A visible light communication method based on precoding and adaptive dynamic compression

Abstract

The present invention discloses a visible light communication method based on precoding and adaptive dynamic compression, comprising the following steps: data layering and precoding processing; signal conversion and clipping; dynamic peak control and signal generation; signal recovery and processing at the receiving end; signal decoding and recovery; and signal separation and demodulation. The present invention adopts a hybrid ACO-OFDM and PAM-DMT modulation method at the transmitting end to layer the original data; pre-processes the layered signal through a precoder to reduce the peak-to-average power ratio of the signal, and inputs the transformed time domain signal into a generative adversarial network-based framework. The framework adaptively performs amplitude compression and dynamic clipping on the signal through a feedback mechanism of a convolutional feature extraction module and a discriminator, effectively reducing the peak-to-average power ratio. The convolutional feature extraction module at the receiving end generates a recovered signal, reducing nonlinear distortion. This significantly improves the signal recovery accuracy and the spectrum efficiency of the communication system, while ensuring the reliability and stability of the communication.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a visible light communication method based on precoding and adaptive dynamic compression. Background Art

[0002] Visible light communication (VLC) uses LEDs for data transmission, avoiding the occupation of radio frequency spectrum resources while also providing high-speed, high-capacity wireless communication services indoors. Orthogonal frequency division multiplexing (OFDM) technology has been widely adopted in VLC systems due to its high spectrum utilization and effective resistance to multipath fading and signal interference.

[0003] Optical wireless communication systems are typically based on intensity modulation / direct detection (IM / DD), which requires the transmitted signal to be real and positive. Although many O-OFDM schemes have been proposed, these schemes still have certain performance defects. Among the many schemes, HACO-OFDM provides a good trade-off between spectral efficiency, computational complexity, and transmission performance. However, the peak-to-average power ratio (PAPR) of the HACO-OFDM system is high, which leads to large fluctuations in the signal amplitude, making it easy for the LED to enter the nonlinear operating region, thereby causing power loss and increased bit error rate problems. These problems limit the performance of HACO-OFDM in practical applications, especially in scenarios with high data rates and low power consumption requirements, making it difficult to effectively improve system performance.

[0004] The Chinese patent publication number is "CN202210773144.5," entitled "A Method for Improving Hybrid HACO-OFDM Performance Using Predistortion Technology." This method, for the first time, utilizes predistortion technology to eliminate interference between ACO-OFDM and PAM-DMT at the transmitter. By performing a Fourier transform on the ACO-OFDM signal, the interference noise is extracted and compensated for on the frequency domain data of the PAM-DMT signal. This technology effectively eliminates the interference of ACO-OFDM on PAM-DMT, avoiding the complex serial interference processing required by traditional methods at the receiver, thus simplifying the processing at the receiver. While this reduces system complexity and improves bit error rate performance to a certain extent, it does not optimize for peak-to-average power ratio (PAPR), which may lead to power amplifier distortion caused by high PAPR, affecting system transmission quality. Therefore, designing a modulation technique that can effectively suppress the peak-to-average power ratio (PAPR) and improve system stability while reducing bit error rate is a key issue that the present invention urgently addresses. Summary of the Invention

[0005] The present invention solves the above-mentioned technical problem by providing a visible light communication method based on precoding and adaptive dynamic compression, comprising the following steps:

[0006] Step 1: Data stratification and precoding processing:

[0007] The signal source is layered, with the first layer of data allocated to odd-numbered subcarriers for ACO-OFDM modulation and the second layer of data allocated to even-numbered subcarriers for PAM-DMT modulation.

