WO2024067248A1 - 一种获取训练数据集的方法和装置 - Google Patents

一种获取训练数据集的方法和装置 Download PDF

Info

Publication number
WO2024067248A1
WO2024067248A1 PCT/CN2023/119695 CN2023119695W WO2024067248A1 WO 2024067248 A1 WO2024067248 A1 WO 2024067248A1 CN 2023119695 W CN2023119695 W CN 2023119695W WO 2024067248 A1 WO2024067248 A1 WO 2024067248A1
Authority
WO
WIPO (PCT)
Prior art keywords
groups
reference signals
information
training
model
Prior art date
Application number
PCT/CN2023/119695
Other languages
English (en)
French (fr)
Inventor
刘礼福
孙琰
陈宏智
庞继勇
李�远
邹菲菲
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from CN202211247927.6A external-priority patent/CN117851819A/zh
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Publication of WO2024067248A1 publication Critical patent/WO2024067248A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting

Definitions

  • the embodiments of the present application relate to the field of communications, and more specifically, to a method and device for obtaining a training data set.
  • AI models can be deployed on training devices (e.g., terminal devices) for training and updating.
  • the network device will continue (e.g., periodically) transmit the training data set to the training device in order to support the training of the AI model by the training device.
  • the network device will not stop sending the training data set to the training device until the training device sends an indication message to the network device that the training model has been completed.
  • the training data set transmitted by the network device to the training device is unnecessary for the training device and will cause a waste of resources. Therefore, when the training device is training the AI model, how to obtain the training data set becomes a technical problem that needs to be solved.
  • the embodiment of the present application provides a method for obtaining a training data set, which can reduce air interface resource waste and air interface overhead, and improve the utilization performance of air interface resources.
  • a method for obtaining a training data set is provided, which can be performed by a training device, or can also be performed by a component (such as a chip or circuit) of the training device, without limitation.
  • the training device can be a terminal device.
  • the method includes: sending first information to a network device, the first information being used to indicate relevant information of a first training data set that a training device requests the network device to send; receiving the first training data set from the network device, the first training data set being a training data set based on the relevant information indicated by the first information, and the first training data set being used for training an artificial intelligence AI model.
  • the training device can request the network to send a training data set, and the request information also indicates the relevant information of the first training data set that the training device needs to be sent to it by the network device.
  • the training device can indicate to the network device which training data sets are needed, and the network device can send the training data set indicated by the training device to the training device, without having to send the training data set all the time. This method can reduce the waste of air interface resources and air interface overhead, and improve the utilization performance of air interface resources.
  • the relevant information includes at least one of the following: information about the size of the first training data set, configuration information of the input of the AI model, and configuration information of a reference signal used for training the AI model.
  • the first information can indicate information about the size of the first training data set, configuration information of the input of the AI model, and configuration information of the reference signal used for training the AI model, thereby displaying or implicitly indicating the size of the first training data set required by the training device, so that the network device can issue the first training data set based on the instruction of the training device, thereby improving the utilization rate of air interface resources.
  • the information about the size of the first training data set is determined by the training device based on the size of the training data set required to complete the training of the AI model.
  • the training device can determine the total number of training data sets required when training the AI model from the initial state of the AI model (for example, the initial state of the AI model is 0) to the convergence state through historical information. For example, the training device determines based on historical experience that a total of 60,000 scans of the full codebook are required to obtain a training data set for training the AI model.
  • the training device can determine the training data set required to train the AI model based on historical experience.
  • the number of training data sets is determined and indicated to the network device, so that the network device sends the training data set based on the instruction, which can reduce the waste of air interface resources.
  • the method before sending the first information to the network device, the method also includes: determining a first performance of the AI model; determining information about the size of the first training data set based on the first performance of the AI model and the second performance of the AI model, wherein the first performance is the current performance of the AI model and the second performance is the target performance of the AI model.
  • the training device monitors the AI model, it compares the performance of the current AI model monitored with historical information and the model performance corresponding to the time when the AI model converges, and according to the performance of the current AI model, it can estimate the size of the first training data set required to achieve the expected model performance.
  • the training device can estimate the number of training data sets required when training the AI model to a convergence state based on the performance of the AI model and indicate it to the network device, so that the network device can send the training data set based on the indication, thereby reducing the waste of air interface resources.
  • the configuration information of the reference signal includes at least one of the following: an identifier of the reference signal, a time domain resource of the reference signal, a frequency domain resource of the reference signal, a transmission period of the reference signal, and a type of the transmitted reference signal.
  • the type of the reference signal is SSB, or CSI-RS, or SRS, etc.
  • the identifier of the reference signal can also be understood as the identifier of the reference signal group.
  • the configuration information of the reference signal includes the group identifiers of N groups of reference signals (N is an integer greater than or equal to 1), wherein each group of reference signals in the N groups of reference signals has the same group identifier, and each group of reference signals includes at least one reference signal.
  • the time domain resources of the reference signal, the frequency domain resources of the reference signal, the transmission period of the reference signal, and the type of the transmitted reference signal can also be understood as the time domain resources of N groups of reference signals, the frequency domain resources of N groups of reference signals, the transmission period of N groups of reference signals, and the type of the transmitted reference signals.
  • the training device can determine the configuration information of the reference signal based on historical information, and the configuration information of the reference signal can also indirectly indicate the number of required training data sets.
  • “configuration information of the input of the AI model” can be understood as, for example, that the training device determines the input information of the AI model based on the historical information of the AI model training, for example, the input information of the AI model is the measurement result of the reference signal corresponding to the sparse beam pattern.
  • the training device can then determine which positions in the full codebook the sparse beam pattern is the beam of.
  • the training device can report information such as the identifier of the sparse beam pattern, the identifier of the reference signal corresponding to the sparse beam pattern, or the measurement result of the reference signal to the first network device, so that the first network device can send down a training data set corresponding to the beam pattern.
  • the training device can determine the input configuration information of the AI model based on historical information, and the input configuration information of the AI model can also indirectly indicate the number of training data sets required.
  • the first information includes at least one of the following: identification information of the AI model, information on the application scenario of the AI model, usage information of the AI model, and information on the computing power capability of the training device.
  • a first mapping relationship may be stored on a network device, where the first mapping relationship is a correspondence between the identifier of each AI model and the size of the training data set corresponding to the AI model identifier.
  • the training device can indicate the size of the training data set required to train the AI model by indicating the identifier of the AI model to the network device.
  • mapping relationship in this application can also be expressed as an “association relationship” or a “correspondence relationship”. It should be understood that the "mapping relationship” in the embodiments of this application can be saved or recorded in the form of a functional relationship, a table, or a mapping relationship. In the following embodiments, the "mapping relationship" mentioned can be configured by a network device, or can be predefined by a protocol, etc., without limitation.
  • the application scenario of the AI model or the purpose of the AI model can be understood as that the AI model is used in a beam management scenario, a CSI feedback scenario, a positioning scenario, etc.
  • the training device can indicate the size of the training data set required to train the AI model by indicating the application scenario of the AI model or the purpose of the AI model to the network device.
  • the training device may also report computing power capabilities.
  • the computing power capabilities reported by the training device include at least one of the following: the processor of the training device (for example, a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), a neural network processor, etc.)
  • the capacity of the training device (neural network processing unit, NPU), field-programmable gate array (FPGA), etc.), the size of the training device storage space, the size of the training device memory, the power of the training device, etc. are not limited.
  • the training device can indicate the maximum number of training data sets that can be processed when training the AI model.
  • the method also includes: training the AI model based on the first training data set and determining the performance of the AI model; sending second information to the network device based on the performance of the AI model, the second information being used to indicate relevant information of the second training data set requested by the training device to be sent by the network device; receiving the second training data set from the network device, the second training data set being a training data set based on the relevant information indicated by the second information, and the second training data set being used for training the AI model.
  • the amount of data in the second training data set may be smaller than the amount of data in the first training data set.
  • the training device may still train the AI based on the second training data set and iterate repeatedly, for example, perform model training based on the second training data set, determine the performance of the AI model again, determine the size of the required third training data set based on the performance of the AI model, and so on.
  • the training device can perform L (L is an integer greater than 1) training times until the training device determines that the AI model has converged (“model convergence" can also be understood as the AI model reaching the target performance).
  • the training device can determine the size of the training data set required for the next training by measuring the performance of the AI model training, thereby improving the efficiency of the AI model training and improving the utilization performance of air interface resources.
  • a method for obtaining a training data set is provided.
  • the method can be executed by a first network device, or can also be executed by a component (such as a chip or circuit) of the first network device, without limitation.
  • the method includes: receiving first information from a training device, the first information being used to indicate relevant information of a first training data set requested to be sent by the first network device; sending the first training data set to the training device according to the relevant information indicated by the first information, the first training data set being used for training the artificial intelligence AI model.
  • the relevant information includes at least one of the following: information about the size of the first training data set, configuration information of the input of the AI model, and configuration information of a reference signal used for training the AI model.
  • the information on the size of the first training data set is determined based on the size of the training data set required to complete the training of the AI model.
  • the configuration information of the reference signal includes at least one of the following: an identifier of the reference signal, a time domain resource of the reference signal, a frequency domain resource of the reference signal, a transmission period of the reference signal, and a type of the transmitted reference signal.
  • the first information includes at least one of the following: identification information of the AI model, information on the application scenario of the AI model, usage information of the AI model, and information on the computing power capability of the training device.
  • the method also includes: obtaining third information from a second network device, the third information being relevant information for training the AI model, wherein the first network device is the target network device to which the training device switches from the second network device; sending the first training data set to the training device according to the relevant information indicated by the first information, includes: sending the first training data set to the training device according to the relevant information indicated by the first information and the third information.
  • the first network device can comprehensively decide whether to support the training of the AI model based on various information.
  • the network device will not continuously send training data sets to the training device, which can reduce unnecessary air interface resource occupancy, save air interface overhead, and improve the utilization performance of air interface resources.
  • the third information includes at least one of the following: information about the size of the training data set that the training device requests the second network device to send, information about the size of the training data set required to complete the AI model training, identification information of the AI model, and information about the computing power capability of the training device.
  • the first network device stores a first mapping relationship, where the first mapping relationship is a mapping relationship between an identifier of an AI model and a size of a training data set corresponding to the identifier of the AI model.
  • the training device requests the second network device to send information about the size of the training data set can be understood as when the second network device is connected to the training device, the training device requests the second network device to send information about the size of the required training data set.
  • the size of the training data set requested from the second network device may also be determined based on historical information.
  • “information on the size of the training data set required to complete the AI model training” may be understood as the size of the training data set required by the training device to train the AI model in total stored on the second network device.
  • the first network device can obtain the information for training the AI model from the second network, so that the first network device can jointly determine the size of the training data that can be sent to the training device based on the training device indication information and the information synchronized from the second network device, thereby improving the utilization performance of air interface resources.
  • the third information also includes: information on the duration of sending the training data set and/or information on the method of sending the training data set, and the method also includes: determining whether the first network device and/or the training device has the ability to support training the AI model based on the third information and resource usage in the first network device.
  • “Information on the method of sending the training data set” can also be understood as, for example, the second network device periodically sends the training data set to the training device.
  • the network device can identify which time period of the day has the least data request volume, which can also be understood as which time period has the least air interface resource occupancy, or which time period has sufficient air interface resources.
  • the training data set can be provided to the AI model.
  • the network device can select this time period every day to support the update of the AI model (here, the number of training devices and/or the number of AI models are not limited).
  • the second network device sends the training data set to the training device at intervals. This solution has some improvement in flexibility relative to periodic transmission.
  • the network device finds and/or judges that the current data request volume is small and the air interface resources are sufficient, then the network device determines that it can support the update of the AI model. In other words, in this implementation, the network device can determine to send a training data set to the training device based on the occupancy of the air interface resources at the current moment, so it does not have obvious periodic characteristics.
  • resource usage can also be understood as “resource occupancy”, “air interface resource occupancy”, etc.
  • the existing protocol framework has defined the maximum number of reference signals (for example, 64 CSI-RS) that can be configured by network devices. If the network device finds that all reference signals have been configured for other functions, it can be determined that the current resources are occupied and reference signal resources cannot be configured for the training device to support the update of the AI model.
  • the first network device can obtain the information for training the AI model from the second network, so that the first network device can jointly determine the size of the training data that can be sent to the training device based on the training device indication information, the information synchronized from the second network device, and the air interface resource occupancy, thereby improving the utilization rate of the air interface resources.
  • a method for obtaining a training data set is provided.
  • the method can be executed by a first network device, or can also be executed by a component (such as a chip or circuit) of the first network device, without limitation.
  • the method includes: a first network device obtains third information from a second network device, wherein the third information is relevant information for training an artificial intelligence (AI) model, wherein the first network device is a target network device to which the training device switches from the second network device; the first network device receives first information from the training device, wherein the first information is used to request the first network device to send a training data set; the first network device determines a first training data set to be sent based on the third information; the first network device sends the first training data set to the training device based on the first information, and the first training data set is used for training the AI model.
  • AI artificial intelligence
  • the first network device can obtain the information for training the AI model from the second network, so that the first network device can jointly determine the size of the training data that can be sent to the training device based on the training device indication information and the information synchronized from the second network device, thereby improving the utilization performance of air interface resources.
  • the third information includes at least one of the following: information about the size of the training data set that the training device requests the second network device to send, information about the size of the training data set required to complete the AI model training, identification information of the AI model, and information about the computing power capability of the training device.
  • the first network device stores a first mapping relationship, where the first mapping relationship is a mapping relationship between an identifier of an AI model and a size of a training data set corresponding to the identifier of the AI model.
  • the third information also includes: information on the duration of sending the training data set and/or information on the method of sending the training data set, and the method also includes: the first network device determines whether the first network device and/or the training device has the ability to support training the AI model based on the third information and resource usage in the first network device.
  • the first network device determines the first training data set to be sent based on the third information, including: the first network device determines the first training data set to be sent based on the first information and the third information, wherein the first information is used to indicate relevant information of the first training data set requested to be sent by the first network device.
  • the relevant information includes at least one of the following: information about the size of the first training data set, Input information of the AI model and configuration information of the reference signal used for training the AI model.
  • the information on the size of the first training data set is determined based on the size of the training data set required to complete the training of the AI model.
  • the configuration information of the reference signal includes at least one of the following: an identifier of the reference signal, a time domain resource of the reference signal, a frequency domain resource of the reference signal, a transmission period of the reference signal, and a type of the transmitted reference signal.
  • the first information includes at least one of the following: identification information of the AI model, information on the application scenario of the AI model, usage information of the AI model, and information on the computing power capability of the training device.
  • the method also includes: receiving second information from the training device, the second information being used to indicate relevant information of a second training data set requested to be sent by the network device, wherein the second information is determined based on the performance of the AI model, and the performance of the AI model is determined based on training of the first training data set; based on the second information, determining the second training data set to be sent.
  • a communication method is provided.
  • the method may be executed by a training device, or may be executed by a component of the training device (eg, a chip or a circuit), without limitation.
  • the method includes: measuring N groups of reference signals, obtaining N groups of measurement results corresponding to the N groups of reference signals, wherein each group of reference signals in the N groups of reference signals includes at least one reference signal, each group of reference signals has the same group identifier, and N is an integer greater than 1; receiving fourth information from a network device, the fourth information is used to indicate M groups of reference signals in the N groups of reference signals; determining first input information of an artificial intelligence AI model according to the fourth information and the N groups of measurement results corresponding to the N groups of reference signals, the first input information includes the M groups of measurement results corresponding to the M groups of reference signals; the AI model is used to obtain first output information based on the first input information, wherein the first output information includes the group identifiers of the K groups of reference signals in the N groups of reference signals, wherein the group identifiers of the K groups of reference signals correspond to the K groups of measurement results with the best channel quality in the N groups of measurement results.
  • Each group of measurement results may include one or more measurement results.
  • the pattern of the sparse beam can still be the beam pattern indicated by the fourth information.
  • the training device can perform a full codebook scan based on the training data set issued by the first network device. Since the channel state (also understood as the channel environment) is time-varying, the measurement results of the reference signal obtained after each full codebook scan are not exactly the same. Therefore, during each training, the corresponding measurement results of the M groups of reference signals in the N groups of reference signals are also different, and the training labels determined by the training device are also different, that is, the input information and training labels of the AI model will change accordingly, but these changes are essentially caused by changes in the channel state, and the beam pattern has not changed.
  • the variable in the AI model training process is only the channel state.
  • the solution provided by the present application can accelerate the convergence speed of the AI model, improve the model training efficiency, and thus reduce the occupancy of air interface resources.
  • a group identifier of a group of reference signals may correspond to a beam identifier
  • N group identifiers of N groups of reference signals may correspond to N beam identifiers
  • the first output information also includes the group identifiers of the remaining (N-K) groups of reference signals in the N groups of measurement results, and the group identifiers of the (N-K) groups of reference signals correspond to the (N-K) groups of measurement results.
  • classification and regression methods can be used for AI model training, and different training methods correspond to different input information and output information of the AI model.
  • the input information of the AI model in the classification method is the measurement result of the reference signal
  • the output information is the K beam identifiers with the best channel quality in the full codebook predicted by the AI model.
  • the input information of the AI model in the regression method is the measurement result of the reference signal (for example, RSRP, RSRQ, SINR of the reference signal)
  • the output information is the measurement results of all reference signals in the full codebook predicted by the AI model.
  • the fourth information includes N fields, the N fields correspond one-to-one to the N groups of reference signals, and the bit values of M fields among the N fields are different from the bit values of the remaining (N-M) fields; the fourth information is used to indicate M groups of reference signals among the N groups of reference signals, specifically including: the M fields in the fourth information are used to indicate the M groups of reference signals.
  • the network device can indicate the sparse beam pattern to the training device by indicating the bit value of each field in the fourth information. That is, the training device can obtain the input information of the AI model by parsing the fourth information, thereby accelerating the convergence speed of the AI model and improving the training efficiency of the AI model.
  • the method also includes: receiving fifth information from a network device, the fifth information being used to indicate P groups of reference signals among the N groups of reference signals; wherein the fifth information includes N fields, the N fields corresponding one-to-one to the N groups of reference signals, and the bit values of P fields among the N fields are different from the bit values of the remaining (N-P) fields; the fifth information is used to indicate P groups of reference signals among the N groups of reference signals, specifically including: the P fields in the fifth information are used to indicate the P groups of reference signals.
  • the network device can indicate multiple sparse beam patterns to the training device by indicating the bit value of each field in the fourth information and the fifth information. That is, the training device can obtain multiple input information of the AI model by parsing the fourth information and the fifth information, thereby accelerating the convergence speed of the AI model and improving the training efficiency of the AI model.
  • the method also includes: receiving configuration information from a network device, wherein the configuration information is used to indicate one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission periods of the N groups of reference signals, and group identifiers of the N groups of reference signals.
  • the configuration information and the fourth information and/or the fifth information may also be sent in the same message, without limitation.
  • the network device can send configuration information of N groups of reference signals to the training device, so that the training device can obtain a training data set by measuring the N groups of reference signals to train the AI model.
  • a communication method is provided.
  • the method can be executed by a network device, or can also be executed by a component of the network device (such as a chip or circuit), without limitation.
  • the method includes: sending N groups of reference signals to a training device, wherein each group of reference signals in the N groups of reference signals includes at least one reference signal, each group of reference signals has the same group identifier, and N is an integer greater than 1; sending fourth information to the training device, wherein the fourth information is used to indicate M groups of reference signals in the N groups of reference signals, wherein the M groups of reference signals are used to determine first input information; the AI model is used to obtain first output information based on the first input information, wherein the first output information includes the group identifiers of the K groups of reference signals in the N groups of reference signals, wherein the group identifiers of the K groups of reference signals correspond to the K groups of measurement results with the best channel quality among the N groups of measurement results corresponding to the N groups of parameter signals.
  • Each group of measurement results may include one or more measurement results.
  • the first output information also includes the group identifiers of the remaining (N-K) groups of reference signals in the N groups of measurement results, and the group identifiers of the (N-K) groups of reference signals correspond to the (N-K) groups of measurement results.
  • the fourth information includes N fields, the N fields correspond one-to-one to the N groups of reference signals, and the bit values of M fields among the N fields are different from the bit values of the remaining (N-M) fields; the fourth information is used to indicate M groups of reference signals among the N groups of reference signals, specifically including: the M fields in the fourth information are used to indicate the M groups of reference signals.
  • the method also includes: sending fifth information to the training device, the fifth information being used to indicate P groups of reference signals among the N groups of reference signals; wherein the fifth information includes N fields, the N fields corresponding one-to-one to the N groups of reference signals, and the bit values of P fields among the N fields are different from the bit values of the remaining (N-P) fields; the fourth information being used to indicate P groups of reference signals among the N groups of reference signals, specifically includes: the P fields in the fourth information being used to indicate the P groups of reference signals.
  • the method also includes: sending configuration information to the training device, wherein the configuration information is used to indicate one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission periods of the N groups of reference signals, and group identifiers of the N groups of reference signals.
  • a communication method is provided, which may be executed by a training device, or may be executed by a component of the training device (eg, a chip or a circuit), without limitation.
  • the method includes: receiving a second reference signal set, wherein the second reference signal set includes N groups of reference signals, each group of reference signals in the N groups of reference signals includes at least one reference signal, and N is an integer greater than 1; receiving second beam indication information, the second beam indication information indicating the beam corresponding to the first reference signal set, wherein the beam corresponding to the first reference signal set is a subset of multiple beams corresponding to the second reference signal set, and the beam corresponding to the first reference signal set is used to determine the first input information of the AI model in the training device, and the first input information is based on the measurement result of the beam corresponding to the first reference signal set, and the first reference signal set includes M groups of reference signals, N is an integer greater than M, and M is an integer greater than or equal to 1; wherein the AI model is used to obtain first output information based on the first input information, and the first output information indicates K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set, and K is an integer greater than or equal to 1
  • the training device can determine the sparse beam pattern through the received second beam indication information. It can also be understood that the training device can determine which beams in the full codebook the sparse beam pattern is composed of based on the second beam indication information, and can also determine the input information of the AI model. At this time, since the sparse beam pattern does not change during the training process, only the channel state changes, this solution can accelerate the convergence of the AI model and improve the training efficiency of the AI model.
  • the first output information indicates K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set, including at least one of the following: information about the K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set; or, group identifiers of K groups of reference signals, wherein the K groups of reference signals correspond to K measurement results predicted to have the best channel quality among N measurement results corresponding to the N groups of reference signals, and the group identifiers of the K groups of reference signals have a predefined or preconfigured correspondence with the K beams; or, multiple beam information corresponding to the N groups of reference signals and N measurement results corresponding to the beam information; or, group identifiers of the N groups of reference signals and N measurement results of the N groups of reference signals, and the N groups of reference signals have a predefined or preconfigured correspondence with the N beams.
  • the output information of the AI model may be slightly different based on the implementation of the algorithm of the AI model.
  • the output information of the AI model in the classification method is the information of the K (K is an integer greater than 0) beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set.
  • the output information of the AI model in the regression method is the N measurement results corresponding to the N groups of reference signals, and so on.
  • the second beam indication information indicating the beam corresponding to the first reference signal set includes: the second beam indication information indicating a position of the beam corresponding to the first reference signal set among the multiple beams corresponding to the second reference signal set.
  • the second beam indication information can indicate the position of the beam corresponding to the first reference signal set in the multiple beams corresponding to the second reference signal set, thereby indicating a sparse beam pattern. It can also be understood that the training device can determine which beams in the full codebook correspond to the reference signal group in the first reference signal set based on the second beam indication information, thereby determining the input information of the AI model.
  • the second beam indication information includes N fields, the N fields correspond one-to-one to multiple beams corresponding to the second reference signal set, and the bit values of M fields among the N fields are different from the bit values of the remaining (N-M) fields; the second beam indication information indicates the beam corresponding to the first reference signal set, including: the M fields in the second beam indication information correspond to the first reference signal set.
  • the sparse beam pattern can be directly indicated by M fields out of N fields, or it can also be understood that M fields out of N fields directly indicate which beams in the full codebook constitute the sparse beam pattern, so that the training device can determine the input information of the AI model.
  • the method also includes: sending first configuration information to the training device, the first configuration information indicating one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission period of the N groups of reference signals, group identifiers of the N groups of reference signals, or beam information of the N groups of reference signals.
  • the first configuration information can be used to indicate how the training device should receive N groups of reference signals. For example, on which time-frequency resources the training device should receive N groups of reference signals.
  • the second beam indication information indicates the beam corresponding to the M groups of reference signals, including: the second beam indication information includes the group identifier or beam information of the M groups of reference signals, the M groups of reference signals are part of the N groups of reference signals, and the N groups of reference signals have a predefined or preconfigured correspondence with the N beams.
  • a sparse beam pattern can be indicated by indicating the group identifier or beam information of M groups of reference signals.
  • the third configuration information of the N groups of reference signals is sent to the training device; when the second beam indication information includes the group identifier of the M groups of reference signals, the third configuration information of the N groups of reference signals includes the group identifiers of the N groups of reference signals and indicates one or more of the following: the time domain resources of the N groups of reference signals, the frequency domain resources of the N groups of reference signals, the transmission period of the N groups of reference signals, or the beams of the N groups of reference signals; and the M groups of reference signals are part of the N groups of reference signals, including: the N group identifiers of the N groups of reference signals include the M group identifiers of the M groups of reference signals; or, when the second beam indication information includes the beam information of the M groups of reference signals, the N groups of reference signals
  • the third configuration information of the reference signal includes the beam information of each of the N groups of reference signals and indicates one or more of the following: the group identifier of the N groups of reference signals, the time domain resources, the frequency domain resources of the N groups
  • the third configuration information may include the identifiers of N groups of reference signals, and M groups of reference signals belong to the N groups of reference signals. Therefore, the second beam indication information may include the identifiers of M groups of reference signals or the beam information of M groups of reference signals, thereby indicating which beams in the full codebook the sparse beam pattern is composed of, so that the training device can determine the input information of the AI model.
  • the method further includes: measuring the N groups of reference signals to obtain N measurement results, where the N measurement results correspond to N beams and the N measurement results include measurement results of beams corresponding to the first reference signal set.
  • the training device can obtain N groups of measurement results by measuring N groups of reference signals, and determine the input information of the AI model based on the second beam information.
  • a communication method is provided, which can be executed by a network device, or can also be executed by a component of the network device (such as a chip or circuit), without limitation.
  • the method includes: sending a second reference signal set to a training device, wherein the second reference signal set includes N groups of reference signals, each group of reference signals in the N groups of reference signals includes at least one reference signal, and N is an integer greater than 1; sending second beam indication information to the training device, the second beam indication information indicating a beam corresponding to a first reference signal set, wherein the beam corresponding to the first reference signal set is a subset of multiple beams corresponding to the second reference signal set, and the beam corresponding to the first reference signal set is used to determine first input information of an AI model in the training device, and the first input information is based on a measurement result of the beam corresponding to the first reference signal set, the first reference signal set includes M groups of reference signals, N is an integer greater than M, and M is an integer greater than or equal to 1; wherein the AI model is used to obtain first output information based on the first input information, and the first output information indicates K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set, and
  • the first output information indicates K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set, including at least one of the following: information about the K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set; or, group identifiers of K groups of reference signals, wherein the K groups of reference signals correspond to K measurement results predicted to have the best channel quality among N measurement results corresponding to the N groups of reference signals, and the group identifiers of the K groups of reference signals have a predefined or preconfigured correspondence with the K beams; or, multiple beam information corresponding to the N groups of reference signals and N measurement results corresponding to the beam information; or, group identifiers of the N groups of reference signals and N measurement results of the N groups of reference signals, and the N groups of reference signals have a predefined or preconfigured correspondence with the N beams.
  • the second beam indication information indicating the beam corresponding to the first reference signal set includes: the second beam indication information indicating a position of the beam corresponding to the first reference signal set among the multiple beams corresponding to the second reference signal set.
  • the second beam indication information includes N fields, the N fields correspond one-to-one to multiple beams corresponding to the second reference signal set, and the bit values of M fields among the N fields are different from the bit values of the remaining (N-M) fields; the second beam indication information indicates the beam corresponding to the first reference signal set, including: the M fields in the second beam indication information correspond to the first reference signal set.
  • the method also includes: sending first configuration information to the training device, the first configuration information indicating one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission period of the N groups of reference signals, group identifiers of the N groups of reference signals, or beam information of the N groups of reference signals.
  • the second beam indication information indicates the beam corresponding to the M groups of reference signals, including: the second beam indication information includes the group identifier or beam information of the M groups of reference signals, the M groups of reference signals are part of the N groups of reference signals, and the N groups of reference signals have a predefined or preconfigured correspondence with the N beams.
  • the method further includes: sending a third configuration of the N groups of reference signals to the training device. information; in the case where the second beam indication information includes the group identifiers of the M groups of reference signals, the third configuration information of the N groups of reference signals includes the group identifiers of the N groups of reference signals and indicates one or more of the following: the time domain resources of the N groups of reference signals, the frequency domain resources of the N groups of reference signals, the transmission period of the N groups of reference signals, or the beams of the N groups of reference signals; and the M groups of reference signals are part of the N groups of reference signals, including: the N group identifiers of the N groups of reference signals include the M group identifiers of the M groups of reference signals; or, in the case where the second beam indication information includes the beam information of the M groups of reference signals, the third configuration information of the N groups of reference signals includes the beam information of the N groups of reference signals and indicates one or more of the following: the group identifiers of the N groups of reference signals, the time domain
  • a communication method is provided, which can be executed by a terminal device, or can also be executed by a component (such as a chip or circuit) of the terminal device, without limitation.
  • the terminal device can be used as an inference device.
  • the method includes: receiving a first reference signal set, wherein the first reference signal set includes M groups of reference signals, each group of reference signals in the M groups of reference signals includes at least one reference signal, and M is an integer greater than or equal to 1; receiving first beam indication information, the first beam indication information indicating a beam corresponding to the first reference signal set, wherein the first reference signal set is used to determine first input information of the AI model, the first input information is based on measurement results of the M groups of reference signals included in the first reference signal set, the beam corresponding to the first reference signal set is a subset of multiple beams corresponding to a second reference signal set, and the second reference signal set includes N groups of reference signals, and N is an integer greater than or equal to M; wherein the AI model is used to obtain first output information based on the first input information, and the first output information indicates K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set, and K is an integer greater than or equal to 1 and K is less than N.
  • the network device can also indicate the input information of the model to the terminal device, so that the terminal device can determine the input information of the model, thereby improving the accuracy of the output information of the terminal device during model reasoning.
  • the first output information indicates K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set, including at least one of the following: information about the K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set; or, group identifiers of K groups of reference signals, wherein the K groups of reference signals correspond to K measurement results predicted to have the best channel quality among N measurement results corresponding to the N groups of reference signals, and the group identifiers of the K groups of reference signals have a predefined or preconfigured correspondence with the K beams; or, multiple beam information corresponding to the N groups of reference signals and N measurement results corresponding to the beam information; or, group identifiers of the N groups of reference signals and N measurement results of the N groups of reference signals, and the N groups of reference signals have a predefined or preconfigured correspondence with the N beams.
  • the inference output information of the AI model may be slightly different based on the implementation of the algorithm of the AI model.
  • the inference output information of the AI model in the classification method is the information of the K (K is an integer greater than 0) beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set.
  • the inference output information of the AI model in the regression method is the N measurement results corresponding to N groups of reference signals, and so on.
  • the first beam indication information indicating the beam corresponding to the first reference signal set includes: the first beam indication information indicating a position of the beam corresponding to the first reference signal set among multiple beams corresponding to the second reference signal set.
  • the second beam indication information can indicate the position of the beam corresponding to the first reference signal set in the multiple beams corresponding to the second reference signal set, thereby indicating a sparse beam pattern. It can also be understood that the terminal device can determine which beams in the full codebook correspond to the reference signal group in the first reference signal set based on the second beam indication information, thereby determining the input information of the AI model.
  • the first beam indication information includes N fields, the N fields correspond one-to-one to multiple beams corresponding to the second reference signal set, and the bit values of M fields among the N fields are different from the bit values of the remaining (N-M) fields; the first beam indication information indicates the beam corresponding to the first reference signal set, including: the M fields in the first beam indication information correspond to the first reference signal set.
  • the sparse beam pattern can be directly indicated by M fields out of N fields, or it can also be understood that M fields out of N fields directly indicate which beams in the full codebook constitute the sparse beam pattern, so that the terminal device can determine the input information of the AI model.
  • the method also includes: receiving first configuration information, wherein the first configuration information indicates one or more of the following: time domain resources of the M groups of reference signals, frequency domain resources of the M groups of reference signals, a transmission period of the M groups of reference signals, a group identifier of the M groups of reference signals, or beam information of the M groups of reference signals.
  • the first configuration information may be used to indicate how the terminal device should receive M groups of reference signals. For example, on which time-frequency resources the terminal device should receive M groups of reference signals.
  • the first beam indication information indicates the beam corresponding to the M groups of reference signals, including: the first beam indication information includes the group identifier or beam information of the M groups of reference signals, the M groups of reference signals are part of the N groups of reference signals, and the N groups of reference signals have a predefined or preconfigured correspondence with the N beams.
  • a sparse beam pattern can be indicated by indicating the group identifier or beam information of M groups of reference signals.
  • the first beam indication information is included in the second configuration information of the M groups of reference signals, and when the first beam indication information includes the group identifier of the M groups of reference signals, the second configuration information also includes one or more of the time domain resources, frequency domain resources, transmission period, or beam information of the M groups of reference signals; or, when the first beam indication information includes the beam information of the M groups of reference signals, the second configuration information also includes one or more of the time domain resources, frequency domain resources, transmission period, or group identifier of the M groups of reference signals.
  • the second configuration information may include first beam indication information, and the first beam indication information includes the group identifier of M groups of reference signals or the beam information of M groups of reference signals, thereby indicating which beams in the full codebook the sparse beam pattern is composed of, so that the terminal device can determine the input information of the AI model.
  • the method further includes: receiving third configuration information of the N groups of reference signals; in a case where the first beam indication information includes the group identifier of the M groups of reference signals, the third configuration information of the N groups of reference signals includes the group identifiers of the N groups of reference signals and indicates one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission periods of the N groups of reference signals, or beams of the N groups of reference signals; and the M groups of reference signals are part of the N groups of reference signals, including: N of the N groups of reference signals
  • the group identifiers include M group identifiers of the M groups of reference signals; or, when the first beam indication information includes the beam information of the M groups of reference signals, the third configuration information of the N groups of reference signals includes the beam information of each of the N groups of reference signals and indicates one or more of the following: the group identifiers of the N groups of reference signals, time domain resources, frequency domain resources of the N groups of reference signals, or
  • the third configuration information may include the identifier of N groups of reference signals, and M groups of reference signals belong to the N groups of reference signals. Therefore, the second beam indication information may include the identifier of M groups of reference signals or the beam information of M groups of reference signals, thereby indicating which beams in the full codebook the sparse beam pattern is composed of, so that the terminal device can determine the input information of the AI model.
  • the method further includes: obtaining the first output information based on the first input information using the AI model; and sending the first output information.
  • the terminal device after the terminal device obtains the inference output information through AI model inference, it can further feed back the output information to the network device, so that the network device can send the corresponding reference signal to the terminal device based on the output information.
  • the terminal device measures the reference signal again, determines the reference signal with the best measurement result, and uses the beam identifier corresponding to the reference signal as the final selected beam, and uses the beam to communicate with the network device.
  • a communication method is provided, which may be executed by a network device, or may be executed by a component of the network device (eg, a chip or a circuit), without limitation.
  • the method includes: sending a first reference signal set to a terminal device, wherein the first reference signal set includes M groups of reference signals, each group of reference signals in the M groups of reference signals includes at least one reference signal, and M is an integer greater than or equal to 1; sending first beam indication information to the terminal device, the first beam indication information indicating the beam corresponding to the first reference signal set, wherein the first reference signal set is used to determine the first input information of the AI model, the first input information is based on the measurement results of the M groups of reference signals included in the first reference signal set, the beam corresponding to the first reference signal set is a subset of multiple beams corresponding to the second reference signal set, and the second reference signal set includes N groups of reference signals, and N is an integer greater than or equal to M; wherein the AI model is used to obtain first output information based on the first input information, and the first output information indicates the second reference signal
  • the K beams predicted to have the best channel quality among the multiple beams corresponding to the signal set, where K is an integer greater than or equal to
  • the first output information indicates K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set, including at least one of the following: information about the K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set; or, group identifiers of K groups of reference signals, wherein the K groups of reference signals correspond to K measurement results predicted to have the best channel quality among N measurement results corresponding to the N groups of reference signals, and the group identifiers of the K groups of reference signals have a predefined or preconfigured correspondence with the K beams; or, multiple beam information corresponding to the N groups of reference signals and N measurement results corresponding to the beam information; or, group identifiers of the N groups of reference signals and N measurement results of the N groups of reference signals, and the N groups of reference signals have a predefined or preconfigured correspondence with the N beams.
  • the first beam indication information indicating the beam corresponding to the first reference signal set includes: the first beam indication information indicating a position of the beam corresponding to the first reference signal set among multiple beams corresponding to the second reference signal set.
  • the first beam indication information includes N fields, the N fields correspond one-to-one to multiple beams corresponding to the second reference signal set, and the bit values of M fields among the N fields are different from the bit values of the remaining (N-M) fields; the first beam indication information indicates the beam corresponding to the first reference signal set, including: the M fields in the first beam indication information correspond to the first reference signal set.
  • the method also includes: sending first configuration information to the terminal device, the first configuration information indicating one or more of the following: time domain resources of the M groups of reference signals, frequency domain resources of the M groups of reference signals, transmission period of the M groups of reference signals, group identifiers of the M groups of reference signals, or beam information of the M groups of reference signals.
  • the first beam indication information indicates the beam corresponding to the M groups of reference signals, including: the first beam indication information includes the group identifier or beam information of the M groups of reference signals, the M groups of reference signals are part of the N groups of reference signals, and the N groups of reference signals have a predefined or preconfigured correspondence with the N beams.
  • the first beam indication information is included in the second configuration information of the M groups of reference signals, and when the first beam indication information includes the group identifier of the M groups of reference signals, the second configuration information also includes one or more of the time domain resources, frequency domain resources, transmission period, or beam information of the M groups of reference signals; or, when the first beam indication information includes the beam information of the M groups of reference signals, the second configuration information also includes one or more of the time domain resources, frequency domain resources, transmission period, or group identifier of the M groups of reference signals.
  • the method further includes: sending third configuration information of the N groups of reference signals to the terminal device; in a case where the first beam indication information includes the group identifier of the M groups of reference signals, the third configuration information of the N groups of reference signals includes the group identifiers of the N groups of reference signals and indicates one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission periods of the N groups of reference signals, or beams of the N groups of reference signals; and the M groups of reference signals are part of the N groups of reference signals, including: the N groups of reference signals
  • the N group identifiers of the N groups of reference signals include the M group identifiers of the M groups of reference signals; or, when the first beam indication information includes the beam information of the M groups of reference signals, the third configuration information of the N groups of reference signals includes the beam information of each of the N groups of reference signals and indicates one or more of the following: the group identifiers of the N groups of reference signals, the time domain resources,
  • the method further includes: receiving the first output information from the terminal device.
  • a communication device is provided, the device being used to execute the method in any possible implementation of the first aspect, the fourth aspect, the sixth aspect, and the eighth aspect.
  • the device may include a unit and/or module, such as a transceiver unit and/or a processing unit, for executing the method in any possible implementation of the first aspect, the fourth aspect, the sixth aspect, and the eighth aspect.
  • the device is a training device, an inference device or a terminal device.
  • the communication unit may be a transceiver, or an input/output interface; the processing unit may be at least one processor.
  • the transceiver may be a transceiver circuit.
  • the input/output interface may be an input/output circuit.
  • the device is a chip, chip system or circuit for a training device, an inference device or a terminal device.
  • the communication unit may be a chip, chip system or circuit on the chip, chip system or circuit.
  • the processing unit can be at least one processor, processing circuit or logic circuit, etc.
  • a communication device is provided, the device being used to execute the method in any possible implementation of the second aspect, the third aspect, the fifth aspect, the seventh aspect, and the ninth aspect.
  • the device may include a unit and/or module, such as a transceiver unit and/or a processing unit, for executing the method in any possible implementation of the second aspect, the third aspect, the fifth aspect, the seventh aspect, and the ninth aspect.
  • the device is a network device or a first network device.
  • the communication unit may be a transceiver, or an input/output interface;
  • the processing unit may be at least one processor.
  • the transceiver may be a transceiver circuit.
  • the input/output interface may be an input/output circuit.
  • the device is a chip, a chip system or a circuit for a network device or a first network device.
  • the communication unit may be an input/output interface, an interface circuit, an output circuit, an input circuit, a pin or a related circuit on the chip, the chip system or the circuit;
