Channel estimation method and device for asynchronous access of large-scale users

Abstract

The invention provides a channel estimation method and device for asynchronous access of large-scale users, and relates to the technical field of wireless communication, the method adopts an unlicensed random access satellite technology to reduce access delay and signaling overhead, and the actual situation of asynchronous access of the users is considered. The maximum delay of a user accessing a satellite is set to expand a pilot frequency sequence, so that a satellite-ground uplink access model conforming to an actual access scene is constructed. Furthermore, the target channel estimation model is a VAMP-based model-data dual-drive network model, a VAMP-based model drive network is used for carrying out channel estimation and carrying out first denoising processing on a channel estimation result, and a data drive network is used for carrying out second denoising processing on the first denoising processing result. Therefore, according to the method, high-precision channel estimation can be realized at the satellite receiving end, and the technical problem of low channel estimation precision of the existing channel estimation method in an unlicensed asynchronous access scene is relieved.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communication, and in particular, to a channel estimation method and apparatus for large-scale user asynchronous access. Background Art

[0002] For the communication requirements of large-scale users to access satellites with high reliability and low latency, the protocol research on user access to satellites has profound research significance and high practical value. The traditional authorization-based random access process is complex, and collisions occur when a large number of users access the channel, resulting in a high probability of access failure, increasing the access latency and signaling overhead. In recent years, the license-free random access technology has become the preferred choice for large-scale users on the ground to access satellites because users can directly send data without waiting for the authorization of the base station after sending the preamble sequence, which can significantly reduce the access latency and signaling overhead of users. At the same time, considering that it is difficult for users to perfectly coordinate the transmission time of accessing the satellite, and due to the high-speed movement of the satellite, the arrival times of different users accessing the satellite are different, and users cannot be completely synchronized. Therefore, the above problems pose new requirements for the satellite to accurately estimate the channel state information.

[0003] The commonly used channel state estimation method in the prior art is the approximate message passing algorithm, but the approximate message passing algorithm usually completes accurate channel estimation based on an independent and identically distributed sub-Gaussian pilot matrix. However, in an actual communication system, due to the asynchronous access of users to the satellite, the orthogonality of the pilot matrix is destroyed. Therefore, when there is a small deviation between the pilot matrix actually received by the satellite and the independent and identically distributed sub-Gaussian pilot matrix, the approximate message passing algorithm is prone to divergence, and the channel estimation performance will be significantly degraded. In summary, the existing channel estimation methods have the technical problem of low channel estimation accuracy in the license-free asynchronous access scenario. Summary of the Invention

[0004] The purpose of the present invention is to provide a channel estimation method and apparatus for large-scale user asynchronous access, so as to alleviate the technical problem of low channel estimation accuracy existing in the existing channel estimation methods in the license-free asynchronous access scenario.

[0005] In a first aspect, the present invention provides a channel estimation method for large-scale user asynchronous access, including: obtaining the received signal at the satellite receiving end, the maximum delay of user access to the satellite, the pilot matrix of all users, and the delay information matrix; wherein, the pilot matrix is composed of the pilot sequences of all users; the delay information matrix is used to characterize the actual delay of each user accessing the satellite; based on the pilot matrix, the delay information matrix, and the maximum delay, determining the actual delay spread pilot matrix of all users; performing format conversion on the received signal at the satellite receiving end and the actual delay spread pilot matrix to obtain the received signal in real number form and the actual delay spread pilot matrix in real number form; using the target channel estimation model to process the received signal in real number form and the actual delay spread pilot matrix in real number form to obtain the channel estimation result in real number form; wherein, the target channel estimation model is a model-data dual-driven network model based on the vectorized approximate message passing algorithm VAMP. The model-driven network based on VAMP is used to perform channel estimation based on its input data and perform first denoising processing on the channel estimation result, and the data-driven network is used to perform second denoising processing on the first denoising processing result of the model-driven network based on VAMP; performing format conversion on the channel estimation result in real number form to obtain the channel estimation result in complex number form.

[0006] Optionally, based on the pilot matrix, the delay information matrix, and the maximum delay, determining the actual delay spread pilot matrix of all users includes: based on the pilot matrix and the maximum delay, determining the target extended pilot matrix of all users at all optional delays; based on the target extended pilot matrix and the delay information matrix, determining the actual delay spread pilot matrix.

[0007] Optionally, it further includes: constructing a training data set; wherein, the training data set includes multiple groups of training data, and each group of training data includes: the actual delay spread pilot matrix sample of all users, the effective channel vector sample, and the received signal sample at the satellite receiving end; performing format conversion on the training data set to obtain the training data set in real number form; using the initial channel estimation model to process the actual delay spread pilot matrix sample and the received signal sample in the target group of training data to obtain the corresponding channel estimation result in real number form; wherein, the target group of training data represents any group of training data in the training data set in real number form; calculating the loss function value based on the effective channel vector sample in the target group of training data and the corresponding channel estimation result in real number form; performing iterative training on the initial channel estimation model based on the loss function value until the preset iteration termination condition is reached to obtain the target channel estimation model.

[0008] Optionally, a training data set is constructed, including: randomly generating a plurality of actual delay spread pilot matrix samples; respectively simulating the transmission of the plurality of actual delay spread pilot matrix samples in satellite multipath fading channels with different signal-to-noise ratios to obtain corresponding received signal samples at a plurality of satellite receivers; wherein, the effective channel vector sample of each satellite multipath fading channel is expressed as: ; represents the effective channel vector sample, represents the active indication matrix of all users, , represents user 's activity, represents user is active, represents user is inactive, represents the number of users, represents the set of complex-valued channel responses of all users, and the complex-valued channel response of user at time and frequency is expressed as , represents the Rice factor, when represents the LoS channel response; when represents the NLoS channel response, represents the number of propagation paths; , represents the complex gain of user on the th path, represents the Doppler shift of user on the th path, represents the transmission delay of user on the th path, represents the antenna array response, represents the azimuth angle of the angle of arrival of user on the th path, represents the elevation angle of the angle of arrival of user on the th path.

[0009] Optionally, the initial channel estimation model includes: an initial VAMP-based model-driven network and an initial data-driven network; the initial VAMP-based model-driven network includes: a linear estimator, a first decoupler, a non-linear estimator, and a second decoupler; processing the actual delay spread pilot matrix samples and received signal samples in the target group training data using the initial channel estimation model includes: performing channel estimation processing on the actual delay spread pilot matrix samples and received signal samples using the linear estimator to obtain a first channel estimation result and a first mean variance; the first mean variance represents the average of the covariance diagonal values of the first channel estimation result; using the first decoupler to process the first channel estimation result and the first mean variance to obtain the mean and mean variance for the non-linear estimator to pass messages; the non-linear estimator performs channel estimation processing and first denoising processing on the mean and mean variance for it to pass messages to obtain a second channel estimation result and a second mean variance; the second mean variance represents the average of the covariance diagonal values of the second channel estimation result; the data-driven network performs second denoising processing on the second channel estimation result to obtain a channel estimation result in real number form; using the second decoupler to process the channel estimation result in real number form, the second mean variance, the mean and mean variance for the non-linear estimator to pass messages to update the mean and mean variance for the linear estimator to pass messages.

