OTFS channel estimation method, system and device and medium

Abstract

The invention discloses an OTFS channel estimation method, system and device and a medium, and relates to the technical field of communication. A regression network model is introduced, after a channel matrix is originally obtained, the channel matrix needs to be input into the regression network model, denoising processing in a model mode is carried out, and the anti-interference performance of the channel matrix on a sending signal of a first signal in the reconstruction processing process is improved; a fitting relation from an original time delay Doppler domain channel estimation matrix to a real channel matrix is constructed, and the estimation precision is improved. According to the method, the input preliminary estimation result is further refined and optimized by using the strong learning and fitting capability of the neural network, and finally a more accurate channel estimation result is output, so that the performance of the whole communication system can be remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to an OTFS channel estimation method, system, device and medium. Background Art

[0002] During the communication process, the transmitter and receiver communicate through the channel, and the Doppler frequency shift method in the orthogonal time-frequency space (OTFS) modulation and demodulation technology is sampled in the form of integers, resulting in a decrease in the accurate representation of the actual fractional Doppler frequency shift during the sampling process, which in turn causes Doppler crosstalk (Inter-Doppler Interference, IDI), thereby changing the signal amplitude at the receiving end, causing the system to incorrectly evaluate the channel parameters and reduce the communication quality.

[0003] Therefore, how to reduce channel interference to improve communication quality is an urgent problem that those skilled in the art need to solve. Summary of the invention

[0004] The object of the present invention is to provide an OTFS channel estimation method, system, device and medium to solve the problem that the conventional accurate representation of fractional Doppler shift is reduced and the system makes an erroneous assessment of channel parameters to reduce the communication quality.

[0005] In order to solve the above technical problems, the present invention provides an OTFS channel estimation method, which is applied to a receiving end and includes:

[0006] receiving a first signal sent by a transmitting end; and performing crosstalk analysis according to the first signal to obtain a delay Doppler domain receiving signal;

[0007] Determine a target path according to the amplitude and threshold value of the delay Doppler domain received signal; and construct a channel matrix according to the path channel state information of the target path;

[0008] The channel matrix is ​​input into a regression network model to obtain a denoised channel matrix, and a transmission signal corresponding to the first signal is reconstructed.

[0009] On the one hand, the training process of the regression network model includes:

[0010] Obtain the initial regression network model, training data in a multipath channel scenario, and the actual channel matrix;

[0011] Determine a first target path according to the amplitude corresponding to the training data in the multipath channel scenario and the threshold value;

[0012] constructing a first channel matrix according to the path channel state information of the first target path;

[0013] Using the first channel matrix as an input signal of the initial regression network model to obtain a first actual channel matrix of the initial regression network model;

[0014] Record the current number of iterations;

[0015] When the difference between the first actual channel matrix and the actual channel matrix does not meet the preset conditions, or the current number of iterations does not reach the preset number of iterations, the model parameters of the initial regression network model are fine-tuned to obtain the fine-tuned initial regression network model; and the method returns to the step of using the first channel matrix as the input signal of the initial regression network model until the difference between the first actual channel matrix and the actual channel matrix meets the preset conditions, or the current number of iterations reaches the preset number of iterations.

[0016] On the other hand, the channel matrix is ​​input into the regression network model to obtain a denoised signal matrix, including:

[0017] Performing feature extraction processing on the data of the channel matrix to obtain first feature data;

[0018] The first feature data is subjected to feature learning residual processing to obtain second feature data; wherein the feature learning residual processing of the regression network model is completed in a feature learning residual module, and the feature learning residual module includes a plurality of enhanced attention modules, and there is a cascade relationship between the plurality of enhanced attention modules, and the enhanced attention module includes a plurality of convolution kernels and a merged convolution layer;

[0019] The second characteristic data is reconstructed to obtain a denoised signal matrix.

[0020] On the other hand, the sizes of multiple convolution kernels in the enhanced attention module are different.

[0021] On the other hand, the difference between the first actual channel matrix and the actual channel matrix does not meet a preset condition, including:

[0022] Obtaining the channel capacity and rank corresponding to the eigenvalues ​​of the first actual channel matrix and the actual channel matrix respectively;

[0023] If the difference between the respective channel capacities is greater than or equal to the first threshold, the ranks corresponding to the respective eigenvalues ​​are compared;

[0024] If the ranks corresponding to the respective eigenvalues ​​are greater than or equal to the second threshold, it is determined that the difference between the first actual channel matrix and the actual channel matrix does not meet a preset condition.

[0025] On the other hand, the difference between the first actual channel matrix and the actual channel matrix does not meet a preset condition, including:

[0026] Obtaining the correlations and estimation errors corresponding to the first actual channel matrix and the actual channel matrix respectively;

[0027] If the difference between the respective corresponding correlations is not within the first preset range, then comparing the respective corresponding estimation errors;

[0028] If the difference between the corresponding estimation errors is not within the second preset range, it is determined that the difference between the first actual channel matrix and the actual channel matrix does not meet a preset condition.

[0029] On the other hand, performing crosstalk analysis according to the first signal to obtain a delay Doppler domain received signal includes:

[0030] Performing SFFT transformation on the first signal to obtain a first delay Doppler domain signal;

[0031] Acquire a first target row and a second target row of the first delay-Doppler domain signal;

[0032] The first delay-Doppler domain signal is intercepted according to the first target row and the second target row, and the signal from the first target row to the second target row is used as the delay-Doppler domain received signal.

[0033] In order to solve the above technical problems, the present invention also provides an OTFS channel estimation system, including a receiving end and a transmitting end:

[0034] Controlling the sending end to send a first signal;

[0035] The receiving end is controlled to perform crosstalk analysis according to the first signal to obtain a delay Doppler domain received signal; a target path is determined according to the amplitude and threshold value of the delay Doppler domain received signal; a channel matrix is ​​constructed according to the path channel state information of the target path; and the channel matrix is ​​denoised to obtain a denoised channel matrix to reconstruct the transmission signal corresponding to the first signal.

[0036] In order to solve the above technical problems, the present invention also provides an OTFS channel estimation device, comprising:

[0037] Memory for storing computer programs;

[0038] A processor is used to implement the steps of the OTFS channel estimation method when executing the computer program.

[0039] In order to solve the above technical problem, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the OTFS channel estimation method are implemented.

