Signal processing method, device, electronic device, storage medium and program product

Abstract

The present application discloses a signal processing method, apparatus, electronic device, storage medium, and program product, relating to the field of communication technology. The method includes first sampling the signal transmission link between the signal source and the load, determining whether there is a signal reflection problem through time domain and frequency domain analysis, and if so, obtaining and executing the signal compensation strategy corresponding to the reflection feature from a pre-established database based on the reflection feature of the sampled signal. This method replaces real-time calculation with an offline database, pre-associating different reflection features with compensation strategies, eliminating the need to run complex algorithms in real time during signal transmission, and significantly reducing hardware costs and power consumption. The matching strategy is directly queried from the database based on the detected reflection feature, eliminating the delay introduced by dynamic calculation and meeting the timeliness requirements of high-frequency and high-speed scenarios.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a signal processing method, device, electronic device, storage medium and program product. Background Art

[0002] In modern electronic communications and high-speed data transmission, signal integrity issues are becoming increasingly prominent as signal rates continue to increase. Signal reflection has become a key factor seriously impacting signal quality. Research has found that signal reflections are primarily caused by impedance discontinuities in transmission lines, and impedance discontinuities are particularly prevalent in high-speed, complex signal topologies. For example, in multilayer printed circuit boards, structural features such as vias between layers, branching of transmission lines, and impedance jumps can all contribute to signal reflections. When reflected signals interfere with the original signal, they can cause signal distortion and intersymbol interference, severely compromising signal accuracy and reliability, significantly reducing system performance.

[0003] In response to the above-mentioned signal reflection problem, related technologies often attempt to use dynamic adjustment algorithms and corresponding circuits to compensate for the signal in real time. This type of dynamic compensation solution monitors signal changes in real time and adjusts the compensation parameters based on the detected reflection conditions to achieve the purpose of offsetting the impact of the reflected signal. However, dynamic compensation has extremely high requirements for hardware performance and usually requires high-performance hardware processors and complex computing circuits, which not only increases hardware costs but also leads to increased system power consumption; real-time calculation and processing processes will introduce delays. In application scenarios such as high-frequency communications and high-speed data acquisition that have strict requirements on signal processing timeliness, this delay can easily cause data loss, communication interruption and other problems, making this type of dynamic compensation solution difficult to promote and apply in the actual productization process. Therefore, it is urgent to propose a new solution to overcome the defects of high cost, high power consumption and large delay of the dynamic compensation solution in the existing technology, effectively solve the signal reflection problem, and improve signal integrity and system performance. Summary of the Invention

[0004] The present application provides a signal processing method, device, electronic device, storage medium and program product to at least solve the signal reflection problem, improve signal integrity, and reduce hardware costs and signal processing delays.

[0005] The present application provides a signal processing method, comprising: sampling on a signal transmission link between a signal source and a load to obtain a sampled signal; performing time domain analysis and frequency domain analysis on the sampled signal to determine whether the sampled signal has a signal reflection problem; if the sampled signal has a signal reflection problem, obtaining a signal compensation strategy corresponding to the reflection feature from a pre-established database based on the reflection feature of the sampled signal; and compensating the signal in the signal transmission link according to the signal compensation strategy.

[0006] The present application also provides a signal processing device, comprising:

[0007] A sampling module is used to perform sampling on the signal transmission link between the signal source and the load to obtain a sampling signal;

[0008] The signal analysis module is used to perform time domain analysis and frequency domain analysis on the sampled signal to determine whether there is a signal reflection problem in the sampled signal;

[0009] A compensation strategy acquisition module is used to obtain a signal compensation strategy corresponding to the reflection feature from a pre-established database based on the reflection feature of the sampled signal when the sampled signal has a signal reflection problem;

[0010] The signal compensation module is used to compensate the signal in the signal transmission link according to the signal compensation strategy.

[0011] The present application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for implementing the steps of the above-mentioned signal processing method when executing the computer program.

[0012] The present application also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned signal processing method are implemented.

[0013] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned signal processing method when executed by a processor.

[0014] This application first samples the signal transmission link between the signal source and the load, and uses time and frequency domain analysis to determine whether there is a signal reflection problem. If so, the signal compensation strategy corresponding to the reflection characteristics of the sampled signal is obtained from a pre-established database and executed. This method replaces real-time calculations with an offline database, pre-associating different reflection characteristics with compensation strategies. This eliminates the need to run complex algorithms in real time during signal transmission, significantly reducing hardware costs and power consumption. Matching strategies are directly queried from the database based on the detected reflection characteristics, eliminating the delay introduced by dynamic calculations and meeting the timeliness requirements of high-frequency and high-speed scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0016] Figure 1A schematic diagram of the hardware architecture relied upon for executing a signal processing method provided in an embodiment of the present application;

[0017] Figure 2A A schematic diagram of a signal processing method provided in an embodiment of the present application Figure 1 ;

[0018] Figure 2B Schematic diagram 2 of a flow chart of a signal processing method provided in an embodiment of the present application;

[0019] Figure 3 A schematic diagram of the process of establishing a database provided in an embodiment of the present application;

[0020] Figure 4 A schematic diagram of the process of training a convolutional neural network provided in an embodiment of the present application;

[0021] Figure 5 A schematic structural diagram of a signal processing device provided in an embodiment of the present application;

[0022] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0024] It should be noted that, in the description of this application, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. The terms "first," "second," etc., in this application are used to distinguish similar objects, and are not used to describe a particular order or sequence.

[0025] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0026] In conjunction with the specific application environment architecture or specific hardware architecture on which the execution of the signal processing method depends, the specific application environment architecture or specific hardware architecture is described here.

[0027] like Figure 1As shown, the hardware architecture on which the execution of the signal processing method relies includes: an experimental sample design module, a sample result learning module, a signal detection module, a signal compensation module and a signal conditioning module.

[0028] The experimental sample design module is responsible for constructing a diverse training sample set. By performing random time offsets and amplitude scaling on time-domain reflection signals, and adding noise and frequency shifts to frequency-domain signals, it simulates signal characteristics in complex scenarios. This provides subsequent modules with sample data covering multiple operating conditions, addressing the sample homogeneity issue of traditional solutions.

[0029] The Sample Result Learning Module learns the mapping between signal characteristics and compensation strategies based on the output of the Experimental Sample Design Module. It associates the topology corresponding to the standard reflection characteristics with the corresponding compensation strategies (such as matching resistors and frequency-domain notch filtering). This builds a database of reflection characteristics, topologies, and signal compensation strategies, providing a basis for decision-making in signal detection and compensation.

[0030] The signal detection module collects and analyzes the actual signal. Relying on the knowledge reserve of the sample result learning module, it identifies the signal reflection characteristics, determines the current signal status, and outputs the detection results to drive the subsequent compensation and conditioning process.

