External, untrusted applications can access core network data through NEF to ensure network security. NEF can provide functions such as QoS capability exposure for external applications, event subscription, and AF (Active Request) distribution.

[0060] NRF network elements can be used for core network element registration, management, and status monitoring, thereby achieving automated management of core network elements. When a core network element starts up, it must register with the NRF to provide services. Registration information may include, for example, the core network element's type, address, and service list.

[0061] NWDAF network elements can collect data from various network elements in the core network and network management systems, and perform big data statistics, analysis, or intelligent data analysis to obtain network-side analysis or prediction data, thereby assisting various network elements in more effectively controlling terminal device access based on the data analysis results.

[0062] It should be understood that the above-mentioned functional entities in the core network can also be referred to as network elements, and this application does not limit this. For example, a UPF entity can also be referred to as a UPF network element, and an AMF entity can also be referred to as an AMF network element, etc.

[0063] It should also be understood that in some embodiments, the xx functional entity or xx network element can also be directly abbreviated as xx. For example, the UPF entity (or UPF network element) can be abbreviated as UPF, and the AMF entity (or AMF network element) can be abbreviated as AMF. For ease of description, the xx (such as UPF, AMF, etc.) mentioned in the embodiments of this application can refer to the xx entity or xx network element, which will not be repeated hereafter.

[0064] Figure 1 In the communication system shown, various parts or functional entities can communicate with each other through interfaces. For example, a terminal device can connect to the AN via the Uu interface for access stratum (AS) communication, exchanging AS messages and radio data transmission; the terminal device can connect to the AMF via the N1 interface for non-access stratum (NAS) communication, exchanging NAS messages; the AN can connect to the AMF via the N2 interface to transmit radio bearer control information from the core network side to the AN; the UPF can transmit data with the AN via the N3 interface and with the DN via the N6 interface, etc. For interfaces connecting other parts or functional entities, please refer to [link to relevant documentation]. Figure 1 This will not be elaborated upon here.

[0065] It should be understood that the network architecture shown above is merely an illustrative example, and the network architecture applicable to the embodiments of this application is not limited thereto. Any network architecture capable of implementing the functions of the above-described functional entities is applicable to the embodiments of this application.

[0066] It should be understood that Figure 1 The access network devices, AMF, SMF, UPF, and PCF shown are merely names and do not limit the devices themselves. In 5G networks and other future networks, the entities corresponding to access network devices, AMF, SMF, UPF, and PCF may also have other names, and this application embodiment does not specifically limit them.

[0067] It should be understood that Figure 1 The interface names between the various functional entities shown are just examples. In specific implementations, the interface names between the various functional entities can also be other names, such as the interface names between functional entities in a 6G network. This application does not specifically limit this.

[0068] It should be understood that all or part of the functions of the communication device in this application can also be implemented by software functions running on hardware, or by virtualization functions instantiated on a platform (such as a cloud platform).

[0069] It should be understood that the network architecture described in the embodiments of this application is for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and does not constitute a limitation on the technical solutions provided in the embodiments of this application. Those skilled in the art will know that as network architectures evolve, the embodiments of this application can also be applied to similar technical problems.

[0070] In recent years, artificial intelligence (AI) research, represented by neural networks, has achieved remarkable results in many fields, and it will play an important role in people's production and life for a long time to come.

[0071] Federated learning (FL) Federated learning can include horizontal federated learning, vertical federated learning (VFL), and federated transfer learning. Federated learning is a new encrypted distributed machine learning paradigm that allows participants to collaborate on AI without leaving their local machines, achieving "knowledge sharing without data sharing" and improving the performance of their respective AI models.

[0072] The following is combined with Figure 2 and Figure 3 This section introduces the vertical federated learning architecture. See also... Figure 2 as well as Figure 3 As shown, vertical federated learning is generally applicable to federated learning scenarios consisting of participants (e.g., node A and node B) with the same sample space but different feature spaces on the dataset. Vertical federated learning can also be understood as federated learning based on feature partitioning.

[0073] Suppose nodes A and B want to collaboratively train a model (e.g., model A and / or model B). Due to user privacy and data security concerns, nodes A and B cannot directly exchange data. Therefore, a third-party coordinator C is needed. Coordinator C can be a semi-honest third party, independent of nodes A and B. Its main function is to assist nodes A and B in secure federated learning. It collects intermediate results from model A and model B during training, calculates gradients and loss values, and then forwards the results to nodes A and B so that they can train their respective models. The information received by coordinator C from nodes A and B is encrypted or obfuscated, so the raw data of each party is not exposed to the other. Furthermore, nodes A and B only receive model parameters relevant to the features they possess.

[0074] Figure 2 This is a schematic diagram illustrating the training process of vertical federated learning applicable to embodiments of this application. See also... Figure 2 As shown, the training process of vertical federated learning generally includes two parts: the first part is to align encrypted sample data with the same ID but distributed among different participants; the second part is to train the model based on the aligned encrypted sample data.

[0075] The first part involves encrypted sample data alignment. Since the users corresponding to the sample data in node A and node B may be different, the system can use an encrypted user ID alignment technique to ensure that nodes A and B can align common users without exposing their original data. During encrypted sample alignment, the system will not expose users belonging to any particular node.

[0076] The second part involves training the model based on aligned encrypted sample data. This process may include steps S1 through S4.

[0077] See Figure 2 As shown in (b), in step S1, coordinator C sends the public key to nodes A and B. This public key is used to encrypt the sample data to be transmitted. In some implementations, homomorphic encryption can be used. That is, homomorphic encryption of two sample data m1 and m2 is equal to the homomorphic encryption of m1 plus the homomorphic encryption of m2. Homomorphic encryption of sample m multiplied by a constant is equal to the homomorphic encryption of that sample multiplied by the constant.

[0078] Step S2: Node A and Node B exchange intermediate results. Typically, the party possessing the sample labels is the active party and the demand party, such as Node B. Node A is the data provider (i.e., the passive party), and this node does not possess the sample data labels. Node A and Node B each use their local data to perform calculations and obtain the model's intermediate results. Node A encrypts the intermediate results and sends them to Node B. Node B, based on its own labels, Node A's model output results, and Node B's model output results, calculates the overall output error of the model and encrypts the output error before sending it to Node A.

[0079] Step S3: Calculate the gradient and loss. Nodes A and B calculate their respective encrypted gradients based on the output error, add masks, and send them to coordinator C. Additionally, node B can calculate the loss and send its encrypted loss to coordinator C.

[0080] Step S4: Update the model. Coordinator C can decrypt the gradient and loss, and send the results back to node A and node B respectively. After removing the mask, nodes A and B can update the model based on the gradient and loss.

[0081] Figure 3 This is a schematic diagram illustrating the inference process of vertical federated learning to which this application's embodiments apply. See also... Figure 3 As shown, the reasoning process may include steps S310 to S340.

[0082] In step S310, the coordinator C can send model inference requests to both node A and node B, requesting them to perform model inference. These model inference requests can also specify the model that nodes A and B should use.

[0083] In step S320, node A and node B input local data into their respective deployed models to perform model inference and obtain intermediate results.

[0084] In step S330, node A and node B encrypt their respective intermediate results and transmit the encrypted intermediate results to coordinator C.

[0085] In step S340, coordinator C aggregates the intermediate results sent by nodes A and B, and inputs the aggregated intermediate results into the local model for model inference, obtaining inference result 1. Since the intermediate results are encrypted and sent by nodes A and B respectively, coordinator C decrypts inference result 1 after obtaining it, obtaining inference result 2.

[0086] In step S350, coordinator C sends inference result 2 to node B.

[0087] Horizontal federated learning, also known as sample-based federated learning, can be applied to scenarios where the datasets of the various participants in federated learning have the same feature space but different sample spaces.

[0088] Related technologies have introduced a horizontal federated architecture into communication systems to improve system performance. The following analysis uses the example of NWDAF collecting connection establishment data from terminal devices to analyze the connection establishment performance of terminal devices within a specific area of ​​interest (AOI). Figure 4 This paper provides an exemplary introduction to the application of horizontal federation architecture in communication systems.

[0089] To analyze the connection establishment performance of terminal devices, NWDAF needs to collect connection establishment information from multiple terminal devices located at the SMF of the AOI. See also Figure 4 The SMFs located within this AOI can include SMF1 to SMFn, and each SMF can serve different terminal devices. That is, each SMF can collect and provide connection establishment information for different terminal devices; for example, SMF1 to SMFn can respectively provide connection establishment information for UE1 to UEn. The SMF network element deploys an NWDAF client, or in other words, the SMF network element and the NWDAF client are jointly configured.

[0090] The same network elements ensure that the feature dimensions of the collected data samples are the same. Different terminal devices provide different data samples. Therefore, the above architecture meets the characteristics of horizontal federation, can collect data from terminal devices from different manufacturers' SMFs, and ensures that local data is not directly shared, thus meeting the requirements of data privacy.