[0040] An OTFS channel estimation method provided by the present invention is applied to a receiving end, receiving a first signal sent by a transmitting end; performing crosstalk analysis based on the first signal to obtain a delay Doppler domain receiving signal; determining a target path based on the amplitude and threshold value of the delay Doppler domain receiving signal; constructing a channel matrix based on the path channel state information of the target path; inputting the channel matrix into a regression network model to obtain a denoised channel matrix, and reconstructing the transmission signal corresponding to the first signal. The present invention introduces a regression network model. After the original channel matrix is ​​obtained, the channel matrix needs to be input into the regression network model to perform model-based denoising processing, thereby improving the anti-interference of the channel matrix to the transmission signal of the first signal during the reconstruction process, so as to construct a fitting relationship between the original delay Doppler domain channel estimation matrix and the real channel matrix, thereby improving the estimation accuracy. By utilizing the powerful learning and fitting capabilities of the neural network, the input preliminary estimation results are further refined and optimized, and finally a more accurate channel estimation result is output, which can significantly improve the performance of the entire communication system.

[0041] In addition, the present invention also provides an OTFS channel estimation system, device and medium, which have the same beneficial effects as the above-mentioned OTFS channel estimation method. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A schematic diagram of transmitting and receiving signals of an integer Doppler system provided in an embodiment of the present invention;

[0044] Figure 2 A schematic diagram of transmitting and receiving signals of a fractional Doppler system provided in an embodiment of the present invention;

[0045] Figure 3 A flow chart of an OTFS channel estimation method provided by an embodiment of the present invention;

[0046] Figure 4 A schematic diagram of OTFS data frame changes between a transmitting end and a receiving end provided in an embodiment of the present invention;

[0047] Figure 5 A schematic diagram of a regression network model provided by an embodiment of the present invention;

[0048] Figure 6 A schematic diagram of the principle of a regression network model provided by an embodiment of the present invention;

[0049] Figure 7 A schematic diagram of a pooling layer operation process provided by an embodiment of the present invention;

[0050] Figure 8 A schematic diagram of another pooling layer operation process provided by an embodiment of the present invention;

[0051] Fig. 9 A schematic diagram of a normalized mean square error curve of RIDChanNet under a five-path channel provided by an embodiment of the present invention;

[0052] Fig.10 A schematic diagram of a normalized mean square error curve of RIDChanNet under an eight-path channel provided by an embodiment of the present invention;

[0053] Fig.11 A schematic diagram of a normalized mean square error curve of RIDChanNet under a ten-path channel provided by an embodiment of the present invention;

[0054] Fig.12 A schematic diagram of a bit error rate curve of RIDChanNet under a five-path channel provided by an embodiment of the present invention;

[0055] Fig.13 A schematic diagram of a bit error rate curve of RIDChanNet in an eight-path channel provided by an embodiment of the present invention;

[0056] Fig.14 A schematic diagram of a bit error rate curve of RIDChanNet in a ten-path channel provided by an embodiment of the present invention;

[0057] Fig.15 A structural diagram of an OTFS channel estimation device provided by an embodiment of the present invention;

[0058] Fig.16 A structural diagram of an OTFS channel estimation device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] The core of the present invention is to provide an OTFS channel estimation method, system, device and medium to solve the problem that the conventional accurate representation of fractional Doppler frequency shift is reduced and the system makes an erroneous evaluation of channel parameters to reduce the communication quality.

[0061] In order to enable those skilled in the art to better understand the solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0062] In the OTFS system, Doppler shift is used to describe the signal frequency change caused by the relative motion between the transmitter and the receiver. According to the numerical characteristics of Doppler shift, the OTFS system can be divided into integer Doppler system and fractional Doppler system, and the ratio of the Doppler shift of each path to the sampling resolution is defined as: ;in, For the The Doppler shift of the path, is the sampling resolution; if P paths correspond to If both are integers, then the system is an integer Doppler system; otherwise, it is a fractional Doppler system. Figure 1 A schematic diagram of a signal receiving and transmitting integer Doppler system provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown in the figure, the sampling point mapping follows the rule of one-to-one mapping, the representation of the signal in the delay-Doppler domain is relatively simple, the complexity of the channel estimation and equalization technology is low, and the commonly used estimation methods include the threshold-based channel estimation algorithm and the compressed sensing-based channel estimation algorithm. In addition, the integer Doppler system can usually better utilize the sparsity characteristics of the OTFS technology to improve the transmission efficiency and performance of the system. In the fractional system, Figure 2 A schematic diagram of a signal transmission and reception of a fractional Doppler system provided by an embodiment of the present invention is shown in FIG. Figure 2 As shown, the Doppler shift is sampled in the form of an integer in the delay-Doppler domain, which results in the sampling unit being unable to accurately represent the actual fractional Doppler shift, thereby causing IDI interference. IDI interference refers to the crosstalk received by the receiving unit from the sampling units corresponding to the same delay but different Doppler parameters within the OTFS data frame. This interference changes the signal amplitude of the receiving unit, causing the system to mis-estimate the channel parameters and reduce the communication quality. The OTFS channel estimation method provided by the present invention can solve the above technical problems.

[0063] Figure 3 A flow chart of an OTFS channel estimation method provided by an embodiment of the present invention, such as Figure 3 As shown, the method is applied to the receiving end, comprising:

[0064] S11: receiving a first signal sent by a transmitting end; and performing crosstalk analysis according to the first signal to obtain a delay Doppler domain receiving signal;

[0065] S12: determining a target path according to the amplitude and threshold value of the received signal in the delay Doppler domain; and constructing a channel matrix according to the path channel state information of the target path;

[0066] S13: Input the channel matrix into the regression network model to obtain the denoised channel matrix, and reconstruct the transmission signal corresponding to the first signal.

[0067] Figure 4 A schematic diagram of OTFS data frame changes between a transmitting end and a receiving end provided by an embodiment of the present invention, such as Figure 4 As shown in the figure, during the OTFS signal demodulation process at the transmitter, the complete delay Doppler domain signal After the Inverse Symplectic Finite Fourier Transform (ISFFT), the time-frequency domain (TF) signal is obtained. , where ISFFT is defined as:

[0068] ;

[0069] in, , are the number of sampling points in the Doppler domain and the delay domain, respectively. , are the indexes of Doppler domain and delay domain respectively; represents the index of the frequency domain dimension, The index of the time dimension.

[0070] Will Perform Heisenberg transform to obtain parallel time domain signals, where The signal transmission waveform is:

[0071] ;

[0072] in, is the time variable, is the sampling period of the OTFS system, is the subcarrier spacing of the system.

