Capacitive coupled dynamic encoding isolation method and system

Abstract

The application discloses a kind of capacitive coupling dynamic encoding isolation method and system, through the dual mechanism of dynamic encoding mapping and adaptive filtering reconstruction, the influence of parasitic pulse string on signal integrity is significantly reduced, ensure that the zero error rate and high reliability of data transmission in high voltage transient environment, while maintaining the high response speed of system, avoid the signal delay caused by traditional filtering means, greatly improve the running stability of power electronic system in complex electromagnetic environment, effectively solve the problem that receiving end misjudges logic level due to fixed threshold and simple filtering cannot cope with dynamic common-mode transient voltage interference in prior art.

Description

Technical Field

[0001] This invention relates to the field of power electronic signal isolation technology, specifically to a capacitive coupling dynamic coding isolation method and system. Background Technology

[0002] In power electronic converters and high-voltage drive systems, signal isolation is a crucial element in ensuring the safe and reliable operation of control circuits. Capacitive coupling isolation technology, due to its advantages such as fast high-frequency response, small size, and high transmission rate, is widely used in the gate drive signal transmission of power devices such as IGBTs and SiC MOSFETs. Traditional capacitive isolation structures typically employ simple digital encoding (such as NRZ codes) combined with a fixed receiver threshold decision mechanism for signal transmission. Its basic working principle is to transmit the logic level from the transmitting end to the receiving end through a capacitive coupling channel. The receiving end uses a comparator to compare the analog voltage signal with a preset fixed reference voltage, thereby recovering the original digital logic signal.

[0003] However, this traditional approach has significant drawbacks when dealing with rapidly changing common-mode transient voltages on the high-voltage side. When power devices (such as IGBTs) turn on or off, they generate extremely high voltage change rates (dv / dt). These drastically changing common-mode transient voltages couple into the isolation channel through parasitic capacitance, inducing high-frequency parasitic pulse trains at the receiver. Because traditional receivers use fixed threshold decisions, these spurious pulses caused by common-mode interference are often misinterpreted as valid logic level flips, leading to incorrect logic states at the receiver output. Furthermore, existing anti-interference methods largely rely on simple low-pass filtering or increasing dead time. While this can filter out some high-frequency noise, it inevitably introduces signal delay, reduces system response speed, and is difficult to adapt to dynamically changing electromagnetic environments. Once the interference frequency components change or the intensity exceeds the filter's design range, the system is highly susceptible to malfunctions, leading to serious control failures or even equipment damage. Summary of the Invention

[0004] The present invention aims to provide a capacitively coupled dynamic coding isolation method and system, which realizes reliable signal isolation with high bandwidth and low bit error rate under strong common-mode transient interference environment.

[0005] To achieve the above objectives, the technical solution adopted by this invention is: a capacitively coupled dynamic coding isolation method, comprising:

[0006] Simultaneously acquire the common-mode transient voltage waveform and the output response of the coupling channel on the high-voltage side, calculate the cross-correlation function to extract the spatiotemporal correlation features, construct a multi-band noise energy distribution map and identify the dominant interference frequency band, and generate a reference noise spectrum vector;

[0007] Extract real-time signal segments and perform short-time energy weighted statistics, calculate real-time feature fingerprint vectors and perform Euclidean distance measurement with the benchmark noise spectrum vector, determine the interference level based on the distance measurement results and update the interference confidence, and generate a dynamic interference fingerprint matching report.

[0008] The dynamic interference fingerprint matching report is analyzed and the optimal coding template under the current working condition is extracted. Based on the optimal coding template, the original logic signal is preprocessed by grouping and polarity reversal. A dynamic Manchester coding variant is introduced and the instantaneous coding symbol sequence is calculated to generate the initial coding mapping data packet.

