OFDM (Orthogonal Frequency Division Multiplexing) signal time domain nonlinear distortion recovery method and device for satellite communication

Abstract

The invention discloses a method and a system for recovering time domain nonlinear distortion of an OFDM (Orthogonal Frequency Division Multiplexing) signal of satellite communication, which mainly solve the problem that the nonlinear distortion of the OFDM signal of the satellite communication is difficult to recover in the prior art. According to the scheme, the method comprises the following steps: acquiring a data set of satellite communication OFDM signal time domain nonlinear distortion recovery; constructing a deep learning network model composed of an optimized multi-layer one-dimensional convolutional neural network and a bidirectional long-short-term memory network module; inputting the acquired data set into a deep learning network model for training, verification and testing so as to recover a time domain distortion signal; and carrying out channel equalization and demodulation on the recovered time domain signal, completing demodulation from a modulation symbol to bit information, and realizing effective recovery and transmission of data. The method has good mean square error and bit error rate performance, the complexity of a network model can be reduced, signal distortion caused by nonlinear distortion can be effectively recovered, the precision and efficiency of signal recovery can be improved, and the method can be used for recovering nonlinear distortion signals in a satellite communication system.

Description

Technical Field

[0001] The present invention belongs to the field of communication technologies, and particularly relates to a method and apparatus for recovering non - linear distortion of OFDM signals, which can be used to efficiently recover the non - linear distortion of OFDM signals caused by non - linear power amplifiers, multipath effects, noise interference, etc. in satellite communication systems, and improve the signal recovery quality and the overall performance of the system. Background Art

[0002] In satellite communication systems, the orthogonal frequency - division multiplexing (OFDM) technology has become one of the important modulation technologies for modern satellite communications due to its strong anti - multipath fading ability and high spectrum utilization rate. However, due to the particularity of the satellite communication environment, the signal will be affected by many factors during transmission, such as non - linear distortion introduced by non - linear power amplifiers, multipath effects, and noise interference, resulting in serious distortion of the received OFDM signal. If this distortion is not effectively recovered, it will have a serious impact on the performance of the communication system. Especially under high - order modulation and low signal - to - noise ratio conditions, the existence of distortion will greatly reduce the demodulation accuracy, leading to a significant increase in the bit error rate (BER).

[0003] To recover the distorted signal, traditional methods usually rely on model - based channel estimation and equalization algorithms, such as the minimum mean - square error (MMSE) algorithm and the zero - forcing (ZF) algorithm, etc. However, model - based methods usually require pre - assuming channel models and distortion models, and their performance may be greatly limited in complex communication environments.

[0004] With the development of deep learning technologies, data - driven deep learning methods have gradually demonstrated their powerful capabilities in complex non - linear problems. Against this background, deep - learning - based OFDM signal recovery methods have emerged. By learning a large amount of distorted signal data, deep - learning models can effectively capture the complex non - linear characteristics of the channel and achieve accurate recovery of time - domain signals without relying on explicit channel models.

[0005] A two-stage LMMSE / DNN receiver disclosed in Document 1 "Z. Huang, D. He and Z. Wang, 'Two-Stage LMMSE / DNN Receiver for High-Order Modulation,' in IEEE Communications Letters, vol. 27, no. 8, pp. 2068-2072, Aug. 2023." is used to improve the demodulation performance of OFDM signals in the presence of multipath effects and nonlinear distortion. This receiver consists of two parts, LMMSE and DNN. Specifically, the received signal is first processed by the LMMSE equalizer to mitigate the multipath effects, and then the DNN is used to further eliminate the nonlinear distortion. However, the designed network model is relatively simple, so it is difficult to adapt to complex nonlinear systems.

[0006] The patent document with the application number CN201310703252.6 discloses an efficient joint method for nonlinear distortion recovery and channel estimation in an OFDM ultra-wideband system. It performs two clippings at the transmitter to make the signal amplitude lower than a preset threshold, and at the receiver, it estimates and compensates for the nonlinear distortion based on compressive sensing sparse channel estimation and equalization. Although this method can suppress the signal nonlinear distortion and the degradation of channel estimation performance caused by the high peak-to-average power ratio in the OFDM ultra-wideband system, it has a high computational complexity and does not consider the nonlinear distortion effect brought by the power amplifier, which limits the nonlinear effect of the system.

[0007] The patent document with the application number CN201610512985.5 discloses an OFDM compressive sensing channel and nonlinear distortion joint estimation algorithm, which simultaneously estimates the nonlinear distortion brought by the channel and power amplification in the receiver to avoid complex peak-to-average power ratio algorithms at the transmitter. However, due to its high dependence on channel characteristics and the requirement for strong sparsity of the signal, it is difficult to process signals with complex nonlinearity and low sparsity characteristics.

