Downlink signal transmission method and device, downlink signal transmission equipment and storage medium

Abstract

The invention relates to a downlink signal transmission method and device, downlink signal transmission equipment and a storage medium. The method comprises the following steps: acquiring user coordinates and base station coordinates, and obtaining feature data according to the user coordinates and the base station coordinates; performing data processing on the feature data according to a forming factor prediction model to obtain a forming factor; the beamforming factor is a parameter of beamforming when downlink signal transmission is carried out between the user terminal of the target user and the base station. By adopting the method, the efficiency of obtaining the forming factor prediction result can be improved.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a downlink signal transmission method, apparatus, downlink signal transmission equipment, storage medium and computer program product. Background Art

[0002] With the development of the field of communication technology, in a large-scale MIMO (Multiple Input Multiple Output,) system, in order to improve the quality of communication services, it is necessary to obtain CSI (Channel State Information) and realize the transmission of downlink signals through beamforming based on the channel state information CSI.

[0003] In the traditional method, the terminal obtains channel estimation based on the pilot method, uses the channel estimation as CSI, and obtains the shaping vector by performing singular value decomposition on the channel matrix in the CSI, that is, the shaping factor used for beamforming. Therefore, the downlink signal can be transmitted under the constraint of the shaping factor.

[0004] However, in traditional methods, since CSI needs to be obtained through pilot signals, pilot signals need to occupy certain resources and bandwidth during data transmission, resulting in large air interface overhead and feedback delays between user terminals and base stations, resulting in low efficiency in determining the shaping factor. Summary of the invention

[0005] Based on this, it is necessary to provide a downlink signal transmission method, apparatus, downlink signal transmission device, computer-readable storage medium and computer program product to address the above technical problems.

[0006] In a first aspect, the present application provides a downlink signal transmission method, including:

[0007] Acquire user coordinates and base station coordinates, and obtain feature data according to the user coordinates and the base station coordinates;

[0008] The characteristic data is processed according to the shaping factor prediction model to obtain a shaping factor; the shaping factor is a beamforming parameter when downlink signal transmission is performed between a user terminal of a target user and a base station.

[0009] In one embodiment, characteristic data is obtained according to the user coordinates and the base station coordinates, including:

[0010] Calculate relative coordinates and relative angles based on user coordinates and base station coordinates;

[0011] Based on the reduction amplitude correction method, the relative coordinates are corrected by the reduction amplitude to obtain the corrected coordinates, and the corrected coordinates and the relative angle are used as feature data.

[0012] In one embodiment, the shaping factor prediction model includes a convolutional structure, an attention structure, a residual structure, and a fully connected structure;

[0013] The characteristic data is processed according to the shaping factor prediction model to obtain shaping factors, including:

[0014] The feature data is input into the shaping factor prediction model, and the feature data is extracted through the convolution structure to obtain the initial feature map;

[0015] Based on the attention structure, the initial feature map is weighted and channel-concatenated to obtain a concatenated feature map;

[0016] Based on the residual structure, the residuals of the concatenated feature maps are added to obtain the target feature map;

[0017] The target feature map is transformed in dimension according to the fully connected structure, and the shaping factor is output.

[0018] In one embodiment, the attention structure includes a channel attention layer and a spatial attention layer including:

[0019] The initial feature map is weighted according to the channel attention layer to obtain a weighted first feature map;

[0020] The first feature map is channel-spliced ​​according to the spatial attention layer to obtain a second feature map after channel splicing, and the second feature map is convolved and activated to obtain a spliced ​​feature map.

[0021] In one of the embodiments, the residual structure includes a first residual layer, a second residual layer, a third residual layer and a fourth residual layer;

[0022] Based on the residual structure, the concatenated feature map is residually added to obtain the target feature map, including:

[0023] Extract features from the concatenated feature map according to the first residual layer to obtain a first initial target feature map, and fuse features of the first initial target feature map with the concatenated feature map through a jump connection to obtain a first target feature map;

[0024] According to the second residual layer, the third residual layer and the fourth residual layer, feature extraction, downsampling processing and feature fusion are performed on the first target feature map respectively to obtain the target feature map.

[0025] In one embodiment, before obtaining the user coordinates and the base station coordinates, the method further includes:

[0026] Acquire multiple sample user coordinates, and use the shaping factor corresponding to each sample user coordinate as a sample label to form a first sample data set of the shaping factor prediction model;

[0027] The same sample user coordinates among the multiple sample user coordinates are divided into the same sample group to obtain multiple sample groups; the sample group includes the sample user coordinates and the sample labels corresponding to the sample user coordinates;

[0028] Based on the sample labels of the sample group, the sample label mean is calculated, and the sample label mean is used as the comprehensive sample label of the target sample group;

[0029] Constructing a second sample data set according to the sample user coordinates of each sample group and the comprehensive sample labels corresponding to the sample groups;

[0030] The shaping factor prediction model is trained according to the first sample data set to obtain a pre-trained model of the shaping factor prediction model;

[0031] The model parameters of the pre-trained model are adjusted according to the second sample data set to obtain a trained shaping factor prediction model.

[0032] In one embodiment, the shaping factor prediction model is trained according to the first sample data set to obtain a pre-trained model of the shaping factor prediction model, including:

[0033] Calculate and obtain sample feature data according to each sample user coordinate and base station coordinate in the first sample data set;

[0034] Based on the shaping factor prediction model and sample characteristic data, the shaping factor prediction result is obtained;

[0035] According to the spectral efficiency maximization loss function and the shaping factor prediction results, the loss value of the shaping factor prediction model is calculated. If the loss value is greater than the preset loss value, the shaping factor prediction result is obtained by executing the shaping factor prediction model and sample feature data until the loss value is less than the preset loss value, and a pre-trained model of the shaping factor prediction model is obtained.