[0010] Optionally, the learning parameters of the linear estimator include: the environmental noise covariance, the left singular matrix obtained by performing singular value decomposition on the actual delay spread pilot matrix in real number form, the singular value matrix, and the right singular matrix.

[0011] Optionally, the initial data-driven network includes: a data preprocessing module, a convolution and pooling module, an upsampling and residual connection module, an SE attention block, and a denoising output module connected in sequence; the data preprocessing module is used to adjust the input data of the data-driven network to a specified dimension; the convolution and pooling module includes: a first sub-module, a second sub-module, and a third sub-module; the first sub-module and the second sub-module have the same structure, and the first sub-module includes: a convolutional layer, an activation layer, a normalization layer, and a pooling layer; the third sub-module includes: a convolutional layer, an activation layer, and a normalization layer; the upsampling and residual connection module includes: a fourth sub-module and a fifth sub-module; the fourth sub-module and the fifth sub-module have the same structure, and the fourth sub-module includes: an upsampling layer, a convolutional layer, an activation layer, a normalization layer, and a residual connection layer; the residual connection layer in the fourth sub-module is used to add and process the output result of the normalization layer in the fourth sub-module and the output result of the second sub-module; the residual connection layer in the fifth sub-module is used to add and process the output result of the normalization layer in the fifth sub-module and the output result of the data preprocessing module; the SE attention block includes: a global feature compression layer, a first fully connected layer, a second fully connected layer, and a feature extraction layer; the feature extraction layer is used to multiply the output result of the second fully connected layer and the output result of the upsampling and residual connection module; the denoising output module includes: a convolutional layer, a denoising layer, and a soft threshold processing layer; the denoising layer is used to subtract the output result of the convolutional layer in the denoising output module from the output result of the data preprocessing module.

[0012] In a second aspect, the present invention provides a channel estimation device for large-scale user asynchronous access, including: an acquisition module, configured to acquire the received signal of the satellite receiving end, the maximum delay of the user accessing the satellite, the pilot matrix of all users, and the delay information matrix; wherein, the pilot matrix is composed of the pilot sequences of all users; the delay information matrix is used to represent the actual delay of each user accessing the satellite; a determination module, configured to determine the actual delay spread pilot matrix of all users based on the pilot matrix, the delay information matrix, and the maximum delay; a first format conversion module, configured to perform format conversion on the received signal of the satellite receiving end and the actual delay spread pilot matrix to obtain a real-form received signal and a real-form actual delay spread pilot matrix; a channel estimation module, configured to process the real-form received signal and the real-form actual delay spread pilot matrix by using a target channel estimation model to obtain a real-form channel estimation result; wherein, the target channel estimation model is a model-data dual-driven network model based on the vectorized approximate message passing algorithm VAMP, the model-driven network based on VAMP is used to perform channel estimation based on its input data and perform first denoising processing on the channel estimation result, and the data-driven network is used to perform second denoising processing on the first denoising processing result of the model-driven network based on VAMP; a second format conversion module, configured to perform format conversion on the real-form channel estimation result to obtain a complex-form channel estimation result.

[0013] In a third aspect, the present invention provides an electronic device, including a memory and a processor. A computer program that can run on the processor is stored on the memory. When the processor executes the computer program, the method for channel estimation for large-scale user asynchronous access described in any one of the foregoing embodiments is implemented.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium. The computer-readable storage medium stores computer instructions. When the computer instructions are executed by a processor, the method for channel estimation for large-scale user asynchronous access described in any one of the foregoing embodiments is implemented.

[0015] Aiming at the problem of the decline in the accuracy of the traditional channel estimation method during large-scale user asynchronous access, the present invention provides a method for channel estimation for large-scale user asynchronous access. First, an unlicensed random access satellite technology is proposed to reduce the access delay and signaling overhead. Considering the actual situation of user asynchronous access, the maximum delay for users to access the satellite is set to expand the pilot sequence, thereby constructing a satellite-ground uplink access model that conforms to the actual access scenario. Further, the target channel estimation model of the present invention is a model-data dual-driven network model based on VAMP. The model-driven network based on VAMP is used for channel estimation and performs a first denoising process on the channel estimation result, and the data-driven network is used for performing a second denoising process on the result of the first denoising process. Therefore, this method can achieve high-precision channel estimation at the satellite receiving end, alleviating the technical problem of low channel estimation accuracy existing in the existing channel estimation method in the unlicensed asynchronous access scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a flowchart of a method for channel estimation for large-scale user asynchronous access provided by an embodiment of the present invention; Figure 2 It is a schematic diagram for determining the target extended pilot matrix of all users at all optional delays according to the maximum delay and the pilot matrices of all users; Figure 3 It is a schematic structural diagram of a channel estimation model provided by an embodiment of the present invention; Figure 4It is a network architecture diagram of a data-driven network provided by an embodiment of the present invention; Figure 5 It is a functional module diagram of a channel estimation device for large-scale user asynchronous access provided by an embodiment of the present invention; Figure 6 It is a schematic diagram of an electronic device provided by an embodiment of the present invention. Specific embodiments

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated herein can be arranged and designed in various different configurations.

[0019] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] The following will describe in detail some embodiments of the present invention with reference to the accompanying drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0021] Embodiment 1 Figure 1 It is a flowchart of a channel estimation method for large-scale user asynchronous access provided by an embodiment of the present invention. As Figure 1 shown, the method specifically includes the following steps: Step S102, obtain the received signal at the satellite receiving end, the maximum delay of user access to the satellite, the pilot matrix of all users, and the delay information matrix of all users.

[0022] Among them, the pilot matrix is composed of the pilot sequences of all users; the delay information matrix is used to characterize the actual delay of each user accessing the satellite.

[0023] Considering the actual situation that different users have different arrival times when accessing the satellite and users cannot be completely synchronized, the embodiment of the present invention proposes to construct a satellite-ground uplink access model that can conform to the actual access scenario. Specifically, the maximum delay of user access to the satellite is preset, and then after obtaining the pilot matrix of all users, the pilot matrix of all users is extended, and then combined with the delay information matrix of all users to construct the actual delay extended pilot matrix of all users. Finally, combined with the received signal at the satellite receiving end, the process of channel estimation for the satellite multipath fading channel is carried out.

[0024] In the embodiments of the present invention, the satellite-ground uplink access model can be expressed as: ; represents the received signal at the satellite receiving end, represents the actual delay spread pilot matrix of all users, represents the effective channel vector (i.e., the object to be estimated in the embodiments of the present invention), represents the noise signal, , represents the target extended pilot matrix of all users at all optional delays determined according to the pilot matrix of all users and the maximum delay of user access to the satellite, represents the pilot length, represents the maximum delay of user access to the satellite, represents the number of users, represents the delay information matrix of all users, , represents user 's delay information matrix, , let represents user 's actual access delay, (i.e., there are optional delays in total), then when , ; when , . That is to say, only one element in takes the value of 1, and the values of the remaining elements are all 0, and the element in that takes the value of 1 is used to characterize the index of the actual access delay situation of the user, that is,

[0025] Step S104, based on the pilot matrix, the delay information matrix and the maximum delay, determine the actual delay spread pilot matrix of all users.

