Signal compensation method and device, electronic equipment and storage medium

Abstract

This application provides a signal compensation method, apparatus, device, and storage medium. The method includes: acquiring a target signal; extracting signal features from the target signal to obtain multi-dimensional signal features; inputting environmental parameters and the multi-dimensional signal features into a first sub-network of a signal compensation model to obtain compensation parameters corresponding to the target signal output by the first sub-network; determining eye diagram data and jitter data corresponding to the target signal based on the multi-dimensional signal features; inputting the eye diagram data, jitter data, and compensation parameters corresponding to the target signal into a second sub-network of the signal compensation model to obtain a target function output by the second sub-network; and compensating the target signal based on the target function to obtain a compensated signal of the target signal.

Description

Technical Field

[0001] This application relates to the field of signal processing technology, and in particular to a signal compensation method, apparatus, electronic device and storage medium. Background Technology

[0002] With the development of computer technology, PCIe (Peripheral Component Interconnect Express) has become an indispensable high-speed data transmission standard in computer technology. As a widely used generation standard, PCIe 4.0 has a data transmission rate of up to 16GT / s, and also places higher demands on signal quality and integrity. Summary of the Invention

[0003] This application provides a signal compensation method, apparatus, electronic device, and storage medium to at least solve the above-mentioned technical problems existing in the prior art.

[0004] A first aspect of this application provides a signal compensation method, comprising:

[0005] Acquire the target signal; The target signal is subjected to signal feature extraction to obtain multi-dimensional signal features; The environmental parameters and the multi-dimensional signal features are input into the first sub-network of the signal compensation model to obtain the compensation parameters corresponding to the target signal output by the first sub-network. Based on the multi-dimensional signal features, determine the eye diagram data and jitter data corresponding to the target signal; The eye diagram data, jitter data and compensation parameters corresponding to the target signal are input into the second sub-network of the signal compensation model to obtain the target function output by the second sub-network. The target signal is compensated based on the objective function to obtain the compensated signal of the target signal.

[0006] In one possible implementation, the step of inputting environmental parameters and the multi-dimensional signal features into the first sub-network of the signal compensation model to obtain compensation parameters corresponding to the target signal output by the first sub-network includes: The environmental parameters include temperature, impedance mismatch, and crosstalk intensity. The compensation parameters include the pre-emphasis coefficient and the equalizer coefficient; The pre-emphasis coefficient is obtained by processing the temperature, impedance mismatch, and crosstalk intensity through the first sub-network of the signal compensation model. The equalizer coefficients are obtained by processing the temperature, impedance mismatch, and multi-dimensional signal characteristics through the first sub-network of the signal compensation model.

[0007] In one possible implementation, the step of inputting the eye diagram data, jitter data, and compensation parameters corresponding to the target signal into the second sub-network of the signal compensation model to obtain the target function output by the second sub-network includes: The compensation parameters are processed by the second sub-network to obtain the power consumption sub-function corresponding to the target signal; The eye diagram data is processed by the second sub-network to obtain the eye diagram sub-function corresponding to the target signal; The jitter data is processed by the second sub-network to obtain the jitter sub-function corresponding to the target signal; The target function is obtained by processing the power consumption sub-function, the eye diagram sub-function, and the jitter sub-function through the second sub-network. The target function is positively correlated with the eye diagram sub-function and the jitter sub-function, and negatively correlated with the power consumption sub-function.

[0008] In one possible implementation, determining the eye diagram data and jitter data corresponding to the target signal based on the multi-dimensional signal features includes: Based on the multi-dimensional signal features, an eye diagram and jitter curve corresponding to the target signal are generated; Based on the eye diagram, a signal quality index corresponding to the target signal is determined, and the signal quality index is used to characterize the quality of the target signal. The eye diagram data is determined based on the signal quality indicators; Based on the jitter curve, a signal integrity index corresponding to the target signal is determined, and the signal integrity index is used to characterize the integrity of the target signal; The jitter data is determined based on the signal integrity index.

[0009] In one possible implementation, signal feature extraction is performed on the target signal to obtain multi-dimensional signal features, including: Transient features are extracted from the target signal to obtain signal transient features, which include: rise time, fall time, and overshoot rate; wherein, the rise time is the time for the target signal to rise from a first preset voltage amplitude to a second preset voltage amplitude, the fall time is the time for the target signal to rise from the second preset voltage amplitude to the first preset voltage amplitude, the first preset voltage amplitude is less than the second preset voltage amplitude, and the overshoot rate is the proportion by which the maximum voltage amplitude of the target signal exceeds the stable voltage amplitude of the target signal; Perform wavelet transform on the target signal to obtain wavelet entropy features; The multidimensional signal features are obtained by fusing the transient features of the signal and the wavelet entropy features.

[0010] In one possible implementation, before acquiring the target signal, the method further includes: Acquire the raw signal; An elliptic low-pass filter is used to filter out the aliasing signal in the original signal to obtain the first signal; The first signal is converted from an analog signal to a digital signal using a signal sampler to obtain the second signal; The second signal is normalized to obtain the target signal.

[0011] A second aspect of this application provides a signal compensation device, the device comprising: The acquisition module is used to acquire the target signal; The feature extraction module is used to extract signal features from the target signal to obtain multi-dimensional signal features; The first obtaining module is used to input environmental parameters and the multi-dimensional signal features into the first sub-network of the signal compensation model to obtain the compensation parameters corresponding to the target signal output by the first sub-network. The determination module is used to determine the eye diagram data and jitter data corresponding to the target signal based on the multi-dimensional signal features; The second obtaining module is used to input the eye diagram data, jitter data and compensation parameters corresponding to the target signal into the second sub-network of the signal compensation model to obtain the target function output by the second sub-network. The compensation module is used to compensate the target signal based on the objective function to obtain the compensated signal of the target signal.

[0012] In one possible implementation, the environmental parameters include temperature, impedance mismatch, and crosstalk intensity; The compensation parameters include the pre-emphasis coefficient and the equalizer coefficient; The first obtaining module includes: The first sub-processing module is used to process the temperature, the impedance mismatch, and the crosstalk intensity through the first sub-network of the signal compensation model to obtain the pre-emphasis coefficient. The second sub-processing module is used to process the temperature, impedance mismatch, and multi-dimensional signal characteristics through the first sub-network of the signal compensation model to obtain the equalizer coefficients.