[0036] In one embodiment, calculating the loss value of the shaping factor prediction model according to the spectrum efficiency maximization loss function and the shaping factor prediction result includes:

[0037] According to the real channel of the strongest flow shaping factor and the shaping factor prediction results, the target spectrum effect and the predicted spectrum effect are calculated respectively;

[0038] Based on the target spectral efficiency, predicted spectral efficiency and spectral efficiency maximization loss function, the loss value of the shaping factor prediction model is obtained.

[0039] In a second aspect, the present application further provides a downlink signal transmission device, including:

[0040] An acquisition module is used to acquire user coordinates and base station coordinates, and obtain feature data according to the user coordinates and the base station coordinates;

[0041] The prediction module is used to process the characteristic data according to the shaping factor prediction model to obtain the shaping factor; the shaping factor is a beamforming parameter when downlink signal transmission is performed between the user terminal of the target user and the base station.

[0042] In one embodiment, the acquisition module is specifically used for:

[0043] Calculate relative coordinates and relative angles based on user coordinates and base station coordinates;

[0044] Based on the reduction amplitude correction method, the relative coordinates are corrected by the reduction amplitude to obtain the corrected coordinates, and the corrected coordinates and the relative angle are used as feature data.

[0045] In one embodiment, the shaping factor prediction model includes a convolutional structure, an attention structure, a residual structure, and a fully connected structure;

[0046] The prediction module is specifically used for:

[0047] The feature data is input into the shaping factor prediction model, and the feature data is extracted through the convolution structure to obtain the initial feature map;

[0048] Based on the attention structure, the initial feature map is weighted and channel-concatenated to obtain a concatenated feature map;

[0049] Based on the residual structure, the residuals of the concatenated feature maps are added to obtain the target feature map;

[0050] The target feature map is transformed in dimension according to the fully connected structure, and the shaping factor is output.

[0051] In one embodiment, the attention structure includes a channel attention layer and a spatial attention layer;

[0052] The prediction module is specifically used for:

[0053] The initial feature map is weighted according to the channel attention layer to obtain a weighted first feature map;

[0054] The first feature map is channel-spliced ​​according to the spatial attention layer to obtain a second feature map after channel splicing, and the second feature map is convolved and activated to obtain a spliced ​​feature map.

[0055] In one of the embodiments, the residual structure includes a first residual layer, a second residual layer, a third residual layer and a fourth residual layer;

[0056] The prediction module is specifically used for:

[0057] Extract features from the concatenated feature map according to the first residual layer to obtain a first initial target feature map, and fuse features of the first initial target feature map with the concatenated feature map through a jump connection to obtain a first target feature map;

[0058] According to the second residual layer, the third residual layer and the fourth residual layer, feature extraction, downsampling processing and feature fusion are performed on the first target feature map respectively to obtain the target feature map.

[0059] In one embodiment, the device further comprises:

[0060] A first construction module is used to obtain a plurality of sample user coordinates, and use a shaping factor corresponding to each sample user coordinate as a sample label to form a first sample data set of a shaping factor prediction model;

[0061] A division module is used to divide the same sample user coordinates among the multiple sample user coordinates into the same sample group to obtain multiple sample groups; the sample group includes the sample user coordinates and the sample labels corresponding to the sample user coordinates;

[0062] A calculation module, used for calculating a sample label mean based on the sample labels of the sample group, and using the sample label mean as a comprehensive sample label of the target sample group;

[0063] A second construction module is used to construct a second sample data set according to the sample user coordinates of each sample group and the comprehensive sample labels corresponding to the sample groups;

[0064] A first training module is used to train a shaping factor prediction model according to a first sample data set to obtain a pre-trained model of the shaping factor prediction model;

[0065] The second training module is used to adjust the model parameters of the pre-trained model according to the second sample data set to obtain a trained shaping factor prediction model.

[0066] In one embodiment, the first training module is specifically used for:

[0067] Calculate and obtain sample feature data according to each sample user coordinate and base station coordinate in the first sample data set;

[0068] Based on the shaping factor prediction model and sample characteristic data, the shaping factor prediction result is obtained;

[0069] According to the spectral efficiency maximization loss function and the shaping factor prediction results, the loss value of the shaping factor prediction model is calculated. If the loss value is greater than the preset loss value, the shaping factor prediction result is obtained by executing the shaping factor prediction model and sample feature data until the loss value is less than the preset loss value, and a pre-trained model of the shaping factor prediction model is obtained.

[0070] In one embodiment, the first training module is specifically used for:

[0071] According to the real channel of the strongest flow shaping factor and the shaping factor prediction results, the target spectrum effect and the predicted spectrum effect are calculated respectively;

[0072] Based on the target spectral efficiency, predicted spectral efficiency and spectral efficiency maximization loss function, the loss value of the shaping factor prediction model is obtained.

[0073] In a third aspect, the present application further provides a downlink signal transmission device, including a transceiver, a memory and a processor, wherein the transceiver is used to receive and send data under the control of the processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0074] Obtain user coordinates and base station coordinates, and obtain feature data based on the user coordinates and the base station coordinates;

[0075] The characteristic data is processed according to the shaping factor prediction model to obtain a shaping factor; the shaping factor is a beamforming parameter when downlink signal transmission is performed between a user terminal of a target user and a base station.