[0009] Extract the original waveform from the receiving end and perform adaptive filtering based on the frequency band mask, calculate the instantaneous signal-to-noise ratio distribution map and determine the dynamic decision threshold sequence, perform dual-edge triggered logic reconstruction and remove single-cycle abnormal pulses, encapsulate the reconstructed logic frame and verify the consistency of the encoding template;

[0010] Parse the redundant check bits in the logical frame and calculate the Hamming distance matrix. Select an adaptive error correction strategy based on the Hamming distance value and perform correction operations. Perform multi-source cross-validation and generate a data credibility score. Output the final clean data stream.

[0011] Aggregate full-process performance indicators and construct multi-dimensional feature evolution curves. Based on the trends of the evolution curves, predict future interference patterns and pre-set parameter adjustment schemes. Perform online fine-tuning of execution parameters and update local non-volatile storage configurations. Archive system runtime snapshots.

[0012] Preferably, in the synchronous acquisition of the high-voltage side common-mode transient voltage waveform and the output response of the coupling channel, a high-bandwidth differential probe and a high-speed oscilloscope interface are used to synchronously capture the common-mode transient voltage waveform generated by the high-voltage side power device at the moment of turn-on and turn-off, and at the same time record the original output response of the capacitive coupling channel within the same time window, so as to ensure that the rising edge of the sampling clock and the high-voltage side drive signal are strictly time-aligned.

[0013] Preferably, the calculation of the cross-correlation function is used to extract the spatiotemporal correlation features, and the cross-correlation function between the common-mode transient voltage waveform and the original output response is calculated. By performing peak detection on the cross-correlation function, the delay time and corresponding peak intensity of the main interference component are determined. The peak intensity reflects the degree to which the common-mode interference weakens the channel signal-to-noise ratio.

[0014] Preferably, in the step of extracting real-time signal segments and performing short-time energy weighted statistics, a sliding window signal segment of length L is extracted from the real-time output stream of the coupled channel, and short-time energy weighted statistics are performed on the segment to enhance the weight of high-frequency components and highlight the influence of transient interference. The weighting coefficient is defined as an exponential decay form to emphasize the contribution of recent samples.

[0015] Preferably, in the step of calculating the real-time feature fingerprint vector and measuring the Euclidean distance with the reference noise spectrum vector, a real-time feature fingerprint vector composed of energy features of multiple frequency bands is constructed, and the Euclidean distance between the real-time feature fingerprint vector and the reference noise spectrum vector is calculated. When the Euclidean distance exceeds a preset safety threshold but does not reach a critical value, a slight interference is indicated.

[0016] Preferably, in the preprocessing of grouping and polarity reversal of the original logic signal according to the optimal encoding template, the original logic signal sequence to be sent is divided according to the grouping length specified by the template. For each group of signals, a preprocessing operation is performed according to the polarity reversal rule in the template. The polarity reversal strategy is used to reduce the occurrence of long consecutive 0s or long consecutive 1s and balance the DC component.

[0017] Preferably, in the step of extracting the original waveform at the receiving end and performing adaptive filtering based on the frequency band mask, an adaptive notch filter is designed using the reference noise spectrum vector as a frequency domain mask. The transfer function of this filter includes a Gaussian attenuation term for the center frequency of the dominant interference band, and the attenuation gain coefficient depends on the Euclidean distance calculated in real time.

[0018] Preferably, in the step of selecting an adaptive error correction strategy based on the Hamming distance value and performing correction calculation, if the Hamming distance is small, linear block codes are used for single-bit or multi-bit correction; if the Hamming distance is large but still within the detectable range, an interleaving decoding strategy is initiated; if the Hamming distance exceeds the error correction limit, it is marked as an unrecoverable error.

[0019] Preferably, in the method of predicting future interference patterns based on evolution curve trends and setting up parameter adjustment schemes, time series analysis is used to predict the interference trend within a short time window in the future. If the prediction results show that the interference intensity will continue to rise, then the parameter adjustment scheme is set up in advance. The new parameter set consists of the old parameter set plus the directional derivative term based on the gradient of the evolution curve.

