Data transmission methods, apparatuses, storage medium and program product

Abstract

Provided are data transmission methods, apparatuses, a storage medium and a program product, relating to the technical field of communications. A method comprises: receiving first information from a second node , the first information being used for indicating an AI data transmission configuration; and transmitting AI data on the basis of the AI data transmission configuration.

Description

Data transmission methods, devices, storage media and software products

[0001] This disclosure claims priority to Chinese patent application No. 202510112895.6, filed on January 22, 2025, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This disclosure relates to the field of communication technology, and in particular to a data transmission method, apparatus, storage medium, and program product. Background Technology

[0003] As the number of digital data applications and services continues to surge, the demands and challenges on network resources and operators will continue to increase. Delivering the diverse network performance characteristics required for future services is one of the major technical challenges facing service providers today. Network performance requirements primarily include connection data rates, latency, quality of service (QoS), security, and availability, and these requirements vary from service to service.

[0004] Therefore, on the one hand, future communication networks (such as 6G) will be deeply integrated with artificial intelligence (AI), becoming an inherent capability of the next-generation network and building a more intelligent, efficient, and secure mobile communication network. In future communication systems, how to improve the transmission efficiency of AI data has become an urgent problem to be solved. Summary of the Invention

[0005] Firstly, a data transmission method is provided, applied to a first node, including:

[0006] Receive first information from the second node, which is used to indicate the AI ​​data transmission configuration;

[0007] AI data is transmitted based on the AI ​​data transmission configuration.

[0008] Secondly, a data transmission method is provided, applied to a second node, including:

[0009] Send the first message to the first node. The first message is used to indicate the AI ​​data transmission configuration.

[0010] Thirdly, a communication device is provided for use in a first node, comprising:

[0011] The receiving unit receives first information from the second node, which is used to indicate the AI ​​data transmission configuration.

[0012] The sending unit is used to transmit AI data based on the AI ​​data transmission configuration.

[0013] Fourthly, a communication device is provided for use in a second node, comprising:

[0014] The sending unit is used to send first information to the first node, which is used to indicate the AI ​​data transmission configuration.

[0015] Fifthly, a communication device is provided, comprising: a processor and a memory; the memory and the processor are coupled; the memory is used to store instructions executable by the processor, the memory storing the processor-executable instructions; when the processor is configured to execute the instructions, the communication device implements the method provided in the first or second aspect above.

[0016] A sixth aspect provides a computer-readable storage medium, including a non-transitory computer-readable storage medium storing computer instructions that, when executed on a computer, cause the computer to perform the method provided in the first or second aspect.

[0017] In a seventh aspect, a computer program product comprising computer instructions is provided, which, when executed on a computer, causes the computer to perform the method provided in the first or second aspect. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this disclosure, the accompanying drawings used in some embodiments of this disclosure will be briefly described below. Obviously, the drawings described below are merely drawings of some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings.

[0019] Figure 1 is a user plane data transmission architecture diagram provided according to an embodiment of the present disclosure.

[0020] Figure 2 is a diagram of an AI-related data transmission architecture provided according to an embodiment of the present disclosure.

[0021] Figure 3 is a diagram of an intra-domain AI-related data interaction architecture provided according to an embodiment of this disclosure.

[0022] Figure 4 is a structural diagram of a communication system provided according to an embodiment of the present disclosure.

[0023] Figure 5 is a flowchart of a data transmission method provided according to an embodiment of the present disclosure.

[0024] Figure 6 is a flowchart of a UE requesting the RAN to train an AI model for the UE according to an embodiment of the present disclosure.

[0025] Figure 7 is a flowchart of a UE requesting the RAN to perform AI inference for the UE according to an embodiment of the present disclosure.

[0026] Figure 8 is a flowchart of an RAN receiving AI data from a UE according to an embodiment of the present disclosure.

[0027] Figure 9 is a flowchart of a UE receiving AI data from a RAN according to an embodiment of the present disclosure.

[0028] Figure 10 is a flowchart of a UE interacting with CDAF through DPRB and data tunnel to train AI model related data according to an embodiment of the present disclosure.

[0029] Figure 11 is a flowchart of a UE interacting with CDAF to obtain AI inference-related data through DPRB and data tunnel according to an embodiment of the present disclosure.

[0030] Figure 12 is a flowchart of a CN collecting AI training data or performing AI analysis from a UE according to an embodiment of this disclosure.

[0031] Figure 13 is a flowchart of an NE1 requesting NE2 to train an AI model for NE1 according to an embodiment of the present disclosure.

[0032] Figure 14 is a flowchart of an NE1 requesting NE2 to perform AI inference according to an embodiment of the present disclosure.

[0033] Figure 15 is a flowchart of the interaction of AI-related data between NE1 and NE2 according to an embodiment of the present disclosure.

[0034] Figure 16 is a flowchart of another data transmission method provided according to an embodiment of the present disclosure.

[0035] Figure 17 is a block diagram of a communication device provided according to an embodiment of the present disclosure.

[0036] Figure 18 is a block diagram of another communication device provided according to an embodiment of the present disclosure.

[0037] Figure 19 is a block diagram of another communication device provided according to an embodiment of the present disclosure. Detailed Implementation

[0038] To enable those skilled in the art to better understand the technical solutions of the embodiments of this disclosure, the technical solutions of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0039] Unless the context otherwise requires, throughout the specification and claims, the term "comprise" and other forms such as the third-person singular "comprises" and the present participle "comprising" are interpreted as open-ended and encompassing, meaning "including, but not limited to." In the description of the specification, terms such as "one embodiment," "some embodiments," "exemplary embodiments," "example," "specific example," or "some examples" are intended to indicate that a particular feature, structure, material, or characteristic associated with that embodiment or example is included in at least one embodiment or example of this disclosure. The illustrative representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics mentioned may be included in any suitable manner in any one or more embodiments or examples.

[0040] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this disclosure, unless otherwise stated, "a plurality of" means two or more.

[0041] In this disclosure, the terms "exemplarily" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplarily" or "for example" in this disclosure should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of the terms "exemplarily" or "for example" is intended to present the relevant concepts in a specific manner.

[0042] In addition, the use of “based on” implies openness and inclusivity, because processes, steps, calculations or other actions “based on” one or more of the stated conditions or values ​​may in practice be based on additional conditions or values ​​beyond those stated.

[0043] Future communication networks supporting AI can be divided into the following two scenarios:

[0044] 1) AI for Network (AI4NET): AI for Net aims to improve network performance, efficiency, and user service experience through AI. AI-enabled network research primarily includes using AI to optimize traditional algorithms (such as air interface channel coding and modulation), network functions (such as mobility optimization and session management optimization), and network operation and maintenance management (such as resource management optimization and planning management optimization).

[0045] 2) Network for AI (NET4AI): This concept leverages the network to provide various support capabilities for AI, enabling more efficient and real-time AI training / inference, and improving data security and privacy. NET4AI expands the traditional network scope from connectivity services to computing power, data, and algorithms.

[0046] With the introduction of these two new AI-enabled next-generation wireless communication network scenarios, 6G communication networks add a large amount of data generated and acquired by communication system network elements on top of traditional user service data transmission. This data can come from terminals, base stations, edge servers, core networks, etc. Compared with fifth-generation mobile communication technology (5G), 6G data exhibits more massive, polymorphic, temporal, and correlated characteristics. The industry generally proposes introducing a data plane into 6G networks to build a unified, reliable, complete, and universal data service framework at the architecture level, thereby improving network and application performance and maximizing data value.

[0047] Referring, Figure 1 illustrates a user plane data transmission architecture according to an embodiment of this disclosure. In a 5G network, user equipment (UE) service data can be mapped to different packet data unit (PDU) sessions based on its associated data network (DN) and single network slice selection assistance information (S-NSSAI). Within a PDU session, there can be various types of data, which can be mapped to individual service data flows according to service data flow (SDF) / traffic flow template (TFT) templates. Furthermore, the user plane function (UPF) / UE maps these service data to different quality of service (QoS) flows in downlink / uplink according to the session management function (SMF) configuration. Between the base station and the UPF, a next-generation tunnel (NG tunnel) is established at the PDU session level for corresponding data transmission. Between the UE and the base station, uplink and downlink QoS flows are associated with data radio bearers (DRBs) according to the base station's access stratum (AS) level mapping rules. The base station can establish one or more DRBs for each QoS flow in a PDU session. At the non-access stratum (NAS) level, the QoS flow is the smallest granularity for QoS differentiation within a PDU session. Each QoS flow within a PDU session is identified by a QoS flow ID (QFI) and is packaged, encapsulated, and transmitted by the next-generation user plane tunnel (NG-U tunnel). At the access stratum (AS) level, QoS guarantees between the UE and the base station are implemented at the DRB granularity.Once the PDU session, NG-U tunnel, DRB, service data to QoS Flow, and QoS Flow to DRB mappings are all configured, the UE can transmit uplink and downlink service data with the DN.

[0048] It should be noted that in Figure 1, NG-RAN stands for Next Generation Radio Access Network (NG-RAN), NR stands for New Radio (NR), NG-UP stands for Next Generation User Plane (NG-UP), E2E stands for End-to-End (E2E), SDF stands for Service Data Flow (SDF), and NG stands for Next Generation (NG).

[0049] Future 6G communication networks will incorporate new AI scenarios on top of traditional user service data transmission. For example, AI learning and AI inference will be introduced to optimize network performance. AI models used for network optimization can be based on centralized learning, distributed learning, or federated learning. Furthermore, AI inference can also be centralized or distributed. Further, future networks can provide a unified computing platform, allowing users and external third-party applications to utilize the computing power of the wireless network for AI learning or AI inference. These diverse AI scenarios will generate massive amounts of data, which can originate from terminals, base stations, edge servers, core networks, and application servers. This means that all network element nodes in the 6G network will have a need for data generation, data collection, and data analysis and processing, and data transmission may occur between any network element nodes within the 6G network.

[0050] Table 1 lists the types of AI-related data that may be transmitted between network elements in a wireless communication network. Specifically, different types of AI-related data need to be transmitted between network elements from the elements shown in rows (C1-C5) to the elements shown in columns (D1-D5). Generally, the AI-related data that needs to be considered in next-generation communication networks are as follows:

[0051] AI Model: The AI ​​model can be generated by UE, RAN, core network (CN), application framework (AF) / application server (AS), or operation administration and maintenance (OAM); while the AI ​​model consumer can be UE, RAN, CN, AF / AS, or OAM. When the AI ​​model generator and consumer are different, it is necessary to transfer the AI ​​model between different network elements.

[0052] For AI for network scenarios, the following AI model transmission methods can be considered: AI models generated by OAM are transmitted to RAN or core network elements, AI models generated by the core network are transmitted to RAN or UE, AI models generated by RAN are transmitted to UE, or AI models generated by AF are transmitted to UE, etc.

[0053] For network for AI scenarios, the following AI model transfer can be considered: The UE, RAN, or CN helps the AF or AS train the AI ​​model, and the trained model needs to be transferred from the UE, RAN, or CN to the AF or AS.

[0054] AI training data: The data producer can be UE, RAN, CN or AF / AS, and the data consumer can be UE, RAN, CN, AF / AS or OAM.

[0055] For AI for network scenarios: AI model creators can collect AI training data from network elements of interest. For example, OAM can collect training data from RAN or core network, RAN can collect training data from UE or core network, and core network can also collect AI training data from UE or RAN, etc.

[0056] For network for AI scenarios: For AI model training initiated by AF / AS, AF / AS can distribute training data from AF / AS to network nodes participating in AI model training, such as CN, RAN, or UE.

[0057] AI Intermediate Models and Model Gradients: For distributed / federated learning, UE, RAN, CN, or AF can act as the initiator, and UE, RAN, CN, or AF can act as the participant. Intermediate computational results of the AI ​​model (such as updates to intermediate results and model gradients) need to be transferred between the initiator and participants.

[0058] For AI for network scenarios, the following approach to transferring intermediate model results can be considered: the User Equipment (UE), as a participant in federated learning, sends AI model gradient updates to the RAN, while the RAN, as the initiator of federated learning, sends intermediate model update results to the UE. Alternatively, intermediate model results can be transferred between different RAN nodes, different CN nodes, or between different RAN and CN nodes.

[0059] For network for AI scenarios, the following approach to transferring intermediate model results can be considered: For distributed learning initiated by the AF or AS, training data needs to be distributed from the AF to network nodes participating in AI model training, such as CN, RAN, or UE. After the CN, RAN, or UE completes its corresponding distributed / federated training, its trained intermediate AI model is then sent from the CN, RAN, or UE to the AF or AS.

[0060] AI performance data: The AI ​​model producer can be a UE, RAN, CN, AF / AS, or OAM; while the AI ​​model consumer can be a UE, RAN, CN, AF / AS, or OAM. If the AI ​​model producer and consumer are not in the same network element, the AI ​​model consumer needs to provide feedback on the AI ​​model's performance to the AI ​​model producer.

[0061] AI analytics: Network elements performing AI model inference, such as UE, RAN, CN, AF, and OAM, can generate AI statistical analysis or predictive data, collectively referred to as AI analytics. Consumers of AI analytics can be UE, RAN, CN, AF, or OAM. Consumers of AI analytics can request or subscribe to AI analytics of interest from the producers of the analytics.

[0062] For AI for network scenarios: the core network can obtain AI analyses of interest from the RAN or AF. Similarly, the RAN can obtain AI analyses of interest from the CN or AF. These AI analyses can help the CN or RAN optimize network configuration.

[0063] For network for AI scenarios: AF can also obtain AI analytics of interest from UE, RAN, or core network. These AI analytics can help AF adjust and optimize applications.

[0064] Intermediate and final AI inference results: For distributed AI inference, such as when a UE can execute a specific layer of AI model inference and send the intermediate AI inference results to the network for subsequent inference, and the network returns the final AI inference result to the UE, it is necessary to transmit the intermediate and final AI inference results between the UE and the network element.

[0065] For network for AI scenarios: The UE can send intermediate AI inference results to the RAN, CN, or AF. The RAN, CN, or AF then performs subsequent AI inference, and finally, the RAN, CN, or AF sends the AI ​​inference results back to the UE. Additionally, scenarios such as distributed inference between RANs, distributed inference between CNs, or the RAN delegating some inference tasks to the CN can also be considered.

[0066] Table 1

[0067] As shown in Figure 2, to support data analysis and processing in any topology of future wireless communication networks, UEs, base stations, core networks, or AF / AS all have data analytics functions (DAFs). DAFs can perform data acquisition, data storage, data processing, model training, model inference, and model monitoring. DAFs can be located in the UE, RAN, CN, AS, or AF. To distinguish these DAFs, they can be called UDAF (for UE), RDAF (for RAN), CDAF (for CN), MDAF (for OAM), and ADAF (for AF). In addition, UEs, base stations, core networks, or AF / AS can also have data plane functions (DPFs), separating AI data acquisition from the DAF. As shown in Figure 2, the UE can directly interact with xNBs, core networks, or AS / AFs through the DAF to exchange AI-related data, or it can complete data acquisition through the DPF, with the DAF subsequently performing AI model training and / or AI model inference and AI analysis generation.

[0068] On the other hand, in actual network deployments, there may be multiple UDAFs, RDAFs, CDAFs, ADAFs, and MDAFs. As shown in Figure 3, information exchange between DAFs of the same type should also be considered. For example, information exchange between UDAFs, RDAFs, CDAFs, ADAFs, and MDAFs.

[0069] Based on the characteristics of AI-related data, it is necessary to design AI data transmission for future communication networks. This includes configuring AI capability interaction, model training / AI inference / AI data reporting tasks, establishing AI data transmission channels, and other processes to achieve on-demand and efficient transmission of AI data.

[0070] Based on this, the present disclosure provides a data transmission method, apparatus, storage medium, and program product. The first node determines the AI ​​data transmission configuration based on the first information sent by the second node, and then performs AI data transmission based on the AI ​​data transmission configuration, thereby improving the transmission efficiency of AI data, that is, realizing efficient transmission and processing of AI data.

[0071] Before describing the technical solutions of the embodiments of this disclosure, the AI-related data transmission process between the UE and RAN in the wireless communication network is introduced.

[0072] According to Table 1, in an AI-enabled wireless network, the UE can send the following AI-related information to the RAN: 1) AI analysis data generated by the UE; 2) Training data of the UE AI model and performance data of the UE AI model monitored by the UE when the RAN performs AI model training for the UE; 3) AI model and AI inference data or intermediate AI inference results of the UE when the RAN performs AI inference or distributed inference for the UE; 4) AI model gradient update information can be sent by the UE to the RAN when the RAN initiates federated learning for the UE.

[0073] Furthermore, in an AI-enabled wireless network, the RAN can send the following AI-related information to the UE: 1) AI analysis data generated by the RAN; 2) UE AI model trained by the RAN in the scenario where the RAN trains the AI ​​model for the UE; 3) the final AI inference result or the AI ​​inference result of certain layers obtained by the RAN when performing AI inference or split AI inference for the UE; 4) the intermediate and final AI model of the UE in the scenario where the RAN initiates federated learning.

