Communication method and apparatus

Abstract

The present application provides a communication method and apparatus, relating to the technical field of communications. A first network function receives a request message from a third network function, wherein the request message is used for requesting an access address of a second network function associated with target artificial intelligence (AI) information; on the basis of the request message, querying an association relation to obtain the access address of the second network function associated with the target AI information, wherein the association relation indicates a relation between at least one piece of AI information supported by at least one second network function and an access address of a second network function associated with each piece of AI information, and the target AI information is one of the at least one piece of AI information; and sending a request response message to the third network function, wherein the request response message is used for indicating the access address of the second network function associated with the target AI information. In this way, an intelligent service can be provided for a device.

Description

Communication method and device

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to the Chinese patent application filed with the State Intellectual Property Office of China on September 20, 2023, with application number 202311216424.7 and application name “A Communication Method and Device”, the entire contents of which are incorporated by reference into this application. Technical Field

[0003] The embodiments of the present application relate to the field of communication technology, and in particular to a communication method and apparatus. Background Art

[0004] In terms of standardization, the Third Generation Partnership Project (3GPP) has begun expanding the systematic intelligence and management architecture of the 5th generation (5G), standardizing intelligent services such as artificial intelligence (AI). However, specific solutions for how these intelligent services will be delivered to devices have yet to be provided.

[0005] Summary of the Invention

[0006] The present application provides a communication method and apparatus to provide intelligent services for devices.

[0007] In a first aspect, the present application provides a communication method, which can be applied to a first network function, wherein the first network function registers AI information of a network element, and the first network function can be an artificial intelligence storage service (artificial intelligence repository function, AIRF), which is specifically used to implement the registration of AI information. It can also be a functional entity set in a network storage function (network repository function, NRF) network element for implementing the registration of AI information. This application is not specifically limited here. When the first network function executes the method, it includes:

[0008] The method includes receiving a request message from a third network function, where the request message is used to request an access address of a second network function associated with target AI information; querying an association relationship according to the request message to obtain an access address of the second network function associated with the target AI information, where the association relationship indicates a relationship between at least one AI information supported by at least one second network function and an access address of the second network function associated with each AI information, where the target AI information is one of the at least one AI information; and sending a request response message to the third network function, where the request response message is used to indicate the access address of the second network function associated with the target AI information.

[0009] In this application, the first network function receives a request message for the access address associated with the target AI information, queries the corresponding access address, and feeds back the access address to the third network function, so that the third network function can obtain intelligent services through the access address. In this way, AI services can be provided to the third network function.

[0010] In an optional manner, when the request message is a subscription request message, sending a request response message to the third network function includes: sending a notification message of successful subscription to the third network function, and sending an access address of the second network function associated with at least one AI information to the third network function.

[0011] In actual application, the notification of successful subscription and the access address of the second network function associated with at least one AI information are not sent at the same time. For example, the notification message of successful subscription is sent at the first time, and the access address of the second network function associated with at least one AI information is sent at the second time. In addition, the above-mentioned notification of successful subscription and the access address of the second network function associated with at least one AI information can be sent using the same signaling or different signaling, which is not specifically limited here. By using a subscription request message as the request message, the present application can reuse the message format of the subscription request message. This approach does not require the message format to be reconstructed, which can reduce the complexity of data processing.

[0012] In an optional manner, when the request message is a query request message, sending the request response message to the third network function includes: sending the access address of the second network function associated with the target AI information to the third network function.

[0013] By adopting a query request message as a request message, the present application can reuse the message format of the query request message. This approach does not require reconstructing the message format, and can reduce the complexity of data processing.

[0014] In an optional manner, the first network function further receives a registration request message from at least one second network function, where the registration request message is used to indicate at least one AI information supported by the second network function and an access address of the second network function associated with each AI information; and determines the association relationship based on the registration request message.

[0015] In the present application, after the first network function receives the AI ​​information registered by the second network function, it can establish an association relationship between the AI ​​information and the access address of the second network function, so that when the third network function requests the access address of the second network function associated with the AI ​​information from the first network function, the first network function can query the access address of the second network function that can provide AI services based on the above association relationship and return it to the third network function.

[0016] In an optional manner, the first network function further sends a monitoring message to the second network function, where the monitoring message is used to monitor the working status of the second network function; and receives a monitoring response message from the second network function.

[0017] In this application, the working status of the second network function can be understood as the change of AI information supported by the second network function, or the actual working status of the second network function, which is not specifically limited in this application.

[0018] In one case, after the second network function registers certain AI information with the first network function, the AI ​​information supported by the second network function may change. For example, the second network function is NWDAF1, and NWDAF1 has registered its AI information supporting image processing with the first network function at the first time. At the second time (i.e., after the first time), the AI ​​information supported by NWDAF1 changes, such as the addition of AI information for audio processing. If the first network function sends a monitoring message to NWDAF1, it can monitor the change message of the AI ​​information supported by the second network function in real time, so as to timely adjust the registration information stored by the first network function.

[0019] In another case, after the second network function registers certain AI information with the first network function, the actual working status of the second network function changes. For example, the second network function is NWDAF1. NWDAF1 has registered its AI information supporting image processing with the first network function at the first time. At the second time, the NWDAF1 device crashes. If the first network function sends a monitoring message to NWDAF1, it can monitor NWDAF1 in real time and determine that NWDAF1 is down. It is determined that NWDAF1 cannot provide AI services for image processing, so that the registration information stored by the first network function can be adjusted in time.

[0020] In an optional manner, the third network function is a terminal; receiving a request message from the third network function includes: the first network function receives a request message sent by the third network function through a fourth network function, the fourth network function being used to convert the format of messages exchanged between the terminal and the first network function, and / or perform identity authentication on the terminal and the first network function.

[0021] In this application, the fourth network function is used as a bridge for information interaction between the first network function and the terminal, so that the terminal can request the access address associated with the AI ​​information from the first network function through the fourth network function. In this way, the terminal can obtain intelligent services. In addition, the fourth network function can convert the format of the messages exchanged between the terminal and the first network function, which can improve data processing efficiency. In addition, identity authentication of the terminal and the first network function can ensure the security of information interaction between the terminal and the first network function.

[0022] In an optional manner, the AI ​​information includes: AI services, and / or AI computing power.

[0023] In this application, AI information may refer to AI services, such as image processing, audio processing, and neural network model building. This is for illustrative purposes only and is not intended to be limiting. AI information may also refer to AI computing power, such as whether the central processing unit (CPU) of a network element can support the calculation of a random forest algorithm. This is for illustrative purposes only and is not intended to be limiting.

[0024] In an optional manner, the AI ​​information is an AI service; the registration request message includes: an identifier of the AI ​​service and an access address of a second network function that provides the AI ​​service.

[0025] In this application, when the AI ​​information is an AI service, the registration request message includes the identifier of the AI ​​service and the access address of the second network function providing the AI ​​service, so that the first network function can know the specific information of the AI ​​service and the access address of the second network function providing the AI ​​service. For example, the AI ​​service identifier is T1 (predefined T1 corresponds to image recognition processing), which means that the second network function registered with the first network function supports image recognition processing function.

[0026] In an optional manner, the registration request information further includes one or more of the following:

[0027] The categories of AI services, the categories of models supported by AI services, the service objects of AI services, the data formats supported by AI services, and the data calculation scale of AI services; among them, the service objects of AI services include terminal device type, APP name, terminal users belonging to the preset number segment, geographic location information, slice information, and access point identification.

[0028] Specifically, the categories of AI services (e.g., cloud photography, image processing, face recognition, translation, chatbot model (GPT Chat) dialogue, etc.), the categories of models supported by AI services (e.g., convolutional neural network (CNN), recurrent neural network (RNN), etc.), network, RNN), random forest, etc.), the service objects of the AI ​​service (i.e., the target objects of the AI ​​service), the data formats supported by the AI ​​service (e.g., if the AI ​​service is image recognition, the data format is RGB (RGB image is a digital image format. Its name refers to both the color space (red, green, blue) used by the image format and the image itself, which contains the pixel values ​​of the red, green, and blue channels) data or YUV (YUV image is a digital image format that separates image brightness (Y) and color (U, V). It is often used in the fields of video encoding and digital image processing. The following introduces common YUV sampling and arrangement methods) data), the data calculation scale of the AI ​​service (e.g., if the AI ​​service is image recognition, the data calculation scale refers to the number and size of image recognition); the service objects of the AI ​​service include the terminal device type (e.g., IoT device, mobile phone, etc.), the APP name (e.g., instant messaging software, video software, etc.), the terminal user belonging to the preset number segment (e.g., international mobile subscriber identity (IMSI) / mobile station International ISDN number) number, MSISD) (i.e., the terminal user of the mobile phone number segment), geographic location information (e.g., located in a specific location), slice information (e.g., receiving services through a certain slice), and access point identifier (e.g., receiving services through transmission reception point (TRP) 1). For example, the identifier of the AI ​​service is AI1, and the service object of AI1 is the Internet of Vehicles equipment located in Shanghai, then AI1 only provides AI services for the Internet of Vehicles equipment in Shanghai. This is only an example explanation, and the service object of the AI ​​service is not specifically limited.

[0029] In an optional manner, the AI ​​information is AI computing power; the registration request information includes: an identifier of the AI ​​computing power and an access address of a second network function that provides the AI ​​computing power.

[0030] In this application, when the AI ​​information is AI computing power, the registration request message includes the identifier of the AI ​​computing power and the access address of the second network function that provides the AI ​​computing power, so that the first network function can know the specific information of the AI ​​computing power and the access address of the second network function that provides the AI ​​computing power. For example, the AI ​​computing power identifier is S1 (pre-defined S1 corresponds to random forest calculation), which means that the second network function registered with the first network function supports the random forest calculation function.

[0031] In an optional manner, the registration request information further includes one or more of the following:

[0032] The AI ​​computing power type, AI computing power specifications, AI framework type indicated by the AI ​​computing power, AI model category indicated by the AI ​​computing power, and operator information of the AI ​​computing power.

[0033] Specifically, the computing power type of AI computing power (such as X86 / ARM CPU, graphics processing unit (GPU) and embedded neural-network processing unit (NPU) and other computing power resources that can run AI tasks), the computing power specifications of AI computing power (such as computing power evaluation indicators such as transactions per second (TPS) and queries per second (QPS)), the AI ​​framework type indicated by AI computing power (such as TensorFlow, Mindspore, SKlearn and other AI framework types), the AI ​​model category indicated by AI computing power (such as CNN, RNN, random forest, etc.) and the operator information of AI computing power (such as vector calculation, matrix calculation, etc.).

[0034] In a second aspect, the present application provides a communication method, which can be applied to a fourth network function. The fourth network function is a proxy network element, which can be used to build a connection bridge between the third network function and the first network function. It can also be used to authenticate the identity of the first network function and the third network function. This application is not specifically limited here. When the fourth network function executes the method, it includes:

[0035] Receive a request message from a third network function, where the request message is used to request the access address of the second network function associated with the target artificial intelligence (AI) information; send a request message to the first network function; receive a request response message from the first network function, where the request response message is used to indicate the access address of the second network function associated with the target AI information; and send a request response message to the third network function.

[0036] In an optional manner, when the request message is a subscription request message, receiving a request response message from the first network function includes: receiving a notification message of successful subscription, and receiving an access address of a second network function associated with at least one AI information from the first network function.

[0037] In an optional manner, when the request message is a query request message, receiving a request response message from the first network function includes: receiving an access address of a second network function associated with target AI information from the first network function.

[0038] In an optional manner, the third network function is a terminal; the fourth network function further performs format conversion on messages exchanged between the third network function and the first network function; and / or performs identity authentication on the third network function and the first network function.

[0039] In an optional manner, the fourth network function also receives data to be processed from the terminal; converts the data to be processed into a data format corresponding to the AI ​​information required by the terminal and sends it to the access address of the second network function associated with the AI ​​information required by the terminal; receives the data processing result from the second network function; and sends the data processing result to the terminal.

[0040] In a third aspect, the present application provides a communication method, which can be applied to a third network function, which is a terminal or other network element in a core network. The present application is not specifically limited here. When the third network function executes the method, it includes:

[0041] Send a request message to the fourth network function, where the request message is used to request the access address of the second network function associated with the target artificial intelligence (AI) information; receive a request response message from the fourth network function, where the request response message is used to indicate the access address of the second network function associated with the target AI information; send the data to be processed to the access address of the second network function associated with the target AI information through the fourth network function; and receive the data processing result from the second network function through the fourth network function.

[0042] In an optional manner, the third network function is a terminal.

[0043] In a fourth aspect, the present application provides a communication method, which can be applied to a second network function. The second network function has the ability to perform AI processing, can perform AI services, or perform AI calculations. This application is not specifically limited here. When the second network function executes the method, it includes:

[0044] Send a registration request message to the first network function, where the registration request message is used to indicate at least one AI information supported by the second network function and the access address of the second network function associated with each AI information; and receive a response message of successful registration from the second network function.

