Signal balanced transmission method and device for high-speed optoelectronic interconnection

Abstract

The present application relates to the technical field of signal balanced transmission, and discloses a signal balanced transmission method and device for high-speed optoelectronic interconnection, the method comprising: collecting bandwidth characteristics and nonlinear characteristics of an on-chip optoelectronic interconnection channel to obtain a channel characteristic data set; measuring the scattering parameters of the on-chip optoelectronic interconnection channel according to the channel characteristic data set, and performing S21 pre-compensation processing to obtain a pre-compensated signal; performing on-chip multi-channel parallel transmission of the pre-compensated signal, and collecting multi-channel transmission state information; inputting the pre-compensated signal and the multi-channel transmission state information into a two-level signal equalization model for linear and nonlinear equalization processing to obtain a first equalized signal; performing a mixed compensation operation of time domain equalization and frequency domain equalization on the first equalized signal to obtain a second equalized signal. The present application effectively suppresses the influence of channel crosstalk and dispersion distortion through real-time monitoring and dynamic compensation, and ensures the stability of signal transmission.

Description

Technical Field

[0001] The present application relates to the technical field of signal balanced transmission, and in particular to a signal balanced transmission method and device for high-speed optoelectronic interconnection. Background Art

[0002] High-speed optoelectronic interconnection systems play an increasingly important role in chip-level communications. On-chip high-speed optoelectronic interconnection channels have the advantages of high bandwidth, low power consumption, and cost-effectiveness, but they face multiple technical challenges in practical applications.

[0003] During high-speed transmission, the signal transmission quality is severely impaired due to the limited bandwidth and nonlinear characteristics of on-chip optoelectronic interconnection channels. Especially in high-speed transmission scenarios, problems such as linear distortion caused by on-chip parasitic effects, signal distortion caused by temperature drift, crosstalk between channels, and dispersion effects are more prominent. These problems will lead to serious signal distortion and inter-symbol interference, significantly reducing the transmission performance of the system. Traditional signal equalization methods often adopt a single equalization strategy, which is difficult to effectively deal with complex signal distortion problems. At the same time, existing equalization technologies lack comprehensive consideration of time domain and frequency domain characteristics, and the equalization effect is limited. Summary of the invention

[0004] The present application provides a signal balanced transmission method and device for high-speed optoelectronic interconnection. The present application effectively suppresses the influence of channel crosstalk and dispersion distortion through real-time monitoring and dynamic compensation, thereby ensuring the stability of signal transmission.

[0005] In a first aspect, the present application provides a signal balanced transmission method for high-speed optoelectronic interconnection, the signal balanced transmission method for high-speed optoelectronic interconnection comprising:

[0006] Collect bandwidth characteristics and nonlinear characteristics of on-chip optoelectronic interconnection channels to obtain a channel characteristic data set;

[0007] Measuring scattering parameters of an on-chip optoelectronic interconnection channel according to the channel characteristic data set, and performing S21 pre-compensation processing through a pre-compensation filter to obtain a pre-compensated signal;

[0008] Performing on-chip multi-channel parallel transmission on the pre-compensated signal, and collecting channel crosstalk and dispersion distortion data to obtain multi-channel transmission state information;

[0009] Inputting the pre-compensation signal and the multi-channel transmission state information into a dual-level signal equalization model for linear and nonlinear equalization processing to obtain a first equalized signal;

[0010] Based on a penalty function algorithm and an iterative optimization algorithm, a mixed compensation operation of time domain equalization and frequency domain equalization is performed on the first equalized signal to obtain a second equalized signal.

[0011] A second aspect of the present application provides a high-speed photoelectric interconnection signal balanced transmission device, the high-speed photoelectric interconnection signal balanced transmission device comprising:

[0012] An acquisition module is used to acquire bandwidth characteristics and nonlinear characteristics of an on-chip optoelectronic interconnection channel to obtain a channel characteristic data set;

[0013] A pre-compensation module, used to measure the scattering parameters of the on-chip optoelectronic interconnection channel according to the channel characteristic data set, and perform S21 pre-compensation processing through a pre-compensation filter to obtain a pre-compensated signal;

[0014] A parallel transmission module, used for performing on-chip multi-channel parallel transmission of the pre-compensated signal, collecting channel crosstalk and dispersion distortion data, and obtaining multi-channel transmission status information;

[0015] An equalization processing module, used for inputting the pre-compensation signal and the multi-channel transmission state information into a dual-level signal equalization model for linear and nonlinear equalization processing to obtain a first equalized signal;

[0016] The hybrid compensation module is used to perform a hybrid compensation operation of time domain equalization and frequency domain equalization on the first equalized signal based on a penalty function algorithm and an iterative optimization algorithm to obtain a second equalized signal.

[0017] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned high-speed optoelectronic interconnection signal equalization transmission method.

[0018] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned signal balanced transmission method for high-speed optoelectronic interconnection.

[0019] Compared with the prior art, the present application has the following beneficial effects: by establishing a two-level signal equalization model, combining the least squares algorithm and the deep neural network, targeted compensation is performed for linear distortion and nonlinear distortion respectively, thereby significantly improving the accuracy of signal equalization; a linkage optimization mechanism based on a penalty function is adopted to realize the coordinated compensation of time domain equalization and frequency domain equalization, effectively solving the problem of unsatisfactory compensation effect of the traditional single equalization method; a multi-channel parallel transmission control strategy is introduced to effectively suppress the influence of channel crosstalk and dispersion distortion through real-time monitoring and dynamic compensation; a complete signal equalization transmission link is constructed by combining S21 pre-compensation and post-equalization processing, achieving a significant expansion of the system bandwidth, and an iterative optimization algorithm is used to adaptively adjust the equalization parameters, thereby improving the system's robustness to environmental changes and ensuring the stability of signal transmission; a mixed penalty form and a soft threshold constraint strategy are adopted to effectively control the parameter divergence problem in the optimization process and ensure the convergence of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 these drawings without paying creative labor.

[0021] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0022] Figure 1 It is a schematic diagram of the flow of a signal balanced transmission method for high-speed optoelectronic interconnection provided by an embodiment of the present invention;

[0023] Figure 2 is a schematic block diagram of the structure of a signal equalization transmission device for high-speed optoelectronic interconnection provided by an embodiment of the present invention;

[0024] Figure 3 It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0027] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0028] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 , an embodiment of the signal balanced transmission method of high-speed optoelectronic interconnection in the embodiment of the present application includes:

[0029] Step 100, collecting bandwidth characteristics and nonlinear characteristics of an on-chip optoelectronic interconnection channel to obtain a channel characteristic data set;