[0073] right After serial-to-parallel conversion, it is sent to the channel as a transmission signal. Gaussian white noise is superimposed on the signal during transmission. Therefore, the signal at the receiving end can be expressed as:

[0074] ;

[0075] in, is the Gaussian white noise superimposed during the transmission process. Assume that the channel has paths, and the channel fading gain, delay and Doppler frequency shift corresponding to each path are , , ; The index of the path.

[0076] Delay-Doppler Domain Channel (DD) under ideal conditions It can be expressed as follows:

[0077] ;

[0078] in, For the The channel fading gain of the path, Represents the Dirac delta function, which is an impulse function.

[0079] It should be noted that the signal is demodulated at the receiving end to convert the serial received signal into a parallel time domain signal. , and then after sampling and Wigner transform, the time-frequency domain received symbol is obtained, that is, the first signal:

[0080] ;

[0081] in, is the time-frequency domain symbol of the received pulse waveform, Upper right corner This represents the conjugate complex number, For the receiving end; Represents the variable in the convolution process and has no practical meaning.

[0082] In some embodiments, performing crosstalk analysis according to the first signal to obtain a delay-Doppler domain received signal includes:

[0083] Performing SFFT transformation on the first signal to obtain a first delay-Doppler domain signal;

[0084] Acquire a first target row and a second target row of a first delay-Doppler domain signal;

[0085] The first delay Doppler domain signal is intercepted according to the first target line and the second target line, and the signal from the first target line to the second target line is used as the delay Doppler domain received signal.

[0086] Specifically, the time-frequency domain symbols are transformed to the delay-Doppler domain through sparse fast Fourier transform (SFFT) to obtain the DD domain received symbols:

[0087] .

[0088] Specifically, the first signal at the transmitting end is mapped into data symbols through quadrature phase shift keying (QPSK) and arranged in Figure 4 Mid-DD domain signal The remaining position is called the guard band, which is distributed at the bottom of the data frame and is used to isolate the interference of data on pilot symbols and channel estimation. The transmitted symbols in this area are The values ​​are:

[0089] ;

[0090] in, represents the pilot symbol, and is the coordinate parameter corresponding to the position of the pilot point. It is worth noting that a guard interval needs to be set on both sides of the pilot symbol in the delay domain. The guard interval between the pilot and the data can prevent the pilot from aliasing with the data symbols affected by the delay; the guard interval on the other side can show the delay of the channel, which can be used for channel estimation.

[0091] In the threshold-based channel estimation process, the receiver intercepts of Row (first target row) to The row (second target row) symbol is used for channel estimation, and the received symbol in this area (delay-Doppler domain received signal) can be expressed as:

[0092] ;

[0093] in, Indicates that there is a delay coefficient , Doppler coefficient The transmission path, , Respectively represent the coordinate information in Doppler domain and delay domain; , are the indexes of Doppler domain and delay domain respectively; is the additive white Gaussian noise in the DD domain, if If the value is 0, it means that the path does not exist and the data is received. Contains only white noise.

[0094] The process of determining the delay-Doppler domain received signal provided in this embodiment is based on the data in the guard band to prevent the pilot from overlapping with the data symbols affected by the delay; the guard interval on the other side can show the delay of the channel and is then used for channel estimation.

[0095] The delay-Doppler domain equivalent channel matrix can be expressed as:

[0096] ;

[0097] in, Indicates The initial phase of each path, , It represents the process variable in the signal convolution operation and has no practical meaning; , are the indexes of Doppler domain and delay domain respectively; represents the equivalent channel Doppler response; the phase offset term and the equivalent channel Doppler response are given by the following formulas, is the length of the cyclic prefix.

[0098] ;

[0099] ;

[0100] Among them, The Doppler domain coordinates corresponding to the paths are , and Represents the integer part and the decimal part respectively.

[0101] In an integer Doppler system, the Doppler domain coordinates corresponding to all paths can be expressed as ,at this time ; In the fractional Doppler system, , at this time regardless of What value to take? Therefore, when the system changes from an integer Doppler system to a fractional Doppler system, many of the previous zeros are The appearance of is transformed into a non-zero point, forming a new fractional Doppler equivalent channel matrix . It consists of two parts, one of which is the original matrix under the integer system , and the other part is interference Therefore, the delay-Doppler domain received signal in the fractional Doppler system can be re-expressed as:

[0102] ;

[0103] in, is the exact value of the received symbol, Doppler crosstalk is the occurrence of pilot energy dissipating into each Doppler unit under a certain delay interval, changing the amplitude of the received symbol. Therefore, the occurrence of Doppler crosstalk will have an adverse effect on subsequent channel estimation and signal detection.

[0104] In order to distinguish the sampling points containing only noise from the sampling points corresponding to the path, a threshold is used to filter the data in the guard band. ,in is the power spectral density of the noise. If the amplitude of the received signal is , then it is considered that there is a path (i.e., the target path) corresponding to the Doppler-delay parameter of the sampling point, then is 1; otherwise, it is considered that there is no corresponding path. is 0, at this time Only the noise amplitude is included. According to the number, position and amplitude of the selected sampling points, the number of multipaths and the delay corresponding to each path can be obtained. , Doppler shift and channel fading gain Such path channel state information can then be used to construct a complete DD domain channel matrix.

[0105] In addition to the Gaussian white noise component, Doppler crosstalk is also introduced into the received signal. This interference term greatly increases the amplitude of the received signal, making it difficult for the receiver to identify the true transmission path through the threshold, affecting the estimation accuracy of the delay Doppler domain channel.

[0106] The conventional technical means is to use the channel matrix in step S12 to reconstruct the transmission signal corresponding to the first signal, but the corresponding signal quality is low. The present invention inputs the channel matrix into the regression network model on the basis of determining the channel matrix. The regression network model can be a conventional network model or a network model improved based on the current channel estimation. It is not limited here and can be set according to the actual situation. The denoised channel matrix is ​​output based on the regression network model to reconstruct the transmission signal corresponding to the first signal.

[0107] An OTFS channel estimation method provided by an embodiment of the present invention is applied to a receiving end, receiving a first signal sent by a transmitting end; performing crosstalk analysis based on the first signal to obtain a delay Doppler domain receiving signal; determining a target path based on the amplitude and threshold value of the delay Doppler domain receiving signal; constructing a channel matrix based on the path channel state information of the target path; inputting the channel matrix into a regression network model to obtain a denoised channel matrix, and reconstructing the transmission signal corresponding to the first signal. The present invention introduces a regression network model. After the original channel matrix is ​​obtained, the channel matrix needs to be input into the regression network model for model-based denoising to improve the anti-interference of the channel matrix to the transmission signal of the first signal during the reconstruction process, so as to construct a fitting relationship between the original delay Doppler domain channel estimation matrix and the real channel matrix, thereby improving the estimation accuracy. By utilizing the powerful learning and fitting capabilities of the neural network, the input preliminary estimation results are further refined and optimized, and finally a more accurate channel estimation result is output, which can significantly improve the performance of the entire communication system.

