A method and apparatus in a node for model training used for wireless communication

Abstract

This application discloses a method and apparatus for training a model in a node used in wireless communication. The node first receives a first dataset, which is used in the first training; then it sends a first reporting message indicating the performance of a first parameter set obtained through the first training; the first reporting message indicates that the performance of the first parameter set meets a first condition, which applies to a target function, and the first training is for the target function. This application improves the reliability of model training and reduces the time consumed by model training by optimizing the training method of AI / ML models, thereby improving transmission performance and spectral efficiency.

Description

Technical Field

[0001] This application relates to signal transmission methods and apparatus in wireless communication systems, and more particularly to methods and apparatus for training models that integrate artificial intelligence and communication. Background Technology

[0002] Leveraging AI / ML (Artificial Intelligence / Machine Learning) technologies to enhance 5G network performance is a crucial component of achieving deep integration of 5G and AI / ML and building intelligent dimensions for 5G-Advanced (5.5G) networks. The 3GPP (3rd Generation Partnership Project) standards organization initiated research on standards for RAN (Radio Access Networks) intelligence starting with Rel-16 (Release-16), primarily focusing on intelligent use cases, enhanced data collection, and the potential impact on RAN nodes and interfaces. Rel-18 formally established a project for AI / ML-based 5G air interface enhancement, initiating international standardization work on the integration of 5G air interface and AI / ML, mainly focusing on research use cases, lifecycle management (LCM), simulation verification, and data collection.

[0003] Currently, AI / ML development has entered the large-scale model stage. Large-scale communication models can achieve autonomous networks and intelligent services, supporting network operation optimization and improving network efficiency. Deep integration of communication and AI is a crucial direction for future communication evolution. AI will empower the development and upgrade of 5G, 5.5G, and 6G, bringing new management models such as automated frequency band and traffic management, real-time analysis of user data and network load, and prediction of network status. Based on the progress of discussions in 3GPP TR38.843, the functional architecture of AI / ML includes multiple parts, one important part being the model training part. This part is responsible for training, validating, and testing AI / ML models, and the performance criteria of the generated models will be part of the model testing process. Generally, only when the performance of the trained model meets certain performance indicators will the model be used for inference. Summary of the Invention

[0004] In the 3GPP RAN1#119 meeting, a training method for the two-sided AI / ML CSI (Channel State Information) compression research of Rel-19 was presented. This method involves the network side sharing a dataset with the terminal, which includes the target CSI and CSI feedback. These serve as the input and output of a normalized encoder, enabling the UE and network side to align the two-sided compression model. Furthermore, how to efficiently and accurately train the model for the aforementioned scenario is a problem that needs to be considered.

[0005] To address the aforementioned scenarios, and considering the characteristics of the dataset as well as the reliability and efficiency of model training, this application discloses a solution. It should be noted that while this application is initially intended for AI / ML scenarios, it can also be applied to other non-AI / ML scenarios. Furthermore, adopting a unified design scheme for different scenarios (such as other non-AI / ML scenarios, including but not limited to sensor integration, Vehicle to Everything (V2X), capacity enhancement systems, short-range communication systems, NTN (Non-Terrestrial Network), IoT (Internet of Things), and URLLC (Ultra-Reliable Low Latency Communication) networks) helps reduce hardware complexity and cost. Where there is no conflict, embodiments and features in any node of this application can be applied to any other node. Where there is no conflict, embodiments and features in any embodiment of this application can be arbitrarily combined with each other.

[0006] In particular, the interpretation of terms, nouns, functions, and variables in this application (unless otherwise specified) can be found in the definitions of the TS38 and TS37 series of 3GPP (3rd Generation Partnership Project) Technical Specifications (TS). Where necessary, reference can be made to TS38.211, TS38.212, TS38.213, TS38.214, TS38.215, TS38.300, TS38.304, TS38.305, TS38.321, TS38.331, TS37.355, and TS38.423 in the 3GPP technical specifications to aid in understanding this application.

[0007] As an example, the interpretation of terms in this application is based on the definitions in the 3GPP specification protocol TS38 series.

[0008] As an example, the interpretation of terms in this application is based on the definitions in the 3GPP specification protocol TS37 series.

[0009] As an example, the interpretation of the terms used in this application is based on the definitions in 3GPP specification protocol Rel-17.

[0010] As an example, the interpretation of the terms used in this application is based on the definitions in 3GPP specification protocol Rel-18.

[0011] This application discloses a method for the first node of a model training process in wireless communication, comprising:

[0012] Receive a first dataset, which is used by the first training.

[0013] Send a first report, the first report indicating the performance of the first parameter set;

[0014] Wherein, the first parameter set is obtained through the first training; the first reported information indicates that the performance of the first parameter set meets a first condition, the first condition is applicable to the target function, and the first training is for the target function.

[0015] As an example, the problem this application aims to solve includes: how to efficiently train a model.

[0016] As an example, the problem this application aims to solve includes: how to make full use of the intermediate results of model training.

[0017] As an example, the problem this application aims to solve includes: how to detect problems in model training as early as possible in order to avoid meaningless continuous training.

[0018] As an example, the problem this application aims to solve includes: how to improve the efficiency and accuracy of model training through distributed training.

[0019] As an example, the problem this application aims to solve includes: how terminals can share the parameter set obtained from training in order to improve model training speed and training accuracy.

[0020] As an example, the features of the above method include: by reporting the first reporting information, the performance of the first parameter set is informed to the network side, thereby updating the training performance of the model in real time and avoiding waste of resources and computing power.

[0021] As an example, the features of the above method include: when the performance of the first parameter set meets the first condition, the network side can perform subsequent operations on the first parameter set, thereby improving the overall performance.

[0022] As an example, the advantages of the above method include: this application supports the deep integration of AI and communication, improves the adaptability and intelligence level of the communication system, and thus enhances the performance, efficiency and user experience of the communication system.

[0023] As an example, the advantages of the above method include: this application supports distributed training of AI models, that is, multiple entities performing training obtain parameter sets through interactive training, so as to improve training accuracy and training speed.

[0024] As an example, the advantages of the above method include: this application supports segmented training of AI models, that is, the entity performing the training reports the performance corresponding to the intermediate training results, so as to flexibly adjust the training method and dataset and avoid ineffective resource waste.

[0025] As an example, the advantages of the above method include: simplified protocol implementation and good compatibility.

[0026] According to one aspect of this application, the above method is characterized by comprising:

[0027] Receive the first signaling, which configures the first condition.

[0028] As an example, the advantages of the above method include: configuring the first condition through the first signaling, thereby enabling flexible configuration of performance parameters for the training parameter set to adapt to the needs of different scenarios and applications.

[0029] According to one aspect of this application, the above method is characterized by comprising:

[0030] Receive the second parameter set;

[0031] The first condition includes a first threshold, and satisfying the first condition means that the performance of the first parameter set is worse than the first threshold; the second parameter set is used for the target function.

[0032] As an example, the above method is characterized by the following: this solution is designed for scenarios where the first dataset is not suitable for training the target function.

[0033] As an example, the advantages of the above method include: when the performance of the first parameter set is poor, the network side will directly configure the second parameter set for the first node to replace the first parameter set obtained by the first node during training, thereby avoiding the waste of resources caused by continued training and improving efficiency.

[0034] As an example, the advantages of the above method include: the second parameter set is the result of training a node associated with the first node or the target function, thereby ensuring that the second parameter set can be applied to the first node or the target function.

[0035] According to one aspect of this application, the above method is characterized in that the first node stops training for the target function.

[0036] As an example, the above method is characterized by: stopping the training of the target function when the performance of the first parameter set is poor to avoid wasting resources.

[0037] According to one aspect of this application, the above method is characterized by comprising:

[0038] Send the second reporting information;

[0039] The first condition includes a second threshold, and satisfying the first condition means that the performance of the first parameter set is better than the second threshold; the second reported information includes the first parameter set.

[0040] As an example, the above method is characterized by the following features: this solution has good performance for the first dataset and is applicable to scenarios on other terminals.

[0041] As an example, the advantages of the above method include: when the performance of the first dataset is good, the first node reports the first parameter set to the network side, and the network side can send the first parameter set to other terminals to make more efficient use of the training results and improve training efficiency and training accuracy.

[0042] According to one aspect of this application, the above method is characterized in that only a portion of the data in the first dataset is used by the first training, and the first reporting information indicates data in the first dataset that was not used for the first training.

[0043] As an example, the advantages of the above method include: indicating unused datasets in the configured dataset to ensure that the network side and the first node have the same understanding of the used training data.

[0044] As an example, the advantages of the above method include: indicating unused datasets in the configured dataset, so that the unused datasets do not need to be reconfigured, thereby improving spectral efficiency and avoiding resource waste.

[0045] According to one aspect of this application, the above method is characterized by comprising:

[0046] In response to the occurrence of the first event, the first training session is stopped;

[0047] Send a second signaling message;

[0048] The second signaling indicates that the first training is terminated; the candidates for the first event include at least one of cell handover, overheating detection, and entering power-saving mode.

[0049] As an example, the advantages of the above method include: indicating the termination of the first training through the second signaling, so that the network side and the first node have the same understanding of the training state.

[0050] As an example, the advantages of the above method include: stopping the first training when the first node experiences cell handover, detects overheating, or enters power-saving mode, in order to avoid wasting resources.

[0051] As an example, the features of the above method include: the first node is a user equipment.

[0052] As an example, the features of the above method include: the first node is a terminal.

[0053] As an example, the features of the above method include: the first node is configured with an entity for AI / ML.

[0054] As an example, the features of the above method include: the first node is configured with an AI / ML model.

[0055] As an example, the features of the above method include: the first node is configured with a Functionality for AI / ML.

[0056] As an example, the features of the above method include: the first node includes an entity for AI / ML.

[0057] As an example, the features of the above method include: the first node includes a core network device that provides AI services to the terminal.

[0058] As an example, the features of the above method include: the first node includes an application layer device that provides AI services to the terminal.

[0059] As an example, the features of the above method include: the first node includes an Agent that provides AI services to the terminal.

[0060] As an example, the features of the above method include: the first node includes a Handset.

[0061] This application discloses a method for a second node in model training for wireless communication, comprising:

[0062] Send the first dataset, which is used by the first training.

[0063] Receive first reported information, the first reported information indicating the performance of the first parameter set;

[0064] Wherein, the first parameter set is obtained through the first training; the first reported information indicates that the performance of the first parameter set meets a first condition, the first condition is applicable to the target function, and the first training is for the target function.

