WO2023077272A1 - 信道数据生成方法、装置、设备及存储介质 - Google Patents

信道数据生成方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2023077272A1
WO2023077272A1 PCT/CN2021/128216 CN2021128216W WO2023077272A1 WO 2023077272 A1 WO2023077272 A1 WO 2023077272A1 CN 2021128216 W CN2021128216 W CN 2021128216W WO 2023077272 A1 WO2023077272 A1 WO 2023077272A1
Authority
WO
WIPO (PCT)
Prior art keywords
channel
channel data
information
model
generation model
Prior art date
Application number
PCT/CN2021/128216
Other languages
English (en)
French (fr)
Inventor
田文强
肖寒
刘文东
Original Assignee
Oppo广东移动通信有限公司
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
Application filed by Oppo广东移动通信有限公司 filed Critical Oppo广东移动通信有限公司
Priority to CN202180100752.3A priority Critical patent/CN117678172A/zh
Priority to PCT/CN2021/128216 priority patent/WO2023077272A1/zh
Publication of WO2023077272A1 publication Critical patent/WO2023077272A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/391—Modelling the propagation channel

Definitions

  • the present application relates to the technical field of wireless communication, and in particular to a method, device, device and storage medium for generating channel data.
  • the channel environment is one of the main issues affecting the wireless transmission performance between communication devices.
  • Embodiments of the present application provide a method, device, device, and storage medium for generating channel data.
  • the scheme can accurately and automatically generate massive channel data, thereby improving the effect of channel modeling and improving the accuracy of wireless communication system research and design. Described technical scheme is as follows:
  • an embodiment of the present application provides a method for generating channel data, the method is executed by a computer device, and the method includes:
  • the channel generation model is a machine learning model obtained by performing machine learning training on channel data samples.
  • an embodiment of the present application provides a channel data processing method, the method is executed by a computer device, and the method includes:
  • the channel data samples are used to characterize channel conditions in the sample channel environment;
  • a channel generation model as a generator and a channel discrimination model as a discriminator, based on the channel data samples, train the channel generation model and the channel discrimination model by means of generative adversarial learning;
  • the channel generation model trained to convergence is used to generate virtual channel data; the virtual channel data is used to characterize channel conditions in a channel environment.
  • an embodiment of the present application provides a device for generating channel data, the device comprising:
  • a generating module configured to generate virtual channel data through a channel generation model; the virtual channel data is used to characterize channel conditions in a channel environment;
  • the channel generation model is a machine learning model obtained by performing machine learning training on channel data samples.
  • an embodiment of the present application provides a channel data processing device, the device comprising:
  • An acquisition module configured to acquire channel data samples; the channel data samples are used to characterize channel conditions in the sample channel environment;
  • a training module configured to use a channel generation model as a generator and a channel discrimination model as a discriminator to train the channel generation model and the channel discrimination model by means of generative confrontation learning based on the channel data samples;
  • the channel generation model trained to convergence is used to generate virtual channel data; the virtual channel data is used to characterize channel conditions in a channel environment.
  • an embodiment of the present application provides a computer device, the computer device is implemented as an information reporting device, and the computer device includes a processor, a memory, and a transceiver;
  • a computer program is stored in the memory, and the processor executes the computer program, so that the computer device implements the channel data generating method or the channel data processing method described above.
  • an embodiment of the present application provides a computer device, the computer device includes a processor, a memory, and a transceiver, the memory stores a computer program, and the computer program is used to be executed by the processor to The above channel data generation method or channel data processing method is implemented.
  • the embodiment of the present application also provides a computer-readable storage medium, where a computer program is stored in the storage medium, and the computer program is loaded and executed by a processor to implement the above channel data generation method or channel data processing method.
  • the present application also provides a chip, which is used to run in a computer device, so that the computer device executes the above channel data generation method or channel data processing method.
  • the present application provides a computer program product comprising computer instructions stored in a computer readable storage medium.
  • the processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the above channel data generation method or channel data processing method.
  • the present application provides a computer program, which is executed by a processor of a computer device, so as to implement the above channel data generating method or channel data processing method.
  • a channel generation model is trained in advance through channel data samples.
  • the virtual channel data corresponding to the channel environment can be automatically generated through simulation and prediction, and the channel data corresponding to the channel environment can be quickly obtained without actual collection. , thereby greatly improving the acquisition efficiency of channel data under various channel environments, thereby improving the effect of channel modeling, and improving the accuracy of research and design of wireless communication systems.
  • FIG. 1 is a schematic diagram of a network architecture provided by an embodiment of the present application.
  • Fig. 2 is a schematic diagram of the principle of a communication system provided by an embodiment of the present application.
  • Fig. 3 is a schematic diagram of a neural network provided by an embodiment of the present application.
  • Fig. 4 is a schematic diagram of a neural network provided by another embodiment of the present application.
  • FIG. 5 is a flowchart of a method for generating channel data provided by an embodiment of the present application.
  • FIG. 6 is a flowchart of a channel data processing method provided by an embodiment of the present application.
  • FIG. 7 is a flow chart of model training and channel data generation provided by an embodiment of the present application.
  • FIG. 8 is a flowchart of a channel data processing and channel data generation method provided by an embodiment of the present application.
  • Fig. 9 is a schematic diagram of a virtual channel data structure involved in the embodiment shown in Fig. 8;
  • FIG. 10 is a schematic diagram of another virtual channel data structure involved in the embodiment shown in FIG. 8;
  • FIG. 11 is a model architecture diagram of a channel generation model and a channel discrimination model involved in the embodiment shown in FIG. 8;
  • Fig. 12 is a block diagram of a channel data generation device provided by an embodiment of the present application.
  • Fig. 13 is a block diagram of a channel data processing device provided by an embodiment of the present application.
  • Fig. 14 is a schematic structural diagram of a computer device provided by an embodiment of the present application.
  • the network architecture and business scenarios described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute limitations on the technical solutions provided by the embodiments of the present application.
  • the evolution of the technology and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.
  • FIG. 1 shows a schematic diagram of a network architecture of a wireless communication system provided by an embodiment of the present application.
  • the network architecture may include: a terminal 10 and a base station 20 .
  • the number of terminals 10 is generally multiple, and one or more terminals 10 may be distributed in a cell managed by each base station 20 .
  • the terminal 10 may include various handheld devices with wireless communication functions, vehicle-mounted devices, wearable devices, computing devices or other processing devices connected to a wireless modem, as well as various forms of user equipment (User Equipment, UE), mobile station ( Mobile Station, MS), terminal device (terminal device) and so on.
  • UE User Equipment
  • MS Mobile Station
  • terminal device terminal device
  • the base station 20 is a device deployed in an access network to provide a wireless communication function for the terminal 10 .
  • the base station 20 may include various forms of macro base stations, micro base stations, relay stations, access points and so on.
  • the names of devices with base station functions may be different, for example, in a 5th-Generation (5G) NR system, it is called gNodeB or gNB.
  • the name "base station” may change as communication technology evolves.
  • the above-mentioned devices that provide the wireless communication function for the terminal 10 are collectively referred to as base stations.
  • the above-mentioned network architecture also includes other network devices, such as: a central control node (Central Network Control, CNC), an access and mobility management function (Access and Mobility Management Function, AMF ) device, session management function (Session Management Function, SMF) or user plane function (User Plane Function, UPF) device, etc.
  • a central control node Central Network Control, CNC
  • AMF Access and Mobility Management Function
  • SMF Session Management Function
  • UPF User Plane Function
  • the "5G NR system" in the embodiments of the present disclosure may also be called a 5G system or a New Radio (New Radio, NR) system, but those skilled in the art can understand its meaning.
  • the technical solution described in the embodiments of the present disclosure can be applied to the 4G system, the 5G NR system, or the subsequent evolution system of the 5G NR system.
  • FIG. 2 shows a schematic schematic diagram of a wireless communication system provided by an embodiment of the present application.
  • the basic workflow is that the transmitter performs operations such as encoding, modulation, and encryption on the information source at the transmitting end to form the transmission information to be transmitted.
  • the sent information is transmitted to the receiving end through the wireless space, and the receiving end performs operations such as decoding, decryption and demodulation on the received received information, and finally restores the source information.
  • the encoding, modulation, encryption, decoding, demodulation, decryption and other operations at the sending end and receiving end are controllable, but the channel environment in the space environment is uncontrollable, complex and changeable.
  • FIG. 3 shows a schematic diagram of a neural network provided by an embodiment of the present application.
  • the basic structure of a simple neural network includes: input layer, hidden layer and output layer.
  • the input layer is responsible for receiving data
  • the hidden layer processes the data
  • the final result is generated in the output layer.
  • each node represents a processing unit, which can be regarded as simulating a neuron.
  • Multiple neurons form a layer of neural network, and multi-layer information transmission and processing constructs an overall neural network.
  • neural network deep learning algorithms have been proposed in recent years, more hidden layers have been introduced, and feature learning is performed through layer-by-layer training of neural networks with multiple hidden layers, which greatly improves the learning of neural networks.
  • processing capabilities and are widely used in pattern recognition, signal processing, optimization combination, anomaly detection, etc.
  • FIG. 4 shows a schematic diagram of a neural network provided by another embodiment of the present application.
  • its basic structure includes: an input layer, multiple convolutional layers, multiple pooling layers, a fully connected layer, and an output layer.
  • the introduction of convolutional layers and pooling layers effectively controls the network
  • the rapid increase of parameters limits the number of parameters and excavates the characteristics of local structures, which improves the robustness of the algorithm.
  • wireless channels The use and understanding of wireless channels is the basis for the construction of wireless communication systems.
  • the most direct way is to collect actual wireless channels, such as through paired signal transmitters and signal receivers.
  • Wireless channel information or obtain wireless channel information by collecting signals from a third-party transmitter (such as a cellular network base station) through a specific receiver.
  • a third-party transmitter such as a cellular network base station
  • the propagation characteristics of the wireless channel can be obtained directly, thereby assisting the design of the wireless communication system.
  • wireless channel modeling related work can extract relevant transmission characteristics of a given channel from limited wireless channel samples (ie, channel data samples), such as large-scale parameters, small-scale Parameters, such as: multipath information, delay power spectral density, transmission launch angle/arrival angle, etc.
  • the frequency band is gradually moving towards high frequency, and the scene is gradually moving towards more complex special environments such as air, space, earth and sea.
  • the expansion of more scenarios makes the wireless channel environment that the current wireless communication system needs to face more and more complex.
  • FIG. 5 shows a flowchart of a method for generating channel data provided by an embodiment of the present application, the method may be executed by a computer device; the method may include the following steps:
  • Step 501 generate virtual channel data through a channel generation model, the virtual channel data is used to characterize channel conditions in a channel environment, and the channel generation model is a machine learning model obtained through machine learning training on channel data samples.
  • the channel environment represented by the above-mentioned virtual channel data can be a simulated channel environment.
  • the channel generation model can generate channel data corresponding to various channel environments by means of simulation and prediction. In this process, there is no need The channel data is collected for the actual channel environment, so that the acquisition of the channel data does not depend on the actual channel environment.
  • the virtual channel data generated by the channel generation model in the embodiment of the present application can be used to construct a channel environment for research and design of wireless communication systems, or can also be used as sample data of machine learning models in wireless communication system design, In order to improve the performance gain of the combination of artificial intelligence and wireless communication.
  • the solution shown in the embodiment of this application can pre-train a channel generation model through channel data samples, and through the channel generation model, virtual channel data corresponding to the channel environment can be automatically generated by means of simulation and prediction.
  • Channel data corresponding to various channel environments can be quickly obtained without actual acquisition, thereby greatly improving the acquisition efficiency of channel data in various channel environments, thereby improving the effect of channel modeling, and improving wireless communication system research. design accuracy.
  • the above-mentioned channel generation model can be trained by means of generative adversarial learning, so that the virtual channel data generated by the channel generation model can accurately simulate and predict channel conditions in various channel environments.
  • FIG. 6 shows a flowchart of a method for processing channel data provided by an embodiment of the present application, the method may be executed by a computer device; the method may include the following steps:
  • step 601 channel data samples are acquired, and the channel data samples are used to characterize channel conditions in the sample channel environment.
  • the above-mentioned channel data samples may be samples obtained by collecting channel data from the actual channel environment; or, the channel data samples may also be samples artificially constructed by technicians based on the actual channel environment; or, the above-mentioned channel data samples may also be It is a sample automatically constructed by technicians through other channel data construction tools.
  • the channel generation model is used as the generator, the channel discrimination model is used as the discriminator, and based on the channel data samples, the channel generation model and the channel discrimination model are trained by means of generative confrontation learning, and the channel generation model after training is used for For generating virtual channel data, the virtual channel data is used to characterize the channel situation in the channel environment.
  • two machine learning models in the training phase of the channel generation model, two machine learning models can be set in the computer device, one machine learning model A (corresponding to the above-mentioned channel generation model) is used to generate channel data, and the other machine learning model A
  • the role of B (corresponding to the above channel identification model) is to judge whether the input channel data is true (or in other words, to judge whether the input channel data is naturally existing channel data or channel data generated by a machine), and the channel data samples are used as training samples , train the two machine learning models by means of generative adversarial learning until both machine learning models are trained to converge.
  • the trained machine learning model B has a certain ability to judge whether the input channel data samples are true, and when the accuracy of the machine learning model B is high enough (that is, converges), if the machine If the channel data generated by learning model A cannot be accurately discriminated by machine learning model B, it is considered that the channel data generated by machine learning model A is close enough to the real channel data, and then machine learning model A also reaches the convergence state. At this time, the machine learning model A can then be used as a converged channel generation model for subsequent generation of channel data.