[0108] In some embodiments, the training process of the regression network model includes:

[0109] Obtain the initial regression network model, training data in a multipath channel scenario, and the actual channel matrix;

[0110] Determine a first target path according to an amplitude and a threshold value corresponding to the training data in a multipath channel scenario;

[0111] constructing a first channel matrix according to the path channel state information of the first target path;

[0112] Using the first channel matrix as an input signal of the initial regression network model to obtain a first actual channel matrix of the initial regression network model;

[0113] Record the current number of iterations;

[0114] When the difference between the first actual channel matrix and the actual channel matrix does not meet the preset conditions, or the current number of iterations does not reach the preset number of iterations, the model parameters of the initial regression network model are fine-tuned to obtain the fine-tuned initial regression network model; and the method returns to the step of using the first channel matrix as the input signal of the initial regression network model until the difference between the first actual channel matrix and the actual channel matrix meets the preset conditions, or the current number of iterations reaches the preset number of iterations.

[0115] Figure 5 A schematic diagram of a regression network model provided by an embodiment of the present invention, such as Figure 5As shown in the figure, in order to effectively reduce the negative impact of Doppler crosstalk and complex Gaussian white noise on channel estimation results and improve the performance and accuracy of the communication system, a deep learning network architecture with excellent performance in the field of image denoising is introduced. The network has a built-in image self-attention mechanism, which can enhance the model's ability to capture complex channel characteristics and accurately construct the fitting relationship between the original delay Doppler domain channel estimation matrix and the real channel matrix, thereby improving the accuracy of the estimation and laying a solid foundation for the performance optimization and practical application of the OTFS communication system.

[0116] The results of the traditional threshold estimation algorithm are used as the input of the deep neural network model. The powerful learning and fitting capabilities of the neural network are used to further refine and optimize the preliminary estimation results of the input, and finally output more accurate channel estimation results, which can significantly improve the performance of the entire communication system. The originally complex delay-Doppler domain channel response matrix fitting problem is transformed into a dual-channel image denoising problem, and the virtual and real parts of the channel matrix are regarded as two channels of the network input, which greatly facilitates the implementation of subsequent processing steps. Figure 5 The model is divided into an offline phase and an online phase. The tasks of the two phases are network training and real-time channel estimation. With its unique network structure and powerful learning ability, the network can efficiently learn the nonlinear mapping relationship between the channel estimation results and the actual channel response. In particular, the introduction of the self-attention module enables the network to focus more accurately on the key feature information in the channel response, while resisting Doppler crosstalk and Gaussian white noise interference, retaining and enhancing the effective components of the channel response.

[0117] In the offline stage, the training data and the known actual channel matrix in the multipath channel scenario are obtained, and the first target path is determined according to the amplitude and threshold value corresponding to the training data in the multipath channel scenario; the first channel matrix is ​​constructed according to the path channel state information of the first target path. The implementation method of constructing the first channel matrix here is the same as the above embodiment, and the above embodiment can be referred to. The first channel matrix is ​​input into the initial regression network model to output the first actual channel matrix, and the first actual channel matrix is ​​compared with the actual channel matrix. When the difference between the first actual channel matrix and the actual channel matrix does not meet the preset conditions, or the current number of iterations does not reach the preset number of iterations, the model parameters of the initial regression network model are fine-tuned to obtain the fine-tuned initial regression network model. The fine-tuning here is based on the fine-tuning of the current model parameters, which can be set according to the actual situation and is not limited here. The fine-tuned initial regression network model continues to train until the difference between the first actual channel matrix and the actual channel matrix meets the preset conditions, or the current number of iterations reaches the preset number of iterations, and the training is terminated to obtain the final regression network model.

[0118] The training process of the regression network model provided in this embodiment can enhance the model's ability to capture complex channel characteristics, and construct a fitting relationship from the original delay-Doppler domain channel estimation matrix to the real channel matrix with high precision, thereby improving the estimation accuracy.

[0119] In some embodiments, the channel matrix is ​​input into a regression network model to obtain a denoised signal matrix, including:

[0120] Performing feature extraction processing on the data of the channel matrix to obtain first feature data;

[0121] Performing feature learning residual processing on the first feature data to obtain second feature data; wherein the feature learning residual processing of the regression network model is completed in a feature learning residual module, and the feature learning residual module includes a plurality of enhanced attention modules, and there is a cascade relationship between the plurality of enhanced attention modules, and the enhanced attention module includes a plurality of convolution kernels and a merged convolution layer;

[0122] The second characteristic data is reconstructed to obtain a denoised signal matrix.

[0123] Figure 6 A schematic diagram of a regression network model provided by an embodiment of the present invention is shown in FIG. Figure 6 As shown in the figure, it mainly includes the following three modules: feature extraction module, feature learning residual module and reconstruction module. The three modules are introduced below.

[0124] 1. Feature extraction module:

[0125] The feature extraction module consists of a convolution layer and an activation layer, which is used to extract initial features from the input noisy image and can capture the basic texture and edge information of the image. In the feature extraction module of RIDChanNet, the convolution layer (Conv) extracts initial features from the network input. :

[0126] ;

[0127] In the formula, represents the convolution operation, Input parameters entered for the network.

[0128] 2. Feature learning residual module:

[0129] The feature learning residual module is the core part of RIDChanNet (regression network model), which consists of multiple Enhancement Attention Modules (EAMs) cascaded (in Figure 6The left figure of the figure is represented by enhanced attention module 1', enhanced attention module 2' and enhanced attention module 3'. Each EAM module learns the difference between the disturbed image and the original image through a specific structural design. In the feature extraction module, the network converts the initial feature Passed to the feature learning residual module, the calculation process of this module is described as :

[0130] .