[0065] As an example, the features of the above method include: the second node includes a core network.

[0066] As an example, the features of the above method include: the second node includes an entity for deploying AI / ML models.

[0067] As an example, the features of the above method include: the second node includes a node for deploying AI / ML models.

[0068] As an example, the features of the above method include: the second node includes a base station.

[0069] As an example, the features of the above method include: the second node is a base station.

[0070] As an example, the features of the above method include: the second node is an eNB.

[0071] As an example, the features of the above method include: the second node is a gNB.

[0072] As an example, the features of the above method include: the second node is a network device, which includes at least one of a core network device and an access network device.

[0073] As an example, the features of the above method include: the second node is a device that provides wireless communication function services, can communicate with terminal devices, and is usually located on the network side.

[0074] As an example, the features of the above method include: the base station in this application includes a core network.

[0075] As an example, the features of the above method include: the base station in this application includes core network equipment.

[0076] As an example, the features of the above method include: the base station in this application includes an entity for deploying AI / ML models.

[0077] As an example, the features of the above method include: the base station in this application includes nodes for deploying AI / ML models.

[0078] As an example, the features of the above method include: the second node includes an OTT (Over-The-Top) Server.

[0079] As an example, the features of the above method include: the second node includes an eNB.

[0080] As an example, the second node in this application includes OAM (Operation Administration and Maintenance).

[0081] According to one aspect of this application, the above method is characterized by comprising:

[0082] Send a first signaling message, which configures the first condition.

[0083] According to one aspect of this application, the above method is characterized by comprising:

[0084] Send the second parameter set;

[0085] The first condition includes a first threshold, and satisfying the first condition means that the performance of the first parameter set is worse than the first threshold; the second parameter set is used for the target function.

[0086] According to one aspect of this application, the above method is characterized in that the sender of the first reporting information stops training for the target function.

[0087] According to one aspect of this application, the above method is characterized by comprising:

[0088] Receive the second reported information;

[0089] The first condition includes a second threshold, and satisfying the first condition means that the performance of the first parameter set is better than the second threshold; the second reported information includes the first parameter set.

[0090] According to one aspect of this application, the above method is characterized in that only a portion of the data in the first dataset is used by the first training, and the first reporting information indicates data in the first dataset that was not used for the first training.

[0091] According to one aspect of this application, the above method is characterized by comprising:

[0092] Receive second signaling;

[0093] In this context, the sender of the first reported information stops the first training as a response to the occurrence of the first event; the second signaling indicates that the first training is terminated; the candidates for the first event include at least one of cell handover, overheating detection, and entering power-saving mode.

[0094] According to one aspect of this application, the method described above is characterized in that the second node is a base station.

[0095] According to one aspect of this application, the above method is characterized in that the second node includes a TRP (transmitter-receiver point).

[0096] This application discloses a first node for model training in wireless communication, comprising:

[0097] A first receiver receives a first dataset, which is used by the first training.

[0098] The first transmitter sends a first report, which indicates the performance of the first parameter set.

[0099] Wherein, the first parameter set is obtained through the first training; the first reported information indicates that the performance of the first parameter set meets a first condition, the first condition is applicable to the target function, and the first training is for the target function.

[0100] This application discloses a second node for model training in wireless communication, comprising:

[0101] The second transmitter sends out the first dataset, which is used by the first training.

[0102] The second receiver receives the first reported information, which indicates the performance of the first parameter set;

[0103] Wherein, the first parameter set is obtained through the first training; the first reported information indicates that the performance of the first parameter set meets a first condition, the first condition is applicable to the target function, and the first training is for the target function.

[0104] As an example, compared with conventional solutions, this application has the following advantages, but is not limited to:

[0105] This application supports the deep integration of AI and communication to improve the adaptability and intelligence of communication systems, thereby enhancing the performance, efficiency, and user experience of communication systems.

[0106] The terminal improves model training speed and accuracy by sharing the parameter set obtained during training;

[0107] By reporting the first reported information, the performance of the first parameter set is informed to the network side, thereby updating the training performance of the model in real time and avoiding the waste of resources and computing power.

[0108] This avoids the waste caused by releasing existing training outputs, improves the efficiency of dataset utilization, and thus improves overall performance. Attached Figure Description

[0109] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0110] Figure 1 A flowchart of the first node transmission according to an embodiment of this application is shown;

[0111] Figure 2 A schematic diagram of a network architecture according to an embodiment of this application is shown;

[0112] Figure 3 A schematic diagram of an embodiment of a wireless protocol architecture for the user plane and control plane according to an embodiment of this application is shown;

[0113] Figure 4 A schematic diagram of a first communication device and a second communication device according to an embodiment of this application is shown;

[0114] Figure 5 A flowchart illustrating a transmission between a first node and a second node according to an embodiment of this application is shown.

[0115] Figure 6 A flowchart of a first signaling transmission according to an embodiment of this application is shown;

[0116] Figure 7 A flowchart illustrating the transmission of a second parameter set according to an embodiment of this application is shown;

[0117] Figure 8 A flowchart of a second reporting information transmission according to an embodiment of this application is shown;

[0118] Figure 9A flowchart of a second signaling transmission according to an embodiment of this application is shown;

[0119] Figure 10 A schematic diagram illustrating the deployment of RAN domain AI / ML functions according to an embodiment of this application is shown;

[0120] Figure 11 A schematic diagram illustrating the deployment of AI / ML functions in a UE according to an embodiment of this application is shown;

[0121] Figure 12 A schematic diagram of an artificial intelligence or machine learning-based processing system according to an embodiment of this application is shown;

[0122] Figure 13 A schematic diagram of artificial intelligence or machine learning according to an embodiment of this application is shown;

[0123] Figure 14 A structural block diagram of a processing apparatus for a first node according to an embodiment of this application is shown;

[0124] Figure 15 A structural block diagram of a processing apparatus for a second node according to an embodiment of this application is shown. Detailed Implementation

[0125] The technical solutions of this application will be further described in detail below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other. Considering performance, flexibility, complexity, overhead, and compatibility, those skilled in the art are motivated to flexibly combine the embodiments in different drawings without conflict, including but not limited to the accompanying drawings. Figure 1 Examples and appendices Figure 5 - Appendix Figure 15 The embodiments in the appendix Figure 5 Examples and appendices Figure 6 - Appendix Figure 15 Examples, etc.

[0126] Example 1

[0127] Example 1 illustrates a flowchart of a first node transmission according to an embodiment of this application, as shown in the attached diagram. Figure 1 As shown. In the appendix Figure 1 In this diagram, each box represents a step. Specifically, the order of the steps within the boxes does not indicate a specific temporal sequence between them.

[0128] The first node receives a first dataset in step 101, which is used by the first training; and sends a first report in step 102, which indicates the performance of the first parameter set.

[0129] In Example 1, the first parameter set is obtained through the first training; the first reported information indicates that the performance of the first parameter set meets a first condition, the first condition is applicable to the target function, and the first training is for the target function.

[0130] As one example, the first node is a user equipment (UE).

[0131] As one example, the first node is a terminal.

[0132] As an example, the first node is the first node in this application.

[0133] As an example, the first dataset is configured for the first training.

[0134] As an example, the first dataset is configured for the target function.

[0135] As an example, the first dataset is a training dataset.

[0136] As an example, the first dataset includes at least one training dataset.

[0137] As an example, the first dataset includes multiple data groups, and one of the data groups includes a target CSI and a CSI feedback.

[0138] As an example, the first dataset includes multiple data groups, and each data group includes a sample of the encoded output and a sample of the input before encoding.

[0139] As an example, the physical layer channels occupied by the first dataset include PDSCH (Physical Downlink Shared Channel).

[0140] As an example, the transmission channels occupied by the first dataset include DL-SCH (Downlink Shared Channel).

[0141] As an example, the first dataset is transmitted via NAS (Non-access stratum) signaling.

[0142] As an example, the first training is for training an AI / ML model.

[0143] As an example, the first training is training for the target function.

[0144] As an example, the first training is associated with an association ID.

[0145] As an example, the first reported information is transmitted via a non-air interface.

[0146] As an example, the physical layer channel occupied by the first reported information includes PUSCH (Physical Uplink Shared Channel).

[0147] As an example, the transmission channel occupied by the first reported information includes UL-SCH (Uplink Shared Channel).

[0148] As an example, the first reported information is reported via RRC (Radio Resource Control) signaling.

[0149] As an example, the first reported information is reported through MAC (Medium Access Control) CE (Control Elements).

[0150] As an example, the first reported information is reported via MAC signaling.

[0151] As an example, the first reported information is reported via PUCCH (Physical Uplink Control Channel).

[0152] As an example, the first reported information is reported via UCI (Uplink Control Information).

[0153] As an example, the first reported information explicitly indicates the performance of the first parameter set.

[0154] As an example, the first reported information implicitly indicates the performance of the first parameter set.

[0155] As an example, the first reported information directly indicates the performance of the first parameter set.

[0156] As an example, the first reported information indirectly indicates the performance of the first parameter set.

[0157] As an example, the first reported information indicates that the performance of the first parameter set meets the first condition.

[0158] As an example, the first reported information indicates the parameters included in the first parameter set.

[0159] As an example, the first reported information indicates the performance of the first parameter set.

[0160] As an example, the first set of parameters is associated with an AI / ML model.

[0161] As an example, the first set of parameters is associated with the target function.

[0162] As an example, the first parameter set is associated with an association ID.

[0163] As an example, the first parameter set includes at least one parameter.

[0164] As one example, the first parameter set includes multiple parameters.

[0165] As an example, the first parameter set corresponds to an Encoder.

[0166] As an example, the first parameter set corresponds to a Reconstructor.

[0167] As an example, the training dataset used in the first training includes multiple data groups, and each data group includes a target CSI and a CSI feedback.

[0168] As an example, the training dataset used in the first training includes multiple data groups, and each data group includes a sample of the encoded output and a sample of the input before encoding.

[0169] As an example, the structure of the AI / ML model corresponding to the first parameter set is one of the following: Transformer structure, RNN (Recurrent Neural Network) structure, or CNN (Conventional Neural Networks) structure.

[0170] As an example, the structure of the AI / ML model corresponding to the first parameter set is the structure of a hybrid model composed of multiple models, including at least one of the Transformer structure, RNN structure, or CNN structure.

[0171] As an example, the first condition is configured for the target function.

[0172] As an example, the first condition is used to trigger a report for the target function.

[0173] As an example, the first condition is used to evaluate the performance of training for the target function.

[0174] As an example, the first training is for training the target function.