[0008] Create a Vandermonde-like matrix, and multiply the ACO-OFDM frequency domain signal and the PAM-DMT frequency domain signal by the Vandermonde-like matrix respectively to obtain a precoded ACO-OFDM signal and a precoded PAM-DMT signal;

[0009] Step 2: Signal transformation and trimming:

[0010] Perform Hermitian symmetric transform and inverse Fourier transform (IFFT) on the two-layer precoded signals to generate ACO-OFDM and PAM-DMT bipolar time domain signals;

[0011] Perform negative signal clipping on the bipolar time domain signal, and superimpose the clipped time domain positive signal to obtain a superimposed signal;

[0012] Step 3: Dynamic peak control and signal generation:

[0013] Perform μ-law mapping on the superimposed signal to obtain a mapped signal, and input the mapped signal into a dynamic peak control module to perform dynamic peak reduction to obtain a reduced and clipped low peak-to-average power ratio signal;

[0014] The processed signal is input into the discriminator 1 in the DPC-GAN framework. The discriminator 1 analyzes the amplitude distribution and characteristics of the signal and optimizes the signal processing process. The convolution feature extraction module generates an output signal with amplitude compression and dynamic peak reduction.

[0015] Step 4: Signal recovery and processing at the receiving end:

[0016] The received time domain signal is input into the dynamic signal recovery module in the DPC-GAN framework to recover the signal that has been cut and adjusted by the dynamic threshold;

[0017] Perform μ-law inverse mapping on the recovered signal to obtain an inverse mapping signal;

[0018] The inverse mapped signal is input to the discriminator 2 for comparative analysis to evaluate the quality and accuracy of signal recovery. The convolution feature extraction module generates the signal that has been recovered and inverse mapped;

[0019] Step 5: Signal decoding and recovery:

[0020] Perform fast Fourier transform and precoding decoding on the received signal, and use the reversibility of the Vandermonde-like precoding matrix to obtain the precoded decoded signal, which contains mixed data of ACO-OFDM and PAM-DMT;

[0021] Step 6: Signal separation and demodulation:

[0022] Directly detect and extract the ACO-OFDM signal from the received signal and restore its original data information;

[0023] The clipped noise of the ACO-OFDM signal is reconstructed and removed from the received mixed signal to obtain the complete PAM-DMT signal, achieving accurate separation and demodulation of the two signals.

[0024] Furthermore, the method of creating a Vandermonde-like matrix in step 1 includes: generating a Vandermonde-like matrix using monomials, and replacing the monomials with polynomials at nodes to create the Vandermonde-like matrix.

[0025] Furthermore, the dynamic peak control module in step 3 includes:

[0026] Determine a dynamic threshold value, and calculate the dynamic threshold value according to the maximum amplitude value of the mapping signal;

[0027] The peak characteristics of the signal are optimized. When the signal amplitude exceeds the threshold, it is reduced to within the threshold range. At the same time, the reduced part is redistributed to the zero value position of the signal to obtain the target signal.

[0028] Furthermore, the DPC-GAN framework in step 3 includes:

[0029] Convolution feature extraction module, used to extract the peak characteristics of the signal and output a generated signal with amplitude compression and dynamic peak reduction;

[0030] The discriminator 1 is used to analyze the amplitude distribution and characteristics of the generated signal and the target signal to improve the quality of the generated signal.

[0031] Furthermore, the convolutional feature extraction module includes a fully connected layer module, a convolution block module and an output layer module, and the discriminator consists of a convolution module, a flattening module and a discrimination module.

[0032] Furthermore, the dynamic signal recovery module in step 4 includes:

[0033] Dynamic threshold estimation, which recovers the corresponding threshold value according to the maximum amplitude value of the received signal;

[0034] Signal classification mechanism, which determines the signal status based on the relationship between the received signal and the dynamic threshold;

[0035] The original signal is restored and reconstructed, and the signal is linearly superimposed using a dynamic threshold according to the signal classification result to restore the original signal.

[0036] Furthermore, the discriminator 2 in step 4 is used to evaluate the quality and accuracy of signal recovery and optimize the feature extraction capability of the convolutional feature extraction module through feedback.

[0037] Furthermore, the precoding and decoding process in step 5 utilizes the invertibility of the Vandermonde-like precoding matrix to perform an inverse operation on the signal to obtain a precoding and decoding signal.

[0038] Furthermore, the signal separation and demodulation process in step 6 includes directly detecting and extracting the ACO-OFDM signal, and reconstructing and removing the clipping noise of the ACO-OFDM signal to obtain a complete PAM-DMT signal.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. The present invention provides a precoding method based on a Vandermonde-like matrix and Chebyshev node selection. By reducing the autocorrelation of matrix elements, it effectively reduces inter-symbol interference and suppresses the increase of peak-to-average power ratio (PARA). It solves the problem that high peak-to-average power ratio and inter-symbol interference affect transmission performance in optical communication systems and improves system stability.