  • a channel generation model is trained based on channel data samples in advance through generative adversarial learning.
  • the channel environment can be automatically generated by means of simulation and prediction.
  • Corresponding virtual channel data can quickly obtain channel data corresponding to various channel environments without actual collection, thus greatly improving the acquisition efficiency of channel data in various channel environments, thereby improving the effect of channel modeling. And improve the accuracy of wireless communication system research and design.
  • the solution proposed in this application includes a model training phase and a model application phase.
  • FIG. 7 shows a flow chart of model training and channel data generation provided by an embodiment of the present application.
  • the above-mentioned model training phase and model application phase may be executed by a model training device and a channel data generating device respectively.
  • the process includes the following steps:
  • Step 1 in the model training phase, the model training device 71 acquires the channel data sample 71a, the initialized machine learning model A, and the initialized machine learning module B.
  • the format of the data output by the initialized machine learning model A and the format of the data input by the initialized machine learning model B may match the data format of the channel data.
  • the format of the data output by the initialized machine learning model A or the format of the data input by the initialized machine learning model B can be the same as the data format of the channel data; or, the format of the data output by the initialized machine learning model A or the format of the initialized
  • the format of the input data of the machine learning model B can be converted into the data format of the channel data through a pre-designed conversion method.
  • the format of the data output by the initialized machine learning model A and the format of the data input by the initialized machine learning model B may be pre-designed by the developer.
  • step 2 the model training device uses the machine learning model A to generate predicted virtual channel data 71b.
  • the model training device may use the machine learning model A to output data satisfying the data format of the channel data as predicted virtual channel data.
  • step 3 the model training device uses the channel data sample 71a and the predicted virtual channel data 71b as positive and negative samples to train the machine learning model A and the machine learning model B in an adversarial learning manner.
  • the above-mentioned predicted virtual channel data can be used as a negative sample, and its corresponding training label is the first label, and the first label can indicate that the predicted virtual channel data is non-naturally existing channel data (or it is generated by simulation prediction). channel data).
  • the above-mentioned channel data sample may be used as a positive sample, and its corresponding training label is a second label, and the second label may indicate that the channel data sample is naturally existing channel data.
  • the model training device can train the machine learning model A and the machine learning model B in turn.
  • the accuracy of the predicted virtual channel data output by machine learning model A at the beginning of training is not high enough, and the accuracy of machine learning model B in judging whether the input channel data is naturally existing channel data is not high enough. It is not high enough.
  • the judgment accuracy of machine learning model B is getting higher and higher.
  • the predicted virtual channel data generated by machine learning model A is getting closer and closer to the naturally existing channel data.
  • machine learning model B can accurately judge that the channel data samples are naturally existing channel data, but cannot accurately distinguish whether the predicted virtual channel data is naturally existing channel data. It can be considered that the predicted virtual channel data generated by the machine learning model A is close enough to the naturally existing channel data.
  • Step 4 when both the machine learning model A and the machine learning model B converge, the model training device outputs the machine learning model A as a channel generation model 72; the channel generation model can be deployed to the channel data generation device 73.
  • Step 5 in the model application phase channel, the channel data generation device 73 generates virtual channel data 72a through the channel generation model 72 .
  • model training device and channel data generating device can be implemented as the same physical device, for example, can be implemented as the same personal computer, workstation or server.
  • the aforementioned model training device and channel data generating device may also be implemented as different physical devices.
  • the aforementioned model training device may be implemented as a personal computer, workstation or server used by developers
  • the aforementioned data generating device may be a personal computer, workstation or server used by designers of wireless communication systems.
  • FIG. 8 shows a flowchart of a method for channel data processing and channel data generation provided by an embodiment of the present application.
  • the method may be executed by a computer device, for example, it may be executed interactively by a model training device and a channel data generation device; the method may include the following steps:
  • Step 801 in the model training phase, the model training device acquires channel data samples; the channel data samples are used to represent channel conditions in the sample channel environment.
  • the developer may pre-collect several channel data samples, and input the collected channel data samples to the model training device.
  • channel data samples may be channel data collected in an actual channel environment, or may be artificially or machine-constructed and considered to be naturally existing channel data.
  • the channel generation model can be used as the generator, and the channel identification model can be used as the discriminator , based on channel data samples, the channel generation model and channel discrimination model are trained by generative adversarial learning.
  • the training process can refer to steps 802 to 807 of the harness.
  • Step 802 in the stage of training the channel discrimination model, the model training device generates predicted virtual channel data through the channel generation model.
  • the above-mentioned channel generation model includes at least one of the following four types of networks: a fully connected network, a convolutional neural network, a residual network, or a self-attention mechanism network.
  • the model training device may input input information into the channel generation model, and obtain predicted virtual channel data output by the channel generation model after processing the input information samples.
  • the channel generation model may have an input port, and in the process of model training and application, the channel generation model may process the input information layer by layer, and finally output data satisfying a certain data format as virtual channel data.
  • the input information includes at least one of the following four types of information:
  • Noise information random number information, channel type indication information, or channel data sample information
  • the channel type indication information is used to indicate the channel type
  • Channel data sample information is information constructed based on channel data samples.
  • the input of the channel generation model can be any input, for example, any noise, or any random number, and the channel generation model is triggered by any input, that is, subsequent step-by-step layer processing, and finally output virtual channel data, wherein the virtual channel data output in the model training stage is the virtual channel data predicted above.
  • the above-mentioned noise information may come from a real environment, or may be artificially generated.
  • the above random number information may be a random number sequence or a pseudo-random number sequence.
  • the format of the above noise information or random number information may be a one-dimensional vector, or a two-dimensional matrix, or a high-dimensional noise or random number set.
  • the format of the above noise information and random number information can be agreed in advance, or be consistent with the format of the virtual channel data to be generated.
  • the input of the channel generation model may also be information with specified meanings, for example, it may be information indicating a certain channel type (the purpose is to make the output virtual channel data simulate The virtual channel corresponding to the channel type), or information derived based on the channel data samples (the purpose is to make the channel environment of the virtual channel corresponding to the output virtual channel data similar to the channel environment corresponding to the channel data samples, or in other words, The channel environment of the virtual channel corresponding to the output virtual channel data is improved on the basis of the channel environment corresponding to the channel data samples).
  • the above channel type indication information is used to indicate the channel type corresponding to the channel data sample.
  • the channel type indicated by the above channel type indication information may be consistent with the channel type corresponding to the channel data sample, so that when random numbers or random noise are mixed in, the During the training process and the subsequent application, the channel generation model is made to generate various virtual channel data matching the input channel type indication information.
  • the channel type indication information includes at least one of the following five types of information:
  • Time domain feature information frequency feature information, space domain feature information, environment feature information, or scene feature information.
  • the above-mentioned channel type indication information may indicate the frequency information, environment information, and scene information corresponding to the channel, for example: high frequency, low frequency, indoor, outdoor, dense residential area, open field, Internet of Things scene, industrial scene, etc. .
  • time-domain feature information may be referred to as channel index feature information, for example: delay power spectrum information, multipath information, angle information, speed information, and the like.
  • the above environmental feature information may indicate indoor environment, outdoor environment, open field and so on.
  • the above scene feature information may indicate scene categories such as line of sight transmission (Line Of Sight, LOS), non-line of sight transmission (Not Line Of Sight, NLOS), high speed, and low speed.
  • LOS Line Of Sight
  • NLOS Near Line Of Sight
  • high speed high speed
  • low speed low speed
  • the channel data sample information includes at least one of the following three types of information:
  • the model training device may mix noise and channel data samples as input to the channel generation model, or mix random numbers and channel data samples as input to the channel generation model, and so on.
  • the virtual channel data includes channel data respectively corresponding to at least one granularity in at least one dimension.
  • the virtual channel data when the virtual channel data includes channel data respectively corresponding to each granularity in at least two dimensions, the virtual channel data includes a matrix of at least two dimensions; or, when the virtual channel data includes a matrix corresponding to each granularity in at least two dimensions
  • the virtual channel data when channel data correspond to each, the virtual channel data includes one-dimensional data obtained by arranging channel data respectively corresponding to each granularity in at least two dimensions.
  • a single sample of the above virtual channel data may be composed of a matrix with a size of M*N, which has M
  • M*N which has M
  • M and N may or may not be equal
  • the specific numerical indication in the matrix represents the channel quality.
  • the two-dimensional data of M*N can also be synthesized into one-dimensional data of size 1*(M*N) or (M*N)*1.
  • the specific transformation can be the first dimension and then the second dimension, or It can be the second dimension first and then the first dimension. This transformation is the difference in the form of expression.
  • At least one dimension includes at least one of the following four dimensions:
  • Frequency domain dimension time domain dimension, space domain dimension, or real and imaginary part dimension.
  • a granularity on the frequency domain dimension includes:
  • At least one radio bearer (Radio Bearer, RB) or at least one subcarrier.
  • a single sample of virtual channel data is composed of a first dimension with a granularity of m.
  • the first dimension may be a frequency domain dimension.
  • the granularity m may be a RB (a greater than or equal to 1, such as 2RB, 4RB, 8RB), or b subcarriers (b is greater than 1, such as 4 subcarriers, 6 subcarriers, or 18 subcarriers).
  • the frequency domain range indicated by a single sample of the virtual channel data is the frequency domain range of M*m.
  • a granularity on the time-domain dimension includes:
  • p1 microseconds, the length of at least one symbol, or the number of sampling points of at least one symbol; wherein, p1 is a positive number.
  • a single sample of virtual channel data may also be composed of a first dimension whose granularity is p.
  • the first dimension may be a time-domain dimension.
  • the granularity p may be a delay granularity, for example, a The delay granularity is p1 microseconds, or p2 symbol length, or the number of sampling points of p3 symbols, and the symbol mentioned here can be an Orthogonal Frequency Division Multiplexing (OFDM) symbol.
  • OFDM Orthogonal Frequency Division Multiplexing
  • a granularity on the spatial domain dimension includes:
  • a single sample of virtual channel data is composed of a second dimension with a granularity of n.
  • the second dimension may be a spatial domain dimension, specifically, an antenna dimension.
  • the second dimension is composed of N antenna pairs, and the second dimension The granularity is a pair of transmit and receive antennas.
  • a single sample of virtual channel data may also be composed of a second dimension with a granularity of q
  • the second dimension may be a space domain dimension, specifically an angle domain dimension, for example, the second dimension is composed of N angles
  • the second dimension The second granularity is the angle interval size between the above N angles.
  • the channel quality indication on a specific combination of dimensions may represent the channel quality indication under the specific combination of dimensions.
  • FIG. 9 shows a schematic diagram of a virtual channel data structure involved in the embodiment of the present application.
  • the indicator value X on the third row and sixth column can be used to represent the channel quality situation on the bandwidth of the third specific granularity on the sixth spatial granularity.
  • FIG. 10 shows a schematic diagram of another virtual channel data structure involved in the embodiment of the present application.
  • the indication value Y on the fourth row and the fifth column can be used to represent the channel quality condition of the delay of the fourth specific granularity on the fifth spatial granularity (for example, angle of arrival).
  • the output of the above channel generation model can have an additional dimension based on the content described above.
  • This dimension is to combine the virtual channel (or virtual channel
  • the imaginary part and real part data of the obtained channel feature information) are presented independently.
  • the output of the above-mentioned channel generation model can also be split and combined on the basis of the above-mentioned first dimension, second dimension, and third dimension.
  • the second dimension is the antenna pair dimension
  • there is also It can be split into a transmitting antenna sub-dimension and a receiving antenna sub-dimension, thereby expanding the dimension of the above-mentioned virtual channel output form.
  • the two-dimensional virtual channel formed by the first dimension and the second dimension is taken as an example, and the dimension of the virtual channel involved in the embodiment of the present application is not limited to two dimensions.
  • the virtual channel data includes at least one of the following two types of information: original channel information, or a channel feature vector;
  • the channel feature vector is obtained by performing data transformation on the original channel information.
  • the original channel information includes channel quality information.
  • the channel feature vector is obtained by performing singular value decomposition (Singular Value Decomposition, SVD) on the original channel information.
  • singular value decomposition Single Value Decomposition
  • the output information of the above-mentioned channel generation model can also be the channel feature information obtained by mathematical transformation of the above-mentioned original channel information, for example, the channel feature vector information obtained by SVD decomposition, which can be single-stream channel feature vector information or multi-stream channel feature information.
  • Channel feature vector information of streams such as 2-stream, 4-stream, and 8-stream channel feature vector information.
  • Step 803 the model training device inputs the predicted virtual channel data and channel data samples into the channel discrimination model to obtain the first discrimination result of the channel discrimination model; the first discrimination result is used to indicate whether the predicted virtual channel data and channel data samples are model Generated channel data.
  • the above channel discrimination model includes one or more of the following four types of networks: a fully connected network, a convolutional neural network, a residual network, or a self-attention mechanism network.
  • FIG. 11 shows a model architecture diagram of a channel generation model and a channel discrimination model related to the embodiment of the present application.
  • Step 804 the model training device updates the model parameters of the channel discrimination model according to the first discrimination result.
  • the model training device in the stage of training the channel identification model, can calculate the loss function value through the first identification result, the predicted virtual channel data and the respective labels of the channel data samples, and then use the loss function value The model parameters of the channel discrimination model are updated.