[0045] In an optional manner, the second network function also receives a monitoring message from the first network function, where the monitoring message is used to monitor the working status of the second network function; and sends a monitoring response message to the first network function, where the monitoring response message includes the working status of the second network function.

[0046] In an optional manner, the AI ​​information includes: AI services, and / or AI computing power.

[0047] In an optional manner, the AI ​​service includes one or more of the following: an identifier of the AI ​​service, an access address of a second network function providing the AI ​​service, a category of the AI ​​service, a category of models supported by the AI ​​service, a service object of the AI ​​service, a data format supported by the AI ​​service, and a data calculation scale of the AI ​​service; wherein the service objects of the AI ​​service include terminal device type, APP name, terminal users belonging to a preset number segment, geographic location information, slice information, and access point identifier.

[0048] In one optional embodiment, AI computing power includes one or more of the following:

[0049] AI computing identifier, AI computing power type, AI computing power specifications, AI framework type indicated by AI computing power, AI model category indicated by AI computing power, and operator information of AI computing power.

[0050] In a fifth aspect, an embodiment of the present application provides a communication device, which may be the above-mentioned first network function or a chip arranged inside the first network function, or the above-mentioned second network function or a chip arranged inside the second network function, or the above-mentioned third network function or a chip arranged inside the third network function, or the above-mentioned fourth network function or a chip arranged inside the fourth network function. The communication device has the function of implementing any one of the above-mentioned first to fourth aspects. For example, the communication device includes a module or unit or means corresponding to the steps involved in any one of the above-mentioned first to fourth aspects. The function or unit or means may be implemented by software, or by hardware, or the corresponding software may be implemented by hardware.

[0051] In one possible design, the communication device includes a processing unit and a transceiver unit, wherein the transceiver unit can be used to send and receive signals to achieve communication between the communication device and other devices; the processing unit can be used to perform some internal operations of the communication device. The transceiver unit can be called an input / output unit, a communication unit, etc., and the transceiver unit can be a transceiver; the processing unit can be a processor. When the communication device is a module (such as a chip) in a communication device, the transceiver unit can be an input / output interface, an input / output circuit, or an input / output pin, etc., and can also be called an interface, a communication interface, or an interface circuit, etc.; the processing unit can be a processor, a processing circuit, or a logic circuit, etc.

[0052] In another possible design, the communication device includes a processor and may also include a transceiver, the transceiver is used to send and receive signals, and the processor executes program instructions to complete the method in any possible design or implementation of the first to fourth aspects above. The communication device may also include one or more memories, the memory is used to couple with the processor, and the memory can store the necessary computer programs or instructions for implementing the functions involved in any aspect of the first to fourth aspects above. The processor can execute the computer program or instructions stored in the memory, and when the computer program or instructions are executed, the communication device implements the method in any possible design or implementation of the first to fourth aspects above.

[0053] In another possible design, the communication device includes a processor, which can be coupled to a memory. The memory can store the necessary computer programs or instructions for implementing the functions of any of the first to fourth aspects described above. The processor can execute the computer programs or instructions stored in the memory. When the computer programs or instructions are executed, the communication device implements the method of any possible design or implementation of the first to fourth aspects described above.

[0054] In another possible design, the communication device includes a processor and an interface circuit, wherein the processor is used to communicate with other devices through the interface circuit and execute the method in any possible design or implementation of the first to fourth aspects above.

[0055] It can be understood that in the fifth aspect above, the processor can be implemented by hardware or by software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor can be a general-purpose processor, which is implemented by reading the software code stored in the memory. In addition, the above processors can be one or more, and the memories can be one or more. The memory can be integrated with the processor, or the memory and the processor can be set separately. In the specific implementation process, the memory can be integrated with the processor on the same chip, or can be set on different chips respectively. The embodiment of the present application does not limit the type of memory and the setting method of the memory and the processor.

[0056] In a sixth aspect, an embodiment of the present application provides a communication system, which includes the first network function, the second network function, the third network function and the third network function in the above-mentioned first to fourth aspects.

[0057] In a seventh aspect, the present application provides a chip system, which includes a processor and may also include a memory, for implementing the method described in any possible design of aspects 1 to 4 above. The chip system may be composed of a chip, or may include a chip and other discrete devices.

[0058] In an eighth aspect, the present application also provides a computer-readable storage medium, in which computer-readable instructions are stored. When the computer-readable instructions are run on a computer, the computer executes a method in any possible design as in the first to fourth aspects.

[0059] In a ninth aspect, the present application provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the methods of the various embodiments of the first to fourth aspects described above.

[0060] For the technical effects that can be achieved in the above-mentioned second to ninth aspects, please refer to the description of the technical effects that can be achieved by the corresponding possible design schemes in the above-mentioned first aspect, and this application will not repeat them here. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] FIG1A shows a schematic diagram of a communication system;

[0062] FIG1B shows a schematic diagram of another communication system;

[0063] FIG2 shows a schematic diagram of yet another communication system;

[0064] FIG3 shows a schematic flow chart of a communication method;

[0065] FIG4 shows a schematic flow chart of a communication method;

[0066] FIG5 shows a schematic flow chart of a communication method;

[0067] FIG6 shows a schematic flow chart of a communication method;

[0068] FIG7 shows a schematic flow chart of a communication method;

[0069] FIG8 shows a schematic flow chart of a communication method;

[0070] FIG9 shows a schematic flow chart of a communication method;

[0071] FIG10 shows a schematic flow chart of a communication method;

[0072] FIG11 shows a schematic structural diagram of a communication device provided in an embodiment of the present application;

[0073] FIG12 shows a schematic structural diagram of another communication device provided in an embodiment of the present application;

[0074] FIG13 shows a schematic structural diagram of another communication device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical solutions and advantages of this application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The specific operating methods in the method embodiments can also be applied to the device embodiments or system embodiments. In the description of this application, unless otherwise specified, "multiple" means two or more. Therefore, the implementation of the device and method can refer to each other, and the repeated parts will not be repeated.

[0076] To meet the challenges of wireless broadband technology and maintain the leading edge of the Third Generation Partnership Project (3GPP) network, the 3GPP standards group has developed the Next Generation System architecture for mobile communications, known as the 5G network architecture. This architecture not only supports access to the 5G core network (CN) using radio access technologies defined by the 3GPP standards group (such as Long Term Evolution (LTE) and RAN), but also supports access to the core network using non-3GPP access technologies via the non-3GPP interworking function (N3IWF) or the next generation packet data gateway (ngPDG).

[0077] Refer to Figure 1A, which shows a schematic diagram of the network architecture for a non-roaming scenario. In this network architecture, the network slice selection function (NSSF), network exposure function (NEF), NRF, policy control function (PCF), unified data management (UDM), application function (AF), authentication server function (AUSF), access and mobility management function (AMF), and session management function (SMF) can communicate with each other based on service methods. Of course, for two network elements to communicate, one network element must expose the corresponding service methods to the other. In Figure 1A, Nnssf can be regarded as the service interface of the NSSF. Similarly, Nnef is the service interface of the NEF, Npcf is the service interface of the PCF, Nudm is the service interface of the UDM, Naf is the service interface of the AF, Nausf is the service interface of the AUSF, Namf is the service interface of the AMF, and Nsmf is the service interface of the SMF. In addition, AMF and terminal equipment can communicate through the N1 interface, AMF and (R)AN can communicate through the N2 interface, SMF and user plane function (UPF) can communicate through the N4 interface, the terminal equipment and (R)AN can communicate through the air interface, (R)AN and UPF can communicate through the N3 interface, and UPF and data network (DN) can communicate through the N6 interface.

[0078] The mobility management function network element is, for example, an MME in a 4G system and an AMF in a 5G system. The embodiments of the present application are not limited thereto and may be implemented through other network elements in other communication systems. Taking the mobility management function network element as an example, the AMF is mainly responsible for wireless docking, terminating the radio access network (RAN) control plane (CP) interface, that is, the N2 interface, terminating the non-access stratum (NAS) and NAS encryption and integrity protection, registration management, connection management, reachability management, mobility management, transmitting session management (SM) messages between user equipment (UE) and SMF, or UE mobility event notification, etc.

[0079] The session management function network element, in the 5G system, is, for example, the SMF, and can also be implemented through other network elements in other communication systems. Taking the SMF as an example, the SMF can provide session management functions such as session establishment, modification, and release, including tunnel maintenance functions between the UPF and the access network (AN) node, UE Internet Protocol (IP) address allocation and management, dynamic host control protocol (DHCP), selection and control of the user plane (UP), configuration of the UPF diversion function, termination of the policy control function interface, billing, roaming functions, or policy control related functions.

[0080] In 5G systems, the user plane function network element is, for example, the UPF. It can also be implemented through other network elements in other communication systems. For example, the UPF is the user plane data forwarding entity. As the external PDU session point for data network interconnection, the UPF has functions such as message routing and forwarding, message inspection, user plane partial policy enforcement, lawful interception, traffic usage reporting, and quality of service (QoS) processing.

[0081] The network function repository functional network element is, for example, NRF in the 5G system, and can also be implemented through other network elements in other communication systems. Taking the network function repository functional network element as NRF as an example, NRF can support service discovery functions, maintain NF configuration of available network function (NF) instances and other functions. In the roaming scenario, it is divided into visit NRF (vNRF) and home NRF (hNRF). Based on the network implementation, it is divided into public land mobile network (PLMN) level NRF, shared slice NRF, or specific slice NRF. Among them, the PLMN-level NRF can be understood as different PLMNs can use different NRFs; the shared slice NRF can be understood as different network slices can share one NRF; the specific slice NRF can be understood as different network slices use different NRFs.

[0082] In FIG1A , Nnef, Npcf, Nudm, Naf, Namf, Nsmf, N1, N2, N3, N4, and N6 are interface serial numbers. The meanings of these interface serial numbers can be found in the 3GPP standard protocol and are not limited here.

[0083] Figure 1B shows a schematic diagram of a 5G network architecture based on point-to-point interfaces. The functions of the network elements described in Figure 1A are similar to those of the corresponding network elements, and are not described in detail here. The main difference between Figure 1B and Figure 1A is that the interfaces between the network elements in Figure 1B are point-to-point interfaces, not service-based interfaces.

[0084] In the architecture shown in Figure 1B, the interface between the UE and the AMF network element is called the N1 interface, the interface between the AMF network element and the RAN device is called the N2 interface, the interface between the RAN device and the UPF network element can be called the N3 interface, the interface between the SMF network element and the UPF network element is called the N4 interface, the interface between the PCF network element and the AF network element is called the N5 interface, the interface between the UPF network element and the DN is called the N6 interface, the interface between the SMF network element and the PCF network element is called the N7 interface, the interface between the AMF network element and the UDM network element is called the N8 interface, the interface between different UPF network elements is called the N9 interface, the interface between the UDM network element and the SMF network element is called the N10 interface, the interface between the AMF network element and the SMF network element is called the N11 interface, the interface between different AMF network elements is called the N14 interface, and the interface between the AMF network element and the PCF network element is called the N15 interface.

[0085] It is understood that the above-mentioned network element or function can be a network element in a hardware device, a software function running on dedicated hardware, or a virtualized function instantiated on a platform (e.g., a cloud platform). Optionally, the above-mentioned network element or function can be implemented by a single device, or by multiple devices, or can be a functional module within a single device, and this is not specifically limited in the embodiments of the present application.

[0086] The mobility management network element, session management network element, policy control network element, application function network element, access network equipment, and network storage network element in this application can be the AMF, SMF, PCF, AF, RAN, and NRF in Figure 1A or Figure 1B, respectively, or can be network elements with the functions of the above-mentioned AMF, SMF, PCF, AF, RAN, and NRF in future communications such as the sixth generation (6G) network. This application is not limited to this. For the convenience of explanation, this application takes the mobility management network element, session management network element, policy control network element, application function network element, access network equipment, and network storage network element as the above-mentioned AMF, SMF, PCF, AF, RAN, and NRF as examples for explanation.

[0087] For ease of explanation, in the embodiments of this application, a base station (such as a 4th generation (4G) eNB, a 5G gNB, or a base station in future communications) is used as an example of an access network device for explanation, and the subsequent "base station" can be replaced with "access network device." In the embodiments of this application, a UE is used as an example of a terminal for explanation, and the subsequent "UE" can be replaced with "terminal."

[0088] It is understandable that the core network may also include other network functional entities, which is not limited in this application.

[0089] The technical solutions provided in the embodiments of this application can be applied to various communication systems. For example, they can be applied to 5G systems, as well as other future-oriented new systems, such as 6G systems. The embodiments of this application do not specifically limit this. In addition, the term "system" and "network" can be used interchangeably.

