Communication system based on combination of deep joint source channel coding and OFDM (Orthogonal Frequency Division Multiplexing)

Abstract

The invention relates to a communication system based on combination of deep joint source channel coding and OFDM (Orthogonal Frequency Division Multiplexing), belongs to the technical field of communication, and aims to improve the performance of the communication system. Serial / parallel conversion: dispersing the high-speed serial data to a plurality of subcarriers; iFFT conversion: realizing conversion from a frequency domain to a time domain; parallel / serial conversion: normalizing a time domain signal; adding a cyclic prefix to resist inter-symbol interference; d / A conversion: analog channel transmission is adapted; a / D conversion: digitally receiving the signal; removing a cyclic prefix; restoring the time domain signal; serial / parallel conversion: preparing for subsequent processing; carrying out FFT (Fast Fourier Transform); balancing: correcting channel distortion; parallel / serial conversion: recovering a serial form; and decoding: restoring the original information. According to the method, the two are effectively combined through the complete process, and the communication system performance is improved.

Description

[0001] The present invention relates to the field of communication technologies, and particularly to a communication system based on the combination of deep joint source-channel coding and OFDM. In particular, it involves reducing the peak-to-average power ratio by introducing a deep residual clipping network to improve communication performance. Background Art

[0002] In high-speed data transmission, traditional single-carrier modulation technologies are affected by multipath propagation. The delay spread is likely to exceed the symbol period, resulting in severe inter-symbol interference and reduced communication reliability. Although orthogonal frequency division multiplexing (OFDM) technology can effectively cope with multipath fading and use cyclic prefixes to eliminate inter-symbol interference, OFDM symbols are orthogonally synthesized from multiple subcarrier signals, resulting in a problem of too high peak-to-average power ratio (PAPR). When high-PAPR signals pass through nonlinear devices such as radio frequency power amplifiers, in-band distortion and out-of-band radiation are easily caused, deteriorating the system bit error rate and decreasing the spectral efficiency. Traditional methods for reducing PAPR, such as clipping, require manual parameter adjustment. Improper threshold setting will introduce serious nonlinear distortion; technologies such as selective mapping (SLM) have problems such as high complexity. At the same time, deep joint source-channel coding (DJSCC) realizes end-to-end joint coding with the help of neural networks, and can optimize the information anti-interference ability for specific tasks. However, the nonlinear mapping of DJSCC will exacerbate the PAPR problem. Therefore, a new communication system is needed that can effectively combine the advantages of OFDM and DJSCC, while solving the problem of too high PAPR and improving the performance of the communication system. Summary of the Invention

[0003] This application proposes a communication system based on the combination of deep joint source-channel coding and OFDM. By optimizing the system structure and signal processing flow, the peak-to-average power ratio is reduced, and the reliability, spectral efficiency, and anti-interference ability of the communication system are improved.

[0004] To achieve the above object, the present application provides the following solutions.

[0005] Use the CIFAR-10 neural network system for training. The CIFAR-10 dataset uses images with a size of 32×32, and the input images are divided by 255 for pixel value normalization.

[0006] Furthermore, use an Encoder encoding module composed of multiple convolutional layers, a generalized divisive normalization layer (GDN), and a PRelu layer to extract features and compress the size of the input image; then convert the high-speed serial data into low-speed parallel data through serial-to-parallel conversion, and distribute it to multiple orthogonal subcarriers for transmission; convert the frequency-domain data into the time domain through the inverse Fourier transform (IFFT); perform parallel-to-serial conversion to regularize the parallel time-domain signal into a serial signal; add a cyclic prefix to prevent inter-symbol interference; perform digital-to-analog conversion (D / A) so that the signal can be transmitted over an analog channel. Among them, the DJSCC encoder is composed of a convolutional layer, a GDN layer, and an activation function layer.

[0007] After channel transmission, it reaches the receiving module, and then the received signal is processed. First, the radio frequency signal is converted into a digital signal through analog-to-digital conversion (A / D), the cyclic prefix is removed to restore the original time-domain signal, and after serial-to-parallel conversion, the time-domain signal is converted back to the frequency domain through fast Fourier transform (FFT). Then, subcarrier demapping and pilot extraction are performed, and channel equalization (using frequency-domain or time-domain calibration algorithms) is used to compensate for channel distortion.