[0175] As an example, the first training is training for the target function.

[0176] As an example, the target function includes a functionality.

[0177] As an example, the target function includes an entity.

[0178] As an example, the target function includes a model that is used for inference.

[0179] As an example, the target function includes at least one model, which is used for inference.

[0180] As one example, the target function includes inference configuration.

[0181] As one example, the target function includes one or more CSI-ReportConfigs.

[0182] As an example, the target function is associated with one or more CSI-ReportConfig IEs.

[0183] As one example, the target function includes one or more CSI-MeasConfigs.

[0184] As an example, the target function is associated with one or more CSI-MeasConfig IEs.

[0185] As one example, the target function includes one or more CSI-ResourceConfigs.

[0186] As an example, the target function is associated with one or more CSI-ResourceConfig IEs.

[0187] As an example, the name of the RRC signaling used to configure the target function includes CSI.

[0188] As an example, the name of the RRC signaling used to configure the target function includes CSI-RS.

[0189] As an example, the name of the RRC signaling used to configure the target function includes Report.

[0190] As an example, the name of the RRC signaling used to configure the target function includes Config.

[0191] As one embodiment, the first signaling configures the target function.

[0192] As an example, the second node in this application includes an OTT (Over-The-Top) Server.

[0193] As an example, the second node in this application includes an eNB.

[0194] As an example, the second node in this application includes OAM.

[0195] As an example, the first node in this application includes a core network device that provides services for terminal AI.

[0196] As an example, the first node in this application includes an application layer device that provides services for terminal AI.

[0197] As an example, the first node in this application includes an Agent that provides services for terminal AI.

[0198] As an example, the first node in this application includes a Handset.

[0199] As one embodiment, the first condition applicable to the target function includes: the first condition includes the performance requirements of the AI ​​model associated with the target function.

[0200] As one embodiment, the first condition applicable to the target function includes: the first condition includes the performance requirements of the AI / ML model associated with the target function.

[0201] As one embodiment, the first condition applicable to the target function includes: the first condition includes performance requirements for the set of parameters associated with the target function.

[0202] As an example, the first condition applicable to the target function includes: the first condition includes performance requirements for the inference results of the AI ​​model associated with the target function.

[0203] As an example, the first condition applicable to the target function includes: the first condition includes performance requirements for the inference results of the AI / ML model associated with the target function.

[0204] As one embodiment, the first condition applicable to the target function includes: the first condition includes performance requirements for the inference results of the parameter set associated with the target function.

[0205] As one embodiment, the first condition applicable to the target function includes: the first condition includes performance requirements of the training results of the AI ​​model associated with the target function.

[0206] As one embodiment, the first condition applicable to the target function includes: the first condition includes performance requirements of the training results of the AI / ML model associated with the target function.

[0207] As one embodiment, the first condition applicable to the target function includes: the first condition includes performance requirements for the training results of the parameter set associated with the target function.

[0208] Example 2

[0209] Example 2 illustrates a schematic diagram of a network architecture according to an embodiment of this application, as shown in the attached diagram. Figure 2 As shown.

[0210] Appendix Figure 2Network architecture 200 is described. Network architecture 200 refers to the network architectures of LTE (Long-Term Evolution), LTE-A (Long-Term Evolution Advanced), 5G systems, 5G-Advanced, and future 6G systems. The network architectures of LTE, LTE-A, 5G systems, 5G-Advanced, and future 6G systems are referred to as EPS (Evolved Packet System). The 5G NR or LTE network architecture may be referred to as 5GS (5G System) / EPS or some other suitable terminology; the 6G network architecture may be referred to as 6GS (6G System) / EPS or some other suitable terminology.

[0211] The network architecture 200 may include one or more UEs 201, a RAN (Radio Access Network) 202, a core network 210, an HSS (Home Subscriber Server) / UDM (Unified Data Management) 220, and an Internet service 230. The network architecture 200 may interconnect with other access networks, but these entities / interfaces are not shown for simplicity.

[0212] As attached Figure 2As shown, the network architecture 200 provides packet switching services; however, those skilled in the art will readily understand that the various concepts presented throughout this application can be extended to networks providing circuit-switched services or other cellular networks. The RAN 202 includes Node B 203 and other nodes 204. Node B 203 provides user and control plane protocol termination toward the UE 201. Node B 203 may be connected to other nodes 204 via an Xn interface (e.g., backhaul). Node B 203 may also be referred to as eNB (evolved Node B), gNB, base station, base transceiver station, wireless base station, wireless transceiver, transceiver function, Basic Service Set (BSS), Extended Service Set (ESS), TRP (Transmitter Receiver Point), or some other suitable term. Node B 203 provides UE 201 with an access point to the core network 210; the core network 210 is a 5GC (5G Core network) / EPC (Evolved Packet Core), or the core network 210 is a 6GC (6G Core network). Examples of the UE 201 include cellular phones, smartphones, Session Initiation Protocol (SIP) phones, laptops, personal digital assistants (PDAs), satellite radios, GPS devices, multimedia devices, video devices, digital audio players (e.g., MP3 players), cameras, game consoles, drones, aircraft, narrowband physical network devices, machine-type communication devices, land vehicles, automobiles, wearable devices, or any other similar functional devices. Those skilled in the art may also refer to the UE 201 as a mobile station, subscriber station, mobile unit, subscriber unit, radio unit, remote unit, mobile device, radio device, wireless communication device, remote device, mobile subscriber station, access terminal, mobile terminal, radio terminal, remote terminal, handheld device, user agent, mobile client, client, or any other suitable term. The Node B 203 is connected to the core network 210 via an S1 / NG interface.The core network 210 includes an MME (Mobility Management Entity) / AMF (Authentication Management Field) / SMF (Session Management Function) 211, other MMEs / AMFs / SMFs 214, an S-GW (Service Gateway) / UPF (User Plane Function) 212, and a P-GW (Packet Data Network Gateway) / UPF 213. The MME / AMF / SMF 211 is the control node that handles signaling between the UE 201 and the core network 210. Generally, the MME / AMF / SMF 211 provides bearer and connection management. All user IP (Internet Protocol) packets are transmitted through the S-GW / UPF 212, which is itself connected to the P-GW / UPF 213. The P-GW provides UE IP address allocation and other functions. The P-GW / UPF 213 is connected to the Internet service 230. The Internet service 230 includes carrier-compliant Internet protocol services, specifically including the Internet, intranet, IMS (IP Multimedia Subsystem), and packet-switched streaming services.

[0213] As an example, the first node in this application includes the UE 201.

[0214] As an example, the second node in this application includes node B 203.

[0215] As an example, node B 203 is a macrocell base station.

[0216] As an example, node B 203 is a microcell base station.

[0217] As an example, node B 203 is a pico cell base station.

[0218] As an example, node B 203 is a femtocell.

[0219] As an example, node B 203 is a base station device that supports large latency differences.

[0220] As an example, node B 203 is a flight platform device.

[0221] As an example, node B 203 is a satellite device.

[0222] As one embodiment, the node B 203 is a test device (e.g., a transceiver device simulating part of the functions of a base station, a signaling tester).

[0223] As an example, the UE 201 includes a mobile phone.

[0224] As an example, the UE 201 is a vehicle including a car.

[0225] As an example, the wireless link from the UE 201 to the node B 203 is an uplink, which is used to perform uplink transmissions.

[0226] As an example, the radio link from the node B 203 to the UE 201 is a downlink, which is used to perform downlink transmissions.

[0227] As an example, the wireless link between the node B 203 and the UE 201 includes a cellular link.

[0228] As an example, the node B 203 and the UE 201 are connected via the Uu air interface.

[0229] As an example, the node B 203 supports the deployment of network-side (NW-side) AI / ML models.

[0230] As an example, the UE 201 supports the deployment of UE-side AI / ML models.

[0231] As an example, the UE 201 supports a 5G system.

[0232] As an example, the node B 203 supports a 5G system.

[0233] As an example, the UE 201 supports at least a 6G system.

[0234] As an example, the node B 203 supports at least a 6G system.

[0235] As an example, the sender of the first dataset in this application includes the node B 203.

[0236] As an example, the recipient of the first dataset in this application includes the UE 201.

[0237] As an example, the sender of the first reported information in this application includes the UE 201.

[0238] As an example, the recipient of the first reported information in this application includes the node B 203.

[0239] As an example, the sender of the first signaling in this application includes the node B 203.

[0240] As an example, the recipient of the first signaling in this application includes the UE 201.

[0241] As an example, the sender of the second parameter set in this application includes the node B 203.

[0242] As an example, the recipient of the second parameter set in this application includes the UE 201.

[0243] As an example, the sender of the second reported information in this application includes the UE 201.

[0244] As an example, the recipient of the second reported information in this application includes the node B 203.

[0245] As an example, the sender of the second signaling in this application includes the UE 201.

[0246] As an example, the recipient of the second signaling in this application includes the node B 203.

[0247] Example 3

[0248] Example 3 illustrates a schematic diagram of an embodiment of a wireless protocol architecture for the user plane and control plane according to an embodiment of this application, as shown in the attached diagram. Figure 3 As shown.

[0249] Figure 3 This is a schematic diagram illustrating an embodiment of a wireless protocol architecture for the user plane 350 and the control plane 300. Figure 3The wireless protocol architecture for the control plane 300 between the first communication node device (UE or RSU in V2X, onboard equipment or onboard communication module) and the second node device (gNB, UE or RSU in V2X, onboard equipment or onboard communication module), or between two UEs, is illustrated using three layers: Layer 1 (L1), Layer 2 (L2), and Layer 3 (L3). L1 is the lowest layer and implements various PHY (Physical layer) signal processing functions. L1 will be referred to as PHY 301 in this document. L2305 sits above PHY 301 and is responsible for the link between the first and second node devices, or between two UEs, via PHY 301. L2305 includes a MAC (Medium Access Control) sublayer 302, an RLC (Radio Link Control) sublayer 303, and a PDCP (Packet Data Convergence Protocol) sublayer 304, which terminate at the second node device. The PDCP sublayer 304 provides multiplexing between different radio bearers and logical channels. It also provides security through encrypted data packets and supports cross-cell mobility between the second communication node devices and the first communication node device. The RLC sublayer 303 provides upper-layer packet segmentation and reassembly, retransmission of lost packets, and packet reordering to compensate for out-of-order reception due to HARQ (Hybrid Automatic Repeat Quest). The MAC sublayer 302 provides multiplexing between logical and transport channels. It is also responsible for allocating various radio resources (e.g., resource blocks) within a cell among the first communication node devices. The MAC sublayer 302 is also responsible for HARQ operations. The RRC (Radio Resource Control) sublayer 306 in L3 of the control plane 300 is responsible for obtaining radio resources (i.e., radio bearers) and using RRC signaling between the second communication node device and the first communication node device to configure the lower layer.The wireless protocol architecture of user plane 350 includes Layer 1 (L1) and Layer 2 (L2). The wireless protocol architecture for the first and second communication node devices in user plane 350 is largely the same as the corresponding layers and sublayers in control plane 300 for Physical Layer 351, PDCP sublayer 354 in L2355, RLC sublayer 353 in L2355, and MAC sublayer 352 in L2355. However, PDCP sublayer 354 also provides header compression for upper-layer packets to reduce wireless transmission overhead. L2355 in user plane 350 also includes SDAP (Service Data Adaptation Protocol) sublayer 356. SDAP sublayer 356 is responsible for mapping between QoS (Quality of Service) streams and Data Radio Bearer (DRB) to support service diversity. Although not illustrated, the first communication node device may have several upper layers above L2355, including a network layer (e.g., IP (Internet Protocol) layer) terminating at the P-GW on the network side and an application layer terminating at the other end of the connection (e.g., remote UE, server, etc.).