[0091] Figure 4 The architecture shown can involve both the initiator of model training / inference and the participants in model training / inference. The initiator could be, for example, NWDAF, and the participants could be, for example, SMF1~SMFn. Figure 4 The model training process for the architecture shown may include the following steps S410 to S440 (partial steps) Figure 4 (Not shown in the image).

[0092] In step S410, NWDAF initiates model training. Simultaneously, NWDAF distributes the model to the participants in the model training process.

[0093] In step S420, each SMF trains the initial model based on local data, such as connection establishment information of the terminal device.

[0094] In step S430, each SMF sends the training results to the NWDAF, the initiator of model training.

[0095] In step S440, NWDAF aggregates the model training results of each SMF and updates the global model.

[0096] based on Figure 4 The model inference process of the architecture shown may include the following steps S450 to S490 (partial steps) Figure 4 (Not shown in the image).

[0097] In step S450, NWDAF initiates model inference. Simultaneously, NWDAF distributes the global model to the participants in the model inference process.

[0098] In step S460, after receiving the global model, each SMF can perform model inference based on local data.

[0099] In step S470, each SMF sends the inference result to the NWDAF, the initiator of the model inference.

[0100] In step S480, NWDAF obtains the final model inference result based on the model inference result of each SMF.

[0101] The aforementioned horizontal federated architecture can only be applied to single network elements, such as scenarios where all elements are SMFs or UPFs, to ensure data feature consistency. However, in many use cases, the data to be collected spans different domains. Collecting this data is crucial for a more comprehensive analysis of the terminal device's characteristics across different dimensions. Here, "domain" can refer to UE, RAN, 5GC, OAM, and application. Different NFs within the 5GC can also be considered different domains.

[0102] User data is typically distributed across various nodes, including terminals, access network equipment, core network equipment, and third-party application servers (usually referring to servers that utilize the operator's network but whose services are provided by a third party outside the operator, also known as "OTT (over-the-top) application servers"). For example, user data can be distributed across... Figure 1 The diagram shows multiple nodes and network elements in the communication system network architecture. Therefore, how to conduct a comprehensive analysis of users based on their different dimensions of characteristics is a problem that needs to be solved.

[0103] To address the aforementioned issues, embodiments of this application provide a method for training a model based on user features across different dimensions (such as user features corresponding to multiple second devices), which helps improve the accuracy of the model training results. It should be understood that this model can be an AI model; of course, it can also be a new AI model introduced into future communication systems. Alternatively, the model can be an ML model.

[0104] The following is combined with Figure 5 The methods provided in the embodiments of this application will be described. Figure 5 The method shown involves the initiator of model training, such as a first device, and the participants in model training, such as one or more second devices. Figure 5 The method shown can be applied to the communication systems mentioned above, and can also be applied to future communication systems.

[0105] by Figure 5 The method shown is applied to Figure 1 Taking the communication system shown as an example, the first device can be Figure 1 In the NWDAF network element, the second device can be Figure 1 This includes SMF network elements, UPF network elements, AMF network elements, etc. It should be noted that one or more second devices can serve the same group of users, thus ensuring the consistency of the sample spaces across different domains.

[0106] Figure 5 The method shown may include step S510. The method provided in the embodiments of this application will be described below from the perspective of the interaction between the first device and one or more second devices.

[0107] In step S510, the first device sends a first request to one or more second devices. Alternatively, one or more second devices receive the first request sent by the first device.

[0108] Step S510 can also be replaced by: the second device receiving the first request sent by the first device. It should be noted that the second device mentioned here can be any one of one or more second devices participating in model training.

[0109] The aforementioned first request can be used to request one or more second devices to train a model associated with the first service. In some embodiments, the first request may include a model training identifier, which may indicate that the first request is used to request one or more second devices to train the model. The model associated with the first service may also be referred to as the model corresponding to the first service, such as a model used to perform the first service, or a model whose inference results can be used to process the first service. The first service mentioned here may be, for example, a signaling storm analysis service, or a connection establishment information analysis service, etc.

[0110] The first request can also be called a model training request. The first request may include one or more of the following: an initial model; data features associated with each of the one or more second devices; the training time of the model associated with the first service; the geographical region associated with the data features; the service identifier of the first service; the training task identifier; and the type of model training.

[0111] In some embodiments, the first request may include an initial model, meaning that an initial model can be issued through the first request. The initial model can be used for model training. Alternatively, the initial model is associated with the model to be trained, or the initial model is an initial model associated with the first business function.

[0112] In some embodiments, the initial models of multiple nodes participating in the model training for the first service association, such as multiple second devices, may be the same or different. The initial model corresponding to each second device can be determined based on usage requirements, and this application does not impose any limitations on this.

[0113] In some embodiments, the first request may indicate the identifier of the initial model. If the initial model is stored locally on the device participating in model training, or on a node with storage capabilities in the communication system, then before model training, the device participating in model training, such as one or more second devices, may look up the corresponding initial model locally based on the model identifier, or download the corresponding initial model from a node with storage capabilities.

[0114] In some embodiments, the first request may include data features required for model training. Data associated with a data feature may refer to a type of user data. For example, the connection establishment information of the terminal device, the user plane data packet information of the terminal device, the mobility management information of the terminal device, the session management information of the terminal device, the policy information associated with the terminal device, and the subscription information associated with the terminal device, as mentioned above, are each data associated with a data feature. Furthermore, data features may also include data sample identifiers, such as terminal device IDs, and the correspondence between samples and data features.

[0115] To generate a model that better meets user needs, model training requires more multi-dimensional data from the user. Therefore, the initial request can include multiple data features needed for model training to improve model accuracy. One type of data feature mentioned here can correspond to one dimension of data used in model training.

[0116] As mentioned earlier, a single device, such as a single network element, can often only provide data for a single feature. Correspondingly, the multiple data features required for model training often need to be provided by multiple participants, such as multiple second devices. Therefore, the first request can also include data features associated with each of the one or more second devices, or in other words, the first request can indicate the association between multiple data features and multiple second devices, that is, the first request can indicate the data features that each second device needs to collect.

[0117] In some embodiments, the first request may include training time for a model associated with a first service, requesting one or more second devices to train the model within that training time. This training time facilitates service scheduling by the one or more second devices and also enables multiple second devices to collaboratively train the model.

[0118] In some embodiments, the first request may include a geographic region associated with data features. Terminal devices may exhibit different features in different geographic regions, and different services or different model training tasks for the same type of service may require acquiring features from different geographic regions. Therefore, including a geographic region associated with data features in the first request can indicate the association between the data features to be collected by the second device and the geographic region.

[0119] In some embodiments, the first request may include a service identifier for the first service. When multiple services are being trained, the service identifier can be used to distinguish the model training information corresponding to different services.

[0120] A single service may include multiple model training tasks. For example, a signaling storm service may include two model training tasks, each targeting different model use cases. Therefore, to further differentiate between different model training tasks within the same service, the first request may include a training task identifier.

[0121] In some embodiments, the first request may also include the type of model training, such as model training based on a horizontal federated learning architecture.

[0122] In some embodiments, Figure 5 The method shown may also include step S520. In step S520, the first device receives one or more training results sent by one or more second devices, or in other words, the second device sends training results to the first device.

[0123] The aforementioned training results may be the training results of one or more second devices training the model associated with the first business based on local data. Local data may, for example, be data collected by one or more second devices based on data features.

[0124] In some embodiments, Figure 5 The method shown may also include step S530. In step S530, the first device sends the loss parameters to one or more second devices, or in other words, the second device receives the loss parameters sent by the first device.

[0125] The aforementioned loss parameters, also known as loss functions, reflect the model's performance. Loss parameters can be determined based on one or more training results. For example, a first device can aggregate one or more training results and calculate the aforementioned loss function based on the labels stored in the first device.

[0126] In some embodiments, the loss parameters can be used by one or more second devices to update the model associated with the first service. For example, one or more second devices can determine their respective loss states based on the loss function, calculate gradients, and then update the model based on the loss states and gradient information. Typically, before updating the model, the first device can send the calculated results of the loss parameters to one or more second devices, or the second devices can receive the calculated results of the loss parameters sent by the first device.

[0127] It should be noted that one or more training results may also include training task identifiers, so that the first device can identify training results associated with the same training task, thereby helping the first device to process training results of the same task, such as training result aggregation, loss parameter calculation, etc.

[0128] After the model is updated on one or more second devices, model training can continue based on local data until a preset condition is met or the model converges, at which point training is complete. The preset condition can be, for example, associated with a preset number of training epochs. Whether the model has converged can be determined based on the loss parameter, such as the relationship between the loss parameter and a preset threshold.

[0129] In some embodiments, the first device may send first indication information to one or more second devices. The first indication information is used to indicate that the training of the model associated with the first service has been completed. For example, the first indication information may be determined based on a preset number of model training epochs and / or a loss parameter. Alternatively, when the number of model training epochs reaches a preset number, or when the loss parameter is less than or equal to a preset threshold, the first device sends the first indication information to one or more second devices.

[0130] In some embodiments, a preset number of training rounds can be included in the first request to save signaling overhead.

