Model failure recovery method and apparatus, and terminal device and network device

Abstract

Provided in the embodiments of the present application are a model failure recovery method and apparatus, and a terminal device and a network device. The method comprises: a terminal device performing performance monitoring on a first AI model, and sending model failure reporting information to a network device on the basis of a monitoring result, wherein the model failure reporting information is used for indicating that a model failure occurs in the first AI model, or for indicating whether a model failure occurs in the first AI model; and the terminal device receiving model recovery information sent by the network device, and performing information transmission with the network device on the basis of the model recovery information.

Description

Model failure recovery method and device, terminal equipment, and network equipment Technical Field

[0001] The embodiments of the present application relate to the field of mobile communication technology, and specifically to a model failure recovery method and apparatus, terminal equipment, and network equipment. Background Art

[0002] For wireless communications based on artificial intelligence (AI) / machine learning (ML), terminal devices can use AI models deployed on the terminal device side to transmit information to and from network devices. However, as the communication environment of the terminal device changes, if the AI ​​model on the terminal device side is not updated, the performance of the AI ​​model may gradually deteriorate, resulting in poor transmission performance of the terminal device, which may even be worse than non-AI / ML methods. How to timely monitor the performance of the AI ​​model and update the model is a problem that needs to be solved.

[0003] Summary of the Invention

[0004] Embodiments of the present application provide a model failure recovery method and apparatus, terminal equipment, network equipment, chip, computer-readable storage medium, computer program product, and computer program.

[0005] In a first aspect, an embodiment of the present application provides a model failure recovery method, the method comprising:

[0006] The terminal device monitors the performance of the first AI model and sends model failure reporting information to the network device based on the monitoring result, where the model failure reporting information is used to indicate that a model failure has occurred in the first AI model, or to indicate whether a model failure has occurred in the first AI model;

[0007] The terminal device receives the model recovery information sent by the network device, and transmits information with the network device according to the model recovery information.

[0008] In a second aspect, an embodiment of the present application provides a model failure recovery method, the method comprising:

[0009] The network device receives model failure reporting information sent by the terminal device, where the model failure reporting information is used to indicate that a model failure has occurred in the first AI model, or to indicate whether a model failure has occurred in the first AI model;

[0010] The network device sends model recovery information to the terminal device, where the model recovery information is used for information transmission between the terminal device and the network device.

[0011] In a third aspect, an embodiment of the present application provides a model failure recovery device, which is applied to a terminal device and includes:

[0012] a monitoring unit, configured to monitor the performance of the first AI model;

[0013] a sending unit, configured to send model failure reporting information to a network device according to a monitoring result, where the model failure reporting information is used to indicate that a model failure has occurred in the first AI model, or to indicate whether a model failure has occurred in the first AI model;

[0014] A receiving unit, configured to receive the model recovery information sent by the network device;

[0015] The sending unit is further configured to transmit information to and from the network device based on the model recovery information.

[0016] In a fourth aspect, an embodiment of the present application provides a model failure recovery device, which is applied to a network device and includes:

[0017] a receiving unit, configured to receive model failure reporting information sent by a terminal device, where the model failure reporting information is used to indicate that a model failure has occurred in the first AI model, or to indicate whether a model failure has occurred in the first AI model;

[0018] The sending unit is used to send model recovery information to the terminal device, where the model recovery information is used for information transmission between the terminal device and the network device.

[0019] In a fifth aspect, an embodiment of the present application provides a terminal device comprising a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to perform the above-mentioned model failure recovery method.

[0020] In a sixth aspect, an embodiment of the present application provides a network device comprising a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to perform the above-mentioned model failure recovery method.

[0021] In a seventh aspect, an embodiment of the present application provides a chip, which includes: a processor for calling and running a computer program from a memory, so that a device equipped with the chip executes the above-mentioned model failure recovery method.

[0022] In an eighth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by at least one processor, it implements the above-mentioned model failure recovery method.

[0023] In a ninth aspect, an embodiment of the present application provides a computer program product, which includes computer program instructions, and the computer program instructions enable a computer to execute the above-mentioned model failure recovery method.

[0024] In a tenth aspect, the computer program provided in the embodiment of the present application, when running on a computer, enables the computer to execute the above-mentioned model failure recovery method.

[0025] The technical solution of the embodiment of the present application proposes a model failure recovery method, in which the terminal device monitors the performance of the first AI model and sends model failure reporting information to the network device based on the monitoring results, and the model failure reporting information indicates that the first AI model has a model failure or indicates whether the first AI model has a model failure; after receiving the model failure reporting information, the network device sends model recovery information to the terminal device, so that the terminal device can update the first AI model in real time according to the model recovery information, and then transmit information between the network device based on the updated AI model, thereby avoiding transmission performance loss caused by the deterioration of the AI ​​model performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0027] FIG1 is a schematic diagram of an application scenario of an embodiment of the present application;

[0028] Figure 2 is a schematic diagram of an example of neuron structure;

[0029] Figure 3 is a schematic diagram of an example of a neural network;

[0030] FIG4 is a schematic diagram of a neural network for CSI feedback;

[0031] FIG5 is a flow chart of a model failure recovery method according to an embodiment of the present application;

[0032] FIG6 is a second flow chart of a model failure recovery method according to an embodiment of the present application;

[0033] FIG7 is a third flow chart of a model failure recovery method according to an embodiment of the present application;

[0034] FIG8 is a fourth flow chart of a model failure recovery method according to an embodiment of the present application;

[0035] FIG9 is a fifth flow chart of a model failure recovery method according to an embodiment of the present application;

[0036] FIG10 is a schematic diagram of the structure of a model failure recovery device provided in an embodiment of the present application;

[0037] FIG11 is a second schematic diagram of the structure of a model failure recovery device provided in an embodiment of the present application;

[0038] FIG12 is a schematic structural diagram of a communication device provided in an embodiment of the present application;

[0039] FIG13 is a schematic structural diagram of a chip according to an embodiment of the present application;

[0040] FIG14 is a schematic block diagram of a communication system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0041] The following will describe the technical solutions in the embodiments of this application in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0042] FIG1 is a schematic diagram of an application scenario of an embodiment of the present application.

[0043] As shown in Figure 1, a communication system 100 may include a terminal device 110 and a network device 120. The network device 120 may communicate with the terminal device 110 via an air interface. The terminal device 110 and the network device 120 support multi-service transmission.

[0044] It should be understood that the embodiments of the present application are only illustrative of the communication system 100, but the embodiments of the present application are not limited thereto. That is, the technical solutions of the embodiments of the present application can be applied to various communication systems, such as: Long Term Evolution (LTE) system, LTE Time Division Duplex (TDD), Universal Mobile Telecommunication System (UMTS), Internet of Things (IoT) system, Narrow Band Internet of Things (NB-IoT) system, enhanced Machine-Type Communications (eMTC) system, 5G communication system (also known as New Radio (NR) communication system), or future communication systems.

[0045] In the communication system 100 shown in Figure 1, the network device 120 may be an access network device that communicates with the terminal device 110. The access network device may provide communication coverage for a specific geographical area and may communicate with the terminal device 110 (eg, UE) located within the coverage area.

[0046] The network device 120 may be an evolved Node B (eNB or eNodeB) in a Long Term Evolution (LTE) system, or a Next Generation Radio Access Network (NG RAN) device, or a base station (gNB) in an NR system, or a wireless controller in a Cloud Radio Access Network (CRAN), or the network device 120 may be a relay station, an access point, an in-vehicle device, a wearable device, a hub, a switch, a bridge, a router, or a network device in a future evolved Public Land Mobile Network (PLMN), etc.

[0047] The terminal device 110 may be any terminal device, including but not limited to a terminal device connected to the network device 120 or other terminal devices by wire or wireless connection.

[0048] For example, the terminal device 110 may refer to an access terminal, user equipment (UE), a subscriber unit, a subscriber station, a mobile station, a mobile station, a remote station, a remote terminal, a mobile device, a user terminal, a terminal, a wireless communication device, a user agent, or a user apparatus. An access terminal may be a cellular phone, a cordless phone, a Session Initiation Protocol (SIP) phone, an IoT device, a satellite handheld terminal, a Wireless Local Loop (WLL) station, a Personal Digital Assistant (PDA), a handheld device with wireless communication capabilities, a computing device or other processing device connected to a wireless modem, an in-vehicle device, a wearable device, a terminal device in a 5G network, or a terminal device in a future evolution network, etc.

[0049] The terminal device 110 can be used for device-to-device (D2D) communication.

[0050] The wireless communication system 100 may further include a core network device 130 for communicating with the base station. The core network device 130 may be a 5G core network (5G Core, 5GC) device, such as an Access and Mobility Management Function (AMF), an Authentication Server Function (AUSF), a User Plane Function (UPF), or a Session Management Function (SMF). Optionally, the core network device 130 may also be an Evolved Packet Core (EPC) device of an LTE network, such as a Session Management Function + Core Packet Gateway (SMF+PGW-C) device. It should be understood that SMF+PGW-C can simultaneously implement the functions that can be implemented by SMF and PGW-C. During the network evolution process, the above-mentioned core network device may also be called other names, or a new network entity may be formed by dividing the functions of the core network, which is not limited in the embodiments of the present application.

[0051] The functional units in the communication system 100 may also establish connections and implement communication via next generation (NG) network interfaces.

[0052] For example, the terminal device establishes an air interface connection with the access network device through the NR interface for transmitting user plane data and control plane signaling; the terminal device can establish a control plane signaling connection with the AMF through the NG interface 1 (referred to as N1); the access network device, such as the next generation wireless access base station (gNB), can establish a user plane data connection with the UPF through the NG interface 3 (referred to as N3); the access network device can establish a control plane signaling connection with the AMF through the NG interface 2 (referred to as N2); the UPF can establish a control plane signaling connection with the SMF through the NG interface 4 (referred to as N4); the UPF can exchange user plane data with the data network through the NG interface 6 (referred to as N6); the AMF can establish a control plane signaling connection with the SMF through the NG interface 11 (referred to as N11); the SMF can establish a control plane signaling connection with the PCF through the NG interface 7 (referred to as N7).

[0053] Figure 1 exemplarily shows a base station, a core network device and two terminal devices. Optionally, the wireless communication system 100 may include multiple base station devices and each base station may include other numbers of terminal devices within its coverage area, which is not limited in this embodiment of the present application.

[0054] It should be noted that Figure 1 is merely an example of a system applicable to this application. Of course, the methods described in the embodiments of this application can also be applied to other systems. Furthermore, the terms "system" and "network" are often used interchangeably herein. The term "and / or" herein simply describes an association relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " generally indicates that the associated objects are in an "or" relationship. It should also be understood that the "indication" mentioned in the embodiments of this application can be a direct indication, an indirect indication, or an indication of an association relationship. For example, "A indicates B" can mean that A directly indicates B, for example, B can obtain information through A; it can also mean that A indirectly indicates B, for example, A indicates C, and B can obtain information through C; or it can mean that A and B have an association relationship. It should also be understood that the "correspondence" mentioned in the embodiments of this application can mean that there is a direct or indirect correspondence between two objects, or that there is an association relationship between the two objects, or a relationship between an indicator and the indicated, a configuration and the configured, and so on. It should also be understood that the “predefined” or “predefined rules” mentioned in the embodiments of the present application can be implemented by pre-saving corresponding codes, tables or other methods that can be used to indicate relevant information in a device (for example, including a terminal device and a network device), and the present application does not limit its specific implementation method. For example, predefined can refer to a definition in a protocol. It should also be understood that in the embodiments of the present application, the “protocol” can refer to a standard protocol in the field of communications, such as an LTE protocol, an NR protocol, and related protocols used in future communication systems, and the present application does not limit this.

[0055] To facilitate understanding of the technical solutions of the embodiments of the present application, the relevant technologies of the embodiments of the present application are described below. The following relevant technologies can be arbitrarily combined with the technical solutions of the embodiments of the present application as optional solutions, and they all fall within the protection scope of the embodiments of the present application.

[0056] 1. Neuronal structure

[0057] A neural network is a computational model consisting of multiple interconnected neuron nodes. The connections between the nodes represent the weighted values ​​from the input signal to the output signal, which are called weights. Each node performs a weighted summation of different input signals and outputs them through a specific activation function. The neuron structure is shown in Figure 2, where a1, a2, ..., an and 1 are the inputs of neurons, w1, w2, ..., w n and b represent weights, Sum represents the summation function, f represents the activation function, and t is the output result.

[0058] 2. Neural Networks

[0059] Figure 3 shows a simple neural network. This neural network consists of an input layer, a hidden layer, and an output layer. By connecting multiple neurons in different ways, combined with different weights and activation functions, different outputs can be generated, thereby fitting the mapping relationship from input to output. For example, each previous-level node can be connected to all of its next-level nodes, forming a fully connected model. This fully connected model can also be called a deep neural network (DNN).

[0060] 3. Model acquisition

[0061] A neural network model can be trained through the process of constructing, training, verifying, and testing a data set. The training of a neural network model can be divided into offline training and online training. It should be noted that offline training and online training are not mutually exclusive. For example, the network can first obtain a static training result through offline training of the data set. This process can be called offline training. During the use of the neural network model by the network or terminal device, as the terminal device further measures and / or reports, the neural network model can continue to collect more data and perform real-time online training to optimize the parameters of the neural network model, thereby achieving better inference and prediction results. After obtaining the neural network model, by inputting the currently obtained information into the neural network model, the corresponding model output can be inferred.

[0062] When neural network models are used for wireless communication between terminals and network devices, they can be categorized as single-ended or dual-ended models. Single-ended models can be deployed on only one side of the terminal device or network device, and model training can be performed on only one side. Dual-ended models require deployment on both the terminal device and the network device, and the models on both sides must be trained together. This means the models deployed on both sides correspond to each other and cannot be used or updated independently.

[0063] For example: the neural network model used for channel state information (CSI) feedback is a typical two-end model. On the terminal device side, there is a neural network model. The terminal device uses the measured channel information (such as eigenvectors, beam information, delay information, etc.) as the input of the model to infer the corresponding CSI quantization bits. On the network device side, there will be a neural network model corresponding to the terminal device side. The network device uses the CSI quantization bits fed back by the terminal device as the input of the model, and can reversely infer the corresponding channel information. For example, Figure 4 shows a typical neural network model for CSI feedback (or for CSI reporting). A neural network model for encoding (encoder) (referred to as the encoding model for short) is deployed on the terminal device side, and a corresponding neural network model for decoding (decoder) (referred to as the decoding model for short) is deployed on the network side. The terminal device outputs CSI quantization bits (such as PMI bits) based on the coding model, and then feeds them back to the network device through the uplink channel (such as the Physical Uplink Shared Channel (PUSCH) / Physical Uplink Control Channel (PUCCH)). The network device uses the CSI quantization bits fed back by the terminal device as the input of the decoding model, thereby outputting channel information corresponding to the input on the terminal device side (such as eigenvectors, beam information, delay information, etc.) for downlink precoding. In addition, the terminal device can also monitor the performance of the coding model, update the coding model when the coding model performance is not good, and report it to the network device, so that the network device can synchronously update the decoding model. Here, since the neural network models on both sides of the terminal device and the network device are matched, if the model is updated, they need to notify each other through signaling to achieve the purpose of synchronous model update.