[0031] Based on the results of the signal detection module, the signal compensation module calls the compensation strategy matched by the sample result learning module and executes compensation through standardized hardware drivers. For example, if the matching resistance compensation strategy is used, the bus controls the digital potentiometer to adjust the resistance value; if the frequency domain notch filter compensation strategy is used, the filter is configured and the register parameters are modified; if the reverse pulse compensation strategy is used, the high-speed pulse generator is driven to generate pulses. This achieves precise mapping from strategy to hardware action.

[0032] Based on the results of the signal compensation module, the signal conditioning module enters a closed-loop optimization phase. If overcompensation is detected, an adaptive filtering algorithm adjusts the filter coefficients. If undercompensation is detected, a programmable gain amplifier increases the amplitude, and an equalization algorithm is used to optimize the waveform. The conditioned signal is continuously compared with the expected signal, and iterative optimization is performed until quality indicators meet requirements, achieving refined control of signal quality.

[0033] The embodiment of the present application provides a signal processing method, and the method is described in detail in conjunction with the execution flow of the signal method. Figure 2A As shown, the following steps S201 to S204 are included:

[0034] S201 : Sampling is performed on a signal transmission link between a signal source and a load to obtain a sampled signal.

[0035] In step S201, a sampling circuit is connected in series with the signal transmission link between the signal source and the load to obtain a sampled signal. The sampling circuit can be a high-speed sampling circuit with good impedance matching with the signal transmission link. The sampling frequency of the sampling circuit is determined based on the highest frequency component of the signal, following the Nyquist sampling theorem and allowing a 20% margin.

[0036] In some embodiments, the sampled signal is preprocessed, including but not limited to denoising and normalization, to improve the data quality of the sampled signal.

[0037] S202: Perform time domain analysis and frequency domain analysis on the sampled signal to determine whether the sampled signal has a signal reflection problem.

[0038] Among them, signal reflection problems include but are not limited to: low-frequency reflections caused by impedance discontinuity, narrow-band high-frequency reflections where the reflected energy is concentrated at a specific frequency point, and reflections with obvious pulse characteristics in the time domain.

[0039] In some embodiments, performing time domain analysis and frequency domain analysis on the sampled signal to determine whether the sampled signal has a signal reflection problem includes: Figure 2B As shown, a time domain analysis is performed on the sampling signal to extract the time domain reflection characteristics of the sampling signal; a frequency domain analysis is performed on the sampling signal to extract the frequency domain reflection characteristics of the sampling signal; a standard reflection characteristic is obtained from a database; and based on the time domain reflection characteristics, the frequency domain reflection characteristics and the standard reflection characteristics, it is determined whether the sampling signal has a signal reflection problem.

[0040] The above-described embodiment extracts time-domain and frequency-domain reflection features by performing time-domain and frequency-domain analysis on the sampled signal, eliminating the need for complex real-time computations and significantly reducing hardware costs and system power consumption. Furthermore, this solution transforms the reflection feature determination process into a comparison with standard reflection features in a database. Compared to traditional solutions that rely on real-time calculation of compensation parameters, this significantly reduces the amount of computation required, avoids delays caused by real-time computation, and effectively addresses data loss and communication interruptions in high-frequency communications and high-speed data acquisition scenarios. The use of a non-real-time, multi-dimensional feature analysis mechanism avoids the reliance on high-performance hardware in traditional dynamic compensation.

[0041] In terms of technical effects, through multi-dimensional feature analysis and standardized comparison, the problems of high hardware cost, high system power consumption, and data loss caused by real-time calculation delay in dynamic compensation solutions are effectively solved, and the accuracy and stability of signal reflection problem detection are improved. Time domain analysis captures the characteristics of the signal such as mutation and oscillation in the time dimension, and frequency domain analysis reveals the spectral characteristics of the reflected signal from the perspective of frequency components. The combination of the two can comprehensively and accurately identify signal reflection phenomena. By comparing with the standard reflection characteristics in the database, it is possible to determine whether the signal is reflected based on quantitative indicators, reduce human or algorithmic errors, and improve detection reliability. In addition, the non-real-time calculation characteristics of this solution enable it to adapt to low-cost hardware, while reducing costs, meeting the stringent requirements of high-frequency and high-speed scenarios for signal processing timeliness, laying the foundation for the precise implementation of subsequent signal compensation strategies, and improving the practicality of the entire signal processing system.

[0042] During time-domain analysis of the sampled signal, edge detection, peak search, and sliding window algorithms are used to extract the sampled signal's time-domain reflection characteristics. These characteristics include, but are not limited to, the time-domain reflection period and the time-domain reflection amplitude. By optimizing the signal feature extraction method, computational complexity and real-time requirements are reduced at the algorithmic level. This addresses the high cost and power consumption of dynamic compensation solutions in related technologies, which rely on high-performance hardware for real-time computing. The delays caused by real-time processing can easily lead to data transmission issues.

[0043] Edge detection algorithms can accurately capture sudden changes in signal waveforms and quickly locate the start and end points of reflected signals. Peak search algorithms can accurately identify the amplitude extremes of reflected signals. Sliding window algorithms process signal data in time segments, reducing the amount of data required for a single calculation while ensuring continuous feature extraction. These algorithms require no complex hardware architecture and can be implemented solely on a basic processor, significantly reducing the need for high-performance processors and complex arithmetic circuits, lowering hardware costs and system power consumption. Furthermore, the algorithms' inherent efficiency avoids lengthy real-time calculations and reduces signal processing latency.

[0044] In terms of technical effectiveness, the algorithm combination achieves efficient and accurate extraction of time-domain reflection features. The edge detection algorithm ensures that the time boundaries of the reflected signal are accurately marked, providing a reliable basis for calculating the time-domain reflection period. The peak search algorithm directly obtains the maximum intensity of the reflected signal and accurately determines the time-domain reflection amplitude. The sliding window algorithm, through segmented processing, enhances the algorithm's adaptability to complex signal environments and improves the stability and anti-interference capabilities of feature extraction. The precise extraction of these time-domain reflection features not only provides quantitative data support for subsequent judgment of signal reflection issues, but also lays a solid foundation for matching compensation strategies in the database, making signal compensation more accurate and efficient, thereby improving the performance and reliability of the entire signal processing system.

[0045] During frequency-domain analysis of the sampled signal, Fast Fourier Transform (FFT) and frequency-domain interpolation algorithms are used to extract frequency-domain reflection characteristics, including but not limited to the resonant peak frequency and amplitude. Specifically, the FFT transform is used to obtain the sampled signal's spectrum, calculate the reflection coefficient, and identify the frequency-domain resonant peak characteristics. By optimizing the frequency-domain analysis process, computational complexity and hardware dependency are reduced at the algorithmic level, effectively addressing the high hardware costs and latency associated with dynamic compensation solutions, which rely on complex, real-time computations.