[0008] In addition, when some existing methods use neural networks for channel estimation and signal recovery, they need to be trained with a large number of simulation data sets to enable the network to gradually learn the nonlinear characteristics and distortion laws of the channel. However, when these methods are applied to the satellite communication field, especially for the signal distortion recovery task brought by the power amplifier, they still face problems such as insufficient data sets, high network complexity, and overfitting. Summary of the Invention

[0009] The object of the present invention aims at the deficiencies of the above-mentioned existing technologies, and proposes a method and device for restoring the time-domain nonlinear distortion of OFDM signals in satellite communication, so as to restore the distortion of OFDM signals caused by factors such as nonlinear power amplifiers, multipath effects, and noise interference in satellite communication systems, and improve the signal quality and the overall performance of satellite communication systems.

[0010] To achieve the above object, the method for restoring the time-domain nonlinear distortion of OFDM signals in satellite communication provided by the present invention includes

[0011] Obtaining a time-domain nonlinear distortion restoration data set for satellite communication OFDM signals;

[0012] Constructing a deep learning network model including an N-layer one-dimensional convolutional neural network and L bidirectional long short-term memory networks;

[0013] Inputting the obtained time-domain nonlinear distortion restoration data set into the constructed deep learning network model for training to restore the time-domain distorted signal

[0014] For the restored time-domain signal Performing channel equalization and demodulation to obtain the final bit stream b.

[0015] Furthermore, the construction of the deep learning network model is realized as follows:

[0016] a) Construct an N (N≥2)-layer one-dimensional convolutional neural network, where the first convolutional layer receives a single-channel input signal y(n) and generates a feature map with m channels. The middle convolutional layers maintain m channels, the convolutional kernel size is i×i, the padding is j, and the ReLU activation function is applied after each convolutional layer for normalization to introduce the nonlinear characteristics of the network. The last convolutional layer converts the feature map with m channels back to a single channel for outputting the preliminarily restored signal sequence y(t);

[0017] b) Input the preliminarily restored signal sequence y(t) into the existing L bidirectional long short-term memory network modules, and perform residual skip connections between the first convolutional layer and the last convolutional layer to obtain the enhanced signal sequence y(m);

[0018] c) Input the enhanced signal sequence y(m) into the fully connected layer, and output the restored signal sequence after passing through the fully connected layer That is, the construction of the deep learning network model is completed.

[0019] Furthermore, the channel equalization and demodulation of the restored signal are realized as follows:

[0020] (a) For the restored signal Perform cyclic prefix CP removal and fast Fourier transform FFT operations to obtain the frequency-domain signal Y at the receiving end K ;

[0021] (b) Extract the pilot symbols from the frequency-domain signal Y K in it Perform minimum mean square error LMMSE channel estimation on it to obtain the channel response matrix at the pilot symbol locations

[0022] (c) Perform linear interpolation on the channel response matrix at the pilot symbol locations to obtain the channel response matrix H of the entire signal K ;

[0023] (d) Perform minimum mean square error MMSE equalization on the channel response matrix H K to obtain the equalized result X K ;

[0024] (e) Perform symbol decision on the equalized result X according to the constellation diagram rules, and demap the decision result into the corresponding bit data, that is, demodulate the bit data b

[0025] To achieve the above object, the satellite communication OFDM signal time-domain non-linear distortion recovery device provided by the present invention includes:

[0026] A digital modulation module for digitally modulating the bit stream and mapping it into frequency-domain complex symbols;

[0027] An OFDM modulation module for converting the frequency-domain complex symbols into a time-domain signal through inverse fast Fourier transform IFFT, and adding a cyclic prefix CP to reduce multipath interference;

[0028] A power amplification module for amplifying the signal after OFDM modulation to ensure the transmission power of the signal in the satellite communication link;

[0029] An up-conversion and RF module for converting the amplified baseband signal into an RF signal and transmitting it to the satellite through an antenna to achieve wireless transmission of the signal;

[0030] A receiving RF and down-conversion module for receiving the RF signal from the satellite and down-converting it back to the baseband signal for subsequent signal processing;

[0031] A time-domain restoration network module for performing preliminary non-linear distortion recovery processing on the received satellite signal;

[0032] An OFDM demodulation module for performing fast Fourier transform FFT on the preliminarily restored signal, removing the pilot symbols, and removing the cyclic prefix at the same time;

[0033] A channel estimation and equalization module, which is used to estimate the channel characteristics and perform equalization processing to compensate for the fading and distortion in the channel and restore the original transmitted signal;

[0034] A digital demodulation module, which is used to demap the equalized signal into a bit stream to complete the demodulation of the data.

[0035] Compared with the prior art, the present invention has the following advantages:

[0036] First, by combining the time-domain characteristics of OFDM signals in satellite communication, the present invention designs a deep learning model composed of an optimized one-dimensional convolutional neural network module and a bidirectional long short-term memory network module, which can effectively capture the time-domain characteristics of the channel and the power amplifier, and improve the accuracy and efficiency of signal distortion recovery.