[0041] 2. The present invention provides a dynamic compression framework based on DPC-GAN adaptation. First, the signal is compressed by the μ-law mapping module, and the part exceeding the threshold is cut in the dynamic peak control module; then, based on the feedback results of the discriminator 1, the convolutional feature extraction module extracts signal features and generates adaptive compression and clipping signals, effectively solving the problems of signal distortion and excessive peak in traditional visible light communication.

[0042] 3. This paper provides a signal recovery framework based on DPC-GAN. First, the received signal is recovered using a dynamic recovery module. Then, a μ-law inverse mapping module is used to generate an inverse mapping signal. Finally, the convolutional feature extraction module is optimized based on the feedback from Discriminator 2 to generate a highly accurate recovered signal. This method significantly improves the accuracy and quality of signal recovery while enhancing the stability of signal transmission, ensuring signal integrity and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0044] Figure 1 This is a flowchart of the steps of the visible light communication method based on precoding and adaptive dynamic compression according to the present invention;

[0045] Figure 2 This is a structural diagram of a visible light communication method based on precoding and adaptive dynamic compression according to the present invention;

[0046] Figure 3 Flowchart of the convolutional feature extraction module of the DPC-GAN framework of the present invention;

[0047] Figure 4 Flowchart of the discriminator of the DPC-GAN framework of the present invention. DETAILED DESCRIPTION

[0048] The present invention proposes a visible light communication method based on precoding and adaptive dynamic compression, aiming to improve the signal recovery accuracy and the spectrum efficiency of the communication system while ensuring the reliability and stability of communication.

[0049] The visible light communication method based on precoding and adaptive dynamic compression proposed by the present invention will be described below in a specific embodiment:

[0050] Example 1:

[0051] A visible light communication method based on precoding and adaptive dynamic compression, such as Figure 1 As shown, the following steps are included:

[0052] Step 1: Data stratification and precoding processing:

[0053] The signal source is layered, with the first layer of data allocated to odd-numbered subcarriers for ACO-OFDM modulation and the second layer of data allocated to even-numbered subcarriers for PAM-DMT modulation.

[0054] Create a Vandermonde-like matrix, and multiply the ACO-OFDM frequency domain signal and the PAM-DMT frequency domain signal by the Vandermonde-like matrix respectively to obtain a precoded ACO-OFDM signal and a precoded PAM-DMT signal;

[0055] Step 2: Signal transformation and trimming:

[0056] Perform Hermitian symmetric transform and inverse Fourier transform (IFFT) on the two-layer precoded signals to generate ACO-OFDM and PAM-DMT bipolar time domain signals;

[0057] Perform negative signal clipping on the bipolar time domain signal, and superimpose the clipped time domain positive signal to obtain a superimposed signal;

[0058] Step 3: Dynamic peak control and signal generation:

[0059] Perform μ-law mapping on the superimposed signal to obtain a mapped signal, and input the mapped signal into a dynamic peak control module to perform dynamic peak reduction to obtain a reduced and clipped low peak-to-average power ratio signal;

[0060] The processed signal is input into the discriminator 1 in the DPC-GAN framework. The discriminator 1 analyzes the amplitude distribution and characteristics of the signal and optimizes the signal processing process. The convolution feature extraction module generates an output signal with amplitude compression and dynamic peak reduction.

[0061] Step 4: Signal recovery and processing at the receiving end:

[0062] The received time domain signal is input into the dynamic signal recovery module in the DPC-GAN framework to recover the signal that has been cut and adjusted by the dynamic threshold;

[0063] Perform μ-law inverse mapping on the recovered signal to obtain an inverse mapping signal;

[0064] The inverse mapped signal is input to the discriminator 2 for comparative analysis to evaluate the quality and accuracy of signal recovery. The convolution feature extraction module generates the signal that has been recovered and inverse mapped;

[0065] Step 5: Signal decoding and recovery:

[0066] Perform fast Fourier transform and precoding decoding on the received signal, and use the reversibility of the Vandermonde-like precoding matrix to obtain the precoded decoded signal, which contains mixed data of ACO-OFDM and PAM-DMT;

[0067] Step 6: Signal separation and demodulation:

[0068] Directly detect and extract the ACO-OFDM signal from the received signal and restore its original data information;

[0069] The clipped noise of the ACO-OFDM signal is reconstructed and removed from the received mixed signal to obtain the complete PAM-DMT signal, achieving accurate separation and demodulation of the two signals.