[0131] Before model training can begin, the initiator of the training, such as the first device, usually needs to identify or discover the participants in the training, namely one or more second devices.

[0132] In some embodiments, before sending a first request to one or more second devices, the first device may send a second request to a third device. The third device may be a device storing device information, such as an NRF network element. This device information may include, for example, device capability information, such as information related to the device's model inference / training capabilities.

[0133] The aforementioned second request can be used to discover a fourth device, which is a device with model training capability, specifically the ability to train a model associated with the first business.

[0134] In some embodiments, the second request may include capability requirements for model training associated with the first business to determine devices that can participate in model training associated with the first business. For example, the second request may include the type of model training, the business identifier of the first business, the data characteristics associated with the first business, and one or more of the geographic regions associated with those data characteristics. The type of model training may include model training based on lateral federated learning.

[0135] In some embodiments, the second request may include the training time of the model associated with the first service. Since a device with the ability to train the model associated with the first service may not be able to allocate resources for model training when communication services are busy, the training time of the model associated with the first service helps to improve the reliability of determining the participants in this model training.

[0136] In some embodiments, the third device may indicate to the first device a device that conforms to the second request, such as a fourth device. For example, the first device may receive a response message sent by the third device, which can be used to indicate the fourth device. This response message is a response to the second request.

[0137] In some embodiments, the response message to the second request may indicate the fourth device via a device identifier or device address. In some cases, the fourth device may implement model training functionality by deploying a client that supports model inference / training, or by interacting with a client that has model inference / training capabilities. Therefore, the response message may also indicate the address of the client associated with the fourth device. The client associated with the fourth device may refer to the aforementioned client deployed on the fourth device that supports model inference / training, or a client that can interact with the fourth device and has model inference / training capabilities.

[0138] In some embodiments, the first device may determine the devices participating in this model training based on the discovered fourth device, that is, the devices participating in the model training of the first business association. For example, one or more second devices may be some or all of the fourth devices.

[0139] There are several methods for determining one or more second devices. For example, the first device can randomly determine one or more second devices from the fourth device. Alternatively, the first device can determine one or more second devices based on device information of the fourth device, such as local information.

[0140] As an example, the first device may send a model training execution request to the fourth device. The model training execution request may include one or more of the following: the type of model training; the business identifier of the first business; the data features associated with the first business; the training time of the model associated with the first business; and the geographical region associated with the data features.

[0141] The fourth device can determine whether it can complete model training under the execution request requirements based on local information. For example, the fourth device can determine whether it can complete model training under the aforementioned execution request requirements based on one or more of the following: the number of services currently running or within the model training time, resource usage, remaining resources, the amount of local data, and the corresponding data characteristics. As an example, if the fourth device has a large number of services running or limited remaining resources within the model training time, it cannot complete model training. As another example, if the amount of local data corresponding to the data characteristics associated with the aforementioned execution request is small, the fourth device cannot complete model training.

[0142] As an example, the fourth device can send a response message to the first device regarding the aforementioned execution request. This response message can indicate whether the fourth device supports model training associated with the current first service, or in other words, whether the fourth device is capable of participating in model training associated with the current first service. This response message can be determined based on the assessment result of whether the fourth device can complete model training.

[0143] As another example, if the fourth device can complete the model training required by the execution request, then the fourth device sends a response message to the first device, indicating that the fourth device can complete the model training required by the execution request. If the fourth device cannot complete the model training required by the execution request, then the fourth device does not send a response message to the first device, thereby helping to save signaling resources.

[0144] In some embodiments, the second device may register its capability information with the third device. This capability information may be information associated with the second device's model inference / training capabilities. For example, the capability information may include one or more of the following: whether it has model training / inference capabilities; the types of model training / inference supported; the business operations that support model training / inference; the characteristics of the data it supports collecting; the time periods during which it supports model training / inference; and the geographical regions where it supports model training / inference.

[0145] In some embodiments, the second device can be a device with AI capabilities. For example, the second device can be a gNB or a terminal device, meaning the second device has model inference / training capabilities. Since the OAM has a management data analytics function (MDAF), this network element also has AI capabilities. That is, the second device can also be an OAM. Alternatively, the second device can be a device that has deployed a client supporting model inference / training. As an example, the second device can be a network element that has deployed a client supporting model inference / training, such as an NWDAF client. Another example is a device that supports interaction with a client that has model inference / training capabilities. As an example, the second device can be an NF that has an internal interface with the NWDAF client; that is, the second device is an NF that can interact with the NWDAF client through an internal interface.

[0146] In some embodiments, the first device may store device information of one or more second devices for later use. For example, the first device may also store device information of one or more second devices, as well as a service identifier of a first service associated with one or more second devices. The stored information of one or more second devices, and / or the service identifier of the first service associated with one or more second devices, may be used for model inference associated with the first service, or for other model training tasks associated with the first service.

[0147] As discussed earlier, training a model by combining data from different features of the same user across various nodes can improve model performance. However, multi-node, multi-domain data sharing poses a significant challenge to data privacy. Figure 1 Taking the communication system network architecture shown as an example, this communication system contains multiple nodes and network elements, each with its own specific functions and data / information. However, in actual deployment, different network elements may come from different manufacturers, and for security and privacy reasons, different manufacturers cannot share their data. Therefore, the initiator of model training, such as the analysis network element in the network, cannot collect user data with different characteristics, thus making it impossible to train the model based on user data with different characteristics.

[0148] The vertical federated learning architecture mentioned above can enable artificial intelligence systems to use local data from multiple nodes efficiently and accurately, while meeting data privacy, security and regulatory requirements. It breaks down data silos and achieves cross-domain multi-node data sharing while ensuring privacy and security.

[0149] Therefore, to address the data security issue during multi-node data sharing, this application introduces a vertical federated learning architecture into the communication system. For example, the aforementioned first request can be used to request one or more second devices to perform model training based on vertical federated learning for the model associated with the first service. Based on this, the types of model training mentioned above can also include model training based on vertical federated learning.

[0150] Figure 6 This is a diagram illustrating an application of a vertical federated learning architecture in a communication system. Figure 6 The model training is initiated by the NWDAF server, and the participants in the model training include UE, gNB, OAM network element, AMF network element, SMF network element, UPF network element, AF network element, etc.

[0151] based on Figure 6 During model training in the vertical federated learning architecture shown, the NWDAF server, the initiator of the model, first determines the different network element nodes that need to participate in this model training. It should be noted that the participants selected by the initiator should serve the same set of terminal devices, which helps to ensure the consistency of the sample spaces in each domain.

[0152] In some embodiments, the NWDAF server can send the initial sub-model to each node, wherein the sub-model of each network element can be the same or different.

[0153] In some embodiments, the NWDAF server can indicate the data features associated with each node (i.e., the participating nodes in model training), that is, it can indicate the data features that need to be collected to each node.

[0154] In some embodiments, NWDAF can send a model training task ID to each node. After receiving the results from each node, the NWDAF server can identify the training results of the same model training task based on the model training task ID. Furthermore, the NWDAF server can aggregate the training results of the same model training task and calculate the loss.

[0155] In some embodiments, each node can collect local data based on the received initial model and the data features required to be collected as indicated by the NWDAF server. Each node can train the model based on the results of the local data collection and send the training results (also known as intermediate results of model training) to the NWDAF server.

[0156] In some embodiments, the NWDAF server can aggregate the training results from each node and calculate the loss function using its stored labels. Additionally, the NWDAF server can send the loss status, such as the calculated loss function result, to each node.

[0157] In some embodiments, each node can calculate gradient information based on the loss state and update the local model (such as the initial model mentioned above) based on the loss state and gradient information.

[0158] In some embodiments, multiple rounds of model training can be performed based on data from each node. Model training terminates upon model convergence or reaching a preset number of training rounds. Model convergence here can refer to the model's loss function falling below a preset threshold.

[0159] It should be noted that in 5G communication systems (such as...) Figure 1 In the communication system shown, each NFx can implement local AI capabilities by deploying an NWDAF client, thus acting as an initiator or participant in model training. The NWDAF client can be integrated into an NF or deployed independently. In the case of a standalone NWDAF client, the NF can interact with the NWDAF client through an internal interface. Furthermore, in 6G or future communication systems, nodes with AI capabilities from the outset, such as nodes with AI capabilities acquired through intrinsic intelligence, can also become initiators or participants in model training. This application does not impose any limitations on this.

[0160] The previous section introduced a method for model training based on multidimensional user data features. The following section will combine... Figure 7 Another embodiment of the wireless communication method provided in this application will be described. Figure 7The method shown helps improve the accuracy of model inference results by performing model inference based on the user's multidimensional data features.

[0161] Figure 7 The method shown can be applied to the communication systems mentioned above, and also to future communication systems. Figure 7 The method shown is applied to Figure 1 Taking the communication system shown as an example, the first device can be Figure 1 In the NWDAF network element, the second device can be Figure 1 The network elements include SMF, UPF, AMF, PCF, UDM, and terminal equipment.

