WO2022042182A1

WO2022042182A1 - Downlink channel estimation method and apparatus, communication device, and storage medium

Info

Publication number: WO2022042182A1
Application number: PCT/CN2021/108806
Authority: WO
Inventors: 高飞飞; 杨玉雯; 菅梦楠
Original assignee: 中兴通讯股份有限公司; 清华大学
Priority date: 2020-08-31
Filing date: 2021-07-28
Publication date: 2022-03-03
Also published as: CN112039807A

Abstract

The present application discloses a downlink channel estimation method and apparatus, a communication device, and a storage medium. The downlink channel estimation method comprises: inputting uplink channel state information and partial downlink channel state information into a preset neural network model; and completing downlink channel estimation according to an output result from the neural network model, the preset neural network model being generated by training on the basis of uplink channel state information samples and partial downlink channel state information samples.

Description

Downlink channel estimation method, apparatus, communication device and storage medium

technical field

The present application relates to the field of wireless communication, for example, to a downlink channel estimation method, apparatus, communication device and storage medium.

Background technique

As a key technology in the fifth generation mobile communication technology (5th Generation Mobile Communication Technology, 5G) or 6G communication system, Massive Multiple Input Multiple Output (MIMO) technology has higher performance than traditional MIMO systems. Spectral efficiency, high energy efficiency, high spatial resolution and other advantages. Therefore, there are higher requirements for the accuracy of the downlink channel state information (Channel State Information, CSI). In the frequency division duplexing (Frequency Division Duplexing, FDD) system, the base station cannot obtain the downlink channel through the uplink channel. The estimation problem of downlink channel in the system becomes particularly important. For the MIMO system, the downlink CSI is trained through a large number of downlink channel pilots, and the downlink CSI is fed back to the base station through the uplink. The overhead required for downlink training and uplink feedback is huge, which is not suitable for massive MIMO systems.

SUMMARY OF THE INVENTION

The present application provides a downlink channel estimation method, apparatus, communication device and storage medium, so as to realize accurate estimation of downlink channel state information and improve the accuracy of data transmission.

The embodiment of the present application provides a downlink channel estimation method, and the method includes:

Input the uplink channel state information and part of the downlink channel state information into a preset neural network model; complete the downlink channel estimation according to the output result of the neural network model; wherein, the preset neural network model is based on the uplink channel state information samples and Part of the downlink channel state information samples are generated by training.

The embodiment of the present application also provides a downlink channel estimation device, the device includes:

The model testing module is used to input the uplink channel state information and part of the downlink channel state information into the preset neural network model; the channel estimation module is used to complete the downlink channel estimation according to the output result of the neural network model; The set neural network model is generated by training the uplink channel state information samples and part of the downlink channel state information samples.

The embodiment of the present application also provides a communication device, the communication device includes:

one or more processors; memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the The downlink channel estimation method described in the embodiment.

Embodiments of the present application further provide a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, implements the downlink information estimation method described in the embodiments of the present application.

In the present application, the input uplink channel state information and part of the downlink channel state information are processed by using a preset neural network model, wherein the preset neural network model is generated by training the uplink channel state information samples and part of the downlink channel state information samples. , the downlink channel estimation is completed based on the output result of the neural network model, the accurate estimation of the downlink channel state information is realized, and the accuracy of data transmission is improved.

Description of drawings

FIG. 1 is a flowchart of a downlink channel estimation method provided by an embodiment of the present application;

2 is an example diagram of a preset neural network model provided by an embodiment of the present application;

3 is a flowchart of another downlink channel estimation method provided by an embodiment of the present application;

FIG. 4 is a flowchart of another downlink channel estimation method provided by an embodiment of the present application;

FIG. 5 is a flowchart of another downlink channel estimation method provided by an embodiment of the present application;

6 is an example diagram of another preset neural network model provided by an embodiment of the present application;

FIG. 7 is a schematic structural diagram of an apparatus for downlink channel estimation provided by an embodiment of the present application;

FIG. 8 is a schematic structural diagram of a communication device provided by an embodiment of the present application.

detailed description

Hereinafter, the embodiments of the present application will be described with reference to the accompanying drawings.

FIG. 1 is a flowchart of a downlink channel estimation method provided by an embodiment of the present application. The embodiment of the present application can be applied to a situation in which downlink channel state information of a downlink channel is predicted, and the method can be performed by a downlink channel estimation device. It can be implemented in software and/or hardware, and can generally be integrated in a base station. Referring to FIG. 1 , the method provided by the embodiment of the present application includes the following steps:

Step 110: Input the uplink channel state information and part of the downlink channel state information into a preset neural network model.