[0037] Second, the present invention only performs distortion recovery at the receiving end without any processing on the transmitting end. Therefore, the computational overhead is greatly reduced, the distortion signal can be recovered in a shorter time, and the real-time performance and processing efficiency of the system are improved.

[0038] Third, after the present invention performs non-linear distortion recovery using a neural network at the receiving end, channel estimation and equalization are implemented to further recover the distortion caused by the channel, avoiding the use of a large amount of data for network training and reducing the probability of overfitting. Description of the Drawings

[0039] Figure 1 is a flowchart for implementing the method for time-domain non-linear distortion recovery of OFDM signals in satellite communication of the present invention;

[0040] Figure 2 is the network structure diagram constructed in the embodiment of the present invention;

[0041] Figure 3 is a block diagram of the structure of the device for time-domain non-linear distortion recovery of OFDM signals in satellite communication of the present invention;

[0042] Figure 4 is a simulation comparison diagram of the mean square error (MSE) between the present invention and the prior non-linear distortion recovery method;

[0043] Figure 5 is a simulation comparison diagram of the bit error rate (BER) between the present invention and the prior non-linear distortion recovery method. Detailed Embodiments

[0044] The following further elaborates on the specific embodiments of the present invention with reference to the accompanying drawings.

[0045] The technical key of the present invention lies in constructing a one-dimensional convolutional bidirectional long short-term memory network 1D-CNN-BiLSTM and applying it to the existing satellite communication OFDM system to achieve the recovery of time-domain non-linear distortion signals.

[0046] Embodiment 1, a method for recovering time-domain non-linear distortion of OFDM signals in satellite communication.

[0047] Refer to Figure 1 , the implementation steps of this example are as follows:

[0048] Step 1, obtain a dataset for recovering time-domain non-linear distortion of OFDM signals in satellite communication.

[0049] The implementation of this step is based on a satellite communication OFDM system, which includes digital modulation at the transmitter, OFDM modulation at the transmitter, power amplifier, satellite channel, OFDM demodulation at the receiver, and digital demodulation at the receiver. The specific steps are as follows:

[0050] 1.1) At the transmitter of the satellite communication system, first use the random function randi() to generate a binary bit stream of length where each a ∈ {0, 1}, Q is the number of symbols, P is the number of subcarriers, and M is the modulation order; in this example, the 16QAM modulation method is adopted, that is, M = 16, the number of symbols Q is set to 60, and the subcarrier format P is set to 512. The generated binary bit stream a is divided into several groups in sequence, and each group contains i bits. According to the 16QAM modulation scheme, map each group of 4 bits to a point on a two-dimensional constellation diagram. The real and imaginary parts of each 16QAM symbol respectively correspond to a coordinate point in the constellation diagram, and the coordinates of this point are determined by a predefined mapping rule. Arrange the mapped 16QAM symbols in sequence to generate the final modulation signal; 1.2) Perform an inverse fast Fourier transform IFFT on each group of Q symbols to convert the frequency-domain signal into a time-domain signal x(s):

[0051]

[0052]

[0053] where X k is the 16QAM symbol on the k-th subcarrier, and P is the number of subcarriers which can also be expressed as the number of points of the IFFT;

[0054] 1.3) In order to eliminate the inter-symbol interference ISI caused by multipath effects, insert vertical pilots in front of the time-domain signal x(s), add a cyclic prefix CP, and perform a serial-to-parallel conversion on the time-domain signal after adding the cyclic prefix to generate a continuous signal x(n), which is used as the label of the dataset;

[0055] 1.4) Calculate the nonlinear characteristics of the power amplifier to obtain the corresponding relationship G(|x(n)|) between the modulus of the output signal and the modulus of the input signal and the phase difference relationship φ(|x(n)|) between the output signal and the input signal:

[0056]

[0057] where α ρ , β ρ , α θ , β θ are four different parameters describing the system, and in this example, the values are but not limited to α ρ = 2, β ρ = 1, β θ = 1;

[0058] 1.5) Amplify the signal x(n) after parallel-to-serial conversion through the nonlinear power amplifier HPA to output the amplified signal z(n):

[0059] z(n) = G(|x(n)|)e jθ(x(n)+φ(|x(n)|))

[0060] 1.6) Pass the amplified signal z(n) through the satellite channel h, add Gaussian white noise w(n) for transmission to obtain the received signal y(n) at the receiving end, and use it as a sample of the data set:

[0061] y(n) = z(n)*h + w(n)

[0062] In this embodiment, the Lutz channel is used, but it is not limited to this channel;

[0063] 1.7) Generate N groups of samples y(n) and labels x(n) by looping N times, and form a data set with the obtained samples and labels.

[0064] Step 2, preprocess the data set composed of samples y(n) and labels x(n), and divide it into a training set, a validation set, and a test set.

