Model management method and apparatus

Abstract

The present application provides a model management method and apparatus. In a model training stage, a network side issues information indicating an offset between time-domain positions of two reference signal resource sets used for model training. Accordingly, a terminal side can determine a correspondence between input data of models and labels of the models, thereby avoiding mismatching between the input data of the models and the labels of the models. In addition, during function alignment in a model inference stage, in the solution of the present application, a process in which the network side queries and feeds back the offset is added, thereby enhancing the interpretation of inference results by the network side and the determination of time-domain resources among monitoring resources in the subsequent model monitoring stage, and thus achieving more scientific monitoring of model performance.

Description

Methods and apparatus for managing models

[0001] This application claims priority to Chinese patent application filed on February 28, 2025, with application number 202510257176.3 and entitled "Method and Apparatus for Management Model", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of artificial intelligence (AI), and more specifically, to a method for training a model and a communication device. Background Technology

[0003] In scenarios where artificial intelligence (AI) models are applied to air interface tasks, such as beam management or channel state information (CSI) prediction, the model training phase typically requires the network to distribute two types of reference signal resources: channel state information-reference signal (CSI-RS), demodulation reference signal (DMRS), or synchronization signal block (SSB). One of these reference signal resources is used to generate the model's input data, and the other is used to generate the model's labels or ground truth. In existing standards, these two sets of reference signal resources are distinguished by different CSI resource configuration identifiers, which can be represented as CSI-resourceConfigID. The terminal side trains or monitors the model based on these two reference signal resources; however, the model's performance still needs improvement. Summary of the Invention

[0004] This application provides a method and communication device for managing models, which helps to improve model performance.

[0005] In a first aspect, a method for managing a model is provided, executed by a communication device or a module for the communication device (e.g., a processor, chip, circuit, AI entity, etc., or a logic module, hardware, and / or software capable of implementing all or part of the functions of the communication device), the communication device corresponding to the terminal-side device in the method embodiment. The method may include: receiving first information from a network side, the first information being used to determine a first offset, the first offset being an offset between a first set of reference signal resources and a second set of reference signal resources; performing at least one of training or monitoring a first model, wherein the dataset used for training the first model includes input data and labels of the first model obtained based on the first set of reference signal resources and the second set of reference signal resources, or the dataset used for monitoring the first model includes a measured true value of the first model and inference data corresponding to the measured true value obtained based on the first set of reference signal resources and the second set of reference signal resources, and one of the first set of reference signal resources and the second set of reference signal resources is determined based on the other and the first information.

[0006] In this technical solution, the network side instructs the terminal side on information used to determine the offset between the two sets of reference signal resources. Based on this offset, the terminal side can pair (or correspond or associate) the two sets of reference signal resources during the model training phase, making the matching of the model's input data and labels more orderly. This avoids the pairing chaos that may occur when the terminal side autonomously matches the reference signal resource sets, and the performance of the trained model is also better.

[0007] Furthermore, during the model monitoring phase, the terminal side uses this offset to determine the corresponding monitoring resources for each inference result and calculates monitoring metrics. Since the temporal location of the monitoring resources is determined based on the offset between the two sets of reference signal resources during model training and the time of effect of the inference result, the mechanism for matching monitoring resources is more scientific, making the model monitoring results more accurate.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, a first reference signal resource set is used to acquire input data of the first model, and a second reference signal resource set is used to acquire labels of the first model; and the training or monitoring of the first model includes: training the first model based on the first reference signal resource set and the second reference signal resource set, wherein the first measurement data obtained based on the first reference signal resource set is one of the input data of the first model, and the second measurement data obtained based on the second reference signal resource set is the label corresponding to the first measurement data.

[0009] During the model training phase, the network distributes two sets of reference signal resources to obtain the input data and labels required for model training. Based on the offset indicated by the network, the terminal determines the one-to-one correspondence between the model's input data and labels obtained from these two sets of reference signal resources. This allows for model training based on the paired input data and labels, avoiding the risks of chaotic or unreasonable pairings that might occur if the terminal arbitrarily pairs the reference signal resource sets, which the network cannot detect. This helps improve model performance.

[0010] In conjunction with the first aspect, in some implementations of the first aspect, a first reference signal resource set is used to acquire inference data of a first model, and a second reference signal resource set is used to acquire measurement truth values ​​for model monitoring; and the training or monitoring of the first model includes: monitoring the first model based on the first reference signal set and the second reference signal resource set, wherein: first inference data is acquired based on the first measurement data obtained based on the first reference signal resource set and the first model; a first measurement truth value is acquired based on the second reference signal resource set; and a first monitoring result is acquired based on the first inference data and the first measurement truth value.

[0011] During the model monitoring phase, the network side distributes two sets of reference signal resources to obtain the inference results and measurement ground truth values ​​required for model monitoring. The terminal side, based on the offset corresponding to the model (i.e., the offset between the two sets of reference signal resources during model training), finds paired monitoring resources for a single inference result to calculate monitoring metrics. In the mechanism for determining monitoring resources for a single inference result in this application, the effective time of the inference result is used as a benchmark to match monitoring resources that are close to it. This mechanism for matching monitoring resources is more scientific, and the monitoring results are more accurate.

[0012] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: saving attribute information of the first model, which includes the first offset corresponding to the first model.

[0013] In this implementation, the offset between the two sets of reference signal resources during model training is stored on the terminal side as one of the model's attributes. Thus, each model has a corresponding offset. This offset, as one of the model's attributes, can be fed back to the network side when querying for model attribute-related information, allowing the network side to obtain more attribute information about the model and contributing to a more accurate and comprehensive interpretation of the inference results.

[0014] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: receiving a first query message, the first query message being used to request a query for the offset corresponding to the first model; and sending a first response message to the first query message, the first query message indicating the first offset.

[0015] During model function alignment in the model inference phase, the network side can query the corresponding offset of the model, which helps to enhance the network side's interpretation of the inference results and determine the temporal location of monitoring resources in the subsequent model monitoring phase.

[0016] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: receiving first configuration information of a first reference signal resource set and second configuration information of a second reference signal resource set, wherein the first configuration information indicates the time domain position of the first reference signal resource set and the second configuration information indicates the time domain position of the second reference signal resource set, and the time domain position of the second reference signal resource set is determined based on the time domain position of the first reference signal resource set and a first offset.

[0017] In this implementation, the terminal side uses the time of effect of the inference result as the benchmark for finding the closest monitoring resource. Through pre-configuration on the network side, the terminal side uses a dual strategy of finding monitoring resources based on the offset of the model and the temporal location of the inference resource. This ensures that the corresponding monitoring resource can be accurately matched based on the time of effect of the inference result, thus achieving more scientific model monitoring.

[0018] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: receiving a second query message from the network side, the second query message being used to request a query for a model corresponding to the first offset; determining a first model corresponding to the first offset from a plurality of models; and sending a second response message in response to the second query message, the second response message indicating the first model.

[0019] In this implementation, the offset corresponding to the model is one of the model attributes and can also be used by the network side to query the model. For example, when the network side is filtering models, it can send information indicating the offset to the terminal side. The terminal side then searches for models that meet the offset requirements from the saved models and feeds them back to the network side, which can enhance the fit between the model and the requirements.

[0020] Secondly, a method for managing a model is provided, executed by a communication device or a module for the communication device (e.g., a processor, chip, circuit, AI entity, etc., or a logic module, hardware, and / or software capable of implementing all or part of the functions of the communication device), the communication device corresponding to the network-side device in the method embodiment. The method may include: determining first information, the first information being used to determine a first offset, the first offset being an offset between a first set of reference signal resources and a second set of reference signal resources; and sending the first information, wherein: the first set of reference signal resources and the second set of reference signal resources are used for training a first model, the dataset used for training the first model including input data and labels of the first model obtained based on the first set of reference signal resources and the second set of reference signal resources; or, the first set of reference signal resources and the second set of reference signal resources are used for monitoring the first model, the dataset used for monitoring the first model including measured ground truth values ​​of the first model and inference data corresponding to the measured ground truth values ​​obtained based on the first set of reference signal resources and the second set of reference signal resources; and one of the first set of reference signal resources and the second set of reference signal resources is determined based on the other and the first offset.

[0021] The second aspect is the network-side method corresponding to the terminal-side method of the first aspect. The beneficial technical effects of the second aspect or any implementation thereof can be referred to the description of the beneficial technical effects of the corresponding solution of the first aspect.

[0022] In conjunction with the second aspect, in some implementations of the second aspect, a first reference signal resource set is used to obtain input data of the first model, and a second reference signal resource set is used to obtain labels of the first model. The first measurement data obtained based on the first reference signal resource set is one of the input data of the first model, and the second measurement data obtained based on the second reference signal resource set is the label corresponding to the first measurement data.

[0023] In conjunction with the second aspect, in some implementations of the second aspect, a first reference signal resource set is used to obtain inference data of a first model, and a second reference signal resource set is used to obtain the measurement truth value of model monitoring. The first inference data obtained based on the first reference signal resource set corresponds to the first measurement truth value obtained based on the second reference signal resource set, and the first inference data and the first measurement truth value are used to obtain a first monitoring result.

[0024] In conjunction with the second aspect, in some implementations of the second aspect, the method further includes: sending a first query message, the first query message being used to request a query for the offset corresponding to the first model; and receiving a first response message in response to the first query message, the first query message indicating the first offset.

[0025] In conjunction with the second aspect, in some implementations of the second aspect, the method further includes: saving the attribute information of the first model, wherein the attribute information of the first model includes the first offset corresponding to the first model.

[0026] In this implementation, the offset between the two sets of reference signal resources during the model training phase, as one of the model's attributes, can be obtained by the network side by querying the terminal side during the model inference phase. Saving the offset corresponding to the model on the network side helps it to have a comprehensive interpretation of the inference results for each model and can be used to monitor the temporal location configuration of resources during the model monitoring phase.

[0027] In conjunction with the second aspect, in some implementations of the second aspect, the method further includes: determining first configuration information of a first reference signal resource set, the first configuration information indicating the time domain position of the first reference signal resource set; determining second configuration information of a second reference signal resource set, the second configuration information indicating the time domain position of the second reference signal resource set, wherein the time domain position of the second reference signal resource set is determined based on the time domain position of the first reference signal resource set plus a first offset; and transmitting the first configuration information and the second configuration information.

[0028] In conjunction with the second aspect, in some implementations of the second aspect, the method further includes: sending a second query message, the second query message being used to request a query for the model corresponding to the first offset; and receiving a second response message in response to the second query message, the second response message indicating the first model.

[0029] In some implementations of the first or second aspect, one of the first and second reference signal resource sets is determined based on the other and the first information, including: the second reference signal resource set is determined from a plurality of candidate second reference signal resource sets based on the first reference signal resource and the first information.

[0030] In some implementations of the first or second aspect, the first information indicates the first offset.

[0031] In this implementation, the network side indicates an explicit numerical value for the offset.

[0032] In some implementations of the first or second aspect, the first information indicates a rule for determining the first offset, the rule including one of the following: determining a second reference signal resource set based on a first reference signal resource set, wherein the second reference signal resource set is one of a plurality of candidate second reference signal resource sets that is closest in the time domain to the first reference signal resource set; or, determining a first reference signal resource set based on the time domain position of the second reference signal resource set, wherein the first reference signal resource set is one of a plurality of candidate first reference signal resource sets that is closest in the time domain to the second reference signal resource set; or, determining a second reference signal resource set based on the first reference signal resource set, wherein the time domain position of the second reference signal resource set is not earlier than the time domain position of the first reference signal resource set; or, determining a first reference signal resource set based on the second reference signal resource set, wherein the time domain position of the second reference signal resource set is not earlier than the time domain position of the first reference signal resource set.

[0033] In this implementation, the network side instructs the terminal side on the rules for determining the offset corresponding to the model, so that the offset and pairing rules between the two reference signal resources of the terminal side and the network side for the model training or monitoring phase are aligned, which helps in the subsequent interpretation of the model inference results and model monitoring.

[0034] In some implementations of the first or second aspect, the first information indicates the range to which the first offset belongs, wherein the offset between the time-domain location of the first set of reference signal resources and the time-domain location of the second reference signal resource is less than or equal to the range to which the first offset belongs.

[0035] In this implementation, the network side indicates the range of offsets corresponding to the model. Based on the model training requirements, such as application scenario requirements and inference latency, the network side can indicate different ranges of offsets to the terminal side, which can improve the adaptability of the model to application requirements.

[0036] In some implementations of the first or second aspect, the first model is one of a plurality of models on the terminal side, each of which has its corresponding offset.

[0037] In some implementations of the first or second aspect, the first model and the second model in the plurality of models both correspond to a first offset, and one or more of the other attribute information of the first model and the second model are different, except for the first offset.

[0038] In this implementation, offset is one of the model's attributes, and the model can also include other attribute information besides offset. When the network side queries the model, it can query other attributes besides offset to obtain more attribute information about the model, which helps to improve the interpretation of the model's inference results and more accurately monitor the model.

[0039] In some implementations of the first or second aspect, the second response message also indicates one or more other attribute information of the first model.

[0040] Thirdly, a method for managing a model is provided, which can be executed by a terminal. The method may include: receiving a first query message, the first query message being used to request a query for an offset corresponding to a first model; and sending a first response message in response to the first query message, the first response message indicating a first offset corresponding to the first model.

[0041] In one possible implementation, the first model is one of a plurality of models on the terminal side, each of which has its corresponding offset.

[0042] Optionally, a method for managing models is also provided, which can be executed by a terminal. The method may include: receiving a second query message requesting a query for a model corresponding to a first offset; determining a first model corresponding to the first offset from a plurality of models; and sending a second response message to the second query message, the second response message indicating the first model.

[0043] In this method, the terminal side stores the correspondence between multiple models and their respective offsets. The network side can query the models on the terminal side based on the offsets during model function alignment in the model inference phase to select the model corresponding to the offset.