[0076] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the following steps are implemented:

[0077] Obtain user coordinates and base station coordinates, and obtain feature data based on the user coordinates and the base station coordinates;

[0078] The characteristic data is processed according to the shaping factor prediction model to obtain a shaping factor; the shaping factor is a beamforming parameter when downlink signal transmission is performed between a user terminal of a target user and a base station.

[0079] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0080] Obtain user coordinates and base station coordinates, and obtain feature data based on the user coordinates and the base station coordinates;

[0081] The characteristic data is processed according to the shaping factor prediction model to obtain a shaping factor; the shaping factor is a beamforming parameter when downlink signal transmission is performed between a user terminal of a target user and a base station.

[0082] The above-mentioned downlink signal transmission method, device, downlink signal transmission equipment, storage medium and computer program product process the feature data through the shaping factor prediction model, establish a nonlinear mapping from the user coordinates of low-dimensional information to the shaping factor prediction results of high-dimensional information, can quickly obtain the shaping factor, avoid the complicated channel estimation and shaping weight calculation process, can reduce the requirements on the base station processing capability, and at the same time reduce the communication delay, thereby improving the efficiency of obtaining the shaping factor prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0084] Figure 1 A schematic diagram of a flow chart of a downlink signal transmission method in an embodiment;

[0085] Figure 2 is a schematic flow chart of another downlink signal transmission method in an embodiment;

[0086] Figure 3 A schematic diagram of the principle of calculating characteristic data in one embodiment;

[0087] Figure 4 A schematic diagram of the structure of a shaping factor prediction model in one embodiment;

[0088] Figure 5 A schematic diagram of a flow chart of data processing by a shaping factor prediction model in one embodiment;

[0089] Figure 6 A schematic diagram of a flow chart of data processing by an attention structure in one embodiment;

[0090] Figure 7 A schematic diagram of a flow chart of data processing of a residual structure in one embodiment;

[0091] Figure 8 A schematic diagram of a process for constructing a training sample in one embodiment;

[0092] Fig. 9 A schematic diagram of a process for training a shaping factor prediction model in one embodiment;

[0093] Fig.10 A schematic diagram of a process for calculating a loss value of a shaping factor prediction model in one embodiment;

[0094] Fig.11is a structural block diagram of a downlink signal transmission device in an embodiment;

[0095] Fig.12 FIG. 4 is an internal structure diagram of a downlink signal transmission device in an embodiment. DETAILED DESCRIPTION

[0096] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0097] In an exemplary embodiment, Figure 1 As shown, a downlink signal transmission method is provided, and the method is applied to a base station as an example for explanation. The present application does not limit the execution device of the downlink signal transmission method, and includes the following steps 102 to 104.

[0098] Step 102, obtaining user coordinates and base station coordinates, and obtaining feature data according to the user coordinates and the base station coordinates.

[0099] In the embodiment of the present application, when the mobile user terminal needs to transmit a signal with the base station, the mobile user terminal transmits a reference signal to the base station, and the reference signal is used by the base station to perform subsequent beamforming processing on the mobile user terminal, that is, the base station receives the reference signal sent by the mobile user terminal through the antenna of the base station, and obtains the user coordinates of the mobile user terminal according to the reference signal. The coordinates of the current base station can be stored in the base station, and the characteristic data between the mobile user terminal and the base station can be calculated and obtained according to the obtained user coordinates and the base station coordinates of the base station.

[0100] Step 104, processing the feature data according to the shaping factor prediction model to obtain the shaping factor.

[0101] The shaping factor is a beamforming parameter when downlink signals are transmitted between the user terminal of the target user and the base station.

[0102] In an embodiment of the present application, a trained shaping factor prediction model is deployed in the base station. The base station processes the feature data between the mobile user terminal and the base station based on the shaping factor prediction model, maps the prediction result of the shaping factor through the position feature between the user coordinates and the base station coordinates, obtains the shaping factor used for beamforming, realizes beamforming for the mobile user terminal, and completes the downlink signal transmission to the mobile user terminal.

[0103] In an optional embodiment, the base station may save the base station coordinates of other base stations. The current base station may calculate and obtain feature data based on the base station coordinates of other base stations and the user coordinates, and implement beamforming of other base stations and mobile user terminals based on the feature data to complete downlink signal transmission to the mobile user terminal.

[0104] In the above-mentioned downlink signal transmission method, the feature data is processed by means of a shaping factor prediction model, and a nonlinear mapping from the user coordinates of low-dimensional information to the shaping factor prediction results of high-dimensional information is established. The shaping factor can be obtained quickly, and the complex channel estimation and shaping weight calculation process can be avoided. The requirements on the base station processing capability can be reduced, and the communication delay can be reduced, thereby improving the efficiency of obtaining the shaping factor prediction results.

[0105] In an exemplary embodiment, since the feature data needs to reflect more of the mapping relationship between the user coordinates of the low-dimensional information and the shaping factors of the high-order information, and the relationship between the information represented by the directly obtained user coordinates and the base station coordinates is relatively divergent, therefore, in the embodiment of the present application, further processing can be performed based on the user coordinates and the base station coordinates to obtain feature data that can reflect more mapping relationships, such as Figure 2 As shown, step 102 includes steps 202 to 204. Among them:

[0106] Step 202, calculating relative coordinates and relative angles according to user coordinates and base station coordinates.