[0065] 2.1) Perform dimensionality conversion on the samples y(n) and labels x(n) in the data set, extract their real and imaginary parts, and flatten them into a one-dimensional sequence, so that each signal sample becomes a long sequence composed of real and imaginary parts to adapt to the input rules of the network;

[0066] 2.2) Perform normalization processing to adjust the numerical range of the real and imaginary parts to the range of [0,1], making the data more stable and helping the network converge faster;

[0067] 2.2) Divide the normalized dataset into a training set, a validation set, and a test set according to a ratio of 7:1.5:1.5;

[0068] The training set is used to train the neural network model and is the basic data for model learning;

[0069] The validation set is used to evaluate the performance of the model during training, help adjust hyperparameters, and prevent overfitting;

[0070] The test set is used to evaluate the final effect of the model after training and simulate the performance of the model in actual applications.

[0071] Step 3, construct a deep learning network model.

[0072] In the existing fully connected deep neural network DNN, due to the excessive number of parameters, it is prone to overfitting and easily falls into local optima, and it cannot analyze the overall logical sequence between input information and the complex temporal correlation between information, so it cannot model the changes in time series. The denoising convolutional neural network DnCNN can effectively extract local features in the time-domain signal through multi-layer convolutional operations. This local feature extraction ability can help the model better identify and suppress the noise and interference caused by non-linear distortion. The bidirectional long short-term memory network BiLSTM captures the complex dependence relationships between signal points in the time-domain signal through its forward and backward bidirectional processing capabilities, thereby significantly improving the recovery accuracy and stability of non-linear distortion signals. Therefore, in this example, a network model composed of a denoising convolutional network DnCNN and a bidirectional long short-term memory network BiLSTM is constructed.

[0073] Reference Figure 2 , the specific implementation method of this step is as follows:

[0074] 3.1) Optimize the existing denoising convolutional network DnCNN model to reduce the computational complexity, that is, simplify the original 17-layer DnCNN into a 7-layer one-dimensional convolutional neural network 1D-CNN, where:

[0075] The first convolutional layer receives a single-channel input signal y(n) and generates a feature map with 32 channels;

[0076] The middle 5-6 convolutional layers maintain 32 channels, the size of each convolutional kernel is 3×3, and the padding is set to 1 to keep the feature map size unchanged. After the convolutional layer, ReLU activation operation and batch normalization are performed in sequence. Through the ReLU activation operation, the convergence speed of the neural network can be accelerated, and overfitting of the network can be alleviated. The batch normalization is to perform reverse normalization processing on the preprocessed data according to the mean and variance to reduce the phenomenon of internal covariate shift and prevent the model from fitting too slowly;

[0077] The last layer, i.e., the 7th convolutional layer, converts the feature map with 32 channels back to a single channel and outputs the preliminarily restored signal sequence y(t).

[0078] 3.2) Input the preliminarily restored signal sequence y(t) into an existing bidirectional long short-term memory network module, and perform a residual skip connection between the first convolutional layer and the last convolutional layer to obtain the enhanced signal sequence y(m). Adding the residual connection is to directly transmit the low-frequency information of the initial input signal to the output layer, improve the training efficiency and model performance, and alleviate the problem of gradient vanishing or explosion;

[0079] 3.3) Select a fully connected layer containing 128 neurons, input the enhanced signal sequence y(m) into this fully connected layer for linear transformation, non-linear activation, and feature extraction, and output the restored signal sequence This sequence is a long string of single-channel signals, which has the same dimension as the input time-domain signal, that is, the construction of the deep learning network model is completed.

[0080] Step 4, Train and validate the constructed deep learning network model.

[0081] 4.1) Set the minimum batch size batch, learning rate η, number of training epochs epoch, and total number of samples N;

[0082] 4.2) Initialize the parameters in the deep learning network model to give it an initial state to avoid the situation of gradient vanishing or gradient explosion in the initial stage of training;

[0083] 4.3) Input the divided training set into the network for forward propagation:

[0084] Forward propagation is the first stage in the neural network calculation process. In this stage, the input data first undergoes a linear transformation through the weight matrix in each convolutional layer; secondly, it undergoes a non-linear transformation through the ReLU activation function; finally, it undergoes batch normalization to normalize the data to generate the input of the next layer. This process is repeated until the last layer of the network, and finally the forward propagation result is output;

[0085] 4.4) After each forward propagation is completed, use the mean squared error MSE as the loss function to calculate the error between the network output and the true label:

[0086]

[0087] where N is the total number of samples, x(n) is the label, is the restored signal sequence;

[0088] 4.5) Calculate the gradient ▽ of the loss function with respect to the network parameters according to the error θJ(θ):

[0089] 4.6) Use the backpropagation algorithm to backpropagate the gradient from the output layer to the input layer to update the gradient, so as to reduce the output error and obtain the current gradient:

[0090] θ t+1 = θ t - η▽ θ J(θ)