The uplink channel state information may be information of the channel state of the uplink communication link, and the uplink channel state information may include channel matrix, multipath delay, Doppler frequency offset, rank of MIMO channel, beamforming vector, and the like. Part of the downlink channel state information may be channel information corresponding to the communication link of a part of the antennas in the antennas corresponding to the base station, and part of the downlink channel state information may be pilot frequencies. In the embodiment of the present application, the antenna used for communication is an array antenna. The number of antenna array antennas is M; h(fd) is the downlink channel state information, fd is the channel frequency of the downlink channel; h′(fd) is the channel state information of some downlink channels, fd is the channel frequency of the downlink channel, len( h'(fd))<len(h(fd)), len(*) represents the length of the vector *, and the partial downlink channel state information may be the channel state information of one or more downlink channels in all downlink channels of the base station. The neural network model can be a pre-trained neural network, and the neural network model can process the input information and determine the value of the corresponding downlink channel state information after training on the uplink channel state information and part of the downlink channel state information.

The uplink channel state information may be attribute information obtained by the base station of the data transmitted by the user in the uplink direction. When performing downlink channel estimation, the acquired uplink channel state information and part of the downlink channel state information may be input as input parameters to the preset channel. In the neural network model, the uplink channel state information and part of the downlink channel state information can be input into the input layer of the preset neural network model, and the amount of input data in the input layer can be determined by the signal source and the array antenna in the wireless communication system. to adjust for the difference. In the embodiment of the present application, the uplink channel state information input into the preset neural network model may be the channel state information of all uplink channels, or may be the channel state information of some uplink channels, and input into the preset neural network model The more uplink channels corresponding to the uplink channel state information in , the more accurate the downlink channel estimation can be.

Step 120: Complete downlink channel estimation according to the output result of the neural network model; wherein, the preset neural network model is generated based on the training of uplink channel state information samples and some downlink channel state information samples.

The output result can be the result of processing the uplink channel state information and part of the downlink channel state information by the neural network model, the output result can correspond to the channel state information of all downlink channels, and the output result can be the identification number of the downlink channel state information or the downlink channel state information. Numerical value of channel state information.

In the embodiment of the present application, the output result output by the neural network model can be obtained, the output result can be a vector, the output result can correspond to all downlink channels, and the output result can be composed of the channel state information of the downlink channel or by the channel state information of the downlink channel. It is composed of an identification number, and the downlink channel state information corresponding to the output result can be searched for downlink channel transmission. The neural network model can be generated by training the uplink channel state information samples and part of the downlink channel state information samples, and the uplink channel state information samples can be a threshold amount of uplink channel state information collected in advance, and the uplink channel state information samples in the uplink channel state information samples The state information can correspond to all the uplink channels of the base station, and some of the downlink channel state information samples can be pre-collected channel state information of the downlink channel. After the neural network model is initially established, the neural network model can be trained according to a large number of training samples.

In this embodiment of the present application, the input uplink channel state information and part of the downlink channel state information are processed by using a preset neural network model, wherein the preset neural network model is trained by the uplink channel state information samples and part of the downlink channel state information samples Generate, complete downlink channel estimation based on the output result of the neural network model, realize accurate estimation of downlink channel state information, and improve the accuracy of data transmission.

Based on the above application embodiments, the neural network model includes at least a downlink module for processing the partial downlink channel state information, an uplink module for processing the uplink channel state information, and an uplink module for merging the uplink module. and a fusion module of the feature vector output by the downlink module.

The downlink module can be a neural network model that processes downlink channel state information, the uplink module can be a neural network model that processes uplink channel state information, and the fusion module can be a neural network model that fuses the information output by the downlink module and the uplink module. In the network model, the feature vector can be a vector composed of feature data, and the feature vector can be the output result of the uplink module and the downlink module.

In the embodiment of the present application, the preset neural network model may be composed of a downlink module, an uplink module, and a fusion module, and each module constituting the preset neural network model may be an independent neural network model, or may be related In the neural network model, for example, when the uplink module processes the uplink channel state information, the output result of the downlink module can be used as an input parameter when the uplink module processes the uplink channel state information. Exemplarily, FIG. 2 is an example diagram of a preset neural network model provided by an embodiment of the present application. Referring to FIG. 2, a neural network model may include an uplink module 201, a downlink module 202, and a fusion module 203. The uplink module 201 and the downlink module 202 may be relatively independent neural network models. The uplink module 201 does not use the information of the downlink module 202 when processing the uplink channel state information. Similarly, the downlink module 202 does not use the information of the uplink module 201 when processing the downlink channel state information. information, the uplink module 201 and the downlink module 202 input the generated feature vectors to the fusion module 203 after processing the uplink channel state information and part of the downlink channel state information respectively, and the fusion module 203 fuses the input feature vectors. The fusion method can be Including the use of deep neural network processing on feature vectors and weighted average processing on feature vectors.