[0030] It is understandable that the execution subject of the present application may be a signal equalization transmission device for high-speed optoelectronic interconnection, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0031] Specifically, the S21 scattering parameter scanning measurement is performed on the on-chip optoelectronic interconnection channel within a preset frequency range, and the frequency response data is obtained by analyzing the amplitude and phase relationship between the channel input and output. By processing the frequency response data, the bandwidth limitation characteristics of the channel are identified, including the high-frequency attenuation degree and the signal amplitude change characteristics, so as to extract the bandwidth-limited characteristic parameters. These parameters directly reflect the transmission capability of the channel at different frequencies. At the same time, in order to fully obtain the nonlinear characteristics of the channel, the bias conditions of the optoelectronic transceiver in the on-chip optoelectronic interconnection channel are scanned. By adjusting the bias voltage or current of the optoelectronic transceiver, the corresponding relationship between the input signal and the output signal is gradually recorded to obtain the transmission characteristic curve. The transmission characteristic curve presents certain nonlinear characteristics, such as gain compression, amplitude saturation or other nonlinear distortion phenomena. Through mathematical modeling or fitting analysis, the nonlinear characteristics are quantitatively described, and the nonlinear distortion characteristic parameters are extracted to reflect the nonlinear performance of the optoelectronic device under actual working conditions. In order to capture the dynamic distortion characteristics of the channel, the on-chip optoelectronic interconnection channel is tested and transmitted based on the high-speed data pattern. In this process, a series of known data patterns are input, the signal waveform data obtained after passing through the channel is collected, and the collected signal waveform is analyzed to reveal the distortion characteristics. By analyzing the eye opening, jitter characteristics and inter-signal interference phenomenon of the waveform, the signal distortion characteristics caused by inter-symbol interference are identified. The distortion characteristic data is quantitatively processed and the inter-symbol interference analysis operation is performed to obtain the inter-symbol interference characteristic parameters, which describe the signal degradation caused by the pattern dependency effect. The bandwidth-limited characteristic parameters, nonlinear distortion characteristic parameters and inter-symbol interference characteristic parameters are feature fused to obtain the channel characteristic feature vector. Using data fusion technology, such as weight allocation, normalization processing or feature dimension reduction, different types of parameters are mapped to a multidimensional vector space to form a channel characteristic feature vector with consistency and expressiveness. The channel characteristic feature vector is subjected to data standardization and structured processing. The purpose of standardization is to eliminate the dimensional differences of different feature parameters and normalize the range of data for easy comparison and analysis; structured processing is to organize the feature vector into a data set format with a logical hierarchy, such as a matrix or tensor structure, to facilitate subsequent modeling and optimization. Generate a channel characteristic data set.

[0032] Step 200, measuring the scattering parameters of the on-chip optoelectronic interconnection channel according to the channel characteristic data set, and performing S21 pre-compensation processing through a pre-compensation filter to obtain a pre-compensated signal;

[0033] Specifically, the S21 scattering parameters of the on-chip optoelectronic interconnection channel are calculated according to the frequency response data in the channel characteristic data set. By analyzing the amplitude and phase changes of the input and output signals, the frequency response function of the channel is obtained, which reflects the behavior of the signal changing with frequency in the frequency domain. The frequency response function is Fourier transformed to convert it into a channel transfer function in the time domain, which describes the time domain characteristics of the channel, including the impulse response and the time delay of signal propagation. Based on the channel transfer function, a pre-compensation filter is designed. The core of the pre-compensation filter is to enhance the high-frequency components in the signal to compensate for the high-frequency attenuation caused by bandwidth limitation in the channel. To achieve this goal, a pre-compensation transfer function matrix is ​​constructed. This matrix is ​​a mathematical description of the filter, indicating the proportion of different frequency components that will be amplified after passing through the filter. By optimizing the design of the transfer function matrix, the specific parameters of the pre-compensation filter, such as the gain, cutoff frequency and phase response of the filter, are determined. Based on the designed filter parameters, the power fading is compensated and calculated to generate a compensation coefficient matrix. The compensation coefficient matrix is ​​a high-dimensional array, each element of which corresponds to a gain adjustment factor at a specific frequency, which is used to correct the power loss of the original signal at different frequencies. The original digital signal is pre-compensated and filtered according to the compensation coefficient matrix. Each frequency component of the original signal is multiplied by the corresponding compensation coefficient to enhance the high-frequency component and adjust the low-frequency component so that the signal can maintain the original waveform and spectrum characteristics to the greatest extent after passing through the channel. After the filtering operation is completed, the compensated signal is obtained. The compensated signal is Fourier transformed to generate a frequency domain compensation signal. The frequency domain compensation signal is inversely Fourier transformed to restore the pre-compensated signal in the time domain. The time domain pre-compensated signal is converted into a modulated drive current through digital-to-analog conversion and current drive circuit. The modulated drive current is used as input to drive the on-chip optoelectronic modulator to generate the final pre-compensated signal.

[0034] Step 300: Perform on-chip multi-channel parallel transmission on the pre-compensated signal, collect channel crosstalk and dispersion distortion data, and obtain multi-channel transmission state information;

[0035] It should be noted that the pre-compensated signal is distributed through multi-channel branching technology so that it enters different transmission paths of the on-chip multi-channel optoelectronic interconnection system respectively, and a multi-channel transmission channel signal is obtained. Each channel carries an independent signal stream, and the parallel processing method can significantly improve the total bandwidth and efficiency of signal transmission. In order to monitor the integrity of the signal in multi-channel transmission, the signal integrity analysis of the multi-channel transmission channel signal is performed. Through high-precision sampling and calculation, the initial transmission characteristic data of each channel is obtained, including key characteristics such as signal amplitude, phase and jitter. Based on the initial transmission characteristic data, the crosstalk effect between multiple channels is collected and analyzed in real time. The extraction of the crosstalk parameters between channels depends on correlation analysis and the identification of interference sources. By capturing the unrelated components between the signals of each channel, the interference intensity and the impact range between channels are quantified. These parameters reflect the cross-interference characteristics of the signal in multi-channel transmission. Based on the crosstalk parameters between channels, the temperature drift effect on the chip is dynamically monitored. The temperature drift distortion data is obtained by real-time measurement of the thermal noise, power fluctuation and transmission delay of the key nodes in the channel. The signal dispersion feature extraction is performed on the temperature drift distortion data. Through joint analysis of the time domain and frequency domain, such as eye diagram analysis and dynamic measurement of spectrum width, the features of signal delay, broadening and waveform distortion caused by dispersion are extracted. For the dispersion distortion features, dynamic compensation calculation is performed, and the distortion is corrected in real time by adjusting the transmission parameters of each channel, such as channel gain, bias voltage or time synchronization offset, to obtain the compensated multi-channel signal. The transmission state features of the compensated multi-channel signal and the crosstalk parameters between channels are extracted. By constructing a feature extraction model, the signal's amplitude, phase, spectrum and jitter and other attributes are comprehensively analyzed, and this information is integrated into a state feature vector. Data encoding and feature fusion operations are performed on the state feature vector. Data encoding reduces data redundancy through compression and standardization, while feature fusion operations integrate multi-dimensional features to form a unified vector with globality and representation, and finally generate multi-channel transmission state information.

[0036] Step 400: Input the pre-compensation signal and the multi-channel transmission state information into a dual-level signal equalization model for linear and nonlinear equalization processing to obtain a first equalized signal;

[0037] Specifically, the pre-compensation signal and the multi-channel transmission state information are sequentially input into the first-level least squares equalizer of the two-level signal equalization model, and the linear distortion caused by parasitic effects is extracted by using the pre-compensation signal. In the analysis of linear distortion, the pre-compensation signal provides the frequency domain and time domain information of the signal, while the multi-channel transmission state information supplements the data of interference effects such as crosstalk and dispersion distortion in the actual channel environment. By combining these two parts of information, a linear equalization input data matrix is ​​constructed. The coefficient optimization operation of the linear equalization input data matrix is ​​performed using the least squares iteration algorithm to solve the optimal linear equalization coefficient. The least squares iteration algorithm approximates the optimal solution through multiple iterations, minimizes the error, and ensures the accuracy and efficiency of the equalization processing. The obtained linear equalization coefficients are used to describe the linear loss and distortion characteristics of the signal when it propagates in the channel. These coefficients reflect the correction requirements of the signal amplitude and phase, and provide parameter support for subsequent convolution operations. Based on the linear equalization coefficients, a linear convolution operation is performed on the linear equalization input data matrix, and the distortion caused by linear factors during the transmission of the signal is corrected by the calculated linear equalization output signal, and the basic form and spectrum distribution of the signal are restored. Linear equalization can only correct the signal degradation caused by linear factors such as parasitic effects and bandwidth limitations. However, in actual on-chip optoelectronic interconnection channels, the complex distortion caused by device nonlinearity and multi-channel interaction cannot be eliminated by linear equalization alone. Therefore, the linear equalization output signal is input into the second level of the two-level signal equalization model, namely the deep neural network equalizer. The deep neural network can perform in-depth analysis and transformation of the signal through its nonlinear feature mapping ability. The multiple hidden layers of the neural network gradually extract the nonlinear features in the signal, model the complex distortion pattern through nonlinear activation functions, generate nonlinear mapping results, and compensate for the high-order nonlinear effects caused by device characteristics, temperature drift and complex interference. The nonlinear mapping results are input into the output layer of the two-level signal equalization model, and the nonlinear equalization output signal is generated through nonlinear transformation. In the output layer, the nonlinear characteristics of the signal are optimized to the optimal state, and the residual distortion is eliminated to maximize the original signal characteristics. Data reconstruction operation is performed on the nonlinear equalization output signal. The signal characteristics are recombined to ensure that the output signal meets the limitations of the physical transmission environment and can accurately express the information content. The first equalized signal finally generated has the compensation effect of both linear and nonlinear distortion, and can maintain signal integrity and transmission quality in the on-chip optoelectronic interconnection channel.