[0250] As an example, Appendix Figure 3 The wireless protocol architecture described herein is applicable to the first node in this application.

[0251] As an example, Appendix Figure 3 The wireless protocol architecture described herein is applicable to the second node in this application.

[0252] As an example, the higher layer mentioned in this application refers to the layer above the physical layer.

[0253] As an example, in this application, the first dataset is generated by the MAC302 or MAC352.

[0254] As an example, the first dataset in this application is generated in the RRC306.

[0255] As an example, in this application, the first dataset is generated on the core network above RRC306.

[0256] As an example, the first reporting information in this application is generated in the RRC306.

[0257] As an example, in this application, the first reported information is generated in the core network above the RRC306.

[0258] As an example, in this application, the first signaling is generated in the PHY301 or PHY351.

[0259] As an example, in this application, the first signaling is generated in MAC302 or MAC352.

[0260] As an example, in this application, the first signaling is generated in the RRC306.

[0261] As an example, in this application, the first signaling is generated in the core network above the RRC306.

[0262] As an example, the second parameter set in this application is generated in MAC302 or MAC352.

[0263] As an example, the second parameter set in this application is generated in the RRC306.

[0264] As an example, in this application, the second parameter set is generated in the core network above the RRC306.

[0265] As an example, in this application, the first reporting information is generated by MAC302 or MAC352.

[0266] As an example, the second reporting information in this application is generated in the PHY301 or PHY351.

[0267] As an example, the second reporting information in this application is generated by the MAC302 or MAC352.

[0268] As an example, the second reporting information in this application is generated in the RRC306.

[0269] As an example, the second reporting information in this application is generated in the core network above the RRC306.

[0270] As an example, the second signaling in this application is generated in the PHY301 or PHY351.

[0271] As an example, in this application, the second signaling is generated in MAC302 or MAC352.

[0272] As an example, the second signaling in this application is generated in the RRC306.

[0273] As an example, in this application, the second signaling is generated in the core network above the RRC306.

[0274] Example 4

[0275] Example 4 illustrates a schematic diagram of a first communication device and a second communication device according to an embodiment of this application, as shown in the attached diagram. Figure 4 As shown. (Attached) Figure 4 This is a block diagram of a first communication device 410 and a second communication device 450 communicating with each other in the access network.

[0276] The first communication device 410 includes a controller / processor 475, a memory 476, a receiver processor 470, a transmitter processor 416, a multi-antenna receiver processor 472, a multi-antenna transmitter processor 471, a transmitter / receiver 418, and an antenna 420.

[0277] The second communication device 450 includes a controller / processor 459, a memory 460, a data source 467, a transmitting processor 468, a receiving processor 456, a multi-antenna transmitting processor 457, a multi-antenna receiving processor 458, a transmitter / receiver 454, and an antenna 452.

[0278] In the transmission from the first communication device 410 to the second communication device 450, at the first communication device 410, upper-layer data packets from the core network are provided to the controller / processor 475. The controller / processor 475 implements L2 functionality. In the DL, the controller / processor 475 provides header compression, encryption, packet segmentation and reordering, multiplexing between logical and transport channels, and radio resource allocation to the second communication device 450 based on various priority metrics. The controller / processor 475 is also responsible for HARQ operation, retransmission of lost packets, and signaling to the second communication device 450. The transmit processor 416 and the multi-antenna transmit processor 471 implement various signal processing functions for L1 (i.e., the physical layer). Transmit processor 416 performs encoding and interleaving to facilitate forward error correction (FEC) at the second communication device 450, and mapping of signal clusters based on various modulation schemes (e.g., Binary Phase Shift Keying (BPSK), Quadrature Phase Shift Keying (QPSK), M-PSK, and M-Quadrature Amplitude Modulation (M-QAM)). Multi-antenna transmit processor 471 performs digital spatial precoding on the encoded and modulated symbols, including codebook-based precoding and non-codebook-based precoding, and beamforming processing, generating one or more parallel streams. The transmit processor 416 then maps each parallel stream to a subcarrier, multiplexes the modulated symbols with a reference signal (e.g., a pilot) in the time and / or frequency domains, and then uses an inverse fast fourier transform (IFFT) to generate a physical channel carrying the time-domain multicarrier symbol stream. The multi-antenna transmit processor 471 then performs transmit analog precoding / beamforming operations on the time-domain multicarrier symbol stream. Each transmitter 418 converts the baseband multicarrier symbol stream provided by the multi-antenna transmit processor 471 into an RF stream, which is then provided to a different antenna 420.

[0279] In the transmission from the first communication device 410 to the second communication device 450, at the second communication device 450, each receiver 454 receives a signal through its corresponding antenna 452. Each receiver 454 recovers the information modulated onto the radio frequency carrier and converts the radio frequency stream into a baseband multicarrier symbol stream, which is then provided to the receiver processor 456. The receiver processor 456 and the multi-antenna receiver processor 458 implement various L1 signal processing functions. The multi-antenna receiver processor 458 performs receive analog precoding / beamforming operations on the baseband multicarrier symbol stream from the receiver 454. The receiver processor 456 uses a Fast Fourier Transform (FFT) to convert the baseband multicarrier symbol stream after the receive analog precoding / beamforming operations from the time domain to the frequency domain. In the frequency domain, the physical layer data signal and the reference signal are demultiplexed by the receiver processor 456, where the reference signal is used for channel estimation, and the data signal is recovered in the multi-antenna receiver processor 458 after multi-antenna detection to recover any parallel stream destined for the second communication device 450. Symbols on each parallel stream are demodulated and recovered in the receive processor 456, generating soft decisions. The receive processor 456 then decodes and deinterleaves the soft decisions to recover the upper-layer data and control signals transmitted by the first communication device 410 over the physical channel. The upper-layer data and control signals are then provided to the controller / processor 459. The controller / processor 459 implements L2 functionality. The controller / processor 459 may be associated with a memory 460 storing program code and data. The memory 460 may be referred to as computer-readable media. In the DL, the controller / processor 459 provides multiplexing, packet reassembly, decryption, header decompression, and control signal processing between the transmission and logical channels to recover upper-layer packets from the core network. The upper-layer packets are then provided to all protocol layers above L2. Various control signals may also be provided to L3 for L3 processing. The controller / processor 459 is also responsible for error detection using Acknowledgement (ACK) and / or Negative Acknowledgement (NACK) protocols to support HARQ operation.

[0280] In the transmission from the second communication device 450 to the first communication device 410, at the second communication device 450, a data source 467 is used to provide upper-layer data packets to the controller / processor 459. The data source 467 represents all protocol layers above L2. Similar to the transmission functions at the first communication device 410 described in the DL, the controller / processor 459 implements header compression, encryption, packet segmentation and reordering, and multiplexing between logical and transport channels based on the radio resource allocation of the first communication device 410, implementing L2 functions for the user plane and control plane. The controller / processor 459 is also responsible for HARQ operations, retransmission of lost packets, and signaling to the first communication device 410. Transmit processor 468 performs modulation mapping and channel coding processing, while multi-antenna transmit processor 457 performs digital multi-antenna spatial precoding, including codebook-based and non-codebook-based precoding, and beamforming processing. Subsequently, transmit processor 468 modulates the generated parallel stream into a multi-carrier / single-carrier symbol stream. After analog precoding / beamforming operations in multi-antenna transmit processor 457, the stream is provided to different antennas 452 via transmitter 454. Each transmitter 454 first converts the baseband symbol stream provided by multi-antenna transmit processor 457 into a radio frequency symbol stream before providing it to antenna 452.

[0281] In the transmission from the second communication device 450 to the first communication device 410, the function at the first communication device 410 is similar to the receiving function at the second communication device 450 described in the transmission from the first communication device 410 to the second communication device 450. Each receiver 418 receives radio frequency signals through its corresponding antenna 420, converts the received radio frequency signals into baseband signals, and provides the baseband signals to the multi-antenna receiving processor 472 and the receiving processor 470. The receiving processor 470 and the multi-antenna receiving processor 472 jointly implement the L1 function. The controller / processor 475 implements the L2 function. The controller / processor 475 may be associated with a memory 476 storing program code and data. The memory 476 may be referred to as computer-readable media. The controller / processor 475 provides multiplexing, packet reassembly, decryption, header decompression, and control signal processing between the transmission and logical channels to recover upper-layer data packets from the second communication device 450. The upper-layer data packets from the controller / processor 475 may be provided to the core network. The controller / processor 475 is also responsible for error detection using ACK and / or NACK protocols to support HARQ operation.

[0282] As one embodiment, the second communication device 450 includes: at least one processor and at least one memory, the at least one memory including computer program code; the at least one memory and the computer program code are configured to be used with the at least one processor. The second communication device 450 apparatus at least first receives a first dataset, the first dataset being used in the first training; subsequently sends a first reporting message, the first reporting message indicating the performance of a first parameter set; the first parameter set being obtained through the first training; the first reporting message indicating that the performance of the first parameter set satisfies a first condition, the first condition being applicable to a target function, the first training being for the target function.

[0283] As one embodiment, the second communication device 450 includes: a memory storing a computer-readable instruction program that produces actions when executed by at least one processor, the actions including: receiving a first dataset used by the first training; and sending a first reporting message indicating the performance of a first parameter set.