FIG. 3 is a flowchart of another downlink channel estimation method provided by an embodiment of the present application. The embodiment of the present application is a description based on the above-mentioned embodiment of the application. When the neural network model is set, the uplink channel state information and part of the downlink channel state information are converted. Referring to FIG. 3, the method provided by the embodiment of the present application includes the following steps:

Step 210: Convert the uplink channel state information and part of the downlink channel state information according to a preset mapping relationship.

The preset mapping relationship may be a mapping relationship that converts part of the downlink channel state information and the uplink channel state information into an input vector. For example, the values of each channel state information may be arranged in a preset order into feature vectors, or each channel state information may be The values of are weighted and arranged into eigenvectors.

Part of the downlink channel state information and the uplink channel state information to be input into the preset neural network model can be converted into feature vectors according to the preset mapping relationship, and part of the downlink channel state information and the uplink channel state information can correspond to one feature vector, or can be separately corresponds to a feature vector.

Step 220, input the converted uplink channel state information and part of the downlink channel state information to the hidden layer of the uplink module and the downlink module respectively, to utilize the activation function corresponding to the hidden layer and the hidden layer to the converted uplink channel state information and part. Downlink channel state information is processed.

The hidden layer may be a layer in the neural network model that processes and calculates data, and the preset neural network model may include one or more hidden layers. The activation function can be a rule used for processing computational data in the hidden layer, and the activation function can include a sigmoid function, a tanh function, a Rectified Linear Unit (ReLU) function, a LeakyReLU function, and the like.

Part of the downlink channel state information and the uplink channel state information converted by the preset mapping relationship can be input into the hidden layers of the uplink module and the downlink module as feature vectors, respectively, and the uplink module and the hidden layer of the downlink module are activated by the activation function. The channel state information and part of the downlink channel state information are processed.

Step 230: Input the processing result of each hidden layer into the fusion module through the output layers of the uplink module and the downlink module, and generate an output result in the fusion module.

In this embodiment of the present application, the uplink module and the downlink module may respectively include multiple hidden layers, the output layers of the uplink module and the downlink module may output respective processing results, and the fusion module may output the processing results corresponding to the uplink module and the downlink module respectively For fusion, the processing result after fusion by the fusion module can be used as the output result of the preset neural network model.

Step 240: Use the channel state information corresponding to the output result as the estimation of the downlink channel.

In this embodiment of the present application, the uplink channel state information and part of the downlink channel state information are converted according to a preset mapping relationship, and the converted uplink channel state information and part of the downlink channel state information are input into the hidden layer of the uplink module and the hidden layer of the downlink module for processing. processing, wherein each hidden layer is provided with a corresponding activation function, the processing result of the hidden layer is input to the fusion module through the output layer of the uplink module and the downlink module for fusion to generate an output result, and the channel state information corresponding to the output result is used as the downlink The channel estimation realizes the accurate estimation of the downlink channel state information in the MIMO system and improves the stability of data transmission.

Based on the above-mentioned embodiments of the application, the preset mapping relationship includes a corresponding relationship between a complex number and a real part and an imaginary part of the complex number.

In the embodiment of the present application, the preset mapping relationship for converting the uplink channel state information and part of the downlink channel state information is the corresponding relationship between the complex number and the real part and the imaginary part of the complex number. The values corresponding to part of the downlink channel state information are used as complex numbers, and the real part and imaginary part corresponding to each complex number are obtained. The real part and imaginary part corresponding to the uplink channel state information and part of the downlink channel state information can be used as the input of the preset neural network model Feature vector.

In an exemplary embodiment, R[x] and S[x] respectively represent the real part and imaginary part of x, and the preset mapping relationship can be expressed as γ: x→(R(x ^T ), S( x ^T )) ^T .

On the basis of the above application embodiment, the processing results of each hidden layer are input into the fusion module through the output layers of the uplink module and the downlink module, and the output results are generated in the fusion module, including:

In the fusion module, the processing results output by the uplink module and the downlink module are spliced according to preset rules; the spliced processing results are passed through at least one hidden layer in the fusion module and an activation function corresponding to the hidden layer. Process to generate output.