[0064] For example, the neural network model used for beam management can be a single-ended model. On the terminal device side, a neural network model measures the beams in the second beam set (called Set B) and uses the measured beam signal quality as input to the model, thereby inferring the optimal beam index and the corresponding signal quality of the optimal beam in the first beam set (called Set A). This type of downlink beam prediction is called spatial-domain downlink beam prediction (DL beam prediction). In addition, there is another type of downlink beam prediction: temporal DL beam prediction.

[0065] Definition of spatial-domain beam prediction (BM-Case 1): Downlink beams in Set A are spatially predicted by measuring beams (pairs) in Set B. Set B is either a subset of Set A or different sets of beams (pairs). Set B can be considered a subset of Set A, while Set A can be considered the full set of beams (pairs).

[0066] Definition of time-domain beam prediction (BM-Case2): The optimal beam (or beam pair) in Set A is predicted F times in the future by measuring the beams (or beam pairs) in Set B K times. Set B can be a subset of Set A or the same as Set A. If Set B is the same as Set A, beam prediction is purely time-domain. If Set B is a subset of Set A, beam prediction is performed in both the spatial and time domains.

[0067] In addition, Set B and Set A can also be two different sets, Set B is a group of a small number of wide beams, and Set A is a group of a large number of narrow beams. The small number of wide beams in Set B can roughly cover the large number of narrow beams in Set A in space.

[0068] It should be understood that the output of the model can also be understood as inference, prediction or estimation results, that is, inference, prediction and estimation can represent the same meaning and can be used interchangeably.

[0069] The neural network model is used for wireless communication between the terminal and the network device, which can also be called AI / ML-based wireless communication. For AI / ML-based wireless communication, the terminal device can use the neural network model deployed on the terminal device side to transmit information with the network device, such as: CSI reporting (or CSI feedback), beam reporting, etc. However, as the communication environment of the terminal device changes, if the model on the terminal device side is not updated, the performance of the model may gradually deteriorate, resulting in the transmission performance of the terminal device being even poor, which may be worse than that of non-AI / ML. To this end, the following technical solutions of the embodiments of the present application are proposed.

[0070] To facilitate understanding of the technical solutions of the embodiments of the present application, the technical solutions of the present application are described in detail below through specific embodiments. The above related technologies can be combined arbitrarily with the technical solutions of the embodiments of the present application as optional solutions, and all of them fall within the scope of protection of the embodiments of the present application. The embodiments of the present application include at least part of the following contents.

[0071] It should be noted that the "AI model" described in the embodiments of the present application can be called a neural network model, an ML model, or an AI / ML model. For example, the AI ​​model in the embodiments of the present application can be used for CSI reporting, and the input and output of the AI ​​model can refer to the aforementioned description of the neural network model for CSI feedback. For example, the AI ​​model in the embodiments of the present application can also be used for beam reporting, and the input and output of the AI ​​model can refer to the aforementioned description of the neural network model for beam management. The AI ​​model in this application can also be used for other communication functions, such as channel coding, modulation, precoding, etc., which are not limited in this application.

[0072] It should be noted that, unless otherwise specified, the "AI model" described in the embodiments of the present application refers to the model on the terminal device side. The AI ​​model on the terminal device side is used for information transmission between the terminal device and the network device.

[0073] FIG5 is a flow chart of a model failure recovery method according to an embodiment of the present application. As shown in FIG5 , the method includes all or part of the following contents:

[0074] Step 501: The terminal device monitors the performance of the first AI model and sends model failure reporting information to the network device based on the monitoring results. The network device receives the model failure reporting information sent by the terminal device. The model failure reporting information is used to indicate that a model failure has occurred in the first AI model, or to indicate whether a model failure has occurred in the first AI model.

[0075] Here, the terminal device monitors the performance of the first AI model and determines whether the first AI model has experienced a model failure based on the monitoring results. Based on whether the first AI model has experienced a model failure, the terminal device sends model failure reporting information to the network device. In response, the network device receives the model failure reporting information sent by the terminal device.

[0076] The above performance monitoring performed by the terminal device may be periodic, semi-continuous, or non-periodic.

[0077] In some embodiments, before the terminal device performs performance monitoring of the first AI model, the following steps may also be included: the network device sends a first signaling to the terminal device; the terminal device receives the first signaling sent by the network device; wherein the first signaling is used to configure the period of the above-mentioned performance monitoring and / or the period of the terminal device sending the above-mentioned model failure reporting information (that is, the reporting period of the model failure reporting information); or, the first signaling is used to activate the terminal device to perform the above-mentioned performance monitoring and / or activate the terminal device to send the above-mentioned model failure reporting information.

[0078] Exemplarily, the above-mentioned performance monitoring is periodic, and the network device configures the performance monitoring period through a first signaling (such as RRC signaling). After the terminal device receives the first signaling, it periodically monitors the performance of the first AI model according to the period configured by the first signaling until the configuration fails and the monitoring is stopped.

[0079] Exemplarily, the above-mentioned performance monitoring is semi-continuous, and the network device activates the terminal device through a first signaling (such as MAC signaling or DCI) to perform periodic performance monitoring. After receiving the first signaling, the terminal device periodically monitors the performance of the first AI model until it receives a second signaling (such as MAC signaling or DCI) sent by the network device to stop monitoring. The second signaling is used to deactivate the terminal device for periodic performance monitoring.

[0080] For example, the above performance monitoring is non-periodic, and the network device activates the terminal device to perform one or more continuous performance monitoring via a first signaling (such as DCI). After receiving the first signaling, the terminal device performs one or more continuous performance monitoring.

[0081] Exemplarily, the reporting of the above-mentioned model failure reporting information is periodic, and the network device configures the period of the terminal device sending the model failure reporting information through the first signaling (such as RRC signaling). After the terminal device obtains the first signaling, it periodically monitors the performance of the first AI model before sending the model failure reporting information each time according to the period configured by the first signaling, and periodically sends the corresponding model failure reporting information based on the monitoring results until the configuration fails and stops the performance monitoring and reporting of the model failure reporting information.

[0082] Exemplarily, the reporting of the above-mentioned model failure reporting information is semi-continuous, and the network device activates the terminal device through the first signaling (such as MAC signaling or DCI) to perform periodic reporting of the model failure reporting information. After the terminal device obtains the first signaling, it periodically monitors the performance of the first AI model and reports the model failure reporting information until it receives the second signaling (such as MAC signaling or DCI) sent by the network device to stop reporting the model failure reporting information. The second signaling is used to deactivate the terminal device to perform semi-continuous reporting of the model failure reporting information. The reporting period of the model failure reporting information can be pre-configured through RRC signaling.

[0083] Exemplarily, the reporting of the above-mentioned model failure reporting information is non-periodic, and the network device activates the terminal device through the first signaling (such as DCI) to perform one or more continuous reporting of the model failure reporting information. After the terminal device obtains the first signaling, it performs one or more continuous performance monitoring and corresponding reporting of the model failure reporting information.

[0084] In some embodiments, the first signaling is also used to indicate the type of monitoring results used for performance monitoring and / or a threshold value for model failure, which is used by the terminal device to determine whether a model failure occurs in the first AI model based on the monitoring results.

[0085] Exemplarily, the first signaling may indicate that the monitoring result types used for the above-mentioned performance monitoring are block error rate (BLER), throughput, CSI error, CSI consistency, beam signal quality measurement error, beam prediction consistency, square cosine similarity (SGCS) of channel information, cosine similarity (GCS) of channel information, positioning error, etc. The signal quality here may include at least one of the following: reference signal receiving power (RSRP) (such as L1-RSRP), signal to interference plus noise ratio (SINR) (such as L1-SINR), received signal strength indicator (RSSI) (such as L1-RSSI), reference signal receiving quality (RSRQ) (such as L1-RSRQ). In the following, the signal quality is explained using RSRP as an example.

[0086] In some embodiments, the first signaling is also used to configure resources for carrying model failure reporting information. The resources for carrying model failure reporting information may be PUCCH resources or PUSCH resources. If the model failure reporting information is reported via PUCCH (i.e., the resources for carrying model failure reporting information are PUCCH resources), the PUCCH resources used by the terminal device may be obtained through the first signaling. If the model failure reporting information is reported via PUSCH (i.e., the resources for carrying model failure reporting information are PUSCH resources), the slot offset of the PUSCH may be obtained through the first signaling. In other embodiments, the first signaling is also used to indicate the CSI reporting configuration corresponding to the model failure reporting information, and the CSI reporting configuration may configure reporting resources (such as PUCCH resources or PUSCH resources), and the terminal device determines the reporting resources used for the model failure reporting information based on the CSI reporting configuration.

[0087] The terminal device monitors the performance of the first AI model by comparing the output of the first AI model with a reference tag. Here, the reference tag may also be referred to as an AI reference tag.

[0088] The terminal device can perform one or more performance monitoring on the first AI model, and each performance monitoring can obtain a corresponding monitoring result. After each performance monitoring of the first AI model, the terminal device can obtain the corresponding monitoring result by comparing the output of the first AI model with the reference label.

[0089] In some embodiments, the reference label is a ground truth label corresponding to the output of the first AI model. The ground truth label can be obtained by the terminal device or notified to the terminal device by the network device. For example, the ground truth label can be beam information obtained by the terminal device based on the relative position between the terminal device and the network device, or the ground truth label can be ideal channel information, or the ground truth label can be accurate positioning information obtained by ideal means. The terminal device monitors the performance of the first AI model by comparing the output of the first AI model with the corresponding ground truth label.

[0090] In some embodiments, the reference tag is an output obtained based on a non-AI / ML method. The terminal device monitors the performance of the first AI model by comparing the output of the first AI model with the output obtained based on the non-AI / ML method.

[0091] The terminal device performs one or more performance monitoring on the first AI model and obtains corresponding monitoring results. Then, the terminal device can determine whether the first AI model has experienced a model failure based on the detection results. Here, the terminal device can determine whether the first AI model has experienced a model failure based on the detection results by the following methods:

[0092] Method 1 (physical layer method): If in multiple performance monitorings of the first AI model, the monitoring results are all worse than the threshold value, or the monitoring results are all different from the preset value, or the statistical value of the monitoring results is worse than the threshold value, the terminal device determines that the first AI model has a model failure.

[0093] Here, the terminal device performs multiple performance monitoring on the first AI model. If the monitoring results of the multiple performance monitoring are all worse than the preset threshold value, or the statistical values ​​of the monitoring results of the multiple performance monitoring are worse than the preset threshold value, or the monitoring results of the multiple performance monitoring are all different from the preset value, the terminal device determines that a model failure has occurred. Exemplarily, the statistical value of the monitoring result can be, for example: the average value of the monitoring result over a period of time, such as the BLER average value, the throughput average value, the beam RSRP measurement error average value, etc. Exemplarily, the statistical value of the monitoring result can be, for example: the proportion of the monitoring result that is a preset value over a period of time, such as the proportion of the monitoring result that is "consistent beam prediction" over a period of time or the proportion of the monitoring result that is "inconsistent beam prediction" over a period of time. Exemplarily, the preset value can be the optimal beam obtained by other means, and the monitoring result of the performance monitoring can be the optimal beam predicted by the first AI model. If the optimal beams predicted multiple times are all different from the preset value, the terminal determines that a model failure has occurred.

[0094] Method 2 (physical layer and high-layer method): In each performance monitoring of the first AI model, if the monitoring result is worse than the threshold value (for example, if the measured SGCS is lower than the preset threshold value) or the monitoring result is different from the preset value (for example, the predicted optimal beam is different from the actual optimal beam), the terminal device reports the model failure sample to the high layer; if the high layer counts that the number of model failure samples reaches the specified maximum value, the terminal device determines that a model failure has occurred.

[0095] Here, in each performance monitoring performed by the terminal device on the first AI model, if the monitoring result is worse than the threshold value or the monitoring result is different from the preset value, the terminal device reports the model failure sample to the upper layer (such as the MAC layer); at the upper layer, when the counter used to count the model failure samples reaches the specified maximum value, the terminal device determines that a model failure has occurred.

[0096] In the embodiment of the present application, the terminal device sends model failure reporting information to the network device based on the monitoring results, which can be achieved by the following method:

[0097] Method a) If the monitoring result is that the first AI model has a model failure, the terminal device sends model failure reporting information to the network device, and the model failure reporting information is used to indicate that the first AI model has a model failure.

[0098] Here, the terminal device only sends model failure reporting information to the network device when a model failure occurs in the first AI model. In other words, the sending of the model failure reporting information is triggered by an event, namely, the terminal device detecting a model failure in the first AI model. If the first AI model does not fail, the terminal device does not need to send the model failure reporting information.

[0099] In some implementations, the above-mentioned model failure reporting information may be sent via a PUCCH resource used for a scheduling request (SR).

[0100] Method b) The terminal device sends model failure reporting information to the network device, where the model failure information is used to indicate whether a model failure occurs in the first AI model.

[0101] Here, the terminal device sends model failure reporting information to the network device, and the model failure reporting information is used to indicate whether a model failure has occurred. The sending of the model failure reporting information can be triggered by a trigger signaling sent by the network device, or periodically sent through an uplink control channel. That is, regardless of whether the first AI model fails, if the terminal device receives the trigger signaling sent by the network device, it will send the model failure reporting information to the network device. Alternatively, the terminal device periodically sends the model failure reporting information on the uplink control channel configured by the network device.

[0102] Exemplarily, whether a model failure occurs may be indicated by 1 bit of information in the model failure reporting information.

[0103] In some embodiments, the above-mentioned model failure reporting information includes first indication information, and the first indication information is used to indicate one or more second AI models (or the first indication information includes information of one or more second AI models, or the model failure reporting information includes information of one or more second AI models).

[0104] Case 1) In some embodiments, the one or more second AI models are one or more AI models recommended by the terminal device after performance monitoring of the candidate AI models.

[0105] The terminal device monitors the performance of the candidate AI models, determines one or more AI models (i.e., one or more second AI models) whose monitoring results meet preset conditions, and reports information about the one or more second AI models (i.e., the first indication information) to the network device via model failure reporting information. Here, the performance monitoring of the candidate AI models can be event-triggered, for example, when a model failure occurs in the first AI model, the performance monitoring of the candidate AI models is performed.