[0074] For UE AI analysis, RAN AI analysis, AI training data, and AI performance data, these data can be transmitted between the UE and RAN using a subscription notification model. The specific information to be subscribed to and the required data transmission method can be pre-configured based on radio resource control (RRC) signaling. For intermediate or final AI inference results, AI model gradient updates, and the UE's intermediate and final AI models, transmission between the UE and RAN can be performed on demand according to the progress of split inference (split inference) or federated learning. For specific AI data transmission methods, transmission can be based on RRC signaling or DPRB. It should be noted that DPRB can be a reused DRB or a new RB type, which can be used for AI data transmission. This guideline applies to all the embodiments below.

[0075] The embodiments of this disclosure will now be described in conjunction with the accompanying drawings.

[0076] The technical solutions provided in this disclosure can be applied to various mobile communication networks, such as 5G NR mobile communication networks, future mobile communication networks (e.g., 6G wireless communication systems), or multiple communication convergence systems, etc. This disclosure does not limit them.

[0077] Figure 4 shows a structural diagram of a communication system provided according to an embodiment of the present disclosure. As shown in Figure 4, the communication system includes, but is not limited to, a terminal 110, an access network node 120, a core network function 130, and an application server (AS) 140. Here, the terminal 110, access network node 120, core network function 130, and AS 140 can transmit and receive wireless signals and perform related interactions. The various devices shown in Figure 4 can be connected via a wired network or a wireless network. Here, the wired network or wireless network can include routers, switches, or other devices that facilitate communication between multiple devices; the present disclosure does not limit this.

[0078] In some embodiments, terminal 110 can be a device with wireless transceiver capabilities, such as a mobile phone, tablet computer, wearable device, in-vehicle device, augmented reality (AR) / virtual reality (VR) device, laptop computer, ultra-mobile personal computer (UMPC), netbook, personal digital assistant (PDA), etc. This disclosure does not limit the specific type of terminal 110.

[0079] In some embodiments, the access network node 120 is a base station, which can be any of the following: an evolved NodeB (eNB), a next-generation NodeB (gNB), a transmission receive point (TRP), a transmission point (TP), a femtocell, or some other type of access node. Based on the size of the service coverage area provided, base stations can be further categorized as macro base stations for providing macrocell coverage, micro base stations for providing microcell coverage, and femto base stations for providing femtocell coverage. As wireless communication technology continues to evolve, future base stations may also adopt other names.

[0080] In some embodiments, core network function 130 includes at least one of the following:

[0081] Core network, some functions / functional entities of the core network, servers, clients, third-party clients, and application layer.

[0082] In one example, core network function 130 includes access and mobility management function (AMF) network elements, session management function (SMF) network elements, UPF network elements, data analytics function (DAF) network elements, data plane function (DPF) network elements, and sensing function (SF) network elements.

[0083] In some embodiments, the DPF network element can be co-located with the UPF network element, or the function of the DPF network element can be included in the function of the UPF network element, that is, the UPF network element can have the function of the DPF network element.

[0084] Core network functions may also have other names, such as core network elements, and this disclosure does not limit this.

[0085] In some embodiments, AS140 is used for at least one of the following:

[0086] Provides the application runtime environment;

[0087] Hosting and managing applications;

[0088] Provide services and functions;

[0089] Integrate with other systems and services, etc.

[0090] In some embodiments, AS140 includes an application function (AF), meaning that AS140 integrates an AF.

[0091] In some embodiments, AS140 is connected to AF, meaning that AF can exist as an independent service component and interact with AS140 through specific interfaces or protocols.

[0092] Figure 4 is an exemplary structural diagram. The number of devices included in the communication system shown in Figure 4 is not limited; for example, the number of terminals 110 and access network nodes 120 is not limited. Furthermore, in addition to the devices shown in Figure 4, the communication system shown in Figure 4 may include other devices, which are not limited thereto.

[0093] Next, as shown in Figure 5, this disclosure provides a data transmission method, which can be applied to a first node, and the method may include the following steps:

[0094] S101, Receive the first information from the second node.

[0095] Here, the first piece of information is used to indicate the AI ​​data transmission configuration. The first piece of information can also have other names, such as AI data configuration information.

[0096] In some embodiments, the first node is a terminal, and the second node is an access network node; or...

[0097] The first node is the terminal, and the second node is the core network function; or...

[0098] The first node is an access network node, and the second node is an access network node; or...

[0099] The first node performs core network functions, and the second node performs core network functions; or...

[0100] The first node is an access network node, and the second node performs core network functions; or...

[0101] The first node is the access network node, and the second node is the Operation, Maintenance, and Management (OAM) node; or...

[0102] The first node is for core network functions, and the second node is for OAM.

[0103] Taking the first node as the terminal and the second node as the access network node as an example, the terminal can be the terminal 110 shown in Figure 4 above, and the access network node can be the access network node 120 shown in Figure 4 above.

[0104] In some embodiments, AI data is transmitted based on at least one of the following:

[0105] Signaling radio bearer (SRB), for example, AI data can be carried on radio resource control (RRC) signaling and non-access stratum (NAS) signaling;

[0106] General Packet Radio Service Tunneling Protocol (GTP) tunneling;

[0107] Data plane adaptation protocol (DPAP) tunnel;

[0108] Internet Protocol (IP) data packets;

[0109] The data plane radio bearer (DPRB) is independent of the PDU session. Different data plane radio bearers can correspond to different QoS parameter requirements. For a description of the data plane radio bearer, please refer to relevant technical documents; it will not be elaborated upon here.

[0110] DRB.

[0111] In some embodiments, AI data includes at least one of the following:

[0112] AI training data, AI models, AI performance data, AI inference data, AI inference intermediate results, AI inference final results, and AI analysis data.

[0113] In some embodiments, the AI ​​data transmission configuration includes at least one of the following:

[0114] Data transmission configuration corresponding to AI training data;

[0115] Data transmission configuration corresponding to the AI ​​model;

[0116] Data transmission configuration for AI inference data;

[0117] Data transmission configuration corresponding to AI inference results;

[0118] Data transmission configuration corresponding to intermediate results of AI inference;

[0119] Data transmission configuration corresponding to AI analysis data;

[0120] Data transmission configuration corresponding to AI performance data.

[0121] In some embodiments, the AI ​​data transmission configuration includes at least one of the following bearer configurations:

[0122] DPRB configuration;

[0123] SRB configuration;

[0124] DRB configuration;

[0125] GTP tunnel configuration;

[0126] DPAP tunnel configuration.

[0127] In some embodiments, for each of the above-mentioned at least one bearer configurations, the bearer configuration includes at least one of the following, that is, the first information also includes the bearer configuration corresponding to AI data transmission, and the bearer configuration includes at least one of the following:

[0128] It carries information for adding, modifying, and releasing information.

[0129] For example, for a DPRB configuration, the DPRB configuration includes at least one of the following:

[0130] DPRB add information, DPRB modify information, DPRB release information.

[0131] In some embodiments, bearer addition information or bearer modification information includes at least one of the following:

[0132] The associated configurations include AI task identifiers, AI data identifiers, AI data types, radio bearer identifiers, packet data convergence protocol (PDCP) related configurations, instructions on whether to rebuild PDCP, instructions on whether to restore PDCP, radio link control (RLC) related configurations, logical channel related configurations, and medium access control (MAC) related configurations.

[0133] For example, DPRB add information or DPRB modify information includes at least one of the following:

[0134] The associated AI task identifier, AI data identifier, AI data type, DPRB identifier, PDCP related configuration, indication of whether to rebuild PDCP, indication of whether to restore PDCP, RLC related configuration, logical channel related configuration, and MAC related configuration.

[0135] Here, AI data identification includes at least one of the following: AI model identification, AI training data identification, AI inference data identification, AI inference intermediate result identification, AI inference final result identification, fragment identification, AI analysis identification, and AI model performance monitoring indicators.

[0136] PDCP-related configurations may include at least one of the following: whether out-of-order delivery is allowed, discard timer, PDCP sequence number (SN) size, t-reordering timer, whether header compression is allowed, and whether integrity protection is allowed. RLC-related configurations may include at least one of the following: automatic repeat request (AM) mode, unacknowledged (UM) mode, SN length, t-reassembly timer, maxRetxThreshold, and T-PollRetransmit timer. Logical channel configurations may further include: priority, prioritized bit rate (PBR), bucket size duration (BSD), logical channel group ID (LCGID), service request ID (SRID), allowed scheduling configuration (SCS), and allowed configured grant. The bearer release information includes one or more radio bearer identifiers that need to be released.

[0137] In some embodiments, the first information includes at least one of the following: AI data type, AI data transmission time window, immediate transmission indication, AI data format, and AI data transmission address.

[0138] In some embodiments, the AI ​​data type includes at least one of the following: AI inference data, AI inference intermediate results, AI inference final results, AI model, AI training data, AI performance data, and AI analysis data.

[0139] In some embodiments, the AI ​​data format includes at least one of the following: whether compression is supported, compression algorithm, whether encryption is supported, encryption algorithm, whether integrity protection is supported, supported integrity protection algorithm, IP or non-IP, protocol identifier, and protocol type.

[0140] In some embodiments, the AI ​​data transmission address includes at least one of the following: target node, target port number, second node, second node port number, and tunnel endpoint identifier. Here, the tunnel endpoint identifier corresponds to the tunnel endpoint identifier (TEID) for tunnel transmission.

[0141] In some embodiments, the first information is also used to indicate AI task configuration; that is, the first information can be used not only to indicate AI data transmission configuration but also to further indicate AI task configuration. When the first information is also used to indicate AI task configuration, the first information (which can also be replaced by AI task configuration) includes at least one of the following: AI task identifier, AI training data identifier, AI model identifier, AI data format, AI training data reporting configuration information, AI inference data identifier, AI inference intermediate result identifier, AI inference final result identifier, number of split layers, and indication of dynamic splitting inference.

[0142] Here, the AI ​​training data reporting configuration information includes at least one of the following:

[0143] AI data identification, AI data format, AI data reporting time window, AI data reporting address, periodic reporting instruction, event-triggered reporting instruction, maximum number of reports, immediate reporting instruction, reporting cycle, and reporting threshold.

[0144] In some embodiments, the first information is further used to indicate the AI ​​data subscribed to by the second node, and the first information further includes at least one of the following:

[0145] AI data identification and AI subscription data reporting configuration information.

[0146] Here, the AI ​​subscription data reporting configuration information includes at least one of the following:

[0147] AI data identifier, AI data format, AI data reporting time window, AI data reporting address, periodic reporting instruction, event-triggered reporting instruction, maximum number of reports, immediate reporting instruction, reporting cycle, and reporting threshold.

[0148] The AI ​​data source address includes at least one of the following: source node identifier, source node port number, and source node tunnel identifier. Here, the source node tunnel identifier corresponds to the TEID of the source node.

[0149] In some embodiments, where the first information is also used to indicate the AI ​​data or AI task configuration subscribed to by the second node, the first node sends a confirmation message to the second node after receiving the first information. The confirmation message includes at least one of the following:

[0150] Accepted AI analysis identifier, accepted AI model identifier, accepted AI model performance monitoring indicators, accepted AI training data identifier, and AI data source address.

[0151] In one example, the first information received from the second node includes:

[0152] A second message is sent to the second node, indicating at least one of the following: requesting AI model training, requesting AI inference, or requesting computing resources. In other words, the second message indicates at least one of the following: requesting the second node to train the AI ​​model for the first node, requesting the second node to perform AI inference, or requesting the second node's computing resources.

[0153] Receive the first message from the second node.

[0154] In other words, the first node sends a request message to the second node, and after receiving the request message, the second node sends the first message back to the first node. In turn, the first node receives the first message from the second node.

[0155] In some embodiments, the second information includes at least one of the following:

[0156] AI task identifier, AI model training request, AI model retraining request, AI model update request, network status information, environment status information, estimated computing power requirements, expected model training completion time, AI training data identifier, AI training data volume, AI model identifier, AI model data volume, reliability requirements, AI task priority, preemptibility indicator, AI model inference request, split inference indicator, number of split inference layers, dynamic split inference, AI model inference computing power requirements, expected model inference latency, AI model type, AI model intermediate inference result data volume, AI model final inference result data volume, AI data transmission quality of service (QoS) requirements, computing power resource preemptibility indicator, address of the first node side of AI data transmission, AI analysis of interest identifier, AI model of interest identifier, AI data sending intention information, whether AI model inference is supported by other nodes, whether AI model training is supported by other nodes, AI data capability indicator, computing power availability time, computing power capacity, computing power type, AI training data, AI task identifier corresponding to AI training data.

[0157] In some embodiments, the QoS requirements for AI data transmission include at least one of the following: latency, reliability, and priority.

[0158] In some embodiments, the address of the first node side of AI data transmission includes at least one of the following: node identifier and port number.

[0159] In some embodiments, the AI ​​data sending intention information includes at least one of the following: AI data format, AI data reporting address, periodic reporting instruction, event-triggered reporting instruction, maximum number of reporting times, immediate reporting instruction, reporting cycle, and reporting threshold.

[0160] In some embodiments, the AI ​​data capability indicator is used to indicate at least one of the following: whether compression is supported, compression algorithm, whether encryption is supported, encryption algorithm, whether integrity protection is supported, integrity protection algorithm, IP or non-IP, protocol identifier, and protocol type.

[0161] In some embodiments, the first information is carried in the RRC reconfiguration information.

[0162] The above example illustrates the concept using the second information as the request information. In another example, the first information received from the second node includes:

[0163] Send a third message to the second node, which indicates the AI ​​capabilities supported by the first node;

[0164] Receive the first message from the second node.

[0165] In other words, the first node sends third information to the second node, representing the AI ​​capabilities it supports. After receiving the third information, the second node, based on the AI ​​capabilities supported by the first node, sends first information back to the first node. Correspondingly, the first node receives the first information from the second node.

[0166] Here, the third information includes at least one of the following:

[0167] Whether federated learning is supported, whether segmented inference is supported, whether AI model inference by other nodes is supported, whether AI model training by other nodes is supported, AI data capability indicators, supported AI models, supported AI training data types, supported AI analysis, whether AI model performance monitoring is supported, and fragment identification.

[0168] In some embodiments, when the second node receives third information indicating the AI ​​capabilities supported by the first node, the first information is further used to indicate the AI ​​data subscribed by the second node. That is, based on the received third information, the second node determines the AI ​​capabilities supported by the first node, and then sends the first information to interested first nodes to subscribe to some or all of the AI ​​capabilities supported by the interested first nodes. In this case, the first information can be referred to as AI data subscription information.

[0169] In some embodiments, the first information further includes AI data reporting configuration information, which includes at least one of the following: AI analysis identifier, AI model identifier, and AI data transmission configuration information.

[0170] Here, the AI ​​data transmission configuration information includes at least one of the following:

[0171] AI data format, AI data source address, periodic sending, event-triggered sending, maximum number of sending times, immediate sending, sending cycle, and sending threshold.

[0172] Taking the first node as UE and the second node as RAN as an example, the first node receiving the first information is illustrated in the following example.

[0173] Example 1-1: The second piece of information is used to request AI model training.

[0174] In the case of Example 1-1, the second information can be called an AI computing power request information, requesting the use of the second node to train the AI ​​model for the first node.

[0175] If the UE wants the RAN to train an AI model for it, the UE can send a second message to the RAN. The second message may include at least one of the following: AI task identifier, AI model training request, AI model retraining request, AI model update request, network status, environmental status, estimated power requirements, expected model training completion time, AI training data identifier, AI training data volume, AI model identifier, AI model data volume, reliability requirements, AI task priority, and whether it is preemptible.

[0176] After receiving the second information sent by the UE, the RAN determines whether it can meet the UE's computing power and completion time requirements for the AI ​​model training task. If it can, the RAN sends RRC reconfiguration information (i.e., the first information) to the UE. The first information may contain at least one of the following: AI task configuration and AI data transmission configuration. That is, the first information is used to indicate the AI ​​data transmission configuration and can also be used to indicate the AI ​​task configuration.

[0177] Here, where the first information is also used to indicate the AI ​​task configuration, the first information (i.e., the AI ​​task configuration) includes at least one of the following: AI task identifier, AI training data identifier, AI model identifier, AI data format, and AI training data reporting configuration. The AI ​​task identifier indicates the AI ​​task accepted by the RAN, such as the task of training an AI model for the UE. The AI ​​training data reporting configuration may include at least one of the following: AI data reporting time window, immediate reporting, and AI data reporting address. The AI ​​data format further includes at least one of the following: whether it is compressed, compression algorithm, whether it is encrypted, encryption algorithm, whether it is integrity protected, integrity protection algorithm, IP or non-IP indication, and protocol identifier or protocol type. The AI ​​data reporting address includes at least one of the following: target node and target port number. The target node can be a target IP address or a target node identifier. For Example 1-1, the target node can be the RAN's IP address or RAN identifier.

