Channel estimation method and system of ATTCSDL-assisted OTFS in satellite communication

Through the ATTCSDL assisted OTFS channel estimation method, combined with multi-pipe embedded pilot and deep neural network, the problems of low channel estimation accuracy and low resource utilization efficiency in LEO satellite communication are solved, and high-precision channel estimation and transmission efficiency are improved. It exhibits superior performance especially in low signal-to-noise ratio environments.

CN119030833AInactive Publication Date: 2024-11-26HANGZHOU DIANZI UNIV

Patent Information

Application Number
CN202411507857.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-28
Publication Date
2024-11-26
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Traditional OFDM technology in LEO satellite communications has increased difficulty in capturing channel state information due to the Doppler effect. The existing OTFS channel estimation method is prone to path miss detection and excessive fading coefficient errors under harsh channel conditions, which is difficult to meet the needs of high-quality communications. .

Method used

Using the ATTCSDL assisted OTFS channel estimation method, the channel estimation is further refined by constructing the OTFS system model, designing a multi-pipe embedded pilot scheme, combining the OMP channel estimation module and deep neural network, deep learning methods using attention mechanisms are used.

Benefits of technology

It significantly improves channel estimation accuracy and resource utilization efficiency, reduces signal distortion and interference caused by the Doppler effect, improves the overall reliability and transmission efficiency of satellite communication systems, and shows excellent performance in low signal-to-noise ratio environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119030833A_ABST
    Figure CN119030833A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of satellite communication, and particularly relates to a channel estimation method and system for ATTCSDL-assisted OTFS in satellite communication. The method comprises the following steps: 1, constructing an OTFS system model, designing a multi-pilot embedded pilot scheme, simulating a signal transmission process in a satellite communication channel, and generating a pilot signal of a receiving end and corresponding channel information; 2, inputting the obtained channel information into a traditional OMP channel estimation module to serve as a channel pre-estimator so as to obtain a preliminary estimation value of a channel; 3, taking the preliminary channel estimation result as input, taking the actual channel response as a label to train the network, and adopting an ATTCSDL deep neural network method to further refine the estimation; and 4, performing real-time estimation on a channel by using the trained deep neural network. According to the method, challenges of satellite communication in a complex dynamic environment are accurately handled, the channel estimation precision is remarkably improved, the resource utilization efficiency is optimized, and stability and high efficiency of communication performance are ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of satellite communications, and in particular relates to a channel estimation method and system for ATTCSDL-assisted OTFS in satellite communications. Background Art

[0002] LEO satellites play a key driving role in the field of satellite communication technology with their small size, low cost and excellent ability to quickly supplement launches. With the in-depth exploration of terahertz communication technology in the 6G era, the communication link between LEO satellites and the ground faces a more complex and changeable environment, especially the extremely high Doppler frequency shift, which greatly increases the difficulty of accurately capturing channel state information. In this situation, the traditional OFDM technology suffers from the loss of subcarrier orthogonality caused by the Doppler effect, and the interference between carriers increases significantly, making it difficult to meet the needs of LEO satellite networks for high-quality communications.

[0003] In order to overcome the limitations of OFDM technology, orthogonal time-frequency-space (OTFS) modulation technology came into being. OTFS converts time-varying channels into the delay-Doppler (DD) domain, showing good adaptability to Doppler shift and delay, thereby showing excellent performance in high-mobility scenarios of LEO satellite communications. However, current OTFS channel estimation methods, such as those based on embedded pilots and guard symbols, are simple to implement and have low complexity, but are prone to path miss detection and excessive fading coefficient errors under harsh channel conditions. In recent years, the application of deep learning in physical layer channel estimation of wireless communications has made significant progress. Through offline training of a large number of data sets, data-driven deep neural networks are able to handle complex noise and show strong capabilities that do not rely on prior knowledge. However, its performance is highly dependent on the quality and quantity of training data, which is often difficult to guarantee in actual communication systems. In contrast, model-driven deep learning cleverly utilizes prior knowledge and only requires a smaller data set to build an effective network, which is more practical.

[0004] In view of the advantages of OTFS technology in solving Doppler interference in LEO satellite communications and the complexity of its channel estimation, it is particularly important to study new channel estimation methods for LEO satellite communication systems based on OTFS. Summary of the invention

[0005] In view of the above problems, the present invention proposes a channel estimation method and system for ATTCSDL-assisted OTFS in satellite communication, aiming to solve the problems of low channel estimation accuracy and low resource utilization efficiency in existing satellite communication technology under complex dynamic environments.