[0091] where θ t+1 represents the currently updated parameter, θ t represents the current parameter, η is the learning rate, and ▽ θ J(θ) is the gradient of the loss function with respect to the parameter θ, which indicates the direction in which the parameter should be updated to reduce the loss;

[0092] 4.7) Adopt the Adam optimizer to gradually update the network parameters according to the calculated gradient and learning rate, and at the same time use the existing ReduceLROnPlateau learning rate optimizer to dynamically adjust the learning rate until the set number of training epochs is reached to complete the training;

[0093] 4.8) Input the divided validation set into the deep learning network model after training for forward propagation, repeat step 4.4), calculate the value of the MSE loss function for each round until the set number of training epochs is reached, and determine whether the value of the MSE loss function for each round is close to the loss value obtained by the network during training:

[0094] If they are close, it indicates that the constructed network model has good robustness and generalization ability. At this time, save the network model parameters of the last round as the optimal network model parameters;

[0095] If there is a large difference between the two, it is necessary to adjust parameters such as the learning rate and loss function during training and retrain and validate;

[0096] In this example, the total number of samples and labels is set to 2,000. Mini-batch gradient descent is used during training, with 32 mini-batches in each round, and a total of 100 rounds of training. The initial learning rate is set to 0.001, and when the learning rate is reduced each time, it will be updated by 0.5 times the current learning rate.

[0097] Step 5 Use the verified network model to recover the signal sequence

[0098] 5.1) Input the test set into the verified network model, load its optimal network model parameters, perform one forward propagation, and calculate the value of the MSE loss function;

[0099] 5.2) Determine whether the value of the MSE loss function is close to the loss value obtained in the last round of network training and validation:

[0100] If they are close, it indicates that the constructed network model has good robustness and generalization ability. At this time, regenerate a new set of distorted samples according to step 1 and input them into the network model to obtain the restored signal sequence.

[0101] If there is a large difference between the two, it is necessary to adjust parameters such as the learning rate and loss function during training and retrain and validate.

[0102] Step 6. For the restored signal sequence Perform channel estimation, equalization, and demodulation.

[0103] 6.1) For the restored signal sequence Perform cyclic prefix CP removal and fast Fourier transform FFT operations to obtain the frequency-domain signal Y at the receiving end K :

[0104] Y K = X K H K + W K

[0105] where X K is the frequency-domain signal transmitted by the transmitting end, H K is the channel frequency response, and W K is the frequency-domain noise.

[0106] 6.2) Extract the pilot signal from the frequency-domain signal Y at the receiving end K

[0107] 6.2.1) In the frequency-domain signal Y K According to the 16QAM modulation method arrangement rule and the signal transmission protocol, find the position of the pilot signal. This pilot signal is a known signal and is usually used for synchronization and channel estimation;

[0108] 6.2.2) Perform indexing and slicing operations according to the found pilot signal position to extract the pilot signal at the receiving end 6.3) Perform minimum linear mean square error LMMSE channel estimation on the pilot signal at the receiving end to obtain its channel response matrix

[0109] 6.3.1) Assume that the channel autocorrelation matrix R HH is the diagonal matrix of the true channel H, which is used to avoid complex matrix inversion operations and thus reduce the computational complexity;

[0110] ​6.3.2) Perform least squares estimation on the received pilot signal to obtain the result of the initial estimation

[0111]

[0112] where is the pilot signal at the transmitter;

[0113] 6.3.3) Set the channel estimation constant β for the 16QAM modulation scheme. In this example, but not limited to, the value of β is 17 / 9;

[0114] 6.3.4) Calculate the channel response matrix of the received pilot signal according to the channel autocorrelation matrix R HH and the result of the initial estimation where SNR is the signal-to-noise ratio and I is the identity matrix.

[0115]

[0116]

[0117] 6.4) Perform linear interpolation on the channel response matrix of the received pilot signal to obtain the channel response matrix H of the entire signal K :

[0118] 6.4.1) Determine the positions of the missing data points in the channel response matrix of the received pilot signal;

[0119] 6.4.2) Find the coordinates of the adjacent data points of the missing data points and use the linear interpolation formula to calculate the y value at x of the missing data point:

[0120] y = y 1 +(x - x 1 ) * (y 2 - y 1 ) / (x 2 - x 1 )

[0121] where (x, y) are the coordinates of a certain missing data point, and (x 1 , y 1 ) and (x 2 , y 2 ) are the coordinates of the adjacent points of the missing data point;

[0122] 6.4.3) Fill in the coordinates of all the obtained missing data points into the corresponding positions to obtain the channel response matrix H of the entire signal K .

[0123] ​6.5) Perform minimum mean square error (MMSE) equalization on the channel response matrix H K to obtain the equalized result X K :

[0124] 6.5.1) Determine the MMSE equalizer coefficient C using the average power of the transmitted signal, the average power of the noise, and the frequency-domain channel response matrix H K : MMSE :

[0125]

[0126] where is the average power of the noise, is the average power of the transmitted signal, H K * denotes the conjugate matrix of H K , and I is the identity matrix;

[0127] 6.5.2) Calculate the equalized result X using the MMSE equalizer coefficient C MMSE : K :

[0128] X K = C MMSE Y K

[0129] where Y K is the frequency-domain signal at the receiving end.