  • Step 805 in the stage of training the channel generation model, the model training device generates predicted virtual channel data through the channel generation model.
  • step 806 the model training device inputs the predicted virtual channel data into the channel identification model to obtain a second identification result of the channel identification model; the second identification result is used to indicate whether the predicted virtual channel data is channel data generated by the model.
  • Step 807 the model training device updates the model parameters of the channel generation model according to the second identification result.
  • the model training device in the stage of training the channel generation model, can calculate the loss function value based on the second identification result and the predicted label of the virtual channel data, and then use the loss function value to the channel generation model The model parameters are updated.
  • Steps 802 to 807 above are iteratively executed until both models converge.
  • the generator can be extracted separately for the generation of virtual channel data.
  • the channel generation model is used to generate virtual channel data
  • the channel identification model is used to judge the difference between the virtual channel and the real channel.
  • the model training device can use Generative Adversarial Networks (GAN) as the basic structure of the channel generation model and the channel identification model, let the input information (such as random numbers) be used as the input of the channel generation model, and let the given neural network structure be used as The basic structure of the channel generative model, which generates the current virtual channel input. Then the channel discrimination model judges the difference between the virtual channel and the real channel. If the difference can be judged, the above process will continue to be circulated, the channel generation model parameters will be updated, and a new virtual channel output will be generated until the channel discrimination model cannot judge the difference between the virtual channel and the real channel. When there is a difference between the real channels, it can be considered that the channel generation model has been constructed, and a virtual channel fitting the real channel can be generated.
  • GAN Generative Adversarial Networks
  • This solution can be used to solve the problem that it is difficult to obtain a large amount of data sets for traditional channel modeling and channel estimation under complex frequencies, scenarios, and environments, and it is difficult to effectively fit complex nonlinear channel models.
  • This scheme only a small amount of actual data can be realized, and the channel generation model can be built, so as to build a large amount of virtual channel data based on the channel generation model, which greatly saves the difficulty and cost of manual actual data collection, and also avoids traditional data modeling Efficiency problems under complex channels.
  • the above virtual channel data can be used for artificial intelligence-based wireless communication solutions to quickly build data sets in multiple frequency bands, multiple scenarios, and multiple environments to support model retraining and updating on demand.
  • Step 808 in the model application stage channel the channel data generation device generates virtual channel data through the channel generation model.
  • virtual channel data is generated through a channel generation model, including:
  • the input information is input into the channel generation model, and the virtual channel data output after the channel generation model processes the input information is obtained.
  • the channel generation model in the channel data generation device may be encapsulated in the channel data generator.
  • the above-mentioned channel data generator may be provided with an input interface, and the input of the input interface is the input information of the above-mentioned channel generation model, which will not be repeated here.
  • the noise generator and/or random number generator may also be encapsulated in the channel data generator, At this time, the channel data generator may have no input, or in other words, the channel data generator does not require additional input of noise information and/or random number information.
  • a noise generator and/or a random number generator may also be encapsulated in the channel generation model, that is, when the processing information of the channel generation model is noise information and/or random number In the case of information, the channel generation model may have no input. In this case, the channel generation model automatically generates noise information and/or random number information, processes the noise information and/or random number information, and outputs virtual channel data.
  • This application provides a method for constructing a virtual channel, which replaces the actual channel with the virtual channel, and replaces the actual environment with the virtual environment, thereby reducing the dependence on the actual channel environment data in the research and development of the integration of artificial intelligence and wireless communication systems.
  • the construction of the virtual channel environment provided by this application depends on the realization of the channel generation model, and the channel generation model can generate wireless channel data corresponding to one or more frequency bands, scenarios, and environments.
  • the wireless channel data generated by the above-mentioned channel generation model can be used to construct a data set of an AI solution for a wireless communication system, or be used for channel analysis and modeling of a wireless communication system.
  • the virtual channel data output by the channel generation model can be used to simulate channel information in different frequencies, environments, and scenarios, such as: high frequency, low frequency, indoors, outdoors, dense communities, open field, IoT scenarios, industrial scenarios, etc.
  • FIG. 12 shows a block diagram of an apparatus for generating channel data provided by an embodiment of the present application.
  • the apparatus has the functions executed by the channel data generating device in the method shown in FIG. 5 or FIG. 8 above.
  • the device may include:
  • a generation module 1201 configured to generate virtual channel data through a channel generation model; the virtual channel data is used to characterize channel conditions in a channel environment;
  • the channel generation model is a machine learning model obtained by performing machine learning training on channel data samples.
  • the generation module 1201 is configured to input input information into the channel generation model, and obtain the virtual channel data output by the channel generation model after processing the input information.
  • the input information includes at least one of the following four types of information:
  • Noise information random number information, channel type indication information, or channel data sample information
  • the channel type indication information is used to indicate the channel type
  • the channel data sample information is information constructed based on the channel data samples.
  • the channel type indication information includes at least one of the following five types of information:
  • Time domain feature information frequency feature information, space domain feature information, environment feature information, or scene feature information.
  • the channel data sample information includes at least one of the following three types of information:
  • the virtual channel data includes channel data respectively corresponding to at least one granularity in at least one dimension.
  • the at least one dimension includes at least one of the following four dimensions:
  • Frequency domain dimension time domain dimension, space domain dimension, or real and imaginary part dimension.
  • a granularity on the frequency domain dimension includes:
  • At least one RB or at least one subcarrier At least one RB or at least one subcarrier.
  • a granularity on the time domain dimension includes:
  • p1 microseconds, the length of at least one symbol, or the number of sampling points of at least one symbol; wherein, p1 is a positive number.
  • a granularity on the spatial domain dimension includes:
  • the virtual channel data includes channel data respectively corresponding to each granularity in at least two dimensions
  • the virtual channel data includes a matrix of at least two dimensions
  • the virtual channel data includes one-dimensional data obtained by arranging channel data corresponding to respective granularities in the at least two dimensions.
  • the virtual channel data includes at least one of the following two types of information: original channel information, or a channel feature vector;
  • the channel feature vector is obtained by performing data transformation on the original channel information.
  • the original channel information includes channel quality information.
  • the channel feature vector includes a singular value decomposition of the original channel information.
  • the channel generation model includes at least one of the following four networks:
  • a channel generation model is trained in advance through channel data samples.
  • virtual channel data corresponding to the channel environment can be automatically generated by means of simulation and prediction.
  • the channel data corresponding to various channel environments can be quickly obtained through actual acquisition, which greatly improves the acquisition efficiency of channel data in various channel environments, thereby improving the effect of channel modeling and improving the research and design of wireless communication systems. accuracy.
  • FIG. 13 shows a block diagram of an apparatus for generating channel data provided by an embodiment of the present application.
  • the apparatus has the functions executed by the model training device in the method shown in FIG. 6 or FIG. 8 above.
  • the device may include:
  • An acquisition module 1301, configured to acquire channel data samples; the channel data samples are used to characterize channel conditions in the sample channel environment;
  • the training module 1302 is configured to use a channel generation model as a generator and a channel discrimination model as a discriminator to train the channel generation model and the channel discrimination model by means of generative confrontation learning based on the channel data samples;
  • the channel generation model trained to convergence is used to generate virtual channel data; the virtual channel data is used to characterize channel conditions in a channel environment.
  • the training module 1302 is configured to:
  • the channel discrimination model Inputting the predicted virtual channel data and the channel data samples into the channel discrimination model to obtain a first discrimination result of the channel discrimination model; the first discrimination result is used to indicate the predicted virtual channel data and Whether the channel data sample is channel data generated by the model;
  • the model parameters of the channel discrimination model are updated according to the first discrimination result.
  • the training module 1302 is configured to:
  • the channel generation model In the stage of training the channel generation model, generate predicted virtual channel data through the channel generation model;
  • the second discrimination result is used to indicate whether the predicted virtual channel data is channel data generated by the model ;
  • the training module 1302 is configured to input input information into the channel generation model, and obtain the predicted virtual output after the channel generation model processes the input information samples. channel data.
  • the input information includes at least one of the following four types of information:
  • Noise information random number information, channel type indication information, or channel data sample information
  • the channel type indication information is used to indicate the channel type
  • the channel data sample information is information constructed based on the channel data samples.
  • the channel type indication information is used to indicate a channel type corresponding to the channel data sample.
  • the channel type indication information includes at least one of the following five types of information:
  • Time domain feature information frequency feature information, space domain feature information, environment feature information, or scene feature information.
  • the channel data sample information includes at least one of the following three types of information:
  • a channel generation model is trained based on channel data samples in advance through generative adversarial learning.
  • the channel environment can be automatically generated by means of simulation and prediction.
  • Corresponding virtual channel data can quickly obtain channel data corresponding to various channel environments without actual collection, thus greatly improving the acquisition efficiency of channel data in various channel environments, thereby improving the effect of channel modeling. And improve the accuracy of wireless communication system research and design.
  • the device provided by the above embodiment realizes its functions, it only uses the division of the above-mentioned functional modules as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional modules according to actual needs. That is, the content structure of the device is divided into different functional modules to complete all or part of the functions described above.
  • Fig. 14 shows a structural block diagram of a computer device 1400 shown in an exemplary embodiment of the present application.
  • the computer device 1400 includes a central processing unit (Central Processing Unit, CPU) 1401, a system memory 1404 including a random access memory (Random Access Memory, RAM) 1402 and a read-only memory (Read-Only Memory, ROM) 1403, and A system bus 1405 that connects the system memory 1404 and the central processing unit 1401 .
  • the computer device 1400 also includes a basic input/output system (Input/Output system, I/O system) 1406 that helps to transmit information between various devices in the computer, and is used to store an operating system 1413, application programs 1414 and other programs The mass storage device 1407 of the module 1415 .
  • I/O system Basic input/output system
  • the basic input/output system 1406 includes a display 1408 for displaying information and input devices 1409 such as a mouse and a keyboard for users to input information. Both the display 1408 and the input device 1409 are connected to the central processing unit 1401 through the input and output controller 1410 connected to the system bus 1405 .
  • the basic input/output system 1406 may also include an input-output controller 1410 for receiving and processing input from a keyboard, a mouse, or an electronic stylus and other devices. Similarly, input output controller 1410 also provides output to a display screen, printer, or other type of output device.
  • the mass storage device 1407 is connected to the central processing unit 1401 through a mass storage controller (not shown) connected to the system bus 1405 .
  • the mass storage device 1407 and its associated computer-readable media provide non-volatile storage for the computer device 1400 . That is to say, the mass storage device 1407 may include a computer-readable medium (not shown) such as a hard disk or a Compact Disc Read-Only Memory (CD-ROM) drive.
  • a computer-readable medium such as a hard disk or a Compact Disc Read-Only Memory (CD-ROM) drive.
  • Computer-readable media may comprise computer storage media and communication media.
  • Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data.
  • Computer storage media include RAM, ROM, Erasable Programmable Read Only Memory (EPROM), Electronically Erasable Programmable Read-Only Memory (EEPROM) flash memory or other Solid state storage technology, CD-ROM, Digital Versatile Disc (DVD) or other optical storage, tape cartridge, tape, disk storage or other magnetic storage device.
  • EPROM Erasable Programmable Read Only Memory
  • EEPROM Electronically Erasable Programmable Read-Only Memory
  • CD-ROM Compact Disc
  • DVD Digital Versatile Disc
  • the computer storage medium is not limited to the above-mentioned ones.
  • the above-mentioned system memory 1404 and mass storage device 1407 may be collectively referred to as memory.
  • the computer device 1400 can also operate on a remote computer connected to a network through a network such as the Internet. That is, the computer device 1400 can be connected to the network 1412 through the network interface unit 1411 connected to the system bus 1405, or in other words, the network interface unit 1411 can also be used to connect to other types of networks or remote computer systems (not shown). ).
  • the memory also includes at least one computer instruction, the at least one computer instruction is stored in the memory, and the central processing unit 1401 implements the above-mentioned embodiments by executing the at least one instruction, at least one section of program, code set or instruction set. In the method, all or part of the steps are performed by the model training device or the channel data generating device.
  • the embodiment of the present application also provides a computer-readable storage medium, where a computer program is stored in the storage medium, and the computer program is loaded and executed by a processor to implement the method shown in FIG. 5 , FIG. 6 or FIG. 8 above. In, all or part of the steps performed by the model training device or the channel data generating device.
  • the present application also provides a chip, which is used to run in a computer device, so that the computer device executes the method performed by the model training device or the channel data generation device in the method shown in Fig. 5, Fig. 6 or Fig. 8 above. All or part of the steps.
  • the present application also provides a computer program product, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium.
  • the processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the method shown in FIG. 5, FIG. 6 or FIG. 8 above, and the model training device or the channel data Generate all or part of the steps performed by the device.
  • the present application also provides a computer program, the computer program is executed by the processor of the computer device, so as to realize all the steps performed by the model training device or the channel data generation device in the method shown in Fig. 5, Fig. 6 or Fig. 8 above. or partial steps.
  • the functions described in the embodiments of the present application may be implemented by hardware, software, firmware or any combination thereof.
  • the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium.
  • Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another.
  • a storage media may be any available media that can be accessed by a general purpose or special purpose computer.