[0284] As one embodiment, the first communication device 410 includes: at least one processor and at least one memory, the at least one memory including computer program code; the at least one memory and the computer program code are configured to be used with the at least one processor. The first communication device 410 firstly transmits a first dataset, which is used in the first training; subsequently receives first reporting information indicating the performance of a first parameter set; the first parameter set is obtained through the first training; the first reporting information indicates that the performance of the first parameter set meets a first condition, the first condition being applicable to a target function, and the first training being for the target function.

[0285] As one embodiment, the first communication device 410 includes: a memory storing a computer-readable instruction program that produces actions when executed by at least one processor, the actions including: sending a first dataset used by the first training; and receiving first reporting information indicating the performance of a first parameter set.

[0286] As an example, the first node in this application includes the second communication device 450.

[0287] As an example, the second node in this application includes the first communication device 410.

[0288] As an example, at least one of {the antenna 420, the transmitter 418, the transmission processor 416, the multi-antenna transmission processor 471, the controller / processor 475, and the memory 476} is used to transmit the first dataset; at least one of {the antenna 452, the receiver 454, the receiving processor 456, the multi-antenna receiving processor 458, the controller / processor 459, the memory 460, and the data source 467} is used to receive the first dataset.

[0289] As an example, at least one of {the antenna 452, the transmitter 454, the transmission processor 468, the multi-antenna transmission processor 457, the controller / processor 459, the memory 460, and the data source 467} is used to transmit the first reporting information; at least one of {the antenna 420, the receiver 418, the receiving processor 470, the multi-antenna receiving processor 472, the controller / processor 475, and the memory 476} is used to receive the first reporting information.

[0290] As an example, at least one of {the antenna 420, the transmitter 418, the transmission processor 416, the multi-antenna transmission processor 471, the controller / processor 475, and the memory 476} is used to transmit the first signaling; at least one of {the antenna 452, the receiver 454, the receiving processor 456, the multi-antenna receiving processor 458, the controller / processor 459, the memory 460, and the data source 467} is used to receive the first signaling.

[0291] As an example, at least one of {the antenna 420, the transmitter 418, the transmission processor 416, the multi-antenna transmission processor 471, the controller / processor 475, and the memory 476} is used to transmit the second parameter set; at least one of {the antenna 452, the receiver 454, the receiving processor 456, the multi-antenna receiving processor 458, the controller / processor 459, the memory 460, and the data source 467} is used to receive the second parameter set.

[0292] As an example, at least one of {the antenna 452, the transmitter 454, the transmitter processor 468, the multi-antenna transmitter processor 457, the controller / processor 459, the memory 460, and the data source 467} is used to transmit the second reporting information; at least one of {the antenna 420, the receiver 418, the receiver processor 470, the multi-antenna receiver processor 472, the controller / processor 475, and the memory 476} is used to receive the second reporting information.

[0293] As an example, at least one of {the antenna 452, the transmitter 454, the transmitter processor 468, the multi-antenna transmitter processor 457, the controller / processor 459, the memory 460, and the data source 467} is used to transmit the second signaling; at least one of {the antenna 420, the receiver 418, the receiver processor 470, the multi-antenna receiver processor 472, the controller / processor 475, and the memory 476} is used to receive the second signaling.

[0294] As an example, at least one of {the antenna 452, the receiver 454, the receiver processor 456, the multi-antenna receiver processor 458, the controller / processor 459, the memory 460, and the data source 467} is used as a response to the occurrence of the first event to stop the first training.

[0295] Example 5

[0296] Example 5 illustrates a flowchart of a transmission between a first node and a second node according to an embodiment of this application, as shown in the attached diagram. Figure 5 As shown. In the appendix Figure 5 In this embodiment, the first node U1 and the second node N2 communicate via a wireless link. It should be noted that the order in this embodiment does not limit the signal transmission order or the order of implementation in this application.

[0297] For the first node U1, the first dataset is received in step S510; and the first reporting information is sent in step S511.

[0298] For the second node N2, the first dataset is sent in step S520; and the first reporting information is received in step S521.

[0299] In Example 5, the first dataset is used by the first training; the first reported information indicates the performance of the first parameter set; the first parameter set is obtained through the first training; the first reported information indicates that the performance of the first parameter set meets a first condition, the first condition being applicable to the target function, and the first training being for the target function.

[0300] As an example, the first node U1 is the first node in this application.

[0301] As an example, the second node N2 is the second node in this application.

[0302] As one embodiment, the air interface between the second node N2 and the first node U1 includes a wireless interface between the base station equipment and the user equipment.

[0303] As one embodiment, the air interface between the second node N2 and the first node U1 includes a wireless interface between the relay node device and the user equipment.

[0304] As one embodiment, the air interface between the second node N2 and the first node U1 includes a wireless interface between user equipment and user equipment.

[0305] As one example, the second node N2 and the first node U1 communicate via the Uu interface.

[0306] As one example, the second node N2 is the maintenance base station of the serving cell of the first node U1.

[0307] As an example, after receiving the first dataset, the first node performs the first training.

[0308] As an example, the first node performs the first training accordingly.

[0309] Typically, only a portion of the data in the first dataset is used in the first training, and the first reported information indicates data in the first dataset that was not used in the first training.

[0310] Typically, the first dataset comprises multiple data subsets, and a portion of these subsets is used in the first training.

[0311] As an example, the first reported information indicates a subset of data in the first dataset that was not used for the first training.

[0312] As one embodiment, the plurality of data subsets are indexed sequentially, and the first reporting information indicates the index corresponding to the data subset in the first dataset that was not used for the first training.

[0313] As an example, the training dataset used in the first training includes multiple data groups, and each data group includes a target CSI and a CSI feedback.

[0314] As an example, the training dataset used in the first training includes multiple data groups, and each data group includes a sample of the encoded output and a sample of the input before encoding.

[0315] Typically, the first reported information indicates the association ID of the training dataset associated with the first parameter set or the association ID of the first parameter set.

[0316] As an example, the associated ID is a non-negative integer.

[0317] As an example, the association ID is associated with at least one RS (Reference Signal) resource set.

[0318] As an example, the association ID is associated with at least one CSI (Channel State Information) report configuration.

[0319] As an example, the training dataset associated with the first parameter set includes the first dataset.

[0320] As an example, the ID mentioned in this application refers to IDentify, proof.

[0321] As an example, the ID mentioned in this application refers to: IDentification, identity verification.

[0322] As an example, the ID mentioned in this application refers to: IDentity, identity, or identifier.

[0323] As an example, the ID mentioned in this application refers to: Identifier, identifier.

[0324] As an example, the ID mentioned in this application refers to: InDex, index.

[0325] As an example, the ID mentioned in this application refers to: InDicator, indicator.

[0326] Example 6

[0327] Example 6 illustrates a flowchart of a first signaling transmission according to an embodiment of this application, as shown in the appendix. Figure 6 As shown. In the appendix Figure 6 In this embodiment, the first node U3 and the second node N4 communicate via a wireless link. It should be noted that the order in this embodiment does not limit the signal transmission order or the order of implementation in this application.

[0328] For the first node U3, the first signaling is received in step S610.

[0329] For the second node N4, the first signaling is sent in step S620.

[0330] In Example 6, the first signaling configures the first condition.

[0331] As an example, the first signaling is transmitted via a non-air interface.

[0332] As one embodiment, the first signaling includes RRC signaling.

[0333] As one example, the first signaling includes one or more RRC IEs.

[0334] As one example, the first signaling includes one or more fields in an RRC IE.

[0335] As an example, the first signaling includes MAC CE.

[0336] As one example, the first signaling includes MAC signaling.

[0337] As one example, the first signaling includes DCI (Downlink Control Information).

[0338] As an example, the physical layer channel occupied by the first signaling includes PDCCH.

[0339] As an example, the first signaling explicitly indicates the first condition.

[0340] As an example, the first signaling implicitly indicates the first condition.

[0341] As an example, the first signaling directly indicates the first condition.

[0342] As an example, the first signaling indirectly indicates the first condition.

[0343] As an example, the first signaling configures the first condition.

[0344] As an example, step S610 is located before step S510 in Example 5.

[0345] As an example, step S610 is located after step S510 and before step S511 in Example 5.

[0346] As an example, step S620 is located before step S510 in Example 5.

[0347] As an example, step S620 is located after step S510 and before step S511 in Example 5.

[0348] Example 7

[0349] Example 7 illustrates a flowchart of a second parameter set transmission according to an embodiment of this application, as shown in the attached diagram. Figure 7 As shown. In the appendix Figure 7 In this embodiment, the first node U5 and the second node N6 communicate via a wireless link. It should be noted that the order in this embodiment does not limit the signal transmission order or the order of implementation in this application.

[0350] For the first node U5, the second parameter set is received in step S710.

[0351] For the second node N6, the second parameter set is sent in step S720.

[0352] In Example 7, the first condition includes a first threshold, and satisfying the first condition means that the performance of the first parameter set is worse than the first threshold; the second parameter set is used for the target function.

[0353] As an example, the second set of parameters is associated with an AI / ML model.

[0354] As an example, the second set of parameters is associated with the target function.

[0355] As an example, the second parameter set is associated with an association ID.

[0356] As an example, the second parameter set includes at least one parameter.

[0357] As one example, the second parameter set includes multiple parameters.

[0358] As an example, the second parameter set corresponds to an Encoder.

[0359] As an example, the second parameter set corresponds to a Reconstructor.

[0360] As an example, the structure of the AI / ML model corresponding to the second parameter set is one of the following: Transformer structure, RNN structure, or CNN structure.

[0361] As an example, the structure of the AI / ML model corresponding to the second parameter set is the structure of a hybrid model composed of multiple models, including at least one of the Transformer structure, RNN structure, or CNN structure.

[0362] As an example, the second parameter set is generated by a third node, which is a node other than the first node.

[0363] As a sub-implementation of this embodiment, the third parameter set is the training result of the third node.

[0364] As a sub-implementation of this embodiment, the third node is a node other than the second node in this application.

[0365] As a sub-example of this embodiment, the third node includes core network equipment that provides services for terminal AI.

[0366] As a sub-example of this embodiment, the third node includes an application layer device that provides services for terminal AI.

[0367] As a sub-example of this embodiment, the third node includes an Agent that provides services for the terminal AI.

[0368] As a sub-implementation of this embodiment, the third node includes a Handset.

[0369] As a sub-example of this embodiment, the first node and the third node belong to the same terminal group.

[0370] As a sub-example of this embodiment, the first node and the third node belong to the same Vendor.