[0130] 6.6) Perform symbol decision on the equalized result X K according to the constellation diagram rule, and demap the decision result into the corresponding bit data, i.e., demodulate the bit data b:

[0131] 6.6.1) Map each signal in the equalized result X K to the constellation diagram of the corresponding modulation method, and calculate the Euclidean distance d between it and each point in the constellation diagram ij :

[0132]

[0133] where (x i , y i ) is the coordinate of each signal in the equalized result X, and (x j , y j ) is the coordinate of each point in the constellation diagram;

[0134] 6.6.2) According to the result of the distance calculation, select the constellation point with the closest distance as the decision result, aiming to minimize the symbol error rate and ensure the accurate recovery of the transmitted symbols as much as possible in the noise and interference environment;

[0135] 6.6.3) Map the decision result points to the original bitstream b, complete the demodulation process from modulation symbols to bit information, and thus achieve the effective recovery and transmission of data.

[0136] Reference Figure 3 A device for recovering the time-domain nonlinear distortion of OFDM signals in satellite communication provided by the present invention includes a digital modulation module 1, an OFDM modulation module 2, a power amplification module 3, an up-conversion and RF module 4, a receiving RF and down-conversion module 5, a time-domain restoration network module 6, an OFDM demodulation module 7, a channel estimation and equalization module 8, and a digital demodulation module 9.

[0137] The working principle of the entire device is as follows:

[0138] The digital modulation module 1 performs 16QAM modulation on the input bitstream, maps the binary data to specific points on the complex plane, forms symbols in the frequency domain, and transmits them to the OFDM modulation module 2; the OFDM modulation module 2 performs an inverse fast Fourier transform IFFT on the frequency-domain symbols, converts them into time-domain signals, adds pilots and a cyclic prefix CP, and then outputs them to the power amplification module 3; the power amplification module 3 amplifies the power of the OFDM-modulated signal to ensure that the signal has sufficient transmission power in the satellite communication link, and transmits the amplified signal to the up-conversion and RF module 4; the up-conversion and RF module 4 up-converts the power-amplified OFDM signal to the RF frequency band, and this RF signal is transmitted to the satellite through an antenna to achieve wireless transmission of the signal.

[0139] The down-conversion module 5 receives the RF signal from the satellite, down-converts it back to the baseband signal, and transmits it to the time-domain restoration network module 6; the time-domain restoration network module 6 performs preliminary nonlinear distortion recovery processing on the received satellite signal through a neural network. The restored signal is transmitted to the subsequent OFDM demodulation module 7; the OFDM demodulation module 7 first performs a fast Fourier transform FFT on the signal preliminarily restored in the time-domain multiplexing network to restore the frequency-domain symbols. At the same time, the cyclic prefix CP is removed and transmitted to the channel estimation and equalization module 8; the channel estimation and equalization module 8 performs channel estimation on the signal after removing the pilots and CP to obtain the characteristic parameters of the channel, such as attenuation, delay, etc., and uses these parameters for equalization processing to compensate for the fading and distortion in the channel and ensure the further accurate recovery of the signal; the equalized signal is transmitted to the digital demodulation module 9; the digital demodulation module 9 restores the frequency-domain symbols to the original binary bit data through the demapping rule of 16QAM corresponding to the modulation mapping rule, so as to accurately obtain the bit information transmitted by the sending end at the receiving end.

[0140] The effects of the present invention can be further illustrated by the following simulation results.

[0141] I. Simulation Conditions

[0142] The modulation method is 16QAM, the number of symbols is 60, the number of pilots is 3, the CP length is 32, the FFT size is 512, the number of training samples is 2,000, the initial learning rate is 0.001, the number of training epochs is 100, and the SNR range is 0 - 20 dB.

[0143] The simulation platforms are MATLAB R2023b, Python 3.7, and the PyTorch - GPU version. The required libraries are: os, numpy, sklearn, matplotlib, torch.

[0144] The training process is based on the Python 3.7 environment and is deployed on an NVIDIA RTX3080 GPU using the PyTorch deep - learning framework.

[0145] II. Simulation Contents

[0146] Simulation 1: Under the above conditions, the proposed method, the existing TSLD, LMMSE, and DnCN are respectively used to perform non - linear recovery processing on the OFDM time - domain distorted signal, and their respective MSE errors are calculated. The results are as Figure 4 .

[0147] From Figure 4 it can be seen that within the SNR range of 0 - 20 dB, the MSE error of the proposed method is the lowest compared with other existing methods, which proves the excellent performance of the proposed method in terms of MSE error.