[0008] The data received after passing through the AWGN channel is restored and reconstructed by the Decoder module composed of a deconvolution layer, an inverse GDN layer, and a PRelu layer. Further, the mean square error between the image restored by the Decoder module and the original image is calculated, and together with the quantization error loss value, it is used for the training of the overall model.

[0009] Among them, the deep residual clipping network module (DRCN) is integrated into the OFDM system. When the complex input signal is received, it is split into real and imaginary parts and combined into a tensor. Adaptive average pooling is performed on the amplitude, and the adjustment factor α is calculated through a fully connected layer to determine the threshold. After subtracting the threshold from the amplitude, it is processed by the ReLU function, and the scaling factor ω is calculated. After multiplying the real and imaginary parts by the scaling factor and splicing, the complex signal is restored, realizing dynamic clipping to reduce PAPR. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0011] Figure 1 It is a schematic flowchart of a communication system based on the combination of deep joint source-channel coding and OFDM provided by the present application;

[0012] Figure 2 It is a block diagram of the OFDM system;

[0013] Figure 3 It is a structural diagram of the DCRN system; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0015] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] S1 Data Processing and Sending Module: Use CIFAR-10 to train the neural network system. The CIFAR-10 dataset has images of size 32×32, with 50,000 training images and 10,000 test images. Divide the input image by 255 to normalize the pixel values. Use an Encoder module composed of a convolutional layer, a Generalized Divisive Normalization (GDN) layer, and a PRelu layer to extract features and compress the size of the input image; use a DJSCC encoder to encode the input information bitstream, configure the serial-to-parallel conversion module according to the data rate and the number of subcarriers, use the IFFT algorithm to perform the conversion from the frequency domain to the time domain, and sequentially perform serial-to-parallel conversion, adding a cyclic prefix, and D / A conversion operations.

[0017] S2 Receiving Module: First, digitize the received radio frequency signal through an A / D conversion module, detect the symbol boundary according to the cyclic prefix length and signal characteristics to remove the cyclic prefix, convert the time-domain signal back to the frequency domain through an FFT module, perform subcarrier demapping and pilot extraction, use a specific channel equalization algorithm (such as the Minimum Mean Square Error Equalization algorithm) to compensate for channel distortion, and then perform serial-to-parallel conversion, decoding, and demodulation to restore the original data.

[0018] S3 DRCN Module: Build the Figure 3 DRCN module structure, including an adaptive average pooling layer, a fully connected layer, etc., initialize the network parameters to ensure that the input complex signal can be processed according to the established process to reduce the PAPR. The PAPR calculation formula is as follows:

[0019]

[0020] S4 Signal Receiving and Recovery: Receive the data after channel transmission, perform A / D conversion, remove the cyclic prefix, serial-to-parallel conversion, FFT transformation, subcarrier demapping and pilot extraction, channel equalization, serial-to-parallel conversion, decoding, and demodulation on the received data in sequence, and adjust relevant parameters according to the signal characteristics and channel state to ensure accurate restoration of the original data.

[0021] S5 Reducing Peak-to-Average Power Ratio: Input the complex signal whose PAPR needs to be reduced into the DRCN module. The module processes the signal according to steps such as splitting, pooling, calculating the adjustment factor, threshold processing, and scaling, and real-time monitors the PAPR of the output signal, and fine-tunes the network parameters according to the results.

[0022] S6 System Training and Evaluation: During the training process on the CIFAR-10 dataset, the Batchsize is set to 16. And to ensure that the model fully learns the features in the data, the Epoch is set to 1200. The Adam optimizer is used with a learning rate of 0.0001. The training process still uses a randomly generated signal-to-noise ratio within a certain range as the input, that is, the signal-to-noise ratio of the AWGN channel passed through in each simulated transmission is randomly selected within the range of [0, 20], thereby improving the generalization ability of the model. Among them, the loss function uses the weighted sum of the MSE loss and the PAPR loss. The loss function is as shown in Equation (2), where λ is set to 0.001 and η is set to 0.0001:

[0023]

[0024] Mean Squared Error

[0025] The calculation formula of MSE (Mean Squared Error) is as follows:

[0026]

[0027] In this application, all actions of obtaining signals, information, or data are carried out on the premise of complying with the corresponding data protection regulations and policies of the country where the device is located and obtaining authorization from the owner of the corresponding device.

