Channel estimation method, apparatus, device, storage medium, and computer program product

Abstract

The application relates to the technical field of satellite communication, and discloses a channel estimation method, a channel estimation device, a channel estimation equipment, a storage medium and a computer program product, which comprises the following steps: acquiring receiving signal information and inputting the receiving signal information into a preset channel estimation model; selecting a current layer cascade layer from a plurality of cascade layers and acquiring a previous layer cascade layer corresponding to the current layer cascade layer; calculating a channel update value through a gradient descent module according to a channel estimation value obtained by the previous layer cascade layer and the receiving signal information; calculating a current channel estimation value through a near-end mapping module according to the channel update value; if the current channel estimation value does not satisfy a preset stop condition, returning to the step of selecting the current layer cascade layer from the plurality of cascade layers until the current channel estimation value satisfies the condition, and taking the current channel estimation value as a target channel estimation value. Channel characteristics are automatically learned through iteration instead of relying on an inaccurate preset model, so that accurate estimation of a complete channel state is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of satellite communication, and particularly relates to a channel estimation method and device, equipment, a storage medium and a computer program product. BACKGROUND

[0002] For the channel estimation problem in a large-scale MIMO LEO (Multiple-Input Multiple-Output Low Earth Orbit) satellite communication system, the prior art mainly adopts a channel estimation method based on compressed sensing. However, this compressed sensing method has a priori dependence problem. Due to the high-speed movement of the satellite and the complexity of the channel propagation environment, the actual channel characteristics often differ significantly from the preset sparse model, which will cause a significant decrease in channel reconstruction accuracy and, in some cases, will cause the algorithm to fail to converge, making it difficult to guarantee the performance of the satellite communication system. SUMMARY

[0003] The main purpose of the present application is to provide a channel estimation method, device, equipment, storage medium and computer program product, which aims to solve the technical problem that the prior art channel estimation method based on compressed sensing cannot accurately estimate the complete channel state due to the high-speed movement of the satellite and the complexity of the channel propagation environment, resulting in a significant decrease in channel reconstruction accuracy.

[0004] To achieve the above purpose, the present application provides a channel estimation method, which comprises:

[0005] Obtaining received signal information and inputting the received signal information into a preset channel estimation model, wherein the preset channel estimation model comprises a plurality of cascaded layers, and the cascaded layers comprise a gradient descent module and a near-end mapping module;

[0006] Selecting a current layer from the plurality of cascaded layers and obtaining a previous layer corresponding to the current layer;

[0007] Calculating a channel update value by the gradient descent module according to a channel estimation value obtained from the previous layer and the received signal information;

[0008] Calculating a current channel estimation value by the near-end mapping module according to the channel update value;

[0009] If the current channel estimation value does not satisfy a preset stopping condition, returning to the step of selecting the current layer from the plurality of cascaded layers until the current channel estimation value satisfies the condition, and taking the current channel estimation value as a target channel estimation value.

[0010] Optionally, the step of obtaining the current channel estimation value by the near-end mapping module according to the channel update value comprises:

[0011] obtaining channel features generated by the last cascade layer;

[0012] fusing the channel features based on a spatial attention mechanism to obtain target fusion features;

[0013] obtaining a current channel estimation value according to the target fusion features and the channel update value.

[0014] Optionally, the step of fusing the channel features based on a spatial attention mechanism to obtain target fusion features comprises:

[0015] connecting the channel features along a channel dimension to obtain an initial feature map;

[0016] processing the initial feature map based on a preset convolution layer to generate a corresponding number of feature matrices, wherein the features represented by the feature matrices are different;

[0017] assigning weights to different spatial positions of the feature matrices, and fusing the feature matrices based on the weights to obtain initial fusion features;

[0018] adjusting the size of the initial fusion features, and performing element-wise addition of the adjustment result and encoded features output by an encoder residual block at a current layer of the channel features to obtain target fusion features.

[0019] Optionally, the feature matrices comprise a first matrix, a second matrix and a third matrix.

[0020] The step of assigning weights to different spatial positions of the feature matrices and fusing the feature matrices based on the weights to obtain initial fusion features comprises:

[0021] multiplying the transpose of the first matrix and the second matrix to obtain a product matrix;

[0022] calculating the product matrix based on a normalized exponential function to generate a spatial attention map, wherein each element in the spatial attention map represents the attention weight of the corresponding spatial position of the feature matrix;

[0023] weighting and aggregating the third matrix according to the spatial attention map to obtain initial fusion features.

[0024] Optionally, the step of obtaining a channel update value by the gradient descent module according to the channel estimation value obtained by the last cascade layer and the received signal information comprises:

[0025] obtaining a step training parameter, wherein the step training parameter is used to learn an optimal step strategy;

[0026] calculating a channel update value based on the gradient descent function and the received signal information as inputs.

[0027] Optionally, before the step of obtaining the received signal information and inputting the received signal information into the preset channel estimation model, the method further comprises:

[0028] initializing trainable parameters in the preset channel estimation model, wherein the trainable parameters include a step training parameter and parameters in the proximal mapping module;

[0029] optimizing the trainable parameters based on a preset training set and a loss function, and updating the optimized results to the preset channel estimation model.

[0030] In addition, to achieve the above object, the present application further provides a channel estimation device, which comprises:

[0031] an information obtaining module, configured to obtain received signal information and input the received signal information into a preset channel estimation model, wherein the preset channel estimation model comprises a plurality of cascade layers, and the cascade layers comprise a gradient descent module and a proximal mapping module;

[0032] a level determining module, configured to select a current cascade layer from the plurality of cascade layers and obtain a previous cascade layer corresponding to the current cascade layer;

[0033] a channel updating module, configured to calculate a channel update value through the gradient descent module based on a channel estimation value obtained from the previous cascade layer and the received signal information;

[0034] a channel estimation module, configured to calculate a current channel estimation value through the proximal mapping module based on the channel update value;

[0035] an estimation determining module, configured to return to the step of selecting the current cascade layer from the plurality of cascade layers if the current channel estimation value does not satisfy a preset stop condition, until the current channel estimation value satisfies the condition, and take the current channel estimation value as a target channel estimation value.