[0090] In order to better adapt to the needs of artificial intelligence, the service-oriented architecture provided by this application introduces an artificial intelligence storage service (artificial intelligence repository function, AIRF) and an artificial intelligence service agent (AI Service Agent) as shown in Figure 2. The AIRF can be set up together with the above-mentioned NRF network element or independently. This application does not specifically limit this. The AIRF is mainly used to receive AI information registrations and provide AI information query services to other network elements. AIRF can receive information registrations from the model training logical function (MTLF) and data analysis function (ANLF) in the network data analytics function (NWDAF), and can also receive information registrations from other network elements. This application does not specifically limit this. In actual application, AIRF can directly provide AI information query services to AI consumers in SMF (i.e., AI Compute), and can also provide AI information query services to AI consumers in UPF. This is not specifically limited here. In addition, UE and core network are usually isolated. UE can receive AIRF's AI information query service through AI Service Agent. In addition, some network elements that do not support direct interaction with AIRF can also establish a communication connection with AIRF through AI Service Agent to obtain AI information query services. The service-oriented architecture provided here is only for schematic illustration and does not specifically limit how to set up AIRF and AI Service Agent. Among them, the interface between NSSF and AMF network element is called N22 interface, the interface between AMF network element and RAN equipment is called N12 interface, the interface between AIRF and AI Service Agent is called Nairf interface, the interface between PDU session anchor UPF (packet data unit session anchor UPF, PSA-UPF) and NWDAF is called Nnf interface, the interface between SMF and NWDAF is called Nnwdaf interface, the interface between PCF and NWDAF is called Nnwdaf interface, and the interface between intermediate UPF (intermediate UPF, I-UPF) and PSA-UPF is N9 interface.

[0091] In terms of standardization, 3GPP SA2 and SA5 have already started and expanded on the systematic intelligence and management architecture of 5G, and standardized the definition of NWDAF and management data analytics function (MDAF) and related mechanism processes. There is no clear definition of how the next-generation 5.5G or 6G intelligent plane will define the architecture, business processes and interactive interfaces of the Network for AI to provide the AI ​​service capabilities or AI computing power built by the core network to terminal users, APPs, or AFs. In other words, the AI ​​service capabilities or AI computing power built by the core network do not have appropriate architectural definitions, functional decomposition, and message interaction processes to provide corresponding AI service capabilities or AI computing power to terminal users UE or application servers AF and other NFs. Based on this, the present application provides a communication method to provide intelligent services for devices.

[0092] FIG3 is a flow chart of a communication method provided by the present application. The method can be implemented through the interaction of the first network function and other network functions. The present application is not specifically limited here. The first network function is registered with the AI ​​information of the network element. The first network function can be an AIRF, which is specifically used to implement the registration of AI information. It can also be a functional entity set in the NRF network element for implementing the registration of AI information. The second network function has the ability to perform AI processing, can execute AI services, or perform AI calculations. The present application is not specifically limited here. The second network function can be an NWDAF, MDAF, NF, etc. The present application is not specifically limited here. The third network function is a terminal or other network element in the core network. The present application is not specifically limited here. The fourth network function is an agent network element such as an AI Service Agent, which can be used to build a connection bridge between the third network function and the first network function. It can also be used to authenticate the identities of the first network function and the third network function and convert the format of the interactive messages. The present application is not specifically limited here. FIG3 takes the third network function as a terminal, NF1, NF2, the fourth network function as an AI Service Agent, the second network function as NWDAF1, and the first network function as AIRF as an example for explanation. The execution is as follows:

[0093] Step 301: NWDAF1 sends a registration request message to AIRF, where the registration request message is used to indicate at least one AI information supported by the second network function and an access address of the second network function associated with each AI information.

[0094] Step 302 : The AIRF determines the association relationship according to the registration request message in step 301 .

[0095] The association relationship indicates the relationship between at least one AI information supported by at least one second network function and the access address of the second network function associated with each AI information. For example, based on AI information 1 and access address 1 carried in the NWDAF1 registration request message, the AIRF may determine that there is an association relationship between AI information 1 and access address 1. If a network element subscribes to or queries AI information 1, the AIRF may return access address 1 of NWDAF1 to the network element.

[0096] In actual applications, the AIRF may receive not only the aforementioned registration request message from the AIRF, but also registration request messages from other second network functions, which is not specifically limited in this application. That is, the AIRF may receive registration request messages from at least one second network function. Furthermore, the second network function may support one or more types of AI information. The second network function may register all supported AI information with the AIRF, or only some of the supported AI information, which is not specifically limited in this application. Furthermore, different second network functions may support the same AI information, which is not specifically limited in this application.

[0097] The AI ​​information includes: AI services (such as image classification, image recognition, audio synthesis, etc., which are only illustrative here and not specifically limited), and / or AI computing power (such as the feature vector of the neural network model used for AI calculation calculated by the CPU, etc.). If the AI ​​information is an AI service; the above-mentioned registration request message includes: the identifier (or name) of the AI ​​service and the access address of the second network function that provides the AI ​​service. The identifier of the AI ​​service can be the ID of the AI ​​service, such as the ID of the image classification task, the ID of the audio synthesis task, etc., which are only illustrative here. The access address of the second network function can be the physical address information of the second network function, or the full qualified domain name (FQDN). Since the second network function may include multiple processing units, and different processing units are isolated from each other, the access address of the second network function can be multiple address information that meet the conditions. When other network elements access the second network element, they can access it based on a specific physical address, or they can access it based on the FQDN query to obtain a specific physical address. In addition, the registration request information also includes one or more of the following: the category of AI service (for example, cloud photography, image processing, face recognition, translation, GPT Chat conversation, etc.), the model category supported by the AI ​​service (for example, CNN, RNN, random forest, etc.), the service object of the AI ​​service (that is, the target object of the AI ​​service), the data format supported by the AI ​​service (for example, if the AI ​​service is image recognition, the data format is RGB data or YUV data), and the data calculation scale of the AI ​​service (for example, if the AI ​​service is image recognition, the data calculation scale is the number and size of image recognition). Among them, the service object of the AI ​​service includes the terminal device type (for example, IoT device, mobile phone, etc.), APP name (for example, instant messaging software, video software, etc.), terminal users belonging to the preset number segment (for example, terminal users in the IMSI / MSISD number segment), geographic location information (for example, located in a specific location), slice information (for example, receiving services through a certain slice), and access point identifier (for example, receiving services through TRP1). For example, if the AI ​​service identifier is AI1 and the service object of AI1 is the Internet of Vehicles devices located in Shanghai, then AI1 only provides AI services to Internet of Vehicles devices in Shanghai. This is just an example and does not specifically limit the service objects of AI services.

[0098] If the AI ​​information is AI computing power; the registration request information includes: the identifier of the AI ​​computing power and the access address of the second network function that provides the AI ​​computing power. In addition, the registration request information also includes one or more of the following: the computing power type of the AI ​​computing power (such as X86 / ARM CPU, NPU, GPU, etc., which are computing power resources that can run AI tasks), the computing power specifications of the AI ​​computing power (such as TPS and QPS, which are evaluation indicators of computing power), the AI ​​framework type indicated by the AI ​​computing power (such as TensorFlow, Mindspore, SKlearn, etc.), the AI ​​model category indicated by the AI ​​computing power (such as CNN, RNN, random forest, etc.), and the operator information of the AI ​​computing power (such as vector calculation, matrix calculation, etc.).

[0099] In the present application, after the first network function receives the AI ​​information registered by the second network function, it can establish an association relationship between the AI ​​information and the access address of the second network function, so that when the third network function requests the access address of the second network function associated with the AI ​​information from the first network function, the first network function can query the access address of the second network function that can provide AI services based on the above association relationship and return it to the third network function.

[0100] In addition, the AI ​​information supported by the second network function may be updated regularly. In order to ensure better adaptation to the user's application needs, AIRF can send a monitoring message to the second network function. The monitoring message is used to monitor the working status of the second network function. After that, AIRF can receive the detection response message of the second network function. Of course, AIRF can regularly send detection and monitoring messages to the second network function, such as sending a monitoring message once a day. Of course, the second network function can also regularly report change information to AIRF, such as reporting the AI ​​information of the second network function that changes when it registers with AIRF once every 4 hours. The second network function can also directly report to AIRF after the supported AI information changes. This application is not specifically limited here.

[0101] In this application, the working status of the second network function can be understood as the change of AI information supported by the second network function, or the actual working status of the second network function, which is not specifically limited in this application.

[0102] In one case, after the second network function registers certain AI information with the first network function, the AI ​​information supported by the second network function may change. For example, the second network function is NWDAF1, and NWDAF1 has registered its AI information supporting image processing with the first network function at the first time. At the second time (i.e., after the first time), the AI ​​information supported by NWDAF1 changes, such as the addition of AI information for audio processing. If the first network function sends a monitoring message to NWDAF1, it can monitor the change message of the AI ​​information supported by the second network function in real time, so as to timely adjust the registration information stored by the first network function.

[0103] In another case, after the second network function registers certain AI information with the first network function, the actual working status of the second network function changes. For example, the second network function is NWDAF1. NWDAF1 has registered its AI information supporting image processing with the first network function at the first time. At the second time, the NWDAF1 device crashes. If the first network function sends a monitoring message to NWDAF1, it can monitor NWDAF1 in real time and determine that NWDAF1 is down. It is determined that NWDAF1 cannot provide AI services for image processing, so that the registration information stored by the first network function can be adjusted in time.

[0104] Among them, the terminal can refer to steps 303a to 307a to obtain the access address for executing the second network function associated with the target AI information (the target AI information is one of the at least one AI information, which is the AI ​​information required by the terminal in this case), NF1 can refer to steps 303b to 307b to obtain the access address for executing the second network function associated with the target AI information, and NF2 can refer to steps 303c to 305c to obtain the access address for executing the second network function associated with the target AI information. The specific implementation is as follows:

[0105] In step 303a, the terminal sends a request message to the AI ​​Service Agent, where the request message is used to request an access address of a second network function associated with the target AI information.

[0106] Step 304a: The AI ​​Service Agent sends a request message to the AIRF.

[0107] Step 305a: The AIRF queries the association relationship according to the request message to obtain the access address of the second network function associated with the target AI information.

[0108] Step 306a: The AIRF sends a request response message to the AI ​​Service Agent, where the request response message is used to indicate the access address of the second network function associated with the target AI information.

[0109] Step 307a: The AI ​​Service Agent sends a request response message to the terminal.

[0110] In addition, the terminal can also send the data to be processed to the AI ​​Service Agent, and the AI ​​Service Agent can also convert the data to be processed into a data format corresponding to the AI ​​information required by the terminal and send it to the access address of the second network function associated with the AI ​​information required by the terminal; and receive the data processing results from the second network function; and then send the data processing results to the terminal.

[0111] In step 303b, NF1 sends a request message to the AI ​​Service Agent, where the request message is used to request the access address of the second network function associated with the target AI information.

[0112] Step 304b: The AI ​​Service Agent sends a request message to the AIRF.

[0113] Step 305b: The AIRF queries the association relationship according to the request message to obtain the access address of the second network function associated with the target AI information.

[0114] Step 306b: The AIRF sends a request response message to the AI ​​Service Agent, where the request response message is used to indicate the access address of the second network function associated with the target AI information.

[0115] Step 307b: AI Service Agent sends a request response message to NF1.

[0116] Step 303c: NF2 sends a request message to AIRF.

[0117] Step 304c: The AIRF queries the association relationship according to the request message to obtain the access address of the second network function associated with the target AI information.

[0118] Step 305c: AIRF sends a request response message to NF2, where the request response message is used to indicate the access address of the second network function associated with the target AI information.

[0119] The request message in steps 303a to 305c above may be a subscription request message or a query request message, which is not specifically limited in this application. If the request message is a subscription request message, before sending the request response message, a notification message of successful subscription may be sent to the terminal, NF1, and NF2, and then the access address of the second network function associated with at least one AI information may be sent. Of course, in actual application, the third network function may not only subscribe to the access address of the second network function associated with the required AI information, but may also subscribe to the access address of the second network function associated with other AI information. The subscription information is not specifically limited here. In actual application, The third network function can carry reference information so that the AIRF can query the access address of the second network function associated with the AI ​​information that better meets the third network function's subscription requirements. For example, if the reference information carried by the third network function is an AI service identifier, the AIRF can query the address of the second network function associated with the AI ​​service identifier and feedback it to the third network function. Alternatively, if the third network function carries the AI ​​service identifier and the app name APP1, the AIRF can query which of the AI ​​service identifiers supports APP1 and select the address of the second network function associated with the AI ​​service that supports APP1 and feedback it to the third network function. This is merely an example and not a specific limitation. In actual applications, the subscription success notification and the access address of the second network function associated with at least one AI information may not be sent simultaneously. For example, a subscription success notification message may be sent at a first time, and the access address of the second network function associated with at least one AI information may be sent at a second time, where the first time is before the second time. Furthermore, the subscription success notification and the access address of the second network function associated with at least one AI information may be sent using the same signaling or different signaling, which is not specifically limited here. By using a subscription request message as the request message, the present application can reuse the message format of the subscription request message. This approach eliminates the need to reconstruct the message format, thus reducing the complexity of data processing. Furthermore, by using a query request message as the request message, the present application can reuse the message format of the query request message. This approach eliminates the need to reconstruct the message format, thus reducing the complexity of data processing.