[0026] After knowing the maximum delay of user access to the satellite, according to this maximum delay and the pilot matrix of all users, the pilot matrix of all users at all optional delays can be determined, denoted as: the target extended pilot matrix of all users at all optional delays. Multiplying the target extended pilot matrix by the delay information matrix of all users can determine the actual delay spread pilot matrix of all users.

[0027] Step S106: Convert the received signal at the satellite receiver and the actual delay spread pilot matrix into real number forms, obtaining the received signal in real number form and the actual delay spread pilot matrix in real number form.

[0028] Both the received signal at the satellite receiver and the actual delay spread pilot matrix mentioned above are in complex number form. To reduce the complexity and computational amount of the model, in the embodiments of the present invention, before performing channel estimation using the target channel estimation model, the received signal at the satellite receiver and the actual delay spread pilot matrix are first converted in format, specifically from complex number form to real number form, and then the results of the format conversion are input into the target channel estimation model.

[0029] The above format conversion from complex number to real number can be expressed as: , , where represents the real part of represents the imaginary part of represents the received signal in real number form, represents the real part of represents the imaginary part of represents the actual delay spread pilot matrix in real number form. In the embodiments of the present invention, , so its matrix dimension is related to the pilot length , the maximum delay and the number of users ; , so its dimension is related to the pilot length and the maximum delay .

[0030] Step S108: Process the received signal in real number form and the actual delay spread pilot matrix in real number form using the target channel estimation model, obtaining the channel estimation result in real number form.

[0031] In the embodiments of the present invention, a large amount of sample data is used to train and test the initial channel estimation model in advance, obtaining a target channel estimation model that can output high-accuracy channel estimation results. Among them, the target channel estimation model is a model-data dual-driven network model based on the vectorized approximate message passing algorithm VAMP. The model-driven network based on VAMP is used to perform channel estimation based on its input data and perform the first denoising process on the channel estimation result, and the data-driven network is used to perform the second denoising process on the result of the first denoising process of the model-driven network based on VAMP.

[0032] That is to say, in the embodiments of the present invention, a deep learning method is combined with the vectorized approximate message passing algorithm VAMP. After expanding the VAMP algorithm into a neural network, a VAMP-based model-driven network is obtained. Then, it is further combined with a data-driven network to achieve multiple denoising after channel estimation, thereby obtaining a high-precision channel estimation result.

[0033] Step S110: Perform format conversion on the channel estimation result in real number form to obtain a channel estimation result in complex number form.

[0034] In the embodiments of the present invention, both the input data and the output data of the target channel estimation model are in real number form. Therefore, after obtaining the channel estimation result in real number form, it is necessary to convert it into a more readable complex number form, that is, concatenate the channel estimation result in real number form into a channel estimation result in complex number form .

[0035] Aiming at the problem of the accuracy decline of traditional channel estimation methods during large-scale user asynchronous access, the embodiments of the present invention provide a channel estimation method for large-scale user asynchronous access. First, an unlicensed random access satellite technology is proposed to reduce access delay and signaling overhead. Considering the actual situation of user asynchronous access, the maximum delay for users to access the satellite is set to expand the pilot sequence, thereby constructing a satellite-ground uplink access model that conforms to the actual access scenario. Further, the target channel estimation model used in the embodiments of the present invention is a model-data dual-driven network model based on VAMP. The VAMP-based model-driven network is used for channel estimation and performs the first denoising process on the channel estimation result, and the data-driven network is used for performing the second denoising process on the result of the first denoising process. Therefore, this method can achieve high-precision channel estimation at the satellite receiving end, alleviating the technical problem of low channel estimation accuracy existing in the existing channel estimation methods in the unlicensed asynchronous access scenario.

[0036] In an optional embodiment, in the above step S104, based on the pilot matrix, the delay information matrix, and the maximum delay, determining the actual delay-expanded pilot matrix of all users specifically includes the following steps: Step S1041: Based on the pilot matrix and the maximum delay, determine the target expanded pilot matrix of all users at all optional delays.

[0037] Step S1042: Based on the target expanded pilot matrix and the delay information matrix, determine the actual delay-expanded pilot matrix.

[0038] Figure 2Schematic diagram for determining the target extended pilot matrix of all users at all optional delays according to the maximum delay and the pilot matrices of all users, as Figure 2 shown, after the maximum delay of the known user accessing the satellite is obtained, according to the maximum delay and the pilot sequence of each user, the extended pilot sequence of each user at each optional delay can be determined, and then the extended pilot matrix of each user at all optional delays can be obtained. The extended pilot matrices of all users at all optional delays constitute the target extended pilot matrix.

[0039] Specifically, the pilot sequence of user is expressed as , the pilot matrix of all users is , then the extended pilot sequence of user at delay is expressed as: , means there are zeros before , means there are zeros after , means the conjugate transpose of . Based on this, the extended pilot matrix of user at all optional delays is expressed as: , and the target extended pilot matrix of all users at all optional delays is expressed as: .

[0040] According to the description in the above text, after calculating and obtaining the delay information matrix of all users, through the formula the actual delay extended pilot matrix of all users can be calculated.

[0041] In an alternative embodiment, the embodiment of the present invention further includes the following steps: Step S201, constructing a training data set; wherein, the training data set includes multiple groups of training data, and each group of training data includes: the actual delay extended pilot matrix sample of all users, the effective channel vector sample, and the received signal sample at the satellite receiving end.

[0042] Step S202, converting the format of the training data set to obtain a training data set in real number form.

[0043] Step S203: Process the actual delay spread pilot matrix samples and received signal samples in the target group of training data using the initial channel estimation model to obtain the corresponding channel estimation results in real number form; where the target group of training data represents any group of training data in the training data set in real number form.

[0044] Step S204: Calculate the loss function value based on the effective channel vector samples in the target group of training data and the corresponding channel estimation results in real number form.

[0045] Step S205: Iteratively train the initial channel estimation model based on the loss function value until the preset iteration termination condition is reached to obtain the target channel estimation model.

[0046] Specifically, each parameter in each group of training data should conform to the satellite uplink access model constructed in the embodiments of the present invention. To ensure the generalization ability and universality of the model, after constructing a large number of actual delay spread pilot matrix samples of all users, different actual delay spread pilot matrix samples should be transmitted in satellite multipath fading channels with different signal-to-noise ratios respectively, so as to obtain a large number of received signal samples at the satellite receiving end.

