Orthogonal frequency division multiplexing system signal recovery method and device and electronic equipment

Abstract

The application relates to the technical field of wireless communication, and provides a signal recovery method and device for an orthogonal frequency division multiplexing system and electronic equipment. The method comprises the following steps: collecting a received signal; the received signal is obtained by receiving a transmitted signal of the orthogonal frequency division multiplexing system; determining a signal recovery result of the received signal; inputting the received signal and a pilot sequence into a conditional generative adversarial network to obtain a signal recovery result output by the conditional generative adversarial network. In the application, the conditional generative adversarial network is used for signal detection and signal recovery of the received signal of the orthogonal frequency division multiplexing system; the multiple cascaded sub-layers of the generator in the conditional generative adversarial network are used for gradient updating based on a rank restriction reverse transmission method with a constant matrix rank during training; the multiple cascaded sub-layers are trained as a whole, and the training of a large dense layer is replaced by the training of a lightweight sub-layer, so that the network training complexity is reduced, the data processing speed is improved, and the calculation efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, specifically to a method, apparatus, and electronic device for signal recovery in an orthogonal frequency division multiplexing system. Background Technology

[0002] Orthogonal Frequency Division Multiplexing (OFDM) is a multi-carrier modulation technique used in advanced wireless communication systems, particularly in high-speed Wireless Local Area Network (WLAN) standards, and is an integral part of 5G cellular networks. A key characteristic of OFDM is its ability to convert frequency-selective fading into a flat fading channel. Due to its high spectral efficiency and strong resistance to multipath interference, OFDM modulation schemes have been adopted as the primary transmission method for high data rate systems. In OFDM systems, binary bitstream data is modulated onto corresponding subcarriers using modulation methods such as Phase Shift Keying (PSK) or Quadrature Amplitude Modulation (QAM).

[0003] In practical orthogonal frequency division multiplexing (OFDM) systems, signal degradation occurs due to carrier frequency offset, timing deviation, and frequency-selective fading, leading to phase shifts and amplitude variations. To recover data at the receiver, it is necessary to obtain an effective estimate of the channel state information (CSI) to ensure the accuracy and reliability of signal detection. The performance of an OFDM system largely depends on the channel estimation and signal detection methods employed. Traditional methods with high signal detection accuracy typically suffer from high computational complexity, resulting in low computational efficiency. Summary of the Invention

[0004] This application provides a method, apparatus, and device for signal recovery in an orthogonal frequency division multiplexing system, which solves the technical problem that the high computational complexity of traditional signal detection methods leads to low computational efficiency.

[0005] In a first aspect, embodiments of this application provide a signal recovery method for an orthogonal frequency division multiplexing system, comprising:

[0006] Acquire and receive signals; the received signals are obtained by receiving the transmitted signals of the orthogonal frequency division multiplexing system;

[0007] Determine the signal recovery result of the received signal: input the received signal and pilot sequence into a conditional generative adversarial network to obtain the signal recovery result output by the conditional generative adversarial network;

[0008] The fully connected hidden layers of the generator in the conditional generative adversarial network include multiple cascaded sub-layers, and the gradients of these multiple cascaded sub-layers are updated during training using a rank-constrained backpropagation method with matrix rank invariance. The conditional generative adversarial network is trained based on sample received signals, real transmitted signals from an orthogonal frequency division multiplexing system, and sample pilot sequences.

[0009] In one embodiment, each fully connected hidden layer of the generator in the conditional generative adversarial network includes three cascaded sub-layers; the three cascaded sub-layers perform gradient updates using a row representation matrix, a column representation matrix, and an intermediate matrix, based on a rank-constrained backpropagation method with matrix rank invariant.

[0010] The invariance of matrix rank indicates that the product of the ranks of the row representation matrix, column representation matrix, and intermediate matrix of the three cascaded sub-layers remains unchanged from the original matrix rank of the fully connected hidden layer.

[0011] In one embodiment, the method further includes the step of training the conditional generative adversarial network; training the conditional generative adversarial network includes:

[0012] The sample received signal and the sample pilot sequence are input into the generator of the conditional generative adversarial network;

[0013] The generator's output signal, the sample pilot sequence, and the actual transmitted signal are input into the discriminator of the conditional generative adversarial network, so that the discriminator can distinguish the output signal, the sample pilot sequence, and the actual transmitted signal.

[0014] Construct a loss function, and adjust the generator's model parameters based on the loss function and the discrimination result of the discriminator.

[0015] In one embodiment, the loss function is constructed based on the generative adversarial network loss function and the mean squared error loss function.