[0131] The traditional EAM module is mainly composed of a feature extraction submodule and a feature attention submodule. The feature extraction submodule extracts and learns image features through two dilated convolutions and a merged convolution layer. The dilated convolution layer skips the convolution operation on some pixels by adding holes in the convolution kernel, thereby expanding the receptive field without increasing the number of parameters and the amount of calculation, which helps to capture a wider range of image context information. In order to avoid the overfitting problem caused by the complexity of the model and reduce the training and inference costs of the model, the model changes the two dilated convolution layers in the feature extraction submodule to convolution kernels. This change enables the network to better adapt to the arrangement relationship of Doppler crosstalk on the channel matrix and enhance the network's ability to suppress Doppler crosstalk; at the same time, there is no need to deal with the complex hole filling and receptive field adjustment problems in the dilated convolution, which reduces the difficulty of network implementation.

[0132] In some embodiments, the sizes of multiple convolution kernels within the enhanced attention module are different.

[0133] It should be noted that the size of the modified convolution kernel in this embodiment is different, so as to increase the arrangement range on the channel matrix and enhance the network's ability to suppress Doppler crosstalk.

[0134] The feature attention submodule first uses global average pooling to transform the feature map from Zoom out to , to capture global context information. Then, the feature attention map is generated using two convolutional layers and Sigmoid activation function through an adaptive threshold control mechanism. The first convolutional layer is used to reduce the number of channels, and the second convolutional layer restores the original number of channels. Finally, the feature weight of each channel is obtained through the Sigmoid activation function. This module dynamically adjusts the feature weight of each channel through the feature attention mechanism, so that the network pays more attention to the features that are beneficial to the denoising task, while suppressing useless or harmful features, which helps to improve the network's interference removal performance and feature expression ability.

[0135] ;

[0136] In the formula, is the output of the global pooling layer, is the pooling layer, , Represents the height and width of the feature map; , Represents the index of the pixel point of the feature image; in the formula describing the network, the feature map and eigenvalue are extracted from the network input matrix. is the position in the feature map In addition, in order to further extract useful features in the image, RIDChanNet introduces an adaptive threshold control mechanism to capture channel dependencies from the descriptors retrieved from the global average pool, learn the nonlinear synergistic effects and mutual exclusive relationships between channels, and use soft shrinkage and Sigmoid functions to implement the gating mechanism. and They are soft contraction operator and Sigmoid operator respectively, where: In the OTFS signal processing process of the above embodiment, the variable represents an impulse function, which in the network means a soft contraction operator, and the gating mechanism can be expressed as:

[0137] ;

[0138] In the formula, and They are channel reduction and channel upsampling operators respectively. Global pooling layer The output of is used as input to the downsampling convolution layer for feature extraction. In order to distinguish channel features, the global pooling layer is activated by a soft shrinkage function. It is then input to the upsampling convolution layer for Sigmoid activation. In order to calculate statistics, the output of the Sigmoid activation function is composed of the input of the channel features. Adaptively rescaled to:

[0139] ;

[0140] EAM The operation of each module can be expressed as:

[0141] ;

[0142] In the formula, It is The output of the two submodules in the EAM is equivalent to Considering the existence of the residual structure, the output of each EAM module is At the same time, the output of the previous EAM module will be used as the input of the next EAM, and multiple EAM modules form a cascade relationship.

[0143] The pooling layer is an important part of the EAM module. Its function is to downsample the input data. The specific operation is to slide a pooling window on the input feature map, and according to the pooling type, the statistical value of the data in the calculation window is used as the output. Figure 7 A schematic diagram of a pooling layer operation process provided by an embodiment of the present invention, Figure 8 Another schematic diagram of the pooling layer operation process provided by an embodiment of the present invention is as follows: Figure 7 , 8 As shown in the figure, the pooling layer reduces the size of the feature map by downsampling, thereby reducing the dimension of the data and reducing redundant information, so that the model can pay more attention to important features, while reducing the input size of the next layer, thereby reducing the amount of calculation and the number of parameters. The pooling operation can also smooth the feature map to reduce the interference of noise, making the model more robust to small changes in the input data. In addition, the pooling layer downsamples so that each output pixel corresponds to a larger area of ​​the input image, expanding the receptive field of the model and helping the model capture features in a larger range.

[0144] The area corresponding to the zero element is marked as the low-frequency area, and the non-zero element corresponds to the high-frequency area. In addition, in the fitting process of the delay-Doppler domain channel matrix, the Doppler crosstalk in a specific scenario is represented as local information. The traditional pooling layer mainly uses local information rather than global information, which may cause interference from local information when estimating the channel matrix. In order to reduce the interference effect of local information, a global average pooling strategy is adopted in the pooling layer. By retaining global features, the robustness of the network system can be improved.

[0145] The EAM module also introduces two activation functions, ReLU and Sigmoid, where the expression of the Sigmoid function is:

[0146] .

[0147] The Sigmoid function converts the input value to the range of (0, 1), and the result can be interpreted as probability. Represents a variable node in the network, which is suitable for model construction with predicted probability as output. In addition, the gradient of the Sigmoid function on most domains is smooth, which can effectively avoid sudden changes in output values ​​and ensure the stability of the training process. Furthermore, the derivative expression of the Sigmoid function is concise and clear, which greatly facilitates the calculation steps in the back propagation algorithm.

[0148] The ReLU function only needs to perform a simple threshold judgment: if the input value exceeds 0, it is retained, otherwise it is set to 0, which significantly reduces the amount of calculation and improves the computing speed of the neural network. The expression of ReLU is: .

[0149] In the positive range, the derivative of the ReLU function remains constant at 1, and the gradient will not weaken due to the accumulation of network layers, which helps to alleviate the gradient vanishing phenomenon in deep neural networks. ReLU enables neural networks to simulate nonlinear mapping relationships, which helps to enhance the network's fitting ability.

[0150] 3. Reconstruction module:

[0151] The reconstruction module consists of a convolutional layer and a long skip connection operation to transform the learned residual features back to the image space and add them to the input image to generate the denoised image.

[0152] The features output by the feature learning residual module are input into the convolutional layer of the reconstruction module to obtain: However, simply cascading residual modules cannot achieve better performance. Therefore, RIDChanNet converts the initial features Superimposed on the output of the convolutional layer of the reconstruction module, we get the final network output: In the formula, represents the convolution operation, and the stacking operation is called Long Skip Connection (LSC), which can simplify the information flow between groups; is the output of the convolutional layer of the reconstruction module. Finally, the network outputs the same number of feature maps as the network input, completing a forward propagation.

[0153] The structure of the regression network model provided in this embodiment focuses on the learning and processing of interference features, so that the network can more effectively separate and suppress the interference components in the channel impulse response matrix, thereby achieving higher quality restoration while retaining the details and structural information of the channel matrix, which not only improves the processing efficiency, but also greatly enhances the robustness of the model.