[0178] AI data transmission configuration can include the data transmission configuration corresponding to AI training data and / or AI models. Specific AI training data and / or AI models can be transmitted via RRC signaling or DPRB. If transmitted via RRC signaling, the UE can send AI training data and the corresponding AI task identifier to the RAN via AI computing power request information. Furthermore, the RAN can send the trained AI model to the UE via RRC reconfiguration messages (i.e., the first information can also include the AI ​​model).

[0179] In addition, the RAN can send the corresponding SRBx (x = 2, 3, 4, 5...n) and its related configuration to the UE via RRC reconfiguration information. If the transmission is via DPRB, the RAN can send the corresponding bearer configuration information for AI data transmission to the UE via RRC reconfiguration information. Taking the bearer configuration as DPRB configuration as an example, the DPRB configuration information can include at least one of the following: DPRB addition information, DPRB modification information, and DPRB release information. The DPRB addition information or DPRB modification information can include at least one of the following: DPRB identifier, associated AI task identifier, AI data identifier, fragment identifier, PDCP related configuration, whether to rebuild PDCP, whether to restore PDCP, RLC related configuration, logical channel related configuration, and MAC related configuration. Here, the associated AI data identifier can be at least one of the following: AI model identifier and AI training data identifier.

[0180] In some embodiments, after receiving RRC reconfiguration information from the RAN, the UE sends RRC reconfiguration confirmation information to the RAN. Simultaneously, the UE configures the AI ​​task and AI data transmission based on the RRC reconfiguration information. Specifically, the UE's AI data transmission configuration includes the following operations: 1) If the AI ​​data transmission configuration information includes DPRB configuration, the UE performs corresponding DPRB addition, modification, or release operations; 2) If the AI ​​data transmission configuration information includes SRBx configuration, the UE performs corresponding SRB addition, modification, or release operations; 3) If the RRC reconfiguration information includes AI task configuration and corresponding AI training data reporting configuration, the UE organizes AI data according to the AI ​​data format and reports it immediately according to the AI ​​training data reporting configuration or according to the reporting time window information.

[0181] Specific AI training data can be transmitted via DPRB or SRBx. If the AI ​​data transmission configuration information includes an SRBx associated with the AI ​​data, the UE can encapsulate the AI ​​data into IP packets or non-IP packets and send them to the base station through the AI ​​container in the SRBx. If the UE receives the DPRB configuration sent by the base station, and the DPRB configuration includes the corresponding AI task and / or AI data identifier, the UE can transmit the corresponding AI data through that DPRB. Specifically, the UE can encapsulate the AI ​​data into corresponding application protocol data. If it is an IP packet, the source / destination IP and source / destination port number of the IP packet can be set according to the corresponding AI data reporting address exchanged between the UE and the base station. In addition, the UE can perform compression, encryption, and / or integrity protection processing at the application layer or DPAP sublayer according to the format requirements of AI data reporting. The UE can further encapsulate the DPAP subheader, which can carry at least one of the following fields: AI task identifier, AI data identifier, source node identifier, one or more target node identifiers, source port number, target port number, protocol identifier, and timestamp. After the UE encapsulates the DPAP subheader, it delivers the DPAP PDU to the PDCP and RLC entities of the DPRB for subsequent data PDCP / RLC / MAC encapsulation, and finally sends it to the base station through the air interface.

[0182] After receiving AI training data from the UE, the RAN can assist the UE in training the AI ​​model, ensuring that model training is completed before the UE's desired completion time. The RAN then sends the trained AI model to the UE via SRBx or DPRB according to the previously configured AI data format. Simultaneously, the RAN can record information such as the type of computing resources consumed during AI model training, the total amount of computing resources, start time, end time, and / or UE identifier. The RAN can then send this information to the UE. Furthermore, the RAN can also send this information to the core network billing function for subsequent UE billing processing.

[0183] After receiving the AI ​​model from the RAN, the UE can use it for AI-based intelligent processing. Furthermore, if the AI ​​computing power request information sent by the UE to the RAN includes an AI model update request, network status, and / or environmental status information, the RAN can select a suitable AI model for the UE upon receiving the request and send the updated model to the UE.

[0184] For Example 1-1 above, before the UE sends the AI ​​computing power request information to the RAN, the UE can exchange their respective AI capability information with the RAN so that the UE and the RAN can know each other's supported AI capabilities.

[0185] For example, in Example 1-1 above, before the UE sends the AI ​​computing power request information to the RAN, a UE with AI capabilities can send AI capability information (i.e., third-party information) to the RAN. This AI capability information includes at least one of the following: whether federated learning is supported, whether split inference is supported, whether AI model inference by other nodes is supported, whether AI model training by other nodes is supported, AI data capability indication, supported AI models, supported AI training data types, supported AI analysis, whether AI model performance monitoring is supported, and fragmentation identification. The AI ​​data capability indication includes at least one of the following: whether compression is supported, supported compression algorithms, whether encryption is supported, supported encryption algorithms, whether integrity protection is supported, supported integrity protection algorithms, IP or non-IP support indication, and supported protocol identifiers or protocol types.

[0186] The RAN can also send AI capability information (i.e., the fourth information) to the UE. The AI ​​capability information may include at least one of the following: supported AI models, supported AI analytics, instructions on whether federated learning is supported, instructions on whether segmented inference is supported, whether AI model training for other nodes is supported, whether AI model training for other nodes is supported, sharding identifier, available computing time, computing capacity, computing type, etc.

[0187] Referring to the description of Example 1-1, Figure 6 shows a flowchart of a UE requesting the RAN to train an AI model for the UE according to an embodiment of this disclosure. Referring to Figure 6, firstly, the UE and RAN exchange AI computing capabilities (i.e., exchange AI capability information). Then, the UE sends its AI computing capability information (i.e., AI computing power request information) to the RAN. After receiving the AI ​​computing capability information, the RAN sends RRC reconfiguration information to the UE, and then the UE sends an RRC reconfiguration complete message to the RAN. Next, the UE sends AI data to the RAN. After receiving the AI ​​data, the RAN prepares an AI model based on the AI ​​data, and then the RAN sends the trained AI model to the UE.

[0188] Example 1-2: The second piece of information is used to request AI inference.

[0189] If the UE wants the RAN to perform AI inference for it, the UE can send an AI computing power request message (i.e., the second message) to the RAN. The AI ​​computing power request message includes at least one of the following: AI task identifier, AI model inference request, whether split inference is required, number of split inference layers, dynamic split inference, AI model inference computing power requirements, expected model inference latency, AI model type or identifier, AI model data volume, AI model intermediate inference result data volume, AI model final inference result data volume, QoS requirements for AI data transmission, AI task priority, whether computing resources are preemptible, and the UE-side address for AI data transmission. Here, the QoS requirements for AI data transmission include at least one of the following: latency, reliability, priority, etc. The UE-side address for AI data transmission includes at least one of the following: node identifier, port number. The node identifier can be an IP address. In Example 1-2, the node can be the UE's IP address or the UE identifier.

[0190] After receiving the AI ​​computing power request information sent by the UE, the RAN determines whether it can meet the UE's AI model inference task computing power and inference latency requirements. If it can, the RAN sends RRC reconfiguration information (i.e., the first information) to the UE. The RRC reconfiguration information may contain at least one of the following: AI task configuration and AI data transmission configuration. Here, the AI ​​task configuration includes at least one of the following: AI task identifier, AI model identifier, AI inference data identifier, AI inference intermediate result identifier, AI inference final result identifier, AI data format, number of split layers, dynamic split inference indication, and AI data transmission configuration. The AI ​​task identifier indicates the AI ​​task accepted by the RAN, such as performing an AI model inference task for the UE. The AI ​​data transmission configuration may include at least one of the following: AI data type, AI data transmission time window, immediate transmission indication, AI data format, and AI data transmission address. The AI ​​data type may include at least one of the following: AI model inference data, AI model inference intermediate result, AI model inference final result, and AI model. The AI ​​data format further includes at least one of the following: whether it is compressed, compression algorithm, whether it is encrypted, encryption algorithm, whether it is integrity protected, integrity protection algorithm, IP or non-IP indication, and protocol identifier or protocol type. The AI ​​data transmission address includes at least one of the following: the target node and the target port number. The target node can be a target IP address or a target node identifier. In this embodiment, the target node can be the IP address or the RAN identifier.

[0191] Different AI data types can be transmitted via RRC signaling or DPRB. If transmitted via RRC signaling, the UE can send the AI ​​model, AI model inference data, AI model inference, and the corresponding AI task identifier to the base station via AI computing power request information. Furthermore, the RAN can send the trained AI model to the UE via RRC reconfiguration messages. Additionally, the RAN can send the corresponding SRBx (x = 2, 3, 4, 5...n) and its related configuration to the UE via RRC reconfiguration information. If transmitted via SRBx, the UE and RAN can send various types of AI data to the peer node through containers corresponding to different AI data types.

[0192] If transmission is via DPRB, the RAN can send the corresponding DPRB configuration information for AI data transmission to the UE via RRC reconfiguration information. The DPRB configuration information may include at least one of the following: DPRB addition information, DPRB modification information, and DPRB release information. The DPRB addition or modification information may include at least one of the following: DPRB identifier, associated AI task identifier, AI data type, AI data identifier, fragment identifier, PDCP related configuration, whether to rebuild PDCP, whether to restore PDCP, RLC related configuration, logical channel related configuration, and MAC related configuration. Here, the AI ​​data type may include at least one of the following: AI model inference data, AI model inference intermediate results, AI model inference final results, and AI model. The AI ​​data identifier may include at least one of the following: AI model identifier, AI inference data identifier, AI inference intermediate result identifier, and AI inference final result identifier. PDCP-related configurations may include at least one of the following: whether out-of-order delivery is allowed, discard timer, PDCP SN size, t-Reordering timer, whether header compression is enabled, and whether integrity protection is enabled. RLC-related configurations may include at least one of the following: AM mode, UM mode, SN length, t-Reassembly, maxRetxThreshold, and T-PollRetransmit. Logical channel configurations may further include: priority, PBR, BSD, LCGID, SRID, allowed SCS, and allowed configured grant. DPRB release information includes one or more DPRB identifiers that need to be released.

[0193] After receiving the RRC reconfiguration information from the RAN, the UE sends an RRC reconfiguration confirmation message to the RAN. Simultaneously, the UE configures the AI ​​task and AI data transmission according to the RRC reconfiguration information. Specifically, the UE's AI data transmission configuration includes the following operations: 1) If the AI ​​data transmission configuration information includes DPRB configuration, the UE performs the corresponding DPRB addition, modification, or release operations; 2) If the AI ​​data transmission configuration information includes SRBx configuration, the UE performs the corresponding SRB addition, modification, or release operations; 3) If the RRC reconfiguration information includes AI inference task configuration and corresponding AI data transmission configuration, the UE organizes AI data according to the AI ​​data format and transmits AI model, AI inference data, or AI model inference intermediate result data according to the AI ​​data transmission configuration.

[0194] For example, the UE can send the AI ​​model to the RAN via the DPRB. The UE can encapsulate the AI ​​data into corresponding protocol data. If it is an IP packet, the source / destination IP and source / destination port number of the IP packet can be set according to the corresponding AI data transmission address between the UE and the base station. In addition, the UE can perform compression, encryption, and / or integrity protection processing at the application layer or DPAP sublayer according to the format requirements of AI data reporting. The UE can further encapsulate the DPAP subheader, which can carry at least one of the following fields: AI task identifier, AI data identifier, inference layer number identifier, source node identifier, one or more target node identifiers, source port number, target port number, protocol identifier, and timestamp. After encapsulating the DPAP subheader, the UE delivers the DPAP PDU to the PDCP and RLC entities of the DPRB for subsequent data PDCP / RLC / MAC encapsulation, and finally sends it to the base station through the air interface.

[0195] When a UE needs to perform AI inference, it can send AI inference data and / or AI model identifiers to the RAN. Upon receiving the AI ​​inference data from the UE, the RAN performs AI inference based on the previously received AI model. After completing the AI ​​inference, the RAN sends the final AI inference result to the UE. It's important to note that for AI split inference, the UE can send intermediate AI inference results and / or the corresponding split layer number to the RAN. The RAN performs subsequent inference based on this intermediate result and the corresponding AI model. After completing the inference, the RAN sends the final AI inference result to the UE. Furthermore, the UE can send intermediate AI inference results to the RAN. Additionally, the UE can send the range of layers for which the RAN needs to perform AI inference. The RAN can then perform AI inference for the corresponding range of layers based on the UE's instructions and finally send the intermediate AI inference results for the corresponding number of layers back to the UE.

[0196] For Examples 1-2 above, before the UE sends the AI ​​computing power request information to the RAN, the UE can exchange their respective AI capability information with the RAN so that the UE and the RAN can know each other's supported AI capabilities.

[0197] For example, in Examples 1-2 above, before the UE sends the AI ​​computing power request information (i.e., the second information) to the RAN, a UE with AI capabilities can send AI capability information (i.e., the third information) to the RAN. The AI ​​capability information includes at least one of the following: whether federated learning is supported, whether split inference is supported, whether AI model inference by other nodes is supported, whether AI model training by other nodes is supported, and AI data capability indications. AI data capability indications include at least one of the following: whether compression is supported, supported compression algorithms, whether encryption is supported, supported encryption algorithms, whether integrity protection is supported, supported integrity protection algorithms, IP or non-IP support indications, and supported protocol identifiers or protocol types.

[0198] The RAN can also send AI capability information (i.e., the fourth information) to the UE. The AI ​​capability information may include at least one of the following: whether federated learning is supported, whether split inference is supported, whether AI model inference is supported for the UE, whether AI model training is supported for the UE, available computing power time, computing power capacity, computing power type, etc.

[0199] Referring to the descriptions of Examples 1-2, Figure 7 illustrates a flowchart of a UE requesting the RAN to perform AI inference for the UE according to an embodiment of this disclosure. Referring to Figure 7, firstly, the UE and RAN exchange AI computing capabilities (i.e., exchange AI capability information). Then, the UE sends its AI computing capability information (i.e., AI computing power request information) to the RAN. After receiving the AI ​​computing capability information, the RAN sends RRC reconfiguration information to the UE, and then the UE sends an RRC reconfiguration completion message to the RAN. Finally, the UE sends the AI ​​model and intermediate AI inference data to the RAN.

[0200] After receiving the AI ​​model and intermediate AI inference data, the RAN performs AI inference based on the AI ​​model and intermediate AI inference data, and then sends the AI ​​inference results to the UE.

[0201] Examples 1-3: The third piece of information is used to indicate the AI ​​capabilities supported by the first node.

[0202] The following example illustrates the signaling flow for the RAN to subscribe to and receive AI analytics and / or AI performance data from the UE, and for the UE to receive AI analytics from the RAN.

[0203] A UE with AI capabilities can send AI capability information (i.e., third information) to the RAN. The AI ​​capability information may include at least one of the following: supported AI models, supported AI training data types, supported AI analytics, whether AI model performance monitoring is supported, whether federated learning is supported, whether split inference is supported, fragment identification, etc.

[0204] After receiving the UE's AI capability information, the RAN can send AI data subscription information (i.e., the first information) to interested UEs. The AI ​​data subscription information may include at least one of the following: AI data identifier, AI data reporting information. Here, the AI ​​data identifier includes at least one of the following: fragment identifier, AI analysis ID, AI model ID, AI model performance monitoring metric, AI training data ID. The AI ​​data reporting information may include at least one of the following: AI data format, AI data reporting address, periodic reporting, event-triggered reporting, maximum reporting frequency, immediate reporting, reporting cycle, reporting threshold, etc. The AI ​​data format further includes at least one of the following: whether it is compressed, compression algorithm, whether it is encrypted, encryption algorithm, whether it has integrity protection, integrity protection algorithm, IP or non-IP indication, protocol identifier or protocol type. The AI ​​data reporting address includes at least one of the following: target node, target port number. The target node can be a target IP address or a target node identifier. For Examples 1-3, the target node can be the RAN's IP address or RAN identifier.

[0205] The AI ​​data subscription information sent by the RAN to the UE may also include AI data transmission configuration. AI data transmission can be achieved via RRC signaling or DPRB. If transmitted via RRC signaling, the RAN can configure the corresponding SRBx (x = 2, 3, 4, 5...n) for the UE. The AI ​​data subscription information may include the SRBx identifier information corresponding to the subscribed AI data. If transmitted via DPRB, the AI ​​data subscription information may include the DPRB configuration information corresponding to the subscribed AI data. It is important to note that AI data transmission configuration can be performed via RRC reconfiguration signaling, a process that can be completed independently of AI data subscription.

[0206] DPRB configuration information may include at least one of the following: DPRB addition information, DPRB modification information, and DPRB release information. DPRB addition or modification information may include at least one of the following: DPRB identifier, associated AI data identifier, fragment identifier, PDCP-related configuration, whether to rebuild PDCP, whether to restore PDCP, RLC-related configuration, logical channel-related configuration, and MAC-related configuration. The associated AI data identifier may include at least one of the following: AI analysis ID, AI model ID, AI model performance monitoring metric, and AI training data ID. PDCP-related configuration may include at least one of the following: whether out-of-order delivery is allowed, discard timer, PDCP SN size, t-Reordering timer, whether header compression is allowed, and whether integrity protection is allowed. RLC-related configuration may include at least one of the following: AM mode, UM mode, SN length, t-Reassembly, maxRetxThreshold, and T-PollRetransmit. Logical channel configuration may further include: priority, PBR, BSD, LCGID, SRID, allowed SCS, and allowed configured grant. The DPRB release information includes one or more DPRB identifiers that need to be released.