[0016] In one embodiment, before determining the signal recovery result of the received signal, the method further includes:

[0017] The received signal is preprocessed based on the least squares method and the zero-forcing detection method.

[0018] In one embodiment, after determining the signal recovery result of the received signal, the method further includes:

[0019] The signal recovery result is then stored on the blockchain.

[0020] Secondly, embodiments of this application also provide a signal detection device for an orthogonal frequency division multiplexing system, comprising:

[0021] The acquisition module is used to acquire received signals; the received signals are obtained by receiving the transmitted signals of the orthogonal frequency division multiplexing system.

[0022] The signal recovery result determination module is used to determine the signal recovery result of the received signal: inputting the received signal and pilot sequence into a conditional generative adversarial network to obtain the signal recovery result output by the conditional generative adversarial network;

[0023] The fully connected hidden layers of the generator in the conditional generative adversarial network include multiple cascaded sub-layers, and the gradients of these multiple cascaded sub-layers are updated during training using a rank-constrained backpropagation method with matrix rank invariance. The conditional generative adversarial network is trained based on sample received signals, real transmitted signals from an orthogonal frequency division multiplexing system, and sample pilot sequences.

[0024] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement the steps of the orthogonal frequency division multiplexing system signal recovery method described in the first aspect.

[0025] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the orthogonal frequency division multiplexing system signal recovery method described in the first aspect.

[0026] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the orthogonal frequency division multiplexing system signal recovery method described in the first aspect.

[0027] The signal recovery method, apparatus, and electronic device for orthogonal frequency division multiplexing (OFDM) systems provided in this application embodiment achieve signal detection and recovery of received signals in an OFDM system by inputting the received signal and pilot sequence into a conditional generative adversarial network (GAN) to obtain the signal recovery result output by the GAN. The method utilizes a conditional generative adversarial network to perform gradient updates during training of multiple cascaded sub-layers of the generator in the GAN, based on a rank-constrained backpropagation method with matrix rank invariance. By training multiple cascaded sub-layers as a whole, replacing the training of large dense layers with the training of lightweight sub-layers, and replacing the calculation of large weight matrices with the calculation of three small weight matrices, the complexity of network training is reduced, data processing speed is increased, and the computational efficiency of signal recovery in the OFDM system is improved. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is one of the flowcharts illustrating the signal recovery method for an orthogonal frequency division multiplexing system provided in this application embodiment;

[0030] Figure 2 This is the second flowchart illustrating the signal recovery method for an orthogonal frequency division multiplexing system provided in this application embodiment;

[0031] Figure 3 This is a schematic diagram of the internal structure of the generator of the conditional generative adversarial network provided in the embodiments of this application;

[0032] Figure 4 This is a schematic diagram of the structure of the conditional generative adversarial network provided in an embodiment of the present invention;

[0033] Figure 5 This is the third flowchart illustrating the signal recovery method for an orthogonal frequency division multiplexing system provided in this application embodiment;

[0034] Figure 6 This is the fourth flowchart illustrating the signal recovery method for an orthogonal frequency division multiplexing system provided in this application embodiment;

[0035] Figure 7 yes Figure 6 A schematic diagram illustrating the implementation principle of step 400;

[0036] Figure 8 This is a schematic diagram of the structure of the signal recovery device for an orthogonal frequency division multiplexing system provided in the embodiments of this application;

[0037] Figure 9 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0039] Figure 1This is a flowchart illustrating the signal recovery method for an orthogonal frequency division multiplexing system provided in an embodiment of this application. Please refer to... Figure 1 This application provides a method for signal recovery in an orthogonal frequency division multiplexing system, which may include:

[0040] Step 100: Acquire and receive signals; the received signals are obtained by receiving the transmitted signals of the orthogonal frequency division multiplexing system.

[0041] The receiver of an Orthogonal Frequency Division Multiplexing (OFDM) system acquires the received signal. The received signal is obtained by receiving the transmitted signal from the OFDM system. It should be noted that, in one embodiment, please refer to... Figure 2 The received signal is obtained in the following way:

[0042] At the OFDM system transmitter, a bitstream signal is first transmitted. After QAM modulation, the signal is transmitted across multiple parallel subcarriers. An Inverse Fast Fourier Transform (IFFT) converts the frequency-domain signal to a time-domain signal. Next, a pre-defined pilot sequence is inserted at equal intervals into the OFDM symbols, arranged in a block-like structure. To eliminate inter-carrier interference, a cyclic prefix (CP) is added before the OFDM symbol, where the length of the CP should be greater than or equal to the maximum delay spread of the channel. The time-domain signal, after OFDM modulation, is then transmitted through the wireless channel. At the OFDM system receiver, the cyclic prefix (CP) is removed, and a Fast Fourier Transform (FFT) is performed to convert the parallel data stream into a serial symbol stream, yielding the received signal.