[0107] In the embodiment of the present application, the base station performs coordinate conversion based on the relative position of the base station and the mobile user terminal. First, the relative coordinates between the mobile user terminal and the base station are calculated. For example, the base station coordinates are (x0, y0, z0), and the user coordinates of the mobile user terminal are (x b , Y b , Z b ), since the difference in the z-axis values ​​in the coordinates is small, the relative positions between the z-axes are discarded, such as Figure 3 As shown in "1. Calculate relative coordinates and angles", the relative coordinates (x1, y1, ang) are obtained based on the base station coordinates and user coordinates. The relative coordinate and angle calculation formula is as follows:

[0108] x1=x0-X b ,y1=y0-Y b ,ang=atan2(x1,y1) (1)

[0109] Among them, x1 and y1 are relative coordinates, and ang represents the relative angle between the base station and the mobile user terminal.

[0110] Step 204, based on the reduction amplitude correction method, the relative coordinates are corrected by reducing the amplitude to obtain corrected coordinates, and the corrected coordinates and the relative angle are used as feature data.

[0111] In the embodiment of the present application, since the amplitude difference between the coordinate value of the relative coordinate directly calculated according to the base station coordinate and the user coordinate and the coordinate value of the angle is large, the base station performs amplitude correction on the relative coordinate based on the amplitude reduction correction method, and reduces the amplitude of the relative coordinate, such as Figure 3 As shown in "2. Coordinate value correction" in the figure, the corrected coordinates (x, y, ang) are obtained. The calculation formula of the corrected coordinates is as follows:

[0112] x=x1 / 10,x=y1 / 10 (2)

[0113] At this time, the amplitude difference between the relative coordinates and the angle is small, so the corrected coordinates are used as feature data between the mobile user terminal and the base station.

[0114] In this embodiment, the relative coordinates and relative angles are calculated by using the user coordinates and the base station coordinates, which can reflect the characteristic relationship between the base station and the mobile user terminal, and perform amplitude correction on the relative coordinates. During the calculation process, the mapping relationship between the characteristic data and the shaping factor is improved, thereby improving the accuracy of the prediction when predicting the shaping factor through the characteristic data.

[0115] In an exemplary embodiment, Figure 4 As shown, Figure 4 is the model structure of the shaping factor prediction model, which includes convolution structure, attention structure, residual structure and full connection structure, such as Figure 5 As shown, step 104 includes steps 502 to 508. Among them:

[0116] Step 502: input the feature data into the shaping factor prediction model, extract the features of the feature data through the convolution structure, and obtain an initial feature map.

[0117] In the embodiment of the present application, in the convolution structure included in the shaping factor prediction model, the kernel size of the convolution kernel may be 3*3, the step size may be 2, and the padding size may be 3. The base station inputs the feature data into the shaping factor prediction model, such as Figure 4As shown, first, the feature data is convolved through the conv1d layer in the convolution structure, and then batch normalization is performed through the Batch Norm1d layer, and pooling is performed through the MaxPool1d layer to obtain an initial feature map, wherein an activation layer is also included between the Batch Norm1d layer and the MaxPool1d layer, and the activation layer can be a mish function (an activation function).

[0118] Step 504: weight and channel concatenate the initial feature map based on the attention structure to obtain a concatenated feature map.

[0119] In the embodiment of the present application, the attention structure included in the shaping factor prediction model includes a channel attention module (Channel Attention) and a spatial attention module (Spatial Attention), that is, Figure 4 As shown in Figure 1, the Channel atten layer and the Space atten layer in the attend1 structure (attention structure). The base station weights the initial feature map through the channel attention module of the shape factor prediction model, and performs channel splicing according to the spatial attention module to obtain the spliced ​​feature map.

[0120] Step 506, performing residual addition on the concatenated feature map based on the residual structure to obtain a target feature map.

[0121] In an embodiment of the present application, the residual structure in the shaping factor prediction model includes multiple residual layers. The base station adds the residuals of the spliced ​​feature map through the multiple residual layers to obtain a target feature map, which includes the features extracted by the convolution structure and the features of the residual information extracted by each residual layer.

[0122] Step 508, transform the dimension of the target feature map according to the fully connected structure and output the shaping factor.

[0123] In the present application embodiment, Figure 4 As shown, the fully connected structure (fc1 structure) in the shaping factor prediction model inherits the AvgPool1d (global average pooling) layer and the Dropout (random inactivation) layer. The fully connected structure connects all neurons in the Dropout layer, where the ratio of the Dropout layer can be set to 0.2. The base station converts the target feature map into an output of 32*128*2 dimensions, that is, a shaping factor of 32*128*2 dimensions, through the fully connected structure. The 32 dimensions of the 32*128*2 dimensions represent the number of antennas in the base station antenna array, the 128 dimensions represent the subcarriers, and the 2 dimensions represent the real and imaginary parts of the channel response.

[0124] In this embodiment, the initial feature map is obtained by feature extraction of the convolution structure, and the features in the feature data used to reflect the mapping relationship between the feature data and the shaping factor are preliminarily extracted. The initial feature map is weighted and channel spliced ​​by the attention structure to obtain a spliced ​​feature map, which further improves the effect of feature extraction. The residual structure is used to add the residuals of the spliced ​​feature map to solve the gradient disappearance and gradient explosion problems, and the residual information is integrated to obtain a target feature map that can significantly reflect the mapping relationship between the feature data and the shaping factor. Then, the target feature map is dimensionally transformed by the fully connected structure, and the shaping factor is output to complete the prediction of the shaping factor based on the user coordinates and the base station coordinates, avoiding the complex channel estimation and shaping weight calculation process, which can reduce the requirements for the base station processing capacity, while reducing communication delays, thereby improving the efficiency of the obtained shaping factor prediction results.

[0125] In an exemplary embodiment, the attention structure includes a channel attention layer and a spatial attention layer, such as Figure 6 As shown, step 504 includes steps 602 to 604. Among them:

[0126] Step 602, weighting the initial feature map according to the channel attention layer to obtain a weighted first feature map.