[0148] Simulation 2: Under the above conditions, the proposed method, the existing TSLD, LMMSE, and DnCN are respectively used to perform non - linear recovery processing on the OFDM time - domain distorted signal, and their bit - error rates are calculated. The results are as Figure 5 .

[0149] From Figure 5It can be seen that when the signal-to-noise ratio is low, the bit error rate performance of the present invention is similar to that of several existing methods. When the signal-to-noise ratio is high, the present invention shows significant advantages in terms of bit error rate performance compared with other methods. Specifically, when the signal-to-noise ratio is 10 dB, the bit error rate of the present invention is about 0.0527, the bit error rate of the existing DnCNN method is 0.0628, the bit error rate of the existing TSLD method is about 0.0636, and the bit error rate of the existing LMMSE method is about 0.0724, which proves that the present invention has the best bit error rate performance. When the signal-to-noise ratio is 20 dB, the bit error rate of the present invention is about 0.0201, the bit error rate of the existing DnCNN method is about 0.0232, the bit error rate of the existing TSLD method is about 0.0362, and the bit error rate of the existing LMMSE method is about 0.0605, which proves that the present invention has a greater improvement in terms of bit error rate performance compared with the existing methods at high signal-to-noise ratios.

[0150] The above simulation results show that the present invention has good mean square error and bit error rate performance, can reduce the complexity of the implementation of the original model, effectively recover the signal distortion caused by non-linear distortion, and improve the accuracy and efficiency of signal recovery.

[0151] The above is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

[0152] It should be noted that the step numbers in the specification and claims of the present invention are only for a clear description of the embodiments of the present invention for easy understanding, and the order of their numbers is not limited.

Claims

1. A method for restoring time-domain nonlinear distortion of OFDM signals in satellite communications, characterized in that: include: Obtain the time domain nonlinear distortion data set of OFDM signals for satellite communications; Construct a deep learning network model including N layers of one-dimensional convolutional neural network and L bidirectional long short-term memory networks; The acquired time-domain nonlinear distortion data set is input into the constructed deep learning network model for training to restore the time-domain distortion signal. The restored time domain signal Channel estimation, equalization and demodulation are performed to obtain the final bit stream b.

2. The method according to claim 1, characterized in that The acquisition of a satellite communication OFDM signal time domain distortion recovery data set includes the following implementation: 2a) In the satellite communication system, the transmitter randomly generates a bit stream a, performs digital modulation on the data stream, performs inverse Fourier transform IFFT calculation on the modulated signal, adds a pilot signal and inserts the time domain signal x(n) after the cyclic prefix CP as a label; 2b) The time domain signal x(n) is amplified by a nonlinear power amplifier HPA to output a signal z(n), which is then passed through a satellite channel h, and after adding noise w(n), a signal y(n) is output as a sample; 2c) Use the obtained samples y(n) and labels x(n) to form a dataset.

3. The method according to claim 1, characterized in that: The construction of the deep learning network model includes the following steps: 3a) Construct an N (N ≥ 2) layer one-dimensional convolutional neural network, where the first convolution layer receives a single-channel input signal y(n) and generates a feature map of m channels. The middle convolution layer maintains m channels, the convolution kernel size is i×i, and the padding is j. ReLU activation and normalization are applied after each convolution layer to introduce the nonlinear characteristics of the network. The last convolution layer converts the feature map of m channels back to a single channel to output a preliminary recovered signal sequence y(t); 3b) inputting the initially recovered signal sequence y(t) into the existing L bidirectional long short-term memory network modules, and performing a residual jump connection between the first convolutional layer and the last convolutional layer to obtain the enhanced signal sequence y(m); 3c) Input the enhanced signal sequence y(m) into the fully connected layer, and output the restored signal sequence after passing through the fully connected layer That is to complete the construction of the deep learning network model.

4. The method according to claim 1, characterized in that: The acquired data set is input into the constructed deep learning network model for training, verification and testing to restore the time domain distorted signal, and its implementation includes the following: 4a) preprocessing the time domain complex signals in the data set by dimension conversion and normalization to adapt to the input rules of the network; 4b) Divide the preprocessed data into training set, validation set and test set according to the ratio of 7:1.5:1.5; 4c) Set the mean square error (MSE) as the loss function and calculate the error between the network output and the true label; 4d) Set the minimum batch size, learning rate η, training round epoch and total number of samples N; 4e) Initialize the parameters in the neural network model so that the network has an initial state; 4f) Input the divided training set into the constructed deep learning network model, start forward propagation, and output the initial results of the model; 4g) After each forward propagation, first calculate the MSE loss between the network output and the true label, and then calculate the gradient of the loss function with respect to the network parameters; 4h) Use the back-propagation algorithm to propagate the gradient from the output layer back to the input layer, and use the Adam optimizer to gradually update the network parameters according to the calculated gradient until the set training round epoch is reached and the training is completed; 4i) Input the divided validation set and test set into the verified network model to verify the robustness of the model; 4j) Generate a new set of distorted data and input it into the tested network model to obtain the restored signal sequence 5. The method according to claim 1, characterized in that: Channel estimation, equalization and demodulation are performed on the recovered signal, and the implementation includes the following: 5a) For the restored signal Remove the cyclic prefix CP and perform fast Fourier transform FFT operation to obtain the frequency domain signal Y at the receiving end K ; 5b) Extract the frequency domain signal Y K The pilot signal in Perform the minimum linear mean square error LMMSE channel estimation to obtain the channel response matrix of the pilot signal 5c) Channel response matrix for pilot signal Do linear interpolation to get the channel response matrix H of the entire signal K ; 5d) Channel response matrix H K Perform minimum mean square error MMSE equalization to obtain the equalized result X K ; 5e) Perform symbol decision on the equalized result X according to the constellation diagram rule, and demap the decision result into corresponding bit data, that is, demodulate the bit data b.