[0046] The time-domain sampling signal is efficiently converted into a frequency-domain spectrum through the FFT transform, avoiding the complex point-by-point real-time processing of the time-domain signal in traditional dynamic compensation. The computational efficiency of the FFT algorithm is much higher than that of the direct Fourier transform, which significantly reduces the processor load. The frequency-domain interpolation algorithm further optimizes the spectral resolution, improves the feature extraction accuracy without increasing the number of original sampling points, reduces the dependence on high-sampling-rate analog-to-digital converters, and reduces hardware costs. By calculating the reflection coefficient and identifying the frequency-domain resonant peak characteristics, the judgment of the reflection problem is converted into a standardized comparison of frequency-domain characteristics, avoiding the real-time iterative calculation of compensation parameters in traditional solutions and reducing calculation delays.

[0047] In terms of technical effects, accurate diagnosis and efficient processing of signal reflection problems are achieved through frequency domain feature extraction. The FFT transform converts the reflection characteristics of the time domain signal into a resonant peak in the frequency domain, making the characteristics of the reflected signal more intuitive and easier to identify; the frequency domain interpolation algorithm improves the measurement accuracy of the resonant peak frequency and amplitude, providing more accurate data support for the matching of subsequent compensation strategies. By comparing the frequency domain reflection characteristics with the standard features in the database, the severity and type of the reflection problem can be quickly determined, and accurate compensation can be achieved. Compared with traditional time domain analysis, frequency domain analysis is more robust to noise and improves the stability of the system in complex electromagnetic environments. In addition, the computational efficiency and hardware compatibility of this solution enable it to achieve high-speed signal processing on a low-cost, low-power hardware platform.

[0048] In some embodiments, the standard reflection feature includes a standard time domain reflection feature and a standard frequency domain reflection feature. Optionally, in the process of determining whether the sampled signal has a signal reflection problem based on the time domain reflection feature, the frequency domain reflection feature, and the standard reflection feature, firstly, a first similarity between the time domain reflection feature and the standard time domain reflection feature is calculated; a second similarity between the frequency domain reflection feature and the standard frequency domain reflection feature is calculated; based on the first similarity and the second similarity, a matching degree is calculated; and reference is made to the Figure 2B As shown, when the matching degree is greater than or equal to the matching threshold, it is determined that the sampling signal has a signal reflection problem; when the matching degree is less than the matching threshold, it is determined that the sampling signal does not have a signal reflection problem.

[0049] The above-described embodiment, through a predefined standard reflection feature library, transforms the real-time detection process into a feature similarity comparison, avoiding the complex real-time modeling and parameter calculation required in traditional dynamic compensation. Specifically, the time-domain reflection features are compared with the standard time-domain features for similarity, while the frequency-domain reflection features are compared with the standard frequency-domain features. This dual-domain feature fusion judgment mechanism significantly improves the accuracy of reflection detection. By pre-building the database, the system only needs to perform simple similarity calculations during runtime rather than deriving compensation parameters in real time, reducing reliance on high-performance processors and lowering hardware costs. In addition, the low complexity of the similarity calculation keeps processing latency at the microsecond level, an order of magnitude lower than traditional solutions, effectively resolving the data loss problem in high-frequency communication scenarios.

[0050] In terms of technical effectiveness, the independent similarity calculation of dual-domain features ensures comprehensive capture of reflection characteristics. Time-domain features reflect the timing characteristics of the signal, while frequency-domain features reveal the frequency components of the signal. The combination of the two improves adaptability to complex reflection scenarios. The matching index derived by weighted fusion of dual-domain similarities enables quantitative assessment of reflection issues. The setting of matching thresholds makes the judgment process more objective and adjustable. Furthermore, the establishment of a database enables the system to self-learn, adapting to new reflection scenarios through continuous database updates, further enhancing the solution's versatility and robustness.

[0051] When calculating the first similarity between the time domain reflectance signature and the standard time domain reflectance signature, a dynamic time warping (DTW) algorithm is used to calculate the first similarity to characterize the degree of match between the time domain reflectance signature and the standard time domain reflectance signature. Optionally, the first similarity is compared with a similarity threshold to determine whether a signal reflection issue exists. The similarity threshold can be set to 0.7, which is not specifically limited in this application. If the first similarity is greater than or equal to the similarity threshold, a reflected signal is determined to be present. The severity of the signal reflection can then be determined based on the reflection coefficient and the reflection threshold.

[0052] The DTW algorithm calculates the similarity between time-domain reflectance signatures and standard signatures, improving the robustness of reflection detection. By finding the optimal nonlinear mapping path between the two time series, the DTW algorithm accurately calculates similarity even when the time-domain reflectance signatures are subject to time offset or waveform distortion. This eliminates the need for complex signal alignment preprocessing, reducing algorithmic complexity. This feature enables the system to utilize analog-to-digital converters with lower sampling rates and more basic processors for similarity calculation, reducing hardware costs. Furthermore, the DTW algorithm offers superior computational efficiency compared to traditional dynamic programming methods, keeping processing latency to microseconds while maintaining detection accuracy. This addresses the issue of data loss caused by latency in high-frequency communication scenarios. By comparing the first similarity with a similarity threshold (e.g., 0.7), rapid identification of reflection issues is achieved, eliminating the real-time iterative calculation of compensation parameters required in traditional solutions and further reducing computational complexity. The reflection coefficient is combined with the reflection severity assessment to provide a quantitative basis for subsequent compensation strategies, enhancing the targeted nature of the compensation.

[0053] In terms of technical effectiveness, the application of the DTW algorithm improves the matching accuracy of time-domain reflection signatures. The DTW algorithm's similarity calculation improves accuracy, effectively identifying weak reflection signals. By setting a reasonable similarity threshold (e.g., 0.7), the system improves reflection detection accuracy and reduces false positives in typical communication scenarios. This high-precision reflection detection provides a reliable foundation for subsequent signal compensation, increasing the matching accuracy of the compensation strategy and significantly improving signal quality.

[0054] When calculating the second similarity between the frequency domain reflection feature and the standard frequency domain reflection feature, the similarity between the frequency domain resonance peak feature and the standard frequency domain resonance peak feature is calculated using cosine similarity as the second similarity.

[0055] Cosine similarity is used to calculate the similarity between frequency-domain reflection features and standard features, avoiding the complex point-by-point comparison and real-time modeling requirements of traditional frequency-domain data. Cosine similarity measures the directional similarity between frequency-domain resonant peak features (such as frequency and amplitude vectors) and standard frequency-domain resonant peak features by calculating the cosine of the angle between two vectors. This eliminates the need for precise amplitude calibration or complex transformations of the frequency-domain data, significantly simplifying the calculation process. This algorithm requires only basic vector operations and can run efficiently on low-cost microcontrollers without relying on high-performance digital signal processors or field-programmable gate arrays, significantly reducing hardware costs and system power consumption. Furthermore, compared to the large number of operations required to calculate compensation parameters in real time in traditional dynamic compensation schemes, the rapidity of cosine similarity calculations effectively reduces processing delays. In high-frequency communication and high-speed data acquisition scenarios, this can avoid data loss and communication interruptions caused by delays, ensuring the timeliness of signal processing.