The preset rule may be a rule for splicing the processing result of the uplink module and the processing result of the downlink module. For example, each processing result is weighted and averaged, and the calculation result can be used as the processing result after splicing, and the processing result of the uplink module can also be combined. The result and the value corresponding to the same channel attribute in the processing result of the downlink module are spliced in the same row or column.

In the embodiment of the present application, a preset rule for splicing the processing result of the uplink module and the processing result of the downlink module may be preset in the fusion module, and the processing results output by the uplink module and the downlink module are spliced according to the preset rules, and the fusion module The activation function in the spliced processing result is processed to generate an output result corresponding to the downlink channel state information.

FIG. 4 is a flowchart of another downlink channel estimation method provided by an embodiment of the present application. The embodiment of the present application is described based on the foregoing embodiment. Referring to FIG. 4 , the method provided by the embodiment of the present application includes the following steps:

Step 310: Train and generate a neural network model based on the uplink channel state information samples and some of the downlink channel state information samples.

The sources of the uplink channel state information samples and the partial downlink channel state information samples may include the local base station, all base stations within the cell range where the local base station is located, or any base station, and the like.

In this embodiment of the present application, the uplink channel state information and part of the downlink channel state information, as well as the downlink channel state information label corresponding to each uplink channel state information and the downlink channel state information label corresponding to each part of the downlink channel state information may be collected in advance, wherein , the downlink channel state information label may be the downlink channel state information actually used by the base station under the uplink channel state information, and the downlink channel state information actually used by the base station under the part of the downlink channel state information. The neural network model is trained by the acquired uplink channel state information samples and some downlink channel state information samples, and the neural network model may include an input layer, an output layer, a hidden layer, and the like.

Step 320: Input the uplink channel state information and part of the downlink channel state information into a preset neural network model.

Step 330, complete downlink channel estimation according to the output result of the neural network model; wherein, the preset neural network model is generated based on the training of uplink channel state information samples and some downlink channel state information samples.

FIG. 5 is a flowchart of another downlink channel estimation method provided by an embodiment of the present application. The embodiment of the present application describes the process of training and generating a neural network model on the basis of the above embodiment. Referring to FIG. 5 , an embodiment of the present application The provided method includes the following steps:

Step 410: Collect some downlink channel state information samples, uplink channel state information samples and corresponding downlink channel state information labels.

The downlink channel state information label may be the downlink channel state information actually used by the base station under the uplink channel state information, and the downlink channel state information actually used by the base station under the part of the downlink channel state information.

The sources of the uplink channel state information samples and some of the downlink channel state information samples may include the channel state information collected by the local base station, all base stations within the cell range where the local base station is located, or any base station.

The uplink channel state information can be collected in the base station and the channel state information of the downlink channel can be determined by the uplink channel state information, and the downlink channel state information can be used as the corresponding downlink channel state information label. Similarly, the downlink channel can be collected in the base station. The channel state information and the channel state information of the downlink channel determined by the part of the downlink channel state information, and the downlink channel state information can be used as the corresponding downlink channel state information label.

Step 420: Convert each part of the downlink channel state information in the part of the downlink channel state information samples and each part of the uplink channel state information in the uplink channel state information sample as the uplink feature vector and the downlink feature vector respectively according to the preset mapping relationship.

In this embodiment of the present application, part of the collected downlink channel state information samples and uplink channel state information samples may be preprocessed, and each channel state information may be converted into a feature vector, and the feature vector corresponding to the uplink channel state information may be an uplink feature vector , the eigenvector corresponding to part of the downlink channel state information may be a downlink eigenvector. Wherein, the preset mapping relationship may include the corresponding relationship between the complex value and its real part and its imaginary part.

Step 430: Combine each uplink feature vector, each downlink feature vector and the corresponding downlink channel state information label into a training set; train the neural network model according to the training set.

In this embodiment of the present application, each uplink feature vector and corresponding downlink channel state information label, and each downlink feature vector and corresponding downlink channel state information label may be used as a training set, and the training set may be used to train and generate a neural network model .

Step 440: Input the uplink channel state information and part of the downlink channel state information into a preset neural network model.

Step 450: Complete downlink channel estimation according to the output result of the neural network model; wherein, the preset neural network model is generated based on the training of uplink channel state information samples and some downlink channel state information samples.