[0013] A third aspect of this application provides an electronic device comprising: At least one processor; and a memory communicatively connected to at least one processor; wherein, The memory stores instructions that can be executed by at least one processor, which enables the at least one processor to perform the method of this application.

[0014] A fourth aspect of this application provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the method of this application.

[0015] The signal compensation method, apparatus, electronic device, and storage medium of this application acquire multi-dimensional signal characteristics of the target signal and determine compensation parameters together with environmental parameters, making the compensation parameters more accurate and comprehensive. At the same time, the objective function integrates the eye diagram data, jitter data, and compensation parameters corresponding to the target signal. The eye diagram data, jitter data, and compensation parameters respectively characterize the quality, integrity, and power consumption of the target signal. By adjusting the environmental parameters and multi-dimensional signal characteristics, the eye diagram data, jitter data, and compensation parameters can be adjusted, thereby achieving synergistic optimization of the quality, integrity, and power consumption of the target signal and completing the compensation of the target signal.

[0016] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0017] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of this application are illustrated in the drawings by way of example and not limitation, in which: In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0018] Figure 1 A flowchart illustrating the signal compensation method provided in this application embodiment; Figure 2 This is the eye diagram corresponding to the target signal in the embodiments of this application; Figure 3 A schematic diagram of the signal compensation device provided in this application; Figure 4 A schematic diagram of the composition structure of an electronic device according to an embodiment of this application is shown. Detailed Implementation

[0019] To make the objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] This application provides a signal compensation method. In this embodiment, the signal compensation method can be executed by an electronic device, which may include, but is not limited to, desktop computers, laptops, smartphones, tablets, video conferencing all-in-one machines, etc.

[0021] Figure 1 A flowchart of the signal compensation method provided in the embodiments of this application is shown below. Figure 1 As shown, the method includes: Step 101: The electronic device acquires the target signal.

[0022] In one embodiment, before acquiring the target signal, the method further includes: Electronic devices acquire raw signals; The electronic device uses an elliptic low-pass filter to filter out aliasing signals in the original signal to obtain the first signal; Electronic devices use a signal sampler to convert the first signal from an analog signal into a digital signal to obtain the second signal; The electronic device normalizes the second signal to obtain the target signal.

[0023] The signals mentioned in this application can be those generated during data transmission via the PCIe bus.

[0024] Specifically, the process begins with acquiring the raw signal. During the signal acquisition phase, if the sampling frequency is less than twice the highest frequency of the signal, high-frequency components are incorrectly reconstructed as low-frequency signals, causing signal aliasing. Signal aliasing leads to severe signal distortion, destroys the integrity of the original information, introduces false low-frequency components, and interferes with data analysis and processing. It generates noise or artifacts in audio, image, and other fields, reducing the system's signal-to-noise ratio; it can even cause control misjudgments (such as communication errors or equipment malfunctions), threatening system stability.

[0025] Therefore, after sampling the original signal, this application needs to filter the original signal to avoid interference and thus obtain the target signal.

[0026] It should be noted that elliptic low-pass filters are commonly used anti-aliasing filters in high-speed signal processing. Their core characteristic is an extremely steep transition band, enabling them to achieve very high bandwidth suppression at relatively low orders. Therefore, in this application, the steep transition band of the elliptic low-pass filter allows for rapid attenuation of high-frequency noise at a relatively low order, effectively suppressing aliasing. Compared to Butterworth or Chebyshev filters, elliptic filters significantly improve bandwidth edge suppression efficiency while maintaining controllable passband and stopband ripples. This ensures that the sampled signal retains effective components within a limited bandwidth while accurately truncending interference, thus improving the fidelity and anti-interference capability of the acquired signal. Therefore, by limiting the signal bandwidth using anti-aliasing filters (such as elliptic low-pass filters), it is possible to ensure that high-frequency interference is filtered out before sampling, avoiding irreversible information distortion.

[0027] An elliptic low-pass filter filters out frequencies higher than the cutoff frequency before sampling. The high-frequency signal has a stopband attenuation capability of ≥60dB, which can strongly suppress signals above the preset cutoff frequency.

[0028] Cutoff frequency ,in, This is the sampling frequency. The cutoff frequency coefficient is 0.8 to reserve a 20% bandwidth margin to prevent edge frequency distortion.

[0029] The advantage of choosing an elliptic low-pass filter is that, at the same order, it has a steeper transition band, making it suitable for high-frequency signals in PCIe 4.0.

[0030] In this application, the first signal is an analog signal. Sampling the first signal converts the continuous analog signal into a discrete digital signal sequence, resulting in the second signal. This conversion can be achieved using a signal sampler. After normalizing the second signal, the target signal is obtained. Normalization ensures that signals with different amplitude ranges are scaled uniformly, facilitating subsequent processing.

[0031] Step 102: The electronic device extracts signal features from the target signal to obtain multi-dimensional signal features.

[0032] In high-speed digital systems (such as high-speed communication links) and precision measurement, efficient extraction of signal features is crucial for fault diagnosis and performance evaluation. These multi-dimensional features can include time-domain transient dimensions and frequency-domain scale dimensions. The time-domain transient dimension focuses on the dynamic changes of the signal, extracting features such as rise time (the time it takes for the target signal to rise from a first preset voltage amplitude to a second preset voltage amplitude), fall time (the time it takes for the target signal to rise from a second preset voltage amplitude to a first preset voltage amplitude), and overshoot rate (the percentage of the signal exceeding its stable value). Time-domain transient features reflect the signal's response characteristics during rapid switching and are a key indicator for evaluating signal edge integrity. For example, an excessively high overshoot rate can lead to signal distortion, and an excessively long rise time can reduce transmission rate. The frequency-domain scale dimension uses wavelet transform to decompose the signal into multiple scales, obtaining wavelet entropy features for different frequency components.