[0162] Figure 7 The method shown involves the initiator of model inference, such as a first device, and the participants in model inference, such as one or more second devices. Figure 7 The method shown may include step S710. The method provided in the embodiments of this application will be described below from the perspective of the interaction between the first device and one or more second devices.

[0163] In step S710, the first device sends a third request to one or more second devices. Alternatively, one or more second devices receive the third request sent by the first device.

[0164] Step S710 can also be replaced by: the second device receiving the third request sent by the first device. It should be noted that the second device mentioned here can be any one of one or more second devices participating in model inference.

[0165] The aforementioned third request can be used to request one or more second devices to perform model inference associated with the first service. In some embodiments, the third request may include a model inference identifier, which may indicate that the third request is used to request one or more second devices to perform model inference. Here, the model inference associated with the first service may refer to the fact that the result of the model inference can be used to process the first service.

[0166] In some embodiments, the third request may include one or more of the following: a service identifier for the first service; an inference task identifier; data characteristics associated with each of the one or more second devices; a time period associated with the first service; a geographical region associated with the first service; a target terminal device associated with the first service; and a model identifier associated with the first service.

[0167] Before performing model inference, the second device needs to determine the target model, i.e. the model to be inferred.

[0168] In some embodiments, the second device may determine the target model based on a model identifier. For example, the third request may include a model identifier associated with the first service, and the second device may determine the target model based on the model identifier associated with the first service.

[0169] In some embodiments, the second device may determine the target model based on the identifier of the first service. For example, the third request may include the service identifier of the first service. Since there is usually a certain mapping relationship between the service identifier and the model identifier, the second device can determine the target model based on the service identifier of the first service and this mapping relationship. The mapping relationship between the service identifier and the model identifier may be stored locally on the second device or may be obtained by the second device from the first device; this application does not limit this.

[0170] The same service may correspond to different model inference tasks. For example, signaling storm analysis may include model inference tasks associated with signaling storms in the first time period and model inference tasks associated with signaling storms in the second time period. Therefore, in order to distinguish different model inference tasks for the same service, the third request may include an inference task identifier. Simultaneously, the inference task identifier can also be used to distinguish the inference results of different model inference tasks.

[0171] During model inference, different second devices need to collect local data with different data characteristics, or the same second device needs to collect local data with different data characteristics in different model inference tasks. Therefore, the third request may also include one or more data characteristics associated with the second device to indicate the type of local data that each second device needs to collect. Data associated with a data characteristic can refer to a type of user data. For example, the connection establishment information of the terminal device, the user plane data packet information of the terminal device, the mobility management information of the terminal device, the session management information of the terminal device, the policy information associated with the terminal device, and the subscription information associated with the terminal device mentioned above are each data associated with a data characteristic.

[0172] In some embodiments, the third request may further include data sample identifiers, such as terminal device IDs, and the correspondence between samples and data features. For example, the data sample identifiers included in the third request are UE1~UEn, meaning that the participants in the model inference need to collect data for UE1~UEn. As another example, the third request indicates data feature 1 corresponding to UE1~UEm, and data feature 2 corresponding to UEm+1~UEn, meaning that the participants in the model inference need to collect data associated with data feature 1 for UE1~UEm, and data associated with data feature 2 for UEm+1~UEn.

[0173] As an example, data features may include sample identifiers and the correspondence between samples and data features. That is, a third request can indicate data sample identifiers and / or the correspondence between samples and data features through data features.

[0174] In some embodiments, the third request may include one or more of the following: a time period associated with the first service, a geographical region associated with the first service, and a target terminal device associated with the first service. The target terminal device is the terminal device associated with the first service, such as the terminal device corresponding to the sample identifier described above.

[0175] In some embodiments, the first device may receive one or more inference results sent by one or more second devices, or in other words, the second devices may send inference results to the first device. The one or more inference results are the results of inference performed by one or more second devices on a model associated with the first service based on local data.

[0176] The aforementioned local data can be local data collected by one or more second devices based on data characteristics associated with the first service. Taking the first service as signaling storm analysis service as an example, the local data may include one or more of the following: mobility management information of the target terminal device; session management information of the target terminal device; user plane data packet information of the target terminal device; policy information associated with the target terminal device; and subscription information associated with the target terminal device. Wherein, the target terminal device is the terminal device associated with the first service, such as the terminal device corresponding to the sample identifier mentioned above.

[0177] The policy information associated with the aforementioned target terminal device may include, for example, the terminal device's routing policy, mobility management policy, session management policy, billing policy, and other related policies.

[0178] In some embodiments, the mobility management information of the target terminal device may be collected by an AMF network element as a second device. The mobility management information of the target terminal device may include, for example, one or more of the following: the number of terminal devices registering with the AMF and the number of signaling messages generated; the type of registration initiated, such as initial registration, periodic registration update, or mobility registration update; the number of terminal devices that successfully registered and the number of signaling messages generated; the number of terminal devices that failed to register and the number of signaling messages generated; and the number of terminal devices served by the AMF network element, such as the number of terminal devices supported by the AMF network element.

[0179] In some embodiments, the session management information of the target terminal device can be collected by the SMF network element as a second device. The session management information of the target terminal device may include one or more of the following: the number of terminal devices making session requests to the SMF network element and the number of signaling requests generated; the session request type, such as initial session establishment, session modification, session release, etc.; the number of terminal devices that successfully established sessions and the number of signaling requests generated; the number of terminal devices that failed to establish sessions and the number of signaling requests generated; and the number of terminal devices served by the SMF network element, such as the number of terminal devices supporting the service.

[0180] In some embodiments, the user plane data packet information of the target terminal device can be collected by the UPF network element as a second device. The user plane data packet information of the target terminal device includes one or more of the following: the number of data packets forwarded by the UPF network element; and the number of N4 session establishment requests made by the UPF network element and the SMF network element through interaction.

[0181] In some embodiments, the routing policy of the target terminal device can be collected by the target terminal device as a second device. The routing policy of the target terminal device can be used by the terminal device to determine whether an application can be associated with an established PDU session or whether it is necessary to trigger the establishment of a new PDU session.

[0182] In some embodiments, other policies of the target terminal device, such as mobility management policies, session management policies, and charging policies, can be collected by the PCF network element as a second device. Other policy information of the target terminal device may include the number of policies provided by the PCF network element and the amount of signaling generated.

[0183] In some embodiments, the subscription information associated with the target terminal device can be collected by a UDM network element as a second device. The subscription information associated with the target terminal device may include, for example, the number of signaling messages generated by the UDM network element interacting with other network elements, such as the number of signaling messages generated by any network element interacting with the UDM network element due to any process. As an example, the subscription information associated with the target terminal device may include the number of signaling messages for user subscription registration with the UDM network element, or the number of signaling messages generated by the AMF network element and SMF network element querying the UDM network element for subscription information in order to perform registration, session establishment, and other processes.

[0184] In some embodiments, the local data of the target terminal device may also include the location information of the terminal device, which can be used to determine whether the terminal device will enter or leave a certain area in the future.

[0185] It should be noted that the local data mentioned above can be local data generated within the geographical region associated with the first business and / or the time period associated with the first business.

[0186] In some embodiments, one or more inference results may include an inference task identifier, thereby facilitating the processing of inference results for the same inference task by the terminal device, such as inference result aggregation.

[0187] The first device can obtain analysis results related to the first service based on one or more inference results. In some embodiments, the first device can send the analysis results related to the first service to a fifth device. The fifth device mentioned here can be a device subscribed to the first service, such as a terminal device or one or more second devices.

[0188] The analysis results for the first service association can include various factors. The following section uses signaling storm as an example to illustrate the content of the analysis results for the first service association.

[0189] In some embodiments, the analysis results of the first service association may include signaling storm levels, such as low, medium, and high. Based on the signaling storm level in the analysis results, the signaling situation within a target time period can be understood, the timing for adjusting or updating the signaling-related strategies can be determined, and the adjustment scheme for the signaling-related strategies can be identified. For example, when the signaling storm level is medium or higher, the signaling-related strategies are adjusted or updated; when the signaling storm level is low, no adjustment is made. Furthermore, different strategy adjustment schemes can be adopted when the signaling storm level is different. For instance, when the signaling storm level is medium, only high-importance signaling requests are executed; when the signaling storm level is high, connection establishment is not performed, or registration requests are not accepted, etc.

[0190] In some embodiments, the analysis results of the first service association may include the node that generated the signaling storm, and / or the proportion of influence of the node that generated the signaling storm on the signaling storm, thereby helping to adjust the signaling association strategy in a targeted manner and improve the effectiveness of the strategy adjustment scheme. For example, when the node that generated the signaling storm is an AMF network element, the signaling association strategy can be adjusted for the AMF network element, such as not accepting registration requests to reduce the signaling quantity of the AMF network element, thereby helping to avoid the signaling storm. As another example, if the AMF network element has an 80% influence, the SMF network element has a 10% influence, and the UDM network element has a 10% influence in the signaling storm, then the signaling association strategy for the corresponding node can be adjusted according to the different influence proportions of the signaling storm. As an example, based on different influence proportions, the degree of strategy adjustment for each node can be determined. For example, for nodes with a large influence proportion, actions related to signaling association can be not executed, and for nodes with a small influence proportion, actions related to signaling association with lower importance can be not executed. As another example, signaling association strategy adjustments can be made for nodes with a large influence proportion, while no signaling association strategy adjustments can be made for nodes with a small influence proportion.