[0207] After receiving the AI ​​data subscription information and / or AI data transmission configuration sent by the RAN, the UE can send AI data confirmation information to the RAN. The AI ​​data confirmation information may include at least one of the following: accepted AI analysis ID, accepted AI model ID, accepted AI model performance monitoring metric, accepted AI training data ID, etc., and the AI ​​data source address. The AI ​​data source address includes at least one of the following: source node identifier and source node port number. The source node identifier can be either a source IP address or a source node identifier. In this embodiment, the source node identifier can be either the UE's IP address or the UE identifier.

[0208] After receiving AI data subscription information and / or AI data transmission configuration information from the RAN, the UE executes AI data transmission related configurations. Specifically, the UE's AI data transmission related configurations include the following operations: 1) If the AI ​​data transmission configuration information contains DPRB configuration information, the UE performs corresponding DPRB add, modify, or release operations; 2) If the AI ​​data subscription information contains an AI analysis ID, the UE generates the corresponding AI analysis; 3) If the AI ​​data subscription information contains an AI model ID and / or AI model performance monitoring indicators, the UE performs performance monitoring on the corresponding AI model and obtains the corresponding performance monitoring indicator data; 4) If the AI ​​data subscription information contains AI data reporting information, when the AI ​​data reporting conditions are triggered, the UE assembles AI data according to the AI ​​data format requirements and transmits the AI ​​data via DPRB or SRBx.

[0209] For AI data transmission, if the AI ​​data transmission configuration information includes an SRBx associated with the AI ​​data, the UE can encapsulate the AI ​​data into an IP packet or non-IP packet and send it to the RAN through the AI ​​container in the SRBx. If the UE receives a DPRB configuration sent by the RAN, and the DPRB configuration includes the corresponding AI data identifier, the UE can transmit the corresponding AI data through that DPRB. Specifically, the UE can encapsulate the AI ​​data into the corresponding application protocol data. If it is an IP packet, the source / destination IP and source / destination port number of the IP packet can be set according to the corresponding AI data reporting address and AI data source address information exchanged between the UE and the RAN. In addition, the UE can perform compression, encryption, and / or integrity protection processing at the application layer or DPAP sublayer according to the format requirements of the perceived data reporting. The UE can further encapsulate the DPAP subheader, which can carry at least one of the following fields: QoS indication, AI data type indication, AI data identifier, source node identifier, one or more target node identifiers, source port number, target port number, protocol identifier, and timestamp. After the UE encapsulates the DPAP subheader, it delivers the DPAP PDU to the PDCP and RLC entities of the DPRB for subsequent data PDCP / RLC / MAC encapsulation, and finally sends it to the RAN through the air interface.

[0210] In some embodiments, for Examples 1-3, before the first node sends third information to the second node indicating the AI ​​capabilities supported by the first node, the first node receives fourth information sent by the second node indicating the AI ​​capabilities supported by the second node.

[0211] In other words, firstly, the first node and the second node exchange AI capability information so that the first node and the second node know the AI ​​capabilities supported by each other, or the second node sends AI capability information to the first node so that the first node knows the AI ​​capabilities supported by the second node.

[0212] After learning about the AI ​​capabilities supported by the second node, the first node sends a second message to the second node to request computing resources from the second node. After receiving the second message, the second node sends a first message back to the first node.

[0213] For example, the RAN can send AI capability information (i.e., the fourth information) to the UE. The AI ​​capability information may include at least one of the following: supported AI models, supported AI analytics, whether federated learning is supported, whether split inference is supported, whether training models for the UE is supported, fragmentation identifier, etc.

[0214] After receiving the AI ​​capability information from the RAN, the UE can send an AI computing power request message (i.e., the second message, used to request computing power resources from the second node) to the RAN. The AI ​​computing power request message includes at least one of the following: an identifier of the AI ​​analysis of interest, an identifier of the AI ​​model of interest, and AI data transmission intention information. Here, the AI ​​data transmission intention information may include at least one of the following: AI data format, AI data transmission address, periodic transmission, event-triggered transmission, maximum number of transmissions, immediate transmission, transmission period, transmission threshold, etc. The AI ​​data format further includes at least one of the following: whether compression is supported, supported compression algorithms, whether encryption is supported, supported encryption algorithms, whether integrity protection is supported, supported integrity protection algorithms, IP or non-IP indication, and supported protocol identifiers or protocol types. The AI ​​data transmission address includes at least one of the following: target node and target port number. The target node can be a target IP address or a target node identifier. For Examples 1-3, the target node can be the UE's IP address or UE identifier.

[0215] After receiving the AI ​​computing power request information (i.e., the second information) sent by the UE, the RAN can send RRC reconfiguration information (i.e., the first information) to the UE. The RRC reconfiguration information may include AI data reporting configuration information. The AI ​​data reporting configuration may include at least one of the following: AI analysis ID, AI model ID, and AI data transmission configuration information. The AI ​​data transmission configuration information may include at least one of the following: AI data format, AI data source address, periodic transmission, event-triggered transmission, maximum number of transmissions, immediate transmission, transmission period, and transmission threshold. The AI ​​data format further includes at least one of the following: whether it is compressed, compression algorithm, whether it is encrypted, encryption algorithm, whether it has integrity protection, integrity protection algorithm, IP or non-IP indication, and protocol identifier or protocol type. The AI ​​data source address includes at least one of the following: source node and source port number. The source node can be a source IP address or a source node identifier. For Examples 1-3, the source node can be the IP address of the base station or the base station identifier.

[0216] The RRC reconfiguration information sent by the RAN to the UE may also include AI data transmission configuration. Specifically, AI data can be transmitted via RRC signaling or via DPRB. If transmitted via RRC signaling, the base station can directly send the AI ​​data of interest to the UE through the RRC reconfiguration information. Furthermore, the base station can configure corresponding SRBx (x = 2, 3, 4, 5...n) for the UE. The AI ​​data transmission configuration may include SRBx identifier information corresponding to the AI ​​analysis or AI model of interest to the UE. If transmitted via DPRB, the AI ​​data transmission configuration may include DPRB configuration information corresponding to the AI ​​analysis or AI model of interest to the UE.

[0217] After receiving the RRC reconfiguration information sent by the RAN, the UE performs the corresponding SRBx and / or DPRB configuration. Further, the UE can send the RRC reconfiguration completion information to the RAN.

[0218] The RAN prepares the AI ​​data that the UE is interested in. Once the data is ready, the RAN sends the AI ​​data to the UE according to the previous AI data transmission configuration. Specifically, the UE receives the AI ​​data via RRC signaling, SRBx, or DPRB, depending on the configuration. For the received AI data, the UE can perform data compression, decryption, or integrity protection verification as needed, according to the configuration. The AI ​​data may encapsulate a DPAP header, which may carry at least one of the following fields: AI data type indicator, AI data identifier, source node identifier, one or more target node identifiers, source port number, target port number, protocol identifier, and timestamp. The UE can parse the AI ​​data based on the information contained in the DPAP header or perform subsequent processing using the data analytics function (DAF) delivered to the UE.

[0219] Referring to the description of Examples 1-3, Figure 8 shows a flowchart of a RAN receiving AI data from a UE according to an embodiment of this disclosure. Referring to Figure 8, firstly, the UE sends AI capability information to the RAN, then the RAN sends AI data subscription information to the UE, and subsequently, the UE sends AI data confirmation information to the RAN. After sending the AI ​​data confirmation information, the UE prepares the AI ​​data and then sends the AI ​​data to the RAN.

[0220] Figure 9 shows a flowchart illustrating a UE receiving AI data from a RAN according to an embodiment of this disclosure. Referring to Figure 9, firstly, the RAN sends AI capability information to the UE, and then the UE sends its AI computing capability information (i.e., AI computing power request information) to the RAN. Here, the AI ​​computing capability information includes AI data request information. After receiving the AI ​​computing capability information, the RAN sends RRC reconfiguration information to the UE, and then the UE sends an RRC reconfiguration complete message to the RAN. The RAN then prepares the AI ​​data and sends it to the UE.

[0221] Taking the first node as UE and the second node as CN as an example, the following example can be used to illustrate how the first node receives the first information.

[0222] Based on the information shown in Table 1, in an AI-enabled wireless network, the UE can send the following AI-related information to the CN: 1) AI analysis data generated by the UE; 2) Training data of the UE's AI model training and performance data of the UE's AI model monitored by the UE when the CN trains the AI ​​model for the UE; 3) The UE's AI model and AI inference data or the UE's intermediate AI inference results when the CN performs AI inference or separate inference for the UE; 4) When the CN initiates federated learning for the UE, the UE can send AI model gradient update information to the CN.

[0223] Furthermore, in an AI-enabled wireless network, the CN can send the following AI-related information to the UE: 1) AI analysis data generated by the CN; 2) UE AI model trained by the CN in the scenario where the CN trains the AI ​​model for the UE; 3) the final AI inference result or the AI ​​inference result of certain layers obtained by the CN when performing AI inference or split AI inference for the UE; 4) the intermediate and final AI model of the UE in the scenario where the CN initiates federated learning.

[0224] For UE AI analysis, CN analysis, AI training data, and AI performance data, this data can be transmitted between the UE and CN using a subscription notification model. The specific information to be subscribed to and the required data transmission method can be pre-configured based on NAS signaling. For intermediate or final AI inference results, AI model gradient updates, and the UE's intermediate and final AI models, transmission between the UE and CN can be performed on demand according to the progress of split inference or federated learning. Specific AI data transmission methods can be based on NAS signaling or DPRB+Data tunneling.

[0225] Example 2-1: The second piece of information is used to request AI model training.

[0226] In the case of Example 2-1, the second information can be called an AI computing power request information, requesting the use of the second node to train the AI ​​model for the first node.

[0227] If the UE wants the CN to train an AI model for it, the UE can send an AI computing power request message (i.e., the second message) to the CN. The AI ​​computing power request message includes at least one of the following: AI task identifier, AI model training request, AI model retraining request, AI model update request, network status, environmental status, estimated computing power requirements, expected model training completion time, AI training data identifier, AI training data volume, AI model identifier, AI model data volume, reliability requirements, AI task priority, and whether it is preemptible.

[0228] After receiving the AI ​​computing power request information sent by the UE, the CN determines whether it can satisfy the UE's AI computing power request. If it can, the CN sends AI computing power response information (i.e., the first information) to the UE. The AI ​​computing power response information may include at least one of the following: AI task configuration and AI data transmission configuration. Here, the AI ​​task configuration includes at least one of the following: AI task identifier, AI training data identifier, AI model identifier, AI model, AI data format, and AI training data reporting configuration. The AI ​​task identifier indicates the AI ​​task accepted by the CN, such as the task of training an AI model for the UE. The AI ​​training data reporting configuration may include at least one of the following: AI data reporting time window, immediate reporting, and AI data reporting address. The AI ​​data format further includes at least one of the following: whether it is compressed, compression algorithm, whether it is encrypted, encryption algorithm, whether it is integrity protected, integrity protection algorithm, IP or non-IP indication, and protocol identifier or protocol type. The AI ​​data reporting address includes at least one of the following: target node and target port number. The target node may be a target IP address or a target node identifier. In this embodiment, the target node may be the IP address or identifier of the CN CDAF.

[0229] AI data transmission configuration can include data transmission configurations corresponding to AI training data and / or AI models. Specific AI training data and / or AI models can be transmitted via NAS signaling or via Uu DPRB+tunnel. If transmitted via NAS signaling, the UE can send AI training data, AI training data identifier, and / or AI task identifier to the CN via NAS information. After receiving the AI ​​training data sent by the UE, the CN can assist the UE in training the AI ​​model, ensuring that model training is completed before the UE's expected model training completion time, and send the trained AI model to the UE via NAS messages according to the previously configured AI data format. Simultaneously, the CN can record information such as the type of computing resources consumed by the AI ​​model training, the total amount of computing resources, the start time, the end time, and / or the UE identifier. The CN can send this information to the UE. Furthermore, the CN can also send this information to the core network billing function for subsequent UE billing processing. After receiving the AI ​​model from the RAN, the UE can use the model for AI-based intelligent processing.

[0230] Furthermore, if the AI ​​computing power request information sent by the UE to the CN includes an AI model update request, network status, and / or environmental status information, the CN can select a suitable AI model for the UE and send the updated model to the UE after receiving the request.

[0231] It is important to note that the exchange of capability information, AI computing power request and response information, and AI training models and data between the UE and CN can be transmitted via NAS signaling. Specifically, this can include the following transmission methods:

[0232] 1) The UE sends AI capability information, AI computing power requests, or AI training data to the base station through the NAS container. The base station then sends the NAS message from the NAS container to the AMF. Furthermore, the AMF can send this information to the CDAF. Similarly, the CDAF can send AI capability information, AI computing power responses, or AI training models to the AMF, which then sends them to the RAN through the NAS container. The RAN then sends this information to the UE through the NAS container in the RRC signaling.

[0233] 2) The UE sends AI capability information, AI computing power requests, or AI training data to the RAN via the NAS container. The RAN then sends the NAS messages in the NAS container to the CDAF. Similarly, the CDAF can send AI capability information, AI computing power responses, or AI training models to the RAN, which then sends the information to the UE via the NAS container in the RRC signaling.

[0234] For Example 2-1 above, before the UE sends the AI ​​computing power request information to the CN, the UE can exchange their respective AI capability information with the CN so that the UE and the CN can know each other's supported AI capabilities.

[0235] For example, in Example 2-1 above, before the UE sends AI computing power request information to the RAN, the UE and CN can exchange AI capability information. Specifically, a UE with AI capabilities can send AI capability information to the CN (e.g., AMF and / or CDAF). The AI ​​capability information includes at least one of the following: whether federated learning is supported, whether split inference is supported, whether AI model inference is supported by the CN, whether AI model training is supported by the CN, and AI data capability indications. AI data capability indications include at least one of the following: whether compression is supported, supported compression algorithms, whether encryption is supported, supported encryption algorithms, whether integrity protection is supported, supported integrity protection algorithms, IP or non-IP support indication, and supported protocol identifiers or protocol types.

[0236] The CN (e.g., AMF and / or CDAF) can also send AI capability information to the UE. The AI ​​capability information may include at least one of the following: whether federated learning is supported, whether split inference is supported, whether AI model inference is supported for the UE, whether AI model training is supported for the UE, available computing power time, computing power capacity, computing power type, etc.

[0237] Referring to the description of Example 2-1, Figure 10 shows a flowchart of a UE interacting with CDAF via DPRB and data tunnel according to an embodiment of this disclosure, involving AI model training data. Referring to Figure 10, firstly, the UE exchanges AI computing capabilities (i.e., AI capability information) with the RAN, AMF, and CDAF. Then, the UE sends an AI computing request to the CDAF via the RAN and AMF, and the CDAF sends an AI computing request response to the UE via the AMF and RAN. After receiving the AI ​​computing request response, the UE sends AI training data to the CDAF via the RAN and AMF. After receiving the AI ​​training data, the CDAF trains an AI model based on the AI ​​training data. After obtaining the AI ​​model, the CDAF sends the AI ​​model to the UE via the AMF and RAN.

[0238] Example 2-2: The second piece of information is used to request AI inference.

[0239] If the UE requests the CN to perform AI inference for it, the UE can send an AI computing power request message (i.e., the second message) to the CN. The AI ​​computing power request message includes at least one of the following: AI task identifier, AI model inference request, whether split inference is required, number of split inference layers, dynamic split inference, AI model inference computing power requirements, expected model inference latency, AI model type or identifier, AI model data volume, AI model intermediate inference result data volume, AI model final inference result data volume, QoS requirements for AI data transmission, AI task priority, whether computing resources are preemptible, and the UE-side address for AI data transmission. Here, dynamic split inference refers to the range of AI inference layers that the UE instructs the CN to complete; the range of layers indicated by the UE each time can be different. QoS requirements for AI data transmission include at least one of the following: latency, reliability, priority, etc. The UE-side address for AI data transmission includes at least one of the following: node identifier, port number. The node identifier can be an IP address. Here, the node can be the UE's IP address or the UE identifier.