[0033] The advantage of wavelet transform lies in its ability to adaptively separate the time-frequency characteristics of different frequency bands of a signal: the DB4 wavelet basis, with its tight support and approximate symmetry, accurately matches the steep edges of high-speed signals. Its multi-level decomposition gradually strips away high-frequency noise and low-frequency fundamental waves, and the energy entropy further quantifies the differences in complexity of each frequency band.

[0034] In one embodiment, signal feature extraction is performed on the target signal to obtain multi-dimensional signal features, including: The electronic device extracts transient features from the target signal to obtain transient features, which include: rise time, fall time, and overshoot rate. The rise time is the time it takes for the target signal to rise from a first preset voltage amplitude to a second preset voltage amplitude, the fall time is the time it takes for the target signal to rise from the second preset voltage amplitude to the first preset voltage amplitude, the first preset voltage amplitude is less than the second preset voltage amplitude, and the overshoot rate is the proportion by which the maximum voltage amplitude of the target signal exceeds the stable voltage amplitude of the target signal. Electronic devices perform wavelet transform on the target signal to obtain wavelet entropy features; Electronic devices integrate signal transient features and wavelet entropy features to obtain multi-dimensional signal features.

[0035] In this application, transient feature extraction captures the dynamic behavior of a signal in the time domain, including rise time. descent time and overshoot rate Rise time is the time interval between a signal transitioning from a first preset voltage amplitude (e.g., 10% amplitude) to a second preset voltage amplitude (e.g., 90% amplitude), representing the switching speed of the signal from "0" to "1". Fall time is the time interval between a signal dropping from the second preset voltage amplitude (e.g., 90% amplitude) to the first preset voltage amplitude (e.g., 10% amplitude), representing the switching speed of the signal from "1" to "0". These two parameters, rise time and fall time, are directly related to the signal transmission rate and edge integrity. Specifically, excessively long rise / fall times can cause "tailing" of the signal during high-speed transmission, leading to inter-symbol interference; excessively short times may cause oscillations due to insufficient circuit bandwidth. Overshoot rate The portion of the signal whose maximum voltage amplitude exceeds the signal's stable voltage amplitude. ) and the stable voltage amplitude of the signal ( The ratio of (e.g.) An excessively high overshoot rate can cause the signal to exceed the circuit's withstand voltage range, leading to hardware damage or additional noise. It is an important indicator for evaluating signal stability.

[0036] Wavelet entropy features include signal components at different scales after the target signal is decomposed. These signal components at different scales are used to characterize the signal features of the target signal at different frequencies and resolutions.

[0037] In this application, DB4 wavelet basis decomposition is used to extract 5-level wavelet entropy features. (j=1,2,3,4,5)

[0038] in This represents the energy entropy (wavelet entropy feature) after the j-th level decomposition. Let be the wavelet coefficients of the j-th layer. Energy entropy is used to quantify signal complexity and distinguish between normal and distorted signals.

[0039] Wavelet basis is a key concept in wavelet transform, and different wavelet bases have different characteristics. The DB4 wavelet basis (Daubechies 4 wavelet basis) is one of the Daubechies wavelet families. It has compact support and symmetry. For PCIE signals, rapid edge changes are an important feature of the signal, and the DB4 wavelet basis can effectively capture and analyze these changes, thereby better extracting the signal features.

[0040] Five-level wavelet decomposition of a signal involves passing the target signal through a series of filters (determined by the chosen wavelet basis) to progressively decompose it into signal components of different scales. Each level of decomposition yields a low-frequency component (approximate signal) and a high-frequency component (detail signal). After five levels of decomposition, the signal is decomposed into multiple sub-signals of different scales, each reflecting the characteristics of the target signal at different frequency ranges and resolutions.

[0041] When performing wavelet transform on a target signal, it is decomposed into a series of components at different scales: low-scale components and high-scale components. The low-scale components correspond to high-frequency signals and can capture high-frequency details such as sudden noise and rapid oscillations. The high-scale components correspond to low-frequency signals and can reflect low-frequency characteristics such as the overall trend of the signal, slow decay, or baseline drift. This application utilizes wavelet entropy features to quickly distinguish between effective and interference components in a signal.

[0042] The multi-dimensional signal features extracted in this application possess key characteristics of signal time-domain dynamics and frequency-domain distribution, thus avoiding the limitations of single-dimensional analysis. By fusing signal transient features and wavelet entropy features, the signal's time-domain robustness, frequency-domain energy distribution, and nonlinear distortion characteristics can be comprehensively characterized, providing highly discriminative input for machine learning models and improving the accuracy and generalization ability of signal anomaly detection (such as reflection and crosstalk).

[0043] Step 103: The electronic device inputs environmental parameters and multi-dimensional signal characteristics into the first sub-network of the signal compensation model to obtain the compensation parameters corresponding to the target signal output by the first sub-network.

[0044] It should be noted that environmental parameters can be parameters affected by the environment, including temperature T and impedance mismatch. Crosstalk intensity, Xtalk, etc. Among them, impedance mismatch is a parameter that measures the degree of inconsistency between the source impedance (transmitter) and the load impedance (receiver) in the signal transmission path; crosstalk intensity refers to the interference intensity generated between adjacent signal lines due to electromagnetic coupling.

[0045] Compensation parameters include pre-weighting coefficient and equalizer coefficients .

[0046] The pre-emphasis factor represents the degree to which the transmitting end actively enhances the high-frequency components of the signal. Since high-frequency components attenuate more severely than low-frequency components during high-speed signal transmission, pre-emphasis increases the amplitude of the high-frequency signal to offset the high-frequency losses caused by the transmission link. The equalizer factor is an adjustable parameter used at the receiving end to compensate for signal distortion. It corrects distorted signals using a digital filter (equalizer) to counteract negative effects such as inter-symbol interference.

[0047] In one embodiment, environmental parameters and multi-dimensional signal features are input into the first sub-network of the signal compensation model to obtain compensation parameters corresponding to the target signal output by the first sub-network, including: The electronic device processes temperature, impedance mismatch, and crosstalk intensity through the first sub-network of the signal compensation model to obtain the pre-emphasis coefficient; Electronic devices process temperature, impedance mismatch, and multi-dimensional signal characteristics through the first sub-network of the signal compensation model to obtain equalizer coefficients.