[0020] On the other hand, the present invention proposes a capacitively coupled dynamic coding isolation system, comprising:

[0021] The synchronous acquisition module is used to synchronously acquire the common-mode transient voltage waveform and the output response of the coupling channel on the high-voltage side, calculate the cross-correlation function to extract the spatiotemporal correlation features, construct a multi-band noise energy distribution map and identify the dominant interference frequency band, and generate a reference noise spectrum vector.

[0022] The real-time feature extraction module is used to extract real-time signal segments and perform short-time energy weighted statistics, calculate real-time feature fingerprint vectors and perform Euclidean distance measurement with the reference noise spectrum vector, determine the interference level based on the distance measurement result and update the interference confidence, and generate a dynamic interference fingerprint matching report.

[0023] The adaptive coding mapping module is used to parse the dynamic interference fingerprint matching report and extract the optimal coding template under the current working condition. Based on the optimal coding template, the original logic signal is preprocessed by grouping and polarity reversal. Dynamic Manchester coding variants are introduced and instantaneous coding symbol sequences are calculated to generate the initial coding mapping data packet.

[0024] The pulse suppression and reconstruction module is used to extract the original waveform at the receiving end and perform adaptive filtering based on the frequency band mask, calculate the instantaneous signal-to-noise ratio distribution map and determine the dynamic decision threshold sequence, perform dual-edge triggered logic reconstruction and remove single-cycle abnormal pulses, encapsulate the reconstructed logic frame and verify the consistency of the encoding template.

[0025] The error correction module is used to parse the redundant check bits in the logical frame and calculate the Hamming distance matrix. Based on the Hamming distance value, it selects an adaptive error correction strategy and performs correction operations. It performs multi-source cross-validation and generates a data credibility score, and outputs the final clean data stream.

[0026] The self-evolutionary optimization module is used to aggregate full-process performance indicators and construct multi-dimensional feature evolution curves. Based on the trend of the evolution curves, it predicts future interference patterns and presets parameter adjustment schemes. It performs online fine-tuning of parameters and updates local non-volatile storage configurations, and archives system running snapshots.

[0027] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0028] This invention achieves accurate identification of the dominant interference frequency band under current operating conditions by synchronously acquiring the common-mode transient voltage waveform and coupled channel output response of the high-voltage side, calculating the cross-correlation function to extract spatiotemporal correlation features, and constructing a reference noise spectrum vector. It extracts real-time signal segments, performs short-time energy weighted statistics, calculates real-time feature fingerprint vectors, and uses Euclidean distance measurement with the reference noise spectrum vector to determine the interference level, thus realizing dynamic perception of the interference environment. It analyzes the interference level to generate the optimal coding template, preprocesses the original logic signal by grouping and polarity reversal, and introduces a dynamic Manchester coding variant, avoiding susceptible signal modes from the transmitting end. It extracts the original waveform at the receiving end, performs adaptive filtering based on a frequency band mask, determines the dynamic decision threshold sequence, and performs dual-edge triggered logic reconstruction, effectively eliminating single-cycle abnormal pulses and false pulses. It analyzes redundant check bits to calculate the Hamming distance matrix, selects an adaptive error correction strategy based on the value, and performs correction operations, further ensuring data integrity. It aggregates full-process performance indicators to construct a multi-dimensional feature evolution curve, predicts future interference modes, and presets parameter adjustment schemes, enabling the system to have self-evolution capabilities.

[0029] This invention significantly reduces the impact of parasitic pulse trains on signal integrity through a dual mechanism of dynamic coding mapping and adaptive filtering reconstruction, ensuring zero bit error rate and high reliability of data transmission under high voltage transient environment, while maintaining the system's high response speed, avoiding signal delay caused by traditional filtering methods, greatly improving the operational stability of power electronic systems in complex electromagnetic environments, and effectively solving the problem of misjudging logic levels at the receiving end due to the inability of fixed thresholds and simple filtering to cope with dynamic common-mode transient voltage interference in the prior art. Attached Figure Description