Landscapes

  • Physics & Mathematics (AREA)
  • Electromagnetism (AREA)
  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

一种信道数据生成方法、装置、设备及存储介质，属于无线通信技术领域。该方法包括：通过信道生成模型生成虚拟信道数据(501)；该虚拟信道数据用于表征信道环境中的信道情况；其中，该信道生成模型是通过信道数据样本进行机器学习训练得到的机器学习模型。通过上述方案，极大的提高了各种信道环境下的信道数据的获取效率，进而提高信道建模的效果，并提高无线通信系统研究设计的准确性。

Description

信道数据生成方法、装置、设备及存储介质 技术领域
本申请涉及无线通信技术领域，特别涉及一种信道数据生成方法、装置、设备及存储介质。
背景技术
在无线通信系统中，信道环境是影响通信设备之间的无线传输性能的主要问题之一。
在相关技术中，为了研究各种信道环境下的无线传输性能，通常需要预先获取到各种信道环境下的信道数据，并根据获取到的信道数据对各种信道环境进行建模，继而辅助进行无线通信系统的研究设计。
然而，随着无线通信的不断发展，无线信道环境也越来越复杂，需要考虑的因素也越来越多，技术人员很难准确的在各种信道环境下采集到足够的信道数据。
发明内容
本申请实施例提供了一种信道数据生成方法、装置、设备及存储介质。该方案能够准确的自动生成海量的信道数据，进而提高信道建模的效果，并提高无线通信系统研究设计的准确性。所述技术方案如下：
一方面，本申请实施例提供了一种信道数据生成方法，所述方法由计算机设备执行，所述方法包括：
通过信道生成模型生成虚拟信道数据；所述虚拟信道数据用于表征信道环境中的信道情况；
其中，所述信道生成模型是通过信道数据样本进行机器学习训练得到的机器学习模型。
一方面，本申请实施例提供了一种信道数据处理方法，所述方法由计算机设备执行，所述方法包括：
获取信道数据样本；所述信道数据样本用于表征样本信道环境中的信道情况；
以信道生成模型为生成器，以信道鉴别模型为判别器，基于所述信道数据样本，通过生成对抗学习的方式对所述信道生成模型和所述信道鉴别模型进行训练；
其中，训练至收敛后的所述信道生成模型用于生成虚拟信道数据；所述虚拟信道数据用于表征信道环境中的信道情况。
另一方面，本申请实施例提供了一种信道数据生成装置，所述装置包括：
生成模块，用于通过信道生成模型生成虚拟信道数据；所述虚拟信道数据用于表征信道环境中的信道情况；
其中，所述信道生成模型是通过信道数据样本进行机器学习训练得到的机器学习模型。
另一方面，本申请实施例提供了一种信道数据处理装置，所述装置包括：
获取模块，用于获取信道数据样本；所述信道数据样本用于表征样本信道环境中的信道情况；
训练模块，用于以信道生成模型为生成器，以信道鉴别模型为判别器，基于所述信道数据样本，通过生成对抗学习的方式对所述信道生成模型和所述信道鉴别模型进行训练；
其中，训练至收敛后的所述信道生成模型用于生成虚拟信道数据；所述虚拟信道数据用于表征信道环境中的信道情况。
另一方面，本申请实施例提供了一种计算机设备，所述计算机设备实现为信息上报设备，所述计算机设备包括处理器、存储器和收发器；
存储器中存储有计算机程序，处理器执行所述计算机程序，以使得计算机设备实现上述信道数据生成方法或者信道数据处理方法。
再一方面，本申请实施例提供了一种计算机设备，所述计算机设备包括处理器、存储器和收发器，所述存储器存储有计算机程序，所述计算机程序用于被所述处理器执行，以实现上述信道数据生成方法或者信道数据处理方法。
又一方面，本申请实施例还提供了一种计算机可读存储介质，所述存储介质中存储有计算机程序，所述计算机程序由处理器加载并执行以实现上述信道数据生成方法或者信道数据处理方法。
又一方面，本申请还提供了一种芯片，所述芯片用于在计算机设备中运行，以使得所述计算机设备执行上述信道数据生成方法或者信道数据处理方法。
又一方面，本申请提供了一种计算机程序产品，该计算机程序产品包括计算机指令，该计算机指令存储在计算机可读存储介质中。计算机设备的处理器从计算机可读存储介质读取该计算机指令，处理器执行该计算机指令，使得该计算机设备执行上述信道数据生成方法或者信道数据处理方法。
又一方面，本申请提供了一种计算机程序，该计算机程序由计算机设备的处理器执行，以实现上述信道数据生成方法或者信道数据处理方法。
本申请实施例提供的技术方案可以带来如下有益效果：
预先通过信道数据样本训练出一个信道生成模型，通过该信道生成模型，可以通过模拟、预测的方式自动生成信道环境对应的虚拟信道数据，不需要进行实际采集即可以快速获得信道环境对应的信道数据，从而极大的提高了各种信道环境下的信道数据的获取效率，进而提高信道建模的效果，并提高无线通信系统研究设计的准确性。
附图说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本申请一个实施例提供的网络架构的示意图；
图2是本申请一个实施例提供的通信系统的原理示意图；
图3是本申请一个实施例提供的神经网络示意图；
图4是本申请另一个实施例提供的神经网络示意图；
图5是本申请一个实施例提供的信道数据生成方法的流程图；
图6是本申请一个实施例提供的信道数据处理方法的流程图；
图7是本申请一个实施例提供的模型训练和信道数据生成的流程框架图；
图8是本申请一个实施例提供的信道数据处理及信道数据生成方法流程图；
图9是图8所示实施例涉及的一种虚拟信道数据结构示意图；
图10是图8所示实施例涉及的另一种虚拟信道数据结构示意图；
图11是图8所示实施例涉及的一种信道生成模型以及信道鉴别模型的模型架构图；
图12是本申请一个实施例提供的信道数据生成装置的框图；
图13是本申请一个实施例提供的信道数据处理装置的框图；
图14是本申请一个实施例提供的计算机设备的结构示意图。
具体实施方式
为使本申请的目的、技术方案和优点更加清楚，下面将结合附图对本申请实施方式作进一步地详细描述。
本申请实施例描述的网络架构以及业务场景是为了更加清楚地说明本申请实施例的技术 方案，并不构成对本申请实施例提供的技术方案的限定，本领域普通技术人员可知，随着网络架构的演变和新业务场景的出现，本申请实施例提供的技术方案对于类似的技术问题，同样适用。
请参考图1，其示出了本申请一个实施例提供的无线通信系统的网络架构的示意图。该网络架构可以包括：终端10和基站20。
终端10的数量通常为多个，每一个基站20所管理的小区内可以分布一个或多个终端10。终端10可以包括各种具有无线通信功能的手持设备、车载设备、可穿戴设备、计算设备或连接到无线调制解调器的其它处理设备，以及各种形式的用户设备(User Equipment，UE)，移动台(Mobile Station，MS)，终端设备(terminal device)等等。为方便描述，本申请实施例中，上面提到的设备统称为终端。
基站20是一种部署在接入网中用以为终端10提供无线通信功能的装置。基站20可以包括各种形式的宏基站，微基站，中继站，接入点等等。在采用不同的无线接入技术的系统中，具备基站功能的设备的名称可能会有所不同，例如在第5代移动通信(5th-Generation，5G)NR系统中，称为gNodeB或者gNB。随着通信技术的演进，“基站”这一名称可能会变化。为方便描述，本申请实施例中，上述为终端10提供无线通信功能的装置统称为基站。
可选的，图1中未示出的是，上述网络架构还包括其它网络设备，比如：中心控制节点(Central Network Control，CNC)、接入和移动性管理功能(Access and Mobility Management Function，AMF)设备、会话管理功能(Session Management Function，SMF)或者用户面功能(User Plane Function，UPF)设备等等。
本公开实施例中的“5G NR系统”也可以称为5G系统或者新空口(New Radio，NR)系统，但本领域技术人员可以理解其含义。本公开实施例描述的技术方案可以适用于4G系统、5G NR系统，也可以适用于5G NR系统后续的演进系统。
为了便于理解，下面对本申请涉及的一些相关名词或者背景概念进行介绍：
一、无线通信
请参考图2，其示出了本申请一个实施例提供的无线通信系统的原理示意图。如图2所示，在无线通信系统之中，基本的工作流程是发送机在发送端对信源进行编码、调制、加密等操作，形成待传输的发送信息。发送信息通过无线空间传输至接收端，接收端对收到的接收信息进行解码、解密解调等操作，最终恢复信源信息。
在上述过程中，发送端和接收端的编码、调制、加密、解码、解调、解密等操作是可控的，但是空间环境中的信道环境则是不可控的，是复杂且多变的。
二、人工智能(Artificial Intelligence，AI)
近年来，以神经网络为代表的人工智能研究在很多领域都取得了非常大的成果，其也将在未来很长一段时间内在人们的生产生活中起到重要的作用。
请参考图3，其示出了本申请一个实施例提供的神经网络示意图。如图3所示，一个简单的神经网络的基本结构包括：输入层，隐藏层和输出层。输入层负责接收数据，隐藏层对数据的处理，最后的结果在输出层产生。在这其中，各个节点代表一个处理单元，可以认为是模拟了一个神经元，多个神经元组成一层神经网络，多层的信息传递与处理构造出一个整体的神经网络。
随着神经网络研究的不断发展，近年来又提出了神经网络深度学习算法，较多的隐层被引入，通过多隐层的神经网络逐层训练进行特征学习，极大地提升了神经网络的学习和处理能力，并在模式识别、信号处理、优化组合、异常探测等方面广泛被应用。
同样，随着深度学习的发展，卷积神经网络也被进一步研究。请参考图4，其示出了本申请另一个实施例提供的神经网络示意图。如图4所示，其基本结构包括：输入层、多个卷 积层、多个池化层、全连接层及输出层，其中，卷积层和池化层的引入，有效地控制了网络参数的剧增，限制了参数的个数并挖掘了局部结构的特点，提高了算法的鲁棒性。
三、信道实采
对于无线信道的利用与认识是无线通信系统构建的基础，要实现对基本的无线通信信道的研究，最直接的方式是对于实际无线信道做采集，例如通过配对的信号发射机和信号接收机获得无线信道信息，或者通过特定的接收机采集第三方发射机(例如蜂窝网络基站)的信号从而获得无线信道信息。通过上述方法，可以直接地获取无线信道的传播特性，从而辅助无线通信系统的设计。
四、传统信道建模
对于无线信道的实采，考虑到采集的困难与成本，很难做到对所有场景、特征的采集。在无线信道实采的基础上，传统无线信道建模类的相关工作可以在有限的无线信道的样本(即信道数据样本)上提取出给定信道的相关传输特征，例如大尺度参数、小尺度参数，诸如：多径信息、时延功率谱密度、传输发射角/到达角等。
随着无线通信系统的发展，在频段上逐渐向高频迈进，在场景上逐步走向更加复杂的空天地海等特殊环境，在应用范围上向人机交互、物联网交互、工业应用、特种应用等更多场景扩展，使得对于当前无线通信系统所需要面对的无线信道环境越来越复杂。
在上述情况下，对于无线信道的实采显得十分困难，这里的困难既有技术层面的困难，也有操作层面的困难。与此同时，对于上述复杂信道的数学建模也面临巨大挑战，频段、环境、场景的复杂性会直接导致信道建模的复杂性，非线性的信道特征以及难以拟合的信道传播特性会对传统数学建模来研究信道的方式带来困难与挑战，例如高频信道建模目前还是一个亟待解决的问题，更进一步地，复杂场景和应用环境下的实际信道环境建模与理想信道环境建模之间的差异在未来无线通信研究中还将继续随着信道环境的复杂程度急剧增加。
由此可见，在复杂频段、复杂环境、复杂场景下，通过信道实采和传统数学建模的方式来获取信道特征信息是一个重大挑战。
与此同时，人工智能在无线通信系统中的应用越来越多，当前大量的研究基于人工智能与无线通信的结合开展，而这些工作对于无线信道本身以及无线信道所关联的数据具有极大的依赖和需求，可以说无线信道数据是决定人工智能与无线通信结合性能增益的关键。
在这种前提下，当需要大量无线信道数据作为基于人工智能的无线通信解决方案所需的数据集时，一方面传统的信道实采和信道建模的方法会在可实现性、可靠性方面存在较大的问题，另一方面，实现成本也是一个需要面对的问题，人工智能解决方案对于模型训练数据集的依赖程度极高，当信道数据需要作为模型训练集时，往往需要几千、几万、几十万甚至更大规模的信道数据，而实采如此规模的数据集信息的成本代价又是极高的。
综上，如何获取、构建有效的信道数据集，用来支持人工智能与无线通信系统融合的研究，是一个亟待解决的关键问题。
请参考图5，其示出了本申请一个实施例提供的信道数据生成方法的流程图，该方法可以由计算机设备执行；该方法可以包括如下步骤：
步骤501，通过信道生成模型生成虚拟信道数据，虚拟信道数据用于表征信道环境中的信道情况，信道生成模型是通过信道数据样本进行机器学习训练得到的机器学习模型。
其中，上述虚拟信道数据表征的信道环境可以是模拟的信道环境，换句话说，信道生成模型可以通过模拟、预测的方式生成各种不同的信道环境对应的信道数据，在此过程中，不需要对实际的信道环境进行信道数据的采集，使得信道数据的获取不依赖于实际的信道环境。
本申请实施例中通过信道生成模型生成的虚拟信道数据，可以用来构建信道环境，用于无线通信系统的研究设计，或者，也可以用作无线通信系统设计中的机器学习模型的样本数 据，以提高人工智能与无线通信结合的性能增益。
综上所述，本申请实施例所示的方案，可以预先通过信道数据样本训练出一个信道生成模型，通过该信道生成模型，可以通过模拟、预测的方式自动生成信道环境对应的虚拟信道数据，不需要进行实际采集即可以快速获得各种信道环境对应的信道数据，从而极大的提高了各种信道环境下的信道数据的获取效率，进而提高信道建模的效果，并提高无线通信系统研究设计的准确性。
其中，上述信道生成模型可以通过生成对抗学习的方式进行训练，从而使得信道生成模型生成的虚拟信道数据能够足够准确的模拟以及预测出各种信道环境的信道情况。
请参考图6，其示出了本申请一个实施例提供的信道数据处理方法的流程图，该方法可以由计算机设备执行；该方法可以包括如下步骤：
步骤601，获取信道数据样本，信道数据样本用于表征样本信道环境中的信道情况。
其中，上述信道数据样本可以是对实际的信道环境进行信道数据采集得到的样本；或者，信道数据样本也可以是技术人员基于实际的信道环境人工构建得到的样本；或者，上述信道数据样本也可以是技术人员通过其它信道数据构建工具自动构建的样本。
步骤602，以信道生成模型为生成器，以信道鉴别模型为判别器，基于信道数据样本，通过生成对抗学习的方式对信道生成模型和信道鉴别模型进行训练，训练至收敛后的信道生成模型用于生成虚拟信道数据，虚拟信道数据用于表征信道环境中的信道情况。
在本申请实施例中，在信道生成模型的训练阶段，计算机设备中可以设置两个机器学习模型，一个机器学习模型A(对应上述信道生成模型)的作用是生成信道数据，另一个机器学习模型B(对应上述信道鉴别模型)的作用是判断输入的信道数据是否为真(或者说，判断输入的信道数据是自然存在的信道数据还是由机器生成的信道数据)，以信道数据样本为训练样本，通过生成对抗学习的方式对两个机器学习模型进行训练，直至两个机器学习模型均训练至收敛。