[0048] Specifically, the electronic device will extract multi-dimensional signal features and temperature T, impedance mismatch, etc. The crosstalk strength Xtalk input signal compensation model (e.g., Deep Reinforcement Learning (DRL) algorithm) is then used by the electronic device to dynamically adjust the pre-emphasis coefficients by calling the first sub-network in the signal compensation model. and equalizer coefficients To maintain the stability of the high-speed signal link, the pre-emphasis coefficient is used. and equalizer coefficients The formula is shown below:

[0049]

[0050] in, and This represents the mapping relationship between input states and output actions established by deep reinforcement learning algorithms through a complex learning and training process.

[0051] Pre-weighting coefficient ∈[0,6]dB; equalizer coefficient ∈[-0.2,0.2]; The measurement range of T is [-40℃,125℃]; Z= ∈[0, 50], where, For load impedance, This is the source impedance.

[0052] Xtalk extracts frequencies using frequency domain analysis, with a frequency range of 1GHz to 10GHz and a resolution of 10MHz.

[0053] Xtalk frequency domain analysis: Crosstalk can be divided into near-end crosstalk (NEXT) and far-end crosstalk (FEXT), and its power spectral density is:

[0054] in, Let be the coupling coefficient between the k-th adjacent channel and the main channel; Let be the interference signal power of the k-th channel.

[0055] Xtalk Time Domain Characterization: Crosstalk waveforms are measured using a Time Domain Reflectometry (TDR) to extract peak voltage. .

[0056]

[0057] in, This represents the voltage of the target signal.

[0058] In this application, the pre-emphasis coefficient compensates for high-frequency attenuation during transmission by enhancing the high-frequency components of the signal. The adjustment is based on the following: increased temperature leads to increased high-frequency loss in the transmission medium (such as cables or PCB lines), necessitating increased pre-emphasis strength; impedance mismatch causes signal reflection, and the reflection intensity is positively correlated with the mismatch degree, requiring pre-emphasis to compensate for amplitude loss caused by reflection; crosstalk between adjacent signals introduces high-frequency noise, requiring pre-emphasis to enhance useful high-frequency components while avoiding excessive noise amplification (it needs to be coordinated with subsequent equalizer coefficients). For example, when the temperature rises sharply and crosstalk is strong, the signal compensation model may choose a combination strategy of "medium-intensity pre-emphasis + high-frequency noise suppression." The equalizer coefficients are a fine-grained compensation based on signal characteristics. The equalizer compensates for channel distortion in reverse, and its coefficient adjustment depends on the characteristics of the signal itself. In this application, the basic equalizer coefficients are determined based on temperature and impedance mismatch, and then fine-grained optimization is performed by combining multi-dimensional signal characteristics. For example, if the wavelet entropy features show severe attenuation of high-frequency components (low energy of high-scale components), then increase the high-frequency gain of the equalizer; if the transient features of the signal show that the rise time is too long (slow signal edges), then increase the steepness adjustment coefficient of the equalizer.

[0059] Signal quality and integrity analysis is a crucial step in ensuring the quality of high-speed signal transmission. Related technologies typically employ fixed pre-emphasis and equalizer parameters to mitigate signal attenuation during transmission, which fails to adapt to dynamically changing channel environments, leading to signal quality degradation. In this application, the pre-emphasis and equalizer parameters can be dynamically adjusted based on environmental parameters and multi-dimensional signal characteristics, thereby adapting to dynamically changing channel environments and improving signal quality.

[0060] In this application, compensation parameters are determined by combining multi-dimensional signal characteristics with environmental parameters, which makes the compensation parameters more accurate and comprehensive.

[0061] Step 104: The electronic device determines the eye diagram data and jitter data corresponding to the target signal based on multi-dimensional signal characteristics; In one embodiment, determining the eye diagram data and jitter data corresponding to the target signal based on multi-dimensional signal features includes: Electronic devices generate eye diagrams and jitter curves corresponding to target signals based on multi-dimensional signal characteristics; Electronic devices determine the signal quality index corresponding to the target signal based on the eye diagram. The signal quality index is used to characterize the quality of the target signal. Electronic devices determine eye diagram data based on signal quality indicators; Electronic devices determine the signal integrity index corresponding to the target signal based on the jitter curve. The signal integrity index is used to characterize the integrity of the target signal. Electronic devices determine jitter data based on signal integrity metrics.

[0062] In this application, the quality and integrity of the target signal will change after each parameter adjustment. Therefore, based on the target signal after each adjustment, the corresponding eye diagram and jitter curve can be generated.

[0063] Figure 2 This is the eye diagram corresponding to the target signal in the embodiments of this application.

[0064] Specifically, such as Figure 2 As shown, in high-speed digital signal transmission, an eye diagram is a graphical method for intuitively evaluating signal quality. Because the symbol period of high-speed digital signals is very short, it is difficult to directly observe their waveform details. Therefore, an eye diagram is created by using multiple symbol periods (here, 10...) to visually assess signal quality. 4 An eye diagram is formed by aligning and superimposing signal waveforms (units of time interval) in time. The shape of the eye diagram reflects many characteristics of the signal, such as inter-symbol interference (ISI), noise, and jitter. If the signal quality is good, the "eye" of the eye diagram will be large and clear, indicating that the rising and falling edges of the signal are steep and the ISI is small. Conversely, if the "eye" is closed or blurry, it indicates that there are large ISI or noise problems.

[0065] After each eye diagram is generated, a signal quality index corresponding to the target signal is determined based on the eye diagram. The signal quality index is used to characterize the quality of the target signal.

[0066] Specifically, the signal quality metric is the eye opening. This refers to eye diagram data. Eye diagram opening. It can be obtained through the following formula:

[0067] in, It represents the vertical height of the open portion of the eye diagram, which reflects the effective voltage range of the signal in a steady state; The peak value of the signal is represented by the difference between its maximum and minimum voltage. By calculating the ratio of these two values ​​and multiplying it by 100%, the percentage of eye diagram opening is obtained, which is used to measure the quality of the signal.