[0036] In addition, to achieve the above object, the present application further provides a channel estimation device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the channel estimation method as described above.

[0037] In addition, to achieve the above-mentioned purpose, the present application also provides a storage medium, which is a computer readable storage medium, and a computer program is stored on the storage medium, and the computer program is executed by a processor to implement the steps of the channel estimation method as described above.

[0038] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of the channel estimation method as described above.

[0039] In the present application, the received signal information is obtained and input into a preset channel estimation model, the preset channel estimation model comprises a plurality of cascade layers, the cascade layers comprise a gradient descent module and a near-end mapping module; a current layer cascade layer is selected from the plurality of cascade layers, and a previous layer cascade layer corresponding to the current layer cascade layer is obtained; a channel update value is calculated by the gradient descent module according to a channel estimation value calculated from the previous layer cascade layer and the received signal information; a current channel estimation value is calculated by the near-end mapping module according to the channel update value; if the current channel estimation value does not satisfy a preset stopping condition, the step of selecting the current layer cascade layer from the plurality of cascade layers is returned until the current channel estimation value satisfies the condition, and the current channel estimation value is taken as a target channel estimation value. The combination of the iterative optimization algorithm and the deep learning enables the network to automatically learn the channel characteristics from the data, instead of relying on the preset model which may not be accurate, thereby fundamentally solving the problem that the traditional compressed sensing method severely depends on the channel prior information, leading to performance degradation in a complex channel environment, and realizing accurate estimation of the complete channel state. BRIEF DESCRIPTION OF DRAWINGS

[0040] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and serve to explain the principles of the present application together with the specification.

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows, and obviously, other drawings can also be obtained by those skilled in the art without any creative labor.

[0042] Figure 1 The flowchart of the first embodiment of the channel estimation method of the present application;

[0043] Figure 2 The schematic diagram of the trainable near-end gradient descent network architecture;

[0044] Figure 3A flowchart of a second embodiment of the channel estimation method of the present application;

[0045] Figure 4 A structural diagram of a cross-layer feature attention fusion module;

[0046] Figure 5 A flowchart of a third embodiment of the channel estimation method of the present application;

[0047] Figure 6 A structural diagram of a channel estimation device of an embodiment of the present application;

[0048] Figure 7 A device structural diagram of a hardware running environment involved in the channel estimation method of an embodiment of the present application.

[0049] The purposes, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0050] It should be understood that the specific embodiments described herein are merely intended to explain the technical solutions of the present application, and are not intended to limit the present application.

[0051] In order to better understand the technical solutions of the present application, the following will be described in detail in combination with the accompanying drawings and specific embodiments.

[0052] The main solution of the embodiment of the present application is: obtaining received signal information, and inputting the received signal information into a preset channel estimation model, the preset channel estimation model including a plurality of cascaded layers, the cascaded layers including a gradient descent module and a near-end mapping module; selecting a current layer from the plurality of cascaded layers, and obtaining a previous layer corresponding to the current layer; calculating a channel update value through the gradient descent module according to a channel estimation value obtained by the previous layer and the received signal information; calculating a current channel estimation value through the near-end mapping module according to the channel update value; if the current channel estimation value does not satisfy a preset stop condition, returning to the step of selecting the current layer from the plurality of cascaded layers, until the current channel estimation value satisfies the condition, and taking the current channel estimation value as a target channel estimation value.

[0053] Since satellite communication systems are equipped with a large number of antenna arrays to improve system performance, two key challenges are brought about: the increase in the number of antennas leads to the need for more pilot signals to obtain complete channel state information, which increases system overhead and reduces effective data transmission rate; and the number of radio frequency links is much less than the number of antenna elements due to cost and complexity considerations, resulting in insufficient channel observation dimension. For channel estimation in large-scale MIMO LEO satellite communication systems, existing technologies mainly use channel estimation methods based on compressed sensing. This method attempts to use the sparse characteristics of the channel for channel reconstruction, and uses compressed sensing theory to recover high-dimensional channel information from low-dimensional observation data. However, this method has the following technical limitations: the compressed sensing method has a priori dependence problem. This method requires prior information such as the sparsity of the channel and the sparse basis to be determined in advance, and is heavily dependent on the basic assumption that the channel has sparse characteristics. In the actual MIMO LEO satellite communication scenario, due to the high-speed motion of the satellite and the complexity of the channel propagation environment, the actual channel characteristics often differ significantly from the preset sparse model. This mismatch between the prior information and the actual channel model will cause the channel reconstruction accuracy to decrease significantly, and in some cases the algorithm may not converge, making it difficult to guarantee system performance.

[0054] The present application provides an efficient channel estimation method that can overcome the problem of insufficient channel observation dimension under the condition of limited number of radio frequency links, and reduce the dependence on pilot signals through an efficient channel estimation algorithm to achieve accurate estimation of the complete channel state.

[0055] It should be noted that the execution subject of the present embodiment can be a computing service device with data processing, network communication and program running functions, such as a computer, or an electronic device capable of realizing the above functions. The present embodiment and the following embodiments will be described below with reference to a satellite communication system.

[0056] Based on this, the present application provides a channel estimation method, which is described in detail below with reference to Figure 1 , Figure 1 The flowchart of the first embodiment of the channel estimation method of the present application is shown in the figure.