[0043] Step 300: Determine the signal recovery result of the received signal: Input the received signal and pilot sequence into the conditional generative adversarial network (GAN) to obtain the signal recovery result output by the GAN; each fully connected hidden layer of the generator in the GAN includes multiple cascaded sub-layers, and the multiple cascaded sub-layers are updated with gradients based on the rank-constrained backpropagation method with matrix rank invariance during training; the GAN is trained based on sample received signals, real transmitted signals based on orthogonal frequency division multiplexing (OFDM) systems, and sample pilot sequences.

[0044] The receiver of an Orthogonal Frequency Division Multiplexing (OFDM) system determines the signal recovery result of the received signal. In this embodiment, a conditional generative adversarial network is used to perform signal recovery on the received signal to obtain the same signal recovery result as the transmitted signal from the OFDM system transmitter.

[0045] Conventional generative adversarial models consist of a generator (i.e., a generative model) and an adversarial model (i.e., a discriminator). Both the generator and discriminator are constructed from deep neural networks. The two models are trained adversarially to learn the mapping from random variables to real data. However, this mapping is random and unstable, increasing the training difficulty, and its performance is generally poor under low signal-to-noise ratio conditions. Therefore, this embodiment of the invention adds a constraint condition, namely a pilot sequence Φ, to the input of the generator as a conditional input, guiding the generator to generate more realistic data and learn the mapping from the conditional input to real data. This network architecture is called a Conditional Generative Adversarial Network (cGAN).

[0046] The existing conditional generative adversarial network model has three fully connected hidden layers. Please refer to [reference needed]. Figure 3 , Figure 3 This diagram illustrates the internal structure of the generator in a conditional generative adversarial network according to an embodiment of this application. Inspired by near-optimal channel matrix estimation methods based on the low-rank structure of large-scale MIMO (Multiple Input Multiple Output) channel matrices, to accelerate the training convergence speed of the generator, each fully connected hidden layer is replaced with multiple cascaded sub-layers. Simultaneously, to adapt to gradient updates during the training process of this architecture, the existing backpropagation (BP) method is modified into a rank-constrained backpropagation method with matrix rank invariance. Multiple cascaded sub-layers are trained as a whole, and the training of large dense layers is replaced with the training of lightweight sub-layers. The calculation of large weight matrices is correspondingly replaced with the calculation of multiple small weight matrices, thereby reducing the network training complexity and data processing speed.

[0047] Specifically, when the fully connected hidden layers of the generator in the conditional generative adversarial network (GAN) model in this embodiment are set to three, the bit error rate performance is optimal. Therefore, the generator model of the GAN in this embodiment consists of an input layer, three hidden layers, and an output layer. Each layer in the network has multiple neurons, and the output of each neuron is a nonlinear function of the weighted sum of the neurons in the previous layer. The nonlinear functions of the first and second hidden layers in the neural network are set to the ReLU function, which is represented as f. R (a) = max(0, a). The nonlinear function of the third layer is set as the Sigmoid function, which is expressed as: In the simulation experiment, a physical frame includes two OFDM symbols, and each OFDM symbol includes 64 subcarriers. Each element of the signal vector in communication is a complex number. The real and imaginary parts of the complex vector are split and concatenated to form a 256*1 real vector. The first 128 elements of the vector are the real part of the original vector, and the last 128 elements are the imaginary part. Therefore, the number of neurons in the input layer of the above conditional generative adversarial network model is set to 256. The three hidden layers in the generator contain 512, 256, and 16 neurons respectively, the three hidden layers in the discriminator contain 512, 256, and 1 neuron respectively, and the number of neurons in the final output layer is set to 1. Finally, the network is trained on a dataset using Tensorflow. Table 1 shows the parameter settings of the OFDM system and GAN (Generative Adversarial Network) in this embodiment of the application.

[0048] Table 1

[0049]

[0050]

[0051] To accelerate the training and convergence speed of the generator, please refer to... Figure 3 In this embodiment, each fully connected hidden layer is replaced with multiple cascaded sub-layers (e.g., 3 cascaded sub-layers). To adapt to gradient updates during the training process of this architecture, the existing back propagation (BP) method is modified into a rank-constrained back propagation method with matrix rank invariance. Multiple cascaded sub-layers are trained as a whole, and the training of large dense layers is replaced with the training of lightweight sub-layers. The calculation of large weight matrices is replaced with the calculation of multiple small weight matrices, thereby reducing the network training complexity and data processing speed.