[0056] In terms of technical effectiveness, the cosine similarity algorithm achieves precise quantitative comparison of frequency-domain reflection characteristics. This algorithm is insensitive to amplitude scaling of frequency-domain data and can focus on the distribution and frequency relationships of frequency-domain resonant peaks, accurately capturing the changes in the frequency-domain characteristics of the reflected signal. Based on accurate second-order similarity calculations, the system can quickly determine signal reflection conditions and provide a precise basis for subsequent signal compensation strategies, improving the matching accuracy of compensation parameters and signal quality.

[0057] In some embodiments, after calculating the matching degree according to the first similarity and the second similarity, the method further includes: obtaining a reflection coefficient of the sampled signal; and determining that a signal reflection problem exists in the sampled signal when the reflection coefficient is greater than a reflection threshold.

[0058] The above embodiment forms a supplementary verification mechanism for similarity matching results by obtaining the reflection coefficient (a physical quantity that characterizes the signal reflection intensity) of the sampled signal and comparing it with the reflection threshold. The reflection coefficient directly reflects the degree of energy loss caused by signal reflection and has a clear physical meaning and quantitative standard. This dual-dimensional judgment mechanism, combining similarity matching with reflection coefficient verification, eliminates the need for complex hardware and can be implemented solely through software algorithm integration. This solves the problem of missed or misjudgment caused by feature distortion in traditional solutions without increasing hardware costs. Furthermore, the introduction of the reflection coefficient provides a quantitative basis for reflection severity, avoiding the problem of over- or under-compensation caused by the inability to accurately assess reflection levels in traditional dynamic compensation. This improves the targeted nature of the compensation strategy and indirectly reduces the hardware resource consumption and power consumption caused by ineffective compensation. Technically, the fusion of reflection coefficient and similarity significantly improves the reliability and accuracy of reflection detection. The quantitative nature of the reflection coefficient enables the system to perform graded processing based on different reflection intensities: prioritizing strong compensation strategies for high-reflection-coefficient signals and employing lightweight compensation strategies for low-reflection-coefficient signals. This avoids the resource waste caused by the one-size-fits-all compensation strategy of traditional solutions and reduces hardware power consumption.

[0059] S203 : When there is a signal reflection problem in the sampled signal, a signal compensation strategy corresponding to the reflection feature is obtained from a pre-established database according to the reflection feature of the sampled signal.

[0060] The pre-established database includes signal compensation strategies corresponding to reflection characteristics. These signal compensation strategies involve the applicable conditions and compensation parameters of various signal compensation strategies, including but not limited to matching resistor compensation strategy, frequency domain notch filter compensation strategy, and reverse pulse compensation strategy. The matching resistor compensation strategy automatically adjusts the resistance value of the matching resistor using a digital potentiometer. The frequency domain notch filter compensation strategy uses real-time configuration of filter parameters to form a notch at the resonant frequency of the reflected signal, suppressing the reflected energy. The reverse pulse compensation strategy uses a high-speed pulse generator to generate a reverse pulse based on calculated parameters, which is superimposed on the reflected pulse to offset the impact of the reflected signal.

[0061] In some embodiments, when a signal reflection problem exists in the sampled signal and the severity of the signal reflection problem is greater than a threshold, a signal compensation strategy corresponding to the reflection feature is obtained from a pre-established database based on the reflection feature of the sampled signal.

[0062] Among them, the database establishment process, such as Figure 3 As shown, the following steps S301 to S304 are included:

[0063] S301: Acquire multiple sample signals with signal reflection problems.

[0064] Among them, multiple sample signals have signal reflection problems and the topological structures corresponding to each sample signal are different, such as star, tree, daisy chain and other circuit topologies.

[0065] When acquiring multiple sample signals with signal reflection issues, printed circuit board (PCB) design software is used to generate circuit board diagrams based on different topological parameters. Based on these diagrams, the circuit board under test (CUT) for testing signal reflection issues can be manufactured using high-precision PCB fabrication techniques (e.g., line width accuracy of ±5μm and impedance control accuracy of ±3Ω). The CUT is then connected to a network analyzer and a high-speed oscilloscope. The network analyzer collects the CUT's scattering parameters, including the first reflection coefficient S11, transmission coefficient S21, reverse transmission coefficient S12, and second reflection coefficient S22. The high-speed oscilloscope then acquires the CUT's time-domain reflection waveform.

[0066] Optionally, calibrate the network analyzer and high-speed oscilloscope before collecting data, and set operating parameters including but not limited to the test frequency band and number of sweep points. You can also collect data multiple times under different environmental conditions, such as varying temperatures (20°C-60°C, in 5°C increments) and humidity (30%-70%, in 10% increments). This allows you to obtain signal feature data under different environmental factors, enriching your sample.

[0067] Multi-topology circuit board diagrams are generated through PCB design software, and high-precision processing technology is used to manufacture the circuit boards to be tested, ensuring the authenticity and diversity of the sample signals from a hardware level. By adjusting the topology parameters, various topologies such as star, tree, and differential pair wiring are generated to cover most actual circuit scenarios, solving the problem of insufficient topological diversity. At the same time, a network analyzer is used to collect scattering parameters and a high-speed oscilloscope is used to collect time domain waveforms. Combined with the calibration process and repeated collection under multiple environmental conditions, data deviations caused by environmental factors are avoided. By sampling across all environmental dimensions, the sample data can reflect signal fluctuations in actual applications, improving the adaptability of the database in complex environments. In addition, high-precision processing technology and instrument calibration ensure the accuracy of data collection such as reflection coefficients and time domain waveforms, avoiding feature extraction errors caused by hardware errors, and indirectly reducing the matching errors of subsequent compensation strategies.

[0068] Among them, different topology parameters are set, such as the number of impedance discontinuity segments (1-8 discrete values), the length of each discontinuity segment (0.1mm-10mm, step size 0.5mm), the impedance value (30Ω-150Ω, interval 10Ω), and the number of vias, branch angle and other factors.

[0069] The number of discrete impedance discontinuities covers scenarios ranging from single-point reflection to multi-point cascade reflection. The millimeter-level precision of discontinuity lengths and impedance values ​​enables the sample to characterize the reflection characteristics of fine structures such as microstrip lines and via stubs, avoiding feature aliasing caused by insufficient parameter granularity. The introduction of key parameters of actual circuits, such as the number of vias and branching angles, overcomes the shortcomings of traditional simulations that ignore the influence of three-dimensional structures. This parameter setting mechanism generates diverse topological combinations by combining discrete values ​​of different dimensions, avoiding the problem of poor adaptability of compensation strategies caused by insufficient sample diversity in the database. When collecting the scattering parameters of the circuit board under test using a network analyzer, the network analyzer is controlled to collect scattering coefficients and time-domain reflection waveforms within a wide frequency band of 10MHz-50GHz. The wide frequency band of 10MHz to 50GHz covers typical application scenarios, from low-frequency digital signals to high-frequency RF signals. Scattering parameter acquisition in high-frequency bands (e.g., above 10GHz) can capture reflection characteristic variations caused by high-frequency characteristics, preventing compensation strategies from failing due to missing frequency bands. Time-domain reflection waveform analysis over a wide frequency band can achieve finer time-domain resolution through inverse Fourier transform. Continuous wide-band scanning ensures the integrity of spectral characteristics, avoiding the omission of resonant peaks caused by large frequency band spacing, and providing accurate frequency points for subsequent strategies such as frequency-domain notch filtering.