[0057] In the present embodiment, the channel estimation method comprises:

[0058] Step S10, acquiring the received signal information and inputting the received signal information into a preset channel estimation model, wherein the preset channel estimation model comprises a plurality of cascaded layers, and the cascaded layers comprise a gradient descent module and a near-end mapping module.

[0059] It should be noted that the received signal information in the satellite communication system is received by the receiving end through the antenna array from the satellite, which can be represented by the received signal matrix. The preset channel estimation model can be a trainable proximal gradient descent network (TPGD-Net), which expands the iterative proximal gradient descent (PGD) algorithm into a layered network structure, which not only retains the theoretical basis of the PGD algorithm, but also introduces the adaptive ability of deep learning. Through this algorithm expansion method, the TPGD-Net not only has interpretability and theoretical guarantee, but also can fully utilize the powerful feature extraction and adaptive learning ability of deep learning.

[0060] It should be understood that the design of multiple cascade layers enables the channel estimation model to be iteratively optimized. Each layer further optimizes and adjusts the signal estimation value based on the previous layer. The gradient descent module of the cascade layer corresponds to the gradient descent step of the PGD algorithm, and the proximal mapping module corresponds to the proximal mapping step in the PGD algorithm.

[0061] Step S20, selecting a current cascade layer from the plurality of cascade layers, and obtaining a previous cascade layer corresponding to the current cascade layer.

[0062] It should be noted that the current cascade layer in the multi-layer cascade layer structure is a certain cascade layer that is currently being processed or focused on, and will be operated to obtain input, calculate output, etc. The previous cascade layer is located before the current cascade layer in the connection order of the cascade layers. The output of the previous cascade layer is usually used as the input of the current cascade layer.

[0063] Step S30, calculating a channel update value by the gradient descent module according to the channel estimation value calculated by the previous cascade layer and the received signal information.

[0064] It should be noted that the channel estimation value calculated by the previous cascade layer is an estimation result of the channel state calculated in the previous cascade layer in the iteration process of the preset channel estimation model, which is an approximate representation of the channel state based on the existing information (such as the received signal matrix, the pilot matrix, and the parameters of the preset channel estimation model itself) in the previous iteration, and will be used as important input information in the calculation of the current cascade layer for further optimization and updating of the channel estimation result.

[0065] It should be understood that the channel update value is an intermediate value for optimizing the channel estimation value, which can reflect the basic principle of gradient descent, i.e. updating the current channel estimation value along the negative direction of the gradient of the objective function.

[0066] In an example, a massive MIMO LEO channel can be estimated by solving the following optimization problem:

[0067]

[0068] where in this optimization problem, is a data fidelity term that ensures consistency between the estimated channel and the observed measurements. This data fidelity term is crucial for guaranteeing the reliability of the estimation result, as it directly reflects the difference between the estimated channel and the actual received signal. A regularization function is introduced to normalize the solution of the equation. The choice of the regularization function has a significant impact on the final estimation result, as it determines how to utilize the prior knowledge of the channel to improve the estimation performance. A weighting parameter is used to balance the data fidelity term and the regularization function, and the selection of the parameter needs to trade off the degree of model fitting to the observed data and the degree of dependence on the prior information. A larger value will make the model more inclined to rely on the prior information, while a smaller value will make the model more dependent on the observed data.

[0069] In this optimization model, represents the received signal matrix (which can be determined by the received signal information), is a known pilot matrix, which is a pre-designed and known matrix, usually composed of a series of specific training sequences, and is usually used to insert pilot signals (signals after multiplying the pilot matrix with the transmitted information) in the transmitted signals and estimate the channel characteristics using the known pilot matrix and the received signals containing the pilot signals, is the channel matrix to be estimated, represents the product of the pilot matrix and the channel matrix.

[0070] An effective method to solve the above optimization problem is the proximal gradient descent algorithm. This algorithm generates the estimated channel by iteratively updating the following steps: In the gradient descent module of TPGD-Net, the first layer takes the channel estimation value of the first layer as input and outputs the channel update value The formula is as follows:

[0071]

[0072] where, is the step size of gradient descent in the first layer, The product of the representative pilot matrix and the channel estimation value.

[0073] Further, in order to accelerate the convergence speed of the preset channel estimation model and improve the channel estimation efficiency, the step S20 can include:

[0074] obtaining a step training parameter, wherein the step training parameter is used to learn an optimal step strategy; and calculating a channel update value based on a gradient descent function with the channel estimation value and the received signal information as inputs.

[0075] It should be noted that, in order to enhance the flexibility of the preset channel estimation model and accelerate the convergence speed, the step is set as a trainable parameter in each layer, allowing the preset channel estimation model to learn the optimal step strategy during the training process, thereby possibly achieving faster convergence than the fixed step method. By changing the traditional fixed step to a trainable parameter, the optimization process can be adaptively adjusted to accelerate the convergence speed. This design not only improves the efficiency of channel estimation, but also reduces the computational burden of the system, making the method more suitable for actual deployment.

[0076] In step S40, a current channel estimation value is calculated by a near-end mapping module based on the channel update value.

[0077] For the convenience of understanding, the following examples are given, but they do not limit the present scheme. In an example, reference is made to Figure 2 , Figure 2 which is a trainable near-end gradient descent network architecture.

[0078] The traditional near-end gradient descent algorithm generates an estimated channel by iteratively updating the following steps:

[0079]

[0080] wherein, is a near-end operator depending on the regularization function , which encapsulates the influence of the regularization function, enabling the algorithm to effectively handle non-smooth regularization terms.