[0106] In some embodiments, the candidate AI models are pre-configured by the network device to the terminal device, or the terminal device pre-trains the basic AI model. For example, each candidate AI model may correspond to a CSI reporting configuration, and the network device indicates the candidate AI model through the CSI reporting configuration, thereby configuring the candidate AI model. For example, the terminal device may have a local pre-stored basic AI model, and the terminal device may pre-train the basic AI model to obtain the candidate AI model.

[0107] In some embodiments, the above-mentioned preset condition may be the best monitoring result or the monitoring result closest to the preset value. Exemplarily, when the first AI model is used for beam reporting, the terminal device selects one or more AI models from the candidate AI models with the highest probability of correctly predicting the optimal beam index (i.e., the highest proportion of consistent beam predictions over a period of time), or selects one or more AI models that correctly predict the optimal beam index from the candidate AI models, or selects one or more models from the candidate AI models with the smallest error in predicting beam RSRP measurement. Exemplarily, when the first AI model is used for CSI reporting, the terminal device selects one or more AI models with the highest SGCS from the candidate AI models, or selects one or more models from the candidate AI models whose corresponding channel capacity is closest to the expected channel capacity.

[0108] The terminal device can report information of multiple second models through model failure reporting information. If model failure occurs in multiple terminal devices and they use the same model, the network device can refer to the information of multiple second models reported by each terminal device and select one of the second models for the multiple terminal devices.

[0109] In some embodiments, the first indication information is further used to indicate the monitoring result corresponding to each second AI model (or the first indication information also includes the monitoring result corresponding to each second AI model). The monitoring result here can be a quantified value of the performance monitoring result. If the terminal device reports multiple second AI models and corresponding monitoring results through model failure reporting information, the network device can select an appropriate AI model for the terminal device from the multiple second AI models based on the monitoring results.

[0110] In some embodiments, if there is no AI model whose monitoring results meet the preset conditions among the candidate AI models, the first indication information is used to indicate that there is no recommended model. For example, the first indication information has multiple indication states, one of which can be used to indicate that there is no recommended model, and the other indication states are used to indicate the information of one or more second AI models reported. After receiving the first indication information indicating that there is no recommended model, the network device can indicate another model for information transmission to the terminal device, or can instruct the terminal device to use a non-AI / ML-based method for information transmission.

[0111] In some implementations, the above-mentioned model failure reporting information may be reported to the network device together with the beam information.

[0112] In some embodiments, the second AI model and the first AI model may be completely different models, i.e., have different model structures and different model parameters. In other embodiments, the second AI model and the first AI model may differ only in model parameters but have the same model structure.

[0113] Case 2) In some embodiments, the one or more second AI models are one or more AI models obtained after the terminal device performs online training on the first AI model.

[0114] The terminal device can perform online training (fine-tuning) on ​​the first AI model to obtain a second AI model, and report information of the second AI model to the network device through model failure reporting information.

[0115] For example, the terminal device can perform online training on the first AI model based on the most recently collected training data, thereby updating the model parameters of the first AI model to obtain a second AI model, and report information about the second AI model to the network device via model failure reporting. As an implementation method, the terminal device can perform online training on the first AI model using different training data to obtain multiple second AI models. As another implementation method, the terminal device can train the first AI model multiple times using the same training data to obtain multiple second AI models.

[0116] Since the second AI model is obtained after the terminal device performs online training on the first AI model, the second AI model and the first AI model may only differ in model parameters but have the same model structure.

[0117] In some embodiments, the first indication information may include information of one or more second AI models, where the information of the second AI model may include at least one of the following information: model functionality, model ID, model index, model structure, model parameters, and ID of the CSI reporting configuration corresponding to the model, where the model index refers to the index corresponding to the second AI model in the candidate AI models.

[0118] Exemplarily, the second AI model differs from the first AI model only in model parameters, but has the same model structure. The information of the second AI model reported by the terminal device may include the model parameters but not the model structure. Optionally, it may also include information of the other second AI models mentioned above.

[0119] In some embodiments, the second AI model may have the same model functions as the first AI model. In other embodiments, the second AI model may have different model functions than the first AI model. The model functions are functions implemented by the AI / ML model, such as: PMI reporting, RI / PMI / CQI reporting, beam information reporting, RSRP reporting, CSI compression, CSI prediction, CSI prediction and compression, channel coding-channel decoding, modulation-demodulation, pilot generation-channel estimation, and transceiver RF signal processing.

[0120] The above solution, by carrying the information of the second AI model recommended by the terminal device in the model failure reporting information, can enable the network device to quickly determine the new model, greatly shorten the time required for the new model application, and thus quickly update the model.

[0121] In some embodiments, the above-mentioned model failure reporting information includes first indication information, and the first indication information is used to indicate whether the terminal device confirms the second AI model, or in other words, the first indication information is used to indicate whether the terminal device obtains an available second AI model.

[0122] Here, the terminal device may not report information about one or more second AI models through model failure reporting information, but instead may indicate through model failure reporting information (e.g., through the second indication information described above) whether the terminal device has confirmed the second AI model, or whether a usable second AI model has been obtained. For example, the terminal device may indicate through 1-bit indication information (i.e., the second indication information) whether the terminal device has confirmed the second AI model, or whether a usable second AI model has been obtained.

[0123] Here, the terminal device may confirm the second AI model (or obtain a usable second AI model) by, but is not limited to, the following methods: 1) the terminal device monitors the performance of candidate AI models to obtain a usable second AI model and confirms the second AI model. The method for obtaining the second AI model may refer to the description of the aforementioned scenario 1). 2) the terminal device obtains a usable second AI model by online training the first AI model and confirms the second AI model. The method for obtaining the second AI model may refer to the description of the aforementioned scenario 2).

[0124] Step 502: The network device sends model recovery information to the terminal device. The terminal device receives the model recovery information sent by the network device and transmits information with the network device according to the model recovery information.

[0125] In some implementations, the model recovery information may be sent via broadcast signaling, high-layer signaling, or physical-layer signaling.

[0126] In some embodiments, the above-mentioned model recovery information is used to indicate whether the terminal device uses the second AI model reported by the terminal device through the above-mentioned model failure reporting information to replace the first AI model, for information transmission between the network device; or, the above-mentioned model recovery information is used to indicate an AI model from multiple second AI models reported by the terminal device through the above-mentioned model failure reporting information to replace the first AI model, for information transmission between the network device; or, the above-mentioned model recovery information is used to indicate whether the terminal device uses the second AI model confirmed by the terminal device to replace the first AI model, for information transmission between the network device; or, the above-mentioned model recovery information is used to indicate an AI model from candidate AI models to update the first AI model, for information transmission between the network device; or, the above-mentioned model recovery information is used to instruct the terminal device to use a non-AI / ML method to transmit information between the network device.

[0127] Exemplarily, when the model failure reporting information includes information about a second AI model, the model recovery information is used to indicate whether the terminal device uses the reported second AI model instead of the first AI model for subsequent information transmission with the network device.

[0128] Exemplarily, when the model failure reporting information includes information of multiple second AI models, the model recovery information is used to specify an AI model from the multiple second AI models reported by the terminal device to replace the first AI model for subsequent information transmission between the network device.

[0129] Exemplarily, when the model failure reporting information includes indication information that the terminal device confirms a second AI model or obtains an available second AI model, the model recovery information is used to indicate whether the terminal device uses the confirmed second AI model instead of the first AI model for subsequent information transmission with the network device.

[0130] Exemplarily, when the model failure reporting information does not include information about the second AI model, or includes information about no recommended model, or includes indication information that the terminal device has not confirmed the second AI model, the model recovery information is used to indicate an AI model from the candidate AI models pre-configured for the terminal device to update the first AI model for subsequent information transmission between the network device.

[0131] Exemplarily, when the model failure reporting information does not include information about the second AI model, or includes information about no recommended model, or includes indication information that the terminal device has not confirmed the second AI model, the model recovery information is used to instruct the terminal device to use a non-AI / ML method to transmit information with the network device.

[0132] In some embodiments, the model recovery information may indicate a CSI reporting configuration, which includes traditional non-AI / ML CSI reporting configuration information (the configuration information is used for non-AI / ML beam reporting or SCI reporting), or includes AI / ML-based CSI reporting configuration information (the configuration information is used for AL / ML-based beam reporting or CSI reporting). The terminal device may continue to perform AI / ML-based beam reporting or CSI reporting based on the configuration information, or adopt the configured non-AI / ML CSI reporting method.

[0133] After receiving the model recovery information, the terminal device can transmit information with the network device based on the model recovery information.

[0134] In some embodiments, the terminal device replaces the first AI model with the target AI model indicated by the model recovery information, and uses the target AI model to transmit information with the network device. In other embodiments, the terminal device updates the model structure of the first AI model with the target model structure indicated by the model recovery information, and / or updates the model parameters of the first AI model with the target model parameters indicated by the model recovery information to obtain the target AI model, and uses the target AI model to transmit information with the network device. In other embodiments, the terminal device uses the non-AI / ML method indicated by the model recovery information to transmit information with the network device.

[0135] Exemplarily, when the model recovery information instructs the terminal device to use the reported second AI model (i.e., the target AI model is the second AI model) instead of the first AI model for subsequent information transmission between the terminal device and the network device, the terminal device uses the second AI model instead of the first AI model for subsequent information transmission between the terminal device and the network device.

[0136] Exemplarily, when the model recovery information instructs the terminal device to use a new AI model (the new AI model has the above-mentioned target model structure and target model parameters) for subsequent information transmission between the terminal device and the network device, the terminal device updates the first AI model with the new AI model, that is, updates the first AI model with the above-mentioned target model structure and target model parameters to obtain the target AI model, and the terminal device adopts the target AI model for subsequent information transmission between the terminal device and the network device.

[0137] Exemplarily, when the model recovery information instructs the terminal device to use a new set of model parameters (the new model parameters are the above-mentioned target model parameters) for subsequent information transmission between the terminal device and the network device, the terminal device updates the first AI model with the new model parameters to obtain the target AI model, and the terminal device uses the target AI model for subsequent information transmission between the terminal device and the network device.

[0138] Exemplarily, when the model recovery information indicates that the terminal device uses a non-AI / ML method to transmit information with the network device, the terminal device uses a traditional non-AI / ML method to transmit information with the network device.

[0139] As a scenario, for the scenario of a single-end model (the meaning of the single-end model can refer to the records of the aforementioned related technologies), the above-mentioned first AI model is used for beam reporting. Accordingly, the information transmission between the terminal device and the network device refers to: the terminal device reports the beam to the network device. As another scenario, for the scenario of a dual-end model (the meaning of the dual-end model can refer to the records of the aforementioned related technologies), the above-mentioned first AI model is used for CSI reporting. Accordingly, the information transmission between the terminal device and the network device refers to: the terminal device reports CSI to the network device.

[0140] In some embodiments, after determining that a model failure has occurred, after sending model failure reporting information, or after receiving model recovery information, and before transmitting information with the network device according to the model recovery information, the terminal device transmits information with the network device in one of the following ways:

[0141] 1) The terminal device transmits information (such as beam reporting or CSI reporting) to the network device in a non-AI / ML manner; accordingly, the network device receives information sent by the terminal device in a non-AI / ML manner.

[0142] Terminal devices use non-AI / ML methods to transmit information to network devices. This can happen in the following situations:

[0143] 1.1) When the first AI model is used for CSI reporting, the terminal device uses the first codebook type for CSI reporting. Accordingly, the network device receives the CSI report sent by the terminal device using the first codebook type. The first codebook type is indicated by the model recovery information or is a codebook type pre-agreed between the terminal device and the network device.

[0144] 1.2) When the first AI model is used for beam reporting, the terminal device uses a non-AI / ML beam reporting method to perform beam reporting. Accordingly, the network device receives the beam reporting sent by the terminal device using the non-AI / ML beam reporting method.

[0145] Here, the terminal device performs beam reporting based on the traditional method (i.e., non-AI / ML beam reporting method). For example, the terminal device measures the beams in the second beam set and reports the optimal beam index and corresponding RSRP therein, without having to report the optimal beam index and corresponding RSRP in the first beam set. In contrast, if an AI / ML-based beam reporting method (i.e., an AI model-based beam reporting method) is adopted, since the AI ​​model can predict the optimal beam index and corresponding RSRP in the first beam set based on the measurement results of the second beam set, the terminal device reports the optimal beam index and corresponding RSRP in the first beam set.

[0146] 2) When the first AI model is used for CSI reporting or beam reporting, the terminal device uses the first CSI reporting configuration to perform CSI reporting or beam reporting. Accordingly, the network device receives the CSI reporting or beam reporting sent by the terminal device using the first CSI reporting configuration. The first CSI reporting configuration is indicated by the model recovery information or is a CSI reporting configuration pre-agreed between the terminal device and the network device.

[0147] 3) The terminal device uses a third AI model to transmit information (such as beamforming or CSI reporting) to and from the network device. Accordingly, the network device receives information sent by the terminal device using the third AI model, which is indicated by model recovery information or is a pre-agreed AI model between the terminal device and the network device. Here, the third AI model can be an AI model with good generalization performance and can be applied to different types of terminal devices and scenarios.

[0148] 4) The terminal device does not transmit information to or from the network device. For example, the terminal device does not report beamforming or CSI. In other words, after a model failure occurs, the terminal device stops reporting beamforming or CSI until a new model is configured and used.

[0149] The above methods can be understood as fallback modes adopted during model failure. By adopting the fallback mode during model failure, the performance loss caused by model performance deterioration can be minimized and the continuity of information transmission can be ensured.

[0150] In some implementations, the aforementioned fallback modes may be used in one of the following time windows:

[0151] 1) After the terminal device determines that a model failure has occurred and before the terminal device transmits information to and from the network device based on the model recovery information. Transmitting information to and from the network device based on the model recovery information may refer to the terminal device applying the new model indicated by the model recovery information to transmit information to and from the network device, or the terminal device transmitting information to and from the network device using a non-AI / ML method based on the model recovery information.

[0152] 2) After the terminal device sends the model failure reporting information, and before the terminal device transmits information with the network device based on the model recovery information.

[0153] 3) After the terminal device receives the model recovery information sent by the network device, and before the terminal device transmits information with the network device according to the model recovery information.

[0154] In some embodiments, the terminal device determines the time when the target AI model indicated by the model recovery information starts to be applied based on the model application time; after the above time, the terminal device uses the target AI model to transmit information between the network device.