[0030] Figure 1 This is a flowchart of the capacitive coupling dynamic coding isolation method of the present invention;

[0031] Figure 2 This is a block diagram of the capacitively coupled dynamic coding isolation system of the present invention. Detailed Implementation

[0032] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0033] like Figure 1 As shown, this invention proposes a capacitively coupled dynamic coding isolation method, which effectively solves the problem in existing technologies where fixed thresholds and simple filtering cannot cope with dynamic common-mode transient voltage interference, leading to misjudgment of logic levels at the receiving end. Specifically, it includes:

[0034] Simultaneously acquire the common-mode transient voltage waveform and the output response of the coupling channel on the high-voltage side, calculate the cross-correlation function to extract the spatiotemporal correlation features, construct a multi-band noise energy distribution map and identify the dominant interference frequency band, and generate a reference noise spectrum vector;

[0035] Furthermore, in the synchronous acquisition of the common-mode transient voltage waveform on the high-voltage side and the output response of the coupling channel, a high-bandwidth differential probe and a high-speed oscilloscope interface are used to simultaneously capture the common-mode transient voltage waveform generated by the high-voltage side power devices at the moment of turn-on and turn-off, and at the same time record the original output response of the capacitive coupling channel within the same time window, to ensure that the rising edge of the sampling clock and the high-voltage side drive signal are strictly time-aligned.

[0036] Furthermore, in the process of calculating the cross-correlation function to extract the spatiotemporal correlation features, the cross-correlation function between the common-mode transient voltage waveform and the original output response is calculated. By performing peak detection on the cross-correlation function, the delay time and corresponding peak intensity of the main interference component are determined. The peak intensity reflects the degree to which the common-mode interference weakens the channel signal-to-noise ratio.

[0037] By strictly aligning the sampling clock with a high-bandwidth differential probe and a high-speed oscilloscope interface, the system can accurately capture the synchronous changes of the high-voltage side common-mode voltage and the coupling channel response within the extremely short transient window of power device turn-on and turn-off, eliminating feature extraction deviations caused by time asynchrony. Peak detection of the cross-correlation function is used to determine the delay time and intensity of the main interference component, quantifying not only the specific degree of signal-to-noise ratio attenuation caused by common-mode interference but also achieving precise locking of the dominant interference frequency band. This provides a high-fidelity spatiotemporal correlation data foundation for subsequent construction of a reference noise spectrum, enabling the system to fundamentally identify and distinguish between real signal transitions and parasitic pulse trains, significantly improving the extraction accuracy of anti-interference features and the system's sensitivity to dynamic electromagnetic environments.

[0038] Extract real-time signal segments and perform short-time energy weighted statistics, calculate real-time feature fingerprint vectors and perform Euclidean distance measurement with the benchmark noise spectrum vector, determine the interference level based on the distance measurement results and update the interference confidence, and generate a dynamic interference fingerprint matching report.

[0039] Furthermore, in the process of extracting real-time signal segments and performing short-time energy weighted statistics, a sliding window signal segment of length L is extracted from the real-time output stream of the coupled channel, and short-time energy weighted statistics are performed on this segment to enhance the weight of high-frequency components and highlight the impact of transient interference. The weighting coefficient is defined in an exponential decay form to emphasize the contribution of recent samples.

[0040] Furthermore, in the process of calculating the real-time feature fingerprint vector and measuring the Euclidean distance with the reference noise spectrum vector, a real-time feature fingerprint vector composed of energy features from multiple frequency bands is constructed. The Euclidean distance between the real-time feature fingerprint vector and the reference noise spectrum vector is calculated. When the Euclidean distance exceeds the preset safety threshold but does not reach the critical value, a slight interference is indicated.