[0052] Therefore, this embodiment of the application achieves signal detection and signal recovery of the received signal of the orthogonal frequency division multiplexing system by inputting the received signal and pilot sequence into a conditional generative adversarial network (GAN) to obtain the signal recovery result output by the GAN. The multiple cascaded sub-layers of the generator in the GAN are trained using a rank-constrained backpropagation method with matrix rank invariance for gradient updates. The multiple cascaded sub-layers are trained as a whole, replacing the training of large dense layers with the training of lightweight sub-layers, and the calculation of large weight matrices is replaced by the calculation of three small weight matrices. This reduces the network training complexity, increases data processing speed, and improves the computational efficiency of signal recovery in the orthogonal frequency division multiplexing system.

[0053] In other aspects of the embodiments of this application, each fully connected hidden layer of the generator in the conditional generative adversarial network includes three cascaded sub-layers; the three cascaded sub-layers perform gradient updates based on the rank-constrained backpropagation method with matrix rank invariance through row representation matrix, column representation matrix and intermediate matrix; the matrix rank invariance indicates that the product of the three matrix ranks of the row representation matrix, column representation matrix and intermediate matrix of the three cascaded sub-layers is unchanged from the original matrix rank of the fully connected hidden layer.

[0054] In this embodiment, the generator of the conditional generative adversarial network divides each fully connected hidden layer into three cascaded sub-layers, and trains and updates these three cascaded sub-layers as a whole. To conform to this model structure, a rank-constrained backpropagation algorithm with matrix rank invariance is proposed for gradient update.

[0055] Please refer to Figure 3 In this embodiment of the application, a high-dimensional weight matrix is ​​used. It is divided into three low-dimensional weight matrices. These three low-dimensional matrices are column representation matrices. intermediate matrix Row representation matrix Where M l N l s l,c s l,r These represent the input dimension, output dimension, sampled column number, and row number of the l-th layer of the cGAN generator, respectively. Let s... l,r <<M l and s l,c <<N l This application trains and updates these three low-dimensional small matrices as a whole, with the overall rank limited to . Where t represents the training epoch of the neural network, i.e., the row representation matrix of the three cascaded sub-layers. Column representation matrix and intermediate matrix The product of the ranks of the three matrices and the original matrix of the fully connected hidden layer The rank remains unchanged.

[0056] The backward gradient update of the column representation matrix and the row representation matrix is ​​expressed as shown in Equation (1) and Equation (2):

[0057]

[0058]

[0059] In formulas (1) and (2), L2 is the mean squared error loss function. Based on this, the row representation matrix and column representation matrix are shown in formula (3):

[0060]

[0061] In formula (3), η is the learning step size, and t represents the number of training iterations of the neural network. The intermediate matrix Φ l Depend on and It is determined that its mathematical expression is shown in formula (4):

[0062]

[0063] in, and These are the row equivalent sampling matrix and the column equivalent sampling matrix, respectively. and Equivalent to respectively and

[0064] By reconstructing the internal structure of the generator, each dense fully connected layer is changed into three cascaded sub-layers. By replacing the training of large dense layers with the training of lightweight sub-layers, the calculation of large weight matrices is replaced with the calculation of three small weight matrices, thereby reducing the network training complexity and improving the data processing speed, and improving the computational efficiency of signal recovery in orthogonal frequency division multiplexing systems.

[0065] In one embodiment, the method further includes the step of training the conditional generative adversarial network; training the conditional generative adversarial network includes:

[0066] Step 10: Input the received sample signal and the sample pilot sequence into the generator of the conditional generative adversarial network;

[0067] Step 20: Input the output signal of the generator, the sample pilot sequence, and the real transmitted signal into the discriminator of the conditional generative adversarial network, so that the discriminator can discriminate the output signal, the sample pilot sequence, and the real transmitted signal.

[0068] Step 30: Construct a loss function, and adjust the model parameters of the generator based on the loss function and the discrimination result of the discriminator.

[0069] The sample received signal refers to the sample received signal obtained from the actual transmitted signal emitted by the orthogonal frequency division multiplexing system during the training of the conditional generative adversarial network. The sample pilot sequence is the pilot sequence used during the training of the conditional generative adversarial network.