[0038] Step 500: Perform a mixed compensation operation of time domain equalization and frequency domain equalization on the first equalized signal based on a penalty function algorithm and an iterative optimization algorithm to obtain a second equalized signal.

[0039] Specifically, the first equalized signal is characterized by decomposing the features, and the penalty function algorithm is used to separate the different features of the signal by constructing a mathematical model for signal distortion, and the important components are extracted to generate a penalty feature vector. These feature vectors can describe the distortion mode existing in the signal and quantify the influence of different distortion factors in the time domain and frequency domain on the overall quality of the signal. On this basis, the parameters of the penalty feature vector are initialized to provide a reasonable starting point for the subsequent optimization process. By setting the initial optimization parameters, it is ensured that the algorithm can quickly converge to the global optimal solution in subsequent iterations. Based on the optimization initial parameters, the gradient calculation of the first equalized signal is performed, and the iterative optimization matrix for optimization is obtained by solving the gradient information of the objective function. The core of the gradient calculation is to determine the direction and amplitude of the signal distortion, so as to provide guidance for the compensation algorithm, so that each iteration is adjusted in the direction of reducing the distortion. The iterative optimization matrix can dynamically adjust the compensation parameters to adapt to the complex distortion mode in the signal. Using this matrix, the time domain distortion compensation calculation is performed, and the time domain error caused by nonlinear distortion and multipath effect is corrected by analyzing the amplitude jitter and time offset of the signal in the time series, and the time domain compensation data is generated. At the same time, the frequency domain conversion is performed through the iterative optimization matrix to obtain the frequency domain compensation data. In the frequency domain compensation, the spectrum of the signal is adjusted, especially the frequency distortion caused by bandwidth limitation and high-frequency attenuation is corrected. The time domain signal is converted to the frequency domain through Fourier transform, and the frequency components are compensated and corrected in combination with the parameters in the iterative optimization matrix to enhance the energy of the high-frequency components and ensure that the spectrum distribution of the signal is more balanced. The time domain compensation data and the frequency domain compensation data are linked and iteratively optimized based on the iterative optimization algorithm, while taking into account the time domain and frequency domain characteristics of the signal to ensure that the compensation process is consistent in the two domains. Through multiple iterative calculations, the optimization compensation results gradually approach the optimal state, and a signal with both time domain and frequency domain equalization characteristics is generated. In order to avoid signal distortion caused by over-compensation, the penalty term constraint calculation is introduced at this stage. The purpose of the penalty term is to limit the parameter fluctuations in the optimization process, balance the gain adjustment and distortion compensation of the signal, and ensure that the final equalized signal is feasible in physical implementation. The optimization compensation results are smoothed to eliminate the small amplitude jitter and discontinuity points, and generate the second equalized signal. Smoothing is achieved through signal filtering and mathematical fitting methods, which has the effect of making the equalized signal smoother and more stable, avoiding negative impact on subsequent signal processing.

[0040] According to the different characteristics of the time domain compensation data and the frequency domain compensation data, the time domain weight coefficient and the frequency domain weight coefficient are set respectively. These weight coefficients are used to measure the relative importance of the two data in the optimization target. For example, the time domain compensation focuses more on the amplitude jitter and time offset correction of the signal, while the frequency domain compensation focuses on the smoothness of the signal spectrum and the recovery of high-frequency components. Based on these weight coefficients, a hybrid optimization function is constructed to take into account the needs of both the time domain and the frequency domain. The optimized gradient matrix is ​​obtained by performing gradient calculation on the hybrid optimization function. The change direction and change amplitude of the objective function in the parameter space are analyzed to provide guidance for parameter updating. Based on the optimized gradient matrix, the linkage optimization parameters are iteratively updated. An iterative optimization algorithm, such as the gradient descent method or the momentum-based optimization method, is used to continuously adjust the parameters to minimize the objective function value. As the iteration proceeds, the optimization parameters gradually approach the global optimal solution, and an iterative optimization parameter set is generated. These parameters can dynamically adapt to the characteristic differences between the time domain and the frequency domain, thereby achieving balanced optimization. The penalty term is calculated based on the iterative optimization parameters. The penalty term limits excessive adjustment of parameters by introducing constraints, such as implementing sparsity constraints through L1 regularization terms or avoiding excessive parameter values ​​through L2 regularization terms, generating a penalty constraint vector to describe the adjustment range of parameters under constraints. The penalty constraint vector is subjected to soft threshold operation and sparse optimization. Soft threshold operation can weaken the smaller values ​​in the parameters, thereby strengthening the influence of important parameters; sparse optimization further reduces redundant parameters, simplifies the model structure, and improves the generalization ability of the optimization results. Based on the constraint optimization results, the first equalized signal is reconstructed and smoothed to obtain the second equalized signal. In the reconstruction operation, by applying the optimized parameters to the signal model, the residual time domain jitter and frequency domain distortion are effectively eliminated, so that the time domain waveform and frequency domain spectrum of the signal reach a more ideal state. Smoothing is used to eliminate sharp changes or discontinuous points in the signal to ensure the stability and consistency of the second equalized signal in physical implementation. The second equalized signal combines the compensation effects of the time domain and the frequency domain, and has high signal integrity and anti-interference ability.

[0041] In the embodiment of the present application, by establishing a two-level signal equalization model, combining the least squares algorithm and the deep neural network, targeted compensation is performed for linear distortion and nonlinear distortion respectively, thereby significantly improving the accuracy of signal equalization; a linkage optimization mechanism based on a penalty function is adopted to realize the coordinated compensation of time domain equalization and frequency domain equalization, effectively solving the problem of unsatisfactory compensation effect of the traditional single equalization method; a multi-channel parallel transmission control strategy is introduced, and the influence of channel crosstalk and dispersion distortion is effectively suppressed through real-time monitoring and dynamic compensation; combined with S21 pre-compensation and post-equalization processing, a complete signal equalization transmission link is constructed, and a significant expansion of the system bandwidth is achieved. The iterative optimization algorithm is used to adaptively adjust the equalization parameters, which improves the system's robustness to environmental changes and ensures the stability of signal transmission; a mixed penalty form and a soft threshold constraint strategy are adopted to effectively control the parameter divergence problem in the optimization process and ensure the convergence of the algorithm.