  • the processing unit may be at least one processor, a processing circuit or a logic circuit.
  • a communication device comprising: at least one processor, configured to execute a computer program or instruction stored in a memory, so as to execute the method in any possible implementation of any of the first aspect, the fourth aspect, the sixth aspect, and the eighth aspect.
  • the device further comprises a memory, configured to store a computer program or instruction.
  • the device further comprises a communication interface, and the processor reads the computer program or instruction stored in the memory through the communication interface.
  • the apparatus is a training device, an inference device or a terminal device.
  • the apparatus is a chip, a chip system or a circuit for a training device, an inference device or a terminal device.
  • a communication device comprising: at least one processor, configured to execute a computer program or instruction stored in a memory, so as to execute the method in any possible implementation of the second aspect, the third aspect, the fifth aspect, the seventh aspect, and the ninth aspect.
  • the device further comprises a memory, configured to store a computer program or instruction.
  • the device further comprises a communication interface, and the processor reads the computer program or instruction stored in the memory through the communication interface.
  • the apparatus is a network device or a first network device.
  • the apparatus is a chip, a chip system or a circuit for a network device or a first network device.
  • the present application provides a processor, comprising: an input circuit, an output circuit, and a processing circuit.
  • the processing circuit is used to receive a signal through the input circuit and transmit a signal through the output circuit, so that the processor executes the method in any possible implementation of any aspect from the first aspect to the ninth aspect.
  • the processor may be one or more chips
  • the input circuit may be an input pin
  • the output circuit may be an output pin
  • the processing circuit may be a transistor, a gate circuit, a trigger, and various logic circuits.
  • the input signal received by the input circuit may be received and input by, for example, but not limited to, a transceiver
  • the signal output by the output circuit may be, for example, but not limited to, output to a transmitter and transmitted by the transmitter
  • the input circuit and the output circuit may be the same circuit, which is used as an input circuit and an output circuit at different times.
  • the embodiments of the present application do not limit the specific implementation methods of the processor and various circuits.
  • a processing device comprising a processor and a memory.
  • the processor is used to read instructions stored in the memory, and can receive signals through a transceiver and transmit signals through a transmitter to execute the method in any possible implementation of any aspect from the first aspect to the ninth aspect.
  • the number of the processors is one or more, and the number of the memories is one or more.
  • the memory may be integrated with the processor, or the memory may be provided separately from the processor.
  • the memory can be a non-transitory memory, such as a read-only memory (ROM), which can be integrated with the processor on the same chip or can be separately set on different chips.
  • ROM read-only memory
  • the embodiments of the present application do not limit the type of memory and the setting method of the memory and the processor.
  • the related data interaction process can be a process of outputting indication information from the processor
  • receiving capability information can be a process of receiving input capability information by the processor.
  • the data output by the processor can be output to the transmitter, and the input data received by the processor can come from the transceiver.
  • the transmitter and the transceiver can be collectively referred to as a transceiver.
  • the processing device in the fifteenth aspect may be one or more chips.
  • the processor in the processing device may be implemented by hardware or by software.
  • the processor When implemented by hardware, the processor may be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor may be a general-purpose processor implemented by reading software codes stored in a memory, which may be integrated in the processor or located outside the processor and exist independently.
  • a computer-readable storage medium which stores a program code for execution by a device, wherein the program code includes a method for executing any possible implementation of the first to ninth aspects above.
  • a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute a method in any possible implementation of the first to ninth aspects.
  • a chip system comprising a processor for calling and running a computer program from a memory, so that a device equipped with the chip system executes the methods in each implementation manner in any one of the first to ninth aspects above.
  • a communication system comprising a training device and a network device.
  • the training device is used to execute any possible implementation method in the first aspect
  • the network device is used to execute any possible implementation method in the second aspect.
  • a communication system comprising a first network device.
  • the first network device is used to execute any possible implementation method of the third aspect.
  • a communication system comprising a training device and a network device.
  • the training device is used to execute any possible implementation method in the fourth aspect
  • the network device is used to execute any possible implementation method in the fifth aspect.
  • a communication system comprising a training device and a network device.
  • the training device is used to execute any possible implementation method in the sixth aspect
  • the network device is used to execute any possible implementation method in the seventh aspect.
  • a communication system comprising an inference device, such as a terminal device, and a network device.
  • the inference device such as a terminal device, is used to execute any possible implementation method in the eighth aspect
  • the network device is used to execute any possible implementation method in the ninth aspect.
  • FIG1 is a schematic diagram of the structure of a communication system
  • Fig. 2 is a schematic diagram of a neuron structure
  • FIG3 is a schematic diagram of the layer relationship of a neural network
  • FIG4 is a schematic diagram of a framework for training and reasoning of an AI model provided in this application.
  • FIG5 is a schematic flow chart of a method 500 for obtaining a training data set provided by the present application.
  • FIG6 is a schematic flow chart of a method 600 for obtaining a training data set provided by the present application.
  • FIG. 7 is a schematic flowchart of a method 700 for obtaining input information of an AI model provided by the present application
  • FIG8 is a schematic flow chart of a communication method 800 provided in the present application.
  • FIG9 is a schematic flow chart of a communication method 900 provided in the present application.
  • FIG10 is a schematic block diagram of a communication device 100 provided in the present application.
  • FIG. 11 is a schematic block diagram of a communication device 200 provided in the present application.
  • the communication system can be a fourth generation (4G) communication system (such as a long term evolution (LTE) system), a fifth generation (5G) communication system, a worldwide interoperability for microwave access (WiMAX) or a wireless local area network (WLAN) system, a satellite communication system, a future communication system, such as a sixth generation (6G) mobile communication system, or a fusion system of multiple systems.
  • 4G fourth generation
  • 5G fifth generation
  • WiMAX worldwide interoperability for microwave access
  • WLAN wireless local area network
  • satellite communication system such as a sixth generation (6G) mobile communication system
  • 6G mobile communication system such as a sixth generation (6G) mobile communication system
  • a fusion system of multiple systems such as a sixth generation (6G) mobile communication system
  • NR new radio
  • Satellite communication system future communication system, such as a sixth generation (6G) mobile communication system, or a fusion system of multiple systems.
  • a device in a communication system can send a signal to another device or receive a signal from another device, wherein the signal may include information, signaling or data, etc.
  • the device can also be replaced by an entity, a network entity, a communication device, a communication module, a node, a communication node, etc.
  • a communication system may include at least one terminal device and at least one network device.
  • a communication system may include a training device and at least one network device. The network device can send a downlink signal to the terminal device, and/or the terminal device can send an uplink signal to the access network device.
  • the communication system includes multiple terminal devices, multiple terminal devices can also send signals to each other, that is, the signal sending network element and the signal receiving network element can both be terminal devices.
  • the terminal device in the present application can be replaced by the first device, and the network device can be replaced by the second device, and the two perform the corresponding communication method in the present disclosure.
  • FIG1 is a simplified schematic diagram of a wireless communication system provided in an embodiment of the present application.
  • the wireless communication system includes a wireless access network 100 (an example of a network device).
  • the wireless access network 100 may be a next generation (e.g., 6G or higher) wireless access network, or a traditional (e.g., 5G, 4G, 3G, or 2G) wireless access network.
  • One or more communication devices 120a-120j, collectively referred to as 120
  • FIG1 is only a schematic diagram, and other devices may also be included in the wireless communication system, such as core network devices, wireless relay devices, and/or wireless backhaul devices, which are not shown in FIG1.
  • the wireless communication system may include multiple network devices (e.g., access network devices) at the same time, or may include multiple communication devices at the same time.
  • a network device may serve one or more communication devices at the same time.
  • a communication device may also access one or more network devices at the same time.
  • the embodiment of the present application does not limit the number of communication devices and network devices included in the wireless communication system.
  • the network device may be an entity on the network side for transmitting or receiving signals.
  • the network device may be an access device for a communication device to access the wireless communication system in a wireless manner, for example, the network device may be a base station.
  • Base station can broadly cover various names as follows, or be replaced with the following names, such as: node B (NodeB), evolved NodeB (eNB), next generation NodeB (gNB), access network equipment in open radio access network (O-RAN), relay station, access point, transmission point (transmitting and receiving point, TRP), transmitting point (transmitting point, TP), master station MeNB, auxiliary station SeNB, multi-standard radio (MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), radio remote unit (RRU), active antenna unit (AAU), radio head (RRH), central unit (CU), distribution unit (DU), radio unit (radio unit, RU), centralized unit control
  • the base station can be a macro base station, a micro base station, a relay node, a donor node or the like, or a combination thereof.
  • the network device may also refer to a communication module, a modem or a chip used to be arranged in the aforementioned device or apparatus.
  • the network device may also be a mobile switching center and a device to device (Device-to-Device, D2D), vehicle external connection (vehicle-to-everything, V2X), a device that performs the base station function in machine-to-machine (machine-to-machine, M2M) communication, a network side device in a 6G network, and a device that performs the base station function in a future communication system.
  • the network device may support networks with the same or different access technologies. The embodiments of the present application do not limit the specific technology and specific device form adopted by the network device.
  • the network equipment can be fixed or mobile.
  • base stations 110a, 110b (examples of network equipment) are stationary and are responsible for wireless transmission and reception in one or more cells from the communication device 120.
  • the helicopter or drone 120i shown in Figure 1 can be configured to act as a mobile base station, and one or more cells can move according to the location of the mobile base station 120i.
  • the helicopter or drone (120i) can be configured to be used as a communication device that communicates with the base station 110b.
  • the communication device used to implement the above network functions may be, for example, an access network device, or a network device having some functions of accessing the network, or a device capable of supporting the implementation of access network functions, such as a chip system, a hardware circuit, a software module, or a hardware circuit plus a software module.
  • the device may be installed in the access network device or used in combination with the access network device.
  • a communication device can be an entity on the user side for receiving or transmitting signals, such as a mobile phone.
  • Communication devices can be used to connect people, objects and machines.
  • Communication devices can communicate with one or more core networks through network devices.
  • Communication devices include handheld devices with wireless connection functions, other processing devices connected to wireless modems, or vehicle-mounted devices.
  • Communication devices can be portable, pocket-sized, handheld, built-in computer or vehicle-mounted mobile devices.
  • the communication device 120 can be widely used in various scenarios, such as cellular communication, device-to-device D2D, vehicle-to-everything V2X, end-to-end P2P, machine-to-machine M2M, machine type communication MTC, Internet of Things IOT, virtual reality VR, augmented reality AR, industrial control, autonomous driving, telemedicine, smart grid, smart furniture, smart Office, smart wearables, smart transportation, smart cities, drones, robots, remote sensing, passive sensing, positioning, navigation and tracking, autonomous delivery and mobility, etc.
  • cellular communication device-to-device D2D, vehicle-to-everything V2X, end-to-end P2P, machine-to-machine M2M, machine type communication MTC, Internet of Things IOT, virtual reality VR, augmented reality AR, industrial control, autonomous driving, telemedicine, smart grid, smart furniture, smart Office, smart wearables, smart transportation, smart cities, drones, robots, remote sensing, passive sensing, positioning, navigation and tracking, autonomous delivery
  • Some examples of the communication device 120 are: user equipment (UE) of 3GPP standard, fixed equipment, mobile equipment, handheld equipment, wearable equipment, cellular phone, smart phone, session initiation protocol (SIP) phone, laptop computer, personal computer, smart book, vehicle, satellite, global positioning system (GPS) equipment, target tracking equipment, drone, helicopter, aircraft, ship, remote control equipment, smart home equipment, industrial equipment, personal communication service (PCS) phone, wireless local loop (WLL) station, personal digital assistant (PDA), wireless network camera, tablet computer, PDA, mobile internet device (MID), wearable equipment such as smart watch, virtual reality (VR) equipment, augmented reality (AR) equipment, wireless terminal in industrial control, terminal in vehicle networking system, wireless terminal in self driving, wireless terminal in smart grid, wireless terminal in transportation safety, smart city, etc.
  • UE user equipment
  • UE user equipment
  • mobile equipment handheld equipment
  • wearable equipment cellular phone
  • smart phone session initiation protocol
  • SIP session initiation protocol
  • laptop computer personal computer
  • the communication device 120 may be a wireless device in the above various scenarios or a device used to be set in a wireless device, for example, a communication module, a modem or a chip in the above device.
  • the communication device may also be referred to as a terminal, a terminal device, a user equipment (UE), a mobile station (MS), a mobile terminal (MT), etc.
  • the communication device may also be a communication device in a future wireless communication system.
  • the communication device may be used in a dedicated network device or a general device. The embodiments of the present application do not limit the specific technology and specific device form adopted by the communication device.
  • the communication device can be used to act as a base station.
  • the UE can act as a scheduling entity that provides sidelink signals between UEs in V2X, D2D, or P2P, etc.
  • the cellular phone 120a and the car 120b communicate with each other using sidelink signals.
  • the cellular phone 120a and the smart home device 120e communicate without relaying the communication signal through the base station 110b.
  • the communication device used to implement the functions of the communication device may be a terminal device, or a terminal device having some of the functions of the above communication device, or a device capable of supporting the functions of the above communication device, such as a chip system, which may be installed in the terminal device or used in combination with the terminal device.
  • the chip system may be composed of a chip, or may include a chip and other discrete devices.
  • a wireless communication system is usually composed of cells, and a base station provides management of the cell.
  • the base station provides communication services to multiple mobile stations (MS) in the cell.
  • the base station includes a baseband unit (BBU) and a remote radio unit (RRU).
  • BBU baseband unit
  • RRU remote radio unit
  • the BBU and RRU can be placed in different places, for example: the RRU is remote and placed in an area with high traffic volume, and the BBU is placed in a central computer room.
  • the BBU and RRU can also be placed in the same computer room.
  • the BBU and RRU can also be different components under one rack.
  • a cell can correspond to one carrier or component carrier.
  • the number and type of each device in the communication system shown in Figure 1 are for illustration only, and the present application is not limited to this.
  • the communication system may also include more terminal devices, more network devices, and other network elements, such as core network devices, and/or network elements for implementing artificial intelligence functions.
  • the AI model is the specific implementation of the AI technology function.
  • the AI model represents the mapping relationship between the model's input and output.
  • the type of AI model can be a neural network, linear regression model, decision tree model, support vector machine (SVM), Bayesian network, Q learning model or other machine learning (ML) model.
  • AI models can also be specifically referred to as machine learning models, deep learning models, or reinforcement learning models.
  • Machine learning is a method for implementing artificial intelligence. The goal of this method is to design and analyze some algorithms (also known as "models") that allow computers to "learn” automatically. The designed algorithms are called “machine learning models.”
  • Machine learning models are a type of algorithm that automatically analyzes data to obtain patterns and uses the patterns to predict unknown data. There are many types of machine learning models. Depending on whether the model training needs to rely on the labels corresponding to the training data, machine learning models can be divided into supervised learning models and unsupervised learning models. The following mainly introduces "supervised learning models.”
  • a “supervised learning model” is a model obtained by determining the parameters of the initial AI model based on the data in a given training data set and the labels corresponding to each data in the training data set.
  • the process of determining the parameters of the initial AI model is also called “supervised learning” (or “supervised training”).
  • the labels of the data in the training data set are usually manually annotated to identify the correct answer to the data for a specific task.
  • Typical supervised learning models include: support vector machines, neural network models, logistic regression models, decision trees, naive Bayes models, Gaussian discriminant models, etc.
  • Supervised learning models are usually used for classification or regression. Among them, quantitative output is called “regression”, which can also be understood as the AI model is “continuous variable prediction”; qualitative output is called “classification”, which can also be understood as the AI model is “discrete variable prediction”.
  • DNN Deep neural network
  • DNN is a specific implementation of machine learning. According to the universal approximation theorem, neural networks can theoretically approximate any continuous function, thus enabling neural networks to learn any mapping.
  • Traditional communication systems require extensive expert knowledge to design communication modules, while DNN-based deep learning communication systems can automatically discover implicit pattern structures from large data sets, establish mapping relationships between data, and achieve performance that is superior to traditional modeling methods.
  • the nonlinear function is max ⁇ 0,x ⁇
  • DNN generally has a multi-layer structure. Each layer of DNN can contain multiple neurons.
  • the input layer passes the received values to the middle hidden layer after processing by neurons.
  • the hidden layer passes the calculation results to the final output layer to generate the final output of DNN, as shown in Figure 3.
  • DNN generally has more than one hidden layer, and the hidden layer often directly affects the ability to extract information and fit functions. Increasing the number of hidden layers of DNN or expanding the width of each layer can improve the function fitting ability of DNN.
  • the weighted value in each neuron is the parameter of the DNN network model. The model parameters are optimized through the training process, so that the DNN network has the ability to extract data features and express mapping relationships.
  • any AI model needs to be trained before it can be used to solve specific technical problems.
  • the training of an AI model refers to the process of using a specified initial model to calculate the training data, and adjusting the parameters in the initial model using a certain method based on the calculation results, so that the model gradually learns certain rules and has specific functions.
  • an AI model with stable functions can be used for reasoning.
  • the reasoning of an AI model is the process of using a trained AI model to calculate the input data and obtain a predicted reasoning result.
  • the training set includes multiple training data, each of which is assigned a label.
  • the label of the training data is the correct answer to a specific question.
  • the label can represent the goal of using the training data to train the deep learning model.
  • the training data can be input into the deep learning model after parameter initialization in batches.
  • the deep learning model calculates the training data (i.e., "inference") to obtain prediction results for the training data.
  • the prediction results obtained through inference and the labels corresponding to the training data are used as data for calculating the loss according to the loss function.
  • the loss function is a function used to calculate the gap between the model's prediction results for the training data and the labels of the training data (i.e., "loss value”) during the model training phase.
  • the loss function can be implemented using different mathematical functions. Commonly used expressions of loss functions include: mean square error loss function, logarithmic loss function, least squares method, etc.
  • Model training is a repeated iterative process. Each iteration infers different training data and calculates the loss value. The goal of multiple iterations is to continuously update the parameters of the deep learning model and find a parameter configuration that minimizes or stabilizes the loss value of the loss function.
  • Training data sets are used for training AI models. Training data sets may include the input of AI models, or the input and target output of AI models. Among them, the training data set includes one or more training data. The training data can be a training sample input to the AI model, or it can be the target output of the AI model. Among them, the target output can also be called a "label” or a "label sample”. Training data sets are one of the important parts of machine learning. Model training is essentially to learn certain features from the training data so that the output of the AI model is as close to the target output as possible, for example, to make the difference between the output of the AI model and the target output as small as possible. The composition and selection of the training data set can determine the performance of the trained AI model to a certain extent. Among them, the performance of the model can be measured, for example, by “loss value”, “inference accuracy” and so on.
  • a loss function can be defined during the training process of an AI model (such as a neural network).
  • the loss function describes the The gap or difference between the output value of the model and the target output value. This application does not limit the specific form of the loss function.
  • the training process of the AI model is the process of adjusting the model parameters of the AI model so that the value of the loss function is less than the threshold, or the value of the loss function meets the target requirements.
  • the AI model is a neural network, and adjusting the model parameters of the neural network includes adjusting at least one of the following parameters: the number of layers, width, weights of neurons, or parameters in the activation function of neurons.
  • “Inference data” can be used as the input of a trained AI model for inference of the AI model. During the model inference process, the inference data is input into the AI model, and the corresponding output is the inference result.
  • beam can also be understood as “spatial filter parameter”, “spatial filter” or “spatial parameters”.
  • a beam used to send signals can be called a transmission beam (Tx beam), which can be called a spatial domain transmit filter or a spatial domain transmit parameter;
  • a beam used to receive signals can be called a reception beam (Rx beam), which can be called a spatial domain receive filter or a spatial domain receive parameter.
  • the beam can be, for example, a spatial filter parameter (e.g., a spatial receive filter parameter, or a spatial transmit filter parameter).
  • a spatial filter parameter e.g., a spatial receive filter parameter, or a spatial transmit filter parameter
  • Beam scanning means that within a specific period or time period, the beam is transmitted in a predefined direction at a fixed period to cover a specific spatial area. For example, during the initial access process, the UE needs to synchronize with the system and receive minimum system information. Therefore, the synchronization signal and the physical broadcast channel (PBCH) block (SSB) are scanned and sent at a fixed period.
  • PBCH physical broadcast channel
  • SSB physical broadcast channel
  • the channel state information-reference signal (CSI-RS) can also use beam scanning technology, but if all predefined beam directions are to be covered, the overhead is too large, so CSI-RS is only transmitted in a specific subset of predefined beam directions based on the location of the terminal device being served.
  • Beam measurement refers to the process in which a network device or terminal device measures the quality and characteristics of the received beamforming signal.
  • the terminal device or network device can obtain the reference signal receiving power (RSRP), reference signal receiving quality (RSRQ), signal to interference plus noise ratio (SINR) and other information of the reference signal through SSB and CSI-RS to identify the best beam.
  • RSRP reference signal receiving power
  • RSRQ reference signal receiving quality
  • SINR signal to interference plus noise ratio
  • the network device or terminal device selects the transmit beam or receive beam it uses, wherein the downlink beam can be determined by the terminal device, for example, the decision criterion is that the maximum received signal strength of the beam should be greater than a specific threshold.
  • the terminal device transmits a sounding reference signal (SRS) according to the direction of the network device, and the network device measures the SRS to determine the best uplink beam.
  • SRS sounding reference signal
  • AI models can be deployed on training devices (for example, terminal devices) for training and updating.
  • the network device will continue (for example, periodically) to transmit the training data set to the training device until the training device sends an indication message to the network device that the training model has been completed.
  • the network device will stop sending the training data set to the training device.
  • the training data set transmitted by the network device to the training device is unnecessary for the training device, which will cause a waste of resources. Therefore, how to obtain the training data set when the training device performs AI model training becomes a technical problem that needs to be solved.
  • the present application proposes a model training method, in which the training device can request the network to send a training data set, and the request information also indicates the relevant information of the training data set that the network device needs to send to it.
  • the training device can indicate to the network device which training data sets are needed, and the network device can send the training data set indicated by the training device to the training device, without having to send the training data set all the time.
  • This method can reduce the waste of air interface resources and air interface overhead, and compared with other methods, the utilization performance of air interface resources is improved.
  • the “training device” in the present application can be understood as a terminal device, that is, the terminal device can be used with The network device communicates, and the terminal device also has the ability to support model training.
  • the "training device” can be understood as a device specifically used for model training. For example, the device can only provide the function of model training. If the device determines that the model training is completed, the trained model can be sent to the required terminal device.
  • model training is illustrated by “AI model training”, and it is assumed that the AI model is deployed on a training device.
  • FIG5 is a schematic flow chart of a method 500 for obtaining a training data set provided by the present application. The steps shown in FIG5 are described below. It should be noted that the steps indicated by dashed lines in FIG5 are optional and will not be described in detail in the following text.
  • the method includes:
  • the training device determines to monitor the performance of the AI model.
  • the training device switches from the second cell to the first cell, wherein the first network device provides services for the first cell and the second network device provides services for the second cell.
  • the training device determines to monitor the performance of the AI model, or the first network device instructs the training device to monitor the performance of the AI model.
  • the training device finds that the input information of the AI model has changed. For example, the training device finds that the sparse beam pattern does not belong to the beam pattern in the full codebook. At this time, the training device determines to monitor the performance of the AI model.
  • the first network device sends a training data set to a training device.
  • the first network device can configure reference signal resources for the training device and transmit N groups of reference signals (i.e., training data sets) to the training device.
  • N groups of reference signals i.e., training data sets
  • the training device can obtain the input information and labels of the AI model through full codebook scanning (also understood as full beam scanning).
  • the first network device sends N groups of reference signals to the training device, and the training device can measure the N groups of reference signals to obtain corresponding N groups of measurement results.
  • each group of measurement results in the N groups of measurement results includes the RSRP of the reference signal, and then the training device can determine any group of measurement results in the N groups of measurement results as the input information of the AI model, or the training device can use the N groups of measurement results as the input information of the AI model.
  • the first network device can indicate to the training device that the measurement results corresponding to the M groups of reference signals in the N groups of reference signals are used as the input information of the AI model (see the description in method 700 for details).
  • the training device can use the RSRP values of the RSRP of all reference signals in the N groups of measurement results, and the RSRP values of the largest RSRP values as the labels for the training of the AI model. After that, the training device can obtain the output information of the AI model (i.e., the inference result of the AI model) through the input information, and the output information is compared with the label, for example, the model prediction performance can be measured by training loss or training accuracy. For example, a threshold value can be set.
  • the training loss (or accuracy) is greater than or equal to the threshold value, it means that the AI model meets the requirements of the new cell or the new input information, that is, the AI model can continue to be used; if the training loss (or accuracy) is lower than the threshold value, the training device can determine that the AI model is not suitable for the requirements of the new cell or the new input information. At this time, the training device determines that the AI model needs to be updated. In this embodiment, it is assumed that the training device determines that the AI model needs to be updated, and it is necessary to continue to perform the following steps 503 to 506.
  • Step 503 The training device sends first information to the first network device, where the first information is used to indicate the relevant information of the first training data set that the training device requests the first network device to send.
  • the first network device receives the first information from the training device.
  • the relevant information of the first training data set may, for example, include at least one of the following: information on the size of the first training data set, configuration information of the input of the AI model, and configuration information of the reference signal used for training the AI model.
  • the training device may determine the "information of the size of the first training data set" in the following manner:
  • the information on the size of the first training data set may be determined by the training device based on the size of the training data set required to complete the training of the AI model.
  • the information on the size of the first training data set may be determined by the training device based on the size of the training data set required to complete the training of the AI model.
  • the training device may determine the total number of training data sets required to train the AI model from the initial state of the AI model (for example, the initial state of the AI model is 0) to a convergence state through historical information.
  • the training device determines based on historical experience that a total of 60,000 scans of the full codebook are required to obtain a training data set for training the AI model.
  • the method further includes: the training device determines the first performance of the AI model; the training device determines the size of the first training data set according to the first performance of the AI model and the second performance of the AI model, wherein the first performance is the current performance of the AI model and the second performance is the target performance of the AI model.
  • the training device monitors the AI model
  • the performance of the current AI model monitored is compared with the performance of the AI model in the historical information.
  • the corresponding model performance when the AI model converges is compared, and according to the performance of the current AI model, the size of the first training data set required to achieve the expected model performance can be estimated.
  • the performance of the AI model can be judged by measuring information such as "training loss” and "training accuracy”. For example, the "training loss” can be compared with a threshold value, and the performance of the AI model training can be measured by comparing the results.
  • the training device can indicate the size of the training data set required by the first network device, so that the first network device sends the training data set based on the instruction of the training device, thereby avoiding the first network device from continuously transmitting data to the training device during the training device model training process, reducing the waste of air interface resources, and improving the utilization performance of air interface resources.
  • reference signal configuration information may include, for example, at least one of the following: reference signal identifier, reference signal time domain resource, reference signal frequency domain resource, reference signal transmission period, and the type of the reference signal transmitted.
  • the reference signal type is SSB, or CSI-RS, or SRS, and so on.
  • the reference signal identifier may also be understood as the identifier of the reference signal group.
  • the reference signal configuration information includes the group identifiers of N groups of reference signals (N is an integer greater than or equal to 1), wherein each group of reference signals in the N groups of reference signals has the same group identifier, and each group of reference signals includes at least one reference signal.
  • the reference signal time domain resource, the reference signal frequency domain resource, the reference signal transmission period, and the type of the reference signal transmitted may also be understood as the time domain resource of N groups of reference signals, the frequency domain resource of N groups of reference signals, the transmission period of N groups of reference signals, and the type of the N groups of reference signals transmitted.
  • resources may be frequency domain resources, time domain resources, resource blocks (RB), physical resource blocks (PRB), etc., and this application is not limited thereto.
  • the training device can determine the configuration information of the reference signal by training the historical information of the AI model, and indicate it to the first network device, so that the network device can configure the reference signal for the training device, and the training device obtains the input information and label of the AI model by measuring the configured reference signal, thereby continuing to train the AI model.
  • the first network device may also indicate the input information of the AI model, thereby accelerating the convergence of the AI model.
  • the specific implementation method can refer to the following method 700. In other words, method 500 can also be combined with method 700.
  • “configuration information of the input of the AI model” can be understood as, for example, that the training device determines the input information of the AI model based on the historical information of the AI model training, for example, the input information of the AI model is the measurement result of the reference signal corresponding to the sparse beam pattern.
  • the training device can then determine which positions in the full codebook the sparse beam pattern is the beam of.
  • the training device can report information such as the identifier of the sparse beam pattern, the identifier of the reference signal corresponding to the sparse beam pattern, or the measurement result of the reference signal to the first network device, so that the first network device can send down a training data set corresponding to the beam pattern.
  • the first information is used to indicate the relevant information of the first training data set that the training device requests the network device to send
  • the first information can display the relevant information of the first training data set that the training device requests the first network device to send.
  • the first information may include information about the size of the first training data set, configuration information of the input of the AI model, and configuration information of the reference signal used for training the AI model.
  • the first information may include information indicating the specific numerical value of the number of scans.
  • the first information may implicitly indicate the relevant information of the first training data set that the training device requests the first network device to send.
  • the first information can indicate the size of the first training data set by including an index, and the first network device can determine the specific numerical value corresponding to the index by querying the index.
  • Other relevant information of the first training data set indicated by the first information can also be understood similarly, and no longer be given examples one by one.
  • the first information also includes at least one of the following: identification information of the AI model, information on the application scenario of the AI model, usage information of the AI model, and information on the computing power capabilities of the training device.
  • a first mapping relationship may be saved on a network device, where the first mapping relationship is a correspondence between the identifier of each AI model and the size of the training data set corresponding to the AI model identifier.
  • the first mapping relationship may be in the form of a table. As shown in Table 1, AI model #1 corresponds to training data set #A, AI model #2 corresponds to training data set #B, and AI model #3 corresponds to training data set #C.
  • the training device can indicate the size of the required training data set by sending the identifier of the AI model.
  • the application scenario of the AI model or the purpose of the AI model can be understood as that the AI model is used in a beam management scenario, a CSI feedback scenario, a positioning scenario, etc.
  • the training device indicates the AI model to the first network device.
  • the application scenario or purpose can indicate the size of the training data set required to train the AI model.
  • the training device may report information about computing power capabilities to the first network device.
  • the computing power capabilities information reported by the training device includes at least one of the following: the capabilities of the training device's processor (e.g., central processing unit (CPU), graphic processing unit (GPU), tensor processing unit (TPU), neural network processing unit (NPU), field-programmable gate array (FPGA), etc.), the size of the training device's storage space, the size of the training device's memory, the power of the training device, etc., are not limited.
  • the training device can indicate the maximum number of training data sets that can be processed when training the AI model. For example, the training device can scan the entire code book up to 40,000 times. At this time, the size of the training data set sent by the first network device to the training device will not exceed the computing power capabilities of the training device.
  • Step 504 The first network device sends a first training data set to the training device according to the relevant information indicated by the first information.
  • the training device receives the first training data set from the first network device.
  • the first training data set is a training data set based on the relevant information of the first training data set indicated by the first information, and the first training data set is used for training the AI model.
  • the first training data set is determined based on the relevant information of the first training data indicated by the first information.
  • the first network device may also obtain third information from the second network device, where the third information is relevant information for training the AI model on the second network device.
  • the third information includes at least one of the following: information on the size of the training data set that the training device requests the second network device to send, information on the size of the training data set required to complete the training of the AI model, identification information of the AI model, information on the computing power of the training device, and so on.
  • the third information also includes: information on the duration of the second network device sending the training data set and/or information on the manner in which the second network device sends the training data set.
  • the first network device may send the first training data set to the training device based on the relevant information indicated by the first information and the third information.
  • the relevant information indicated by the first information and the third information please refer to the description of method 600 below.
  • step 505 the training device trains the AI model based on a first training data set.
  • the training device can determine the input information and the label.
  • the first training data set is N groups of reference signals
  • the terminal device can measure the N groups of reference signals to obtain the corresponding N groups of measurement results.
  • each group of measurement results in the N groups of measurement results includes the RSRP of the reference signal, and then the training device can determine any group of measurement results in the N groups of measurement results as the input information of the AI model.
  • the network device can indicate to the terminal device that the measurement results corresponding to the M groups of reference signals in the N groups of reference signals are used as the input information of the AI model (see the description in method 700 for details).
  • the training device can use the measurement results of the RSRP of all reference signals in the N groups of measurement results as the label of the AI model. After that, the training device can obtain the output information of the AI model through the input information, and the output information is compared with the label to obtain the training loss of the AI model.
  • the above process can be understood as a training of the AI model.
  • the training device can measure the performance of the AI model based on the training loss and training accuracy of the model training, and determine the training data set required for the next model training. For example, the size of the training data set can be determined by the evaluation results of the training device based on the model performance. For example, if the training device determines that the performance of the AI model has significantly improved after the first round of training, the amount of data in the training data set can be reduced.
  • the steps include:
  • the training device can train the AI model based on the first training data set and determine the performance of the AI model; the training device can send second information to the network device based on the performance of the AI model, and the second information is used to indicate the relevant information of the second training data set that the training device requests the network device to send; the training device receives the second training data set from the network device, and the second training data set is a training data set based on the relevant information indicated by the second information, and the second training data set is used for training the AI model.
  • the amount of data in the second training data set can be smaller than the amount of data in the first training data set.
  • the training device can still train the AI based on the second training data set and iterate repeatedly, for example, perform model training based on the second training data set, determine the performance of the AI model again, determine the size of the required third training data set based on the performance of the AI model, and so on. False Assume that the training device can perform L (L is an integer greater than 1) training times until the training device determines that the AI model has converged ("model convergence" can also be understood as the AI model reaching the target performance).
  • the training device sends an indication message indicating that model training is completed to the first network device.
  • the training device After the training device completes AI model training, it can enter the model inference stage.
  • the training device can indicate the relevant information of the required training data set to the network, so that the network device can send the training data set to the training device based on the indication, without having to send the training data set all the time.
  • This method can reduce the waste of air interface resources and air interface overhead, and improve the utilization performance of air interface resources.
  • Method 500 mainly introduces the relevant information of the training data set required to be determined by the training device, so that the network device can send the training data set based on the request of the training device, thereby reducing the waste of air interface resources and improving the performance of air interface resources.
  • the following method 600 mainly introduces that if the training device performs a cell switch, the new network device after the switch is completed can obtain the relevant information for training the AI model from the old network device before the switch, and determine the training data set that needs to be sent to the training device based on the information.
  • FIG6 is a schematic flow chart of a method 600 for obtaining a training data set provided by the present application.
  • the training device performs a cell switching, at this time, the training device determines that model monitoring is required. It is assumed that the training device determines through model monitoring that the AI model needs to be updated.
  • the method 600 includes:
  • steps 601 to 602 the training device determines to monitor the performance of the AI model, and in this embodiment, it is still assumed that the training device determines that the AI model needs to be updated, and then it is necessary to continue to perform the following steps 603 to 608.
  • the implementation of steps 601 to 602 can refer to steps 501 to 502 in method 500, and will not be repeated here.
  • Step 603 The training device sends first information to the first network device, where the first information is used to request the first network device to send a training data set.
  • the training device determines that the AI model needs to be updated by monitoring the performance of the AI model, and then sends the first information to the first network device.
  • Step 604 The first network device obtains third information from the second network device, where the third information is relevant information for training the AI model.
  • the first network device may send a request message to the second network device, the request message being used to request information for training the AI model, and the second network device may synchronize the relevant information of the training AI model to the first network device based on the request message.
  • the second network device may actively provide the first network device with relevant information for training the AI model.
  • the third information may include at least one of the following: information about the size of the training data set that the training device requests the second network device to send, information about the size of the training data set required to complete AI model training, identification information of the AI model, and information about the computing power capabilities of the training device.
  • the information of the size of the training data set that the training device requests the second network device to send can be understood as the information of the size of the training data set that the training device requests from the second network device when the second network device is connected to the training device.
  • the training device can also determine the size of the training data set requested from the second network device based on historical information.
  • the information of the size of the training data set required to complete the AI model training can be understood as the size of the training data set required for the training device to train the AI model in total is stored on the second network device.
  • the network devices can jointly maintain a first mapping relationship
  • the first mapping relationship is a mapping relationship between the identifier of the AI model and the size of the training data set corresponding to the identifier of the AI model.
  • the first mapping relationship can be in the form of a table, as shown in Table 1. That is, the number of training data required for each AI model to complete the training of the AI model once in this cell is counted based on historical experience.
  • the information in Table 1 can be synchronized between network devices.
  • Table 1 is stored on both the first network device and the second network device.
  • the first network device can determine the size of the training data set for this model training based on the size of the training data set sent to the training device from the second network device. It should be noted that for some AI models, they may not be suitable for the environment of this cell at all.
  • the first network device may indicate that the AI model cannot be trained and recommend that the training device replace the AI model.
  • the second network device may also synchronize the computing power information of the training device to the first network device so that the first network can determine the size of the training data set that should be sent to the training device.
  • mapping relationship in this application can also be expressed as “association relationship” or “correspondence relationship”. It should be understood that the "mapping relationship” mentioned in the embodiments of this application can be saved or recorded in the form of a function relationship, a table, or a mapping relationship. In the following embodiments, the "mapping relationship" mentioned may be configured by a network device, or may be predefined by a protocol, etc., without limitation.
  • the third information also includes: information about the duration of the second network device sending the training data set and/or information about the manner in which the second network device sends the training data set.
  • information about the duration of the second network device sending the training data set can also be understood as the time for training the AI model when the training device is connected to the second network device. In other words, the time required to train the AI model until it completes convergence.
  • “Information about the manner in which the second network device sends the training data set” can also be understood as, for example, the second network device periodically sends the training data set to the training device.
  • the network device can identify which time period of the day has the least data request volume, which can also be understood as which time period has the least air interface resource occupancy, or which time period has sufficient air interface resources, and then the AI model can be provided with a training data set.
  • the network device can select this time period every day to support the update of the AI model (here, the number of training devices and/or the number of AI models are not limited).
  • the second network device sends the training data set to the training device at intervals. This solution has some improvement in flexibility relative to periodic transmission.
  • the network device finds and/or determines that the current data request volume is small and the air interface resources are sufficient, then the network device determines that it can support the update of the AI model. In other words, in this implementation, the network device can determine to send a training data set to the training device based on the occupancy of the air interface resources at the current moment, so it does not have obvious periodic characteristics.
  • the first network device can determine whether it has the ability to support the training of the AI model based on the information, and the first network can also determine whether the training device has the ability to support the training of the AI model based on the information (for example, information about the computing power of the training device). For example, for some AI models, the first network device determines based on historical information that the training device is indeed unable to train to a convergence state. At this time, it can be understood that the training device does not support the training of the AI model. For another example, the first network device determines that there are insufficient air interface resources to send a training data set to the training device based on the information synchronized by the second network device. At this time, it can be understood that the first network device does not support the training of the AI model.
  • steps 603 and 604. there is no limitation on the order of steps 603 and 604.
  • steps 603 and 604 may be performed simultaneously.
  • Step 605 The first network device determines a first training data set to be sent according to the third information.
  • the first network obtains relevant information about the training device training the AI model from the second network device, and determines the size of the first training data sent to the training device.
  • the first network device may provide services for multiple training devices or terminal devices at the same time.
  • the first network device may need to send training data sets to other training devices, and the first network device may also need to transmit control information to multiple terminal devices, and so on.
  • the air interface resources of the first network device are very scarce. Therefore, the first network needs to comprehensively determine the size of the training data set sent to the training device based on the current usage of the air interface resources.
  • the first network device determines that the training data set that needs to be sent to the training device is training data set #A based on the third information. However, since there are insufficient time-frequency resources on the first network device to transmit training data set #A, at this time, the first network device can determine that only part of the training data set is transmitted.
  • resource usage can also be understood as “resource occupancy”, “air interface resource occupancy”, etc.
  • the existing protocol framework has defined the maximum number of reference signals (for example, 64 CSI-RS) that can be configured by network devices. If the network device finds that all reference signals have been configured for other functions, it can be determined that the current resources are occupied and reference signal resources cannot be configured for the training device to support the update of the AI model.
  • the first information indicates relevant information of the first training data set that the training device requests the first network device to send.
  • the first information may include at least one of the following: information on the size of the requested training data set, configuration information of the input of the AI model, configuration information of the reference signal used for training the AI model, and identification information of the AI model.
  • the first information also includes at least one of the following: identification information of the AI model, information on the application scenario of the AI model, information on the use of the AI model, and information on the computing power capability of the training device.
  • this embodiment can also be combined with method 500, that is, the training device can indicate the size of the requested first training data set to the network device.
  • the first network device can comprehensively determine the size of the training data set that can be sent to the training device based on the third information, the first information, and the usage of the air interface resources.
  • Step 606 The first network device sends a first training data set to the training device.
  • the training device receives the first training data set from the first network device.
  • the first training data set is used for training the AI model.
  • Step 607 The training device trains the AI model based on the first training data set.
  • step 505 in method 500 the process of the training device in this embodiment training the AI model based on the first training data can be referred to step 505 in method 500 and will not be repeated here.
  • the training device sends an indication message indicating that model training is completed to the network device.
  • the training device After the training device completes AI model training, it can enter the model inference stage.
  • the network device can combine various information to comprehensively decide whether to support the training of the AI model.
  • the network device will not continuously send training data sets to the training device, which can reduce unnecessary air interface resource occupation, save air interface overhead, and improve the utilization performance of air interface resources.
  • the above method 500 and method 600 respectively introduce methods for obtaining training data sets from the perspective of a training device and a perspective of a network device.
  • FIG. 7 is a schematic flow chart of a method 700 for obtaining input information of an AI model provided in the present application, and method 700 includes:
  • steps 701 to 702 the training device determines to monitor the performance of the AI model, and in this embodiment, it is still assumed that the training device determines that the AI model needs to be updated, and then the following steps 703 to 710 need to be continued.
  • the implementation of steps 701 to 702 can refer to steps 501 to 502 in method 500, and will not be repeated here.
  • the training device sends first information to the first network device, where the first information is used to request the first network device to send a training data set.
  • the first network device receives the first information from the training device.
  • the first information indicates relevant information of the first training data set that the training device requests the first network device to send.
  • the first information may include at least one of the following: information on the size of the requested training data set, configuration information of the input of the AI model, configuration information of the reference signal used for training the AI model, and identification information of the AI model.
  • the first information also includes at least one of the following: identification information of the AI model, information on the application scenario of the AI model, information on the purpose of the AI model, and information on the computing power capability of the training device.
  • this embodiment can also be combined with method 500, that is, the training device can indicate the size of the requested training data set to the network device.
  • the first network device determines the size of a first training data set to be sent to the training device.
  • the first network device can obtain the third information from the second network device, and the third information is the relevant information of the training of the AI model.