[0081] In existing millimeter wave channel estimation, most studies usually use norm as the regularization function to promote the sparsity of the beam space channel. The norm as a proxy for sparsity has been proven to be effective in many applications. However, using this hand-designed regularization function to express the true prior information of the channel may be oversimplified, which can lead to a decrease in estimation performance. This is because the

[0082] norm assumes that the channel coefficients are independent, while the actual millimeter wave channel can have more complex structures and correlations.To address this issue, the present scheme unfolds the PGD algorithm into a deep neural network and equips it with a data-driven proximal operator. This approach allows the deep neural network to extract prior information from channel data and perform implicit proximal operations. By learning complex patterns and structures in the data, this approach can overcome the limitations of hand-designed regularization functions, providing more accurate and robust channel estimation.

[0083] As shown in Figure 2 . The TPGD-Net consists of layers of cascaded layers, each corresponding to one iteration of the PGD algorithm. Each layer is composed of a gradient descent module (GDM) and a proximal mapping module (PMM). The GDM corresponds to the gradient descent step in the PGD algorithm. The output of the layer GDM, with as input, outputs the channel update value as follows:

[0084]

[0085] The PMM corresponds to the proximal mapping step in the PGD algorithm, which aims to estimate the channel from the input . The regularization function is introduced to obtain the estimated channel , and only appears in the proximal operator. Therefore, instead of learning a regularization function and explicitly calculating its proximal operator, a neural network is proposed to replace the proximal operator, which directly learns complex prior information from data without pre-specifying the form of the regularization function.

[0086] where the neural network in the PMM is a U-shaped network with encoder-decoder blocks and skip connections. This U-shaped network architecture design can capture and utilize multi-scale feature information, making it more effective in handling the complexity of the channel estimation problem. Furthermore, the neural network uses channel attention blocks (CAB) to extract important features from the GDM output, and encoder-decoder blocks for down-sampling (Down) and up-sampling (Up) to extract the encoder output and decoder output of the previous layer., to obtain multi-scale feature maps. The channel attention mechanism used allows the neural network to adaptively emphasize the most relevant feature channels, thereby improving the efficiency and effectiveness of feature extraction. Residual blocks (RBs) are used to extract features at each scale. The use of residual connections can alleviate the gradient vanishing problem in deep network training and allow the learning of identity mappings, preserving input information. Max-pooling and bilinear interpolation are used for downsampling and upsampling, respectively, enabling the processing of information at different spatial resolutions and thus capturing multi-scale features in the channel. The final output is a channel estimate and optimized features In addition, to address the information transmission problem in deep networks, a cross-layer feature attention fusion (CLFAF) module is introduced. This module allows information to flow freely between different depth levels by establishing direct connections between different network layers. Through an attention mechanism, the CLFAF module can intelligently identify and select the most relevant and important features, greatly improving the efficiency of feature information utilization. This design not only effectively alleviates the information bottleneck problem in deep networks, but also significantly enhances the adaptability to complex channel environments.

[0087] Therefore, the proposed neural network bypasses the explicit hand-designed regularization function, explores the inherent prior information of the millimeter wave channel, and implicitly performs proximal operations. The advantage of this approach is that it can learn complex prior structures from training data without being limited by the form of the predefined regularization function.

[0088] Step S50, if the current channel estimate does not meet the preset stopping condition, return to the step of selecting the current layer from the plurality of cascaded layers until the current channel estimate meets the condition, and take the current channel estimate as the target channel estimate.

[0089] It should be noted that the preset stopping condition can be that the number of iterations of the channel estimate reaches a maximum value, or that the current number of iterations exceeds the number of layers of the preset cascaded layer. The preset stopping condition can also be that the accuracy of the channel estimate reaches a certain threshold.

[0090] It can be understood that the channel estimation value is gradually optimized through multi-layer iteration. Each layer improves on the basis of the previous layer, so that the channel estimation value is closer and closer to the true value. The number of iteration layers can also be reasonably controlled to prevent overfitting problems caused by excessive iteration of the preset channel estimation model, reducing unnecessary computational overhead. When the current channel estimation value does not meet the preset stopping condition, the calculated channel estimation value is input as input into the next layer of the preset channel estimation model. When the current channel estimation value meets the preset stopping condition, it indicates that the preset channel estimation model has learned and adapted to the characteristics of the channel to a certain extent, and the last updated channel estimation value can be considered as a relatively stable and reliable channel estimation result that can be obtained under the current network structure and training conditions.

[0091] In the embodiment, the received signal information is obtained and input into a preset channel estimation model, the preset channel estimation model including a plurality of cascade layers, the cascade layers including a gradient descent module and a near-end mapping module; a current cascade layer is selected from the plurality of cascade layers, and a previous cascade layer corresponding to the current cascade layer is obtained; a channel update value is calculated by the gradient descent module according to a channel estimation value calculated from the previous cascade layer and the received signal information; a current channel estimation value is calculated by the near-end mapping module according to the channel update value; if the current channel estimation value does not meet a preset stopping condition, the step of selecting the current cascade layer from the plurality of cascade layers is returned until the current channel estimation value meets the condition, and the current channel estimation value is taken as a target channel estimation value. The combination of iterative optimization algorithm and deep learning enables the preset channel estimation model to automatically learn channel characteristics from data, rather than relying on a possibly inaccurate preset model, fundamentally solving the problem of severe dependence of traditional compressed sensing methods on channel prior information, leading to performance degradation in complex channel environments, and achieving accurate estimation of complete channel state.

[0092] Reference Figure 3 , Figure 3 The flowchart of the second embodiment of the channel estimation method of the present application is based on the first embodiment described above, and the second embodiment of the channel estimation method of the present application is proposed.

[0093] In the second embodiment, the step S40 includes:

[0094] In step S401, the channel features generated by the previous cascade layer are obtained.

