WO2025241999A1 - Communication method and communication apparatus - Google Patents

Abstract

The present application provides a communication method and a communication apparatus. The method comprises: receiving first information, the first information indicating one or more of the following: performance information of a first model, generalization information of the first model, or related information of a first data set, and the first data set being used for training the first model; and determining a monitoring means of the first model according to the first information. The solution of the embodiments of the present application helps improve the reliability and efficiency of model monitoring.

Description

Communication methods and communication devices

This application claims priority to Chinese Patent Application No. 202410627977.X, filed on May 20, 2024, entitled “Method and Apparatus for Communication”, the entire contents of which are incorporated herein by reference.

Technical Field

This application relates to the field of communications, and more specifically, to a method and apparatus for communication.

Background Technology

In wireless communication networks, such as mobile communication networks, the services supported by the network are becoming increasingly diverse, thus requiring increasingly diverse demands. To support these diverse services, artificial intelligence (AI) technology can be introduced into wireless communication networks to achieve network intelligence. Taking the use of an AI model for compressed feedback of channel information as an example, the user equipment (UE) obtains downlink channel information based on a reference signal. This downlink channel information is then used as input to the UE-side encoder model to obtain the feedback quantity of the channel information. The UE feeds back the feedback quantity of the channel information to the base station, which then inputs the feedback quantity into its base station-side decoder model to recover the downlink channel information.

However, the performance of AI models in real-world applications may not be stable. For example, as environmental conditions change, AI models may no longer adapt to the current communication environment, making it difficult to achieve the expected communication functions or guarantee stable communication performance.

Summary of the Invention

This application provides a communication method and communication device, which are intended to improve the reliability and efficiency of model monitoring.

In a first aspect, a method of communication is provided, which can be performed by a communication device or a module (e.g., a chip or circuit) applied to the communication device, wherein the communication device can be a first device in the method embodiment.

The first device can be a device on the terminal device side or a device on the network device side. The terminal device side can include at least one of a terminal device or an AI entity on the terminal device side. The AI entity on the terminal device side can be the terminal device itself or an AI entity serving the terminal device, such as a server, such as an over-the-top (OTT) server or a cloud server. The network device side can include at least one of a network device or an AI entity on the network device side. The AI entity on the network device side can be the network device itself or an AI entity serving the network device, such as a radio access network (RAN) intelligent controller (RIC), operation administration and maintenance (OAM), or a server, such as an OTT server or a cloud server.

The method includes: receiving first information, the first information indicating one or more of the following: performance information of a first model, generalization information of a first model, or relevant information of a first dataset, the first dataset being used for training the first model; and determining a monitoring method for the first model based on the first information.

In the embodiments of this application, the model's performance information, the model's generalization information, or the relevant information of the dataset can reflect the model's expected performance and/or the expected generalization ability of the first model. This is beneficial for determining a suitable monitoring method for the model, that is, determining a monitoring method that matches the model's performance and/or generalization, so as to achieve effective monitoring of models with different performance and/or generalization, thereby improving the reliability and efficiency of model monitoring.

For example, the first model may be deployed partially or entirely in the first device. The first device can monitor the first model according to the monitoring method of the first model.

Optionally, the method may further include: receiving third information, the third information indicating model monitoring of the first model.

The first device can monitor the first model based on the third information.

In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: sending a second message, the second message indicating the monitoring method of the first model.

For example, the first model may be partially or entirely deployed on the second device. The first device may send second information to the second device, and the second device may monitor the first model according to the monitoring method indicated by the second information.

In conjunction with the first aspect, in some implementations of the first aspect, the first information indicates one or more of the following: performance information of the first model, generalization information of the first model, or relevant information of the first dataset, through an identifier indicating the first model.

In conjunction with the first aspect, in some implementations of the first aspect, the performance information of the first model includes the performance level (i.e., performance grade) of the first model, which is related to the expected performance of the first model, and/or, the generalization information of the first model includes the generalization level of the first model, which is related to the expected generalization ability of the first model.

Optionally, the performance information of the first model may include the expected performance of the first model or the expected performance range of the first model.

Optionally, the generalization information of the first model may include the expected generalization ability of the first model or the expected generalization range of the first model.

In the embodiments of this application, the performance information of the first model can be used to reflect the expected performance of the first model. Determining the monitoring method of the first model based on its performance information is equivalent to determining the monitoring method based on its expected performance, which helps to obtain a monitoring method that matches the performance of the first model, thereby improving the reliability of model monitoring. The generalization information of the first model can be used to reflect its expected generalization ability. Determining the monitoring method of the first model based on its generalization information is equivalent to determining the monitoring method based on its expected generalization ability, which helps to obtain a monitoring method that matches the generalization ability of the first model, thereby improving the reliability and efficiency of model monitoring.

In conjunction with the first aspect, in some implementations of the first aspect, the first information also indicates the model category of the first model.

In conjunction with the first aspect, in some implementations of the first aspect, the model category of the first model includes a basic general model or a cell-specific model.

In the scheme of this application embodiment, the monitoring method of the first model is determined according to the model category of the first model. The relationship between the model category and the expected performance and/or expected generalization ability of the model is considered, so as to make the classification of expected performance and/or expected generalization ability more accurate and refined. This is conducive to selecting a monitoring method that matches the performance and/or generalization ability for models with different generalization, thereby improving the reliability of model monitoring.

In conjunction with the first aspect, in some implementations of the first aspect, the relevant information of the first dataset includes the identifier of the first dataset.

In the embodiments of this application, the dataset can reflect the expected performance and/or expected generalization ability of the model trained on the dataset. This is beneficial for matching the model's performance and/or generalization with the monitoring method, thereby facilitating effective monitoring of models with different performance and/or generalization, and improving the reliability and efficiency of monitoring.

In conjunction with the first aspect, in some implementations of the first aspect, the relevant information of the first dataset includes one or more of the following: performance information corresponding to the first dataset or generalization information corresponding to the first dataset.

In the scheme of this application embodiment, the performance information corresponding to the first dataset can be used to reflect the expected performance of the model trained on the first dataset. Determining the monitoring method of the first model based on the performance information corresponding to the first dataset is equivalent to determining the monitoring method of the first model based on the expected performance of the model trained on the first dataset. This is beneficial for obtaining a monitoring method that matches the performance of the first model, thereby improving the reliability of model monitoring. The generalization information corresponding to the first dataset can be used to reflect the expected generalization ability of the model trained on the first dataset. Determining the monitoring method of the first model based on the generalization information corresponding to the first dataset is equivalent to determining the monitoring method of the first model based on the expected generalization ability of the model trained on the first dataset. This is beneficial for obtaining a monitoring method that matches the generalization ability of the first model, thereby improving the reliability and efficiency of model monitoring.

In conjunction with the first aspect, in some implementations of the first aspect, the performance information corresponding to the first dataset includes the performance level of the model trained on the first dataset, the performance level of the model trained on the first dataset being related to the expected performance of the model trained on the first dataset, and/or, the generalization information corresponding to the first dataset includes the generalization level of the model trained on the first dataset, the generalization level of the model trained on the first dataset being related to the expected generalization ability of the model trained on the first dataset.

In conjunction with the first aspect, in some implementations of the first aspect, the monitoring parameters used in the monitoring method of the first model include at least one of the following: the performance threshold of the first model, the monitoring frequency of the first model, the monitoring duration of the first model, the number of monitoring sessions of the first model, the monitoring error tolerance of the first model, or the switching threshold of the first model.

Secondly, a communication method is provided, which can be executed by a communication device or a module (e.g., a chip or circuit) applied to the communication device, wherein the communication device can be a second device in the method embodiment.

The second device can be a device on the terminal device side or a device on the network device side. The terminal device side can include at least one of a terminal device or an AI entity on the terminal device side. The AI entity on the terminal device side can be the terminal device itself or an AI entity serving the terminal device, such as a server, like an OTT server or a cloud server. The network device side can include at least one of a network device or an AI entity on the network device side. The AI entity on the network device side can be the network device itself or an AI entity serving the network device, such as a RIC, OAM, or a server, like an OTT server or a cloud server.

The method includes: sending first information, the first information indicating one or more of the following: performance information of a first model, generalization information of a first model, or, relevant information of a first dataset, the first dataset being used for training the first model, and the first information being used to determine the monitoring method of the first model; and receiving second information, the second information indicating the monitoring method of the first model.

In the embodiments of this application, the model's performance information, the model's generalization information, or the relevant information of the dataset can reflect the model's expected performance and/or the expected generalization ability of the first model. This is beneficial for determining a suitable monitoring method for the model, that is, determining a monitoring method that matches the model's performance and/or generalization, so as to achieve effective monitoring of models with different performance and/or generalization, thereby improving the reliability and efficiency of model monitoring.

In conjunction with the second aspect, in some implementations of the second aspect, the first information indicates one or more of the following: performance information of the first model, generalization information of the first model, or relevant information of the first dataset, through an identifier indicating the first model.

In conjunction with the second aspect, in some implementations of the second aspect, the performance information of the first model includes the performance level of the first model, which is related to the expected performance of the first model, and/or, the generalization information of the first model includes the generalization level of the first model, which is related to the expected generalization ability of the first model.

In conjunction with the second aspect, in some implementations of the second aspect, the first information also indicates the model category of the first model.

In conjunction with the second aspect, in some implementations of the second aspect, the model category of the first model includes a basic general model or a cell-specific model.

In conjunction with the second aspect, in some implementations of the second aspect, the relevant information of the first dataset includes the identifier of the first dataset.

In conjunction with the second aspect, in some implementations of the second aspect, the relevant information of the first dataset includes one or more of the following: performance information corresponding to the first dataset or generalization information corresponding to the first dataset.

In conjunction with the second aspect, in some implementations of the second aspect, the performance information corresponding to the first dataset includes the performance level of the model trained on the first dataset, the performance level of the model trained on the first dataset is related to the expected performance of the model trained on the first dataset, and/or, the generalization information corresponding to the first dataset includes the generalization level of the model trained on the first dataset, the generalization level of the model trained on the first dataset is related to the expected generalization ability of the model trained on the first dataset.

In conjunction with the second aspect, in some implementations of the second aspect, the monitoring parameters used in the monitoring method of the first model include at least one of the following: the performance threshold of the first model, the monitoring frequency of the first model, the monitoring duration of the first model, the number of monitoring sessions of the first model, the monitoring error tolerance of the first model, or the switching threshold of the first model.

Thirdly, it can be performed by a communication device or a module (e.g., a chip or circuit) applied to the communication device, which can be the second device in the method embodiment.

The second device can be a device on the terminal device side or a device on the network device side. The terminal device side can include at least one of a terminal device or an AI entity on the terminal device side. The AI entity on the terminal device side can be the terminal device itself or an AI entity serving the terminal device, such as a server, like an OTT server or a cloud server. The network device side can include at least one of a network device or an AI entity on the network device side. The AI entity on the network device side can be the network device itself or an AI entity serving the network device, such as a RIC, OAM, or a server, like an OTT server or a cloud server.

The method includes: sending first information, the first information indicating one or more of the following: performance information of a first model, generalization information of a first model, or, relevant information of a first dataset, the first dataset being used for training the first model, and the first information being used to determine the monitoring method of the first model.

In the embodiments of this application, the model's performance information, the model's generalization information, or the relevant information of the dataset can reflect the model's expected performance and/or the expected generalization ability of the first model. This is beneficial for determining a suitable monitoring method for the model, that is, determining a monitoring method that matches the model's performance and/or generalization, so as to achieve effective monitoring of models with different performance and/or generalization, thereby improving the reliability and efficiency of model monitoring.

In conjunction with the third aspect, in some implementations of the third aspect, the method further includes: sending third information, which instructs the first model to be monitored.

In conjunction with the third aspect, in some implementations of the third aspect, the first information indicates one or more of the following: performance information of the first model, generalization information of the first model, or relevant information of the first dataset, through an identifier indicating the first model.

In conjunction with the third aspect, in some implementations of the third aspect, the performance information of the first model includes the performance level of the first model, which is related to the expected performance of the first model, and/or, the generalization information of the first model includes the generalization level of the first model, which is related to the expected generalization ability of the first model.

In conjunction with the third aspect, in some implementations of the third aspect, the first information also indicates the model category of the first model.

In conjunction with the third aspect, in some implementations of the third aspect, the model category of the first model includes a basic general model or a cell-specific model.

In conjunction with the third aspect, in some implementations of the third aspect, the relevant information of the first dataset includes the identifier of the first dataset.

In conjunction with the third aspect, in some implementations of the third aspect, the relevant information of the first dataset includes one or more of the following: performance information corresponding to the first dataset or generalization information corresponding to the first dataset.

In conjunction with the third aspect, in some implementations of the third aspect, the performance information corresponding to the first dataset includes the performance level of the model trained on the first dataset, the performance level of the model trained on the first dataset is related to the expected performance of the model trained on the first dataset, and/or, the generalization information corresponding to the first dataset includes the generalization level of the model trained on the first dataset, the generalization level of the model trained on the first dataset is related to the expected generalization ability of the model trained on the first dataset.

In conjunction with the third aspect, in some implementations of the third aspect, the monitoring parameters used in the monitoring method of the first model include at least one of the following: the performance threshold of the first model, the monitoring frequency of the first model, the monitoring duration of the first model, the number of monitoring sessions of the first model, the monitoring error tolerance of the first model, or the switching threshold of the first model.

Fourthly, a communication device is provided. This communication device can be a terminal device, or a device, module, circuit, or chip configured within a terminal device, or a device compatible with a terminal device. In one design, the communication device may include modules corresponding to the methods/operations/steps/actions described in the first aspect. These modules can be hardware circuits, software, or a combination of hardware circuits and software. In another design, the communication device may include a processing module and a communication module.

The sending module is used to perform the sending action in the method described in the first aspect above, the processing module is used to perform the processing action in the method described in any one of the first to third aspects above, and the receiving module is used to perform the receiving action in the method described in any one of the first to third aspects above.

Fifthly, a communication device is provided. This communication device can be a network device, or a device, module, circuit, or chip configured within a network device, or a device compatible with a network device. In one design, the communication device may include modules corresponding to each of the methods/operations/steps/actions described in the second aspect. These modules can be hardware circuits, software, or a combination of hardware circuits and software. In another design, the communication device may include a processing module and a communication module.

The receiving module is used to perform the receiving action in the method described in the second aspect above, the processing module is used to perform the processing action in the method described in any one of the first to third aspects above, and the sending module is used to perform the sending action in the method described in any one of the first to third aspects above.