  • the third information may include at least one of the following: information on the size of the training data set requested by the training device to be sent by the second network device, information on the size of the training data set required to complete the AI model training, identification information of the AI model, and information on the computing power of the training device.
  • the third information also includes: information on the duration of the second network device sending the training data set and/or information on the manner in which the second network device sends the training data set.
  • the first network device can determine the first training data set sent to the training device based on the third information, and the specific implementation method can refer to the description of the above method 600.
  • the first network device can determine the first training data set sent to the training device based on the relevant information indicated by the first information and the third information.
  • the first network device can determine the first training data set sent to the training device based on the relevant information indicated by the first information, the third information, and the resource usage of the first network device.
  • this embodiment can be combined with method 600, that is, the network device can comprehensively determine the size of the first training data sent to the training device.
  • the first network device sends N (N is an integer greater than 1) groups of reference signals to the training device.
  • the training device receives N groups of reference signals from the first network device.
  • the first network device sends N groups of reference signals to the training device can also be understood as the first network device sending the first training data set to the training device.
  • the first network device can configure reference signal resources for the training device, and send reference signal resources (for example, N groups of reference signals) to the training device.
  • Each group of reference signals in the N groups of reference signals includes at least one reference signal, and each group of reference signals has the same group identifier.
  • the group identifier corresponding to each group of reference signals can also be understood as a beam identification number, for example, a beam identifier; or, the group identifier corresponding to each group of reference signals can also be understood as an identifier of the resource of each group of reference signals.
  • the first network device can instruct the training device to perform a full codebook beam scan. That is, the first network device instructs the training device to measure N groups of reference signals.
  • the first network device sends configuration information to the training device
  • the configuration information may be used to indicate one or more of the following: the time domain resources of the N groups of reference signals, the frequency domain resources of the N groups of reference signals, the transmission period of the N groups of reference signals, and the group identifier of the N groups of reference signals.
  • the configuration information may include the size of the time domain resources of the N groups of reference signals, the size of the frequency domain resources of the N groups of reference signals, the information of the transmission period of the N groups of reference signals, etc. In this case, it can also be understood that the configuration information is displayed to indicate.
  • the configuration information may carry the index of the time domain resources of the N groups of reference signals, the index of the frequency domain resources of the N groups of reference signals, the index of the transmission period of the N groups of reference signals, etc. In this case, it can also be understood that the configuration information is implicitly indicated.
  • Step 706 The training device measures N groups of reference signals and obtains N groups of measurement results corresponding to the N groups of reference signals.
  • each group of measurement results in N groups of measurement results may include at least one measurement quantity.
  • each reference signal in each group of reference signals one or more of its RSRP, RSRQ, SINR, etc. can be measured. That is, each group of measurement results may include one or more of the measurement results of RSRP, RSRQ, and SINR measured for each reference signal in the group of reference signals.
  • each group of measurement results corresponds to the same group identifier.
  • reference signal group #A includes reference signal #A1 and reference signal #A2, and the training device can measure reference signal #A1 and reference signal #A2 respectively, for example, measuring the RSRP and SINR of reference signal #A1 and the RSRP and SINR of reference signal #A2, then the reference signal #A1 measurement result includes the measurement result of the RSRP of reference signal #A1 and the measurement result of the SINR, then the reference signal #A2 measurement result includes the measurement result of the RSRP of reference signal #A2 and the measurement result of the SINR.
  • the training device can determine the training label based on the beam scanning result of the full codebook. Specifically, if the AI model training adopts the regression method for model training, for example, the training device can determine to use the RSRP measurement results of N groups of reference signals as the training label; if the AI model training adopts the classification method for model training, for example, the training device can determine to use the beam identifier corresponding to the K groups of measurement results with the best channel quality among the N groups of reference signals as the training label.
  • Step 707 The first network device sends fourth information to the training device, where the fourth information is used to indicate M groups of reference signals among the N groups of reference signals.
  • the fourth information can be used to indicate a sparse beam pattern.
  • the first network device can indicate the input information of the AI model to the training device.
  • the first network device can indicate to the training device which beams in the full codebook constitute the beam pattern of the input information of the AI model.
  • the fourth information can indicate which beams of the 64 beams constitute the input information.
  • the beam identifier in the sparse beam pattern (“beam identifier” can also be understood as "group identifier of the reference signal group”) is unified with the identifier of the beam in the full code book. For example, there are 64 beams in the full code book, and the beam identifiers are beam #1 to beam #64, respectively, and the sparse beam pattern sent by the first network device contains 16 beams, namely beam #16 to beam #32. For example, in method 1, the beam identifier and the reference signal group identifier are one-to-one corresponding.
  • the network device indicates the sparse beam pattern by indicating the beam identifier (an example of beam information) and/or the group identifier of the reference signal group to the training device.
  • the beam identifiers in the sparse beam pattern are not consistent with the beam identifiers in the full codebook. It can also be understood that the beam identifiers in the sparse beam pattern do not correspond to the beam identifiers in the full codebook. For example, there are 64 beams in the full codebook, and the beam identifiers are beam #1 to beam #64, and the sparse beam pattern sent by the first network device contains 16 beams, beam #1 to beam #16, and the training device cannot parse beam #1 to beam #16 in the sparse beam pattern as beam #1 to beam #16 in the full codebook.
  • the sparse beam pattern contains 16 beams, beam #1 to beam #16, which should actually correspond to beam #1, Beam #4, Beam #8, Beam #12, Beam #16, Beam #20, Beam #24, Beam #28, Beam #32, Beam #36, Beam #40, Beam #44, Beam #48, Beam #52, Beam #56, Beam #60.
  • the fourth information includes N fields, and the N fields correspond one-to-one to the N groups of reference signals, wherein the bit values of M fields in the N fields are different from the bit values of the remaining (NM) fields, and the first network device can indicate the M groups of reference signals through the M fields.
  • the bit values of the M fields are all "1", and the bit values of the remaining (NM) fields are all "0", at which time the training device can determine which reference signal groups the input information corresponds to the measurement results. Therefore, the network device can specifically indicate through the field that the sparse beam pattern is those beams in the full codebook.
  • the first network device may also send fifth information to the training device, and the fifth information is used to indicate the P groups of reference signals in the N groups of reference signals. It can also be understood that the first network device can indicate to the training device the measurement results of the reference signal corresponding to another beam pattern as the input information of the AI model. In other words, in the present application, the first network device can indicate to the training device the measurement results of the reference signals corresponding to multiple beam patterns as the input information of the AI model, respectively, to perform model training, so that the trained AI model can converge for all beam patterns.
  • step 705 and step 707 may not be performed in any order.
  • the configuration information and the fourth information may also be sent in the same message, which is not limited.
  • Step 708 The training device determines the first input information of the AI model based on the fourth information and the N groups of measurement results corresponding to the N groups of reference signals.
  • the training device Since the training device has measured N groups of reference signals and obtained N groups of measurement results corresponding to the N groups of reference signals in step 705, the training device can determine which groups of measurement results among the N groups of measurement results can be used as input information of the AI model based on the fourth information.
  • the training device can use the measurement results of the second group of reference signals, the measurement results of the fourth group of reference signals, and the measurement results of the eighth group of reference signals as input information of the AI model.
  • the first input information can be the measurement results of the RSRP of the M groups of reference signals.
  • Step 709 The training device obtains first output information based on the first input information.
  • the "first output information" can also be understood as the inference output result of the AI model.
  • the first input information can be the measurement result of the RSRP of M groups of reference signals.
  • the first output information can include N group identifiers corresponding to the measurement results of the RSRP of N groups of reference signals; if the AI model training adopts the classification method for model training, for example, the first input information can be the measurement result corresponding to M groups of reference signals.
  • the first output information can be K group identifiers corresponding to K groups of reference signals with the best channel quality measurement results of N groups of reference signals, for example, which can be understood as K beam identifiers.
  • the training device can compare the first output information with the label to obtain the training loss of the AI model.
  • the first output information is the K group identifiers corresponding to the K groups of reference signals with the best channel quality measurement results of the N groups of reference signals inferred by the training device, and the training labels determined by the training device are assumed to be the K group identifiers corresponding to the K groups of reference signals with the best channel quality measurement results of the N groups of reference signals during the full codebook scan.
  • the training device can compare the output results with the training labels, determine the performance of the AI model, and adjust the model parameters.
  • the above process can be understood as a training of the AI model.
  • the training device can measure the performance of the AI model based on the training loss and training accuracy of the model training, and determine the training data set required for the next model training. For example, the size of the training data set can be determined by the training device based on the evaluation results of the model performance. For example, if the training device determines that the performance of the AI model has significantly improved after the first round of training, the amount of data in the training data set can be reduced. Specifically, it also includes the steps of:
  • the training device can train the AI model based on the first training data set and determine the performance of the AI model; the training device can send a second information to the network device based on the performance of the AI model, and the second information is used to indicate the relevant information of the second training data set that the training device requests the network device to send; the training device receives a second training data set from the network device, and the second training data set is a training data set based on the relevant information indicated by the second information, and the second training data set is used for training the AI model.
  • the amount of data in the second training data set can be smaller than the amount of data in the first training data set.
  • the training device can still train the AI based on the second training data set and iterate repeatedly, assuming that the training device performs Q (Q is an integer greater than 1) training times until the training device determines that the AI model converges ("model convergence" can also be understood as the AI model reaches the target performance).
  • the training device when the training device subsequently performs model training based on the requested training data set, the AI The sparse beam pattern of the model can be fixed, or it can also be understood that in the subsequent Q training processes, the sparse beam pattern is still the beam pattern indicated by the fourth information in step 707. It should be understood that during each training, the training device will perform a full codebook scan based on the training data set issued by the first network device. Since the channel state (also understood as the channel environment) is time-varying, the measurement results of the reference signal obtained after each full codebook scan are not exactly the same.
  • the corresponding measurement results of the M groups of reference signals in the N groups of reference signals are also different, and the training labels determined by the training device are also different, that is, the input information and training labels of the AI model will change accordingly, but these changes are essentially caused by changes in the channel state, and the beam pattern has not changed. That is, in the solution provided in this embodiment, the only variable in the AI model training process is the channel state.
  • the input information of the AI model is the measurement results of all reference signals obtained after the full codebook scan, that is, the beam pattern and channel state are changing each time the training is performed, and the model convergence performance is poor during AI model training.
  • the network device can indicate which measurement results obtained in the full codebook scan are the input information of the AI model, that is, only the channel state changes. Compared with the aforementioned other scheme, the convergence speed of the AI model can be accelerated, the model training efficiency can be improved, and the occupation of air interface resources can be reduced.
  • the training device sends a model training completion indication message to the first network device.
  • the training device After the training device completes the AI model training, it can enter the model reasoning stage. For example, the first network device can subsequently send a sparse beam pattern and a corresponding reference signal to the training device, and the training device obtains the input information of the AI model by measuring the reference signal. Assuming that the AI model adopts a classification method, the input of the AI model is the measurement result of the reference signal. At this time, the AI model can output K beam identifiers after reasoning. The K beam identifiers are the beams corresponding to the K measurement results with the best channel quality in the reference signal measurement results in the full codebook inferred by the training device.
  • the training device can feed back the K beam identifiers to the first network device, and the first network device sends the K groups of reference signals corresponding to the K beams to the training device again.
  • the training device measures the K groups of reference signals again, and determines one of the groups of reference signals with the best measurement results, and uses the beam identifier corresponding to the group of reference signals as the final selected beam to communicate with the first network device.
  • the network device can indicate the input information of the AI model to the training device, so that only the channel state changes during the AI model training process, thereby accelerating the convergence speed of the AI model, improving the model training efficiency, and thus reducing the occupancy of air interface resources.
  • the above method 700 provides a method for obtaining AI model input information.
  • the following method 800 provides a communication method, which describes in more detail the method for obtaining AI model input information during the training phase. This method can be implemented independently of the above input information acquisition method or can be used in combination. As shown in Figure 8, the method 800 includes:
  • Step 801 A network device sends a second reference signal set to a training device.
  • the training device receives a second reference signal set from the network device.
  • the second reference signal set includes N groups of reference signals, and each group of reference signals in the N groups of reference signals includes at least one reference signal.
  • “Second reference signal set” can also be understood as a reference signal set corresponding to a full codebook beam, for example, each beam in the full codebook beam can correspond to one of the reference signals in the second reference signal set.
  • the method may also include: the network device sends first configuration information to the training device, and the first configuration information can be used to indicate one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission periods of the N groups of reference signals, group identifiers of the N groups of reference signals, or beams of the N groups of reference signals.
  • Each group of reference signals has a group identifier.
  • the group identifier of the group of reference signals is the identifier of the reference signal, that is, the group identifier can be replaced by the identifier of the reference signal.
  • the method may also include: the network device sends first configuration information to the training device, the first configuration information including the group identifier of each of the N groups of reference signals, and the third configuration information can be used to indicate one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission periods of the N groups of reference signals, or beams of the N groups of reference signals.
  • the method may further include: the network device sends first configuration information to the training device, the first configuration information includes beam information of each of the N groups of reference signals, and the third configuration information may be used to indicate one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission periods of the N groups of reference signals, Or the group identifier of N groups of reference signals.
  • Step 802 The network device sends second beam indication information to the training device, where the second beam indication information indicates the beam corresponding to the first reference signal set.
  • the training device receives second beam indication information from the network device.
  • the beam corresponding to the first reference signal set is a subset of the multiple beams corresponding to the second reference signal set, and the first reference signal set includes M groups of reference signals, where N is an integer greater than M, and M is an integer greater than or equal to 1.
  • the network device sending the second beam indication information to the training device can also be understood as the network device indicating the sparse beam pattern to the training device.
  • the second beam indication information can indicate to the training device which beams in the full codebook the sparse beam pattern that the training device needs to scan is composed of.
  • the second beam indication information indicates the beam corresponding to the first reference signal set.
  • the second beam indication information indicates the position of the beam corresponding to the first reference signal set in the multiple beams corresponding to the second reference signal set.
  • the beam identifiers are beam #1 to beam #64, respectively, and the sparse beam pattern sent by the first network device includes 16 beams.
  • the second beam information can indicate which beams in the full code book the 16 beams are (that is, which beams in the full code book the first reference signal set corresponds to).
  • the second beam information includes N fields, and the N fields correspond one-to-one to the multiple beams corresponding to the second reference signal set, wherein the bit values of M fields in the N fields are different from the bit values of the remaining (N-M) fields, and the network device can indicate the first reference signal set through the M fields.
  • the bit values of the M fields are all "1"
  • the bit values of the remaining (N-M) fields are all "0".
  • there are 64 beams in the full code book and the beam identifiers are beam #1 to beam #64, respectively, and the sparse beam pattern sent by the network device contains 16 beams, respectively, beam #1 to beam #16, and the fields #1 to #16 in the second beam indication information respectively indicate beam #1 to beam 16 in the full code book.
  • the fields #1 to #16 in the second beam indication information respectively indicate reference signal group #1 to reference signal group #16 in the first reference signal set.
  • Mode A can also be understood as that the network device can directly indicate the beam position.
  • the second beam indication information includes group identifiers or beam information (e.g., beam identifiers) of M groups of reference signals, wherein the M groups of reference signals are part of the N groups of reference signals, and the N groups of reference signals have a predefined or preconfigured correspondence with the N beams.
  • the N groups of reference signals correspond one-to-one to the N beams.
  • the group identifiers of the N groups of reference signals correspond one-to-one to the N beam identifiers.
  • there are 64 beams in the full codebook and the beam identifiers are beam #1 to beam #64, respectively, and the sparse beam pattern sent by the first network device includes 16 beams, respectively, beam #16 to beam #32.
  • the beam identifier and the group identifier of the reference signal are in one-to-one correspondence.
  • the network device can indicate the sparse beam pattern by indicating the beam identifier (an example of beam information) and/or the group identifier of the reference signal to the training device.
  • the beam position is indirectly indicated by the group identifier or beam information of the reference signal. Since the beam information corresponds to the group identifier of the reference signal, and the relationship between the group identifier/beam information of M groups of reference signals and the group identifier/beam information of N groups of reference signals is fixed.
  • Step 803 The training device determines the first input information of the AI model.
  • the training device can measure N groups of reference signals in the second reference signal set and obtain measurement results of the N groups of reference signals. At this time, the training device can determine the measurement result corresponding to the beam indicated by the second beam indication information based on the second beam indication information and the measurement results of the N groups of reference signals in the second reference signal set, and use the measurement result corresponding to the beam indicated by the second beam indication information of the AI model as the first input information. That is, the measurement result of the beam corresponding to the sparse beam pattern is used as the first input information.
  • the training device can determine which beam positions in the full codebook correspond to the measurement results of the reference signals of the input information of the AI model. For example, in the above method A, the training device can use the measurement results of reference signal group #16 to reference signal group #32 corresponding to beam #16 to beam #32 in the full codebook as the input information of the AI model.
  • the training device can use the measurement results of reference signal group #1, reference signal group #4, reference signal group #8, reference signal group #12, reference signal group #16, reference signal group #20, reference signal group #24, reference signal group #28, reference signal group #32, reference signal group #36, reference signal group #40, reference signal group #44, reference signal group #48, reference signal group #52, reference signal group #56, and reference signal group #60 as the input information of the AI model.
  • the training device performs model training based on the determined first input information and the AI model to obtain first output information.
  • the first output information indicates K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set, where K is an integer greater than or equal to 1, and K is less than N.
  • the first output information may include information about the K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set.
  • the first output information may include group identifiers or beam information of each of the K groups of reference signals, wherein the K groups of reference signals correspond to the K measurement results predicted to have the best channel quality among the N measurement results corresponding to the N groups of reference signals, and the group identifiers of the K groups of reference signals and the K beams, that is, the K beam information, have a predefined or preconfigured correspondence.
  • the first output information may include multiple beam information corresponding to the N groups of reference signals and the predicted N measurement results corresponding to the beam information.
  • the first output information may include group identifiers of the N groups of reference signals and the predicted N measurement results of the N groups of reference signals, and the N groups of reference signals have a predefined or preconfigured correspondence with the N beam information.
  • the "first output information" can also be understood as the training output result of the AI model.
  • the first input information can be the measurement result of RSRP of M groups of reference signals.
  • the first output information can include the measurement results of RSRP of the reference signals corresponding to the predicted N beams and their respective corresponding N group identifiers;
  • the AI model training adopts classification method for model training for example, the first input information can be the measurement results of M groups of reference signals corresponding to M beams.
  • the first output information can be the beam identifiers of the K beams with the best channel quality measurement results of the predicted N beams or the K group identifiers corresponding to the K beams.
  • the training device can compare the first output information with the label to obtain the training loss of the AI model.
  • the first output information is the K group identifiers corresponding to the K beams with the best channel quality measurement results of the N beams inferred by the training device
  • the training label determined by the training device is assumed to be the K group identifiers corresponding to the K groups of reference signals with the best channel quality measurement results of the N groups of reference signals corresponding to the N beams during the full codebook scan.
  • the training device can compare the output results with the training labels, determine the performance of the AI model, and adjust the model parameters.
  • the above process can be understood as a training of the AI model.
  • the training device can measure the performance of the AI model based on the training loss and training accuracy of the model training, and iterate repeatedly until the model converges.
  • the network device can indicate the input information of the AI model to the training device, so that for the same sparse beam pattern, the channel state changes during the AI model training process, thereby accelerating the convergence speed of the AI model and improving the model training efficiency, thereby reducing the occupancy of air interface resources.
  • the above methods 700 and 800 mainly introduce the specific implementation scheme of obtaining input information in the model training stage.
  • the method 900 in Figure 9 below mainly introduces that this method can also be used in the model reasoning stage.
  • the AI model has been trained.
  • the methods in methods 700 and 800 can be used for model training; for another example, the existing scheme can be used to train the AI model.
  • a reasoning device such as a terminal device, can be directly pre-configured with multiple trained AI models.
  • each AI model is trained by one or more sparse beam patterns to complete the AI model. Therefore, the model reasoning stage can be directly accessed. It can be understood that the method described in Figure 9 can be applied independently with Figure 8, or it can be applied in combination with the method in Figure 8.
  • X are the configuration information of the training process and the configuration information of the reasoning process, respectively, for the purpose of distinction.
  • configuration information they are the configuration information of the training process and the configuration information of the reasoning process, respectively.
  • Step 901 A network device sends a first reference signal set to a terminal device.
  • the terminal device receives the first reference signal set from the network device.
  • the first reference signal set includes M groups of reference signals, each of the M groups of reference signals includes at least one reference signal, where M is an integer greater than or equal to 1.
  • the "first reference signal set” may also be understood as a reference signal set corresponding to a sparse beam pattern, for example, the sparse beam pattern is a subset of a full codebook beam, or the reference signal set corresponding to the sparse beam pattern belongs to a subset of a second reference signal set corresponding to a full codebook beam (that is, the first reference signal set is a subset of the second reference signal set), where the second reference signal set includes N groups of reference signals, each of the N groups of reference signals includes at least one reference signal.
  • the method may further include: the network device sends first configuration information to the terminal device, and the first configuration information can be used to indicate one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission periods of the N groups of reference signals, group identifiers of the N groups of reference signals, or beam information of the N groups of reference signals.
  • the method may further include: the network device sends second configuration information to the terminal device, the second configuration information includes first beam information, the first beam information includes a group identifier of M groups of reference signals, and the second configuration information includes a first beam information including a group identifier of M groups of reference signals.
  • the configuration information may also include one or more of the time domain resources of the M groups of reference signals, the frequency domain resources of the M groups of reference signals, the transmission period of the M groups of reference signals, or the beam information.
  • the method may further include: the network device sends second configuration information to the terminal device, the second configuration information includes first beam information, the first beam information includes beam information of M groups of reference signals, and the second configuration information may also include one or more of time domain resources of M groups of reference signals, frequency domain resources of M groups of reference signals, transmission periods of M groups of reference signals, or group identifiers.
  • the method may further include: the network device sends third configuration information to the terminal device, the third configuration information includes the group identifiers of the N groups of reference signals, and the third configuration information may be used to indicate one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission periods of the N groups of reference signals, or beams of the N groups of reference signals.
  • the N group identifiers of the N groups of reference signals include the M group identifiers of the M groups of reference signals.
  • the method may further include: the network device sends third configuration information to the terminal device, the third configuration information includes beam information of each of the N groups of reference signals, and the third configuration information may be used to indicate one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission periods of the N groups of reference signals, or group identifiers of the N groups of reference signals.
  • the N beam information of the N groups of reference signals includes the M beam information of the M groups of reference signals.
  • Step 902 The network device sends first beam indication information to the terminal device, where the first beam indication information indicates a beam corresponding to a first reference signal set.
  • the terminal device receives the first beam indication information from the network device.
  • the beam corresponding to the first reference signal set is a subset of the multiple beams corresponding to the second reference signal set.
  • the network device sending the first beam indication information to the terminal device can also be understood as the network device indicating the sparse beam pattern to the terminal device.
  • the first beam indication information can indicate to the terminal device which beams in the full codebook the sparse beam pattern that the terminal device needs to scan is composed of.
  • the second beam indication information indicates the beam corresponding to the first reference signal set.
  • the second beam indication information indicates the position of the beam corresponding to the first reference signal set in the multiple beams corresponding to the second reference signal set.
  • the beam identifiers are beam #1 to beam #64, respectively, and the sparse beam pattern sent by the first network device includes 16 beams.
  • the second beam information can indicate which beams in the full code book the 16 beams are (that is, which beams in the full code book the first reference signal set corresponds to).
  • the first beam information includes N fields, and the N fields correspond one-to-one to the multiple beams corresponding to the second reference signal set, wherein the bit values of M fields in the N fields are different from the bit values of the remaining (N-M) fields, and the network device can indicate the first reference signal set through the M fields.
  • the bit values of the M fields are all "1"
  • the bit values of the remaining (N-M) fields are all "0".
  • there are 64 beams in the full code book and the beam identifiers are beam #1 to beam #64, respectively, and the sparse beam pattern sent by the network device contains 16 beams, respectively, beam #1 to beam #16, and the fields #1 to #16 in the second beam indication information respectively indicate beam #1 to beam 16 in the full code book.
  • the fields #1 to #16 in the second beam indication information respectively indicate reference signal group #1 to reference signal group #16 in the first reference signal set.
  • Mode A can also be understood as that the network device can directly indicate the beam position.
  • the first beam indication information includes group identifiers or beam information (e.g., beam identifiers) of M groups of reference signals, wherein the M groups of reference signals are part of the N groups of reference signals, and the N groups of reference signals have a predefined or preconfigured correspondence with the N beams.
  • the N groups of reference signals correspond one-to-one to the N beams.
  • the group identifiers of the N groups of reference signals correspond one-to-one to the N beam identifiers.
  • there are 64 beams in the full codebook and the beam identifiers are beam #1 to beam #64, respectively, and the sparse beam pattern sent by the first network device includes 16 beams, respectively, beam #16 to beam #32.
  • the beam identifier and the group identifier of the reference signal are in one-to-one correspondence.
  • the network device can indicate the sparse beam pattern by indicating the beam identifier (an example of beam information) and/or the group identifier of the reference signal group to the terminal device.
  • method B indirectly indicates the beam position through the group identifier or beam information of the reference signal. Since the beam information corresponds to the group identifier of the reference signal, and the relationship between M groups of reference signal group identifiers/beam information and N groups of reference signal identifiers/beam information can be fixed.
  • the terminal device determines the AI model based on the received first beam indication information, that is, the sparse beam pattern.
  • the terminal device can determine the AI model based on the sparse beam pattern. For example, the terminal device can identify the received sparse beam pattern, and through the identification of the beam pattern, it can determine an AI model corresponding to the sparse beam pattern from multiple AI models that have been preconfigured locally. It can also be understood that the terminal device can determine the best matching AI model from multiple AI models based on the sparse beam pattern.
  • the terminal device identifying the sparse beam pattern can be understood as the terminal device needs to determine which positions of the beams in the full codebook the received beam pattern belongs to.
  • the terminal device needs to establish a connection between the received sparse beam pattern and the beams in the full codebook, that is, it needs to determine which beams in the full codebook the beams in the sparse beam pattern belong to.
  • the terminal device can determine the sparse beam pattern based on method A and/or method B.
  • Step 904 The terminal device determines the first input information of the AI model.
  • the terminal device can measure M groups of reference signals in the first reference signal set and obtain measurement results of the M groups of reference signals. At this time, the terminal device can determine the first input information of the AI model based on the measurement results of the M groups of reference signals.
  • the terminal device can determine which beam positions in the full codebook the input information of the AI model correspond to the measurement results of the reference signals. For example, in the above method A, the terminal device can use the measurement results of reference signal group #16 to reference signal group #32 corresponding to beam #16 to beam #32 in the full codebook as the input information of the AI model.
  • the terminal device can use the measurement results of reference signal group #1, reference signal group #4, reference signal group #8, reference signal group #12, reference signal group #16, reference signal group #20, reference signal group #24, reference signal group #28, reference signal group #32, reference signal group #36, reference signal group #40, reference signal group #44, reference signal group #48, reference signal group #52, reference signal group #56, and reference signal group #60 as the input information of the AI model.
  • step 905 the terminal device performs model inference based on the determined first input information and the AI model to obtain first output information.
  • the first output information indicates K beams predicted to have the best channel quality among the multiple beams corresponding to the second reference signal set, where K is an integer greater than or equal to 1, and K is less than N.
  • This step can refer to the description of the first output information in the reasoning process involved in the training process in step 804 in Figure 8, which will not be repeated here.
  • the difference between this step and step 804 is that this step does not need to obtain the first output information and the label through loss, and the first output information is a usable prediction result.
  • step 906 the terminal device sends the first output information to the network device.
  • the network device may send the reference signal corresponding to the first output information to the terminal device again, and the terminal device measures the reference signal again, determines the reference signal with the best measurement result, and uses the beam identifier corresponding to the reference signal as the final selected beam to communicate with the network device.
  • the optimal measurement result may include the maximum RSRP value, or the maximum SINR value, or other evaluation criteria, which are not limited here.
  • the AI model adopts a classification method.
  • the AI model can output K beam identifiers after reasoning, and the K beam identifiers are the beams corresponding to the K measurement results with the best channel quality in the reference signal measurement results in the full codebook inferred by the terminal device.
  • the terminal device can feed back the K beam identifiers to the first network device, and the first network device again sends the K groups of reference signals corresponding to the K beams to the terminal device.
  • the terminal device measures the K groups of reference signals again, and determines one of the groups of reference signals with the best measurement results, and uses the beam identifier corresponding to the group of reference signals as the final selected beam to communicate with the first network device.
  • the terminal device can identify the sparse beam pattern and further determine the input information of the AI model, so that the model reasoning results are more accurate.
  • predefined in this application may be understood as definition, pre-definition, storage, pre-storage, pre-negotiation, pre-configuration, fixed configuration, etc. or pre-fired.
  • a and/or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural.
  • the character "/” generally indicates that the previous and next associated objects are in an "or” relationship; in the formula of this application, the character "/" indicates that the previous and next associated objects are in a "division" relationship.
  • each node such as a training device, a network device
  • each node includes a hardware structure and/or software module corresponding to each function in order to realize the above functions.
  • the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.
  • the network device and the training device include hardware structures and/or software modules corresponding to the execution of each function.
  • the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application scenario and design constraints of the technical solution.
  • Figures 10 and 11 are schematic diagrams of the structures of possible communication devices provided by embodiments of the present application. These communication devices can be used to implement the functions of the training device or network device in the above method embodiments, and thus can also achieve the beneficial effects possessed by the above method embodiments.
  • the communication device can be one of the terminal devices (an example of a training device, or an example of an inference device) 120a-120j as shown in Figure 1, or it can be a network device 110a or 110b as shown in Figure 1, or it can be a module (such as a chip) applied to a terminal device or a network device.
  • the communication device 100 includes a processing unit 120 and a transceiver unit 110.
  • the communication device 100 is used to implement the functions of the training device or network device in the method embodiments shown in Figs. 5 to 9 above.
  • the transceiver unit 110 is used to send a first information, and the first information is used to indicate the relevant information of the first training data set requested to be sent by the device; the transceiver unit 110 is also used to receive the first training data set, and the first training data set is a training data set based on the relevant information indicated by the first information, and the first training data set is used for training an artificial intelligence AI model.
  • the processing unit 120 is used to train the AI model based on the first training data set and determine the performance of the AI model; the processing unit 120 is also used to control the transceiver unit 110 to send second information based on the performance of the AI model, and the second information is used to indicate relevant information of the second training data set requested to be sent by the device; the transceiver unit 110 is used to receive the second training data set, and the second training data set is a training data set based on the relevant information indicated by the second information, and the second training data set is used for training the AI model.
  • the transceiver unit 110 is used to receive first information, and the first information is used to indicate relevant information of a first training data set requested to be sent by the device; the processing unit 120 is used to control the transceiver unit 110 to send the first training data set according to the relevant information indicated by the first information, and the first training data set is used for training the artificial intelligence AI model.
  • the transceiver unit 110 is used to obtain third information, where the third information is relevant information for training the AI model, and the processing unit 120 is used to control the transceiver unit 110 to send the first training data set according to the relevant information indicated by the first information, including: the processing unit 120 is used to control the transceiver unit 110 to send the first training data set according to the relevant information indicated by the first information and the third information.
  • the transceiver unit 110 is used to receive second information, where the second information is used to indicate relevant information of a second training data set requested to be sent, wherein the second information is determined based on the performance of the AI model, and the AI model performance is determined based on training of the first training data set; the processing unit 120 is used to control the transceiver unit 110 to send the second training data set according to the relevant information indicated by the second information, and the second training data set is used for training the AI model.
  • the transceiver unit 110 is used to obtain the third information, and the third information is the relevant information for training the artificial intelligence AI model; the transceiver unit 110 is used to receive the first information, The first information is used to request the sending of a training data set; the processing unit 120 is used to determine the first training data set to be sent according to the third information; the processing unit 120 is used to control the transceiver unit 110 to send the first training data set based on the first information, and the first training data set is used for training the AI model.
  • the third information also includes: information on the duration of sending the training data set and/or information on the method of sending the training data set, and the processing unit 120 is used to determine whether the device and/or training equipment has the ability to support training the AI model based on the third information and the resource usage of the device.
  • the processing unit 120 is used to determine the first training data set to be sent based on the third information, including: the processing unit 120 is used to determine the first training data set to be sent based on the first information and the third information, wherein the first information is used to indicate relevant information of the first training data set requested to be sent by the first network device.
  • the transceiver unit 110 is used to receive second information, where the second information is used to indicate relevant information of a second training data set requested to be sent by the device, wherein the second information is determined based on the performance of the AI model, and the performance of the AI model is determined based on training of the first training data set; the processing unit 120 is used to determine the second training data set to be sent based on the second information.
  • the processing unit 120 is used to measure N groups of reference signals and obtain N groups of measurement results corresponding to the N groups of reference signals, wherein each group of reference signals in the N groups of reference signals includes at least one reference signal, and each group of reference signals has the same group identifier, and N is an integer greater than 1;
  • the transceiver unit 110 is used to receive fourth information, and the fourth information is used to indicate M groups of reference signals in the N groups of reference signals;
  • the processing unit 110 is used to determine first input information of an artificial intelligence AI model according to the fourth information and the N groups of measurement results corresponding to the N groups of reference signals, and the first input information includes the M groups of measurement results corresponding to the M groups of reference signals;
  • the AI model is used to obtain first output information based on the first input information, wherein the first output information includes the group identifiers of the K groups of reference signals in the N groups of reference signals, and the group identifiers of the K groups of reference signals correspond to
  • the transceiver unit 110 is used to receive configuration information, where the configuration information is used to indicate one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission periods of the N groups of reference signals, and group identifiers of the N groups of reference signals.
  • the transceiver unit 110 is used to send N groups of reference signals to the training device, wherein each group of reference signals in the N groups of reference signals includes at least one reference signal, and each group of reference signals has the same group identifier, and N is an integer greater than 1; the transceiver unit 110 is used to send fourth information to the training device, and the fourth information is used to indicate M groups of reference signals in the N groups of reference signals, wherein the M groups of reference signals are used to determine first input information; the AI model is used to obtain first output information based on the first input information, wherein the first output information includes the group identifiers of K groups of reference signals in the N groups of reference signals, wherein the group identifiers of the K groups of reference signals correspond to K groups of measurement results with the best channel quality among the N groups of measurement results corresponding to the N groups of parameter signals.
  • the transceiver unit 110 is used to send configuration information, where the configuration information is used to indicate one or more of the following: time domain resources of the N groups of reference signals, frequency domain resources of the N groups of reference signals, transmission periods of the N groups of reference signals, and group identifiers of the N groups of reference signals.
  • the transceiver unit 110 is used to receive a second reference signal set, wherein the second reference signal set includes N groups of reference signals, each group of reference signals in the N groups of reference signals includes at least one reference signal, and N is an integer greater than 1; the transceiver unit 110 is also used to receive second beam indication information, the second beam indication information indicates the beam corresponding to the first reference signal set, wherein the beam corresponding to the first reference signal set is a subset of the multiple beams corresponding to the second reference signal set, and the beam corresponding to the first reference signal set is used to determine the first input information of the AI model in the training device, the first input information is based on the measurement result of the beam corresponding to the first reference signal set, the first reference signal set includes M groups of reference signals, N is an integer greater than M, and M is an integer greater than or equal to 1.
  • the transceiver unit 110 is further configured to send first configuration information.
  • the transceiver unit 110 is further configured to send third configuration information.
  • the processing unit 120 is further configured to measure the N groups of reference signals to obtain N measurement results.
  • the transceiver unit 110 is used to receive the first Two reference signal sets, wherein the second reference signal set includes N groups of reference signals, each group of reference signals in the N groups of reference signals includes at least one reference signal, and N is an integer greater than 1; the transceiver unit 110 is also used to receive second beam indication information, the second beam indication information indicates the beam corresponding to the first reference signal set, wherein the beam corresponding to the first reference signal set is a subset of the multiple beams corresponding to the second reference signal set, the beam corresponding to the first reference signal set is used to determine the first input information of the AI model in the training device, the first input information is based on the measurement result of the beam corresponding to the first reference signal set, the first reference signal set includes M groups of reference signals, N is an integer greater than M, and M is an integer greater than or equal to 1.
  • the transceiver unit 110 is configured to send first configuration information.
  • the transceiver unit 110 is configured to send third configuration information.
  • the transceiver unit 110 is used to receive a first reference signal set, wherein the first reference signal set includes M groups of reference signals, each group of reference signals in the M groups of reference signals includes at least one reference signal, and M is an integer greater than or equal to 1; the transceiver unit 110 is also used to receive first beam indication information, and the first beam indication information indicates the beam corresponding to the first reference signal set, wherein the first reference signal set is used to determine the first input information of the AI model, the first input information is based on the measurement results of the M groups of reference signals included in the first reference signal set, the beam corresponding to the first reference signal set is a subset of the multiple beams corresponding to the second reference signal set, and the second reference signal set includes N groups of reference signals, and N is an integer greater than or equal to M.
  • the transceiver unit 110 is configured to receive first configuration information.
  • the transceiver unit 110 is configured to receive second configuration information.
  • the transceiver unit 110 is configured to receive third configuration information.
  • the processing unit 120 is used to obtain first output information based on the first input information, and the transceiver unit 110 is used to send the first output information.
  • the transceiver unit 110 is used to send a first reference signal set, wherein the first reference signal set includes M groups of reference signals, each group of reference signals in the M groups of reference signals includes at least one reference signal, and M is an integer greater than or equal to 1; the transceiver unit 110 is used to send first beam indication information, and the first beam indication information indicates the beam corresponding to the first reference signal set, wherein the first reference signal set is used to determine the first input information of the AI model, the first input information is based on the measurement results of the M groups of reference signals included in the first reference signal set, the beam corresponding to the first reference signal set is a subset of the multiple beams corresponding to the second reference signal set, and the second reference signal set includes N groups of reference signals, and N is an integer greater than or equal to M.
  • the transceiver unit 110 is configured to send first configuration information.
  • the transceiver unit 110 is configured to send the second configuration information.
  • the transceiver unit 110 is configured to send third configuration information.
  • the transceiver unit 110 is configured to receive first output information.
  • processing unit 110 and the transceiver unit 120 can be directly obtained by referring to the relevant descriptions in the method embodiments shown in FIG. 5 to FIG. 9 , and will not be repeated here.
  • the communication device 200 includes a processor 210 and an interface circuit 220.
  • the processor 210 and the interface circuit 220 are coupled to each other.
  • the interface circuit 220 may be a transceiver or an input/output interface.
  • the communication device 200 may further include a memory 230 for storing instructions executed by the processor 210 or storing input data required by the processor 210 to execute instructions or storing data generated after the processor 210 executes instructions.
  • the processor 210 is used to implement the function of the processing unit 120
  • the interface circuit 220 is used to implement the function of the transceiver unit 110 .
  • the processor 210 is used to implement the function of the processing unit 120
  • the interface circuit 220 is used to implement the function of the transceiver unit 110 .
  • the processor 210 is used to implement the function of the processing unit 120
  • the interface circuit 220 is used to implement the function of the transceiver unit 110 .
  • the processor 210 is used to implement the function of the processing unit 120
  • the interface circuit 220 is used to implement the function of the transceiver unit 110 .
  • the processor 210 is used to implement the function of the processing unit 120
  • the interface circuit 220 is used to implement the function of the transceiver unit 110 .
  • processor shown in FIG. 11 may include at least one processor, and the interface circuit may also include multiple interface circuits.
  • the chip of the training device implements the functions of the training device (or terminal device) in the above method embodiment.
  • the chip of the training device (or terminal device) receives information from other modules (such as a radio frequency module or an antenna) in the training device (or terminal device), and the information is sent by the network device to the training device (or terminal device); or, the chip of the training device (or terminal device) sends information to other modules (such as a radio frequency module or an antenna) in the training device (or terminal device), and the information is sent by the training device (or terminal device) to the network device.
  • modules such as a radio frequency module or an antenna
  • the processor in the embodiments of the present application may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof.
  • the general-purpose processor may be a microprocessor or any conventional processor.
  • the present application also provides a computer program product, on which computer program code is stored.
  • the computer program code runs on a computer, the computer executes the methods performed by the training device (or terminal device) or network device in the embodiments of methods 500 to 900.
  • the present application also provides a computer-readable medium, which stores a program code.
  • the program code runs on a computer, the computer executes the methods 500 to 900 performed by the training device (or terminal device) or the network device.
  • the present application also provides a communication system, which includes a training device and a first network device.
  • the training device is used to perform the steps corresponding to the training device in the above method 500
  • the first network device is used to perform the steps corresponding to the first network device in the above method 500.
  • the present application also provides a communication system, which includes a training device and a first network device.
  • the training device is used to perform the steps corresponding to the training device in the above method 600
  • the network device is used to perform the steps corresponding to the first network device in the above method 600.
  • the present application also provides a communication system, which includes a training device and a first network device.
  • the training device is used to perform the steps corresponding to the training device in the above method 700
  • the network device is used to perform the steps corresponding to the first network device in the above method 700.
  • the present application also provides a communication system, which includes a training device and a network device.
  • the training device is used to perform the steps corresponding to the training device in the above method 800
  • the network device is used to perform the steps corresponding to the network device in the above method 800.
  • the present application also provides a communication system, which includes an inference device, such as a terminal device, and a network device.
  • the terminal device is used to execute the steps corresponding to the terminal device in the above method 900
  • the network device is used to execute the steps corresponding to the network device in the above method 900.
  • the method steps in the embodiments of the present application can be implemented in hardware or in software instructions that can be executed by a processor.
  • the software instructions can be composed of corresponding software modules, and the software modules can be stored in random access memory, flash memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, register, hard disk, mobile hard disk, CD-ROM or any other form of storage medium known in the art.
  • An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium.
  • the storage medium can also be a component of the processor.
  • the processor and the storage medium can be located in an ASIC.
  • the ASIC can be located in a base station or a terminal.
  • the processor and the storage medium can also be present in a base station or a terminal as discrete components.
  • all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof.
  • all or part of the embodiments may be implemented in the form of a computer program product.
  • the computer program product includes one or more computer programs or instructions.
  • the computer may be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user device or other programmable device.
  • the computer program or instructions may be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium.
  • the computer program or instructions may be downloaded from a website, The computer, server or data center transmits to another website, computer, server or data center via wired or wireless means.
  • the computer-readable storage medium can be any available medium that can be accessed by the computer or a data storage device such as a server or data center that integrates one or more available media.
  • the available medium can be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; it can also be an optical medium, such as a digital video disk; it can also be a semiconductor medium, such as a solid-state hard disk.
  • the computer-readable storage medium can be a volatile or non-volatile storage medium, or can include both volatile and non-volatile types of storage media.