[0047] Combined with the description in the above text, the input data of the channel estimation model are the actual delay spread pilot matrix samples and received signal samples in real number form, and the output data are the corresponding channel estimation results in real number form. Therefore, before performing channel estimation using the initial channel estimation model, the training data set in complex number form should be adjusted to real number form, and then the initial channel estimation model is used to perform channel estimation on the sample data. After the model outputs the prediction result (channel estimation result in real number form), based on the labels of each group of training data (i.e., the effective channel vector samples), the loss function value predicted by the model can be calculated, and then the initial channel estimation model is iteratively trained using the loss function value until the preset iteration termination condition is reached.

[0048] The embodiments of the present invention do not specifically limit the function expression of the loss function, as long as the error between the channel estimation result in real number form and the corresponding effective channel vector sample is positively correlated with the loss function value. In the embodiments of the present invention, the preset iteration termination condition can be that the loss function value converges below the preset threshold, or it can be that the specified number of training times is reached, and the user can set it according to actual needs.

[0049] In an optional implementation manner, the above step S201, constructing a training data set, specifically includes the following steps: Step S2011: Randomly generate a plurality of actual delay spread pilot matrix samples.

[0050] Specifically, referring to the method for determining the actual delay spread pilot matrix of all users based on the maximum delay of the user accessing the satellite, the pilot matrix of all users, and the delay information matrix in the above text, by customizing the setting of the maximum delay, randomly generating the pilot matrix of all users using Gaussian distribution, and randomly generating the delay information matrix, multiple actual delay spread pilot matrix samples can be generated.

[0051] Step S2012: Simulate the transmission of multiple actual delay spread pilot matrix samples in satellite multipath fading channels with different signal-to-noise ratios to obtain corresponding received signal samples at multiple satellite receivers.

[0052] Among them, the effective channel vector sample of each satellite multipath fading channel is expressed as: ; represents the effective channel vector sample, , represents the active indication matrix of all users, , represents user 's activity level, represents user is active, represents user is inactive, represents the number of users, represents the set of complex-valued channel responses of all users.

[0053] In the embodiment of the present invention, considering that most satellite communications are line-of-sight transmissions, it is assumed that obeys the Rice fading distribution with a Rice factor of , so the complex-valued channel response of user at time and frequency is expressed as , represents the Rice factor, when represents the LoS channel response; when represents the NLoS channel response, represents the number of propagation paths.

[0054] , represents the complex gain of user on the th path, represents the Doppler frequency shift of user on the th path, represents the transmission delay of user on the th path, represents the antenna array response, represents the user at the azimuth angle of the angle of arrival on the represents the user at the elevation angle of the angle of arrival on the

[0055] Antenna array response , where , represents the response of the antenna decomposition on the x-axis, , represents the response of the antenna decomposition on the y-axis, represents the number of antennas of the satellite along the x-axis, represents the number of antennas of the satellite along the y-axis, represents the total number of antennas in the antenna array. If only considering constructing a satellite channel model with a single antenna and a single beam, then there is , .

[0056] In the embodiment of the present invention, by expanding the vector approximate message passing algorithm (VAMP) into a neural network, using VAMP to perform singular value decomposition on the aligned orthogonal pilot matrix (i.e., the actual delay spread pilot matrix in real number form), and iteratively learning the measurement noise variance by setting learnable parameters, optimizing the noise adjustment term of the linear estimator to adapt to the noise in the actual observation environment, and at the same time iteratively learning the key parameters of the non-linear shrinkage function to optimize the denoising effect of the non-linear estimator, then introducing neural networks such as CNN (i.e., data-driven network) to further denoise the noisy data, enhancing the adaptability of the channel estimation method to complex noise, and finally improving the channel estimation accuracy at the satellite receiver.

[0057] VAMP adapts to channel estimation under a wide class of random matrices by performing singular value decomposition on the known pilot matrix. The specific steps are according to Bayes' formula , introducing the variable to be estimated is split into and . In one iteration, first calculate the message passing of and perform belief estimation at , where respectively represent the mean and average variance used by the linear estimator to pass messages in the k-th iteration, is the noise covariance in the k-th iteration. Then, by the variable node Transmit a message to , Transmit a message to , and then estimate at . Among them, the prior is a Bernoulli Gaussian distribution, , respectively represent the mean and average variance used by the nonlinear estimator to transmit messages in the k-th iteration. Finally, the message is transmitted back in the reverse direction and iterated continuously. Among them, both represent Gaussian distributions, represents the identity matrix.

[0058] In an alternative embodiment, the initial channel estimation model includes: an initial VAMP-based model-driven network and an initial data-driven network; the initial VAMP-based model-driven network includes: a linear estimator, a first decoupler, a nonlinear estimator, and a second decoupler.

[0059] Figure 3 is a schematic structural diagram of a channel estimation model provided by an embodiment of the present invention. As Figure 3 shown, the model-driven network mainly uses a linear estimator and a nonlinear estimator . The linear estimator depends on the parameter , and the nonlinear estimator depends on the characteristics of the shrinkage function parameter . In the embodiment of the present invention, the linear estimation stage is parameterized as . Among them, , that is, after performing singular value decomposition on , is the left singular matrix, is the singular value matrix, is the right singular matrix, is the environmental noise covariance. Through multiple rounds of iterative learning and , the noise adjustment term of the linear estimator is optimized to adapt to the global noise of the actual observation environment, and at the same time, the denoising effect of the nonlinear estimator on local noise is optimized.

[0060] Based on the above description, it can be known that in the embodiment of the present invention, the learning parameters of the linear estimator include: the environmental noise covariance, the left singular matrix, the singular value matrix, and the right singular matrix obtained after performing singular value decomposition on the actual delay spread pilot matrix in real number form.

[0061] ​The embodiment of the present invention introduces a data-driven network on the basis of a VAMP-based model-driven network, and iteratively trains its model parameters along with the model-driven network , so as to directly denoise the signal to be estimated. After the initial channel estimation model performs the k-th round of training, the channel estimation result output by the linear estimator is expressed as: , and the channel estimation result output by the non-linear estimator is expressed as: , and the channel estimation result output by the data-driven network is expressed as .

[0062] The following details the output processing flow of the initial channel estimation model. In the above step S203, the initial channel estimation model is used to process the actual delay spread pilot matrix sample and the received signal sample in the target group training data, which specifically includes the following steps: Step S2031, use the linear estimator to perform channel estimation processing on the actual delay spread pilot matrix sample and the received signal sample to obtain a first channel estimation result and a first average variance; the first average variance represents the average of the diagonal values of the covariance of the first channel estimation result.

[0063] Specifically, when the initial channel estimation model performs the k-th round of training, , that is, estimate Essentially These two Gaussians are multiplied approximately to obtain , so the mean is obtained from the mean of and ; the average variance is obtained from the average variance of and calculated.