[0070] S302: Extract standard reflection features of multiple sample signals.

[0071] In some embodiments, a plurality of sample signals are first pre-processed, including but not limited to denoising and normalization, to improve the quality of the sample signal data.

[0072] In some embodiments, when extracting standard reflection features of multiple sample signals, the time domain reflection waveforms of the multiple sample signals are first decomposed into sub-signals of different scales to extract the standard time domain reflection features; the time domain reflection waveforms are converted to the frequency domain to extract the standard frequency domain reflection features.

[0073] The multi-scale time-domain decomposition in the above-mentioned embodiment simultaneously captures both global trends and local mutations in the reflected signal, converting the time-domain waveform to the frequency domain and extracting resonant peak features. This overcomes the limitation of single-time-domain analysis, which cannot directly reflect frequency components. The combination of the two forms a three-dimensional feature space of time, frequency, and scale, comprehensively covering the time-frequency characteristics of the reflected signal. This cross-domain feature extraction mechanism leverages the multi-resolution analysis capabilities of the wavelet transform to suppress ambient noise while preserving signal details, reducing feature extraction errors.

[0074] When decomposing the time domain reflection waveforms of multiple sample signals into sub-signals of different scales, wavelet transform can be used to decompose the time domain reflection waveforms into sub-signals of different scales, and the extracted standard time domain reflection features include but are not limited to the timestamp of the reflection position, the peak amplitude of the reflection pulse, and the reflection period.

[0075] In terms of technical effectiveness, multi-scale analysis using wavelet transforms significantly improves the resolution and reliability of time-domain reflection features. The time-frequency localization of wavelet transforms allows for simultaneous location of the reflection timestamp and capture of the peak amplitude of the reflected pulse. Multi-scale decomposition separates high-frequency noise from low-frequency reflection features, suppresses environmental interference through threshold denoising, reduces feature extraction errors, and avoids misjudgment of compensation parameters due to noise. Reflection period extraction is achieved through inter-scale feature correlation, accurately characterizing periodic reflections.

[0076] When converting the time domain reflection waveform to the frequency domain, the signal is converted to the frequency domain through fast Fourier transform, and the standard frequency domain reflection characteristics such as the frequency value, amplitude and bandwidth of the resonance peak in the S11 curve are extracted.

[0077] In terms of technical effectiveness, the S11 curve after FFT transformation can intuitively display the frequency domain resonance peak characteristics of the reflection signal. By extracting the frequency value, amplitude, and bandwidth of the resonance peak, the frequency domain characteristics of impedance mismatch or structural defects can be accurately characterized. The standardized extraction of frequency domain features avoids the subjectivity of manual analysis, equips the frequency domain features in the database with a unified quantitative standard, and improves the accuracy of subsequent strategy matching. FFT transformation and frequency domain feature extraction significantly improve the diagnostic accuracy of reflection problems and the adaptability of compensation strategies.

[0078] S303: Determine the topological structure and signal compensation strategy corresponding to the standard reflection feature based on the standard reflection feature and the pre-trained convolutional neural network model.

[0079] The pre-trained convolutional neural network can infer the corresponding topological structure and signal compensation strategy based on the input reflection characteristics. It can reflect the nonlinear mapping relationship between reflection characteristics and topological structure, and determine the signal compensation strategy that can solve the signal reflection problem corresponding to different reflection characteristics.

[0080] In some embodiments, the training process of the convolutional neural network model, such as Figure 4 As shown, the following steps S401 to S405 are included:

[0081] S401: Obtain training samples.

[0082] The training samples include reflection signals and corresponding annotated topological structures and signal compensation strategies.

[0083] Signal compensation strategies were developed through a combination of simulation and experimentation to address reflection issues arising from different topologies. Specifically, models of resistors, filters, and high-frequency pulse generators were built in the simulator. By adjusting key parameters such as resistor values, filter center frequencies, bandwidth parameters, pulse amplitudes, and delays, these models were fitted to experimental results to confirm the topology and compensation parameters used in the compensation strategy.

[0084] The simulator builds hardware models of resistors, filters, and high-frequency pulse generators to simulate the characteristics of real circuit components. By adjusting variables such as resistance, frequency, bandwidth, and pulse parameters, reflection scenarios across multiple topologies can be covered. Simulation parameters are then matched to experimental results to verify the feasibility of the compensation strategy in actual hardware. This ensures that the compensation strategy is both consistent with the physical model and adaptable to real hardware environments.

[0085] Optionally, the process of obtaining training samples includes: obtaining original training samples, which include time domain reflection signals and frequency domain reflection signals; performing time domain enhancement on the time domain reflection signals to obtain time domain samples; performing frequency domain enhancement on the frequency domain reflection signals to obtain frequency domain samples; and using the time domain samples and frequency domain samples as training samples. When performing time domain enhancement on the time domain reflection signals, random time offset and amplitude scaling are performed on the original time domain data. When performing frequency domain enhancement on the frequency domain reflection signals, random noise is added and frequency shift is performed.

[0086] The random time offset in time domain enhancement simulates the differences in reflection delay caused by different topologies, and the amplitude scaling covers the dynamic range of reflection intensity, which can characterize the fluctuations in reflection position and intensity in actual circuits. The random noise injection in frequency domain enhancement simulates the influence of environmental factors such as electromagnetic interference, and the frequency shift operation covers the scenario of resonance peak shift caused by temperature drift, which makes up for the deficiency of traditional samples that do not consider environmental disturbances. This data enhancement mechanism allows a single sample to be transformed to generate dozens of times more samples, thereby increasing the training sample size.

[0087] This optional implementation constructs a large and diverse training sample by performing enhancement operations such as random time shifting and amplitude scaling on the original time-domain reflection signals, and by adding random noise and frequency shifting to the frequency-domain reflection signals. This diverse enhancement enables the model to learn more robust reflection features. This time- and frequency-domain data augmentation improves the model's anti-interference capabilities and generalization performance.

[0088] S402: Divide the training samples into a training set, a validation set, and a test set.

[0089] The training samples are divided into training set, validation set and test set. Through scientific data set division, model overfitting is avoided, solving the problem of low sample utilization in traditional solutions.

[0090] S403. Build a basic model based on the machine learning framework.

[0091] Among them, the initial model parameters of the basic model are the converged model parameters of other fields of transfer learning.

[0092] Optionally, use Python and the TensorFlow framework to build a basic model by setting the number of network layers, convolution kernel size, pooling layer parameters, etc.