[0070] Furthermore, the method of creating a Vandermonde-like matrix in step 1 includes: generating a Vandermonde-like matrix using monomials, and replacing the monomials with polynomials at nodes to create the Vandermonde-like matrix.

[0071] Furthermore, the dynamic peak control module in step 3 includes:

[0072] Determine a dynamic threshold value, and calculate the dynamic threshold value according to the maximum amplitude value of the mapping signal;

[0073] The peak characteristics of the signal are optimized. When the signal amplitude exceeds the threshold, it is reduced to within the threshold range. At the same time, the reduced part is redistributed to the zero value position of the signal to obtain the target signal.

[0074] Furthermore, the DPC-GAN framework in step 3 includes:

[0075] Convolution feature extraction module, used to extract the peak characteristics of the signal and output a generated signal with amplitude compression and dynamic peak reduction;

[0076] The discriminator 1 is used to analyze the amplitude distribution and characteristics of the generated signal and the target signal to improve the quality of the generated signal.

[0077] Furthermore, the convolutional feature extraction module includes a fully connected layer module, a convolution block module and an output layer module, and the discriminator consists of a convolution module, a flattening module and a discrimination module.

[0078] Furthermore, the dynamic signal recovery module in step 4 includes:

[0079] Dynamic threshold estimation, which recovers the corresponding threshold value according to the maximum amplitude value of the received signal;

[0080] Signal classification mechanism, which determines the signal status based on the relationship between the received signal and the dynamic threshold;

[0081] The original signal is restored and reconstructed, and the signal is linearly superimposed using a dynamic threshold according to the signal classification result to restore the original signal.

[0082] Furthermore, the discriminator 2 in step 4 is used to evaluate the quality and accuracy of signal recovery and optimize the feature extraction capability of the convolutional feature extraction module through feedback.

[0083] Furthermore, the precoding and decoding process in step 5 utilizes the invertibility of the Vandermonde-like precoding matrix to perform an inverse operation on the signal to obtain a precoding and decoding signal.

[0084] Furthermore, the signal separation and demodulation process in step 6 includes directly detecting and extracting the ACO-OFDM signal, and reconstructing and removing the clipping noise of the ACO-OFDM signal to obtain a complete PAM-DMT signal.

[0085] Example 2:

[0086] A visible light communication method based on precoding and adaptive dynamic compression, such as Figure 2 The method comprises the following steps:

[0087] Step 1: Data stratification and precoding processing:

[0088] First, the signal source is layered. Then, the first layer of data is assigned to odd subcarrier positions and ACO-OFDM modulation is performed. The second layer of data is assigned to even subcarrier positions and PAM-DMT modulation is performed on the signal. Finally, the two layers of signals are input into the precoding module respectively to obtain two layers of precoded signals.

[0089] The specific steps include:

[0090] First, create a Vandermonde-like matrix using the monomial (1, x, x 2 ,…,x n ) generates the Vandermonde matrix at the node Replace the monomial with a polynomial (p0(x), p1(x), ... p n (x)) to create a Vandermonde matrix. The expression of the Vandermonde matrix with m rows and n columns is as follows:

[0091]

[0092] Finally, the ACO-OFDM frequency domain signal and the PAM-DMT frequency domain signal are multiplied by the Vandermonde matrix X pre =X VLM =P(m,n)X, the precoded ACO-OFDM signal is represented by X pre , precoded PAM-DMT is represented by Y pre .