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Small-Scale Networks (AREA)

Abstract

本申请提供了一种获取训练数据集的方法和装置，训练设备可以向网络请求发送训练数据集，并且该请求信息中也指示了训练设备需要的网络设备向其发送的第一训练数据集的相关信息。换句话说，本申请中，训练设备可以向网络设备指示需要哪些训练数据集，网络设备可以向训练设备发送训练设备指示的训练数据集，不需要一直下发训练数据集。该方法可以减少空口资源浪费和空口开销，提高了空口资源的使用性能。

Description

一种获取训练数据集的方法和装置
本申请要求于2022年9月30日提交中国专利局、申请号为202211214685.0、申请名称为“一种获取训练数据集的方法和装置”，以及，要求于2022年10月12日提交中国专利局、申请号为202211247927.6、申请名称为“一种获取训练数据集的方法和装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请实施例涉及通信领域，并且更具体的，涉及一种获取训练数据集的方法和装置。
背景技术
目前，人工智能(artificial intelligent，AI)模型可以部署在训练设备(例如，终端设备)上进行训练、更新。训练设备在训练AI模型时，网络设备为了支撑训练设备对该AI模型的训练，会持续(例如，周期性地)向训练设备传输训练数据集，直到训练设备向网络设备发送完成训练模型的指示信息后，网络设备才会停止向训练设备发送训练数据集。然而，训练设备在训练该AI模型的过程中，网络设备向训练设备传输的训练数据集对训练设备而言是不必要的，会造成资源浪费。因此，在训练设备训练AI模型时，如何获取训练数据集，成为需要解决的技术问题。
发明内容
本申请实施例提供一种获取训练数据集的方法，可以减少空口资源浪费和空口开销，空口资源的使用性能得到提高。
第一方面，提供了一种获取训练数据集的方法，该方法可以由训练设备执行，或者，也可以由训练设备的组成部件(例如芯片或者电路)执行，对此不作限定。例如，该训练设备可以为终端设备。
该方法包括：向网络设备发送第一信息，所述第一信息用于指示训练设备请求所述网络设备发送的第一训练数据集的相关信息；接收来自所述网络设备的所述第一训练数据集，所述第一训练数据集是基于所述第一信息所指示的所述相关信息的训练数据集，所述第一训练数据集用于人工智能AI模型的训练。
基于上述技术方案，本申请中，训练设备可以向网络请求发送训练数据集，并且该请求信息中也指示了训练设备需要的网络设备向其发送的第一训练数据集的相关信息。换句话说，本申请中，训练设备可以向网络设备指示需要哪些训练数据集，网络设备可以向训练设备发送训练设备指示的训练数据集，不需要一直下发训练数据集。该方法可以减少空口资源浪费和空口开销，提高了空口资源的使用性能。
在一种可能的实现方式中，所述相关信息包括以下至少一项：所述第一训练数据集的大小的信息、所述AI模型的输入的配置信息、用于所述AI模型的训练的参考信号的配置信息。
基于上述技术方案，本申请中，第一信息可以指示第一训练数据集的大小的信息、该AI模型的输入的配置信息、用于该AI模型的训练的参考信号的配置信息，从而可以显示或者隐式的指示出训练设备需要的第一训练数据集的大小，使得网络设备可以基于训练设备的指示下发第一训练数据集，提高空口资源的利用率。
在一种可能的实现方式中，所述第一训练数据集的大小的信息是所述训练设备基于完成所述AI模型的训练所需要的训练数据集的大小确定的。
例如，训练设备可以通过历史信息确定从AI模型的初始状态(例如，AI模型的初始状态为0)训练该AI模型到收敛状态时，总共需要的训练数据集的数量。例如，训练设备基于历史经验确定训练该AI模型共需要扫描6万次全码本获得训练数据集。
基于上述技术方案，本申请中，训练设备可以基于历史经验确定训练该AI模型需要的训练数据集 的数量并指示给网络设备，从而使得网络设备基于该指示下发训练数据集，可以减少空口资源的浪费。
在一种可能的实现方式中，所述向网络设备发送第一信息之前，所述方法还包括：确定所述AI模型的第一性能；根据所述AI模型的第一性能和所述AI模型的第二性能，确定所述第一训练数据集的大小的信息，其中，所述第一性能为所述AI模型的当前性能，所述第二性能为所述AI模型的目标性能。
例如，训练设备对该AI模型监控时，通过将监控到的当前AI模型的性能与历史信息与该AI模型收敛时对应的模型性能相比较，并且按照当前AI模型的性能可以估计出达到预期的模型性能时需要的第一训练数据集的大小。
基于上述技术方案，本申请中，训练设备可以基于AI模型的性能估计训练该AI模型到收敛状态时，需要的训练数据集的数量并指示给网络设备，从而使得网络设备基于该指示下发训练数据集，可以减少空口资源的浪费。
在一种可能的实现方式中，所述参考信号的配置信息包括以下至少一项：所述参考信号的标识、所述参考信号的时域资源、所述参考信号的频域资源、所述参考信号的传输周期、传输的所述参考信号的类型。
例如，参考信号的类型为SSB，或者CSI-RS，或者SRS，等等。其中，参考信号的标识也可以理解为参考信号组的标识，例如，参考信号的配置信息包含N(N为大于或者等于1的整数)组参考信号的组标识，其中，N组参考信号中的每组参考信号具有相同的组标识，每组参考信号包括至少一个参考信号。类似的，参考信号的时域资源、参考信号的频域资源、参考信号的传输周期、传输的该参考信号的类型分别也可以理解为，N组参考信号的时域资源、N组参考信号的频域资源、N组参考信号的传输周期、传输的该N组参考信号的类型。
基于上述技术方案，本申请中，由于训练数据集可以是参考信号的测量结果，因此训练设备可以基于历史信息确定参考信号的配置信息，通过该参考信号的配置信息也可以间接的指示需要的训练数据集的数量。
本申请中，“AI模型的输入的配置信息”例如可以理解为，训练设备基于该AI模型训练的历史信息，确定该AI模型的输入信息，例如，该AI模型的输入信息为稀疏波束图样对应的参考信号的测量结果。则训练设备可以确定该稀疏波束图样是全码本中的哪些位置的波束。例如，训练设备可以将该稀疏波束图样的标识，或者将该稀疏波束图样对应的参考信号的标识，或者将参考信号的测量结果等信息上报给第一网络设备，从而可以使得第一网络设备下发与该波束图样对应的训练数据集。
基于上述技术方案，本申请中，由于AI模型的输入信息可以是参考信号的测量结果，因此训练设备可以基于历史信息确定AI模型的输入的配置信息，通过该AI模型的输入的配置信息也可以间接的指示需要的训练数据集的数量。
在一种可能的实现方式中，所述第一信息包括以下至少一项：所述AI模型的标识信息、所述AI模型的应用场景的信息、所述AI模型的用途信息、所述训练设备的算力能力的信息。
本申请中，例如，网络设备上可以保存第一映射关系，第一映射关系为各个AI模型的标识以及该AI模型标识对应的训练数据集的大小之间的对应关系。训练设备通过向网络设备指示该AI模型的标识便可以指示出训练该AI模型需要的训练数据集的大小。
需要说明的是，本申请中的“映射关系”也可以表述为“关联关系”、“对应关系”。应理解，本申请实施例中所说的“映射关系”可以通过函数关系、或表格、或映射关系等方式保存或被记录。下述实施例中，提到的“映射关系”可以是网络设备配置的，也可以是协议预定义的，等等，不予限定。
AI模型的应用场景或者该AI模型的用途可以理解为，该AI模型是用于波束管理场景，或者CSI反馈场景，或者定位场景，等等。训练设备通过向网络设备指示该AI模型的应用场景或者AI模型的用途便可以指示出训练该AI模型需要的训练数据集的大小。
本申请中，训练设备还可以上报算力能力，例如，训练设备上报的算力能力的信息至少包括以下一项：训练设备的处理器(例如，中央处理器(center processing unit，CPU)、图形处理器(graphic processing unit，GPU)、张量处理器(tensor processing unit，TPU)、神经网络处理器 (neural network processing unit，NPU)、现场可编程门阵列(field-programmable gate array，FPGA)等等)的能力、训练设备存储空间的大小、训练设备内存的大小、训练设备的电量等等，不予限定。训练设备通过向网络设备上报自己的算力能力，便可以指示出训练该AI模型时最多可以处理的训练数据集的数量。
在一种可能的实现方式中，所述方法还包括：根据所述第一训练数据集训练所述AI模型，并确定所述AI模型的性能；根据所述AI模型的性能，向所述网络设备发送第二信息，所述第二信息用于指示所述训练设备请求所述网络设备发送的第二训练数据集的相关信息；接收来自所述网络设备的所述第二训练数据集，所述第二训练数据集是基于所述第二信息所指示的所述相关信息的训练数据集，所述第二训练数据集用于所述AI模型的训练。
例如，第二训练数据集的数据量可以小于第一训练数据集的数据量。后续，训练设备仍然可以基于第二训练数据集对该AI进行训练，并反复迭代，例如，基于第二训练数据集进行模型训练，再次确定该AI模型的性能，基于该AI模型的性能确定需要的第三训练数据集的大小，等等。假设训练设备可以进行L(L为大于1的整数)次训练，直到训练设备确定该AI模型收敛(“模型收敛”也可以理解为该AI模型达到目标性能)。
基于上述技术方案，本申请中，训练设备可以通过衡量AI模型训练的性能，确定下一次训练时需要的训练数据集的大小，可以提高AI模型训练的效率并且提高空口资源的使用性能。
第二方面，提供了一种获取训练数据集的方法，该方法可以由第一网络设备执行，或者，也可以由第一网络设备的组成部件(例如芯片或者电路)执行，对此不作限定。
其中网络侧技术方案对应的相同的有益效果可以参照训练设备侧的有益效果的描述，此处不再赘述。
该方法包括：接收来自训练设备的第一信息，所述第一信息用于指示请求所述第一网络设备发送的第一训练数据集的相关信息；根据所述第一信息所指示的所述相关信息，向所述训练设备发送所述第一训练数据集，所述第一训练数据集用于所述人工智能AI模型的训练。
在一种可能的实现方式中，所述相关信息包括以下至少一项：所述第一训练数据集的大小的信息、所述AI模型的输入的配置信息、用于所述AI模型的训练的参考信号的配置信息。
在一种可能的实现方式中，所述第一训练数据集的大小的信息是基于完成所述AI模型的训练所需要的训练数据集的大小确定的。
在一种可能的实现方式中，所述参考信号的配置信息包括以下至少一项：所述参考信号的标识、所述参考信号的时域资源、所述参考信号的频域资源、所述参考信号的传输周期、传输的所述参考信号的类型。
在一种可能的实现方式中，所述第一信息包括以下至少一项：所述AI模型的标识信息、所述AI模型的应用场景的信息、所述AI模型的用途信息、所述训练设备的算力能力的信息。
在一种可能的实现方式中，所述方法还包括：从第二网络设备获取第三信息，所述第三信息为所述训练所述AI模型的相关信息，其中，所述第一网络设备为训练设备从第二网络设备切换至的目标网络设备；所述根据所述第一信息所指示的所述相关信息，向所述训练设备发送所述第一训练数据集，包括：根据所述第一信息所指示的所述相关信息和所述第三信息，向所述训练设备发送所述第一训练数据集。
基于上述技术方案，本申请中，第一网络设备可以结合多方面的信息综合决定是否支撑该AI模型的训练，网络设备不会持续向训练设备下发训练数据集，可以减少不必要的空口资源的占用，节省空口开销，提高空口资源的使用性能。
在一种可能的实现方式中，所述第三信息包括以下至少一项：所述训练设备请求所述第二网络设备发送的训练数据集的大小的信息、完成所述AI模型训练需要的训练数据集的大小的信息、所述AI模型的标识信息、所述训练设备的算力能力的信息。
例如，所述第一网络设备保存有第一映射关系，所述第一映射关系为AI模型的标识与所述AI模型的标识对应的训练数据集的大小之间的映射关系。
例如，“训练设备请求第二网络设备发送的训练数据集的大小的信息”可以理解为，第二网络设备与训练设备连接时，训练设备向第二网络设备请求需要的训练数据集的大小的信息。例如，训练设备 也可以基于历史信息确定向第二网络设备请求的训练数据集的大小。例如，“完成AI模型训练需要的训练数据集的大小的信息”可以理解为，第二网络设备上保存有训练设备训练该AI模型时总共需要的训练数据集的大小。
基于上述技术方案，本申请中，第一网络设备可以从第二网络获取该训练该AI模型的信息，使得第一网络设备可以基于训练设备指示信息和从第二网络设备同步的信息共同确定能够给训练设备下发的训练数据的大小，提高空口资源的使用性能。
在一种可能的实现方式中，所述第三信息还包括：发送训练数据集的时长的信息和/或发送训练数据集的方式的信息，所述方法还包括：根据所述第三信息和所述第一网络设备中的资源使用情况，确定所述第一网络设备和/或所述训练设备是否具备支持训练所述AI模型的能力。
“发送训练数据集的方式的信息”，也可以理解为，例如，第二网络设备是周期性地向该训练设备发送训练数据集。示例性的，网络设备可以识别到一天当中哪个时间段数据请求量是最少的，也可以理解为，哪个时间段空口资源占用最少，或者哪个时间段空口资源充足。此时就可以对AI模型提供训练数据集。例如，网络设备可以每天都选择该时间段支撑AI模型的更新(此处，不限制训练设备的数量和/或AI模型的数量)。又例如，第二网络设备是间隔向该训练设备发送训练数据集。该方案相对周期发灵活度有一些提升，如果网络设备发现和/或判断当前数据请求量较少，空口资源充足时，那么此时网络设备就确定可以支撑AI模型的更新。换句话说，该实现方式中，网络设备可以基于当前时刻空口资源的占用情况，确定向训练设备发送训练数据集，所以其不具有明显周期性的特征。
本申请中，“资源使用情况”也可以理解为“资源占用情况”、“空口资源占用”等等。例如，现有协议框架中已经定义了网络设备最多可以配置的参考信号(例如，64个CSI-RS)的数量，如果网络设备发现全部参考信号都已经被配置给了其他功能，那么就可以确定当前资源被占用，无法给训练设备配置参考信号资源以支撑该AI模型的更新。
基于上述技术方案，本申请中，第一网络设备可以从第二网络获取该训练该AI模型的信息，使得第一网络设备可以基于训练设备指示信息、从第二网络设备同步的信息以及空口资源占用情况共同确定能够给训练设备下发的训练数据的大小，提高空口资源的利用率。
第三方面，提供了一种获取训练数据集的方法，该方法可以由第一网络设备执行，或者，也可以由第一网络设备的组成部件(例如芯片或者电路)执行，对此不作限定。
该方法包括：第一网络设备从第二网络设备获取第三信息，所述第三信息为训练人工智能AI模型的相关信息，其中，所述第一网络设备为训练设备从第二网络设备切换至的目标网络设备；所述第一网络设备接收来自所述训练设备的第一信息，所述第一信息用于请求所述第一网络设备发送训练数据集；所述第一网络设备根据所述第三信息，确定待发送的第一训练数据集；所述第一网络设备基于所述第一信息向所述训练设备发送所述第一训练数据集，所述第一训练数据集用于所述AI模型的训练。
基于上述技术方案，本申请中，第一网络设备可以从第二网络获取该训练该AI模型的信息，使得第一网络设备可以基于训练设备指示信息和从第二网络设备同步的信息共同确定能够给训练设备下发的训练数据的大小，提高空口资源的使用性能。
在一种可能的实现方式中，所述第三信息包括以下至少一项：所述训练设备请求所述第二网络设备发送的训练数据集的大小的信息、完成所述AI模型训练需要的训练数据集的大小的信息、所述AI模型的标识信息、所述训练设备的算力能力的信息。
在一种可能的实现方式中，所述第一网络设备保存有第一映射关系，所述第一映射关系为AI模型的标识与所述AI模型的标识对应的训练数据集的大小之间的映射关系。
在一种可能的实现方式中，所述第三信息还包括：发送训练数据集的时长的信息和/或发送训练数据集的方式的信息，所述方法还包括：所述第一网络设备根据所述第三信息和所述第一网络设备中的资源使用情况，确定所述第一网络设备和/或所述训练设备是否具备支持训练所述AI模型的能力。
在一种可能的实现方式中，所述第一网络设备根据所述第三信息，确定待发送的第一训练数据集，包括：所述第一网络设备根据所述第一信息和所述第三信息，确定待发送的第一训练数据集，其中，所述第一信息用于指示请求所述第一网络设备发送的第一训练数据集的相关信息。
在一种可能的实现方式中，所述相关信息包括以下至少一项：所述第一训练数据集的大小的信息、 所述AI模型的输入信息、用于所述AI模型的训练的参考信号的配置信息。
在一种可能的实现方式中，所述第一训练数据集的大小的信息是基于完成所述AI模型的训练所需要的训练数据集的大小确定的。
在一种可能的实现方式中，所述参考信号的配置信息包括以下至少一项：所述参考信号的标识、所述参考信号的时域资源、所述参考信号的频域资源、所述参考信号的传输周期、传输的所述参考信号的类型。
在一种可能的实现方式中，所述第一信息包括以下至少一项：所述AI模型的标识信息、所述AI模型的应用场景的信息、所述AI模型的用途信息、所述训练设备的算力能力的信息。
在一种可能的实现方式中，所述方法还包括：接收来自所述训练设备的第二信息，所述第二信息用于指示请求所述网络设备发送的第二训练数据集的相关信息，其中，所述第二信息是基于所述AI模型的性能确定的，所述AI模型的性能是基于所述第一训练数据集训练确定的；根所述第二信息，确定待发送的第二训练数据集。
第四方面，提供一种通信方法，该方法可以由训练设备执行，或者，也可以由训练设备的组成部件(例如芯片或者电路)执行，对此不作限定。
该方法包括：测量N组参考信号，获取所述N组参考信号对应的N组测量结果，其中，所述N组参考信号中的每组参考信号包括至少一个参考信号，所述每组参考信号具有相同的组标识，所述N为大于1的整数；接收来自网络设备的第四信息，所述第四信息用于指示所述N组参考信号中的M组参考信号；根据所述第四信息和所述N组参考信号对应的N组测量结果，确定人工智能AI模型的第一输入信息，所述第一输入信息包括所述M组参考信号对应的M组测量结果；所述AI模型用于基于所述第一输入信息，获取第一输出信息，其中，所述第一输出信息包括所述N组参考信号中的K组参考信号各自的组标识，其中，所述K组参考信号各自的组标识对应于所述N组测量结果中信道质量最好的K组测量结果。其中，每组测量结果可以包括一个或多个测量结果。
基于上述技术方案，本申请中，在后续多次的训练过程中，稀疏波束的图样仍然可以为第四信息指示的波束图样。每次训练时，训练设备可以基于第一网络设备下发的训练数据集进行一次全码本扫描，由于信道状态(也可以理解为，信道环境)是时变的，每次全码本扫描后获得的参考信号的测量结果也是不完全相同的。因此，每次训练时，N组参考信号中的M组参考信号对应测量结果也是不同，并且训练设备确定的训练标签也不相同，即AI模型的输入信息和训练标签都会相应发生变化，然而这些变化本质上都是由于信道状态变化引起的，波束图样并未变化。即，本申请提供的方案中，AI模型训练过程中的变量仅仅为信道状态。与另一方案中，AI模型训练过程中波束图样和信道状态都发生变化相比，本申请提供的方案可以加速AI模型的收敛速度，提高模型训练效率，从而也可以减少空口资源的占用。
本申请中，一组参考信号的组标识可以对应一个波束标识，N组参考信号的N个组标识对应N个波束标识。
在一种可能的实现方式中，所述第一输出信息还包括所述N组测量结果中剩余的(N-K)组参考信号各自的组标识，所述(N-K)组参考信号各自的组标识对应于(N-K)组测量结果。
本申请中，AI模型训练时可以采用分类法和回归法，不同的训练方法分别对应不同的AI模型的输入信息和输出信息。例如，分类法中AI模型的输入信息为参考信号的测量结果，输出信息为AI模型预测的全码本中信道质量最好的K个波束标识。又例如，回归法中AI模型的输入信息为参考信号的测量结果(例如，参考信号的RSRP、RSRQ、SINR)，输出信息为AI模型预测的全码本中所有参考信号的测量结果。
在一种可能的实现方式中，所述第四信息包括N个字段，所述N个字段与所述N组参考信号一一对应，所述N个字段中的M个字段的比特值与剩余的(N-M)个字段的比特值不同；所述第四信息用于指示N组参考信号中的M组参考信号，具体包括：所述第四信息中的所述M个字段用于指示所述M组参考信号。
基于上述技术方案，本申请中，网络设备可以通过指示第四信息中每个字段的比特值，向训练设备指示稀疏波束图样。即，训练设备可以通过解析第四信息获取AI模型的输入信息，从而可以加速AI模型的收敛速度，提高AI模型的训练效率。
在一种可能的实现方式中，所述方法还包括：接收来自网络设备的第五信息，所述第五信息用于指示所述N组参考信号中的P组参考信号；其中，所述第五信息包括N个字段，所述N个字段与所述N组参考信号一一对应，所述N个字段中的P个字段的比特值与剩余(N-P)个字段的比特值不同；所述第五信息用于指示N组参考信号中的P组参考信号，具体包括：所述第五信息中的所述P个字段用于指示所述P组参考信号。
基于上述技术方案，本申请中，网络设备可以通过指示第四信息、第五信息中每个字段的比特值，向训练设备指示多个稀疏波束图样。即，训练设备可以通过解析第四信息、第五信息获取AI模型的多个输入信息，从而可以加速AI模型的收敛速度，提高AI模型的训练效率。
在一种可能的实现方式中，所述方法还包括：接收来自网络设备的配置信息，所述配置信息用于指示以下中的一项或多项：所述N组参考信号的时域资源、所述N组参考信号的频域资源、所述N组参考信号的传输周期、所述N组参考信号的组标识。
例如，配置信息与第四信息和/或第五信息也可以在同一条消息中进行发送，不予限定。
基于上述技术方案，本申请中，网络设备可以向训练设备发送N组参考信号的配置信息，使得训练设备可以通过测量该N组参考信号获取训练数据集，进行AI模型的训练。
第五方面，提供一种通信方法，该方法可以由网络设备执行，或者，也可以由网络设备的组成部件(例如芯片或者电路)执行，对此不作限定。
该方法包括：向训练设备发送N组参考信号，其中，所述N组参考信号中的每组参考信号包括至少一个参考信号，所述每组参考信号具有相同的组标识，所述N为大于1的整数；向所述训练设备发送第四信息，所述第四信息用于指示所述N组参考信号中的M组参考信号，其中，所述M组参考信号用于确定第一输入信息；所述AI模型用于基于所述第一输入信息，获取第一输出信息，其中，所述第一输出信息包括所述N组参考信号中的K组参考信号各自的组标识，其中，所述K组参考信号各自的组标识对应于所述N组参量信号对应的N组测量结果中信道质量最好的K组测量结果。其中，每组测量结果可以包括一个或多个测量结果。
在一种可能的实现方式中，所述第一输出信息还包括所述N组测量结果中剩余的(N-K)组参考信号各自的组标识，所述(N-K)组参考信号各自的组标识对应于(N-K)组测量结果。
在一种可能的实现方式中，所述第四信息包括N个字段，所述N个字段与所述N组参考信号一一对应，所述N个字段中的M个字段的比特值与剩余的(N-M)个字段的比特值不同；所述第四信息用于指示N组参考信号中的M组参考信号，具体包括：所述第四信息中的所述M个字段用于指示所述M组参考信号。
在一种可能的实现方式中，所述方法还包括：向训练设备发送第五信息，所述第五信息用于指示所述N组参考信号中的P组参考信号；其中，所述第五信息包括N个字段，所述N个字段与所述N组参考信号一一对应，所述N个字段中的P个字段的比特值与剩余(N-P)个字段的比特值不同；所述第四信息用于指示N组参考信号中的P组参考信号，具体包括：所述第四信息中的所述P个字段用于指示所述P组参考信号。
在一种可能的实现方式中，所述方法还包括：向所述训练设备发送配置信息，所述配置信息用于指示以下中的一项或多项：所述N组参考信号的时域资源、所述N组参考信号的频域资源、所述N组参考信号的传输周期、所述N组参考信号的组标识。
第六方面，提供了一种通信方法，该方法可以由训练设备执行，或者，也可以由训练设备的组成部件(例如芯片或者电路)执行，对此不作限定。
该方法包括：接收第二参考信号集，其中，所述第二参考信号集包括N组参考信号，所述N组参考信号中的每组参考信号包括至少一个参考信号，所述N为大于1的整数；接收第二波束指示信息，所述第二波束指示信息指示第一参考信号集对应的波束，其中，所述第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，所述第一参考信号集对应的波束用于所述训练设备中的AI模型的第一输入信息的确定，所述第一输入信息基于所述第一参考信号集对应的波束的测量结果，所述第一参考信号集包括M组参考信号，所述N为大于M的整数，所述M为大于或等于1的整数；其中，所述AI模型用于基于所述第一输入信息，获得第一输出信息，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，所述K为大于或等于1的整数且所述K小 于所述N，所述AI模型的标签为所述第二参考信号集的测量结果中信道质量最好的K个波束。
基于上述技术方案，本申请中，训练设备可以通过接收的第二波束指示信息确定稀疏波束图样。也可以理解为，训练设备可以基于第二波束指示信息确定出稀疏波束图样是由全码本中的哪些波束组成的，并且还可以确定该AI模型的输入信息。此时，由于在训练过程中稀疏波束图样不发生变化，只有信道状态发生变化，因此本方案可以加速AI模型的收敛，提高AI模型的训练效率。
在一种可能的实现方式中，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，包括如下中的至少一项：所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束的信息；或者，K组参考信号各自的组标识，其中，所述K组参考信号对应于所述N组参考信号对应的N个测量结果中被预测为信道质量最好的K个测量结果，且所述K组参考信号各自的组标识与K个波束具有预先定义或预先配置的对应关系；或者，所述N组参考信号对应的多个波束信息及所述波束信息对应的N个测量结果；或者，所述N组参考信号各自的组标识及所述N组参考信号的N个测量结果，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
基于上述技术方案，本申请中，基于AI模型的算法的实现，AI模型的输出信息可以略有不同。例如，分类法中AI模型的输出信息为第二参考信号集对应的多个波束中被预测为信道质量最好的K(K为大于0的整数)个波束的信息。又例如，回归法中AI模型的输出信息为N组参考信号对应的N个测量结果，等等。
在一种可能的实现方式中，所述第二波束指示信息指示所述第一参考信号集对应的波束包括：所述第二波束指示信息指示所述第一参考信号集对应的波束在所述第二参考信号集对应的多个波束中的位置。
基于上述技术方案，本申请中，第二波束指示信息可以指示第一参考信号集对应的波束在所述第二参考信号集对应的多个波束中的位置，从而指示稀疏波束图样。也可以理解为，训练设备可以基于第二波束指示信息确定出应该第一参考信号集中的参考信号组是全码本中哪些波束上对应的参考信号，从而可以确定AI模型的输入信息。
在一种可能的实现方式中，所述第二波束指示信息包括N个字段，所述N个字段与所述第二参考信号集对应的多个波束一一对应，所述N个字段中的M个字段的比特值与剩余(N-M)个字段的比特值不同；所述第二波束指示信息指示所述第一参考信号集对应的波束，包括：所述第二波束指示信息中的所述M个字段对应所述第一参考信号集。
基于上述技术方案，本申请中，可以通过N个字段中的M个字段直接指示稀疏波束图样，或者也可以理解为，通过N个字段中的M个字段直接指示出全码本中的哪些波束组成了该稀疏波束图样，从而使得训练设备可以确定AI模型的输入信息。
在一种可能的实现方式中，所述方法还包括：向所述训练设备发送第一配置信息，所述第一配置信息指示以下中的一项或多项：所述N组参考信号的时域资源、所述N组参考信号的频域资源、所述N组参考信号的传输周期、所述N组参考信号的组标识、或所述N组参考信号的波束信息。
基于上述技术方案，本申请中，可以通过第一配置信息指示出训练设备应该如何接收N组参考信号。例如，训练设备应该在哪些时频资源上接收N组参考信号。
在一种可能的实现方式中，所述第二波束指示信息指示所述M组参考信号对应的波束，包括：所述第二波束指示信息包括所述M组参考信号的组标识或波束信息，所述M组参考信号为所述N组参考信号中的部分，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
基于上述技术方案，由于N组参考信号和N个波束之间具有一一对应关系，因此，可以通过指示M组参考信号的组标识或波束信息便可以指示出稀疏波束图样。
在一种可能的实现方式中，向所述训练设备发送所述N组参考信号的第三配置信息；所述第二波束指示信息包括所述M组参考信号的组标识的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的组标识并指示以下中的一项或多项：所述N组参考信号的时域资源，所述N组参考信号的频域资源，所述N组参考信号的传输周期，或，所述N组参考信号的波束；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识；或者，所述第二波束指示信息包括所述M组参考信号的波束信息的情况下，所述N组参 考信号的第三配置信息包括所述N组参考信号各自的波束信息并指示以下中的一项或多项：所述N组参考信号的组标识，时域资源，所述N组参考信号的频域资源，或，所述N组参考信号的传输周期；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识。
基于上述技术方案，本申请中，第三配置信息中可以包括N组参考信号的标识，并且M组参考信号属于该N组参考信号，因此，第二波束指示信息中可以包括M组参考信号的标识或者M组参考信号的波束信息，从而指示出稀疏波束图样是由全码本中的哪些波束组成的，从而使得训练设备可以确定AI模型的输入信息。
在一种可能的实现方式中，所述方法还包括：对所述N组参考信号进行测量，获得N个测量结果，所述N个测量结果对应于N个波束且所述N个测量结果包括所述第一参考信号集对应的波束的测量结果。
基于上述技术方案，本申请中，训练设备可以通过测量N组参考信号获得N组测量结果，并且基于第二波束信息确定AI模型的是输入信息。
第七方面，提供了一种通信方法，该方法可以由网络设备执行，或者，也可以由网络设备的组成部件(例如芯片或者电路)执行，对此不作限定。
该方法包括：向训练设备发送第二参考信号集，其中，所述第二参考信号集包括N组参考信号，所述N组参考信号中的每组参考信号包括至少一个参考信号，所述N为大于1的整数；向所述训练设备发送第二波束指示信息，所述第二波束指示信息指示第一参考信号集对应的波束，其中，所述第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，所述第一参考信号集对应的波束用于所述训练设备中的AI模型的第一输入信息的确定，所述第一输入信息基于所述第一参考信号集对应的波束的测量结果，所述第一参考信号集包括M组参考信号，所述N为大于M的整数，所述M为大于或等于1的整数；其中，所述AI模型用于基于所述第一输入信息，获得第一输出信息，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，所述K为大于或等于1的整数且所述K小于所述N，所述AI模型的标签为所述第二参考信号集的测量结果中信道质量最好的K个波束。
在一种可能的实现方式中，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，包括如下中的至少一项：所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束的信息；或者，K组参考信号各自的组标识，其中，所述K组参考信号对应于所述N组参考信号对应的N个测量结果中被预测为信道质量最好的K个测量结果，且所述K组参考信号各自的组标识与K个波束具有预先定义或预先配置的对应关系；或者，所述N组参考信号对应的多个波束信息及所述波束信息对应的N个测量结果；或者，所述N组参考信号各自的组标识及所述N组参考信号的N个测量结果，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
在一种可能的实现方式中，所述第二波束指示信息指示所述第一参考信号集对应的波束包括：所述第二波束指示信息指示所述第一参考信号集对应的波束在所述第二参考信号集对应的多个波束中的位置。
在一种可能的实现方式中，所述第二波束指示信息包括N个字段，所述N个字段与所述第二参考信号集对应的多个波束一一对应，所述N个字段中的M个字段的比特值与剩余(N-M)个字段的比特值不同；所述第二波束指示信息指示所述第一参考信号集对应的波束，包括：所述第二波束指示信息中的所述M个字段对应所述第一参考信号集。
在一种可能的实现方式中，所述方法还包括：向所述训练设备发送第一配置信息，所述第一配置信息指示以下中的一项或多项：所述N组参考信号的时域资源、所述N组参考信号的频域资源、所述N组参考信号的传输周期、所述N组参考信号的组标识、或所述N组参考信号的波束信息。
在一种可能的实现方式中，所述第二波束指示信息指示所述M组参考信号对应的波束，包括：所述第二波束指示信息包括所述M组参考信号的组标识或波束信息，所述M组参考信号为所述N组参考信号中的部分，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
在一种可能的实现方式中，所述方法还包括：向所述训练设备发送所述N组参考信号的第三配置 信息；所述第二波束指示信息包括所述M组参考信号的组标识的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的组标识并指示以下中的一项或多项：所述N组参考信号的时域资源，所述N组参考信号的频域资源，所述N组参考信号的传输周期，或，所述N组参考信号的波束；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识；或者，所述第二波束指示信息包括所述M组参考信号的波束信息的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的波束信息并指示以下中的一项或多项：所述N组参考信号的组标识，时域资源，所述N组参考信号的频域资源，或，所述N组参考信号的传输周期；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识。
第八方面，提供了一种通信方法，该方法可以由终端设备执行，或者，也可以由终端设备的组成部件(例如芯片或者电路)执行，对此不作限定。其中，该终端设备可以作为推理设备。
该方法包括：接收第一参考信号集，其中，所述第一参考信号集包括M组参考信号，所述M组参考信号中的每组参考信号包括至少一个参考信号，所述M为大于或等于1的整数；接收第一波束指示信息，所述第一波束指示信息指示所述第一参考信号集对应的波束，其中，所述第一参考信号集用于所述AI模型的第一输入信息的确定，所述第一输入信息基于所述第一参考信号集所包括的所述M组参考信号的测量结果，所述第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，所述第二参考信号集包括N组参考信号，所述N为大于或等于M的整数；其中，所述AI模型用于基于所述第一输入信息，获得第一输出信息，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，所述K为大于或等于1的整数且所述K小于所述N。
基于上述技术方案，本申请中，在模型推理阶段，网络设备也可以向终端设备指示模型的输入信息，使得终端设备确定出模型的输入信息，从而可以提高终端设备在模型推理时输出信息的准确性。
在一种可能的实现方式中，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，包括如下中的至少一项：所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束的信息；或者，K组参考信号各自的组标识，其中，所述K组参考信号对应于所述N组参考信号对应的N个测量结果中被预测为信道质量最好的K个测量结果，且所述K组参考信号各自的组标识与K个波束具有预先定义或预先配置的对应关系；或者，所述N组参考信号对应的多个波束信息及所述波束信息对应的N个测量结果；或者，所述N组参考信号各自的组标识及所述N组参考信号的N个测量结果，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