[0064] The model parameters of the linear estimator are , therefore, after inputting the actual delay spread pilot matrix sample and the received signal sample, the following calculations are specifically performed: , ; where represents the first channel estimation result output by the linear estimator during the k-th round of training, represents the first average variance output by the linear estimator during the k-th round of training, represents the average variance used by the linear estimator to transmit messages during the k-th round of training, represents the mean used by the linear estimator to transmit messages during the k-th round of training. During the k-th round of training, , , , that is, represents rank of indicating during the k-th round of training positive singular values in

[0065] Step S2032: Use the first decoupler to process the first channel estimation result and the first mean variance to obtain the mean and mean variance for the nonlinear estimator to transmit messages.

[0066] In the embodiment of the present invention, the first decoupler specifically performs the following calculations: , ; where represents the mean for the nonlinear estimator to transmit messages during the k-th round of training, represents the mean variance for the nonlinear estimator to transmit messages during the k-th round of training.

[0067] Step S2033: The nonlinear estimator performs channel estimation processing and first denoising processing on the mean and mean variance for it to transmit messages to obtain a second channel estimation result and a second mean variance; the second mean variance represents the average of the covariance diagonal values of the second channel estimation result.

[0068] Specifically, during the k-th round of training of the initial channel estimation model, , estimate is the same as the estimation described above and will not be elaborated here. The specific operations performed by the nonlinear estimator need to be based on the shrinkage function it adopts, and users can select according to actual needs. The shrinkage function of the nonlinear estimator is based on its model parameters in the k-th round outputs the second channel estimation result and the second mean variance, , represents the second channel estimation result output by the nonlinear estimator during the k-th round of training, represents the second mean variance output by the nonlinear estimator during the k-th round of training.

[0069] Step S2034: The data-driven network performs second denoising processing on the second channel estimation result to obtain a real-number form channel estimation result. This data processing flow can be expressed as: , represents the real-number form channel estimation result output by the data-driven network during the k-th round of training.

[0070] Step S2035: Use the second decoupler to process the real-number form channel estimation result, the second mean variance, the mean and mean variance for the nonlinear estimator to transmit messages to update the mean and mean variance for the linear estimator to transmit messages.

[0071] In the embodiment of the present invention, the second decoupler specifically performs the following calculations: , ; where, represents the mean used by the linear estimator to transmit messages during the (k + 1)-th round of training, represents the average variance used by the linear estimator to transmit messages during the (k + 1)-th round of training.

[0072] Figure 4 is a network architecture diagram of a data-driven network provided by an embodiment of the present invention. The data-driven network learns with the iterative training of the model-driven network parameters such as CNN convolution kernels and biases in Figure 4 As shown, the initial data-driven network includes: a data preprocessing module, a convolution and pooling module, an upsampling and residual connection module, an SE attention block, and a denoising output module connected in sequence. Based on the above network structure, local denoising is performed on the output by the non-linear estimator in the model-driven network during the k-th training, and finally is output.

[0073] The data preprocessing module is used to adjust the input data of the data-driven network to a specified dimension.

[0074] Optionally, six-layer convolution weights and biases are defined and initialized, and the defined parameters are added to the parameter list cnn_paras. Then, two max pooling layers are defined. Then, variables lam for controlling the soft threshold shrinkage intensity and tex for gradient scaling are defined and initialized, and they are also added to the parameter list cnn_paras. Finally, the two-dimensional tensor , which is the signal to be denoised, is input, and the input data is dimensionally expanded by adding an additional dimension after to align with the input dimension of the convolution layer, and a four-dimensional tensor cnn_in is output.

[0075] The convolution and pooling module includes: a first sub-module, a second sub-module, and a third sub-module; the first sub-module and the second sub-module have the same structure. The first sub-module includes: a convolution layer, an activation layer, a normalization layer, and a pooling layer; the third sub-module includes: a convolution layer, an activation layer, and a normalization layer; Optionally, the convolution and pooling module is divided into 3 sub-modules (i.e., Figure 4Among the three layers marked, first, the input cnn_in is convolved using cnn_w1 in the convolutional layer to extract local features. Then, the leaky_relu activation function is used to introduce non-linearity. Next, batch normalization is performed on the output of the convolutional layer to accelerate the training process of the model, and the output data is bn1. In the pooling layer, downsampling is performed on the data bn1 to reduce the feature map dimension and retain important features of the data. The data processing steps of the other two sub-modules are the same, to further extract data features, and the output data of the convolutional and pooling modules is bn3.

[0076] The upsampling and residual connection module includes: a fourth sub-module and a fifth sub-module; the structures of the fourth sub-module and the fifth sub-module are the same. The fourth sub-module includes: an upsampling layer, a convolutional layer, an activation layer, a normalization layer, and a residual connection layer.

[0077] The residual connection layer in the fourth sub-module is used to add and process the output result of the normalization layer in the fourth sub-module and the output result of the second sub-module.

[0078] The residual connection layer in the fifth sub-module is used to add and process the output result of the normalization layer in the fifth sub-module and the output result of the data preprocessing module.

[0079] Optionally, the upsampling and residual connection module is divided into 2 sub-modules. In the fourth sub-module, first, the data is upsampled using the nearest neighbor interpolation method to restore the dimension of the feature map. Then, to adjust the number of channels, a convolution operation is performed on the data, and the leaky_relu activation function is used to introduce non-linearity. Next, batch normalization is performed on the output of the convolutional layer to accelerate the training process of the model, and the output data is bn4. Then, to alleviate the vanishing gradient, data features are extracted, and residual connection is used to add bn4 and the output result of the second sub-module in the convolutional and pooling module to enhance the training effect of the model. In the fifth sub-module, the same operations of upsampling, convolution, activation, and normalization are performed on the data, and the output data is bn5. Then, residual connection is used to add bn5 and the output result cnn_in of the data preprocessing module to alleviate the vanishing gradient and improve the training stability.

[0080] The SE attention block includes: a global feature compression layer, a first fully connected layer, a second fully connected layer, and a feature extraction layer.

[0081] The feature extraction layer is used to multiply the output result of the second fully connected layer and the output result of the upsampling and residual connection module.

[0082] In the embodiments of the present invention, the SE attention block includes a global feature compression layer, two fully-connected layers, and a feature extraction layer, which dynamically adjusts the channel attention weighting of the data and enhances the model's feature extraction ability in different channels. First, global statistical information of each channel is extracted in the global feature compression layer to eliminate the interference of the spatial dimension. Then, channel attention weights are generated in the two fully-connected layers. Finally, in the feature extraction layer, the original input feature map is multiplied by the channel weights channel by channel to enhance the features of important channels.

[0083] The denoising output module includes: a convolutional layer, a denoising layer, and a soft threshold processing layer.

[0084] The denoising layer is used to perform subtraction processing on the output result of the data preprocessing module and the output result of the convolutional layer in the denoising output module.

[0085] The data processing flow of the denoising output module is as follows: First, the convolutional layer in the denoising output module integrates all feature output data conv6. Then, the denoising layer subtracts the convolutional result conv6 from the output result cnn_in of the data preprocessing module to obtain the denoised feature map. Then, the feature map is reconstructed into a two-dimensional tensor xhat, and soft threshold processing is performed on xhat to calculate the gradient to remove the noise of the input two-dimensional tensor data, and finally the function outputs . In actual use, the number of network layers and the network parameters of each layer can be modified according to the actual situation to improve the generalization ability and robustness of the network.