[0154] In some embodiments, the difference between the first actual channel matrix and the actual channel matrix does not meet a preset condition, including:

[0155] Obtaining the channel capacity and rank corresponding to the eigenvalues ​​of the first actual channel matrix and the actual channel matrix respectively;

[0156] If the difference between the respective channel capacities is greater than or equal to the first threshold, the ranks corresponding to the respective eigenvalues ​​are compared;

[0157] If the ranks corresponding to the respective eigenvalues ​​are greater than or equal to the second threshold, it is determined that the difference between the first actual channel matrix and the actual channel matrix does not meet a preset condition.

[0158] Specifically, channel capacity is an important indicator to measure the transmission capacity of a channel, and they are compared by calculating the capacity of each channel. By comparing the channel capacities corresponding to different channel matrices, the difference in their transmission efficiency can be evaluated. If the difference between the respective channel capacities is greater than or equal to the first threshold, the ranks corresponding to the respective eigenvalues ​​are compared, and the eigenvalues ​​of the channel matrix can reflect the rank and diversity of the channel. The higher the rank of the channel matrix, the more independent paths the channel has and the greater the capacity of the channel. By comparing the eigenvalues ​​of different channel matrices, their differences in spatial diversity can be understood. If the ranks corresponding to the respective eigenvalues ​​are greater than or equal to the second threshold, it is determined that the difference between the first actual channel matrix and the actual channel matrix does not meet the preset conditions.

[0159] The method provided in this embodiment for determining the difference between the first actual channel matrix and the actual channel matrix based on the rank corresponding to the channel capacity and the eigenvalue does not meet the preset condition, thereby improving the accuracy of the comparison relationship between different channel matrices.

[0160] In some other embodiments, the difference between the first actual channel matrix and the actual channel matrix does not meet a preset condition, including:

[0161] Obtaining the correlations and estimation errors corresponding to the first actual channel matrix and the actual channel matrix respectively;

[0162] If the difference between the respective corresponding correlations is not within the first preset range, then comparing the respective corresponding estimation errors;

[0163] If the difference between the corresponding estimation errors is not within the second preset range, it is determined that the difference between the first actual channel matrix and the actual channel matrix does not meet the preset condition.

[0164] Specifically, the correlation of the channel matrix can affect the transmission quality of the signal. A channel matrix with lower correlation can provide better signal demodulation performance. By analyzing the correlation of the channel matrix, the performance of different channel matrices can be compared. If the difference between the corresponding correlations is not within the first preset range, the corresponding estimation errors are compared. The channel estimation error is an important indicator for measuring the performance of the channel estimation algorithm. By comparing the estimation errors corresponding to different channel matrices, the performance of the channel estimation algorithm under different channel conditions can be evaluated. For example, the minimum mean square error (MMSE) algorithm takes into account the influence of noise, and its estimation error can be compared with the least squares (LS) algorithm to evaluate the performance difference under different channel conditions. If the difference between the corresponding estimation errors is not within the second preset range, it is determined that the difference between the first actual channel matrix and the actual channel matrix does not meet the preset conditions.

[0165] The method provided in this embodiment for determining the difference between the first actual channel matrix and the actual channel matrix based on the correlation and the corresponding estimation errors does not satisfy the preset condition, thereby improving the accuracy of the comparison relationship between different channel matrices.

[0166] As an important indicator to measure signal quality, the physical meaning of signal-to-noise ratio is the ratio of signal power to noise power. In a wireless communication environment with multipath propagation and mobility, changes in signal-to-noise ratio will directly affect the estimation accuracy of the channel matrix. This measure can help the model better understand and simulate the law of channel matrix changes with signal-to-noise ratio, thereby improving the accuracy and robustness of channel estimation.

[0167] In a wireless communication environment with multipath propagation and mobility, the number of multipaths and the mobile speed are important factors that affect the change of the channel matrix. In order to simplify the complexity of the problem and enhance the model's ability to reconstruct a certain type of mathematical relationship, the two parameters of the number of multipaths and the maximum mobile speed are fixed, and training data is generated in a scenario where the number of multipaths is 5 and the maximum mobile speed is 1000km / h. This can make the training data more targeted, help the model learn the optimal solution for fitting in this state, deepen the model's understanding of the multipath fast time-varying channel scenario, and then show transferability in the same type of problem.

[0168] In terms of quantity, 5,000 sets of estimated channel matrices and true channel matrices were collected as training output and input respectively, and the training set and validation set were divided in a ratio of 9:1. These data were used to train the channel estimation model, making the communication system more adaptive. When the channel environment changes, the system can quickly and accurately adjust the channel estimation parameters to ensure the stability and reliability of communication. This is of great significance for improving the overall performance of the communication system and user experience. In order to prevent unreasonable training data from affecting the model, the L1 loss function was selected when training the channel estimation network based on RIDChanNet. This loss function has a linear penalty on errors, which reduces the impact of large errors, reduces the difficulty of training, and is conducive to enhancing the robustness of the model.

[0169] In order to compare the performance of channel estimation algorithms based on RIDChanNet with those based on convolutional neural networks (CNN), residual networks (ResNet), and thresholds, and to observe the estimation performance of the four detection algorithms under different signal-to-noise ratios and multipath numbers, Matlab Monte Carlo simulations were performed. Table 1 is a simulation parameter table of OTFS channel estimation based on RIDChanNet. As shown in Table 1, the ratio of data symbol to noise power is defined as: The ratio of pilot power to noise power is defined as: The relationship between the pilot signal-to-noise ratio and the data signal-to-noise ratio can be described as: .

[0170] is the ratio of pilot power to data symbol power. Since the power of pilot symbols is much higher than that of data symbols, Also much higher than .

[0171] Table 1

[0172]

[0173] In order to verify the estimation accuracy of the RIDChanNet estimation scheme and evaluate the performance of RIDChanNet in practical applications, the normalized mean square error is used as a measure of channel estimation accuracy. At the same time, three comparison schemes are selected, namely the threshold-based channel estimation algorithm given in the above embodiment, the existing residual neural network-based estimation algorithm, and the existing CNN-based channel estimation algorithm, which correspond to "Conventional", "ResNet" and "SR" in the legend respectively, and the legend corresponding to the proposed algorithm is "RID".