[0006] In order to achieve the above purpose, the present invention adopts the following technical solutions: The channel estimation method of ATTCSDL assisted OTFS in satellite communication includes the following steps: S1, build the OTFS system model and design a multi-pilot embedded pilot scheme to simulate the signal transmission process in the LEO satellite communication channel and generate the pilot signal and corresponding channel information at the receiving end; S2, inputting the acquired channel information into a conventional OMP channel estimation module as a channel pre-estimator to obtain a preliminary estimation value of the channel; S3, using the preliminary estimated value of the channel as input, the actual channel response as a label to train the deep neural network, and further refining the estimation using the ATTCSDL deep neural network method; S4, uses the trained deep neural network to estimate the channel in real time.

[0007] Among them, ATTCSDL stands for Attention Compressive Sensing Deep Learning, which refers to a deep learning method based on compressed sensing that integrates attention mechanism.

[0008] As a preferred solution, in step S1, constructing the OTFS system model specifically includes the following steps: S11, at the sending end, uses Indicates the delay-Doppler (DD) domain sent data symbols, where and Represent the Doppler dimension and the delay dimension respectively, is the number of data symbols, is the number of subcarriers; the data symbols are then mapped to the time-frequency domain through an inverse symplectic finite Fourier transform: (1); Where j is the imaginary unit; , after Heisenberg transformation, the time-frequency domain symbols Convert to time domain symbols : (2); in, To send a pulse waveform, is the subcarrier spacing, is the symbol duration; The LEO satellite channel is represented as a sparse multipath delay-Doppler channel: (3); in, Indicates the number of paths, and denote the channel gain, delay and Doppler shift of the i-th path respectively, represents the impulse function; S12, at the receiving end, the time domain signal is transmitted through the satellite channel to obtain the received signal: (4); in, It is additive white Gaussian noise. The received signal is transformed by Wigner to obtain the time-frequency domain symbol , and then after the symplectic finite Fourier transform, we finally get the symbol in the DD domain: (5); in, is Gaussian white noise.

[0009] As a preferred solution, in step S1, designing a multi-pilot embedded pilot solution specifically includes the following steps: S13, on the delay-Doppler grid, the pilot symbol is set in the middle, and the guard interval is set around the pilot area; in different dimensions, the interval between the pilot symbol and the data symbol is greater than the maximum Doppler frequency shift and the maximum delay frequency shift respectively; the signal sent by the transmitter is expressed as: (6); in, and Represent pilot symbols In the Doppler and delay dimensions, Indicates data symbol, and Respectively represent the maximum Doppler tap parameter and the maximum delay tap parameter, and Respectively represent the number of pilots in Doppler and delay dimensions; S14, in order to meet the minimum pilot overhead, the size of the channel estimation area is set to: (7); in, and They represent the channel estimation range of Doppler dimension and delay dimension respectively; finally, the total number of symbols in the channel estimation area is .

[0010] As a preferred solution, in step S1, the LEO satellite communication channel includes: Delay: By normalizing and scaling the delay in the LEO satellite channel, the system-specific path delay is obtained, namely: (8); in, is the normalized delay of the ith path, It is delay spread; Doppler: Set the Doppler shift to: (9); in, , , , , They represent carrier frequency shift, light speed, terminal moving speed, elevation angle between satellite and ground terminal, and angle between terminal moving direction and satellite projection plane on the ground respectively; is the carrier frequency shift caused by satellite motion, expressed as: (10); in, , , They represent the radius of the earth, the satellite's orbiting altitude, and the satellite's moving speed respectively.

[0011] As a preferred solution, in step S2, the OMP channel estimation module includes the following process: S21, initialization: the perception matrix And receive signal As input, initialize the residual , support set is an empty set, the number of iterations ; S22, iterative process: in the In the iteration, calculate the The residual error of the iteration The atom with the highest correlation with the perception matrix ;Will Incorporate support set And calculate the reconstructed signal under the new support set , and finally calculate the residual of the reconstructed signal , keep repeating the above process until the number of iterations is greater than the channel sparsity, then stop the iteration; S23, output result: final reconstructed signal This is the initial estimation result of the OMP algorithm for the channel. In , the number of non-zero elements is the number of multipaths of the current channel, the position of the non-zero elements corresponds to the delay and Doppler tap coefficient of the channel in the delay-Doppler domain, and the value of the non-zero element is the channel gain of the corresponding path.