[0068] Jitter manifests as timing deviation in high-speed digital signals and is a core indicator for measuring signal integrity. With transmission rates exceeding GHz levels and clock cycles compressed to nanosecond levels, even sub-picosecond jitter can cause bit errors. Its causes encompass physical effects such as phase noise, power supply interference, and crosstalk, directly impacting bit error rate, eye diagram quality, and system timing margin. In fields such as 5G communication, high-speed signal transmission, and high-speed storage, jitter control has become a critical challenge in signal integrity engineering, requiring coordinated optimization through low-jitter clock sources, precise impedance matching, and advanced equalization algorithms.

[0069] In jitter analysis, plot the signal integrity index. As time evolves, signal jitter is calculated using the TIE (Time Interval Error) algorithm. TIE represents the time deviation between the signal edge and the ideal clock edge. By calculating the root mean square (RMS) of these deviations, the root mean square jitter of the signal can be obtained, which is the signal integrity index. Signal integrity indicators This refers to jitter data. Signal integrity indicator. It can be obtained through the following formula:

[0070] in, It is the error of the i-th time interval (TIE). It is all The average value is N, where N is the total number of samples.

[0071] Step 105: The electronic device inputs the eye diagram data and jitter data corresponding to the target signal into the signal compensation model, and determines the target function corresponding to the target signal by combining the compensation parameters.

[0072] In one embodiment, eye diagram data and jitter data corresponding to the target signal are input into a signal compensation model, and a target function corresponding to the target signal is determined by combining compensation parameters, including: The electronic device processes the compensation parameters through a second sub-network to obtain the power consumption sub-function corresponding to the target signal; The electronic device processes the eye diagram data through a second sub-network to obtain the eye diagram sub-function corresponding to the target signal; The electronic device processes the jitter data through a second sub-network to obtain the jitter sub-function corresponding to the target signal; The electronic device processes the power consumption sub-function, eye diagram sub-function, and jitter sub-function through the second sub-network to obtain the objective function. The objective function is positively correlated with the eye diagram sub-function and the jitter sub-function, and negatively correlated with the power consumption sub-function.

[0073] Specifically, the electronic device inputs eye diagram data, jitter data, and compensation parameters into the second sub-network of the signal compensation model, and then calls the second sub-network to process the data to obtain the objective function.

[0074] The objective function R can be obtained by the following formula:

[0075] in, For eye diagram subfunctions, For the jitter subfunction, For power consumption subfunction, This refers to the eye diagram opening (eye diagram data). This represents the theoretical maximum value of the eye diagram opening. This refers to the root mean square jitter of the signal (jitter data). This is the theoretical maximum value of the root mean square jitter of the signal, which is generally equal to 0.1UI (UnitInterval). , and These are the weighting coefficients for the eye diagram subfunction, jitter subfunction, and power consumption subfunction, respectively.

[0076] The objective function R is used to comprehensively evaluate the performance of a system (usually in communication systems). It obtains a single value by weighting and summing different factors to measure the overall performance of the system in terms of signal quality and power consumption.

[0077] Based on the above formula, if wavelet entropy features Influence on pre-weighting coefficient and equalizer coefficients According to and The functional properties of wavelet entropy can increase or decrease wavelet entropy features. To change the pre-weighting coefficient and equalizer coefficients The value of is then used to adjust the objective function R. If the wavelet entropy feature... With pre-weighting coefficient Positive correlation reduces wavelet entropy features This can make the pre-weighting coefficient To reduce the size of the objective function R, Decreasing this term may increase the R value. Similarly, based on the functional relationship, the effect of adjusting parameters such as rise time, overshoot rate, mismatch resistance, and temperature on the preweighting coefficient can be observed. and equalizer coefficients The influence of this, in turn, adjusts the R value.

[0078] The signal compensation model in this application includes a two-layer network: an Actor network (first sub-network) and a Critic network (second sub-network).

[0079] The Actor network consists of a 3-layer fully connected neural network (input layer -> 64 nodes -> 32 nodes -> action output layer), which outputs corresponding actions (i.e., pre-emphasis coefficients and equalizer coefficients) based on the input state. The input layer includes a state space (multi-dimensional signal features and environmental parameters); the hidden layer consists of 64 nodes; and the output layer includes an action space (pre-emphasis coefficients and equalizer coefficients).

[0080] The Critic network comprises a dual-input structure (state + action -> 64 nodes -> 32 nodes -> R-value output) used to evaluate the action value output by the Actor network. Specifically, based on the current state and the action taken, a Q-value (the R-value in this embodiment) is calculated to represent the merit of the action in the current state, providing guidance for parameter updates in the Actor network. The input layer includes a state space and an action space; the hidden layer includes 64 nodes. The output layer includes the Q-value. This application uses the Critic network to evaluate action value, achieving collaborative optimization between the physical layer (pre-emphasis) and the link layer.

[0081] This application trains the signal compensation model using preset training parameters such as learning rate, discount factor, exploration rate, and experience replay buffer capacity, thereby improving the model's stability and convergence.

[0082] The learning rate determines the step size of parameter updates during network training. A suitable learning rate ensures the convergence speed and stability of the network. The learning rate for an Actor network can be 10. -4 The learning rate of the Critic network can be 10. -3 .

[0083] The discount factor δ represents the importance of future rewards. The closer δ is to 1, the more the algorithm values ​​future rewards and tends to seek the long-term optimal strategy. In this application, the discount factor δ can be 0.99.

[0084] In the early stages of training, a larger exploration rate allows the network to try more different actions and explore new strategies. As training progresses, the exploration rate gradually decreases, and the network becomes more inclined to choose the optimal action it has already learned. In this application, the exploration rate decays from 0.5 to 0.01.

[0085] The experience replay mechanism can break the correlation between data, improving the stability and convergence of the algorithm. The cache capacity determines the amount of historical experience that can be stored, while the batch size determines the amount of data retrieved from the cache for training each time. In this application, the cache capacity can be 10. 5 The batch size can be 128.