[0086] Optionally, the networks of the channel estimation model in the embodiments of the present invention are all implemented by TensorFlow. In the k-th training, the initial learning rate of the channel estimation model is 0.001, and the Adam optimizer is used for training to minimize the Normalized Mean Square Error (NMSE). . To achieve accurate channel estimation, the present invention trains the channel estimation model from 0 to 30 dB with a step size of 5 dB.

[0087] To verify the performance of the method provided in the embodiments of the present invention, the following experimental cases are provided. For the dataset generated based on key parameters such as the number of users being 300, the user activity probability being 0.08, the pilot length being 80, the maximum delay number being 10, and the number of satellite antennas being 1, it is verified that the model-data dual-driven channel estimation method of the embodiments of the present invention improves the NMSE performance by about 5.19 dB compared with the single-driven network algorithm at an SNR of 30 dB.

[0088] When comparing the embodiments of the present invention with the method of channel estimation that relies on a single data-driven or model-driven approach, the single-driven approach lacks targeted optimization and it is difficult to take into account the complex and changing channel environment as well as the massive noisy data. The embodiments of the present invention innovate from the principles of traditional algorithms, break through the limitations of the single-driven channel estimation model, and propose a model-data dual-driven channel estimation method based on VAMP. It uses VAMP to perform singular value decomposition on the quasi-orthogonal pilot matrix, expands VAMP into a neural network to learn the prior information of user access, and then introduces a denoising module to perform learnable denoising on the noisy data, aiming to improve the channel estimation accuracy at the satellite receiving end. In addition, the present invention has scalability and can be migrated to channel estimation scenarios such as satellite multi-antenna.

[0089] Embodiment 2 The embodiments of the present invention also provide a channel estimation device for large-scale user asynchronous access. This device is mainly used to execute the channel estimation method for large-scale user asynchronous access provided in the above Embodiment 1. The following is a specific introduction to the channel estimation device for large-scale user asynchronous access provided by the embodiments of the present invention.

[0090] Figure 5 is a functional module diagram of a channel estimation device for large-scale user asynchronous access provided by the embodiments of the present invention. As Figure 5 shown, the device mainly includes: an acquisition module 10, a determination module 20, a first format conversion module 30, a channel estimation module 40, and a second format conversion module 50, where: The acquisition module 10 is used to acquire the received signal at the satellite receiving end, the maximum delay of users accessing the satellite, the pilot matrix of all users, and the delay information matrix; where the pilot matrix is composed of the pilot sequences of all users; the delay information matrix is used to characterize the actual delay of each user accessing the satellite.

[0091] The determination module 20 is used to determine the actual delay spread pilot matrix of all users based on the pilot matrix, the delay information matrix, and the maximum delay.

[0092] The first format conversion module 30 is used to perform format conversion on the received signal at the satellite receiving end and the actual delay spread pilot matrix to obtain a real-form received signal and a real-form actual delay spread pilot matrix.

[0093] A channel estimation module 40 is configured to process a received signal in real number form and an actual delay spread pilot matrix in real number form by using a target channel estimation model, so as to obtain a channel estimation result in real number form. The target channel estimation model is a model-data dual-driven network model based on the vectorized approximate message passing algorithm VAMP. The model-driven network based on VAMP is used to perform channel estimation based on its input data and perform first denoising processing on the channel estimation result, and the data-driven network is used to perform second denoising processing on the first denoising processing result of the model-driven network based on VAMP.

[0094] A second format conversion module 50 is configured to perform format conversion on the channel estimation result in real number form to obtain a channel estimation result in complex number form.

[0095] Aiming at the problem of the accuracy decline of traditional channel estimation methods in the case of large-scale user asynchronous access, an embodiment of the present invention provides a channel estimation device for large-scale user asynchronous access. First, the unlicensed random access satellite technology is proposed to reduce the access delay and signaling overhead. Considering the actual situation of user asynchronous access, the maximum delay of user access to the satellite is set to expand the pilot sequence, so as to construct a satellite-ground uplink access model that conforms to the actual access scenario. Further, the target channel estimation model of the present invention is a model-data dual-driven network model based on VAMP. The model-driven network based on VAMP is used to perform channel estimation and perform first denoising processing on the channel estimation result, and the data-driven network is used to perform second denoising processing on the first denoising processing result. Therefore, this method can achieve high-precision channel estimation at the satellite receiving end, and alleviates the technical problem of low channel estimation accuracy existing in the existing channel estimation methods in the unlicensed asynchronous access scenario.

[0096] Optionally, the determining module 20 is specifically configured to: Based on the pilot matrix and the maximum delay, determine the target extended pilot matrix of all users at all optional delays.

[0097] Based on the target extended pilot matrix and the delay information matrix, determine the actual delay spread pilot matrix.

[0098] Optionally, the device further includes: A construction module is configured to construct a training data set. The training data set includes multiple groups of training data, and each group of training data includes: an actual delay spread pilot matrix sample of all users, an effective channel vector sample, and a received signal sample at the satellite receiving end.

[0099] A third format conversion module is configured to perform format conversion on the training data set to obtain a training data set in real number form.

[0100] A processing module, configured to process the actual delay spread pilot matrix samples and received signal samples in the target group of training data by using an initial channel estimation model, so as to obtain corresponding real-form channel estimation results; wherein, the target group of training data represents any group of training data in the training data set in real form.

[0101] A calculation module, configured to calculate a loss function value based on the effective channel vector samples in the target group of training data and the corresponding real-form channel estimation results.

[0102] A training module, configured to perform iterative training on the initial channel estimation model based on the loss function value until a preset iteration termination condition is reached, so as to obtain a target channel estimation model.

[0103] Optionally, the construction module is specifically configured to: Randomly generate a plurality of actual delay spread pilot matrix samples.

[0104] Simulate the transmission of the plurality of actual delay spread pilot matrix samples in satellite multipath fading channels with different signal-to-noise ratios respectively, so as to obtain corresponding received signal samples of a plurality of satellite receivers.

[0105] Wherein, the effective channel vector sample of each satellite multipath fading channel is expressed as: ; represents the effective channel vector sample, represents the active indication matrix of all users, , represents user 's activity, represents user is active, represents user is inactive, represents the number of users, represents the set of complex-valued channel responses of all users, and the complex-valued channel response of user at time, frequency is expressed as , represents the Rice factor, when, represents the LoS channel response; when, represents the NLoS channel response, represents the number of propagation paths.

[0106] , represents user on the th path complex gain, represents user The Doppler frequency shift on the th path, indicating the user on the th path, indicating the antenna array response, indicating the user on the th path, indicating the azimuth angle of the angle of arrival on the th path, and indicating the elevation angle of the angle of arrival on the

[0107] Optionally, the initial channel estimation model includes: an initial VAMP-based model-driven network and an initial data-driven network; the initial VAMP-based model-driven network includes: a linear estimator, a first decoupler, a non-linear estimator, and a second decoupler.