[0120] In this application, the first network function receives a request message for the access address associated with the target AI information, queries the corresponding access address, and feeds back the access address to the third network function, so that the third network function can obtain intelligent services through the access address. In this way, AI services can be provided to the third network function.

[0121] To better illustrate the solution of this application, the registration process of AI information is described below with reference to FIG4 . This registration process can be executed through data interaction between NWDAF and AIRF, which include AI service instances or computing power such as ANLF and MTLF, and NFs that can provide AI computing power (such as UPF, SMF, and other NF-native AI computing power resources or independently deployed AI computing power). The execution is as follows:

[0122] The following describes the registration process for AI services through steps 401A to 403A, and the following describes the registration process for AI computing power through steps 401B to 403B.

[0123] Step 401A: NWDAF sends a registration request message to AIRF.

[0124] The registration request message can be sent through Nairf_AIManage_AIServiceRegister_Request (artificial intelligence storage service request, artificial intelligence management request, artificial intelligence service registration request), which can carry the identifier of the AI ​​service, the instance providing the AI ​​service (that is, the access address of NWDAF), the category of the AI ​​service, the model category supported by the AI ​​service, the service object of the AI ​​service, the data format supported by the AI ​​service, and the data calculation scale of the AI ​​service. You can refer to the above to understand it and will not repeat it here. The instances and the access addresses of the second network functions in this application can be replaced with each other unless otherwise specified.

[0125] Step 402A: AIRF stores the registration request message.

[0126] Step 403A: AIRF sends a registration success response message to NWDAF.

[0127] The successful registration response message can be sent via Nairf_AIManage_AIServiceRegister_Response (AI storage service request response, AI management request response, AI service registration response). A registered AI service can provide AI service consumers with the ability to query and subscribe to the AI ​​service instance. The query key information carried can be the AI ​​service ID or name. Other NFs that can provide AI service capabilities (such as UPF, SMF, and other NF-native AI services or independent AI services) can also register AI services using Nairf_AIManage_AIServiceRegister_Request, but this process is not explained in detail here.

[0128] Step 401B: NF sends a registration request message to AIRF.

[0129] This registration request message can be sent via Nairf_AIManage_AIServiceRegister_Request and can carry the identifier of the AI ​​computing power, the instance providing the AI ​​computing power (i.e., the logical address of the NF), the computing power type of the AI ​​computing power, the computing power specification of the AI ​​computing power, the AI ​​framework type indicated by the AI ​​computing power, the AI ​​model category indicated by the AI ​​computing power, and the operator information of the AI ​​computing power. Please refer to the above for understanding and will not be repeated here.

[0130] Step 402B: AIRF stores the registration request message.

[0131] Step 403B: AIRF sends a registration success response message to NF.

[0132] The successful registration response message can be sent via Nairf_AIManage_AIServiceRegister_Response. Registered AI computing power can provide AI computing power consumers with query and subscription capabilities for the AI ​​computing power instance. Key query information carried can include AI computing resource type, AI framework type, AI model category, operator capabilities, and more.

[0133] AIRF supports proactively initiating heartbeat messages (also known as monitoring messages) to NFs (or NWDAFs) that have registered AI services or AI computing power to detect the status of the NF (or NWDAF). NFs that have registered AI services or AI computing power can also refresh the NF (or NWDAF) status by updating registration information. AIRF supports providing consumers with the Nairf_AIManage_AIServiceDiscoveryRequest / Response message to enable AI service instance discovery. It also supports providing consumers with the Nairf_AIManage_AIComputeResouceDiscoveryRequest / Response message to enable AI computing power instance discovery.

[0134] In addition to supporting the discovery capabilities of AI service instances and AI computing power instances, AIRF also supports providing Nairf_AIManage_AIServiceStatusSubscribe / Nairf_AIManage_AIServiceStatusNotify (artificial intelligence storage service notification, artificial intelligence management notification, artificial intelligence service status subscription notification, artificial intelligence storage service notification, artificial intelligence management notification, artificial intelligence service status notification), Nairf_AIManage_AIComputeStatusSubscribe / Nairf_AIManage_AIComputeStatusNotify (artificial intelligence storage service notification, artificial intelligence management notification, artificial intelligence computing power status subscription notification, artificial intelligence storage service notification, artificial intelligence management notification, artificial intelligence computing power status notification) to consumers of AI services or AI computing power.

[0135] The AIRF provides a registration and discovery framework for AI services or computing power, including registration of AI services and computing power, status monitoring of AI services or AI computing power instances, discovery of AI services or AI computing power instances, and status subscription process for AI services or AI computing power. This provides a basis for UE, AF or other NFs to discover and use AI services or computing power of the core network as consumers.

[0136] Next, Figures 5 and 6 are used to introduce how the UE uses AI services and AI computing power. The UE can use AI services with reference to Figure 5, and the UE can use AI computing power with reference to Figure 6. This can be done through data interaction between the UE, RAN, AMF, SMF, AI Service Agent, UPF, AIRF, and PCF. The UE interacts with the AI ​​Service Agent through the user plane to query and use network AI service capabilities.

[0137] Case 1: UE uses AI services

[0138] Refer to Figure 5 and execute as follows:

[0139] Step 501: After being activated, the UE initiates a session establishment request to the SMF through the RAN and AMF.

[0140] Step 502: The SMF sends a session policy control request message to the PCF.

[0141] The session policy control request message can be sent via Npcf_SMPolicyControl_Create Request to obtain contract information with the subscribed user, such as control policies and billing rules. In addition, the contract information can also include the address information of the AI ​​Service Agent service.

[0142] Step 503: The PCF sends a response message to the session policy control request message to the SMF.

[0143] The response message can be returned to the SMF via the Npcf_SMPolicyControl_Create Response message. If the contract information includes the address information of the AI ​​Service Agent service, the AMF can record the address information of the AI ​​Service Agent service and can also configure the address information of the AIServiceAgent service locally, such as: AMF configures the AI ​​Service Agent address information on the SMF or AMF through a configuration command or file, and carries it in the corresponding message, without the need to follow the PCF-》SMF-》AMF-》RAN-》UE path from the PCF to the UE in sequence. This application does not specifically limit this.

[0144] Step 504: SMF selects UPF to establish a session for the UE.

[0145] In step 505, the UPF establishes a session for the UE based on the session establishment request message of the SMF, installs the PDR rules (message detection rules) issued by the SMF, performs corresponding perception, statistics and control policies on the user's service flow, and returns an N4 session establishment success response to the SMF.

[0146] Step 506: The SMF returns the address information of the AI ​​Service Agent service issued by the PCF or locally configured to the UE in the session context establishment response Nsmf_PDUSession_CreateSMContext Request.

[0147] After the UE receives the address information of the AI ​​Service Agent service issued by the SMF during the session establishment process, it can interact with the AI ​​Service Agent through the data plane channel of the session to query and use the AI ​​services provided by the core network. Here, the UE can also configure the address information of the AI ​​Service Agent service, such as directly configuring the address information of the AIServiceAgent through the setting interface on the terminal side.

[0148] In step 507, the UE establishes a connection with the AI ​​Service Agent through the data plane channel, and sends an AIServiceAgent_AIServiceDiscovery_Request message through the data plane, which is forwarded to the AI ​​Service Agent via the UPF to query available AI services.

[0149] The UE in step 507 above can call the channel and data interface provided by the UE through an APP. The AIServiceAgent_AIServiceDiscovery_Request message is a query request message or a subscription message (the query request message is used as an example below), which can carry some conditional parameters for the AI ​​service required by the terminal (for example, globally defined AI service ID or name, AI service type, APP name and other parameters), so that the core network can filter out the AI ​​services that the terminal is interested in and provide them to the UE. In addition, in actual applications, the AI ​​Service Agent can also be co-located with the UPF, which is not specifically limited here.

[0150] Step 508: The AI ​​Service Agent receives the AI ​​service query request message initiated by the UE, converts it into AIRF's Nairf_AIManage_AIServiceDiscovery_Request, and initiates the AI ​​service query process to the AIRF.

[0151] Step 509: AIRF receives the AI ​​service query request message initiated by the AI ​​Service Agent and determines the AI ​​service list and the instance information corresponding to the AI ​​service.

[0152] Specifically, AIRF can match the conditional parameters of the AI ​​service carried in the query request message, the attribute parameters of the consumer terminal (i.e., UE) itself (for example, IMSI, MSISDN, terminal type, APP name or ID, etc.) and other information with the registration information of each AI service, and feed back the list of available AI services and their instance information that meet the conditions of the AI ​​service provider and consumer to AIServiceAgent.

[0153] Step 510: AIRF returns the AI ​​service query request response message to the AI ​​Service Agent.

[0154] The query request response message can be sent through Nairf_AIManage_AIServiceDiscovery_Response, and the parameters that can be carried include: AI service ID or name, AI service type, AI service function description, AI model type, instance information (the address or fully qualified domain name providing the AI ​​service), and the data format corresponding to the AI ​​model.

[0155] In step 511, the AI ​​Service Agent records the AI ​​service instance information and results carried in the AI ​​service query request response message returned by AIRF locally, and returns the AI ​​service ID or name and the data format corresponding to the AI ​​model, and returns the available AI service information to the UE through AIServiceAgent_AIServiceDiscovery_Response, which carries the AI ​​service ID or name, AI service type, AI service function description, AI model type, instance information, and the data format corresponding to the AI ​​model.

[0156] The data format corresponding to a specific AI model may be an example of a json file definition as shown below. Through hierarchical domain name and domain value pairs, it describes the data items that the model needs to use, as well as the relationship between each data item and the original data, including what feature engineering (including summation, mean, variance, weighting, matrix operations, etc.) calculations and processing need to be performed based on the original data.

[0157] The UE receives the available AI services returned by the AIServiceAgent, and based on the parameters of each AI service, when it needs to use the AI ​​service, it carries the AI ​​service ID or name, original data or normalized data samples, AI training or reasoning operation tasks, and initiates the corresponding AI service training or reasoning application process through the AIServiceAgent and obtains the AI ​​service.

[0158] Specifically, if the UE wants to train an AI model that meets the requirements of accuracy, robustness, and other indicators based on an AI service provided by the core network and local data, it selects an available AI service based on the AI ​​service information returned by the core network (i.e., AIRF) and initiates an AI service training request message to the AIServiceAgent through the AIServiceAgent_AIServiceTrain_Request message. The information carried includes: AI service ID or name, AI training task ID, original data or data samples that have been normalized, AI training operation code, and training task specifications (for example: the minimum number of samples for training startup, training startup time, training duration, number of training iterations, precision, accuracy, and recall rate requirements for successful training, etc.). The AIServiceAgent receives the AI ​​service training request message initiated by the UE, finds the address information of the corresponding AIService based on the AI ​​service ID or name carried in the training request message, and then performs feature engineering on the original data carried in the UE message or verifies the data samples after feature engineering according to the data format of the AI ​​service. After processing the data samples, the Nnwdaf_AIManage_AIServiceTrain_Request is sent to the corresponding AIService instance (NWDAF can also be replaced by an AI service provided by another NF), carrying the AI ​​service ID or name, the AI ​​training task ID, the normalized data samples, and the training task specifications. Upon receiving the AI ​​training request from the AIServiceAgent, the NWDAF or other NF-provided AIService instance allocates the corresponding training resources based on the AI ​​service ID or name, validates the data samples (if the AIServiceAgent and AIService are co-located, data normalization and validation may be combined), and then initiates the training task at an appropriate time based on the training task specifications. After the training task completes and meets the required specifications, the trained AI model (the trained model can also be stored in the AIService instance and assigned an AI model ID and returned to the UE, which can subsequently use this ID to initiate training or inference tasks) is returned to the AIServiceAgent via Nnwdaf_AIManage_AIServiceTrain_Response. (If the training task fails to meet the required specifications after multiple attempts, a training failure result may also be returned to the AIServiceAgent.)