[0093] Step 2: Signal transformation and trimming:

[0094] The two-layer precoded signals are subjected to Hermitian symmetric transform and inverse Fourier transform (IFFT) respectively to generate ACO-OFDM and PAM-DMT bipolar time domain signals. Then, the bipolar time domain signals are subjected to negative signal clipping and the clipped time domain positive signals are superimposed. Finally, the superimposed signals are subjected to μ-law mapping to obtain the mapped signal.

[0095] The specific steps include:

[0096] Firstly, the precoded ACO-OFDM frequency domain signal and the precoded PAM-DMT frequency domain signal are Hermitian symmetric to obtain the ACO-OFDM frequency domain signal and the PAM-DMT frequency domain signal respectively.

[0097] Secondly, the two-layer frequency domain signals are subjected to inverse Fourier transform (IFFT) to obtain the ACO-OFDM time domain bipolar signal x pre,n and PAM-DMT time domain bipolar signal pre,n , the time domain symmetry of ACO-OFDM and PAM-DMT signals is as follows:

[0098] x pre,n =-x pre,n+N / 2 ,n=0,1,…,N / 2-1;

[0099] y pre,n =-y pre,N-n ,n=0,1,…,N / 2-1;

[0100] In order to facilitate the subsequent design of the present invention, combined with the symmetry of the signal, the index variable k is used to represent the signal side. Assuming it is the mth symbol, the ACO-OFDM signal is represented by x m,k , using side information as the first side of the signal within the range of n, denoted as x m,1 ; The second side using side information as a signal in the range of n+N / 2 is represented as x m,2 ; PAM-DMT signal is represented by y m,k , using side information for the first side of the signal within n, denoted as y m,1 ; Use side information for the second side of the signal within Nn, denoted as y m,2 .

[0101] Finally, the time domain positive signal is obtained by clipping the negative signal, and the clipped time domain positive signal is superimposed to obtain the superimposed time domain signal x haco .

[0102] Step 3: Dynamic peak control and signal generation:

[0103] The time domain signal is input into the DPC-GAN framework designed by the present invention, and after processing, an output signal with amplitude compression and dynamic peak reduction is generated. The time domain signal is input into the μ-law mapping module to obtain a mapped signal; then, the mapped signal is input into the dynamic peak control module to obtain a reduced and clipped low peak-to-average power ratio signal, which is then input into the discriminator 1. Specifically, the steps include:

[0104] (1) Nonlinear compression and dynamic peak reduction process:

[0105] The input time domain signal first passes through the μ-law mapping module to perform nonlinear compression on the signal amplitude, reduce the dynamic range, and thus reduce the peak-to-average power ratio (PARA) of the signal, and obtain the signal x μhaco As shown below:

[0106]

[0107] The mapped signal x μhaco The signal is input to the dynamic peak control module of the present invention, and the peak characteristics of the signal are optimized by the module. The specific steps include:

[0108] Determine the dynamic threshold γ m According to the maximum amplitude value x of the mapping signal max,m Calculate a dynamic threshold, dynamic threshold γ m The expression is as follows:

[0109] γ m =(1-α)·x max,m ;

[0110] Among them, α is the weight parameter, which controls the influence weight of the maximum amplitude value, and its value range is α∈[0,1]. amax,m is the maximum amplitude value of the mth signal.

[0111] The input signal amplitude is dynamically adjusted to ensure it does not exceed a preset threshold. When the signal amplitude is less than or equal to the threshold, the signal is output directly. When the signal amplitude exceeds the threshold, it is clipped to within the threshold range and the clipped portion is redistributed to the signal's zero position, thus limiting the signal amplitude while maintaining an overall balance of signal energy.

[0112] (2) Signal feature extraction process:

[0113] The original time domain signal is input into the convolution feature extraction module to extract the peak characteristics of the signal and generate a signal that has undergone dynamic compression and reduction processing to ensure that it meets the quality requirements; then, the generated signal and the target signal after nonlinear compression and dynamic peak reduction processing are input into the discriminator 1 for comparative analysis. The discriminator 1 analyzes the amplitude distribution and characteristics of the two, evaluates the amplitude and characteristics of the generated signal, and thus optimizes the signal processing process and improves the generation effect.