[0042] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0043] Performing S21 scattering parameter scanning measurement on the on-chip optoelectronic interconnection channel within a preset frequency range to obtain frequency response data, and performing bandwidth limitation characteristic analysis on the frequency response data to obtain bandwidth limitation characteristic parameters;

[0044] Scan the bias conditions of the optoelectronic transceiver in the on-chip optoelectronic interconnection channel, record the input-output correspondence, obtain the transmission characteristic curve, and perform nonlinear characteristic analysis on the transmission characteristic curve to obtain the nonlinear distortion characteristic parameters;

[0045] Based on the high-speed data pattern, the on-chip optoelectronic interconnection channel is tested and transmitted, the signal waveform data is collected, the distortion characteristic data is obtained, and the inter-symbol interference analysis operation is performed on the distortion characteristic data to obtain the inter-symbol interference characteristic parameters;

[0046] The bandwidth limitation characteristic parameters, nonlinear distortion characteristic parameters and inter-symbol interference characteristic parameters are feature fused to obtain a channel characteristic feature vector, and the channel characteristic feature vector is subjected to data standardization and structured processing to generate a channel characteristic data set.

[0047] Specifically, the S21 scattering parameter scanning measurement is performed on the on-chip optoelectronic interconnection channel within a preset frequency range. S21 represents the transmission coefficient of the signal from the channel input port to the output port, which is used to describe the gain or loss characteristics of the channel. The channel is scanned and measured by a network analyzer to obtain the frequency response data. ,in is the frequency response function of the channel, including the amplitude and Phase ,and Indicates frequency. , the amplitude attenuation of the high frequency part is observed, which is the performance of bandwidth limitation. In order to quantify this characteristic, the bandwidth limitation characteristic parameter is defined as:

[0048] ;

[0049] in is the cutoff frequency of the channel, and is the starting frequency. Through this formula, the frequency range and bandwidth limitation of the channel are characterized to obtain the specific value Finally, the compensation algorithm is combined to pre-compensate the high-frequency attenuation of the channel. The bias conditions of the optoelectronic transceiver in the on-chip optoelectronic interconnect channel are scanned. The operating characteristics of the optoelectronic transceiver depend largely on its bias voltage or current. For example, in the operation of a photodiode or modulator, different bias conditions will affect the gain, linearity, and dynamic range of the signal. By adjusting the bias conditions (indicates bias voltage) and record the input signal and output signal The relationship between the two is used to draw the transmission characteristic curve, for example:

[0050] ;

[0051] in is the linear gain coefficient, and are the second and third order nonlinear distortion coefficients, which respectively describe the nonlinear effects in the system. By fitting and analyzing these coefficients, the nonlinear characteristics of the optoelectronic device are quantified, and these coefficients are used as nonlinear distortion characteristic parameters to indicate the nonlinear performance of the device under specific bias conditions. At the same time, in order to capture dynamic signal distortion, the on-chip optoelectronic interconnect channel is tested and transmitted based on the high-speed data pattern. A known high-speed data pattern is input into the channel. , and collect the output signal obtained after passing through the channel During transmission, due to the dispersion and interference effects of the channel, the output signal is different from the input signal. By collecting signal waveform data, an eye diagram is drawn to analyze the distortion characteristics. Based on the following inter-symbol interference analysis formula, the characteristics of inter-symbol interference (ISI) are quantified:

[0052] ;

[0053] in Indicates the code type The integral value of the impulse response is Is the code type time boundary, and is the time domain impulse response function. By calculating ISI, the degree of signal distortion caused by pattern dependence is quantitatively described, and it is used as the inter-symbol interference characteristic parameter for subsequent analysis. After obtaining the above characteristic parameters, the bandwidth-limited characteristic parameter , nonlinear distortion characteristic parameters The feature fusion is performed with the inter-symbol interference characteristic parameter ISI to generate the channel characteristic feature vector. The mathematical expression of feature fusion is:

[0054] ;

[0055] in is the channel characteristic feature vector. In order to make the feature vector suitable for further model training or optimization, it is subjected to data standardization, for example, by normalization method:

[0056] ;

[0057] in is the eigenvalue, is the mean of the feature, is the standard deviation of the feature. By standardization, the influence of dimension is eliminated. By structural processing, the feature vector Organized into matrix or tensor form to construct a complete channel feature dataset. For example, assuming there are multiple channels, the channel feature dataset is represented as:

[0058] ;

[0059] Each of these lines Indicates The feature vector of each channel.

[0060] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0061] The S21 scattering parameters of the on-chip optoelectronic interconnection channel are calculated according to the frequency response data in the channel characteristic data set to obtain the channel frequency response function, and the channel frequency response function is processed by Fourier transform to obtain the channel transfer function;

[0062] A pre-compensation filter is designed based on the channel transfer function, a pre-compensation transfer function matrix is ​​constructed, and pre-compensation filter parameters are obtained. Power fading compensation is calculated based on the pre-compensation filter parameters to obtain a compensation coefficient matrix.

[0063] Performing a pre-compensation filtering operation on the original digital signal according to the compensation coefficient matrix to obtain a compensated signal, and performing Fourier transform on the compensated signal to obtain a frequency domain compensated signal;

[0064] The frequency domain compensation signal is subjected to inverse Fourier transform processing to obtain a time domain pre-compensation signal, and the time domain pre-compensation signal is converted into a modulated driving current and input into an on-chip optoelectronic modulator to obtain a pre-compensation signal.

[0065] Specifically, the S21 scattering parameters of the on-chip optoelectronic interconnection channel are calculated according to the frequency response data in the channel characteristic data set to obtain the frequency response function of the channel: S21 parameters is the channel at a given frequency The gain or loss characteristics under this condition are measured by a network analyzer to obtain its amplitude and phase components, which can be expressed as Based on these data, construct the frequency response function of the channel ,in Contains the frequency domain transmission characteristics of the channel, reflecting the input signal With output signal The relationship between In order to analyze the transmission characteristics of the channel in the time domain, Perform inverse Fourier transform to obtain the time domain transfer function of the channel

[0066] ;

[0067] in, Describing the transient response of the channel to the input signal can reveal the characteristics of the signal in time, such as delay, reflection and attenuation. Based on the channel transfer function , design a pre-compensation filter to perform appropriate high-frequency enhancement and phase correction on the signal to compensate for the loss and distortion in the channel. The core of the pre-compensation filter is to construct a transfer function matrix , so that in the channel and compensation filter After the combined effect, the overall transmission characteristics tend to the ideal state, that is, For design , define the objective function as:

[0068] ;

[0069] in, and are the starting frequency and the cutoff frequency respectively. By optimizing the above objective function, the parameter matrix of the compensation filter is determined .according to Calculate the power fading compensation coefficient matrix , whose expression is:

[0070] ;

[0071] in, Represents the frequency-dependent adjustment of the signal power, which is used to compensate for the power attenuation of different frequency components. Using the compensation coefficient matrix , for the original digital signal Perform pre-compensation filtering operation. Convert the original signal to the frequency domain through fast Fourier transform to obtain , and then perform compensation calculation on each frequency component to obtain the compensated frequency domain signal :

[0072] ;

[0073] Through inverse Fourier transform, the compensated frequency domain signal Convert back to the time domain to generate the compensated signal . For the compensated signal Perform Fourier transform to obtain frequency domain compensation signal , to verify the effect of adjusting the frequency distribution. By observing the recovery of its high-frequency components, it is confirmed whether the compensation filter design has achieved the expected goal. Perform an inverse Fourier transform on the frequency domain compensation signal and return it to the time domain to obtain the time domain pre-compensation signal. The signal is converted into an analog signal through digital-to-analog conversion and then input into the drive circuit to generate a modulated drive current. The drive current matches the transfer function of the optoelectronic modulator, thereby completing the modulation of the optical signal and generating a pre-compensation signal. .