[0159] The AIServiceAgent receives a training completion response message from the AIService instance provided by the NWDAF or other NF, which carries the AI ​​service ID or name, AI training task ID, and training result code. If successful, it can also carry the trained AI model file (of course, the trained AI model can also be delivered to the terminal side through the AIServiceAgent via an independent message) or the assigned AI model ID. The UE determines to continue training or reasoning based on the AI ​​model based on the AI ​​service ID or name, AI training task ID, training result code, AI model, or AI model ID carried in the received training completion response message (local reasoning or using the AIService instance to perform reasoning tasks).

[0160] Specifically, if the UE wants to infer a result based on an AI service provided by the core network and local data, it selects an available AI service based on the AI ​​service information returned by the core network (AI service ID or name (mandatory), AI service type, function description, AI model type, and normalized format definition of the data sample corresponding to the AI ​​model). It sends an AI service inference request message to the AIServiceAgent through the AIServiceAgent_AIServiceInference_Request message. The message carries information including: AI service ID or name, AI inference task ID, raw data or data samples that have been normalized, AI inference opcode, and inference task specifications (for example, the number of batch samples for inference startup, inference startup time, and confidence assessment requirements for inference results). The AIServiceAgent receives the AI ​​service inference request message initiated by the UE, finds the address information of the corresponding AIService based on the AI ​​service ID or name carried in the inference request message, and then performs feature engineering on the raw data carried in the UE message or verifies the data samples after feature engineering according to the data format of the AI ​​service. After processing the data samples, the Nnwdaf_AIManage_AIServiceInference_Request is initiated to the corresponding AIService instance (NWDAF can also be replaced by AI services provided by other NFs), carrying the AI ​​service ID or name, the AI ​​inference task ID, the normalized data sample and its ID (an ID is assigned here to associate the inference result), and the inference task specifications. Upon receiving the AI ​​inference request initiated by the AIServiceAgent, the NWDAF or other AIService instance provided by the NF allocates the corresponding inference resources based on the AI ​​service ID or name carried in the request, verifies the data sample (if the AIServiceAgent and AIService are co-located, data normalization and verification may be combined), and then initiates the inference task at the appropriate time based on the inference task specifications. After the inference task is completed and meets the required specifications, the AI ​​inference task ID and the inference results corresponding to each sample ID are returned to the AIServiceAgent via Nnwdaf_AIManage_AIServiceInference_Response. (If the specification requirements cannot be met after launching the inference task multiple times, the inference failure result of a certain sample ID can also be returned to the AIServiceAgent. Each message can carry multiple sample IDs and their inference results).

[0161] The AIServiceAgent receives an inference completion response message from the AIService instance provided by the NWDAF or other NF, carrying the AI ​​service ID or name, the AI ​​inference task ID, and the inference results corresponding to each sample ID. The UE then performs subsequent processing based on the AI ​​service ID or name, the AI ​​inference task ID, and the inference results corresponding to each sample ID carried in the inference completion response message, completing the interaction process for the AIService inference task.

[0162] In this way, the UE can use the AIServiceAgent to realize functions such as discovery of AI services deployed in the core network, execution of AI training and reasoning tasks, processing of raw data and its normalization, acquisition of AI service training results and models, and reasoning results of each data, thereby achieving the goal of realizing local intelligent data processing by using AI services deployed in the core network.

[0163] Case 2: UE uses AI computing power

[0164] Step 601: After being activated, the UE initiates a session establishment request to the SMF through the RAN and AMF.

[0165] Step 602: The SMF sends a session policy control request message to the PCF.

[0166] Step 603: The PCF sends a response message to the session policy control request message to the SMF.

[0167] Step 604: SMF selects UPF to establish a session for the UE.

[0168] In step 605, the UPF establishes a session for the UE based on the session establishment request message of the SMF, installs the PDR rules (message detection rules) issued by the SMF, performs corresponding perception, statistics and control policies on the user's service flow, and returns an N4 session establishment success response to the SMF.

[0169] Among them, steps 601 to 605 can be understood by referring to steps 501 to 505 in FIG. 5 , and are not described in detail here.

[0170] Step 606: The SMF returns the address information of the AI ​​Service Agent service issued by the PCF or locally configured to the UE in the session context establishment response Nsmf_PDUSession_CreateSMContext Request.

[0171] After the UE receives the address information of the AI ​​Service Agent service issued by the SMF during the session establishment process, it can interact with the AI ​​Service Agent service through the data plane channel of the session to query and use the AI ​​computing power provided by the core network. Here, the UE can also configure the address information of the AI ​​Service Agent service.

[0172] In step 607, the UE establishes a connection with the AI ​​Service Agent through the data plane channel, and sends an AIServiceAgent_AIServiceDiscovery_Request message through the data plane, which is forwarded to the AI ​​Service Agent via the UPF to query the available AI computing power.

[0173] The AIServiceAgent_AIServiceDiscovery_Request message is a query request message or a subscription message (the query request message is used as an example below), which can carry some conditional parameters for the AI ​​computing power required by the terminal (for example: AI computing resource type (X86 / ARM CPU, NPU, GPU, etc., computing resources that can run AI tasks), AI computing power specifications, AI framework types (such as TensorFlow, Mindspore, SKlearn, etc.), AI model categories (such as CNN, RNN, random forest, etc.), operator capabilities (such as vector calculation, matrix calculation, etc.) and other parameters), so that the core network can filter out the AI ​​computing power required by the terminal and provide it to the UE.

[0174] In step 608, the AI ​​Service Agent receives the AI ​​computing power query request message initiated by the UE, converts it into the AIRF's AI computing power query request Nairf_AIManage_AIServiceDiscovery_Request, and initiates the AI ​​computing power query process to the AIRF.

[0175] In step 609, the AIRF receives the AI ​​service query request message initiated by the AI ​​Service Agent and determines the AI ​​computing power provider and the instance information corresponding to the AI ​​computing power.

[0176] Specifically, AIRF matches the information such as the AI ​​computing power condition parameters, AI computing power specifications, AI framework type, AI model category, operator capabilities and other parameters carried in the query request message with the registration information of each AI computing power instance, and provides the AI ​​computing power and its instance information that are available and meet the conditions of AI computing power providers and consumers to AIServiceAgent.

[0177] Step 610: AIRF returns the AI ​​computing power query request response message to the AI ​​Service Agent.

[0178] This query request response message can be sent via Nairf_AIManage_AIServiceDiscovery_Response. Parameters that can be carried include: AI computing power ID or name, AI computing resource type, AI computing power specifications, AI framework type, AI model category, operator capabilities (such as vector calculation, matrix calculation, etc.), and instance information.

[0179] In step 611, the AIServiceAgent returns the AI ​​computing power instance information and results carried in the AI ​​computing power query response message returned by AIRF, and returns the AI ​​computing power ID or name, instance information, and the list of available AI computing power to the UE through AIServiceAgent_AIComputeDiscovery_Response, which carries parameters such as the AI ​​computing power instance ID or name, AI computing resource type, AI computing power specifications, AI framework type, AI model category, and operator capabilities.

[0180] The UE receives the list of available AI computing power returned by the AIServiceAgent, and selects a suitable AI computing power instance based on the parameters of each AI service, including the AI ​​computing power ID or name, AI computing resource type, AI computing power specifications, AI framework type, AI model category, operator capabilities, and other parameters. Then, when the AI ​​computing power instance needs to be used, it carries the AI ​​computing instance ID or name, AI model or AI model ID (at this time, the AI ​​model has been saved in the previous AI computing power instance), the original data (at this time, the normalized format definition description of the data sample corresponding to the AI ​​model is required), or the data sample that has been normalized, and other information to initiate the corresponding AI training or inference application process to the corresponding AI computing power instance through the AIServiceAgent, and obtain the AI ​​processing results.

[0181] Specifically, if the UE wants to train an AI model that meets the requirements of accuracy, robustness and other indicators based on an AI computing power instance, initial model and local data provided by the core network, it selects an available AI computing power instance based on the AI ​​computing power instance information returned by the core network (AI computing power ID or name, AI computing resource type, AI computing power specifications, AI framework type, AI model category, operator capabilities and other parameters), and initiates an AI computing power training request message to the AIServiceAgent through the AIServiceAgent_AIComputeTrain_Request message. The information carried includes: AI computing power instance ID or name, AI training task ID, AI model or AI model ID, original data or data samples that have been normalized, AI training operation code, and training task specifications (for example: the minimum number of samples for training startup, training startup time, training duration, number of training iterations, precision, accuracy, recall rate indicator requirements for successful training, etc.).

[0182] The AIServiceAgent receives the AI ​​service training request message initiated by the UE, finds the address information of the corresponding AIService according to the AI ​​service ID or name carried in the training request message, and then performs feature engineering processing on the original data carried in the UE message according to the data sample normalization format definition description of the AI ​​service, or verifies the data sample after the feature engineering processing. After completing the data sample processing, it initiates an AIManage_AIServiceTrain_Request to the corresponding AIService instance (NWDAF can also be replaced with AI services provided by other NFs), AI computing instance ID or name, AI model or AI model ID (in this case, the AI ​​model has been saved in the previous AI computing instance), AI training task ID, original data (in this case, the data sample normalization format definition description corresponding to the AI ​​model must be carried) or data samples that have been normalized, specification parameters of the training task, and other information to initiate a training task to the corresponding AI computing instance.

[0183] The AICompute instance provided by NWDAF or other NF receives the AI ​​training request initiated by AIServiceAgent, allocates the corresponding training resources based on the AI ​​computing power ID or name it carries, verifies the data samples (if AIServiceAgent and AIService are set up together, data normalization and verification may be integrated), and then starts the training task at the appropriate time based on the AI ​​model or AI model ID, AI training task ID, and training task specification parameters specified in the training task. After the training task is completed and meets the specification parameter requirements, the trained AI model (the trained model can also be saved in the AICompute instance and assigned an AI model ID and returned to the UE. The UE can subsequently start training or inference tasks through this ID) will be returned to the AIServiceAgent through AIManage_AIServiceTrain_Response. (If the training task cannot meet the specification requirements after multiple starts, the result of training failure can also be returned to the AIServiceAgent). The AIServiceAgent receives a training completion response message from the AICompute instance provided by the NWDAF or other NF, which carries the AI ​​service ID or name, AI training task ID, and training result code. If successful, it can also carry the trained AI model file (of course, a separate message can also be used to pass the trained AI model to the terminal side through the AIServiceAgent) or the assigned AI model ID. The UE decides to continue training or reasoning based on the AI ​​model based on the AI ​​service ID or name, AI model ID or training task ID, training result code, AI model or AI model ID carried in the received training completion response message (local reasoning or using the AIService instance to perform reasoning tasks).

[0184] Specifically, if the UE wants to infer a result based on an AI computing power instance provided by the core network and local data, it selects an available AI computing power instance based on the AI ​​computing power instance information returned by the core network (AI computing power ID or name, AI computing resource type, AI computing power specifications, AI framework type, AI model category, operator capabilities and other parameters), and initiates an AI service inference request message to the AIServiceAgent through the AIServiceAgent_AIComputeInference_Request message. The information carried includes: AI computing power instance ID or name, AI model or AI model ID, AI inference task ID, original data (if it is the first time to send, the data sample normalization format definition description information must be carried) or data samples that have been normalized, AI inference operation code, and specification parameters of the inference task (for example: the number of batch samples to start inference, the inference start time, the confidence assessment result requirements of the inference result, etc.).

[0185] The AIServiceAgent receives the AI ​​computing power inference request message initiated by the UE, finds the address information of the corresponding AIService based on the AI ​​computing power instance ID or name carried in the inference request message, and then performs feature engineering on the original data carried in the UE message or verifies the data samples after feature engineering according to the data sample normalization format definition description of the AI ​​model corresponding to the AI ​​inference task. After completing the data sample processing, it initiates an AIManage_AIComputeInference_Request to the corresponding AICompute instance (NWDAF can also be replaced by AI services provided by other NFs), carrying the AI ​​computing power instance ID or name, AI inference task ID, normalized data samples and their IDs (an ID is assigned here to associate inference results), and the specification parameters of the inference task (for example: the number of batch samples to start inference, the inference start time, the confidence assessment result requirements of the inference results, etc.). The AICompute instance provided by NWDAF or other NF receives the AI ​​inference request initiated by AIServiceAgent, allocates the corresponding inference resources based on the AI ​​computing instance ID or name it carries, verifies the data samples (if AIServiceAgent and AIService are set up together, data normalization and verification may be unified), and then starts the inference task at the appropriate time based on the AI ​​model of the inference task and the inference specification parameters (for example: the number of batch samples for inference startup, inference startup time, confidence assessment result requirements for inference results, etc.). After the inference task is completed and meets the specification parameter requirements, the AI ​​inference task ID and the inference result corresponding to each sample ID will be returned to AIServiceAgent through AIManage_AIServiceInference_Response. (If the inference task cannot meet the specification requirements after multiple inference starts, the inference failure result of a certain sample ID can also be returned to AIServiceAgent. Each message can carry multiple sample IDs and their inference results)

[0186] The AIServiceAgent receives an inference completion response message from the AICompute instance provided by the NWDAF or other NF, carrying the AI ​​computing instance ID or name, AI model ID, AI inference task ID, and the inference results corresponding to each sample ID, and returns it to the UE. The UE performs subsequent processing based on the AI ​​computing instance ID or name, AI inference task ID, AI model ID or inference task ID, and the inference results corresponding to each sample ID carried in the received inference completion response message, completing the interaction process for the AIService inference task.