[0191] In some embodiments, the analysis results of the first service association may include the cause of the signaling storm. The cause of a signaling storm may include, for example, the movement of a large number of users. Because the movement of a large number of users leads to a large number of users simultaneously switching, registering, or establishing connections, network nodes receive a large number of signaling requests at the same time, thus generating a signaling storm.

[0192] In some embodiments, the analysis results of the first service association may include terminal devices that are affected by signaling storms. Based on the information of terminal devices that are affected by signaling storms, the signaling policy of those terminal devices can be adjusted, thereby helping to avoid the impact of signaling storms.

[0193] In some embodiments, the analysis results of the first business association may include the recovery time of the signaling storm, or the duration of the signaling storm, thereby helping to adjust the signaling strategy based on the duration of the signaling storm.

[0194] It should be noted that the analysis results of the first business association can include one or more of the above content.

[0195] The analysis results of the first service association can be used to determine the signaling association policies in the fifth device, such as the signaling policy adjustment scheme mentioned above, or the updated policy. Taking the fifth device as a terminal device as an example, the signaling association policy may include not initiating a registration request during a signaling storm, and / or not initiating a connection establishment. Taking the fifth device as a core network element as an example, the signaling association policy may include rejecting signaling requests during a signaling storm, and / or deploying a backup fifth device.

[0196] Once the signaling storm has subsided or recovered, the fifth device can restore or update the policy associated with the signaling to the policy in effect before the signaling storm occurred.

[0197] The embodiments of this application can adjust the strategy associated with the first service based on the reasoning results of the model associated with the first service. For example, based on the analysis results of the signaling storm service, the signaling association strategy can be adjusted, thereby helping to avoid network paralysis caused by the signaling storm and helping to avoid the impact of the signaling storm on the user experience.

[0198] In some embodiments, before performing model inference for the first service association, such as before the first device sends a third request to one or more second devices, the first device may receive first information sent by a fifth device. This first information can be used to subscribe to a first service, which is associated with the model inference of one or more second devices.

[0199] In some embodiments, the first information may include one or more of the following: a service identifier for the first service; an inference task identifier; a time period associated with the first service; a geographical region associated with the first service; a target terminal device associated with the first service; and a model identifier associated with the first service.

[0200] The first piece of information can indicate the target model to be used for model inference through the business identifier of the first service or the model identifier associated with the first service. The business identifier of the first service can have a mapping relationship with the target model.

[0201] In some embodiments, the target terminal device may be all terminal devices, a portion of terminal devices, or a specific terminal device within the geographical area associated with the first service. For example, the target terminal devices associated with the first service may be indicated by one or more of the following: a terminal device list, a terminal device group, and a terminal device identifier.

[0202] Upon receiving a third request, the first device needs to identify one or more second devices participating in the model inference for the first service association. For example, one or more second devices can be identified based on second information pre-stored by the first device. This second information can be information about devices that have historically participated in model training for the first service association. Alternatively, it can be determined based on the model training results for the first service association. As an example, if the first device does not have pre-stored second information, or if the first device has not initiated model training for the first service association, then the first device can train the model for the first service association and identify the first or more second devices based on the training results.

[0203] In some embodiments, the second device can be a device with AI capabilities. For example, the second device can be a gNB or a terminal device, meaning the second device has model inference / training capabilities. Since the OAM (Operational Data Analytics) function (MDAF) exists, this network element also has AI capabilities. That is, the second device can also be an OAM. Alternatively, the second device can be a device that has deployed a client supporting model inference / training. As an example, the second device can be a network element that has deployed a client supporting model inference / training, such as an NWDAF client. Another example is a device that supports interaction with a client that has model inference / training capabilities. As an example, the second device can be an NF (Network Function) that has an internal interface with the NWDAF client; that is, the second device is an NF that can interact with the NWDAF client through an internal interface.

[0204] The method provided in this application embodiment can achieve more comprehensive model reasoning through multi-dimensional data features (such as data associated with multiple data features provided by multiple second devices), thereby improving the accuracy of model reasoning results.

[0205] In some embodiments, the third request is used to request one or more second devices to perform model inference for the first business association based on a vertical federated learning architecture. That is, the model inference method for the first business association provided in this application embodiment can be a model inference method based on a vertical federated learning architecture. Since the vertical federated learning architecture allows artificial intelligence systems to efficiently and accurately use local data from multiple nodes while meeting data privacy, security, and regulatory requirements, model inference based on a vertical federated learning architecture can break down data isolation between different device vendors and help solve data security issues during multi-node data sharing.

[0206] It should be noted that the two wireless communication methods provided in this application embodiment can be used independently or in combination. That is, the model training method and the model inference method provided in this application embodiment can be used independently or in combination. For example, upon receiving a service subscription request, model training associated with the service can be performed first based on the model training method provided in this application embodiment, and then model inference associated with the service can be performed based on the model training results and the model inference method provided in this application embodiment.

[0207] The following text combines Figure 8 The training process of a model based on a vertical federated learning architecture is introduced. Figure 8The model training initiator shown is the NWDAF server, i.e., the first device, and the participant in the model training is the NF (Network Functions) co-located with the NWDAF client, i.e., the second device. Additionally, Figure 8 The model training process shown also involves NRF network elements, i.e., third-party devices, used to store or register NF capability information.

[0208] Figure 8 The method shown includes steps S801 to S810.

[0209] In step S801, the second device registers the capability information with the third device.

[0210] For example, capability information may include whether the second device supports operations based on vertical federated learning, such as model training and inference based on vertical federated learning. Alternatively, capability information may include the services (service X) that the second device supports for vertical federated learning operations and the local data features supported by the second device. Furthermore, capability information may also include the effective time period during which the second device supports vertical federated learning operations.

[0211] In step S802, the first device sends a discovery request to the third device.

[0212] This discovery request is used to discover a second device via a third device, specifically a second device that supports service X (such as the first service mentioned earlier). This discovery request can instruct a vertical federated learning operation, service-X.

[0213] In response to the discovery request, the third device may return device information that meets the discovery request, such as the address of the second device or the address of a client co-located with the second device, to the first device.

[0214] In step S803, the first device may send an execution request for a vertical federated learning operation to the second device. For example, the execution request may include model training based on vertical federated learning. Alternatively, the execution request may indicate the data features that the second device needs to collect; the time required to perform the vertical federated learning operation; and the region where the vertical federated learning operation will be performed.

[0215] In step S804, the second device performs feature alignment, or in other words, the second device determines whether a vertical federated learning operation can be performed.

[0216] For example, the second device can determine, based on local information, whether it can collect data features that meet the conditions within the time and area required by the request, and perform the corresponding longitudinal federated learning operations.

[0217] In step S805, the second device sends a response message to the first device regarding the execution request. This response message can be used to indicate that the second device can support the vertical federated learning operation.

[0218] In step S806, the first device determines one or more second devices to participate in this longitudinal federated learning operation.

[0219] For example, the first device can determine one or more second devices participating in this longitudinal federated learning operation based on the response message of the execution request.

[0220] In step S807, the first device sends a first request to the second device.

[0221] The first request can be used to request the second device to perform vertical federated learning operations. For example, the first request may include the initial model associated with service-X; model training identifiers; training task identifiers (also known as association identifiers); and one or more data features.

[0222] In step S808, a longitudinal federated learning operation, such as model training, is performed.

[0223] Taking model training as an example, the second device can collect local data based on the received initial model and the required data features, and then train the model based on the local data collection results. The second device can send the intermediate results of the model training to the first device, which can then aggregate the results from all the second devices and calculate the loss function using its stored labels. Furthermore, the first device can send the loss state to each of the second devices, allowing each second device to calculate the gradient and update the model based on the received loss state.

[0224] It should be noted that step S808 can be executed once or multiple times during a single model training process.

[0225] In step S809, the first device sends a training completion indication message to the second device.

[0226] For example, the first device can determine whether the model has converged based on the loss function. When the model converges, the first device sends a training completion indication message to the second device. Alternatively, when a preset number of training epochs is reached, the first device can send a training completion indication message to the second device.

[0227] The model training process terminates when the second device receives a training completion indication.

[0228] In step S810, the first device stores information about the second device.

[0229] For example, the first device can store information about the second device participating in this vertical federated learning operation, as well as the mapping relationship between the second device and the first service associated with this vertical federated learning operation. When performing subsequent vertical federated learning operations associated with the first service, the information of the second device can be directly retrieved, making it convenient to use.