A sixth aspect provides a communication device including one or more processors coupled to one or more storage media, the one or more storage media storing instructions that, when executed by the one or more processors, cause a method as in the first aspect or any possible implementation thereof to be implemented, cause a method as in the second aspect or any possible implementation thereof to be implemented, or cause a method as in the third aspect or any possible implementation thereof to be implemented.

A seventh aspect provides a communication apparatus comprising one or more processors, the one or more processors being configured to process data and/or information such that a method as in the first aspect or any possible implementation thereof is implemented, a method as in the second aspect or any possible implementation thereof is implemented, or a method as in the third aspect or any possible implementation thereof is implemented.

Optionally, the communication device may further include a communication interface for receiving data and/or information and transmitting the received data and/or information to the processor. Optionally, the communication interface may also be used to output data and/or information processed by the processor.

Eighthly, a chip is provided, including a processor, the processor being configured to execute a program or instructions to cause the method as described in the first aspect or any possible implementation thereof to be implemented, to cause the method as described in the second aspect or any possible implementation thereof to be implemented, or to cause the method as described in the third aspect or any possible implementation thereof to be implemented.

Optionally, the chip may further include a memory for storing programs or instructions. Optionally, the chip may further include the transceiver.

Optionally, the chip is an application-specific integrated circuit (ASIC) or a system-on-chip (SoC).

A ninth aspect provides a computer-readable storage medium comprising instructions that, when executed by a processor, cause a method as described in the first aspect or any possible implementation thereof to be implemented, a method as described in the second aspect or any possible implementation thereof to be implemented, or a method as described in the third aspect or any possible implementation thereof to be implemented.

In a tenth aspect, a computer program product is provided, the computer program product comprising computer program code or instructions that, when the computer program code or instructions are executed, cause the method as in the first aspect or any possible implementation thereof to be implemented, cause the method as in the second aspect or any possible implementation thereof to be implemented, or cause the method as in the third aspect or any possible implementation thereof to be implemented.

Eleventhly, a communication system is provided, comprising one or more of the following means: a communication device for performing the method of the first aspect or any possible implementation thereof, a communication device for performing the method of the second aspect or any possible implementation thereof, and a communication device for performing the method of the third aspect or any possible implementation thereof. For example, the communication system may include the communication device provided by the fourth aspect, and/or, the communication device provided by the fifth aspect.

Attached Figure Description

Figure 1 is a schematic diagram of a communication system applicable to an embodiment of this application;

Figure 2 is a schematic diagram of another communication system applicable to an embodiment of this application;

Figure 3 is a schematic diagram of another communication system applicable to an embodiment of this application;

Figure 4 is a schematic diagram of an application framework applicable to a communication system according to an embodiment of this application;

Figure 5 is a schematic flowchart of a communication method provided in an embodiment of this application;

Figure 6 is a schematic flowchart of another communication method provided in an embodiment of this application;

Figure 7 is a schematic flowchart of another communication method provided in an embodiment of this application;

Figure 8 is a schematic flowchart of another communication method provided in an embodiment of this application;

Figure 9 is a schematic flowchart of another communication method provided in an embodiment of this application;

Figure 10 is a schematic flowchart of another communication method provided in an embodiment of this application;

Figure 11 is a schematic block diagram of a communication device provided in an embodiment of this application;

Figure 12 is a schematic block diagram of another communication device provided in an embodiment of this application.

Detailed Implementation

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

The technical solutions provided in this application can be applied to various communication systems, such as: 5th generation (5G) or new radio (NR) systems, long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, wireless local area network (WLAN) systems, satellite communication systems, future communication systems such as future communication network mobile communication systems, or integrated systems of multiple systems. The technical solutions provided in this application can also be applied to device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, machine-to-machine (M2M) communication, machine-type communication (MTC), and Internet of Things (IoT) communication systems or other communication systems.

In a communication system, a device can send signals to or receive signals from another device. These signals can include information, signaling, or data. The device can also be replaced by an entity, network entity, network element, communication equipment, communication module, node, communication node, etc. This disclosure uses a device as an example. For instance, a communication system can include at least one terminal device and at least one network device. In this communication system, the network device can send downlink signals to the terminal device, the terminal device can send uplink signals to the network device, the network device can send signals to another network device, and the terminal device can send sidelink signals to another terminal device.

In the embodiments of this application, the terminal device may also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent, or user device.

Terminal devices can be devices that provide voice/data, such as handheld devices with wireless connectivity, in-vehicle devices, etc. Currently, some examples of terminals include: mobile phones, tablets, laptops, PDAs, mobile internet devices (MIDs), wearable devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in telemedicine, wireless terminals in smart grids, wireless terminals in transportation safety, and wireless terminals in smart cities. The embodiments of this application do not limit the scope to wireless terminals in smart homes, cellular phones, cordless phones, session initiation protocol (SIP) phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), handheld devices with wireless communication capabilities, computing devices or other processing devices connected to a wireless modem, wearable devices, terminal devices in 5G networks, or terminal devices in future evolved public land mobile networks (PLMNs).

By way of example and not limitation, in this embodiment, the terminal device can also be a wearable device. Wearable devices, also known as wearable smart devices, are a general term for devices that utilize wearable technology to intelligently design and develop everyday wearables, such as glasses, gloves, watches, clothing, and shoes. Wearable devices are portable devices that are worn directly on the body or integrated into the user's clothing or accessories. Wearable devices are not merely hardware devices, but also achieve powerful functions through software support, data interaction, and cloud interaction. Broadly speaking, wearable smart devices include those that are feature-rich, large in size, and can achieve complete or partial functions without relying on a smartphone, such as smartwatches or smart glasses, as well as those that focus on a specific type of application function and require the use of other devices such as smartphones, such as various smart bracelets and smart jewelry for vital sign monitoring.

In this embodiment, the device for implementing the functions of the terminal device can be the terminal device itself, or it can be any device capable of supporting the terminal device in implementing those functions, such as a chip system. This device can be installed in or used in conjunction with the terminal device. In this embodiment, the chip system can be composed of chips or may include chips and other discrete components. This embodiment only uses the terminal device as an example to illustrate the device for implementing the functions of the terminal device, and does not constitute a limitation on the solution of this embodiment.

The network device in this application embodiment may include a device for communicating with a terminal device. For example, the network device may include an access network device or a wireless access network device, such as a base station. The wireless access network device in this application embodiment may refer to a RAN node (or device) that connects the terminal device to the wireless network. A base station can broadly encompass various names such as, or be replaced by, the following: NodeB, evolved NodeB (eNB), next generation NodeB (gNB), relay station, access point, transmitting and receiving point (TRP), transmitting point (TP), master station, auxiliary station, motor slide retainer (MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed unit (DU), radio unit (RU), positioning node, etc. A base station can be a macro base station, micro base station, relay node, donor node, or a combination thereof. A base station can also refer to a communication module, modem, or chip installed within the aforementioned equipment or apparatus. A base station can also be a mobile switching center and equipment performing base station functions in D2D, V2X, and M2M communications, network-side equipment in future communication networks, or equipment performing base station functions in future communication systems. A base station can support networks using the same or different access technologies. Optionally, a RAN node can also be a server, wearable device, vehicle, or in-vehicle equipment. For example, access network equipment in V2X technology can be a roadside unit (RSU). The embodiments of this application do not limit the specific technologies or equipment forms used in the network equipment.

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

In some deployments, the network devices mentioned in the embodiments of this application may be devices including CU, DU, or CU and DU, or devices with control plane CU nodes (central unit-control plane (CU-CP)) and user plane CU nodes (central unit-user plane (CU-UP)) and DU nodes. For example, the network devices may include gNB-CU-CP, gNB-CU-UP, and gNB-DU.

In some deployments, multiple RAN nodes collaborate to assist terminals in achieving wireless access, with different RAN nodes each implementing some of the base station's functions. For example, RAN nodes can be CUs, DUs, CU-CPs, CU-UPs, or RUs. CUs and DUs can be configured separately or included in the same network element, such as a BBU. RUs can be included in radio frequency equipment or radio frequency units, such as RRUs, AAUs, or RRHs.

RAN nodes can support one or more types of fronthaul interfaces, and different fronthaul interfaces correspond to DU and RU with different functions.

If the fronthaul interface between the DU and RU is a common public radio interface (CPRI), the DU is configured to implement one or more baseband functions, and the RU is configured to implement one or more radio frequency functions.

If the fronthaul interface between DU and RU is a different interface, relative to CPRI, some baseband functions for downlink and/or uplink, such as, for downlink, precoding, digital beamforming (BF), or one or more of inverse fast fourier transform (IFFT)/cyclic prefix addition (CP), are moved from DU to RU; and for uplink, digital beamforming (BF), or one or more of fast fourier transform (FFT)/cyclic prefix removal (CP) are moved from DU to RU.

One possible implementation is that the interface can be an enhanced common public radio interface (eCPRI). In the eCPRI architecture, the segmentation between DU and RU differs, corresponding to different categories (Cat) of eCPRI, such as eCPRI Cat A, B, C, D, E, and F.

Taking eCPRI Cat A as an example, for downlink transmission, the DU is configured to implement one or more functions before and after the layer mapping (i.e., coding, rate matching, scrambling, modulation, layer mapping). Other functions after the layer mapping (e.g., resource element (RE) mapping, digital beamforming (BF), or one or more of inverse fast Fourier transform (IFFT)/adding a cyclic prefix (CP)) are moved to the RU for implementation. For uplink transmission, the de-RE mapping is used as the dividing line. The DU is configured to implement one or more functions preceding de-mapping (i.e., decoding, rate matching de-matching, descrambling, demodulation, inverse discrete Fourier transform (IDFT), channel equalization, and one or more functions in de-RE mapping). Other functions following de-mapping (e.g., one or more functions in digital BF or fast Fourier transform (FFT)/de-CP) are moved to the RU. It is understood that descriptions of the functions of the DU and RU corresponding to various types of eCPRI can be found in the eCPRI protocol and will not be elaborated upon here.

In one possible design, the processing unit in the BBU used to implement baseband functions is called the baseband high (BBH) unit, and the processing unit in the RRU/AAU/RRH used to implement baseband functions is called the baseband low (BBL) unit.

In different systems, CU (or CU-CP and CU-UP), DU, or RU may have different names, but those skilled in the art will understand their meaning. For example, in an open RAN (ORAN) system, CU can also be called O-CU (open CU), DU can also be called O-DU, CU-CP can also be called O-CU-CP, CU-UP can also be called O-CU-UP, and RU can also be called O-RU. Any of the units among CU (or CU-CP, CU-UP), DU, and RU in this application can be implemented through software modules, hardware modules, or a combination of software modules and hardware modules.

In this embodiment, the apparatus for implementing the functions of a network device can be a network device itself; it can also be an apparatus capable of supporting the network device in implementing those functions, such as a chip system, hardware circuit, software module, or a hardware circuit plus a software module. This apparatus can be installed in the network device or used in conjunction with the network device. In this embodiment, the example of a network device being used to implement the functions of a network device is provided only and does not constitute a limitation on the solutions described in this embodiment.

Network devices and/or terminal devices can be deployed on land, including indoors, outdoors, handheld, and/or vehicle-mounted; they can also be deployed on water (such as ships); and they can also be deployed in the air (such as airplanes, balloons, and/or satellites). The embodiments of this application do not limit the scenarios in which the network devices and terminal devices are located.

Furthermore, terminal devices and network devices can be hardware devices, software functions running on dedicated hardware, software functions running on general-purpose hardware, such as virtualization functions instantiated on a platform (e.g., a cloud platform), or entities that include dedicated or general-purpose hardware devices and software functions. This application does not limit the specific form of terminal devices and network devices.

In wireless communication networks, such as mobile communication networks, the services supported by the networks are becoming increasingly diverse, thus requiring increasingly diverse demands. For example, networks need to support ultra-high speeds, ultra-low latency, and/or massive connectivity. This characteristic makes network planning, network configuration, and/or resource scheduling increasingly complex. Furthermore, as network functions become more powerful, such as supporting higher spectrum levels, supporting higher-order multiple-input multiple-output (MIMO) technologies, supporting beamforming, and/or supporting beam management, network energy efficiency has become a hot research topic. These new demands, new scenarios, and new characteristics bring unprecedented challenges to network planning, operation, and efficient operation. To meet these challenges, artificial intelligence technology can be introduced into wireless communication networks to achieve network intelligence.

To support artificial intelligence (AI) technology in wireless networks, AI nodes (also known as AI entities) may be introduced into the network.

Optionally, the AI entity can be deployed in one or more of the following locations within the communication system: access network devices, terminal devices, or core network devices, or the AI entity can be deployed independently, for example, in a location other than any of the aforementioned devices, such as the host or cloud server of an OTT system. The AI entity can communicate with other devices in the communication system, which can be one or more of the following: network devices, terminal devices, or network elements of the core network. Based on the object served by the AI entity, the AI entity can include an AI entity on the network device side, an AI entity on the terminal device side, or an AI entity on the core network side.

It is understood that this application does not limit the number of AI entities. For example, when there are multiple AI entities, they can be divided based on function, such as different AI entities being responsible for different functions.

It can also be understood that AI entities can be independent devices, or they can be integrated into the same device to achieve different functions. Alternatively, they can be network components in hardware devices, software functions running on dedicated hardware, or virtualization functions instantiated on a platform (e.g., a cloud platform). This application does not limit the specific form of the aforementioned AI entities.

AI entities can be AI network elements or AI modules. AI entities are used to implement corresponding AI functions. AI modules deployed in different network elements can be the same or different. Depending on the different parameter configurations, the AI model within an AI entity can achieve different functions. The AI model within an AI entity can be configured based on one or more of the following parameters: structural parameters (e.g., at least one of the following: number of neural network layers, neural network width, inter-layer connections, neuron weights, neuron activation function, or biases in the activation function), input parameters (e.g., the type and/or dimension of the input parameters), or output parameters (e.g., the type and/or dimension of the output parameters). The biases in the activation function can also be referred to as the biases of the neural network.

An AI entity can have one or more models. A model can infer an output that includes one or more parameters. The learning, training, or inference processes of different models can be deployed on different entities or devices, or they can be deployed on the same entity or device.

Figure 1 is a schematic diagram of a communication system applicable to the communication method of this application embodiment. As shown in Figure 1, the communication system 100 may include at least one network device, such as network device 110 shown in Figure 1; the communication system 100 may also include at least one terminal device, such as terminal device 120 and terminal device 130 shown in Figure 1. Network device 110 and terminal devices (such as terminal device 120 and terminal device 130) can communicate via a wireless link. The communication devices in this communication system, for example, network device 110 and terminal device 120, can communicate via multi-antenna technology.