[0070] Based on this, the steps for training the conditional generative adversarial network include the following: using the conditional generative adversarial network to learn the mapping relationship from the sample received signal Y and the sample pilot sequence Φ to the actual transmitted signal, thereby recovering the transmitted signal from the transmitter. Please refer to... Figure 4As shown, Figure 4 This diagram illustrates the structure of the conditional generative adversarial network (GAN) provided in this embodiment of the invention. During offline training, the GAN consists of two neural networks: a generator and a discriminator. The generator is responsible for recovering the transmitted signal based on the conditional input (i.e., the sample received signal Y and the sample pilot sequence Φ), and its output is the signal recovered by the trained neural network. The discriminator identifies real and fake data; its input is the real transmitted signal X and the generator's output signal. And the sample pilot sequence Φ as conditional input. If the discriminator input matches the actual transmitted signal X and the sample pilot sequence Φ, the output is marked as a true "1"; if the discriminator input is the output signal generated by the generator... If the sample pilot sequence Φ is obtained, the output is marked as a false "0". The generator's model parameters are adjusted based on the loss function and the discriminator's discrimination results. When the generator's network model converges during offline training, its weights can be saved. In online applications, the trained conditional generative adversarial network generator is used for signal detection to recover the transmitted signal from the transmitter.

[0071] In other aspects of the embodiments of this application, the loss function is constructed based on the generative adversarial network loss function and the mean squared error loss function.

[0072] Specifically, Conditional Generative Adversarial Networks (cGANs) are an extension of the original GAN ​​network. The only difference between the two is that both the generator and discriminator in the cGAN model have added conditional inputs, thereby improving the performance and efficiency of adversarial training of the neural network. The training objective of the conditional generative adversarial network is to make the generator produce an output signal that is infinitely close to the transmitted signal to deceive the discriminator. At the same time, the discriminator needs to learn not to be easily deceived. The two networks train against each other to obtain the best performance. Therefore, this scheme introduces the generative adversarial network loss function to achieve this optimization. The generative adversarial network loss function is shown in Equation (5):

[0073] L GAN (G Ψ D θ ,Y,X,Φ)=E[logD θ (X,Φ)]+E[log(1-D θ (G Ψ (Y,Φ)))],#

[0074] Formula (5)

[0075] Among them, L GAN (G Ψ D θ (Y, X, Φ) represents the loss function of the generative adversarial network; G Ψ (i.e. G) Ψ(Y,Φ)) represents the generator loss parameterized by Ψ, which aims to generate signal values ​​more similar to the true value of X; D θ Let θ represent the discriminator loss, parameterized by θ. This discriminator aims to distinguish between real transmitted signal values ​​and generated signal values ​​(if the discriminator's input comes from a real transmitted signal, it is labeled as "1" for real data; X represents the real transmitted signal; Y represents the sample received signal; Φ represents the sample pilot sequence. If the discriminator's input comes from the generator's output signal, it is labeled as "0" for fake data). To minimize the GAN loss, two neural networks are trained adversarially, i.e., minimizing the generator loss G... Ψ Maximize the discriminator loss D θ As shown in formula (6):

[0076]

[0077] To ensure that the generator generates data in the correct direction, this embodiment adds an L2 loss to the generative adversarial network loss. The L2 loss, or mean squared error (MSE) loss, represents minimizing the sum of squares of the differences between the predicted and true values, as shown in formula (7):

[0078]

[0079] Finally, the overall loss function of this application embodiment is expressed as shown in formula (8):

[0080]

[0081] This application's embodiments introduce a generative adversarial network loss function (GAN Loss) to correct certain aspects of neural network optimization, addressing the problem that loss functions in deep learning methods are not well designed to generate the necessary data, thereby enabling better signal detection and recovery of the transmitted signal from the orthogonal frequency division multiplexing system.

[0082] For other aspects of the embodiments of this application, please refer to Figure 5 Before step 300, which involves determining the signal recovery result of the received signal, the method further includes:

[0083] Step 200: Preprocess the received signal based on the least squares method and the zero-forcing detection method.

[0084] Specifically, each OFDM system simulation randomly generates a data bitstream consisting of 0s and 1s as input. Wireless communication systems typically use frame structures to represent transmitted symbols. In the simulation experiment of this invention, one physical frame consists of two OFDM symbols, and each OFDM symbol contains 64 subcarriers. Multiple preset pilot training sequences are inserted into the data stream from the time domain in a block-shaped pilot equally spaced manner.

[0085] Please refer to Figure 2 Before the signal is fed into the conditional generative adversarial network (GAN), the received signal is first roughly estimated using the least squares method (LS) and zero forcing (ZF) method. This rough estimate is then used as one of the initial input values ​​for the GAN. After preprocessing, the real and imaginary parts of the complex data stream are concatenated to obtain the initial input of the neural network. The GAN is then used to refine the coarse input for further signal detection.