[0012] As a preferred solution, in step S3, the ATTCSDL deep neural network method comprises the following steps: S31, data preprocessing: the channel result estimated in step S23 Construct a two-dimensional image with complex values , the two-dimensional image is specifically represented as a two-dimensional image consisting of a real part and an imaginary part: (11); in, and Respectively represent the imaginary part of the real part of the channel result; S32, deep neural network training: using the result of data preprocessing in step S31 as input and the actual channel response as a label to train the deep neural network; The structure design of the deep neural network is as follows: The input layer is used to extract shallow features of the input through a combination of a 3*3 convolution kernel Conv and a ReLU activation function; There are several intermediate layers stacked together, each consisting of a convolution kernel Conv, a batch normalization layer BN, and a ReLU activation function, which are used to further extract deep features; A convolutional attention module is embedded after shallow and deep feature extraction to improve the network's sensitivity to noise information; In the last layer of the deep neural network, a convolution operation is used to reconstruct the output to obtain a channel estimation result that matches the input dimension.

[0013] As a preferred solution, in step S32, during the entire process of training the deep neural network, the network structure is optimized by continuously adjusting the parameters of the deep neural network, and a deep neural network model with optimal performance is constructed by minimizing the loss function.

[0014] As a preferred solution, in step S32, the idea of ​​residual learning is used, that is, the final estimate of the channel matrix is ​​obtained by the residual between the rough estimate of the input and the noise value extracted by the deep neural network.

[0015] The present invention also provides an ATTCSDL-assisted OTFS channel estimation system in satellite communications, comprising: The signal transmission simulation module is used to build the OTFS system model and design a multi-pilot embedded pilot scheme to simulate the signal transmission process in the LEO satellite communication channel and generate the pilot signal and corresponding channel information at the receiving end; An estimation value acquisition module, used for inputting the acquired channel information into a conventional OMP channel estimation module as a channel pre-estimator, for obtaining a preliminary estimation value of the channel; A model training module, for taking the preliminary estimated value of the channel as input and the actual channel response as a label to train the deep neural network, and further refining the estimate using the ATTCSDL deep neural network method; The real-time estimation module is used to perform real-time estimation of the channel using a trained deep neural network.

[0016] Compared with the prior art, the present invention has the following technical effects: (1) Based on the idea of ​​inserting multiple pilot symbols in the OTFS frame, this paper proposes a deep learning channel estimation method based on compressed sensing. This method effectively reduces the signal distortion and interference caused by the Doppler effect through the sparse reconstruction capability of the OMP algorithm and the powerful nonlinear mapping capability of the improved DNCNN network, and significantly improves the overall reliability and transmission efficiency of the satellite communication system.

[0017] (2) The present invention cleverly introduces the attention mechanism module, making full use of its ability to efficiently filter key information in massive data, greatly enhancing the network's noise suppression and removal effect. The application of this mechanism makes the channel estimation process more accurate and efficient, can more accurately capture channel characteristics, and output high-precision channel estimation values;

[0018] (3) The present invention shows overwhelming advantages by comparing its performance with the channel estimation scheme in the existing OTFS literature. In a wide range of signal-to-noise ratios, especially in low signal-to-noise ratio environments (such as -10dB to -5dB), the present invention can still maintain excellent channel estimation performance, significantly surpassing the existing schemes and achieving a leap-forward improvement in channel estimation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flow chart of the channel estimation method of ATTCSDL aiding OTFS in satellite communication in the present invention; Figure 2 A schematic diagram of the OTFS system framework in the present invention; Figure 3 A schematic diagram of a design of a multi-pilot embedded pilot scheme in the present invention; Figure 4 A schematic diagram of the architecture of the ATTCSDL model in the present invention; Figure 5 is a NMSE performance diagram of the present invention under different signal-to-noise ratios (SNR); Figure 6 FIG. 4 is a NMSE performance diagram of the present invention under different pilot signal-to-noise ratios (SNRp). DETAILED DESCRIPTION

[0020] The following describes the embodiments of the present invention through specific embodiments, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0021] like Figure 1 As shown, this embodiment provides an ATTCSDL-assisted OTFS channel estimation method in satellite communications, which achieves high-precision channel estimation by inserting pilots in the OTFS frame and fusing the OMP coarse estimation and the DNCNN network enhanced by the CBAM attention mechanism.