基于上述技术方案，本申请中，基于AI模型的算法的实现，AI模型的推理输出信息可以略有不同。例如，分类法中AI模型的推理输出信息为第二参考信号集对应的多个波束中被预测为信道质量最好的K(K为大于0的整数)个波束的信息。又例如，回归法中AI模型的推理输出信息为N组参考信号对应的N个测量结果，等等。
在一种可能的实现方式中，所述第一波束指示信息指示所述第一参考信号集对应的波束包括：所述第一波束指示信息指示所述第一参考信号集对应的波束在所述第二参考信号集对应的多个波束中的位置。
基于上述技术方案，本申请中，第二波束指示信息可以指示第一参考信号集对应的波束在所述第二参考信号集对应的多个波束中的位置，从而指示稀疏波束图样。也可以理解为，终端设备可以基于第二波束指示信息确定出应该第一参考信号集中的参考信号组是全码本中哪些波束上对应的参考信号，从而可以确定AI模型的输入信息。
在一种可能的实现方式中，所述第一波束指示信息包括N个字段，所述N个字段与所述第二参考信号集对应的多个波束一一对应，所述N个字段中的M个字段的比特值与剩余(N-M)个字段的比特值不同；所述第一波束指示信息指示所述第一参考信号集对应的波束，包括：所述第一波束指示信息中的所述M个字段对应所述第一参考信号集。
基于上述技术方案，本申请中，可以通过N个字段中的M个字段直接指示稀疏波束图样，或者也可以理解为，通过N个字段中的M个字段直接指示出全码本中的哪些波束组成了该稀疏波束图样，从而使得终端设备可以确定AI模型的输入信息。
在一种可能的实现方式中，所述方法还包括：接收第一配置信息，所述第一配置信息指示以下中的一项或多项：所述M组参考信号的时域资源、所述M组参考信号的频域资源、所述M组参考信号的传输周期、所述M组参考信号的组标识、或所述M组参考信号的波束信息。
基于上述技术方案，本申请中，可以通过第一配置信息指示出终端设备应该如何接收M组参考信号。例如，终端设备应该在哪些时频资源上接收M组参考信号。
在一种可能的实现方式中，所述第一波束指示信息指示所述M组参考信号对应的波束，包括：所述第一波束指示信息包括所述M组参考信号的组标识或波束信息，所述M组参考信号为所述N组参考信号中的部分，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
基于上述技术方案，由于N组参考信号和N个波束之间具有一一对应关系，因此，可以通过指示M组参考信号的组标识或波束信息便可以指示出稀疏波束图样。
在一种可能的实现方式中，所述第一波束指示信息包括在所述M组参考信号的第二配置信息中，所述第一波束指示信息包括所述M组参考信号的组标识的情况下，所述第二配置信息还包括所述M组参考信号的时域资源、频域资源、传输周期、或、波束信息中的一项或多项；或者，所述第一波束指示信息包括所述M组参考信号的波束信息的情况下，所述第二配置信息还包括所述M组参考信号的时域资源、频域资源、传输周期、或、组标识中的一项或多项。
基于上述技术方案，本申请中，第二配置信息中可以包括第一波束指示信息，并且第一波束指示信息包括M组参考信号的组标识或M组参考信号的波束信息，从而指示出稀疏波束图样是由全码本中的哪些波束组成的，从而使得终端设备可以确定AI模型的输入信息。
在一种可能的实现方式中，所述方法还包括：接收所述N组参考信号的第三配置信息；所述第一波束指示信息包括所述M组参考信号的组标识的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的组标识并指示以下中的一项或多项：所述N组参考信号的时域资源，所述N组参考信号的频域资源，所述N组参考信号的传输周期，或，所述N组参考信号的波束；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识；或者，所述第一波束指示信息包括所述M组参考信号的波束信息的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的波束信息并指示以下中的一项或多项：所述N组参考信号的组标识，时域资源，所述N组参考信号的频域资源，或，所述N组参考信号的传输周期；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识。
基于上述技术方案，本申请中，第三配置信息中可以包括N组参考信号的标识，并且M组参考信号属于该N组参考信号，因此，第二波束指示信息中可以包括M组参考信号的标识或者M组参考信号的波束信息，从而指示出稀疏波束图样是由全码本中的哪些波束组成的，从而使得终端设备可以确定AI模型的输入信息。
在一种可能的实现方式中，所述方法还包括：利用所述AI模型基于所述第一输入信息获得所述第一输出信息；发送所述第一输出信息。
基于上述技术方案，本申请中，终端设备通过AI模型推理获得推理输出信息后，还可以进一步向网络设备反馈该输出信息，使得网络设备可以基于该输出信息为终端设备发送对应的参考信号。终端设备再次测量参考信号，并且确定测量结果最优的参考信号，并将该参考信号对应的波束标识作为最终选择的波束，采用该波束与网络设备进行通信。
第九方面，提供了一种通信方法，该方法可以由网络设备执行，或者，也可以由网络设备的组成部件(例如芯片或者电路)执行，对此不作限定。
该方法包括：向终端设备发送第一参考信号集，其中，所述第一参考信号集包括M组参考信号，所述M组参考信号中的每组参考信号包括至少一个参考信号，所述M为大于或等于1的整数；向所述终端设备发送第一波束指示信息，所述第一波束指示信息指示所述第一参考信号集对应的波束，其中，所述第一参考信号集用于所述AI模型的第一输入信息的确定，所述第一输入信息基于所述第一参考信号集所包括的所述M组参考信号的测量结果，所述第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，所述第二参考信号集包括N组参考信号，所述N为大于或等于M的整数；其中，所述AI模型用于基于所述第一输入信息，获得第一输出信息，所述第一输出信息指示所述第二参考信 号集对应的多个波束中被预测为信道质量最好的K个波束，所述K为大于或等于1的整数且所述K小于所述N。
在一种可能的实现方式中，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，包括如下中的至少一项：所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束的信息；或者，K组参考信号各自的组标识，其中，所述K组参考信号对应于所述N组参考信号对应的N个测量结果中被预测为信道质量最好的K个测量结果，且所述K组参考信号各自的组标识与K个波束具有预先定义或预先配置的对应关系；或者，所述N组参考信号对应的多个波束信息及所述波束信息对应的N个测量结果；或者，所述N组参考信号各自的组标识及所述N组参考信号的N个测量结果，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
在一种可能的实现方式中，所述第一波束指示信息指示所述第一参考信号集对应的波束包括：所述第一波束指示信息指示所述第一参考信号集对应的波束在所述第二参考信号集对应的多个波束中的位置。
在一种可能的实现方式中，所述第一波束指示信息包括N个字段，所述N个字段与所述第二参考信号集对应的多个波束一一对应，所述N个字段中的M个字段的比特值与剩余(N-M)个字段的比特值不同；所述第一波束指示信息指示所述第一参考信号集对应的波束，包括：所述第一波束指示信息中的所述M个字段对应所述第一参考信号集。
在一种可能的实现方式中，所述方法还包括：向所述终端设备发送第一配置信息，所述第一配置信息指示以下中的一项或多项：所述M组参考信号的时域资源、所述M组参考信号的频域资源、所述M组参考信号的传输周期、所述M组参考信号的组标识、或所述M组参考信号的波束信息。
在一种可能的实现方式中，所述第一波束指示信息指示所述M组参考信号对应的波束，包括：所述第一波束指示信息包括所述M组参考信号的组标识或波束信息，所述M组参考信号为所述N组参考信号中的部分，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
在一种可能的实现方式中，所述第一波束指示信息包括在所述M组参考信号的第二配置信息中，所述第一波束指示信息包括所述M组参考信号的组标识的情况下，所述第二配置信息还包括所述M组参考信号的时域资源、频域资源、传输周期、或、波束信息中的一项或多项；或者，所述第一波束指示信息包括所述M组参考信号的波束信息的情况下，所述第二配置信息还包括所述M组参考信号的时域资源、频域资源、传输周期、或、组标识中的一项或多项。
在一种可能的实现方式中，所述方法还包括：向所述终端设备发送所述N组参考信号的第三配置信息；所述第一波束指示信息包括所述M组参考信号的组标识的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的组标识并指示以下中的一项或多项：所述N组参考信号的时域资源，所述N组参考信号的频域资源，所述N组参考信号的传输周期，或，所述N组参考信号的波束；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识；或者，所述第一波束指示信息包括所述M组参考信号的波束信息的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的波束信息并指示以下中的一项或多项：所述N组参考信号的组标识，时域资源，所述N组参考信号的频域资源，或，所述N组参考信号的传输周期；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识。
在一种可能的实现方式中，所述方法还包括：接收来自所述终端设备的所述第一输出信息。
第十方面，提供了一种通信装置，该装置用于执行上述第一方面、第四方面、第六方面、第八方面中任一种可能实现方式中的方法。具体地，该装置可以包括用于执行第一方面、第四方面、第六方面、第八方面中任一种可能实现方式中的方法的单元和/或模块，如收发单元和/或处理单元。
在一种实现方式中，该装置为训练设备，推理设备或终端设备。当该装置为通信设备时，通信单元可以是收发器，或，输入/输出接口；处理单元可以是至少一个处理器。可选地，收发器可以为收发电路。可选地，输入/输出接口可以为输入/输出电路。
在另一种实现方式中，该装置为用于训练设备，推理设备或终端设备的芯片、芯片系统或电路。当该装置为用于通信设备的芯片、芯片系统或电路时，通信单元可以是该芯片、芯片系统或电路上的 输入/输出接口、接口电路、输出电路、输入电路、管脚或相关电路等；处理单元可以是至少一个处理器、处理电路或逻辑电路等。
第十一方面，提供了一种通信装置，该装置用于执行上述第二方面、第三方面、第五方面、第七方面、第九方面中任一种可能实现方式中的方法。具体地，该装置可以包括用于执行第二方面、第三方面、第五方面、第七方面、第九方面中任一种可能实现方式中的方法的单元和/或模块，如收发单元和/或处理单元。
在一种实现方式中，该装置为网络设备或者第一网络设备。当该装置为通信设备时，通信单元可以是收发器，或，输入/输出接口；处理单元可以是至少一个处理器。可选地，收发器可以为收发电路。可选地，输入/输出接口可以为输入/输出电路。
在另一种实现方式中，该装置为用于网络设备或者第一网络设备的芯片、芯片系统或电路。当该装置为用于通信设备的芯片、芯片系统或电路时，通信单元可以是该芯片、芯片系统或电路上的输入/输出接口、接口电路、输出电路、输入电路、管脚或相关电路等；处理单元可以是至少一个处理器、处理电路或逻辑电路等。
第十二方面，提供了一种通信装置，该装置包括：至少一个处理器，用于执行存储器存储的计算机程序或指令，以执行上述第一方面、第四方面、第六方面、第八方面中任一方面中任一种可能实现方式中的方法。可选地，该装置还包括存储器，用于存储的计算机程序或指令。可选地，该装置还包括通信接口，处理器通过通信接口读取存储器存储的计算机程序或指令。
在一种实现方式中，该装置为训练设备，推理设备或终端设备。
在另一种实现方式中，该装置为用于训练设备，推理设备或终端设备的芯片、芯片系统或电路。
第十三方面，提供了一种通信装置，该装置包括：至少一个处理器，用于执行存储器存储的计算机程序或指令，以执行上述第二方面、第三方面、第五方面、第七方面、第九方面中任一种可能实现方式中的方法。可选地，该装置还包括存储器，用于存储的计算机程序或指令。可选地，该装置还包括通信接口，处理器通过通信接口读取存储器存储的计算机程序或指令。
在一种实现方式中，该装置为网络设备或者第一网络设备。
在另一种实现方式中，该装置为用于网络设备或者第一网络设备的芯片、芯片系统或电路。
第十四方面，本申请提供一种处理器，包括：输入电路、输出电路和处理电路。所述处理电路用于通过所述输入电路接收信号，并通过所述输出电路发射信号，使得所述处理器执行第一方面至第九方面中任一方面中任一种可能实现方式中的方法。
在具体实现过程中，上述处理器可以为一个或多个芯片，输入电路可以为输入管脚，输出电路可以为输出管脚，处理电路可以为晶体管、门电路、触发器和各种逻辑电路等。输入电路所接收的输入的信号可以是由例如但不限于收发器接收并输入的，输出电路所输出的信号可以是例如但不限于输出给发射器并由发射器发射的，且输入电路和输出电路可以是同一电路，该电路在不同的时刻分别用作输入电路和输出电路。本申请实施例对处理器及各种电路的具体实现方式不做限定。
对于处理器所涉及的发送和获取/接收等操作，如果没有特殊说明，或者，如果未与其在相关描述中的实际作用或者内在逻辑相抵触，则可以理解为处理器输出和接收、输入等操作，也可以理解为由射频电路和天线所进行的发送和接收操作，本申请对此不做限定。
第十五方面，提供了一种处理设备，包括处理器和存储器。该处理器用于读取存储器中存储的指令，并可通过收发器接收信号，通过发射器发射信号，以执行第一方面至第九方面中任一方面中任一种可能实现方式中的方法。
可选地，所述处理器为一个或多个，所述存储器为一个或多个。
可选地，所述存储器可以与所述处理器集成在一起，或者所述存储器与处理器分离设置。
在具体实现过程中，存储器可以为非瞬时性(non-transitory)存储器，例如只读存储器(read only memory，ROM)，其可以与处理器集成在同一块芯片上，也可以分别设置在不同的芯片上，本申请实施例对存储器的类型以及存储器与处理器的设置方式不做限定。
应理解，相关的数据交互过程例如发送指示信息可以为从处理器输出指示信息的过程，接收能力信息可以为处理器接收输入能力信息的过程。具体地，处理器输出的数据可以输出给发射器，处理器接收的输入数据可以来自收发器。其中，发射器和收发器可以统称为收发器。
上述第十五方面中的处理设备可以是一个或多个芯片。该处理设备中的处理器可以通过硬件来实现也可以通过软件来实现。当通过硬件实现时，该处理器可以是逻辑电路、集成电路等；当通过软件来实现时，该处理器可以是一个通用处理器，通过读取存储器中存储的软件代码来实现，该存储器可以集成在处理器中，可以位于该处理器之外，独立存在。
第十六方面，提供一种计算机可读存储介质，该计算机可读介质存储用于设备执行的程序代码，该程序代码包括用于执行上述第一方面至第九方面任一种可能实现方式中的方法。
第十七方面，提供一种包含指令的计算机程序产品，当该计算机程序产品在计算机上运行时，使得计算机执行上述第一方面至第九方面任一种可能实现方式中的方法。
第十八方面，提供一种芯片系统，包括处理器，用于从存储器中调用并运行计算机程序，使得安装有该芯片系统的设备执行上述第一方面至第九方面中任一方面中各实现方式中的方法。
第十九方面，提供一种通信系统，该通信系统包括训练设备和网络设备。所述训练设备用于执行上述第一方面中的任一种可能实现方法，所述网络设备用于执行上述第二方面中的任一种可能实现的方法。
第二十方面，提供一种通信系统，该通信系统包括第一网络设备。所述第一网络设备用于执行上述第三方面中的任一种可能实现方法。
第二十一方面，提供一种通信系统，该通信系统包括训练设备和网络设备。所述训练设备用于执行上述第四方面中的任一种可能实现方法，所述网络设备用于执行上述第五方面中的任一种可能实现的方法。
第二十二方面，提供一种通信系统，该通信系统包括训练设备和网络设备。所述训练设备用于执行上述第六方面中的任一种可能实现方法，所述网络设备用于执行上述第七方面中的任一种可能实现的方法。
第二十三方面，提供一种通信系统，该通信系统包括推理设备，如终端设备，和网络设备。所述推理设备，如终端设备，用于执行上述第八方面中的任一种可能实现方法，所述网络设备用于执行上述第九方面中的任一种可能实现的方法。
附图说明
图1为一种通信系统的结构示意图；
图2为神经元结构的一种示意图；
图3为神经网络的层关系的一种示意图；
图4为本申请提供的一种AI模型的训练与推理的框架示意图；
图5是本申请提供的一种获取训练数据集的方法500的示意性流程图；
图6是本申请提供的一种获取训练数据集的方法600的示意性流程图；
图7是本申请提供的一种获取AI模型的输入信息的方法700的示意性流程图；
图8是本申请提供的一种通信方法800的示意性流程图；
图9是本申请提供的一种通信方法900的示意性流程图；
图10是本申请提供的一种通信装置100的示意性框图；
图11是本申请提供的一种通信装置200的示意性框图。
具体实施方式
下面将结合附图，对本申请实施例中的技术方案进行描述。
本申请提供的技术可以应用于各种通信系统，例如，该通信系统可以是第四代(4th generation，4G)通信系统(例如长期演进(long term evolution，LTE)系统)、第五代(5th generation，5G)通信系统、全球互联微波接入(worldwide interoperability for microwave access，WiMAX)或者无线局域网(wireless local area network，WLAN)系统、卫星通信系统、未来的通信系统，如第六代(6th generation，6G)移动通信系统，或者多种系统的融合系统等。其中，5G通信系统还可以称为新无线(new radio，NR)系统。卫星通信系统、未来的通信系统，如第六代(6th generation，6G)移动通信系统，或者多种系统的融合系统等。
通信系统中的一个设备可以向另一个设备发送信号或从另一个设备接收信号，其中，信号可以包括信息、信令或者数据等。其中，设备也可以被替换为实体、网络实体、通信设备、通信模块、节点、通信节点等等。例如，通信系统可以包括至少一个终端设备和至少一个网络设备。又例如，通信系统可以包括一个训练设备和至少一个网络设备。网络设备可以向终端设备发送下行信号，和/或终端设备可以向接入网设备发送上行信号此外可以理解的是，若通信系统中包括多个终端设备，多个终端设备之间也可以互发信号，即信号的发送网元和信号的接收网元均可以是终端设备。可以理解的是，本申请中的终端设备可以替换为第一设备，网络设备可以替换为第二设备，二者执行本公开中相应的通信方法。
本申请实施例提供的方法可以应用于5G、6G、卫星通信等无线通信系统中。参见图1，图1是本申请实施例提供的无线通信系统的一简化示意图。如图1所示，该无线通信系统包括无线接入网100(网络设备的示例)。无线接入网100可以是下一代(例如6G或更高版本)无线接入网，或传统(例如5G、4G、3G或2G)无线接入网。一个或多个通信设备(120a-120j，统称为120)可以相互连接或连接到无线接入网100中的一个或多个网络设备(110a、110b，统称为110)。可选的，图1只是示意图，该无线通信系统中还可以包括其它设备，如还可以包括核心网设备、无线中继设备和/或无线回传设备等，在图1中未画出。
可选的，在实际应用中，该无线通信系统可以同时包括多个网络设备(例如，接入网设备)，也可以同时包括多个通信设备。一个网络设备可以同时服务于一个或多个通信设备。一个通信设备也可以同时接入一个或多个网络设备。本申请实施例对该无线通信系统中包括的通信设备和网络设备的数量不做限定。
其中，网络设备可以是网络侧的一种用于发射或接收信号的实体。网络设备可以为通信设备通过无线方式接入到该无线通信系统中的接入设备，例如，网络设备可以是基站。基站可以广义的覆盖如下中的各种名称，或与如下名称进行替换，比如：节点B(NodeB)、演进型基站(evolved NodeB，eNB)、下一代基站(next generation NodeB，gNB)、开放无线接入网(open radio access network，O-RAN)中的接入网设备、中继站、接入点、传输点(transmitting and receiving point，TRP)、发射点(transmitting point，TP)、主站MeNB、辅站SeNB、多制式无线(MSR)节点、家庭基站、网络控制器、接入节点、无线节点、接入点(AP)、传输节点、收发节点、基带单元(BBU)、射频拉远单元(RRU)、有源天线单元(AAU)、射频头(RRH)、中心单元(CU)、分布单元(DU)、无线单元(radio unit，RU)、集中单元控制面(CU control plane，CU-CP)节点、集中单元用户面(CU user plane，CU-UP)节点、定位节点等。基站可以是宏基站、微基站、中继节点、施主节点或类似物，或其组合。网络设备还可以指用于设置于前述设备或装置内的通信模块、调制解调器或芯片。网络设备还可以是移动交换中心以及设备到设备(Device-to-Device，D2D)、车辆外联(vehicle-to-everything，V2X)、机器到机器(machine-to-machine，M2M)通信中承担基站功能的设备、6G网络中的网络侧设备、未来的通信系统中承担基站功能的设备等。网络设备可以支持相同或不同接入技术的网络。本申请的实施例对网络设备所采用的具体技术和具体设备形态不做限定。
网络设备可以是固定的，也可以是移动的。例如，基站110a、110b(网络设备的示例)是静止的，并负责来自通信设备120的一个或多个小区中的无线传输和接收。图1中示出的直升机或无人机120i可以被配置成充当移动基站，并且一个或多个小区可以根据移动基站120i的位置移动。在其他示例中，直升机或无人机(120i)可以被配置成用作与基站110b通信的通信设备。
本申请中，用于实现如上网络功能的通信装置例如可以是接入网设备，也可以是具有接入网络的部分功能的网络设备，也可以是能够支持实现接入网络功能的装置，例如芯片系统，硬件电路、软件模块、或硬件电路加软件模块，该装置可以被安装在接入网设备中或者和接入网设备匹配使用。
通信设备可以是用户侧的一种用于接收或发射信号的实体，如手机。通信设备可以用于连接人、物和机器。通信设备可通过网络设备与一个或多个核心网进行通信。通信设备包括具有无线连接功能的手持式设备、连接到无线调制解调器的其他处理设备或车载设备等。通信设备可以是便携式、袖珍式、手持式、计算机内置的或者车载的移动装置。通信设备120可以广泛应用于各种场景，例如蜂窝通信、设备到设备D2D、车到所有V2X、端到端P2P、机器到机器M2M、机器类型通信MTC、物联网IOT、虚拟现实VR、增强现实AR、工业控制、自动驾驶、远程医疗、智能电网、智能家具、智能 办公、智能穿戴、智能交通、智慧城市、无人机、机器人、遥感、被动传感、定位、导航与跟踪、自主交付与移动等。通信设备120的一些举例为：3GPP标准的用户设备(UE)、固定设备、移动设备、手持设备、可穿戴设备、蜂窝电话、智能电话、会话发起协议(SIP)电话、笔记本电脑、个人计算机、智能书、车辆、卫星、全球定位系统(GPS)设备、目标跟踪设备、无人机、直升机、飞行器、船只、遥控设备、智能家居设备、工业设备、个人通信业务(personal communication service，PCS)电话、无线本地环路(wireless local loop，WLL)站、个人数字助理(personal digital assistant，PDA)、无线网络摄像头、平板电脑、掌上电脑、移动互联网设备(mobile internet device，MID)、可穿戴设备如智能手表、虚拟现实(virtual reality，VR)设备、增强现实(augmented reality，AR)设备、工业控制(industrial control)中的无线终端、车联网系统中的终端、无人驾驶(self driving)中的无线终端、智能电网(smart grid)中的无线终端、运输安全(transportation safety)中的无线终端、智慧城市(smart city)中的无线终端如智能加油器，高铁上的终端设备以及智慧家庭(smart home)中的无线终端，如智能音响、智能咖啡机、智能打印机等。通信设备120可以为以上各种场景中的无线设备或用于设置于无线设备的装置，例如，上述设备中的通信模块、调制解调器或芯片等。通信设备也可以称为终端、终端设备、用户设备(user equipment,UE)、移动台(mobile station，MS)、移动终端(mobile terminal，MT)等。通信设备还可以是未来的无线通信系统中的通信设备。通信设备可以用于专用网设备或者通用设备中。本申请的实施例对通信设备所采用的具体技术和具体设备形态不做限定。
可选的，通信设备可以用于充当基站。例如，UE可以充当调度实体，其在V2X、D2D或P2P等中的UE之间提供侧行链路信号。如图1所示，蜂窝电话120a和汽车120b利用侧行链路信号彼此通信。蜂窝电话120a和智能家居设备120e之间通信，而无需通过基站110b中继通信信号。
本申请中，用于实现通信设备功能的通信装置可以是终端设备，也可以是具有以上通信设备的部分功能的终端设备，也可以是能够支持实现以上通信设备的功能的装置，例如芯片系统，该装置可以被安装在终端设备中或者和终端设备匹配使用。本申请中，芯片系统可以由芯片构成，也可以包括芯片和其他分立器件。
可选的，无线通信系统通常由小区组成，基站提供小区的管理，基站向小区中多个移动台(mobile station，MS)提供通信服务。其中基站包含基带单元(baseband unit，BBU)和远端射频单元(remote radio unit，RRU)。BBU和RRU可以放置在不同的地方，例如：RRU拉远，放置于高话务量的区域，BBU放置于中心机房。BBU和RRU也可以放置在同一机房。BBU和RRU也可以为一个机架下的不同部件。可选的，一个小区可以对应于一个载波或成员载波。
应理解，图1所示的通信系统中各个设备的数量、类型仅作为示意，本申请并不限于此，实际应用中在通信系统中还可以包括更多的终端设备、更多的网络设备，还可以包括其它网元，例如可以包括核心网设备，和/或用于实现人工智能功能的网元。
为了便于理解本申请提供的技术方案，下面首先对本申请涉及专业术语的进行简单的介绍。可以理解的是，该介绍并不作为对本申请的限定。
一、人工智能(artificial Intelligence，AI)技术
1、AI模型
AI模型是AI技术功能的具体实现，AI模型表征了模型的输入和输出之间的映射关系。AI模型的类型可以是神经网络、线性回归模型、决策树模型、支持向量机(support vector machine，SVM)、贝叶斯网络、Q学习模型或者其他机器学习(machine learning,ML)模型。
根据实现人工智能的具体方法和/或技术的不同，AI模型具体也可以称为机器学习模型、深度学习模型或强化学习模型。其中，机器学习是实现人工智能的一种方法，该方法的目标是设计和分析一些让计算机可以自动“学习”的算法(也即“模型”)，所设计的算法称为“机器学习模型”。机器学习模型是一类从数据中自动分析获得规律，并利用规律对未知数据进行预测的算法。机器学习模型包括多种多样，根据模型训练时是否需要依赖训练数据对应的标签，机器学习模型可以分为：有监督学习模型和无监督学习模型，下面主要介绍一下“有监督学习模型”。
2、有监督学习模型
“有监督学习模型”是根据给定的训练数据集中的数据以及该训练数据集中各数据对应的标签，确定初始AI模型的参数后得到的模型，其中，利用训练数据集中的数据及数据对应的标签 确定初始AI模型的参数的过程也称为“有监督学习”(或者“有监督训练”)。训练数据集中的数据的标签通常是由人工标注的，用于标识该数据在特定任务上的正确答案。典型的有监督学习模型包括：支持向量机、神经网络模型、逻辑回归模型、决策树、朴素贝叶斯模型、高斯判别模型等。有监督学习模型通常用于分类或回归。其中，定量输出称为“回归”，也可以理解为AI模型是“连续变量预测”的；定性输出称为“分类”，也可以理解为AI模型是“离散变量预测”。
3、深度神经网络(deep neural network，DNN)
DNN是机器学习的一种具体实现形式，根据通用近似定理，神经网络理论上可以逼近任意连续函数，从而使得神经网络具备学习任意映射的能力。传统通信系统需要借助丰富的专家知识来设计通信模块，而基于DNN的深度学习通信系统可以从大量的数据集中自动发现隐含的模式结构，建立数据之间的映射关系，获得优于传统建模方法的性能。
DNN的思想来源于大脑组织的神经元结构，每个神经元都对其输入值做加权求和运算，并将加权求和结果通过一个非线性函数产生输出，如图2所示。具体的，假设神经的输入为x＝[x0,…,xn]，与输入对应的权值为d＝[d0,…,dn]，加权求和的偏置为b，非线性函数的形式可以多样化，例如，非线性函数为max{0,x}，则一个神经元的执行的效果可以是此时，权值d＝[d0,…,dn]和偏置b便可以理解为模型的参数。
DNN一般具有多层结构，DNN的每一层都可包含多个神经元，输入层将接收到的数值经过神经元处理后，传递给中间的隐藏层。类似的，隐藏层再将计算结果传递给最后的输出层，产生DNN的最后输出，如图3所示。DNN一般具有多于一个的隐藏层，隐藏层往往直接影响提取信息和拟合函数的能力。增加DNN的隐藏层数或扩大每一层的宽度都可以提高DNN的函数拟合能力。每个神经元中加权值即为DNN网络模型的参数。模型参数通过训练过程得到优化，从而使得DNN网络具备提取数据特征、表达映射关系的能力。
4、AI模型训练与推理
任何一个AI模型在用于解决特定的技术问题之前，都需要经过训练。如图4所示AI模型的训练是指利用指定初始模型对训练数据进行计算，根据计算的结果采用一定的方法对初始模型中的参数进行调整，使得该模型逐渐学习到一定的规律，具备特定的功能的过程。经过训练后具有稳定功能的AI模型即可用于推理。AI模型的推理是利用训练完成的AI模型对输入的数据进行计算，获得预测的推理结果的过程。
在训练阶段，首先需要基于目标构建针对深度学习模型的训练集，训练集中包括多个训练数据，每个训练数据设置有标签，训练数据的标签是该训练数据在特定问题上的正确答案，标签可以表示利用训练数据对深度学习模型进行训练的目标。
对深度学习模型进行训练时，训练数据可以分批地输入至经过参数初始化后的深度学习模型，深度学习模型对训练数据进行计算(即“推理”)，获得针对训练数据的预测结果。经过推理获得的预测结果以及训练数据对应的标签作为根据损失(loss)函数计算损失的数据。损失函数是用于在模型训练阶段计算模型针对训练数据的预测结果和该训练数据的标签之间的差距(即“损失值”)的函数，损失函数可以采用不同数学函数实现，常用的损失函数的表达式有：均方误差损失函数、对数损失函数、最小二乘法等。模型的训练是一个重复迭代的过程，每次迭代对不同的训练数据进行推理，并计算损失值，多次迭代的目标是不断地更新深度学习模型的参数，找到使损失函数的损失值最低或者趋于平稳的参数配置。
5、训练数据集和推理数据
“训练数据集”用于AI模型的训练，训练数据集可以包括AI模型的输入，或者包括AI模型的输入和目标输出。其中，训练数据集包括一个或多个训练数据，训练数据可以是输入至AI模型的训练样本，也可以是AI模型的目标输出。其中，目标输出也可以被称为“标签”或者“标签样本”。训练数据集是机器学习重要的部分之一，模型训练本质上就是从训练数据中学习它的某些特征，使得AI模型的输出尽可能接近目标输出，例如，使得AI模型的输出与目标输出之间的差异尽可能地小。训练数据集的构成与选取，在一定程度上可以决定训练出来的AI模型的性能。其中，模型的性能例如可以通过“损失值”、“推理准确性”等来衡量。
另外，在AI模型(如神经网络)的训练过程中，可以定义损失函数。损失函数描述了AI模 型的输出值与目标输出值之间的差距或差异。本申请并不限制损失函数的具体形式。AI模型的训练过程就是通过调整AI模型的模型参数，使得损失函数的取值小于门限，或者使得损失函数的取值满足目标需求的过程。例如，AI模型为神经网络，调整神经网络的模型参数包括调整如下参数中的至少一种：神经网络的层数、宽度、神经元的权值、或神经元的激活函数中的参数。
“推理数据”可以作为已训练好的AI模型的输入，用于AI模型的推理。在模型推理过程中，将推理数据输入AI模型，可以得到对应的输出即为推理结果。
二、波束管理
1、波束
本申请中，“波束”也可以理解为“空域滤波参数”、“空间滤波器(spatial filter)”或者“空间参数(spatial parameters)”。一般用于发送信号的波束可以称为发射波束(transmission beam，Tx beam)，可以称为空域发送滤波器(spatial domain transmit filter)或空域发射参数(spatial domain transmit parameter)；用于接收信号的波束可以称为接收波束(reception beam，Rx beam)，可以称为空域接收滤波器(spatial domain receive filter)或空域接收参数(spatial domain receive parameter)。
在新无线(new radio，NR)协议中，波束例如可以是空域滤波参数(例如，空域接收滤波参数，或者；空域发送滤波参数)。但应理解，本申请并不排除在未来的协议中定义其他的术语来表示相同或相似的含义的可能。
2、波束扫描
“波束扫描”是指在特定周期或者时间段内，波束在预定义的方向上以固定的周期进行传送，以覆盖特定空间区域。例如，初始接入过程中，UE需要与系统进行同步并接收最小系统信息。因此，采用承载同步信号和物理广播信道(physical broadcast channel，PBCH)块(synchronization signal and PBCH block，SSB)以固定周期进行扫描和发送。信道状态信息参考信号(channel state information-reference signal，CSI-RS)也可以采用波束扫描技术，但是如果要对所有预定义的波束方向进行覆盖的话，其开销太大，因此CSI-RS仅根据所服务的终端设备的位置，在预定义波束方向的特定子集中进行传送。
3、波束测量
“波束测量”是指网络设备或者终端设备对所接收到的波束赋形的信号的质量和特性进行测量的过程。波束管理过程中，终端设备或者网络设备可以通过SSB、CSI-RS获取参考信号的参考信号接收功率(seference signal receiving power，RSRP)、参考信号接收质量(reference signal receiving quality，RSRQ)、信号与干扰加噪声比(signal to interference plus noise ratio，SINR)等信息，从而识别最好波束。
4、波束确定
网络设备或者终端设备选择其所使用的发送波束或者接收波束，其中，下行波束可以由终端设备来确定，例如，其判决准则是波束的最大接收信号强度应大于特定的门限。上行方向上，终端设备根据网络设备的方向传送探测参考信号(sounding reference signal，SRS)，网络设备对SRS进行测量以确定最好的上行波束。