[0093] A basic model is built based on the machine learning framework, and transfer learning is used to load the initial parameters of convergent models in other fields, replacing the high computational cost mode of training from scratch, shortening the model training time, and at the same time leveraging the feature extraction capabilities of the pre-trained model to improve the ability to abstract the time-frequency characteristics of the reflected signal.

[0094] S404: Adjust the initial model parameters according to the loss function values ​​of the training set and the validation set.

[0095] Optionally, the Adaptive Moment Estimation (Adam) optimizer is used to update the initial model parameters, adjusting the learning rate based on the loss function values ​​of the training and validation sets. The initial model parameters are iteratively optimized using the loss function of the training and validation sets. When the loss function value of the validation set converges, the model generalization ability is evaluated using the test set. This standardized training process avoids the subjectivity of manual parameter adjustment and enables the model to automatically learn the mapping relationship between reflective features, topology, and policy.

[0096] S405. When the loss function value of the validation set converges, the model is evaluated based on the test set to obtain a convolutional neural network model.

[0097] When the loss function value of the validation set no longer decreases significantly, the model training is stopped and the test set is used to evaluate the current model to obtain a converged convolutional neural network model.

[0098] This training convolutional neural network model overcomes the limitations of manual feature engineering by acquiring training samples that annotate reflection signals, topological structures, and compensation strategies, combined with transfer learning and a phased training mechanism. This combination of transfer learning and a convolutional neural network (CNN) model significantly improves the accuracy of reflection feature recognition and the efficiency of database construction.

[0099] The trained convolutional neural network model takes the standard reflection features of multiple sample signals as input and outputs the topological structure and signal compensation strategy corresponding to the standard reflection features.

[0100] S304: Store the standard reflection characteristics of the plurality of sample signals, the corresponding topological structures and the signal compensation strategies in a database.

[0101] The converged convolutional neural network model and compensation algorithm are burned into the load-side chip in the form of firmware, such as a field-programmable gate array (FPGA) or an application-specific integrated circuit (ASIC) chip.

[0102] The database is established, storing the corresponding standard reflection characteristics, topology, and signal compensation strategies. Before actual signal processing, the signal compensation strategies are already stored in the database. During actual signal processing, the corresponding signal compensation strategies are directly retrieved from the database, reducing hardware processing burden and signal processing latency.

[0103] By acquiring sample signals from multiple topologies (each with a different topology), the CNN model is used to automatically extract standard reflection features and associate corresponding topologies with compensation strategies. This automated construction mechanism reduces database maintenance costs and addresses the poor adaptability of compensation strategies in traditional solutions due to incomplete topology coverage. Furthermore, the CNN model's deep feature abstraction capabilities enable the database to adapt to new topologies, avoiding the need to redesign compensation strategies for unknown scenarios and indirectly reducing system upgrade costs. The database construction process can cover a variety of topologies and signal reflection scenarios, enabling the system to adapt to different signal transmission environments and effectively address signal reflection issues.

[0104] In some embodiments, when executing step S203, the topology corresponding to the standard reflection feature is first determined from a database; then, a signal compensation strategy corresponding to the topology is determined. The association between the reflection feature and the topology is based on the actual physical characteristics of the circuit, making the selection of the compensation strategy physically explainable and avoiding the blind strategy matching in traditional solutions. The topology serves as an intermediate mapping layer, integrating compensation experience from similar topologies, thus resolving the problem of misjudgment of compensation due to topological differences in similar features in traditional solutions.

[0105] By first determining the topological structure corresponding to the standard reflection characteristics from the database and then matching the signal compensation strategy associated with the topological structure, a hierarchical mapping logic is constructed. This hierarchical mapping does not require high-performance hardware to calculate the strategy parameters in real time; it can be completed only through database queries, avoiding the microsecond delay introduced by real-time calculations and solving the problem of data loss in high-frequency scenarios.

[0106] In terms of technical effectiveness, this application, based on a pre-established database, shifts the complexity of real-time calculations in traditional dynamic compensation to the preprocessing stage. The database stores pre-calculated results, and real-time processing only requires feature matching to invoke a signal compensation strategy that can resolve signal reflection issues. This avoids complex real-time calculations and addresses hardware cost and latency issues. Experimental data shows that compared to traditional dynamic compensation solutions, this application reduces hardware resource consumption by over 60% and signal processing latency by over 80%, significantly improving the system's practicality and efficiency, making it more suitable for large-scale applications.

[0107] S204: Compensate the signal in the signal transmission link according to the signal compensation strategy.

[0108] For example, assuming that the signal compensation strategy is a matching resistor compensation strategy, the resistance adjustment of the digital potentiometer is controlled by the bus to adjust the matching resistor in real time; assuming that the signal compensation strategy is a frequency domain notch filter compensation strategy, the parameters of the filter are changed by configuring the control register of the filter; assuming that the signal compensation strategy is a reverse pulse compensation strategy, a control signal is generated to drive the high-speed pulse generator to generate a reverse pulse.

[0109] A standardized hardware-driven process is developed based on different compensation strategies (matching resistors, frequency-domain notch filtering, and reverse pulses). This direct mapping mechanism between strategy and hardware eliminates the need for complex intermediate calculations and can be executed by querying the control parameters corresponding to the strategy in the database. This solves the high hardware cost and latency issues caused by real-time calculations in traditional solutions, and improves the efficiency and reliability of compensation execution.

[0110] In some embodiments, after compensating the sampled signal according to the signal compensation strategy, resampling is performed to obtain a resampled signal; a quality index of the resampled signal is calculated; a signal compensation condition is determined based on the quality index; and signal conditioning is performed based on the signal compensation condition until the quality index meets the requirements.

[0111] During resampling to obtain a resampled signal, an analog-to-digital converter (ADC) performs quantization sampling to obtain the resampled signal. A digital signal processor (DSP) chip then calculates quality metrics for the resampled signal. Quality metrics for the resampled signal include, but are not limited to, signal-to-noise ratio (SNR) and bit error rate (BER).

[0112] The above embodiment captures the compensation effect in real time by resampling and quantizing the compensated signal using an ADC; calculates the quality index and compares it with the threshold to automatically determine whether the compensation meets the standard; it iteratively adjusts the compensation parameters based on the evaluation results until the quality index meets the requirements, forming an adaptive optimization mechanism to achieve dynamic optimization of the compensation effect and improve signal quality and system robustness.

[0113] Optionally, signal conditioning is performed according to the signal compensation situation, including: in the case of over-compensation of the signal, adjusting the filter coefficients in the signal compensation strategy; in the case of under-compensation of the signal, adjusting the signal amplitude and equalizer coefficients in the signal compensation strategy.

[0114] Among them, signal overcompensation includes signal oscillation or distortion, while signal undercompensation includes signal quality that does not meet the expected standard.