[0187] In this way, the UE can use the AIServiceAgent to discover the AI ​​computing power deployed in the core network, execute AI training and reasoning tasks, process raw data and its normalization, obtain the training results and models of AI services, and obtain the reasoning results of each data, thereby achieving the goal of local intelligent data processing using AI services deployed in the core network.

[0188] Next, Figures 7 to 10 illustrate how AF / NEF / NF uses AI services and AI computing power. This can be achieved through data interaction between NF, AI Service Agent, UPF, and AIRF. In actual application, there are two different implementation methods: one is to use AI services and AI computing power through AI Service Agent, and the other is to use AI services and AI computing power directly. The details are as follows:

[0189] Method 1: AF / NEF / NF uses AI services and AI computing power through AI Service Agent

[0190] Case 1: AF / NEF / NF using AI services

[0191] Here we take NF as an example to illustrate, refer to Figure 7 and execute as follows:

[0192] In step 701, the NF establishes a connection with the AI ​​Service Agent and initiates the process of querying or subscribing to AI services provided by the core network. The NF sends an AIServiceAgent_AIServiceDiscovery_Request message to the AI ​​Service Agent through the data plane to query available AI services.

[0193] The query message may also carry some conditional parameters for the AI ​​services required by the NF (for example, globally defined AI service ID or name, AI service type, APP name or ID, etc.), so that the core network can filter out the AI ​​services that the terminal is concerned about and provide them to the AF / NEF or other NFs.

[0194] Step 702: AIServiceAgent receives the AI ​​service discovery request message initiated by the NF, converts it into AIRF's AI service discovery request Nairf_AIManage_AIServiceDiscovery_Request, and initiates the service discovery process to the AIRF.

[0195] The service discovery request message carries the conditional parameters of the AI ​​service required by the AF / NEF or other NFs, and the attribute parameters of the NF itself (for example, APP name or ID, etc.).

[0196] In step 703, AIRF receives the AI ​​service discovery request message initiated by AIServiceAgent, and matches it with the information registered by each AI service based on the condition parameters of the AI ​​service, the attribute parameters of the NF itself, and other information, and determines the list of available AI services that meet the conditions of AI service providers and consumers and their instance information to AIServiceAgent.

[0197] In step 704, the AI ​​service discovery response message Nairf_AIManage_AIServiceDiscovery_Response returned by AIRF carries the following parameters: AI service ID or name, AI service type, function description, AI model type, instance information (the address or fully qualified domain name providing the AI ​​service), and the normalized format definition description of the data sample corresponding to the AI ​​model.

[0198] In step 705, the AIServiceAgent records the AI ​​service ID or name and the normalized format definition description of the data sample corresponding to the AI ​​model locally based on the AI ​​service instance information and results carried in the AI ​​service discovery response message returned by the AIRF, and returns the AI ​​service that can be provided to the AF / NEF or other NF through the AIServiceAgent_AIServiceDiscovery_Response, carrying the AI ​​service ID or name, AI service type, function description, AI model type, instance information, and the normalized format definition description of the data sample corresponding to the AI ​​model.

[0199] After receiving the available AI services returned by the AIServiceAgent, the NF selects the appropriate AI service based on the parameters of each AI service. Then, when the AI ​​service needs to be used, it carries the AI ​​service ID or name, raw data or normalized data samples, and AI training or inference operation tasks. The NF executes the corresponding AI service training or inference process through the AIServiceAgent and obtains the AI ​​service.

[0200] Specifically, the NF can refer to the process of UE using AI services in Figure 5 above to obtain the AI ​​training model and processing results, which will not be repeated here.

[0201] In this way, AF / NEF or other NFs can use AIServiceAgent to realize functions such as discovery of AI services deployed in the core network, execution of AI training and inference tasks, processing of raw data and its normalization, acquisition of AI service training results and models, and inference results of each data, thereby achieving the goal of realizing local intelligent data processing by using AI services deployed in the core network.

[0202] Case 2: AF / NEF / NF uses AI computing power

[0203] Here we take NF as an example to illustrate, refer to Figure 8 and execute as follows:

[0204] In step 801, the NF establishes a connection with the AI ​​Service Agent and initiates the process of querying or subscribing to the AI ​​computing power provided by the core network. The NF sends the AIServiceAgent_AIComputeDiscovery_Request message to the AI ​​Service Agent through the data plane to query the available AI computing power.

[0205] The query message may also carry some conditional parameters for the AI ​​computing power required by AF / NEF or other NFs (for example, AI computing resource type (computing resources such as X86 / ARM CPU, NPU, GPU, etc. that can run AI tasks), AI computing power specifications, AI framework type (such as TensorFlow, Mindspore, SKlearn, etc.), AI model category (such as CNN, RNN, random forest, etc.), operator capabilities (such as vector calculation, matrix calculation, etc.) and instance information (the address or fully qualified domain name providing AI services), so that the core network can filter out the AI ​​computing power required by AF / NEF or other NFs and provide it to AF / NEF or other NFs for use.

[0206] In step 802, AIServiceAgent receives the AI ​​computing power discovery request message initiated by NF, converts it into AIRF's AI computing power discovery request Nairf_AIManage_AIComputeDiscovery_Request, and initiates the service discovery process to AIRF.

[0207] The computing power discovery request message carries the conditional parameters of the AI ​​computing power required by the terminal.

[0208] In step 803, AIRF receives the AI ​​computing power discovery request message initiated by AIServiceAgent, matches the AI ​​computing power condition parameters carried by it with the registration information of each AI computing power instance, and determines to provide the AI ​​computing power and its instance information that meet the conditions of AI computing power providers and consumers to AIServiceAgent.

[0209] Step 804: The AI ​​computing power discovery response message Nairf_AIManage_AIComputeDiscovery_Response returned by AIRF carries the conditional parameters of the AI ​​computing power.

[0210] In step 805, based on the AI ​​service instance information and results carried in the AI ​​service discovery response message returned by AIRF, AIServiceAgent returns the AI ​​service ID or name, instance information, and the list of available AI computing power to NF through AIServiceAgent_AIComputeDiscovery_Response, which carries parameters such as AI computing power instance ID or name, AI computing resource type, AI computing power specification, AI framework type, AI model category, and operator capability.

[0211] NF receives the list of available AI computing power returned by AIServiceAgent and selects a suitable AI computing power instance based on the condition parameters of each AI computing power. Then, when the AI ​​computing power instance needs to be used, it carries the AI ​​computing power instance ID or name, AI model or AI model ID (in this case, the AI ​​model has been saved in the previous AI computing power instance), original data (in this case, the normalized format definition description of the data sample corresponding to the AI ​​model is required) or data samples that have been normalized, etc., to initiate the corresponding AI training or inference application process to the corresponding AI computing power instance through AIServiceAgent and obtain the AI ​​processing results.

[0212] Specifically, if the NF can refer to the process of UE using AI computing power in Figure 6 above to obtain the AI ​​training model and processing results, it will not be repeated here.

[0213] In this way, NF can use AIServiceAgent to realize functions such as discovering AI computing power deployed in the core network, executing AI training and inference tasks, processing raw data and its normalization, obtaining AI service training results and models, and inference results for each data, thereby achieving the goal of realizing local intelligent data processing by using AI services deployed in the core network.

[0214] Method 1: AF / NEF / NF directly uses AI services and AI computing power

[0215] Case 1: AF / NEF / NF using AI services

[0216] Here we take NF as an example, and refer to Figure 9 to perform the following:

[0217] Step 901: NF establishes a connection with AIRF, and AI service discovery request Nairf_AIManage_AIServiceDiscovery_Request initiates the service discovery process to AIRF.

[0218] The service discovery request message carries the conditional parameters of the AI ​​service required by the NF (for example, globally defined AI service ID or name, AI service type, and other parameters) and the attribute parameters of the NF itself (for example, APP name or ID, etc.), so that the core network can filter out the AI ​​services that the terminal is interested in and provide them to the NF.

[0219] In step 902, the AIRF receives the AI ​​service discovery request message initiated by the NF, and matches it with the information registered by each AI service based on the condition parameters of the AI ​​service carried by it, the attribute parameters of the NF itself, and other information, and determines the list of available AI services that meet the conditions of AI service providers and consumers and their instance information to the NF.

[0220] Step 903: AIRF returns an AI service discovery response message.

[0221] The service discovery response message is sent through Nairf_AIManage_AIServiceDiscovery_Response, and can carry parameters including: AI service ID or name, AI service type, function description, AI model type, instance information (the address or fully qualified domain name providing the AI ​​service), and the normalized format definition description of the data sample corresponding to the AI ​​model.

[0222] Based on the AI ​​service instance information and results carried in the AI ​​service discovery response message returned by AIRF, NF records the AI ​​service ID or name, and the normalized format definition description of the data sample corresponding to the AI ​​model locally, and selects the appropriate AI service based on the parameters of each AI service. Then, when the AI ​​service needs to be used, it carries the AI ​​service ID or name, original data or normalized data samples, AI training or reasoning operation tasks to initiate the corresponding AI service training or reasoning application process and obtain the AI ​​service.

[0223] Specifically, if the NF wants to train an AI model that meets the requirements of accuracy, robustness, and other indicators based on an AI service provided by the core network and local data, it selects an available AI service based on the AI ​​service information returned by the core network (AI service ID or name, AI service type, function description, AI model type, and the normalized format definition of the data sample corresponding to the AI ​​model), and performs feature engineering on the original data according to the normalized format definition of the data sample of the AI ​​service. After completing the data sample processing, it initiates Nnwdaf_AIManage_AIServiceTrain_Request to the corresponding AIService instance (NWDAF can also be replaced by AI services provided by other NFs), carrying the AI ​​service ID or name, AI training task ID, normalized data samples, and training task specifications (for example: the minimum number of samples for training startup, training startup time, training duration, number of training iterations, precision, accuracy, and recall rate requirements for successful training, etc.).

[0224] The AIService instance provided by NWDAF or other NF receives the AI ​​training request initiated by AF / NEF or other NF, allocates the corresponding training resources based on the AI ​​service ID or name it carries, verifies the data samples, and then starts the training task at the appropriate time based on the specifications of the training task (for example: the minimum number of samples for training startup, training startup time, training duration, number of training iterations, accuracy, precision, recall rate requirements for successful training, etc.). After the training task is completed and meets the specification parameter requirements, the trained AI model (the trained model can also be saved in the AIService instance and assigned an AI model ID and returned to AF / NEF or other NF. Subsequently, AF / NEF or other NF can use this ID to start training or inference tasks) will be returned to AF / NEF or other NF through Nnwdaf_AIManage_AIServiceTrain_Response. (If the training task cannot meet the specification requirements after multiple starts, the training failure result can also be returned to AF / NEF or other NF)

[0225] The AF / NEF or other NF decides whether to continue training or reasoning based on the AI ​​model (local reasoning or using the AIService instance to perform reasoning tasks) based on the AI ​​service ID or name, AI training task ID, training result code, AI model or AI model ID carried in the received training completion response message.

[0226] Specifically, if AF / NEF or other NF wants to infer a result based on an AI service provided by the core network and local data, it selects an available AI service based on the AI ​​service information returned by the core network (AI service ID or name, AI service type, function description, AI model type, and the normalized format definition of the data sample corresponding to the AI ​​model). After completing the data sample processing, it initiates Nnwdaf_AIManage_AIServiceInference_Request to the corresponding AIService instance (NWDAF can also be replaced with AI services provided by other NFs), carrying the AI ​​service ID or name, AI reasoning task ID, normalized data samples and their IDs (an ID is assigned here to associate reasoning results), and the specification parameters of the reasoning task (for example: the number of batch samples to start reasoning, the reasoning start time, the confidence assessment result requirements of the reasoning results, etc.).