[0240] After receiving the AI ​​computing power request information sent by the UE, the CN determines whether it can meet the UE's AI model inference task computing power and inference latency requirements. If it can, the CN sends AI computing power response information (i.e., the first information) to the UE. The AI ​​computing power response information may contain at least one of the following: AI task configuration and AI data transmission configuration. Here, the AI ​​task configuration includes at least one of the following: AI task identifier, AI model identifier, AI inference data identifier, AI inference intermediate result identifier, AI inference final result identifier, number of split inference layers, and dynamic split inference indication. The AI ​​task identifier indicates the AI ​​task accepted by the CN, such as performing an AI model inference task for the UE. The AI ​​data transmission configuration may include at least one of the following: AI data type, AI data transmission time window, immediate transmission indication, AI data format, and AI data transmission address. The AI ​​data type may include at least one of the following: AI model inference data, AI model inference intermediate result, AI model inference final result, and AI model. The AI ​​data format further includes at least one of the following: whether it is compressed, compression algorithm, whether it is encrypted, encryption algorithm, whether it is integrity protected, integrity protection algorithm, IP or non-IP indication, and protocol identifier or protocol type. The AI ​​data transmission address includes at least one of the following: the target node and the target port number. The target node can be either a target IP address or a target node identifier. For Example 2-2, the target node can be either the IP address of the CDAF of the CN or the CDAF identifier.

[0241] Different AI data types can be transmitted via DPRB and data tunnels. Data tunnels and DPRBs carrying this AI data must be established between the DPF and RAN, and between the RAN and UE, respectively. It is important to note that DPF stands for Data Plane Function; the DPF's functionality can reside within the UPF, or it can be deployed independently of the UPF. This guideline applies to all the embodiments described below.

[0242] Taking the data tunnel establishment / modification process as an example, the CDAF can send a data tunnel establishment / modification request to the DPF to initiate the establishment or modification of a data tunnel for AI data transmission between the RAN and UPF. It's important to note that the data tunnel establishment / modification request process can be completed before or after the CDAF sends the AI ​​computing power response information to the UE. The data tunnel establishment / modification request may contain at least one of the following information: AI task identifier, fragment identifier, UE identifier, AI model identifier, AI inference data identifier, AI inference intermediate result identifier, AI inference final result identifier, AI data type, data port number, data IP address, and data flow QoS information. Here, the data IP address and data port number correspond to the UE's IP address and the UE's AI data port number. After receiving the data tunnel establishment / modification request, the DPF configures the data tunnel between the RAN and DPF. For example, the DPF may establish different data tunnels for different fragments and / or different UEs and / or different AI data types. Furthermore, the DPF stores the mapping information of data IP addresses and data ports corresponding to specific data tunnels. The data tunnel establishment / modification response sent by the DPF to the CDAF may contain at least one of the following information: AI task identifier, fragment identifier, AI data identifier, and data tunnel transport layer information. Here, the data tunnel transport layer information includes the DPF's IP address and the tunnel endpoint identifier (TEID).

[0243] On the other hand, the AMF sends AI request information to the RAN. Alternatively, the CDAF can also send AI request information to the RAN via the AMF. In other words, the first node is the access network node, and the second node is the core network function, which includes the CDAF and AMF; sending the first information to the first node includes: the CDAF sending the first information to the access network node via the AMF.

[0244] The AI ​​request information includes at least one of the following: AI task identifier, AI model identifier, AI inference data identifier, AI inference intermediate result identifier, AI inference final result identifier, and AI data transmission configuration. The AI ​​data transmission configuration may include at least one of the following: AI data type, AI data transmission time window, AI data transmission QoS requirements, and AI data tunnel configuration. The AI ​​data type may include at least one of the following: AI model inference data, AI model inference intermediate results, AI model inference final result, and AI model. The QoS requirements for AI data transmission include at least one of the following: latency, reliability, priority, etc. The AI ​​data tunnel configuration includes at least one of the following: fragmentation information, DPF-side IP address, and DPF-side TEID.

[0245] After receiving the AI ​​request information, the RAN initiates the RRC reconfiguration process to the UE. The RRC reconfiguration information sent by the RAN to the UE includes DPRB configuration information and / or the mapping between AI data and DPRB.

[0246] DPRB configuration information may include at least one of the following: DPRB addition information, DPRB modification information, and DPRB release information. DPRB addition or modification information may include at least one of the following: DPRB identifier, fragment identifier, PDCP-related configuration, whether to rebuild PDCP, whether to restore PDCP, RLC-related configuration, logical channel-related configuration, and MAC-related configuration. PDCP-related configuration may include at least one of the following: whether out-of-order delivery is allowed, discard timer, PDCP SN size, t-Reordering timer, whether header compression is allowed, whether integrity protection is allowed, etc.; RLC-related configuration may include at least one of the following: AM mode, UM mode, SN length, t-Reassembly, maxRetxThreshold, T-PollRetransmit, etc.; logical channel configuration may further include: priority, PBR, BSD, LCGID, SR ID, allowed SCS, allowed configured grant, etc. DPRB release information includes one or more DPRB identifiers that need to be released. The mapping information between AI data and DPRB may include at least one of the following: DPRB identifier, associated AI task identifier, AI data type, AI data identifier, and fragment identifier. Here, the AI ​​data type may include at least one of the following: AI model inference data, AI model inference intermediate results, AI model inference final results, and AI model. The AI ​​data identifier may include at least one of the following: AI model identifier, AI inference data identifier, AI inference intermediate result identifier, and AI inference final result identifier.

[0247] After receiving the RRC reconfiguration information from the RAN, the UE sends an RRC reconfiguration confirmation message to the RAN. Upon receiving the RRC reconfiguration confirmation message, the RAN sends an AI response message to the AMF. Additionally, the RAN can send AI response information to the CDAF through the AMF. The AI ​​response message may contain at least one of the following: the accepted AI task identifier, the unacceptable AI task identifier and reason value, and the AI ​​data tunnel configuration. Here, the AI ​​data tunnel configuration includes at least one of the following: fragmentation information, the RAN-side IP address, and the RAN-side TEID. After receiving the RAN-side data tunnel configuration information, the CDAF or AMF can initiate a data tunnel establishment / modification process to the DPF, informing the DPF of the corresponding RAN-side data tunnel configuration information.

[0248] If the RRC reconfiguration information includes DPRB configuration, the UE performs the corresponding DPRB addition, modification, or release operation. Furthermore, if the AI ​​computing power response information received by the UE via NAS messages contains AI inference task configuration and corresponding AI data transmission configuration, the UE organizes the AI ​​data according to the AI ​​data format and then maps the AI ​​data (such as AI models, AI inference data, or intermediate results of AI model inference) onto DPRBs before sending it to the RAN. Specifically, the UE assembles the AI ​​data into IP packets, with the destination address and destination port number being the CDAF's IP address and port number. The UE then delivers the IP packets encapsulating the AI ​​data to the DPAP sublayer for processing, and the DPAP sublayer maps the IP data packets to the corresponding DPRBs for transmission. When the RAN receives the data sent by the UE via the DPRB, it maps the DPRB data packets to the corresponding data tunnel, encapsulates the data tunnel header, which may contain the RAN-side IP address, RAN-side TEID information, DPF-side IP address, DPF-side TEID information, and / or AI data identifier. The data packet is then delivered to the DPF. After receiving the data packet, DPF removes the data tunnel header and routes the awareness IP packet in the data tunnel header to CDAF via IP routing. It's important to note that the data tunnel can be a GTP tunnel, a DPAP tunnel, or a Quick UDP Internet Connection (QUIC) tunnel. After receiving the AI ​​inference data from the UE, CDAF performs AI inference. Once the AI ​​inference is complete, CDAF sends the final AI inference result to the UE via the data tunnel + DPRB method.

[0249] It is important to note that the data tunnel between the RAN and DPF is optional for AI-related data packets transmitted between the UE and CDAF. In other words, when the RAN receives AI-related data from the UE via the DPRB, if the RAN can directly communicate with the CDAF network element in the CN, the RAN can directly transmit the AI-related data packets to the CDAF, without needing to send them through a data tunnel to the DPF and then from the DPF to the CDAF.

[0250] For Example 2-2 above, before the UE sends the AI ​​computing power request information to the CN, the UE can exchange their respective AI capability information with the CN so that the UE and the CN can know each other's supported AI capabilities.

[0251] For example, in Example 2-2 above, before the UE sends the AI ​​computing power request information to the RAN, a UE with AI capabilities can send AI capability information to the CN (e.g., AMF and / or CDAF). The AI ​​capability information includes at least one of the following: whether federated learning is supported, whether split inference is supported, whether AI model inference is supported by the CN, whether AI model training is supported by the CN, and AI data capability indications. AI data capability indications include at least one of the following: whether compression is supported, supported compression algorithms, whether encryption is supported, supported encryption algorithms, whether integrity protection is supported, supported integrity protection algorithms, IP or non-IP support indication, and supported protocol identifiers or protocol types.

[0252] The CN (e.g., AMF and / or CDAF) can also send AI capability information to the UE. The AI ​​capability information may include at least one of the following: whether federated learning is supported, whether split inference is supported, whether AI model inference is supported for the UE, whether AI model training is supported for the UE, available computing power time, computing power capacity, computing power type, etc.

[0253] Referring to the description of Example 2-2, Figure 11 shows a flowchart of a UE interacting with CDAF via DPRB and data tunnel according to an embodiment of this disclosure, involving AI inference-related data. Referring to Figure 11, the UE exchanges AI computing capabilities (i.e., AI capability information) with the RAN and AMF. Then, the UE sends an AI computing request to the CDAF through the RAN and AMF. The CDAF then sends a data tunnel establishment / modification request to the DPF, followed by a data tunnel establishment / modification response from the DPF to the CDAF. The CDAF then sends an AI computing response to the UE through the AMF and RAN. Afterwards, the CDAF can send an AI request to the RAN through the AMF. Upon receiving the AI ​​request, the RAN sends RRC reconfiguration information to the UE. After receiving the RRC reconfiguration information, the UE can send an RRC reconfiguration completion message to the RAN. Upon receiving the RRC reconfiguration completion message, the RAN sends an AI response to the CDAF through the AMF. Finally, the CDAF sends the AI ​​model, intermediate AI inference results, and final AI inference results to the UE through the DPF, AMF, and RAN.

[0254] Example 2-3: The third piece of information is used to indicate the AI ​​capabilities supported by the first node.

[0255] The following example illustrates the signaling flow from which the CN subscribes to and receives AI analytics and / or AI performance data from the UE.

[0256] A UE with AI capabilities can send AI capability information (i.e., third information) to the CN (Network Controller). This AI capability information may include at least one of the following: supported AI models, supported AI training data types, supported AI analytics, whether AI model performance monitoring is supported, whether federated learning is supported, whether split inference is supported, fragment identifier, etc. After receiving the UE's AI capability information, the CN can send AI data subscription information (i.e., first information) to interested UEs. This AI data subscription information may include at least one of the following: AI data identifier, AI data reporting information. Here, the AI ​​data identifier includes at least one of the following: fragment identifier, AI analytics ID, AI model ID, AI model performance monitoring metric, AI training data ID. The AI ​​data reporting information may include at least one of the following: AI data format, AI data reporting address, periodic reporting, event-triggered reporting, maximum reporting frequency, immediate reporting, reporting cycle, reporting threshold, etc. The AI ​​data format further includes at least one of the following: whether it is compressed, compression algorithm, whether it is encrypted, encryption algorithm, whether it has integrity protection, integrity protection algorithm, IP or non-IP indication, protocol identifier or protocol type. The AI ​​data reporting address includes at least one of the following: target node, target port number. The target node can be a target IP address or a target node identifier. In this embodiment, the target node can be the IP address or the CDAF identifier.

[0257] After receiving the AI ​​data subscription information and / or AI data transmission configuration sent by the CN, the UE can send AI data subscription confirmation information (i.e., confirmation information) to the CN. The AI ​​data confirmation information may include at least one of the following: accepted AI analysis ID, accepted AI model ID, accepted AI model performance monitoring metric, accepted AI training data ID, etc., and the AI ​​data source address. The AI ​​data source address includes at least one of the following: source node identifier and source node port number. The source node identifier can be either a source IP address or a source node identifier. For Example 2-3, the source node identifier can be either the UE's IP address or the UE identifier.

[0258] Furthermore, the UE can transmit the subscribed AI-related data to the CN via NAS information or through a DPRB+ data tunnel. If transmitted via NAS information, the AI ​​data transmission mechanism described in Example 2-1 can be used. If transmitted via DPRB+ data tunnel, the AI ​​data transmission mechanism described in Example 2-2 can be used, and the corresponding data transmission can be performed. Alternatively, the AI ​​data can be transmitted between the UE and RAN via DPRB as described in Example 2-2. When the RAN receives the AI-related data sent by the UE via DPRB, it directly transmits the AI-related data packet to the CDAF, without needing to send it through a data tunnel to the DPF and then from the DPF to the CDAF.

[0259] Referring to the description of Examples 2-3, Figure 12 shows a flowchart of a CN collecting AI training data or performing AI analysis from a UE according to an embodiment of this disclosure. Referring to Figure 12, firstly, the UE sends AI capability information to the AMF / CDAF. The AMF / CDAF then sends AI data subscription information to interested UEs. In response to the AI ​​data subscription information, the UE sends AI data confirmation information to the AMF / CDAF. The UE then prepares the AI ​​data and sends it to the AMF / CDAF.

[0260] Taking the first node as RAN and the second node as RAN, or the first node as CN and the second node as CN, or the first node as RAN and the second node as CN, or the first node as RAN and the second node as OAM, or the first node as CN and the second node as OAM as examples, the first node receiving the first information is illustrated by an example. For example, the following examples may be included.

[0261] Based on the information shown in Table 1, in an AI-enabled wireless network, the following AI-related information can be sent between RANs: 1) AI analysis data generated by the RAN; 2) In the scenario where RAN node 1 trains an AI model for RAN node 2, RAN node 2 sends training data for the AI ​​model training and performance data of the AI ​​model monitored by RAN node 2 to RAN node 1, and RAN node 1 sends the trained AI model to RAN node 2; 3) In the scenario where RAN node 1 performs AI inference or split inference for RAN node 2, RAN node 2 sends the AI ​​model and AI inference data or intermediate AI inference results to RAN node 1, and RAN node 1 sends the final AI inference results to RAN node 2; 4) In the scenario where RAN initiates federated learning between neighboring RANs, AI model gradient updates, intermediate and final AI models are sent.

[0262] The following AI-related information can be sent between CNs: 1) AI analysis data generated by CN; 2) In the scenario where CN1 trains an AI model for CN2, CN2 sends training data of the AI ​​model training and performance data of the AI ​​model monitored by CN2 to CN1, and CN1 sends the trained AI model to CN2; 3) In the scenario where CN1 performs AI inference or split inference for CN2, CN2 sends the AI ​​model and AI inference data or intermediate AI inference results to CN1, and CN1 sends the final AI inference results to CN2; 4) In the scenario where CN initiates federated learning between CNs, AI model gradient updates, intermediate and final AI models.

[0263] The RAN can send the following AI-related information to the CN: 1) AI analysis data generated by the RAN; 2) Training data of the RAN AI model training and performance data of the RAN AI model monitored by the RAN when the CN is training the RAN AI model; 3) AI model and AI inference data or intermediate AI inference results of the RAN when the CN is performing AI inference or split inference for the RAN; 4) AI model gradient update information sent by the RAN to the CN when the CN initiates federated learning for the RAN.

[0264] The CN can send the following AI-related information to the RAN: 1) AI analysis data generated by the CN; 2) RAN AI model trained by the CN in the scenario where the CN trains the AI ​​model for the RAN; 3) The final AI inference result or the AI ​​inference result of certain layers obtained by the CN when performing AI inference or split AI inference for the RAN; 4) The intermediate and final AI models of the RAN in the scenario where the CN initiates federated learning for the RAN.

[0265] The RAN can send the following AI-related information to the OAM: 1) AI analysis data generated by the RAN; 2) training data of the RAN AI model training and performance data of the RAN AI model monitored by the RAN when the OAM trains the RAN AI model; 3) AI model gradient update information sent by the RAN to the OAM when the OAM initiates federated learning for the RAN.

[0266] OAM can send the following AI-related information to RAN: 1) AI analysis data generated by OAM; 2) RAN AI model trained by OAM in scenarios where OAM trains AI models for RAN; 3) RAN intermediate and final AI models in scenarios where OAM initiates federated learning.

[0267] The CN can send the following AI-related information to the OAM: 1) AI analysis data generated by the CN; 2) training data of the CN AI model training and performance data of the CN AI model monitored by the CN in scenarios where the OAM trains the CN AI model; 3) AI model gradient update information sent by the CN to the OAM in scenarios where the OAM initiates federated learning for the CN.

[0268] OAM can send the following AI-related information to CN: 1) AI analysis data generated by OAM; 2) In scenarios where OAM trains AI models for CN, the CN AI model trained by OAM; 3) In scenarios where OAM initiates federated learning for CN, the intermediate and final AI models of CN.

[0269] For RAN / CN / OAM AI analysis, AI training data, and AI performance data, these data can be transmitted between various network elements in RAN / CN / OAM using a subscription notification model. For AI inference data, intermediate or final AI inference results, AI model gradient updates, and intermediate and final AI models of the UE, these data can be transmitted between various network elements in RAN / CN / OAM on demand, based on the progress of AI model training, split inference, or federated learning.

[0270] Example 3-1: The second piece of information is used to request AI model training.