On the basis of the above application embodiments, the training of the neural network model according to the training set includes:

Input the uplink feature vector and the downlink feature vector in the training set into the neural network model, and obtain the predicted value of the output downlink channel state information; determine the predicted value and the loss value of the corresponding downlink channel state information label according to the preset loss function; It is determined that the training of the neural network model is completed when the loss value is less than the preset threshold, otherwise, the upward feature vector and the downward feature vector in the training set are continued to be input to the neural network module for training.

The uplink feature vector and the downlink feature vector are respectively input into the neural network model for processing, and the processing result output by the neural network model can be obtained. The processing result can be the predicted value of the downlink channel state information, and the predicted value and the corresponding downlink can be determined. The loss value between the state information labels. The calculation process of the loss value can be determined by the preset loss function. When the loss value is less than the preset threshold, it can be considered that the neural network training is completed. Otherwise, continue to obtain the channel uplink feature vector and downlink feature. The vector is input to the neural network model for training, and the training process can repeat the above process until the loss value is less than the preset threshold.

In an exemplary implementation, FIG. 6 is an example diagram of another preset neural network model provided by the embodiment of the present application, and FIG. 6 shows the structure of the preset deep neural network model. in,

represent the real and imaginary parts of x, respectively,

Indicates to perform a mapping operation.

The working process of the downlink module is as follows:

First, put

Denoted as x _P , input x _P to the downlink module through the input layer, and use the activation function corresponding to each hidden layer and the hidden layer to obtain the feature vector processed by the hidden layer. The downlink module has a total of _LP layers, including 1 input layer, 1 output layer and LP _-2 hidden layer. _LP can be set to 5, that is, including three hidden layers. The feature vector of the hidden layer can be expressed as:

in,

are the parameters that the network needs to train,

is the nonlinear transformation function of the lth layer, which can be improved as:

in,

and

are the weight matrix, bias parameter and nonlinear activation function of the corresponding layer, respectively. The activation function is a LeakyReLU function, that is, [g _re (z)] _p =max{[z] _p /a, [z] _p }, a>1, where [z] _p represents the p-th vector z elements, p=1, 2, . . . , len(z), and len(z) represents the length of the vector z.

The feature vector processed by the hidden layer is input to the output layer of the downlink module, and the output feature vector is obtained by using the activation function of the output layer.

Based on the content of the above embodiment, the processing process of the uplink module is the same as that of the downlink module. The role of the uplink module is to process the uplink channel state information, and obtain the output feature vector corresponding to the uplink channel state information, including: performing a preset on the vector including the uplink channel state information. The corresponding feature vector is obtained by the mapping operation of ; the feature vector is input to the upstream module to obtain the corresponding output feature vector.

Optionally, the preset mapping is to map the uplink channel information h(f _U ) as

The network structure of the uplink module is similar to that of the downlink module, and the input vector is

The working process will not be repeated here.

Based on the content of the above embodiment, the function of the fusion module is to process the output feature vectors of the downlink module and the uplink module after fusion, and obtain the output vector. The parameters of the downlink channel state information can be predicted according to the output vector to complete the downlink channel estimation, and the output feature vector of the downlink module and the output feature vector of the uplink module can be fused with a preset fusion function to obtain a preset feature vector; The vector is input to the preset fusion module network, the output vector is obtained, the downlink channel parameters are predicted according to the output vector, and the downlink channel estimation is completed.

The fusion module can be implemented as follows:

Feature fusion is performed on the output feature vector of the downlink module and the uplink module:

Among them, F _P (x _P ) is the output feature vector of the downlink module;

is the connection function; F _U (x _U ) is the output feature vector of the uplink module; x _Fus is the fused feature vector. After the fused feature vector x _Fus is input to the fusion module network, the vector z _Fus is obtained after passing through the L _Fus layer hidden layer, and the output vector can be obtained by inputting z _Fus to the output layer after the activation function of the output layer is used. . Predict downlink channel parameters according to the output vector to complete downlink channel estimation.

Based on the content of the foregoing embodiment, before inputting the uplink channel state information and part of the downlink channel state information into the preset deep neural network model, the method further includes: acquiring a plurality of information samples of the uplink channel state information and part of the downlink channel state information, and The downlink channel state information label corresponding to each information sample; after performing the above-mentioned preset mapping operation on each uplink channel state information and part of the downlink channel state information, the eigenvectors corresponding to the uplink channel state information and part of the downlink channel state information are obtained; The combination of the feature vector corresponding to each uplink channel state information and part of the downlink channel state information and the label of the downlink channel state information is used as a training sample, so as to obtain multiple training samples. to train.