[0095] It should be noted that the channel feature is related information capable of reflecting the channel characteristics, and each cascade layer optimizes the channel feature generated by the previous layer for subsequent further analysis, estimation of the channel state, etc., to help better restore the original signal of the sending end or accurately grasp the real situation of the channel.

[0096] It should be understood that when the current cascade layer generates the current channel estimation value, the optimized channel feature is also input into the next layer.

[0097] Step S402, based on the spatial attention mechanism, the channel features are fused to obtain the target fusion features.

[0098] It should be understood that since the PGD algorithm generates an estimated channel in each iteration, the TPGD-Net in the preset channel estimation model is designed to accept these channels as input and output the estimated channel at each layer. This design allows TPGD-Net to utilize and improve previous estimation results at each step. However, the limited feature information transmission between layers in TPGD-Net leads to a decline in channel estimation performance. This is because each layer mainly relies on the output of the previous layer without fully utilizing the information of all previous layers. In order to solve this limitation and considering the characteristics of millimeter wave channels (characterized by sparsity and concentration), a spatial attention mechanism can be used to selectively fuse features from different layers, helping the preset channel estimation model to focus on the most relevant spatial regions and features.

[0099] In an example, a cross-layer feature attention fusion module (CLFAF) based on a spatial attention mechanism can be proposed in TPGD-Net. The introduction of this module aims to enhance the information flow between the layers of the preset channel estimation model, thereby improving the overall estimation performance. The core idea of the CLFAF module is to use a spatial attention mechanism to selectively fuse features from different layers. This method is particularly suitable for handling the sparsity and concentration characteristics of millimeter wave channels, as it can help the preset channel estimation model to focus on the most relevant spatial regions and features. Specifically, the CLFAF module can perform the following operations: collect feature maps from multiple previous layers; use a spatial attention mechanism to assign weights to different spatial locations of each feature map; fuse features from different layers based on these weights; combine the fused features with the features of the current layer to produce a more rich and informative representation.

[0100] Further, in order to enable the preset channel estimation model to focus on key feature regions, enhance the representation ability of channel features, and improve the quality of feature fusion. The step S402 can include:

[0101] The channel features are connected along the channel dimension to obtain an initial feature map; the initial feature map is processed based on a preset convolution layer to generate a corresponding number of feature matrices, wherein the features represented by the feature matrices are different; weights are assigned to different spatial positions of the feature matrices, and the feature matrices are fused based on the weights to obtain initial fusion features; the size of the initial fusion features is adjusted, and the adjustment result is element-wise added to the encoded features output by the current layer of the encoder residual block in the channel features to obtain target fusion features.

[0102] It should be noted that the channel dimension is the number of channels in the feature map represented by the convolutional neural network, and each channel can be regarded as a specific feature representation of the input data. The feature matrix refers to the result output after the convolution layer processing, which represents the new feature information extracted or converted from the initial feature map in the form of a matrix. Different feature matrices carry different aspects of feature representation, i.e., each feature matrix reflects the different dimensions and different angles of the characteristics of the channel features contained in the initial feature map, for example, some feature matrices may focus on the spatial distribution characteristics of the channel, and some may focus on the frequency-related characteristics.

[0103] It should be understood that the number of feature matrices generated by connecting the channel features along the channel dimension can be determined according to the types of features, or can be determined by the number of convolution kernels.

[0104] It can be understood that the spatial attention mechanism is used to assign weights to different spatial positions of each feature map, and based on these weights, the features from different layers are fused, and then the fused features are combined with the features of the current layer. In this way, the preset channel estimation model can more effectively utilize the information accumulated during the entire estimation process, allowing the preset channel estimation model to dynamically determine the importance of spatial regions and features of previous layers to the current estimation, thereby potentially improving the accuracy and robustness of the estimation.

[0105] Further, in order to allow the preset channel estimation model to more effectively utilize and propagate information during the entire estimation process. The step of assigning weights to different spatial positions of the feature matrices and fusing the feature matrices based on the weights to obtain initial fusion features can include:

[0106] The transpose of the first matrix is multiplied by the second matrix to obtain a product matrix; the product matrix is calculated based on a normalized exponential function to generate a spatial attention map, wherein each element in the spatial attention map represents the attention weight of the corresponding spatial position of the feature matrix; the third matrix is weighted and aggregated according to the spatial attention map to obtain initial fusion features.

[0107] For ease of understanding, the following is illustrated by way of example, but does not limit the present solution. In an example, referring to Figure 4 , Figure 4 is a schematic diagram of a cross-layer feature attention fusion module structure. The CLFAF module in TPGD-Net first concatenates the feature maps along the channel dimension and , obtaining an initial feature map with a size of . Among them, C, H and W are the channel number, height and width of the feature map, respectively. This feature concatenation operation is the first step of the CLFAF module, allowing the CLFAF module to consider information from the current layer , the encoder output and the decoder output of the previous layer at the same time. In this way, the CLFAF module can comprehensively utilize the information of different levels and different processing stages to provide rich and diverse inputs for subsequent feature fusion, while reflecting an important insight that the accuracy of channel estimation can be improved by integrating information from multiple processing stages.

[0108] The initial feature map obtained above is input into a convolutional layer (conv) to obtain an intermediate feature with a size of . The use of convolution has several important roles: first, it allows cross-channel information fusion, enabling information exchange between different feature maps; second, it can adjust the number of channels of the feature map, controlling the computational complexity while keeping the spatial dimension unchanged; finally, it introduces nonlinearity, enhancing the model's expressive power. This step can be seen as a preliminary fusion and compression of the concatenated features, preparing more compact and information-rich feature representations for the subsequent attention mechanism.