[0371] As a sub-example of this embodiment, the first node and the third node belong to the same Vendorlist.

[0372] As a sub-example of this embodiment, the first node and the third node are served by the same serving cell.

[0373] As one example, the first condition includes a first threshold.

[0374] As a sub-implementation of this embodiment, the first threshold is predefined or configurable.

[0375] As a sub-implementation of this embodiment, the first condition includes GCS (Generalized Cosine Similarity), and the first threshold corresponds to a GCS value.

[0376] As a sub-implementation of this embodiment, the first condition includes SGSC (Squared Generalized Cosine Similarity), and the first threshold corresponds to an SGSC value.

[0377] As a sub-implementation of this embodiment, the first condition includes NMSE (NormalizedMeanSquaredError), and the first threshold corresponds to an NMSE value.

[0378] As a sub-implementation of this embodiment, the first condition includes equivalent MSE (equivalent Mean Squared Error), and the first threshold corresponds to an equivalent MSE value.

[0379] As a sub-implementation of this embodiment, the first condition includes the numerical spectral efficiency gap, and the first threshold corresponds to a numerical spectral efficiency gap value.

[0380] As a sub-implementation of this embodiment, the first condition includes throughput, and the first threshold corresponds to a throughput value.

[0381] As a sub-implementation of this embodiment, the first condition includes BLER (Block Error Rate), and the first threshold corresponds to a BLER value.

[0382] As a sub-implementation of this embodiment, the first condition includes a hypothetical BLER, and the first threshold corresponds to a hypothetical BLER value.

[0383] As a sub-implementation of this embodiment, the first condition includes truth ground CSI, and the first threshold corresponds to a truth ground CSI value.

[0384] As a sub-implementation of this embodiment, the first condition includes MSE (Mean Squared Error), and the first threshold corresponds to an MSE value.

[0385] As a sub-implementation of this embodiment, the first condition includes MAE (Mean Absolute Error), and the first threshold corresponds to an MAE value.

[0386] As a sub-implementation of this embodiment, the first condition includes RMSE (Root Mean Squared Error), and the first threshold corresponds to an RMSE value.

[0387] As a sub-implementation of this embodiment, the first condition includes cosine similarity, and the first threshold corresponds to a cosine similarity value.

[0388] As a sub-implementation of this embodiment, the first condition includes convergence speed, and the first threshold corresponds to a convergence speed value.

[0389] As an example, satisfying the first condition means that the performance of the first parameter set is higher than the first threshold included in the first condition.

[0390] As an example, satisfying the first condition means that the performance of the first parameter set is lower than the first threshold included in the first condition.

[0391] As an example, the second parameter set is used for training the target function.

[0392] As an example, the second parameter set is used to generate the parameter set of the target function.

[0393] As an example, the second parameter set is used as the parameter set for the target function.

[0394] As an example, the second parameter set is used as the parameter set of the model for the target function.

[0395] As an example, the second parameter set is used for inference of the model of the target function.

[0396] As an example, the second set of parameters is used to predict the model of the target function.

[0397] As an example, the second parameter set is used for inference of the target function.

[0398] As an example, the second set of parameters is used for the prediction of the target function.

[0399] Typically, the first node stops training for the target function.

[0400] As an example, when the first node receives the second parameter set, the first node stops training for the target function.

[0401] As an example, the signaling carrying the second parameter set is used to trigger the first node to stop training for the target function.

[0402] As an example, the second node indicates via signaling that the second parameter set is used for the target function.

[0403] As a sub-implementation of this embodiment, the signaling and the second parameter set are simultaneously sent by the second node.

[0404] As a sub-implementation of this embodiment, the signaling indicates the target function.

[0405] As an example, step S710 is located after step S511 in Example 5.

[0406] As an example, step S720 is located after step S521 in Example 5.

[0407] Example 8

[0408] Example 8 illustrates a flowchart of a second reporting information transmission according to an embodiment of this application, as shown in the attached diagram. Figure 8 As shown. In the appendix Figure 8 In the middle, in the appendix Figure 8 In this embodiment, the first node U7 and the second node N8 communicate via a wireless link. It should be noted that the order in this embodiment does not limit the signal transmission order or the order of implementation in this application.

[0409] For the first node U7, the second reporting information is sent in step S810.

[0410] For the second node N8, the second reporting information is received in step S820.

[0411] In Example 8, the first condition includes a second threshold, and satisfying the first condition means that the performance of the first parameter set is better than the second threshold; the second reported information includes the first parameter set.

[0412] As one embodiment, the physical layer channel occupied by the second reported information includes PUSCH.

[0413] As one embodiment, the transmission channel occupied by the second reported information includes UL-SCH.

[0414] As one example, the second reported information is reported via RRC signaling.

[0415] As an example, the second reported information is reported via MAC CE.

[0416] As one example, the second reported information is reported via MAC signaling.

[0417] As one example, the first condition includes a second threshold.

[0418] As a sub-implementation of this embodiment, the second threshold is predefined or configurable.

[0419] As a sub-implementation of this embodiment, the first condition includes GCS, and the second threshold corresponds to a GCS value.

[0420] As a sub-example of this embodiment, the first condition includes SGSC, and the second threshold corresponds to an SGSC value.

[0421] As a sub-implementation of this embodiment, the first condition includes NMSE, and the second threshold corresponds to an NMSE value.

[0422] As a sub-implementation of this embodiment, the first condition includes equivalent MSE, and the second threshold corresponds to an equivalent MSE value.

[0423] As a sub-implementation of this embodiment, the first condition includes the numerical spectral efficiency gap, and the second threshold corresponds to a numerical spectral efficiency gap value.

[0424] As a sub-implementation of this embodiment, the first condition includes throughput, and the second threshold corresponds to a throughput value.

[0425] As a sub-implementation of this embodiment, the first condition includes BLER, and the second threshold corresponds to a BLER value.

[0426] As a sub-implementation of this embodiment, the first condition includes a hypothetical BLER, and the second threshold corresponds to a hypothetical BLER value.

[0427] As a sub-implementation of this embodiment, the first condition includes truth ground CSI, and the second threshold corresponds to a truth ground CSI value.

[0428] As a sub-implementation of this embodiment, the first condition includes MSE, and the second threshold corresponds to an MSE value.

[0429] As a sub-implementation of this embodiment, the first condition includes MAE, and the second threshold corresponds to an MAE value.

[0430] As a sub-implementation of this embodiment, the first condition includes RMSE, and the second threshold corresponds to an RMSE value.

[0431] As a sub-implementation of this embodiment, the first condition includes cosine similarity, and the second threshold corresponds to a cosine similarity value.

[0432] As a sub-implementation of this embodiment, the first condition includes convergence speed, and the second threshold corresponds to a convergence speed value.

[0433] As an example, satisfying the first condition means that the performance of the first parameter set is higher than the second threshold included in the first condition.

[0434] As an example, satisfying the first condition means that the performance of the first parameter set is lower than the second threshold included in the first condition.

[0435] As an example, the first threshold and the second threshold correspond to the same performance metric.

[0436] As an example, the first threshold and the second threshold correspond to different performance metrics.

[0437] As an example, the first threshold is different from the second threshold.

[0438] As an example, the second threshold is superior to the first threshold.

[0439] As an example, step S810 is located after step S511 in Example 5.

[0440] As an example, step S820 is located after step S521 in Example 5.

[0441] Example 9

[0442] Example 9 illustrates a flowchart of a second signaling transmission according to an embodiment of this application, as shown in the attached diagram. Figure 9As shown. In the appendix Figure 9 The first node U9 and the second node N10 communicate via a wireless link. It should be noted that the order in this embodiment does not limit the signal transmission order or the order of implementation in this application.

[0443] For the first node U9, in step S910, as a response to the occurrence of the first event, the first training is stopped; in step S911, the second signaling is sent.

[0444] For the second node N10, the second signaling is received in step S920.

[0445] In Example 9, the second signaling indicates that the first training is terminated; the candidates for the first event include at least one of cell handover, overheating detection, and entering power-saving mode.

[0446] As an example, the physical layer channel occupied by the second signaling includes PUSCH.

[0447] As one example, the transmission channel occupied by the second signaling includes UL-SCH.

[0448] As an example, the second signaling is reported via RRC signaling.

[0449] As an example, the second signaling is reported via MAC CE.

[0450] As an example, the second signaling is reported via MAC signaling.

[0451] As an example, the second signaling indicates that the first performance requirement was not met when the first training stopped.

[0452] As an example, the second signaling indicates that the first dataset was not fully used when the first training stopped.

[0453] As one embodiment, the first reported information and the second signaling are transmitted through a first message.

[0454] As a sub-implementation of this embodiment, the first reporting information and the second signaling are respectively two fields of the first message.

[0455] As a sub-implementation of this embodiment, the first message includes RRC signaling.

[0456] As a sub-implementation of this embodiment, the first message includes MAC CE.

[0457] As a sub-implementation of this embodiment, the first message includes MAC signaling.

[0458] As a sub-implementation of this embodiment, the first message includes a NAS message.

[0459] As one embodiment, the second reporting information and the second signaling are transmitted through the first message.

[0460] As a sub-implementation of this embodiment, the second reporting information and the second signaling are respectively two fields of the first message.

[0461] As a sub-implementation of this embodiment, the first message includes RRC signaling.

[0462] As a sub-implementation of this embodiment, the first message includes MAC CE.

[0463] As a sub-implementation of this embodiment, the first message includes MAC signaling.

[0464] As a sub-implementation of this embodiment, the first message includes a NAS message.

[0465] As an example, candidates for the first event include the cell handover.

[0466] As an example, candidates for the first event include the detection of overheating.

[0467] As a sub-example of this embodiment, the overheating detection refers to the detection of overheating of the first node.

[0468] As a sub-example of this embodiment, the overheating detection refers to the detection of overheating of the Handset corresponding to the first node.

[0469] As a sub-example of this embodiment, the overheating detection refers to the detection of overheating of the terminal corresponding to the first node.

[0470] As an example, candidates for the first event include entering power-saving mode.

[0471] As a sub-implementation of this embodiment, entering the power-saving mode means that the first node enters the power-saving mode.

[0472] As a sub-implementation of this embodiment, entering the power-saving mode means that the Handset corresponding to the first node enters the power-saving mode.

[0473] As a sub-example of this embodiment, entering the power-saving mode means that the terminal corresponding to the first node enters the power-saving mode.

[0474] Typically, the target cell in the cell handover does not support the first dataset or the first function.

[0475] As one example, the second signaling indicates the cell handover.