[0041] Employing a sliding window weighted statistical approach with exponential decay effectively highlights the high-frequency characteristics of transient interference and strengthens the weight of recent samples, enabling the system to keenly capture rapidly changing common-mode noise. By constructing a multi-band energy feature vector and measuring it with Euclidean distance from a benchmark spectrum, quantitative classification and dynamic updating of interference intensity and confidence levels are achieved. This mechanism not only provides timely warnings in the early stages of interference (slight interference phase) but also adaptively adjusts subsequent processing strategies based on real-time matching results, avoiding misjudgments caused by fixed thresholds in complex electromagnetic environments. This significantly improves the system's robustness against interference and its response sensitivity under dynamic conditions.

[0042] The dynamic interference fingerprint matching report is analyzed and the optimal coding template under the current working condition is extracted. Based on the optimal coding template, the original logic signal is preprocessed by grouping and polarity reversal. A dynamic Manchester coding variant is introduced and the instantaneous coding symbol sequence is calculated to generate the initial coding mapping data packet.

[0043] Furthermore, in the preprocessing of grouping and polarity reversal of the original logic signal according to the optimal coding template, the original logic signal sequence to be sent is divided according to the grouping length specified by the template. For each group of signals, the preprocessing operation is performed according to the polarity reversal rule in the template. The polarity reversal strategy is used to reduce the occurrence of long consecutive 0s or long consecutive 1s and balance the DC component.

[0044] By using an optimal coding template based on dynamic matching of interference fingerprints, adaptive adjustment of the transmitter signal processing strategy is achieved. The polarity reversal strategy effectively eliminates long streaks of 0s or 1s, significantly balancing the DC component of the signal and suppressing low-frequency drift and baseline wandering at the source. Introducing a dynamic Manchester coding variant not only enhances the signal's self-synchronization capability but also ensures the distinguishability of high-frequency transition characteristics from noise characteristics under strong common-mode transient interference, significantly reducing the risk of bit errors caused by deterioration of signal spectral characteristics and improving the robustness of data transmission.

[0045] Extract the original waveform from the receiving end and perform adaptive filtering based on the frequency band mask, calculate the instantaneous signal-to-noise ratio distribution map and determine the dynamic decision threshold sequence, perform dual-edge triggered logic reconstruction and remove single-cycle abnormal pulses, encapsulate the reconstructed logic frame and verify the consistency of the encoding template;

[0046] Furthermore, in extracting the original waveform from the receiving end and performing adaptive filtering based on the frequency band mask, an adaptive notch filter is designed using the reference noise spectrum vector as the frequency domain mask. The transfer function of this filter includes a Gaussian attenuation term for the center frequency of the dominant interference band, and the attenuation gain coefficient depends on the Euclidean distance calculated in real time.

[0047] An adaptive notch filter is constructed using a reference noise spectrum, which can accurately lock onto and attenuate the dominant interference frequency band, achieving targeted suppression of specific common-mode interference and avoiding the accidental damage to the effective signal spectrum caused by traditional fixed filtering. By dynamically adjusting the attenuation gain through real-time Euclidean distance, the filtering depth automatically matches the interference intensity, ensuring signal purity under strong interference while maintaining low delay characteristics under weak interference. Combined with dual-edge triggering and abnormal pulse rejection mechanisms, logical misjudgments caused by parasitic pulse trains are completely eliminated, significantly improving the waveform reconstruction accuracy and data integrity at the receiver in complex electromagnetic environments.

[0048] Parse the redundant check bits in the logical frame and calculate the Hamming distance matrix. Select an adaptive error correction strategy based on the Hamming distance value and perform correction operations. Perform multi-source cross-validation and generate a data credibility score. Output the final clean data stream.

[0049] Furthermore, in selecting an adaptive error correction strategy and performing correction operations based on the Hamming distance value, if the Hamming distance is small, linear block codes are used for single-bit or multi-bit correction; if the Hamming distance is large but still within the detectable range, an interleaving decoding strategy is initiated; if the Hamming distance exceeds the error correction limit, it is marked as an unrecoverable error.