Figure 2 is a schematic diagram of another communication system applicable to the communication method of this application embodiment. Compared with the communication system 100 shown in Figure 1, the communication system 200 shown in Figure 2 further includes an AI network element 140. The AI network element 140 is used to perform AI-related operations, such as building training datasets or training AI models.

In one possible implementation, network device 110 can send data related to the training of the AI model to AI network element 140, which then constructs a training dataset and trains the AI model. For example, the data related to the training of the AI model may include data reported by the terminal device. AI network element 140 can send the results of operations related to the AI model to network device 110, which then forwards them to the terminal device. For example, the results of operations related to the AI model may include at least one of the following: a trained AI model, model evaluation results, or test results. Exemplarily, a portion of the trained AI model may be deployed on network device 110, and another portion on the terminal device. Alternatively, the trained AI model may be deployed on network device 110. Or, the trained AI model may be deployed on the terminal device.

It should be understood that Figure 2 is only used as an example of the AI network element 140 being directly connected to the network device 110. In other scenarios, the AI network element 140 can also be connected to a terminal device. Alternatively, the AI network element 140 can be connected to both the network device 110 and a terminal device simultaneously. Alternatively, the AI network element 140 can also be connected to the network device 110 through a third-party network element. This application embodiment does not limit the connection relationship between the AI network element and other network elements.

The AI network element 140 can also be configured as a module in network devices and/or terminal devices, for example, in network device 110 or terminal device as shown in Figure 1. One or more AI modules can be deployed in network device 110. One or more AI modules can be deployed in terminal device.

It should be noted that Figures 1 and 2 are simplified schematic diagrams for ease of understanding. For example, the communication system may also include other devices, such as wireless relay devices and/or wireless backhaul devices, which are not shown in Figures 1 and 2. In practical applications, the communication system may include multiple network devices or multiple terminal devices. The embodiments of this application do not limit the number of network devices and terminal devices included in the communication system.

Figure 3 is a schematic diagram of a possible application framework of a communication system according to an embodiment of this application. As shown in Figure 3, network elements in the communication system are connected through interfaces (e.g., NG, Xn) or air interfaces. These network element nodes, such as core network equipment, access network nodes (RAN nodes), terminal equipment, or one or more devices in operation administration and maintenance (OAM), are equipped with one or more AI modules (only one is shown in Figure 3 for clarity). The access network node can be a single RAN node or can include multiple RAN nodes, for example, including CU and DU. CU and/or DU can also be equipped with one or more AI modules. Optionally, CU can also be split into CU-CP and CU-UP. One or more AI models are provided in CU-CP and/or CU-UP. Exemplarily, CU and DU are connected through an F1 interface. CU and CU are connected through an Xn interface.

The network device can be a network device equipped with one or more AI modules. The network device can be one or more devices in the core network, access network (RAN) node, or OAM as shown in Figure 3. For example, the AI module can be the RIC shown in Figure 4, such as a near real-time RIC or a non-real-time RIC. For example, the near real-time RIC is set in the RAN node (e.g., in CU, DU), while the non-real-time RIC is set in the OAM, cloud server, core network device, or other network device. The RIC can obtain subsets from multiple terminal devices from the RAN node (e.g., CU, CU-CP, CU-UP, DU, and/or RU), reassemble them into a training dataset #2, and train based on the training dataset #2. Exemplarily, the near real-time RIC and the non-real-time RIC can also be set up separately as a network element; the network device can be a near real-time RIC or a non-real-time RIC.

Figure 4 illustrates a possible application framework in a communication system. As shown in Figure 4, the communication system includes a RAN intelligent controller (RIC). For example, the RIC can be the AI module shown in Figure 3, used to implement AI-related functions. RICs include near-real-time RICs (near-RT RICs) and non-real-time RICs (non-RT RICs). Non-real-time RICs primarily process non-real-time information, such as data that is not sensitive to latency, with latency in the order of seconds. Real-time RICs primarily process near-real-time information, such as data that is relatively sensitive to latency, with latency in the order of tens of milliseconds.

Near real-time (NRT) RICs are used for model training and inference. For example, they are used to train AI models and then use those models for inference. NRT RICs can obtain network-side and/or terminal-side information from RAN nodes (e.g., CUs, CU-CPs, CU-UPs, DUs, and/or RUs) and/or terminals. This information can be used as training data or inference data. Optionally, the NRT RIC can deliver inference results to RAN nodes and/or terminals. Optionally, inference results can be exchanged between CUs and DUs, and/or between DUs and RUs. For example, the NRT RIC delivers inference results to a DU, which then forwards them to an RU.

Non-real-time RICs are also used for model training and inference. For example, they can be used to train AI models and then use those models for inference. Non-real-time RICs can obtain network-side and/or terminal-side information from RAN nodes (e.g., CUs, CU-CPs, CU-UPs, DUs, and/or RUs) and/or terminals. This information can be used as training data or inference data, and the inference results can be delivered to RAN nodes and/or terminals. Optionally, inference results can be exchanged between CUs and DUs, and/or between DUs and RUs; for example, a non-real-time RIC delivers inference results to a DU, which then forwards them to an RU.

Near real-time RICs and non-real-time RICs can also be configured as separate network elements. Optionally, near real-time RICs and non-real-time RICs can also be part of other devices. For example, near real-time RICs can be set in RAN nodes (e.g., CU, DU), while non-real-time RICs can be set in OAM, cloud servers, core network devices, or other network devices.

To facilitate understanding of the solutions in the embodiments of this application, the terms that may be involved in the embodiments of this application are explained below.

(1) Neural network (NN):

Neural networks are a specific implementation of AI or machine learning (ML). According to the general approximation theorem, neural networks can theoretically approximate any continuous function, thus enabling them to learn arbitrary mappings.

Taking neural networks as an example, the AI model disclosed herein can be a deep neural network (DNN). Traditional communication systems require extensive expert knowledge to design communication modules, while deep learning communication systems based on deep neural networks (DNN) can automatically discover hidden pattern structures from large datasets, establish mapping relationships between data, and achieve performance superior to traditional modeling methods.

A neural network can be composed of neurons, each of which performs a weighted summation of its input values, and the result is then passed through a non-linear function to produce the output. DNNs typically have a multi-layered structure, with each layer containing multiple neurons. The input layer processes the received values through neurons and then passes them to the hidden layers. Similarly, the hidden layers then pass the calculation results to the final output layer, producing the final output of the DNN.

DNNs typically have more than one hidden layer, and these hidden layers often directly affect the ability to extract information and fit functions. Increasing the number of hidden layers or widening the width of each layer can improve the function fitting ability of a DNN. The weights in each neuron are the parameters of the DNN network model. The model parameters are optimized through the training process, enabling the DNN network to extract data features and express mapping relationships. DNNs generally use supervised or unsupervised learning strategies to optimize model parameters.

Depending on how the network is constructed, DNNs can include feedforward neural networks (FNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs), etc.

CNNs are neural networks specifically designed to process data with a grid-like structure. For example, time-series data (discrete sampling along the time axis) and image data (two-dimensional discrete sampling) can both be considered grid-like data. CNNs do not use all the input information at once for computation; instead, they use a fixed-size window to extract a portion of the information for convolution operations, which significantly reduces the computational cost of model parameters. Furthermore, depending on the type of information extracted by the window (such as people and objects in an image representing different types of information), each window can use different convolution kernels, allowing CNNs to better extract features from the input data.

Recurrent Neural Networks (RNNs) are a type of distributed neural network (DNN) that utilizes feedback time-series information. Their input includes the current input value and their own output value from the previous time step. RNNs are well-suited for acquiring temporally correlated sequence features, and are particularly applicable to applications such as speech recognition and channel coding/decoding.

The characteristic of FNN networks is that neurons in adjacent layers are completely connected to each other, which makes FNNs typically require a large amount of storage space and result in high computational complexity.

The FNN, CNN, and RNN mentioned above are all constructed based on neurons. As mentioned earlier, each neuron performs a weighted summation operation on its input values, and the result of the weighted summation is used to generate the output through a nonlinear function. The weights of the weighted summation operation of neurons in a neural network and the nonlinear function are called the parameters of the neural network. The parameters of all neurons in a neural network constitute the parameters of that neural network.

(2) Two-side model:

A two-sided model, also known as a bilateral model, collaborative model, or dual model, refers to a model composed of multiple sub-models. These sub-models need to be mutually compatible and can be deployed on different nodes.

Taking the encoder used to compress channel information and the decoder used to recover channel information in the CSI feedback process as an example, the encoder and decoder are used in pairs, which can be understood as complementary AI models. An encoder may include one or more AI models, and the decoder matched with the encoder also includes one or more AI models. The number of AI models included in the encoder and decoder used in the matching process is the same and corresponds one-to-one. The encoder may also include a quantization module, which can be used to quantize the output of the AI model in the encoder. The decoder may include an inverse quantization module, which can be used to inverse quantize the feedback information of the received channel information to obtain the input of the AI model in the decoder. Inverse quantization processing can also be called dequantization processing.

In one possible design, a set of matched encoders and decoders can be two parts of the same autoencoder (AE). An AE model where the encoder and decoder are deployed on different nodes is a typical bilateral model. In other AE models, the encoder and decoder are usually co-trained and used in combination. An autoencoder is an unsupervised learning neural network that uses input data as labeled data; therefore, it can also be understood as a self-supervised learning neural network. Autoencoders can be used for data compression and reconstruction. For example, the encoder in an autoencoder can compress (encode) data A to obtain data B; the decoder in the autoencoder can decompress (decode) data B to recover data A. Alternatively, the decoder can be understood as the inverse operation of the encoder.

The AI model in this application embodiment may include an encoder deployed on the terminal device side and a decoder deployed on the network device side, or an encoder deployed on the terminal device side and a decoder deployed on another terminal device side, or an encoder deployed on the network device side and a decoder deployed on another network device side.

With the development of AI technology, the application of AI models in communication systems is constantly expanding. Taking the use of AI models for compressed feedback of channel information as an example, the UE obtains downlink channel information based on the reference signal, uses the downlink channel information as input to the UE-side encoder model, and obtains the feedback quantity of the channel information. The UE feeds back the feedback quantity of the channel information to the base station, and the base station inputs the feedback quantity into the base station-side decoder model to recover the downlink channel information.

By training an AI model, it can be enabled to perform the expected communication functions. One method to enable the model to perform the expected communication functions is to determine the training dataset for the model and train it to match the expected functions with the actual functions performed.

The performance of AI models in real-world applications may be unstable. For example, changes in the environment in which an AI model is used can affect its performance. During the use of an AI model, it can be monitored, and the model adjusted based on the monitoring results to ensure network performance. However, different models exhibit variations in performance or generalization metrics, and the monitoring process may be mismatched with the model's performance, leading to ineffective monitoring results.

In view of this, this application provides a communication method and communication device that uses appropriate monitoring parameters for different models to perform model monitoring, thereby improving the reliability and efficiency of model monitoring. For example, for models with different performance and/or different generalization abilities, monitoring parameters that match their performance and/or generalization ability can be used for model monitoring, thereby improving the reliability and efficiency of model monitoring.

It should be understood that in this application, the indication includes direct indication (also known as explicit indication) and implicit indication. Direct indication information A refers to information A being included; implicit indication information A refers to information A being indicated through the correspondence between information A and information B, and through direct indication information B. The correspondence between information A and information B can be predefined, pre-stored, pre-burned, or pre-configured.

It should be understood that in this application, information C is used to determine information D, including both situations where information D is determined solely based on information C and situations where it is determined based on information C and other information. Furthermore, information C can also be used to determine information D indirectly, for example, where information D is determined based on information E, and information E is determined based on information C.

Furthermore, in the embodiments of this application, "network element A sends information A to network element B" can be understood as network element B being the destination of information A or an intermediate network element in the transmission path between the destination and network element B, which may include sending information directly or indirectly to network element B. "Network element B receives information A from network element A" can be understood as network element A being the source of information A or an intermediate network element in the transmission path between the source and network element A, which may include receiving information directly or indirectly from network element A. Information may undergo necessary processing between the source and destination, such as format changes, but the destination can understand the valid information from the source. Similar expressions in this application can be understood in a similar way and will not be elaborated further here.

Figure 5 is a schematic flowchart of a communication method provided in this application.

As shown in Figure 5, method 500 may include the following steps.

510, the first device acquires one or more of the following: performance information of the first model, generalization information of the first model, or relevant information of the first dataset.

520. The first device determines the first monitoring method based on one or more of the performance information of the first model, the generalization information of the first model, or the relevant information of the first dataset.

The first dataset is used to train the first model.

The first monitoring method can be used for monitoring the first model. The first monitoring method is the monitoring method for the first model.

The model monitoring of the first model can also be replaced with the performance monitoring of the first model.

The first model can be either an AI model or a non-AI model.

The first model can be a one-sided model, a two-sided model, or a sub-model within a two-sided model.

For example, the first model can be a two-sided model, including a first sub-model and a second sub-model. In this case, monitoring the first model can include monitoring the first sub-model and/or the second sub-model. Taking the first sub-model and the second sub-model as deployed on a first device and a second device respectively, monitoring the first model by the first device can include monitoring the first sub-model, or monitoring both the first and second devices together. Monitoring the first model by the second device can include monitoring the second sub-model, or monitoring both the first and second devices together.

For ease of description, the embodiments of this application mainly use AI models as examples for illustration, and do not constitute a limitation on the solutions of the embodiments of this application.

Optionally, the first device can also acquire the category of the first model. In this case, the first device can also determine the first monitoring method based on the category of the first model.

The category of a model can also be called the type of a model.

Optionally, method 500 may also include step 530.

530, the first device sends second information to the second device. The second information is used to indicate the first monitoring method.

Optionally, method 500 may also include step 540.

540, the second device sends a third message to the first device. The third message is used to instruct the first model to be monitored.

In this embodiment, device A sends information to device B, either directly or via a forwarding mechanism from another device. Similarly, device A receives information from device B, either directly or via a forwarding mechanism from another device.

In the embodiments of this application, the model's performance information, the model's generalization information, or the relevant information of the dataset can reflect the model's expected performance and/or the expected generalization ability of the first model. This is beneficial for determining a suitable monitoring method for the model, that is, determining a monitoring method that matches the model's performance and/or generalization, so as to achieve effective monitoring of models with different performance and/or generalization, thereby improving the reliability and efficiency of model monitoring.