[0115] Specifically, in the case of overcompensated signals, an adaptive filtering algorithm adjusts the filter coefficients to suppress excess compensation signals. In the case of undercompensated signals, a programmable gain amplifier (PGA) appropriately boosts the signal amplitude, and an equalization algorithm is used to further optimize the signal waveform. After signal conditioning, the signal is compared with the original desired signal until quality indicators meet requirements. By identifying overcompensated and undercompensated states, differentiated conditioning is performed using strategies such as adaptive filtering and programmable gain amplification. This state-based differentiated conditioning mechanism quantifies compensation deviations using quality indicators, refines parameter adjustment steps, and avoids delays caused by repeated adjustments, improving signal integrity and system adaptability. This signal conditioning creates a closed-loop feedback loop from detection, compensation, and optimization. Through continuous adjustment and optimization, signal quality is optimized, effectively improving signal integrity and system stability, and ensuring reliable signal transmission.

[0116] In summary, the present application provides a signal processing method, which first samples the signal transmission link between the signal source and the load to obtain a sampled signal; then performs time domain analysis and frequency domain analysis on the sampled signal to determine whether there is a signal reflection problem. If there is a signal reflection problem, the signal compensation strategy corresponding to the reflection feature is obtained from a pre-established database based on the reflection feature of the sampled signal; finally, the signal in the signal transmission link is compensated according to the signal compensation strategy. This method reduces hardware dependence and delay through a non-real-time computing architecture based on a pre-built database. In the reflection feature sampling and analysis link, only basic sampling circuits and analysis modules are required for link sampling and analysis, and no high-performance processor is required for real-time calculation, which effectively reduces hardware cost and power consumption; in the database matching compensation strategy link, a reflection feature and compensation strategy database is pre-established, and the corresponding strategy is directly called according to the detected reflection feature, avoiding real-time complex calculations, converting the calculation into a database query, and solving the timeliness problem.

[0117] At the same time, this method adopts a phased processing mechanism to reduce the amount of real-time calculations and optimize power consumption and latency. In the preprocessing stage, a database is established through extensive testing in advance, binding different reflection characteristics with compensation strategies. This eliminates the need to run complex algorithms in real time during signal transmission, reducing processor performance requirements and power consumption. In the real-time processing stage, only a simple process of sampling, analysis, querying, and compensation is executed. The core calculations are completed in the preprocessing stage, so the real-time calculations are extremely small and the latency can be controlled to the nanosecond level, making it suitable for high-frequency and high-speed scenarios and avoiding data loss.

[0118] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method.

[0119] The embodiment of the present application also provides a signal processing device, such as Figure 5 As shown, the device includes:

[0120] The sampling module 501 is used to perform sampling on the signal transmission link between the signal source and the load to obtain a sampled signal;

[0121] The signal analysis module 502 is used to perform time domain analysis and frequency domain analysis on the sampled signal to determine whether there is a signal reflection problem in the sampled signal;

[0122] The compensation strategy acquisition module 503 is used to acquire a signal compensation strategy corresponding to the reflection feature from a pre-established database according to the reflection feature of the sampled signal when the sampled signal has a signal reflection problem;

[0123] The signal compensation module 504 is configured to compensate the signal in the signal transmission link according to the signal compensation strategy.

[0124] As an optional implementation in the embodiment of the present application, the signal analysis module 502 is specifically used to: perform time domain analysis on the sampled signal to extract the time domain reflection characteristics of the sampled signal; perform frequency domain analysis on the sampled signal to extract the frequency domain reflection characteristics of the sampled signal; obtain standard reflection characteristics from the database; and determine whether the sampled signal has a signal reflection problem based on the time domain reflection characteristics, frequency domain reflection characteristics and standard reflection characteristics.

[0125] As an optional implementation in an embodiment of the present application, the standard reflection feature includes: a standard time domain reflection feature and a standard frequency domain reflection feature; the signal analysis module 502 is specifically used to: calculate a first similarity between the time domain reflection feature and the standard time domain reflection feature; calculate a second similarity between the frequency domain reflection feature and the standard frequency domain reflection feature; calculate a matching degree based on the first similarity and the second similarity; if the matching degree is greater than or equal to a matching threshold, determine that a signal reflection problem exists in the sampled signal; if the matching degree is less than the matching threshold, determine that no signal reflection problem exists in the sampled signal.

[0126] As an optional implementation in the embodiment of the present application, the signal analysis module 502 is further used to: obtain a reflection coefficient of the sampling signal; and determine that a signal reflection problem exists in the sampling signal when the reflection coefficient is greater than a reflection threshold.

[0127] As an optional implementation in the embodiment of the present application, the compensation strategy acquisition module 503 is specifically used to: determine the topological structure corresponding to the standard reflection feature from the database; and determine the signal compensation strategy corresponding to the topological structure.

[0128] As an optional implementation in the embodiment of the present application, the device also includes a database establishment module, which is used to: obtain multiple sample signals with signal reflection problems, where the multiple sample signals have signal reflection problems and the topological structures corresponding to the various sample signals are different; extract standard reflection features of the multiple sample signals; determine the topological structure and signal compensation strategy corresponding to the standard reflection features based on the standard reflection features and a pre-trained convolutional neural network model; and store the standard reflection features of the multiple sample signals and the corresponding topological structures and signal compensation strategies in a database.

[0129] As an optional implementation in the embodiment of the present application, the database establishment module is specifically used to: decompose the time domain reflection waveforms of multiple sample signals into sub-signals of different scales, and extract standard time domain reflection features; convert the time domain reflection waveforms into the frequency domain, and extract standard frequency domain reflection features.

[0130] As an optional implementation in the embodiment of the present application, the device also includes a model training module, which is used to: obtain training samples, the training samples include reflected signals and corresponding annotated topological structures and signal compensation strategies; divide the training samples into training sets, validation sets and test sets; build a basic model based on a machine learning framework, and the initial model parameters of the basic model are the converged model parameters of transfer learning in other fields; adjust the initial model parameters according to the loss function values ​​of the training set and the validation set; when the loss function value of the validation set converges, perform model evaluation according to the test set to obtain a convolutional neural network model.

[0131] As an optional implementation in the embodiment of the present application, the model training module is specifically used to: obtain original training samples, which include time domain reflection signals and frequency domain reflection signals; perform time domain enhancement on the time domain reflection signals to obtain time domain samples; perform frequency domain enhancement on the frequency domain reflection signals to obtain frequency domain samples; and use the time domain samples and frequency domain samples as training samples.

[0132] As an optional implementation in the embodiment of the present application, the device also includes a signal conditioning module, which is used to: after compensating the sampled signal according to the signal compensation strategy, resample to obtain a resampled signal; calculate the quality index of the resampled signal; determine the signal compensation status based on the quality index; and perform signal conditioning based on the signal compensation status until the quality index meets the requirements.

[0133] As an optional implementation in the embodiment of the present application, the signal conditioning module is specifically used to: adjust the filter coefficients in the signal compensation strategy when the signal is over-compensated; and adjust the signal amplitude and equalizer coefficients in the signal compensation strategy when the signal is under-compensated.

[0134] For the description of the features in the embodiments corresponding to the signal processing device, reference can be made to the relevant description of the embodiments corresponding to the signal processing method, which will not be repeated here.