目前，AI模型可以部署在训练设备(例如，终端设备)上进行训练、更新。训练设备在训练AI模型时，为了支撑训练设备对该AI模型的训练，网络设备会持续(例如，周期性地)向训练设备传输训练数据集，直到训练设备向网络设备发送完成训练模型的指示信息后，网络设备才会停止向训练设备发送训练数据集。然而，训练设备在训练该AI模型的过程中，网络设备向训练设备传输的训练数据集对训练设备而言是不必要的，会造成资源浪费。因此，训练设备进行AI模型训练时如何获取训练数据集，成为需要解决的技术问题。
有鉴于此，本申请提出一种模型训练方法，训练设备可以向网络请求发送训练数据集，并且该请求信息中也指示了训练设备需要的网络设备向其发送的训练数据集的相关信息。换句话说，本申请中，训练设备可以向网络设备指示需要哪些训练数据集，网络设备可以向训练设备发送训练设备指示的训练数据集，不需要一直下发训练数据集。该方法可以减少空口资源浪费和空口开销，与其他方法相比，空口资源的使用性能得到提高。
需要说明的是，本申请中的“训练设备”例如可以理解为终端设备，即，该终端设备既可以与 网络设备进行通信，同时该终端设备也具备支持模型训练的能力。又例如，该“训练设备”可以理解为，专门用于模型训练的设备，例如，该设备上可以只提供模型训练的功能，如果该设备确定模型训练完毕，则可以将训练好的模型发送给需要的终端设备。
本申请中，“模型训练”以“AI模型训练”进行举例说明，并且假设AI模型部署在训练设备上。
图5是本申请提供的一种获取训练数据集的方法500的示意性流程图。下面对图5所示的各步骤进行说明。需要说明的是，图5中用虚线表示的步骤是可选的，在后文中不多赘述，该方法包括：
可选的，步骤501，训练设备确定对该AI模型的性能进行监控。
例如，训练设备从第二小区切换到第一小区，其中，第一网络设备为第一小区提供服务，第二网络设备为第二小区提供服务。此时，训练设备确定对该AI模型的性能进行监控，或者第一网络设备指示训练设备对AI模型的性能进行监控。
又例如，训练设备发现AI模型的输入信息发生变化。例如，训练设备发现稀疏波束图样不属于全码本中的波束图样。此时，训练设备确定对该AI模型的性能进行监控。
可选的，步骤502，第一网络设备向训练设备发送训练数据集。
当进入模型监控阶段后，第一网络设备可以为训练设备配置参考信号资源，并向训练设备传送N组参考信号(即，训练数据集)。例如，在波束管理场景下，训练设备可以通过全码本扫描(也可以理解为，全波束扫描)，获取该AI模型的输入信息和标签。
具体的，在一种可能的实现方式中，第一网络设备向训练设备发送N组参考信号，训练设备可以测量N组参考信号获得对应的N组测量结果。例如，该N组测量结果中的每组测量结果包括参考信号的RSRP，然后训练设备可以将N组测量结果中的任意一组测量结果确定为该AI模型的输入信息，或者，训练设备可以将N组测量结果作为该AI模型的输入信息。在另一种可能的实现方式中，第一网络设备可以向训练设备指示N组参考信号中的M组参考信号对应的测量结果作为该AI模型的输入信息(具体参见方法700中的描述)。基于该AI模型的配置，例如，训练设备可以将N组测量结果中的所有参考信号的RSRP中，RSRP值最大的几个RSRP的值作为该AI模型训练的标签。之后，训练设备便可以通过输入信息获取该AI模型的输出信息(即，AI模型的推理结果)，该输出信息与标签进行比较，例如，可以通过训练损失或者训练准确度衡量模型预测性能。例如，可以设置一个门限值，若训练损失(或者，准确性)大于或者等于该门限值，则说明该AI模型符合新小区的要求或者新输入信息的要求，即可以继续使用该AI模型；若训练损失(或者，准确性)低于该门限值，则训练设备可以判定该AI模型不适合新小区的要求或者新输入信息的要求，此时，训练设备确定该AI模型需要进行模型更新。本实施例中假设训练设备确定该AI模型需要更新，则需要继续执行下述步骤503～步骤506。
步骤503，训练设备向第一网络设备发送第一信息，该第一信息用于指示训练设备请求第一网络设备发送的第一训练数据集的相关信息。
对应的，第一网络设备接收来自训练设备的第一信息。
本申请中，第一训练数据集的相关信息，例如可以包括下至少一项：第一训练数据集的大小的信息、AI模型的输入的配置信息、用于AI模型的训练的参考信号的配置信息。
本申请中，例如训练设备可以通过以下方式确定“第一训练数据集的大小的信息”：
在一种可能的实现方式中，第一训练数据集的大小的信息可以是由训练设备基于完成该AI模型的训练所需要的训练数据集的大小确定的。示例性的，第一训练数据集的大小的信息可以是训练设备基于完成所述AI模型的训练所需要的训练数据集的大小确定的。例如，训练设备可以通过历史信息确定从AI模型的初始状态(例如，AI模型的初始状态为0)训练该AI模型到收敛状态时，总共需要的训练数据集的数量。例如，训练设备基于历史经验确定训练该AI模型共需要扫描6万次全码本获得训练数据集。
在另一种可能的实现方式中，在步骤501之前，该方法还包括：训练设备确定该AI模型的第一性能；训练设备根据该AI模型的第一性能和该AI模型的第二性能，确定该第一训练数据集的大小的信息，其中，第一性能为该AI模型的当前性能，该第二性能为该AI模型的目标性能。示例性的，训练设备对该AI模型监控时，通过将监控到的当前AI模型的性能与历史信息中该 AI模型收敛时对应的模型性能相比较，并且按照当前AI模型的性能可以估计出达到预期的模型性能时需要的第一训练数据集的大小。
需要说明的是，本申请中，AI模型的性能可以通过衡量“训练损失”、“训练准确度”等信息进行判断。例如，可以将“训练损失”与一个门限值相比较，通过比较结果可以衡量该AI模型训练的性能。
本申请中，可以由训练设备指示第一网络设备需要的训练数据集的大小，使得第一网络设备基于训练设备的指示下发训练数据集，从而可以避免第一网络设备在训练设备模型训练过程中持续向训练设备传输数据，减少空口资源的浪费，提高空口资源的使用性能。
本申请中，“参考信号的配置信息”例如可以包括以下至少一项：参考信号的标识、参考信号的时域资源、参考信号的频域资源、参考信号的传输周期、传输的该参考信号的类型。例如，参考信号的类型为SSB，或者CSI-RS，或者SRS，等等。其中，参考信号的标识也可以理解为参考信号组的标识，例如，参考信号的配置信息包含N(N为大于或者等于1的整数)组参考信号的组标识，其中，N组参考信号中的每组参考信号具有相同的组标识，每组参考信号包括至少一个参考信号。类似的，参考信号的时域资源、参考信号的频域资源、参考信号的传输周期、传输的该参考信号的类型分别也可以理解为，N组参考信号的时域资源、N组参考信号的频域资源、N组参考信号的传输周期、传输的该N组参考信号的类型。
本申请中“资源”可以是频域资源、时域资源、资源块(resource block，RB)、物理资源块(physical resource block，PRB)等等，本申请不予限定。
本申请中，训练设备可以通过训练该AI模型的历史信息确定参考信号的配置信息，并指示给第一网络设备，使得网络设备可以为该训练设备配置参考信号，训练设备通过测量配置的参考信号获取AI模型的输入信息和标签，从而继续对该AI模型进行训练。进一步的，本实施例中，在该AI模型多次迭代训练过程中，也可以由第一网络设备指示该AI模型的输入信息，从而加速该AI模型的收敛。具体的实现方式可以参见下述方法700，换句话说，方法500也可以与方法700进行结合。
本申请中，“AI模型的输入的配置信息”例如可以理解为，训练设备基于该AI模型训练的历史信息，确定该AI模型的输入信息，例如，该AI模型的输入信息为稀疏波束图样对应的参考信号的测量结果。则训练设备可以确定该稀疏波束图样是全码本中的哪些位置的波束。例如，训练设备可以将该稀疏波束图样的标识，或者将该稀疏波束图样对应的参考信号的标识，或者将参考信号的测量结果等信息上报给第一网络设备，从而可以使得第一网络设备下发与该波束图样对应的训练数据集。
本申请中，“第一信息用于指示训练设备请求网络设备发送的第一训练数据集的相关信息”，具体可以有以下实现方式。在一种可能的实现方式中，第一信息可以显示的指示训练设备请求第一网络设备发送的第一训练数据集的相关信息。例如，第一信息中可以包括第一训练数据集的大小的信息、该AI模型的输入的配置信息、用于该AI模型的训练的参考信号的配置信息。具体的，假设训练设备确定第一训练数据集大小是需要对全码本扫描6万次，此时第一信息中可以包含指示该扫描次数的具体数值的信息。在另一种可能的实现方式中，第一信息可以隐式的指示训练设备请求第一网络设备发送的第一训练数据集的相关信息。例如，第一信息可以通过包含索引的方式指示第一训练数据集的大小，第一网络设备可以通过查询该索引确定该索引对应的具体数值。对于第一信息指示的第一训练数据集的其它相关信息也可以类似理解，不再一一举例说明。
可选的，本申请中，第一信息还包括以下至少一项：该AI模型的标识信息、该AI模型的应用场景的信息、该AI模型的用途信息以及训练设备的算力能力的信息。
例如，网络设备上可以保存第一映射关系，第一映射关系为各个AI模型的标识以及该AI模型标识对应的训练数据集的大小之间的对应关系。例如，该第一映射关系可以是表格的形式。如表格1所示，AI模型#1对应训练数据集#A，AI模型#2对应训练数据集#B，AI模型#3对应训练数据集#C。训练设备通过发送该AI模型的标识便可以指示需要的训练数据集的大小。
又例如，AI模型的应用场景或者该AI模型的用途可以理解为，该AI模型是用于波束管理场景，或者CSI反馈场景，或者定位场景，等等。训练设备通过向第一网络设备指示该AI模型 的应用场景或用途便可以指示出训练该AI模型需要的训练数据集的大小。
再例如，训练设备可以向第一网络设备上报算力能力的信息。本申请中，训练设备上报的算力能力的信息至少包括以下一项：训练设备的处理器(例如，中央处理器(center processing unit，CPU)、图形处理器(graphic processing unit，GPU)、张量处理器(tensor processing unit，TPU)、神经网络处理器(neural network processing unit，NPU)、现场可编程门阵列(field-programmable gate array，FPGA)等等)的能力、训练设备的存储空间的大小、训练设备的内存的大小、训练设备的电量等等，不予限定。训练设备通过向网络设备上报自己的算力能力，便可以指示出训练该AI模型时最多可以处理的训练数据集的数量。例如，训练设备最大可以对全码本扫4万次，此时，第一网络设备向训练设备发送的训练数据集的大小不会超过训练设备的算力能力。
表格1
步骤504，第一网络设备根据第一信息所指示的相关信息，向训练设备发送第一训练数据集。
对应的，训练设备接收来自第一网络设备的第一训练数据集。
本申请中，第一训练数据集是基于第一信息所指示的该第一训练数据集的相关信息的训练数据集，第一训练数据集用于AI模型的训练。换句话说，第一训练数据集是基于第一信息所指示的第一训练数据的相关信息确定的。
可选的，本申请中，第一网络设备还可以从第二网络设备获取第三信息，第三信息为第二网络设备上训练该AI模型的相关信息。例如，第三信息包括以下至少一项：训练设备请求第二网络设备发送的训练数据集的大小的信息、完成该AI模型训练需要的训练数据集的大小的信息、该AI模型的标识信息、训练设备的算力能力的信息，等等。可选的，第三信息还包括：第二网络设备发送训练数据集的时长的信息和/或第二网络设备发送训练数据集的方式的信息。具体的，第一网络设备可以根据第一信息所指示的相关信息和第三信息，向训练设备发送第一训练数据集，具体的实现方式可以参见下述方法600的描述。
可选的，步骤505，训练设备基于第一训练数据集训练该AI模型。
例如，训练设备接收到第一训练数据集后，可以确定出输入信息和标签。在一种可能的实现方式中，第一训练数据集为N组参考信号，终端设备可以测量N组参考信号获得对应的N组测量结果。例如，该N组测量结果中的每组测量结果包括参考信号的RSRP，然后训练设备可以将N组测量结果中的任意一组测量结果确定为该AI模型的输入信息。在另一种可能的实现方式中，网络设备可以向终端设备指示N组参考信号中的M组参考信号对应的测量结果作为该AI模型的输入信息(具体参见方法700中的描述)。基于该AI模型的配置，例如，训练设备可以将N组测量结果中的所有参考信号的RSRP的测量结果作为该AI模型的标签。之后，训练设备便可以通过输入信息获取该AI模型的输出信息，该输出信息与标签进行比较获得AI模型的训练损失，以上过程可以理解为AI模型的一次训练。训练设备可以基于该次模型训练的训练损失、训练准确度衡量该AI模型的性能，并确定下一次模型训练需要的训练数据集。例如，该训练数据集的大小可由训练设备基于模型性能的评估结果决定，例如，如果训练设备确定该AI模型性能经过第一轮训练之后模型性能显著变好，则可以减少训练数据集的数据量。具体的，还包括步骤：
训练设备可以根据第一训练数据集训练该AI模型，并确定该AI模型的性能；训练设备可以根据AI模型的性能，向网络设备发送第二信息，该第二信息用于指示训练设备请求网络设备发送的第二训练数据集的相关信息；训练设备接收来自所网络设备的第二训练数据集，该第二训练数据集是基于第二信息所指示的相关信息的训练数据集，该第二训练数据集用于AI模型的训练。例如，第二训练数据集的数据量可以小于第一训练数据集的数据量。后续，训练设备仍然可以基于第二训练数据集对该AI进行训练，并反复迭代，例如，基于第二训练数据集进行模型训练，再次确定该AI模型的性能，基于该AI模型的性能确定需要的第三训练数据集的大小，等等。假 设训练设备可以进行L(L为大于1的整数)次训练，直到训练设备确定该AI模型收敛(“模型收敛”也可以理解为该AI模型达到目标性能)。
可选的，步骤506，训练设备向第一网络设备发送模型训练完成的指示信息。
训练设备完成AI模型训练之后，便可以进入模型推理阶段。
基于上述技术方案，本申请中，训练设备可以向网络指示需要的训练数据集的相关信息，使得网络设备可以基于该指示向训练设备发送训练数据集，不需要一直下发训练数据集。该方法可以减少空口资源浪费和空口开销，空口资源的使用性能得到提高。
方法500中主要介绍了由训练设备确定需要的训练数据集的相关信息，从而使得网络设备可以基于训练设备的请求下发训练数据集，以此减小空口资源的浪费，提高空口资源的使用性能。下述方法600主要介绍了，如果训练设备进行小区切换，则切换完成后的新网络设备可以从切换前的旧网络设备获取训练该AI模型的相关信息，并且基于该信息确定需要给训练设备发送的训练数据集。
图6是本申请提供的一种获取训练数据集的方法600的示意性流程图，方法600中考虑到，如果训练设备进行了小区切换，此时，训练设备确定需要进行模型监控，假设训练设备通过模型监控确定该AI模型需要进行模型更新。方法600包括：
可选的，步骤601～步骤602，训练设备确定进行对该AI模型的性能进行监控，并且本实施例中依然假设训练设备确定该AI模型需要更新，则需要继续执行下述步骤603～步骤608。具体的，步骤601～步骤602的实现方式可以参见方法500中的步骤501～步骤502，不再赘述。
步骤603，训练设备向第一网络设备发送第一信息，该第一信息用于请求第一网络设备发送训练数据集。
例如，训练设备通过对该AI模型性能的监控，确定需要对该AI模型进行更新，则可以向第一网络设备发送第一信息。
步骤604，第一网络设备从第二网络设备获取第三信息，第三信息为训练该AI模型的相关信息。
在一种可能的实现方式中，第一网络设备可以向第二网络设备发送请求信息，该请求信息用于请求获取训练AI模型的信息，第二网络设备可以基于该请求信息向第一网络设备同步训练AI模型的相关信息。在另一种可能的实现方式中，第二网络设备可以主动向第一网络设备提供训练该AI模型的相关信息。
例如，第三信息可以包括以下至少一项：训练设备请求第二网络设备发送的训练数据集的大小的信息、完成AI模型训练需要的训练数据集的大小的信息、AI模型的标识信息、训练设备的算力能力的信息。
其中，“训练设备请求第二网络设备发送的训练数据集的大小的信息”可以理解为，第二网络设备与训练设备连接时，训练设备向第二网络设备请求需要的训练数据集的大小的信息。例如，训练设备也可以基于历史信息确定向第二网络设备请求的训练数据集的大小。“完成AI模型训练需要的训练数据集的大小的信息”可以理解为，第二网络设备上保存有训练设备训练该AI模型时总共需要的训练数据集的大小。例如，网络设备之间(例如，第一网络设备、第二网络设备)可以共同维护第一映射关系，第一映射关系为AI模型的标识与该AI模型标识对应的训练数据集的大小之间的映射关系。例如，该第一映射关系可以是表格的形式，如表格1所示。即，根据历史经验统计每个AI模型在本小区完成一次AI模型的训练所需的训练数据的数量。换句话说，网络设备之间可同步有关表格1中的信息。例如，第一网络设备和第二网络设备上都保存了表格1。若表格1中没有该AI模型的标识，则第一网络设备可以根据第二网络设备中下发给训练设备的训练数据集的大小决定本次模型训练时训练数据集的大小。需要说明的是，对于某些AI模型，可能完全不适配本小区的环境，第一网络设备可以指示该AI模型无法训练，建议训练设备更换AI模型。第二网络设备也可以向第一网络设备同步训练设备的算力能力的信息，以便于第一网络确定应该向该训练设备发送的训练数据集的大小。
需要说明的是，本申请中的“映射关系”也可以表述为“关联关系”、“对应关系”。应理解，本申请实施例中所说的“映射关系”可以通过函数关系、或表格、或映射关系等方式保存或被记录。 下述实施例中，提到的“映射关系”可以是网络设备配置的，也可以是协议预定义的，等等，不予限定。
可选的，第三信息还包括：第二网络设备发送训练数据集的时长的信息和/或第二网络设备发送训练数据集的方式的信息。其中，“第二网络设备发送训练数据集的时长的信息”也可以理解为，训练设备与第二网络设备连接时，训练该AI模型的时间。换句话说，训练该AI模型直到其完成收敛需要的时间。
“第二网络设备发送训练数据集的方式的信息”也可以理解为，例如，第二网络设备是周期性地向该训练设备发送训练数据集。示例性的，网络设备可以识别到一天当中哪个时间段数据请求量是最少的，也可以理解为，哪个时间段空口资源占用最少，或者哪个时间段空口资源充足，此时就可以对AI模型提供训练数据集。例如，网络设备可以每天都选择该时间段支撑AI模型的更新(此处，不限制训练设备的数量和/或AI模型的数量)。又例如，第二网络设备是间隔向该训练设备发送训练数据集。该方案相对周期发灵活度有一些提升，如果网络设备发现和/或判断当前数据请求量较少，空口资源充足时，那么此时网络设备就确定可以支撑AI模型的更新。换句话说，该实现方式中，网络设备可以基于当前时刻空口资源的占用情况，确定向训练设备发送训练数据集，所以其不具有明显周期性的特征。
第一网络设备可以基于该信息确定自身是否具备支持该AI模型训练的能力，并且，第一网络也可以基于该信息(例如，训练设备的算力能力的信息)确定该训练设备是否具备支持该AI模型训练的能力。例如，对于某些AI模型，第一网络设备基于历史信息确定该训练设备确实无法训练到收敛状态，此时，可以理解为该训练设备不支持该AI模型的训练。又例如，第一网络设备基于第二网络设备同步的信息确定没有充足的空口资源为训练设备下发训练数据集，此时，可以理解为，第一网络设备不支持该AI模型的训练。
需要说明的是，本申请中，步骤603与步骤604之间没有先后顺序的限定，例如，步骤603和步骤604还可以同时执行。
步骤605，第一网络设备根据第三信息，确定待发送的第一训练数据集。
如步骤604所述，第一网络从第二网络设备获取该训练设备训练该AI模型的相关信息，确定为该训练设备下发的第一训练数据的大小。进一步的，第一网络设备可能同时为多个训练设备或者终端设备提供服务。例如，第一网络设备可能需要向其它的训练设备下发训练数据集，并且，第一网络设备可能还需要向多个终端设备传输控制信息，等等。此时，第一网络设备的空口资源十分紧缺，因此，第一网络需要基于当前空口资源的使用情况综合确定为该训练设备下发的训练数据集的大小。例如，第一网络设备基于第三信息确定需要为训练设备发送的训练数据集为训练数据集#A，但是，由于第一网络设备上没有充足的时频资源传输训练数据集#A，此时，第一网络设备可以确定只传输部分训练数据集。
本申请中，“资源使用情况”也可以理解为“资源占用情况”、“空口资源占用”等等。例如，现有协议框架中已经定义了网络设备最多可以配置的参考信号(例如，64个CSI-RS)的数量，如果网络设备发现全部参考信号都已经被配置给了其他功能，那么就可以确定当前资源被占用，无法给训练设备配置参考信号资源以支撑该AI模型的更新。
可选的，第一信息指示了训练设备请求第一网络设备发送的第一训练数据集的相关信息。例如，第一信息可以包括下至少一项：请求的训练数据集的大小的信息、AI模型的输入的配置信息、用于AI模型的训练的参考信号的配置信息、该AI模型的标识信息。可选的，第一信息还包括以下至少一项：该AI模型的标识信息、该AI模型的应用场景的信息、该AI模型的用途信息、该训练设备的算力能力的信息。换句话说，本实施例中还可以和方法500进行结合，即，训练设备可以向网络设备指示请求的第一训练数据集的大小。此时，第一网络设备可以基于第三信息、第一信息以及空口资源的使用情况，综合判断可以给训练设备发送的训练数据集的大小。
步骤606，第一网络设备向训练设备发送第一训练数据集。
对应的，训练设备接收来自第一网络设备的第一训练数据集。
本申请中，第一训练数据集用于该AI模型的训练。
步骤607，训练设备基于第一训练数据集训练该AI模型。
具体的，本实施例中训练设备基于第一训练数据训练该AI模型的过程可以参见方法500中的步骤505，不再赘述。
可选的，步骤608，训练设备向网络设备发送模型训练完成的指示信息。
训练设备完成AI模型训练之后，便可以进入模型推理阶段。
本实施例中，可以由网络设备结合多方面的信息综合决定是否支撑该AI模型的而训练，网络设备不会持续向训练设备下发训练数据集，可以减少不必要的空口资源的占用，节省空口开销，提高空口资源的使用性能。
上述方法500和方法600分别从训练设备的角度和网络设备的角度介绍了获取训练数据集的方法。
本申请中，还考虑到模型训练时具体如何对AI模型进行训练的问题。在一种可能的方案中，网络设备向训练设备发送训练数据集，训练设备基于网络设备发送的训练数据集获得输入信息和标签。在该方案中，在模型训练时将训练数据集全部输入到AI模型中进行训练，这种模型训练方法需要较长的训练时间才会使得模型达到收敛。下述方法700中进一步考虑到具体在训练该AI模型如何使得该AI模型可以快速收敛。图7是本申请提供的一种获取AI模型的输入信息的方法700的示意性流程图，方法700包括：
可选的，步骤701～步骤702，训练设备确定进行对该AI模型的性能进行监控，并且本实施例中依然假设训练设备确定该AI模型需要更新，则需要继续执行下述步骤703～步骤710。具体的，步骤701～步骤702的实现方式可以参见方法500中的步骤501～步骤502，不再赘述。
可选的，步骤703，训练设备向第一网络设备发送第一信息，该第一信息用于请求第一网络设备发送训练数据集。
对应的，第一网络设备接收来自训练设备的第一信息。
可选的，第一信息指示了训练设备请求第一网络设备发送的第一训练数据集的相关信息。例如，第一信息可以包括下至少一项：请求的训练数据集的大小的信息、AI模型的输入的配置信息、用于AI模型的训练的参考信号的配置信息、该AI模型的标识信息。可选的，第一信息还包括以下至少一项：该AI模型的标识信息、该AI模型的应用场景的信息、该AI模型的用途信息、该训练设备的算力能力的信息。换句话说，本实施例中还可以和方法500进行结合，即，训练设备可以向网络设备指示请求的训练数据集的大小。
可选的，步骤704，第一网络设备确定向训练设备发送的第一训练数据集的大小。
例如，第一网络设备接收到接收来自训练设备的第一信息后，可以从第二网络设备获取第三信息，第三信息为该训练该AI模型的相关信息。例如，第三信息可以包括以下至少一项：训练设备请求第二网络设备发送的训练数据集的大小的信息、完成AI模型训练需要的训练数据集的大小的信息、AI模型的标识信息、训练设备的算力能力的信息可选的，第三信息还包括：第二网络设备发送训练数据集的时长的信息和/或第二网络设备发送训练数据集的方式的信息。例如，第一网络设备可以根据第三信息，确定向训练设备发送的第一训练数据集，具体的实现方式可以参见上述方法600的描述。又例如，第一网络设备可以根据第一信息所指示的相关信息和第三信息，确定向训练设备发送的第一训练数据集。再例如，第一网络设备可以根据第一信息所指示的相关信息、第三信息以及第一网络设备的资源使用情况，确定向训练设备发送的第一训练数据集。换句话说，本实施例可以和方法600进行结合，即，网络设备可以综合确定向训练设备发送的第一训练数据的大小。
可选的，步骤705，第一网络设备向训练设备发送N(N为大于1的整数)组参考信号。
对应的，训练设备接收来自第一网络设备的N组参考信号。
本实施例中，“第一网络设备向训练设备发送N组参考信号”也可以理解为，第一网络设备向训练设备发送第一训练数据集。示例性的，在波束管理场景中，第一网络设备可以为训练设备配置参考信号资源，并向训练设备发送参考信号资源(例如，N组参考信号)。其中，N组参考信号中的每组参考信号包括至少一个参考信号，并且每组参考信号具有相同的组标识。每组参考信号对应的组标识也可以理解为波束识别号，例如，波束标识；或者，每组参考信号对应的组标识也可以理解为每组参考信号的资源的标识。换句话说，第一网络设备可以指示训练设备进行全码本波束扫描。即，第一网络设备指示训练设备测量N组参考信号。
步骤705之前，还可以包括：第一网络设备向训练设备发送配置信息，该配置信息可以用于指示以下中的一项或多项：该N组参考信号的时域资源、该N组参考信号的频域资源、该N组参考信号的传输周期以及该N组参考信号的组标识。例如，在一种可能的实现方式中，配置信息中可以包括该N组参考信号的时域资源的大小、该N组参考信号的频域资源的大小、该N组参考信号的传输周期的信息等等，此时，也可以理解为配置信息显示的进行指示。在另一种可能的实现方式中，配置信息可以携带N组参考信号的时域资源的索引、该N组参考信号的频域资源的索引、该N组参考信号的传输周期的索引等等，此时，也可以理解为配置信息隐式的进行指示。
步骤706，训练设备测量N组参考信号，获取该N组参考信号对应的N组测量结果。
例如，N组测量结果中的每组测量结果可以包括至少一个测量量。具体的，针对每组参考信号中每个参考信号都可以测量其RSRP、RSRQ、SINR等中的一项或多项。也就是说，每组测量结果中可以包含针对该组参考信号中的每个参考信号测量的RSRP、RSRQ、SINR的测量结果中的一项或多项。其中，每组测量结果对应相同的组标识。如下表格2所示，参考信号组#A中包含参考信号#A1和参考信号#A2，训练设备可以分别测量参考信号#A1和参考信号#A2，例如，测量参考信号#A1的RSRP、SINR和参考信号#A2的RSRP和SINR，则参考信号#A1测量结果中包含参考信号#A1的RSRP的测量结果以及SINR的测量结果，则参考信号#A2测量结果中包含参考信号#A2的RSRP的测量结果以及SINR的测量结果。
表格2
本实施例中，训练设备基于全码本的波束扫描结果可以确定训练标签。具体的，如果该AI模型训练采用回归法进行模型训练，例如，训练设备可确定将N组参考信号的RSRP的测量结果作为训练标签；如果该AI模型训练采用分类法进行模型训练，例如，训练设备可以确定将N组参考信号中信道质量最优的K组测量结果对应的波束标识作为训练标签。
步骤707，第一网络设备向训练设备发送第四信息，第四信息用于指示该N组参考信号中的M组参考信号。
示例性的，在波束管理场景中，例如，第四信息可以用于指示稀疏波束图样。也可以理解为，本实施例中，第一网络设备可向训练设备指示AI模型的输入信息。例如，第一网络设备可以向训练设备指示该AI模型的输入信息具体是全码本中的哪些波束组成的波束图样，假设全码本中有64个波束，则第四信息可以指示输入信息为64个波束中哪些波束组成的图样。具体的，可以有下述两种实现方式：
方式1
在一种可能实现方式中，稀疏波束图样中的波束标识(“波束标识”也可以理解为“参考信号组的组标识”)与全码本中波束的标识是统一的。例如，全码本中有64个波束，波束标识分别为波束#1～波束#64，而第一网络设备发送的稀疏波束图样中包含16个波束分别为波束#16～波束#32。例如，方式1中波束标识和参考信号组标识是一一对应的。网络设备向训练设备指示出波束标识(波束信息的一种示例)和/或参考信号组的组标识就可以指示出稀疏波束图样。
方式2
在另一种可能的实现方式中，假设，稀疏波束图样中的波束标识与全码本中波束的标识不统一，也可以理解为，稀疏波束图样中波束的标识与全码本中波束的标识不是对应的。例如，全码本中有64个波束，波束标识分别为波束#1～波束#64，而第一网络设备发送的稀疏波束图样中包含16个波束分别为波束#1～波束#16，而该稀疏波束图样中的波束#1～波束#16，训练设备不能解析为全码本中的波束#1～波束#16，例如，稀疏波束图样中包含16个波束分别为波束#1～波束#16其实应该对应全码本中波束#1、 波束#4、波束#8、波束#12、波束#16、波束#20、波束#24、波束#28、波束#32、波束#36、波束#40、波束#44、波束#48、波束#52、波束#56、波束#60。此时，该场景下可以采用如下方案指示稀疏波束图样：例如，第四信息包括N个字段，该N个字段与该N组参考信号一一对应，其中，N个字段中的M个字段的比特值与剩余的(N-M)个字段的比特值不同，第一网络设备可以通过M个字段指示出M组参考信号。示例性的，M个字段的比特值均为“1”，剩余的(N-M)个字段的比特值均为“0”，此时训练设备便可以确定输入信息为哪些参考信号组对应的测量结果。因此，可以网络设备可以通过字段具体指示出稀疏波束图样为全码本中的那些波束。
在一种可能的实现方式中，第一网络设备还可以向训练设备发送第五信息，第五信息用于指示该N组参考信号中的P组参考信号。也可以理解为，第一网络设备可以向训练设备指示另外一个波束图样对应的参考信号的测量结果作为AI模型的输入信息。换句话说，本申请中第一网络设备可以向训练设备指示多个波束图样对应的参考信号的测量结果分别作为AI模型的输入信息，进行模型训练，从而实现训练得到的AI模型可以对所有波束图样都达到收敛。
可选的，步骤705和步骤707可以没有先后顺序。例如，配置信息和第四信息也可以在同一条消息中进行发送，不予限定。
步骤708，训练设备根据第四信息和该N组参考信号对应的N组测量结果，确定该AI模型的第一输入信息。
由于，步骤705中训练设备已经测量N组参考信号并获得N组参考信号对应的N组测量结果，因此，训练设备可以基于第四信息确定该N组测量结果中的哪些组的测量结果可以作为该AI模型的输入信息。
假设，训练设备测量了64组参考信号，第一网络设备指示其中第二组参考信号对应的测量结果、第四组参考信号对应的测量结果、第八组参考信号对应的测量结果、第十六组参考信号对应的测量结果作为AI模型的输入，则训练设备可以将第二组参考信号的测量结果、第四组参考信号的测量结果以及第八组参考信号的测量结果作为AI模型的输入信息。例如，第一输入信息可以为M组参考信号的RSRP的测量结果。
步骤709，训练设备基于第一输入信息，获取第一输出信息。
本实施例中该“第一输出信息”也可以理解，该AI模型的推理输出结果，如果该AI模型训练采用回归法进行模型训练，例如，第一输入信息可以为M组参考信号的RSRP的测量结果，此时，第一输出信息可以包括N组参考信号的RSRP的测量结果对应的N个组标识；如果该AI模型训练采用分类法进行模型训练，例如，第一输入信息可以为M组参考信号对应的测量结果，此时，第一输出信息可以为N组参考信号的信道质量测量结果最优的K组参考信号对应的K个组标识，例如，可以理解为K个波束标识。
训练设备可以将第一输出信息与标签进行比较获得AI模型的训练损失，例如，以模型训练的分类法为例，第一输出信息为训练设备推测的N组参考信号的信道质量测量结果最优的K组参考信号对应的K个组标识，而训练设备确定的训练标签假设为全码本扫描时N组参考信号的信道质量测量结果最优的K组参考信号对应的K个组标识，此时，训练设备可以比较输出结果与训练标签，确定该AI模型的性能，并且调整模型参数，以上过程可以理解为AI模型的一次训练。训练设备可以基于该次模型训练的训练损失、训练准确度衡量该AI模型的性能，并确定下一次模型训练需要的训练数据集。例如，该训练数据集的大小可由训练设备基于模型性能的评估结果决定，例如，如果训练设备确定该AI模型性能经过第一轮训练之后模型性能显著变好，则可以减少训练数据集的数据量。具体的，还包括步骤：
训练设备可以根据第一训练数据集训练该AI模型，并确定该AI模型的性能；训练设备可以根据AI模型的性能，向网络设备发送第二信息，该第二信息用于指示训练设备请求网络设备发送的第二训练数据集的相关信息；训练设备接收来自所网络设备的第二训练数据集，该第二训练数据集是基于第二信息所指示的相关信息的训练数据集，该第二训练数据集用于AI模型的训练。例如，第二训练数据集的数据量可以小于第一训练数据集的数据量。后续，训练设备仍然可以基于第二训练数据集对该AI进行训练，并反复迭代，假设训练设备进行Q(Q为大于1的整数)次训练，直到训练设备确定该AI模型收敛(“模型收敛”也可以理解为该AI模型达到目标性能)。
需要说明的是，本实施例中，训练设备在后续基于请求的训练数据集进行模型训练时，AI 模型的稀疏波束图样可以是固定的，或者，也可以理解为，在后续Q次的训练过程中，稀疏波束的图样仍然为步骤707中第四信息指示的波束图样。应理解，每次训练时，训练设备都会基于第一网络设备下发的训练数据集进行一次全码本扫描，由于信道状态(也可以理解为，信道环境)是时变的，每次全码本扫描后获得的参考信号的测量结果也是不完全相同的。因此，每次训练时，N组参考信号中的M组参考信号对应测量结果也是不同，并且训练设备确定的训练标签也不相同，即AI模型的输入信息和训练标签都会相应发生变化，然而这些变化本质上都是由于信道状态变化引起的，波束图样并未变化。即，本实施例提供的方案中，AI模型训练过程中的变量仅仅为信道状态。
另一方案中，在AI模型训练时，AI模型的输入信息是全码本扫描后获得的所有参考信号的测量结果，即，每次训练时波束图样和信道状态均在发生变化，AI模型训练时模型收敛性能较差。而按照本申请提出的方法700，在模型训练时，网络设备可以指示全码本扫描中获得的测量结果中的哪些测量结果为AI模型的输入信息，即，只有信道状态发生变化，与前述另一方案相比可以加速AI模型的收敛速度，提高模型训练效率，从而也可以减少空口资源的占用。
可选的，步骤710，训练设备向第一网络设备发送模型训练完成指示信息。
训练设备完成AI模型训练之后，便可以进入模型推理阶段。例如，后续第一网络设备可以向训练设备发送稀疏波束图样以及对应的参考信号，训练设备通过测量该参考信号，获得AI模型的输入信息。假设，AI模型采用分类法，则AI模型的输入为参考信号的测量结果，此时AI模型经过推理可以输出的K个波束标识，该K波束标识为训练设备推测的全码本中参考信号测量结果中信道质量最好的K个测量结果对应的波束。训练设备可以将该K波束标识反馈给第一网络设备，第一网络设备再次向训练设备发送该K个波束对应的K组参考信号，训练设备再次测量该K组参考信号，并且确定测量结果最优的其中一组参考信号，并将该组参考信号对应的波束标识作为最终选择的波束与第一网络设备进行通信。
基于上述技术方案，本申请中，网络设备可以向训练设备指示AI模型的输入信息，使得在AI模型训练过程中只保持信道状态发生变化，从而可以加速AI模型收敛速度，提高模型训练效率，从而也可以减少空口资源的占用。
上述方法700提供了一种获取AI模型输入信息的方法，下面方法800中提供了一种通信方法，该方法更为详细的说明了在训练阶段获取AI模型输入信息的方法，该方法可以与前述输入信息的获取方法独立实施，也可以结合应用。如图8所示，该方法800包括：
步骤801，网络设备向训练设备发送第二参考信号集。
对应的，训练设备接收来自网络设备的第二参考信号集。
本实施例中，该第二参考信号集包括N组参考信号，N组参考信号中的每组参考信号包括至少一个参考信号。“第二参考信号集”也可以理解为全码本波束对应的参考信号集合，例如，全码本波束中的每个波束都可以对应第二参考信号集合中的其中一组参考信号。