[0044] Optionally, the second query message may also request one or more other attribute information of the model corresponding to the first offset.

[0045] Fourthly, a method for managing models is provided, which can be executed by the network side. The method may include: sending a first query message, the first query message being used to request a query for an offset corresponding to a first model; and receiving a first response message in response to the first query message, the first response message indicating a first offset corresponding to the first model.

[0046] Optionally, a method for managing models is also provided, which can be performed by the network side. The method may include: sending a second query message requesting a query for the model corresponding to a first offset; and receiving a second response message in response to the second query message, the second response message indicating the first model.

[0047] Optionally, the second response message may also indicate one or more other attribute information of the first model.

[0048] Fifthly, a communication device is provided, the communication device having the function of implementing the methods of the first aspect or the third aspect, or any possible implementation of these aspects. The function can be implemented by hardware, by software, or by hardware executing corresponding software. The hardware or software includes one or more units corresponding to the above-described functions.

[0049] Sixthly, a communication device is provided, the communication device having the function of implementing the methods of the second or fourth aspect, or any possible implementation of these aspects. The function can be implemented by hardware, by software, or by hardware executing corresponding software. The hardware or software includes one or more units corresponding to the above-described functions.

[0050] A seventh aspect provides a communication device including at least one processor configured to cause the communication device to perform a method of the first aspect, the third aspect, or any possible implementation thereof; or to perform a method of the second aspect, the fourth aspect, or any possible implementation thereof. Optionally, the at least one processor is coupled to at least one memory for storing a computer program or instructions, and the at least one processor is configured to call and run the computer program or instructions from the at least one memory, causing the communication device to perform a method of the first aspect, the third aspect, or any possible implementation thereof; or to perform a method of the second aspect, the fourth aspect, or any possible implementation thereof. Optionally, the at least one processor may be included in the communication device or may be configured externally to the communication device. Optionally, the communication device further includes the at least one memory. Optionally, the communication device further includes a communication interface.

[0051] Eighthly, a communication device is provided, comprising a communication circuit and a processing circuit. The communication circuit is configured to receive a signal to be processed and transmit the signal to the processing circuit. The processing circuit is configured to process the signal to perform a method as described in the first aspect, the third aspect, or any possible implementation thereof; or to perform a method as described in the second aspect, the fourth aspect, or any possible implementation thereof. Optionally, the communication circuit is further configured to output the processed signal. As an example, the communication circuit may be a transceiver, hardware circuit, bus, module, pin, or other type of communication interface. The signal includes information and / or data. Optionally, the communication device may be a chip or a chip system.

[0052] Ninth aspect, a computer-readable storage medium is provided, wherein computer program code or instructions are stored therein, which, when executed on a computer, cause the method of any one of the first to fourth aspects, or any possible implementation thereof, to be implemented.

[0053] In a tenth aspect, a computer program product is provided, the computer program product comprising computer program code or instructions, which, when executed on a computer, cause the method in any one of the first to fourth aspects, or any possible implementation thereof, to be implemented; or, as in the second aspect or any possible implementation thereof, to be implemented.

[0054] Eleventh aspect: A wireless communication system is provided, including the communication device as described in the fifth aspect and the communication device as described in the sixth aspect. Attached Figure Description

[0055] Figure 1 is a schematic diagram of a communication system applicable to an embodiment of this application.

[0056] Figure 2 is another schematic diagram of a communication system applicable to an embodiment of this application.

[0057] Figure 3 is a schematic diagram of a possible application framework in a communication system.

[0058] Figure 4 is a schematic diagram of another possible application framework in a communication system.

[0059] Figure 5 is a schematic diagram of the prediction process for BM case 1.

[0060] Figure 6 is a schematic diagram of the prediction process for BM case 2.

[0061] Figure 7 is a schematic diagram of two sets of reference signal resources.

[0062] Figure 8 is a schematic diagram of the model inference stage.

[0063] Figure 9 is a schematic diagram of the model monitoring phase.

[0064] Figure 10 is a schematic flowchart of the management model method 200 provided in this application.

[0065] Figure 11 is a schematic flowchart illustrating the application of the management model method provided in this application to the model reasoning stage.

[0066] Figure 12 is a schematic flowchart illustrating the application of the management model method provided in this application to the model monitoring phase.

[0067] Figure 13 is a schematic diagram of matching monitoring resources for inference results in the model monitoring stage of existing schemes and the scheme of this application.

[0068] Figure 14 is a schematic diagram of an application example 1 of the technical solution of this application.

[0069] Figure 15 is a schematic diagram of application example 2 of the technical solution of this application.

[0070] Figure 16 is a schematic diagram of application example 3 of the technical solution of this application.

[0071] Figure 17 shows an application example of the technical solution of this application under another possible architecture.

[0072] Figure 18 is a schematic block diagram of the communication device 1000 provided in this application.

[0073] Figure 19 is a schematic block diagram of another communication device 1100 provided in this application.

[0074] Figure 20 is a schematic structural diagram of the chip provided in this application.

[0075] Figure 21 is a schematic diagram of the structure of the AI ​​processor provided in an embodiment of this application. Detailed Implementation

[0076] The technical solutions in this application will now be described with reference to the accompanying drawings.

[0077] The technical solutions provided in this application can be applied to various communication systems, such as: 5th generation (5G) or new radio (NR) systems, long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, wireless local area network (WLAN) systems, satellite communication systems, future communication systems, or integrated systems of multiple systems, etc.

[0078] In a communication system, one network element can send signals to or receive signals from another network element. These signals can include information, signaling, or data. A network element can also be replaced by an entity, network entity, device, communication equipment, communication module, node, or communication node; this application uses a device as an example. For instance, a communication system can include at least one terminal device and at least one network device. The network device can send downlink signals to the terminal device, and / or the terminal device can send uplink signals to the network device.

[0079] Figure 1 is a schematic diagram of a communication system applicable to an embodiment of this application. As shown in Figure 1, the communication system 100 may include at least one network device, such as network device 110 in Figure 1; the communication system 100 may also include at least one terminal device, such as terminal device 120 and terminal device 130 shown in Figure 1. Network device 110 and terminal devices (such as terminal devices 120 and 130) can communicate via a wireless link. The communication devices in this communication system, for example, network device 110 and terminal device 120, can communicate via multi-antenna technology.

[0080] Optionally, the communication system may also include at least one AI node.

[0081] Figure 2 is another schematic diagram of a communication system applicable to embodiments of this application. Compared to the communication system 100 shown in Figure 1, the communication system 100 shown in Figure 2 further includes an AI node 140. The AI ​​node 140 is used to perform AI-related operations, such as building training datasets, training AI models, inference, or monitoring.

[0082] In one implementation, network device 110 can send data related to AI model training to AI node 140, whereby AI node 140 constructs a training dataset and trains the AI ​​model. As an example, the data related to AI model training may include data reported by terminal devices. AI node 140 can send the results of AI model-related operations to network device 110, which then forwards them to the terminal devices. For example, the results of AI model-related operations may include at least one of the following: a trained AI model, model evaluation results, or test results. Exemplarily, a portion of the trained AI model may be deployed on network device 110, and another portion on the terminal devices. Optionally, the trained AI model may be deployed on network device 110, or it may be deployed on the terminal devices.

[0083] It should be understood that Figure 2 is only used as an example of AI node 140 being directly connected to network device 110. In other scenarios, AI node 140 can also be connected to terminal device. Alternatively, AI node 140 can be connected to both network device 110 and terminal device simultaneously. Alternatively, AI node 140 can also be connected to one or more of network device 110 and terminal device through a third-party network element. This application embodiment does not limit the connection relationship between AI network element and other network elements.

[0084] Alternatively, in another implementation, the AI ​​node 140 can also be configured as a module in a network device and / or a terminal device, for example, in the network device 110, terminal device 120, or terminal device 130 shown in FIG1.

[0085] It should be noted that Figures 1 and 2 are schematic diagrams for ease of understanding only. The communication system may also include other devices, such as wireless relay devices and / or wireless backhaul devices, which are not shown in Figures 1 and 2. Furthermore, in practical applications, the communication system may include multiple network devices or multiple terminal devices. This application does not limit the number of network devices and terminal devices.

[0086] In the embodiments of this application, the terminal device may also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent, or user apparatus. The terminal device can be a device that provides voice / data, such as a handheld device or vehicle-mounted device with wireless connectivity. Currently, examples of terminals include: mobile phones, tablets, laptops, PDAs, mobile internet devices (MIDs), wearable devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical surgery, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, cellular phones, cordless phones, session initiation protocol (SIP) phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), handheld devices with wireless communication capabilities, computing devices or other processing devices connected to wireless modems, wearable devices, terminal devices in 5G networks, or terminal devices in future communication systems, etc., and this application does not limit these examples.

[0087] In this embodiment, the device used to implement terminal-side functions can be a terminal device or a device capable of supporting the terminal device in implementing corresponding functions, such as a processor, circuit, or chip. This device can be configured in the terminal device or used in conjunction with the terminal device. This embodiment only uses the terminal side as an example for illustration, and generally refers to a terminal device or a device deployed on the terminal side, and does not constitute a limitation on the solutions in this embodiment.

[0088] The network devices in this application embodiment may include radio access network (RAN) nodes that connect terminal devices to wireless networks, such as base stations. Base stations can broadly encompass various names as follows, or be replaced by the following names: NodeB, evolved NodeB (eNB), next-generation NodeB (gNB), relay station, access point, transmitting and receiving point (TRP), transmitting point (TP), master station, auxiliary station, motor slide retainer (MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed unit (DU), radio unit (RU), etc. A base station can be a macro base station, micro base station, relay node, donor node, or a combination thereof. A base station can also refer to a communication module, modem, or chip installed within the aforementioned equipment or apparatus. A base station can also be a mobile switching center, equipment performing base station functions in D2D, V2X, and M2M communications, or equipment performing base station functions in future communication systems. A base station can support networks using the same or different access technologies. Optionally, a RAN node can also be a server, wearable device, vehicle, or in-vehicle equipment. For example, the access network equipment in vehicle-to-everything (V2X) technology can be a roadside unit (RSU). The embodiments of this application do not limit the specific technologies or equipment forms used in the network equipment.

[0089] Base stations can be fixed or mobile. For example, a helicopter or drone can be configured to act as a mobile base station, and one or more cells can move depending on the location of the mobile base station. In other examples, a helicopter or drone can be configured as a device to communicate with another base station.

[0090] In some deployments, the network devices mentioned in the embodiments of this application may be devices including CU, DU, or CU and DU, or devices with control plane CU nodes (central unit-control plane (CU-CP)) and user plane CU nodes (central unit-user plane (CU-UP)) and DU nodes. For example, the network devices may include gNB-CU-CP, gNB-CU-UP, and gNB-DU.

[0091] In some deployments, multiple RAN nodes collaborate to assist terminals in achieving wireless access, with different RAN nodes each implementing some of the base station's functions. For example, RAN nodes can be CUs, DUs, CU-CPs, CU-UPs, or RUs. CUs and DUs can be configured separately or included in the same network element, such as a BBU. RUs can be included in radio frequency equipment or radio frequency units, such as RRUs, AAUs, or RRHs.

[0092] In one possible design, the processing unit in the BBU used to implement baseband functions is called the baseband high (BBH) unit, and the processing unit in the RRU / AAU / RRH used to implement baseband functions is called the baseband low (BBL) unit.

[0093] In different systems, CU (or CU-CP and CU-UP), DU, or RU may have different names, but those skilled in the art will understand their meaning. For example, in an open radio access network (ORAN / O-RAN) system, CU can also be called O-CU (open CU), DU can also be called O-DU, CU-CP can also be called O-CU-CP, CU-UP can also be called O-CU-UP, and RU can also be called O-RU. Any of the units among CU (or CU-CP, CU-UP), DU, and RU in this application can be implemented through software modules, hardware modules, or a combination of software modules and hardware modules.

[0094] In this embodiment, the device used to implement network-side functions can be a network device; it can also be a device capable of supporting the network device in implementing corresponding functions, such as a processor, circuit, or chip. This device can be configured within the network device or used in conjunction with the network device. In this embodiment, "network side" is used only as an example, generally referring to network devices or devices deployed on the network side, and does not constitute a limitation on the solutions in this embodiment.

[0095] Network devices and / or terminal devices can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on water; and they can also be deployed in the air on airplanes, balloons, and satellites. This application does not limit the scenario in which the network devices and terminal devices are located. Furthermore, terminal devices and network devices can be hardware devices, or software functions running on dedicated hardware or general-purpose hardware, such as virtualization functions instantiated on a platform (e.g., a cloud platform), or entities that include dedicated or general-purpose hardware devices and software functions. This application does not limit the specific form of the terminal devices and network devices.

[0096] Optionally, the AI ​​node can be deployed in one or more of the following locations within the communication system: access network equipment, terminal equipment, or core network equipment, etc. Alternatively, the AI ​​node can be deployed independently, for example, in a location other than any of the aforementioned devices, such as in the host or cloud server of an over-the-top (OTT) system. The AI ​​node can communicate with other devices in the communication system, which can be, for example, one or more of the following: access network equipment, terminal equipment, or core network equipment, etc.

[0097] This application does not limit the number of AI nodes. For example, when there are multiple AI nodes, they can be divided based on function, such as different AI nodes being responsible for different functions.

[0098] Optionally, AI nodes can be independent devices, integrated into the same device to implement different functions, or they can be network elements in hardware devices, software functions running on dedicated hardware, or virtualization functions instantiated on a platform (e.g., a cloud platform). This application does not limit the specific form of the AI ​​nodes described above. AI nodes can also be called AI network elements or AI modules.