[0230] In some embodiments, the NWDAF server can also interact with the gNB, OAM, or terminal devices, meaning the gNB, OAM, or terminal devices can also participate in the vertical federated learning operation. Since the terminal devices and gNB themselves possess certain AI capabilities, there is no need to deploy an NWDAF client. The OAM contains the MDAF (which has AI capabilities), therefore, the NWDAF server can interact with the MDAF to perform vertical federated learning operations.

[0231] The surge in extreme mobile scenarios, such as subways during peak hours and unexpected network outages, could trigger signaling storms. The following section combines... Figure 9 Taking signaling storm analysis as an example, this paper introduces the model inference process based on the vertical federated learning architecture. Figure 9 The model training initiator shown is the NWDAF server, i.e., the first device, and the participant in the model training is the NF (Network Functions) co-located with the NWDAF client, i.e., the second device. Additionally, Figure 9 The model inference process shown also involves the subscription device of the signaling storm, namely the fifth device, such as the terminal device.

[0232] Figure 9 The method shown includes steps S901 to S909.

[0233] In step S901, the terminal device and the second device establish a connection.

[0234] Because a connection establishment process has occurred between the terminal equipment and the core network, the core network possesses the historical data needed to determine signaling storms.

[0235] In step S902, the fifth device subscribes to signaling storm analysis from the first device. The fifth device can be either a second device or a terminal device.

[0236] Subscription messages may include, for example, a service identifier (signaling storm identifier), the time period and region where the signaling storm occurred, and the analysis target (i.e., the target terminal device). The target terminal device can be all terminal devices within that region, or a subset of them. If the target terminal device includes multiple terminal devices, it can be represented by a terminal device list or a terminal device group. If the target terminal device includes a single terminal device, it can be represented by a terminal device identifier.

[0237] In step S903, the first device selects a second device to participate in the signaling storm analysis.

[0238] If the first device stores information about a second device that supports signaling storm analysis, then the first device determines the second device to participate in the signaling storm analysis based on that information. If the first device does not store information about a second device that supports signaling storm analysis, such as if the first device has not performed signaling storm analysis model training, then the first device can proceed according to... Figure 8 The method shown is used to train the model, and then the second device to participate in the signaling storm analysis is determined based on the training results.

[0239] In step S904, the first device sends an analysis request to the second device.

[0240] The analysis request may include, for example, an inference task identifier (also known as an association identifier); a business identifier; the characteristics of the data to be collected; the time period to be analyzed; the geographical area or location to be analyzed; and the terminal device to be analyzed for signaling storm analysis.

[0241] In step S905, the second device performs model inference, such as performing model inference based on local data. The second device can perform model inference according to the content of the analysis request.

[0242] In step S906, the second device sends the intermediate inference result to the first device. The intermediate inference result may also include an inference task ID.

[0243] In step S907, the first device acquires the analysis results. For example, the first device can obtain the analysis results of the signaling storm by aggregating the intermediate inference results sent by the second device.

[0244] In step S908, the first device replies the analysis results to the fifth device, namely the subscription device for signaling storm analysis.

[0245] In step S909, the fifth device updates the policy based on the analysis results. The policy mentioned here can be a signaling-related policy.

[0246] The fifth device, such as a terminal device, can refrain from making registration requests or establishing connections after receiving the analysis results of a signaling storm. Instead, it can wait for the signaling storm to subside before resuming registration requests and connection establishment. The fifth device, such as the second device (i.e., core network elements), can reject certain signaling requests or temporarily deploy backup network elements during or before a signaling storm occurs to expand network load capacity and thus prevent network paralysis caused by the signaling storm.

[0247] In some embodiments, the local data mentioned in step S905 above may include mobility management information of the terminal device; session management information of the terminal device; user plane data packet information of the terminal device; policy information associated with the terminal device; and subscription information associated with the terminal device.

[0248] Mobility management information for terminal devices can be collected by AMF network elements as a second device. This mobility management information may include, for example, one or more of the following: the number of terminal devices registering with the AMF and the number of signaling messages generated; the type of registration initiated, such as initial registration, periodic registration update, or mobility registration update; the number of terminal devices that successfully registered and the number of signaling messages generated; the number of terminal devices that failed to register and the number of signaling messages generated; and the number of terminal devices served by the AMF network element, such as the number of terminal devices for which the AMF network element supports services.

[0249] Session management information of terminal devices can be collected by the SMF network element as a second device. This session management information may include one or more of the following: the number of terminal devices making session requests to the SMF network element and the amount of signaling generated; the session request type, such as initial session establishment, session modification, session release, etc.; the number of terminal devices successfully establishing sessions and the amount of signaling generated; the number of terminal devices failing to establish sessions and the amount of signaling generated; and the number of terminal devices served by the SMF network element, such as the number of terminal devices supporting the service.

[0250] User plane data packet information of the terminal device can be collected by the UPF network element as a second device. The user plane data packet information of the terminal device includes one or more of the following: the number of data packets forwarded by the UPF network element; and the number of N4 session establishment requests made by the UPF network element and the SMF network element.

[0251] The routing policy of the terminal device can be collected by the terminal device as a secondary device. The routing policy of the terminal device can be used to determine whether the application can be associated with an established PDU session or whether it is necessary to trigger the establishment of a new PDU session.

[0252] Other policies of the terminal equipment, such as mobility management policies, session management policies, and charging policies, can be collected by the PCF network element as a secondary device. This other policy information of the terminal equipment may include the number of policies provided by the PCF network element and the amount of signaling generated.

[0253] The subscription information associated with the terminal device can be collected by the UDM network element as a second device. This subscription information may include, for example, the number of signaling messages generated by interactions between the UDM network element and other network elements, such as the number of signaling messages generated by any network element interacting with the UDM network element due to any process. As an example, the subscription information associated with the terminal device may include the number of signaling messages for user registration with the UDM network element, or the number of signaling messages generated by the AMF network element and SMF network element querying the UDM network element for subscription information to perform registration, session establishment, and other processes.

[0254] The local data of the terminal device may also include the location information of the terminal device, which can be used to determine whether the terminal device will enter or leave a certain area in the future.

[0255] It should be noted that the local data mentioned above can be local data generated within the geographical region associated with the first business and / or the time period associated with the first business.

[0256] In some embodiments, the analysis results mentioned in step S907 above may include the signaling storm level of the entire communication network, such as low, medium, and high. The analysis results may also include the nodes in the network that generate the signaling storm, such as AMF network elements. For example, the interaction between terminal devices and AMF network elements generates a large amount of signaling, which may lead to a signaling storm and network paralysis. The analysis results may also include the proportion of influence of the nodes that generate the signaling storm on the signaling storm. For example, in the generation of a signaling storm, the influence proportion of AMF network elements is 80%, the influence proportion of SMF network elements is 10%, and the influence proportion of UDM network elements is 10%. The analysis results may also include the causes of the signaling storm in the network, such as the movement of a large number of users, simultaneous handover, simultaneous registration, or connection establishment, causing network nodes to receive a large number of signaling requests at the same time, thus generating a signaling storm. The analysis results may also include the terminal devices affected by the signaling storm and the time when the signaling storm recovers or disappears.

[0257] It should be noted that the model inference / training mentioned in the embodiments of this application can refer to model inference or model training.

[0258] The above text combined Figures 1 to 9 The method embodiments of this application are described in detail below, in conjunction with... Figures 10 to 14The present application provides a detailed description of the apparatus embodiments. It should be understood that the descriptions of the method embodiments correspond to the descriptions of the apparatus embodiments; therefore, any parts not described in detail can be found in the foregoing method embodiments.

[0259] Figure 10 This is a schematic diagram of a communication device provided in an embodiment of this application. Figure 10 The communication device 1000 shown can be any of the first devices mentioned above, and the communication device 1000 may include a first transmitting unit 1010.

[0260] The first sending unit 1010 is configured to send a first request to one or more second devices, the first request being used to request the one or more second devices to train a model associated with the first service.

[0261] In some embodiments, the first request includes one or more of the following: an initial model; data features associated with each of the one or more second devices; the training time of the model associated with the first service; the geographical region associated with the data features; the service identifier of the first service; a training task identifier; and the type of model training.

[0262] In some embodiments, the device further includes: a first receiving unit, configured to receive one or more training results sent by the one or more second devices, wherein the one or more training results are training results obtained by the one or more second devices respectively training the model associated with the first service based on local data.

[0263] In some embodiments, the local data is data collected by the one or more second devices based on the data characteristics.

[0264] In some embodiments, the device further includes: a second transmitting unit, configured to transmit loss parameters to the one or more second devices, the loss parameters being determined based on the one or more training results.

[0265] In some embodiments, the loss parameter is used by the one or more second devices to update the model associated with the first service.

[0266] In some embodiments, the device further includes: a third sending unit, configured to send first indication information to the one or more second devices, the first indication information being used to indicate that the training of the model associated with the first service is complete.

[0267] In some embodiments, the first indication information is determined based on a preset number of model training epochs and / or the loss parameter.