In one possible implementation, the second device can be an AI entity. Part or all of the first model can be deployed on the second device. For example, the first model can be a one-sided model, which can be deployed on the second device. Alternatively, the first model can be a two-sided model, including a first sub-model and a second sub-model, with the first and second sub-models deployed on the second and first devices respectively. The first sub-model can be an encoder or decoder, and the second sub-model can be a decoder or encoder.

In this case, method 500 may include step 530. After determining the first monitoring method, the first device may notify the second device, which can then perform model monitoring on the first model according to the monitoring method indicated by the first device.

As an example, the second device can be an AI entity on the terminal device side. Devices on the terminal device side include the terminal device itself, or other devices that communicate with the terminal device, such as devices controlled by or serving the terminal device. The AI entity can be the terminal device itself, or an AI entity that communicates with the terminal device. For example, the second device can be a server, such as an OTT server or a cloud server.

For example, the first device can be any device on the terminal device side other than the second device. For instance, the first device can be a server, and the second device can be a terminal device.

Alternatively, the first device can be a network device-side device. Network device-side devices include network devices themselves, or other devices that communicate with the network device, such as devices controlled by or serving the network device. For example, the first device can be an AI entity on the network device side. This AI entity can be the network device itself, or an AI entity that communicates with the network device. For example, the first device can be a RIC, OAM, or a server, such as an OTT server or a cloud server. Near real-time RICs are set up in RAN nodes, such as in CU/DU.

As another example, the second device can be an AI entity on the network device side.

For example, the first device can be a device on the terminal device side. For instance, the first device can be an AI entity on the terminal device side.

In another possible implementation, the first device can be an AI entity, and the first model can be entirely deployed on the first device. For example, the first model can be a one-sided model, which can be deployed on the first device.

In this case, method 500 may not include step 530. After determining the monitoring method for the first model, the first device can perform model monitoring on the first model according to the monitoring method.

The first device can autonomously initiate model monitoring. Alternatively, method 500 may include step 540, in which the first device initiates model monitoring based on third information sent by the second device.

For example, the first device can be an AI entity on the terminal device side.

Alternatively, the first device can be an AI entity on the network device side.

The method for determining the first monitoring method is explained below.

The monitoring method of the model can be represented by the monitoring parameters under that monitoring method. Accordingly, the monitoring method in the embodiments of this application can also be replaced by monitoring parameters.

Monitoring parameters can include at least one of the following types: performance metric, performance threshold, monitoring frequency, monitoring cycle, monitoring duration, number of monitoring cycles, monitoring error tolerance, or switching threshold. For example, the performance threshold is a threshold for the performance metric.

Optionally, the second information can be used to indicate the first monitoring parameter in the first monitoring method, i.e., the monitoring parameter of the first model. The first monitoring parameter may include at least one of the following: a first performance threshold, a first monitoring frequency, a first monitoring period, a first monitoring duration, a first monitoring duration count, a first monitoring error tolerance, or a first switching threshold.

Performance thresholds are used for model performance monitoring, such as determining whether a model has failed or whether its performance meets the required standards. For example, when model performance fails to meet the standards, any of the following actions can be triggered: model switching, model fine-tuning, or accumulating the number of failures.

The performance of a model can be reflected by one or more performance metrics. Taking channel compression feedback as an example, the model's performance can be reflected by one performance metric, such as the squared generalized cosine similarity (SGCS). Alternatively, the model's performance can be reflected by two performance metrics, including SGCS and the normalized mean square error (NMSE). Correspondingly, the model's performance threshold can include the performance thresholds corresponding to these one or more performance metrics. That is, a model can have one or more performance thresholds. For descriptive purposes, this application embodiment only uses one performance metric as an example and does not constitute a limitation on the scheme of this application embodiment. Performance metrics can also be replaced by performance parameters.

For example, for some performance metrics, the larger the value of the performance metric, the better the model's performance. For this type of performance metric, if the model's performance metric value is less than or equal to the corresponding performance threshold, the model is considered to have failed. Alternatively, if the model's performance metric value is less than the corresponding performance threshold, the model is considered to have failed. Alternatively, for some performance metrics, the smaller the value of the performance metric, the better the model's performance. For this type of performance metric, if the model's performance metric value is greater than or equal to the corresponding performance threshold, the model is considered to have failed. Alternatively, if the model's performance metric value is greater than the corresponding performance threshold, the model is considered to have failed. For ease of description, this application embodiment only uses the example of a larger performance metric value indicating better model performance for illustration, and does not constitute a limitation on the solution of this application embodiment.

Using a model to perform inference and calculating a performance index value based on the inference results can be considered as monitoring the model. Taking channel compression feedback as an example, the network device sends a reference signal for monitoring, the terminal device determines the channel information based on the reference signal, inputs the channel information into the encoder, and calculates a performance index value based on the encoder's output and the channel information input to the encoder. This process can be considered as monitoring the encoder.

Monitoring frequency indicates how frequently the model is monitored. Monitoring period indicates the period during which the model is monitored. A monitoring period of T means that model monitoring is performed at time intervals of T. That is, after every time interval T, model performance monitoring begins, for example, to determine whether the model has failed. T is a positive number.

Monitoring duration refers to the number of times the model is repeatedly monitored, or in other words, the number of times the performance metric value is calculated.

Monitoring duration refers to the length of time the model is continuously monitored. Within this duration, the model may be monitored once, or it may be monitored repeatedly multiple times.

By monitoring the model N times, we can obtain N performance index values. N is a positive integer.

Monitoring error tolerance refers to the number of times a model's performance falls below a performance threshold out of N performance metrics that is allowed. In other words, it's the number of times the model's performance fails to meet standards across N monitoring sessions, or the percentage of such instances. For example, if the number of values below the performance threshold among the N performance metrics falls within the monitoring error tolerance range, the model does not need to be processed. For instance, if the monitoring error tolerance is 3 and N is 10, it means that 3 out of 10 performance metrics are allowed to fall below the performance threshold. If the number of values below the performance threshold among the 10 performance metrics is less than or equal to 3, then the model is not processed.

The switching threshold refers to the threshold that determines the switching of models.

This threshold can be a performance-related threshold.

For example, model switching is performed when the statistical values of multiple performance metrics do not meet the threshold. For example, the statistical values can be at least one of the average, maximum, or minimum values.

This threshold can also be a threshold related to the number of monitoring sessions.

For example, if the model fails to meet performance standards for n consecutive times, a model switch is performed. In this case, n is the switch threshold. n is a positive integer.

When multiple factors, such as the performance information of the first model, the generalization information of the first model, the relevant information of the first dataset, or the category of the first model, are used to determine multiple monitoring parameters in the first monitoring method, these multiple factors can be used to determine different monitoring parameters among the multiple monitoring parameters, or the multiple factors can be used to jointly determine the multiple monitoring parameters.

For example, the performance information of the first model can be used to determine the first performance threshold, and the generalization information of the first model can be used to determine the first monitoring frequency, the first monitoring duration, and the first switching threshold.

For example, relevant information from the first dataset can be used to determine the first performance threshold, and generalization information from the first model can be used to determine the first monitoring frequency, the first number of monitoring sessions, and the first switching threshold.

For example, the performance information of the first model can be used to determine the first performance threshold, and the relevant information of the first dataset can be used to determine the first monitoring frequency, the first number of monitoring sessions, and the first switching threshold.

For example, the performance information and category of the first model can be used to determine the first performance threshold, and the generalization information and category of the first model can be used to determine the first monitoring frequency, the first monitoring duration, and the first switching threshold.

For a detailed description, please refer to the examples below.

As one possible approach, the model monitoring method is related to the model's expected performance; that is, the model monitoring method can be determined based on the model's expected performance.

The expected performance of a model is the performance that the model is expected to achieve when it is working normally.

In this embodiment, the model's performance information can be used to reflect the model's expected performance. In this case, the model's performance information can also be understood as the model's expected performance information. Determining the model's monitoring method based on its performance information is equivalent to determining the model's monitoring method based on its expected performance.

The expected performance of a model is related to the dataset used to train it. In other words, the expected performance of a model is determined by the dataset used to train it. Taking the expected function of a model as channel state information (CSI) compression as an example, i.e., the model is a CSI compression model, and the dataset is used to train the CSI compression model. This dataset can include the channel input, the corresponding CSI compression information of the channel input, and the reconstructed channel information corresponding to the channel input. The compression performance of the CSI compression model is limited by the channel input in the dataset and the corresponding CSI compression information. For example, the larger the number of training samples in the dataset, the better the expected performance of the model trained on that dataset is likely to be.

Correspondingly, datasets can also be used to reflect the expected performance of models trained on those datasets. Determining the model monitoring method based on relevant information from the dataset is equivalent to determining the model monitoring method based on its expected performance.

The expected performance of a model can be used to determine the model's performance threshold.

For example, the expected performance of a model and its performance threshold can be positively correlated. That is, the better the expected performance of a model, the higher its performance threshold; conversely, the worse the expected performance of a model, the lower its performance threshold. For instance, if model A has a performance threshold of threshold A and model B has a performance threshold of threshold B, and model A's expected performance is higher than model B's, then threshold A is greater than threshold B. Since model A has a better expected performance, in actual use, model A needs to meet a higher performance threshold (threshold A) to be considered to be operating normally, while model B only needs to meet a lower performance threshold (threshold B) to be considered to be operating normally.

In the scheme of this application embodiment, the performance threshold of the model is related to the expected performance of the model. This helps to ensure that the performance threshold of the model matches the performance of the model, thereby improving the reliability of model monitoring.

As one possible approach, the model monitoring method is related to the model's expected generalization ability; that is, the model monitoring method can be determined based on the model's expected generalization ability.

The expected generalization ability of a model is the generalization that the model is expected to achieve when it is working normally.

In this embodiment, the model's generalization information can be used to reflect the model's expected generalization ability. In this case, the model's generalization information can also be understood as the model's expected generalization ability information. Determining the model's monitoring method based on its generalization information is equivalent to determining the model's monitoring method based on its expected generalization ability.

A model's expected generalization ability is related to the dataset used to train it. In other words, a model's expected generalization ability is determined by the dataset used to train it. Take, for example, a dataset used to train a CSI-compressed model. The generalization ability of a CSI-compressed model is limited by the channel inputs in the dataset and the corresponding CSI compression information. For instance, the more diverse the training samples in the dataset, the better the expected performance of the model trained on that dataset is likely to be.

Correspondingly, datasets can also be used to reflect the expected generalization ability of models trained on those datasets. Determining the model monitoring method based on relevant information from the dataset is equivalent to determining the model monitoring method based on its expected generalization ability.

The model’s expected generalization ability can be used to determine at least one of the following: monitoring frequency, monitoring cycle, monitoring duration, number of monitoring cycles, monitoring error tolerance, or switching threshold.

For example, the expected generalization ability of a model and the monitoring frequency of the model can be negatively correlated. That is, the higher the expected generalization ability of a model, the lower the monitoring frequency can be, and the lower the expected generalization ability of a model, the higher the monitoring frequency can be.

In the embodiments of this application, the monitoring frequency of the model is related to the expected generalization ability of the model. This helps to ensure that the monitoring frequency matches the generalization ability of the model, thereby improving the reliability and efficiency of model monitoring. For example, for models with high expected generalization ability, their adaptability may be stronger, and they may be able to perform well in different environments. A lower monitoring frequency can be used to monitor such models, which helps to improve the efficiency of model monitoring. Conversely, for models with low expected generalization ability, their adaptability may be weaker, and their performance may vary greatly when the environment changes. A higher monitoring frequency can be used to monitor such models, which helps to ensure the reliability of model monitoring.

For example, the expected generalization ability of a model and the number of monitoring sessions can be negatively correlated. That is, the higher the expected generalization ability of a model, the fewer monitoring sessions are needed to monitor the model, and the lower the expected generalization ability of a model, the more monitoring sessions are needed to monitor the model.

For example, there is a positive correlation between a model's expected generalization ability and its switching threshold. That is, the higher a model's expected generalization ability, the higher its switching threshold; conversely, the lower a model's expected generalization ability, the lower its switching threshold.

For specific examples of the above monitoring parameters, please refer to the descriptions in Scheme #1 and Scheme #2.

In the embodiments of this application, the model monitoring method is related to the model's expected performance and/or the model's expected generalization ability. This is beneficial for providing suitable monitoring methods for models with different performance and/or different generalization abilities, so that the model monitoring method matches the model's performance and/or generalization ability, thereby improving the reliability and efficiency of model monitoring.

The performance information of a model can be represented in various forms.

Optionally, the performance information of the first model may include the expected performance of the first model or the expected performance range of the first model.

Optionally, the performance information of the first model may include the performance level of the first model, which is related to the expected performance of the first model. In other words, the performance level of the first model is the level corresponding to the expected performance of the first model.

The expected performance of a model can be divided into multiple performance levels, which can be used to reflect the model's expected performance. Different performance levels correspond to different expected performances. For example, performance levels can be represented by numerical values. The lower the performance level, the better the expected performance. Alternatively, the higher the performance level, the better the expected performance. For ease of description, the embodiments in this application are only used as examples.

In this embodiment of the application, the performance level can also be replaced with the performance grade or performance level, etc.

The generalization information of a model can be represented in various forms.

Optionally, the generalization information of the first model may include the expected generalization ability of the first model or the range of its expected generalization ability. For example, the expected generalization ability of the model may be the number of scenarios the model is expected to be applicable to. Similarly, the range of the model's expected generalization ability may be the range of the number of scenarios the model is expected to be applicable to. Or, the range of the model's expected generalization ability may include the scenarios the model is expected to be applicable to.

Optionally, the generalization information of the first model may include the generalization level of the first model, which is related to the expected generalization ability of the first model. In other words, the generalization level of the first model is the level corresponding to the expected generalization ability of the first model.

The expected generalization ability of a model can be divided into multiple generalization levels, which can be used to reflect the model's expected generalization ability. Different generalization levels correspond to different expected generalization abilities. For example, generalization levels can be represented numerically. The lower the generalization level, the better the expected generalization ability. Alternatively, the higher the generalization level, the better the expected generalization ability. For ease of description, this application's embodiments are only used as examples.

In this embodiment, the generalization level can also be replaced with generalization grade or generalization level, etc.

The relevant information in the first dataset can be represented in various forms.

For example, relevant information about the first dataset can be used to indicate the first dataset.

Optionally, the relevant information for the first dataset may include the identifier (ID) of the first dataset. The identifier of the first dataset may also be replaced with the index of the first dataset.

Alternatively, the relevant information for a dataset can include other content, as long as it distinguishes different datasets. For example, the relevant information for the first dataset can include any one or more of the following: the number of training samples in the first dataset, the data format of the training samples in the first dataset, or the identifier of the provider of the first dataset, etc.