[0271] The following combinations of NE1 and NE2 are possible:

[0272] 1) NE1 is RAN node1, and NE2 is RAN node2;

[0273] 2) NE1 is CN entity 1, and NE2 is CN entity 2;

[0274] 3) NE1 is RAN, NE2 is CN;

[0275] 4) NE1 is RAN, NE2 is OAM;

[0276] 5) NE1 is CN, and NE2 is OAM.

[0277] If NE1 wants NE2 to train an AI model for it, NE1 can send an AI model training request message (i.e., the second message) to NE2. The AI ​​model training request message includes at least one of the following: AI model training request, AI model retraining request, AI model update request, network status, environment status, estimated computational power requirement, expected model training completion time, AI training data identifier, AI training data volume, AI model identifier, AI model data volume, reliability requirements, AI task priority, and whether preemption is possible.

[0278] After receiving the AI ​​model training request information sent by NE1, NE2 can send AI model training response information (i.e., the first information) to NE1. The AI ​​model training response information may contain at least one of the following: AI task configuration, AI data transmission configuration. Here, the AI ​​task configuration includes at least one of the following: AI task identifier, AI training data identifier, AI model identifier, AI data format, AI training data reporting configuration. The AI ​​task identifier indicates the AI ​​task accepted by NE2, such as the task of training an AI model for NE1. The AI ​​training data reporting configuration may include at least one of the following: AI data reporting time window, immediate reporting, AI data reporting address. The AI ​​data format further includes at least one of the following: whether it is compressed, compression algorithm, whether it is encrypted, encryption algorithm, whether it has integrity protection, integrity protection algorithm, IP or non-IP indication, protocol identifier or protocol type. The AI ​​data reporting address includes at least one of the following: target node, target port number, TEID. The target node can be a target IP address or a target node identifier. For Example 3-1, the target node can be the IP address of NE2 or the NE2 identifier.

[0279] AI data transmission configuration can include data transmission configurations corresponding to AI training data and / or AI models. Specifically, AI training data and / or AI models can be transmitted via control signaling between NE1 and NE2, or via a GTP tunnel, a DPAP tunnel, or directly via IP packets. Specifically, different types of AI data can be transmitted between NE1 and NE2 in the following ways:

[0280] 1) Data transmission using DPAP: NE1 can encapsulate AI data according to the protocol type and optionally perform compression / encryption / integrity protection / IP header addition on the AI ​​data according to the pre-configured AI data format. Furthermore, it can encapsulate the AI ​​data with a DPAP sub-header, which may contain at least one of the following fields: QoS indicator, split inference start layer indicator, split inference end layer indicator, AI data type indicator, AI data identifier, source node identifier, one or more destination node identifiers, source port number, destination port number, protocol identifier, and timestamp. After encapsulating the DPAP sub-header, NE1 further encapsulates a UDP / IP sub-header, where the source and destination IP addresses in the IP sub-header correspond to the IP addresses of NE2 and NE1, respectively. Finally, NE1 transmits the IP packet to NE2.

[0281] 2) AI Data Transmission Using GTP: NE1 can encapsulate AI data according to the protocol type and optionally perform compression / encryption / integrity protection / IP header addition on the AI ​​data according to the pre-configured AI data format. Furthermore, NE1 can encapsulate AI data with a GTP sub-header. NE1 can find the corresponding AI data transmission address based on the corresponding AI data type, and encapsulate the corresponding destination address and TEID in the GTP sub-header. The GTP sub-header can also contain at least one of the following fields: QoS indicator, split inference start layer indicator, split inference end layer indicator, AI data type indicator, AI data identifier, source node identifier, one or more destination node identifiers, source port number, destination port number, protocol identifier, and timestamp. After encapsulating the GTP sub-header, NE1 further encapsulates a UDP / IP sub-header, where the source and destination IP addresses in the IP sub-header correspond to the IP addresses of NE1 and NE2, respectively. Finally, NE1 transmits the IP packet to NE2.

[0282] 3) Transmitting AI data via IP: NE1 can encapsulate AI data according to the protocol type and optionally perform compression / encryption / integrity protection on the AI ​​data according to the pre-configured AI data format. Furthermore, NE1 can encapsulate the AI ​​data with an IP header, where the source IP address is NE1's IP address and the destination IP address corresponds to NE2's IP address. Finally, NE1 transmits the IP packet to NE2.

[0283] 4) AI data is transmitted using control signaling: NE1 can encapsulate AI data according to the protocol type and optionally perform compression / encryption / integrity protection on the AI ​​data according to the pre-configured AI data format. Then, NE1 can transmit the AI ​​data to NE2 through the control signaling interface between NE1 and NE2.

[0284] After receiving the AI ​​training data from NE1, NE2 can assist NE1 in training the AI ​​model, ensuring that training is completed before NE1's desired completion time, and then send the trained AI model back to NE1 as described above. Simultaneously, NE2 can record information such as the type of computing resources consumed during AI model training, the total amount of computing resources, start time, end time, and / or NE1 identifier. NE2 can then send this information to NE1. Furthermore, NE2 can also send this information to the core network billing function for subsequent billing processing by NE1.

[0285] After receiving the AI ​​model from NE2, NE1 can use the model for AI-based intelligent processing. Furthermore, if the AI ​​model training request information sent by NE1 to NE2 includes an AI model update request, network status, and / or environment status information, NE2 can select a suitable AI model for NE1 upon receiving the request and send the updated model to NE1.

[0286] For example 3-1 above, before NE1 sends the AI ​​computing power request information to NE2, NE1 can exchange their respective AI capability information with NE2 so that NE1 and NE2 can know each other's supported AI capabilities.

[0287] For example, NE1, which possesses AI capabilities, can send AI capability information to NE2. This AI capability information includes at least one of the following: whether it supports federated learning, whether it supports split inference, whether it supports AI model inference by other nodes, whether it supports AI model training by other nodes, and AI data capability indicators. AI data capability indicators include at least one of the following: whether it supports compression, supported compression algorithms, whether it supports encryption, supported encryption algorithms, whether it supports integrity protection, supported integrity protection algorithms, IP or non-IP support, and supported protocol identifiers or protocol types.

[0288] NE2 can also send AI capability information to NE1. The AI ​​capability information may include at least one of the following: whether it supports federated learning, whether it supports split inference, whether it supports AI model inference for other nodes, whether it supports training AI models for other nodes, available computing power time, computing power capacity, computing power type, etc.

[0289] Referring to the description of Example 3-1, Figure 13 shows a flowchart of an embodiment of this disclosure whereby NE1 requests NE2 to train an AI model for NE1. Referring to Figure 13, firstly, NE1 and NE2 exchange AI capability information. Then, NE1 sends an AI model training request to NE2. After receiving the AI ​​model training request, NE2 sends an AI model training response to NE1. Next, NE1 sends AI training data to NE2. After receiving the AI ​​training data, NE2 trains the AI ​​model based on the AI ​​training data. Once the trained AI model is obtained, it sends the AI ​​model back to NE1.

[0290] Example 3-2: The second piece of information is used to request AI inference.

[0291] If NE1 requests NE2 to perform AI inference for it, NE1 can send an AI inference request message (i.e., the second message) to NE2. The AI ​​inference request message may include at least one of the following: AI model inference request, whether to perform split inference, the range of split inference layers, the dynamic number of split inference layers, AI model inference computing power requirements, the expected model inference latency, AI model type or identifier, AI model data volume, AI model intermediate inference result data volume, AI model final inference result data volume, QoS requirements for AI data transmission, AI inference priority, whether computing resources are preemptible, and the NE1-side address for AI data transmission. Here, the range of split inference layers refers to the range of AI inference layers that NE1 expects NE2 to complete, such as from layer n to the end, or from layer n to layer m. The dynamic number of split inference layers means that the number of AI inference layers that NE1 expects NE2 to help complete each time is not fixed and can be dynamically indicated to NE2 when sending intermediate AI inference results. The QoS requirements for AI data transmission include at least one of the following: latency, reliability, priority, etc. The NE1-side address for AI data transmission includes at least one of the following: node identifier, port number, TEID. The node identifier can be either the IP address of NE1 or the NE1 identifier.

[0292] After receiving the AI ​​computing power request information sent by NE1, NE2 determines whether it can meet NE1's AI model inference task computing power and inference latency requirements. If it can, NE2 sends AI inference response information (i.e., the first information) to NE1. The AI ​​inference response information may contain at least one of the following: AI task configuration and AI data transmission configuration. Here, the AI ​​task configuration includes at least one of the following: AI task identifier, AI model identifier, AI inference data identifier, AI inference intermediate result identifier, AI inference final result identifier, AI data format, split inference layer range, dynamic split inference indication, and AI data transmission configuration. The AI ​​task identifier indicates the AI ​​model inference task accepted by NE2. The AI ​​data transmission configuration may include at least one of the following: AI data type, AI data transmission time window, AI model reporting indication, AI data format, and AI data transmission address. The AI ​​data type may include at least one of the following: AI model inference data, AI model inference intermediate result, AI model inference final result, and AI model. The AI ​​data format further includes at least one of the following: whether it is compressed, compression algorithm, whether it is encrypted, encryption algorithm, whether it is integrity protected, integrity protection algorithm, IP or non-IP indication, and protocol identifier or protocol type. The AI ​​model reporting instruction indicates that NE2 requests NE1 to send the AI ​​model that needs to be used for AI inference to NE2. The AI ​​data transmission address includes at least one of the following: target node, target port number, and TEID. The target node can be a target IP address or a target node identifier. In this embodiment, the target node can be the IP address of NE2 or the identifier of NE2.

[0293] Different types of AI data can be transmitted between NE1 and NE2 in the following ways:

[0294] 1) AI data transmission using DPAP: NE1 can encapsulate AI data according to the protocol type and optionally perform compression / encryption / integrity protection / IP header addition on the AI ​​data according to the pre-configured AI data format. Furthermore, it can encapsulate the AI ​​data with a DPAP sub-header, which can contain at least one of the following fields: QoS indicator, split inference start layer indicator, split inference end layer indicator, AI data type indicator, AI data identifier, source node identifier, one or more destination node identifiers, source port number, destination port number, protocol identifier, and timestamp. After encapsulating the DPAP sub-header, NE1 further encapsulates a UDP / IP sub-header, where the source and destination IP addresses in the IP sub-header correspond to the IP addresses of NE2 and NE1, respectively. Finally, NE1 transmits the IP packet to NE2.

[0295] 2) AI Data Transmission Using GTP: NE1 can encapsulate AI data according to the protocol type and optionally perform compression / encryption / integrity protection / IP header addition on the AI ​​data according to the pre-configured AI data format. Furthermore, NE1 can encapsulate AI data with a GTP sub-header. NE1 can find the corresponding AI data transmission address based on the corresponding AI data type, and encapsulate the corresponding destination address and TEID in the GTP sub-header. The GTP sub-header can also contain at least one of the following fields: QoS indicator, split inference start layer indicator, split inference end layer indicator, AI data type indicator, AI data identifier, source node identifier, one or more destination node identifiers, source port number, destination port number, protocol identifier, and timestamp. After encapsulating the GTP sub-header, NE1 further encapsulates a UDP / IP sub-header, where the source and destination IP addresses in the IP sub-header correspond to the IP addresses of NE1 and NE2, respectively. Finally, NE1 transmits the IP packet to NE2.

[0296] 3) Transmitting AI data via IP: NE1 can encapsulate AI data according to the protocol type and optionally perform compression / encryption / integrity protection on the AI ​​data according to the pre-configured AI data format. Furthermore, NE1 can encapsulate the AI ​​data with an IP header, where the source IP address is NE1's IP address and the destination IP address corresponds to NE2's IP address. Finally, NE1 transmits the IP packet to NE2.

[0297] 4) AI data is transmitted using control signaling: NE1 can encapsulate AI data according to the protocol type and optionally perform compression / encryption / integrity protection on the AI ​​data according to the pre-configured AI data format. Then, NE1 can transmit the AI ​​data to NE2 through the control signaling interface between NE1 and NE2.

[0298] NE1 can send the AI ​​model used for subsequent AI inference to NE2 using one of the methods described above, based on NE2's configuration. When NE1 needs to perform AI inference, it can send AI inference data and / or AI model identifiers to NE2. After receiving the AI ​​inference data from NE1, NE2 performs AI inference based on the previously received AI model. After completing AI inference, NE2 sends the final AI inference result back to NE1. It's important to note that for AI split inference, NE1 can send intermediate AI inference results and / or the corresponding splitting layer number to NE2. NE2 then performs subsequent inference based on these intermediate results and the corresponding AI model. After completing the inference, NE2 sends the final AI inference result back to NE1. Furthermore, NE1 can also send the range of layers that NE2 needs to perform AI inference to NE2. NE2 can then perform AI inference within the corresponding range of inference layers based on NE1's instructions, and finally send the intermediate AI inference results for the corresponding number of inference layers back to NE1.

[0299] In Example 3-2 above, before NE1 sends the AI ​​computing power request information to NE2, NE1 can exchange their respective AI capability information with NE2 so that NE1 and NE2 can know each other's supported AI capabilities.

[0300] For example, NE1, which possesses AI capabilities, can send AI capability information (i.e., third-party information) to NE2. This AI capability information includes at least one of the following: whether it supports federated learning, whether it supports split inference, whether it supports AI model inference by other nodes, whether it supports AI model training by other nodes, and AI data capability indicators. AI data capability indicators include at least one of the following: whether it supports compression, the supported compression algorithm, whether it supports encryption, the supported encryption algorithm, whether it supports integrity protection, the supported integrity protection algorithm, IP or non-IP support, and supported protocol identifiers or protocol types.

[0301] NE2 can also send AI capability information (i.e., the fourth information) to NE1. The AI ​​capability information may include at least one of the following: whether it supports federated learning, whether it supports split inference, whether it supports AI model inference for other nodes, whether it supports training AI models for other nodes, available computing power time, computing power capacity, computing power type, etc.

[0302] Referring to the description of Example 3-2, Figure 14 shows a flowchart of an NE1 requesting NE2 to perform AI inference according to an embodiment of this disclosure. Referring to Figure 14, firstly, NE1 and NE2 exchange AI computing capabilities (i.e., exchange AI capability information). Then, NE1 sends its AI inference request information to NE2. After receiving the AI ​​inference request information, NE2 sends AI inference response information to NE1. After receiving the AI ​​inference response information, NE1 sends an AI model to NE2. Subsequently, NE1 can perform AI inference on a specific layer to obtain intermediate AI inference data, and then send the intermediate AI inference data to NE2. After receiving the intermediate AI inference data, NE2 performs AI inference on the remaining layers based on the intermediate AI inference data to obtain the AI ​​inference result, and then sends the AI ​​inference result to NE1.

[0303] Example 3-3: The third piece of information is used to indicate the AI ​​capabilities supported by the first node.

[0304] The following example illustrates the signaling flow for NE1 to subscribe to and receive AI analytics and / or AI performance data from NE2.

[0305] NE1 with AI capabilities can send AI capability information (i.e., third information) to NE2. The AI ​​capability information may include at least one of the following: supported AI models, supported AI training data types, supported AI analysis, whether AI model performance monitoring is supported, whether federated learning is supported, whether split inference is supported, fragment identification, etc.

[0306] After receiving the AI ​​capability information from NE1, NE2 can send AI data request information (i.e., the first information) to interested NE1s. The AI ​​data request information may include at least one of the following: AI data identifier, AI data reporting information. Here, the AI ​​data identifier includes at least one of the following: fragment identifier, AI analysis ID, AI model ID, AI model performance monitoring metric, AI training data ID. The AI ​​data reporting information may include at least one of the following: AI data format, AI data reporting address, periodic reporting, event-triggered reporting, maximum reporting frequency, immediate reporting, reporting cycle, reporting threshold, etc. The AI ​​data format further includes at least one of the following: whether it is compressed, compression algorithm, whether it is encrypted, encryption algorithm, whether it has integrity protection, integrity protection algorithm, IP or non-IP indication, protocol identifier or protocol type. The AI ​​data reporting address includes at least one of the following: target node, target port number, TEID. The target node can be a target IP address or a target node identifier. For Example 3-3, the target node can be the IP address of NE2 or the NE2 identifier.

[0307] The AI ​​data request information sent by NE2 to NE1 may also include AI data transmission configuration information (i.e., AI data transmission configuration). The specific AI data can be transmitted through control signaling between NE1 and NE2, or through a GTP tunnel, a DPAP tunnel, or directly through IP packets.

[0308] After receiving the AI ​​data request information and / or AI data transmission configuration information sent by NE2, NE1 can send AI data response information (i.e., the aforementioned confirmation information) to NE2. The AI ​​data response information may include at least one of the following: accepted AI analysis ID, accepted AI model ID, accepted AI model performance monitoring metric, accepted AI training data ID, etc., and the AI ​​data source address. The AI ​​data source address includes at least one of the following: source node identifier, source node port number, and TEID. The source node identifier can be either a source IP address or a source node identifier. For Example 3-3, the source node identifier can be either the IP address of NE1 or the NE1 identifier.