[0050] By constructing a Hamming distance matrix and implementing a hierarchical adaptive error correction strategy, intelligent switching from precise single-bit correction to interleaved decoding in the event of sudden interference is achieved. This mechanism effectively balances error correction efficiency and computational overhead: it provides rapid linear correction for minor errors, avoiding overprocessing; and it initiates high-order interleaved decoding when strong interference causes multi-bit errors, maximizing the recovery of data integrity. Combined with a credibility score generated by multi-source cross-validation, the system can automatically identify and isolate unrecoverable severe errors, preventing dirty data from contaminating downstream services and significantly improving the final data throughput quality and system reliability of the communication link in complex electromagnetic environments.

[0051] Aggregate full-process performance indicators and construct multi-dimensional feature evolution curves. Based on the trends of the evolution curves, predict future interference patterns and pre-set parameter adjustment schemes. Perform online fine-tuning of execution parameters and update local non-volatile storage configurations. Archive system runtime snapshots.

[0052] Furthermore, in predicting future interference patterns based on evolution curve trends and pre-setting parameter adjustment schemes, time series analysis is used to predict interference trends within a short time window. If the prediction results show that the interference intensity will continue to rise, a parameter adjustment scheme is pre-set. The new parameter set consists of the old parameter set plus the directional derivative term based on the gradient of the evolution curve.

[0053] By introducing a time-series-based trend prediction and gradient-guided parameter presetting mechanism, the system response is upgraded from passive adaptation to active defense. By predicting the inflection point of interference intensity in advance and fine-tuning parameters along the gradient direction of the evolution curve, the system can achieve optimal configuration before interference worsens, significantly shortening the transient oscillation period during dynamic adjustments. Combined with online updates and runtime snapshot archiving using local non-volatile storage, continuous evolution of interference response strategies and fault backtracking capabilities are achieved, greatly improving the system's long-term robustness and self-healing efficiency in complex and variable electromagnetic environments.

[0054] On the other hand, this invention proposes a capacitively coupled dynamic coding isolation system, such as Figure 2 As shown, it includes:

[0055] The synchronous acquisition module is used to synchronously acquire the common-mode transient voltage waveform and the output response of the coupling channel on the high-voltage side, calculate the cross-correlation function to extract the spatiotemporal correlation features, construct a multi-band noise energy distribution map and identify the dominant interference frequency band, and generate a reference noise spectrum vector.

[0056] The real-time feature extraction module is used to extract real-time signal segments and perform short-time energy weighted statistics, calculate real-time feature fingerprint vectors and perform Euclidean distance measurement with the reference noise spectrum vector, determine the interference level based on the distance measurement result and update the interference confidence, and generate a dynamic interference fingerprint matching report.

[0057] The adaptive coding mapping module is used to parse the dynamic interference fingerprint matching report and extract the optimal coding template under the current working condition. Based on the optimal coding template, the original logic signal is preprocessed by grouping and polarity reversal. Dynamic Manchester coding variants are introduced and instantaneous coding symbol sequences are calculated to generate the initial coding mapping data packet.

[0058] The pulse suppression and reconstruction module is used to extract the original waveform at the receiving end and perform adaptive filtering based on the frequency band mask, calculate the instantaneous signal-to-noise ratio distribution map and determine the dynamic decision threshold sequence, perform dual-edge triggered logic reconstruction and remove single-cycle abnormal pulses, encapsulate the reconstructed logic frame and verify the consistency of the encoding template.

[0059] The error correction module is used to parse the redundant check bits in the logical frame and calculate the Hamming distance matrix. Based on the Hamming distance value, it selects an adaptive error correction strategy and performs correction operations. It performs multi-source cross-validation and generates a data credibility score, and outputs the final clean data stream.

[0060] The self-evolutionary optimization module is used to aggregate full-process performance indicators and construct multi-dimensional feature evolution curves. Based on the trend of the evolution curves, it predicts future interference patterns and presets parameter adjustment schemes. It performs online fine-tuning of parameters and updates local non-volatile storage configurations, and archives system running snapshots.