其中，由于信道数据样本的存在，经过训练的机器学习模型B具有一定的判断输入的信道数据样本是否为真的能力，而当机器学习模型B的准确性足够高(即收敛)时，如果机器学习模型A生成的信道数据无法被机器学习模型B准确判别，则认为机器学习模型A生成的信道数据足够接近真实存在的信道数据，则机器学习模型A也达到收敛状态，此时，机器学习模型A即可以作为收敛后的信道生成模型，用于后续的信道数据的生成。
综上所述，本申请实施例所示的方案，预先通过生成对抗学习的方式，基于信道数据样本训练出一个信道生成模型，通过该信道生成模型，可以通过模拟、预测的方式自动生成信道环境对应的虚拟信道数据，不需要进行实际采集即可以快速获得各种信道环境对应的信道数据，从而极大的提高了各种信道环境下的信道数据的获取效率，进而提高信道建模的效果，并提高无线通信系统研究设计的准确性。
如本申请上述图5和图6所示，本申请提出的方案包括模型训练阶段和模型应用阶段。请参考图7，其示出了本申请一个实施例提供的模型训练和信道数据生成的流程框架图。如图7所示，上述模型训练阶段和模型应用阶段可以分别由模型训练设备和信道数据生成设备执行。如图7所示，该流程包括如下几个步骤：
步骤1，在模型训练阶段，模型训练设备71获取信道数据样本71a、初始化的机器学习模型A、以及初始化的机器学习模块B。
其中，初始化的机器学习模型A输出的数据的格式，以及初始化的机器学习模型B输入的数据的格式，可以与信道数据的数据格式相匹配。比如，初始化的机器学习模型A输出的数据的格式或者初始化的机器学习模型B输入的数据的格式，可以与信道数据的数据格式相同；或者，初始化的机器学习模型A输出的数据的格式或者初始化的机器学习模型B输入的 数据的格式，可以通过预先设计的转化方式转化为信道数据的数据格式。
上述初始化的机器学习模型A输出的数据的格式，以及初始化的机器学习模型B输入的数据的格式可以由开发人员预先设计。
步骤2，模型训练设备通过机器学习模型A生成预测的虚拟信道数据71b。
在本申请实施例中，模型训练设备可以通过机器学习模型A输出满足信道数据的数据格式的数据，作为预测的虚拟信道数据。
步骤3，模型训练设备以信道数据样本71a以及预测的虚拟信道数据71b作为正负样本，对机器学习模型A和机器学习模型B进行对抗学习方式的训练。
其中，上述预测的虚拟信道数据可以作为负样本，其对应的训练标签为第一标签，该第一标签可以指示该预测的虚拟信道数据是非自然存在的信道数据(或者说是通过模拟预测生成的信道数据)。
相应的，上述信道数据样本可以作为正样本，其对应的训练标签为第二标签，该第二标签可以指示信道数据样本是自然存在的信道数据。
在训练过程中，模型训练设备可以对机器学习模型A和机器学习模型B进行轮流训练。在对抗学习过程中，机器学习模型A在训练初始时输出的预测的虚拟信道数据的准确性不够高，并且，机器学习模型B对输入的信道数据是否为自然存在的信道数据进行判断的准确性也不够高，随着对抗学习的推进，机器学习模型B的判断准确性越来越高，相应的，机器学习模型A生成的预测的虚拟信道数据也越来越接近自然存在的信道数据，当两个机器学习模型都趋近于收敛时，机器学习模型B能够准确的判断出信道数据样本为自然存在的信道数据，但是无法准确的区分预测的虚拟信道数据是否为自然存在的信道数据，此时可以认为机器学习模型A生成的预测的虚拟信道数据足够接近自然存在的信道数据。
步骤4，当机器学习模型A和机器学习模型B都收敛后，模型训练设备将机器学习模型A输出为信道生成模型72；该信道生成模型可以部署至信道数据生成设备73中。
步骤5，在模型应用阶段信道，信道数据生成设备73通过该信道生成模型72生成虚拟信道数据72a。
其中，上述模型训练设备和信道数据生成设备可以实现为同一个实体设备，比如，可以实现为同一台个人电脑、工作站或者服务器。
或者，上述模型训练设备和信道数据生成设备也可以实现为不同的实体设备。比如，上述模型训练设备可以实现为开发人员使用的个人电脑、工作站或者服务器，上述数据生成设备可以是无线通信系统的设计人员使用的个人电脑、工作站或者服务器。
请参考图8，其示出了本申请一个实施例提供的信道数据处理及信道数据生成的方法流程图。该方法可以由计算机设备执行，比如，可以由模型训练设备和信道数据生成设备交互执行；该方法可以包括如下几个步骤：
步骤801，在模型训练阶段，模型训练设备获取信道数据样本；信道数据样本用于表征样本信道环境中的信道情况。
在本申请实施例中，在模型训练阶段，开发人员可以预先收集若干信道数据样本，并将收集到的信道数据样本输入至模型训练设备。
其中，上述信道数据样本可以是实际信道环境中采集得到的信道数据，也可以是通过人工或者机器构建，且被认为是自然存在的信道数据。
在获取到信道数据样本，以及开发人员预先构建并进行参数初始化(比如随机设置参数)之后的信道生成模型以及信道鉴别模型之后，即可以以信道生成模型为生成器，以信道鉴别模型为判别器，基于信道数据样本，通过生成对抗学习的方式对信道生成模型和信道鉴别模型进行训练。该训练过程可以参考线束步骤802至步骤807。
步骤802，在对信道鉴别模型进行训练的阶段，模型训练设备通过信道生成模型生成预 测的虚拟信道数据。
在一种可能的实现方式中，上述信道生成模型包括以下四种网络中的至少一种：全连接网络、卷积神经网络、残差网络、或者自注意力机制网络。
在一种可能的实现方式中，模型训练设备可以将输入信息输入至信道生成模型，获得信道生成模型对输入信息样本进行处理后输出的预测的虚拟信道数据。
其中，信道生成模型可以具有输入端口，在模型训练和应用过程中，信道生成模型可以对输入的信息进行逐层处理，最终输出满足一定数据格式的数据，作为虚拟信道数据。
在一种可能的实现方式中，输入信息包括以下四种信息中的至少一种：
噪声信息、随机数信息、信道类型指示信息、或者信道数据样本信息；
其中，信道类型指示信息用于指示信道类型；
信道数据样本信息是基于信道数据样本构建的信息。
在本申请实施例的一个示例性的方案中，信道生成模型的输入可以是任意输入，比如，任意的噪声，或者任意的随机数，信道生成模型通过任意输入进行触发，即可以进行后续的逐层处理，最终输出虚拟信道数据，其中，在模型训练阶段输出的虚拟信道数据，即为上述预测的虚拟信道数据。
其中，上述的噪声信息可以来自于真实环境，也可以由人工产生。
上述的随机数信息可以是随机数序列，或者伪随机数序列。
其中，上述噪声信息或者随机数信息的格式可以是一维向量，或者二维矩阵，或者高维的噪声或者随机数集合。上述噪声信息和随机数信息的格式可以提前约定，或者和待生成的虚拟信道数据格式一致。
在本申请实施例的另一个示例性的方案中，信道生成模型的输入也可以是具有指定含义的信息，比如，可以是指示某种信道类型的信息(目的是使得输出的虚拟信道数据可以模拟对应信道类型的虚拟信道)，或者，也可以是基于信道数据样本衍生出的信息(目的是使得输出的虚拟信道数据对应的虚拟信道的信道环境与信道数据样本对应的信道环境相近，或者说，使得输出的虚拟信道数据对应的虚拟信道的信道环境是在信道数据样本对应的信道环境的基础上改进得到的)。
在一种可能的实现方式中，在模型训练阶段，上述信道类型指示信息用于指示与信道数据样本相对应的信道类型。
在本申请实施例中，在模型训练阶段，上述信道类型指示信息所指示的信道类型，可以与信道数据样本相对应的信道类型相一致，这样在混入随机数或者随机噪声的情况下，可以在训练过程以及后续的应用中，使得信道生成模型生成多种与输入的信道类型指示信息相匹配的虚拟信道数据。
在一种可能的实现方式中，信道类型指示信息包括以下五种信息中的至少一种：
时域特征信息、频率特征信息、空间域特征信息、环境特征信息、或者场景特征信息。
在本申请实施例中，上述信道类型指示信息可以指示信道对应的频率信息、环境信息、场景信息，例如：高频、低频、室内、室外、密集小区、空旷外场、物联网场景、工业场景等。
其中，上述时域特征信息、频率特征信息、空间域特征信息可以称为信道的指标特征信息，例如：时延功率谱信息、多径信息、角度信息、速度信息等。
上述环境特征信息可以指示室内环境、室外环境、空旷野外等等。
上述场景特征信息可以指示视线传输(Line Of Sight，LOS)、非视线传输(Not Line Of Sight，NLOS)、高速、低速等场景类别。
在一种可能的实现方式中，信道数据样本信息包括以下三种信息中的至少一种：
在信道数据样本中混合噪声后得到的信息；
在信道数据样本中混合随机数后得到的信息；
或者，在信道数据样本中混合噪声和随机数后得到的信息。
例如，模型训练设备可以混合噪声和信道数据样本作为信道生成模型的输入，或者混合随机数和信道数据样本作为信道生成模型的输入等等。
在一种可能的实现方式中，虚拟信道数据包括与至少一个维度上的至少一个粒度分别对应的信道数据。
其中，当虚拟信道数据包括与至少两个维度上的各个粒度分别对应的信道数据时，虚拟信道数据包括至少两个维度的矩阵；或者，当虚拟信道数据包括与至少两个维度上的各个粒度分别对应的信道数据时，虚拟信道数据包括与至少两个维度上的各个粒度分别对应的信道数据排列得到的一维数据。
其中，以虚拟信道数据包括与两个维度上的各个粒度分别对应的信道数据为例，上述虚拟信道数据的单个样本可以由大小为M*N的矩阵构成，其在第一维度上有M个第一粒度，在第二维度上有N个第二粒度，M和N可以相等也可以不相等，矩阵内具体的数值指示代表信道质量。此外，也可以将M*N的两维数据合成成为1*(M*N)大小或者(M*N)*1大小的一维数据，具体变换可以是先第一维度再第二维度，也可以是先第二维度再第一维度，这种变换是表述形式上的区别。
在一种可能的实现方式中，至少一个维度包括以下四种维度中的至少一种：
频域维度、时域维度、空间域维度、或者实虚部维度。
在一种可能的实现方式中，当至少一个维度包括频域维度时，频域维度上的一个粒度包括：
至少一个无线承载(Radio Bearer，RB)或者至少一个子载波。
示例性的，虚拟信道数据的单个样本由粒度为m的第一维度构成，第一维度可以是频域维度，当第一维度是频域维度时，粒度m可以是a个RB(a大于等于1，例如2RB，4RB，8RB)，或者可以是b个子载波(b大于1，例如4个子载波，6个子载波，18个子载波)。当第一维度是频域维度时，虚拟信道数据的单个样本所指示的频域范围是M*m的频域范围。
在一种可能的实现方式中，当至少一个维度包括时域维度时，时域维度上的一个粒度包括：
p1微秒、至少一个符号长度或者至少一个符号的采样点个数；其中，p1为正数。
示例性的，虚拟信道数据的单个样本也可由粒度为p的第一维度构成，第一维度可以是时域维度，当第一维度是时域维度时，粒度p可以是时延粒度，例如一个时延粒度是p1个微秒、或者p2个符号长度、或者p3个符号的采样点个数，这里说的符号可以是一个正交频分复用(Orthogonal Frequency Division Multiplexing，OFDM)符号。当第一维度是时域维度时，训练集合的单个样本所指示的时域范围(或者说时延范围)是M*p的时域范围。
在一种可能的实现方式中，当至少一个维度包括空间域维度时，空间域维度上的一个粒度包括：
至少一对收发天线、至少一个接收天线、至少一个发射天线、或者目标角度间隔的角度范围。
示例性的，虚拟信道数据的单个样本由粒度为n的第二维度构成，第二维度可以是空间域维度，具体地可以是天线维度，例如第二维度上由N个天线对构成，第二粒度是一对收发天线。
示例性的，虚拟信道数据的单个样本也可由粒度为q的第二维度构成，第二维度可以是空间域维度，具体地可以是角度域维度，例如第二维度上由N个角度构成，第二粒度是上述N个角度之间的角度间隔大小。
虚拟信道数据的单个样本中，某一个特定维度组合上的信道质量指示，可以代表该特定维度组合下的信道质量指示情况。
例如，请参考图9，其示出了本申请实施例涉及的一种虚拟信道数据结构示意图。在M*N 的矩阵中，第3行第6列上的指示值X可以用来表示第六个空间粒度上的第三个特定粒度带宽上的信道质量情况。
再例如，请参考图10，其示出了本申请实施例涉及的另一种虚拟信道数据结构示意图。在M*N的矩阵中，第4行第5列上的指示值Y可以用来表示第5个空间粒度(例如到达角度)上的第4个特定粒度时延的信道质量情况。
由于虚拟信道以及虚拟信道得到的信道特征信息都可以是通过复数来呈现的，所以上述信道生成模型的输出可以在上述描述的内容基础上额外多一个维度，该维度是将虚拟信道(或者虚拟信道得到的信道特征信息)的虚部和实部数据独立呈现所造成的。例如上述除了第一维度和第二维度外，还可以有第三维度，第三维度来自于信道的实部和虚部。
此外，还需要注意的是，上述信道生成模型的输出还可以是在上述第一维度、第二维度、第三维度基础上的拆分与组合，例如当第二维度是天线对维度时，还可以拆分成为发送天线子维度和接收天线子维度，从而扩展上述虚拟信道输出形式的维度。
上述描述中，为了描述简单起见，都以第一维度和第二维度构成的两维虚拟信道作为举例，本申请实施例涉及的虚拟信道的维度不局限在二维。
在一种可能的实现方式中，虚拟信道数据包括以下两种信息中的至少一种：原始的信道信息，或者，信道特征向量；
信道特征向量是对原始的信道信息进行数据变换得到的。
在一种可能的实现方式中，原始的信道信息包括信道质量信息。
在一种可能的实现方式中，信道特征向量包括对原始的信道信息进行奇异值分解(Singular Value Decomposition，SVD)得到的。
上述信道生成模型的输出信息，还可以是上述原始信道信息通过数学变换后得到的信道特征信息，例如通过SVD分解得到的信道特征向量信息，可以是单流的信道特征向量信息，也可以是多流的信道特征向量信息，例如2流、4流、8流信道特征向量信息。
步骤803，模型训练设备将预测的虚拟信道数据和信道数据样本输入信道鉴别模型，获得信道鉴别模型的第一鉴别结果；第一鉴别结果用于指示预测的虚拟信道数据和信道数据样本是否为模型生成的信道数据。
在一种可能的实现方式中，上述信道鉴别模型包括以下四种网络中的一种或者多种：全连接网络、卷积神经网络、残差网络、或者自注意力机制网络。
请参考图11，其示出了本申请实施例涉及的一种信道生成模型以及信道鉴别模型的模型架构图。
步骤804，模型训练设备根据第一鉴别结果对信道鉴别模型的模型参数进行更新。
在本申请实施例中，对信道鉴别模型进行训练的阶段，模型训练设备可以通过第一鉴别结果，以及预测的虚拟信道数据和信道数据样本各自的标签计算出损失函数值，然后通过损失函数值对信道鉴别模型的模型参数进行更新。
步骤805，在对信道生成模型进行训练的阶段，模型训练设备通过信道生成模型生成预测的虚拟信道数据。
步骤806，模型训练设备将预测的虚拟信道数据输入信道鉴别模型，获得信道鉴别模型的第二鉴别结果；第二鉴别结果用于指示预测的虚拟信道数据是否为模型生成的信道数据。
步骤807，模型训练设备根据第二鉴别结果对信道生成模型的模型参数进行更新。
在本申请实施例中，对信道生成模型进行训练的阶段，模型训练设备可以通过第二鉴别结果，以及预测的虚拟信道数据的标签计算出损失函数值，然后通过损失函数值对信道生成模型的模型参数进行更新。