[0108] Specifically, the processing module is configured to: Use the linear estimator to perform channel estimation processing on the actual delay spread pilot matrix samples and the received signal samples, to obtain a first channel estimation result and a first average variance; the first average variance represents the average of the diagonal values of the covariance of the first channel estimation result.

[0109] Use the first decoupler to process the first channel estimation result and the first average variance, to obtain the mean and average variance for the non-linear estimator to transmit messages.

[0110] The non-linear estimator performs channel estimation processing and a first denoising process on the mean and average variance for it to transmit messages, to obtain a second channel estimation result and a second average variance; the second average variance represents the average of the diagonal values of the covariance of the second channel estimation result.

[0111] The data-driven network performs a second denoising process on the second channel estimation result, to obtain a channel estimation result in real number form.

[0112] Use the second decoupler to process the channel estimation result in real number form, the second average variance, the mean and average variance for the non-linear estimator to transmit messages, to update the mean and average variance for the linear estimator to transmit messages.

[0113] Optionally, the learning parameters of the linear estimator include: the environmental noise covariance, the left singular matrix, the singular value matrix, and the right singular matrix obtained by performing singular value decomposition on the actual delay spread pilot matrix in real number form.

[0114] Optionally, the initial data-driven network includes: a data preprocessing module, a convolution and pooling module, an upsampling and residual connection module, an SE attention block, and a denoising output module connected in sequence.

[0115] The data preprocessing module is used to adjust the input data of the data-driven network to a specified dimension.

[0116] The convolution and pooling module includes: a first sub-module, a second sub-module, and a third sub-module; the first sub-module and the second sub-module have the same structure, and the first sub-module includes: a convolutional layer, an activation layer, a normalization layer, and a pooling layer; the third sub-module includes: a convolutional layer, an activation layer, and a normalization layer.

[0117] The upsampling and residual connection module includes: a fourth sub-module and a fifth sub-module; the fourth sub-module and the fifth sub-module have the same structure, and the fourth sub-module includes: an upsampling layer, a convolutional layer, an activation layer, a normalization layer, and a residual connection layer.

[0118] The residual connection layer in the fourth sub-module is used to perform an addition process on the output result of the normalization layer in the fourth sub-module and the output result of the second sub-module.

[0119] The residual connection layer in the fifth sub-module is used to perform an addition process on the output result of the normalization layer in the fifth sub-module and the output result of the data preprocessing module.

[0120] The SE attention block includes: a global feature compression layer, a first fully connected layer, a second fully connected layer, and a feature extraction layer.

[0121] The feature extraction layer is used to perform a multiplication process on the output result of the second fully connected layer and the output result of the upsampling and residual connection module.

[0122] The denoising output module includes: a convolutional layer, a denoising layer, and a soft threshold processing layer.

[0123] The denoising layer is used to perform a subtraction process on the output result of the data preprocessing module and the output result of the convolutional layer in the denoising output module.

[0124] Embodiment III See Figure 6 , this embodiment of the present invention provides an electronic device, which includes: a processor 60, a memory 61, a bus 62, and a communication interface 63, and the processor 60, the communication interface 63, and the memory 61 are connected through the bus 62; the processor 60 is used to execute an executable module stored in the memory 61, such as a computer program.

[0125] Among them, the memory 61 may include high-speed random access memory (RAM), or may also include non-volatile memory, such as at least one disk memory. The communication connection between this system network element and at least one other network element is realized through at least one communication interface 63 (which can be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. can be used.

[0126] The bus 62 can be an ISA bus, a PCI bus, an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 6 only a bidirectional arrow is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0127] Among them, the memory 61 is used to store programs. After receiving an execution instruction, the processor 60 executes the program. The methods executed by the devices defined by the processes disclosed in any of the foregoing embodiments of the present invention can be applied to or implemented by the processor 60.

[0128] The processor 60 may be an integrated circuit chip with signal processing capabilities. In the implementation process, the steps of the above method can be completed by the integrated logic circuit in the hardware of the processor 60 or the instructions in software form. The above-mentioned processor 60 can be a general-purpose processor, including a central processing unit (CPU for short), a network processor (NP for short), etc.; it can also be a digital signal processor (DSP for short), an application specific integrated circuit (ASIC for short), a field-programmable gate array (FPGA for short), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present invention can be directly embodied as being executed and completed by a hardware decoding processor, or executed and completed by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the art such as random access memory, flash memory, read-only memory, programmable read-only memory, or electrically erasable programmable memory, register, etc. This storage medium is located in the memory 61, and the processor 60 reads the information in the memory 61 and combines its hardware to complete the steps of the above method.

[0129] A computer program product of a channel estimation method and device for asynchronous access of a large number of users provided by an embodiment of the present invention includes a computer-readable storage medium storing non-volatile program code executable by a processor. The instructions included in the program code can be used to execute the method described in the foregoing method embodiments. For specific implementation, reference can be made to the method embodiments and will not be elaborated herein.

[0130] In addition, in each embodiment of the present invention, each functional unit can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0131] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program code.

[0132] It should be noted that similar reference numerals and letters indicate similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0133] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the present invention is usually placed during use. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, the terms "first", "second", "third", etc. are only used for descriptive distinction and cannot be understood as indicating or implying relative importance.

[0134] In addition, terms such as "horizontal", "vertical", "hanging", etc. do not mean that the components are required to be absolutely horizontal or hanging, but can be slightly inclined. For example, "horizontal" only means that its direction is more horizontal relative to "vertical", and does not mean that the structure must be completely horizontal, but can be slightly inclined.

[0135] In the description of the present invention, it should also be noted that unless otherwise clearly specified and defined, the terms "set", "installed", "connected", "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A channel estimation method for asynchronous access of a large number of users, characterized in that Including: Obtaining the received signal of the satellite receiving end, the maximum delay for a user to access the satellite, the pilot matrix of all users, and the delay information matrix; wherein, the pilot matrix is composed of the pilot sequences of all users; the delay information matrix is used to represent the actual delay for each user to access the satellite; Based on the pilot matrix, the delay information matrix, and the maximum delay, determining the actual delay spread pilot matrix of all users; Performing format conversion on the received signal of the satellite receiving end and the actual delay spread pilot matrix to obtain the received signal in real number form and the actual delay spread pilot matrix in real number form; Processing the received signal in real number form and the actual delay spread pilot matrix in real number form by using a target channel estimation model to obtain a channel estimation result in real number form; wherein, the target channel estimation model is a model-data dual-driven network model based on the vectorized approximate message passing algorithm VAMP. The model-driven network based on VAMP is used to perform channel estimation based on its input data and perform a first denoising process on the channel estimation result, and the data-driven network is used to perform a second denoising process on the first denoising result of the model-driven network based on VAMP; Performing format conversion on the channel estimation result in real number form to obtain a channel estimation result in complex number form.