[0086] In high-speed signal processing systems, dynamic changes in environmental parameters (such as temperature, impedance mismatch, and crosstalk intensity) can cause continuous fluctuations in signal quality (eye diagram, jitter). Fixed learning rate strategies struggle to adapt to this complexity. If the learning rate is too high, the signal compensation model is prone to oscillations or divergence due to sudden environmental changes; if the learning rate is too low, the convergence speed is too slow, failing to meet real-time requirements. Dynamic learning rate adjustment intelligently adjusts the parameter update step size by monitoring the rate of change in signal quality—accelerating convergence during stable phases and suppressing oscillations during periods of severe fluctuation, thus balancing training speed and stability. This strategy ensures that the signal compensation model can efficiently optimize pre-emphasis coefficients and equalizer parameters under complex operating conditions, maintaining signal integrity in high-speed links, and is one of the core mechanisms for achieving adaptive closed-loop control in the system.

[0087] In one embodiment, the method further includes: Determine the rate of change of signal quality based on signal quality indicators and / or signal integrity indicators; Based on the rate of change of signal quality, the learning rate corresponding to the signal compensation model is determined. The learning rate is related to the adjustment step size of environmental parameters and multi-dimensional signal features. If the rate of change of signal quality is less than a first preset threshold, the learning rate is adjusted to the initial value multiplied by a first coefficient, where the first coefficient is greater than a preset value. If the rate of change of signal quality is greater than or equal to a first preset threshold and less than a second preset threshold, the learning rate is kept at the initial value. If the rate of change of signal quality is greater than or equal to a second preset threshold, the learning rate is adjusted to the initial value multiplied by a second coefficient, where the second coefficient is less than a preset value.

[0088] Specifically, the core of the dynamic learning rate adjustment strategy is based on the rate of change in signal quality. Automatically adjust the learning rate of the signal compensation model To balance convergence speed and stability, the learning rate is appropriately increased when signal quality changes are small to accelerate convergence. When signal quality changes are large, the learning rate is decreased to avoid oscillations or dispersion. In this disclosure, the preset value can be 1.

[0089] Based on signal quality change rate The learning rate of the signal compensation model is adjusted according to the following rules, and the mathematical expression is:

[0090] in, The value of the rate of change of signal quality is [0,1]. Score the signal quality at the current moment (e.g., a score based on signal quality metrics and / or signal integrity metrics). The current learning rate, with a value in the range

[10] . -5 10 -3 ], As the initial value, it can be set to 10. -4 The first coefficient is 1.1 and the second coefficient is 0.9. The first and second coefficients are used to increase and decrease the learning rate, respectively, in order to accelerate convergence and stabilize training. The first preset threshold is 0.01 and the second preset threshold is 0.05. Of course, it is understandable that the specific values ​​of the first coefficient, the second coefficient, the first preset threshold and the second preset threshold can be changed according to actual needs.

[0091] Step 106: The electronic device compensates the target signal based on the objective function to obtain the compensated signal of the target signal.

[0092] The electronic device compensates for the target signal based on the objective function to obtain a compensated signal, including: Electronic devices determine the values ​​of compensation parameters corresponding to the target signal based on environmental parameters and multi-dimensional signal characteristics. The electronic device determines the value of the target function based on the eye diagram data, jitter data, and compensation parameter values ​​corresponding to the target signal. The electronic device determines whether the function value of the target function meets the preset requirements; in response to the fact that the function value of the target function does not meet the preset requirements, it adjusts the values ​​of environmental parameters and multi-dimensional signal characteristics to adjust the function value of the target function until the function value of the target function meets the preset requirements. The electronic device generates a compensation signal for the target signal based on the multi-dimensional signal characteristics corresponding to the function value of the target function after the preset requirements are met.

[0093] Specifically, such as Figure 1As shown, for example, initial values ​​for environmental parameters and multi-dimensional signal features are first selected. These initial values ​​are then substituted into the formula corresponding to the compensation parameters to obtain their initial values. Based on the initial values ​​of the multi-dimensional signal features, initial values ​​for eye diagram data and jitter data can be obtained. Substituting these values ​​into the objective function yields its initial value. The initial value of the objective function is then determined to meet preset requirements, such as whether it exceeds a certain threshold. If it does, signal compensation is complete. If not, the process returns to adjusting the values ​​of the environmental parameters and multi-dimensional signal features to readjust the objective function's value until it meets the preset requirements, or until a preset number of iterations is reached. Once the objective function's value meets the preset requirements, the signal compensation process is complete. The compensation signal for the target signal can then be generated based on the values ​​of the multi-dimensional signal features obtained during this adjustment.

[0094] The objective function is obtained by weighted summation of different factors to obtain a single function value R, which measures the overall performance of the system in terms of signal quality and power consumption. In this application, eye diagram data uses eye height and eye width as core indicators; higher values ​​indicate better signal amplitude integrity. Jitter data uses total jitter and random jitter as core indicators; lower values ​​indicate better signal timing stability. Power consumption data is positively correlated with compensation parameters (pre-emphasis intensity, equalizer complexity) (e.g., stronger pre-emphasis and higher equalizer order result in higher power consumption); lower values ​​indicate better energy efficiency.

[0095] For example, after initial compensation, if the eye diagram or jitter does not meet the standard, the signal compensation model will indirectly adjust the pre-emphasis coefficient and equalizer coefficient by adjusting environmental parameters (such as controlling the temperature within a reasonable range) or optimizing multi-dimensional signal characteristics (such as further suppressing noise through filtering), thereby reducing the value of power consumption data. The reduction of power consumption data will increase the overall value of the objective function (because they are negatively correlated), driving the model to iterate towards "meeting the quality standard and achieving optimal power consumption". When the objective function is optimized to reach the preset threshold, the iteration stops and the result is output.

[0096] In one embodiment, after obtaining the signal integrity index, the method further includes: By inputting multi-dimensional signal characteristics and signal integrity indicators into the fault detection model, the probability of the target signal failing within a preset time period can be obtained. Based on the probability of failure, determine whether to issue an alarm and the alarm level.