Optionally, the relevant information for the first dataset may include the performance information corresponding to the first dataset.

The performance information corresponding to a dataset can be understood as the performance information of the model trained on that dataset, and can be used to reflect the expected performance of the model trained on that dataset. For example, if the first dataset is used to train the first model, then the performance information corresponding to the first dataset can be used to reflect the expected performance of the first model.

Optionally, the performance information corresponding to the first dataset may include the expected performance or expected performance range corresponding to the first dataset.

The expected performance or expected performance range of a dataset can be understood as the expected performance or expected performance range of the model trained on that dataset.

Optionally, the performance information corresponding to the first dataset may include the performance level corresponding to the first dataset, which is related to the expected performance of the model trained on the first dataset. In other words, the performance level corresponding to the first dataset is the same as the expected performance level of the model trained on the first dataset.

The expected performance of a model can be divided into multiple performance levels. The performance level corresponding to a dataset can be used to reflect the expected performance of the model trained on that dataset. Different performance levels correspond to different expected performances. For example, performance levels can be represented numerically. The lower the performance level, the better the expected performance. Alternatively, the higher the performance level, the better the expected performance. For ease of description, this application's embodiment is only used as an example for illustration.

Optionally, the relevant information of the first dataset may include generalization information corresponding to the first dataset.

The generalization information corresponding to a dataset can be understood as the generalization information of the model trained on that dataset, and can be used to reflect the expected generalization ability of the model trained on that dataset. For example, if the first dataset is used to train the first model, then the generalization information corresponding to the first dataset can be used to reflect the expected generalization ability of the first model.

Optionally, the generalization information corresponding to the first dataset may include the expected generalization ability or the expected generalization ability range corresponding to the first dataset.

The expected generalization ability or expected generalization range of a dataset can be understood as the expected generalization ability or expected generalization range of the model trained on that dataset.

Optionally, the generalization information corresponding to the first dataset may include the generalization level corresponding to the first dataset, which is related to the expected generalization ability of the model trained on the first dataset. In other words, the generalization level corresponding to the first dataset is the level corresponding to the expected generalization ability of the model trained on the first dataset.

The expected generalization ability of a model can be divided into multiple generalization levels. The generalization level corresponding to a dataset can be used to reflect the expected generalization ability of the model trained on that dataset. Different generalization levels can correspond to different expected generalization abilities. For example, the generalization level can be represented numerically. The lower the generalization level, the better the expected generalization ability. Alternatively, the higher the generalization level, the better the expected generalization ability. For ease of description, this application embodiment is only used as an example for illustration.

Optionally, the model categories may include basic general models and cell-specific models.

That is, the first model can be either a basic general model or a community-specific model.

The basic general model is the foundational model. It can be applied to various scenarios or multiple cells. This model is not associated with any specific scenario; that is, it can be configured in any cell within any scenario.

Cell-specific models, also known as dedicated models, are applicable only to specific cells or scenarios. In other words, they are only matched/associated with specific cells or scenarios; that is, a user can only configure this model when accessing a specific cell or entering a specific scenario. In a matched cell or scenario, a cell-specific model may perform better than a base model applied to that cell or scenario. However, in a mismatched cell or scenario, a cell-specific model may perform worse than a base model applied to that cell or scenario; that is, performance degradation due to cell or scenario mismatch is more severe.

The following examples (methods #1 to #6) illustrate how to determine monitoring parameters. Monitoring parameters may include monitoring metrics.

Method #1:

As an example, the first network element can determine the first monitoring method based on the performance information of the first model.

For example, the first network element can determine the first monitoring parameter in the first monitoring method based on the performance information of the first model. The first monitoring parameter may include a first monitoring indicator, such as a first performance threshold.

That is, there is a correspondence between the performance information of the first model and the first monitoring parameter. In this embodiment, the correspondence can also be replaced by an association relationship. The first monitoring parameter is a monitoring parameter related to the performance information of the first model.

Alternatively, the performance information of the first model has a first correspondence with the first monitoring parameter, and the first network element can determine the first monitoring parameter based on the correspondence between the performance information and the monitoring parameter, which includes the first correspondence.

Table 1 shows an example of the correspondence between performance information and monitoring parameters.

Table 1

For example, the performance information of the first model is performance information #A. According to the correspondence shown in Table 1, the first monitoring parameter includes parameter value #A.

The following example illustrates this using a channel information feedback scenario. The first model is the channel compression feedback model.

Table 2 shows an example of the correspondence between performance levels, expected performance ranges, and performance thresholds.

Table 2

For example, as shown in Table 2, the expected performance of the channel compression feedback model can be divided into three performance levels. Different performance levels correspond to different expected performance ranges. Different expected performances correspond to different monitoring parameters.

In Table 2, the performance of the channel compression feedback model is characterized by the channel recovery accuracy, such as the squared generalized cosine similarity (SGCS). The monitoring parameter is the performance threshold.

In Table 2, there is a positive correlation between performance level and expected performance. That is, the higher the performance level of a model, the better its expected performance. Correspondingly, performance level and performance threshold can also be positively correlated.

The following explanation of Table 2 will primarily use performance level 1 as an example.

In Table 2, the expected performance range of the channel compression feedback model at performance level 1 during normal operation is SGCS of 0.6-0.7, with an associated performance threshold of 0.6- _Δ1 . For the channel compression feedback model at performance level 1, if the SGCS of the model is less than 0.6- _Δ1 during model monitoring, the model can be considered to have failed once. As shown in Table 2, the higher the performance level, the better the associated expected performance, and correspondingly, the higher the performance threshold.

Taking Table 2 as an example, the performance information of the first model may include performance level 1 or SGCS: [0.6-0.7]. According to the correspondence indicated in Table 2, its associated performance threshold (i.e., the first performance threshold) can be determined to be 0.6- _Δ1 .

This association method helps ensure the matching of model performance and monitoring methods, that is, it helps to select monitoring methods that match the performance of models with different performance levels, thereby improving the reliability of model monitoring.

_Δ1 , _Δ2 , and _Δ3 represent the deviation values of the performance thresholds associated with different expected performances. For different expected performances, these deviation values can be the same or different; that is, _Δ1 , _Δ2 , and _Δ3 can be the same or different.

The deviation value of the performance threshold can be determined by the first device, the second device, or it can be predefined.

It should be understood that Tables 1 and 2 are merely examples and do not constitute a limitation on the solutions of the embodiments of this application.

In the scheme of this application embodiment, the performance information of the first model can be used to reflect the expected performance of the first model. Determining the first monitoring method based on the performance information of the first model is equivalent to determining the first monitoring method based on the expected performance of the first model. This is beneficial to obtaining a monitoring method that matches the performance of the first model, thereby improving the reliability of model monitoring.

Method #2:

As an example, the first network element can determine the first monitoring method based on the generalization information of the first model.

For example, the first network element can determine the first monitoring parameter in the first monitoring method based on the generalization information of the first model. For example, the first monitoring parameter may include at least one of the following: first monitoring frequency, first monitoring period, first monitoring duration, first monitoring duration number of times, first monitoring error tolerance, or first switching threshold.

That is, there is a correspondence between the generalization information of the first model and the first monitoring parameter. The first monitoring parameter is a monitoring parameter related to the performance information of the first model.

Alternatively, the generalization information of the first model has a second correspondence with the first monitoring parameter. The first network element can determine the first monitoring parameter based on the correspondence between the generalization information and the monitoring parameter, and this correspondence includes the second correspondence.

Table 3 shows an example of the correspondence between generalization information and monitoring parameters.

Table 3

For example, the generalization information of the first model is generalization information #A. According to the correspondence shown in Table 3, the first monitoring parameter includes parameter value #D.

The following example illustrates this using a channel information feedback scenario. The first model is the channel compression feedback model.

Table 4 shows an example of the relationship between generalization level and monitoring frequency, monitoring duration, and switching threshold.

Table 4

For example, as shown in Table 4, the expected generalization capability of the channel compression feedback model can be divided into three generalization levels. Different generalization levels correspond to different expected generalization capabilities. Different expected generalization capabilities correspond to different monitoring parameters. In Table 4, the monitoring parameters include monitoring frequency, monitoring duration, and handover threshold.

In Table 4, there is a positive correlation between the generalization level and the expected generalization ability. That is, the higher the generalization level of a model, the better its expected generalization ability.

As shown in Table 4, the generalization level and the monitoring frequency can be negatively correlated.

Models with high generalization ability have higher performance robustness and can be monitored at a lower monitoring frequency. As shown in Table 4, during model monitoring, for channel compression feedback models at generalization level 3, performance monitoring can be carried out at a period of 60 minutes (min), while for channel compression feedback models at generalization level 1, performance monitoring needs to be carried out at a period of 1 minute.

As shown in Table 4, the generalization level and the number of monitoring sessions can be negatively correlated.

Models with higher generalization ability exhibit greater robustness and require fewer performance monitoring cycles. As shown in Table 4, during model monitoring, for a channel compression feedback model at generalization level 3, the model is considered to be in normal working condition if it achieves the performance target twice consecutively. However, for a channel compression feedback model at generalization level 1, the model is considered to be in normal working condition only if it achieves the performance target 10 times consecutively.

As shown in Table 4, the generalization level and the switching threshold can be positively correlated.

Models with higher generalization ability exhibit greater robustness and tolerance for performance fluctuations, allowing for the use of larger switching thresholds. As shown in Table 4, during model monitoring, for a channel compression feedback model at generalization level 3, model switching is only required if the model fails to meet performance standards four times consecutively. For a channel compression feedback model at generalization level 1, model switching is required if the model fails to meet performance standards twice consecutively.

Taking Table 4 as an example, the generalization information of the first model may include generalization level 1. According to the correspondence indicated in Table 4, it can be determined that the monitoring frequency associated with generalization level 1 is <1min, the associated monitoring duration is 10, and the associated switching threshold is 2.

This association method helps ensure the generalization of the model and the matching of the monitoring method. In other words, it helps to select a matching monitoring method for models with different generalization, thereby improving the efficiency of model monitoring while ensuring the reliability of model monitoring and avoiding resource waste.

It should be understood that Tables 3 and 4 are merely examples and do not constitute a limitation on the solutions of the embodiments of this application.

In the scheme of this application embodiment, the generalization information of the first model can be used to reflect the expected generalization ability of the first model. Determining the first monitoring method based on the generalization information of the first model is equivalent to determining the first monitoring method based on the expected generalization ability of the first model. This is beneficial to obtaining a monitoring method that matches the generalization of the first model, thereby improving the reliability and efficiency of model monitoring.

Method #3:

As an example, the first network element can determine the first monitoring method based on the identifier of the first dataset.

For example, the first network element can determine the first monitoring parameter in the first monitoring method based on the identifier of the first dataset. For example, the first monitoring parameter may include at least one of the following: a first performance threshold, a first monitoring frequency, a first monitoring cycle, a first monitoring duration, a first monitoring duration count, a first monitoring error tolerance, or a first switching threshold.

That is, there is a correspondence between the identifier of the first dataset and the first monitoring parameter. The first monitoring parameter is the monitoring parameter related to the identifier of the first dataset.

Alternatively, the identifier of the first dataset has a third correspondence with the first monitoring parameter. The first network element can determine the first monitoring parameter based on the correspondence between the identifier of the dataset and the monitoring parameter. This correspondence includes the third correspondence.

Table 5 shows an example of the correspondence between dataset identifiers and monitoring parameters.

Table 5

For example, the identifier of the first dataset is #A, and according to the correspondence shown in Table 5, the first monitoring parameter can be determined to include the parameter value #G.

The monitoring parameter values associated with the identifiers of different datasets may be the same or different. That is, parameter values #G, #H, and #I may be the same or different.

The following example illustrates this using a channel information feedback scenario. The first dataset is used to train the channel compression feedback model.

Table 6 shows an example of the correspondence between dataset identifiers, performance thresholds, and monitoring frequencies.

Table 6

As shown in Table 6, the dataset identifier is associated with the monitoring parameters. The relevant monitoring parameters can be determined based on the dataset identifier.

Taking dataset 1 as an example, the performance threshold associated with dataset 1 is 0.6- _Δ1 , and the monitoring frequency associated with dataset 1 is <60 minutes. For the model trained from dataset 1, if the SGCS of the model is less than 0.6- _Δ1 during the model monitoring process, it can be determined that the model has failed once, and performance monitoring can be carried out on it at a cycle of 60 minutes.

Taking Table 6 as an example, the identifier for the first dataset can be dataset 1. Based on the correspondence indicated in Table 6, the performance threshold associated with dataset 1 can be determined to be 0.6- _Δ1 , and the monitoring frequency associated with dataset 1 is <60min.

It should be understood that Tables 5 and 6 are merely examples and do not constitute a limitation on the solutions of the embodiments of this application.

A dataset can reflect the expected performance and/or expected generalization ability of a model trained on that dataset. When establishing the correlation between a dataset and monitoring parameters, the expected performance and/or expected generalization ability of the dataset can be considered. This helps to match the model's performance and/or generalization with the monitoring method, thereby facilitating effective monitoring of models with different performance and/or generalization, and improving the reliability and efficiency of monitoring.

Method #4:

As an example, the first network element can determine the first monitoring method based on the performance information corresponding to the first dataset.

For example, the first network element can determine the first monitoring parameter in the first monitoring method based on the performance information corresponding to the first dataset. For instance, the first monitoring parameter may include a first performance threshold.

This means there is a correspondence between the performance information corresponding to the first dataset and the first monitoring parameter. The first monitoring parameter is the monitoring parameter related to the performance information corresponding to the first dataset.

Alternatively, the performance information corresponding to the first dataset has a fourth correspondence with the first monitoring parameter. The first network element can determine the first monitoring parameter based on the correspondence between the performance information and the monitoring parameter, and this correspondence includes the fourth correspondence.

Examples of the correspondence between performance information and monitoring parameters for the dataset can be found in Tables 1 and 2 of Method #1. Simply replace the model's performance information with the performance information corresponding to the dataset.

For example, the performance information corresponding to the first dataset is performance information #A. According to the correspondence shown in Table 1, the first monitoring parameter includes parameter value #A.

Taking Table 2 as an example, the dataset corresponds to performance level 1, meaning the expected performance range of the model trained on this dataset during normal operation is SGCS of 0.6-0.7, and the associated performance threshold is 0.6- _Δ1 . For the model trained on this dataset, if the SGCS of the model is less than 0.6- _Δ1 during model monitoring, it can be determined that the model has failed once. As shown in Table 2, the higher the performance level, the better the associated expected performance, and correspondingly, the higher the performance threshold.