[0476] As a sub-implementation of this embodiment, the second signaling includes a measurement report message.

[0477] As a sub-implementation of this embodiment, the second signaling indicates the target cell for the cell handover.

[0478] As one example, the second signaling indicates that overheating has been detected.

[0479] As a sub-implementation of this embodiment, the second signaling includes a UEAssistanceInformation message.

[0480] As a sub-implementation of this embodiment, the second signaling includes OverheatingAssistance IE.

[0481] As one embodiment, the second signaling indicates that the power-saving mode has been entered.

[0482] As a sub-implementation of this embodiment, the name of the message used to carry the second signaling includes Energy.

[0483] As a sub-implementation of this embodiment, the name of the message used to carry the second signaling includes Saving.

[0484] As a sub-implementation of this embodiment, the name of the message used to carry the second signaling includes Handset.

[0485] As a sub-implementation of this embodiment, the name of the message used to carry the second signaling includes UE.

[0486] As an example, step S910 is located after step S510 and before step S511 in Example 5.

[0487] As an example, step S911 is located after step S510 and before step S511 in Example 5.

[0488] As an example, step S911 occurs simultaneously with step S511 in Example 5.

[0489] As an example, step S920 is located after step S520 and before step S521 in Example 5.

[0490] As an example, step S920 occurs simultaneously with step S521 of Example 5.

[0491] Example 10

[0492] Example 10 illustrates a schematic diagram of RAN domain AI / ML function deployment according to an embodiment of this application, as shown in the attached diagram. Figure 10 As shown. In the appendix Figure 10 In this context, gNB can be replaced with network equipment such as eNB or 6G base stations.

[0493] In Example 10, the management of the ML inference functions of multiple base stations is completed by the RAN domain management function 1002, that is, data interaction with the RAN domain MnS (Management Service) consumer / cross-domain management 1001 (as shown in the attached document). Figure 10 (As shown by the dashed arrow in the diagram). The RAN domain ML training function 1003 is located in the RAN domain management function 1002; while the ML inference function is located in the base station, that is, the AI / ML inference function 1004 is located in gNB 1005, the AI / ML inference function 1006 is located in gNB 1007, and so on.

[0494] AI / ML related functions include ML training (also known as AI training or AI / ML training), ML testing, and ML inference (also known as AI inference or AI / ML inference), etc. ML training, ML testing, and ML inference functions can be deployed independently or co-located. Deployment of AI / ML related functions can be implemented through software, such as downloading and / or running executable files; or it can be implemented through a combination of software and hardware, such as accelerating specific computing units through hardware to improve computing speed or save power.

[0495] ML training functions can be deployed in a cross-domain management system or a domain-specific management system; the domain-specific management system is used to manage the RAN domain or the CN (Core Network) domain. For example, ML training functions for MDA (Management Data Analytics) can be deployed in MDAF (Management Data Analytic Function); ML training for network data analytics can be deployed in NWDAF (Network Data Analytics Function), meaning the ML training function is an MTLF (Model Training Logical Function).

[0496] The ML inference function can also be deployed in a cross-domain management system or a domain-specific management system; for example, the ML inference function is MDAF, or the ML inference function is AnLF (Analytics Logical Function) located in NWDAF.

[0497] Similarly, ML testing capabilities can also be deployed in cross-domain management systems or domain-specific management systems.

[0498] Optionally, the management of ML inference function can also be completed by the base station itself, that is, each base station can independently interact with the RAN domain MnS consumer / cross-domain management 1001.

[0499] It should be noted that Embodiment 10 is merely a non-limiting implementation method; optionally, the ML training function of the RAN domain may also be deployed in the base station; or optionally, some base stations may deploy both the ML inference function and the ML training function of the RAN domain, while some base stations may only deploy the ML inference function.

[0500] As an example, one of the gNBs (or base stations) in Example 10 is the second node of this application.

[0501] Example 11

[0502] Example 11 illustrates a schematic diagram of the deployment of AI / ML functions in a UE according to an embodiment of this application, as shown in the attached diagram. Figure 11 As shown. In the appendix Figure 11 In this context, the RAN domain ML training function 1104 is optional.

[0503] UE function 1103 is deployed in the first node of this application, and the UE function 1103 includes AI / ML inference function 1105; the AI / ML inference function 1105 uses an ML model (also called an AI model) for inference; an ML model is typically trained before being used for AI / ML inference.

[0504] As an example, the UE function 1103 includes a RAN domain ML training function 1104, which runs training data through an ML model to obtain a relevant loss and adjusts the parameters of the ML model based on the calculated loss; the ML training includes at least one of ML initial training, ML re-training, and reinforcement learning.

[0505] The above embodiments can reduce the complexity of the base station, or save air interface resources caused by reporting training data; however, the above embodiments place high demands on the processing capabilities of the UE side.

[0506] Optionally, the UE function 1103 also includes a CN domain ML training function ( Figure 11 (Not included in the text).

[0507] Optionally, the UE function 1103 also includes an AI / ML deployment function. Figure 11 It is not included in the list, which is used to load ML models and data.

[0508] As an example, the first node indicates whether it supports ML training function (RAN domain or CN domain) through capability reporting. The capability reporting is RRC signaling or NAS (Non-Access Stratum) signaling.

[0509] As an example, the ML model and the associated metadata are loaded by the first node from a network device or a remote server.

[0510] Optionally, the UE function 1103 is an MnS producer that provides data to the CN domain MnF (Management Function) and / or the RAN domain MnF and / or the cross-domain management system 1101 for management or analysis (as shown by the double arrow 1102).

[0511] Optionally, the UE function 1103 is an MnS consumer that loads data from the CN domain MnF and / or RAN domain MnF and / or cross-domain management system 1101 for AI / ML-related management, such as managing data requests, ML model activation, and / or ML training (as shown by double arrow 1102).

[0512] As an example, the ML model is based on NN (Neural Networks).

[0513] As an example, the ML model is based on ANN (Artificial Neural Networks).

[0514] As an example, the ML model is based on CNN.

[0515] As an example, the ML model is based on the LLM (Large Language Model) architecture.

[0516] As an example, the ML model is based on the Transformer architecture.

[0517] As an example, the ML model is based on the GPT (Generative Pre-Trained) architecture.

[0518] As an example, the ML model is based on LSTM (Long Short-Term Memory network).

[0519] As an example, the ML model is based on MLP (MultiLayer Perceptron).

[0520] As an example, the ML model is based on GAN (Generative Adversarial Networks).

[0521] As an example, the ML model is based on a lightweight neural network.

[0522] As a sub-example of this embodiment, the lightweight neural network includes one or more of MobileNet, ShuffleNet, and SqueezeNet.

[0523] Example 12

[0524] Example 12 illustrates a schematic diagram of a processing system based on artificial intelligence or machine learning according to an embodiment of this application, as shown in the attached diagram. Figure 12 As shown. In the appendix Figure 12 In this context, the processing system based on artificial intelligence or machine learning includes a first processor, a second processor, a third processor, and a fourth processor.

[0525] In Example 12, the first processor sends a first dataset to the second processor and a second dataset to the third processor; the second processor generates a target first-class parameter set based on the first dataset, and sends the generated target first-class parameter set to the third processor; the third processor processes the second dataset using the target first-class parameter set to obtain a first-class output, optionally, the third processor sends the first-class output to the fourth processor. (See Appendix...) Figure 12 In this configuration, the first type of feedback and the second type of feedback are optional; the second processor includes ML training functionality; and the third processor includes ML inference functionality.

[0526] As one embodiment, the fourth processor includes ML testing functionality.

[0527] As one embodiment, the fourth processor includes performance monitoring / evaluation of the ML model.

[0528] As an example, the third processor sends a first type of feedback to the second processor; the first type of feedback is used to trigger the recalculation or update of the target first type of parameter set, that is, to trigger ML initial training or ML retraining.

[0529] As one embodiment, the fourth processor sends a second type of feedback to the first processor; the second type of feedback is used to generate the first dataset or the second dataset, or the second type of feedback is used to trigger the sending of the first dataset or the sending of the second dataset.

[0530] As one embodiment, the first processor generates the first dataset and the second dataset based on the measurement of the reference signal.

[0531] As one embodiment, the third processor belongs to the first node, and the fourth processor belongs to the second node.

[0532] As an example, the third processor belongs to the first node.

[0533] As an example, the first dataset includes training data.

[0534] As one embodiment, the second processor is used to train an ML model, and the trained model is described by the target first class of parameter sets.

[0535] As an example, the second processor belongs to the first node; the above method avoids passing the first dataset to the second node.

[0536] As an example, the second processor belongs to the second node in this application; the above method supports joint training and optimizes system performance.

[0537] As an example, the second processor belongs to the core network; the above method supports network-wide joint training, further optimizing system performance.

[0538] As an example, the second dataset includes inference data.

[0539] As an example, the third processor constructs a model based on the target first type of parameter group, and then inputs the second dataset into the constructed model to obtain the first type of output.

[0540] As one embodiment, the output of the third processor includes the performance parameters described in this application.

[0541] As an example, the third processor generates a recovery dataset based on the first type of output, and the error between the recovery dataset and the second dataset is used to generate the first type of feedback.

[0542] As an example, the first type of feedback is used to reflect the performance of the trained model; when the performance of the trained model fails to meet the requirements, the second processing opportunity will recalculate the target first type of parameter set.

[0543] As an example, when the error is too large or the update has not been performed for too long, the performance of the trained model is considered to be unsatisfactory.

[0544] As an example, the target first type of parameter group includes one or more of the following: convolution kernel, pooling kernel, pooling function, activation function, parameters of the pooling function, or parameters of the activation function.

[0545] As an example, the target first type of parameter group includes one or more of the following: convolution kernel size, number of convolution layers, convolution stride, pooling kernel size, pooling kernel stride, pooling function, activation function, or number of feature maps.

[0546] As an example, the target first type of parameter group corresponds to the first parameter set in this application.

[0547] As an example, the target first type of parameter group corresponds to the second parameter set in this application.

[0548] Example 13

[0549] Example 13 illustrates a schematic diagram based on artificial intelligence or machine learning according to an embodiment of this application, as shown in the attached diagram. Figure 13 As shown. In the appendix Figure 13 In this process, the first and second operations belong to the first stage, the third operation belongs to the second stage, the fourth operation belongs to the third stage, and the fifth operation belongs to the fourth stage; the arrowed lines indicate the sequence of the process.