[0155] Here, the model application time is reported by the terminal device to the network device via UE capabilities, or configured by the network device to the terminal device. The model application time is used by the terminal device to determine the time when the target AI model indicated by the model recovery information begins to be applied. For example, the model application time can be indicated to the terminal device through the model recovery information together with other information about the target AI model.

[0156] Exemplarily, the terminal device reports the model application time to the network device through the UE capability, or determines the moment when the target AI model indicated by the model recovery information (i.e., the updated AI model) starts to be applied according to the model application time configured by the network device, and adopts the target AI model for beam reporting or CSI reporting after that moment. In this way, by negotiating the model application time, the network device can know whether the information reported by the terminal device is based on the updated AI model, so as to perform corresponding processing. For the dual-end model, the network device can update the model synchronously with the terminal device to ensure the accuracy of the recovery information output by the network side.

[0157] In some embodiments, after the network device sends the model recovery information to the terminal device, the method further includes: the network device receiving information sent by the terminal device using the target AI model indicated by the model recovery information. In some embodiments, for a dual-end model scenario, the network device uses the network-side model corresponding to the target AI model to process the information. Exemplarily, the network device uses the network-side model corresponding to the target AI model to decode the information.

[0158] The technical solution of the embodiment of the present application proposes a model failure recovery method, in which the terminal device monitors the performance of the first AI model and sends model failure reporting information to the network device based on the monitoring results, and the model failure reporting information indicates that the first AI model has a model failure or indicates whether the first AI model has a model failure; after receiving the model failure reporting information, the network device sends model recovery information to the terminal device, so that the terminal device can update the first AI model in real time according to the model recovery information, and then transmit information between the network device based on the updated AI model, thereby avoiding transmission performance loss caused by the deterioration of the AI ​​model performance.

[0159] The technical solutions of the embodiments of the present application are illustrated below with reference to specific application examples.

[0160] Application Example 1

[0161] This application example is applied to beam management in a single-end model scenario. In this scenario, a first AI model is deployed on the terminal device side. The first AI model is used for beam management (or beam reporting), which can be called AI / ML-based beam management (or beam reporting). The input and output of the first AI model can refer to the aforementioned description of the neural network model for beam management. Specifically, the terminal device uses the measured RSRP of multiple beams in the second beam set as the input of the first AI model, and infers the optimal beam index and corresponding RSRP in the first beam set based on the first AI model; the terminal device reports the inference result to the network device (that is, the terminal device performs beam reporting).

[0162] FIG6 is a second flow chart of a model failure recovery method provided in an embodiment of the present application. As shown in FIG6 , the method includes all or part of the following contents:

[0163] Step 601: The terminal device monitors the performance of the first AI model and determines whether a model failure occurs based on the monitoring results.

[0164] Here, the above performance monitoring may be periodic, non-semi-continuous, or non-periodic.

[0165] In some embodiments, before the terminal device performs performance monitoring of the first AI model, the following steps may also be included: the terminal device receives a first signaling sent by the network device.

[0166] In one implementation, the first signaling is used to configure the period of the above-mentioned performance monitoring and / or the period of the terminal device sending the above-mentioned model failure reporting information (that is, the reporting period of the model failure reporting information). Here, the above-mentioned performance monitoring can be periodic. The reporting of the above-mentioned model failure reporting information can be periodic. In some embodiments, the period of the above-mentioned performance monitoring can be the same as the period for beam reporting, or the same as the reporting period of the model failure reporting information.

[0167] In another implementation, the first signaling is used to activate the terminal device to perform the above-mentioned performance monitoring and / or activate the terminal device to send the above-mentioned model failure reporting information. Here, the above-mentioned performance monitoring can be semi-continuous or non-periodic. If it is semi-continuous, the first signaling (such as MAC signaling) is used to activate the terminal device to perform periodic performance monitoring; if it is non-periodic, the first signaling (such as DCI) is used to trigger the terminal device to perform one or more continuous performance monitoring. Here, the reporting of the above-mentioned model failure reporting information can be semi-continuous or non-periodic. If it is semi-continuous, the first signaling (such as MAC signaling) is used to activate the terminal device to report periodic model failure reporting information; if it is non-periodic, the first signaling (such as DCI) is used to trigger the terminal device to report one or more continuous model failure reporting information.

[0168] In some embodiments, the first signaling is also used to indicate the type of monitoring results used for performance monitoring and / or a threshold value for model failure, which is used by the terminal device to determine whether a model failure occurs in the first AI model based on the monitoring results.

[0169] Exemplarily, the first signaling may indicate that the monitoring result type used for the above-mentioned performance monitoring is BLER, throughput, beam RSRP measurement error, beam prediction consistency, etc.

[0170] In some embodiments, the first signaling is also used to configure resources for carrying model failure reporting information. The resources for carrying model failure reporting information may be PUCCH resources or PUSCH resources. If the model failure reporting information is reported via PUCCH (i.e., the resources for carrying model failure reporting information are PUCCH resources), the PUCCH resources used by the terminal device may be obtained through the first signaling. If the model failure reporting information is reported via PUSCH (i.e., the resources for carrying model failure reporting information are PUSCH resources), the slot offset of the PUSCH may be obtained through the first signaling. In other embodiments, the first signaling is also used to indicate the CSI reporting configuration corresponding to the model failure reporting information, and the CSI reporting configuration may configure reporting resources (such as PUCCH resources or PUSCH resources), and the terminal device determines the reporting resources used for the model failure reporting information based on the CSI reporting configuration.

[0171] The manner in which the terminal device performs performance monitoring may refer to the description related to FIG5 above.

[0172] The following are several examples to illustrate that the terminal device monitors the performance of the first AI model by comparing the output of the first AI model with the true label.

[0173] In one example, the monitoring result type is beam prediction consistency, the true label is the optimal beam index in the first beam set, and the output of the first AI model is the optimal beam index estimated by the terminal device. The terminal device can obtain the monitoring result by comparing whether the two are consistent, that is, the monitoring result is "beam prediction consistency" or "beam prediction inconsistency" (hereinafter referred to as "beam prediction consistency / inconsistency"). It should be noted that the monitoring result of "beam prediction consistency" means that the output of the AI ​​model is consistent with the reference label, and the monitoring result of "beam prediction inconsistency" means that the output of the AI ​​model is inconsistent with the reference label. The beam index here can also be replaced by a downlink reference signal resource index or a downlink reference signal index or a spatial filter index. Among them, the downlink reference signal resource index can be a CSI-RS resource indication (CRI) or an SSB resource indication (SSBRI).

[0174] In one example, the monitoring result type is beam RSRP measurement error, the true label is the RSRP of the optimal beam in the first beam set, and the output of the first AI model is the optimal RSRP estimated by the terminal device. The terminal device can obtain the monitoring result by calculating the difference between the two, that is, the monitoring result is the optimal beam RSRP measurement error.

[0175] The following are several examples to illustrate that the terminal device monitors the performance of the first AI model by comparing the output of the first AI model with the output obtained based on non-AI / ML methods.

[0176] In one example, the monitoring result types include beam prediction consistency and optimal beam RSRP measurement error. The process of obtaining the output based on the non-AI / ML method is: the terminal device measures the beams in the first beam set once every period of time to obtain the optimal beam index and the corresponding RSRP. The terminal device compares the output obtained based on the non-AI / ML method with the output of the first AI model to obtain the monitoring result. The output of the first AI model includes the optimal beam index estimated by the terminal device and the corresponding RSRP. The monitoring result includes "beam prediction consistency / inconsistency" and the optimal beam RSRP measurement error.

[0177] In one example, the monitoring result type is beam prediction consistency. The process of obtaining the output based on a non-AI / ML method is as follows: the network device measures the uplink receive beam to determine the optimal downlink transmit beam, and notifies the terminal device of the optimal downlink transmit beam index. The terminal device compares the output obtained based on the non-AI / ML method with the output of the first AI model to obtain the monitoring result. The output of the first AI model is the optimal downlink transmit beam index estimated by the terminal device, and the monitoring result is "beam prediction consistency / inconsistency".

[0178] The manner in which the terminal device determines whether a model failure occurs may refer to the description related to FIG. 5 .

[0179] Several examples are given below to illustrate how a terminal device determines whether a model failure occurs through the physical layer.

[0180] In one example, the terminal device performs multiple performance monitoring on the first AI model. In each performance monitoring, the monitoring result (i.e., beam prediction consistency / inconsistency) is obtained by comparing whether the optimal beam index output by the first AI model is consistent with the reference label. If the monitoring results obtained for N consecutive performance monitorings are all "beam prediction inconsistency" (or not "beam prediction consistency", that is, the above preset value is "beam prediction consistency"), the terminal device determines that the first AI model has a model failure.

[0181] In one example, the terminal device performs multiple performance monitoring on the first AI model. In each performance monitoring, the monitoring result (i.e., beam prediction consistency / inconsistency) is obtained by comparing whether the optimal beam index output by the first AI model is consistent with the reference label. If the proportion of inconsistent beam predictions within a period of time is higher than a preset threshold value (such as 40%), it is determined that a model failure has occurred; or if the proportion of consistent beam predictions within a period of time is lower than a preset threshold value (such as 60%), it is determined that a model failure has occurred.

[0182] The following are several examples to illustrate that the terminal device determines whether a model failure occurs through the physical layer and the upper layer.

[0183] In one example, during each performance monitoring of the first AI model, the terminal device obtains a monitoring result (i.e., beam prediction consistency / inconsistency) by comparing whether the optimal beam index output by the first AI model is consistent with the reference label; if the monitoring result is that the beam prediction is inconsistent, the model failure sample is reported to the MAC layer; at the MAC layer, when the counter used to count model failure samples reaches the specified maximum value, the terminal device determines that a model failure has occurred.

[0184] In one example, during each performance monitoring of the first AI model, the terminal device obtains a monitoring result (i.e., beam prediction consistency / inconsistency) by comparing whether the optimal beam index output by the first AI model is consistent with the reference label; the terminal device counts the proportion of monitoring results that are consistent with the beam prediction within a monitoring period, and if the proportion is lower than a preset threshold value, the model failure sample is reported to the MAC layer; at the MAC layer, when the counter used to count the model failure samples reaches the specified maximum value, the terminal device determines that a model failure has occurred.

[0185] In one example, during each performance monitoring of the first AI model, the terminal device obtains a monitoring result (i.e., the optimal beam RSRP measurement error) by comparing the difference between the optimal RSRP output by the first AI model and the reference tag; if the optimal beam RSRP measurement error is higher than a preset threshold value, the model failure sample is reported to the MAC layer; at the MAC layer, when the counter used to count the model failure samples reaches the specified maximum value, the terminal device determines that a model failure has occurred.

[0186] Step 602: The terminal device sends model failure reporting information to the network device based on whether a model failure occurs.

[0187] Here, for the specific implementation of model failure reporting information, please refer to the description related to Figure 5 above.

[0188] In some implementations, the above-mentioned model failure reporting information may be reported to the network device together with the beam information.

[0189] Step 603: The network device sends model recovery information to the terminal device.

[0190] Here, for the specific implementation of the model recovery information, please refer to the description related to FIG5 above.

[0191] Step 604: The terminal device reports the beam to the network device according to the model recovery information.

[0192] Here, the terminal device reports the beam to the network device based on the model recovery information. You can refer to the description related to Figure 5 above. You only need to understand the "information transmission between the network device" in the description related to Figure 5 as "reporting the beam to the network device (i.e., reporting beam information)".

[0193] In addition, after determining that a model failure has occurred and before beam reporting is performed according to the model recovery information, or after sending the model failure reporting information and before beam reporting is performed according to the model recovery information, or after receiving the model recovery information and before beam information reporting is performed according to the model recovery information, the terminal device may perform beam reporting in one of the following ways:

[0194] Method 1: The terminal device uses a non-AI / ML beam reporting method for beam reporting. In other words, the terminal device performs beam reporting based on traditional methods. For example, the terminal device measures the beams (CSI-RS resources) in the second beam set and reports the optimal beam index and corresponding RSRP.

[0195] Method 2: The terminal device uses a first CSI reporting configuration to perform beam reporting. The first CSI reporting configuration is indicated by model recovery information, or is a CSI reporting configuration pre-agreed between the terminal device and the network device.

[0196] Method 3: The terminal device uses a third AI model for beam reporting. The third AI model is indicated by model recovery information, or is an AI model pre-agreed between the terminal device and the network device.

[0197] Method 4: The terminal device does not report the beam. That is, after a model failure occurs, the terminal stops reporting the beam until a new model is configured and used.

[0198] Application Example 2

[0199] This application example is applied to CSI feedback in a dual-end model scenario. In this scenario, a first AI model is deployed on the terminal device side. The first AI model is used for CSI feedback (or CSI reporting), which can be called AI / ML-based CSI feedback (or CSI reporting). The input and output of the first AI model can refer to the aforementioned description of the neural network model for CSI feedback. Specifically, the terminal device uses the measured channel information (such as feature vectors, spatial domain information, delay information, etc.) as the input of the first AI model, and infers the corresponding CSI quantization bits based on the first AI model; the terminal device reports the inference result to the network device (that is, the terminal device performs CSI reporting). On the network device side, there will be an AI model corresponding to the first AI model. The network device uses the CSI quantization bits reported by the terminal device as the input of its AI model, and can reversely infer the corresponding channel information.

[0200] FIG7 is a flow chart of a third method for recovering from model failure according to an embodiment of the present application. As shown in FIG7 , the method includes all or part of the following:

[0201] Step 701: The terminal device monitors the performance of the first AI model and determines whether a model failure occurs based on the monitoring results.

[0202] Here, the above performance monitoring may be periodic, non-semi-continuous, or non-periodic.

[0203] In some embodiments, before the terminal device performs performance monitoring of the first AI model, the following steps may also be included: the terminal device receives a first signaling sent by the network device.

[0204] In one implementation, the first signaling is used to configure the period of the above-mentioned performance monitoring and / or the period of the terminal device sending the above-mentioned model failure reporting information (i.e., the reporting period of the model failure reporting information). Here, the above-mentioned performance monitoring can be periodic. The reporting of the above-mentioned model failure reporting information can be periodic.