[0074] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0075] Perform multi-channel demultiplexing on the pre-compensated signal, and distribute the pre-compensated signal to an on-chip multi-channel optoelectronic interconnection system to obtain a multi-channel transmission channel signal;

[0076] Perform signal integrity analysis on multi-channel transmission channel signals to obtain initial transmission characteristic data, and collect crosstalk effects in multiple channels on the chip in real time based on the initial transmission characteristic data to obtain crosstalk parameters between channels;

[0077] Based on the crosstalk parameters between channels, the signal distortion caused by the on-chip temperature drift is dynamically monitored to obtain the temperature drift distortion data, and the signal dispersion feature is extracted from the temperature drift distortion data to obtain the dispersion distortion feature;

[0078] Perform dynamic compensation calculation on the dispersion distortion characteristics, adjust the transmission parameters of each channel, and obtain compensated multi-channel signals;

[0079] The transmission state features of the compensated multi-channel signals and the crosstalk parameters between channels are extracted to obtain the state feature vector, and data encoding and feature fusion operations are performed on the state feature vector to output the multi-channel transmission state information.

[0080] Specifically, the pre-compensation signal Multi-channel distribution is performed through the splitter, and the signal received by each channel is expressed as ,in is the channel index, indicating the The transmission signal of the channel. This allocation process is expressed as:

[0081] ;

[0082] in is the branching coefficient, which is used to control the power allocation ratio of each channel signal. The branched signals are transmitted to multiple channels of the on-chip optoelectronic interconnection system to obtain multi-channel transmission channel signals. Signal integrity analysis is performed on the multi-channel transmission channel signals, including amplitude, jitter and phase measurements. Assume that the received signal is , define signal integrity metrics such as eye opening and jitter :

[0083] ;

[0084] Where Threshold is the signal decision threshold, Indicates the standard deviation of the signal over time. By analyzing these indicators, the initial transmission characteristic data of each channel is obtained, including signal amplitude and jitter characteristics. Based on the initial transmission characteristic data, the crosstalk effect between channels is collected in real time. Crosstalk effect It refers to the channel The signal of other channels The interference is expressed mathematically as:

[0085] ;

[0086] in It is a channel For Channel The crosstalk transfer function describes the distribution of interference in time. Measurement to extract crosstalk parameters between channels , which is used to characterize the mutual interference in the system. At the same time, the temperature drift on the chip will further cause signal distortion, especially in high-speed transmission environments. , monitor the transmission performance fluctuations caused by temperature changes. For example, temperature changes It affects the phase and gain characteristics of the channel, and its distortion effect is calculated by the following formula:

[0087] ;

[0088] in and is the temperature-dependent phase drift and gain variation coefficient. By measuring and fitting and , generate temperature drift distortion data, and extract the signal dispersion characteristics related to it. Dynamic compensation calculation is performed on the dispersion distortion characteristics. The core of compensation is to adjust the transmission parameters of each channel. , so that the distortion effect is minimized. Assume that the compensation objective function is:

[0089] ;

[0090] in is the target eye opening, is the regularization coefficient used to balance crosstalk and signal integrity. , and get the compensated multi-channel signal . For the compensated multi-channel signal and channel-to-channel crosstalk parameters Extract transmission state features. Construct state feature vector :

[0091] ;

[0092] And encode the data, for example by normalizing it:

[0093] ;

[0094] The normalized feature vector The state features are fused into a matrix to generate the final multi-channel transmission state information.

[0095] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0096] The pre-compensation signal and the multi-channel transmission state information are sequentially input into the first-level least square equalizer in the double-level signal equalization model, and the linear distortion caused by the parasitic effect is feature extracted to obtain a linear equalization input data matrix;

[0097] Performing coefficient optimization operation on the linear equalization input data matrix through the least squares iterative algorithm to obtain linear equalization coefficients, and performing linear convolution operation on the linear equalization input data matrix based on the linear equalization coefficients to obtain a linear equalization output signal;

[0098] Inputting the linear equalization output signal into the second-level deep neural network equalizer in the dual-level signal equalization model for nonlinear feature mapping and nonlinear transformation to obtain a nonlinear mapping result;

[0099] The nonlinear mapping result is input into the output layer of the double-level signal equalization model for nonlinear transformation to obtain a nonlinear equalization output signal, and a data reconstruction operation is performed on the nonlinear equalization output signal to output a first equalization signal.

[0100] Specifically, the pre-compensation signal and multi-channel transmission status information Input the first level of the two-level signal equalization model, namely the least squares equalizer. The linear distortion caused by parasitic effects is corrected by the least squares fitting method. In the linear equalization part, the linear equalization input data matrix is ​​constructed. , the matrix consists of the pre-compensated signal The sampling value and multi-channel status information The time domain projection is composed of:

[0101] ;

[0102] in It's time point The pre-compensation signal value, Indicates the status information The eigenvalues ​​at time The weight, is the number of sampling points, is the characteristic number of state information. Input the least squares iterative algorithm to calculate the optimal linear equalization coefficient , and its optimization goal is:

[0103] ;

[0104] in, is the target vector of the ideal output signal, is the linear equalization weight vector that needs to be optimized. By solving the optimization problem, the optimal parameters of linear equalization are obtained. Applied to a linear convolution operation to generate a linear equalized output signal :

[0105] ;

[0106] in It is the linear equalization weights, is the corresponding state information component. The linear equalization output signal can effectively suppress linear distortion. The input is sent to the second layer of the two-layer signal equalization model, namely the deep neural network equalizer. The deep neural network has a multi-layer structure, through the weights between layers and bias Optimize the nonlinear distortion and perform feature mapping and transformation. Assuming that the deep neural network has Layer structure, the first Output of the layer Expressed as:

[0107] ;

[0108] in, is a nonlinear activation function, such as ReLU or Sigmoid, and They are The weight matrix and bias vector of the layer, It is The final output of the network is the nonlinear equalization mapping result. , and optimized compensation was performed in the frequency domain and amplitude. Input to the output layer, perform nonlinear transformation to eliminate residual distortion and generate nonlinear equalized output signal The output layer optimizes the loss function , the nonlinear equalization signal approaches the ideal target signal. Perform data reconstruction operations to restore the time domain and frequency domain consistency of the signal and obtain the first equalized signal .

[0109] The pre-compensation signal and the multi-channel transmission state information are sequentially input into the first-level least squares equalizer in the two-level signal equalization model, and the linear distortion caused by the parasitic effect is feature extracted to obtain a linear equalization input data matrix, including: inputting the pre-compensation signal into a 5th-order finite impulse response filter for signal sampling, with the number of sampling points set to 1024, to obtain pre-compensation sampling data; performing 32-bit quantization encoding on the pre-compensation sampling data, with the quantization step set to 0.01, to obtain quantized pre-compensation data; inputting the quantized pre-compensation data and the multi-channel transmission state information into a 25×25-dimensional feature mapping matrix for data mapping, to obtain an initial feature mapping result; the initial feature mapping result is coded by sliding The moving window method is used to segment the data, the window length is set to 32, the overlap rate is set to 50%, and a segmented data matrix is ​​obtained; the segmented data matrix is ​​input into the least squares feature extraction unit to perform parasitic effect feature separation operation, the feature dimension is set to 64, and the parasitic effect features are obtained; a 128×64-dimensional linear distortion feature matrix is ​​constructed according to the parasitic effect features, and the data normalization coefficient is set to 0.1 to obtain linear distortion data; a singular value decomposition operation is performed on the linear distortion data, and the feature components with a contribution rate greater than 0.01 are retained to obtain a reduced-dimensional feature matrix; the reduced-dimensional feature matrix is ​​reconstructed through orthogonal transformation, the transformation matrix dimension is set to 64×32, and the linear equalization input data matrix is ​​output.