[0086] This application embodiment uses a Rayleigh channel to generate training data, and the multipath channel is represented by a complex random variable. The mathematical expression for the time-domain signal at the receiving end is shown in formula (9):

[0087]

[0088] Here, Let represent the convolution operation, and x(n) and w(n) represent the transmitted signal and additive white Gaussian noise, respectively. After removing the cyclic prefix and performing a fast Fourier transform on the signal sequence, the signal at the receiver in the frequency domain can be expressed as shown in equation (10):

[0089] Y(k)=X(k)H(k)+W(k),#Formula (10)

[0090] Where Y(k), X(k), H(k), and W(k) are the frequency domain signal values ​​obtained by performing Fourier transform on the time domain signals y(n), x(n), h(n), and w(n), respectively.

[0091] Please refer to Figure 2 As shown, to obtain effective neural network training samples at the receiving end, offline training and online application are combined. Before offline training, the received signal is preprocessed using the least squares method and zero-forcing detection method. This involves effectively initializing the data before feeding it into the input of the conditional generative adversarial network (GAN) to achieve better data training performance. Specifically, this compensates for the effects of channel fading distortion, thereby improving signal detection capabilities.

[0092] The least squares method is equivalent to a linear estimator. Preprocessing the received signal using the least squares method and zero-forcing detection method results in low overall algorithm complexity at the receiver. The output of the least squares method is obtained through the least squares channel estimation method, as shown in formula (11):

[0093]

[0094] Among them, X P (k) and Y P (k) represents the pilot symbol value and pilot received signal value at the k-th subcarrier. The channel response value at the pilot is obtained. The total response value of the entire channel is then obtained using linear interpolation. That is, the frequency response value of the data subchannel is obtained from the response values ​​at the two pilot points before and after, which can be specifically expressed as follows: where N represents the number of subcarriers and P represents the number of pilot subcarriers. The zero-forcing detection method is used for channel equalization and signal compensation, and its output is shown in formula (12):

[0095]

[0096] 0≤m≤P-1,#Formula (12)

[0097] Where N represents the number of subcarriers and P represents the number of pilot subcarriers. The zero-forcing detection method performs channel equalization and signal compensation, and its output is shown in formula (13):

[0098]

[0099] The preprocessed received signal is The real and imaginary parts of the preprocessed received signal are concatenated and used as the input sample for the conditional generative adversarial network. That is, the input data of the conditional generative adversarial network is expressed as shown in formula (14):

[0100] input={Re(pilot),lm(pilot),Re(Y),lm(Y)}#Formula (14)

[0101] The embodiments of this application adopt a combination of model-driven and data-driven approaches, which not only utilize the training advantages of rapid learning in neural networks, but also use the prior knowledge of traditional communication to recover the signal from the transmitting end. This avoids the disadvantage of high complexity caused by using data-driven training entirely, and improves training efficiency and performance.

[0102] This application employs a combination of offline training and online application. Offline training comprises two stages: a data preprocessing stage using least squares and zero-forcing detection methods, and a training stage. During the preprocessing stage, both the generator and discriminator utilize the Adam optimization algorithm, incorporating quadratic gradient correction. This ensures that the training parameters of the conditional generative adversarial network (GAN) are close to their optimal values, effectively avoiding overfitting and gradient vanishing problems caused by initial parameter randomization during GAN training. This improves signal detection accuracy and accelerates the convergence speed of the GAN. After the aforementioned preprocessing and offline training stages, the GAN possesses good online detection capabilities and can be applied to OFDM systems to recover transmitted signals.

[0103] In one embodiment, please refer to Figure 6 and Figure 7 After step 300, which involves determining the signal recovery result of the received signal, the method further includes:

[0104] Step 400: Perform data uplink storage operation on the signal recovery result.

[0105] This application uses the bit error rate (BER) to represent the error between the recovered signal and the actual transmitted signal. This difference is expressed as shown in formula (15):

[0106]

[0107] Among them, y true Let y be the true value of the received signal at the receiving end. pred This represents the signal recovery result at the receiver after passing through a conditional generative adversarial network.

[0108] The receiver of the orthogonal frequency division multiplexing (OFDM) system stores the recovered signal data on the blockchain. Leveraging the immutability, forgery resistance, and traceability of blockchain technology, storing the recovered signal data on the chain ensures the accuracy of the bit error rate calculation, enabling proper evaluation of the training and learning performance of the generative adversarial network. It should be noted that... (Please refer to...) Figure 2 Alternatively, the transmitted signal can be uploaded to the blockchain at the transmitting end of the orthogonal frequency division multiplexing system to ensure the authenticity and reliability of the signal data, thereby guaranteeing the accuracy of the bit error rate performance and identifying the training performance of the neural network.

[0109] The orthogonal frequency division multiplexing (OFDM) system signal recovery device provided in the embodiments of this application is described below. The OFDM system signal recovery device described below and the OFDM system signal recovery method described above can be referred to each other.