[0109] The intermediate feature map is further processed by three convolutional layers (conv) to obtain feature maps and with a size of , which are then reshaped into a feature matrix with a size of . This design draws inspiration from the self-attention mechanism, especially the concepts of Query, Key and Value in the Transformer. By generating three different feature representations, the module prepares the necessary components for subsequent attention calculations. This design allows TPGD-Net to learn complex feature interaction patterns, rather than being limited to simple linear combinations.

[0110] Then the transpose of matrix is multiplied with matrix , and the resulting matrix is passed through a softmax function to obtain the spatial attention map :

[0111]

[0112] where each element in represents the attention weight of the corresponding spatial position in matrix and matrix . By computing the similarity between and , the spatial attention mechanism is realized, dynamically determining which spatial positions are important. The application of the softmax function ensures the non-negativity and sum of 1 of the attention weights, making the attention allocation more explicit and interpretable. For millimeter wave channels, the energy of the channel often concentrates in certain spatial directions or regions, and the spatial attention mechanism effectively captures the sparsity and clustering characteristics of the channel.

[0113] Then the transpose of matrix is multiplied with matrix , and the resulting matrix (initial fused features) is reshaped to size after scaling by learnable parameter . This step can be understood as weighted aggregation of matrix features according to attention weights. The learnable parameter provides an additional flexibility, allowing the preset channel estimation model to adjust the strength of the attention mechanism. This reshaped matrix is element-wise added to the feature map , resulting in the final feature fusion output :

[0114]

[0115] The final fusion operation combines attention-weighted features and original features, forming a residual structure. The use of residual connections helps to alleviate the problem of gradient vanishing in deep networks and allows the preset channel estimation model to retain original feature information when necessary. The resulting feature is a selective fusion of the current layer and the previous layer features. As the fusion feature of the current layer, it participates in the feature optimization process of the current layer.

[0116] Therefore, the CLFAF module significantly improves the cross-layer feature information transmission by introducing cross-layer feature fusion and attention mechanisms, thereby improving the channel estimation performance. This design not only allows the preset channel estimation model to more effectively utilize and propagate information throughout the estimation process, but also enables the preset channel estimation model to adaptively focus on the most relevant features and spatial locations. This capability is invaluable for complex tasks such as millimeter wave channel estimation, as it allows the preset channel estimation model to handle the time-varying nature and spatial selectivity of the channel.

[0117] In step S403, a current channel estimation value is obtained according to the target fusion feature and the channel update value.

[0118] In an example, the channel estimation value output by the proximal mapping module can be represented as:

[0119]

[0120] wherein denotes the parameters of the layer PMM. is the proximal operator. and are the features extracted from the layer at the scale in the encoder residual block and the decoder residual block. This expression reflects how the proximal mapping module integrates information from different scales and layers to generate the final channel estimation, demonstrating how the TPGD-Net transforms the proximal mapping step in traditional optimization algorithms into a learnable neural network operation. The advantage of this approach is that it can adaptively learn complex prior information without the need to explicitly define regularization functions.

[0121] In this embodiment, the channel features generated by the previous cascade layer are obtained; the channel features are fused based on a spatial attention mechanism to obtain target fusion features; and a current channel estimation value is obtained according to the target fusion features and the channel update value. The spatial attention mechanism is used to selectively fuse features of different layers, helping the preset channel estimation model to focus on the most relevant spatial regions and features, thereby improving the channel estimation performance.

[0122] Referring to Figure 5 , Figure 5 is a flowchart of a third embodiment of the channel estimation method of the present application, based on the second embodiment described above, a third embodiment of the channel estimation method of the present application is proposed.

[0123] In the third embodiment, before step S10, the method further comprises:

[0124] Step S01, initializing the trainable parameters in the preset channel estimation model, wherein the trainable parameters include a step training parameter and parameters in the proximal mapping module.

[0125] It should be noted that the trainable parameters are parameters in the channel estimation model that can be automatically adjusted during the training process according to the input data and the expected output results (such as accurate signal estimation values). By continuously optimizing these parameters, the model can better fit the data and improve the accuracy and performance of signal estimation.

[0126] It can be understood that the trainable parameters can include a step training parameter in the gradient descent module and an optimization in the proximal mapping module, such as a parameter for scaling the matrix in the cross-layer feature attention fusion module.

[0127] Step S02, optimizing the trainable parameters based on a preset training set and a loss function, and updating the optimized results to the preset channel estimation model.

[0128] It can be understood that the training process of optimizing the trainable parameters through the loss function and the preset training set can be regarded as reinterpreting the traditional iterative optimization algorithm under the deep learning framework. The interpretability and theoretical basis of the traditional method are retained, while the powerful expression ability and adaptability of deep learning are utilized, which not only improves the performance of the algorithm, but also enhances its robustness and adaptability under various complex channel conditions.

[0129] In an example, the TPGD-Net can be trained by using a stochastic gradient descent algorithm to optimize its network parameters to minimize the loss function :

[0130]

[0131] wherein, represents the step training parameter, represents the parameters of the layer PMM, and this loss function measures the Euclidean distance between the estimated channel and the real channel , which is accumulated on all layers. By minimizing this loss function, the preset channel estimation model can learn how to produce more and more accurate channel estimates at each layer.

[0132] In the embodiment, trainable parameters in the preset channel estimation model are initialized, wherein the trainable parameters include a step training parameter and parameters in the proximal mapping module; the trainable parameters are optimized based on a preset training set and a loss function, and the optimized result is updated to the preset channel estimation model. By changing the traditional fixed step to trainable parameters, the preset channel estimation model can adaptively adjust the optimization process, accelerate the convergence speed, and reduce the calculation burden of the system.

[0133] It should be noted that the above examples are only used for understanding the present application and do not constitute a limitation on the channel estimation method of the present application. More forms of simple transformation based on this technical concept are within the protection scope of the present application.