[0268] In some embodiments, the device further includes: a fourth sending unit, configured to send a second request to a third device before the first device sends a first request to one or more second devices, the second request being used to discover the fourth device, the fourth device being a device with model training capability, the model training capability being the ability to train a model associated with the first service.

[0269] In some embodiments, the second request includes one or more of the following: the type of model training; the business identifier of the first service; the data features associated with the first service; the training time of the model associated with the first service; and the geographical region associated with the data features.

[0270] In some embodiments, the device further includes: a second receiving unit, configured to receive a response message of the second request sent by the third device, wherein the response message of the second request is used to instruct the fourth device.

[0271] In some embodiments, the response message to the second request includes the address of the fourth device and / or the address of the client associated with the fourth device.

[0272] In some embodiments, the one or more second devices are some or all of the fourth devices.

[0273] In some embodiments, the device further includes a storage unit for storing information of the one or more second devices.

[0274] In some embodiments, the second device is one or more of the following: a device with model inference / training capabilities; a device with a client deployed to support model inference / training; and a device that supports interaction with a client having model inference / training capabilities.

[0275] In some embodiments, the first request is used to request the one or more second devices to perform model training based on vertical federated learning for the model associated with the first service.

[0276] Figure 11 This is a schematic diagram of a communication device provided in another embodiment of this application. Figure 11 The communication device 1100 shown can be any of the second devices mentioned above, and the communication device 1100 may include a first receiving unit 1110.

[0277] The first receiving unit 1110 is used to receive a first request sent by the first device, wherein the first request is used to request the second device to train the model associated with the first service.

[0278] In some embodiments, the first request includes one or more of the following: an initial model; data features associated with the second device; training time of the model associated with the first service; a geographical region associated with the data features; a service identifier of the first service; a training task identifier; and a type of model training.

[0279] In some embodiments, the device further includes: a first sending unit, configured to send training results to the first device, wherein the training results are training results obtained by the second device training the model associated with the first service based on local data.

[0280] In some embodiments, the local data is data collected by the second device based on the data characteristics.

[0281] In some embodiments, the device further includes a second receiving unit for receiving loss parameters sent by the first device, the loss parameters being determined based on the training results.

[0282] In some embodiments, the loss parameter is used by the second device to update the model associated with the first service.

[0283] In some embodiments, the device further includes: a third receiving unit, configured to receive first indication information sent by the first device, the first indication information being used to indicate that the training of the model associated with the first service is complete.

[0284] In some embodiments, the first indication information is determined based on a preset number of model training epochs and / or the loss parameter.

[0285] In some embodiments, the second device is one or more of the following: a device with model inference / training capabilities; a device with a client deployed to support model inference / training; and a device that supports interaction with a client having model inference / training capabilities.

[0286] In some embodiments, the device further includes a registration unit for registering capability information with a third device, wherein the capability information is information related to the model inference / training capabilities of the second device.

[0287] In some embodiments, the capability information includes one or more of the following: whether it has model training / inference capability; business that supports model training / inference; data characteristics that it supports collecting; types of model training / inference that it supports; time periods that it supports model training / inference; and geographical regions that it supports model training / inference.

[0288] In some embodiments, the first request is used to request the second device to perform model training based on vertical federated learning for the model associated with the first service.

[0289] Figure 12 This is a schematic diagram of a communication device provided in another embodiment of this application. Figure 12 The communication device 1200 shown can be any of the first devices mentioned above, and the communication device 1200 may include a first transmitting unit 1210.

[0290] The first sending unit 1210 is configured to send a third request to one or more second devices, the third request being used to request the one or more second devices to perform model inference for the first service association.

[0291] In some embodiments, the third request includes one or more of the following: a service identifier for the first service; an inference task identifier; data characteristics associated with each of the one or more second devices; a time period associated with the first service; a geographical region associated with the first service; a target terminal device associated with the first service; and a model identifier associated with the first service.

[0292] In some embodiments, the device further includes: a first receiving unit, configured to receive one or more inference results sent by the one or more second devices, wherein the one or more inference results are the results of the one or more second devices inferring the model associated with the first service based on local data.

[0293] In some embodiments, the first service is signaling storm analysis, and the local data includes one or more of the following: mobility management information of the target terminal device; session management information of the target terminal device; user plane data packet information of the target terminal device; policy information associated with the target terminal device; and subscription information associated with the target terminal device; the target terminal device is a terminal device associated with the first service.

[0294] In some embodiments, the policy information associated with the target terminal device includes the routing policy of the target terminal device.

[0295] In some embodiments, the one or more inference results include an inference task identifier.

[0296] In some embodiments, the device further includes: a second sending unit, configured to send analysis results associated with the first service to a fifth device, the analysis results being determined based on the one or more inference results.

[0297] In some embodiments, the first service is signaling storm analysis, and the analysis results include one or more of the following: the signaling storm level; the node that generated the signaling storm; the proportion of influence of the node that generated the signaling storm on the signaling storm; the cause of the signaling storm; the terminal devices that will be affected by the signaling storm; and the recovery time of the signaling storm.

[0298] In some embodiments, the analysis results are used to determine the signaling-related policies in the fifth device.

[0299] In some embodiments, the signaling-associated strategy includes performing one or more of the following during the signaling storm: not initiating a registration request; not initiating a connection establishment; rejecting a signaling request; and deploying a backup fifth device.

[0300] In some embodiments, the device further includes: a second receiving unit, configured to receive first information sent by the fifth device before the first device sends a third request to one or more second devices, the first information being used to subscribe to a first service, the first service being associated with model inference of the one or more second devices.

[0301] In some embodiments, the first information includes one or more of the following: the service identifier of the first service; the inference task identifier; the time period associated with the first service; the geographical region associated with the first service; the target terminal device associated with the first service; and the model identifier associated with the first service.

[0302] In some embodiments, the one or more second devices are determined based on second information pre-stored by the first device, or the one or more second devices are determined based on the model training results of the first service association, wherein the second information includes information about the devices participating in the model training of the first service association.

[0303] In some embodiments, the second device is one or more of the following: a device with model inference / training capabilities; a device with a client deployed to support model inference / training; and a device that supports interaction with a client having model inference / training capabilities.

[0304] Figure 13 This is a schematic diagram of a communication device provided in another embodiment of this application. Figure 13 The communication device 1300 shown can be any of the second devices mentioned above, and the communication device 1300 may include a receiving unit 1310.

[0305] The receiving unit 1310 is used to receive a third request sent by the first device, the third request being used to request the second device to perform model inference of the first service association.

[0306] In some embodiments, the third request includes one or more of the following: a service identifier of the first service; an inference task identifier; data characteristics associated with the second device; a time period associated with the first service; a geographical region associated with the first service; a target terminal device associated with the first service; and a model identifier associated with the first service.

[0307] In some embodiments, the device further includes: a sending unit, configured to send an inference result to the first device, the inference result being the result of the second device inferring the model associated with the first service based on local data.

[0308] In some embodiments, the first service is signaling storm analysis, and the local data includes one or more of the following: mobility management information of the target terminal device; session management information of the target terminal device; user plane data packet information of the target terminal device; policy information associated with the target terminal device; and subscription information associated with the target terminal device; the target terminal device is a terminal device associated with the first service.

[0309] In some embodiments, the policy information associated with the target terminal device includes the routing policy of the target terminal device.

[0310] In some embodiments, the reasoning result includes a reasoning task identifier.

[0311] In some embodiments, the first service is signaling storm analysis, and the analysis results associated with the first service include one or more of the following: the signaling storm level; the node that generated the signaling storm; the proportion of influence of the node that generated the signaling storm on the signaling storm; the cause of the signaling storm; the terminal devices that will be affected by the signaling storm; and the recovery time of the signaling storm.

[0312] In some embodiments, the analysis results are used to determine the signaling-related policies in a fifth device, which is a subscription device for the first service.

[0313] In some embodiments, the signaling-associated strategy includes performing one or more of the following during the signaling storm: not initiating a registration request; not initiating a connection establishment; rejecting a signaling request; and deploying a backup of the fifth device.

[0314] In some embodiments, the second device is determined based on second information pre-stored by the first device, or the second device is determined based on the model training results of the first service association, wherein the second information includes information about the devices participating in the model training of the first service association.

[0315] In some embodiments, the second device is one or more of the following: a device with model inference / training capabilities; a device with a client deployed to support model inference / training; and a device that supports interaction with a client having model inference / training capabilities.

[0316] In an optional embodiment, the first transmitting unit 1010, the first receiving unit 1110, the first transmitting unit 1210, and the receiving unit 1310 described above can be transceivers 1430. The communication device 1400 may further include a processor 1410 and a memory 1420, specifically as follows... Figure 14 As shown.

[0317] Figure 14 This is a schematic structural diagram of a communication device according to an embodiment of this application. Figure 14 The dashed lines indicate that the unit or module is optional. The device 1400 can be used to implement the methods described in the above method embodiments. The device 1400 can be a chip, a terminal device, or a network device.