First, obtain multiple partial downlink channel state information samples

And, multiple uplink channel state information samples h(f _U ), and obtain the corresponding downlink channel state information h(f _D ),

as the corresponding part of the downlink channel information

and the label of the upstream channel information h(f _U ).

Second, the

h(f _U ) and the corresponding

The combination of labels is used as a sample, resulting in multiple training samples. in each sample

h(f _U ) is input to the constructed deep neural network model after mapping, and the relevant parameters of the deep neural network model are adjusted according to the output results, and the training of the deep neural network model is completed, thereby obtaining the above-mentioned preset deep data network Model.

Based on the content of the above embodiment, the training of the deep neural network model by using the plurality of training samples includes: inputting any part of the downlink channel information and the feature vector of the uplink channel information samples into the deep neural network model, output the predicted value of the downlink channel state information parameter; use the preset loss function to calculate the loss value with the downlink channel state information label corresponding to the sample according to the predicted value of the downlink channel state information parameter; if the loss value is less than the preset threshold, then The training of the deep neural network model is completed.

First, select any part of the downlink channel state information parameters from the pilot signal and the uplink channel information samples

and the uplink channel state information parameter h(f _U ), the

After h(f _U ) completes the corresponding mapping, it is input to the deep neural network model, and is output from the output layer through nonlinear transformation of each hidden layer and its activation function.

predicted value. according to

The predicted value of , respectively corresponds to the label corresponding to the training sample, and calculates the loss value corresponding to the loss function.

For example, for a regression network, the loss function can be the L2 _norm function, ie:

where V is the number of samples in a single batch, v represents the serial number of the sample in the batch,

is the prediction vector output by the network, y(v) represents the label corresponding to the training sample, that is

In the training phase, the deep neural network model performs layer-by-layer nonlinear transformation on the input sample data to obtain the predicted output, and calculates the loss value corresponding to the loss function according to the corresponding label. The loss function can be gradually optimized through the adaptive matrix estimation (Adam) algorithm, so as to continuously optimize and update the parameters of the network until the loss function converges.

After training, the parameters of the deep neural network model remain unchanged, and the performance evaluation is achieved by calculating the error between the predicted value and the label value using the predicted value obtained from the input of the test set.

Before training the deep neural network model, the network of the downlink module, the network of the uplink module, and the network of the fusion module need to be trained separately. The training process of the downlink module network, the uplink module network, and the fusion module network is similar to the training process of the deep neural network model, and will not be repeated here. Among them, the input value of the downlink module network is

Labeled as

The input values and labels of the upstream module network are

The input values of the fusion module network are F _P (x _P ), F _U (x _U ), and the labels are

The training sequence of all the above deep neural network models is as follows: firstly, the training process of the downlink module network, the uplink module network, and the fusion module network is completed, and then the training process of the neural network model is completed.

FIG. 7 is a schematic structural diagram of a downlink channel estimation apparatus provided by an embodiment of the present application. The apparatus shown in FIG. 7 can execute the downlink channel estimation method provided by the embodiment of the present application, and execute the corresponding functional modules and beneficial effects of the method. The apparatus can be implemented by software and/or hardware, and includes: a model testing module 501 and a channel estimation module 502 .

The model testing module 501 is configured to input the uplink channel state information and part of the downlink channel state information into a preset neural network model.

The channel estimation module 502 is configured to complete downlink channel estimation according to the output result of the neural network model; wherein, the preset neural network model is generated through training of uplink channel state information samples and part of downlink channel state information samples.

In this embodiment of the present application, the input uplink channel state information and part of the downlink channel state information are processed through a neural network model preset by the model testing module, wherein the preset neural network model passes through the uplink channel state information samples and part of the downlink channel state information. The information sample is trained and generated, and the channel estimation module completes the downlink channel estimation based on the output result of the neural network model, which realizes the accurate estimation of the downlink channel state information and improves the accuracy of data transmission.

On the basis of the above application embodiments, the neural network model in the downlink channel estimation device includes at least a downlink module for processing the partial downlink channel state information, an uplink module for processing the uplink channel state information, and an uplink module for fusion A fusion module of feature vectors output by the uplink module and the downlink module.

On the basis of the above application embodiments, the model testing module 501 includes:

A parameter conversion unit, configured to convert the uplink channel state information and the part of the downlink channel state information according to a preset mapping relationship.

A model processing unit for inputting the converted uplink channel state information and the partial downlink channel state information to the hidden layers of the uplink module and the downlink module respectively, so as to utilize the hidden layer and the hidden layer The activation function corresponding to the layer processes the converted uplink channel state information and part of the downlink channel state information.