[0127] In an embodiment of the present application, the shaping factor prediction model in the base station passes through the channel attention module, and the shaping factor prediction model respectively subjects the initial feature map to global maximum pooling and average pooling operations, wherein the global maximum pooling and average pooling operations are respectively parallel pooling operations, and the two parallel pooling operations are input into a shared fully connected layer, and the results obtained by the global maximum pooling and the average pooling are respectively weighted through the shared fully connected layer, and the weighted first feature map is activated.

[0128] Step 604, performing channel splicing on the first feature map according to the spatial attention layer to obtain a second feature map after channel splicing, and performing convolution and activation on the second feature map to obtain a spliced ​​feature map.

[0129] In an embodiment of the present application, after the shaping factor prediction model obtains the channel attention feature map, the base station uses the channel attention feature map as the input of the spatial attention module (i.e., the Space atten layer), and performs global maximum pooling and average pooling operations on the channel attention feature map in the channel dimension through the spatial attention module, and splices the output result of the global maximum pooling with the output result of the average pooling operation to obtain a spliced ​​second feature map and performs convolution and activation operations to obtain a spliced ​​feature map.

[0130] In this embodiment, by performing feature extraction on the initial feature map in the channel dimension and the spatial dimension respectively, the feature extraction capability of the shaping factor prediction model can be further improved, thereby improving the accuracy of the shaping factor prediction model for shaping factor prediction.

[0131] In an exemplary embodiment, the residual structure includes a first residual layer, a second residual layer, a third residual layer and a fourth residual layer, such as Figure 7 As shown, step 506 includes steps 702 to 704. Among them:

[0132] Step 702: extract features from the concatenated feature map according to the first residual layer to obtain a first initial target feature map, and fuse features of the first initial target feature map with the concatenated feature map through a jump connection to obtain a first target feature map.

[0133] In an embodiment of the present application, each residual layer in the residual structure is composed of two convolutional layers and a jump connection. In the first residual layer, the two convolution modules are the same, the convolution kernel size of the convolution module is 3*3, the step size is 1, and no downsampling operation is performed. The base station extracts features from the spliced ​​feature map through the convolution module of the first residual layer to obtain a first initial target feature map, the first initial target feature map contains residual information, and the first initial target feature map is feature fused with the spliced ​​feature map through a jump connection. Specifically, the base station connects the first initial feature map with the spliced ​​feature map in a channel splicing manner, performs a convolution operation on the spliced ​​feature map, obtains a higher-level feature representation, and performs pixel-by-pixel addition or splicing operation on the feature representation after the convolution operation and the first initial feature map to obtain the first target feature map.

[0134] Step 704, performing feature extraction, downsampling processing and feature fusion on the first target feature map according to the second residual layer, the third residual layer and the fourth residual layer, respectively, to obtain a target feature map.

[0135] In the embodiment of the present application, the second residual layer, the third residual layer and the third residual layer respectively include two different convolution modules, such as Figure 4 As shown in the figure, the step size of the first convolution module is 2, and downsampling operation will be performed; the step size of the second convolution module is 1, and no downsampling operation will be performed. In the second residual layer, the third residual layer, and the fourth residual layer, the first target feature map is respectively extracted and downsampled by the first convolution module, and the feature map is extracted by the second convolution module, and the first target feature map after feature extraction and downsampling is fused with the concatenated feature map through jump connection to obtain the target feature map.

[0136] In this embodiment, feature extraction, downsampling and feature fusion are performed on the spliced ​​feature map through multiple residual layers, so that features in the residual information can be extracted and fused with the spliced ​​feature map, which can solve the problems of gradient disappearance and gradient explosion, and improve the effect of the target feature map on the mapping relationship between feature data and the shaping factor, thereby improving the accuracy of the shaping factor prediction model in predicting the shaping factor.

[0137] In an exemplary embodiment, the process of training the shaping factor prediction model by the base station includes two stages: model training and model fine-tuning. Therefore, it is necessary to construct a first sample data set for model training and a second sample data set for model fine-tuning through the full training samples. Figure 8 As shown, before step 102, the method further includes steps 802 to 812. Among them:

[0138] Step 802: Acquire multiple sample user coordinates, and use the shaping factor corresponding to each sample user coordinate as a sample label to form a first sample data set of the shaping factor prediction model.

[0139] In the embodiment of the present application, the base station obtains a sufficient amount of training samples, which can be channel response data at several discrete locations collected in advance. For example, the training samples can be 8000 geographic coordinates (loc), that is, sample user coordinates, and the shaping factor corresponding to each sample user coordinate. In model training, in order to improve the convergence effect, a sample data set with sufficient and rich data features is required. Therefore, the base station uses the full amount of training samples as the first sample data set of the shaping factor prediction model.

[0140] Step 804 , dividing the same sample user coordinates among the multiple sample user coordinates into the same sample group to obtain multiple sample groups.

[0141] The sample group includes sample user coordinates and sample labels corresponding to the sample user coordinates.

[0142] In an embodiment of the present application, the base station determines a sample group with the same user coordinates in the sample user coordinates. For example, among 8,000 training samples, if every 40 sample user coordinates are the same, then 200 sample groups with different sample user coordinates are included, and the sample user coordinates in each sample group may include different sample labels.

[0143] Step 806, based on the sample labels of the sample group, calculate the sample label mean, and use the sample label mean as the comprehensive sample label of the target sample group.