[0110] Please refer to Figure 8A signal recovery device for an orthogonal frequency division multiplexing system, comprising:

[0111] Acquisition module 201 is used to acquire received signals; the received signals are obtained by receiving the transmitted signals of the orthogonal frequency division multiplexing system.

[0112] The signal recovery result determination module 202 is used to determine the signal recovery result of the received signal by inputting the received signal and pilot sequence into a conditional generative adversarial network to obtain the signal recovery result output by the conditional generative adversarial network.

[0113] The fully connected hidden layers of the generator in the conditional generative adversarial network include multiple cascaded sub-layers, and the gradients of these multiple cascaded sub-layers are updated during training using a rank-constrained backpropagation method with matrix rank invariance. The conditional generative adversarial network is trained based on sample received signals, real transmitted signals from an orthogonal frequency division multiplexing system, and sample pilot sequences.

[0114] By inputting the received signal and pilot sequence into a Conditional Generative Adversarial Network (CGAN), the signal recovery result output by the CGAN is obtained, enabling signal detection and recovery of the received signal in an Orthogonal Frequency Division Multiplexing (OFDM) system. The CGAN employs a rank-constrained backpropagation method based on matrix rank invariance during training, updating gradients across multiple cascaded sublayers of the generator. By training these multiple cascaded sublayers as a whole, replacing the training of large, dense layers with training of lightweight sublayers, and replacing the calculation of large weight matrices with the calculation of three small weight matrices, the network training complexity is reduced, data processing speed is increased, and the computational efficiency of signal recovery in the OFDM system is improved.

[0115] In one embodiment, each fully connected hidden layer of the generator in the conditional generative adversarial network includes three cascaded sub-layers; the three cascaded sub-layers perform gradient updates using a row representation matrix, a column representation matrix, and an intermediate matrix, based on a rank-constrained backpropagation method with matrix rank invariant.

[0116] The invariance of matrix rank indicates that the product of the ranks of the row representation matrix, column representation matrix, and intermediate matrix of the three cascaded sub-layers remains unchanged from the original matrix rank of the fully connected hidden layer.

[0117] In one embodiment, the orthogonal frequency division multiplexing system signal recovery device further includes a training module; the training module includes:

[0118] The first input module is used to input the sample received signal and the sample pilot sequence into the generator of the conditional generative adversarial network;

[0119] The second input module is used to input the output signal of the generator, the sample pilot sequence, and the real transmitted signal into the discriminator of the conditional generative adversarial network, so that the discriminator can discriminate the output signal, the sample pilot sequence, and the real transmitted signal.

[0120] The loss function construction module is used to construct a loss function and adjust the model parameters of the generator based on the loss function and the discrimination result of the discriminator.

[0121] In one embodiment, the loss function is constructed based on the generative adversarial network loss function and the mean squared error loss function.

[0122] In one embodiment, the orthogonal frequency division multiplexing system signal recovery device further includes a preprocessing module; the preprocessing module is used to preprocess the received signal based on the least squares method and the zero-forcing detection method.

[0123] In one embodiment, the orthogonal frequency division multiplexing system signal recovery device further includes: a data uplink module, used to perform data uplink storage operation on the signal recovery result.

[0124] Figure 9 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 9 As shown, the electronic device may include a processor 910, a communication interface 920, a memory 930, and a communication bus 940, wherein the processor 910, the communication interface 920, and the memory 930 communicate with each other via the communication bus 940. The processor 910 can call a computer program in the memory 930 to execute steps of a signal recovery method for an orthogonal frequency division multiplexing (OFDM) system, such as: acquiring a received signal; the received signal is obtained by receiving the transmitted signal of the OFDM system; determining the signal recovery result of the received signal; inputting the received signal and pilot sequence into a conditional generative adversarial network (GAN) to obtain the signal recovery result output by the GAN; each fully connected hidden layer of the generator in the GAN includes multiple cascaded sub-layers, and the multiple cascaded sub-layers are updated with gradients during training based on a rank-constrained backpropagation method with matrix rank invariance; the GAN is trained based on sample received signals, actual transmitted signals of the OFDM system, and sample pilot sequences.

[0125] Furthermore, the logical instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0126] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the steps of the orthogonal frequency division multiplexing system signal recovery method provided in the above embodiments, such as: acquiring a received signal; the received signal is obtained by receiving the transmitted signal of the orthogonal frequency division multiplexing system; determining the signal recovery result of the received signal; inputting the received signal and pilot sequence into a conditional generative adversarial network to obtain the signal recovery result output by the conditional generative adversarial network; each fully connected hidden layer of the generator in the conditional generative adversarial network includes multiple cascaded sub-layers, and the multiple cascaded sub-layers are updated with gradients based on the rank-constrained backpropagation method with matrix rank invariance during training; the conditional generative adversarial network is trained based on sample received signals, the actual transmitted signals of the orthogonal frequency division multiplexing system, and sample pilot sequences.