2. The channel estimation method for large-scale user asynchronous access according to claim 1, characterized in that Based on the pilot matrix, the delay information matrix, and the maximum delay, determining the actual delay spread pilot matrix of all users includes: Based on the pilot matrix and the maximum delay, determining the target extended pilot matrix of all users at all optional delays; Based on the target extended pilot matrix and the delay information matrix, determining the actual delay spread pilot matrix.

3. The channel estimation method for large-scale user asynchronous access according to claim 1, characterized in that Also including: Constructing a training data set; wherein, the training data set includes multiple groups of training data, and each group of training data includes: the actual delay spread pilot matrix sample of all users, the effective channel vector sample, and the received signal sample of the satellite receiving end; Performing format conversion on the training data set to obtain a training data set in real number form; Processing the actual delay spread pilot matrix sample and the received signal sample in the target group of training data by using an initial channel estimation model to obtain a corresponding channel estimation result in real number form; wherein, the target group of training data represents any group of training data in the training data set in real number form; Calculating a loss function value based on the effective channel vector sample in the target group of training data and the corresponding channel estimation result in real number form; Performing iterative training on the initial channel estimation model based on the loss function value until a preset iteration termination condition is reached to obtain the target channel estimation model.

4. The channel estimation method for asynchronous access of a large number of users according to claim 3, characterized in that Constructing a training data set includes: Randomly generating multiple actual delay spread pilot matrix samples; Simulating the transmission of the multiple actual delay spread pilot matrix samples in satellite multipath fading channels with different signal-to-noise ratios to obtain corresponding multiple received signal samples of the satellite receiving end; Among them, the effective channel vector samples of each satellite multipath fading channel are expressed as: ; represents the effective channel vector samples, represents the active indication matrix of all users, , represents user 's activity, represents user is active, represents user is inactive, represents the number of users, represents the set of complex-valued channel responses of all users. The complex-valued channel response of user at time and frequency is expressed as , represents the Rice factor, When represents the LoS channel response; When represents the NLoS channel response, represents the number of propagation paths; , represents the complex gain of the user on the th path, represents the Doppler shift of the user on the th path, represents the propagation delay of the user on the th path, represents the antenna array response, represents the azimuth angle of the angle of arrival of the user on the th path, represents the elevation angle of the angle of arrival of the user on the th path.

5. The channel estimation method for large-scale user asynchronous access according to claim 3, characterized in that The initial channel estimation model includes: an initial VAMP-based model-driven network and an initial data-driven network; the initial VAMP-based model-driven network includes: a linear estimator, a first decoupler, a non-linear estimator, and a second decoupler; Processing the actual delay spread pilot matrix samples and received signal samples in the target group training data using the initial channel estimation model includes: Performing channel estimation processing on the actual delay spread pilot matrix samples and the received signal samples using the linear estimator to obtain a first channel estimation result and a first average variance; the first average variance represents the average of the diagonal values of the covariance of the first channel estimation result; Processing the first channel estimation result and the first average variance using the first decoupler to obtain the mean and average variance for the non-linear estimator to transmit messages; The non-linear estimator performs channel estimation processing and first denoising processing on the mean and average variance for it to transmit messages to obtain a second channel estimation result and a second average variance; the second average variance represents the average of the diagonal values of the covariance of the second channel estimation result; The data-driven network performs second denoising processing on the second channel estimation result to obtain the real-form channel estimation result; Processing the real-form channel estimation result, the second average variance, the mean and average variance for the non-linear estimator to transmit messages using the second decoupler to update the mean and average variance for the linear estimator to transmit messages.

6. The channel estimation method for asynchronous access of a large number of users according to claim 5, characterized in that The learning parameters of the linear estimator include: the environmental noise covariance, the left singular matrix obtained by performing singular value decomposition on the real-form actual delay spread pilot matrix, the singular value matrix, and the right singular matrix.

7. The channel estimation method for asynchronous access of a large number of users according to claim 5, characterized in that The initial data-driven network includes: a data preprocessing module, a convolution and pooling module, an upsampling and residual connection module, an SE attention block, and a denoising output module connected in sequence; The data preprocessing module is used to adjust the input data of the data-driven network to a specified dimension; The convolution and pooling module includes: a first sub-module, a second sub-module, and a third sub-module; the structures of the first sub-module and the second sub-module are the same, and the first sub-module includes: a convolutional layer, an activation layer, a normalization layer, and a pooling layer; the third sub-module includes: a convolutional layer, an activation layer, and a normalization layer; The upsampling and residual connection module includes: a fourth sub-module and a fifth sub-module; the structures of the fourth sub-module and the fifth sub-module are the same, and the fourth sub-module includes: an upsampling layer, a convolutional layer, an activation layer, a normalization layer, and a residual connection layer; The residual connection layer in the fourth sub-module is used to perform summation processing on the output result of the normalization layer in the fourth sub-module and the output result of the second sub-module; The residual connection layer in the fifth sub-module is used to perform summation processing on the output result of the normalization layer in the fifth sub-module and the output result of the data preprocessing module; The SE attention block includes: a global feature compression layer, a first fully connected layer, a second fully connected layer, and a feature extraction layer; The feature extraction layer is used to multiply the output result of the second fully connected layer by the output result of the upsampling and residual connection module; The denoising output module includes: a convolutional layer, a denoising layer, and a soft threshold processing layer; The denoising layer is used to subtract the output result of the convolutional layer in the denoising output module from the output result of the data preprocessing module.

8. A channel estimation device for asynchronous access of a large number of users, characterized in that, Comprising: An acquisition module, configured to acquire the received signal of the satellite receiving end, the maximum delay of the user accessing the satellite, the pilot matrices of all users, and the delay information matrix; wherein, the pilot matrix is composed of the pilot sequences of all users; the delay information matrix is used to represent the actual delay of each user accessing the satellite; A determination module, configured to determine the actual delay spread pilot matrices of all users based on the pilot matrix, the delay information matrix, and the maximum delay; A first format conversion module, configured to perform format conversion on the received signal of the satellite receiving end and the actual delay spread pilot matrices to obtain a real-form received signal and a real-form actual delay spread pilot matrix; A channel estimation module, configured to process the real-form received signal and the real-form actual delay spread pilot matrices by using a target channel estimation model to obtain a real-form channel estimation result; wherein, the target channel estimation model is a model-data dual-driven network model based on the vectorized approximate message passing algorithm VAMP. The model-driven network based on VAMP is used to perform channel estimation based on its input data and perform first denoising processing on the channel estimation result, and the data-driven network is used to perform second denoising processing on the first denoising processing result of the model-driven network based on VAMP; A second format conversion module, configured to perform format conversion on the real-form channel estimation result to obtain a complex-form channel estimation result.

9. An electronic device, comprising a memory and a processor, wherein a computer program capable of running on the processor is stored on the memory, characterized in that When the processor executes the computer program, it implements the channel estimation method for large-scale user asynchronous access according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by the processor, it implements the channel estimation method for large-scale user asynchronous access according to any one of claims 1 to 7.

Images