[0309] After receiving AI data request information and / or AI data transmission configuration information from NE2, NE1 executes the relevant configuration for AI data transmission. Specifically, if the AI ​​data request information contains an AI analysis ID, NE1 generates the corresponding AI analysis; if the AI ​​data request information contains an AI model ID and / or AI model performance monitoring indicators, NE1 performs performance monitoring on the corresponding AI model and obtains the corresponding performance monitoring indicator data; if the AI ​​data request information contains AI data reporting information, when the AI ​​data reporting conditions are triggered, AI data is assembled according to the AI ​​data format requirements and transmitted via control signaling between NE1 and NE2, or via a GTP tunnel, a DPAP tunnel, or directly via IP packets.

[0310] Referring to the description of Example 3-3, Figure 15 shows a flowchart of the interaction of AI-related data between NE1 and NE2 according to an embodiment of this disclosure. Referring to Figure 15, firstly, NE1 can send AI capability information to NE2 so that NE2 knows the AI ​​capabilities supported by NE1. Then, NE2 can send AI data request information to NE1. After receiving the AI ​​data request information, NE1 sends AI data response information to NE2. Afterwards, NE1 prepares AI data and then sends the AI ​​data to NE2.

[0311] S102. Based on the AI ​​data transmission configuration, transmit AI data.

[0312] As described above, AI data includes at least one of the following: AI training data, AI model, AI inference data, intermediate results of AI model inference, and final results of AI inference. Based on this, AI data is transmitted according to the AI ​​data transmission configuration, including: transmitting at least one of the following: AI training data, AI model, AI inference data, intermediate results of AI model inference, and final results of AI inference.

[0313] Based on the embodiment shown in Figure 5, the first node determines the AI ​​data transmission configuration based on the first information sent by the second node, and then performs AI data transmission based on the AI ​​data transmission configuration, thereby improving the transmission efficiency of AI data, that is, realizing efficient transmission and processing of AI data.

[0314] Furthermore, in this embodiment, the AI ​​data transmission of the future communication network is designed, including AI capability interaction, configuration of model training / AI inference / AI data reporting tasks, establishment of AI data transmission channels, and AI data transmission processes, which improves the transmission efficiency of AI data and enables efficient on-demand transmission of AI data.

[0315] Next, as shown in Figure 16, this embodiment of the present disclosure also provides a data transmission method, which is applied to a second node, and includes the following steps:

[0316] S201, Send the first message to the first node.

[0317] Here, the first piece of information is used to indicate the AI ​​data transmission configuration.

[0318] In some embodiments, the first node is a terminal, and the second node is an access network node; or...

[0319] The first node is the terminal, and the second node is the core network function; or...

[0320] The first node is an access network node, and the second node is an access network node; or...

[0321] The first node performs core network functions, and the second node performs core network functions; or...

[0322] The first node is an access network node, and the second node performs core network functions; or...

[0323] The first node is the access network node, and the second node is the OAM; or...

[0324] The first node is for core network functions, and the second node is for OAM.

[0325] The following example uses the first node as the terminal and the second node as the access network node.

[0326] In some embodiments, AI data is transmitted based on at least one of the following:

[0327] SRB;

[0328] GTP tunnel;

[0329] DPAP tunnel;

[0330] IP packets;

[0331] DPRB;

[0332] DRB.

[0333] In some embodiments, AI data includes at least one of the following:

[0334] AI training data, AI models, AI performance data, AI inference data, intermediate results of AI model inference, final results of AI inference, and AI analysis data.

[0335] In some embodiments, the AI ​​data transmission configuration includes at least one of the following bearer configurations:

[0336] DPRB configuration;

[0337] SRB configuration;

[0338] DRB configuration;

[0339] GTP tunnel configuration;

[0340] DPAP tunnel configuration.

[0341] In some embodiments, the AI ​​data transmission configuration includes at least one of the following bearer configurations:

[0342] DPRB configuration, SRB configuration, DRB configuration, GTP tunnel configuration, DPAP tunnel configuration.

[0343] In some embodiments, the bearer configuration includes at least one of the following:

[0344] It carries information for adding, modifying, and releasing information.

[0345] In some embodiments, the bearer addition information or the bearer modification information includes at least one of the following:

[0346] The associated AI task identifier, AI data identifier, AI data type, radio bearer identifier, PDCP related configuration, PDCP indication, PDCP restoration indication, RLC related configuration, logical channel related configuration, and MAC related configuration.

[0347] In some embodiments, the first information includes at least one of the following: AI data type, AI data transmission time window, immediate transmission indication, AI data format, and AI data transmission address.

[0348] In some embodiments, the AI ​​data type includes at least one of the following: AI inference data, AI inference intermediate results, AI inference final results, AI model, and AI training data.

[0349] In some embodiments, the AI ​​data format includes at least one of the following:

[0350] Whether compression is supported, compression algorithm, whether encryption is supported, encryption algorithm, whether integrity protection is supported, supported integrity protection algorithms, IP or non-IP, protocol identifier, protocol type.

[0351] In some embodiments, the AI ​​data transmission address includes at least one of the following: target node, target port number, second node, second node port number, and tunnel end-side identifier.

[0352] In some embodiments, the first information is further used to indicate AI task configuration, and the first information further includes at least one of the following: AI task identifier, fragment identifier, AI training data identifier, AI model identifier, AI data format, AI training data reporting configuration information, AI inference data identifier, AI inference intermediate result identifier, AI inference final result identifier, number of split layers, and indication of dynamic split inference.

[0353] In some embodiments, the AI ​​training data reporting configuration information includes at least one of the following:

[0354] AI data identifier, AI data format, AI data reporting time window, AI data reporting address, periodic reporting instruction, event-triggered reporting instruction, maximum number of reports, immediate reporting instruction, reporting cycle, and reporting threshold.

[0355] In some embodiments, the first information is further used to indicate the AI ​​data subscribed by the second node, and the first information further includes at least one of the following: AI data identifier and AI subscription data reporting configuration information.

[0356] In some embodiments, the AI ​​subscription data reporting configuration information includes at least one of the following:

[0357] AI data identifier, AI data format, AI data reporting time window, AI data reporting address, periodic reporting instruction, event-triggered reporting instruction, maximum number of reports, immediate reporting instruction, reporting cycle, and reporting threshold.

[0358] In some embodiments, the AI ​​data source address includes at least one of the following: source node identifier, source node port number, and source node tunnel identifier.

[0359] In some embodiments, where the first information is also used to indicate the AI ​​data subscribed to by the second node, after sending the first information, the second node receives confirmation information from the first node, the confirmation information including at least one of the following: accepted AI analysis identifier, accepted AI model identifier, accepted AI model performance monitoring metric, accepted AI training data identifier, and AI data source address.

[0360] In some embodiments, before sending the first information, a second information is received from the first node, the second information being used to indicate at least one of the following: requesting AI model training, requesting AI inference, or requesting computing resources.

[0361] In some embodiments, the second information includes at least one of the following:

[0362] AI task identifier, AI model training request, AI model retraining request, AI model update request, network status information, environment status information, estimated computing power requirements, expected model training completion time, AI training data identifier, AI training data volume, AI model identifier, AI model data volume, reliability requirements, AI task priority, preemptibility indicator, AI model inference request, split inference indicator, number of split inference layers, dynamic split inference, AI model inference computing power requirements, expected model inference latency, AI model type, AI model intermediate inference result data volume, AI model final inference result data volume, QoS requirements for AI data transmission, preemptibility indicator for computing resources, address of the first node side of AI data transmission, AI analysis of interest identifier, AI model of interest identifier, AI data transmission intention information.

[0363] In some embodiments, the QoS requirements for AI data transmission include at least one of the following: latency, reliability, and priority.

[0364] In some embodiments, the address of the first node side of AI data transmission includes at least one of the following: node identifier, port number, and tunnel end-side identifier.

[0365] In some embodiments, the AI ​​data sending intention information includes at least one of the following: AI data format, AI data reporting address, periodic reporting instruction, event-triggered reporting instruction, maximum number of reporting times, immediate reporting instruction, reporting cycle, and reporting threshold.

[0366] In some embodiments, before sending the first information to the first node, the second node receives third information from the first node, the third information indicating the AI ​​capabilities supported by the first node. The second node then sends the first information to the first node.

[0367] In some embodiments, the third information includes at least one of the following:

[0368] Whether federated learning is supported, whether segmented inference is supported, whether AI model inference by other nodes is supported, whether AI model training by other nodes is supported, AI data capability indicators, supported AI models, supported AI training data types, supported AI analysis, whether AI model performance monitoring is supported, and fragment identification.

[0369] In some embodiments, before sending the first information to the first node, or after receiving the third information from the first node, the second node sends the fourth information to the first node, the fourth information being used to indicate the AI ​​capabilities supported by the second node.

[0370] In some embodiments, after the second node sends the fourth information to the first node, the second node receives the second information from the first node, and then the second node sends the first information to the first node.

[0371] In some embodiments, the fourth information includes at least one of the following:

[0372] Supported AI models, supported AI analytics, instructions on whether federated learning is supported, instructions on whether split inference is supported, whether training AI models for other nodes is supported, whether training AI models for other nodes is supported, sharding identifier, available computing time, computing capacity, and computing type.

[0373] In some embodiments, the second node sends a fifth message to the third node, the fifth message being used to request the establishment or modification of a data tunnel for AI data transmission between the first node and the second node; then, the second node receives a sixth message from the third node. Here, the sixth message is the third node's response to the fifth message.

[0374] In some embodiments, the third node is a DPF or UPF. For example, taking a second node including a CADF and a UPF, and a first node being an access network node (RAN), the fifth message sent to the third node includes:

[0375] CADF sends a fifth message to DPF, which is used to request the establishment or modification of a data tunnel for AI data transmission between the access network node and UPF.

[0376] In some embodiments, the fifth information includes at least one of the following:

[0377] AI task identifier, fragment identifier, terminal identifier, AI model identifier, AI inference data identifier, AI inference intermediate result identifier, AI inference final result identifier, AI data type, data port number, data IP address, and data stream QoS information.

[0378] In some embodiments, the sixth information includes at least one of the following:

[0379] AI task identifier, fragment identifier, AI data identifier, and data tunnel transmission layer information.

[0380] Here, the data tunnel transport layer information includes at least one of the following:

[0381] The network address of the third node;

[0382] Tunnel endpoint identifier.

[0383] The fifth and sixth pieces of information will be introduced below.

[0384] As described above for Example 2-2, data of different AI data types can be transmitted via DPRB and data tunnels. Data tunnels and DPRBs carrying this AI data must be established between the DPF and RAN, and between the RAN and UE, respectively.

[0385] Taking the data tunnel establishment / modification process as an example, the CDAF (i.e., the second node) can send a data tunnel establishment / modification request (i.e., the fifth information) to the DPF (i.e., the third node) to initiate the establishment or modification of a data tunnel for AI data transmission between the RAN and UPF. It is important to note that the data tunnel establishment / modification request process can be completed before or after the CDAF sends the AI ​​computing power response information to the UE. The data tunnel establishment / modification request (i.e., the fifth information) may contain at least one of the following: AI task identifier, fragment identifier, UE identifier, AI model identifier, AI inference data identifier, AI inference intermediate result identifier, AI inference final result identifier, AI data type, data port number, data IP address, and data flow QoS information. Here, the data IP address and data port number correspond to the UE's IP address and the UE's AI data port number. After receiving the data tunnel establishment / modification request, the DPF configures the data tunnel between the RAN and the DPF. For example, the DPF establishes different data tunnels for different fragments and / or different UEs and / or different AI data types. Furthermore, the DPF stores the mapping information of the data IP address and data port corresponding to the specific data tunnel. The data tunnel establishment / modification response (i.e., the sixth message) sent by the DPF to the CDAF may include at least one of the following: AI task identifier, fragment identifier, AI data identifier, and data tunnel transport layer information. Here, the data tunnel transport layer information includes the DPF-side IP address and tunnel endpoint identifier information.

[0386] In the embodiment shown in Figure 16, the second node sends first information to the first node to indicate the AI ​​data transmission configuration, so that the first node can transmit AI data based on the AI ​​data transmission configuration, thereby improving the transmission efficiency of AI data, that is, realizing efficient transmission and processing of AI data.

[0387] The foregoing primarily describes the solution provided in this disclosure from the perspective of the interaction between various nodes. Each node, such as the first node or the second node, includes corresponding hardware structures and / or software modules to perform the aforementioned functions. Those skilled in the art should readily recognize that, based on the algorithmic steps of the examples described in conjunction with the embodiments disclosed herein, this disclosure can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0388] This disclosure embodiment can divide the first node or the second node into functional modules according to the above method embodiment. For example, each function can be divided into a separate functional module, or two or more functions can be integrated into one functional module. The integrated module can be implemented in hardware or software. It should be noted that the module division in this disclosure embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods. The following description uses the example of dividing each functional module according to each function.

[0389] Figure 17 is a block diagram of a communication device according to an embodiment of the present disclosure. As shown in Figure 17, the communication device 30 includes a receiving unit 301 and a transmitting unit 302. The communication device 30 can be the first node described above or a chip in the first node. When the communication device 30 is used to implement the function of the first node in the above embodiment, each unit is specifically used to implement the following functions.

[0390] Receiving unit 301 is used to receive first information from the second node, the first information being used to indicate the AI ​​data transmission configuration;

[0391] The sending unit 302 is used to transmit AI data based on the AI ​​data transmission configuration.

[0392] In some embodiments, the sending unit 302 is further configured to send confirmation information to the second node, the confirmation information including at least one of the following:

[0393] Accepted AI analysis identifier, accepted AI model identifier, accepted AI model performance monitoring indicators, accepted AI training data identifier, and AI data source address.

[0394] In some embodiments, the sending unit 302 is further configured to send second information to the second node, the second information indicating at least one of the following: requesting AI model training, requesting AI inference, or requesting computing resources.

[0395] In some embodiments, the sending unit 302 is further configured to send third information to the second node, the third information being used to indicate the AI ​​capabilities supported by the first node.

[0396] In some embodiments, the receiving unit 301 is further configured to receive fourth information sent by the second node, the fourth information being used to indicate the AI ​​capabilities supported by the second node.

[0397] Figure 18 is a block diagram of another communication device provided according to an embodiment of the present disclosure. As shown in Figure 18, the communication device 40 includes a transmitting unit 401. In some embodiments, the communication device 40 further includes a receiving unit 402.

[0398] The communication device 40 can be the second node or a chip within the second node. When the communication device 40 is used to implement the functions of the second node in the above embodiments, each unit is specifically used to implement the following functions.

[0399] The sending unit 401 is used to send first information to the first node, and the first information is used to indicate the AI ​​data transmission configuration.

[0400] In some embodiments, the receiving unit 402 is configured to receive confirmation information from the first node, the confirmation information including at least one of the following: accepted AI analysis identifier, accepted AI model identifier, accepted AI model performance monitoring index, accepted AI training data identifier, and AI data source address.

[0401] In some embodiments, the receiving unit 402 is configured to receive second information from the first node, the second information being used to indicate at least one of the following: requesting AI model training, requesting AI inference, or requesting computing resources.

[0402] In some embodiments, the receiving unit 402 is configured to receive third information from the first node, the third information being used to indicate the AI ​​capabilities supported by the first node.

[0403] In some embodiments, the sending unit 401 is further configured to send fourth information to the first node, the fourth information being used to indicate the AI ​​capabilities supported by the second node.

[0404] In some embodiments, the sending unit 401 is further configured to send fifth information to the third node, the fifth information being used to request the establishment or modification of a data tunnel for AI data transmission between the first node and the second node; the receiving unit 402 is further configured to receive sixth information from the third node.

[0405] It should be noted that the units in Figure 17 or Figure 18 can also be called modules; for example, the transmitting unit can be called a transmitting module. Furthermore, in the embodiments shown in Figure 17 or Figure 18, the names of the various units may not be those shown in the figures; for example, the transmitting unit can also be called a communication unit, and the receiving unit can also be called a communication unit.

[0406] If the units in Figure 17 or Figure 18 are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this disclosure, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this disclosure. Storage media for storing computer software products include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0407] In the case where the communication device 30 or communication device 40 implements the functions of the integrated module in hardware, a block diagram of another communication device is also provided according to an embodiment of this disclosure. As shown in FIG19, the communication device 50 includes: a processor 502, a communication interface 503, and a bus 504. In some embodiments, the communication device 50 may further include a memory 501.

[0408] Processor 502 may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with this disclosure. Processor 502 may be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with this disclosure. Processor 502 may also be a combination that implements computational functions, such as a combination of one or more microprocessors, a digital signal processor (DSP), and a microprocessor.

[0409] Communication interface 503 is used to connect to other devices via a communication network. This communication network can be Ethernet, wireless access network, wireless local area network (WLAN), etc.

[0410] The memory 501 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0411] In some embodiments, the memory 501 may exist independently of the processor 502. The memory 501 may be connected to the processor 502 via a bus 504 and is used to store instructions or program code. When the processor 502 calls and executes the instructions or program code stored in the memory 501, it can implement the data transmission method provided in the embodiments of this disclosure.