[0205] In another implementation, the first signaling is used to activate the terminal device to perform the above-mentioned performance monitoring and / or activate the terminal device to send the above-mentioned model failure reporting information. Here, the above-mentioned performance monitoring can be semi-continuous or non-periodic. If it is semi-continuous, the first signaling (such as DCI) is used to activate the terminal device to perform periodic performance monitoring; if it is non-periodic, the first signaling (such as DCI) is used to trigger the terminal device to perform one or more continuous performance monitoring. Exemplarily, the network device can trigger the terminal device to perform performance monitoring of the first AI model using the same DCI while triggering the non-periodic CSI reporting through DCI. Here, the reporting of the above-mentioned model failure reporting information can be semi-continuous or non-periodic. If it is semi-continuous, the first signaling (such as DCI) is used to activate the terminal device to perform periodic model failure reporting information reporting; if it is non-periodic, the first signaling (such as DCI) is used to trigger the terminal device to perform one or more continuous model failure reporting information reporting. Exemplarily, the network device may trigger the aperiodic CSI reporting through DCI and, at the same time, use the same DCI to trigger the terminal device to perform performance monitoring of the first AI model and report corresponding model failure reporting information. In one embodiment, the DCI may include CSI request information for triggering aperiodic CSI reporting, and also include 1 bit of indication information for indicating whether the terminal device includes model failure reporting information in the triggered CSI reporting.

[0206] In some embodiments, the first signaling is also used to indicate the type of monitoring results used for performance monitoring and / or a threshold value for model failure, which is used by the terminal device to determine whether a model failure occurs in the first AI model based on the monitoring results.

[0207] Exemplarily, the first signaling may indicate that the monitoring result type used for the above-mentioned performance monitoring is BLER, throughput, CSI error, CSI consistency, SGCS of channel information, GCS of channel information, etc. Exemplarily, the first signaling may carry a CSI reporting configuration, which includes a performance monitoring period, a monitoring result type used for performance monitoring, and a threshold value corresponding to the monitoring result for determining whether a model failure occurs.

[0208] In some embodiments, the first signaling is also used to configure resources for carrying model failure reporting information. The resources for carrying model failure reporting information may be PUCCH resources or PUSCH resources. If the model failure reporting information is reported via PUCCH (i.e., the resources for carrying model failure reporting information are PUCCH resources), the PUCCH resources used by the terminal device may be obtained through the first signaling. If the model failure reporting information is reported via PUSCH (i.e., the resources for carrying model failure reporting information are PUSCH resources), the slot offset of the PUSCH may be obtained through the first signaling.

[0209] The manner in which the terminal device performs performance monitoring may refer to the description related to FIG5 above.

[0210] The following are several examples to illustrate that the terminal device monitors the performance of the first AI model by comparing the output of the first AI model with the true label.

[0211] In one example, the monitoring result type is SGCS, and the true label is the channel information measured by the terminal device. The terminal device can use the corresponding model on the network device side to restore the output of the first AI model, and use the SGCS between the restored channel information and the channel information measured by the terminal device as the monitoring result.

[0212] In one example, the monitoring result type includes channel capacity or throughput, and the true label is expected channel capacity or expected throughput. The terminal device can use the corresponding model on the network device side to restore the output of the first AI model to obtain the corresponding precoding matrix, calculate the corresponding channel capacity or throughput based on the precoding matrix, and compare the calculated channel capacity with the expected channel capacity, or compare the calculated throughput with the expected throughput, as the monitoring result.

[0213] Several examples are given below to illustrate that the terminal device monitors the performance of the first AI model by comparing the output of the first AI model with the output obtained based on non-AI / ML methods.

[0214] In one example, if the monitoring result type is CSI error or CSI consistency, the terminal device may use the first AI model to output RI and CQI, and obtain RI and CQI based on a traditional codebook. The terminal device may compare the RI / CQI output by the first AI model with the RI / CQI obtained by the traditional method, and use the difference between the two as the monitoring result. For example, the monitoring result may be whether two RIs are consistent, or the difference between two CQIs.

[0215] The manner in which the terminal device determines whether a model failure occurs may refer to the description related to FIG. 5 .

[0216] Several examples are given below to illustrate how a terminal device determines whether a model failure occurs through the physical layer.

[0217] In one example, the terminal device performs multiple performance monitorings on the first AI model. In each performance monitoring, the SGCS between the output of the first AI model and the reference label is calculated. If the SGCS obtained from multiple consecutive monitorings is lower than a preset threshold value (such as 0.8), the terminal device determines that a model failure has occurred; or, if the proportion of SGCS above the threshold value within a period of time is lower than a preset value (such as 90%), the terminal device determines that a model failure has occurred; or, if the average value of the SGCS obtained from multiple monitorings is lower than a preset threshold value (such as 0.78), the terminal device determines that a model failure has occurred.

[0218] In one example, the terminal device performs multiple performance monitoring on the first AI model. In each performance monitoring, the difference between the CQI output by the first AI model and the CQI obtained by the traditional method is calculated. If the CQI difference obtained by multiple consecutive monitorings is greater than the preset threshold value, the terminal determines that a model failure has occurred.

[0219] The following are several examples to illustrate that the terminal device determines whether a model failure occurs through the physical layer and the upper layer.

[0220] In one example, during each performance monitoring of the first AI model, the terminal device calculates the SGCS between the output of the first AI model and the reference tag. As long as the monitored SGCS is lower than a preset threshold, the terminal device reports a model failure sample to the MAC layer; alternatively, as long as the average SGCS within a monitoring period is lower than a preset threshold, the terminal device reports a model failure sample to the MAC layer; at the MAC layer, when the counter used to count model failure samples reaches a specified maximum value, the terminal device determines that a model failure has occurred.

[0221] In one example, during each performance monitoring of the first AI model, the terminal device calculates the difference between the channel capacity corresponding to the output of the first AI model and the expected channel capacity. If the difference between the channel capacity corresponding to the output of the first AI model and the expected channel capacity is higher than a preset threshold value (such as 1dB), the model failure sample is reported to the MAC layer; at the MAC layer, when the counter used to count the model failure samples reaches the specified maximum value, the terminal device determines that a model failure has occurred.

[0222] Step 702: The terminal device sends model failure reporting information to the network device based on whether a model failure occurs.

[0223] Here, for the specific implementation of model failure reporting information, please refer to the description related to Figure 5 above.

[0224] In some embodiments, the above-mentioned model failure reporting information can be reported to the network device together with the CSI feedback information obtained by the terminal device (such as the CSI feedback information obtained based on the first AI model). If the model failure reporting information is reported together with the CSI feedback information, the two need to be multiplexed and reported. Furthermore, when CSI omission is required, the priority of the model failure reporting information is higher than the CSI feedback information. That is, when the code rate of the PUSCH / PUCCH carrying the CSI is higher than the configured value and the CSI needs to be discarded, the CSI feedback information is discarded first, and then the model failure reporting information is discarded. The advantage of doing this is that the reporting of the model failure reporting information can be guaranteed as a priority. If a model failure occurs, the reported CSI feedback information will become unreliable. At this time, the importance of the CSI feedback information will be lower than the model failure reporting information.

[0225] Step 703: The network device sends model recovery information to the terminal device.

[0226] Here, for the specific implementation of the model recovery information, please refer to the description related to FIG5 above.

[0227] It should be noted that for models deployed on both ends, such as models used for CSI reporting, if a model on the network device side corresponds to models on multiple terminal device sides, the network device can select a suitable model for each terminal device based on the monitoring results of multiple second AI models reported by these multiple terminal devices, ensuring that the network device can still correspond to models on multiple terminal device sides through the same model, thereby reducing complexity.

[0228] As an implementation, network devices can perform online training based on collected data, determine a new available model, and indicate this to the terminal device via model recovery information. This online training can be performed based on the existing model, thereby updating the original model to ensure performance.

[0229] Step 704: The terminal device reports CSI to the network device based on the model recovery information.

[0230] Here, the terminal device reports CSI to the network device based on the model recovery information. You can refer to the description related to Figure 5 above. You only need to understand "transmitting information with the network device" in the description related to Figure 5 as "reporting CSI to the network device".

[0231] As a case, when the model recovery information indicates that the terminal device uses the updated target AI model instead of the first AI model for CSI reporting, the terminal device uses the updated target AI model to report CSI to the network device; accordingly, the network device uses the network side model corresponding to the updated target AI model to decode the CSI report.

[0232] As another case, when the model recovery information indicates that the terminal device uses a non-AI / ML method to report CSI, the terminal device uses a non-AI / ML method to report CSI to the network device; accordingly, the network device uses a traditional method to receive the CSI report from the terminal device.

[0233] In addition, after determining that a model failure has occurred and before beam reporting is performed according to the model recovery information, or after sending the model failure reporting information and before beam reporting is performed according to the model recovery information, or after receiving the model recovery information and before beam information reporting is performed according to the model recovery information, the terminal device may report CSI in one of the following ways:

[0234] Method 1: The terminal device uses a non-AI / ML beamforming reporting method to report CSI. In other words, the terminal device reports CSI based on a traditional method. For example, the terminal device uses a first codebook type for CSI reporting. The first codebook type is indicated by the model recovery information, or is a codebook type pre-agreed between the terminal device and the network device. For example, the first codebook type can be a Type 1 codebook, a Type 2 codebook, or an enhanced Type 2 codebook specified in the protocol. The configuration of the first codebook type can also be indicated by the model recovery information.

[0235] Method 2: The terminal device reports CSI using a first CSI reporting configuration, where the first CSI reporting configuration is indicated by model recovery information or is a CSI reporting configuration pre-agreed between the terminal device and the network device. For example, the model recovery information may indicate a CSI reporting configuration ID, or the terminal device and the network device may pre-agreed on a CSI reporting configuration ID.

[0236] Method 3: The terminal device uses a third AI model to report CSI. The third AI model is indicated by model recovery information or is an AI model pre-agreed between the terminal device and the network device.

[0237] Method 4: The terminal device does not report CSI. That is, after a model failure occurs, the terminal stops reporting CSI until a new model is configured and used.

[0238] Application Example 3

[0239] This application example can be applied to beam management in a single-end model scenario. In this scenario, a first AI model is deployed on the terminal device side. This first AI model is used for beam management (or beam reporting), which can be called AI / ML-based beam management (or beam reporting).

[0240] The terminal device uses the measured RSRP of multiple beams in the second beam set as the input of the first AI model, and obtains the optimal beam index and corresponding RSRP in the first beam set based on the inference of the first AI model. The terminal device reports the inference result to the network device (that is, the terminal device performs beam reporting). Among them, the second beam set can be understood as a measurement set, and the first beam set can be understood as a prediction set. Whether it is a second beam set or a second beam set, the beam set can contain multiple beams, and the beam here can be replaced by a downlink reference signal, such as a channel state information-reference signal (Channel State Information-Reference Signal, CSI-RS); accordingly, the beam index can be replaced by a downlink reference signal resource index, such as CRI.

[0241] FIG8 is a fourth flow chart of a model failure recovery method provided in an embodiment of the present application. As shown in FIG8 , the method includes all or part of the following contents:

[0242] Step 801: The terminal device receives RRC signaling sent by the network device, where the RRC signaling is used to configure the reporting period and PUCCH resources used by the terminal device to send model failure reporting information.

[0243] Here, the model failure reporting information is periodically reported via the PUCCH.

[0244] Step 802: The terminal device periodically monitors the performance of the first AI model. The performance monitoring process includes: 1) the terminal device measures N (N>1) beams in the second beam set to obtain the RSRP of the N beams; 2) the terminal device inputs the RSRP of the N beams into the first AI model, and outputs the optimal beam index and corresponding RSRP in the first beam set through reasoning of the first AI model; 3) the terminal device compares the optimal beam index output by the first AI model with the AI ​​reference tag to determine whether they are consistent, and obtains the monitoring result of this performance monitoring, that is, the monitoring result is "consistent" or "inconsistent".

[0245] Here, the AI ​​reference tag is obtained by the terminal device through a traditional beam measurement method. Specifically, the terminal device measures M (M>1) beams in the first beam set to obtain the RSRP of the M beams; the beam index corresponding to the beam with the best RSRP among the M beams is used as the AI ​​reference tag. Here, the performance monitoring period can be the same as the reporting period of the model failure reporting information. Alternatively, the reporting period of the model failure reporting information is an integer multiple of the performance monitoring period.

[0246] Step 803: During each performance monitoring process, if the monitoring result is "inconsistent", the physical layer of the terminal device reports a model failure sample to the MAC layer; after the MAC layer of the terminal device receives the model failure sample reported by the physical layer, it adds 1 to the value of the counter. When the value of the counter reaches the specified maximum value (such as 10), the terminal device determines that a model failure has occurred in the first AI model.

[0247] Here, the MAC layer of the terminal device has a dedicated counter for counting the number of model failure samples.

[0248] Step 804: After the first AI model fails, the terminal device performs performance monitoring on P (P≥1) candidate AI models. The performance monitoring process can refer to the above step 802.

[0249] Here, P candidate AI models can be pre-configured to the terminal device by the network device. For example, each candidate AI model can correspond to a CSI reporting configuration, and the network device indicates the candidate AI model to the terminal device through the CSI reporting configuration.

[0250] Step 805: The terminal device selects L (L≥1) AI models with “consistent” monitoring results from the P candidate AI models based on the monitoring results of the P candidate AI models, which are referred to as L second AI models.

[0251] Step 806: The terminal device sends model failure reporting information to the network device according to the reporting period and PUCCH resources configured by the RRC signaling. The model failure reporting information includes information of L second AI models.

[0252] In some embodiments, the model failure reporting information may also include 1-bit indication information, which is used to indicate whether a model failure has occurred. For example, a 1-bit value of 1 indicates that a model failure has occurred, and a value of 0 indicates that a model failure has not occurred. Or, for example, a 1-bit value of 0 indicates that a model failure has occurred, and a value of 1 indicates that a model failure has not occurred.

[0253] In some embodiments, the model failure reporting information may also include the RSRP corresponding to the optimal beam index inferred by each second AI model in the L second AI models, which can be used by the network device to select a suitable AI model from the L second AI models to replace or update the first AI model.

[0254] Here, the information of the second AI model may be a model index or a model ID of the second AI model, where the model index refers to an index corresponding to the second AI model in the candidate AI models.

[0255] Step 807: The network device receives the model failure reporting information sent by the terminal device, and sends model recovery information to the terminal device through RRC signaling. The model recovery information is used to indicate an AI model from the L second AI models reported by the terminal device as the target AI model for subsequent beam reporting by the terminal device.

[0256] Here, the model recovery information includes the model index or model ID of the target AI model, where the model index refers to the corresponding index of the target AI model in the candidate AI models.

[0257] As a case, if the terminal device reports multiple second AI models (i.e., L>1), the network device can select an appropriate AI model for the terminal device as the target AI model for subsequent beam reporting by the terminal device based on the RSRP corresponding to the optimal beam index inferred by each second AI model contained in the model failure reporting information.

[0258] As another case, if the terminal device reports a second AI model (i.e., L=1), the network device can directly instruct the terminal device through the model recovery information whether to use the reported second AI model as the target AI model for subsequent terminal devices to perform beam reporting.