[0022] The specific steps include: 1. Build an OTFS system model, design a multi-pilot embedded pilot scheme, simulate the signal transmission process in the LEO satellite communication channel, and generate the pilot signal and its corresponding channel information at the receiving end; Furthermore, in step 1, the construction of the OTFS system model includes the following steps: like Figure 2 As shown, at the sending end, using Indicates the delay-Doppler (DD) domain sent data symbols, where and Represent the Doppler dimension and the delay dimension respectively, is the number of data symbols, is the number of subcarriers; the data symbols are then mapped to the time-frequency domain through an inverse symplectic finite Fourier transform: ; in, Then, after Heisenberg transformation, the time-frequency domain symbols Convert to time domain symbols : ; in, To send a pulse waveform, is the subcarrier spacing, is the symbol duration; The LEO satellite channel can be represented as a sparse multipath delay-Doppler channel: ; in, Indicates the number of paths, and denote the channel gain, delay and Doppler shift of the i-th path respectively, represents the impulse function; At the receiving end, after the time domain signal is transmitted through the satellite channel, the received signal can be obtained: ; in, is additive white Gaussian noise. The received signal is transformed by Wigner to obtain the time-frequency domain symbol , and then after the symplectic finite Fourier transform, we finally get the symbol in the DD domain: ; in, is Gaussian white noise.

[0023] Furthermore, in step 1, a multi-pilot embedded pilot scheme is designed, including the following process: On the delay-Doppler grid, the pilot symbol is set in the middle and the guard interval is set around the pilot area to avoid interference with the pilot symbol. In different dimensions, the interval between the pilot symbol and the data symbol will be larger than the maximum Doppler shift and the maximum delay shift, respectively. Figure 3 As shown, the signal sent by the transmitter can be expressed as: ; in, and Represent pilot symbols In the Doppler and delay dimensions, Indicates data symbol, and Respectively represent the maximum Doppler tap parameter and the maximum delay tap parameter, and Respectively represent the number of pilots in Doppler and delay dimensions; In order to meet the minimum pilot overhead, the red area symbols are used for channel estimation, and their size is: ; in, and They represent the channel estimation range of Doppler dimension and delay dimension respectively. Therefore, the total number of symbols in the channel estimation area is .

[0024] Further, in step 1, the LEO satellite communication channel includes: Delay: Due to the long distance between the satellite and the ground UTs, the propagation delay is much greater than the propagation delay in the ground communication system. By normalizing and scaling the delay in the LEO satellite channel, the system-specific path delay can be obtained, that is: ; in, is the normalized delay of the ith path, It is delay spread; Doppler: In LEO satellite communication systems, the Doppler effect is caused by the movement of the satellite and the movement of the terminal. When the terminal is stationary, due to the high altitude of the satellite orbit, all path components reaching the satellite will have the same Doppler frequency shift. Since the satellite's motion trajectory is known, the Doppler frequency shift caused by the satellite motion can be pre-compensated. Therefore, the Doppler frequency shift can be expressed as: ; in, , , , , They represent carrier frequency shift, light speed, terminal moving speed, elevation angle between satellite and ground terminal, and angle between terminal moving direction and satellite projection plane on the ground respectively; is the carrier frequency shift caused by satellite motion, expressed as: ; in, , , They represent the radius of the earth, the satellite's orbiting altitude, and the satellite's moving speed respectively.

[0025] 2. Input the acquired channel information into the traditional OMP channel estimation module as a channel pre-estimator to obtain a preliminary estimate of the channel; Further, in step 2, the OMP channel estimation module comprises the steps of: 2-1. Initialization: The perception matrix And receive signal As input, initialize the residual , support set is an empty set, the number of iterations ; 2-2. Iteration process: In the In the iteration, calculate the The residual error of the iteration The atom with the highest correlation with the perception matrix .Will Incorporate support set And calculate the reconstructed signal under the new support set , and finally calculate the residual of the reconstructed signal , keep repeating the above process until the number of iterations is greater than the channel sparsity, then stop the iteration; 2-3. Output: Final reconstructed signal This is the initial estimation result of the OMP algorithm for the channel. In , the number of non-zero elements is the number of multipaths of the current channel, the position of the non-zero elements corresponds to the delay and Doppler tap coefficient of the channel in the delay-Doppler domain, and the value of the non-zero element is the channel gain of the corresponding path.