[0134] The present application also provides a channel estimation device, please refer to Figure 6 , the channel estimation device comprises:

[0135] The information acquisition module 10 is configured to acquire the received signal information and input the received signal information into the preset channel estimation model, wherein the preset channel estimation model comprises a plurality of cascade layers, and the cascade layers comprise a gradient descent module and a proximal mapping module.

[0136] The level determination module 20 is configured to select a current cascade layer from the plurality of cascade layers and acquire a previous cascade layer corresponding to the current cascade layer.

[0137] The channel update module 30 is configured to calculate a channel update value by the gradient descent module according to a channel estimation value obtained from the previous cascade layer and the received signal information.

[0138] The channel estimation module 40 is configured to calculate a current channel estimation value by the proximal mapping module according to the channel update value.

[0139] The estimation determination module 50 is configured to return to the step of selecting the current cascade layer from the plurality of cascade layers if the current channel estimation value does not satisfy a preset stop condition, until the current channel estimation value satisfies the condition, and take the current channel estimation value as a target channel estimation value.

[0140] The channel estimation device provided by the present application adopts the channel estimation method in the above embodiment, which can solve the technical problem that the existing channel estimation method based on compressed sensing cannot accurately estimate the complete channel state due to the high-speed motion of the satellite and the complexity of the channel propagation environment, resulting in a significant decrease in channel reconstruction accuracy. Compared with the prior art, the channel estimation device provided by the present application has the same beneficial effects as the channel estimation method provided by the above embodiment, and the other technical features in the channel estimation device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0141] The application provides a channel estimation device, comprising: at least one processor; and a memory connected with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the channel estimation method in the above embodiment one.

[0142] Reference will now be made to the following description Figure 7 which shows a structural diagram of a channel estimation device suitable for implementing the embodiments of the application. The channel estimation device in the embodiments of the application can include, but is not limited to, mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (Personal Digital Assistant), PADs (Portable Application Description), PMPs (Portable Media Player), vehicle-mounted terminals (for example, vehicle-mounted navigation terminals), and the like, and fixed terminals such as digital TVs, desktop computers, and the like. Figure 7 The channel estimation device shown is only an example, and should not bring any limitation to the functions and use range of the embodiments of the application.

[0143] As Figure 7As shown, the channel estimation device can include a processing apparatus 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a read only memory (ROM) 1002 or a program loaded from a storage apparatus 1003 into a random access memory (RAM) 1004. Various programs and data required for operation of the channel estimation device are also stored in the RAM 1004. The processing apparatus 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input apparatus 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output apparatus 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; the storage apparatus 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication apparatus 1009. The communication apparatus 1009 can allow the channel estimation device to communicate wirelessly or by wire with other devices to exchange data. Although the channel estimation device having various systems is shown in the figure, it should be understood that all of the shown systems are not required to be implemented or possessed. More or less systems can be alternatively implemented or possessed.

[0144] In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by a communication apparatus, or installed from the storage apparatus 1003, or installed from the ROM 1002. When the computer program is executed by the processing apparatus 1001, the above-mentioned functions defined in the methods of embodiments of the present disclosure are performed.

[0145] The channel estimation device provided by the present disclosure adopts the channel estimation method in the above-mentioned embodiments, and can solve the technical problem that the channel reconstruction precision is significantly reduced due to the high-speed movement of the satellite and the complexity of the channel propagation environment, and the complete channel state cannot be accurately estimated by the existing channel estimation method based on compressed sensing. Compared with the prior art, the channel estimation device provided by the present disclosure has the same beneficial effects as the channel estimation method provided by the above-mentioned embodiments, and other technical features in the channel estimation device are the same as the features disclosed in the previous embodiment method, which will not be repeated here.

[0146] It should be understood that portions of the application disclosed can be implemented in hardware, software, firmware, or combinations thereof. In the description of the embodiments above, specific features, structures, materials or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0147] The above description is merely illustrative of the application and is not intended to limit the scope of the application. Any changes and modifications that can be made to the application in accordance with the principles of the application would be readily apparent to those skilled in the art and the present application is therefore not limited to the description and examples contained herein but is only limited by the claims.

[0148] The application provides a computer readable storage medium having stored thereon computer readable program instructions (i.e., a computer program) for performing the channel estimation method in the above-described embodiments.

[0149] The computer readable storage medium provided by the application may, for example, be a U disk, but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more conductive wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present embodiment, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer readable storage medium can be transmitted in any suitable medium, including but not limited to electrical wire, optical cable, RF (Radio Frequency), etc., or any suitable combination of the above.

[0150] The above computer readable storage medium can be included in the channel estimation device; or can exist separately and not be assembled into the channel estimation device.

[0151] The above computer readable storage medium carries one or more programs, which, when executed by the channel estimation device, cause the channel estimation device to perform the channel estimation method described above.

[0152] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0153] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0154] The modules involved in the embodiments of the present application can be implemented in software or hardware. In some cases, the names of the modules do not constitute a limitation on the modules themselves.

[0155] The readable storage medium provided by the application is a computer readable storage medium, which stores computer readable program instructions (i.e. computer programs) for executing the above channel estimation method, and can solve the technical problem that the existing channel estimation method based on compressed sensing cannot achieve accurate estimation of the complete channel state due to the high-speed movement of the satellite and the complexity of the channel propagation environment, resulting in a significant decrease in channel reconstruction accuracy. Compared with the prior art, the computer readable storage medium provided by the application has the same beneficial effects as the channel estimation method provided by the above embodiments, which will not be repeated here.

[0156] The application also provides a computer program product comprising a computer program, which, when executed by a processor, implements the steps of the channel estimation method as described above.