[0227] The AIService instance provided by NWDAF or another NF receives an AI inference request initiated by an AF / NEF or other NF, allocates the corresponding inference resources based on the AI ​​service ID or name carried in the request, verifies the data samples (if the AIServiceAgent and AIService are co-located, data normalization and verification may be combined), and then initiates the inference task at the appropriate time based on the inference task specifications (for example, the number of batch samples to initiate inference, the inference startup time, and the confidence assessment requirements for the inference results). After the inference task is completed and meets the required specifications, the AI ​​inference task ID and the inference results corresponding to each sample ID are returned to the AF / NEF or other NF via Nnwdaf_AIManage_AIServiceInference_Response. (If the inference task cannot meet the specifications after multiple inference task launches, the inference failure result for a specific sample ID can also be returned to the AIServiceAgent. Each message can carry multiple sample IDs and their inference results.) AF / NEF or other NFs perform subsequent processing based on the AI ​​service ID or name, AI reasoning task ID, and reasoning results corresponding to each sample ID carried in the received reasoning completion response message to complete the interaction process of the AIService reasoning task.

[0228] Case 2: AF / NEF / NF uses AI computing power

[0229] Here we take NF as an example, and refer to Figure 10 to perform the following operations:

[0230] Step 1001: NF establishes a connection with AIRF and initiates a service discovery process to AIRF through the AI ​​computing power discovery request Nairf_AIManage_AIComputeDiscovery_Request.

[0231] The service discovery request message carries the conditional parameters of the AI ​​computing power required by the terminal (for example, AI computing resource type (X86 / ARM CPU, NPU, GPU, etc., computing power resources that can run AI tasks), AI computing power specifications, AI framework type (such as TensorFlow, Mindspore, SKlearn, etc.), AI model category (such as CNN, RNN, random forest, etc.), operator capabilities (such as vector calculation, matrix calculation, etc.), and other parameters).

[0232] In step 1002, AIRF receives the AI ​​computing power discovery request message initiated by NF, and matches it with the information of each AI computing power instance registration based on the AI ​​service condition parameters, AI computing power specifications, AI framework type, AI model category, operator capabilities and other parameters carried by the message, and determines the AI ​​computing power and its instance information that are available and meet the conditions of AI computing power providers and consumers to AIServiceAgent.

[0233] Step 1003: AIRF returns an AI computing power discovery response message.

[0234] The AI ​​computing power discovery response message is sent through Nairf_AIManage_AIComputeDiscovery_Response and can carry parameters including: AI computing power ID or name, AI computing resource type (computing resources that can run AI tasks, such as X86 / ARM CPU, NPU, GPU, etc.), AI computing power specifications, AI framework type (such as TensorFlow, Mindspore, SKlearn, etc.), AI model category (such as CNN, RNN, random forest, etc.), operator capabilities (such as vector calculation, matrix calculation, etc.), and instance information (the address or fully qualified domain name providing AI services).

[0235] After receiving the list of available AI computing power returned by AIRF, NF selects a suitable AI computing power instance based on the parameters of each AI service, including AI computing power ID or name, AI computing resource type (X86 / ARM CPU, NPU, GPU, etc., which can run AI tasks), AI computing power specifications, AI framework type (such as TensorFlow, Mindspore, SKlearn, etc.), AI model category (such as CNN, RNN, Random Forest, etc.), operator capabilities (such as vector calculation, matrix calculation, etc.). Then, when the AI ​​computing power instance needs to be used, it carries the AI ​​computing power instance ID or name, AI model or AI model ID (in this case, the AI ​​model has been saved in the previous AI computing power instance), raw data (in this case, the normalized format definition description of the data sample corresponding to the AI ​​model is required), or data samples that have undergone normalization processing, etc., to initiate the corresponding AI training or inference application process to the corresponding AI computing power instance and obtain the AI ​​processing results.

[0236] Specifically, if AF / NEF or other NF wants to train an AI model that meets the requirements of accuracy, robustness and other indicators based on an AI computing power instance, initial model and local data provided by the core network, based on the AI ​​computing power instance information returned by the core network (AI computing power ID or name, AI computing resource type (X86 / ARM The system selects an available AI computing instance based on parameters such as CPU, NPU, GPU, etc., AI computing power resources that can run AI tasks, AI computing power specifications, AI framework types (such as TensorFlow, MindSpore, SKlearn, etc.), AI model categories (such as CNN, RNN, Random Forest, etc.), and operator capabilities (such as vector calculation, matrix calculation, etc.). After completing data sample processing, it initiates an AIManage_AIServiceTrain_Request to the corresponding AIService instance (NWDAF can also be replaced with AI services provided by other NFs). The AI ​​computing instance ID or name, AI model or AI model ID (in this case, the AI ​​model has been saved in the previous AI computing power instance), AI training task ID, raw data (in this case, the normalized format definition of the data samples corresponding to the AI ​​model must be included) or data samples that have undergone normalization processing, and training task specifications (for example, the minimum number of samples for training startup, training startup time, training duration, number of training iterations, and the precision, accuracy, and recall rate requirements for successful training) are used to initiate a training task to the corresponding AI computing power instance.

[0237] The AICompute instance provided by NWDAF or other NF receives the AI ​​training request initiated by AF / NEF or other NF, allocates the corresponding training resources based on the AI ​​computing power ID or name it carries, verifies the data samples, and then starts the training task at the appropriate time based on the AI ​​model or AI model ID, AI training task ID, and training task specification parameters specified in the training task (for example: the minimum number of samples for training startup, training startup time, training duration, number of training iterations, accuracy, precision, recall rate requirements for successful training, etc.). After the training task is completed and meets the specification parameter requirements, the trained AI model (the trained model can also be saved in the AICompute instance and assigned an AI model ID and returned to the UE. The UE can subsequently start training or inference tasks through this ID) will be returned to the AF / NEF or other NF through AIManage_AIServiceTrain_Response. (If the training task cannot meet the specification requirements after multiple starts, the training failure result can also be returned to the AF / NEF or other NF)

[0238] AF / NEF or other NFs decide whether to continue training or reasoning based on the AI ​​model based on the AI ​​service ID or name, AI model ID or training task ID, training result code, AI model or AI model ID carried in the received training completion response message (local reasoning or using AI computing power instances to perform reasoning tasks).

[0239] Specifically, if AF / NEF or other NF wants to infer a result based on an AI computing power instance provided by the core network and local data, it selects an available AI computing power instance based on the AI ​​computing power instance information returned by the core network (AI computing power ID or name, AI computing resource type (X86 / ARM CPU, NPU, GPU, etc. that can run AI tasks), AI computing power specifications, AI framework type (such as TensorFlow, Mindspore, SKlearn, etc.), AI model category (such as CNN, RNN, random forest, etc.), operator capabilities (such as vector calculation, matrix calculation, etc.) and other parameters). After completing the data sample processing, it initiates AIManage_AIComputeInference_Request to the corresponding AICompute instance (NWDAF can also be replaced by AI services provided by other NFs), carrying the AI ​​computing power instance ID or name, AI inference task ID, normalized data sample and its ID (an ID is assigned here to associate the inference result), and the specification parameters of the inference task (for example: the number of batch samples for inference startup, inference startup time, confidence assessment result requirements for inference results, etc.).

[0240] The AICompute instance provided by NWDAF or other NF receives the AI ​​inference request initiated by AF / NEF or other NF, allocates the corresponding inference resources based on the AI ​​computing instance ID or name it carries, verifies the data samples (if AIServiceAgent and AIService are set up together, data normalization and verification may be unified), and then starts the inference task at the appropriate time based on the AI ​​model of the inference task and the inference specification parameters (for example: the number of batch samples for inference startup, inference startup time, confidence assessment result requirements for inference results, etc.). After the inference task is completed and meets the specification parameter requirements, the AI ​​inference task ID and the inference result corresponding to each sample ID will be returned to AF / NEF or other NF through AIManage_AIServiceInference_Response. (If the specification requirements cannot be met after starting the inference task multiple times, the inference failure result of a certain sample ID can also be returned to AF / NEF or other NF. Each message can carry multiple sample IDs and their inference results)

[0241] AF / NEF or other NFs perform subsequent processing based on the AI ​​computing instance ID or name, AI inference task ID, AI model ID or inference task ID carried in the received inference completion response message, and the inference result corresponding to each sample ID to complete the interaction process of the AIService inference task.

[0242] The above mainly introduces the solution provided by the embodiment of the present application from the perspective of device interaction. It is understandable that, in order to implement the above functions, each device may include a hardware structure and / or software module that performs each function. Those skilled in the art should easily appreciate that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the embodiments of the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in a hardware or computer software driven hardware manner depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0243] In the embodiments of the present application, the functional units of the device can be divided according to the above method examples. For example, each functional unit can be divided according to each function, or two or more functions can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or software functional units.

[0244] In the case of adopting an integrated unit, Figure 11 shows a possible exemplary block diagram of the communication device involved in the embodiments of the present application. As shown in Figure 11, the communication device 1100 may include: a processing unit 1101 and a transceiver unit 1102. The processing unit 1101 is used to control and manage the actions of the communication device 1100. The transceiver unit 1102 is used to support communication between the communication device 1100 and other devices. Optionally, the transceiver unit 1102 may include a receiving unit and / or a sending unit, which are used to perform receiving and sending operations respectively. Optionally, the communication device 1100 may also include a storage unit for storing program code and / or data of the communication device 1100. The transceiver unit may be referred to as an input / output unit, a communication unit, etc., and the transceiver unit may be a transceiver; the processing unit may be a processor. When the communication device is a module (such as a chip) in a communication device, the transceiver unit may be an input / output interface, an input / output circuit, or an input / output pin, etc., and may also be referred to as an interface, a communication interface, or an interface circuit, etc.; the processing unit may be a processor, a processing circuit, or a logic circuit, etc. Specifically, the device may be the first network function, the second network function, the third network function, the fourth network function, etc. mentioned above.

[0245] In one embodiment, the communication device 1100 is a first network function, and the transceiver unit 1102 is used to receive a request message from a third network function, where the request message is used to request the access address of the second network function associated with the target artificial intelligence (AI) information; the processing unit 1101 is used to query the association relationship according to the request message to obtain the access address of the second network function associated with the target AI information, where the association relationship indicates the relationship between at least one AI information supported by at least one second network function and the access address of the second network function associated with each AI information, and the target AI information is one of the at least one AI information; the transceiver unit 1102 is also used to send a request response message to the third network function, where the request response message is used to indicate the access address of the second network function associated with the target AI information.

[0246] In an optional manner, the request message is a subscription request message; the transceiver unit 1102 is configured to send a notification message of successful subscription to the third network function, and send an access address of the second network function associated with at least one AI information to the third network function.

[0247] In an optional manner, the request message is a query request message; the transceiver unit 1102 is configured to send the access address of the second network function associated with the target AI information to the third network function.

[0248] In an optional manner, the transceiver unit 1102 is configured to receive a registration request message from at least one second network function, where the registration request message is used to indicate at least one AI information supported by the second network function and an access address of the second network function associated with each AI information; and the processing unit 1101 is configured to determine an association relationship based on the registration request message.

[0249] In an optional manner, the transceiver unit 1102 is further used to send a monitoring message to the second network function, where the monitoring message is used to monitor the working status of the second network function; the transceiver unit 1102 is further used to receive a monitoring response message from the second network function.

[0250] In an optional manner, the third network function is a terminal; the transceiver unit 1102 is further used to receive a request message sent by the third network function through a fourth network function, and the fourth network function is used to convert the format of the message exchanged between the terminal and the first network function, and / or perform identity authentication on the terminal and the first network function.

[0251] In an optional manner, the AI ​​information includes: AI services, and / or AI computing power.

[0252] In an optional manner, the AI ​​information is an AI service; the registration request message includes: an identifier of the AI ​​service and an access address of a second network function that provides the AI ​​service.

[0253] In an optional manner, the registration request information further includes one or more of the following:

[0254] AI service categories, model categories supported by AI services, service objects of AI services, data formats supported by AI services, and data computing scale of AI services;

[0255] Among them, the service objects of AI services include terminal device type, APP name, terminal users belonging to the preset number segment, geographic location information, slice information and access point identification.

[0256] In an optional manner, the AI ​​information is AI computing power; the registration request information includes: an identifier of the AI ​​computing power and an access address of a second network function that provides the AI ​​computing power.

[0257] In an optional manner, the registration request information further includes one or more of the following:

[0258] The AI ​​computing power type, AI computing power specifications, AI framework type indicated by the AI ​​computing power, AI model category indicated by the AI ​​computing power, and operator information of the AI ​​computing power.

[0259] In another embodiment, the communication device 1100 is a fourth network function, and the transceiver unit 1102 is used to receive a request message from a third network function, where the request message is used to request the access address of the second network function associated with the target artificial intelligence AI information; the transceiver unit 1102 is also used to send a request message to the first network function; receive a request response message from the first network function, where the request response message is used to indicate the access address of the second network function associated with the target AI information; and send a request response message to the third network function.

[0260] In an optional manner, the request message is a subscription request message; the transceiver unit 1102 is configured to receive a notification message of successful subscription and receive an access address of a second network function associated with at least one AI information from the first network function.