[0097] Signal integrity is susceptible to dynamic factors such as temperature fluctuations, impedance mismatch, and crosstalk, leading to a gradual degradation of signal quality (e.g., eye diagram closure, increased jitter) over time. Traditional fault detection methods rely on threshold alarms, failing to capture the temporal evolution of signal characteristics and hindering early warning. By constructing a fault detection model using Long Short-Term Memory (LSTM) networks, we can effectively model the long-term dependencies of historical signal characteristics (e.g., wavelet entropy features, jitter), identify potential signal degradation patterns from time-series data, and predict future fault probabilities (e.g., link failure risk within 10 seconds). Combined with a signal compensation model, the predictions from the fault detection model can trigger parameter adjustments or maintenance strategies in advance, preventing communication interruptions or equipment damage due to sudden failures and significantly improving system reliability and pre-maintenance capabilities.

[0098] In this application, a Long Short-Term Memory (LSTM) network is used to process the historical signal feature sequence { Modeling is used to predict the probability of failure within a future time T (e.g., T=10 seconds) in the following way. .

[0099]

[0100] in, n The length of the time window (e.g.) n =120 seconds).

[0101] It should be noted that wavelet entropy features and signal integrity metrics are used here to determine the failure probability, but other parameters may be used to determine the failure probability in other embodiments.

[0102] In this application, in order to intuitively reflect the signal integrity status and fault prediction results, the visualization report needs to integrate multi-dimensional data for dynamic display, which may include eye diagrams and jitter curves.

[0103] Eye diagram matrix: Overlay 10 4 An eye diagram is generated at individual UI intervals, annotating the vertical eye opening (Heye) and horizontal eye opening (Weye). The changes in eye diagram morphology over different time periods are displayed using a heatmap format, for example: Green area (Heyeye≥70%): Excellent signal quality, no alarms.

[0104] Yellow area (50%≤Heye<70%): Signal quality deteriorates, requiring attention.

[0105] Red zone (Heye < 50%): Signal is severely degraded, triggering an alarm.

[0106] Jitter timing curve: plotting As the curve evolves over time, a theoretical threshold Jmax = 0.1 × UI is added. When the curve exceeds the threshold, the abnormal time point is marked and analyzed in conjunction with environmental parameters (such as temperature T, impedance mismatch ΔZ).

[0107] Wavelet entropy feature spectrum: showing the wavelet entropy features after 5-level decomposition of the DB4 wavelet. (j=1,2,…,5), the spatiotemporal distribution of entropy values ​​in each frequency band is presented in the form of a waterfall plot. A sudden increase in entropy value in the high-frequency layer (e.g., j=1,2) (e.g., Ej>0.8) indicates an increase in transient noise or crosstalk, while the fluctuation of entropy value in the low-frequency layer (e.g., j=4,5) reflects the stability of the fundamental frequency.

[0108] LSTM Failure Probability Prediction Curve: Dynamically displays the failure probability within the next 10 seconds. And indicate the DRL compensation actions (such as pre-weighting coefficient adjustment) for... The impact. For example, when When the equalizer coefficient was reduced from 0.75 to 0.6, the curve was marked "DRL compensation effective".

[0109] In this application, the alarm strategy of the fault detection model needs to be deeply coordinated with the signal compensation model to realize a complete link from early warning to closed-loop control.

[0110] when When the value is greater than 0.8, a Level 1 alarm is triggered. Level 1 alarms can be handled in the following ways: Immediately switch to redundant communication links to ensure continuous operation of services; at the same time, freeze the current deep reinforcement learning (DRL) parameters and start the safety mode to limit the pre-emphasis coefficient to ≤3dB to prevent oscillations caused by overcompensation.

[0111] It supports manual intervention processes, including automatically generating fault diagnosis reports, accurately locating potential fault causes, such as impedance mismatch ΔZ > 30Ω or temperature exceeding the limit T > 100℃, and issuing prompts to maintenance personnel to check physical connections or the heat dissipation system.

[0112] When 0.6 At a value of 0.8, a level 2 alarm is triggered. Level 2 alarms can be handled in the following ways: Increase the weight α of eye diagram data in the objective function, for example, from 0.3 to 0.5, thereby strengthening the optimization priority of eye diagram opening (Heye). Reducing the exploration rate, for example from 0.2 to 0.05, can help the signal compensation model converge quickly to the optimal strategy under the current environment. An adaptive sampling strategy is adopted to increase the sampling frequency during periods of high signal activity (such as the PCIe burst transmission phase) in order to capture transient distortion details more accurately.

[0113] When 0.4 When the value is less than 0.6, a Level 3 alarm is triggered. Level 3 alarms can be handled in the following ways: The current multi-dimensional signal characteristics and environmental parameters are recorded in the database to provide data support for subsequent offline analysis. Activate the Long Short-Term Memory (LSTM) network to make long-term predictions (e.g., the next hour) to assess the system's degradation rate and provide a reliable basis for preventative maintenance.

[0114] when No alarm is triggered when the value is less than 0.4.

[0115] This application also provides a signal compensation device. Figure 3 A schematic diagram of the signal compensation device provided in this application is shown below. Figure 3 As shown, the device includes: Acquisition module 301 is used to acquire the target signal; Feature extraction module 302 is used to extract signal features from the target signal to obtain multi-dimensional signal features; The first obtaining module 303 is used to input environmental parameters and the multi-dimensional signal features into the first sub-network of the signal compensation model to obtain the compensation parameters corresponding to the target signal output by the first sub-network. The determination module 304 is used to determine the eye diagram data and jitter data corresponding to the target signal based on the multi-dimensional signal features; The second obtaining module 305 is used to input the eye diagram data, jitter data and compensation parameters corresponding to the target signal into the second sub-network of the signal compensation model to obtain the target function output by the second sub-network. The compensation module 306 is used to compensate the target signal based on the objective function to obtain the compensated signal of the target signal.

[0116] In one embodiment, environmental parameters include temperature, impedance mismatch, and crosstalk intensity; The compensation parameters include the pre-emphasis coefficient and the equalizer coefficient; The first module 303 includes: The first sub-processing module is used to process temperature, impedance mismatch and crosstalk intensity through the first sub-network of the signal compensation model to obtain the pre-emphasis coefficient. The second sub-processing module is used to process temperature, impedance mismatch, and multi-dimensional signal characteristics through the first sub-network of the signal compensation model to obtain the equalizer coefficients.