[0144] In the embodiment of the present application, the base station calculates the label mean based on a preset number of sample groups, and uses the label mean as the comprehensive sample label of the preset number of samples. For example, among 200 sample groups, the label mean is calculated for the labels of each of the 10 sample groups, and the label mean is used as the comprehensive sample label of the 10 sample groups.

[0145] Step 808: construct a second sample data set according to the sample user coordinates of each sample group and the comprehensive sample labels corresponding to the sample groups.

[0146] In an embodiment of the present application, the base station determines the comprehensive sample labels corresponding to the multiple groups of sample user coordinates and the Meizu sample user coordinates as a second sample data set, which is used to perform model fine-tuning on the shaping factor prediction model.

[0147] Step 810: Train the shaping factor prediction model according to the first sample data set to obtain a pre-trained model of the shaping factor prediction model.

[0148] In an embodiment of the present application, the base station pre-trains the shaping factor prediction model according to the first sample data to obtain a shaping factor prediction model that meets preset training conditions, wherein the preset training condition may be that the loss value meets a preset loss condition, or the preset training condition may be that the number of training times meets a preset training quantity.

[0149] Step 812, adjusting the model parameters of the pre-trained model according to the second sample data set to obtain a trained shaping factor prediction model.

[0150] In an embodiment of the present application, the second sample data set is a training sample after the sample label is smoothed. The base station fine-tunes the model parameters of the pre-trained model according to the second sample data set, and its feature extraction capability for the main features, thereby obtaining a trained shaping factor prediction model.

[0151] In this embodiment, the shaped silver prediction model is trained by using a feature-rich first sample data set, which can improve the model convergence effect and convergence speed; by smoothing the training sample labels to obtain a second sample data set, and using the second sample data set to fine-tune the model parameters of the shaped silver prediction model, the model generalization effect can be improved, thereby improving the accuracy of the shaped factor prediction model in predicting shaped silver.

[0152] In an exemplary embodiment, Fig. 9 As shown, step 810 includes steps 902 to 906 .

[0153] in:

[0154] Step 902: Calculate and obtain sample feature data based on each sample user coordinate and base station coordinate in the first sample data set.

[0155] In the embodiment of the present application, the method for calculating the sample feature data in the training of the shaping factor prediction model is similar to the method for calculating the feature data in step 202 and step 204. The method for calculating the sample feature data is not repeated in this embodiment.

[0156] Step 904, obtaining a shaping factor prediction result based on the shaping factor prediction model and the sample feature data.

[0157] In the embodiment of the present application, the base station inputs the sample feature data corresponding to each sample into the shaping factor prediction model to obtain the shaping factor prediction result corresponding to each sample.

[0158] Step 906, calculate the loss value of the shaping factor prediction model according to the spectral efficiency maximization loss function and the shaping factor prediction result. If the loss value is greater than the preset loss value, execute the shaping factor prediction model and sample feature data to obtain the shaping factor prediction result until the loss value is less than the preset loss value, and obtain the pre-trained model of the shaping factor prediction model.

[0159] In an embodiment of the present application, after obtaining the shaping factor prediction result based on the first sample data set, the base station calculates and obtains the loss value corresponding to the shaping factor prediction model under the current model parameters according to the spectral efficiency maximization loss function. If the loss value does not meet the preset loss condition, that is, the loss value is greater than the preset loss value, the shaping factor prediction model continues to be trained, and the sample feature data is repeatedly input into the shaping factor prediction model. Back propagation is performed according to the gradient descent method to update the model parameters until the loss value corresponding to the shaping factor prediction result output by the shaping factor prediction model is less than the preset loss value, and the current shaping factor prediction model is determined as a pre-trained model.

[0160] In this embodiment, by training the shaping factor prediction model, the model parameters are updated until the loss value of the shaping factor prediction model is less than the preset loss value, so that a shaping factor prediction model with a smaller output error can be obtained, thereby improving the accuracy of the output results of the shaping factor prediction model.

[0161] In an exemplary embodiment, Fig.10 As shown, step 906 includes steps 1002 to 1004. Among them:

[0162] Step 1002, calculate the target spectral efficiency and the predicted spectral efficiency respectively according to the real channel of the strongest flow shaping factor and the shaping factor prediction result.

[0163] In the embodiment of the present application, the loss function calculation formula based on spectral efficiency maximization is shown in the following formula (3):

[0164]

[0165] Where R represents the target spectrum effect, Represents the predicted spectrum effect. The base station calculates the target spectrum effect and the predicted spectrum effect respectively through the shaping factor prediction result corresponding to each sample, the strongest flow shaping factor and the real channel, and calculates the loss value of the shaping factor prediction model based on the target spectrum effect and the predicted spectrum effect. The target spectrum effect calculation method is shown in the following formula (4), and the predicted spectrum effect calculation method is shown in the following formula (5):

[0166]

[0167]

[0168] Among them, H and W are the real channel and the strongest flow shaping factor respectively, is the prediction result of the shaping factor, He is the equivalent channel, is to predict the equivalent channel, and σ is the noise variance.

[0169] Step 1004: Based on the target spectral efficiency, the predicted spectral efficiency and the spectral efficiency maximization loss function, the loss value of the shaping factor prediction model is obtained.

[0170] In the embodiment of the present application, the base station substitutes the calculated target spectral efficiency and predicted spectral efficiency into the spectral efficiency maximization loss function, that is, formula (3), and calculates and obtains the loss value of the shaping factor prediction model.

[0171] In this embodiment, the loss value of the shaping factor prediction model is calculated by using the target spectral efficiency, predicted spectral efficiency and spectral efficiency maximization loss function, thereby realizing the loss value calculation for the purpose of spectral efficiency. The loss value can better reflect the output accuracy of the current shaping factor prediction model and improve the accuracy of the loss value calculation.