[0259] Step 808: The terminal device receives the model recovery information sent by the network device, and performs beam reporting according to the target AI model indicated by the model recovery information. The beam reporting process includes: 1) the terminal device measures N (N>1) beams in the second beam set to obtain the RSRP of the N beams; 2) the terminal device inputs the RSRP of the N beams into the target AI model, and outputs the optimal beam index and corresponding RSRP in the first beam set through reasoning of the target AI model; 3) the terminal device reports the optimal beam index and corresponding RSRP obtained by reasoning of the target AI model to the network device.

[0260] After step 804 and before step 808, the terminal device may perform beam reporting in a non-AI / ML manner, that is, the terminal device performs beam reporting based on a traditional manner. The process of performing beam reporting in a non-AI / ML manner (or a traditional manner) includes: 1) the terminal device measures N (N>1) beams in the second beam set to obtain the RSRP of the N beams; 2) the terminal device reports the beam with the best RSRP among the N beams and the corresponding RSRP to the network device.

[0261] Step 809: The network device receives the beam report sent by the terminal device.

[0262] The method for determining model failure described in step 803 above may also have other alternatives. For example, in multiple consecutive performance monitoring processes, if the proportion of monitoring results that are "inconsistent" exceeds a specified proportion threshold (such as 30%), the terminal device determines that the first AI model has a model failure. Exemplarily, the terminal device performs 8 consecutive performance monitoring processes, of which 3 detection results are "inconsistent", then the proportion of monitoring results that are "inconsistent" is 3 / 8 = 37.5%, which exceeds the proportion threshold of 30%, and the terminal device determines that the first AI model has a model failure.

[0263] The performance monitoring process described in step 802 above may also have other alternatives, and other alternatives may refer to the descriptions related to Figures 5 and 6. As an example, other alternatives may be as follows, and the performance monitoring process includes: 1) the terminal device measures N (N>1) beams in the second beam set to obtain the RSRPs of the N beams; 2) the terminal device inputs the RSRPs of the N beams into the first AI model, and outputs the optimal beam index and corresponding RSRP in the first beam set through reasoning of the first AI model; 3) the terminal device compares the RSRP corresponding to the optimal beam index output by the first AI model with the difference between the AI ​​reference tag, and obtains the monitoring result of this performance monitoring, that is, the monitoring result is "optimal RSRP error".

[0264] Here, the AI ​​reference tag is obtained by the terminal device through a traditional beam measurement method. Specifically, the terminal device measures M (M>1) beams in the first beam set to obtain the RSRP of the M beams; and the optimal RSRP among the M beams is used as the AI ​​reference tag.

[0265] Accordingly, the method for determining model failure described in step 803 above may also have other alternatives, and other alternatives may refer to the descriptions related to Figures 5 and 6. As an example, other alternatives may be as follows: during each performance monitoring process, if the monitoring result "optimal RSRP error" is greater than or equal to the set threshold value, the physical layer of the terminal device reports a model failure sample to the MAC layer; after the MAC layer of the terminal device receives the model failure sample reported by the physical layer, it adds 1 to the value of the counter. When the value of the counter reaches the specified maximum value (such as 10), the terminal device determines that the first AI model has a model failure.

[0266] The PUCCH resources described in the above steps 801 and 806 may also have other alternatives, such as PUSCH resources.

[0267] It should be noted that, in this application example, the terminal device periodically performs performance monitoring of the first AI model and reports model failure reporting information as an example, and there may be other alternative schemes, for example: the terminal device semi-continuously or non-periodically performs performance monitoring of the first AI model and reports model failure reporting information. For semi-continuous or non-periodic situations, step 801 may have other alternative schemes, for example: for the semi-continuous situation, the terminal device receives MAC signaling sent by the network device, and the MAC signaling is used to activate the terminal device to perform periodic performance monitoring and reporting of model failure reporting information; for example, for the non-periodic situation, the terminal device receives DCI sent by the network device, and the DCI is used to trigger the terminal device to perform one or more continuous performance monitoring and reporting of model failure reporting information.

[0268] Application Example 4

[0269] This application example is applied to CSI feedback in a dual-end model scenario. In this scenario, a first AI model is deployed on the terminal device side. This first AI model is used for CSI feedback (or CSI reporting), which can be called AI / ML-based CSI feedback (or CSI reporting).

[0270] The terminal device uses the measured channel information (such as channel eigenvectors, channel spatial domain information, channel delay information, etc.) as the input of the first AI model, and infers the corresponding CSI quantization bits (such as PMI quantization bits) based on the first AI model; the terminal device reports the inference results to the network device (i.e., the terminal device reports CSI). On the network device side, there will be an AI model corresponding to the first AI model. The network device uses the CSI quantization bits reported by the terminal device as the input of its AI model, and can reversely infer the corresponding channel information.

[0271] FIG9 is a fifth flow chart of a model failure recovery method provided in an embodiment of the present application. As shown in FIG9 , the method includes all or part of the following contents:

[0272] Step 901: The terminal device receives the DCI sent by the network device, where the DCI is used to trigger the terminal device to perform model monitoring of the first AI model and report corresponding model failure reporting information.

[0273] Here, the DCI may also include CSI request information for triggering CSI reporting. In addition, it may also include 1-bit indication information, which is used to indicate whether the terminal device includes model failure reporting information in the CSI reporting triggered by the DCI. For example, a 1-bit value of 1 indicates that the model failure reporting information is included in the CSI reporting triggered by the DCI, or, for example, a 1-bit value of 0 indicates that the model failure reporting information is included in the CSI reporting triggered by the DCI.

[0274] In some implementations, the DCI may further indicate a time interval between the DCI and the PUSCH carrying the model failure reporting information.

[0275] Step 902: The terminal device performs performance monitoring on the first AI model. The performance monitoring process includes: 1) the terminal device performs CSI measurement to obtain corresponding channel information (such as channel eigenvectors, channel spatial domain information, channel delay information, etc.). The measured channel information is the channel information obtained by the non-AI / ML method; 2) the terminal device inputs the measured channel information into the first AI model, and performs encoding processing through the first AI model to obtain CSI quantized bits (such as PMI quantized bits); 3) the terminal device inputs the CSI quantized bits obtained by the first AI model into a third AI model, which is a decoding model corresponding to the first AI model, and obtains estimated channel information through decoding processing by the third AI model; 4) the terminal device compares the estimated channel information with the channel information obtained by the non-AI / ML method to obtain a monitoring result of this performance monitoring, which is "SGCS of the estimated channel information and the channel information obtained by the non-AI / ML method".

[0276] Step 903: If the SGCS obtained from the performance monitoring for H consecutive times (H≥1) is lower than a preset threshold value (such as 0.8), the terminal device determines that the first AI model has a model failure.

[0277] Here, H times of performance monitoring may be the performance monitoring performed between the terminal device receiving the above-mentioned DCI and sending the corresponding model failure reporting information.

[0278] Step 904: After the first AI model fails, the terminal device performs online training on the first AI model to obtain an optimized AI model, which is called the second AI model.

[0279] Here, the terminal device can perform online training on the first AI model based on the most recently collected training data, thereby obtaining a new set of model parameters. These new model parameters are the model parameters of the second AI model. Since the second AI model is obtained based on the online training of the first AI model, the second AI model and the first AI model have the same model structure but different model parameters.

[0280] Step 905: The terminal device sends model failure reporting information to the network device through PUSCH according to the above-mentioned DCI. The model failure reporting information includes indication information 1, which is used to indicate whether the terminal device has confirmed the second AI model (through online training).

[0281] Here, the indication information 1 may be 1-bit indication information, used to indicate whether the terminal device has confirmed the second AI model (through online training), or whether the terminal device has obtained a usable second AI model (through online training). For example, a 1-bit value of 1 indicates that the terminal device has confirmed the second AI model or has obtained a usable second AI model, while a 1-bit value of 0 indicates that the terminal device has not confirmed the second AI model or has not obtained a usable second AI model.

[0282] In some embodiments, the model failure reporting information may further include 1-bit indication information 2, where the indication information 2 is used to indicate whether a model failure has occurred. For example, a 1-bit value of 1 indicates that a model failure has occurred, and a value of 0 indicates that no model failure has occurred.

[0283] In some implementations, the above-mentioned model failure reporting information may be reported to the network device together with the CSI quantization bits (ie, CSI feedback information) obtained based on the first AI model.

[0284] Step 906: The network device receives the model failure reporting information sent by the terminal device, and sends the model recovery information to the terminal device via DCI.

[0285] In step 906, if the model failure reporting information indicates that the terminal device has confirmed the second AI model, the network device may instruct the terminal device, through the model recovery information, whether to use the confirmed second AI model instead of the first AI model for subsequent CSI reporting. If the model failure reporting information indicates that the terminal device has not confirmed the second AI model, the network device may, through the model recovery information, indicate an AI model from candidate AI models pre-configured for the terminal device for subsequent CSI reporting, or may instruct the terminal device, through the model recovery information, to use a non-AI / ML method for CSI reporting.

[0286] Step 907: The terminal device receives the model recovery information sent by the network device, and reports the CSI according to the model recovery information.

[0287] In step 907, if the model recovery information indicates that the terminal device uses the second AI model instead of the first AI model for subsequent CSI reporting, the terminal device uses the second AI model for CSI reporting. If the model recovery information indicates that the terminal device uses a non-AI / ML method for CSI reporting, the terminal device uses a non-AI / ML method for CSI reporting.

[0288] In some embodiments, for the above situation 1), the terminal device reports the model application time to the network device, for example, by reporting the model application time to the network device through UE capabilities. The terminal device determines the moment when the second AI model starts to be applied based on the model application time, and adopts the second AI model to report CSI after that moment. The network device can update the model synchronously with the terminal device based on the model application time, thereby ensuring that the AI ​​model adopted on the network device side matches the second AI model on the terminal device side. In this way, when the AI ​​model on the network device side decodes the CSI quantization bits (such as PMI quantization bits) from the second AI model, more accurate channel information can be restored.

[0289] Here, the process of the terminal device using the second AI model to report CSI includes: 1) the terminal device performs CSI measurement to obtain corresponding channel information (such as channel eigenvectors, channel spatial domain information, channel delay information, etc.); 2) the terminal device inputs the measured channel information into the second AI model, and encodes it through the second AI model to obtain CSI quantization bits (such as PMI quantization bits); 3) the terminal device reports the CSI quantization bits obtained by the second AI model to the network device.

[0290] Here, the process of the terminal device reporting CSI using a non-AI / ML method includes: 1) the terminal device performs CSI measurement to obtain the corresponding channel information (such as channel eigenvectors, channel spatial domain information, channel delay information, etc.); 2) based on the channel information and a pre-agreed codebook, the corresponding CSI information (such as RI / PMI / CQI) is obtained; 3) the terminal device reports the CSI information (such as RI / PMI / CQI) to the network device.

[0291] After step 903 and before step 907, the terminal device does not report CSI. That is, after the first AI model fails, if the terminal device does not receive model recovery information sent by the network device, the terminal device stops reporting CSI until the terminal device receives model recovery information sent by the network device, and the model recovery information indicates that the second AI model should be used instead of the first AI model for subsequent CSI reporting or indicates that a non-AI / ML method should be used for CSI reporting.

[0292] Step 908: The network device receives the CSI report made by the terminal device based on the above model recovery information.

[0293] In step 908, if the model recovery information indicates that the terminal device uses the second AI model instead of the first AI model for subsequent CSI reporting, the network device uses the network device-side AI model corresponding to the second AI model to decode the received CSI information. If the model recovery information indicates that the terminal device uses a non-AI / ML method for CSI reporting, the network device uses a traditional method to receive the CSI information sent by the terminal device.

[0294] The method for determining model failure described in step 903 above may also have other alternatives, and other alternatives may refer to the descriptions related to Figures 5 and 7. As an example, another alternative may be as follows: if the ratio of SGCS lower than a preset first threshold value (such as 0.8) in H consecutive performance monitorings is higher than a preset second threshold value (such as 0.4), the terminal device determines that the first AI model has a model failure. For example: in H performance monitorings, there are F times when the monitored SGCS is lower than the first threshold value (such as 0.8). Then, if the ratio F / H is higher than the preset second threshold value (such as 0.4), it is determined that a model failure has occurred. As an example, another alternative may be as follows: during each performance monitoring process, if the monitored SGCS is lower than the preset threshold value (such as 0.8), the physical layer of the terminal device reports a model failure sample to the MAC layer; after the MAC layer of the terminal device receives the model failure sample reported by the physical layer, it adds 1 to the counter value. When the counter value reaches the specified maximum value (such as 10), the terminal device determines that the first AI model has a model failure. Here, the MAC layer of the terminal device has a dedicated counter for counting the number of model failure samples.

[0295] The method for obtaining the second AI model in step 904 above may also have other alternatives, which can be referred to in the description related to Figures 5 and 6. As an example, another alternative may be as follows: based on the monitoring results of P candidate AI models, the terminal device selects L (L ≥ 1) AI models from the P candidate AI models whose monitoring results SGCS are greater than or equal to a preset threshold value (e.g., 0.8), which are referred to as L second AI models.

[0296] Accordingly, the content indicated by the model failure reporting information in the above step 905 may also have other alternative solutions, for example, the model failure reporting information includes information of L second AI models.

[0297] Accordingly, the content indicated by the model recovery information in the above step 906 may also have other replacement schemes, for example, the model recovery information is used to indicate an AI model from the L second AI models reported by the terminal device as the target AI model for subsequent CSI reporting by the terminal device.

[0298] It should be noted that in this application example, the terminal device performs non-periodic performance monitoring of the first AI model and reporting of model failure reporting information as an example. There may be other alternatives, for example: the terminal device semi-continuously or periodically performs performance monitoring of the first AI model and reporting of model failure reporting information. For semi-continuous or periodic situations, step 901 may have other alternatives, for example: for the semi-continuous situation, the terminal device receives MAC signaling sent by the network device, and the MAC signaling is used to activate the terminal device to perform periodic performance monitoring and reporting of model failure reporting information; for example, for the periodic situation, the terminal device receives RRC signaling sent by the network device, and the RRC signaling is used to configure the reporting period and PUCCH or PUSCH resources used by the terminal device to send model failure reporting information.

[0299] The preferred embodiments of the present application are described in detail above in conjunction with the accompanying drawings. However, the present application is not limited to the specific details in the above embodiments. Within the technical concept of the present application, the technical solution of the present application can be subjected to a variety of simple modifications, and these simple modifications all fall within the scope of protection of the present application. For example, the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present application will no longer describe the various possible combinations separately. For another example, the various different embodiments of the present application can also be arbitrarily combined, as long as they do not violate the idea of ​​the present application, they should also be regarded as the contents disclosed in the present application. For another example, under the premise of no conflict, the various embodiments and / or the technical features in each embodiment described in the present application can be arbitrarily combined with the prior art, and the technical solution obtained after the combination should also fall within the scope of protection of the present application.