Taking Table 2 as an example, the performance information corresponding to the first dataset may include performance level 1 or SGCS: [0.6-0.7]. Based on the correspondence indicated in Table 2, the associated performance threshold (i.e., the first performance threshold) can be determined to be 0.6- _Δ1 .

In the scheme of this application embodiment, the performance information corresponding to the first dataset can be used to reflect the expected performance of the model trained by the first dataset. Determining the first monitoring method based on the performance information corresponding to the first dataset is equivalent to determining the first monitoring method based on the expected performance of the model trained by the first dataset. This is beneficial to obtaining a monitoring method that matches the performance of the first model, thereby improving the reliability of model monitoring.

Method #5:

As an example, the first network element can determine the first monitoring method based on the generalization information corresponding to the first dataset.

For example, the first network element can determine the first monitoring parameter in the first monitoring method based on the generalization information corresponding to the first dataset. For example, the first monitoring parameter may include at least one of the following: first monitoring frequency, first monitoring period, first monitoring duration, first monitoring duration number of times, first monitoring error tolerance, or first switching threshold.

That is, there is a correspondence between the generalization information corresponding to the first dataset and the first monitoring parameter. The first monitoring parameter is a monitoring parameter related to the generalization information corresponding to the first dataset.

Alternatively, the generalization information corresponding to the first dataset has a fifth correspondence with the first monitoring parameter. The first network element can determine the first monitoring parameter based on the correspondence between the generalization information and the monitoring parameter, and this correspondence includes the fifth correspondence.

Examples of the correspondence between the generalization information of the dataset and the monitoring parameters can be found in Tables 3 and 4 of Method #2. Simply replace the generalization information of the model with the generalization information corresponding to the dataset.

For example, the generalization information corresponding to the first dataset is generalization information #A. According to the correspondence shown in Table 3, the first monitoring parameter includes parameter value #D.

Taking Table 4 as an example, the generalization level corresponding to a dataset is generalization level 1. Generalization level 1 is associated with a monitoring frequency of <1 minute, a monitoring duration of 10 times, and a switching threshold of 2. For the model trained on this dataset, performance monitoring is performed at 1-minute intervals. If the model meets the performance standard for 10 consecutive times, it is considered to be in normal working condition. If the model fails to meet the performance standard for 2 consecutive times, model switching is required.

Taking Table 4 as an example, the generalization information corresponding to the first dataset may include generalization level 1. According to the correspondence indicated in Table 4, it can be determined that the monitoring frequency associated with generalization level 1 is <1min, the associated monitoring duration is 10, and the associated switching threshold is 2.

In the scheme of this application embodiment, the generalization information corresponding to the first dataset can be used to reflect the expected generalization ability of the model trained by the first dataset. Determining the first monitoring method based on the generalization information corresponding to the first dataset is equivalent to determining the first monitoring method based on the expected generalization ability of the model trained by the first dataset. This is beneficial to obtaining a monitoring method that matches the generalization of the first model, thereby improving the reliability and efficiency of model monitoring.

Method #6:

As an example, the first network element can also determine the first monitoring method based on the category of the first model.

That is, any of the methods #1 to #5 mentioned above can be used in combination with the category of the first model.

Taking the combination of mode #1 and the category of the first model as an example, the first network element can determine the first monitoring parameter in the first monitoring mode based on the performance information of the first model and the category of the first model. For example, the first monitoring parameter may include a first performance threshold.

That is, there is a correspondence between the performance information of the first model, the category of the first model, and the first monitoring parameter. The first monitoring parameter is a monitoring parameter related to the performance information and category of the first model.

Alternatively, the performance information of the first model, the category of the first model, and the first monitoring parameter have a sixth correspondence. The first network element can determine the first monitoring parameter based on the correspondence between the performance information, the category, and the monitoring parameter. This correspondence includes the sixth correspondence.

Table 7 shows an example of the correspondence between performance information, categories, and monitoring parameters.

Table 7

For example, the performance information of the first model is performance information #A, the category of the first model is category #A, and according to the correspondence shown in Table 1, the first monitoring parameter can be determined to include parameter value #A1.

The following example illustrates this using a channel information feedback scenario. The first model is the channel compression feedback model.

Table 8 shows an example of the correspondence between category, performance level, expected performance range, and performance threshold.

Table 8

For example, as shown in Table 8, the expected performance of a model is related to the model category and the model performance level. Accordingly, the performance threshold can be determined by the performance level and the category.

Taking Table 8 as an example, the performance information of the first model may include performance level 1, and the category of the first model may be a basic general model. According to the correspondence indicated in Table 8, its associated performance threshold (i.e., the first performance threshold) can be determined to be 0.6- _Δ1 .

This association method helps to further ensure the matching of model performance and monitoring methods. It takes into account the relationship between model category and expected model performance, so as to make the classification of expected performance more accurate and refined. This is conducive to selecting monitoring methods that match the performance of models with different performance levels, thereby improving the reliability of model monitoring.

_δ1 and _δ2 represent the deviation values of the performance thresholds associated with different performance levels corresponding to the cell-specific model. For different performance levels, these deviation values can be the same or different, that is, _δ1 and _δ2 can be the same or different.

The deviation value of the performance threshold can be determined by the first device, the second device, or it can be predefined.

It should be understood that the number of performance tiers can be the same or different for different categories. For example, in Table 8, the basic general model and the cell-specific model have the same number of performance tiers, both corresponding to two performance tiers. Furthermore, in some scenarios, the expected performance range of the basic general model may be large, so it can be divided into three performance tiers, meaning the basic general model corresponds to three performance tiers. Conversely, the expected performance range of the cell-specific model may be smaller, so it can be divided into two performance tiers, meaning the cell-specific model corresponds to two performance tiers.

Taking the combination of method #2 and the category of the first model as an example, the first network element can determine the first monitoring parameter in the first monitoring method based on the generalization information of the first model and the category of the first model. For example, the first monitoring parameter may include a first performance threshold.

That is, there is a correspondence between the generalization information of the first model, the category of the first model, and the first monitoring parameter. The first monitoring parameter is a monitoring parameter related to the generalization information and category of the first model.

Alternatively, the generalization information of the first model, the category of the first model, and the first monitoring parameter have a seventh correspondence. The first network element can determine the first monitoring parameter based on the correspondence between the generalization information, the category, and the monitoring parameter. This correspondence includes an eighth correspondence.

Table 9 shows an example of the correspondence between generalization information, categories, and monitoring parameters.

Table 9

For example, the performance information of the first model is generalization information #A, the category of the first model is category #A, and according to the correspondence shown in Table 1, the first monitoring parameter can be determined to include parameter value #A2.

The following example illustrates this using a channel information feedback scenario. The first model is the channel compression feedback model.

Table 10 shows an example of the relationship between category, generalization level, monitoring frequency, monitoring duration, and switching threshold.

Table 10

For example, as shown in Table 10, the expected generalization ability of a model is related to the model's category and its generalization level. Accordingly, the monitoring parameters can be determined jointly by the generalization level and the category.

Taking Table 10 as an example, the generalization information of the first model may include generalization level 1, the category of the first model is a basic general model, the associated monitoring frequency is <1min, the associated monitoring duration is 5, and the associated switching threshold is 2.

This association method helps to further ensure the generalization of the model and the matching of the monitoring method. That is, it takes into account the relationship between the model category and the expected generalization ability of the model, so as to make the classification of expected generalization ability more accurate and refined. This is conducive to selecting a monitoring method that matches the generalization ability for models with different generalization abilities, thereby improving the reliability of model monitoring.

It should be understood that the number of generalization levels can be the same or different for different categories. For example, in Table 10, the number of generalization levels for category 1 is different from the number of generalization levels for category 2; the basic general model has 3 generalization levels, while the cell-specific model has 1 generalization level. In other implementations, the number of generalization levels for the basic general model and the cell-specific model can also be the same.

The combination of methods #3 to #5 and method #6 can be found in the previous text and will not be repeated here.

Furthermore, some or all of the methods #1 to #6 mentioned above can be used in combination. That is, the first monitoring method can be determined based on multiple factors, including the performance information of the first model, the generalization information of the first model, the identifier of the first dataset, the performance information corresponding to the first dataset, the generalization information corresponding to the first dataset, or the category of the first model. These multiple factors can be used together to determine the same type of monitoring parameter, or they can be used separately to determine different types of monitoring parameters.

Taking the combination of methods #1 and #2 as an example, the first network element can determine the first monitoring method based on the performance information and generalization information of the first model. For instance, the performance information and generalization information of the first model can be used to determine different types of monitoring parameters in the first monitoring method.

Taking a combination of methods #1, #2, and #6 as an example, the first network element can determine the first monitoring method based on the performance information and generalization information of the first model. For example, the performance information and category of the first model can be used to determine a first performance threshold. The category and generalization information of the first model can be used to determine at least one of the following: first monitoring frequency, first monitoring cycle, first monitoring duration, first monitoring duration count, first monitoring error tolerance, or first handover threshold.

Taking the combination of methods #4 and #5 as an example, the first network element can determine the first monitoring method based on the performance information and generalization information corresponding to the first dataset. For instance, the performance information and generalization information corresponding to the first dataset can be used to determine different types of monitoring parameters in the first monitoring method.

Step 510 will be explained below.

The first device can obtain one or more of the following in a variety of ways: performance information of the first model, generalization information of the first model, or relevant information of the first dataset.

As one possible implementation, step 510 may include: the first device receiving first information. The first information indicates one or more of the following: performance information of the first model, generalization information of the first model, or relevant information about the first dataset.

Figure 6 shows a schematic flowchart of a communication method according to an embodiment of this application. The method shown in Figure 6 can be regarded as a specific implementation of the method 500 shown in Figure 5.

As shown in Figure 6, step 510 may include: the first device receiving first information from the second device.

Before the second device sends the first information to the first device, method 500 may further include: the second device acquiring the first information or the second device determining the first information. In other words, the second device acquires or determines the content indicated by the first information.

The first information can indicate the performance information of the first model in various forms.

Optionally, the first information may include an identifier of the first model. The identifier of the first model may also be replaced with an index of the first model.

That is, the performance information of the first model is indirectly indicated by the identifier of the first model.

The first device can determine the performance information of the first model based on the identifier of the first model. There is a correspondence between the identifier of the first model and the performance information of the first model. The performance information of the first model is the performance information associated with the identifier of the first model.

Table 11 shows an example of the correspondence between model identifiers and performance levels.

Table 11

For example, the first information includes a model ID of model 1. The first device determines the performance level associated with model 1 (i.e., the performance level of the first model) as performance level 1 according to the correspondence shown in Table 11.

The correspondence between the model identifier and performance information, such as Table 11, can be determined by the first device itself, or it can be received by the first device from other devices, such as from the node that provides the first model, or from the second device, or it can be predefined.

Optionally, the first information may include performance information of the first model, that is, the first information may directly indicate the performance information of the first model. For example, the first information may include the performance level of the first model.

The second device can determine the performance information of the first model in various ways. For example, the second device can determine the performance information of the first model based on the correspondence between the model's identifier and performance information, as shown in Table 11. This correspondence can be determined by the second device itself, or it can be received by the second device from other devices, such as the first device, or the node providing the first model, or it can be predefined. Alternatively, the second device can receive the performance information of the first model from other devices, such as the node providing the first model.

The first information can indicate the generalization information of the first model in various forms.

Optionally, the first information may include the identifier of the first model.

That is, the generalization information of the first model is indirectly indicated by the identifier of the first model.

The first device can determine the generalization information of the first model based on the identifier of the first model. There is a correspondence between the identifier of the first model and the generalization information of the first model. The generalization information of the first model is the generalization information associated with the identifier of the first model.

Table 11 also shows an example of the correspondence between model identifiers and generalization levels.

For example, the first information includes a model ID of model 1. The first device determines the generalization level associated with model 1 (i.e., the generalization level of the first model) as generalization level 3 according to the correspondence shown in Table 11.

The correspondence between the model identifier and the generalization information, such as Table 11, can be determined by the first device itself, or it can be received by the first device from other devices, such as from the node that provides the first model, or from the second device, or it can be predefined.

Optionally, the first information may include generalization information of the first model, that is, the first information may directly indicate the generalization information of the first model. For example, the first information may include the generalization level of the first model.

The second device can determine the generalization information of the first model in several ways. For a detailed description, please refer to the method for determining the performance information of the first model mentioned earlier, and simply replace the performance information of the first model with its generalization information.

The first information can indicate the performance information corresponding to the first dataset in various forms.

Optionally, the first information may include the identifier of the first dataset.

That is, the performance information corresponding to the first dataset is indirectly indicated by the identifier of the first dataset.

The first device can determine the performance information corresponding to the first dataset based on the identifier of the first dataset. There is a correspondence between the identifier of the first dataset and the performance information corresponding to the first dataset. The performance information corresponding to the first dataset is the performance information associated with the identifier of the first dataset.

Table 12 shows an example of the correspondence between dataset identifiers and performance tiers.

Table 12

For example, the dataset ID included in the first information is dataset 1. The first device determines the performance level associated with dataset 1 (i.e., the performance level corresponding to the first dataset) as performance level 1 according to the correspondence shown in Table 12.

The correspondence between the dataset identifier and performance information, such as Table 12, can be determined by the first device itself, or it can be received by the first device from other devices, such as from the node that provides the first dataset, or from the second device, or it can be predefined.

Optionally, the first information may include performance information corresponding to the first dataset; that is, the first information may directly indicate the performance information corresponding to the first dataset. For example, the first information may include the performance level corresponding to the first dataset.

Optionally, the first information may include the identifier of the first model.

That is, the performance information corresponding to the first dataset is indicated by the identifier of the first model.

The first device can determine the performance information corresponding to the first dataset based on the identifier of the first model. There is a correspondence between the identifier of the first model and the performance information corresponding to the first dataset. The performance information corresponding to the first dataset is the performance information corresponding to the dataset associated with the identifier of the first model.

The second device can determine the performance information corresponding to the first dataset in several ways. For a detailed description, please refer to the previous description of the performance information of the first model; simply replace the performance information of the first model with the performance information corresponding to the first dataset.

The first piece of information can indicate the generalization information corresponding to the first dataset in various forms.

Optionally, the first information may include the identifier of the first dataset.

That is, the generalization information corresponding to the first dataset is indirectly indicated by the identifier of the first dataset.

The first device can determine the generalization information corresponding to the first dataset based on the identifier of the first dataset. There is a correspondence between the identifier of the first dataset and the generalization information corresponding to the first dataset. The generalization information corresponding to the first dataset is the generalization information associated with the identifier of the first dataset.