[0318] Apparatus 1400 may include one or more processors 1410. The processor 1410 may support apparatus 1400 in implementing the methods described in the preceding method embodiments. The processor 1410 may be a general-purpose processor or a special-purpose processor. For example, the processor may be a central processing unit (CPU). Alternatively, the processor may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0319] The apparatus 1400 may further include one or more memories 1420. The memories 1420 store a program that can be executed by the processor 1410, causing the processor 1410 to perform the methods described in the preceding method embodiments. The memories 1420 may be independent of the processor 1410 or integrated within the processor 1410.

[0320] The device 1400 may also include a transceiver 1430. The processor 1410 can communicate with other devices or chips via the transceiver 1430. For example, the processor 1410 can send and receive data with other devices or chips via the transceiver 1430.

[0321] This application also provides a computer-readable storage medium for storing a program. This computer-readable storage medium can be applied to a first or second device provided in this application, and the program causes a computer to perform the methods executed by the first or second device in various embodiments of this application.

[0322] This application also provides a computer program product. The computer program product includes a program. The computer program product can be applied to a first device or a second device provided in this application embodiment, and the program causes a computer to perform the methods executed by the first device or the second device in various embodiments of this application.

[0323] This application also provides a computer program. This computer program can be applied to the first or second device provided in this application, and causes the computer to perform the methods executed by the first or second device in various embodiments of this application.

[0324] It should be understood that the terms "system" and "network" in this application can be used interchangeably. Furthermore, the terminology used in this application is only for explaining specific embodiments of the application and is not intended to limit the application. The terms "first," "second," "third," and "fourth," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. In addition, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0325] In the embodiments of this application, the term "instruction" can be a direct instruction, an indirect instruction, or an indication of a relationship. For example, A instructing B can mean that A directly instructs B, such as B being able to obtain information through A; it can also mean that A indirectly instructs B, such as A instructing C, so B can obtain information through C; or it can mean that there is a relationship between A and B.

[0326] In the embodiments of this application, "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean that B is determined solely based on A; B can also be determined based on A and / or other information.

[0327] In the embodiments of this application, the term "correspondence" can indicate a direct or indirect correspondence between two things, or an association between two things, or a relationship such as instruction and being instructed, configuration and being configured.

[0328] In this application embodiment, "predefined" or "preconfigured" can be implemented by pre-storing corresponding codes, tables, or other means that can be used to indicate relevant information in the device (e.g., including terminal devices and network devices). This application does not limit the specific implementation method. For example, predefined can refer to what is defined in the protocol.

[0329] In this application embodiment, the "protocol" may refer to a standard protocol in the field of communication, such as the LTE protocol, the NR protocol, and related protocols applied to future communication systems. This application does not limit this.

[0330] In the embodiments of this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0331] In the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0332] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0333] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0334] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0335] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can read or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0336] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for wireless communication, characterized in that, include: A first device sends a first request to one or more second devices, the first request being used to request the one or more second devices to train a model associated with a first service; The first request includes one or more of the following: Initial model; Data characteristics associated with each of the one or more second devices; The training time of the model associated with the first business; The geographical region associated with the data features; The service identifier of the first service; Training task identifiers; and Types of model training.

2. The method according to claim 1, characterized in that, The method further includes: The first device receives one or more training results sent by the one or more second devices, wherein the one or more training results are training results obtained by the one or more second devices from training the model associated with the first service based on local data; wherein the local data is data collected by the one or more second devices based on the data features.

3. The method according to claim 2, characterized in that, The method further includes: The first device sends loss parameters to the one or more second devices, the loss parameters being determined based on the one or more training results, and the loss parameters being used by the one or more second devices to update the model associated with the first service.

4. The method according to claim 3, characterized in that, The method further includes: The first device sends a first indication message to the one or more second devices. The first indication message is used to indicate that the training of the model associated with the first service is completed, and the first indication message is determined based on a preset number of model training rounds and / or the loss parameter.

5. The method according to any one of claims 1-4, characterized in that, Before the first device sends a first request to one or more second devices, the method further includes: The first device sends a second request to the third device. The second request is used to discover a fourth device, which is a device with model training capability, namely the ability to train the model associated with the first service. The second request includes one or more of the following: Types of model training; The service identifier of the first service; The data characteristics associated with the first business; The training time of the model associated with the first business.

6. The method according to any one of claims 1-5, characterized in that, The first request is used to request the one or more second devices to perform model training based on vertical federated learning for the model associated with the first service.

7. The method according to claim 1, characterized in that, The method further includes: The first device sends a third request to the one or more second devices, the third request being used to request the one or more second devices to perform model inference associated with the first service; The third request includes one or more of the following: The service identifier of the first service; Reasoning task identifier; Data characteristics associated with each of the one or more second devices; The time period associated with the first service; The geographical region associated with the first service; The target terminal device associated with the first service; and The model identifier associated with the first business.

8. The method according to claim 7, characterized in that, The method further includes: The first device receives one or more inference results sent by the one or more second devices, wherein the one or more inference results are the results of the one or more second devices inferring the model associated with the first service based on local data.

9. The method according to claim 8, characterized in that, The method further includes: The first device sends analysis results associated with the first service to the fifth device, the analysis results being determined based on the one or more inference results.

10. The method according to claim 9, characterized in that, Before the first device sends a third request to one or more second devices, the method further includes: The first device receives first information sent by the fifth device, the first information being used to subscribe to a first service, the first service being associated with the model inference of the one or more second devices; The first information includes one or more of the following: The service identifier of the first service; Reasoning task identifier; The time period associated with the first service; The geographical region associated with the first service; and The target terminal device associated with the first service; and The model identifier associated with the first business.

11. The method according to any one of claims 7-10, characterized in that, The one or more second devices are determined based on second information pre-stored by the first device, or the one or more second devices are determined based on the model training results of the first service association, wherein the second information includes information about the devices that participated in the model training of the first service association.

12. The method according to any one of claims 7-11, characterized in that, The second device is a device that has deployed a client that supports model inference / training.

13. A method for wireless communication, characterized in that, include: The second device receives a first request sent by the first device, the first request being used to request the second device to train a model associated with the first service; The first request includes one or more of the following: Initial model; The data characteristics associated with the second device; The training time of the model associated with the first business; The geographical region associated with the data features; The service identifier of the first service; Training task identifiers; and Types of model training.

14. The method according to claim 13, characterized in that, The method further includes: The second device sends the training result to the first device. The training result is the training result of the second device training the model associated with the first service based on local data.

15. The method according to claim 14, characterized in that, The method further includes: The second device receives loss parameters sent by the first device, the loss parameters being determined based on the training results, and the loss parameters being used by the second device to update the model associated with the first service.

16. The method according to claim 15, characterized in that, The method further includes: The second device receives a first indication message sent by the first device. The first indication message is used to indicate that the training of the model associated with the first service is completed, and the first indication message is determined based on a preset number of model training rounds and / or the loss parameter.

17. The method according to any one of claims 13-16, characterized in that, The method further includes: The second device registers capability information with the third device, wherein the capability information is information related to the model inference / training capabilities of the second device; The capability information includes one or more of the following: Does it have model training / inference capabilities? Supports business operations involving model training / inference; Supports the collection of data characteristics; Supported model training / inference types; The supported time period for model training / inference; and Supports geographical regions for model training / inference.

18. The method according to claim 13, characterized in that, The method further includes: The second device receives a third request sent by the first device, the third request being used to request the second device to perform model inference associated with the first service; The third request includes one or more of the following: The service identifier of the first service; Reasoning task identifier; The data characteristics associated with the second device; The time period associated with the first service; The geographical region associated with the first service; The target terminal device associated with the first service; and The model identifier associated with the first business.

19. The method according to claim 18, characterized in that, The method further includes: The second device sends the inference result to the first device. The inference result is the result of the second device inferring the model associated with the first service based on local data.

20. A communication device, characterized in that, The device is a first device, and the device includes: The first sending unit is configured to send a first request to one or more second devices, wherein the first request is configured to request the one or more second devices to train a model associated with the first service; The first request includes one or more of the following: Initial model; Data characteristics associated with each of the one or more second devices; The training time of the model associated with the first business; The geographical region associated with the data features; The service identifier of the first service; Training task identifiers; and Types of model training.

21. A communication device, characterized in that, The device is a second device, and the device includes: The first receiving unit is configured to receive a first request sent by the first device, wherein the first request is configured to request the second device to train a model associated with the first service; The first request includes one or more of the following: Initial model; The data characteristics associated with the second device; The training time of the model associated with the first business; The geographical region associated with the data features; The service identifier of the first service; Training task identifiers; and Types of model training.

22. A communication device, characterized in that, The device includes a transceiver, a memory, and a processor. The memory stores a program, and the processor invokes the program in the memory and controls the transceiver to receive or send signals so that the terminal device performs the method as described in any one of claims 1-12.

23. A communication device, characterized in that, The device includes a transceiver, a memory, and a processor. The memory stores a program, and the processor invokes the program in the memory and controls the transceiver to receive or send signals so that the terminal device performs the method as described in any one of claims 13-19.

Images