[0026] 3. The preliminary channel estimation results are used as input, the actual channel response is used as the label to train the network, and the ATTCSDL deep neural network method is used to further refine the estimation; In step 3, the ATTCSDL deep neural network method includes: 3-1. Data preprocessing: Current neural network models are mainly used to process real data, and there is no complex neural network designed specifically for complex data. Therefore, the channel result estimated in step 2-3 Can be constructed into a two-dimensional image with complex values , can be represented as a two-dimensional image consisting of a real part and an imaginary part: ; in, and Respectively represent the imaginary part of the real part of the channel result; 3-2. Deep neural network training: Use the result of data preprocessing in step 3-1 as input and the actual channel response as the label to train the deep neural network.

[0027] Network architecture such as Figure 4As shown in the figure, the input layer first extracts the shallow features of the input through a combination of a 3*3 convolution kernel (Conv) and a ReLU activation function. Subsequently, multiple intermediate layers consisting of Conv, batch normalization (BN) and ReLU are stacked to further extract deep features. In order to improve the network's sensitivity to noise information, the CBAM attention module is embedded after the shallow and deep features are extracted. This module significantly improves the model's perception and performance by introducing channel attention and spatial attention mechanisms without increasing the complexity of the network. In the last layer of the network, a convolution operation is used to reconstruct the output to obtain a channel estimation result that matches the input dimension. During the entire training process, the parameters of the ATTCSDL network are continuously adjusted to optimize the network structure and build the model with the best performance by minimizing the loss function. It is worth noting that in order to obtain the final channel estimation matrix, the idea of ​​residual learning is used, that is, the final estimate of the channel matrix is ​​obtained by the residual between the rough estimate of the input and the noise value extracted by the network. This approach not only improves the accuracy of the estimation, but also alleviates the gradient problem in deep network training.

[0028] 4. Use the trained deep neural network to estimate the channel in real time.

[0029] In addition, the present invention also provides an ATTCSDL-assisted OTFS channel estimation system in satellite communications, comprising: The signal transmission simulation module is used to build the OTFS system model and design a multi-pilot embedded pilot scheme to simulate the signal transmission process in the LEO satellite communication channel and generate the pilot signal and corresponding channel information at the receiving end; An estimation value acquisition module, used for inputting the acquired channel information into a conventional OMP channel estimation module as a channel pre-estimator, for obtaining a preliminary estimation value of the channel; A model training module, for taking the preliminary estimated value of the channel as input and the actual channel response as a label to train the deep neural network, and further refining the estimate using the ATTCSDL deep neural network method; The real-time estimation module is used to perform real-time estimation of the channel using a trained deep neural network.

[0030] The usage scenario considered in the present invention is satellite communication. The simulation channel adopts the satellite-to-ground communication link level simulation channel model NTN-TDL-D defined by the 3GPP protocol. The satellite altitude is , the satellite speed is , the maximum speed of the mobile terminal is set to , the delay extension time is The OTFS system parameters described in step 1 are set as follows: , , , , using 4-QAM modulation. For the ATTCSDL deep neural network described in step 3, Adam is selected as the optimizer, the batch size is set to 32, and the learning rate is set to 0.001. In order to evaluate the channel estimation performance, the proposed ATTCSDL algorithm is compared with CSDL, the traditional least squares LS method and the OMP method, and the normalized mean square error (NMSE) is used as the performance indicator of channel estimation, which is defined as: ; in, and denote the estimated and true channel responses respectively.

[0031] Figure 5 The NMSE performance of the present invention under different SNRs is shown. Figure 5 ATTCSDL 1Pilot and ATTCSDL2Pilot represent the estimation results of the ATTCSDL model using one pilot symbol and two pilot symbols, respectively. It can be observed that the ATTCSDL model shows the best NMSE performance under all signal-to-noise ratio conditions, and its signal-to-noise ratio gain is particularly significant when the same NMSE level is achieved. When the SNR is 5dB, the NMSE performance of ATTCSDL is less than , far exceeding the existing channel estimation scheme of OTFS. At the same level, ATTCSDL achieves about 2dB signal-to-noise ratio gain compared to CSDL and about 5dB compared to OMP. The gain is more significant for the traditional LS scheme, which proves the excellent estimation performance of the present invention and its robustness in low signal-to-noise ratio scenarios.