迭代执行上述步骤802至步骤807，直至两个模型均收敛。
上述生成器和判别器通过神经网络训练达到稳定状态时，可单独提取出生成器用于虚拟信道数据的生成。
对于信道生成模型的构建，需要同时构建信道生成模型和信道鉴别模型两部分，其中信道生成模型用于生成虚拟信道数据，信道鉴别模型用于判断虚拟信道与真实信道之间的差别。
模型训练设备可利用生成对抗网络(Generative Adversarial Networks，GAN)作为信道生成模型和信道鉴别模型的基础结构，让输入信息(例如随机数)作为信道生成模型的输入，让给定的神经网络结构作为信道生成模型的基本结构，产生当前的虚拟信道输入。继而由信道鉴别模型判断虚拟信道与真实信道之间的差别，如果能判断出差别，则继续循环上述过程，更新信道生成模型参数，产生新的虚拟信道输出，直至信道鉴别模型无法判断虚拟信道与真实信道之间差别时，既可认为信道生成模型完成构建，可生成拟合真实信道的虚拟信道。
基于上述信道生成模型的构建，可以仅利用少量真实信道信息协助信道鉴别模型构建，最终基于信道生成模型产生大量虚拟信道数据。
本方案可以用于解决在复杂频率、场景、环境下传统信道建模和信道估计难以获得大量数据集，难以对复杂非线性信道模型实现有效拟合的问题。通过本方案的设计，可实现仅依赖少量实际数据，既可以构建信道生成模型，从而基于信道生成模型构建大量虚拟信道数据，大大节省人工实采数据的难度与开销，也规避了传统数据建模在复杂信道下的实效问题。上述虚拟信道数据可以用于基于人工智能的无线通信解决方案在多频段、多场景、多环境下快速构建数据集以支持模型按需再训练与更新。
步骤808，在模型应用阶段信道，信道数据生成设备通过该信道生成模型生成虚拟信道数据。
在一种可能的实现方式中，通过信道生成模型生成虚拟信道数据，包括：
将输入信息输入至信道生成模型，获得信道生成模型对输入信息进行处理后输出的虚拟信道数据。
其中，上述信道生成模型的输入和输出可以参考上述步骤802下的描述，此处不再赘述。
在本申请实施例中，信道数据生成设备中的信道生成模型可以封装在信道数据生成器中。
在一种可能的实现方式中，上述信道数据生成器可以设置有输入接口，该输入接口的输入为上述信道生成模型的输入信息，此处不再赘述。
在另一种可能的实现方式中，当信道生成模型的输入信息为噪声信息和/或随机数信息的情况下，也可以在信道数据生成器中封装噪声生成器和/或随机数生成器，此时，信道数据生成器可以是无输入的，或者说，信道数据生成器无需额外输入噪声信息和/或随机数信息。
或者，在另一种可能的实现方式中，也可以在信道生成模型中封装噪声生成器和/或随机数生成器，也就是说，当信道生成模型的处理信息为噪声信息和/或随机数信息的情况下，该信道生成模型可以是无输入的，此时，信道生成模型自动生成噪声信息和/或随机数信息，并对噪声信息和/或随机数信息进行处理并输出虚拟信道数据。
本申请提供了一种构建虚拟信道的方法，用虚拟信道替代实际信道，用虚拟环境替代实际环境，从而降低人工智能与无线通信系统融合研究和开发时对于实际信道环境数据的依赖程度。
本申请提供的虚拟信道环境的构建依赖信道生成模型实现，信道生成模型可以生成一种或者多种频段、场景、环境下所对应的无线信道数据。上述通过信道生成模型生成的无线信道数据可用于构建无线通信系统AI解决方案的数据集、或者用于无线通信系统的信道分析与建模。
其中，信道生成模型输出的虚拟信道数据可用于模拟不同频率、环境、场景下的信道信息，例如：高频、低频、室内、室外、密集小区、空旷外场、物联网场景、工业场景等。
请参考图12，其示出了本申请一个实施例提供的信道数据生成装置的框图。该装置具有实现上述图5或图8所示的方法中，由信道数据生成设备执行的功能。如图12所示，该装置可以包括：
生成模块1201，用于通过信道生成模型生成虚拟信道数据；所述虚拟信道数据用于表征信道环境中的信道情况；
其中，所述信道生成模型是通过信道数据样本进行机器学习训练得到的机器学习模型。
在一种可能的实现方式中，所述生成模块1201，用于将输入信息输入至所述信道生成模型，获得所述信道生成模型对所述输入信息进行处理后输出的所述虚拟信道数据。
在一种可能的实现方式中，所述输入信息包括以下四种信息中的至少一种：
噪声信息、随机数信息、信道类型指示信息、或者信道数据样本信息；
其中，所述信道类型指示信息用于指示信道类型；
所述信道数据样本信息是基于所述信道数据样本构建的信息。
在一种可能的实现方式中，所述信道类型指示信息包括以下五种信息中的至少一种：
时域特征信息、频率特征信息、空间域特征信息、环境特征信息、或者场景特征信息。
在一种可能的实现方式中，所述信道数据样本信息包括以下三种信息中的至少一种：
在所述信道数据样本中混合噪声后得到的信息；
在所述信道数据样本中混合随机数后得到的信息；
或者，在所述信道数据样本中混合噪声和随机数后得到的信息。
在一种可能的实现方式中，所述虚拟信道数据包括与至少一个维度上的至少一个粒度分别对应的信道数据。
在一种可能的实现方式中，所述至少一个维度包括以下四种维度中的至少一种：
频域维度、时域维度、空间域维度、或者实虚部维度。
在一种可能的实现方式中，当所述至少一个维度包括频域维度时，所述频域维度上的一个粒度包括：
至少一个RB或者至少一个子载波。
在一种可能的实现方式中，当所述至少一个维度包括时域维度时，所述时域维度上的一个粒度包括：
p1微秒、至少一个符号长度或者至少一个符号的采样点个数；其中，p1为正数。
在一种可能的实现方式中，当所述至少一个维度包括空间域维度时，所述空间域维度上的一个粒度包括：
至少一对收发天线、至少一个接收天线、至少一个发射天线、或者目标角度间隔的角度范围。
在一种可能的实现方式中，当所述虚拟信道数据包括与至少两个维度上的各个粒度分别对应的信道数据时，
所述虚拟信道数据包括至少两个维度的矩阵；
或者，所述虚拟信道数据包括与所述至少两个维度上的各个粒度分别对应的信道数据排列得到的一维数据。
在一种可能的实现方式中，所述虚拟信道数据包括以下两种信息中的至少一种：原始的信道信息，或者，信道特征向量；
所述信道特征向量是对所述原始的信道信息进行数据变换得到的。
在一种可能的实现方式中，所述原始的信道信息包括信道质量信息。
在一种可能的实现方式中，所述信道特征向量包括对所述原始的信道信息进行奇异值分解得到的。
在一种可能的实现方式中，所述信道生成模型包括以下四种网络中的至少一种：
全连接网络、卷积神经网络、残差网络、或者自注意力机制网络。
综上所述，本申请实施例所示的方案，预先通过信道数据样本训练出一个信道生成模型，通过该信道生成模型，可以通过模拟、预测的方式自动生成信道环境对应的虚拟信道数据，不需要进行实际采集即可以快速获得各种信道环境对应的信道数据，从而极大的提高了各种 信道环境下的信道数据的获取效率，进而提高信道建模的效果，并提高无线通信系统研究设计的准确性。
请参考图13，其示出了本申请一个实施例提供的信道数据生成装置的框图。该装置具有实现上述图6或图8所示的方法中，由模型训练设备执行的功能。如图13所示，该装置可以包括：
获取模块1301，用于获取信道数据样本；所述信道数据样本用于表征样本信道环境中的信道情况；
训练模块1302，用于以信道生成模型为生成器，以信道鉴别模型为判别器，基于所述信道数据样本，通过生成对抗学习的方式对所述信道生成模型和所述信道鉴别模型进行训练；
其中，训练至收敛后的所述信道生成模型用于生成虚拟信道数据；所述虚拟信道数据用于表征信道环境中的信道情况。
在一种可能的实现方式中，所述训练模块1302，用于，
在对所述信道鉴别模型进行训练的阶段，通过所述信道生成模型生成预测的虚拟信道数据；
将所述预测的虚拟信道数据和所述信道数据样本输入所述信道鉴别模型，获得所述信道鉴别模型的第一鉴别结果；所述第一鉴别结果用于指示所述预测的虚拟信道数据和所述信道数据样本是否为模型生成的信道数据；
根据所述第一鉴别结果对所述信道鉴别模型的模型参数进行更新。
在一种可能的实现方式中，所述训练模块1302，用于，
在对所述信道生成模型进行训练的阶段，通过所述信道生成模型生成预测的虚拟信道数据；
将所述预测的虚拟信道数据输入所述信道鉴别模型，获得所述信道鉴别模型的第二鉴别结果；所述第二鉴别结果用于指示所述预测的虚拟信道数据是否为模型生成的信道数据；
根据所述第二鉴别结果对所述信道生成模型的模型参数进行更新。
在一种可能的实现方式中，所述训练模块1302，用于将输入信息输入至所述信道生成模型，获得所述信道生成模型对所述输入信息样本进行处理后输出的所述预测的虚拟信道数据。
在一种可能的实现方式中，所述输入信息包括以下四种信息中的至少一种：
噪声信息、随机数信息、信道类型指示信息、或者信道数据样本信息；
其中，所述信道类型指示信息用于指示信道类型；
所述信道数据样本信息是基于所述信道数据样本构建的信息。
在一种可能的实现方式中，所述信道类型指示信息用于指示与所述信道数据样本相对应的信道类型。
在一种可能的实现方式中，所述信道类型指示信息包括以下五种信息中的至少一种：
时域特征信息、频率特征信息、空间域特征信息、环境特征信息、或者场景特征信息。
在一种可能的实现方式中，所述信道数据样本信息包括以下三种信息中的至少一种：
在所述信道数据样本中混合噪声后得到的信息；
在所述信道数据样本中混合随机数后得到的信息；
或者，在所述信道数据样本中混合噪声和随机数后得到的信息。
综上所述，本申请实施例所示的方案，预先通过生成对抗学习的方式，基于信道数据样本训练出一个信道生成模型，通过该信道生成模型，可以通过模拟、预测的方式自动生成信道环境对应的虚拟信道数据，不需要进行实际采集即可以快速获得各种信道环境对应的信道数据，从而极大的提高了各种信道环境下的信道数据的获取效率，进而提高信道建模的效果，并提高无线通信系统研究设计的准确性。
需要说明的一点是，上述实施例提供的装置在实现其功能时，仅以上述各个功能模块的划分进行举例说明，实际应用中，可以根据实际需要而将上述功能分配由不同的功能模块完成，即将设备的内容结构划分成不同的功能模块，以完成以上描述的全部或者部分功能。
关于上述实施例中的装置，其中各个模块执行操作的具体方式已经在有关该方法的实施例中进行了详细描述，此处将不做详细阐述说明。
图14示出了本申请一示例性实施例示出的计算机设备1400的结构框图。所述计算机设备1400包括中央处理单元(Central Processing Unit，CPU)1401、包括随机存取存储器(Random Access Memory，RAM)1402和只读存储器(Read-Only Memory，ROM)1403的系统存储器1404，以及连接系统存储器1404和中央处理单元1401的系统总线1405。所述计算机设备1400还包括帮助计算机内的各个器件之间传输信息的基本输入/输出系统(Input/Output系统，I/O系统)1406，和用于存储操作系统1413、应用程序1414和其他程序模块1415的大容量存储设备1407。
所述基本输入/输出系统1406包括有用于显示信息的显示器1408和用于用户输入信息的诸如鼠标、键盘之类的输入设备1409。其中所述显示器1408和输入设备1409都通过连接到系统总线1405的输入输出控制器1410连接到中央处理单元1401。所述基本输入/输出系统1406还可以包括输入输出控制器1410以用于接收和处理来自键盘、鼠标、或电子触控笔等多个其他设备的输入。类似地，输入输出控制器1410还提供输出到显示屏、打印机或其他类型的输出设备。
所述大容量存储设备1407通过连接到系统总线1405的大容量存储控制器(未示出)连接到中央处理单元1401。所述大容量存储设备1407及其相关联的计算机可读介质为计算机设备1400提供非易失性存储。也就是说，所述大容量存储设备1407可以包括诸如硬盘或者只读光盘(Compact Disc Read-Only Memory，CD-ROM)驱动器之类的计算机可读介质(未示出)。
不失一般性，所述计算机可读介质可以包括计算机存储介质和通信介质。计算机存储介质包括以用于存储诸如计算机可读指令、数据结构、程序模块或其他数据等信息的任何方法或技术实现的易失性和非易失性、可移动和不可移动介质。计算机存储介质包括RAM、ROM、可擦除可编程只读寄存器(Erasable Programmable Read Only Memory，EPROM)、电子抹除式可复写只读存储器(Electrically-Erasable Programmable Read-Only Memory，EEPROM)闪存或其他固态存储其技术，CD-ROM、数字多功能光盘(Digital Versatile Disc，DVD)或其他光学存储、磁带盒、磁带、磁盘存储或其他磁性存储设备。当然，本领域技术人员可知所述计算机存储介质不局限于上述几种。上述的系统存储器1404和大容量存储设备1407可以统称为存储器。
根据本公开的各种实施例，所述计算机设备1400还可以通过诸如因特网等网络连接到网络上的远程计算机运行。也即计算机设备1400可以通过连接在所述系统总线1405上的网络接口单元1411连接到网络1412，或者说，也可以使用网络接口单元1411来连接到其他类型的网络或远程计算机系统(未示出)。
所述存储器还包括至少一条计算机指令，所述至少一条计算机指令存储于存储器中，中央处理器1401通过执行该至少一条指令、至少一段程序、代码集或指令集来实现上述各个实施例所示的方法中，由模型训练设备或者信道数据生成设备执行的全部或者部分步骤。
本申请实施例还提供了一种计算机可读存储介质，所述存储介质中存储有计算机程序，所述计算机程序由处理器加载并执行以实现上述图5、图6或者图8所示的方法中，由模型训练设备或者信道数据生成设备执行的全部或者部分步骤。
本申请还提供了一种芯片，该芯片用于在计算机设备中运行，以使得计算机设备执行上 述图5、图6或者图8所示的方法中，由模型训练设备或者信道数据生成设备执行的全部或者部分步骤。
本申请还提供了一种计算机程序产品，该计算机程序产品或计算机程序包括计算机指令，该计算机指令存储在计算机可读存储介质中。计算机设备的处理器从计算机可读存储介质读取该计算机指令，处理器执行该计算机指令，使得计算机设备执行上述图5、图6或者图8所示的方法中，由模型训练设备或者信道数据生成设备执行的全部或者部分步骤。
本申请还提供了一种计算机程序，该计算机程序由计算机设备的处理器执行，以实现上述图5、图6或者图8所示的方法中，由模型训练设备或者信道数据生成设备执行的全部或者部分步骤。
本领域技术人员应该可以意识到，在上述一个或多个示例中，本申请实施例所描述的功能可以用硬件、软件、固件或它们的任意组合来实现。当使用软件实现时，可以将这些功能存储在计算机可读介质中或者作为计算机可读介质上的一个或多个指令或代码进行传输。计算机可读介质包括计算机存储介质和通信介质，其中通信介质包括便于从一个地方向另一个地方传送计算机程序的任何介质。存储介质可以是通用或专用计算机能够存取的任何可用介质。
以上所述仅为本申请的示例性实施例，并不用以限制本申请，凡在本申请的精神和原则之内，所作的任何修改、等同替换、改进等，均应包括在本申请的保护范围之内。