[0300] It should also be understood that in the various method embodiments of the present application, the sequence numbers of the above-mentioned processes do not imply a precedence in the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. In addition, in the embodiments of the present application, the terms "downlink," "uplink," and "sidelink" are used to indicate the transmission direction of signals or data, where "downlink" is used to indicate the first direction of transmission of signals or data from a site to a user equipment in a cell, "uplink" is used to indicate the second direction of transmission of signals or data from a user equipment in a cell to a site, and "sidelink" is used to indicate the third direction of transmission of signals or data from user equipment 1 to user equipment 2. For example, "downlink signal" indicates that the transmission direction of the signal is the first direction. In addition, in the embodiments of the present application, the term "and / or" is merely a description of the association relationship between associated objects, indicating that three relationships can exist. Specifically, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0301] FIG10 is a schematic diagram of the structure of a model failure recovery device provided in an embodiment of the present application. The device is applied to a terminal device. As shown in FIG10 , the device includes:

[0302] A monitoring unit 1001 is configured to monitor the performance of a first AI model;

[0303] A sending unit 1002 is configured to send model failure reporting information to a network device based on a monitoring result, where the model failure reporting information is used to indicate that a model failure has occurred in the first AI model, or to indicate whether a model failure has occurred in the first AI model;

[0304] The receiving unit 1003 is configured to receive the model recovery information sent by the network device;

[0305] The sending unit 1002 is further configured to transmit information to and from a network device based on the model recovery information.

[0306] In some embodiments, the receiving unit 1003 is used to receive a first signaling sent by the network device, where the first signaling is used to configure a period of the performance monitoring, and / or the first signaling is used to activate the terminal device to perform the performance monitoring.

[0307] In some embodiments, the first signaling is used to configure the period of the performance monitoring and / or the period of the terminal device sending the model failure reporting information; or, the first signaling is used to activate the terminal device to perform the performance monitoring and / or activate the terminal device to send the model failure reporting information.

[0308] In some embodiments, the monitoring unit 1001 is used to monitor the performance of the first AI model by comparing the output of the first AI model with a reference label; the reference label is a real label corresponding to the output of the first AI model, or the reference label is an output obtained based on a non-AI / ML method.

[0309] In some embodiments, the monitoring unit 1001 is used to determine that a model failure has occurred in the first AI model if, in multiple performance monitoring operations for the first AI model, the monitoring results are all worse than a threshold value, or the monitoring results are all different from a preset value, or the statistical value of the monitoring results is worse than a threshold value.

[0310] In some embodiments, the monitoring unit 1001 is configured to report model failure samples to a higher layer if, in each performance monitoring of the first AI model, the monitoring result is worse than a threshold value or is different from a preset value; and if the number of model failure samples counted by the higher layer reaches a specified maximum value, it is determined that a model failure has occurred.

[0311] In some embodiments, the sending unit 1002 is configured to send model failure reporting information to a network device if a monitoring result indicates that a model failure has occurred in the first AI model, where the model failure reporting information is used to indicate that a model failure has occurred in the first AI model; or, to send model failure reporting information to a network device, where the model failure information is used to indicate whether a model failure has occurred in the first AI model.

[0312] In some implementations, the sending of the model failure reporting information is triggered by an event, or is triggered by a trigger signaling sent by a network device, or is periodically sent through an uplink control channel.

[0313] In some embodiments, the model failure reporting information includes first indication information, where the first indication information is used to indicate one or more second AI models, or to indicate whether the terminal device confirms the second AI model; the second AI model is an AI model recommended by the terminal device after performance monitoring of the candidate AI model, or is an AI model obtained by the terminal device after online training of the first AI model.

[0314] In some embodiments, the candidate AI model is configured by the network device to the terminal device, or is obtained by pre-training a basic AI model by the terminal device.

[0315] In some embodiments, the first indication information is further used to indicate the monitoring results corresponding to each second AI model.

[0316] In some embodiments, if there is no AI model among the candidate AI models whose monitoring results meet the preset conditions, the first indication information is used to indicate that there is no recommended model.

[0317] In some embodiments, the model recovery information is used to indicate whether the terminal device uses the second AI model reported by the terminal device through the above-mentioned model failure reporting information to replace the first AI model for information transmission between the terminal device and the network device; or, the model recovery information is used to indicate an AI model from multiple second AI models reported by the terminal device through the above-mentioned model failure reporting information to replace the first AI model for information transmission between the terminal device and the network device; or, the model recovery information is used to indicate whether the terminal device uses the second AI model confirmed by the terminal device to replace the first AI model for information transmission between the terminal device and the network device; or, the model recovery information is used to indicate an AI model from candidate AI models to update the first AI model for information transmission between the terminal device and the network device; or, the model recovery information is used to instruct the terminal device to use a non-AI / ML method to transmit information between the terminal device and the network device.

[0318] In some embodiments, the sending unit 1002 is configured to replace the first AI model with the target AI model indicated by the model recovery information, and use the target AI model to transmit information with the network device; or, update the model structure of the first AI model with the target model structure indicated by the model recovery information, and / or update the model parameters of the first AI model with the target model parameters indicated by the model recovery information to obtain the target AI model, and use the target AI model to transmit information with the network device; or, use the non-AI / machine learning ML method indicated by the model recovery information to transmit information with the network device.

[0319] In some embodiments, the apparatus further includes: a determining unit, configured to determine, based on a model application time, a time at which the target AI model indicated by the model recovery information starts to be applied; the model application time is reported by the terminal device to the network device via UE capabilities, or configured by the network device to the terminal device;

[0320] The sending unit 1002 is used to use the target AI model to transmit information between the network device after the moment.

[0321] In some embodiments, the sending unit 1002 is configured to, after determining that a model failure has occurred, after sending the model failure reporting information, or after receiving the model recovery information, and before transmitting information with the network device according to the model recovery information, transmit information with the network device in one of the following ways:

[0322] Use non-AI / ML methods to transmit information to and from network devices; or

[0323] When the first AI model is used for channel state information CSI reporting or beam reporting, a first CSI reporting configuration is used for CSI reporting or beam reporting, where the first CSI reporting configuration is indicated by the model recovery information or is a CSI reporting configuration pre-agreed between the terminal device and the network device; or

[0324] using a third AI model to transmit information to and from the network device, the third AI model being indicated by the model recovery information, or being an AI model pre-agreed between the terminal device and the network device; or

[0325] No information is transmitted to or from network devices.

[0326] In some embodiments, the sending unit 1002 is configured to, when the first AI model is used for CSI reporting, use a first codebook type for CSI reporting, where the first codebook type is indicated by the model recovery information, or is a codebook type pre-agreed between the terminal device and the network device; or, when the first AI model is used for beam reporting, use a non-AI / ML beam reporting method for beam reporting.

[0327] Those skilled in the art should understand that the relevant description of the above-mentioned model failure recovery device in the embodiment of the present application can be understood with reference to the relevant description of the model failure recovery method in the embodiment of the present application.

[0328] FIG11 is a second schematic diagram of the structure of a model failure recovery device provided in an embodiment of the present application, which is applied to a network device. As shown in FIG11 , the device includes:

[0329] A receiving unit 1101 is configured to receive model failure reporting information sent by a terminal device, where the model failure reporting information is used to indicate that a model failure has occurred in the first AI model, or to indicate whether a model failure has occurred in the first AI model;

[0330] The sending unit 1102 is configured to send model recovery information to the terminal device, where the model recovery information is used for information transmission between the terminal device and the network device.

[0331] In some embodiments, the sending unit 1102 is used to send a first signaling to the terminal device, wherein the first signaling is used to configure the period of the performance monitoring and / or the period of the terminal device sending the model failure reporting information; or, the first signaling is used to activate the terminal device to perform the performance monitoring and / or activate the terminal device to send the model failure reporting information.

[0332] In some embodiments, the first signaling is further used to indicate the type of monitoring results used for the performance monitoring and / or the threshold value for model failure, and the threshold value is used by the terminal device to determine whether the first AI model has a model failure based on the monitoring results.

[0333] In some implementations, the sending of the model failure reporting information is triggered by an event, or is triggered by a trigger signaling sent by a network device, or is periodically sent through an uplink control channel.

[0334] In some embodiments, the event includes the terminal device detecting that a model failure of the first AI model occurs.

[0335] In some embodiments, the model failure reporting information includes first indication information, which is used to indicate one or more second AI models, or to indicate whether the terminal device confirms the second AI model; the second AI model is a model recommended by the terminal device after performance monitoring of the candidate AI model, or is an AI model obtained by the terminal device after online training of the first AI model.

[0336] In some embodiments, the candidate AI model is configured by the network device to the terminal device, or is obtained by pre-training a basic AI model by the terminal device.

[0337] In some embodiments, the first indication information is further used to indicate the monitoring results corresponding to each second AI model.

[0338] In some embodiments, if there is no AI model among the candidate AI models whose monitoring results meet the preset conditions, the first indication information indicates that there is no recommended model.

[0339] In some embodiments, the model recovery information is used to indicate whether the terminal device uses the second AI model reported by the terminal device through the above-mentioned model failure reporting information to replace the first AI model for information transmission between the terminal device and the network device; or, the model recovery information is used to indicate an AI model from multiple second AI models reported by the terminal device through the above-mentioned model failure reporting information to replace the first AI model for information transmission between the terminal device and the network device; or, the model recovery information is used to indicate whether the terminal device uses the second AI model confirmed by the terminal device to replace the first AI model for information transmission between the terminal device and the network device; or, the model recovery information is used to indicate an AI model from candidate AI models to update the first AI model for information transmission between the terminal device and the network device; or, the model recovery information is used to instruct the terminal device to use a non-AI / ML method to transmit information between the terminal device and the network device.

[0340] In some embodiments, the receiving unit 1101 is used to receive the model application time reported by the terminal device; or the sending unit 1102 is used to configure the model application time for the terminal device, and the model application time is used by the terminal device to determine the moment when the target AI model indicated by the model recovery information starts to be applied.

[0341] In some embodiments, the receiving unit 1101 is configured to receive information sent by the terminal device using the target AI model indicated by the model recovery information;

[0342] The device also includes: a processing unit 1103, configured to process the information using a network-side model corresponding to the target AI model.

[0343] In some embodiments, the receiving unit 1101 is configured to, after receiving the model failure reporting information or after sending the model recovery information, and before receiving information transmitted by the terminal device according to the model recovery information, receive information sent by the terminal device in one of the following ways:

[0344] Receiving information sent by the terminal device using a non-AI / ML method; or

[0345] When the first AI model is used for CSI reporting or beam reporting, receiving a CSI report or beam reporting sent by the terminal device using a first CSI reporting configuration, where the first CSI reporting configuration is indicated by the model recovery information or is a CSI reporting configuration pre-agreed between the terminal device and the network device; or

[0346] Receive information sent by the terminal device using a third AI model, where the third AI model is indicated by the model recovery information or is an AI model pre-agreed between the terminal device and the network device.

[0347] In some embodiments, the receiving unit 1101 is configured to, when the first AI model is used for CSI reporting, receive a CSI report sent by the terminal device using a first codebook type, where the first codebook type is indicated by the model recovery information or is a codebook type pre-agreed between the terminal device and the network device; or, when the first AI model is used for beam reporting, receive a beam report sent by the terminal device using a non-AI / ML beam reporting method.

[0348] Those skilled in the art should understand that the relevant description of the above-mentioned model failure recovery device in the embodiment of the present application can be understood with reference to the relevant description of the model failure recovery method in the embodiment of the present application.

[0349] Figure 12 is a schematic diagram of a communication device 1200 provided in an embodiment of the present application. The communication device can be a terminal device or a network device. The communication device 1200 shown in Figure 12 includes a processor 1210, which can call and execute a computer program from a memory to implement the method in the embodiment of the present application.

[0350] In some implementations, as shown in FIG12 , the communication device 1200 may further include a memory 1220. The processor 1210 may call and execute a computer program from the memory 1220 to implement the method in the embodiment of the present application.

[0351] The memory 1220 may be a separate device independent of the processor 1210 , or may be integrated into the processor 1210 .

[0352] In some embodiments, as shown in FIG12 , the communication device 1200 may further include a transceiver 1230 , and the processor 1210 may control the transceiver 1230 to communicate with other devices. Specifically, the transceiver 1230 may send information or data to other devices, or receive information or data sent by other devices.

[0353] The transceiver 1230 may include a transmitter and a receiver. The transceiver 1230 may further include an antenna, and the number of antennas may be one or more.

[0354] In some embodiments, the communication device 1200 may specifically be a network device of an embodiment of the present application, and the communication device 1200 may implement the corresponding processes implemented by the network device in each method of the embodiment of the present application. For the sake of brevity, they will not be repeated here.

[0355] In some embodiments, the communication device 1200 may specifically be a mobile terminal / terminal device of an embodiment of the present application, and the communication device 1200 may implement the corresponding processes implemented by the mobile terminal / terminal device in each method of the embodiment of the present application. For the sake of brevity, they will not be repeated here.

[0356] Figure 13 is a schematic structural diagram of a chip according to an embodiment of the present application. The chip 1300 shown in Figure 13 includes a processor 1310, which can call and run a computer program from a memory to implement the method according to the embodiment of the present application.

[0357] In some embodiments, as shown in FIG13 , the chip 1300 may further include a memory 1320 , wherein the processor 1310 may call and execute a computer program from the memory 1320 to implement the method in the embodiment of the present application.

[0358] The memory 1320 may be a separate device independent of the processor 1310 , or may be integrated into the processor 1310 .

[0359] In some embodiments, the chip 1300 may further include an input interface 1330. The processor 1310 may control the input interface 1330 to communicate with other devices or chips, and specifically, may obtain information or data sent by other devices or chips.

[0360] In some embodiments, the chip 1300 may further include an output interface 1340. The processor 1310 may control the output interface 1340 to communicate with other devices or chips, and specifically, may output information or data to other devices or chips.

[0361] In some embodiments, the chip can be applied to the network device in the embodiments of the present application, and the chip can implement the corresponding processes implemented by the network device in each method of the embodiments of the present application. For the sake of brevity, they will not be repeated here.

[0362] In some embodiments, the chip can be applied to the mobile terminal / terminal device in the embodiments of the present application, and the chip can implement the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of the present application. For the sake of brevity, they will not be repeated here.

[0363] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0364] FIG12 is a schematic block diagram of a communication system 1200 provided in an embodiment of the present application. As shown in FIG12 , the communication system 1200 includes a terminal device 1210 and a network device 1220 .