[0261] In an optional manner, the request message is a query request message; the transceiver unit 1102 is configured to receive an access address of a second network function associated with target AI information from a first network function.

[0262] In an optional manner, the third network function is a terminal; the processing unit 1101 is configured to convert the format of messages exchanged between the third network function and the first network function; and / or perform identity authentication on the third network function and the first network function.

[0263] In an optional manner, the transceiver unit 1102 is used to receive data to be processed from the terminal; convert the data to be processed into a data format corresponding to the AI ​​information required by the terminal and then send it to the access address of the second network function associated with the AI ​​information required by the terminal; receive the data processing result from the second network function; and send the data processing result to the terminal.

[0264] In another embodiment, the communication device 1100 is a third network function, and the transceiver unit 1102 is used to send a request message to a fourth network function, where the request message is used to request the access address of the second network function associated with the target artificial intelligence AI information; the transceiver unit 1102 is also used to receive a request response message from the fourth network function, where the request response message is used to indicate the access address of the second network function associated with the target AI information; send the data to be processed to the access address of the second network function associated with the target AI information through the fourth network function; and receive the data processing result from the second network function through the fourth network function.

[0265] In an optional manner, the communication device 1100 is a terminal.

[0266] In another embodiment, the communication device 1100 is a second network function, and the transceiver unit 1102 is used to send a registration request message to the first network function, where the registration request message is used to indicate at least one AI information supported by the second network function and an access address of the second network function associated with each AI information; the transceiver unit 1102 is also used to receive a response message of successful registration from the second network function.

[0267] In an optional manner, the transceiver unit 1102 is also used to receive a monitoring message from the first network function, which is used to monitor the working status of the second network function; and send a monitoring response message to the first network function, which includes the working status of the second network function.

[0268] In an optional manner, the AI ​​information includes: AI services, and / or AI computing power.

[0269] In an optional manner, the AI ​​service includes one or more of the following: an identifier of the AI ​​service, an access address of a second network function providing the AI ​​service, a category of the AI ​​service, a category of models supported by the AI ​​service, a service object of the AI ​​service, a data format supported by the AI ​​service, and a data calculation scale of the AI ​​service; wherein the service objects of the AI ​​service include terminal device type, APP name, terminal users belonging to a preset number segment, geographic location information, slice information, and access point identifier.

[0270] In one optional embodiment, AI computing power includes one or more of the following:

[0271] AI computing identifier, AI computing power type, AI computing power specifications, AI framework type indicated by AI computing power, AI model category indicated by AI computing power, and operator information of AI computing power.

[0272] As shown in Figure 12, this application also provides a communication device 1200. Communication device 1200 can be a chip or a chip system. The communication device can be located in the device involved in any of the above method embodiments, such as an access network device or a first core network device, to perform the corresponding actions of the device.

[0273] Optionally, the chip system may consist of the chip, or may include the chip and other discrete devices.

[0274] The communication device 1200 includes a processor 1210 .

[0275] The processor 1210 is configured to execute the computer program stored in the memory 1220 to implement the actions of each device in any of the above method embodiments.

[0276] The communication device 1200 may further include a memory 1220 for storing computer programs.

[0277] Optionally, the memory 1220 and the processor 1210 are coupled. Coupling is an indirect coupling or communication connection between devices, units, or modules, which can be electrical, mechanical, or other forms, and is used for information exchange between the devices, units, or modules. Optionally, the memory 1220 and the processor 1210 are integrated.

[0278] The processor 1210 and the memory 1220 can be one or more without limitation.

[0279] Optionally, in actual applications, the communication device 1200 may or may not include a transceiver 1230, as illustrated by a dashed box in the figure. The communication device 1200 can exchange information with other devices via the transceiver 1230. The transceiver 1230 can be a circuit, a bus, a transceiver, or any other device capable of exchanging information.

[0280] In a possible implementation, the communication device 1200 may be an access network device or a first core network device in the implementation of the above methods.

[0281] The specific connection medium between the transceiver 1230, processor 1210, and memory 1220 is not limited in the embodiments of the present application. In FIG12 , the memory 1220, processor 1210, and transceiver 1230 are connected via a bus. The bus is represented by a bold line in FIG10 . The connection between other components is for illustrative purposes only and is not intended to be limiting. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, FIG12 uses only a single bold line, but this does not imply that there is only one bus or a single type of bus. In the embodiments of the present application, the processor can be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component, and can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of the present application can be directly executed by a hardware processor or by a combination of hardware and software modules within the processor.

[0282] In an embodiment of the present application, the memory may be a non-volatile memory, such as a hard disk drive (HDD) or a solid-state drive (SSD), etc., or a volatile memory (volatile memory), such as a random-access memory (RAM). The memory may also be any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory in an embodiment of the present application may also be a circuit or any other device that can implement a storage function, for storing computer programs, program instructions and / or data.

[0283] Based on the above embodiments, referring to FIG13 , the embodiment of the present application also provides another communication device 1300, including: an interface circuit 1310 and a logic circuit 1320; the interface circuit 1310 can be understood as an input and output interface, which can be used to execute the receiving and sending steps of each device in any of the above method embodiments, and the logic circuit 1320 can be used to run code or instructions to execute the method executed by each device in any of the above embodiments, which will not be repeated.

[0284] Based on the above embodiments, embodiments of the present application further provide a computer-readable storage medium storing instructions that, when executed, cause the method executed by each device in any of the above method embodiments to be implemented. The computer-readable storage medium may include any medium capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory, a random access memory, a magnetic disk, or an optical disk.

[0285] Based on the above embodiments, an embodiment of the present application provides a communication system, which includes the first network function, the second network function, the third network function and the fourth network function mentioned in any of the above method embodiments, and can be used to execute the method executed by each device in any of the above method embodiments.

[0286] In addition, the above-mentioned communication system may further include a terminal (eg, UE), which may perform data interaction with an access network device and a core network device to execute a related method in any of the above-mentioned method embodiments.

[0287] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, compact disc read-only memory (CD-ROM), optical storage, etc.) containing computer-usable program code.

[0288] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0289] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0290] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

Claims

1. A communication method, applied to a first network function, characterized in that: include: Receiving a request message from a third network function, the request message being used to request an access address of a second network function associated with target artificial intelligence AI information; querying the association relationship according to the request message to obtain an access address of the second network function associated with the target AI information, wherein the association relationship indicates a relationship between at least one AI information supported by at least one second network function and an access address of the second network function respectively associated with each of the AI ​​information, and the target AI information is one of the at least one AI information; Send a request response message to the third network function, where the request response message is used to indicate an access address of the second network function associated with the target AI information.

2. The method according to claim 1, characterized in that The request message is a subscription request message; The sending the request response message to the third network function includes: A notification message of successful subscription is sent to the third network function, and an access address of the second network function associated with the at least one AI information is sent to the third network function.

3. The method according to claim 1, characterized in that The request message is a query request message; The sending the request response message to the third network function includes: The access address of the second network function associated with the target AI information is sent to the third network function.

4. The method according to any one of claims 1 to 3, characterized in that: The method further comprises: receiving a registration request message from at least one of the second network functions, where the registration request message is used to indicate the at least one AI information supported by the second network function and an access address of the second network function respectively associated with each of the AI ​​information; The association relationship is determined according to the registration request message.

5. The method according to claim 4, characterized in that The method further comprises: Sending a monitoring message to the second network function, where the monitoring message is used to monitor the working status of the second network function; A monitoring response message is received from the second network function.

6. The method according to any one of claims 1 to 5, characterized in that: The third network function is a terminal; The receiving a request message from a third network function includes: and receiving a request message sent by the third network function through a fourth network function, where the fourth network function is used to perform format conversion on messages exchanged between the terminal and the first network function, and / or to perform identity authentication on the terminal and the first network function.

7. The method according to any one of claims 1 to 6, characterized in that: The AI ​​information includes: AI services, and / or AI computing power.

8. The method according to claim 4, characterized in that The AI ​​information is an AI service; the registration request message includes: The identifier of the AI ​​service and the access address of a second network function that provides the AI ​​service.

9. The method according to claim 8, characterized in that The registration request information may also include one or more of the following: The category of the AI ​​service, the model category supported by the AI ​​service, the service object of the AI ​​service, the data format supported by the AI ​​service, and the data calculation scale of the AI ​​service; Among them, the service objects of the AI ​​service include terminal device type, APP name, terminal users belonging to a preset number segment, geographic location information, slice information and access point identification.

10. The method according to claim 4, characterized in that The AI ​​information is AI computing power; the registration request information includes: an identifier of the AI ​​computing power and an access address of a second network function that provides the AI ​​computing power.

11. The method according to claim 10, characterized in that The registration request information may also include one or more of the following: The computing power type of the AI ​​computing power, the computing power specification of the AI ​​computing power, the AI ​​framework type indicated by the AI ​​computing power, the AI ​​model category indicated by the AI ​​computing power, and the operator information of the AI ​​computing power.

12. A communication method, applied to a fourth network function, characterized in that: include: Receiving a request message from a third network function, the request message being used to request an access address of a second network function associated with target artificial intelligence AI information; Sending the request message to the first network function; receiving a request response message from the first network function, where the request response message is used to indicate an access address of the second network function associated with the target AI information; The request response message is sent to the third network function.

13. The method according to claim 12, characterized in that The request message is a subscription request message; The receiving a request response message from the first network function includes: A notification message of successful subscription is received, and an access address of a second network function associated with at least one AI information from the first network function is received.

14. The method according to claim 12, characterized in that The request message is a query request message; The receiving a request response message from the first network function includes: An access address of a second network function associated with the target AI information is received from the first network function.

15. The method according to any one of claims 12 to 14, characterized in that: The third network function is a terminal; The method further comprises: performing format conversion on messages exchanged between the third network function and the first network function; and / or, Perform identity authentication on the third network function and the first network function.

16. The method according to claim 15, characterized in that Also includes: receiving data to be processed from the terminal; Converting the data to be processed into a data format corresponding to the AI ​​information required by the terminal and then sending the data to an access address of a second network function associated with the AI ​​information required by the terminal; receiving a data processing result from the second network function; The data processing result is sent to the terminal.

17. A communication method, applied to a third network function, characterized in that: include: Sending a request message to the fourth network function, wherein the request message is used to request an access address of the second network function associated with the target artificial intelligence AI information; receiving a request response message from the fourth network function, where the request response message is used to indicate an access address of the second network function associated with the target AI information; Sending the data to be processed to the access address of the second network function associated with the target AI information through the fourth network function; The data processing result from the second network function is received through the fourth network function.

18. The method according to claim 17, characterized in that The third network function is a terminal.

19. A communication method, applied to a second network function, characterized in that: include: Sending a registration request message to the first network function, where the registration request message is used to indicate at least one AI information supported by the second network function and an access address of the second network function associated with each AI information; A registration success response message is received from the second network function.

20. The method according to claim 19, characterized in that The method further comprises: receiving a monitoring message from the first network function, where the monitoring message is used to monitor a working status of the second network function; A monitoring response message is sent to the first network function, where the monitoring response message includes the working status of the second network function.

21. The method according to claim 19 or 20, characterized in that The AI ​​information includes: AI services, and / or AI computing power.

22. The method according to claim 21, characterized in that The AI ​​service includes one or more of the following: an identifier of the AI ​​service, an access address of a second network function providing the AI ​​service, a category of the AI ​​service, a model category supported by the AI ​​service, a service object of the AI ​​service, a data format supported by the AI ​​service, and a data calculation scale of the AI ​​service; Among them, the service objects of the AI ​​service include terminal device type, APP name, terminal users belonging to a preset number segment, geographic location information, slice information and access point identification.

23. The method according to claim 21, characterized in that The AI ​​computing power includes one or more of the following: The identifier of the AI ​​computing power, the computing power type of the AI ​​computing power, the computing power specification of the AI ​​computing power, the AI ​​framework type indicated by the AI ​​computing power, the AI ​​model category indicated by the AI ​​computing power, and the operator information of the AI ​​computing power.

24. A communication device, characterized in that: include: A functional module for implementing the method according to any one of claims 1 to 23.

25. A communication device, characterized in that: include: at least one processor and memory; The memory is used to store computer programs or instructions; The at least one processor is configured to execute the computer program or instructions so that the method according to any one of claims 1 to 23 is performed.

26. A chip system, characterized in that: The chip system comprises: a processing circuit; the processing circuit is coupled to a storage medium; The processing circuit is used to execute part or all of the computer programs or instructions in the storage medium, and when the part or all of the computer programs or instructions are executed, it is used to implement the method according to any one of claims 1 to 23.

27. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed by a computer, the method according to any one of claims 1 to 23 is executed.

28. A computer program product comprising a computer program or instructions, characterized in that When it is run on a computer, the method described in any one of claims 1 to 23 is executed.

Images