[0172] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0173] Based on the same inventive concept, the embodiment of the present application also provides a downlink signal transmission device for implementing the downlink signal transmission method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more downlink signal transmission device embodiments provided below can refer to the limitations on the downlink signal transmission method above, and will not be repeated here.

[0174] In an exemplary embodiment, Fig.11 As shown, a downlink signal transmission device 1100 is provided, including: an acquisition module 1101 and a prediction module 1102, wherein:

[0175] The acquisition module 1101 is used to acquire the user coordinates and the base station coordinates, and obtain the feature data according to the user coordinates and the base station coordinates;

[0176] The prediction module 1102 is used to process the characteristic data according to the shaping factor prediction model to obtain the shaping factor; the shaping factor is a beamforming parameter when downlink signal transmission is performed between the user terminal of the target user and the base station.

[0177] In one embodiment, the acquisition module 1101 is specifically used for:

[0178] Calculate relative coordinates and relative angles based on user coordinates and base station coordinates;

[0179] Based on the reduction amplitude correction method, the relative coordinates are corrected by the reduction amplitude to obtain the corrected coordinates, and the corrected coordinates and the relative angle are used as feature data.

[0180] In one embodiment, the shaping factor prediction model includes a convolutional structure, an attention structure, a residual structure, and a fully connected structure;

[0181] The prediction module 1102 is specifically used for:

[0182] The feature data is input into the shaping factor prediction model, and the feature data is extracted through the convolution structure to obtain the initial feature map;

[0183] Based on the attention structure, the initial feature map is weighted and channel-concatenated to obtain a concatenated feature map;

[0184] Based on the residual structure, the residuals of the concatenated feature maps are added to obtain the target feature map;

[0185] The target feature map is transformed in dimension according to the fully connected structure, and the shaping factor is output.

[0186] In one embodiment, the attention structure includes a channel attention layer and a spatial attention layer;

[0187] The prediction module 1102 is specifically used for:

[0188] The initial feature map is weighted according to the channel attention layer to obtain a weighted first feature map;

[0189] The first feature map is channel-spliced ​​according to the spatial attention layer to obtain a second feature map after channel splicing, and the second feature map is convolved and activated to obtain a spliced ​​feature map.

[0190] In one of the embodiments, the residual structure includes a first residual layer, a second residual layer, a third residual layer and a fourth residual layer;

[0191] The prediction module 1102 is specifically used for:

[0192] Extract features from the concatenated feature map according to the first residual layer to obtain a first initial target feature map, and fuse features of the first initial target feature map with the concatenated feature map through a jump connection to obtain a first target feature map;

[0193] According to the second residual layer, the third residual layer and the fourth residual layer, feature extraction, downsampling processing and feature fusion are performed on the first target feature map respectively to obtain the target feature map.

[0194] In one embodiment, the apparatus 1100 further includes:

[0195] A first construction module is used to obtain a plurality of sample user coordinates, and use a shaping factor corresponding to each sample user coordinate as a sample label to form a first sample data set of a shaping factor prediction model;

[0196] A division module is used to divide the same sample user coordinates among the multiple sample user coordinates into the same sample group to obtain multiple sample groups; the sample group includes the sample user coordinates and the sample labels corresponding to the sample user coordinates;

[0197] A calculation module, used for calculating a sample label mean based on the sample labels of the sample group, and using the sample label mean as a comprehensive sample label of the target sample group;

[0198] A second construction module is used to construct a second sample data set according to the sample user coordinates of each sample group and the comprehensive sample labels corresponding to the sample groups;

[0199] A first training module is used to train a shaping factor prediction model according to a first sample data set to obtain a pre-trained model of the shaping factor prediction model;

[0200] The second training module is used to adjust the model parameters of the pre-trained model according to the second sample data set to obtain a trained shaping factor prediction model.

[0201] In one embodiment, the first training module is specifically used for:

[0202] Calculate and obtain sample feature data according to each sample user coordinate and base station coordinate in the first sample data set;

[0203] Based on the shaping factor prediction model and sample characteristic data, the shaping factor prediction result is obtained;

[0204] According to the spectral efficiency maximization loss function and the shaping factor prediction results, the loss value of the shaping factor prediction model is calculated. If the loss value is greater than the preset loss value, the shaping factor prediction result is obtained by executing the shaping factor prediction model and sample feature data until the loss value is less than the preset loss value, and a pre-trained model of the shaping factor prediction model is obtained.

[0205] In one embodiment, the first training module is specifically used for:

[0206] According to the real channel of the strongest flow shaping factor and the shaping factor prediction results, the target spectrum effect and the predicted spectrum effect are calculated respectively;

[0207] Based on the target spectral efficiency, predicted spectral efficiency and spectral efficiency maximization loss function, the loss value of the shaping factor prediction model is obtained.

[0208] Each module in the above-mentioned downlink signal transmission device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the downlink signal transmission device in the form of hardware, or can be stored in the memory in the downlink signal transmission device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0209] In an exemplary embodiment, Fig.12 As shown, a downlink signal transmission device is provided, including: a transceiver 1200, a processor 1210 and a memory 1220; wherein:

[0210] The transceiver 1200 is used to receive and send data under the control of the processor 1210 .

[0211] The processor 1210 is configured to implement the steps in the above method embodiments when executing a computer program.

[0212] The memory 1220 is used to store data used by the processor 1210 when performing operations.