[0365] Among them, the terminal device 1210 can be used to implement the corresponding functions implemented by the terminal device in the above method, and the network device 1220 can be used to implement the corresponding functions implemented by the network device in the above method. For the sake of brevity, they are not repeated here.

[0366] It should be understood that the processor of the embodiments of the present application may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiment can be completed by hardware integrated logic circuits in the processor or software instructions. The above processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The various methods, steps, and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of the present application can be directly embodied as being executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0367] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as 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 link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0368] It should be understood that the above-mentioned memories are exemplary but not restrictive. For example, the memories in the embodiments of the present application may also be 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 link dynamic random access memory (SLDRAM), and direct RAM RAM (DR RAM), etc. In other words, the memories in the embodiments of the present application are intended to include, but are not limited to, these and any other suitable types of memories.

[0369] An embodiment of the present application also provides a computer-readable storage medium for storing a computer program.

[0370] Optionally, the computer-readable storage medium can be applied to the network device in the embodiments of the present application, and the computer program enables the computer to execute the corresponding processes implemented by the network device in the various methods of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0371] In some embodiments, the computer-readable storage medium can be applied to the mobile terminal / terminal device in the embodiments of the present application, and the computer program enables the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of the present application. For the sake of brevity, they will not be repeated here.

[0372] An embodiment of the present application also provides a computer program product, including computer program instructions.

[0373] Optionally, the computer program product can be applied to the network device in the embodiments of the present application, and the computer program instructions enable the computer to execute the corresponding processes implemented by the network device in the various methods of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0374] In some embodiments, the computer program product can be applied to the mobile terminal / terminal device in the embodiments of the present application, and the computer program instructions enable the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of the present application. For the sake of brevity, they will not be repeated here.

[0375] The embodiment of the present application also provides a computer program.

[0376] Optionally, the computer program can be applied to the network device in the embodiments of the present application. When the computer program runs on a computer, the computer executes the corresponding processes implemented by the network device in the various methods of the embodiments of the present application. For the sake of brevity, they are not described here.

[0377] In some embodiments, the computer program can be applied to the mobile terminal / terminal device in the embodiments of the present application. When the computer program runs on a computer, the computer executes the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of the present application. For the sake of brevity, they are not further described here.

[0378] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0379] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0380] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0381] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0382] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0383] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0384] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A model failure recovery method, the method comprising: The terminal device performs performance monitoring of the first artificial intelligence AI model, and sends model failure reporting information to the network device according to the monitoring result, where the model failure reporting information is used to indicate that a model failure occurs in the first AI model, or to indicate whether a model failure occurs in the first AI model; The terminal device receives the model recovery information sent by the network device, and transmits information with the network device according to the model recovery information.

2. The method according to claim 1, wherein: The method further comprises: The terminal device receives a first signaling sent by the network device; The first signaling is used to configure the period of the performance monitoring and / or the period of the terminal device sending the model failure reporting information; or, The first signaling is used to activate the terminal device to perform the performance monitoring and / or activate the terminal device to send the model failure reporting information.

3. The method according to claim 2, wherein: The first signaling is also used to indicate the type of monitoring results used for the performance monitoring and / or the threshold value of the model failure, and the threshold value is used by the terminal device to determine whether the first AI model fails based on the monitoring results.

4. The method according to any one of claims 1 to 3, wherein: The terminal device performs performance monitoring of the first AI model, including: The terminal device monitors the performance of the first AI model by comparing the output of the first AI model with a reference label; the reference label is a real label corresponding to the output of the first AI model, or the reference label is an output obtained based on a non-AI / ML method.

5. The method according to any one of claims 1 to 4, wherein: The method further comprises: If, in multiple performance monitorings of the first AI model, the monitoring results are all worse than the threshold value, or the monitoring results are all different from the preset value, or the statistical value of the monitoring results is worse than the threshold value, the terminal device determines that the first AI model has a model failure.

6. The method according to any one of claims 1 to 4, wherein: The method further comprises: In each performance monitoring of the first AI model, if the monitoring result is worse than the threshold value or the monitoring result is different from the preset value, the terminal device reports the model failure sample to the upper layer; if the number of model failure samples counted by the upper layer reaches a specified maximum value, the terminal device determines that a model failure has occurred.

7. The method according to any one of claims 1 to 6, wherein: The sending of model failure reporting information to the network device according to the monitoring result includes: If the monitoring result is that the first AI model has a model failure, the terminal device sends model failure reporting information to the network device, where the model failure reporting information is used to indicate that the first AI model has a model failure; or, The terminal device sends model failure reporting information to the network device, where the model failure information is used to indicate whether a model failure occurs in the first AI model.

8. The method according to any one of claims 1 to 7, wherein: The sending of the model failure reporting information is triggered by an event, or is triggered by a trigger signaling sent by a network device, or is periodically sent through an uplink control channel.

9. The method according to any one of claims 1 to 8, wherein: The model failure reporting information includes first indication information, where the first indication information is used to indicate one or more second AI models, or to indicate whether the terminal device confirms the second AI model; Among them, the second AI model is the AI ​​model recommended by the terminal device after performance monitoring of the candidate AI model, or is the AI ​​model obtained after the terminal device performs online training on the first AI model.

10. The method according to claim 9, wherein: The candidate AI model is configured by the network device to the terminal device, or is obtained by pre-training a basic AI model by the terminal device.

11. The method according to claim 9 or 10, wherein: The first indication information is also used to indicate the monitoring results corresponding to each of the second AI models.

12. The method according to any one of claims 9 to 11, wherein: If there is no AI model among the candidate AI models whose monitoring results meet the preset conditions, the first indication information is used to indicate that there is no recommended model.

13. The method according to any one of claims 1 to 12, wherein: The model recovery information is used to indicate whether the terminal device uses the second AI model reported by the terminal device through the model failure reporting information to replace the first AI model for information transmission with the network device; or The model recovery information is used to indicate an AI model from a plurality of second AI models reported by the terminal device through the model failure reporting information to replace the first AI model, for information transmission with the network device; or The model recovery information is used to indicate whether the terminal device uses a second AI model confirmed by the terminal device to replace the first AI model for information transmission with the network device; or The model recovery information is used to instruct an AI model from candidate AI models to update the first AI model, and is used to transmit information with the network device; or, The model recovery information is used to instruct the terminal device to use a non-AI / ML method to transmit information with the network device.

14. The method according to any one of claims 1 to 13, wherein: The recovering information according to the model and transmitting information between the network device includes: The terminal device replaces the first AI model with the target AI model indicated by the model recovery information, and uses the target AI model to transmit information with the network device; or The terminal device updates the model structure of the first AI model with the target model structure indicated by the model recovery information, and / or updates the model parameters of the first AI model with the target model parameters indicated by the model recovery information to obtain a target AI model, and uses the target AI model to transmit information with the network device; or The terminal device transmits information with the network device in a non-AI / machine learning ML manner indicated by the model recovery information.

15. The method according to any one of claims 1 to 14, wherein: The recovering information according to the model and transmitting information with the network device comprises: The terminal device determines, according to the model application time, the time at which the target AI model indicated by the model recovery information starts to be applied; the model application time is reported by the terminal device to the network device through the UE capability, or is configured by the network device to the terminal device; After the moment, the terminal device uses the target AI model to transmit information with the network device.

16. The method according to any one of claims 1 to 15, wherein: The method further comprises: After determining that a model failure has occurred, or after sending the model failure reporting information, or after receiving the model recovery information, and before transmitting information with the network device according to the model recovery information, the terminal device transmits information with the network device in one of the following ways: The terminal device uses a non-AI / ML method to transmit information to the network device; or When the first AI model is used for channel state information CSI reporting or beam reporting, the terminal device adopts a first CSI reporting configuration for CSI reporting or beam reporting, and the first CSI reporting configuration is indicated by the model recovery information, or is a CSI reporting configuration pre-agreed between the terminal device and the network device; or, The terminal device uses a third AI model to transmit information with the network device, and the third AI model is indicated by the model recovery information, or is an AI model pre-agreed between the terminal device and the network device; or, The terminal device does not transmit information with the network device.

17. The method according to claim 16, wherein: The terminal device transmits information with the network device in a non-AI / ML manner, including: When the first AI model is used for CSI reporting, the terminal device uses a first codebook type for CSI reporting, where the first codebook type is indicated by the model recovery information or is a codebook type pre-agreed between the terminal device and the network device; or, When the first AI model is used for beam reporting, the terminal device uses a non-AI / ML beam reporting method to perform beam reporting.

18. A model failure recovery method, the method comprising: The network device receives model failure reporting information sent by the terminal device, where the model failure reporting information is used to indicate that a model failure occurs in the first AI model, or is used to indicate whether a model failure occurs in the first AI model; The network device sends model recovery information to the terminal device, and the model recovery information is used for the terminal device to transmit information with the network device.

19. The method according to claim 18, wherein: The method further comprises: The network device sends a first signaling to the terminal device; The first signaling is used to configure the period of the performance monitoring and / or the period of the terminal device sending the model failure reporting information; or, The first signaling is used to activate the terminal device to perform the performance monitoring and / or activate the terminal device to send the model failure reporting information.

20. The method according to claim 19, wherein: The first signaling is also used to indicate the type of monitoring results used for the performance monitoring and / or the threshold value of the model failure, and the threshold value is used by the terminal device to determine whether the first AI model fails based on the monitoring results.

21. The method according to any one of claims 18 to 20, wherein: The sending of the model failure reporting information is triggered by an event, or is triggered by a trigger signaling sent by a network device, or is periodically sent through an uplink control channel.

22. The method according to claim 21, wherein: The event includes the terminal device detecting that a model failure occurs in the first AI model.

23. The method according to any one of claims 18 to 22, wherein: The model failure reporting information includes first indication information, where the first indication information is used to indicate one or more second AI models, or to indicate whether the terminal device confirms the second AI model; Among them, the second AI model is the AI ​​model recommended by the terminal device after performance monitoring of the candidate AI model, or is the AI ​​model obtained after the terminal device performs online training on the first AI model.

24. The method according to claim 23, wherein: The candidate AI model is configured by the network device to the terminal device, or is obtained by pre-training a basic AI model by the terminal device.

25. The method according to claim 23 or 24, wherein: The first indication information is also used to indicate the monitoring results corresponding to each of the second AI models.

26. The method according to any one of claims 23 to 25, wherein: If there is no AI model among the candidate AI models whose monitoring results meet the preset conditions, the first indication information indicates that there is no recommended model.

27. The method according to any one of claims 18 to 26, wherein: The model recovery information is used to indicate whether the terminal device uses the second AI model reported by the terminal device through the model failure reporting information to replace the first AI model for information transmission with the network device; or The model recovery information is used to indicate an AI model from a plurality of second AI models reported by the terminal device through the model failure reporting information to replace the first AI model, for information transmission with the network device; or The model recovery information is used to indicate whether the terminal device uses a second AI model confirmed by the terminal device to replace the first AI model for information transmission with the network device; or The model recovery information is used to instruct an AI model from candidate AI models to update the first AI model, and is used to transmit information with the network device; or, The model recovery information is used to instruct the terminal device to use a non-AI / ML method to transmit information with the network device.

28. The method according to any one of claims 18 to 27, wherein: The method further comprises: The network device receives the model application time reported by the terminal device or configures the model application time for the terminal device, and the model application time is used by the terminal device to determine the moment when the target AI model indicated by the model recovery information starts to be applied.

29. The method according to any one of claims 18 to 28, wherein: The method further comprises: The network device receives information sent by the terminal device using the target AI model indicated by the model recovery information; The network device uses a network-side model corresponding to the target AI model to process the information.

30. The method according to any one of claims 18 to 29, wherein: The method further comprises: After receiving the model failure reporting information or after sending the model recovery information, and before receiving the information transmitted by the terminal device according to the model recovery information, the network device receives the information sent by the terminal device in one of the following ways: The network device receives information sent by the terminal device in a non-AI / ML manner; or, When the first AI model is used for CSI reporting or beam reporting, the network device receives a CSI report or beam report sent by the terminal device using a first CSI reporting configuration, where the first CSI reporting configuration is indicated by the model recovery information or is a CSI reporting configuration pre-agreed between the terminal device and the network device; or, The network device receives information sent by the terminal device using a third AI model, where the third AI model is indicated by the model recovery information or is an AI model pre-agreed between the terminal device and the network device.

31. The method according to claim 30, wherein: The network device receives information sent by the terminal device in a non-AI / ML manner, including: When the first AI model is used for CSI reporting, the network device receives a CSI report sent by the terminal device using a first codebook type, where the first codebook type is indicated by the model recovery information or is a codebook type pre-agreed between the terminal device and the network device; or, When the first AI model is used for beam reporting, the network device receives the beam reporting sent by the terminal device using a non-AI / ML beam reporting method.

32. A model failure recovery device, applied to a terminal device, comprising: A monitoring unit, used to monitor the performance of the first AI model; a sending unit, configured to send model failure reporting information to a network device according to a monitoring result, wherein the model failure reporting information is used to indicate that a model failure occurs in the first AI model, or to indicate whether a model failure occurs in the first AI model; A receiving unit, configured to receive the model recovery information sent by the network device; The sending unit is further used to transmit information with the network device according to the model recovery information.

33. A model failure recovery device, applied to a network device, comprising: A receiving unit, configured to receive model failure reporting information sent by a terminal device, where the model failure reporting information is used to indicate that a model failure occurs in the first AI model, or to indicate whether a model failure occurs in the first AI model; A sending unit is used to send model recovery information to the terminal device, and the model recovery information is used for the terminal device to transmit information with the network device.

34. A terminal device, comprising: A processor and a memory, the memory being used to store a computer program, the processor being used to call and run the computer program stored in the memory, so that the terminal device executes the method as claimed in any one of claims 1 to 17.

35. A network device comprising: A processor and a memory, the memory being used to store a computer program, the processor being used to call and run the computer program stored in the memory so that the network device executes the method as claimed in any one of claims 18 to 31.

36. A chip, comprising: A processor, configured to call and run a computer program from a memory, so that a device equipped with the chip executes a method as claimed in any one of claims 1 to 17, or a method as claimed in any one of claims 18 to 31.

37. A computer-readable storage medium for storing a computer program, wherein the computer program causes a computer to execute the method according to any one of claims 1 to 17, or the method according to any one of claims 18 to 31.

38. A computer program product comprising computer program instructions, the computer program instructions causing a computer to execute the method of any one of claims 1 to 17, or any one of claims 18 to 31.

39. A computer program, the computer program causing a computer to execute the method of any one of claims 1 to 17, or any one of claims 18 to 31.

Images