[0412] In another possible implementation, the memory 501 can also be integrated with the processor 502.

[0413] Bus 504 can be an extended industry standard architecture (EISA) bus, etc. Bus 504 can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in Figure 19, but this does not mean that there is only one bus or one type of bus.

[0414] Through the above description of the implementation methods, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the first node or the second node can be divided into different functional modules to complete all or part of the functions described above.

[0415] This disclosure also provides a computer-readable storage medium, including a non-transitory computer-readable storage medium on which computer instructions are stored. All or part of the processes in the above method embodiments can be executed by computer instructions instructing related hardware. The program can be stored in the above computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. The above computer-readable storage medium can also be an external storage device for the first or second node, such as a pluggable hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the first or second node. Further, the above computer-readable storage medium can include both internal storage units of the first or second node and external storage devices. The above computer-readable storage medium is used to store the computer program and other programs and data required by the first or second node. The above computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0416] This disclosure also provides a computer program product comprising a computer program that, when run on a computer, causes the computer to perform any of the data transmission methods provided in the above embodiments.

[0417] Although this disclosure has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, the disclosure, and the appended claims in carrying out the claimed disclosure. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce a good effect.

[0418] Although this disclosure has been described in conjunction with specific features and embodiments, it will be apparent that various modifications and combinations can be made therein without departing from the spirit and scope of this disclosure. Accordingly, this specification and drawings are merely exemplary illustrations of the disclosure as defined by the appended claims and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this disclosure. It is obvious that those skilled in the art can make various alterations and modifications to this disclosure without departing from its spirit and scope. Thus, this disclosure is also intended to include any such modifications and modifications that fall within the scope of the claims of this disclosure and their equivalents.

[0419] The above are merely specific embodiments of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any changes or substitutions within the technical scope disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A data transmission method, wherein, Applied to the first node, the method includes: Receive first information from the second node, the first information being used to indicate the AI ​​data transmission configuration; AI data is transmitted based on the aforementioned AI data transmission configuration.

2. The method according to claim 1, wherein, The AI ​​data is transmitted based on at least one of the following: Signaling Radio Bearer (SRB); General Packet Radio Service (GTP) tunneling; Data plane adaptation protocol DPAP tunnel; Internet Protocol (IP) packets; Data plane radio bearer DPRB; Data Radio Bearer (DRB).

3. The method according to claim 1 or 2, wherein, The AI ​​data includes at least one of the following: AI training data, AI models, AI performance data, AI inference data, AI inference intermediate results, AI inference final results, and AI analysis data.

4. The method according to any one of claims 1 to 3, wherein, The AI ​​data transmission configuration includes at least one of the following bearer configurations: DPRB configuration; SRB configuration; DRB configuration; GTP tunnel configuration; DPAP tunnel configuration.

5. The method according to claim 4, wherein, The bearer configuration includes at least one of the following: It carries information for adding, modifying, and releasing information.

6. The method according to claim 5, wherein, The bearer addition information or the bearer modification information includes at least one of the following: The associated configurations include AI task identifier, AI data identifier, AI data type, radio bearer identifier, Packet Data Convergence Protocol (PDCP) related configurations, indications for whether to rebuild PDCP, indications for whether to restore PDCP, Radio Link Control (RLC) related configurations, Logical Channel (LC) related configurations, and Media Access Control (MAC) related configurations.

7. The method according to any one of claims 1 to 6, wherein, The first information includes at least one of the following: AI data type, AI data transmission time window, immediate transmission instruction, AI data format, and AI data transmission address.

8. The method according to claim 7, wherein, The AI ​​data types include at least one of the following: AI inference data, AI inference intermediate results, AI inference final results, AI models, AI training data, AI performance data, and AI analysis data.

9. The method according to claim 7 or 8, wherein, The AI ​​data format includes at least one of the following: Whether compression is supported, compression algorithm, whether encryption is supported, encryption algorithm, whether integrity protection is supported, supported integrity protection algorithms, IP or non-IP, protocol identifier, protocol type.

10. The method according to any one of claims 7 to 9, wherein, The AI ​​data transmission address includes at least one of the following: target node, target port number, second node, second node port number, and tunnel end-side identifier.

11. The method according to any one of claims 1 to 10, wherein, The first information is also used to indicate the AI ​​task configuration, and the first information includes at least one of the following: AI task identifier, fragment identifier, AI training data identifier, AI model identifier, AI data format, AI training data reporting configuration information, AI inference data identifier, AI inference intermediate result identifier, AI inference final result identifier, number of split layers, and indication of dynamic split inference.

12. The method according to claim 11, wherein, The AI ​​training data reporting configuration information includes at least one of the following: AI data identification, AI data format, AI data reporting time window, AI data reporting address, periodic reporting instruction, event-triggered reporting instruction, maximum number of reports, immediate reporting instruction, reporting cycle, and reporting threshold.

13. The method according to any one of claims 1 to 12, wherein, The first information is also used to indicate the AI ​​data subscribed to by the second node, and the first information further includes at least one of the following: AI data identification and AI subscription data reporting configuration information.

14. The method according to claim 13, wherein, The AI ​​subscription data reporting configuration information includes at least one of the following: AI data identifier, AI data format, AI data reporting time window, AI data reporting address, periodic reporting instruction, event-triggered reporting instruction, maximum number of reports, immediate reporting instruction, reporting cycle, and reporting threshold.

15. The method according to claim 11 or 13, wherein, The method further includes: Send an acknowledgment message to the second node, the acknowledgment message including at least one of the following: Accepted AI analysis identifier, accepted AI model identifier, accepted AI model performance monitoring indicators, accepted AI training data identifier, and AI data source address.

16. The method according to claim 15, wherein, The AI ​​data source address includes at least one of the following: source node identifier, source node port number, first node identifier, first node port number, and source node tunnel identifier.

17. The method according to any one of claims 1 to 16, wherein, Before receiving the first information from the second node, the method further includes: Send a second message to the second node, the second message indicating at least one of the following: requesting AI model training, requesting AI inference, or requesting computing resources.

18. The method according to claim 17, wherein, The second information includes at least one of the following: AI task identifier, AI model training request, AI model retraining request, AI model update request, network status information, environment status information, estimated computing power requirements, expected model training completion time, AI training data identifier, AI training data volume, AI model identifier, AI model data volume, reliability requirements, AI task priority, preemptibility indicator, AI model inference request, split inference indicator, number of split inference layers, dynamic split inference, AI model inference computing power requirements, expected model inference latency, AI model type, AI model intermediate inference result data volume, AI model final inference result data volume, AI data transmission QoS requirements, preemptibility indicator, address of the first node side of AI data transmission, AI analysis of interest identifier, AI model of interest identifier, AI data transmission intention information.

19. The method according to claim 18, wherein, The QoS requirements for AI data transmission include at least one of the following: latency, reliability, and priority.

20. The method according to claim 18 or 19, wherein, The address of the first node side of the AI ​​data transmission includes at least one of the following: node identifier, port number, and tunnel end-side identifier.

21. The method according to any one of claims 18 to 20, wherein, The AI ​​data sending intention information includes at least one of the following: AI data format, AI data reporting address, periodic reporting instruction, event-triggered reporting instruction, maximum number of reporting times, immediate reporting instruction, reporting cycle, and reporting threshold.

22. The method according to any one of claims 1 to 21, wherein, Before receiving the first information from the second node, the method further includes: Send a third message to the second node, the third message being used to indicate the AI ​​capabilities supported by the first node.

23. The method according to claim 22, wherein, The third information includes at least one of the following: Whether federated learning is supported, whether segmented inference is supported, whether AI model inference by other nodes is supported, whether AI model training by other nodes is supported, AI data capability indicators, supported AI models, supported AI training data types, supported AI analysis, whether AI model performance monitoring is supported, and fragment identification.

24. The method according to any one of claims 1 to 23, wherein, The method further includes: Receive a fourth message sent by the second node, the fourth message being used to indicate the AI ​​capabilities supported by the second node.

25. The method according to claim 24, wherein, The fourth piece of information includes at least one of the following: Supported AI models, supported AI analytics, whether federated learning is supported, whether segmented inference is supported, whether AI models can be trained for other nodes, whether AI models can be trained for other nodes, shard identification, available computing time, computing capacity, and computing type.

26. The method according to any one of claims 1 to 25, wherein, The first node is a terminal, and the second node is an access network node; or, The first node is a terminal, and the second node is a core network function; or... The first node is an access network node, and the second node is an access network node; or... The first node performs core network functions, and the second node performs core network functions; or... The first node is an access network node, and the second node is a core network function; or... The first node is an access network node, and the second node is an Operation, Maintenance and Management (OAM) node; or, The first node is the core network function, and the second node is the OAM (Operational Information Management) function.

27. A data transmission method, wherein, Applied to the second node, the method includes: Send the first message to the first node, which is used to indicate the AI ​​data transmission configuration.

28. The method according to claim 27, wherein, The AI ​​data is transmitted based on at least one of the following: SRB, GTP tunnel, DPAP tunnel, IP packet, DPRB, DRB.

29. The method according to claim 27 or 28, wherein, The AI ​​data includes at least one of the following: AI training data, AI models, AI performance data, AI inference data, intermediate results of AI model inference, final results of AI inference, and AI analysis data.

30. The method according to any one of claims 27 to 29, wherein, The AI ​​data transmission configuration includes at least one of the following bearer configurations: DPRB configuration, SRB configuration, DRB configuration, GTP tunnel configuration, DPAP tunnel configuration.

31. The method according to claim 30, wherein, The bearer configuration includes at least one of the following: It carries information for adding, modifying, and releasing information.

32. The method according to claim 31, wherein, The bearer addition information or the bearer modification information includes at least one of the following: The associated AI task identifier, AI data identifier, AI data type, radio bearer identifier, PDCP related configuration, PDCP indication, PDCP restoration indication, RLC related configuration, logical channel related configuration, and MAC related configuration.

33. The method according to any one of claims 27 to 32, wherein, The first information includes at least one of the following: AI data type, AI data transmission time window, immediate transmission instruction, AI data format, and AI data transmission address.

34. The method according to claim 33, wherein, The AI ​​data types include at least one of the following: AI inference data, AI inference intermediate results, AI inference final results, AI models, and AI training data.

35. The method according to claim 33 or 34, wherein, The AI ​​data format includes at least one of the following: Whether compression is supported, compression algorithm, whether encryption is supported, encryption algorithm, whether integrity protection is supported, supported integrity protection algorithms, IP or non-IP, protocol identifier, protocol type.

36. The method according to any one of claims 33 to 35, wherein, The AI ​​data transmission address includes at least one of the following: target node, target port number, second node, second node port number, and tunnel end-side identifier.

37. The method according to any one of claims 27 to 36, wherein, The first information is also used to indicate AI task configuration, and the first information includes at least one of the following: AI task identifier, fragment identifier, AI training data identifier, AI model identifier, AI data format, AI training data reporting configuration information, AI inference data identifier, AI inference intermediate result identifier, AI inference final result identifier, number of split layers, and indication of dynamic split inference.

38. The method according to claim 37, wherein, The AI ​​training data reporting configuration information includes at least one of the following: AI data identifier, AI data format, AI data reporting time window, AI data reporting address, periodic reporting instruction, event-triggered reporting instruction, maximum number of reports, immediate reporting instruction, reporting cycle, and reporting threshold.

39. The method according to claim 37 or 38, wherein, The first information is also used to indicate the AI ​​data subscribed to by the second node, and the first information further includes at least one of the following: AI data identification and AI subscription data reporting configuration information.

40. The method according to claim 39, wherein, The AI ​​subscription data reporting configuration information includes at least one of the following: AI data identifier, AI data format, AI data reporting time window, AI data reporting address, periodic reporting instruction, event-triggered reporting instruction, maximum number of reports, immediate reporting instruction, reporting cycle, and reporting threshold.

41. The method according to claim 37 or 39, wherein, The method further includes: Receive confirmation information from the first node, the confirmation information including at least one of the following: Accepted AI analysis identifier, accepted AI model identifier, accepted AI model performance monitoring indicators, accepted AI training data identifier, and AI data source address.

42. The method according to claim 41, wherein, The AI ​​data source address includes at least one of the following: source node identifier, source node port number, and source node tunnel identifier.

43. The method according to any one of claims 27 to 42, wherein, Before sending the first information to the first node, the method further includes: Receive second information from the first node, the second information indicating at least one of the following: Request AI model training, request AI inference, request computing resources.

44. The method according to claim 43, wherein, The second information includes at least one of the following: AI task identifier, AI model training request, AI model retraining request, AI model update request, network status information, environment status information, estimated computing power requirements, expected model training completion time, AI training data identifier, AI training data volume, AI model identifier, AI model data volume, reliability requirements, AI task priority, preemptibility indicator, AI model inference request, split inference indicator, number of split inference layers, dynamic split inference, AI model inference computing power requirements, expected model inference latency, AI model type, AI model intermediate inference result data volume, AI model final inference result data volume, QoS requirements for AI data transmission, preemptibility indicator for computing resources, address of the first node side of AI data transmission, AI analysis of interest identifier, AI model of interest identifier, AI data transmission intention information.

45. The method according to claim 44, wherein, The QoS requirements for AI data transmission include at least one of the following: latency, reliability, and priority.

46. ​​The method according to claim 44 or 45, wherein, The address of the first node side of the AI ​​data transmission includes at least one of the following: node identifier, port number, and tunnel end-side identifier.

47. The method according to any one of claims 44 to 46, wherein, The AI ​​data sending intention information includes at least one of the following: AI data format, AI data reporting address, periodic reporting instruction, event-triggered reporting instruction, maximum number of reporting times, immediate reporting instruction, reporting cycle, and reporting threshold.

48. The method according to any one of claims 27 to 47, wherein, Before sending the first information to the first node, the method further includes: Receive third information from the first node, the third information being used to indicate the AI ​​capabilities supported by the first node.

49. The method according to claim 48, wherein, The third information includes at least one of the following: Whether federated learning is supported, whether segmented inference is supported, whether AI model inference by other nodes is supported, whether AI model training by other nodes is supported, AI data capability indicators, supported AI models, supported AI training data types, supported AI analysis, whether AI model performance monitoring is supported, and fragment identification.

50. The method according to any one of claims 27 to 49, wherein, The method further includes: A fourth message is sent to the first node, the fourth message being used to indicate the AI ​​capabilities supported by the second node.

51. The method according to claim 50, wherein, The fourth piece of information includes at least one of the following: Supported AI models, supported AI analytics, instructions on whether federated learning is supported, instructions on whether split inference is supported, whether training AI models for other nodes is supported, whether training AI models for other nodes is supported, sharding identifier, available computing time, computing capacity, and computing type.

52. The method according to any one of claims 27 to 51, wherein, The method further includes: Send a fifth message to the third node, the fifth message being used to request the establishment or modification of a data tunnel for AI data transmission between the first node and the second node; Receive the sixth message from the third node.

53. The method according to claim 52, wherein, The third node is a Data Plane Function (DPF) or a User Plane Function (UPF), the first node is an access network node, the second node is a core network function, and the core network function includes the Core Network Data Analysis Function (CDAF) and the UPF. Sending the fifth message to the third node includes: The CDAF sends the fifth information to the DPF, which is used to request the establishment or modification of a data tunnel for AI data transmission between the access network node and the UPF.

54. The method according to claim 52 or 53, wherein, The fifth piece of information includes at least one of the following: AI task identifier, fragment identifier, terminal identifier, AI model identifier, AI inference data identifier, AI inference intermediate result identifier, AI inference final result identifier, AI data type, data port number, data IP address, and data stream QoS information.

55. The method according to any one of claims 52 to 54, wherein, The sixth piece of information includes at least one of the following: AI task identifier, fragment identifier, AI data identifier, and data tunnel transmission layer information.

56. The method according to claim 55, wherein, The data tunnel transport layer information includes at least one of the following: the network address of the third node and the tunnel endpoint identifier.

57. The method according to any one of claims 27 to 56, wherein, The first node is an access network node, and the second node is a core network function, which includes CDAF and Mobility Management Function (AMF).

58. The method according to any one of claims 27 to 56, wherein, The first node is a terminal, and the second node is an access network node; or, The first node is a terminal, and the second node is a core network function; or... The first node is an access network node, and the second node is an access network node; or... The first node performs core network functions, and the second node performs core network functions; or... The first node is an access network node, and the second node is a core network function; or... The first node is an access network node, and the second node is an OAM (Operational Access Management) node; or, The first node is the core network function, and the second node is the OAM (Operational Information Management) function.

59. A communication device, wherein, include: Memory and processor; Memory and processor are coupled; The memory is used to store instructions that can be executed by the processor; When the processor executes the instructions, it performs the method as described in any one of claims 1 to 58.

60. A computer-readable storage medium, wherein, The computer-readable storage medium includes a non-transitory computer-readable storage medium on which computer instructions are stored, which, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 58.

61. A computer program product, wherein, The computer program product includes computer instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 58.

Images