[0032] Figure 6 The NMSE performance of the present invention under different SNRp is shown. Figure 6 ATTCSDL 1Pilot and ATTCSDL2Pilot represent the estimation results of the ATTCSDL model using one pilot symbol and two pilot symbols, respectively. Obviously, the proposed ATTCSDL algorithm is significantly better than other methods. And the NMSE of different channel estimation algorithms decreases with the increase of pilot signal-to-noise ratio SNRp, that is, the channel estimation accuracy increases. It is worth noting that the use of two pilots shows significant superiority in NMSE performance compared to a single pilot. The reduced NMSE value clearly confirms the effectiveness of the multi-pilot channel estimation scheme in improving the accuracy of channel estimation, and fully verifies the important role of multi-pilot configuration in enhancing system robustness.

[0033] The above description is only a detailed description of the preferred embodiments and principles of the present invention. For ordinary technicians in this field, according to the ideas provided by the present invention, there will be changes in the specific implementation methods, and these changes should also be regarded as the protection scope of the present invention.

Claims

1. ATTCSDL assisted OTFS channel estimation method in satellite communication, characterized in that: The following steps are involved: S1, build the OTFS system model and design a multi-pilot embedded pilot scheme to simulate the signal transmission process in the LEO satellite communication channel and generate the pilot signal and corresponding channel information at the receiving end; S2, inputting the acquired channel information into a conventional OMP channel estimation module as a channel pre-estimator to obtain a preliminary estimation value of the channel; S3, using the preliminary estimated value of the channel as input, the actual channel response as a label to train the deep neural network, and further refining the estimation using the ATTCSDL deep neural network method; S4, uses the trained deep neural network to estimate the channel in real time.

2. The channel estimation method of ATTCSDL assisted OTFS in satellite communication according to claim 1, characterized in that: In step S1, the OTFS system model is constructed, specifically comprising the following steps: S11, at the sending end, uses Indicates the delay-Doppler DD domain sent data symbols, where and Represent the Doppler dimension and the delay dimension respectively, is the number of data symbols, is the number of subcarriers; the data symbols are then mapped to the time-frequency domain through an inverse symplectic finite Fourier transform: (1); Where j is the imaginary unit; , after Heisenberg transformation, the time-frequency domain symbols Convert to time domain symbols : (2); in, To send a pulse waveform, is the subcarrier spacing, is the symbol duration; The LEO satellite channel is represented as a sparse multipath delay-Doppler channel: (3); in, Indicates the number of paths, and denote the channel gain, delay and Doppler shift of the i-th path respectively, represents the impulse function; S12, at the receiving end, the time domain signal is transmitted through the satellite channel to obtain the received signal: (4); in, It is additive white Gaussian noise. The received signal is transformed by Wigner to obtain the time-frequency domain symbol , and then after the symplectic finite Fourier transform, we finally get the symbol in the DD domain: (5); in, is Gaussian white noise.

3. The channel estimation method of ATTCSDL assisted OTFS in satellite communication according to claim 1, characterized in that: In step S1, the design of a multi-pilot embedded pilot scheme specifically includes the following steps: S13, on the delay-Doppler grid, the pilot symbol is set in the middle, and the guard interval is set around the pilot area; in different dimensions, the interval between the pilot symbol and the data symbol is greater than the maximum Doppler frequency shift and the maximum delay frequency shift respectively; the signal sent by the transmitter is expressed as: (6); in, and Represent pilot symbols In the Doppler and delay dimensions, Indicates data symbol, and Respectively represent the maximum Doppler tap parameter and the maximum delay tap parameter, and Respectively represent the number of pilots in Doppler and delay dimensions; S14, in order to meet the minimum pilot overhead, the size of the channel estimation area is set to: (7); in, and They represent the channel estimation range of Doppler dimension and delay dimension respectively; finally, the total number of symbols in the channel estimation area is .

4. The channel estimation method of ATTCSDL assisted OTFS in satellite communication according to claim 1, characterized in that: In step S1, the LEO satellite communication channel includes: Delay: By normalizing and scaling the delay in the LEO satellite channel, the system-specific path delay is obtained, namely: (8); in, is the normalized delay of the ith path, It is delay spread; Doppler: Set the Doppler shift to: (9); in, , , , , They represent carrier frequency shift, light speed, terminal moving speed, elevation angle between satellite and ground terminal, and angle between terminal moving direction and satellite projection plane on the ground respectively; is the carrier frequency shift caused by satellite motion, expressed as: (10); in, , , They represent the radius of the earth, the satellite's orbiting altitude, and the satellite's moving speed respectively.