[0114] like Figure 3As shown in the figure, all convolutional feature extraction modules are composed of a fully connected layer module, a convolution block module, and an output layer module. The fully connected layer module contains a fully connected layer, which is used to convert the features of the input signal into a high-dimensional representation. The convolution block module consists of five convolution blocks, each of which contains upsampling, a 1D convolution layer, batch normalization, and a ReLU activation function. Local features of the signal are extracted through upsampling and downsampling operations. Batch normalization improves training stability, and the ReLU activation function increases the nonlinear ability of the network and helps extract complex features. The output layer module consists of a 1D convolution layer and a ReLU activation function, which is used to further process the extracted features to ensure efficient and accurate representation of the signal.

[0115] like Figure 4 As shown in the figure, all discriminators consist of a convolution module, a flattening module, and a discrimination module. The convolution module contains six convolution blocks, each of which contains a 1D convolution layer, a batch normalization layer, and a LeakyReLU activation function. This design can effectively extract the time domain features of the signal and enhance the nonlinear expression ability of the model through nonlinear activation functions. After the output of the convolution module is converted into a one-dimensional vector through the flattening module, it is passed to the discrimination module. The discrimination module consists of two fully connected layers, where the first layer uses the LeakyReLU activation function for feature processing, and the second layer uses the Sigmoid activation function to output the signal discrimination result, indicating the similarity between the generated signal and the target signal. This structure not only improves the ability to extract signal features, but also enhances the discriminator's signal discrimination accuracy.

[0116] (3) Feedback optimization process:

[0117] The discriminator's feedback is used to optimize the convolutional feature extraction function, thereby improving the quality of signal generation. First, the discriminator 1 calculates its discrimination loss based on the input generated signal and the true signal, evaluating the difference between the generated signal and the true signal. The discriminator continuously improves its discrimination ability by maximizing the probability of distinguishing the true signal and minimizing the probability of distinguishing the generated signal. The discriminator's feedback is passed to the generator through a backpropagation mechanism. The generator updates the parameters of its convolutional feature extraction module based on this feedback. The goal of the convolutional feature extraction module is to maximize the probability that the generated signal will be identified as the target signal by the discriminator, prompting the convolutional feature extraction module to produce an output that is closer to the true signal.

[0118] The loss function of the convolutional feature extraction module is composed of the mean square error loss function between the generated signal and the true signal, which is defined as follows:

[0119]

[0120] in, To generate a signal, D is the real signal, n is the number of signal samples, for each sample is the signal generated by the convolutional feature extraction module, and d i is the target signal.

[0121] The loss function of the discriminator is composed of a weighted combination of the class loss and the similarity loss of the generated signal. The class loss is directly calculated by cross entropy, while the similarity loss is calculated by the similarity between the generated signal and the original signal. The loss function is defined as follows:

[0122]

[0123] Among them, λ1,λ2 are weight factors, L ce (p ic ) is the cross entropy loss, which is calculated as follows:

[0124]

[0125] Among them, p ic The predicted probability that observation sample i belongs to category c, Structural similarity loss, calculated as follows:

[0126]

[0127] Step 4: Signal recovery and processing at the receiving end:

[0128] At the receiving end, the received signal is input into the DPC-GAN framework designed by the present invention, and the output is restored to the signal that has been dynamically thresholded and adjusted at the sending end.

[0129] The specific steps include:

[0130] (1) Dynamic recovery and inverse mapping process:

[0131] The received time domain signal is input into the dynamic signal recovery module, which performs reverse recovery based on the dynamic threshold reduction and zero-value signal relocation operations performed by the transmitter, reconstructs the peak characteristics of the signal, and obtains a reconstructed signal. The specific steps of the dynamic signal recovery module of the present invention are as follows:

[0132] Dynamic threshold estimation: Since the transmitter sets different dynamic thresholds γ for the signal m The receiving end needs to restore the corresponding threshold value based on the signal received by the photoelectric intensity modulator By detecting the maximum amplitude value in the received signal, the receiver can infer the threshold used by the sender. and the threshold γ at the sender m near.