[0174] Fig. 9 A schematic diagram of a normalized mean square error curve of RIDChanNet under a five-path channel provided by an embodiment of the present invention is shown in FIG. Fig. 9 As shown in the figure, in the five-path fast time-varying channel, thanks to the powerful feature extraction and data fitting capabilities of RIDChanNet, the proposed channel estimation scheme shows the performance of several other optimization schemes. This advantage is mainly due to the powerful feature extraction and noise suppression capabilities of the EAM module. By learning and optimizing the network parameters, the proposed scheme can more accurately fit the real DD domain channel characteristics, greatly improve the channel estimation accuracy, and significantly reduce the communication bit error rate.

[0175] Compared with other estimation schemes, the traditional channel estimation scheme performs the worst because it does not process IDI interference and noise, which seriously affects the estimation accuracy of the channel response. Compared with the traditional estimation scheme, the CNN-assisted OTFS channel estimation method has improved performance, but the lack of residual connections between the layers of this network makes it prone to overfitting during training, so the improvement is small. The ResNet estimation method shows a relatively superior performance, but its simple structural characteristics make it difficult to extract the edge information of the image, so the estimation accuracy is slightly inferior to RIDChanNet.

[0176] In order to verify the versatility of RIDChanNet, the simulation scenario is extended to eight-path and ten-path fast time-varying channels. Fig.10 A schematic diagram of a normalized mean square error curve of RIDChanNet under an eight-path channel provided by an embodiment of the present invention is shown in FIG. Fig.10 As shown in the figure, the IDI crosstalk of the system becomes more serious under the eight-path channel. The estimation accuracy of all schemes decreases under low signal-to-noise ratio; in a more complex channel environment, the feature extraction ability of the residual network decreases slightly; at this time, the performance of RIDChanNet still shows absolute superiority compared with other schemes. Fig.11 A schematic diagram of a normalized mean square error curve of RIDChanNet under a ten-path channel provided by an embodiment of the present invention is shown in FIG. Fig.11 As shown in the figure, under the ten-path channel, channel features become more difficult to extract. Among them, CNN is prone to overfitting due to the lack of residual connections between network layers, and its generalization ability is poor. Although the estimation accuracy of ResNet and RIDChanNet has decreased, they still maintain good stability. Thanks to RIDChanNet's unique feature extraction mechanism and self-attention module, it has stronger feature extraction and generalization capabilities than ResNet, and can well fit the mathematical expression of IDI interference, so it has superior and stable performance in different channel environments.

[0177] In order to further evaluate the impact of the proposed scheme on the entire communication system, the bit error rate (BER) is used as an indicator to measure the performance of the optimization scheme. Fig.12 A schematic diagram of a bit error rate curve of RIDChanNet under a five-path channel provided by an embodiment of the present invention is shown in FIG. Fig.12 As shown in the figure, under the five-path channel, the traditional estimation scheme is close to the bit error rate floor at high signal-to-noise ratio. The reason is that under high signal-to-noise ratio, IDI interference replaces noise as the main factor affecting communication performance, and this part of the influence cannot be removed by simply improving the signal-to-noise ratio. The neural network can extract the invisible features of the interference part and use the nonlinear mapping relationship between a large number of parameters to fit IDI and noise interference. Fig.12 Among them, RIDChanNet maintains a relatively superior performance with its diverse network architecture and powerful feature extraction capabilities. Fig.13 A schematic diagram of a bit error rate curve of RIDChanNet under an eight-path channel provided by an embodiment of the present invention is shown in FIG. Fig.13 As shown in Figure 2, the difficulty of channel estimation increases further in the eight-path channel. Fig.12 and Fig.13 From the ideal channel matrix bit error rate curve, it can be seen that the signal detection error has also increased. At this time, RIDChanNet still maintains a relatively stable estimation performance, showing the same change trend as the curve under the ideal state, which further proves the stability and accuracy of the method. Fig.14 A schematic diagram of a bit error rate curve of RIDChanNet under a ten-path channel provided by an embodiment of the present invention is shown in FIG. Fig.14 As shown in the figure, when the number of paths further increases to 10, the performance gap between RIDChanNet and the other network structures becomes larger. The reason is that the feature extraction module effectively extracts the deep features of IDI interference, so it can maintain stable estimation performance in channel environments with different numbers of paths. The structures of the other two networks are relatively simple compared to RIDChanNet, and it is difficult to maintain the original performance in more complex environments.

[0178] Furthermore, the present invention also provides an OTFS channel estimation system, comprising a receiving end and a transmitting end:

[0179] Controlling the sending end to send a first signal;

[0180] The receiving end is controlled to perform crosstalk analysis according to the first signal to obtain a delay-Doppler domain receiving signal; a target path is determined according to the amplitude and threshold value of the delay-Doppler domain receiving signal; a channel matrix is ​​constructed according to the path channel state information of the target path; and the channel matrix is ​​denoised to obtain a denoised channel matrix to reconstruct the transmission signal corresponding to the first signal.

[0181] For an introduction to an OTFS channel estimation system provided by the present invention, please refer to the above method embodiment, which will not be described in detail herein. The system has the same beneficial effects as the above OTFS channel estimation method.

[0182] The above describes in detail various embodiments corresponding to the OTFS channel estimation method. On this basis, the present invention also discloses an OTFS channel estimation device corresponding to the above method. Fig.15 A structural diagram of an OTFS channel estimation device provided by an embodiment of the present invention. Fig.15 As shown, the OTFS channel estimation device includes:

[0183] The receiving module 11 is used to receive a first signal sent by a transmitting end; and to perform crosstalk analysis on the first signal to obtain a delay Doppler domain receiving signal;

[0184] A determination module 12 is used to determine the target path according to the amplitude and threshold of the delay Doppler domain received signal; and to construct a channel matrix according to the path channel state information of the target path;

[0185] The reconstruction processing module 13 is used to input the channel matrix into the regression network model to obtain the denoised channel matrix, and to reconstruct the transmission signal corresponding to the first signal.

[0186] Since the embodiments of the device part correspond to the above embodiments, please refer to the description of the embodiments of the method part for the embodiments of the device part, and will not be repeated here.

[0187] For an introduction to an OTFS channel estimation device provided by the present invention, please refer to the above method embodiment, and the present invention will not be repeated here. It has the same beneficial effects as the above OTFS channel estimation method.

[0188] Fig.16 A structural diagram of an OTFS channel estimation device provided by an embodiment of the present invention, such as Fig.16 As shown, the device comprises:

[0189] A memory 21, used for storing computer programs;

[0190] The processor 22 is configured to implement the steps of the OTFS channel estimation method when executing a computer program.