[0127] On the other hand, embodiments of this application also provide a processor-readable storage medium storing a computer program. The computer program is used to cause a processor to execute the steps of the orthogonal frequency division multiplexing (OFDM) system signal recovery method provided in the above embodiments. For example, it includes: acquiring a received signal; the received signal is obtained by receiving the transmitted signal of the OFDM system; determining the signal recovery result of the received signal; inputting the received signal and pilot sequence into a conditional generative adversarial network (GAN) to obtain the signal recovery result output by the GAN; each fully connected hidden layer of the generator in the GAN includes multiple cascaded sub-layers, and the multiple cascaded sub-layers are updated with gradients during training based on a rank-constrained backpropagation method with matrix rank invariance; the GAN is trained based on sample received signals, the actual transmitted signal of the OFDM system, and sample pilot sequences.

[0128] The processor-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0129] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0130] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for signal recovery in an orthogonal frequency division multiplexing system, characterized in that, include: Acquire and receive signals; The received signal is obtained by receiving the transmitted signal from the orthogonal frequency division multiplexing system; Determine the signal recovery result of the received signal: input the received signal and pilot sequence into a conditional generative adversarial network to obtain the signal recovery result output by the conditional generative adversarial network; The fully connected hidden layers of the generator in the conditional generative adversarial network include three cascaded sub-layers. The three cascaded sub-layers perform gradient updates based on the rank-constrained backpropagation method with matrix rank invariance through row representation matrix, column representation matrix and intermediate matrix. The matrix rank invariance means that the product of the ranks of the row representation matrix, column representation matrix and intermediate matrix of the three cascaded sub-layers is unchanged from the original matrix rank of the fully connected hidden layer. The conditional generative adversarial network is trained based on sample received signals, real transmitted signals from an orthogonal frequency division multiplexing system, and sample pilot sequences.

2. The signal recovery method for an orthogonal frequency division multiplexing system according to claim 1, characterized in that, The method further includes the step of training the conditional generative adversarial network; Training the conditional generative adversarial network includes: The sample received signal and the sample pilot sequence are input into the generator of the conditional generative adversarial network; The generator's output signal, the sample pilot sequence, and the actual transmitted signal are input into the discriminator of the conditional generative adversarial network, so that the discriminator can distinguish the output signal, the sample pilot sequence, and the actual transmitted signal. Construct a loss function, and adjust the generator's model parameters based on the loss function and the discrimination result of the discriminator.

3. The signal recovery method for an orthogonal frequency division multiplexing system according to claim 2, characterized in that, The loss function is constructed based on the generative adversarial network loss function and the mean squared error loss function.

4. The signal recovery method for an orthogonal frequency division multiplexing system according to claim 1, characterized in that, Before determining the signal recovery result of the received signal, the method further includes: The received signal is preprocessed based on the least squares method and the zero-forcing detection method.

5. The signal recovery method for an orthogonal frequency division multiplexing system according to claim 1, characterized in that, After determining the signal recovery result of the received signal, the method further includes: The signal recovery result is then stored on the blockchain.

6. A signal detection device for an orthogonal frequency division multiplexing system, characterized in that, include: The acquisition module is used to acquire received signals; The received signal is obtained by receiving the transmitted signal from the orthogonal frequency division multiplexing system; The signal recovery result determination module is used to determine the signal recovery result of the received signal: inputting the received signal and pilot sequence into a conditional generative adversarial network to obtain the signal recovery result output by the conditional generative adversarial network; The fully connected hidden layers of the generator in the conditional generative adversarial network include three cascaded sub-layers. The three cascaded sub-layers perform gradient updates based on the rank-constrained backpropagation method with matrix rank invariance through row representation matrix, column representation matrix and intermediate matrix. The matrix rank invariance means that the product of the ranks of the row representation matrix, column representation matrix and intermediate matrix of the three cascaded sub-layers is unchanged from the original matrix rank of the fully connected hidden layer. The conditional generative adversarial network is trained based on sample received signals, real transmitted signals from an orthogonal frequency division multiplexing system, and sample pilot sequences.

7. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the orthogonal frequency division multiplexing system signal recovery method according to any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the signal recovery method for an orthogonal frequency division multiplexing system as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the orthogonal frequency division multiplexing system signal recovery method according to any one of claims 1 to 5.

Images