[0099] Figure 3 illustrates a possible application framework in a communication system. As shown in Figure 3, network elements in the communication system are connected via interfaces (e.g., NG, Xn) or air interfaces. These network elements, such as core network equipment, access network equipment (RAN nodes), terminals, or one or more devices in the operation administration and maintenance (OAM) system, are equipped with one or more AI modules. Access network equipment can be a single RAN node or can include multiple RAN devices, such as CUs and DUs. The CUs and / or DUs can also be equipped with one or more AI modules. Optionally, the CU can be further divided into CU-CP and CU-UP. One or more AI models are configured in the CU-CP and / or CU-UP.

[0100] The AI ​​module is used to implement corresponding AI functions. AI modules deployed in different network elements can be the same or different. Depending on the parameter configuration, the AI ​​module can implement different functions. The AI ​​module model can be configured based on one or more of the following parameters: structural parameters (e.g., at least one of the following: number of neural network layers, neural network width, inter-layer connections, neuron weights, neuron activation function, or bias in the activation function), input parameters (e.g., type and / or dimension of input parameters), or output parameters (e.g., type and / or dimension of output parameters). The bias in the activation function can also be referred to as the neural network bias.

[0101] An AI module can have one or more models. A model can infer an output, which includes one or more parameters. The learning, training, or inference processes of different models can be deployed on different nodes or devices, or they can be deployed on the same node or device.

[0102] The network device can be a network device equipped with one or more AI modules. For example, the network device can be one or more devices in the core network, access network, or OAM as shown in Figure 3. The AI ​​module can be the RAN intelligent controller (RIC) shown in Figure 4, such as a near real-time RIC or a non-real-time RIC. For example, a near real-time RIC is set in a RAN node (e.g., in a CU or DU), while a non-real-time RIC is set in the OAM, cloud server, core network device, or other network device.

[0103] Figure 4 illustrates another possible application framework in a communication system. As shown in Figure 4, the communication system includes a Resource Interchange (RIC). For example, the RIC could be the AI ​​module in the RAN device shown in Figure 4, used to implement AI-related functions. RICs include near-real-time RICs (near-RT RICs) and non-real-time RICs (non-RT RICs). Non-real-time RICs primarily process non-real-time information, such as data that is not sensitive to latency, with latency in the order of seconds. Real-time RICs primarily process near-real-time information, such as data that is relatively sensitive to latency, with latency in the order of tens of milliseconds.

[0104] Near real-time RICs are used for model training and inference. For example, they are used to train AI models and then use those models for inference. Near real-time RICs can obtain network-side and / or terminal-side information from RAN devices (e.g., CU, CU-CP, CU-UP, DU, and / or RU) and / or terminal devices. This information can be used as training data or as data for inference.

[0105] Optionally, near real-time RIC can deliver inference results to RAN devices and / or terminal devices.

[0106] Optionally, inference results can be exchanged between CU and DU, and / or between DU and RU. For example, in near real-time RIC, the inference result is submitted to DU, and DU sends it to RU.

[0107] Non-real-time RICs are also used for model training and inference. For example, they can be used to train AI models and then use those models for inference. Non-real-time RICs can obtain network-side and / or terminal-side information from RAN devices (e.g., CU, CU-CP, CU-UP, DU, and / or RU) and / or terminals. This information can be used as training data or as inference data, and the inference results can be delivered to RAN nodes and / or terminals.

[0108] Optionally, inference results can be exchanged between CU and DU, and / or between DU and RU. For example, a non-real-time RIC can submit inference results to DU, which in turn can send them to RU.

[0109] Near real-time RICs and non-real-time RICs can also be configured as separate devices. Alternatively, near real-time RICs and non-real-time RICs can also be part of other devices. For example, near real-time RICs can be configured in RAN nodes (e.g., CU, DU), while non-real-time RICs can be configured in OAM, cloud servers, core network devices, or other devices.

[0110] Optionally, the AI ​​model can be implemented as hardware circuitry, software, or a combination of both, without limitation. Non-limiting examples of software include: program code, program, subroutine, instruction, instruction set, code, code segment, software module, application program, or software application, etc.

[0111] The following section introduces some related technologies involved in the technical solution of this application.

[0112] The application of AI models in beam management (BM) mainly focuses on two aspects: spatial domain prediction and temporal domain prediction, generally referred to as BM case 1 and BM case 2. For ease of understanding, a brief introduction is provided below:

[0113] 1) BM case 1

[0114] Figure 5 illustrates the prediction process for BM case 1. The AI ​​model's input is the beam information of a specific pattern scanned at a certain moment, typically the reference signal received power (RSRP) value. The set of beam information for this specific pattern is called set B. After prediction by the AI ​​model, the output is the prediction information for each beam in the complete beam set (called set A). Based on the prediction information for each beam, the terminal selects the top-K beams in set A and reports these top-K beams and their related information to the network side. The top-K beams can be the optimal top-K beams (see the explanation of optimal beams below; for example, they could be the top-K beams that maximize received or transmitted energy).

[0115] 2) BM case 2

[0116] Figure 6 illustrates the prediction process for BM case 2. A sliding time window T1 is used to collect input data for the AI ​​model, such as the RSRP of set B from time (t-N+1) to time (t). The AI ​​model processes the input data and outputs the prediction results for the beams within the future time window T2. Time window T2 is shown in Figure 6 as time (t+1) to time (t+M). Taking the AI ​​model as a regression model as an example, the prediction result is the predicted RSRP of each beam in set A. The terminal side determines the optimal top-K beams by comparing the predicted RSRPs (refer to the explanation of optimal beams below; for example, it could be the top-K beams that maximize receive or transmit energy). The terminal side reports the predicted RSRPs and beam identifiers (IDs) of these top-K beams to the network side. Taking the AI ​​model as a classification model as another example, the prediction result is the probability that each beam in set A will become the optimal beam. By comparing the probabilities, the optimal top-K beams are determined. The terminal side reports the beam IDs of these top-K beams to the network side. In Figure 6, top-K(t+1) represents the prediction result of the AI ​​model at time (t+1), and top-K(t+M) represents the prediction result of the AI ​​model at time (t+M).

[0117] The following uses beam management as an example to illustrate the training, inference, and monitoring processes of the AI ​​model in this application embodiment.

[0118] To facilitate understanding of the solution, a brief introduction to the relevant concepts involved in model training, inference, and monitoring will be provided.

[0119] Training data used to train an AI model includes training samples and sample labels. For example, in an AI-based beam management scenario, the training samples are measured values ​​of channel state information corresponding to each reference signal resource in reference signal resource set B, and the sample labels are measured values ​​of channel state information corresponding to each reference signal resource in reference signal resource set A. Reference signal resource set B is a subset of reference signal resource set A, or the signal angle corresponding to each reference signal resource in reference signal resource set B is greater than the signal angle corresponding to each reference signal resource in reference signal resource set A. For example, each reference signal resource in reference signal resource set B is a wide beam, and each reference signal resource in reference signal resource set A is a narrow beam.

[0120] The specific training process is as follows: The model training node uses the measured values ​​of the channel state information corresponding to each reference signal resource in the reference signal resource set B to determine the input data of the initial AI model, and inputs this input data into the initial AI model to obtain the inference result of the initial AI model. This input data includes the measured values ​​of the channel state information corresponding to each reference signal resource in the reference signal resource set B, or it includes data obtained by performing interpolation, normalization, or filtering on the measured values ​​of the channel state information corresponding to each reference signal resource in the reference signal resource set B. The inference result includes the predicted values ​​of the channel state information corresponding to each reference signal resource in the reference signal resource set A. The model training node calculates the difference between the predicted value and the measured value of the channel state information corresponding to at least one reference signal resource in the reference signal resource set A (i.e., calculates the difference between the inference result of the initial AI model and the corresponding sample label), that is, calculates the value of the loss function. The model training node updates the parameters in the initial AI model according to the value of the loss function, minimizing the difference between the inference result obtained by the updated AI model and the corresponding sample label, that is, minimizing the loss function. For example, the loss function can be the minimum mean square error (MSE) or cosine similarity, etc. Repeating the above steps yields an AI model that meets the target requirements. The training nodes for this model can be terminal devices, network devices, or other network elements with AI capabilities in a communication system.

[0121] Beam management refers to the process by which terminal and network devices periodically identify the optimal beam. The optimal beam can be the beam that maximizes received or transmitted energy. For example, if a receiver uses different receive beams to receive a reference signal, the optimal beam can include the beam with the highest measured value of the corresponding reference signal among multiple different receive beams. Similarly, if a transmitter uses different transmit beams to transmit a signal, the optimal beam can include the beam with the highest measured value of the corresponding reference signal when the transmitted reference signal arrives at the receiver. The measured value of the reference signal can be, for example, the measured reference signal received power (RSRP), signal-to-interference-plus-noise ratio (SINR), or other possible estimates.

[0122] A beam is a communication resource. In the NR protocol, a beam can be represented as a spatial filter, or spatial parameters. The beam used to transmit signals can be called a transmission beam (Tx beam), or a spatial domain transmit filter, or a spatial domain transmit parameter; the beam used to receive signals can be called a reception beam (Rx beam), or a spatial domain receiver filter, or a spatial domain receive parameter. The transmission beam refers to the distribution of signal strength in different directions in space after the signal is transmitted through the antenna, while the reception beam refers to the distribution of signal strength in different directions in space of the wireless signal received from the antenna. Beams can be identified by their identifier (ID). For example, a beam ID can be a Channel State Information Reference Signal Resource Indicator (CSI-RS, CRI), an SSB Resource Indicator (SSBRI), or a bit in a bitmap corresponding to the beam, or an index of the beam within a beam set. For instance, the number of bits in the bitmap is equal to the total number of beams associated with the network device in a single beam management inference task. Beams can be categorized as wide beams and narrow beams. A wide beam is a beam with a relatively large radiation range of the transmitting or receiving antenna when transmitting or receiving signals. Wide beams are typically used in applications requiring broadcasting signals to a large area or providing wide coverage. Wide beams can provide a wider coverage area, but the signal strength is relatively weaker. A narrow beam is a beam with a relatively small radiation range of the transmitting or receiving antenna. Narrow beams are typically used in applications requiring focusing signals onto a specific target or area. Narrow beams can provide higher signal strength and higher directivity, but the coverage area is relatively smaller.

[0123] The above measurements can also be called information characterizing the channel state, that is, they can be called measurements of channel state information.

[0124] In this application, the channel state information may include one or more of the following: rank indication (RI) information, channel quality indicator (CQI) information, precoding matrix indicator (PMI) information, or reference signal receiver power (RSRP), such as layer 1 reference signal receiver power (L1-RSRP), reference signal receiver quality (RSRQ), and signal to interference plus noise ratio (SINR).

[0125] In this application, the reference signal resource is used to carry a reference signal, which, for example, includes a synchronizing signal block (SSB) and / or a channel state information reference signal (CSI-RS). The reference signal resource may include a beam, or have a corresponding relationship with a beam. Furthermore, the reference signal resource may also include time-domain resources and / or frequency-domain resources corresponding to the beam, such as time-frequency resources. The beam can also be referred to as a spatial domain resource.

[0126] Alternatively, the beam can be replaced with a first signal, downlink beam, transmit beam, transmit beam, thin beam, narrow beam, wide beam, spatial filter, spatial filter, spatial parameters, spatial transmit filter, port, etc.

[0127] In this application, the information used to indicate the beam used for transmission can be called beam indication information. Beam indication information can be one or more of the following: beam number (or index, identifier, ID, etc.), identifier of signal resources (e.g., identifier of reference signal resources, such as index or number, where the signal resources can be one or more of uplink signal resources, downlink signal resources, or sidelink signal resources, where the index or number can be absolute, relative, or logical, and can include one or more of the following: group or set index or number, or index or number of resources within a group or set, or index or number of resources), absolute index of the beam, relative index of the beam, logical index of the beam, index of the antenna port corresponding to the beam, index of the antenna port group corresponding to the beam, index of the signal (e.g., downlink signal, uplink signal, or sidelink signal, etc.) corresponding to the beam, time index of the SSB corresponding to the beam, beam pair link (BPL) information, transmit parameters (Tx parameter) corresponding to the beam, and receive parameters (Rx parameter) corresponding to the beam. The beam indication information includes at least one of the following: beam parameter, beam-corresponding transmit weight, beam-corresponding weight matrix, beam-corresponding weight vector, beam-corresponding receive weight, beam-corresponding transmit weight index, beam-corresponding weight matrix index, beam-corresponding weight vector index, beam-corresponding receive weight index, beam-corresponding receive codebook, beam-corresponding transmit codebook, beam-corresponding receive codebook index, and beam-corresponding transmit codebook index. The absolute index of the beam includes, for example, the index of the beam in beam set A, and the relative index of the beam includes, for example, the index of the beam in a subset M of beam set A. The logical index of the beam includes, for example, the bit corresponding to the beam in the bitmap. Beam indication information can also be represented as a transmission configuration index (TCI) or a TCI status. A TCI status includes one or more quasi-co-location (QCL) information, each QCL information including a reference signal (or synchronization signal block) ID and a QCL type. For example, a terminal device may need to determine the beam to receive the physical downlink shared channel (PDSCH) based on the TCI status indicated by the network device (typically carried by the physical downlink control channel, PDCCH). In this application, the beam index information, i.e., the beam ID, is a typical example of beam indication information. The beam index can also be replaced with other beam indication information that can indicate a beam.In this application, the identification information of the reference signal resource can be replaced with the identification information of the beam corresponding to the reference signal resource or the index information of the beam (such as one or more of absolute index, relative index, or logical index). The reference signal resource can also be replaced with a beam.

[0128] In this application, prediction information (i.e., predicted values) refers to the prediction results directly output by the AI ​​model, or the result obtained after data processing of the prediction results directly output by the AI ​​model. A single piece of prediction information refers to the prediction result directly output by the AI ​​model in a single prediction process, or the result obtained after processing the prediction result. The AI ​​model can be deployed on a terminal device or on an OTT device on the terminal device side. When the AI ​​model is deployed on an OTT device, the terminal device can receive the prediction results output by the AI ​​model from the OTT device.