[0213] Among them, Fig.12In the embodiment, the processor 1210 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or a complex programmable logic device (CPLD), and the processor may also adopt a multi-core architecture. The bus architecture may include any number of interconnected buses and bridges, and various circuits of one or more processors represented by the processor and the memory represented by the memory are linked together. The bus architecture may also link together various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface. The transceiver may be a plurality of components, namely, a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium, which may include a wireless channel, a wired channel, an optical cable, and the like. The processor is responsible for managing the bus architecture and general processing.

[0214] Those skilled in the art will understand that Fig.12 The structure shown in the figure is only a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the downlink signal transmission device to which the scheme of the present application is applied. The specific downlink signal transmission device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0215] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0216] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0217] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0218] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0219] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0220] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A downlink signal transmission method, characterized in that: The method comprises: Acquire user coordinates and base station coordinates, and obtain feature data according to the user coordinates and the base station coordinates; The characteristic data is processed according to a shaping factor prediction model to obtain a shaping factor; the shaping factor is a beamforming parameter when downlink signal transmission is performed between a user terminal of a target user and a base station.

2. The method according to claim 1, characterized in that The obtaining of feature data according to the user coordinates and the base station coordinates includes: Calculate relative coordinates and relative angles according to the user coordinates and the base station coordinates; The relative coordinates are corrected by reducing the amplitude based on the reduction amplitude correction method to obtain corrected coordinates, and the corrected coordinates and the relative angle are used as feature data.

3. The method according to claim 1, characterized in that The shaping factor prediction model includes a convolution structure, an attention structure, a residual structure and a fully connected structure; The step of processing the feature data according to the shaping factor prediction model to obtain the shaping factor includes: Inputting the feature data into a shaping factor prediction model, performing feature extraction on the feature data through the convolution structure, and obtaining an initial feature map; Based on the attention structure, weighting and channel-joining are performed on the initial feature map to obtain a joint feature map; Based on the residual structure, residual addition is performed on the concatenated feature map to obtain a target feature map; The target feature map is dimensionally transformed according to the fully connected structure, and a shaping factor is output.

4. The method according to claim 3, characterized in that The attention structure includes a channel attention layer and a spatial attention layer; The weighting and channel splicing of the initial feature map based on the attention structure to obtain a spliced ​​feature map includes: The initial feature map is weighted according to the channel attention layer to obtain a weighted first feature map; Channel splicing is performed on the first feature map according to the spatial attention layer to obtain a second feature map after channel splicing, and convolution and activation are performed on the second feature map to obtain a spliced ​​feature map.

5. The method according to claim 3, characterized in that: The residual structure includes a first residual layer, a second residual layer, a third residual layer and a fourth residual layer; The step of performing residual addition on the concatenated feature map based on the residual structure to obtain a target feature map includes: Extracting features from the concatenated feature map according to the first residual layer to obtain a first initial target feature map, and fusing features of the first initial target feature map with the concatenated feature map through a jump connection to obtain a first target feature map; According to the second residual layer, the third residual layer, and the fourth residual layer, feature extraction, downsampling processing, and feature fusion are performed on the first target feature map respectively to obtain a target feature map.

6. The method according to claim 1, characterized in that Before obtaining the user coordinates and the base station coordinates, the method further includes: Acquire multiple sample user coordinates, and use the shaping factor corresponding to each of the sample user coordinates as a sample label to form a first sample data set of the shaping factor prediction model; The same sample user coordinates among the multiple sample user coordinates are divided into the same sample group to obtain multiple sample groups; the sample group includes the sample user coordinates and the sample labels corresponding to the sample user coordinates; Based on the sample labels of the sample group, a sample label mean is calculated, and the sample label mean is used as a comprehensive sample label of the target sample group; constructing a second sample data set according to the sample user coordinates of each of the sample groups and the comprehensive sample labels corresponding to the sample groups; Training the shaping factor prediction model according to the first sample data set to obtain a pre-trained model of the shaping factor prediction model; The model parameters of the pre-trained model are adjusted according to the second sample data set to obtain a trained shaping factor prediction model.

7. The method according to claim 6, characterized in that The step of training the shaping factor prediction model according to the first sample data set to obtain a pre-trained model of the shaping factor prediction model includes: Calculate and obtain sample feature data according to each sample user coordinate and base station coordinate in the first sample data set; Based on the shaping factor prediction model and the sample characteristic data, obtaining a shaping factor prediction result; According to the spectral efficiency maximization loss function and the shaping factor prediction result, the loss value of the shaping factor prediction model is calculated. If the loss value is greater than the preset loss value, the shaping factor prediction result is obtained by executing the shaping factor prediction model and the sample feature data until the loss value is less than the preset loss value, thereby obtaining a pre-trained model of the shaping factor prediction model.

8. The method according to claim 7, characterized in that The step of calculating the loss value of the shaping factor prediction model according to the spectral efficiency maximization loss function and the shaping factor prediction result includes: According to the real channel of the strongest flow shaping factor and the shaping factor prediction results, the target spectrum effect and the predicted spectrum effect are calculated respectively; Based on the target spectral efficiency, the predicted spectral efficiency and the spectral efficiency maximization loss function, the loss value of the shaping factor prediction model is obtained.

9. A downlink signal transmission device, characterized in that: The device comprises: An acquisition module, used to acquire user coordinates and base station coordinates, and obtain feature data according to the user coordinates and the base station coordinates; The prediction module is used to process the characteristic data according to the shaping factor prediction model to obtain the shaping factor; the shaping factor is a parameter of beamforming when downlink signal transmission is performed between the user terminal of the target user and the base station.

10. A downlink signal transmission device, comprising a transceiver, a memory and a processor, wherein the transceiver is used to receive and send data under the control of the processor, and the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images