On the basis of the above application embodiments, the preset mapping relationship in the parameter conversion unit includes a corresponding relationship between a complex number and the real part and imaginary part of the complex number.

On the basis of the above application embodiments, the channel estimation module 502 includes:

The result fusion unit is configured to input the processing result of each hidden layer into the fusion module through the output layers of the uplink module and the downlink module, and generate an output result in the fusion module.

The channel estimation unit is configured to use the channel state information corresponding to the output result as the estimation of the downlink channel.

On the basis of the above application embodiment, the result fusion unit is used for:

On the basis of the above application embodiments, it also includes:

A training module, configured to train and generate the neural network model based on the uplink channel state information samples and part of the downlink channel state information samples.

On the basis of the above application embodiment, the training module includes:

The information collection unit is used for collecting part of the downlink channel state information samples, the uplink channel state information samples and the corresponding downlink channel state information labels.

The sample conversion unit is configured to convert each part of the downlink channel state information in the partial downlink channel state information samples and each part of the uplink channel state information in the uplink channel state information as the uplink feature vector and the downlink feature vector respectively according to the preset mapping relationship.

A sample generating unit, configured to combine each of the uplink feature vectors, each of the downlink feature vectors and corresponding downlink channel state information labels into a training set.

A training execution unit, configured to train the neural network model according to the training set.

On the basis of the above application embodiment, the training execution unit is configured to: input the uplink feature vector and the downlink feature vector in the training set into the neural network model, and obtain the predicted value of the output downlink channel state information Determine the predicted value and the loss value of the corresponding downlink channel state information label according to the preset loss function; determine that the neural network model training is completed when the loss value is less than the preset threshold, otherwise continue to use the uplink in the training set. The feature vector and the descending feature vector are input to the neural network module for training.

FIG. 8 is a schematic structural diagram of a communication device provided by an embodiment of the present application. As shown in FIG. 8 , the communication device includes a processor 80, a memory 81, an input device 82, and an output device 83; the number of processors 80 in the communication device There may be one or more, and a processor 80 is taken as an example in FIG. 8; the communication device processor 80, the memory 81, the input device 82 and the output device 83 can be connected through a bus or in other ways. In FIG. 8, the connection through the bus is example.

The memory 81, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the modules (model testing module 501 and channel estimation module 502) corresponding to the downlink channel estimation apparatus in the embodiments of the present application. The processor 80 executes various functional applications and data processing of the communication device by running the software programs, instructions and modules stored in the memory 81 , that is, to implement the above-mentioned method.

The memory 81 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the terminal, and the like. Additionally, memory 81 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, memory 81 may include memory located remotely from processor 80, which may be connected to a communication device through a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

The input device 82 may be used to receive input numerical or character information and to generate key signal input related to user settings and function control of the communication device. The output device 83 may include a display device such as a display screen.

Embodiments of the present application further provide a computer-readable storage medium, where the computer-executable instructions are used to execute a downlink channel estimation method when executed by a computer processor, the method comprising: converting uplink channel state information and part of downlink channel state information A preset neural network model is input; downlink channel estimation is completed according to the output result of the neural network model; wherein, the preset neural network model is generated based on the training of uplink channel state information samples and some downlink channel state information samples.

An embodiment of the present application provides a storage medium containing computer-executable instructions. The computer-executable instructions are not limited to the above method operations, and can also perform related operations in the downlink channel estimation method provided by any embodiment of the present application.

The term user terminal covers any suitable type of wireless user equipment, such as a mobile telephone, portable data processing device, portable web browser or vehicle mounted mobile station.

In general, the various embodiments of the present application may be implemented in hardware or special purpose circuits, software, logic, or any combination thereof. For example, some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that may be executed by a controller, microprocessor or other computing device, although the application is not limited thereto.

Embodiments of the present application may be implemented by the execution of computer program instructions by a data processor of a mobile device, eg in a processor entity, or by hardware, or by a combination of software and hardware. Computer program instructions may be assembly instructions, Instruction Set Architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or written in any combination of one or more programming languages source or object code.