[0129] To achieve beam management, one possible approach is to reduce beam scanning overhead through methods such as layered scanning. This involves scanning a wide beam first, followed by scanning a portion of narrow beams within the wide beam, thus reducing overhead. A schematic diagram of wide and narrow beams is shown in Figure 4. As shown in Figure 4, compared to narrow beams, wide beams have a wider beam angle and can transmit signals over a wider range of directions. Conversely, narrow beams have a narrower beam angle and can transmit signals over a smaller range of directions. When the sum of the beam angles of a set of wide beams is the same as the sum of the beam angles of a set of narrow beams, the wide beams contain fewer beams than the narrow beams.

[0130] Beam selection is primarily accomplished through reference signals and corresponding beam measurements. Specifically, the reference signal can include a synchronization signal block (SSB, also known as a synchronization signal / physical broadcast channel block (SS / PBCH block) or a synchronization signal block (SS block)), or one or more of the following: CSI-RS. The SSB can be a cell broadcast signal, comprising the primary synchronization signal (PSS), secondary synchronization signal (SSS), physical broadcast channel (PBCH), and demodulation reference signal (DMRS). The SSB can be transmitted periodically according to the cell configuration, and can be considered a wide-beam signal. Correspondingly, the CSI-RS signal can be a user equipment-level signal, and can be understood as a narrow-beam signal.

[0131] AI technology can also be applied to beam scanning, thereby reducing overhead. In AI model-based beam management, the AI ​​model can be a regression model or a classification model. The terminal device or network device determines the input data of the AI ​​model based on the measured values ​​of the channel state information corresponding to each reference signal resource in the reference signal resource set B. In existing schemes, the number of reference signal resources in the reference signal resource set B used by the AI ​​model during the training or inference phase is usually fixed.

[0132] When the AI ​​model is a regression model, during the training phase, the input to the AI ​​model is input data A. The inference result obtained by inputting this input data A into the untrained AI model is the inference result A. The true value of the AI ​​model is the true value A, and the loss function of the AI ​​model is the loss function A. Input data A is determined based on the measured values ​​of the channel state information corresponding to each reference signal resource in the reference signal resource set B. For example, input data A can be the measured values ​​of the channel state information corresponding to each reference signal resource in the reference signal resource set B, or it can be data obtained after processing the measured values ​​of the channel state information corresponding to each reference signal resource in the reference signal resource set B, such as through normalization or filtering. This filtering process includes removing one or more measured values ​​(e.g., one or more smaller measured values) from the measured values ​​of the channel state information corresponding to each reference signal resource in the reference signal resource set B. The inference result A includes the predicted values ​​of the channel state information corresponding to each reference signal resource in the reference signal resource set A (set A). The true value A is determined based on the measured values ​​of the channel state information corresponding to each reference signal resource in the reference signal resource set A. For example, the true value A is the measured value of the channel state information corresponding to each reference signal resource in the reference signal resource set A, or the true value A is the data obtained after processing the measured value of the channel state information corresponding to each reference signal resource in the reference signal resource set A, such as by normalization. The loss function A is determined based on the inference result A and the true value A. For example, the loss function is the average of the differences between the measured value and the predicted value of the channel state information corresponding to at least one reference signal resource in the reference signal resource set A, the MSE of the difference, or the cosine similarity of the difference, etc. The AI ​​model needs to be trained using multiple input data A during the training process, but the number of reference signal resources in the reference signal resource set B corresponding to these multiple input data A is the same. Based on the loss function A, one or more methods such as gradient descent, momentum method, and adaptive learning rate method are used to adjust the parameters in the untrained AI model, so that the difference between the inference result A obtained using the adjusted AI model and input data A and the true value A is minimized, that is, the loss function A is minimized, thereby obtaining a trained AI model. It is understood that "well trained" in this application may mean that the performance of the AI ​​model has met the requirements, for example, the loss function is less than the required threshold.

[0133] It should be understood that the methods used to adjust the parameters of an AI model, such as gradient descent, momentum, and adaptive learning rate, are similar to existing methods for adjusting AI model parameters. The following explanation uses gradient descent as an example to illustrate the method for adjusting AI model parameters to obtain a well-trained AI model.

[0134] For example, the steps for adjusting the parameters in an AI model using gradient descent on a terminal device or training device are as follows:

[0135] (1) Initialize the parameters of the AI ​​model, that is, determine at least one initial value for each parameter in at least one parameter of the initial AI model.

[0136] (2) Use the current parameters of the AI ​​model to process the current input of the AI ​​model and obtain the output data of the AI ​​model.

[0137] (3) Determine the value of the loss function based on the output data obtained in the previous step and the real data corresponding to the output data.

[0138] (4) Determine the gradient of the loss function based on the value of the loss function.

[0139] (5) Update the parameters of the AI ​​model based on the gradient of the loss function.

[0140] (6) Repeat steps (2)-(5) above until the loss function converges or the preset number of iterations is reached. The convergence of the loss function includes the loss function reaching its minimum value. The AI ​​model that converges when the loss function converges is a pre-trained AI model, or the AI ​​model that reaches the preset number of iterations is a pre-trained AI model.

[0141] In scenarios where AI models are applied to air interface tasks, such as beam management or CSI prediction, the training phase of the model typically requires the network side to provide two types of reference signal resources:

[0142] 1) Reference signal resources used to acquire model input data;

[0143] 2) Reference signal resources used to obtain the label or truth value of the model.

[0144] These two sets of reference signal resources can be identified by two different CSI-resourceConfigIDs. For example, the set of the first set of reference signal resources is represented as set B, and the set of the second set of reference signal resources is represented as set A.

[0145] Figure 7 is a schematic diagram of two reference signal resource sets. As shown in Figure 7, one reference signal resource set is set B, as shown in set B#1 to set B#5, which can be used by the terminal side to obtain the input data of the model; the other reference signal resource set is set A, as shown in set A#1 to set A#5, which can be used by the terminal side to obtain the label of the model.

[0146] During the model training phase, the terminal autonomously selects the correspondence between set B and set A to generate input data and labels for model training. For example, in Figure 7, the terminal can autonomously select a one-to-one correspondence between {set B#1, ..., set B#5} and {set A#1, ..., set A#5}. The offset between the corresponding set B and set A, during the model inference phase, is the time interval between the time of set B's RS and the applicable time of the inference result, as shown in Figure 8.

[0147] Figure 8 is a schematic diagram of the model inference phase. The time at which set A occurs during the model training phase is the time at which the model inference results take effect.

[0148] Similar to model training, during the model monitoring phase, the network side distributes a set of reference signal resources for model inference, represented as set B for the terminal side to obtain inference data and reference signal RS for the terminal side to obtain measurement truth values, called monitoring RS. The terminal side determines the pairing relationship between the inference results and the measurement truth values, and compares the paired inference results with the measurement truth values ​​to obtain the values ​​of the monitoring indicators, as shown in Figure 9.

[0149] Figure 9 is a schematic diagram of the model monitoring phase. As shown in Figure 9, the pairing relationship determined on the terminal side can be the pairing of the inference result of set B#1 and the monitoring RS#1.

[0150] In existing solutions, there are two methods for the terminal side to select the pairing relationship between the inference result and the monitoring RS: 1) select the monitoring RS pairing closest to the time when set B was issued; 2) select the monitoring RS pairing closest to the time when the inference result was reported (not shown in Figure 9).

[0151] In known schemes, the pairing relationship between the two sets of reference signal resources (such as set B and set A) is determined autonomously by the terminal during both the model training and monitoring phases. However, during the research process of this application, careful analysis revealed that the scheme of autonomously selecting reference signal resources for pairing by the terminal has the following drawbacks:

[0152] 1) Since the pairing relationship between reference signal resources is determined by the terminal side, the network side lacks awareness of model training, which is not conducive to the interpretation of inference results and the monitoring of the model.

[0153] 2) When the terminal autonomously determines the pairing relationship between set B and set A, pairing confusion may occur. Taking the application of the AI ​​model in beam management (BM) task (also known as BM case 1) as an example, refer to any one of Figures 7 to 9. If set B#1 and set A#3 are paired, the interval between the time domain position of set B#1 and the time domain position of set A#3 will reach 45ms. This is obviously too large for BM case 1 and may cause the inference results to be unusable or unreferenced.

[0154] 3) During the model monitoring phase, the existing two pairing relationships cannot accurately determine the effective time of the inference results, while the effective time of the inference results can often correctly reflect the location of the RS predicted by the model.

[0155] The aforementioned shortcomings mean that the accuracy of the trained model still has room for improvement. The model's performance monitoring mechanism is unscientific, resulting in low accuracy of the monitoring results. Furthermore, because the terminal autonomously pairs the reference signal resource set, the network is unaware of the pairing details, leading to inaccurate interpretation of the inference results.

[0156] Based on the aforementioned technological status, this application provides a method for managing models. By determining the correspondence (or pairing relationship) between the model's input data and its labels during the training phase, it avoids pairing confusion. Furthermore, it enhances the network's interpretation of the model's inference results and improves the accuracy of model performance monitoring.

[0157] Overall, the technical solution of this application mainly involves the following aspects:

[0158] 1. During model training, the network instructs the terminal to determine the offset between two sets of reference signal resources (e.g., set B and set A as mentioned above). Based on this offset information, the terminal can define the pairing relationship between a single set B and a single set A. During model training, the input data and labels of the model are paired according to this offset, and model training is performed based on the paired data. Furthermore, the terminal can store the offset as attribute information of the model itself.

[0159] 2. During the model inference phase, the network side can query the offset corresponding to the model. The terminal side responds with the offset queried by the network side. That is, during the model inference phase, a query-response process for the model's "offset" is added. This process is executed when the network side is unaware of the model's offset.

[0160] 3. During the model monitoring phase, for each inference result (or prediction result), the terminal calculates the effective time of the inference result based on the offset corresponding to the model, and then finds the closest monitoring resource in the time domain based on the effective time of the inference result to calculate the monitoring result. Optionally, the aforementioned offset can come from an indication from the network side. Optionally, the network side can determine the time domain location of the monitoring resource based on the offset corresponding to the model.

[0161] It is understandable that one or more of the above aspects can be applied in combination or implemented independently.

[0162] The technical solution of this application will be described in detail below.

[0163] Figure 10 is a schematic flowchart of the management model method 200 provided in this application. Method 200 involves network-side devices and terminal-side devices, and method 200 can be implemented by the network-side devices and terminal-side devices each performing corresponding steps.

[0164] Optionally, the network-side device in method 200 can refer to any network device and any apparatus used with the network device (e.g., referred to as the first apparatus). The first apparatus can be a chip, processor, circuit, or AI entity serving the network device. The AI ​​entity can be deployed on the network device or outside of the network device. If the first apparatus is deployed outside the network device, air interface interaction between the first apparatus and the network device may also be involved, such as data exchange or information exchange related to model training, inference, or monitoring. This facilitates the first apparatus to perform relevant processing based on the data set or information provided by the network device to complete the model training, inference, or monitoring method in this application. This description applies to any embodiment of this application, and will not be repeated below.

[0165] Similarly, "terminal-side device" generally refers to a terminal device or a device used in a terminal device (e.g., referred to as a second device). The second device can be a chip, processor, circuit, or AI entity used in a terminal device, or serving a network device.

[0166] When the first device is an AI entity, it can be a deployment device for a network-side AI model collectively referred to as an intelligent network element, such as the near real-time RIC or non-real-time RIC in Figure 4 above. When the second device is an AI entity, it can be a host or cloud server of an over-the-top (OTT) system. The deployment of the network-side or terminal-side AI model can be located outside the same physical entity as other network-side or terminal-side devices. The following embodiments use network-side devices and terminal-side devices as examples.

[0167] As described above, network-side devices may include network devices and network-side AI entities, such as intelligent network elements; terminal-side devices may include terminal devices and terminal-side AI entities, such as OTT servers. Therefore, method 200 can be implemented by network devices and terminal devices performing corresponding steps; or, in addition to network devices and terminal devices, it may also include intelligent network elements, OTT servers, etc., each of which performs corresponding steps to implement method 200.

[0168] 210. The network side determines the first information, which is used to determine the first offset, which is the offset between the first reference signal resource set and the second parameter signal resource set.

[0169] Optionally, "offset" can be represented as offset or "time interval," correspondingly as time gap. When expressed as time interval, the first information indicates the first time interval, which is the interval between the time-domain positions of the first reference signal resource set and the second reference signal resource set. Alternatively, it can be said that the first time interval is the interval between the time-domain resources of the first reference signal resource set and the time-domain resources of the second reference signal resource set.

[0170] It can be understood that the "offset" or "time interval" can be calculated from the time domain position of the first reference signal resource set and the time domain position of the second reference signal resource set.

[0171] In the embodiments of this application, "time domain location" can be understood as one or more of time, time slot, subframe, frame, or orthogonal frequency division multiplexing (OFDM) symbol; for example, the time domain location of A can represent the time slot, subframe, frame, or OFDM symbol where A is located, or the first time slot, subframe, frame, or OFDM symbol where A is located, or the last time slot, subframe, frame, or OFDM symbol where A is located, etc., without limitation.

[0172] 220. The network side sends the first information. The terminal side receives the first information.

[0173] As an example, the first information indicates the first offset.

[0174] In this example, the network side explicitly indicates the value of the first offset, such as 5ms, 10ms, -15ms, etc.

[0175] As shown in Table 1, the scheme of this application supports positive and negative offsets. It can be understood that the direction of the offset can also be fixed as positive or negative. In this case, if the direction is positive or negative as predefined by the protocol, the offset can only include positive values.