在一种可能的实现方式中，步骤801之前，该方法还可以包括：网络设备向训练设备发送第一配置信息，该第一配置信息可以用于指示以下中的一项或多项：该N组参考信号的时域资源、该N组参考信号的频域资源、该N组参考信号的传输周期、该N组参考信号的组标识，或，该N组参考信号的波束。
其中，每组参考信号具有组标识。
需要说明的，本申请中，如果N组参考信号中每组参考中只包含一个参考信号，则该组参考信号的组标识为该参考信号的标识，也即，组标识可以被替换为该参考信号的标识。
在另一种可能的实现方式中，步骤801之前，该方法还可以包括：网络设备向训练设备发送第一配置信息，该第一配置信息包括N组参考信号各自的组标识，并且第三配置信息可以用于指示以下中的一项或多项：N组参考信号的时域资源，N组参考信号的频域资源，N组参考信号的传输周期，或N组参考信号的波束。
在又一种可能的实现方式中，步骤801之前，该方法还可以包括：网络设备向训练设备发送第一配置信息，该第一配置信息包括N组参考信号各自的波束信息，并且第三配置信息可以用于指示以下中的一项或多项：N组参考信号的时域资源，N组参考信号的频域资源，N组参考信号的传输周期， 或N组参考信号的组标识。
步骤802，网络设备向训练设备发送第二波束指示信息，第二波束指示信息指示第一参考信号集对应的波束。
对应的，训练设备接收来自网络设备的第二波束指示信息。
本实施例中，第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，第一参考信号集包括M组参考信号，其中，N为大于M的整数，M为大于或等于1的整数。
网络设备向训练设备发送第二波束指示信息也可以理解为，网络设备向训练设备指示稀疏波束图样。换句话说，第二波束指示信息可以向训练设备指示训练设备需要扫描的稀疏波束图样为全码本中的哪些波束组成的图样。
本实施例中，第二波束指示信息指示第一参考信号集对应的波束，例如可以是，第二波束指示信息指示第一参考信号集对应的波束在第二参考信号集对应的多个波束中的位置。例如，全码本中有64个波束，波束标识分别为波束#1～波束#64，而第一网络设备发送的稀疏波束图样中包含16个波束。此时，第二波束信息可以指示出该16个波束到底是全码本波束中的哪些波束(即，第一参考信号集对应的是全码本中的哪些波束)。具体的，可以有下述两种实现方式：
方式A
第二波束信息包括N个字段，该N个字段与该第二参考信号集对应的多个波束一一对应，其中，N个字段中的M个字段的比特值与剩余的(N-M)个字段的比特值不同，网络设备可以通过M个字段指示出第一参考信号集。示例性的，M个字段的比特值均为“1”，剩余的(N-M)个字段的比特值均为“0”。例如，全码本中有64个波束，波束标识分别为波束#1～波束#64，而网络设备发送的稀疏波束图样中包含16个波束分别为波束#1～波束#16，第二波束指示信息中的字段#1～字段#16分别指示全码本中的波束#1～波束16。例如，第二波束指示信息中的字段#1～字段#16分别指示第一参考信号集中的参考信号组#1～参考信号组#16。方式A也可以理解为，网络设备可以直接指示波束位置。
方式B
第二波束指示信息包括M组参考信号的组标识或波束信息(例如，波束标识)，其中，M组参考信号为N组参考信号中的部分，且N组参考信号与N个波束具有预先定义或预先配置的对应关系。例如，N组参考信号与N个波束一一对应。例如，N组参考信号的组标识与N个波束标识一一对应。示例性的，全码本中有64个波束，波束标识分别为波束#1～波束#64，而第一网络设备发送的稀疏波束图样中包含16个波束分别为波束#16～波束#32。例如，方式B中波束标识和参考信号的组标识是一一对应的。网络设备向训练设备指示出波束标识(波束信息的一种示例)和/或参考信号的组标识就可以指示出稀疏波束图样。与方式A相比，方式B中，通过参考信号的组标识或波束信息间接指示波束位置，由于波束信息与参考信号的组标识对应，并且，M组参考信号的组标识/波束信息与N组参考信号的组标识/波束信息的关系是固定的。
步骤803，训练设备确定AI模型的第一输入信息。
例如，训练设备接收到第二参考信号集后，可以测量第二参考信号集中的N组参考信号，并获得N组参考信号的测量结果。此时，训练设备可以根据第二波束指示信息和对第二参考信号集中N组参考信号的测量结果确定该第二波束指示信息所指示的波束对应的测量结果，并将该AI模型的第二波束指示信息所指示的波束对应的测量结果第一输入信息。也即，将稀疏波束图样对应的波束的测量结果作为该第一输入信息。
对应于步骤802中的两种方式，训练设备确定稀疏波束图样后，便可以确定AI模型的输入信息是全码本中哪些波束位置的参考信号对应的测量结果。例如，上述方式A中，训练设备可以将全码本中的波束#16～波束#32对应的参考信号组#16～参考信号组#32的测量结果作为AI模型的输入信息。又例如，上述方式B中训练设备可以将参考信号组#1、参考信号组#4、参考信号组#8、参考信号组#12、参考信号组#16、参考信号组#20、参考信号组#24、参考信号组#28、参考信号组#32、参考信号组#36、参考信号组#40、参考信号组#44、参考信号组#48、参考信号组#52、参考信号组#56、参考信号组#60的测量结果作为AI模型的输入信息。
可选的，步骤804，训练设备基于确定的第一输入信息和AI模型进行模型训练，获得第一输出信息。
本申请中，第一输出信息指示第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，其中，K为大于或等于1的整数，且K小于N。
示例性的，如前所述，按照训练AI模型的不同的算法实现可以采用分类法和回归法，例如，采用分类法训练AI模型时，第一输出信息可以包括第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束的信息。或者，第一输出信息可以包括K组参考信号各自的组标识或波束信息，其中，K组参考信号对应于N组参考信号对应的N个测量结果中被预测为信道质量最好的K个测量结果，且K组参考信号各自的组标识与K个波束，也即，K个波束信息，具有预先定义或预先配置的对应关系。例如，采用回归法训练AI模型时，第一输出信息可以包括N组参考信号对应的多个波束信息及预测的该波束信息对应的N个测量结果。或者，第一输出信息可以包括N组参考信号各自的组标识及预测的N组参考信号的N个测量结果，且N组参考信号与N个波束信息具有预先定义或预先配置的对应关系。
本实施例中该“第一输出信息”也可以理解，该AI模型的训练输出结果，如果该AI模型训练采用回归法进行模型训练，例如，第一输入信息可以为M组参考信号的RSRP的测量结果，此时，第一输出信息可以包括预测的N个波束对应的参考信号的RSRP的测量结果及其各自对应的N个组标识；如果该AI模型训练采用分类法进行模型训练，例如，第一输入信息可以为M个波束对应的M组参考信号的测量结果，此时，第一输出信息可以为预测的N个波束的信道质量测量结果最优的K个波束的波束标识或K个波束对应的K个组标识。
训练设备可以将第一输出信息与标签进行比较获得AI模型的训练损失，例如，以模型训练的分类法为例，第一输出信息为训练设备推测的N个波束的信道质量测量结果最优的K个波束对应的K个组标识，而训练设备确定的训练标签假设为全码本扫描时N个波束对应的N组参考信号的信道质量测量结果最优的K组参考信号对应的K个组标识，此时，训练设备可以比较输出结果与训练标签，确定该AI模型的性能，并且调整模型参数，以上过程可以理解为AI模型的一次训练。训练设备可以基于该次模型训练的训练损失、训练准确度衡量该AI模型的性能，并且反复迭代，直到模型收敛。
基于上述技术方案，网络设备可以向训练设备指示AI模型的输入信息，使得针对同一个稀疏波束图样，在AI模型训练过程中信道状态发生变化，从而可以加速AI模型收敛速度，提高模型训练效率，从而也可以减少空口资源的占用。
应理解，针对多个稀疏波束图样，则重复上述描述的步骤即可。
上述方法700、800均主要介绍了在模型训练阶段的获取输入信息的具体实现方案。下述图9中的方法900主要介绍了在模型推理阶段也可以使用该方法。方法900中假设AI模型已经完成训练，例如，可以采用方法700、方法800中的方法进行模型训练；又例如，可以采用现有的方案对AI模型进行训练。例如，推理设备，如终端设备，上可以直接预配置多个训练好的AI模型，例如，每个AI模型是通过一种或者多种稀疏波束图样训练完成了AI模型。因此，可以直接接入模型推理阶段。可以理解的是，图9所描述的方法可以与图8中的独立应用，也可以与图8中的方法结合应用。结合应用时，各自所涉及的相同用语，以X代表，分别为训练过程的配置信息以及推理过程的配置信息，以便区分。比如针对配置信息，分别为训练过程的配置信息以及推理过程的配置信息。
步骤901，网络设备向终端设备发送第一参考信号集。
对应的，终端设备接收来自网络设备的第一参考信号集。
本实施例中，第一参考信号集包括M组参考信号，M组参考信号中的每组参考信号包括至少一个参考信号，其中，M为大于或等于1的整数。“第一参考信号集”也可以理解为稀疏波束图样对应的参考信号集合，例如，该稀疏波束图样为全码本波束的子集，或者，该稀疏波束图样对应的参考信号集合属于全码本波束对应第二参考信号集合中的子集(也即，第一参考信号集为第二参考信号集的子集)，其中，第二参考信号集包括N组参考信号，N组参考信号中的每组参考信号包括至少一个参考信号。
在一种可能的实现方式中，步骤901之前，所述方法还可以包括：网络设备向终端设备发送第一配置信息，该第一配置信息可以用于指示以下中的一项或多项：该N组参考信号的时域资源、该N组参考信号的频域资源、该N组参考信号的传输周期、该N组参考信号的组标识，或，该N组参考信号的波束信息。
在另一种可能的实现方式中，步骤901之前，所述方法还可以包括：网络设备向终端设备发送第二配置信息，第二配置信息包括第一波束信息，该第一波束信息包括M组参考信号的组标识，第二配 置信息还可以包括M组参考信号的时域资源、M组参考信号频域资源、M组参考信号传输周期、或波束信息中的一项或多项。
在一种可能的实现方式中，步骤901之前，所述方法还可以包括：网络设备向终端设备发送第二配置信息，第二配置信息包括第一波束信息，该第一波束信息包括M组参考信号的波束信息，第二配置信息还可以包括M组参考信号的时域资源、M组参考信号频域资源、M组参考信号传输周期、或组标识中的一项或多项。
在又一种可能的实现方式中，步骤901之前，所述方法还可以包括：网络设备向终端设备发送第三配置信息，该第三配置信息包括N组参考信号各自的组标识，并且第三配置信息可以用于指示以下中的一项或多项：N组参考信号的时域资源，N组参考信号的频域资源，N组参考信号的传输周期，或N组参考信号的波束。其中，N组参考信号的N个组标识包括所述M组参考信号的M个组标识。
在又一种可能的实现方式中，步骤901之前，所述方法还可以包括：网络设备向终端设备发送第三配置信息，该第三配置信息包括N组参考信号各自的波束信息，并且第三配置信息可以用于指示以下中的一项或多项：N组参考信号的时域资源，N组参考信号的频域资源，N组参考信号的传输周期，或N组参考信号的组标识。其中，N组参考信号的N个波束信息包括所述M组参考信号的M个波束信息。
步骤902，网络设备向终端设备发送第一波束指示信息，第一波束指示信息指示第一参考信号集对应的波束。
对应的，终端设备接收来自网络设备的第一波束指示信息。
本实施例中，第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集。
网络设备向终端设备发送第一波束指示信息也可以理解为，网络设备向终端设备指示稀疏波束图样。换句话说，第一波束指示信息可以向终端设备指示终端设备需要扫描的稀疏波束图样为全码本中的哪些波束组成的图样。
本实施例中，第二波束指示信息指示第一参考信号集对应的波束，例如可以是，第二波束指示信息指示第一参考信号集对应的波束在第二参考信号集对应的多个波束中的位置。例如，全码本中有64个波束，波束标识分别为波束#1～波束#64，而第一网络设备发送的稀疏波束图样中包含16个波束。此时，第二波束信息可以指示出该16个波束到底是全码本波束中的哪些波束(即，第一参考信号集对应的是全码本中的哪些波束)。具体的，可以有下述两种实现方式：
方式A
第一波束信息包括N个字段，该N个字段与该第二参考信号集对应的多个波束一一对应，其中，N个字段中的M个字段的比特值与剩余的(N-M)个字段的比特值不同，网络设备可以通过M个字段指示出第一参考信号集。示例性的，M个字段的比特值均为“1”，剩余的(N-M)个字段的比特值均为“0”。例如，全码本中有64个波束，波束标识分别为波束#1～波束#64，而网络设备发送的稀疏波束图样中包含16个波束分别为波束#1～波束#16，第二波束指示信息中的字段#1～字段#16分别指示全码本中的波束#1～波束16。例如，第二波束指示信息中的字段#1～字段#16分别指示第一参考信号集中的参考信号组#1～参考信号组#16。方式A也可以理解为，网络设备可以直接指示波束位置。
方式B
第一波束指示信息包括M组参考信号的组标识或波束信息(例如，波束标识)，其中，M组参考信号为N组参考信号中的部分，且N组参考信号与N个波束具有预先定义或预先配置的对应关系。例如，N组参考信号与N个波束一一对应。例如，N组参考信号的组标识与N个波束标识一一对应。示例性的，全码本中有64个波束，波束标识分别为波束#1～波束#64，而第一网络设备发送的稀疏波束图样中包含16个波束分别为波束#16～波束#32。例如，方式B中波束标识和参考信号的组标识是一一对应的。网络设备向终端设备指示出波束标识(波束信息的一种示例)和/或参考信号组的组标识就可以指示出稀疏波束图样。与方式A相比方式B中是通过参考信号的组标识或波束信息间接指示波束位置，由于波束信息与参考信号的组标识对应，并且，M组参考信号组标识/波束信息与N组参考信号标识/波束信息的关系可以是固定的。
可选的，步骤903，终端设备根据接收的第一波束指示信息，也即，稀疏波束图样，确定AI模型。
如果假设终端设备上预配置了多个AI模型，则终端设备可以基于稀疏波束图样确定AI模型。例 如，终端设备可以对接收的稀疏波束图样进行识别，通过对波束图样的识别，可以在本地已经预配置的多个AI模型中确定与该稀疏波束图样对应的一个AI模型。也可以理解为，终端设备可以基于稀疏波束图样在多个AI模型中确定一个匹配性最好的AI模型。
本申请中，终端设备识别稀疏波束图样可以理解为，终端设备需要确定接收的波束图样是全码本波束中哪些位置的波束。换句话说，终端设备需要将接收的稀疏波束图样与全码本中的波束建立联系，即，需要确定出稀疏波束图样中的波束具体是全码本中的哪些波束。基于网络设备和终端设备的配置，终端设备可以基于方式A和/或方式B确定稀疏波束图样。
步骤904，终端设备确定AI模型的第一输入信息。
例如，终端设备接收到第一参考信号集后，可以测量第一参考信号集中的M组参考信号，并获得M组参考信号的测量结果。此时，终端设备可以根据M组参考信号的测量结果确定AI模型的第一输入信息。
对应于步骤902中的两种方式，终端设备确定稀疏波束图样后，便可以确定AI模型的输入信息是全码本中哪些波束位置的参考信号对应的测量结果。例如，上述方式A中，终端设备可以将全码本中的波束#16～波束#32对应的参考信号组#16～参考信号组#32的测量结果作为AI模型的输入信息。又例如，上述方式B中终端设备可以将参考信号组#1、参考信号组#4、参考信号组#8、参考信号组#12、参考信号组#16、参考信号组#20、参考信号组#24、参考信号组#28、参考信号组#32、参考信号组#36、参考信号组#40、参考信号组#44、参考信号组#48、参考信号组#52、参考信号组#56、参考信号组#60的测量结果作为AI模型的输入信息。
可选的，步骤905，终端设备基于确定的第一输入信息和AI模型进行模型推理，获得第一输出信息。
本申请中，第一输出信息指示第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，其中，K为大于或等于1的整数，且K小于N。
本步骤可以参考图8中步骤804中训练过程所涉及的推理过程中第一输出信息的描述，在此不予赘述。本步骤与步骤804的不同在于，本步骤无需将第一输出信息和标签进行损失获得，第一输出信息为可用的预测结果。
可选的，步骤906，终端设备将第一输出信息发送给网络设备。
后续，网络设备可以再次向终端设备发送该第一输出信息对应的参考信号，终端设备再次测量参考信号，并且确定测量结果最优的参考信号，并将该参考信号对应的波束标识作为最终选择的波束与网络设备进行通信。
应理解，本申请中，测量结果最优，可以包括RSRP值最大，或是，SINR值最大。也可以是其他评价标准，在此不予限定。
示例性的，假设，AI模型采用分类法，此时AI模型经过推理可以输出的K个波束标识，该K波束标识为终端设备推测的全码本中参考信号测量结果中信道质量最好的K个测量结果对应的波束。终端设备可以将该K波束标识反馈给第一网络设备，第一网络设备再次向终端设备发送该K个波束对应的K组参考信号，终端设备再次测量该K组参考信号，并且确定测量结果最优的其中一组参考信号，并将该组参考信号对应的波束标识作为最终选择的波束与第一网络设备进行通信。
基于上述技术方案，本申请中，终端设备可以识别出稀疏波束图样，并且进一步确定AI模型的输入信息，使得模型推理结果更为准确。
可以理解，本申请实施例中的方法500～方法900中的例子仅仅是为了便于本领域技术人员理解本申请实施例，并非要将本申请实施例限于例示的具体场景。本领域技术人员根据方法500～方法900中的例子，显然可以进行各种等价的修改或变化，这样的修改或变化也落入本申请实施例的范围内。
还可以理解，本申请的各实施例中的一些可选的特征，在某些场景下，可以不依赖于其他特征，也可以在某些场景下，与其它特征进行结合，不作限定。
还可以理解，本申请中描述的各个实施例可以为独立的方案，也可以根据内在逻辑进行组合，这些方案都落入本申请的保护范围中。并且实施例中出现的各个术语的解释或说明可以在各个实施例中互相参考或解释，对此不作限定。
应该理解，本申请中的预定义可以理解为定义、预先定义、存储、预存储、预协商、预配置、固 化、或预烧制。
可以理解，在本申请中，“若”以及“如果”均指在某种客观情况下装置会做出相应的处理，并非是限定时间，且也不要求装置实现时一定要有判断的动作，也不意味着存在其它限定。
可以理解，本文中术语“和/或”，描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B的情况，其中A,B可以是单数或者复数。在本申请的文字描述中，字符“/”，一般表示前后关联对象是一种“或”的关系；在本申请的公式中，字符“/”，表示前后关联对象是一种“相除”的关系。
上述主要从各个节点之间交互的角度对本申请实施例提供的方案进行了介绍。可以理解的是，各个节点，例如训练设备、网络设备，为了实现上述功能，其包含了执行各个功能相应的硬件结构和/或软件模块。本领域技术人员应该可以意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，本申请能够以硬件或硬件和计算机软件的结合形式来实现。某个功能究竟以硬件还是计算机软件驱动硬件的方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
可以理解的是，为了实现上述实施例中功能，网络设备和训练设备包括了执行各个功能相应的硬件结构和/或软件模块。本领域技术人员应该很容易意识到，结合本申请中所公开的实施例描述的各示例的单元及方法步骤，本申请能够以硬件或硬件和计算机软件相结合的形式来实现。某个功能究竟以硬件还是计算机软件驱动硬件的方式来执行，取决于技术方案的特定应用场景和设计约束条件。
图10和图11为本申请的实施例提供的可能的通信装置的结构示意图。这些通信装置可以用于实现上述方法实施例中训练设备或网络设备的功能，因此也能实现上述方法实施例所具备的有益效果。在本申请的实施例中，该通信装置可以是如图1所示的终端设备(训练设备的一个示例，或者推理设备的一个示例)120a-120j中的一个，也可以是如图1所示的网络设备110a或110b，还可以是应用于终端设备或网络设备的模块(如芯片)。
如图10所示，通信装置100包括处理单元120和收发单元110。通信装置100用于实现上述图5～图9中所示的方法实施例中训练设备或网络设备的功能。
当通信装置100用于实现图5所示的方法实施例中训练设备的功能时：收发单元110用于发送第一信息，所述第一信息用于指示该装置请求发送的第一训练数据集的相关信息；收发单元110还用于接收来所述第一训练数据集，所述第一训练数据集是基于所述第一信息所指示的所述相关信息的训练数据集，所述第一训练数据集用于人工智能AI模型的训练。
在一种可能的实现方式中，处理单元120用于根据所述第一训练数据集训练所述AI模型，并确定所述AI模型的性能；处理单元120还用于根据所述AI模型的性能，控制收发单元110发送第二信息，所述第二信息用于指示所述装置请求发送的第二训练数据集的相关信息；收发单元110用于接收所述第二训练数据集，所述第二训练数据集是基于所述第二信息所指示的所述相关信息的训练数据集，所述第二训练数据集用于所述AI模型的训练。
当通信装置100用于实现图5所示的方法实施例中第一网络设备的功能时：收发单元110用于接收第一信息，所述第一信息用于指示该装置请求发送的第一训练数据集的相关信息；处理单元120用于根据所述第一信息所指示的所述相关信息，控制收发单元110发送所述第一训练数据集，所述第一训练数据集用于所述人工智能AI模型的训练。
在一种可能的实现方式中，收发单元110用于获取第三信息，所述第三信息为训练所述AI模型的相关信息，所述处理单元120用于根据所述第一信息所指示的所述相关信息，控制收发单元110发送所述第一训练数据集，包括：所述处理单元120用于根据所述第一信息所指示的所述相关信息和所述第三信息，控制收发单元110发送所述第一训练数据集。
在一种可能的实现方式中，收发单元110用于接收第二信息，所述第二信息用于指示请求发送的第二训练数据集的相关信息，其中，所述第二信息是基于所述AI模型的性能确定的，所述AI模型性能是基于所述第一训练数据集训练确定的；处理单元120用于根据所述第二信息所指示的所述相关信息，控制收发单元110发送所述第二训练数据集，所述第二训练数据集用于所述AI模型的训练。
当通信装置100用于实现图6所示的方法实施例中第一网络设备的功能时：收发单元110用于获取第三信息，所述第三信息为训练人工智能AI模型的相关信息；收发单元110用于接收第一信息，所 述第一信息用于请求发送训练数据集；处理单元120用于根据所述第三信息，确定待发送的第一训练数据集；处理单元120用于基于所述第一信息，控制收发单元110发送所述第一训练数据集，所述第一训练数据集用于所述AI模型的训练。
在一种可能的实现方式中，所述第三信息还包括：发送训练数据集的时长的信息和/或发送训练数据集的方式的信息，处理单元120用于根据所述第三信息和所述装置的资源使用情况，确定所述装置和/或训练设备是否具备支持训练所述AI模型的能力。
在一种可能的实现方式中，处理单元120用于根据所述第三信息，确定待发送的第一训练数据集，包括：所述处理单元120用于根据所述第一信息和所述第三信息，确定待发送的第一训练数据集，其中，所述第一信息用于指示请求所述第一网络设备发送的第一训练数据集的相关信息。
在一种可能的实现方式中，收发单元110用于接收第二信息，所述第二信息用于指示请求所述装置发送的第二训练数据集的相关信息，其中，所述第二信息是基于所述AI模型的性能确定的，所述AI模型的性能是基于所述第一训练数据集训练确定的；所述处理单元120用于根据所述第二信息，确定待发送的第二训练数据集。
当通信装置100用于实现图7所示的方法实施例中训练设备的功能时，处理单元120用于测量N组参考信号，获取所述N组参考信号对应的N组测量结果，其中，所述N组参考信号中的每组参考信号包括至少一个参考信号，所述每组参考信号具有相同的组标识，所述N为大于1的整数；收发单元110用于接收第四信息，所述第四信息用于指示所述N组参考信号中的M组参考信号；处理单元110用于根据所述第四信息和所述N组参考信号对应的N组测量结果，确定人工智能AI模型的第一输入信息，所述第一输入信息包括所述M组参考信号对应的M组测量结果；所述AI模型用于基于所述第一输入信息，获取第一输出信息，其中，所述第一输出信息包括所述N组参考信号中的K组参考信号各自的组标识，其中，所述K组参考信号各自的组标识对应于所述N组测量结果中信道质量最好的K组测量结果。
在一种可能的实现方式中，收发单元110用于接收配置信息，所述配置信息用于指示以下中的一项或多项：所述N组参考信号的时域资源、所述N组参考信号的频域资源、所述N组参考信号的传输周期、所述N组参考信号的组标识。
当通信装置100用于实现图7所示的方法实施例中网络设备的功能时，收发单元110用于向训练设备发送N组参考信号，其中，所述N组参考信号中的每组参考信号包括至少一个参考信号，所述每组参考信号具有相同的组标识，所述N为大于1的整数；收发单元110用于向所述训练设备发送第四信息，所述第四信息用于指示所述N组参考信号中的M组参考信号，其中，所述M组参考信号用于确定第一输入信息；所述AI模型用于基于所述第一输入信息，获取第一输出信息，其中，所述第一输出信息包括所述N组参考信号中的K组参考信号各自的组标识，其中，所述K组参考信号各自的组标识对应于所述N组参量信号对应的N组测量结果中信道质量最好的K组测量结果。
在一种可能的实现方式中，收发单元110用于发送配置信息，所述配置信息用于指示以下中的一项或多项：所述N组参考信号的时域资源、所述N组参考信号的频域资源、所述N组参考信号的传输周期、所述N组参考信号的组标识。
当通信装置100用于实现图8所示的方法实施例中训练设备的功能时，收发单元110用于接收第二参考信号集，其中，所述第二参考信号集包括N组参考信号，所述N组参考信号中的每组参考信号包括至少一个参考信号，所述N为大于1的整数；收发单元110还用于接收第二波束指示信息，所述第二波束指示信息指示第一参考信号集对应的波束，其中，所述第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，所述第一参考信号集对应的波束用于所述训练设备中的AI模型的第一输入信息的确定，所述第一输入信息基于所述第一参考信号集对应的波束的测量结果，所述第一参考信号集包括M组参考信号，所述N为大于M的整数，所述M为大于或等于1的整数。
在一种可能的实现方式中，收发单元110还用于发送第一配置信息。
在一种可能的实现方式中，收发单元110还用于发送第三配置信息。
在一种可能的实现方式中，处理单元120还用于对所述N组参考信号进行测量，获得N个测量结果。
当通信装置100用于实现图8所示的方法实施例中网络设备的功能时，收发单元110用于接收第 二参考信号集，其中，所述第二参考信号集包括N组参考信号，所述N组参考信号中的每组参考信号包括至少一个参考信号，所述N为大于1的整数；收发单元110还用于接收第二波束指示信息，所述第二波束指示信息指示第一参考信号集对应的波束，其中，所述第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，所述第一参考信号集对应的波束用于所述训练设备中的AI模型的第一输入信息的确定，所述第一输入信息基于所述第一参考信号集对应的波束的测量结果，所述第一参考信号集包括M组参考信号，所述N为大于M的整数，所述M为大于或等于1的整数。
在一种可能的实现方式中，收发单元110用于发送第一配置信息。
在一种可能的实现方式中，收发单元110用于发送第三配置信息。
当通信装置100用于实现图9所示的方法实施例中终端设备的功能时，收发单元110用于接收第一参考信号集，其中，所述第一参考信号集包括M组参考信号，所述M组参考信号中的每组参考信号包括至少一个参考信号，所述M为大于或等于1的整数；收发单元110还用于接收第一波束指示信息，所述第一波束指示信息指示所述第一参考信号集对应的波束，其中，所述第一参考信号集用于所述AI模型的第一输入信息的确定，所述第一输入信息基于所述第一参考信号集所包括的所述M组参考信号的测量结果，所述第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，所述第二参考信号集包括N组参考信号，所述N为大于或等于M的整数。
在一种可能的实现方式中，收发单元110用于接收第一配置信息。
在一种可能的实现方式中，收发单元110用于接收第二配置信息。
在一种可能的实现方式中，收发单元110用于接收第三配置信息。
在一种可能的实现方式中，处理单元120用于基于第一输入信息获取第一输出信息，收发单元110用于发送第一输出信息。
当通信装置100用于实现图9所示的方法实施例中网络设备的功能时，收发单元110用于发送第一参考信号集，其中，所述第一参考信号集包括M组参考信号，所述M组参考信号中的每组参考信号包括至少一个参考信号，所述M为大于或等于1的整数；收发单元110用于发送第一波束指示信息，所述第一波束指示信息指示所述第一参考信号集对应的波束，其中，所述第一参考信号集用于所述AI模型的第一输入信息的确定，所述第一输入信息基于所述第一参考信号集所包括的所述M组参考信号的测量结果，所述第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，所述第二参考信号集包括N组参考信号，所述N为大于或等于M的整数。
在一种可能的实现方式中，收发单元110用于发送第一配置信息。
在一种可能的实现方式中，收发单元110用于发送第二配置信息。
在一种可能的实现方式中，收发单元110用于发送第三配置信息。
在一种可能的实现方式中，收发单元110用于接收第一输出信息。
有关上述处理单元110和收发单元120更详细的描述可以直接参考图5～图9所示的方法实施例中相关描述直接得到，这里不加赘述。
如图11所示，通信装置200包括处理器210和接口电路220。处理器210和接口电路220之间相互耦合。可以理解的是，接口电路220可以为收发器或输入输出接口。可选的，通信装置200还可以包括存储器230，用于存储处理器210执行的指令或存储处理器210运行指令所需要的输入数据或存储处理器210运行指令后产生的数据。
当通信装置200用于实现图5所示的方法时，处理器210用于实现上述处理单元120的功能，接口电路220用于实现上述收发单元110的功能。
当通信装置200用于实现图6所示的方法时，处理器210用于实现上述处理单元120的功能，接口电路220用于实现上述收发单元110的功能。
当通信装置200用于实现图7所示的方法时，处理器210用于实现上述处理单元120的功能，接口电路220用于实现上述收发单元110的功能。
当通信装置200用于实现图8所示的方法时，处理器210用于实现上述处理单元120的功能，接口电路220用于实现上述收发单元110的功能。
当通信装置200用于实现图9所示的方法时，处理器210用于实现上述处理单元120的功能，接口电路220用于实现上述收发单元110的功能。
应理解，图11示出的处理器可以包含至少一个处理器，接口电路也可以包括多个接口电路。
上述提供的任一种装置中相关内容的解释及有益效果均可参考上文提供的对应的方法实施例，此处不再赘述。
当上述通信装置为应用于训练设备(或终端设备)的芯片时，该训练设备(或终端设备)的芯片实现上述方法实施例中训练设备(或终端设备)的功能。该训练设备(或终端设备)的芯片从训练设备(或终端设备)中的其它模块(如射频模块或天线)接收信息，该信息是网络设备发送给训练设备(或终端设备)的；或者，该训练设备(或终端设备)的芯片向训练设备(或终端设备)中的其它模块(如射频模块或天线)发送信息，该信息是训练设备(或终端设备)发送给网络设备的。
可以理解的是，本申请的实施例中的处理器可以是中央处理单元(Central Processing Unit，CPU)，还可以是其它通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field Programmable Gate Array，FPGA)或者其它可编程逻辑器件、晶体管逻辑器件，硬件部件或者其任意组合。通用处理器可以是微处理器，也可以是任何常规的处理器。
根据本申请实施例提供的方法，本申请还提供一种计算机程序产品，该计算机程序产品上存储有计算机程序代码，当该计算机程序代码在计算机上运行时，使得该计算机执行方法500～方法900实施例中由训练设备(或终端设备)或者网络设备执行的方法。
根据本申请实施例提供的方法，本申请还提供一种计算机可读介质，该计算机可读介质存储有程序代码，当该程序代码在计算机上运行时，使得该计算机执行上述方法500～方法900中由训练设备(或终端设备)或者网络设备执行的方法。
根据本申请实施例提供的方法，本申请还提供一种通信系统，该通信系统包括训练设备和第一网络设备。该训练设备用于执行上述方法500中训练设备对应的步骤，该第一网络设备用于执行上述方法500中第一网络设备对应的步骤。
根据本申请实施例提供的方法，本申请还提供一种通信系统，该通信系统包括训练设备和第一网络设备。该训练设备用于执行上述方法600中训练设备对应的步骤，该网络设备用于执行上述方法600中第一网络设备对应的步骤。
根据本申请实施例提供的方法，本申请还提供一种通信系统，该通信系统包括训练设备和第一网络设备。该训练设备用于执行上述方法700中训练设备对应的步骤，该网络设备用于执行上述方法700中第一网络设备对应的步骤。
根据本申请实施例提供的方法，本申请还提供一种通信系统，该通信系统包括训练设备和网络设备。该训练设备用于执行上述方法800中训练设备对应的步骤，该网络设备用于执行上述方法800中网络设备对应的步骤。
根据本申请实施例提供的方法，本申请还提供一种通信系统，该通信系统包括推理设备，如终端设备，和网络设备。该终端设备用于执行上述方法900中终端设备对应的步骤，该网络设备用于执行上述方法900中网络设备对应的步骤。
本申请的实施例中的方法步骤可以在硬件中实现，也可以在可由处理器执行的软件指令中实现。软件指令可以由相应的软件模块组成，软件模块可以被存放于随机存取存储器、闪存、只读存储器、可编程只读存储器、可擦除可编程只读存储器、电可擦除可编程只读存储器、寄存器、硬盘、移动硬盘、CD-ROM或者本领域熟知的任何其它形式的存储介质中。一种示例性的存储介质耦合至处理器，从而使处理器能够从该存储介质读取信息，且可向该存储介质写入信息。存储介质也可以是处理器的组成部分。处理器和存储介质可以位于ASIC中。另外，该ASIC可以位于基站或终端中。处理器和存储介质也可以作为分立组件存在于基站或终端中。
在上述实施例中，可以全部或部分地通过软件、硬件、固件或者其任意组合来实现。当使用软件实现时，可以全部或部分地以计算机程序产品的形式实现。所述计算机程序产品包括一个或多个计算机程序或指令。在计算机上加载和执行所述计算机程序或指令时，全部或部分地执行本申请实施例所述的流程或功能。所述计算机可以是通用计算机、专用计算机、计算机网络、网络设备、用户设备或者其它可编程装置。所述计算机程序或指令可以存储在计算机可读存储介质中，或者从一个计算机可读存储介质向另一个计算机可读存储介质传输，例如，所述计算机程序或指令可以从一个网站站点、 计算机、服务器或数据中心通过有线或无线方式向另一个网站站点、计算机、服务器或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存取的任何可用介质或者是集成一个或多个可用介质的服务器、数据中心等数据存储设备。所述可用介质可以是磁性介质，例如，软盘、硬盘、磁带；也可以是光介质，例如，数字视频光盘；还可以是半导体介质，例如，固态硬盘。该计算机可读存储介质可以是易失性或非易失性存储介质，或可包括易失性和非易失性两种类型的存储介质。
在本申请的各个实施例中，如果没有特殊说明以及逻辑冲突，不同的实施例之间的术语和/或描述具有一致性、且可以相互引用，不同的实施例中的技术特征根据其内在的逻辑关系可以组合形成新的实施例。
可以理解的是，在本申请的实施例中涉及的各种数字编号仅为描述方便进行的区分，并不用来限制本申请的实施例的范围。上述各过程的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定。