[0028] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0029] In this article, specific examples are used to elaborate on the principle and implementation method of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A communication system based on the combination of deep joint source-channel coding and OFDM, characterized in that, It includes: a transmitting module: First, extract features from the input information through convolution operations, and then encode the extracted feature information bitstream; Convert high-speed serial data into low-speed parallel data and distribute it to multiple subcarriers; perform an inverse Fourier transform to convert frequency-domain data into time-domain data; convert the parallel time-domain signal into a serial signal; add a cyclic prefix before the time-domain signal; convert the digital signal into an analog signal for transmission, and model the channel as a series of untrainable layers and incorporate it into the neural network architecture; a receiving module: used to receive the analog signal and convert it into a digital signal; remove the cyclic prefix; Convert the serial data into parallel data; perform a fast Fourier transform to convert the time-domain signal back to the frequency domain; perform subcarrier demapping and pilot extraction on the signal; perform channel equalization to compensate for channel distortion; convert the parallel data into serial data; perform channel decoding and demodulation to recover the original information bitstream; a deep residual clipping network module (DRCN): integrated into the OFDM system, used to reduce the peak-to-average power ratio (PAPR), with the input being a complex signal, dynamically calculate the threshold through adaptive average pooling and fully connected layers, and perform a shrinking process on the signal amplitude.

2. The communication system according to claim 1, characterized in that, In the said transmitting module, DJSCC encoder is used for encoding, and QAM modulation technology is used for digital modulation.

3. The communication system according to claim 1, characterized in that, In the said receiving module, frequency-domain or time-domain calibration algorithms are used for channel equalization to compensate for the amplitude response and phase offset of each subcarrier.

4. The communication system according to claim 1, wherein When the DRCN module works, first split the input complex signal into real and imaginary parts and combine them into a specific tensor, perform an adaptive average pooling operation on the amplitude, calculate the adjustment factor α through a sequence of fully connected layers, calculate the threshold according to α, set the values less than 0 to 0 using the ReLU function after subtracting the threshold from the amplitude, calculate the scaling factor ω, and splice and restore the real and imaginary parts into a complex signal after multiplying them by the scaling factor respectively.

5. A communication method based on the combination of deep joint source-channel coding and OFDM, applied to the communication system according to any one of claims 1-4, characterized in that, It includes the following steps: Transmitting step: Encode the input information bitstream; Perform serial-to-parallel conversion to disperse the high-speed serial data to multiple subcarriers; convert the frequency-domain data into time-domain data through an inverse Fourier transform; perform parallel-to-serial conversion; Add a cyclic prefix; Perform digital-to-analog conversion and then transmit; Receiving step: Receive the radio frequency signal and perform analog-to-digital conversion; Remove the cyclic prefix; Perform serial-to-parallel conversion; convert the time-domain signal back to the frequency domain through a fast Fourier transform; perform subcarrier demapping and pilot extraction; perform channel equalization; perform parallel-to-serial conversion; perform channel decoding and demodulation to recover the original data; Reduce the peak-to-average power ratio: Use the DRCN module to process the input complex signal according to a specific process, dynamically calculate the threshold and shrink the signal amplitude to reduce the PAPR.

6. The DJSCC encoder according to claim 2, characterized in that, The DJSCC encoder consists of a series of convolutional layers, normalization layers, and parametric ReLU (PReLU) activation function layers. The convolutional layers are responsible for extracting image features, the PReLU activation function enables the model to learn non-linear mappings, and the normalization layers ensure that the output satisfies the power constraint; the DJSCC decoder consists of transposed convolutional layers, normalization layers, activation functions, and an output layer. The decoder maps the received corrupted complex signals to an estimate of the original input image. The decoder reverses the encoder operations through a series of transposed convolutional layers (with non-linear activation functions), gradually converts the corrupted image features into an estimate of the original image, and upsamples to the correct resolution. Finally, the final output is obtained through an inverse normalization layer.

Images