[0176] Table 1

[0177] For example: offset = 5ms means that set A is 5ms after set B (corresponding to set A#1 and set B#1 in Table 1); offset = -15ms means that set A is 15ms before set B (corresponding to set A#1 and set B#2 in Table 1).

[0178] In another example, the first information indicates the rule used to determine the first offset, the rule including any of the following:

[0179] A second reference signal resource set is determined based on a first reference signal resource set, wherein the second reference signal resource set is the one among multiple candidate second reference signal resource sets whose time-domain location is closest to the first reference signal resource set; or,

[0180] The first reference signal resource set is determined based on the second reference signal resource set, wherein the first reference signal resource set is one of the second reference signal resource sets whose time-domain location is closest among multiple candidate first reference signal resource sets; or,

[0181] A second reference signal resource set is determined based on a first reference signal resource set, wherein the time-domain location of the second reference signal resource set is no earlier than that of the first reference signal resource set; or,

[0182] The first reference signal resource set is determined based on the second reference signal resource set, and the time domain position of the second reference signal resource set is no earlier than the time domain position of the first reference signal resource set.

[0183] In this example, the network side indicates the rules for determining the first offset. Based on these rules, the terminal side selects a corresponding (i.e., paired with the first reference signal resource set) second reference signal resource set, using the first reference signal resource set as a reference; or, using the second reference signal resource set as a reference, selects a corresponding first reference signal resource set. Once the first and second reference signal resource sets are determined, i.e., paired, the first offset can be determined by the temporal position of the first and second reference signal resource sets.

[0184] In yet another example, the first information indicates the range to which the first offset belongs.

[0185] In this example, the network side indicates the range of the offset between the first and second reference signal resource sets. For example, the first information indicates 30ms, meaning the offset between the time-domain positions of the first and second reference signal resource sets is no greater than 30ms. Based on this first information, assuming the terminal side determines the second reference signal resource set based on the first reference signal resource set, the terminal side uses the time-domain position of the first reference signal resource set as a reference to select a suitable second reference signal resource set from multiple candidate second reference signal resource sets. This is equivalent to the network side indirectly indicating the first offset. Because once the second reference signal resource set is determined, the offset between the time-domain positions of the first and second reference signal resource sets can be determined, thus obtaining the first offset.

[0186] Optionally, when multiple sets of second reference signal resources that meet the requirements are determined based on the range indicated by the first information, the terminal can select one of the multiple sets as the final set of second reference signal resources paired with the first set of reference signal resources. In this case, the rule for the terminal to select from the multiple sets can be predefined by the protocol or indicated by the first information; that is, the first information can include both information indicating the range and information indicating the rule. For example, the information indicating the range and the information indicating the rule can be carried in the same or different information cells in the same message, or in information cells in different messages, which is not limited here.

[0187] As can be seen, based on the first information, the terminal can determine the first offset, thereby determining the second reference signal resource set paired with the first reference signal resource set, and perform model training or monitoring based on the paired first and second reference signal resource sets. Furthermore, the first offset can also be used in the model's inference phase, which will be discussed in detail later.

[0188] Both the first and second reference signal resource sets mentioned above support three configuration modes: periodic, semi-persistent, and aperiodic. In the periodic configuration mode, the network side configures the transmission period and offset (e.g., symbol offset within the period) of the reference signal resource set and notifies the terminal side. The network side then transmits the reference signal resource set according to the configured period and offset. In the semi-persistent configuration mode, the network side configures the transmission period and offset (e.g., symbol offset within the period) of the reference signal resource set and notifies the terminal side. However, whether or not the reference signal resource set is actually transmitted can be determined by the medium access control-control element (MAC CE). The MAC CE can activate or deactivate the transmission of the reference signal resource set and notify the terminal side. In the aperiodic configuration method, the network side notifies the terminal side of the transmission of each reference signal resource set through downlink control information (DCI) signaling. The aperiodic reference signal resource set also supports the configuration of multiple reference signal resource sets for transmission at one time. For example, it can be configured through a set of parameters [m, K], where m is the time offset between reference signal resource sets and K is the number of reference signal resource sets transmitted at one time.

[0189] 230. The terminal side performs training or monitoring of the first model.

[0190] The dataset used for training the first model includes measurement data or inference data corresponding to the measurement data, obtained based on the first reference signal resource set and the second reference signal resource set.

[0191] In one possible scenario, the technical solution of this application is applied to the model training stage on the terminal side.

[0192] In this case, in step 230, the first model is trained on the terminal side.

[0193] Specifically, the dataset used for training the first model includes: input data and labels of the first model obtained based on a first reference signal resource set and a second reference signal resource set. For example, the first reference signal resource set is used to obtain the input data of the first model, and the second reference signal resource set is used to obtain the labels of the first model.

[0194] During the training phase of the first model, the terminal measures the first reference signal resource set to obtain the input data for the first model. Since the first information indicates the offset between the first and second reference signal resource sets, the terminal can pair the first and second reference signal resource sets based on the first offset. For example, if the first reference signal resource set is used as a reference and combined with the first offset, the second reference signal resource set can be selected; or, if the second reference signal resource set is used as a reference and combined with the first offset, the first reference signal resource set can be selected. Which of the first and second reference signal resource sets is used as the reference to determine the other pairing can be determined by protocol definition, pre-configuration, or signaling interaction, and is not limited.

[0195] The first reference signal resource set is one of multiple reference signal resource sets used to obtain input data for the first model. Therefore, the first measurement data (also referred to as the first input data) obtained based on the first reference signal resource set is one of the input data of the first model. Similarly, the second reference signal resource set is one of multiple reference signal resource sets used to obtain labels for the first model. The second measurement data obtained based on the second reference signal resource set is one of the labels for the first model, hereinafter referred to as the first label. Based on the technical solution provided in this application, the first measurement data and the first label are corresponding or paired for training the first model.

[0196] Taking Figure 7 as an example, the reference signal resource set used to acquire the model's input data is denoted as set B, and the reference signal resource set used to acquire the model's labels is denoted as set A. Figure 7 shows five sets B, denoted as set B#1 to set B#5. The first reference signal resource set can refer to any one of set B#1 to set B#5. Similarly, Figure 7 shows five sets A, denoted as set A#1 to set A#5. The second reference signal resource set can refer to any one of set A#1 to set A#5. Therefore, it can be understood that the first offset indicates the correspondence between a single set B and a single set A. The terminal side can determine the set A paired with a set B based on the temporal location of a set B, combined with the first offset; or, it can determine the set B paired with a set A based on the temporal location of a set A, combined with the first offset. When determining the second reference signal resource set paired with a first reference signal resource set as a reference, it can be understood as determining the second reference signal resource set paired with the first reference signal resource set from multiple candidate second reference signal resource sets. Taking set B#1 as an example, sets A#1 to A#5 can be understood as candidate sets of second reference signal resources. If the first offset is 5ms, it means that the second set of reference signal resources is set A that follows the first set of reference signal resources and has an offset of 5ms from set B#1, which is set A#1. If the offset is 25ms, the set A paired with set B#1 is set A#2. For each set B, the paired set A can be determined in this way. Conversely, if set A is used as a reference to determine the paired set B, the process is similar. Taking set A#3 as an example, sets B#1 to B#5 are candidate sets of first reference signal resources. If the first offset is -5ms, it means that the set B paired with set A#3 is set B that precedes set A#3 and has an offset of 3ms from set A#3, which is set B#3.

[0197] Therefore, in the embodiments of this application, the candidate first reference signal resource set can be understood as a plurality of reference signal resource sets configured on the network side for obtaining input data of the model when determining the first reference signal resource set paired with it based on the second reference signal resource set; the candidate second reference signal resource set can be understood as a plurality of reference signal resource sets configured on the network side for obtaining labels of the model when determining the second reference signal resource set paired with it based on the first reference signal resource set.

[0198] Furthermore, when pairing set B and set A based on the first offset indicated by the first information and the rules or range for determining the first offset, generally one set B is paired with one set A. If multiple sets A meet the conditions for pairing with a set B, for example, if the first information indicates that the range of the first offset is 30ms, and set A#1 and set A#2 both meet the conditions, the terminal can select one set A from the multiple sets A that meet the conditions to pair with set B#1. For example, it can select the set A that is closest to set B#1 in the time domain from the multiple sets A that meet the conditions, thus ensuring that set B and set A are paired one-to-one. Conversely, when determining the corresponding set B based on set A, if a "one-to-many" situation occurs, the same processing method can be used, which will not be elaborated further.

[0199] According to the method of this application, each set B is paired with a set A. Based on multiple sets of paired reference signal resources, multiple sets of training data can be obtained, for example, represented as (input data B#1, label B#1), (input data B#2, label B#2), ..., (input data B#m, label B#m), where m equals the number of sets B. Taking Figure 7 as an example, m equals 5. The model is trained based on these multiple sets of training data to obtain the first model.

[0200] In one possible scenario, among multiple sets A, there might be sets A that are not paired with any set B. Taking Figure 7 as an example, when determining the paired set A based on set B, if the first offset is 25ms, then set B#1 is paired with set A#2, set B#2 is paired with set A#3, and so on. Set A#1 is not paired with any set B. In other words, it is permissible for sets A to be unused. As an example, it is allowed for sets A to be unused, while each set B is used.

[0201] In another possible scenario, the above-described technical solution of this application can also be applied to model monitoring.

[0202] In this case, in step 230, the terminal side monitors the first model.

[0203] During the model monitoring phase, the network also distributes two sets of reference signal resources. Unlike the model training phase, the two sets of reference signal resources serve different purposes. For example, in model monitoring, the first set of reference signal resources is used to acquire the inference data of the first model, while the second set of reference signal resources is used to acquire the measured ground truth for model monitoring. The terminal measures the first set of reference signal resources, and the obtained measurement data serves as the input to the first model, thus acquiring the inference data of the first model, hereinafter referred to as the first inference data. Similarly, a second set of reference signal resources is determined by a first offset, thus the first inference data obtained based on the first set of reference signal resources and the first measured ground truth obtained based on the second set of reference signal resources are paired.

[0204] Since the first set of reference signal resources is used for model inference, it can also be called inference resources, while the second set of reference signal resources can be called monitoring resources.

[0205] As can be seen, in model monitoring, the terminal side pairs inference resources and monitoring resources through the first offset. This is the same process as pairing resources used to obtain model input with resources used to obtain model labels during model training, and the pairing process will not be described again.

[0206] The terminal measures the first set of reference signal resources and uses the obtained measurement data as input to the first model to obtain the corresponding first inference result. The terminal then measures the second set of reference signal resources to obtain the measurement truth value (referred to as the first measurement truth value) corresponding to the first inference result. It is understood that there are multiple inference resources and monitoring resources; therefore, the inference result obtained based on each inference resource corresponds to a measurement truth value. Similar to the model training process, the monitoring resource corresponding to the inference resource can be determined based on the inference resource, or vice versa; there is no limitation. It is understood that the correspondence between the inference resource and the monitoring resource is determined based on the first offset, thus the inference result determined based on the inference resource also corresponds to the monitoring result obtained based on the monitoring resource. The terminal calculates the monitoring index based on the first inference result and the first measurement truth value to obtain the first monitoring result.

[0207] Furthermore, the first set of reference signal resources is any one of the inference resources in model monitoring, and the second set of reference signal resources is any one of the monitoring resources. Accordingly, the first inference result and the first monitoring result are for a single model monitoring session. Any single model monitoring session can use the method provided in this application to pair inference resources and monitoring resources, and then calculate monitoring results based on the paired inference resources and monitoring resources to evaluate model performance.

[0208] In existing solutions, two sets of reference signal resources are distributed by the network side during the model training or monitoring phases. However, the pairing of these two sets depends on the terminal side's implementation. In contrast, in this application's solution, the network side not only configures two sets of reference signal resources but also explicitly indicates the offset between these sets to the terminal side. This ensures that the pairing of reference signal resource sets is aligned between the network and the terminal side during model training or monitoring. Consequently, the network side can have a more comprehensive understanding of the inference results, and during the model monitoring phase, the scientific rigor of the model performance monitoring mechanism is improved, leading to more accurate model performance monitoring.

[0209] In the technical solution of this application, after the terminal side completes the model training, it can save the offset between the two sets of reference signal resources as the attribute information of the model.

[0210] Taking model training as an example, in different application scenarios, the network indicates different offsets to the terminal. In each application scenario, the terminal pairs the two sets of reference signal resources based on the offsets indicated by the network, and trains the model based on the paired input data and labels to obtain the corresponding model. Thus, each model can have a corresponding (or associated) offset. The offset can serve as one of the model's attribute information, reflecting the pairing relationship between the sets of reference signal resources during model training.

[0211] Optionally, the terminal can train the model using the method for training the first model described in method 200 above, thereby obtaining multiple models, each with an offset. Furthermore, each model may also have other attribute information besides the offset. For example, in BM case 1, the other attribute information of each model may include one or more of the following: the number of reference signal resources contained in set B, and the number of reference signal resources contained in set A. As another example, in BM case 2, the other attribute information of each model may include one or more of the following: the size of the observation time window, the size of the prediction time window, etc.

[0212] Optionally, multiple models stored on the terminal can also be associated with application scenarios. For example, model #1 corresponds to offset #1, model #2 corresponds to offset #2, and offset #1 is less than offset #2. Model #1 corresponds to a high-speed movement scenario, and model #2 corresponds to a low-speed movement scenario. As another example, both model #1 and model #4 correspond to high-speed movement scenarios, but model #1 can correspond to both high-speed movement and line-of-sight (LOS) scenarios, while model #4 can correspond to both high-speed movement and non-line-of-sight (NLOS) scenarios. In Table 2, the scenario #1 corresponding to model #3 and the scenario #2 corresponding to model #5 can represent other defined scenarios without limitation. Optionally, LOS or NLOS can also be stored as one of the other attribute information of the model, rather than as an application scenario.