[0117] It should be noted that the signal compensation device in this application embodiment is similar in principle to the aforementioned signal compensation method in solving the problem. Therefore, the implementation process, implementation principle, and beneficial effects of the signal compensation device can be found in the description of the implementation process, implementation principle, and beneficial effects of the aforementioned method. Repeated descriptions will not be repeated.

[0118] According to embodiments of this application, this application also provides an electronic device and a readable storage medium.

[0119] Figure 4 A schematic block diagram of an example electronic device 400 that can be used to implement embodiments of this application is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0120] like Figure 4 As shown, device 400 includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 402 or a computer program loaded from storage unit 408 into random access memory (RAM) 403. RAM 403 may also store various programs and data required for the operation of device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0121] Multiple components in device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of monitors, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0122] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as signal compensation methods. For example, in some embodiments, the signal compensation method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of the signal compensation method described above may be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to perform the signal compensation method by any other suitable means (e.g., by means of firmware).

[0123] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0124] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0125] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0126] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0127] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0128] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0129] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this application can be achieved, and this is not limited herein.

[0130] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0131] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A signal compensation method, characterized in that, include: Acquire the target signal; The target signal is subjected to signal feature extraction to obtain multi-dimensional signal features; The environmental parameters and the multi-dimensional signal features are input into the first sub-network of the signal compensation model to obtain the compensation parameters corresponding to the target signal output by the first sub-network. Based on the multi-dimensional signal features, determine the eye diagram data and jitter data corresponding to the target signal; The eye diagram data, jitter data and compensation parameters corresponding to the target signal are input into the second sub-network of the signal compensation model to obtain the target function output by the second sub-network. The target signal is compensated based on the objective function to obtain the compensated signal of the target signal.

2. The signal compensation method according to claim 1, characterized in that, The first sub-network of the signal compensation model, which inputs environmental parameters and multi-dimensional signal features into the signal compensation model, obtains compensation parameters corresponding to the target signal output by the first sub-network, including: The environmental parameters include temperature, impedance mismatch, and crosstalk intensity. The compensation parameters include the pre-emphasis coefficient and the equalizer coefficient; The pre-emphasis coefficient is obtained by processing the temperature, impedance mismatch, and crosstalk intensity through the first sub-network of the signal compensation model. The equalizer coefficients are obtained by processing the temperature, impedance mismatch, and multi-dimensional signal characteristics through the first sub-network of the signal compensation model.

3. The signal compensation method according to claim 1, characterized in that, The step of inputting the eye diagram data, jitter data, and compensation parameters corresponding to the target signal into the second sub-network of the signal compensation model to obtain the target function output by the second sub-network includes: The compensation parameters are processed by the second sub-network to obtain the power consumption sub-function corresponding to the target signal; The eye diagram data is processed by the second sub-network to obtain the eye diagram sub-function corresponding to the target signal; The jitter data is processed by the second sub-network to obtain the jitter sub-function corresponding to the target signal; The target function is obtained by processing the power consumption sub-function, the eye diagram sub-function, and the jitter sub-function through the second sub-network. The target function is positively correlated with the eye diagram sub-function and the jitter sub-function, and negatively correlated with the power consumption sub-function.

4. The signal compensation method according to claim 1, characterized in that, The step of determining the eye diagram data and jitter data corresponding to the target signal based on the multi-dimensional signal features includes: Based on the multi-dimensional signal features, an eye diagram and jitter curve corresponding to the target signal are generated; Based on the eye diagram, a signal quality index corresponding to the target signal is determined, and the signal quality index is used to characterize the quality of the target signal. The eye diagram data is determined based on the signal quality indicators; Based on the jitter curve, a signal integrity index corresponding to the target signal is determined, and the signal integrity index is used to characterize the integrity of the target signal; The jitter data is determined based on the signal integrity index.

5. The signal compensation method according to claim 1, characterized in that, The target signal is subjected to signal feature extraction to obtain multi-dimensional signal features, including: Transient features are extracted from the target signal to obtain signal transient features, which include: rise time, fall time, and overshoot rate; wherein, the rise time is the time for the target signal to rise from a first preset voltage amplitude to a second preset voltage amplitude, the fall time is the time for the target signal to rise from the second preset voltage amplitude to the first preset voltage amplitude, the first preset voltage amplitude is less than the second preset voltage amplitude, and the overshoot rate is the proportion by which the maximum voltage amplitude of the target signal exceeds the stable voltage amplitude of the target signal; Perform wavelet transform on the target signal to obtain wavelet entropy features; The multidimensional signal features are obtained by fusing the transient features of the signal and the wavelet entropy features.

6. The signal compensation method according to claim 1, characterized in that, Before acquiring the target signal, the method further includes: Acquire the raw signal; An elliptic low-pass filter is used to filter out the aliasing signal in the original signal to obtain the first signal; The first signal is converted from an analog signal to a digital signal using a signal sampler to obtain the second signal; The second signal is normalized to obtain the target signal.

7. A signal compensation device, characterized in that, The device includes: The acquisition module is used to acquire the target signal; The feature extraction module is used to extract signal features from the target signal to obtain multi-dimensional signal features; The first obtaining module is used to input environmental parameters and the multi-dimensional signal features into the first sub-network of the signal compensation model to obtain the compensation parameters corresponding to the target signal output by the first sub-network. The determination module is used to determine the eye diagram data and jitter data corresponding to the target signal based on the multi-dimensional signal features; The second obtaining module is used to input the eye diagram data, jitter data and compensation parameters corresponding to the target signal into the second sub-network of the signal compensation model to obtain the target function output by the second sub-network. The compensation module is used to compensate the target signal based on the objective function to obtain the compensated signal of the target signal.

8. The signal compensation device according to claim 7, characterized in that, The environmental parameters include temperature, impedance mismatch, and crosstalk intensity. The compensation parameters include the pre-emphasis coefficient and the equalizer coefficient; The first obtaining module includes: The first sub-processing module is used to process the temperature, the impedance mismatch, and the crosstalk intensity through the first sub-network of the signal compensation model to obtain the pre-emphasis coefficient. The second sub-processing module is used to process the temperature, impedance mismatch, and multi-dimensional signal characteristics through the first sub-network of the signal compensation model to obtain the equalizer coefficients.

9. An electronic device, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-6.

Images