[0133] Signal classification mechanism: Based on the received signal and dynamic threshold The relationship between the two signals is used to determine the signal status and classify the received signal. The signal classification mechanism steps are as follows: Judgment of uncut signals: If the amplitude of the signal is less than or equal to the threshold, it means that the symbol has not been cut, and the receiving end directly uses the received signal.

[0134] Determining the signal being clipped: If the signal amplitude is greater than the threshold, it indicates that the signal has been clipped and adjusted, and the receiver needs to restore the signal to its original state. By comparing the received signal with the threshold, it can be determined which side is clipped.

[0135] Restore the original signal and reconstruct it: The receiver determines the type of signal through the signal classification mechanism, and uses the dynamic threshold to linearly superimpose the signal based on the classification result to restore the original signal. Then, the reconstructed signal z is obtained by reconstructing the restored signal with the information in the zero value area. n .

[0136] For the reconstructed signal z n Perform μ-law inverse mapping to restore the compressed signal to the approximate value of the original signal and obtain the inverse mapping signal. The inverse mapping signal expression is z μn As shown below:

[0137]

[0138] (2) Signal feature extraction process:

[0139] The time domain signal at the receiving end is first input into the convolution feature extraction module, which is responsible for extracting the time domain amplitude characteristics of the signal and generating a generated signal after dynamic recovery and inverse mapping processing; then, the generated signal is input into the discriminator 2 for comparison and analysis with the inverse mapping signal obtained through dynamic recovery and inverse mapping processing; discriminator 2 analyzes the characteristics and amplitude distribution of the two to evaluate the quality and accuracy of signal recovery.

[0140] (3) Discriminator feedback optimization process:

[0141] The feedback result of discriminator 2 is used to optimize the feature extraction capability of the convolution feature extraction module, thereby improving the quality of signal generation; first, discriminator 2 receives the generated signal from the convolution feature extraction module, and compares and analyzes it with the inverse mapping signal after dynamic recovery and inverse mapping, and evaluates the differences in amplitude distribution and characteristics between the two; based on the results of the comparative analysis, discriminator 2 inputs the feedback signal into the convolution feature extraction module through the back propagation mechanism, prompting it to adjust the feature extraction process to generate an output that is closer to the inverse mapping signal; with the iterative update of the feedback, the convolution feature extraction module gradually improves its generation capability, and finally outputs the restored signal after dynamic recovery and inverse mapping.

[0142] Step 5: Signal decoding and recovery:

[0143] The signal is subjected to fast Fourier transform and precoding decoding, wherein the Vandermonde precoding matrix (VLM) is reversible. The precoding decoding signal can be obtained by performing inverse operation on the signal. The precoding decoding signal contains mixed data of ACO-OFDM and PAM-DMT. n .

[0144] Step 6: Signal separation and demodulation:

[0145] When receiving the signal r n In the first step, the ACO-OFDM signal is directly detected and extracted to restore its original data information x aco-ofdm Then, the clipping noise of the ACO-OFDM signal is reconstructed and removed from the received mixed signal to eliminate the interference to the signal, and finally the complete PAM-DMT signal y is obtained. pam-dmt , to achieve accurate separation and demodulation of the two signals.