[0213] In addition, taking model training as an example, models #1 and #4 also have other attribute information, such as the number of reference signal resources (e.g., beams) in set B and set A under BM case 1.

[0214] Table 2

[0215] The terminal side stores the offset between the reference signal resource sets as one of the model's attribute information. During the model inference phase, the network side queries the terminal side's model using the offset to enhance the network side's interpretation of the inference results, as shown in Figure 11.

[0216] Figure 11 is a schematic flowchart illustrating the application of the management model method provided in this application to the model inference stage. It should be understood that the process shown in Figure 11 can be an optional operation in the inference stage after model training is completed based on method 200. Furthermore, model training generally refers to the training of one or more models; for example, multiple models can be trained using method 200 provided in this application. This results in multiple models being obtained on the terminal side, with each model's attribute information including its corresponding offset. In addition, each model may also have other attribute information besides the offset.

[0217] Based on the correspondence between models and offsets, the network side can query the offset corresponding to a certain model during the model inference stage, as in case 1; or query models that meet the conditions based on the offset, as in case 2.

[0218] Case 1

[0219] 310. The network side sends a first query message, which is used to request the offset corresponding to the first model.

[0220] The terminal receives the first query message.

[0221] As an example, the first query message may carry an identifier for the first model and an offset field to indicate a query for the "offset" of one of the attributes of the first model. For example, the offset field may be set to 1 bit; when 1 bit is set to 0, it indicates that the network side is requesting a query for the offset.

[0222] The terminal queries the attribute information of the first model to determine the first offset corresponding to the first model.

[0223] 320. The terminal sends a response message to the first query message (denoted as the first response message), wherein the first response message indicates the first offset.

[0224] The network side receives the first response message.

[0225] Optionally, the first query message may also be used to request other attribute information of the first model besides the offset. In this case, the first response message may also indicate other attribute information of the first model. Optionally, the other attribute information may be one or more items.

[0226] Optionally, step 330 is also included.

[0227] 330. The network side stores the correspondence between the first offset and the first model.

[0228] During the model inference phase, the network side obtains the corresponding offset of the model through querying and saves the correspondence, which can be used in the subsequent model monitoring phase, making model performance monitoring more scientific.

[0229] Case 2

[0230] 340. The network side sends a second query message, which is used to request a query for the model corresponding to the first offset.

[0231] The second query message indicates the first offset. As an example, the second query message can be the first information in method 200 above, which can explicitly indicate the first offset, or indicate information used to determine the first offset, such as rules or ranges, for the terminal side to filter out offsets that meet the conditions, as well as the model corresponding to each offset that meets the conditions.

[0232] Optionally, the second query message may also indicate one or more other attribute information besides the first offset.

[0233] The terminal receives the second query message and determines the first model corresponding to the first offset based on the second query message. Optionally, the terminal determines other attribute information requested by the network side.

[0234] 350. The terminal sends a second response message, which indicates the first model.

[0235] Optionally, the second response message may also indicate one or more other attribute information of the first model, i.e., respond to other attribute information requested by the network side.

[0236] After the query-feedback process in case 1 or case 2, the model inference-related process is executed.

[0237] 360. The terminal side and network side execute the model inference-related processes.

[0238] For example, the network side issues inference resources (set B#1 to set B#5 in Figure 8), the terminal side performs model inference based on the inference resources, and reports the inference results, etc.

[0239] Optionally, the query-feedback process in case 1 or case 2 can be categorized as part of the model inference process, or it can be considered as a process of aligning the model or its attribute information before model inference.

[0240] It is understandable that either scenario 1 or scenario 2 may occur during the model inference phase. For example, when scenario 1 occurs, method 300 includes steps 1 and 360 of scenario 1. For instance, the offset corresponding to the first model on the terminal side is unknown to the network side. In order for the network side to understand the offset between the reference signal resource sets during the model training phase and obtain more attribute information of the model to improve the comprehensiveness of the interpretation of the inference results, the network side can request the terminal side to query the first offset corresponding to the first model through the process of scenario 1. When scenario 2 occurs, method 300 includes steps 2 and 360 of scenario 2. For instance, the network side determines that the offset between the reference signal resource sets should not be greater than 15ms in a certain application scenario, but the network side does not know whether there is a model that meets the conditions on the terminal side. In this case, the network uses the process of scenario 2 to query the model on the terminal side. Optionally, in addition to the offset attribute, the network side can also indicate one or more other attributes to filter out the models that meet the requirements with finer granularity and improve the adaptability of the model to the inference requirements (e.g., distinguished by application scenarios, such as high-speed mobile scenarios and low-speed mobile scenarios).

[0241] In method 300, based on the concept proposed in this application of taking the offset between reference signal resource sets as one of the attributes of the model, the network side queries the model or the offset corresponding to the model on the terminal side based on the correspondence between the model and the offset during the model inference stage, which can enhance the network side's subsequent interpretation of the model inference results.

[0242] The following section introduces the application of using the model's offset as one of the model's attributes in the model monitoring phase, as shown in Figure 12.

[0243] Figure 12 is a schematic flowchart illustrating the application of the management model method provided in this application to the model monitoring phase.

[0244] 510. Obtain the first offset corresponding to the first model from the network side.

[0245] Step 510 is optional. If the network side knows the offset corresponding to the currently monitored model (e.g., the first offset corresponding to the first model, which may have been obtained through querying during the model inference phase), or if the network side has stored the correspondence between the offsets and models of the two sets of reference signal resources configured for model training during the model training phase, this step can be omitted. However, if the offset corresponding to the currently monitored model is unknown to the network side, the network side, during the model monitoring phase, first determines the offset corresponding to the currently monitored first model, for example, by obtaining the first offset corresponding to the first model through a query-feedback process similar to that used in the model inference phase.

[0246] 520. The network side generates first configuration information and second configuration information, wherein the first configuration information indicates the time-domain resources of the first reference signal resource set, and the second configuration information indicates the time-domain resources of the second reference signal resource set. The time-domain resources of the second reference signal resource set are determined based on the first reference signal resource set and the first offset.

[0247] Similar to the above, a set of reference signal resources used for inference is denoted as the first reference signal resource set, which can correspond to set B in Figure 9, such as any one of set B#1 to set B#5; a set of reference signal resources used for monitoring is denoted as the second reference signal resource set, which can correspond to the Monitoring RS in Figure 9, such as any one of RS#1 to RS#5. The offset between the time domain positions of the first reference signal set and its paired second reference signal resource set is called the first offset.

[0248] As an example, the network side generates first configuration information, which indicates the time-domain resources of the first reference signal resource set. Based on the first offset, the network side determines the time-domain resources of the second reference signal resource set paired with the first reference signal resource set, specifically: the time-domain position corresponding to the time-domain resource of the first reference signal resource set plus the first offset, as the time-domain position corresponding to the time-domain resource of the second reference signal resource, thereby generating the second configuration information.

[0249] 530. The network side sends the first reference signal resource set.

[0250] For example, corresponding to set B#1 to set B#5 in Figure 9.

[0251] 540. The network side transmits the second reference signal resource set.

[0252] For example, corresponding to monitoring RS#1 to RS#5 in Figure 9.

[0253] Taking a first offset of 5ms as an example, set B#1 corresponds to monitoring RS#1, set B#2 corresponds to monitoring RS#2, and so on.

[0254] The terminal measures the first reference signal resource set to obtain inference data; the terminal measures the second reference signal resource set to obtain the measurement truth value.

[0255] 550. The terminal side determines the second reference signal resource set corresponding to the first reference signal resource set based on the first offset.

[0256] For the first set of reference signal resources, such as set B#1 in Figure 9, the terminal side finds the matching monitoring RS based on the first offset, which is monitoring RS#1.

[0257] 560. The terminal side determines the monitoring results.

[0258] The terminal side obtains the first measurement data from measurement set B#1, and uses this first measurement data as input to the first model to obtain the first inference data. The terminal side then pairs the first inference data with the first measurement true value obtained based on monitoring RS#1. Based on the first inference data and the first measurement true value, the terminal side calculates the monitoring index and obtains the first monitoring result.

[0259] Similarly, the terminal side obtains the second inference data based on set B#2, and pairs the second inference data with the second measurement truth value obtained based on monitoring RS#2 to calculate the second monitoring result, and so on.

[0260] 570. The terminal side reports the monitoring results.

[0261] Optionally, 550-570 can also be replaced with:

[0262] 550' The terminal side reports the measurement results of the second reference signal resource set and the inference results obtained based on the first reference signal resource set.

[0263] 560' The network side determines the measurement result corresponding to the aforementioned reasoning result based on the first offset.

[0264] 570', The network side determines the monitoring results.

[0265] In Method 500, the network side and the terminal side align the model offsets to make the mechanism for selecting matching RS in model monitoring more scientific and the monitoring results more accurate, as shown in Figure 13.

[0266] Figure 13 is a schematic diagram of matching monitoring resources for inference results in the model monitoring stage of existing schemes and the scheme of this application.

[0267] During the model monitoring phase, inference results need to be matched with monitoring resources to calculate monitoring metrics. Regarding how to determine which monitoring resource a single inference result matches, existing model monitoring solutions typically use the time of the set BRS, the time of the inference report, or the reference resource associated with the inference report as a benchmark, searching for the monitoring resource closest in time for matching. For example, as shown in Figure 13, when the inference report is associated with only one inference resource, the associated inference resource is at the same time as the set BRS. However, when the inference report is associated with multiple inference resources (e.g., in BM case 2 above), when using the associated inference resource as a benchmark to determine the matching monitoring resource, the first or last inference resource among these multiple resources can be selected as the benchmark. Then, the monitoring resource closest in time to this benchmark inference resource is selected and matched with the inference result in that inference report. Typically, in BM case 1, one inference report is associated with one inference resource; in BM case 2, one inference report may be associated with one or more inference resources.

[0268] However, this application proposes a new mechanism for selecting monitoring resources that match the inference results, making the selected monitoring resources more closely matched with the inference results. Specifically, the applicable time of one inference result is used as the benchmark for finding the closest monitoring resource. The applicable time of the inference result is determined by calculating t__set B + offset, where t__set B represents the temporal location of the inference resource corresponding to the inference result, and offset represents the offset corresponding to the current monitoring model. Taking BM case 2 in Figure 6 as an example, offset can be the offset or interval between the start time of T2 and the end time of T1. If the network side does not align the offset with the terminal side during the model inference stage, alignment is required through a process of network side query and terminal side reporting. It can be seen that the existing solutions and the solution of this application use different benchmarks to determine the monitoring RS that matches the inference results.

[0269] Based on the above embodiments, it can be understood that the reason for using the `t__set B+offset` method to select a monitoring resource matching the inference result during the model monitoring phase is that the temporal location of the monitoring resource within the effective time of the inference result is the most natural and scientific matching relationship. This matching relationship is guaranteed during the model training phase by configuring the offset between the two sets of reference signal resources used to train the model. In other words, the method proposed in this application for finding the monitoring resource that best matches the inference result requires that the offset between the two sets of reference signal resources used to obtain the model's input data and labels be clearly defined during model training. Thus, during the model monitoring phase, when selecting a monitoring resource that matches the inference result, the terminal side uses the effective time of the inference result as a time reference and selects the monitoring resource closest to this time reference as the monitoring resource that best matches the inference result. Based on this monitoring resource and the inference result, the model's monitoring metrics are calculated, making model performance monitoring more scientific and accurate.

[0270] The following examples illustrate the technical solutions of this application.

[0271] Example 1

[0272] Figure 14 is a schematic diagram of an application example 1 of the technical solution of this application.

[0273] 1) Training Phase

[0274] 601. Configure the network side for the resources set A and set B used for training model #1, as well as the offset between set B and set A. For example, the offset configured on the network side is offset #1.

[0275] 602. The network side indicates the configured offset#1 and sends set A and set B.

[0276] 603. On the terminal side, measure set A and set B to obtain the input data and labels of the model respectively. Then, pair the input data and labels of the model according to offset #1 to train the model and obtain model #1.

[0277] 2) Reasoning stage

[0278] 604. The network side requests the model corresponding to offset#1.

[0279] Optionally, the query message may also include one or more other attribute information of the model to be queried.

[0280] 605. The terminal side determines that the model that matches offset#1 is model #1.

[0281] When the query message on the network side also includes other attribute information of the model, the terminal side obtains one or more other attribute information of the model corresponding to offset#1, in addition to the model corresponding to offset#1.

[0282] 606. The terminal side returns a response message to the network side, and the response message indicates the model #1 corresponding to offset #1.

[0283] Optionally, if the query message also includes other attribute information of the model, the response message will also include one or more of the other attribute information.

[0284] Optionally, 607, the network side records the correspondence between offset#1 reported by the terminal side and model#1.

[0285] Optionally, when the terminal side also reports other attribute information of the model, the network side can also record offset#1, other attribute information besides offset#1, and the correspondence between model#1.

[0286] 608. Model Reasoning.

[0287] In the above process, the network side can configure the query offset to be one or more, which can be candidates on the terminal side. In step 606, the terminal side selects a suitable offset (e.g., offset #1) and feeds it back to the network side. In steps 604 and 606, the terminal side and the network side can exchange information through radio resource control (RRC) signaling. In step 607, the network side can record the value of the offset as one of the attribute information of the terminal side model. In addition, in BM case 2 and CSI prediction applications, because the input RS set and output result of the model may not correspond one-to-one (e.g., the number of RS contained in the observation window and the prediction window are different), as an example, the offset can satisfy one of the following: 1) the offset corresponds to the time from the first reference signal resource set of the observation window to the first reference signal resource set of the prediction window; 2) the offset corresponds to the time from the last reference signal resource set of the observation window to the first reference signal resource set of the prediction window; 3) the offset corresponds to the time from the first reference signal resource set of the observation window to the last reference signal resource set of the prediction window; 4) the offset corresponds to the time from the last reference signal resource set of the observation window to the last reference signal resource set of the prediction window.