Claims (51)

  1. 一种信道数据生成方法，其特征在于，所述方法包括：
    通过信道生成模型生成虚拟信道数据；所述虚拟信道数据用于表征信道环境中的信道情况；
    其中，所述信道生成模型是通过信道数据样本进行机器学习训练得到的机器学习模型。
  2. 根据权利要求1所述的方法，其特征在于，所述通过信道生成模型生成虚拟信道数据，包括：
    将输入信息输入至所述信道生成模型，获得所述信道生成模型对所述输入信息进行处理后输出的所述虚拟信道数据。
  3. 根据权利要求2所述的方法，其特征在于，所述输入信息包括以下四种信息中的至少一种：
    噪声信息、随机数信息、信道类型指示信息、或者信道数据样本信息；
    其中，所述信道类型指示信息用于指示信道类型；
    所述信道数据样本信息是基于所述信道数据样本构建的信息。
  4. 根据权利要求3所述的方法，其特征在于，所述信道类型指示信息包括以下五种信息中的至少一种：
    时域特征信息、频率特征信息、空间域特征信息、环境特征信息、或者场景特征信息。
  5. 根据权利要求3所述的方法，其特征在于，所述信道数据样本信息包括以下三种信息中的至少一种：
    在所述信道数据样本中混合噪声后得到的信息；
    在所述信道数据样本中混合随机数后得到的信息；
    或者，在所述信道数据样本中混合噪声和随机数后得到的信息。
  6. 根据权利要求1至5任一所述的方法，其特征在于，所述虚拟信道数据包括与至少一个维度上的至少一个粒度分别对应的信道数据。
  7. 根据权利要求6所述的方法，其特征在于，所述至少一个维度包括以下四种维度中的至少一种：
    频域维度、时域维度、空间域维度、或者实虚部维度。
  8. 根据权利要求7所述的方法，其特征在于，当所述至少一个维度包括频域维度时，所述频域维度上的一个粒度包括：
    至少一个RB或者至少一个子载波。
  9. 根据权利要求7所述的方法，其特征在于，当所述至少一个维度包括时域维度时，所述时域维度上的一个粒度包括：
    p1微秒、至少一个符号长度或者至少一个符号的采样点个数；其中，p1为正数。
  10. 根据权利要求7所述的方法，其特征在于，当所述至少一个维度包括空间域维度时，所述空间域维度上的一个粒度包括：
    至少一对收发天线、至少一个接收天线、至少一个发射天线、或者目标角度间隔的角度范围。
  11. 根据权利要求6所述的方法，其特征在于，当所述虚拟信道数据包括与至少两个维度上的各个粒度分别对应的信道数据时，
    所述虚拟信道数据包括至少两个维度的矩阵；
    或者，所述虚拟信道数据包括与所述至少两个维度上的各个粒度分别对应的信道数据排列得到的一维数据。
  12. 根据权利要求1至11任一所述的方法，其特征在于，所述虚拟信道数据包括以下两种信息中的至少一种：原始的信道信息，或者，信道特征向量；
    所述信道特征向量是对所述原始的信道信息进行数据变换得到的。
  13. 根据权利要求12所述的方法，其特征在于，所述原始的信道信息包括信道质量信息。
  14. 根据权利要求12所述的方法，其特征在于，所述信道特征向量包括对所述原始的信道信息进行奇异值分解得到的。
  15. 根据权利要求1至14任一所述的方法，其特征在于，所述信道生成模型包括以下四种网络中的至少一种：
    全连接网络、卷积神经网络、残差网络、或者自注意力机制网络。
  16. 一种信道数据处理方法，其特征在于，所述方法包括：
    获取信道数据样本；所述信道数据样本用于表征样本信道环境中的信道情况；
    以信道生成模型为生成器，以信道鉴别模型为判别器，基于所述信道数据样本，通过生成对抗学习的方式对所述信道生成模型和所述信道鉴别模型进行训练；
    其中，训练至收敛后的所述信道生成模型用于生成虚拟信道数据；所述虚拟信道数据用于表征信道环境中的信道情况。
  17. 根据权利要求16所述的方法，其特征在于，所述以信道生成模型为生成器，以信道鉴别模型为判别器，基于所述信道数据样本，通过生成对抗学习的方式对所述信道生成模型和所述信道鉴别模型进行训练，包括：
    在对所述信道鉴别模型进行训练的阶段，通过所述信道生成模型生成预测的虚拟信道数据；
    将所述预测的虚拟信道数据和所述信道数据样本输入所述信道鉴别模型，获得所述信道鉴别模型的第一鉴别结果；所述第一鉴别结果用于指示所述预测的虚拟信道数据和所述信道数据样本是否为模型生成的信道数据；
    根据所述第一鉴别结果对所述信道鉴别模型的模型参数进行更新。
  18. 根据权利要求16所述的方法，其特征在于，所述以信道生成模型为生成器，以信道鉴别模型为判别器，基于所述信道数据样本，通过生成对抗学习的方式对所述信道生成模型和所述信道鉴别模型进行训练，包括：
    在对所述信道生成模型进行训练的阶段，通过所述信道生成模型生成预测的虚拟信道数据；
    将所述预测的虚拟信道数据输入所述信道鉴别模型，获得所述信道鉴别模型的第二鉴别结果；所述第二鉴别结果用于指示所述预测的虚拟信道数据是否为模型生成的信道数据；
    根据所述第二鉴别结果对所述信道生成模型的模型参数进行更新。
  19. 根据权利要求17或18所述的方法，其特征在于，所述通过所述信道生成模型生成预测的虚拟信道数据，包括：
    将输入信息输入至所述信道生成模型，获得所述信道生成模型对所述输入信息样本进行处理后输出的所述预测的虚拟信道数据。
  20. 根据权利要求19所述的方法，其特征在于，所述输入信息包括以下四种信息中的至少一种：
    噪声信息、随机数信息、信道类型指示信息、或者信道数据样本信息；
    其中，所述信道类型指示信息用于指示信道类型；
    所述信道数据样本信息是基于所述信道数据样本构建的信息。
  21. 根据权利要求20所述的方法，其特征在于，所述信道类型指示信息用于指示与所述信道数据样本相对应的信道类型。
  22. 根据权利要求20所述的方法，其特征在于，所述信道类型指示信息包括以下五种信息中的至少一种：
    时域特征信息、频率特征信息、空间域特征信息、环境特征信息、或者场景特征信息。
  23. 根据权利要求20所述的方法，其特征在于，所述信道数据样本信息包括以下三种信 息中的至少一种：
    在所述信道数据样本中混合噪声后得到的信息；
    在所述信道数据样本中混合随机数后得到的信息；
    或者，在所述信道数据样本中混合噪声和随机数后得到的信息。
  24. 一种信道数据生成装置，其特征在于，所述装置包括：
    生成模块，用于通过信道生成模型生成虚拟信道数据；所述虚拟信道数据用于表征信道环境中的信道情况；
    其中，所述信道生成模型是通过信道数据样本进行机器学习训练得到的机器学习模型。
  25. 根据权利要求24所述的装置，其特征在于，所述生成模块，用于将输入信息输入至所述信道生成模型，获得所述信道生成模型对所述输入信息进行处理后输出的所述虚拟信道数据。
  26. 根据权利要求25所述的装置，其特征在于，所述输入信息包括以下四种信息中的至少一种：
    噪声信息、随机数信息、信道类型指示信息、或者信道数据样本信息；
    其中，所述信道类型指示信息用于指示信道类型；
    所述信道数据样本信息是基于所述信道数据样本构建的信息。
  27. 根据权利要求26所述的装置，其特征在于，所述信道类型指示信息包括以下五种信息中的至少一种：
    时域特征信息、频率特征信息、空间域特征信息、环境特征信息、或者场景特征信息。
  28. 根据权利要求26所述的装置，其特征在于，所述信道数据样本信息包括以下三种信息中的至少一种：
    在所述信道数据样本中混合噪声后得到的信息；
    在所述信道数据样本中混合随机数后得到的信息；
    或者，在所述信道数据样本中混合噪声和随机数后得到的信息。
  29. 根据权利要求24至28任一所述的装置，其特征在于，所述虚拟信道数据包括与至少一个维度上的至少一个粒度分别对应的信道数据。
  30. 根据权利要求29所述的装置，其特征在于，所述至少一个维度包括以下四种维度中的至少一种：
    频域维度、时域维度、空间域维度、或者实虚部维度。
  31. 根据权利要求30所述的装置，其特征在于，当所述至少一个维度包括频域维度时，所述频域维度上的一个粒度包括：
    至少一个RB或者至少一个子载波。
  32. 根据权利要求30所述的装置，其特征在于，当所述至少一个维度包括时域维度时，所述时域维度上的一个粒度包括：
    p1微秒、至少一个符号长度或者至少一个符号的采样点个数；其中，p1为正数。
  33. 根据权利要求30所述的装置，其特征在于，当所述至少一个维度包括空间域维度时，所述空间域维度上的一个粒度包括：
    至少一对收发天线、至少一个接收天线、至少一个发射天线、或者目标角度间隔的角度范围。
  34. 根据权利要求29所述的装置，其特征在于，当所述虚拟信道数据包括与至少两个维度上的各个粒度分别对应的信道数据时，
    所述虚拟信道数据包括至少两个维度的矩阵；
    或者，所述虚拟信道数据包括与所述至少两个维度上的各个粒度分别对应的信道数据排列得到的一维数据。
  35. 根据权利要求24至34任一所述的装置，其特征在于，所述虚拟信道数据包括以下两 种信息中的至少一种：原始的信道信息，或者，信道特征向量；
    所述信道特征向量是对所述原始的信道信息进行数据变换得到的。
  36. 根据权利要求35所述的装置，其特征在于，所述原始的信道信息包括信道质量信息。
  37. 根据权利要求35所述的装置，其特征在于，所述信道特征向量包括对所述原始的信道信息进行奇异值分解得到的。
  38. 根据权利要求24至37任一所述的装置，其特征在于，所述信道生成模型包括以下四种网络中的至少一种：
    全连接网络、卷积神经网络、残差网络、或者自注意力机制网络。
  39. 一种信道数据处理装置，其特征在于，所述装置包括：
    获取模块，用于获取信道数据样本；所述信道数据样本用于表征样本信道环境中的信道情况；
    训练模块，用于以信道生成模型为生成器，以信道鉴别模型为判别器，基于所述信道数据样本，通过生成对抗学习的方式对所述信道生成模型和所述信道鉴别模型进行训练；
    其中，训练至收敛后的所述信道生成模型用于生成虚拟信道数据；所述虚拟信道数据用于表征信道环境中的信道情况。
  40. 根据权利要求39所述的装置，其特征在于，所述以信道生成模型为生成器，以信道鉴别模型为判别器，基于所述信道数据样本，通过生成对抗学习的方式对所述信道生成模型和所述信道鉴别模型进行训练，包括：
    在对所述信道鉴别模型进行训练的阶段，通过所述信道生成模型生成预测的虚拟信道数据；
    将所述预测的虚拟信道数据和所述信道数据样本输入所述信道鉴别模型，获得所述信道鉴别模型的第一鉴别结果；所述第一鉴别结果用于指示所述预测的虚拟信道数据和所述信道数据样本是否为模型生成的信道数据；
    根据所述第一鉴别结果对所述信道鉴别模型的模型参数进行更新。
  41. 根据权利要求39所述的装置，其特征在于，所述以信道生成模型为生成器，以信道鉴别模型为判别器，基于所述信道数据样本，通过生成对抗学习的方式对所述信道生成模型和所述信道鉴别模型进行训练，包括：
    在对所述信道生成模型进行训练的阶段，通过所述信道生成模型生成预测的虚拟信道数据；
    将所述预测的虚拟信道数据输入所述信道鉴别模型，获得所述信道鉴别模型的第二鉴别结果；所述第二鉴别结果用于指示所述预测的虚拟信道数据是否为模型生成的信道数据；
    根据所述第二鉴别结果对所述信道生成模型的模型参数进行更新。
  42. 根据权利要求40或41所述的装置，其特征在于，所述通过所述信道生成模型生成预测的虚拟信道数据，包括：
    将输入信息输入至所述信道生成模型，获得所述信道生成模型对所述输入信息样本进行处理后输出的所述预测的虚拟信道数据。
  43. 根据权利要求42所述的装置，其特征在于，所述输入信息包括以下四种信息中的至少一种：
    噪声信息、随机数信息、信道类型指示信息、或者信道数据样本信息；
    其中，所述信道类型指示信息用于指示信道类型；
    所述信道数据样本信息是基于所述信道数据样本构建的信息。
  44. 根据权利要求43所述的装置，其特征在于，所述信道类型指示信息用于指示与所述信道数据样本相对应的信道类型。
  45. 根据权利要求43所述的装置，其特征在于，所述信道类型指示信息包括以下五种信息中的至少一种：
    时域特征信息、频率特征信息、空间域特征信息、环境特征信息、或者场景特征信息。
  46. 根据权利要求43所述的装置，其特征在于，所述信道数据样本信息包括以下三种信息中的至少一种：
    在所述信道数据样本中混合噪声后得到的信息；
    在所述信道数据样本中混合随机数后得到的信息；
    或者，在所述信道数据样本中混合噪声和随机数后得到的信息。
  47. 一种计算机设备，其特征在于，所述计算机设备包括处理器、存储器和收发器；
    所述存储器中存储有计算机程序，所述处理器执行所述计算机程序，以使得所述计算机设备实现如上述权利要求1至15任一所示的信道数据生成方法，或者，实现如上述权利要求16至23任一所示的信道数据处理方法。
  48. 一种计算机可读存储介质，其特征在于，所述存储介质中存储有计算机程序，所述计算机程序用于被处理器执行，以实现如权利要求1至15任一所示的信道数据生成方法，或者，实现如上述权利要求16至23任一所示的信道数据处理方法。
  49. 一种芯片，其特征在于，所述芯片用于在计算机设备中运行，以使得所述计算机设备执行如权利要求1至15任一所示的信道数据生成方法，或者，实现如上述权利要求16至23任一所示的信道数据处理方法。
  50. 一种计算机程序产品，其特征在于，所述计算机程序产品包括计算机指令，所述计算机指令存储在计算机可读存储介质中；计算机设备的处理器从所述计算机可读存储介质读取所述计算机指令，并执行所述计算机指令，使得所述计算机设备执行如权利要求1至15任一所示的信道数据生成方法，或者，实现如上述权利要求16至23任一所示的信道数据处理方法。
  51. 一种计算机程序，其特征在于，所述计算机程序由计算机设备的处理器执行，以实现如权利要求1至15任一所示的信道数据生成方法，或者，实现如上述权利要求16至23任一所示的信道数据处理方法。
PCT/CN2021/128216 2021-11-02 2021-11-02 信道数据生成方法、装置、设备及存储介质 WO2023077272A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202180100752.3A CN117678172A (zh) 2021-11-02 2021-11-02 信道数据生成方法、装置、设备及存储介质
PCT/CN2021/128216 WO2023077272A1 (zh) 2021-11-02 2021-11-02 信道数据生成方法、装置、设备及存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2021/128216 WO2023077272A1 (zh) 2021-11-02 2021-11-02 信道数据生成方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2023077272A1 true WO2023077272A1 (zh) 2023-05-11