[0157] The computer program product provided by the application can solve the technical problem that the existing channel estimation method based on compressed sensing cannot achieve accurate estimation of the complete channel state due to the high-speed movement of the satellite and the complexity of the channel propagation environment, resulting in a significant decrease in channel reconstruction accuracy. Compared with the prior art, the computer program product provided by the application has the same beneficial effects as the channel estimation method provided by the above embodiments, which will not be repeated here.

[0158] The above only describes some embodiments of the application, and does not limit the patent scope of the application, and any equivalent structural transformation, direct / indirect application in other related technical fields based on the technical concept of the application, and the content of the specification and drawings are included in the patent protection scope of the application.

Claims

1. A method of channel estimation, characterized by, The channel estimation method comprises: obtaining receiving signal information and inputting the receiving signal information into a preset channel estimation model, wherein the preset channel estimation model comprises a plurality of cascaded layers, and the cascaded layers comprise a gradient descent module and a near-end mapping module; selecting a current layer from the plurality of cascaded layers, and obtaining a previous layer corresponding to the current layer; calculating a channel update value by the gradient descent module according to a channel estimation value calculated from the previous layer and the receiving signal information; calculating a current channel estimation value by the near-end mapping module according to the channel update value; if the current channel estimation value does not satisfy a preset stop condition, returning to the step of selecting the current layer from the plurality of cascaded layers until the current channel estimation value satisfies the condition, and taking the current channel estimation value as a target channel estimation value; In the gradient descent module, the first... Layer and first Channel estimation of layer Input, output channel update value The formula is as follows: wherein, is the step size of the gradient descent in the layer, and , is a pilot matrix that is pre-designed and known, represents the product of the pilot matrix and the channel estimate value, denotes a received signal matrix; the step of calculating the current channel estimation value by the near-end mapping module according to the channel update value comprises: obtaining channel features generated by the previous layer; fusing the channel features based on a spatial attention mechanism to obtain target fusion features; calculating the current channel estimation value according to the target fusion features and the channel update value.

2. The channel estimation method of claim 1, wherein, the step of fusing the channel features based on the spatial attention mechanism to obtain the target fusion features comprises: connecting the channel features along a channel dimension to obtain an initial feature map; processing the initial feature map based on a preset convolution layer to generate a corresponding number of feature matrices, wherein the feature matrices represent different features; allocating weights to different spatial positions of the feature matrices, and fusing the feature matrices based on the weights to obtain initial fusion features; adjusting the size of the initial fusion features, and performing element-wise addition of the adjustment result and encoder residual block current layer output encoding features in the channel features to obtain target fusion features.

3. The channel estimation method of claim 2 wherein, The feature matrices comprise a first matrix, a second matrix and a third matrix. The step of allocating weights to different spatial positions of the feature matrices and fusing the feature matrices based on the weights to obtain initial fusion features comprises: multiplying the transpose of the first matrix and the second matrix to obtain a product matrix; calculating the product matrix based on a normalization exponential function to generate a spatial attention map, wherein each element in the spatial attention map represents the attention weight of the corresponding spatial position of the feature matrix; weighting and aggregating the third matrix according to the spatial attention map to obtain initial fusion features.

4. The channel estimation method of claim 1, wherein, The step of calculating a channel update value by the gradient descent module according to a channel estimation value calculated from the previous layer and the receiving signal information comprises: obtaining a step training parameter, wherein the step training parameter is used to learn an optimal step strategy; calculating a channel update value by inputting the channel estimation value and the receiving signal information into a gradient descent function.

5. The channel estimation method of any one of claims 1 to 4, wherein, Before the step of obtaining receiving signal information and inputting the receiving signal information into a preset channel estimation model, the method further comprises: initializing trainable parameters in the preset channel estimation model, wherein the trainable parameters include a step training parameter and parameters in the proximal mapping module; optimizing the trainable parameters based on a preset training set and a loss function, and updating the optimized results to the preset channel estimation model.

6. A channel estimation apparatus characterized by comprising: The device comprises: An information acquisition module is configured to acquire receive signal information and input the receive signal information into a preset channel estimation model, wherein the preset channel estimation model comprises a plurality of cascaded layers, and the cascaded layers comprise a gradient descent module and a proximal mapping module. A level determination module is configured to select a current cascaded layer from the plurality of cascaded layers and acquire a previous cascaded layer corresponding to the current cascaded layer. A channel update module is configured to calculate a channel update value by using the gradient descent module based on a channel estimation value obtained by the previous cascaded layer and the receive signal information. A channel estimation module is configured to calculate a current channel estimation value by using the proximal mapping module based on the channel update value. An estimation determination module is configured to return to the step of selecting the current cascaded layer from the plurality of cascaded layers if the current channel estimation value does not satisfy a preset stop condition, until the current channel estimation value satisfies the condition, and take the current channel estimation value as a target channel estimation value. In the gradient descent module, the first... Layer and first Channel estimation of layer Input, output channel update value The formula is as follows: wherein, is the step size of the gradient descent in the layer, is the step size of the gradient descent in the layer, and , is a pilot matrix that is pre-designed and known, represents the product of the pilot matrix and the channel estimate value, denotes a received signal matrix; The channel estimation module is further configured to acquire a channel feature generated by the previous cascaded layer, fuse the channel feature based on a spatial attention mechanism to obtain a target fused feature, and calculate the current channel estimation value based on the target fused feature and the channel update value.

7. A channel estimation device characterized by comprising: The device comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the channel estimation method according to any one of claims 1 to 5.

8. A storage medium, characterized by The storage medium is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the channel estimation method according to any one of claims 1 to 5.

9. A computer program product, characterised in that, The computer program product comprises a computer program, and the computer program is executed by the processor to implement the steps of the channel estimation method according to any one of claims 1 to 5.

Images