[0550] As an example, the first operation includes AI / ML training, the second operation includes AI / ML testing, the third operation includes AI / ML emulation, the fourth operation includes AI / ML entity loading, and the fifth operation includes AI / ML inference.

[0551] As one embodiment, the first stage includes a training phase, the second stage includes an emulation phase, the third stage includes a deployment phase, and the fourth stage includes an inference phase.

[0552] As an example, the first stage includes AI / ML model training.

[0553] As an example, the first stage includes AI / ML model training and AI / ML testing.

[0554] As an example, the AI / ML model training includes initial training and re-training of one or a group of AI / ML entities.

[0555] As an example, the training of the AI / ML model depends on training data.

[0556] As an example, the AI / ML model training includes AI / ML entity validation.

[0557] As an example, the AI / ML entity verification is used to evaluate the performance of the AI / ML entity.

[0558] As an example, the AI / ML entity verification relies on verification data.

[0559] As an example, if the AI / ML entity verification results do not meet expectations, the AI / ML model will be retrained.

[0560] As an example, the AI / ML testing includes testing the validated AI / ML entities to estimate the performance of the trained AI / ML model.

[0561] As an example, if the AI / ML test results meet expectations, the AI / ML entity proceeds to the next stage; otherwise, the AI / ML model will be retrained.

[0562] As an example, the AI / ML test relies on test data.

[0563] As one embodiment, the second stage includes AI / ML simulation, which performs AI / ML entity reasoning in a simulation environment.

[0564] As an example, the AI / ML simulation estimates the performance of AI / ML entity reasoning in a simulation environment before using AI / ML entities.

[0565] As one embodiment, the second stage is optional.

[0566] As an example, the third stage includes AI / ML entity loading, which is to obtain trained AI / ML entities to obtain the desired AI / ML inference function.

[0567] As an example, the third stage is optional.

[0568] As an example, the third stage is no longer needed when the training and inference functions are co-located.

[0569] As an example, the fourth stage includes AI / ML inference.

[0570] Example 14

[0571] Example 14 illustrates a structural block diagram of a processing apparatus for a first node according to an embodiment of this application, as shown in the attached diagram. Figure 14 As shown. In the appendix Figure 14In the first node, the processing device 1400 includes a first receiver 1401 and a first transmitter 1402.

[0572] The first receiver 1401 receives the first dataset, which is used by the first training.

[0573] The first transmitter 1402 sends a first reporting message, which indicates the performance of the first parameter set;

[0574] In Example 14, the first parameter set is obtained through the first training; the first reported information indicates that the performance of the first parameter set meets a first condition, the first condition applies to the target function, and the first training is for the target function.

[0575] As an example, the first receiver 1401 receives a first signaling, which configures the first condition.

[0576] As an example, the first receiver 1401 receives a second set of parameters; the first condition includes a first threshold, wherein satisfying the first condition means that the performance of the first set of parameters is worse than the first threshold; the second set of parameters is used for the target function.

[0577] As an example, the first node stops training for the target function.

[0578] As an example, the first transmitter 1402 sends second reporting information; the first condition includes a second threshold, and satisfying the first condition means that the performance of the first parameter set is better than the second threshold; the second reporting information includes the first parameter set.

[0579] As an example, only a portion of the data in the first dataset is used in the first training, and the first reporting information indicates data in the first dataset that was not used in the first training.

[0580] As an example, in response to the occurrence of the first event, the first receiver 1401 stops the first training; the first transmitter 1402 sends a second signaling; the second signaling indicates that the first training is terminated; the candidates for the first event include at least one of cell handover, overheating detection, and entering a power-saving mode.

[0581] As an example, the first node 1400 is a user equipment.

[0582] As an example, the first node 1400 is a Handset.

[0583] As an example, the first node 1400 is a terminal.

[0584] As an example, the first receiver 1401 includes at least one of the following in embodiment 4: the antenna 452, the receiver 454, the receiver processor 456, the multi-antenna receiver processor 458, the controller / processor 459, the memory 460, and the data source 467.

[0585] As an example, the first transmitter 1402 includes at least one of the following in embodiment 4: the antenna 452, the transmitter 454, the transmission processor 468, the multi-antenna transmission processor 457, the controller / processor 459, the memory 460, and the data source 467.

[0586] Example 15

[0587] Example 15 illustrates a structural block diagram of a processing apparatus for a second node according to an embodiment of this application, as shown in the attached diagram. Figure 15 As shown. In the appendix Figure 15 In the second node, the processing device 1500 includes a second transmitter 1501 and a second receiver 1502.

[0588] The second transmitter 1501 transmits the first dataset, which is used by the first training.

[0589] The second receiver 1502 receives first reported information, which indicates the performance of the first parameter set;

[0590] In Example 15, the first parameter set is obtained through the first training; the first reported information indicates that the performance of the first parameter set meets a first condition, the first condition is applicable to the target function, and the first training is for the target function.

[0591] As an example, the second transmitter 1501 sends a first signaling message, which configures the first condition.

[0592] As an example, the second transmitter 1501 transmits a second set of parameters; the first condition includes a first threshold, wherein satisfying the first condition means that the performance of the first set of parameters is worse than the first threshold; the second set of parameters is used for the target function.

[0593] As an example, the sender of the first reporting information stops training for the target function.

[0594] As an example, the second receiver 1502 receives second reporting information; the first condition includes a second threshold, and satisfying the first condition means that the performance of the first parameter set is better than the second threshold; the second reporting information includes the first parameter set.

[0595] As an example, only a portion of the data in the first dataset is used in the first training, and the first reporting information indicates data in the first dataset that was not used in the first training.

[0596] As one embodiment, the second receiver 1502 receives a second signaling; the sender of the first reporting information stops the first training in response to the occurrence of the first event; the second signaling indicates that the first training is terminated; the candidates for the first event include at least one of cell handover, overheating detection, and entering a power-saving mode.

[0597] As an example, the second node 1500 is a base station device.

[0598] As one embodiment, the second node 1500 is a user equipment.

[0599] As an example, the second node 1500 is a TRP.

[0600] As an example, the second transmitter 1501 includes at least one of the following in embodiment 4: the antenna 420, the transmitter 418, the transmission processor 416, the multi-antenna transmission processor 471, the controller / processor 475, and the memory 476.

[0601] As one embodiment, the second receiver 1502 includes at least one of the following in embodiment 4: the antenna 420, the receiver 418, the receiver processor 470, the multi-antenna receiver processor 472, the controller / processor 475, and the memory 476.

[0602] Those skilled in the art will understand that all or part of the steps in the above methods can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium, such as a read-only memory, hard disk, or optical disk. Optionally, all or part of the steps in the above embodiments can also be implemented using one or more integrated circuits. Correspondingly, each module unit in the above embodiments can be implemented in hardware or in the form of software functional modules. This application is not limited to any specific combination of software and hardware. The user equipment, terminal, and UE in this application include, but are not limited to, drones, communication modules on drones, remote-controlled aircraft, aircraft, small aircraft, mobile phones, tablets, laptops, vehicle-mounted communication equipment, vehicles, RSUs, wireless sensors, internet cards, IoT terminals, RFID (Radio Frequency Identification) terminals, NB-IoT (Narrow Band Internet of Things) terminals, MTC (Machine Type Communication) terminals, eMTC (enhanced MTC) terminals, data cards, internet cards, vehicle-mounted communication equipment, low-cost mobile phones, low-cost tablets, and other wireless communication devices. The base stations or system equipment in this application include, but are not limited to, macrocell base stations, microcell base stations, small cell base stations, home base stations, relay base stations, eNB (evolved Node B), gNB, TRP, GNSS (Global Navigation Satellite System), relay satellites, satellite base stations, airborne base stations, RSUs, unmanned aerial vehicles, and test equipment, such as transceivers or signaling testers that simulate some functions of a base station, and other wireless communication equipment.

[0603] Those skilled in the art will understand that the present invention can be practiced in other specified forms without departing from its core or essential characteristics. Therefore, the embodiments disclosed herein should in any way be considered descriptive rather than restrictive. The scope of the invention is defined by the appended claims rather than the foregoing description, and all modifications within their equivalent meaning and scope are considered to be included therein.

Claims

1. A first node for model training in wireless communication, characterized in that... include: A first receiver receives a first dataset, which is used by a first training set. The first transmitter sends a first report, which indicates the performance of the first parameter set. The first parameter set is obtained through the first training; The first reported information indicates that the performance of the first parameter set meets a first condition, the first condition applies to the target function, and the first training is for the target function.

2. The first node according to claim 1, characterized in that... include: The first receiver receives the first signaling, which configures the first condition.

3. The first node according to claim 1 or 2, characterized in that... include: The first receiver receives the second parameter set; The first condition includes a first threshold, and satisfying the first condition means that the performance of the first parameter set is worse than the first threshold; the second parameter set is used for the target function.

4. The first node according to claim 3, characterized in that, The first node stops training for the target function.

5. The first node according to claim 1 or 2, characterized in that... include: The first transmitter sends the second reporting information; The first condition includes a second threshold, and satisfying the first condition means that the performance of the first parameter set is better than the second threshold; the second reported information includes the first parameter set.

6. The first node according to any one of claims 1 to 5, characterized in that, Only a portion of the data in the first dataset was used in the first training, and the first reported information indicates data in the first dataset that was not used in the first training.

7. The first node according to any one of claims 1 to 6, characterized in that... include: The first receiver, in response to the occurrence of the first event, stops the first training. The first transmitter sends the second signaling; The second signaling indicates that the first training is terminated; the candidates for the first event include at least one of cell handover, overheating detection, and entering power-saving mode.

8. A second node used for model training in wireless communication, characterized in that, include: The second transmitter sends the first dataset, which is used by the first training unit. The second receiver receives the first reported information, which indicates the performance of the first parameter set; The first parameter set is obtained through the first training; The first reported information indicates that the performance of the first parameter set meets a first condition, the first condition applies to the target function, and the first training is for the target function.

9. A method for the first node in model training for wireless communication, characterized in that, include: Receive the first dataset, which is used in the first training. Send a first report, the first report indicating the performance of the first parameter set; The first parameter set is obtained through the first training; The first reported information indicates that the performance of the first parameter set meets a first condition, the first condition applies to the target function, and the first training is for the target function.

10. A method for a second node in model training for wireless communication, characterized in that, include: Send the first dataset, which is used for the first training. Receive first reported information, the first reported information indicating the performance of the first parameter set; The first parameter set is obtained through the first training; The first reported information indicates that the performance of the first parameter set meets a first condition, the first condition applies to the target function, and the first training is for the target function.

Images