The block diagrams of any logic flow in the figures of this application may represent program steps, or may represent interconnected logic circuits, modules, and functions, or may represent a combination of program steps and logic circuits, modules, and functions. Computer programs can be stored on memory. The memory may be of any type suitable for the local technical environment and may be implemented using any suitable data storage technology, such as, but not limited to, Read-Only Memory (ROM), Random Access Memory (RAM), optical Memory devices and systems (Digital Video Disc (DVD) or Compact Disk (CD)), etc. Computer-readable media may include non-transitory storage media. The data processor may be of any type suitable for the local technical environment, such as, but not limited to, a general purpose computer, a special purpose computer, a microprocessor, a Digital Signal Processing (DSP), an Application Specific Integrated Circuit (ASIC) ), programmable logic devices (Field-Programmable Gate Array, FPGA) and processors based on multi-core processor architecture.

Claims

A downlink channel estimation method, comprising:

Input the uplink channel state information and part of the downlink channel state information into the preset neural network model;

Complete downlink channel estimation according to the output result of the neural network model;

Wherein, the preset neural network model is generated by training based on uplink channel state information samples and some downlink channel state information samples.
The method according to claim 1, wherein the neural network model includes at least a downlink module for processing the partial downlink channel state information, an uplink module for processing the uplink channel state information, and an uplink module for fusing the A fusion module of the feature vectors output by the uplink module and the downlink module.
The method according to claim 1 or 2, wherein the inputting the uplink channel state information and part of the downlink channel state information into a preset neural network model comprises:

Converting the uplink channel state information and the part of the downlink channel state information according to a preset mapping relationship;

The converted uplink channel state information and the partial downlink channel state information are respectively input into the hidden layers of the uplink module and the downlink module, so as to use the hidden layer and the activation function corresponding to the hidden layer to The converted uplink channel state information and the partial downlink channel state information are processed.
The method according to claim 3, wherein the preset mapping relationship includes a corresponding relationship between a complex number and a real part and an imaginary part of the complex number.
The method according to claim 3, wherein the performing downlink channel estimation according to the output result of the neural network model comprises:

Input the processing results of the multi-layer hidden layers of the uplink module and the downlink module into the fusion module through the output layers of the uplink module and the downlink module, and generate an output result in the fusion module;

The channel state information corresponding to the output result is used as the estimation of the downlink channel.
The method of claim 5, wherein the generating an output result in the fusion module comprises:

In the fusion module, the processing results output by the uplink module and the downlink module are spliced according to preset rules;

The spliced processing result is processed by at least one hidden layer in the fusion module and an activation function corresponding to the at least one hidden layer to generate the output result.
The method of claim 1, further comprising:

The neural network model is generated by training based on the uplink channel state information samples and the partial downlink channel state information samples.
The method according to claim 7, wherein the partial downlink channel state information samples include a plurality of downlink channel state information, the uplink channel state information samples include a plurality of uplink channel state information, and the uplink channel state information is based on the uplink channel state information. Information samples and the partial downlink channel state information samples are trained to generate the neural network model, including:

Collect the partial downlink channel state information samples, the uplink channel state information samples, the downlink channel state information label corresponding to each downlink channel state information in the partial downlink channel state information samples, and the uplink channel state information samples. The downlink channel state information label corresponding to each uplink channel state information of ;

Converting the downlink channel state information in the partial downlink channel state information samples and the uplink channel state information in the uplink channel state information samples into an uplink feature vector and a downlink feature vector respectively according to a preset mapping relationship;

combining the uplink feature vector, the downlink feature vector, and the downlink channel state information label corresponding to the uplink feature vector and the downlink feature vector into a training set;

The neural network model is trained according to the training set.
The method of claim 8, wherein the training the neural network model according to the training set comprises:

Input the uplink feature vector and the downlink feature vector in the training set into the neural network model, and obtain the predicted value of the output downlink channel state information;

Determine the loss value of the predicted value and the downlink channel state information label corresponding to the uplink feature vector and the downlink feature vector in the training set according to a preset loss function;

In the case that the loss value is less than the preset threshold, the training of the neural network model is completed, and in the case that the loss value is not less than the preset threshold, continue to use the uplink feature vector and the downlink in the training set The feature vector is input to the neural network module for training.
A downlink channel estimation device, comprising:

The model testing module is set to input the uplink channel state information and part of the downlink channel state information into the preset neural network model;

a channel estimation module, configured to complete downlink channel estimation according to the output result of the neural network model;

Wherein, the preset neural network model is generated through training of uplink channel state information samples and part of downlink channel state information samples.
A communication device comprising:

at least one processor;

a memory, arranged to store at least one program;

When the at least one program is executed by the at least one processor, the at least one processor implements the downlink channel estimation method according to any one of claims 1-9.
A computer-readable storage medium storing a computer program, wherein when the program is executed by a processor, the downlink channel estimation method according to any one of claims 1-9 is implemented.