5. The channel estimation method of ATTCSDL assisted OTFS in satellite communication according to claim 1, characterized in that: In step S2, the OMP channel estimation module includes the following process: S21, initialization: the perception matrix And receive signal As input, initialize the residual , support set is an empty set, the number of iterations ; S22, iterative process: in the In the iteration, calculate the The residual error of the iteration The atom with the highest correlation with the perception matrix ;Will Incorporate support set And calculate the reconstructed signal under the new support set , and finally calculate the residual of the reconstructed signal , keep repeating the above process until the number of iterations is greater than the channel sparsity, then stop the iteration; S23, output result: final reconstructed signal This is the initial estimation result of the OMP algorithm for the channel. In , the number of non-zero elements is the number of multipaths of the current channel, the position of the non-zero elements corresponds to the delay and Doppler tap coefficient of the channel in the delay-Doppler domain, and the value of the non-zero element is the channel gain of the corresponding path.

6. The channel estimation method of ATTCSDL assisted OTFS in satellite communication according to claim 5, characterized in that: In step S3, the ATTCSDL deep neural network method comprises the following steps: S31, data preprocessing: the channel result estimated in step S23 Construct a two-dimensional image with complex values , the two-dimensional image is specifically represented as a two-dimensional image consisting of a real part and an imaginary part: (11); in, and Respectively represent the imaginary part of the real part of the channel result; S32, deep neural network training: using the result of data preprocessing in step S31 as input and the actual channel response as a label to train the deep neural network; The structure design of the deep neural network is as follows: The input layer is used to extract shallow features of the input through a combination of a 3*3 convolution kernel Conv and a ReLU activation function; There are several intermediate layers stacked together, each consisting of a convolution kernel Conv, a batch normalization layer BN, and a ReLU activation function, which are used to further extract deep features; A convolutional attention module is embedded after shallow and deep feature extraction to improve the network's sensitivity to noise information; In the last layer of the deep neural network, a convolution operation is used to reconstruct the output to obtain a channel estimation result that matches the input dimension.

7. The channel estimation method of ATTCSDL assisted OTFS in satellite communication according to claim 6, characterized in that: In step S32, during the entire process of training the deep neural network, the network structure is optimized by continuously adjusting the parameters of the deep neural network, and a deep neural network model with optimal performance is constructed by minimizing the loss function.

8. The channel estimation method of ATTCSDL assisted OTFS in satellite communication according to claim 1, characterized in that: In step S32, the idea of ​​residual learning is used, that is, the final estimate of the channel matrix is ​​obtained by the residual between the rough estimate of the input and the noise value extracted by the deep neural network.

9. ATTCSDL assisted OTFS channel estimation system in satellite communication, used to implement the ATTCSDL assisted OTFS channel estimation method in satellite communication according to any one of claims 1 to 8, characterized in that: The ATTCSDL-assisted OTFS channel estimation system in satellite communications includes: The signal transmission simulation module is used to build the OTFS system model and design a multi-pilot embedded pilot scheme to simulate the signal transmission process in the LEO satellite communication channel and generate the pilot signal and corresponding channel information at the receiving end; An estimation value acquisition module, used for inputting the acquired channel information into a conventional OMP channel estimation module as a channel pre-estimator, for obtaining a preliminary estimation value of the channel; A model training module, for taking the preliminary estimated value of the channel as input and the actual channel response as a label to train the deep neural network, and further refining the estimate using the ATTCSDL deep neural network method; The real-time estimation module is used to perform real-time estimation of the channel using a trained deep neural network.

Citation Information

Patent Citations

  • OTFS channel estimation method based on model-driven deep learning

    CN116055261A

  • Channel estimation method for OTFS system in Internet of Vehicles based on improved convolutional neural network

    CN116248444A

  • OTFS fractional Doppler channel estimation method based on DCNN

    CN116886472A

  • Channel estimation method of low earth orbit satellite system based on improved GOMP

    CN117749241A

  • Single-peak multi-pilot structure in OTFS system and channel estimation method thereof

    CN118233257A

Cited By

  • Joint detection method of satellite-ground downlink NOMA system

    CN119835122A