Family

ID=86240473

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/128216 WO2023077272A1 (zh) 2021-11-02 2021-11-02 信道数据生成方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN117678172A (zh)
WO (1) WO2023077272A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117320047A (zh) * 2023-11-29 2023-12-29 中国信息通信研究院 一种网络工作参数处理方法和设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110289927A (zh) * 2019-07-01 2019-09-27 上海大学 基于条件生成对抗网络的信道模拟实现方法
CN110875790A (zh) * 2019-11-19 2020-03-10 上海大学 基于生成对抗网络的无线信道建模实现方法
US20200334542A1 (en) * 2018-01-02 2020-10-22 Nokia Technologies Oy Channel modelling in a data transmission system
CN112422208A (zh) * 2020-11-06 2021-02-26 西安交通大学 未知信道模型下基于对抗式学习的信号检测方法
CN112821971A (zh) * 2021-02-01 2021-05-18 之江实验室 一种基于对抗式学习的时变信道信号检测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200334542A1 (en) * 2018-01-02 2020-10-22 Nokia Technologies Oy Channel modelling in a data transmission system
CN110289927A (zh) * 2019-07-01 2019-09-27 上海大学 基于条件生成对抗网络的信道模拟实现方法
CN110875790A (zh) * 2019-11-19 2020-03-10 上海大学 基于生成对抗网络的无线信道建模实现方法
CN112422208A (zh) * 2020-11-06 2021-02-26 西安交通大学 未知信道模型下基于对抗式学习的信号检测方法
CN112821971A (zh) * 2021-02-01 2021-05-18 之江实验室 一种基于对抗式学习的时变信道信号检测方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
LIU LIU, ZHANG JIANHUA, YUANYUAN FAN, LI Y U, JIACHI ZHANG: "Survey of application of machine learning in wireless channel modeling", JOURNAL ON COMMUNICATIONS, vol. 42, no. 2, 28 December 2021 (2021-12-28), pages 134 - 153, XP093063564, DOI: 10.11959/j.issn.1000−436x.2021001 *
YANG YANG; LI YANG; ZHANG WUXIONG; QIN FEI; ZHU PENGCHENG; WANG CHENG-XIANG: "Generative-Adversarial-Network-Based Wireless Channel Modeling: Challenges and Opportunities", IEEE COMMUNICATIONS MAGAZINE., IEEE SERVICE CENTER, PISCATAWAY., US, vol. 57, no. 3, 1 March 2019 (2019-03-01), US , pages 22 - 27, XP011713901, ISSN: 0163-6804, DOI: 10.1109/MCOM.2019.1800635 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117320047A (zh) * 2023-11-29 2023-12-29 中国信息通信研究院 一种网络工作参数处理方法和设备
CN117320047B (zh) * 2023-11-29 2024-02-06 中国信息通信研究院 一种网络工作参数处理方法和设备

Also Published As

Publication number Publication date
CN117678172A (zh) 2024-03-08

Similar Documents

Publication Publication Date Title
Huang et al. Deep learning for physical-layer 5G wireless techniques: Opportunities, challenges and solutions
Wang et al. Transfer learning promotes 6G wireless communications: Recent advances and future challenges
Pei et al. Federated learning encounters 6g wireless communication in the scenario of internet of things
Najla et al. Predicting device-to-device channels from cellular channel measurements: A learning approach
WO2023077272A1 (zh) 信道数据生成方法、装置、设备及存储介质
EP4443939A1 (en) Communication method and device
Koufos et al. Trends in intelligent communication systems: review of standards, major research projects, and identification of research gaps
Eller et al. A deep learning network planner: Propagation modeling using real-world measurements and a 3D city model
CN115243293A (zh) 网络优化模型的确定方法、装置、电子设备及存储介质
Wang et al. Super resolution of wireless channel characteristics: A multi-task learning model
Huangfu et al. Wair-d: Wireless ai research dataset
Hassan Wireless and mobile networking
Sharma et al. Sixth Generation (6G) to the Waying Seventh (7G) Wireless Communication Visions and Standards, Challenges, Applications
Brandes et al. RF waveform synthesis guided by deep reinforcement learning
Huang et al. An energy harvesting algorithm for UAV-assisted TinyML consumer electronic in low-power IoT networks
Gao et al. Localization-Oriented Digital Twinning in 6G: A New Indoor-Positioning Paradigm and Proof-of-Concept
Zhao et al. Design of jiutian intelligent network simulation platform
Boas et al. Machine learning for CSI recreation in the digital twin based on prior knowledge
Fischione et al. Data sets for machine learning in wireless communications and networks
da Costa et al. 6G: Vision, Applications, and Challenges
Bora et al. Big data and deep learning for Stochastic Wireless channel
Chen et al. Over-the-air computing aided federated learning and analytics via belief propagation based stochastic bandits
Shafik An Overview of Computational Modeling and Simulations in Wireless Communication Systems
Feng et al. Game theoretic based intelligent multi-user millimeter-wave MIMO systems under uncertain environment and unknown interference
Wang et al. Application of improved Naive Bayes classification algorithm in 5G signaling analysis

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: 21962793

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 202180100752.3

Country of ref document: CN

NENP Non-entry into the national phase

Ref country code: DE