Claims (85)

  1. 一种获取训练数据集的方法，其特征在于，所述方法由训练设备或者由被配置为设置于训练设备的芯片或电路执行，所述方法包括：
    向网络设备发送第一信息，所述第一信息用于指示训练设备请求所述网络设备发送的第一训练数据集的相关信息；
    接收来自所述网络设备的所述第一训练数据集，所述第一训练数据集是基于所述第一信息所指示的所述相关信息的训练数据集，所述第一训练数据集用于人工智能AI模型的训练。
  2. 根据权利要求1所述的方法，其特征在于，所述相关信息包括以下至少一项：所述第一训练数据集的大小的信息、所述AI模型的输入的配置信息、用于所述AI模型的训练的参考信号的配置信息。
  3. 根据权利要求2所述的方法，其特征在于，
    所述第一训练数据集的大小的信息是所述训练设备基于完成所述AI模型的训练所需要的训练数据集的大小确定的。
  4. 根据权利要求2所述的方法，其特征在于，所述向网络设备发送第一信息之前，所述方法还包括：
    确定所述AI模型的第一性能；
    根据所述AI模型的第一性能和所述AI模型的第二性能，确定所述第一训练数据集的大小的信息，其中，所述第一性能为所述AI模型的当前性能，所述第二性能为所述AI模型的目标性能。
  5. 根据权利要求2至4中任一项所述的方法，其特征在于，所述参考信号的配置信息包括以下至少一项：所述参考信号的标识、所述参考信号的时域资源、所述参考信号的频域资源、所述参考信号的传输周期、传输的所述参考信号的类型。
  6. 根据权利要求1至5中任一项所述的方法，其特征在于，所述第一信息包括以下至少一项：所述AI模型的标识信息、所述AI模型的应用场景的信息、所述AI模型的用途信息、所述训练设备的算力能力的信息。
  7. 根据权利要求1至6中任一项所述的方法，其特征在于，所述方法还包括：
    根据所述第一训练数据集训练所述AI模型，并确定所述AI模型的性能；
    根据所述AI模型的性能，向所述网络设备发送第二信息，所述第二信息用于指示所述训练设备请求所述网络设备发送的第二训练数据集的相关信息；
    接收来自所述网络设备的所述第二训练数据集，所述第二训练数据集是基于所述第二信息所指示的所述相关信息的训练数据集，所述第二训练数据集用于所述AI模型的训练。
  8. 一种获取训练数据集的方法，其特征在于，所述方法由第一网络设备或者由被配置为设置于第一网络设备的芯片或电路执行，所述方法包括：
    接收来自训练设备的第一信息，所述第一信息用于指示请求所述第一网络设备发送的第一训练数据集的相关信息；
    根据所述第一信息所指示的所述相关信息，向所述训练设备发送所述第一训练数据集，所述第一训练数据集用于所述人工智能AI模型的训练。
  9. 根据权利要求8所述的方法，其特征在于，所述相关信息包括以下至少一项：所述第一训练数据集的大小的信息、所述AI模型的输入的配置信息、用于所述AI模型的训练的参考信号的配置信息。
  10. 根据权利要求9所述的方法，其特征在于，所述第一训练数据集的大小的信息是基于完成所述AI模型的训练所需要的训练数据集的大小确定的。
  11. 根据权利要求9或10所述的方法，其特征在于，所述参考信号的配置信息包括以下至少一项：所述参考信号的标识、所述参考信号的时域资源、所述参考信号的频域资源、所述参考信号的传输周期、传输的所述参考信号的类型。
  12. 根据权利要求8至11中任一项所述的方法，其特征在于，所述第一信息包括以下至少一项：所述AI模型的标识信息、所述AI模型的应用场景的信息、所述AI模型的用途信息、所述训练设备的算力能力的信息。
  13. 根据权利要求8至12中任一项所述的方法，其特征在于，所述方法还包括：
    从第二网络设备获取第三信息，所述第三信息为训练所述AI模型的相关信息，其中，所述第一网络设备为所述训练设备从第二网络设备切换至的目标网络设备；
    所述根据所述第一信息所指示的所述相关信息，向所述训练设备发送所述第一训练数据集，包括：
    根据所述第一信息所指示的所述相关信息和所述第三信息，向所述训练设备发送所述第一训练数据集。
  14. 根据权利要求13所述的方法，其特征在于，所述第三信息包括以下至少一项：
    所述训练设备请求所述第二网络设备发送的训练数据集的大小的信息、完成所述AI模型训练需要的训练数据集的大小的信息、所述AI模型的标识信息、所述训练设备的算力能力的信息。
  15. 根据权利要求13或14所述的方法，其特征在于，所述第一网络设备保存有第一映射关系，所述第一映射关系为AI模型的标识与所述AI模型的标识对应的训练数据集的大小之间的映射关系。
  16. 根据权利要求13至15中任一项所述的方法，其特征在于，所述第三信息还包括：发送训练数据集的时长的信息和/或发送训练数据集的方式的信息，所述方法还包括：
    根据所述第三信息和所述第一网络设备中的资源使用情况，确定所述第一网络设备和/或所述训练设备是否具备支持训练所述AI模型的能力。
  17. 根据权利要求8至16中任一项所述的方法，其特征在于，所述方法还包括：
    接收来自所述训练设备的第二信息，所述第二信息用于指示请求所述第一网络设备发送的第二训练数据集的相关信息，其中，所述第二信息是基于所述AI模型的性能确定的，所述AI模型性能是基于所述第一训练数据集训练确定的；
    根据所述第二信息所指示的所述相关信息，向所述训练设备发送所述第二训练数据集，所述第二训练数据集用于所述AI模型的训练。
  18. 一种获取训练数据集的方法，其特征在于，所述方法由第一网络设备或者由被配置为设置于第一网络设备的芯片或电路执行，所述方法包括：
    从第二网络设备获取第三信息，所述第三信息为训练人工智能AI模型的相关信息，其中，所述第一网络设备为训练设备从第二网络设备切换至的目标网络设备；
    接收来自所述训练设备的第一信息，所述第一信息用于请求所述第一网络设备发送训练数据集；
    根据所述第三信息，确定待发送的第一训练数据集；所述第一网络设备基于所述第一信息向所述训练设备发送所述第一训练数据集，所述第一训练数据集用于所述AI模型的训练。
  19. 根据权利要求18所述的方法，其特征在于，所述第三信息包括以下至少一项：所述训练设备请求所述第二网络设备发送的训练数据集的大小的信息、完成所述AI模型训练需要的训练数据集的大小的信息、所述AI模型的标识信息、所述训练设备的算力能力的信息。
  20. 根据权利要求18或19所述的方法，其特征在于，所述第一网络设备保存有第一映射关系，所述第一映射关系为所述AI模型的标识与所述AI模型的标识对应的训练数据集的大小之间的映射关系。
  21. 根据权利要求18至20中任一项所述的方法，其特征在于，所述第三信息还包括：发送训练数据集的时长的信息和/或发送训练数据集的方式的信息，所述方法还包括：所述第一网络设备根据所述第三信息和所述第一网络设备中的资源使用情况，确定所述第一网络设备和/或所述训练设备是否具备支持训练所述AI模型的能力。
  22. 根据权利要求18至21中任一项所述的方法，其特征在于，所述根据所述第三信息，确定待发送的第一训练数据集，包括：
    根据所述第一信息和所述第三信息，确定待发送的第一训练数据集，其中，所述第一信息用于指示请求所述第一网络设备发送的第一训练数据集的相关信息。
  23. 根据权利要求22所述的方法，其特征在于，所述相关信息包括以下至少一项：所述第一训练数据集的大小的信息、所述AI模型的输入信息、用于所述AI模型的训练的参考信号的配置信息。
  24. 根据权利要求23所述的方法，其特征在于，所述第一训练数据集的大小的信息是基于完成所述AI模型的训练所需要的训练数据集的大小确定的。
  25. 根据权利要求23所述的方法，其特征在于，所述参考信号的配置信息包括以下至少一项：所述参考信号的标识、所述参考信号的时域资源、所述参考信号的频域资源、所述参考信号的传输周期、 传输的所述参考信号的类型。
  26. 根据权利要求22至25中任一项所述的方法，其特征在于，所述第一信息包括以下至少一项：所述AI模型的标识信息、所述AI模型的应用场景的信息、所述AI模型的用途信息、所述训练设备的算力能力的信息。
  27. 根据权利要求18至26中任一项所述的方法，其特征在于，所述方法还包括：
    接收来自所述训练设备的第二信息，所述第二信息用于指示请求所述网络设备发送的第二训练数据集的相关信息，其中，所述第二信息是基于所述AI模型的性能确定的，所述AI模型的性能是基于所述第一训练数据集训练确定的；
    根据所述第二信息，确定待发送的第二训练数据集。
  28. 一种通信方法，其特征在于，所述方法由训练设备或者由被配置为设置于训练设备的芯片或电路执行，所述方法包括：
    测量N组参考信号，获取所述N组参考信号对应的N组测量结果，其中，所述N组参考信号中的每组参考信号包括至少一个参考信号，所述每组参考信号具有相同的组标识，所述N为大于1的整数；
    接收来自网络设备的第四信息，所述第四信息用于指示所述N组参考信号中的M组参考信号；
    根据所述第四信息和所述N组参考信号对应的N组测量结果，确定人工智能AI模型的第一输入信息，所述第一输入信息包括所述M组参考信号对应的M组测量结果；
    所述AI模型用于基于所述第一输入信息，获取第一输出信息，其中，所述第一输出信息包括所述N组参考信号中的K组参考信号各自的组标识，其中，所述K组参考信号各自的组标识对应于所述N组测量结果中信道质量最好的K组测量结果，每组测量结果包括一个或多个测量结果。
  29. 根据权利要求28所述的方法，其特征在于，所述第一输出信息还包括所述N组测量结果中剩余的(N-K)组参考信号各自的组标识，所述(N-K)组参考信号各自的组标识对应于(N-K)组测量结果。
  30. 根据权利要求28或29所述的方法，其特征在于，所述第四信息包括N个字段，所述N个字段与所述N组参考信号一一对应，所述N个字段中的M个字段的比特值与剩余的(N-M)个字段的比特值不同；
    所述第四信息用于指示N组参考信号中的M组参考信号，包括：
    所述第四信息中的所述M个字段用于指示所述M组参考信号。
  31. 根据权利要求28至30中任一项所述的方法，其特征在于，所述方法还包括：
    接收来自网络设备的第五信息，所述第五信息用于指示所述N组参考信号中的P组参考信号；其中，所述第五信息包括N个字段，所述N个字段与所述N组参考信号一一对应，所述N个字段中的P个字段的比特值与剩余(N-P)个字段的比特值不同；
    所述第五信息用于指示N组参考信号中的P组参考信号，包括：
    所述第五信息中的所述P个字段用于指示所述P组参考信号。
  32. 根据权利要求28至31中任一项所述的方法，其特征在于，所述方法还包括：
    接收来自网络设备的配置信息，所述配置信息用于指示以下中的一项或多项：所述N组参考信号的时域资源、所述N组参考信号的频域资源、所述N组参考信号的传输周期、所述N组参考信号的组标识。
  33. 一种通信方法，其特征在于，所述方法由网络设备或者由被配置为设置于网络设备的芯片或电路执行，所述方法包括：
    向训练设备发送N组参考信号，其中，所述N组参考信号中的每组参考信号包括至少一个参考信号，所述每组参考信号具有相同的组标识，所述N为大于1的整数；
    向所述训练设备发送第四信息，所述第四信息用于指示所述N组参考信号中的M组参考信号，其中，所述M组参考信号用于确定第一输入信息；
    AI模型用于基于所述第一输入信息，获取第一输出信息，其中，所述第一输出信息包括所述N组参考信号中的K组参考信号各自的组标识，其中，所述K组参考信号各自的组标识对应于所述N组参量信号对应的N组测量结果中信道质量最好的K组测量结果，每组测量结果包括一个或多个测量结果。
  34. 根据权利要求33所述的方法，其特征在于，所述输出信息还包括所述N组测量结果中剩余的(N-K)组参考信号各自的组标识，所述(N-K)组参考信号各自的组标识对应于(N-K)组测量结果。
  35. 根据权利要求33或34所述的方法，其特征在于，所述第四信息包括N个字段，所述N个字段与所述N组参考信号一一对应，所述N个字段中的M个字段的比特值与剩余的(N-M)个字段的比特值不同；
    所述第四信息用于指示N组参考信号中的M组参考信号，包括：
    所述第四信息中的所述M个字段用于指示所述M组参考信号。
  36. 根据权利要求33至35中任一项所述的方法，其特征在于，所述方法还包括：
    向训练设备发送第五信息，所述第五信息用于指示所述N组参考信号中的P组参考信号；其中，所述第五信息包括N个字段，所述N个字段与所述N组参考信号一一对应，所述N个字段中的P个字段的比特值与剩余(N-P)个字段的比特值不同；
    所述第五信息用于指示N组参考信号中的P组参考信号，包括：
    所述第五信息中的所述P个字段用于指示所述P组参考信号。
  37. 根据权利要求33至36中任一项所述的方法，其特征在于，所述方法还包括：
    向所述训练设备发送配置信息，所述配置信息用于指示以下中的一项或多项：所述N组参考信号的时域资源、所述N组参考信号的频域资源、所述N组参考信号的传输周期、所述N组参考信号的组标识。
  38. 一种通信方法，其特征在于，所述方法由训练设备或者由被配置为设置于训练设备的芯片或电路执行，所述方法包括：
    接收第二参考信号集，其中，所述第二参考信号集包括N组参考信号，所述N组参考信号中的每组参考信号包括至少一个参考信号，所述N为大于1的整数；
    接收第二波束指示信息，所述第二波束指示信息指示第一参考信号集对应的波束，其中，所述第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，所述第一参考信号集对应的波束用于所述训练设备中的AI模型的第一输入信息的确定，所述第一输入信息基于所述第一参考信号集对应的波束的测量结果，所述第一参考信号集包括M组参考信号，所述N为大于M的整数，所述M为大于或等于1的整数；
    其中，所述AI模型用于基于所述第一输入信息，获得第一输出信息，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，所述K为大于或等于1的整数且所述K小于所述N，所述AI模型的标签为所述第二参考信号集的测量结果中信道质量最好的K个波束。
  39. 根据权利要求38所述的方法，其特征在于，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，包括如下中的至少一项：
    所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束的信息；或者，
    K组参考信号各自的组标识，其中，所述K组参考信号对应于所述N组参考信号对应的N个测量结果中被预测为信道质量最好的K个测量结果，且所述K组参考信号各自的组标识与K个波束具有预先定义或预先配置的对应关系；或者，
    所述N组参考信号对应的多个波束信息及所述波束信息对应的N个测量结果；或者，
    所述N组参考信号各自的组标识及所述N组参考信号的N个测量结果，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
  40. 根据权利要求38或39所述的方法，其特征在于，所述第二波束指示信息指示所述第一参考信号集对应的波束包括：
    所述第二波束指示信息指示所述第一参考信号集对应的波束在所述第二参考信号集对应的多个波束中的位置。
  41. 根据权利要求38至40中任一项所述的方法，其特征在于，所述第二波束指示信息包括N个字段，所述N个字段与所述第二参考信号集对应的多个波束一一对应，所述N个字段中的M个字段的比特值与剩余(N-M)个字段的比特值不同；
    所述第二波束指示信息指示所述第一参考信号集对应的波束，包括：
    所述第二波束指示信息中的所述M个字段对应所述第一参考信号集。
  42. 根据权利要求38至41中任一项所述的方法，其特征在于，所述方法还包括：
    向所述训练设备发送第一配置信息，所述第一配置信息指示以下中的一项或多项：所述N组参考信号的时域资源、所述N组参考信号的频域资源、所述N组参考信号的传输周期、所述N组参考信号的组标识、或所述N组参考信号的波束信息。
  43. 根据权利要求38至42中任一项所述的方法，其特征在于，所述第二波束指示信息指示所述M组参考信号对应的波束。
  44. 根据权利要求43所述的方法，其特征在于，所述第二波束指示信息指示所述M组参考信号对应的波束，包括：
    所述第二波束指示信息包括所述M组参考信号的组标识或波束信息，所述M组参考信号为所述N组参考信号中的部分，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
  45. 根据权利要求44所述的方法，其特征在于，所述的方法还包括：
    向所述训练设备发送所述N组参考信号的第三配置信息；
    所述第二波束指示信息包括所述M组参考信号的组标识的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的组标识并指示以下中的一项或多项：所述N组参考信号的时域资源，所述N组参考信号的频域资源，所述N组参考信号的传输周期，或，所述N组参考信号的波束；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识；或者，
    所述第二波束指示信息包括所述M组参考信号的波束信息的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的波束信息并指示以下中的一项或多项：所述N组参考信号的组标识，时域资源，所述N组参考信号的频域资源，或，所述N组参考信号的传输周期；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识。
  46. 根据权利要求38至45中任一项所述的方法，其特征在于，所述方法还包括：对所述N组参考信号进行测量，获得N个测量结果，所述N个测量结果对应于N个波束且所述N个测量结果包括所述第一参考信号集对应的波束的测量结果。
  47. 一种通信方法，其特征在于，所述方法由网络设备或者由被配置为设置于网络设备的芯片或电路执行，所述方法包括：
    向训练设备发送第二参考信号集，其中，所述第二参考信号集包括N组参考信号，所述N组参考信号中的每组参考信号包括至少一个参考信号，所述N为大于1的整数；
    向所述训练设备发送第二波束指示信息，所述第二波束指示信息指示第一参考信号集对应的波束，其中，所述第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，所述第一参考信号集对应的波束用于所述训练设备中的AI模型的第一输入信息的确定，所述第一输入信息基于所述第一参考信号集对应的波束的测量结果，所述第一参考信号集包括M组参考信号，所述N为大于M的整数，所述M为大于或等于1的整数；
    其中，所述AI模型用于基于所述第一输入信息，获得第一输出信息，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，所述K为大于或等于1的整数且所述K小于所述N，所述AI模型的标签为所述第二参考信号集的测量结果中信道质量最好的K个波束。
  48. 根据权利要求47所述的方法，其特征在于，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，包括如下中的至少一项：
    所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束的信息；或者，
    K组参考信号各自的组标识，其中，所述K组参考信号对应于所述N组参考信号对应的N个测量结果中被预测为信道质量最好的K个测量结果，且所述K组参考信号各自的组标识与K个波束具有预先定义或预先配置的对应关系；或者，
    所述N组参考信号对应的多个波束信息及所述波束信息对应的N个测量结果；或者，
    所述N组参考信号各自的组标识及所述N组参考信号的N个测量结果，且所述N组参考信号与N 个波束具有预先定义或预先配置的对应关系。
  49. 根据权利要求47或48所述的方法，其特征在于，所述第二波束指示信息指示所述第一参考信号集对应的波束包括：所述第二波束指示信息指示所述第一参考信号集对应的波束在所述第二参考信号集对应的多个波束中的位置。
  50. 根据权利要求47至49中任一项所述的方法，其特征在于，所述第二波束指示信息包括N个字段，所述N个字段与所述第二参考信号集对应的多个波束一一对应，所述N个字段中的M个字段的比特值与剩余(N-M)个字段的比特值不同；
    所述第二波束指示信息指示所述第一参考信号集对应的波束，包括：
    所述第二波束指示信息中的所述M个字段对应所述第一参考信号集。
  51. 根据权利要求47至50中任一项所述的方法，其特征在于，所述方法还包括：
    向所述训练设备发送第一配置信息，所述第一配置信息指示以下中的一项或多项：所述N组参考信号的时域资源、所述N组参考信号的频域资源、所述N组参考信号的传输周期、所述N组参考信号的组标识、或所述N组参考信号的波束信息。
  52. 根据权利要求47至51中任一项所述的方法，其特征在于，所述第二波束指示信息指示所述M组参考信号对应的波束。
  53. 根据权利要求52所述的方法，其特征在于，所述第二波束指示信息指示所述M组参考信号对应的波束，包括：
    所述第二波束指示信息包括所述M组参考信号的组标识或波束信息，所述M组参考信号为所述N组参考信号中的部分，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
  54. 根据权利要求53所述的方法，其特征在于，所述方法还包括：
    向所述训练设备发送所述N组参考信号的第三配置信息；
    所述第二波束指示信息包括所述M组参考信号的组标识的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的组标识并指示以下中的一项或多项：所述N组参考信号的时域资源，所述N组参考信号的频域资源，所述N组参考信号的传输周期，或，所述N组参考信号的波束；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识；或者，
    所述第二波束指示信息包括所述M组参考信号的波束信息的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的波束信息并指示以下中的一项或多项：所述N组参考信号的组标识，时域资源，所述N组参考信号的频域资源，或，所述N组参考信号的传输周期；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识。
  55. 一种通信方法，其特征在于，所述方法由终端设备或者由被配置为设置于终端设备的芯片或电路执行，所述方法包括：
    接收第一参考信号集，其中，所述第一参考信号集包括M组参考信号，所述M组参考信号中的每组参考信号包括至少一个参考信号，所述M为大于或等于1的整数；
    接收第一波束指示信息，所述第一波束指示信息指示所述第一参考信号集对应的波束，其中，所述第一参考信号集用于所述AI模型的第一输入信息的确定，所述第一输入信息基于所述第一参考信号集所包括的所述M组参考信号的测量结果，所述第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，所述第二参考信号集包括N组参考信号，所述N为大于或等于M的整数；
    其中，所述AI模型用于基于所述第一输入信息，获得第一输出信息，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，所述K为大于或等于1的整数且所述K小于所述N。
  56. 根据权利要求55所述的方法，其特征在于，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，包括如下中的至少一项：
    所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束的信息；或者，
    K组参考信号各自的组标识，其中，所述K组参考信号对应于所述N组参考信号对应的N个测量结果中被预测为信道质量最好的K个测量结果，且所述K组参考信号各自的组标识与K个波束具有预 先定义或预先配置的对应关系；或者，
    所述N组参考信号对应的多个波束信息及所述波束信息对应的N个测量结果；或者，
    所述N组参考信号各自的组标识及所述N组参考信号的N个测量结果，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
  57. 根据权利要求55或56所述的方法，其特征在于，所述第一波束指示信息指示所述第一参考信号集对应的波束包括：所述第一波束指示信息指示所述第一参考信号集对应的波束在所述第二参考信号集对应的多个波束中的位置。
  58. 根据权利要求55至57中任一项所述的方法，其特征在于，所述第一波束指示信息包括N个字段，所述N个字段与所述第二参考信号集对应的多个波束一一对应，所述N个字段中的M个字段的比特值与剩余(N-M)个字段的比特值不同；
    所述第一波束指示信息指示所述第一参考信号集对应的波束，包括：
    所述第一波束指示信息中的所述M个字段对应所述第一参考信号集。
  59. 根据权利要求55至58中任一项所述的方法，其特征在于，所述方法还包括：
    接收第一配置信息，所述第一配置信息指示以下中的一项或多项：所述M组参考信号的时域资源、所述M组参考信号的频域资源、所述M组参考信号的传输周期、所述M组参考信号的组标识、或所述M组参考信号的波束信息。
  60. 根据权利要求55至59中任一项所述的方法，其特征在于，所述第一波束指示信息指示所述M组参考信号对应的波束。
  61. 根据权利要求60所述的方法，其特征在于，所述第一波束指示信息指示所述M组参考信号对应的波束，包括：
    所述第一波束指示信息包括所述M组参考信号的组标识或波束信息，所述M组参考信号为所述N组参考信号中的部分，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
  62. 根据权利要求61所述的方法，其特征在于，所述第一波束指示信息包括在所述M组参考信号的第二配置信息中，
    所述第一波束指示信息包括所述M组参考信号的组标识的情况下，所述第二配置信息还包括所述M组参考信号的时域资源、频域资源、传输周期、或、波束信息中的一项或多项；或者，
    所述第一波束指示信息包括所述M组参考信号的波束信息的情况下，所述第二配置信息还包括所述M组参考信号的时域资源、频域资源、传输周期、或、组标识中的一项或多项。
  63. 根据权利要求61或62所述的方法，其特征在于，所述方法还包括：
    接收所述N组参考信号的第三配置信息；
    所述第一波束指示信息包括所述M组参考信号的组标识的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的组标识并指示以下中的一项或多项：所述N组参考信号的时域资源，所述N组参考信号的频域资源，所述N组参考信号的传输周期，或，所述N组参考信号的波束；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识；或者，
    所述第一波束指示信息包括所述M组参考信号的波束信息的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的波束信息并指示以下中的一项或多项：所述N组参考信号的组标识，时域资源，所述N组参考信号的频域资源，或，所述N组参考信号的传输周期；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识。
  64. 根据权利要求55至63中任一项所述的方法，其特征在于，所述方法还包括：
    利用所述AI模型基于所述第一输入信息获得所述第一输出信息；
    发送所述第一输出信息。
  65. 一种通信方法，其特征在于，所述方法由网络设备或者由被配置为设置于终端设备的芯片或电路执行，所述方法包括：
    向终端设备发送第一参考信号集，其中，所述第一参考信号集包括M组参考信号，所述M组参考信号中的每组参考信号包括至少一个参考信号，所述M为大于或等于1的整数；
    向所述终端设备发送第一波束指示信息，所述第一波束指示信息指示所述第一参考信号集对应的波束，其中，所述第一参考信号集用于所述AI模型的第一输入信息的确定，所述第一输入信息基于所述第一参考信号集所包括的所述M组参考信号的测量结果，所述第一参考信号集对应的波束为第二参考信号集对应的多个波束的子集，所述第二参考信号集包括N组参考信号，所述N为大于或等于M的整数；
    其中，所述AI模型用于基于所述第一输入信息，获得第一输出信息，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，所述K为大于或等于1的整数且所述K小于所述N。
  66. 根据权利要求65所述的方法，其特征在于，所述第一输出信息指示所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束，包括如下中的至少一项：
    所述第二参考信号集对应的多个波束中被预测为信道质量最好的K个波束的信息；或者，
    K组参考信号各自的组标识，其中，所述K组参考信号对应于所述N组参考信号对应的N个测量结果中被预测为信道质量最好的K个测量结果，且所述K组参考信号各自的组标识与K个波束具有预先定义或预先配置的对应关系；或者，
    所述N组参考信号对应的多个波束信息及所述波束信息对应的N个测量结果；或者，所述N组参考信号各自的组标识及所述N组参考信号的N个测量结果，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
  67. 根据权利要求65或66所述的方法，其特征在于，所述第一波束指示信息指示所述第一参考信号集对应的波束包括：所述第一波束指示信息指示所述第一参考信号集对应的波束在所述第二参考信号集对应的多个波束中的位置。
  68. 根据权利要求65至67中任一项所述的方法，其特征在于，所述第一波束指示信息包括N个字段，所述N个字段与所述第二参考信号集对应的多个波束一一对应，所述N个字段中的M个字段的比特值与剩余(N-M)个字段的比特值不同；
    所述第一波束指示信息指示所述第一参考信号集对应的波束，包括：
    所述第一波束指示信息中的所述M个字段对应所述第一参考信号集。
  69. 根据权利要求65至68中任一项所述的方法，其特征在于，所述方法还包括：
    向所述终端设备发送第一配置信息，所述第一配置信息指示以下中的一项或多项：所述M组参考信号的时域资源、所述M组参考信号的频域资源、所述M组参考信号的传输周期、所述M组参考信号的组标识、或所述M组参考信号的波束信息。
  70. 根据权利要求65至69中任一项所述的方法，其特征在于，所述第一波束指示信息指示所述M组参考信号对应的波束。
  71. 根据权利要求70所述的方法，其特征在于，所述第一波束指示信息指示所述M组参考信号对应的波束，包括：
    所述第一波束指示信息包括所述M组参考信号的组标识或波束信息，所述M组参考信号为所述N组参考信号中的部分，且所述N组参考信号与N个波束具有预先定义或预先配置的对应关系。
  72. 根据权利要求71所述的方法，其特征在于，
    所述第一波束指示信息包括在所述M组参考信号的第二配置信息中，
    所述第一波束指示信息包括所述M组参考信号的组标识的情况下，所述第二配置信息还包括所述M组参考信号的时域资源、频域资源、传输周期、或、波束信息中的一项或多项；或者，
    所述第一波束指示信息包括所述M组参考信号的波束信息的情况下，所述第二配置信息还包括所述M组参考信号的时域资源、频域资源、传输周期、或、组标识中的一项或多项。
  73. 根据权利要求71或72所述的方法，其特征在于，所述方法还包括：
    向所述终端设备发送所述N组参考信号的第三配置信息；
    所述第一波束指示信息包括所述M组参考信号的组标识的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的组标识并指示以下中的一项或多项：所述N组参考信号的时域资源，所述N组参考信号的频域资源，所述N组参考信号的传输周期，或，所述N组参考信号的波束；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组 参考信号的M个组标识；或者，
    所述第一波束指示信息包括所述M组参考信号的波束信息的情况下，所述N组参考信号的第三配置信息包括所述N组参考信号各自的波束信息并指示以下中的一项或多项：所述N组参考信号的组标识，时域资源，所述N组参考信号的频域资源，或，所述N组参考信号的传输周期；且所述M组参考信号为所述N组参考信号中的部分包括：所述N组参考信号的N个组标识包括所述M组参考信号的M个组标识。
  74. 根据权利要求65至73中任一项所述的方法，其特征在于，所述方法还包括：
    接收来自所述终端设备的所述第一输出信息。
  75. 一种通信装置，其特征在于，用于实现如权利要求1至7中任一项所述的方法，或者，用于实现如权利要求28至32中任一项所述的方法，或者，用于实现如权利要求38至46中任一项所述的方法。
  76. 一种通信装置，其特征在于，用于实现如权利要求8至17中任一项所述的方法，或者，用于实现如权利要求18至27中任一项所述的方法。
  77. 一种通信装置，其特征在于，用于实现如权利要求33至37中任一项所述的方法，或者，用于实现如权利要求47至54中任一项所述的方法。
  78. 一种通信装置，其特征在于，用于实现如权利要求55至64中任一项所述的方法，或者，用于实现如权利要求65至74中任一项所述的方法。
  79. 一种通信装置，其特征在于，包括：
    处理器，所述处理器和存储器耦合，所述处理器用于调用所述存储器存储的计算机程序指令，以执行如权利要求1至7中任一项所述的方法，或者，用于实现如权利要求28至32中任一项所述的方法，或者，用于实现如权利要求38至46中任一项所述的方法。
  80. 一种通信装置，其特征在于，包括：
    处理器，所述处理器和存储器耦合，所述处理器用于调用所述存储器存储的计算机程序指令，以执行如权利要求8至17中任一项所述的方法，或者，用于实现如权利要求18至27中任一项所述的方法。
  81. 一种通信装置，其特征在于，包括：处理器，所述处理器和存储器耦合，所述处理器用于调用所述存储器存储的计算机程序指令，以执行如权利要求33至37中任一项所述的方法，或者，用于实现如权利要求47至54中任一项所述的方法。
  82. 一种通信装置，其特征在于，包括：处理器，所述处理器和存储器耦合，所述处理器用于调用所述存储器存储的计算机程序指令，以执行如权利要求55至64中任一项所述的方法，或者，用于实现如权利要求65至74中任一项所述的方法。
  83. 一种通信系统，其特征在于，包括用于执行如权利要求1至7中任一项所述的方法的通信装置以及用于实现如权利要求8至17中任一项所述方法的通信装置；
    或者，包括用于执行如权利要求1至7中任一项所述方法的通信装置以及用于实现如权利要求8至27中任一项所述方法的通信装置；
    或者，包括用于实现如权利要求28至32中任一项所述方法的通信装置以及用于实现如权利要求33至37中任一项所述方法的通信装置；
    或者，包括用于实现如权利要求38至46中任一项所述方法的通信装置以及用于实现如权利要求47至54中任一项所述方法的通信装置；
    或者，包括用于实现如权利要求55至64中任一项所述方法的通信装置以及用于实现如权利要求65至74中任一项所述方法的通信装置。
  84. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质上存储有指令，当所述指令在计算机上运行时，使得计算机执行如权利要求1至74任一项所述的方法。
  85. 一种计算机程序产品，其特征在于，包括指令，当所述指令在计算机上运行时，使得计算机执行如权利要求1至74任一项所述的方法。
PCT/CN2023/119695 2022-09-30 2023-09-19 一种获取训练数据集的方法和装置 WO2024067248A1 (zh)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
CN202211214685.0 2022-09-30
CN202211214685 2022-09-30
CN202211247927.6 2022-10-12
CN202211247927.6A CN117851819A (zh) 2022-09-30 2022-10-12 一种获取训练数据集的方法和装置

Publications (1)

Publication Number Publication Date
WO2024067248A1 true WO2024067248A1 (zh) 2024-04-04

Family

ID=90476178

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/119695 WO2024067248A1 (zh) 2022-09-30 2023-09-19 一种获取训练数据集的方法和装置

Country Status (1)

Country Link
WO (1) WO2024067248A1 (zh)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110430068A (zh) * 2018-04-28 2019-11-08 华为技术有限公司 一种特征工程编排方法及装置
CN111954228A (zh) * 2019-05-16 2020-11-17 北京三星通信技术研究有限公司 波束管理方法、装置、电子设备及计算机可读存储介质
CN112073106A (zh) * 2020-08-14 2020-12-11 清华大学 毫米波波束预测方法及装置、电子设备、可读存储介质
US20210336682A1 (en) * 2020-04-24 2021-10-28 Qualcomm Incorporated Reporting quantized user equipment (ue) orientation for beam selection
CN113994598A (zh) * 2019-04-17 2022-01-28 诺基亚技术有限公司 无线网络的波束预测
WO2022061940A1 (zh) * 2020-09-28 2022-03-31 华为技术有限公司 一种模型数据传输方法及通信装置
WO2022174461A1 (zh) * 2021-02-22 2022-08-25 北京小米移动软件有限公司 波束测量方法、波束测量装置及存储介质

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110430068A (zh) * 2018-04-28 2019-11-08 华为技术有限公司 一种特征工程编排方法及装置
CN113994598A (zh) * 2019-04-17 2022-01-28 诺基亚技术有限公司 无线网络的波束预测
CN111954228A (zh) * 2019-05-16 2020-11-17 北京三星通信技术研究有限公司 波束管理方法、装置、电子设备及计算机可读存储介质
US20210336682A1 (en) * 2020-04-24 2021-10-28 Qualcomm Incorporated Reporting quantized user equipment (ue) orientation for beam selection
CN112073106A (zh) * 2020-08-14 2020-12-11 清华大学 毫米波波束预测方法及装置、电子设备、可读存储介质
WO2022061940A1 (zh) * 2020-09-28 2022-03-31 华为技术有限公司 一种模型数据传输方法及通信装置
WO2022174461A1 (zh) * 2021-02-22 2022-08-25 北京小米移动软件有限公司 波束测量方法、波束测量装置及存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
HUAWEI, HISILICON: "Discussion on AI/ML for beam management", 3GPP DRAFT; R1-2203143, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG1, no. e-Meeting; 20220509 - 20220520, 29 April 2022 (2022-04-29), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France, XP052143961 *

Similar Documents

Publication Publication Date Title
Brik et al. Deep learning for B5G open radio access network: Evolution, survey, case studies, and challenges
WO2022022334A1 (zh) 基于人工智能的通信方法和通信装置
US11838787B2 (en) Functional architecture and interface for non-real-time ran intelligent controller
CN113661727B (zh) 用于无线网络的无线电接入网(ran)节点的神经网络的配置
US20230016595A1 (en) Performing a handover procedure
EP4099635A1 (en) Method and device for selecting service in wireless communication system
US11617094B2 (en) Machine learning in radio access networks
WO2023143267A1 (zh) 一种模型配置方法及装置
US20220321421A1 (en) Systems, methods and apparatuses for automating context specific network function configuration
WO2022134854A1 (zh) 信息共享方法及通信装置
CN115428383B (zh) 基于频带间参考信号来估计无线电频带的特征
WO2024012331A1 (zh) 一种确定人工智能ai模型的方法及装置
WO2022038760A1 (ja) 通信品質を予測する装置、方法及びプログラム
EP4158931A1 (en) Dynamic labeling for machine learning models for use in dynamic radio environments of a communications network
WO2024067248A1 (zh) 一种获取训练数据集的方法和装置
WO2022206513A1 (zh) 模型处理的方法、通信装置和系统
US11863354B2 (en) Model transfer within wireless networks for channel estimation
CN117851819A (zh) 一种获取训练数据集的方法和装置
WO2021114192A1 (zh) 一种网络参数调整方法及网络管理设备
WO2024109682A1 (zh) 一种用于定位的方法及装置
WO2023011655A1 (zh) 一种通信方法及装置
WO2024082274A1 (zh) Ai任务指示的方法、通信装置和系统
WO2024022007A1 (zh) 一种无线局域网中通信的方法和装置
US20230354063A1 (en) Method and apparatus for configuring artificial neural network for wireless communication in mobile communication system
US20240098575A1 (en) Methods and devices for determination of an update timescale for radio resource management algorithms

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23870480

Country of ref document: EP

Kind code of ref document: A1