[0110] Among them, a coefficient optimization operation is performed on a linear equalization input data matrix through a least squares iterative algorithm to obtain a linear equalization coefficient, and a linear convolution operation is performed on the linear equalization input data matrix based on the linear equalization coefficient to obtain a linear equalization output signal, including: the linear equalization input data matrix is ​​divided into blocks according to 25×25 dimensions, the principal component characteristics are extracted through a singular value decomposition operation, and the eigenvalues ​​are arranged in descending order to obtain a characteristic decomposition data matrix; the characteristic decomposition data matrix is ​​input into a least squares iterative optimization unit, and the iteration termination condition is set to be that the error change rate of 1000 consecutive iterations is less than 0.001, and the parameters are initialized to obtain iterative optimization initial parameters; the objective function is constructed for the iterative optimization initial parameters, the least squares error function is established through 25-dimensional vector space mapping, and the L2 regularization term is introduced for constraint to obtain the optimization objective function; based on the optimization objective function, The gradient is calculated in the parameter space, the chain derivative rule is used to construct the error back propagation path, and the gradient value of each parameter is calculated to obtain the parameter gradient matrix; the parameter gradient matrix is ​​input into the dynamic learning rate adjustment unit, the learning rate is adaptively adjusted according to the gradient value, and the learning rate range is set to 0.01 to 0.001 to obtain the linear equalization coefficient; a 5th-order finite impulse response filter is constructed according to the linear equalization coefficient, the filter coefficient matrix dimension is set to 5×5, and the filter coefficient is normalized to obtain the linear convolution kernel; the linear equalization input data matrix is ​​input into the sliding convolution processing unit, and a sliding window with 32-point sampling is used for convolution operation, the window sliding step is set to 1, and the signal is linearly filtered to obtain the filtering processing result; the filtering processing result is subjected to DC offset elimination and signal amplitude normalization, and the data is calibrated by the zero mean normalization method to output the linear equalization output signal.

[0111] The linear equalization output signal is input into the second-level deep neural network equalizer in the double-level signal equalization model for nonlinear feature mapping and nonlinear transformation to obtain a nonlinear mapping result, including: performing a 16-bit dislocation filling operation on the linear equalization output signal, expanding the data dimension to 128×64, constructing a feature matrix by zero-filling, and obtaining the first preprocessed data; inputting the first preprocessed data into the data standardization unit for maximum and minimum value normalization operation, scaling the value range to the interval [-1, 1], and performing mean centering processing to obtain standardized feature data; based on the standardized feature data, data segmentation is performed according to a batch size of 256 samples, and each batch of data is shuffled to obtain training batch data; the training batch data is input into the first hidden layer of the neural network, and the training batch data is obtained through 6 Four neurons perform linear mapping of the input features in 64×64 dimensions, and use the ReLU activation function for nonlinear transformation to obtain the output data of the first hidden layer; a neuron inactivation rate of 0.3 is set for the output data of the first hidden layer to perform random discarding to increase the robustness of the network and obtain the denoised feature data; the denoised feature data is input into the second hidden layer of the neural network, and 32 neurons perform feature compression in 64×32 dimensions, and use the ReLU activation function for nonlinear mapping to obtain the output data of the second hidden layer; the output data of the second hidden layer is batch normalized, the data distribution offset is adjusted, and the mean variance is calibrated to obtain the normalized output data; a residual connection path is constructed based on the normalized output data, and the input features and output features are skip-layer connected and weighted fused to obtain the nonlinear mapping result.

[0112] In a specific embodiment, the process of executing step 500 may specifically include the following steps:

[0113] Based on the penalty function algorithm, the first equalized signal is characterized by decomposing the first equalized signal to obtain a penalty characteristic vector, and the penalty characteristic vector is parameterized to obtain an optimized initial parameter;

[0114] Performing gradient calculation on the first equalization signal according to the optimized initial parameters to obtain an iterative optimization matrix;

[0115] Perform time domain distortion compensation calculation based on iterative optimization matrix to obtain time domain compensation data, and perform frequency domain conversion through iterative optimization matrix to obtain frequency domain compensation data;

[0116] Based on the iterative optimization algorithm, the time domain compensation data and the frequency domain compensation data are linked and iteratively optimized to obtain an optimized compensation result, and penalty item constraint calculation and smoothing processing are performed according to the optimized compensation result to obtain a second equalized signal.

[0117] Specifically, for the first equalized signal Perform eigendecomposition to extract the main distortion mode in the signal. Eigendecomposition uses a penalty function algorithm, which introduces a regularization term to limit the amplitude or sparsity of the characteristic components, thereby separating the main features more effectively. The sampled data is represented as a vector , the eigendecomposition is performed through the following objective function:

[0118] ;

[0119] in, is the characteristic basis matrix, which represents the basis function set of the signal. is the penalty eigenvector to be sought, describing the projection coefficients of the signal on different basis functions. Regularization parameter Controlling the sparsity constraints, the larger The value tends to generate sparse solutions. By solving the above optimization problem, we get the eigenvector , extracts the representative distortion pattern in the signal. Perform parameter initialization to generate optimization initial parameters The goal of initialization is to provide a reasonable starting point for the subsequent optimization process. Initialization methods include uniform distribution, normal distribution, or eigenvector-based normalization methods, such as:

[0120] ;

[0121] in, Represents the norm of the eigenvector, the initialized parameter Ensure proper scale and directionality. Based on the optimization of initial parameters , for the first equalized signal Perform gradient calculations to build an iterative optimization matrix Gradient calculation is the core of the optimization algorithm. By calculating the partial derivative of the objective function with respect to the parameter, the direction and magnitude of the parameter update are determined. Suppose the objective function is , then the gradient calculation formula is:

[0122] ;

[0123] Gradient Matrix Each element of It means the objective function is about signal components and the The partial derivatives of the parameters are used to guide the subsequent parameter updates. Based on the iterative optimization matrix , respectively, to perform compensation calculations in the time domain and frequency domain. In the time domain, using Correct the amplitude and phase distortion of the signal to obtain time domain compensation data :

[0124] ;

[0125] in is the learning rate, which controls the step size of the gradient update. Similarly, after converting the signal to the frequency domain through Fourier transform, we use Compensate the frequency domain components to obtain frequency domain compensation data :

[0126] ;

[0127] in, represents the Fourier transform operation. The frequency domain compensation data can restore the high-frequency signal energy while reducing the spectrum distortion caused by channel characteristics. In order to make full use of the compensation results in the time domain and frequency domain, an iterative optimization algorithm is used to and The goal of linkage optimization is to minimize the errors in both time domain and frequency domain at the same time. The optimization goal is expressed as:

[0128] ;

[0129] in, is the target signal, is the inverse Fourier transform. Through multiple iterations, the parameters are gradually optimized , and get the optimized compensation result After the optimization compensation results are generated, penalty term constraint calculation and smoothing are performed to avoid overfitting or introducing false signals. The penalty term avoids overcompensation by controlling the amplitude or gradient of the optimization parameters, for example:

[0130] ;

[0131] in and is the penalty coefficient, which controls the sparsity and smoothness constraints respectively. Smoothing further eliminates sharp changes in the signal, which is done using a low-pass filter. Combining the optimization compensation result and the penalty term constraint, the second equalized signal is obtained. .

[0132] In a specific embodiment, the execution step performs linkage iterative optimization on the time domain compensation data and the frequency domain compensation data based on an iterative optimization algorithm to obtain an optimized compensation result, and performs penalty term constraint calculation and smoothing processing according to the optimized compensation result to obtain the second equalized signal. The process may specifically include the following steps:

[0133] According to the time domain compensation data and the frequency domain compensation data, a time domain weight coefficient and a frequency domain weight coefficient are set respectively, and a hybrid optimization function is constructed;

[0134] Based on the hybrid optimization function, gradient calculation and parameter update are performed to obtain an optimized gradient matrix, and the linkage optimization parameters are iteratively updated according to the optimized gradient matrix to obtain iterative optimization parameters;

[0135] The penalty term is calculated based on the iterative optimization parameters to obtain the penalty constraint vector, and the penalty constraint vector is subjected to soft threshold operation and sparse optimization to obtain the constraint optimization result;

[0136] The first equalized signal is reconstructed and smoothed according to the constraint optimization result to obtain a second equalized signal.