[0288] In Example 1, the network distributes offsets during the training phase, allowing the terminal to determine the correspondence between the model's input data and its labels, avoiding mismatched pairings. Furthermore, during the inference phase, an offset query and feedback process is added, providing the network with more attribute information about the model. This enhances the network's interpretation of the inference results and allows for the determination of the temporal location of monitored resources based on the model's offset during subsequent model monitoring, thereby improving the accuracy of model performance monitoring.

[0289] Example 2

[0290] Compared to Example 1, in Example 2, during the model training process, the network side sends out one or more rules or ranges to determine the offset, rather than an exact value.

[0291] As an example, the rules for determining the offset could be: select the nearest set A pair based on set B; select the nearest set B pair based on set A; set A needs to be paired later than set B.

[0292] As an example, the offset range can be: given an offset range, offsets between set B and set A that satisfy this range can be paired. Taking Table 1 above as an example, if the offset range sent by the network side is 0 to 30ms, then pairing set B#1 with set A#1 or set A#2 satisfies this range because the offsets are all within 30ms. As introduced above, the terminal side can choose an offset of 5ms or 25ms between set B and set A, so that set B and set A satisfy a one-to-one pairing relationship, and can indicate the offset determined by the terminal side to the network side.

[0293] Figure 15 is a schematic diagram of application example 2 of the technical solution of this application.

[0294] 1) Training Phase

[0295] 701. Configure the network side for the resources set A and set B used for model training, as well as the range of rules or offsets that need to be satisfied for pairing set B and set A.

[0296] 702. The network side instructs the determination of the rules or range of offset and sends set A and set B.

[0297] 703. The terminal side determines offset #1 according to the rules or range indicated by the network side, and pairs the input data and labels of the model obtained by measuring set A and set B according to offset #1, and performs model training to obtain model #1.

[0298] For details on the implementation of the rules or ranges for determining the offset, please refer to the description in step 210.

[0299] 2) Reasoning stage

[0300] 704. The network side requests a query for the model corresponding to one or more offsets (optionally, other attribute information of the model may also be included).

[0301] The following explanation uses an offset (represented as offset#1) as an example.

[0302] 705. The terminal side determines that the model that matches offset#1 is model #1.

[0303] 706. The terminal side returns a response message to the network side, and the response message indicates the model #1 corresponding to offset #1.

[0304] 707. The network side records the correspondence between offset#1 reported by the terminal side and model#1.

[0305] 708. Model Reasoning.

[0306] As an example, in step 701 above, the offset range is the range of offsets selected from set A based on set B, for example, -15 to 5ms. The terminal can select a set A from 15ms before to 5ms after set B according to the range requirements and match it, and then report the finally selected offset. In step 704, the offset requested by the network side can be one or more, as candidates for the terminal side; in step 705, the terminal side selects a suitable offset (e.g., offset #1) and feeds it back to the network side. Other processes are similar to those in Example 1 and will not be described again.

[0307] In Examples 1 and 2, the sending of set B and set A can be done periodically as shown in Figure 7, or in a non-periodic or semi-static manner, without limitation.

[0308] Example 3

[0309] Figure 16 is a schematic diagram of application example 3 of the technical solution of this application. The main process is as follows:

[0310] 801. At the start of the monitoring phase, the network side queries the offset corresponding to the running terminal-side model.

[0311] Assume the model currently running on the terminal side is model #1.

[0312] 802. The terminal side provides feedback offset#1, where offset#1 is the offset corresponding to model #1.

[0313] 803. Configure set B on the network side, and configure the monitoring RS according to offset#1 and set B. The time domain position corresponding to the time domain resource of the monitoring RS is t__set B+offset#1.

[0314] 804. The network side sends set B to monitor RS.

[0315] 805. The terminal side measures set B. Based on the measurement results of set B and model #1, the inference result #1 is obtained. The terminal side uses t__set B+offset #1 as the benchmark to find the monitoring RS that matches the inference result #1. The measurement truth value #1 obtained based on the monitoring RS is paired with the inference result #1.

[0316] 806. The terminal side calculates the monitoring indicators of the model based on the measured true value #1 and the inference result #1, and obtains the monitoring results.

[0317] 807. The terminal side reports the monitoring results.

[0318] Step 801 is only included when the offset is unknown on the network side during the inference phase. For steps 803 and 805, the time-domain position of the monitoring RS configured on the network side is ideally within the applicable time of the inference result calculated on the terminal side. In this case, the applicable time of inference result #1 and the offset of the matched monitoring RS should be 0.

[0319] The solution proposed in this application, in the model monitoring on the terminal side, adopts a dual strategy of pre-configuring offsets on the network side and determining the matching monitoring RS based on the offsets as the inference result on the terminal side. This ensures that the corresponding monitoring RS can be accurately matched based on the effect time of the inference result, thereby achieving more scientific monitoring of model performance.

[0320] Figure 17 illustrates an application example of the technical solution of this application under another possible architecture. The AI ​​model on the terminal side can be implemented on a chip inside the terminal device, or it can be deployed outside the terminal device, such as in the host of an OTT system or a cloud server. As shown in Figure 17, taking model monitoring as an example, the network queries the model's offset. After configuring set B, it configures the monitoring RS based on the offset. The network side issues set B and the monitoring RS. The terminal device measures set B and sends the measurement data to the OTT. The OTT uses the received measurement data as model input for inference and obtains the inference result (or prediction result). The OTT returns the inference result to the terminal device. The terminal device determines the monitoring RS that matches the inference result based on the offset corresponding to the model and measures the monitoring RS to obtain the measurement truth value. The process of monitoring the RS based on the offset is described in the above embodiment and will not be repeated here. Finally, after calculating the monitoring result, the terminal device reports the monitoring result to the network side.

[0321] The above provides a detailed description of the management model method provided in this application. The following describes the corresponding communication device.

[0322] Figure 18 is a schematic block diagram of the communication device 1000 provided in this application. As shown in Figure 18, the communication device 1000 may include a processing module 1001 and a communication module 1002. The communication device 1000 may be a terminal device, or a communication device applied to or used in conjunction with a terminal device to achieve the corresponding functions of the terminal device, such as a processor, chip, circuit, or AI entity. Alternatively, the communication device 1000 may be a network device, or a communication device applied to or used in conjunction with a network device to achieve the corresponding functions of the network device, such as a processor, chip, circuit, or AI entity.

[0323] The communication module can also be called a transceiver module, transceiver, transceiver machine, or transceiver device. The processing module can also be called a processor, processing board, processing unit, or processing device. Optionally, the communication module is used to perform the sending and receiving operations of the terminal-side device or network-side device in the above method. The device in the communication module that implements the receiving function can be regarded as a receiving unit, and the device in the communication module that implements the sending function can be regarded as a sending unit. That is, the communication module includes a receiving unit and a sending unit. When the communication device 1000 is applied to the network-side device or terminal-side device, the processing module 1001 can be used to implement the processing functions of the network-side device or terminal-side device in the embodiments of Figures 10 to 17, and the communication module 1002 can be used to implement the sending and receiving functions of the network-side device or terminal-side device. For example, when applied to the network side, the processing module 1001 can be used to: determine first information, determine first configuration information, determine second configuration information, etc.; the communication module 1002 can be used to: send first information, send a first query message, receive a first response message, send a second query message, receive a second response message, etc. When the communication device 1000 is applied to a terminal-side device, the communication module 1002 can be used to: receive first information, receive a first query message, send a first response message, receive a second query message, receive a second response message, etc.; the processing module 1001 can be used to: train a first model, monitor the first model, determine a first offset, etc. Optionally, when the communication device 1000 is applied to a network side or a terminal side, it may further include a storage module for storing attribute information of the first model, including the first offset corresponding to the first model.

[0324] Furthermore, it should be noted that the aforementioned communication module and / or processing module can be implemented through virtual modules. For example, the processing module can be implemented through software functional units or virtual devices, and the communication module can be implemented through software functions or virtual devices. Alternatively, the processing module or communication module can also be implemented through physical devices. For example, if the device is implemented using a chip / chip circuit, the communication module can be an input / output circuit and / or a communication interface, performing input operations (corresponding to the aforementioned receiving operation) and output operations (corresponding to the aforementioned sending operation); the processing module is an integrated processor, microprocessor, or integrated circuit.

[0325] The module division in this application is illustrative and represents only one logical functional division. In actual implementation, other division methods are possible. Furthermore, the functional modules in the various examples of this application can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware, as software functional modules, or a combination of hardware and software.

[0326] Figure 19 is a schematic block diagram of another communication device 1100 provided in this application. Optionally, the communication device 1100 may be a chip or a chip system. Optionally, in this application, the chip system may be composed of chips or may include chips and other discrete devices.

[0327] The communication device 1100 can be used to implement the functions of any of the network elements (e.g., network-side devices or terminal-side devices) described in the foregoing embodiments. The communication device 1100 may include at least one processor 1110. Optionally, the processor 1110 is coupled to a memory, which may be located within the communication device 1100, integrated with the processor, or located outside the communication device 1100. As an example, the communication device 1100 may also include at least one memory 1120. The memory 1120 stores the necessary computer programs (or computer instructions) and / or data for implementing the corresponding functions of any of the network elements in any of the above method embodiments; the processor 1110 may execute the computer programs stored in the memory 1120 to complete the methods implemented by any of the network elements in any of the above method embodiments.

[0328] The communication device 1100 may also include a communication interface 1130, through which the communication device 1100 can interact with other devices. For example, the communication interface 1130 may be a transceiver, circuit, bus, module, pin, or other type of communication interface. When the communication device 1100 is a chip-based device or circuit, the communication interface 1130 in the device 1100 may also be an input / output circuit, capable of inputting information (or receiving information) and outputting information (or sending information). The processor may be an integrated processor, a microprocessor, an integrated circuit, or a logic circuit, and the processor can determine the output information based on the input information.

[0329] The coupling in this application refers to indirect coupling or communication connection between devices, units, or modules, which can be electrical, mechanical, or other forms, used for information exchange between devices, units, or modules. The processor 1110 may operate in conjunction with the memory 1120 and the communication interface 1130. This application does not limit the specific connection medium between the processor 1110, the memory 1120, and the communication interface 1130.

[0330] Optionally, as shown in FIG19, the processor 1110, the memory 1120, and the communication interface 1130 are interconnected via a bus 1140. The bus 1140 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one line is used to represent the bus 1140 in FIG19, but this does not indicate that there is only one bus or one type of bus.

[0331] Figure 20 is a schematic structural diagram of the chip provided in this application. Chip 30 includes a processing circuit 31 and a communication circuit 32. The processing circuit 31 can be a logic circuit, integrated circuit, etc., and the communication circuit 32 can be an input / output circuit, input / output interface, interface circuit, etc., capable of inputting information (or receiving information) or outputting information (or sending information). Chip 30 can execute the methods performed by the network-side device or the terminal-side device in the various embodiments of this application. The processing circuit 31 can be one or more processors, or all or part of the circuitry used for control or processing in one or more processors. Optionally, the functions on the terminal side or the network side can be deployed in different parts of the chip.

[0332] It is understood that the processor in the embodiments of this application may be one or more of the following devices, or all or part of the circuitry of the following devices for processing functions: a central processing unit (CPU), a processor for AI, or other general-purpose processors, digital signal processors (DSPs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor may be a microprocessor or any conventional processor.

[0333] For example, the processor used for AI can be one or more of the following: graphics processing unit (GPU), neural processing unit (NPU), tensor processing unit (TPU), and data processing unit (DPU).

[0334] For example, one possible implementation of a processor for AI could be the AI ​​processor 2100 shown in Figure 21.

[0335] Figure 21 is a schematic diagram of the structure of the AI ​​processor provided in an embodiment of this application. As shown in Figure 21, the AI ​​processor 2100 may include one or more of the following: an AI core, a digital vision pre-processing (DVPP) module, a task scheduler (TS), an L3 cache, an AI CPU, a control CPU, an L2 cache, a universal serial bus (USB) interface, a network card, a peripheral component interconnect express (PCIe) interface (PCIe is a high-speed serial computer expansion bus standard), a double data rate (DDR) / high bandwidth memory (HBM) interface, a generational purpose input / output (GPIO) / inter-integrated circuit (I2C) bus, etc. It is understood that the specific meanings of these terms are well known to those skilled in the art and will not be elaborated here.

[0336] In addition, this application also provides a computer-readable storage medium storing computer instructions, which, when executed on a computer, cause operations and / or processes performed by a terminal-side device or a network-side device in the various method embodiments of this application to be executed.

[0337] This application also provides a computer program product, which includes computer program code or instructions. When the computer program code or instructions are run on a computer, the operations and / or processes performed by the terminal-side device or the network-side device in the various method embodiments of this application are executed.

[0338] This application also provides a chip including a processor, and a memory for storing a computer program is provided independently of the chip. The processor executes the computer program stored in the memory, such that operations and / or processes performed by a terminal-side device or a network-side device in any method embodiment are executed. Further, the chip may also include a communication interface. The communication interface may be an input / output interface or an interface circuit, etc. Further, the chip may also include a memory.

[0339] This application also provides a chip, which may include circuitry and an input / output interface. The circuitry may be logic circuitry, integrated circuits, etc., and exemplaryly, the circuitry may be one or more processors, or all or part of the circuitry in one or more processors used to implement one or more processing, control, or computing functions. The input / output interface may also be an input / output circuit, or an interface circuit, capable of inputting information (or receiving information) and / or outputting information (or sending information). The chip may include a chip system. Optionally, the chip system may be composed of chips or may include chips and other discrete devices. The chip can be used to execute the methods implemented by terminal-side devices or network-side devices in the various embodiments of this application. Optionally, the chip may be a baseband chip, also known as a modem.