Table 12 also shows an example of the correspondence between dataset identifiers and generalization levels.

For example, the dataset ID included in the first information is dataset 1. The first device determines the generalization level associated with dataset 1 (i.e., the generalization level corresponding to the first dataset) as generalization level 3 according to the correspondence shown in Table 12.

The correspondence between the dataset identifier and the generalization information, such as Table 12, can be determined by the first device itself, or it can be received by the first device from other devices, such as from the node that provides the first dataset, or from the second device, or it can be predefined.

Optionally, the first information may include generalization information corresponding to the first dataset. For example, the first information may include the generalization level corresponding to the first dataset.

Optionally, the first information may include the identifier of the first model.

That is, the generalization information corresponding to the first dataset is indicated by the identifier of the first model.

The first device can determine the generalization information corresponding to the first dataset based on the identifier of the first model. There is a correspondence between the identifier of the first model and the generalization information corresponding to the first dataset. The performance information corresponding to the first dataset is the generalization information corresponding to the dataset associated with the identifier of the first model.

The second device can determine the generalization information corresponding to the first dataset in several ways. For details, please refer to the previous text; simply replace the performance information corresponding to the first dataset with the generalization information corresponding to the first dataset.

Furthermore, the first information can also be used to indicate the category of the first model.

The first piece of information can indicate the category of the first model in various forms.

Optionally, the first information may include the identifier of the first model.

That is, the category of the first model is indirectly indicated by the identifier of the first model.

The first device can determine the category of the first model based on its identifier. There is a correspondence between the identifier of the first model and its category. The category of the first model is the category associated with its identifier.

Table 13 shows an example of the correspondence between model identifiers and model categories.

Table 13

For example, the ID of the model included in the first information is model 1. The first device determines the category of the model associated with model 1 (i.e., the category of the first model) based on the correspondence shown in Table 3 as the basic general model.

The correspondence between model identifiers and model categories, for example, Table 13, can be determined by the first device itself, or it can be received by the first device from other devices, such as from the node that provides the first model, or from the second device, or it can be predefined.

Optionally, the first information may include the identifier of the first dataset.

That is, the category of the first model is indirectly indicated by the identifier of the first dataset.

The first device can determine the category of the model trained on the first dataset based on the identifier of the first dataset; this category can then be considered the category of the first model. There is a correspondence between the identifier of the first dataset and the category of the model trained on the first dataset. The category of the model trained on the first dataset is the category associated with the identifier of the first dataset.

The correspondence between the dataset identifier and the model category can be determined by the first device itself, or it can be received by the first device from other devices, such as from the node that provides the first dataset.

Optionally, the first information may include an identifier of the category of the first model. That is, the first information may directly indicate the category of the first model.

The second device can determine the category of the first model in several ways. For example, the second device can determine the category of the first model based on the correspondence between the model's identifier and the model's category. This correspondence can be determined by the second device itself, or it can be received by the second device from other devices, such as from the first device, or from the node providing the first model. Alternatively, the second device can receive the category of the first model from other devices, such as from the node providing the first model. Alternatively, the second device can determine the category of the model trained on the first dataset based on the identifier of the first dataset; this category can then be considered the category of the first model. The correspondence between the dataset identifier and the model's category can be determined by the second device itself, or it can be received by the second device from other devices, such as from the node providing the first dataset. Alternatively, the second device can receive the categories associated with the first dataset from other devices.

Furthermore, as mentioned above, the first model can be deployed partially or entirely in the second device, and the first information can also come from a third device other than the second device. That is, in step 510, the first device can receive the first information from the third device. The third device can be a node providing the first model and/or the first dataset.

Alternatively, the category of the first model can also be indicated by other information besides the first information.

As another possible implementation, step 510 may include: the first device determining one or more of the following: performance information of the first model, generalization information of the first model, or relevant information of the first dataset.

For example, the first device can determine the performance information corresponding to the identifier of the first model, i.e., the performance information of the first model, based on the correspondence between the model's identifier and performance information.

The specific determination method can be found in the descriptions of the above items. To avoid repetition, it will not be repeated here.

It should be understood that the methods for obtaining the performance information of the first model, the generalization information of the first model, or the relevant information of the first dataset can be the same or different. For example, all of the above can be indicated by the first information. Alternatively, some of the above can be indicated by the first information, while the rest can be obtained through other means. For instance, the rest can be indicated by other information. Furthermore, the other content can be determined by the first device itself. Alternatively, all of the above can be determined by the first device itself.

The following uses two scenarios (Scenario #1 and Scenario #2) as examples to illustrate method 500.

Scene #1:

In one possible scenario, the model can be obtained by acquiring a dataset. That is, a model capable of performing the relevant functions can be obtained by acquiring a dataset. Acquiring the model can also be replaced by deploying the model.

For example, taking an AI model as an example, the model can be trained based on a dataset to enable it to achieve the expected function or complete function matching. The trained model can then be deployed on the corresponding nodes to implement the corresponding function. Specifically, after acquiring the dataset, the device can train the model to perform the expected function based on that dataset.

Taking the model on the terminal device side as an example, the terminal device, such as the terminal device or the cloud server of the terminal device, can obtain the dataset and train the model based on the dataset in order to enable the model to achieve the expected function.

In scenario #1, different datasets can be used to distinguish different models. The dataset identifier can serve as a model identifier. The dataset ID can also be replaced with the associated ID.

The following uses the two-sided model (Example #1) and the one-sided model (Example #2) as examples to illustrate method 500 in scenario #1.

Example #1:

In Example #1, the bilateral model may include a first sub-model and a second sub-model, deployed on a first device and a second device, respectively.

Figure 7 shows a schematic flowchart of a communication method according to an embodiment of this application. The method 700 shown in Figure 7 is a specific implementation of method 500. For a detailed description, please refer to method 500. To avoid repetition, some descriptions are omitted when describing method 700.

Method 700 is primarily illustrated using a CSI feedback scenario as an example, where the second sub-model is an encoder for compressing channel information, and the first sub-model is a decoder for recovering channel information. Exemplarily, the first device can be an AI entity on the network device side, and the second device can be an AI entity on the terminal device side. Alternatively, the first device can be an AI entity on the terminal device side, and the second device can be an AI entity on the network device side. Method 700 is illustrated using only the example of the first device being a network device and the second device being a terminal device, and does not constitute a limitation on the solutions of this application embodiment.

As shown in Figure 7, method 700 includes the following steps.

710, The terminal device obtains the first dataset.

The terminal device can train a first encoder based on the first dataset, and the first encoder can be regarded as the first model.

Alternatively, the terminal device can train a first encoder and a first decoder based on a first dataset, meaning the encoder and decoder of the bilateral model are trained together. This first encoder and first decoder can be considered as the first model.

For example, a terminal device can acquire multiple datasets and train multiple models based on each dataset. For instance, these datasets could be acquired from a network device. In this way, the network device can distinguish the multiple models by the IDs of the datasets.

720, The terminal device sends the first information to the network device, the first information indicating the relevant information of the first dataset.

The training dataset for the first encoder is the first dataset. If the terminal device needs to perform the CSI feedback process (i.e., compress channel information) through the first encoder, it can instruct the network device on the relevant information of the first dataset.

For example, the first piece of information may include the ID of the first dataset.

For example, the first information may include performance information corresponding to the first dataset and/or generalization information corresponding to the first dataset.

730. The network device determines the first monitoring method based on the relevant information in the first dataset.

Taking the ID of the first dataset as an example, after receiving the ID of the first dataset, the network device can determine the first monitoring method according to method #3. For example, the network device can determine the monitoring parameters associated with the ID of the first dataset based on the correspondence between the dataset ID and the monitoring parameters (such as Table 5 or Table 6).

Alternatively, after receiving the ID of the first dataset, the network device can determine the first monitoring method according to method #4 and/or method #5. For example, the network device can determine the performance information and/or generalization information associated with the ID of the first dataset based on the correspondence between the dataset ID and performance information and/or generalization information (as shown in Table 12), and then determine the relevant monitoring parameters based on the correspondence between the performance information and/or generalization information and monitoring parameters (as shown in any one or more of Tables 1 to 4).

Taking the first information including performance information corresponding to the first dataset and/or generalization information corresponding to the first dataset as an example, the network device can determine the first monitoring method according to method #4 and/or method #5. Taking Tables 2 and 4 as examples, for instance, the first information may include performance level 2 and generalization level 2. The network device determines the following values for the monitoring parameters associated with performance level 2 according to Table 2: the performance threshold is 0.7 - _Δ2. According to Table 4, the network device determines the following values for the monitoring parameters associated with generalization level 2: monitoring frequency less than 10 minutes, monitoring duration count of 4, and switching threshold of 3. The first monitoring parameter may include any one or more of the above parameter values.

740. The network device sends a second message to the terminal device, which indicates the first monitoring method.

750, The terminal device performs model monitoring according to the first monitoring method.

The terminal device can perform model monitoring on the first encoder according to the first monitoring method.

For example, the second information could indicate a first performance threshold and a first monitoring frequency. The terminal device can perform monitoring based on the first monitoring threshold. If the performance of the first encoder on the terminal device side is lower than the first performance threshold, a model switch can be triggered. The terminal device can then perform model monitoring at the first monitoring frequency.

Alternatively, the terminal device and the network device can jointly perform model monitoring on the first encoder and the first decoder according to the first monitoring method.

In addition, in steps 720 to 750, the terminal device can also be replaced by a network device, and the network device can also be replaced by a terminal device.

It should be understood that the above is only an example of the CSI feedback scenario. Method 700 can also be applied to the monitoring of bilateral models in other scenarios.

Example #2:

In Example #2, the first model is a one-sided model and is deployed on the first device.

Figure 8 shows a schematic flowchart of a communication method according to an embodiment of this application. The method 800 shown in Figure 8 is a specific implementation of method 500. For a detailed description, please refer to method 500. To avoid repetition, some descriptions are omitted when describing method 800.

In method 800, only the first device is used as a terminal device and the second device is used as a network device for illustration, and it does not constitute a limitation on the solution of the embodiments of this application.

As shown in Figure 8, method 800 includes the following steps.

810, The terminal device obtains the first dataset.

The terminal device can train the first model based on the first dataset.

For example, the first dataset may come from a network device. Alternatively, the first dataset may also come from other devices, such as a cloud server.

820, the network device sends the first information to the terminal device, the first information indicating the relevant information of the first dataset.

830, The terminal device determines the first monitoring method based on the relevant information in the first dataset.

840, The terminal device performs model monitoring according to the first monitoring method.

Optionally, before step 840, method 800 may also include step 850.

850. The network device sends third information to the terminal device, which can be used to instruct the terminal device to perform model monitoring.

In this case, the model monitoring operation of the terminal device can be triggered by the indication information sent by the network device.

Method 800 may also exclude step 850. After determining the first monitoring method, the terminal device can independently carry out monitoring operations without the need for instruction information from the network device to trigger it.

For a detailed description, please refer to Example #1, which will not be repeated here.

It should be understood that the above explanation only uses the example of a one-sided model deployed on the first device. In other scenarios, the one-sided model can also be deployed on the second device. For example, the first device can be a network device, and the second device can be a terminal device. Alternatively, the first device can be a terminal device, and the second device can be a network device. Other possible implementations can be found in the description of method 500 above, and will not be elaborated upon here.

Scene #2:

In one possible scenario, the model can be obtained by receiving the model. Obtaining the model can also be replaced by deploying the model.

Taking the model on the terminal device side as an example, the terminal device, such as the terminal device itself, can receive the model sent by the network device or third-party device, such as receiving the model parameters.

In scenario #2, different models can be distinguished by their model identifier (ID).

The following uses the two-sided model (Example #3) and the one-sided model (Example #4) as examples to illustrate method 500 in scenario #2.

Example #3:

In Example #3, the bilateral model may include a first sub-model and a second sub-model, deployed on a first device and a second device, respectively.

Figure 9 shows a schematic flowchart of a communication method according to an embodiment of this application. The method 900 shown in Figure 9 is a specific implementation of method 500. For a detailed description, please refer to method 500. To avoid repetition, some descriptions are omitted when describing method 900.

Method 900 is primarily illustrated using a CSI feedback scenario, where the second sub-model is an encoder for compressing channel information, and the first sub-model is a decoder for recovering channel information. The main difference between Method 900 and Method 700 lies in the method by which the terminal device acquires the first model and the content indicated by the first information. The description of Method 900 focuses on these differences; other details can be found in Method 700.

As shown in Figure 9, method 900 includes the following steps.

910, the terminal device obtains the first model.

For example, the terminal device may receive model parameters from a first model from another device.

For example, the terminal device can receive model parameters from a first encoder, which can be regarded as a first model.

For example, the terminal device can receive model parameters of the first encoder and the first decoder, which can be regarded as the first model.

For example, the first model can be obtained from a network device. Within the network device, multiple models can be distinguished by their respective IDs.

920, the terminal device sends first information to the network device, the first information indicating the performance information of the first model and/or the generalization information of the first model.

For example, the first information may include the ID of the first model.

For example, the first information may include the performance information of the first model and/or the generalization information of the first model.

930, the network device determines the first monitoring method based on the first information.

Taking the ID of the first model as an example, after receiving the ID of the first model, the network device can determine the first monitoring method according to method #1 and/or method #2. For example, the network device can determine the performance information and/or generalization information associated with the ID of the first model according to the correspondence between the model ID and performance information and/or generalization information (as shown in Table 11), and then determine the relevant monitoring parameters according to the correspondence between the performance information and/or generalization information and monitoring parameters (as shown in any one or more of Tables 1 to 4).

Taking the first information including the performance information and/or generalization information of the first model as an example, the network device can determine the first monitoring method according to method #1 and/or method #2. Taking Tables 2 and 4 as examples, for instance, the first information may include performance level 2 and generalization level 2. The network device determines the following values for the monitoring parameters associated with performance level 2 according to Table 2: the performance threshold is 0.7 - Δ2 _. According to Table 4, the following values are determined for the monitoring parameters associated with generalization level 2: monitoring frequency less than 10 minutes, monitoring duration count 4, and switching threshold 3. The first monitoring parameter may include any one or more of the above parameter values.

Furthermore, the first information can also indicate the category of the first model.

In this case, the first monitoring method is also related to the category of the first model. For example, the first device can determine the first monitoring method based on the performance information of the first model, the generalization information of the first model, and the category of the first model. For instance, the first device can determine the relevant monitoring parameters based on Tables 7 to 10.