[0137] Specifically, based on time domain compensation data and frequency domain compensation data , respectively set the time domain weight coefficient and frequency domain weight coefficient These weight coefficients are used to balance the optimization requirements between the time domain and the frequency domain. The specific values ​​are adjusted according to the priorities of different application scenarios. For example, if the amplitude jitter of the time domain signal requires a higher priority, then The application of weights is reflected in the construction of the hybrid optimization function, whose objective function is:

[0138] ;

[0139] in, represents the parameter vector to be optimized, is the target signal, is the Fourier transform, which is used to convert the target signal into the frequency domain for comparison. The two terms of measure the errors in the time domain and frequency domain respectively, through the weight and Achieve a dynamic balance between the two optimization requirements. Perform gradient calculation to obtain the optimized gradient matrix Each element of the gradient matrix represents the partial derivative of the objective function with respect to the corresponding parameter, and the calculation formula is:

[0140] ;

[0141] in, and Respectively represent the time domain and frequency domain compensation matrices, which are related to the signal characteristics and channel characteristics and are generated through measurement and modeling. After that, the parameters Perform iterative updates to gradually optimize the objective function. The parameter update formula is:

[0142] ;

[0143] in, is the learning rate, which controls the step size of each iteration, Represents the number of iterations. Through multiple iterations, the parameter Will gradually approach the global optimal solution and generate the optimization parameter vector Based on the optimization parameters , penalty term calculation is performed to constrain the optimization results and prevent overfitting or signal distortion. The penalty term is defined as follows:

[0144] ;

[0145] in, is the sparsity regularization term of the parameters, used to encourage the sparsity of the solution, is the smoothness regularization term of the parameter, which is used to limit the excessive fluctuation of the parameter. and is the regularization coefficient. By substituting the optimization parameters into the penalty term calculation, the penalty constraint vector is generated In order to optimize the signal, the penalty constraint vector Perform soft threshold operation and sparse optimization. The formula for soft threshold operation is:

[0146] ;

[0147] in, is the penalty vector elements, is the soft threshold parameter, which controls the degree of sparseness. After soft threshold processing, the sparse optimization result is generated. Based on the sparse optimization results , for the first equalized signal Perform reconstruction and smoothing to obtain the final second equalized signal The signal reconstruction is achieved by the following formula:

[0148] ;

[0149] in, It is the inverse Fourier transform, which is used to convert the frequency domain signal back to the time domain. Smoothing is done through a low-pass filter to eliminate sharp fluctuations and ensure the physical feasibility and consistency of the signal.

[0150] The above describes the signal balanced transmission method of the high-speed optoelectronic interconnection in the embodiment of the present application. The following describes the signal balanced transmission device 10 of the high-speed optoelectronic interconnection in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of a signal equalization transmission device 10 for high-speed optoelectronic interconnection includes:

[0151] The acquisition module 11 is used to acquire bandwidth characteristics and nonlinear characteristics of the on-chip optoelectronic interconnection channel to obtain a channel characteristic data set;

[0152] A pre-compensation module 12 is used to measure the scattering parameters of the on-chip optoelectronic interconnection channel according to the channel characteristic data set, and perform S21 pre-compensation processing through a pre-compensation filter to obtain a pre-compensated signal;

[0153] The parallel transmission module 13 is used to perform on-chip multi-channel parallel transmission of the pre-compensated signal, collect channel crosstalk and dispersion distortion data, and obtain multi-channel transmission status information;

[0154] An equalization processing module 14 is used to input the pre-compensation signal and the multi-channel transmission state information into a double-level signal equalization model for linear and nonlinear equalization processing to obtain a first equalized signal;

[0155] The hybrid compensation module 15 is used to perform a hybrid compensation operation of time domain equalization and frequency domain equalization on the first equalized signal based on a penalty function algorithm and an iterative optimization algorithm to obtain a second equalized signal.

[0156] Through the coordinated cooperation of the above-mentioned components, a two-level signal equalization model is established, combined with the least squares algorithm and deep neural network, targeted compensation is performed for linear distortion and nonlinear distortion respectively, and the accuracy of signal equalization is significantly improved; a linkage optimization mechanism based on penalty function is adopted to realize the coordinated compensation of time domain equalization and frequency domain equalization, effectively solving the problem of unsatisfactory compensation effect of traditional single equalization method; a multi-channel parallel transmission control strategy is introduced, and the influence of channel crosstalk and dispersion distortion is effectively suppressed through real-time monitoring and dynamic compensation; a complete signal equalization transmission link is constructed by combining S21 pre-compensation and post-equalization processing, and a significant expansion of system bandwidth is achieved; an iterative optimization algorithm is used to adaptively adjust the equalization parameters, which improves the system's robustness to environmental changes and ensures the stability of signal transmission; a mixed penalty form and soft threshold constraint strategy are adopted to effectively control the parameter divergence problem in the optimization process and ensure the convergence of the algorithm.

[0157] See also Figure 3 , Figure 3 This is a schematic block diagram of the structure of an electronic device 300 provided in an embodiment of the present application. The electronic device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected via a device bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0158] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned signal balanced transmission methods for high-speed optoelectronic interconnection.

[0159] The processor 301 is used to provide computing and control capabilities to support the operation of the entire electronic device 300 .

[0160] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned signal balanced transmission methods for high-speed optoelectronic interconnection.

[0161] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a partial structure related to the present application scheme, and does not constitute a limitation on the electronic device 300 involved in the present application scheme. The specific electronic device 300 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0162] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0163] It should be noted that technicians in the relevant field can clearly understand that for the convenience and simplicity of description, the specific working process of the electronic device 300 described above can refer to the corresponding process of the aforementioned high-speed optoelectronic interconnection signal equalization transmission method, and will not be repeated here.

[0164] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the signal balanced transmission method for high-speed optoelectronic interconnection provided in the embodiment of the present application.

[0165] The computer-readable storage medium may be an internal storage unit of the electronic device 300 in the aforementioned embodiment, such as a hard disk or memory of the electronic device 300. The computer-readable storage medium may also be an external storage device of the electronic device 300, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped with the electronic device 300.