[0135] An embodiment of the present application further provides an electronic device, including a memory 601 and a processor 602, wherein the memory 601 stores a computer program, and the processor 602 is configured to run the computer program to perform the steps in any of the above signal processing method embodiments.

[0136] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored. The computer program is configured to execute the steps of any one of the above-mentioned signal processing method embodiments when run.

[0137] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0138] An embodiment of the present application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps in any one of the above signal processing method embodiments are implemented.

[0139] An embodiment of the present application further provides another computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above-mentioned signal processing method embodiments are implemented.

[0140] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0141] The above is a detailed introduction to a signal processing method, device, electronic device, storage medium and program product provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and core ideas of the present application. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of the present application.

Claims

1. A signal processing method, characterized in that: include: Sampling is performed on the signal transmission link between the signal source and the load to obtain a sampling signal; Performing time domain analysis and frequency domain analysis on the sampled signal to determine whether the sampled signal has a signal reflection problem; In the case where the sampled signal has a signal reflection problem, obtaining a signal compensation strategy corresponding to the reflection feature from a pre-established database according to the reflection feature of the sampled signal; Compensating the signal in the signal transmission link according to the signal compensation strategy; The performing time domain analysis and frequency domain analysis on the sampled signal to determine whether the sampled signal has a signal reflection problem includes: performing time domain analysis on the sampled signal to extract a time domain reflection feature of the sampled signal; performing frequency domain analysis on the sampled signal to extract a frequency domain reflection feature of the sampled signal; obtaining a standard reflection feature from the database; and determining whether the sampled signal has a signal reflection problem based on the time domain reflection feature, the frequency domain reflection feature, and the standard reflection feature. The method also includes: obtaining multiple sample signals with signal reflection problems, wherein the multiple sample signals have signal reflection problems and the topological structures corresponding to the respective sample signals are different; extracting standard reflection features of the multiple sample signals; determining the topological structures and signal compensation strategies corresponding to the standard reflection features based on the standard reflection features and a pre-trained convolutional neural network model; and storing the standard reflection features of the multiple sample signals, the corresponding topological structures and the signal compensation strategies in the database.

2. The method according to claim 1, characterized in that The standard reflection characteristics include: standard time domain reflection characteristics and standard frequency domain reflection characteristics; The determining, based on the time domain reflection feature, the frequency domain reflection feature, and the standard reflection feature, whether the sampled signal has a signal reflection problem includes: Calculating a first similarity between the time domain reflectometry feature and the standard time domain reflectometry feature; Calculating a second similarity between the frequency domain reflection feature and the standard frequency domain reflection feature; Calculating a matching degree according to the first similarity and the second similarity; When the matching degree is greater than or equal to a matching threshold, determining that a signal reflection problem exists in the sampling signal; When the matching degree is less than the matching threshold, it is determined that there is no signal reflection problem in the sampling signal.

3. The method according to claim 2, characterized in that After calculating the matching degree according to the first similarity and the second similarity, the method further includes: Obtaining a reflection coefficient of the sampled signal; When the reflection coefficient is greater than the reflection threshold, it is determined that a signal reflection problem exists in the sampling signal.

4. The method according to claim 1, wherein The acquiring, based on the reflection characteristic of the sampled signal, a signal compensation strategy corresponding to the reflection characteristic from a pre-established database includes: determining a topological structure corresponding to the standard reflection feature from the database; A signal compensation strategy corresponding to the topology is determined.

5. The method according to claim 1, wherein The extracting standard reflection features of the plurality of sample signals includes: Decomposing the time domain reflection waveforms of the multiple sample signals into sub-signals of different scales and extracting standard time domain reflection features; The time domain reflection waveform is converted to the frequency domain, and the standard frequency domain reflection feature is extracted.

6. The method according to claim 1, characterized in that The training process of the convolutional neural network model includes: Acquire a training sample, wherein the training sample includes a reflection signal and a corresponding annotated topological structure and a signal compensation strategy; Dividing the training samples into a training set, a validation set, and a test set; Building a basic model based on a machine learning framework, where the initial model parameters of the basic model are the convergent model parameters of other fields of transfer learning; Adjusting the initial model parameters according to the loss function values ​​of the training set and the validation set; When the loss function value of the validation set converges, a model evaluation is performed based on the test set to obtain the convolutional neural network model.

7. The method according to claim 6, characterized in that The obtaining of training samples includes: Obtaining original training samples, where the original training samples include time domain reflection signals and frequency domain reflection signals; Performing time domain enhancement on the time domain reflection signal to obtain a time domain sample; Performing frequency domain enhancement on the frequency domain reflected signal to obtain a frequency domain sample; The time domain samples and the frequency domain samples are used as the training samples.

8. The method according to claim 1, characterized in that The method further comprises: After compensating the sampled signal according to the signal compensation strategy, resampling to obtain a resampled signal; Calculating a quality indicator of the resampled signal; determining a signal compensation condition according to the quality indicator; Signal conditioning is performed according to the signal compensation condition until the quality index meets the requirement.

9. The method according to claim 8, characterized in that Performing signal conditioning according to the signal compensation condition includes: In the event of signal overcompensation, adjusting the filter coefficients in the signal compensation strategy; In the case of under-compensation of the signal, the signal amplitude and equalizer coefficients in the signal compensation strategy are adjusted.

10. A signal processing device, characterized in that: include: A sampling module is used to perform sampling on the signal transmission link between the signal source and the load to obtain a sampling signal; A signal analysis module is used to perform time domain analysis and frequency domain analysis on the sampled signal to determine whether the sampled signal has a signal reflection problem; a compensation strategy acquisition module, configured to, when a signal reflection problem exists in the sampled signal, acquire, from a pre-established database, a signal compensation strategy corresponding to the reflection feature according to the reflection feature of the sampled signal; A signal compensation module, configured to compensate the signal in the signal transmission link according to the signal compensation strategy; The signal analysis module is specifically used to perform time domain analysis on the sampled signal to extract the time domain reflection characteristics of the sampled signal; perform frequency domain analysis on the sampled signal to extract the frequency domain reflection characteristics of the sampled signal; Obtaining a standard reflection feature from the database; determining whether the sampled signal has a signal reflection problem based on the time domain reflection feature, the frequency domain reflection feature, and the standard reflection feature; A database establishment module is used to obtain multiple sample signals with signal reflection problems, where the multiple sample signals have signal reflection problems and the topological structures corresponding to the various sample signals are different; extract standard reflection features of the multiple sample signals; determine the topological structures and signal compensation strategies corresponding to the standard reflection features based on the standard reflection features and a pre-trained convolutional neural network model; and store the standard reflection features of the multiple sample signals, along with the corresponding topological structures and signal compensation strategies, in the database.

11. An electronic device, characterized in that: include: memory for storing computer programs; A processor, configured to implement the steps of the signal processing method according to any one of claims 1 to 9 when executing the computer program.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of the signal processing method according to any one of claims 1 to 9.

13. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the signal processing method according to any one of claims 1 to 9 are implemented.

Images