[0191] The OTFS channel estimation device provided in this embodiment may include but is not limited to a smart phone, a tablet computer, a laptop computer, or a desktop computer.

[0192] Among them, the processor 22 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 22 may be implemented in at least one hardware form of a digital signal processor (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 22 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 22 may be integrated with a graphics processing unit (GPU), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 22 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning.

[0193] The memory 21 may include one or more computer-readable storage media, which may be non-transitory. The memory 21 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 21 is at least used to store the following computer program 211, wherein the computer program can implement the relevant steps of the OTFS channel estimation method disclosed in any of the aforementioned embodiments after being loaded and executed by the processor 22. In addition, the resources stored in the memory 21 may also include an operating system 212 and data 213, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 212 may include Windows, Unix, Linux, etc. Data 213 may include, but is not limited to, data involved in the OTFS channel estimation method, etc.

[0194] In some embodiments, the OTFS channel estimation device may further include a display screen 23 , an input / output interface 24 , a communication interface 25 , a power supply 26 , and a communication bus 27 .

[0195] Those skilled in the art can understand that Fig.16 The structure shown in the figure does not constitute a limitation on the OTFS channel estimation device, and may include more or less components than those shown in the figure.

[0196] The processor 22 implements the OTFS channel estimation method provided by any of the above embodiments by calling the instructions stored in the memory 21 .

[0197] For an introduction to an OTFS channel estimation device provided by the present invention, please refer to the above method embodiment, which will not be described in detail herein. The device has the same beneficial effects as the above OTFS channel estimation method.

[0198] Furthermore, the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by the processor 22, the steps of the above-mentioned OTFS channel estimation method are implemented.

[0199] It is understandable that if the method in the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium to execute all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0200] For an introduction to a computer-readable storage medium provided by the present invention, please refer to the above method embodiment, which will not be described in detail herein. The present invention has the same beneficial effects as the above OTFS channel estimation method.

[0201] The above is a detailed introduction to an OTFS channel estimation method, system, device and medium provided by the present invention. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the embodiments can refer to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can refer to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, the present invention can also be improved and modified in several ways, and these improvements and modifications also fall within the scope of protection of the present invention.

[0202] It should also be noted that, in this specification, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

Claims

1. An OTFS channel estimation method, characterized in that: Applied to the receiving end, including: receiving a first signal sent by a transmitting end; and performing crosstalk analysis according to the first signal to obtain a delay Doppler domain received signal; Determine a target path according to the amplitude and threshold value of the delay Doppler domain received signal; and construct a channel matrix according to the path channel state information of the target path; The channel matrix is ​​input into a regression network model to obtain a denoised channel matrix, and a transmission signal corresponding to the first signal is reconstructed.

2. The OTFS channel estimation method according to claim 1, characterized in that: The training process of the regression network model includes: Obtain the initial regression network model, training data in a multipath channel scenario, and the actual channel matrix; Determine a first target path according to the amplitude corresponding to the training data in the multipath channel scenario and the threshold value; constructing a first channel matrix according to the path channel state information of the first target path; Using the first channel matrix as an input signal of the initial regression network model to obtain a first actual channel matrix of the initial regression network model; Record the current number of iterations; When the difference between the first actual channel matrix and the actual channel matrix does not meet the preset conditions, or the current number of iterations does not reach the preset number of iterations, the model parameters of the initial regression network model are fine-tuned to obtain the fine-tuned initial regression network model; and the method returns to the step of using the first channel matrix as the input signal of the initial regression network model until the difference between the first actual channel matrix and the actual channel matrix meets the preset conditions, or the current number of iterations reaches the preset number of iterations.

3. The OTFS channel estimation method according to claim 2, characterized in that: Inputting the channel matrix into the regression network model to obtain a denoised signal matrix includes: Performing feature extraction processing on the data of the channel matrix to obtain first feature data; The first feature data is subjected to feature learning residual processing to obtain second feature data; wherein the feature learning residual processing of the regression network model is completed in a feature learning residual module, and the feature learning residual module includes a plurality of enhanced attention modules, and there is a cascade relationship between the plurality of enhanced attention modules, and the enhanced attention module includes a plurality of convolution kernels and a merged convolution layer; The second characteristic data is reconstructed to obtain a denoised signal matrix.

4. The OTFS channel estimation method according to claim 3, characterized in that: The sizes of multiple convolution kernels in the enhanced attention module are different.

5. The OTFS channel estimation method according to claim 2, characterized in that: The difference between the first actual channel matrix and the actual channel matrix does not meet a preset condition, including: Obtaining the channel capacity and rank corresponding to the eigenvalues ​​of the first actual channel matrix and the actual channel matrix respectively; If the difference between the respective channel capacities is greater than or equal to the first threshold, the ranks corresponding to the respective eigenvalues ​​are compared; If the ranks corresponding to the respective eigenvalues ​​are greater than or equal to the second threshold, it is determined that the difference between the first actual channel matrix and the actual channel matrix does not meet a preset condition.

6. The OTFS channel estimation method according to claim 2, characterized in that: The difference between the first actual channel matrix and the actual channel matrix does not meet a preset condition, including: Obtaining the correlations and estimation errors corresponding to the first actual channel matrix and the actual channel matrix respectively; If the difference between the respective corresponding correlations is not within the first preset range, then comparing the respective corresponding estimation errors; If the difference between the corresponding estimation errors is not within the second preset range, it is determined that the difference between the first actual channel matrix and the actual channel matrix does not meet the preset condition.

7. The OTFS channel estimation method according to claim 1, characterized in that: The performing crosstalk analysis according to the first signal to obtain a delay Doppler domain received signal includes: Performing SFFT transformation on the first signal to obtain a first delay Doppler domain signal; Acquire a first target row and a second target row of the first delay-Doppler domain signal; The first delay-Doppler domain signal is intercepted according to the first target row and the second target row, and the signal from the first target row to the second target row is used as the delay-Doppler domain received signal.

8. An OTFS channel estimation system, characterized in that: Including the receiving end and the sending end: Controlling the sending end to send a first signal; Controlling the receiving end to perform crosstalk analysis according to the first signal to obtain a delay Doppler domain received signal; determining a target path according to the amplitude and threshold value of the delay Doppler domain received signal; Constructing a channel matrix according to the path channel state information of the target path; The channel matrix is ​​denoised to obtain a denoised channel matrix to reconstruct a transmission signal corresponding to the first signal.

9. An OTFS channel estimation device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of the OTFS channel estimation method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the OTFS channel estimation method according to any one of claims 1 to 7 are implemented.

Images