940, the network device sends a second message to the terminal device, the second message indicating the first monitoring method.

950, the terminal device performs model monitoring according to the first monitoring method.

Example #4:

In Example #4, the first model is a one-sided model and is deployed on the first device.

Figure 10 shows a schematic flowchart of a communication method according to an embodiment of this application. The method 1000 shown in Figure 10 is a specific implementation of method 500. For a detailed description, please refer to method 500. To avoid repetition, some descriptions are omitted when describing method 1000.

In method 1000, only the first device is described as a terminal device and the second device as a network device, and this does not constitute a limitation on the solution of the embodiments of this application. The main difference between method 1000 and method 800 is that the terminal device obtains the first model in a different way and the content indicated by the first information is different. When describing method 1000, the main focus is on describing the differences, and other descriptions can be found in method 800.

As shown in Figure 10, method 1000 includes the following steps.

1010, the terminal device obtains the first model.

1020, the network device sends first information to the terminal device, the first information indicating the performance information of the first model and/or the generalization information of the first model.

1030, the terminal device determines the first monitoring method based on the first information.

1040, The terminal device performs model monitoring according to the first monitoring method.

Optionally, before step 1040, method 1000 may also include step 1050.

1050. The network device sends third information to the terminal device, which can be used to instruct the terminal device to perform model monitoring.

It is understood that the information names involved in some of the above embodiments are merely examples and do not limit the scope of protection of the embodiments of this application.

It should also be understood that the formulas involved in the various embodiments of this application are merely illustrative and do not limit the scope of protection of the embodiments of this application. In calculating the parameters involved above, calculations can also be performed based on the above formulas, or on variations of the above formulas, or in other ways to satisfy the results of the formula calculations.

It is also understood that some optional features in the various embodiments of this application may not depend on other features in some scenarios, or may be combined with other features in some scenarios, without limitation.

It is also understood that the solutions in the various embodiments of this application can be used in reasonable combinations, and the explanations or descriptions of the various terms appearing in the embodiments can be referenced or explained to each other in the various embodiments, without limitation.

It should also be understood that the various numerical sequences in the embodiments of this application do not imply the order of execution, but are merely a distinction for the convenience of description, and should not constitute any limitation on the implementation process of the embodiments of this application.

It can also be understood that the methods and operations implemented by the device in the above-described method embodiments can also be implemented by components of the device (such as chips or circuits).

Corresponding to the methods described in the above embodiments, this application also provides corresponding apparatuses, which include modules for executing the methods described above. These modules can be software, hardware, or a combination of both. It is understood that the technical features described in the above method embodiments are also applicable to the following apparatus embodiments.

Figure 11 is a schematic diagram of a communication device 1900 provided in an embodiment of this application. The device 1900 includes a transceiver unit 1910 and a processing unit 1920. The transceiver unit 1910 can be used to implement corresponding communication functions. The transceiver unit 1910 can also be referred to as a communication interface or communication unit, etc. The processing unit 1920 can be used to implement corresponding processing or control functions, such as configuring resources.

Optionally, the device 1900 further includes a storage unit that can be used to store instructions and/or data. The processing unit 1920 can read the instructions and/or data in the storage unit to enable the device to perform the operation of the device or network element in the foregoing method embodiments.

The device 1900 may be a second device, or a communication device applied to or used in conjunction with a second device to realize a communication method executed on the second device side; or, the device 1900 may be a first device, or a communication device applied to or used in conjunction with a first device to realize a communication method executed on the first device side.

When the device 1900 is applied to the second device, the device 1900 can implement the steps or processes corresponding to those performed by the second device in the above method embodiments. Specifically, the transceiver unit 1910 can be used to perform transceiver-related operations of the second device in the above method embodiments, and the processing unit 1920 can be used to perform processing-related operations of the second device in the above method embodiments.

When the device 1900 is applied to the first device, the device 1900 can implement the steps or processes performed by the first device corresponding to those in the above method embodiments. The transceiver unit 1910 can be used to perform transceiver-related operations of the first device in the above method embodiments, and the processing unit 1920 can be used to perform processing-related operations of the first device in the above method embodiments.

It should be understood that the specific process of each unit performing the above-mentioned corresponding steps has been described in detail in the above-mentioned method embodiments, and will not be repeated here for the sake of brevity.

It should also be understood that the device 1900 here is embodied in the form of a functional unit. The term "unit" here can refer to an ASIC, electronic circuitry, a processor (e.g., a shared processor, a proprietary processor, or a group processor, etc.) and memory for executing one or more software or firmware programs, integrated logic circuitry, and/or other suitable components supporting the described functions. In an alternative example, those skilled in the art will understand that device 1900 may be specifically a first device in the above embodiments, used to execute the various processes and/or steps corresponding to the first device in the above method embodiments; or, device 1900 may be specifically a second device in the above embodiments, used to execute the various processes and/or steps corresponding to the second device in the above method embodiments. To avoid repetition, further details are omitted here.

The apparatus 1900 of each of the above-described schemes has the function of implementing the corresponding steps performed by the device (such as the first device, or the second device) in the above-described methods. The function can be implemented by hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions; for example, the transceiver unit can be replaced by a transceiver (e.g., the transmitting unit in the transceiver unit can be replaced by a transmitter, and the receiving unit in the transceiver unit can be replaced by a receiver), and other units, such as processing units, can be replaced by processors, respectively executing the transceiver operations and related processing operations in each method embodiment.

Furthermore, the aforementioned transceiver unit 1910 can also be a transceiver circuit (e.g., it may include a receiving circuit and a transmitting circuit), and the processing unit 1920 can be a processing circuit. The processing circuit may include one or more processors, or circuits in one or more processors used for processing or control functions, etc.

It should be noted that the device in Figure 11 can be a network element or device as described in the foregoing embodiments, or it can be a chip or chip system, such as a SoC. The transceiver unit can be an input/output circuit or a communication interface; the processing unit is a processor, microprocessor, or integrated circuit integrated on the chip. No limitations are imposed here.

Figure 12 is a schematic diagram of another communication apparatus 2000 provided in an embodiment of this application. The apparatus 2000 includes a processor 2010, which is used to execute computer programs or instructions stored in a memory 2020, or to read data/signaling stored in the memory 2020, to perform the methods in the above-described method embodiments. Optionally, there may be one or more processors 2010.

Optionally, as shown in FIG12, the device 2000 further includes a memory 2020 for storing computer programs or instructions and/or data. The memory 2020 may be integrated with the processor 2010 or may be disposed separately. Optionally, there may be one or more memories 2020.

Optionally, as shown in FIG12, the device 2000 further includes a transceiver circuit 2030 for receiving and/or transmitting signals. For example, a processor 2010 is used to control the transceiver circuit 2030 to receive and/or transmit signals. The processor 2010 may also be replaced by a processing circuit.

The device 2000 can be a network element or device as described in the foregoing embodiments, or it can be a chip or chip system. When the device 2000 is a network element or device as described in the foregoing embodiments, the transceiver circuit 2030 can be a transceiver. When the device 2000 is a chip or chip system, the transceiver circuit 2030 can be an interface circuit or an input/output interface.

As one approach, the device 2000 can be applied to a second device. Specifically, the device 2000 can be the second device itself, or it can be any device capable of supporting the second device and implementing the functions of the second device in any of the examples described above. The device 2000 is used to implement the operations performed by the second device in the various method embodiments described above.

For example, processor 2010 is used to execute computer programs or instructions stored in memory 2020 to implement the relevant operations of the second device in the various method embodiments described above.

Alternatively, the device 2000 can be applied to the first device. Specifically, the device 2000 can be the first device itself, or it can be any device capable of supporting the first device and implementing the functions of the first device in any of the examples mentioned above. The device 2000 is used to implement the operations performed by the first device in the various method embodiments described above.

For example, processor 2010 is used to execute computer programs or instructions stored in memory 2020 to implement the relevant operations of the first device in the various method embodiments described above.

It should be understood that the processor mentioned in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), ASICs, field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

It should also be understood that the memory mentioned in the embodiments of this application can be volatile memory and/or non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM). For example, RAM can be used as an external cache. By way of example and not limitation, RAM includes the following forms: static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

It should be noted that when the processor is a general-purpose processor, DSP, ASIC, FPGA, or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, the memory (storage module) can be integrated into the processor.

It should also be noted that the memory described herein is intended to include, but is not limited to, these and any other suitable types of memory.

This application also provides a computer-readable storage medium storing computer instructions for implementing the methods executed by the communication device in the above-described method embodiments.

For example, when the computer program is executed by the computer, it enables the computer to implement the methods executed by the first device in the various embodiments of the above methods.

For example, when the computer program is executed by the computer, it enables the computer to implement the methods executed by the second device in the various embodiments of the above methods.

This application also provides a computer program product comprising instructions which, when executed by a computer, implement the methods performed by a device (such as a first device or a second device) in the above-described method embodiments.

This application also provides a communication system, including the aforementioned first device and second device. The first device and second device can implement the communication method shown in any of the foregoing examples.

Optionally, the system may also include a device that communicates with the first device and/or the second device described above.

The explanations and beneficial effects of the relevant contents in any of the devices provided above can be found in the corresponding method embodiments provided above, and will not be repeated here.

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

In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. For example, the computer can be a personal computer, a server, or a network device, etc. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)). For example, the aforementioned available media include, but are not limited to, various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

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

Claims (29)

A communication method, characterized in that it includes: Receive first information, the first information indicating one or more of the following: Performance information of the first model, The generalization information of the first model, or Information related to the first dataset, which is used for training the first model; The monitoring method for the first model is determined based on the first information. The method according to claim 1, characterized in that the method further comprises: Send a second message, which indicates the monitoring method for the first model. The method according to claim 1 or 2 is characterized in that the first information indicates one or more of the following: the performance information of the first model, the generalization information of the first model, or the relevant information of the first dataset, by indicating the identifier of the first model. The method according to any one of claims 1 to 3, characterized in that the performance information of the first model includes the performance level of the first model, the performance level of the first model is related to the expected performance of the first model, and/or The generalization information of the first model includes the generalization level of the first model, which is related to the expected generalization ability of the first model. The method according to any one of claims 1 to 4 is characterized in that the first information further indicates the model category of the first model. According to the method described in claim 5, the model category of the first model includes a basic general model or a cell-specific model. The method according to any one of claims 1 to 6 is characterized in that the relevant information of the first dataset includes the identifier of the first dataset. The method according to any one of claims 1 to 7 is characterized in that the relevant information of the first dataset includes one or more of the following: performance information corresponding to the first dataset or generalization information corresponding to the first dataset. According to the method of claim 8, the performance information corresponding to the first dataset includes the performance level of the model trained on the first dataset, wherein the performance level of the model trained on the first dataset is related to the expected performance of the model trained on the first dataset, and/or, The generalization information corresponding to the first dataset includes the generalization level of the model trained on the first dataset, and the generalization level of the model trained on the first dataset is related to the expected generalization ability of the model trained on the first dataset. The method according to any one of claims 1 to 9 is characterized in that the monitoring parameters used in the monitoring method of the first model include at least one of the following: the performance threshold of the first model, the monitoring frequency of the first model, the monitoring duration of the first model, the number of monitoring sessions of the first model, the monitoring error tolerance of the first model, or the switching threshold of the first model. A communication method, characterized in that it includes: Send a first message, which indicates one or more of the following: Performance information of the first model, The generalization information of the first model, or, Information related to the first dataset, which is used for training the first model. The first information is used to determine the monitoring method of the first model. The method according to claim 11, characterized in that the method further comprises: Receive second information, which indicates the monitoring method of the first model. The method according to claim 11, characterized in that the method further comprises: Send a third message, which instructs the first model to be monitored. The method according to any one of claims 11 to 13 is characterized in that the first information indicates one or more of the following: performance information of the first model, generalization information of the first model, or relevant information of the first dataset, by indicating the identifier of the first model. The method according to any one of claims 11 to 14, characterized in that the performance information of the first model includes the performance level of the first model, the performance level of the first model being related to the expected performance of the first model, and/or The generalization information of the first model includes the generalization level of the first model, which is related to the expected generalization ability of the first model. The method according to any one of claims 11 to 15 is characterized in that the first information further indicates the model category of the first model. According to the method of claim 16, the model category of the first model includes a basic general model or a cell-specific model. The method according to any one of claims 11 to 17 is characterized in that the relevant information of the first dataset includes the identifier of the first dataset. The method according to any one of claims 11 to 18 is characterized in that the relevant information of the first dataset includes one or more of the following: performance information corresponding to the first dataset or generalization information corresponding to the first dataset. According to the method of claim 19, the performance information corresponding to the first dataset includes the performance level of the model trained on the first dataset, wherein the performance level of the model trained on the first dataset is related to the expected performance of the model trained on the first dataset, and/or, The generalization information corresponding to the first dataset includes the generalization level of the model trained on the first dataset, and the generalization level of the model trained on the first dataset is related to the expected generalization ability of the model trained on the first dataset. The method according to any one of claims 11 to 20 is characterized in that the monitoring parameters used in the monitoring method of the first model include at least one of the following: the performance threshold of the first model, the monitoring frequency of the first model, the monitoring duration of the first model, the number of monitoring sessions of the first model, the monitoring error tolerance of the first model, or the switching threshold of the first model. A computer-readable storage medium, characterized in that the computer-readable storage medium includes instructions that, when executed by a processor, cause the method of any one of claims 1 to 10 to be implemented, or cause the method of any one of claims 11 to 21 to be implemented. A communication device, characterized in that the communication device includes a processor coupled to a storage medium, the storage medium storing instructions, which, when executed by the processor, cause the communication device to perform the method as described in any one of claims 1 to 10, or to perform the method as described in any one of claims 11 to 21. A communication device, characterized in that it includes a module that performs the method as described in any one of claims 1 to 10. A communication device, characterized in that it includes a module that performs the method as described in any one of claims 11 to 21. A communication device, characterized in that it includes one or more processors, said one or more processors being configured to process data and/or information such that the method of any one of claims 1 to 10 is implemented, or that the method of any one of claims 11 to 21 is implemented. A chip, characterized in that it includes a processor for running a program or instructions to cause the method of any one of claims 1 to 10 to be implemented, or to cause the method of any one of claims 11 to 21 to be implemented. A computer program product, characterized in that it includes computer program code or instructions, which, when the computer program code or instructions are executed, cause the method as described in any one of claims 1 to 10 to be implemented, or cause the method as described in any one of claims 11 to 21 to be implemented. A communication system, characterized in that it includes the communication device as described in claim 24, and/or the communication device as described in claim 25.