6. The method according to claim 5, characterized in that: Step 5b) takes out the pilot signal of the receiving end Perform minimum linear mean square error LMMSE channel estimation to obtain the channel response matrix of the pilot signal Its implementation includes: 5b1) Assume that the channel autocorrelation matrix R HH is the diagonal matrix of the real channel H; 5b2) Pilot signal to the receiving end Do the least squares estimation and get the result of the initial estimate in is the pilot signal at the transmitting end; 5b3) Setting the modulation constant β; 5b4) According to the autocorrelation matrix R HH and the results of the initial estimate Calculate the channel response matrix of the pilot signal Among them, SNR is the signal-to-noise ratio of the signal, and I is the unit matrix.

7. The method according to claim 5, characterized in that: The channel response matrix for the pilot signal in step 5c) is Do linear interpolation to get the channel response matrix H of the entire signal K , whose implementation includes: 5c1) Determine the channel response matrix of the pilot signal The location of the points with missing numbers in the ; 5c2) Find the coordinates of the adjacent data points of the missing data point and use the linear interpolation formula to calculate the y value of the missing data point at x: y=y1+(x-x1)*(y2-y1) / (x2-x1) Among them, (x, y) is the coordinate of a missing data point, (x1, y1) and (x2, y2) are the coordinates of the adjacent points of the missing data point; 5c3) Fill the coordinates of all missing data points into the corresponding positions to obtain the channel response matrix H of the entire signal K .

8. The method according to claim 5, characterized in that: In step 5d), the channel response matrix H K Perform minimum mean square error MMSE equalization to obtain the equalized result X K , whose implementation includes: 5d1) Using the average power of the transmitted signal, the average power of the noise and the frequency domain channel response matrix H K , determine the MMSE equalizer coefficient C MMSE : in, is the average power of the noise, is the average power of the transmitted signal, H K * Indicates H K The conjugate matrix of , I is the unit matrix; 5d2) Using MMSE equalizer coefficient C MMSE , calculate the balanced result X K : X K =C MMSE Y K Among them, Y K is the frequency domain signal at the receiving end.

9. The method according to claim 5, characterized in that: In step 5e), the equalized result X K Symbol decision is made according to the constellation diagram rules to demodulate the bit stream b, which includes: 5e1) The equilibrium result X K Each signal in is mapped to the constellation diagram of the corresponding modulation mode, and the Euclidean distance d between it and each point in the constellation diagram is calculated. ij ; Among them, (x i ,y i ) is the equilibrium result X K Each signal coordinate in (x j ,y j ) are the coordinates of each point in the constellation diagram; 5e2) According to the result of distance calculation, the constellation point with the closest distance is selected as the decision result; 5e3) Map the decision result point to the original bit stream b.

10. A device for restoring time-domain nonlinear distortion of OFDM signals for satellite communication, characterized in that ,include: A digital modulation module, used for digitally modulating a bit stream and mapping it into a frequency domain complex symbol; OFDM modulation module, used to convert frequency domain complex symbols into time domain signals through inverse fast Fourier transform IFFT, and add cyclic prefix CP to reduce multipath interference; The power amplifier module is used to amplify the power of the OFDM modulated signal to ensure the transmission power of the signal in the satellite communication link; The up-conversion and RF module is used to convert the amplified baseband signal into a RF signal and transmit it to the satellite through the antenna to achieve wireless transmission of the signal; The receiving RF and down-conversion module is used to receive the RF signal from the satellite and down-convert it back to the baseband signal for subsequent signal processing; A time domain restoration network module is used to perform preliminary nonlinear distortion restoration processing on the received satellite signals; The OFDM demodulation module is used to perform fast Fourier transform (FFT) on the initially recovered signal, remove the pilot symbol, and remove the cyclic prefix; The channel estimation and equalization module is used to estimate the channel characteristics and perform equalization processing to compensate for the fading and distortion in the channel and restore the original transmission signal; The digital demodulation module is used to demap the equalized signal into a bit stream and complete data demodulation.

Images