[0340] Furthermore, this application provides a communication system, including a terminal-side device and a network-side device as described in any embodiment of this application. This communication system can implement the training model method provided in any of the embodiments shown in Figures 10 to 17.

[0341] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0342] The processor in this application embodiment has signal processing capabilities and can be a central processing unit (CPU), or a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. It can implement or execute the methods, steps, and logic block diagrams disclosed in this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in this application can be directly embodied in the execution of the hardware processor, or executed by a combination of hardware and software modules within the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0343] In the embodiments of this application, memory is any other medium capable of carrying or storing desired program code in the form of instructions or data structures, and accessible by a computer, but is not limited thereto. The memory in this application can also be a circuit or any other means capable of implementing a storage function for storing computer programs and / or data; or, it can also be a circuit or any other means capable of implementing a storage function for storing computer programs and / or data. As an example, memory can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which serves as an external cache. By way of example, but 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 linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, the types described above or any other suitable types of memory.

[0344] The technical solutions provided in this application can be implemented in whole or in part through software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented in whole or in part as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a terminal device, an access network device, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video disc (DVD)), or a semiconductor medium, etc.

[0345] In the embodiments of this application, "at least one" refers to one or more items. "More than one" means two or more items. "And / or" is used to describe the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0346] The term "comprising" and any variations thereof used in the embodiments of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0347] In this application, examples may reference each other without logical contradiction. For example, methods and / or terms between method embodiments may reference each other, functions and / or terms between device embodiments may reference each other, and functions and / or terms between device examples and method examples may reference each other.

[0348] Those skilled in the art will recognize that the units and algorithm steps of the various examples 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 implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0349] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0350] The units described as separate components may or may not be physically separate. 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 the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0351] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0352] If the aforementioned functions are implemented as 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 this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0353] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for a management model, characterized in that, include: Receive first information from the network side, the first information is used to determine a first offset, the first offset is the offset between a first reference signal resource set and a second parameter signal resource set; At least one of training or monitoring the first model is performed, wherein the dataset used for training the first model includes input data and labels of the first model obtained based on the first reference signal resource set and the second reference signal resource set; or the dataset used for monitoring the first model includes the measured true value of the first model and the inference data corresponding to the measured true value obtained based on the first reference signal resource set and the second reference signal resource set; and one of the first reference signal resource set and the second reference signal resource set is determined based on the other and the first information.

2. The method according to claim 1, characterized in that, One of the first reference signal resource set and the second reference signal resource set is determined based on the other and the first information, including: The second set of reference signal resources is determined from multiple candidate sets of second reference signal resources based on the first reference signal resources and the first information.

3. The method according to claim 1 or 2, characterized in that, The first information indicates the first offset.

4. The method according to claim 1 or 2, characterized in that, The first information indicates a rule for determining the first offset, the rule including the following: The second reference signal resource set is determined based on the first reference signal resource set, wherein the second reference signal resource set is the one among a plurality of candidate second reference signal resource sets that is closest to the first reference signal resource set in the time domain; or... The first reference signal resource set is determined based on the time-domain location of the second reference signal resource set, wherein the first reference signal resource set is the second reference signal resource set that is closest in time-domain location among a plurality of candidate first reference signal resource sets; or... The second reference signal resource set is determined based on the first reference signal resource set, wherein the time-domain position of the second reference signal resource set is no earlier than the time-domain position of the first reference signal resource set; or, The first reference signal resource set is determined based on the second reference signal resource set, wherein the time domain position of the second reference signal resource set is no earlier than the time domain position of the first reference signal resource set.

5. The method according to claim 1 or 2, characterized in that, The first information indicates the range to which the first offset belongs, wherein the offset between the time-domain position of the first reference signal resource set and the time-domain position of the second reference signal resource is less than or equal to the range to which the first offset belongs.

6. The method according to any one of claims 1-5, characterized in that, The first reference signal resource set is used to acquire the input data of the first model, and the second reference signal resource set is used to acquire the labels of the first model; And, the training or monitoring of the first model includes: The first model is trained based on the first reference signal resource set and the second reference signal resource set, wherein the first measurement data obtained based on the first reference signal resource set is one of the input data of the first model, and the second measurement data obtained based on the second reference signal resource set is the label corresponding to the first measurement data.

7. The method according to any one of claims 1-5, characterized in that, The first reference signal resource set is used to acquire the inference data of the first model, and the second reference signal resource set is used to acquire the measurement truth value of the model monitoring. And, the training or monitoring of the first model includes: Based on the first reference signal set and the second reference signal resource set, the first model is monitored, wherein: First inference data is obtained based on the first measurement data obtained from the first reference signal resource set and the first model; Based on the second reference signal resource set, the first measurement true value is obtained; Based on the first inference data and the first measurement true value, obtain the first monitoring result.

8. The method according to any one of claims 1-7, characterized in that, The method further includes: Save the attribute information of the first model, the attribute information including the first offset corresponding to the first model.

9. The method according to claim 8, characterized in that, The method further includes: Receive a first query message, which is used to request a query for the offset corresponding to the first model; Send a first response message in response to the first query message, the first response message indicating the first offset.

10. The method according to claim 9, characterized in that, The method further includes: The system receives first configuration information of the first reference signal resource set and second configuration information of the second reference signal resource set. The first configuration information indicates the time domain position of the first reference signal resource set, and the second configuration information indicates the time domain position of the second reference signal resource set. The time domain position of the second reference signal resource set is determined based on the time domain position of the first reference signal resource set and the first offset.

11. The method according to any one of claims 8-10, characterized in that, The first model is one of multiple models on the terminal side, and each of the multiple models has its corresponding offset.

12. The method according to claim 11, characterized in that, The method further includes: Receive a second query message from the network side, the second query message being used to request a query for the model corresponding to the first offset; Determine the first model corresponding to the first offset from the plurality of models; and, Send a second response message in response to the second query message, the second response message indicating the first model.

13. The method according to claim 11 or 12, characterized in that, The first model and the second model in the plurality of models both correspond to the first offset, and one or more of the other attribute information of the first model and the second model are different from the first offset.

14. The method according to claim 12 or 13, characterized in that, The second response message also indicates one or more of the other attribute information of the first model.

15. A method for a management model, characterized in that, include: First information is determined, which is used to determine a first offset, wherein the first offset is the offset between a first reference signal resource set and a second parameter signal resource set; as well as, Send the first information, wherein: The first reference signal resource set and the second reference signal resource set are used for training the first model. The dataset used for training the first model includes the input data and labels of the first model obtained based on the first reference signal resource set and the second reference signal resource set, or... The first reference signal resource set and the second reference signal resource set are used for monitoring the first model. The dataset used for monitoring the first model includes the measured true value of the first model and the inference data corresponding to the measured true value, obtained based on the first reference signal resource set and the second reference signal resource set. Furthermore, one of the first reference signal resource set and the second reference signal resource set is determined based on the other and the first offset.

16. The method according to claim 15, characterized in that, One of the first reference signal resource set and the second reference signal resource set is determined based on the other and the first information, including: The second set of reference signal resources is determined from multiple candidate sets of second reference signal resources based on the first reference signal resources and the first information.

17. The method according to claim 15 or 16, characterized in that, The first information indicates the first offset.

18. The method according to claim 15 or 16, characterized in that, The first information indicates a rule for determining the first offset, the rule including the following: The second reference signal resource set is determined based on the first reference signal resource set, wherein the second reference signal resource set is the one among a plurality of candidate second reference signal resource sets that is closest to the first reference signal resource set in the time domain; or... The first reference signal resource set is determined based on the time-domain location of the second reference signal resource set, wherein the first reference signal resource set is the second reference signal resource set that is closest in time-domain location among a plurality of candidate first reference signal resource sets; or... The second reference signal resource set is determined based on the first reference signal resource set, wherein the time-domain position of the second reference signal resource set is no earlier than the time-domain position of the first reference signal resource set; or, The first reference signal resource set is determined based on the second reference signal resource set, wherein the time domain position of the second reference signal resource set is no earlier than the time domain position of the first reference signal resource set.

19. The method according to claim 15 or 16, characterized in that, The first information indicates the range to which the first offset belongs, wherein the offset between the time-domain position of the first reference signal resource set and the time-domain position of the second reference signal resource is less than or equal to the range to which the first offset belongs.

20. The method according to any one of claims 15-19, characterized in that, The first reference signal resource set is used to acquire the input data of the first model, and the second reference signal resource set is used to acquire the labels of the first model, wherein, The first measurement data obtained based on the first reference signal resource set is one of the input data of the first model, and the second measurement data obtained based on the second reference signal resource set is the label corresponding to the first measurement data.

21. The method according to any one of claims 15-19, characterized in that, The first reference signal resource set is used to acquire the inference data of the first model, and the second reference signal resource set is used to acquire the measurement truth values ​​monitored by the model. The first inference data obtained based on the first reference signal resource set corresponds to the first measurement truth value obtained based on the second reference signal resource set. The first inference data and the first measurement truth value are used to obtain the first monitoring result.

22. The method according to any one of claims 15-21, characterized in that, The method further includes: Send a first query message, which is used to request the offset corresponding to the first model; Receive a first response message in response to the first query message, the first response message indicating the first offset.

23. The method according to claim 22, characterized in that, The method further includes: The attribute information of the first model is saved, including the first offset corresponding to the first model.

24. The method according to claim 23, characterized in that, The method further includes: First configuration information of the first reference signal resource set is determined, wherein the first configuration information indicates the time domain location of the first reference signal resource set; A second configuration information for the second reference signal resource set is determined, the second configuration information indicating the time domain position of the second reference signal resource set, wherein the time domain position of the second reference signal resource set is determined based on the time domain position of the first reference signal resource set plus the first offset; Send the first configuration information and the second configuration information.

25. The method according to any one of claims 15-24, characterized in that, The method further includes: Send a second query message, which is used to request a query for the model corresponding to the first offset; Receive a second response message in response to the second query message, the second response message indicating the first model.

26. The method according to claim 25, characterized in that, The second query message also requests one or more other attribute information. The second response message also indicates one or more of the other attribute information of the first model besides the first offset.

27. A method for a management model, characterized in that, Applied to the terminal side, including: Receive a first query message, which requests the offset corresponding to the first model; and, Send a first response message in response to the first query message, the first response message indicating the first offset corresponding to the first model. Wherein, the first offset is the offset between the first reference signal resource set and the second reference signal resource set; the dataset used for training the first model includes the input data and labels of the first model obtained based on the first reference signal resource set and the second reference signal resource set, or the dataset used for monitoring the first model includes the measured true value of the first model and the inference data corresponding to the measured true value obtained based on the first reference signal resource set and the second reference signal resource set.

28. The method according to claim 27, characterized in that, The first model is one of multiple models on the terminal side, and each of the multiple models corresponds to an offset.

29. A method for a management model, characterized in that, Applied to the terminal side, including: Receive a second query message, which is used to request a query for the model corresponding to the first offset; Determine the first model corresponding to the first offset from multiple models; and, Send a second response message in response to the second query message, the second response message indicating the first model; Wherein, the first offset is the offset between the first reference signal resource set and the second reference signal resource set; the dataset used for training the first model includes the input data and labels of the first model obtained based on the first reference signal resource set and the second reference signal resource set, or the dataset used for monitoring the first model includes the measured true value of the first model and the inference data corresponding to the measured true value obtained based on the first reference signal resource set and the second reference signal resource set.

30. The method according to claim 29, characterized in that, The second response message also indicates one or more other attribute information of the first model, in addition to the first offset.

31. A method for a management model, characterized in that, Applied to the terminal side, including: Send a first query message, the first query message being used to request a query for the offset corresponding to the first model; and, Receive a first response message in response to the first query message, wherein the first response message indicates a first offset corresponding to the first model; Wherein, the first offset is the offset between the first reference signal resource set and the second reference signal resource set; the dataset used for training the first model includes the input data and labels of the first model obtained based on the first reference signal resource set and the second reference signal resource set, or the dataset used for monitoring the first model includes the measured true value of the first model and the inference data corresponding to the measured true value obtained based on the first reference signal resource set and the second reference signal resource set.

32. The method according to claim 31, characterized in that, The first model is one of multiple models on the terminal side, and each of the multiple models corresponds to an offset.

33. A method for a management model, characterized in that, Applied to the terminal side, including: Send a second query message, which requests a query for the model corresponding to the first offset; and, Receive a second response message in response to the second query message, the second response message indicating the first model; Wherein, the first offset is the offset between the first reference signal resource set and the second reference signal resource set; the dataset used for training the first model includes the input data and labels of the first model obtained based on the first reference signal resource set and the second reference signal resource set, or the dataset used for monitoring the first model includes the measured true value of the first model and the inference data corresponding to the measured true value obtained based on the first reference signal resource set and the second reference signal resource set.

34. The method according to claim 33, characterized in that, The second response message also indicates one or more other attribute information of the first model, in addition to the first offset.

35. A communication device, characterized in that, It includes modules or units for performing the method as described in any one of claims 1-14, 27-30; or, it includes modules or units for performing the method as described in any one of claims 15-26, 31-34.

36. A communication device, characterized in that, The device includes a processor coupled to a memory, the processor being configured to execute a computer program or instructions stored in the memory to cause the communication device to perform the method as described in any one of claims 1-14, 27-30, or to perform the method as described in any one of claims 15-26, 31-34.

37. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed on a computer, implement the method as described in any one of claims 1-14, 27-30, or the method as described in any one of claims 15-26, 31-34.

38. A computer program product, characterized in that, When the computer program product is run, it causes the method as described in any one of claims 1-14, 27-30 to be implemented, or causes the method as described in any one of claims 15-26, 31-34 to be implemented.

39. A communication system, characterized in that, Includes means for performing the method as described in any one of claims 1-14, 27-30; or includes means for performing the method as described in any one of claims 15-26, 31-34.

Images