[0166] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0167] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0168] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A signal balanced transmission method for high-speed optoelectronic interconnection, characterized in that: The method comprises: Collect bandwidth characteristics and nonlinear characteristics of on-chip optoelectronic interconnection channels to obtain a channel characteristic data set; According to the channel characteristic data set, the scattering parameters of the on-chip optoelectronic interconnection channel are measured, and S21 pre-compensation processing is performed through the pre-compensation filter to obtain a pre-compensation signal; specifically including: according to the frequency response data in the channel characteristic data set, the S21 scattering parameters of the on-chip optoelectronic interconnection channel are calculated to obtain a channel frequency response function, and the channel frequency response function is Fourier transformed to obtain a channel transfer function; based on the channel transfer function, a pre-compensation filter is designed, a pre-compensation transfer function matrix is ​​constructed, and pre-compensation filter parameters are obtained, and power fading compensation calculation is performed based on the pre-compensation filter parameters to obtain a compensation coefficient matrix; according to the compensation coefficient matrix, a pre-compensation filtering operation is performed on the original digital signal to obtain a compensated signal, and the compensated signal is Fourier transformed to obtain a frequency domain compensation signal; the frequency domain compensation signal is inversely Fourier transformed to obtain a time domain pre-compensation signal, and the time domain pre-compensation signal is converted into a modulated drive current, which is input into the on-chip optoelectronic modulator to obtain a pre-compensation signal, wherein S21 represents the transmission coefficient of the signal from the channel input port to the output port, which is used to describe the gain or loss characteristics of the channel; Performing on-chip multi-channel parallel transmission on the pre-compensated signal, and collecting channel crosstalk and dispersion distortion data to obtain multi-channel transmission state information; The pre-compensation signal and the multi-channel transmission state information are input into a two-level signal equalization model for linear and nonlinear equalization processing to obtain a first equalization signal; specifically including: inputting the pre-compensation signal and the multi-channel transmission state information into the first-level least squares equalizer in the two-level signal equalization model in sequence, extracting features of linear distortion caused by parasitic effects, and obtaining a linear equalization input data matrix; performing coefficient optimization operation on the linear equalization input data matrix through a least squares iterative algorithm to obtain linear equalization coefficients, and performing linear convolution operation on the linear equalization input data matrix based on the linear equalization coefficients to obtain a linear equalization output signal; inputting the linear equalization output signal into the second-level deep neural network equalizer in the two-level signal equalization model for nonlinear feature mapping and nonlinear transformation to obtain a nonlinear mapping result; inputting the nonlinear mapping result into the output layer of the two-level signal equalization model for nonlinear transformation to obtain a nonlinear equalization output signal, and performing data reconstruction operation on the nonlinear equalization output signal to output a first equalization signal; Based on a penalty function algorithm and an iterative optimization algorithm, a mixed compensation operation of time domain equalization and frequency domain equalization is performed on the first equalized signal to obtain a second equalized signal.

2. The signal balanced transmission method for high-speed optoelectronic interconnection according to claim 1, characterized in that: The collecting of bandwidth characteristics and nonlinear characteristics of the on-chip optoelectronic interconnection channel to obtain a channel characteristic data set includes: Performing S21 scattering parameter scanning measurement on the on-chip optoelectronic interconnection channel within a preset frequency range to obtain frequency response data, and performing bandwidth limitation characteristic analysis on the frequency response data to obtain bandwidth limitation characteristic parameters; Scanning bias conditions of the optoelectronic transceiver in the on-chip optoelectronic interconnection channel, recording the input-output correspondence, obtaining a transmission characteristic curve, and performing nonlinear characteristic analysis on the transmission characteristic curve to obtain nonlinear distortion characteristic parameters; Based on the high-speed data pattern, the on-chip optoelectronic interconnection channel is tested and transmitted, signal waveform data is collected to obtain distortion characteristic data, and inter-symbol interference analysis operation is performed on the distortion characteristic data to obtain inter-symbol interference characteristic parameters; The bandwidth limitation characteristic parameter, the nonlinear distortion characteristic parameter and the inter-symbol interference characteristic parameter are feature-fused to obtain a channel characteristic feature vector, and the channel characteristic feature vector is data-standardized and structured to generate a channel characteristic data set.

3. The signal balanced transmission method for high-speed optoelectronic interconnection according to claim 1, characterized in that: The performing on-chip multi-channel parallel transmission of the pre-compensated signal, and collecting channel crosstalk and dispersion distortion data to obtain multi-channel transmission state information, includes: Performing multi-channel branching on the pre-compensation signal, distributing the pre-compensation signal to an on-chip multi-channel optoelectronic interconnection system to obtain a multi-channel transmission channel signal; Performing signal integrity analysis on the multi-channel transmission channel signals to obtain initial transmission characteristic data, and collecting crosstalk effects in the on-chip multi-channels in real time based on the initial transmission characteristic data to obtain inter-channel crosstalk parameters; Dynamically monitoring the signal distortion caused by the on-chip temperature drift based on the inter-channel crosstalk parameter to obtain temperature drift distortion data, and extracting signal dispersion characteristics from the temperature drift distortion data to obtain dispersion distortion characteristics; Performing dynamic compensation calculation on the dispersion distortion characteristics, adjusting the transmission parameters of each channel, and obtaining a compensated multi-channel signal; Transmission state feature extraction is performed on the compensated multi-channel signal and the inter-channel crosstalk parameter to obtain a state feature vector, and data encoding and feature fusion operations are performed on the state feature vector to output multi-channel transmission state information.

4. The signal balanced transmission method for high-speed optoelectronic interconnection according to claim 1, characterized in that: The method of performing a mixed compensation operation of time domain equalization and frequency domain equalization on the first equalized signal based on the penalty function algorithm and the iterative optimization algorithm to obtain a second equalized signal includes: Performing feature decomposition on the first equalized signal based on a penalty function algorithm to obtain a penalty feature vector, and initializing parameters of the penalty feature vector to obtain an optimized initial parameter; Performing gradient calculation on the first equalized signal according to the optimization initial parameters to obtain an iterative optimization matrix; Performing time domain distortion compensation calculation based on the iterative optimization matrix to obtain time domain compensation data, and performing frequency domain conversion through the iterative optimization matrix to obtain frequency domain compensation data; Based on an iterative optimization algorithm, the time domain compensation data and the frequency domain compensation data are linked and iteratively optimized to obtain an optimized compensation result, and penalty term constraint calculation and smoothing processing are performed according to the optimized compensation result to obtain a second equalized signal.

5. The signal balanced transmission method for high-speed optoelectronic interconnection according to claim 4, characterized in that: The method of performing linkage iterative optimization on the time domain compensation data and the frequency domain compensation data based on an iterative optimization algorithm to obtain an optimized compensation result, and performing penalty term constraint calculation and smoothing processing according to the optimized compensation result to obtain a second equalized signal includes: According to the time domain compensation data and the frequency domain compensation data, a time domain weight coefficient and a frequency domain weight coefficient are respectively set, and a hybrid optimization function is constructed; Performing gradient calculation and parameter update based on the hybrid optimization function to obtain an optimized gradient matrix, and iteratively updating the linkage optimization parameters according to the optimized gradient matrix to obtain iterative optimization parameters; Calculating penalty items based on the iterative optimization parameters to obtain a penalty constraint vector, and performing soft threshold operation and sparse optimization on the penalty constraint vector to obtain a constraint optimization result; The first equalized signal is reconstructed and smoothed according to the constraint optimization result to obtain a second equalized signal.

6. A signal equalization transmission device for high-speed optoelectronic interconnection, characterized in that: Used to perform the signal balanced transmission method of high-speed optoelectronic interconnection according to any one of claims 1 to 5, the signal balanced transmission device of high-speed optoelectronic interconnection comprises: An acquisition module is used to acquire bandwidth characteristics and nonlinear characteristics of an on-chip optoelectronic interconnection channel to obtain a channel characteristic data set; A pre-compensation module, used to measure the scattering parameters of the on-chip optoelectronic interconnection channel according to the channel characteristic data set, and perform S21 pre-compensation processing through a pre-compensation filter to obtain a pre-compensated signal; A parallel transmission module, used for performing on-chip multi-channel parallel transmission of the pre-compensated signal, collecting channel crosstalk and dispersion distortion data, and obtaining multi-channel transmission status information; An equalization processing module, used for inputting the pre-compensation signal and the multi-channel transmission state information into a dual-level signal equalization model for linear and nonlinear equalization processing to obtain a first equalized signal; The hybrid compensation module is used to perform a hybrid compensation operation of time domain equalization and frequency domain equalization on the first equalized signal based on a penalty function algorithm and an iterative optimization algorithm to obtain a second equalized signal.

7. An electronic device, characterized in that: The electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electronic device executes the signal equalization transmission method for high-speed optoelectronic interconnection according to any one of claims 1-5.

8. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the signal balanced transmission method for high-speed optoelectronic interconnection according to any one of claims 1 to 5 is implemented.

Images