[0146] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A visible light communication method based on precoding and adaptive dynamic compression, characterized in that: The following steps are involved: Step 1: Data stratification and precoding processing: The signal source is layered, with the first layer of data allocated to odd-numbered subcarriers for ACO-OFDM modulation and the second layer of data allocated to even-numbered subcarriers for PAM-DMT modulation. Create a Vandermonde-like matrix, and multiply the ACO-OFDM frequency domain signal and the PAM-DMT frequency domain signal by the Vandermonde-like matrix respectively to obtain a precoded ACO-OFDM signal and a precoded PAM-DMT signal; Step 2: Signal transformation and trimming: Perform Hermitian symmetric transform and inverse Fourier transform (IFFT) on the two-layer precoded signals to generate ACO-OFDM and PAM-DMT bipolar time domain signals; Perform negative signal clipping on the bipolar time domain signal, and superimpose the clipped time domain positive signal to obtain a superimposed signal; Step 3: Dynamic peak control and signal generation: Perform μ-law mapping on the superimposed signal to obtain a mapped signal, and input the mapped signal into a dynamic peak control module to perform dynamic peak reduction to obtain a reduced and clipped low peak-to-average power ratio signal; The processed signal is input into the discriminator 1 in the DPC-GAN framework. The discriminator 1 analyzes the amplitude distribution and characteristics of the signal and optimizes the signal processing process. The convolution feature extraction module generates an output signal with amplitude compression and dynamic peak reduction. Step 4: Signal recovery and processing at the receiving end: The received time domain signal is input into the dynamic signal recovery module in the DPC-GAN framework to recover the signal that has been cut and adjusted by the dynamic threshold; Perform μ-law inverse mapping on the recovered signal to obtain an inverse mapping signal; The inverse mapped signal is input to the discriminator 2 for comparative analysis to evaluate the quality and accuracy of signal recovery. The convolution feature extraction module generates the signal that has been recovered and inverse mapped; Step 5: Signal decoding and recovery: Perform fast Fourier transform and precoding decoding on the received signal, and use the reversibility of the Vandermonde-like precoding matrix to obtain the precoded decoded signal, which contains mixed data of ACO-OFDM and PAM-DMT; Step 6: Signal separation and demodulation: Directly detect and extract the ACO-OFDM signal from the received signal and restore its original data information; The clipped noise of the ACO-OFDM signal is reconstructed and removed from the received mixed signal to obtain the complete PAM-DMT signal, achieving accurate separation and demodulation of the two signals.

2. The visible light communication method based on precoding and adaptive dynamic compression according to claim 1, characterized in that: The method of creating a Vandermonde-like matrix in step 1 includes: generating a Vandermonde-like matrix using monomials, and replacing the monomials with polynomials at nodes to create the Vandermonde-like matrix.

3. The visible light communication method based on precoding and adaptive dynamic compression according to claim 1, characterized in that: The dynamic peak control module in step 3 includes: Determine a dynamic threshold value, and calculate the dynamic threshold value according to the maximum amplitude value of the mapping signal; The peak characteristics of the signal are optimized. When the signal amplitude exceeds the threshold, it is reduced to within the threshold range. At the same time, the reduced part is redistributed to the zero value position of the signal to obtain the target signal.

4. The visible light communication method based on precoding and adaptive dynamic compression according to claim 1, characterized in that: The DPC-GAN framework in step 3 includes: Convolution feature extraction module, used to extract the peak characteristics of the signal and output a generated signal with amplitude compression and dynamic peak reduction; The discriminator 1 is used to analyze the amplitude distribution and characteristics of the generated signal and the target signal to improve the quality of the generated signal.

5. The visible light communication method based on precoding and adaptive dynamic compression according to claim 4, characterized in that: The convolutional feature extraction module includes a fully connected layer module, a convolution block module and an output layer module, and the discriminator consists of a convolution module, a flattening module and a discrimination module.

6. The visible light communication method based on precoding and adaptive dynamic compression according to claim 1, characterized in that: The dynamic signal recovery module in step 4 includes: Dynamic threshold estimation, which recovers the corresponding threshold value according to the maximum amplitude value of the received signal; Signal classification mechanism, which determines the signal status based on the relationship between the received signal and the dynamic threshold; The original signal is restored and reconstructed, and the signal is linearly superimposed using a dynamic threshold according to the signal classification result to restore the original signal.

7. The visible light communication method based on precoding and adaptive dynamic compression according to claim 1, characterized in that: The discriminator 2 in step 4 is used to evaluate the quality and accuracy of signal recovery and optimize the feature extraction ability of the convolutional feature extraction module through feedback.

8. The visible light communication method based on precoding and adaptive dynamic compression according to claim 1, characterized in that: The precoding and decoding process in step 5 utilizes the invertibility of the Vandermonde-like precoding matrix to perform an inverse operation on the signal to obtain a precoding and decoding signal.

9. The visible light communication method based on precoding and adaptive dynamic compression according to claim 1, characterized in that: The signal separation and demodulation process in step 6 includes directly detecting and extracting the ACO-OFDM signal, and reconstructing and removing the clipping noise of the ACO-OFDM signal to obtain the complete PAM-DMT signal.

Images