[0061] Furthermore, the modules in the above system perform other steps to implement the aforementioned capacitive coupling dynamic coding isolation method, as follows:

[0062] Step 1: Construction of a reference noise spectrum based on multi-source transient sensing:

[0063] This step aims to perform a deep scan and feature capture of the electromagnetic environment of the coupling channel during system power-on initialization and early operation, thereby establishing a benchmark spectrum that reflects the noise distribution characteristics under current operating conditions. This process is not a simple static sampling, but rather combines the periodic characteristics of high-voltage side switch operations with multi-dimensional data acquisition and analysis to provide reliable reference coordinates for subsequent dynamic coding. Without this benchmark, subsequent signal processing would be like calibrating a compass in a windless environment, making it difficult to distinguish between real signals and environmental disturbances. Specifically, this includes:

[0064] 1.1 Synchronously acquire the common-mode transient voltage waveform on the high-voltage side and the output response of the coupling channel:

[0065] First, using a high-bandwidth differential probe and a high-speed oscilloscope interface, the common-mode transient voltage waveforms generated by high-voltage side power devices (such as IGBTs) during turn-on and turn-off are captured simultaneously. Simultaneously, the original output response of the capacitively coupled channel within the same time window is recorded. Here, it is necessary to ensure that the sampling clock and the rising edge of the high-voltage side drive signal are strictly time-aligned to eliminate phase errors caused by time drift. The acquired common-mode transient voltage waveform is represented as follows:

[0066] ;

[0067] in, This represents the transient voltage amplitude generated by the k-th switching event. This marks the starting point of the event. The decay time constant characterizing the voltage waveform. The high-frequency oscillation frequency that accompanies the transient. Let be the initial phase angle, and N be the total number of switching events within a single observation window. The output response of the synchronously recorded coupling channel. This includes real digital logic reversal information and information derived from... The parasitic interference component generated by induction. Through this synchronous acquisition, the correspondence between the two in the time domain can be obtained, laying the foundation for subsequent separation of the real signal and noise.

[0068] 1.2 Extracting the spatiotemporal correlation features of common-mode interference and channel response based on cross-correlation analysis:

[0069] In obtaining synchronous data collection and After obtaining the data, the next step is to calculate the cross-correlation function between the two to quantify the specific impact of the common-mode transient voltage on the output response of the coupled channel and its time delay characteristics. This step is crucial for understanding the noise propagation path, revealing how interference signals traverse the capacitor dielectric and form spurious pulses at the receiver. Cross-correlation function Defined as:

[0070] ;

[0071] In the formula, This represents the time offset, used to find the time difference between two signals where the correlation is strongest. By... Peak detection can determine the delay time of the main interference component. Its corresponding peak intensity This reflects the degree to which common-mode interference weakens the channel's signal-to-noise ratio. If If the preset warning threshold is exceeded, it indicates that the risk of generating parasitic pulse trains under the current operating conditions is extremely high, and a higher level of coding protection strategy must be activated. The extraction of this correlation feature directly depends on the raw waveform data acquired in step 1.1. Without high-precision synchronous acquisition, the cross-correlation analysis will not be able to accurately reflect the true physical coupling effect.

[0072] 1.3 Constructing multi-band noise energy distribution maps and identifying dominant interference frequency bands:

[0073] Based on the spatiotemporal correlation features extracted in step 1.2, the output response of the coupled channel is further analyzed. A short-time Fourier transform (STFT) is performed to convert the noise from the time domain to the time-frequency domain, thereby constructing a multi-band noise energy distribution map. This map can intuitively display the energy density distribution of each frequency component over different time periods, especially those interference bands that coincide with the high-voltage side switching frequency and its harmonics. The energy density function S(f,t) in the map can be expressed as:

[0074] ;

[0075] in, Let f be a sliding window function used to limit the integration range to obtain local spectral characteristics, where f is the frequency variable and t is the time variable. By analyzing the evolution trend of S(f,t), frequency bands with energy significantly higher than the background noise level can be identified; these frequency bands are the current dominant interference bands. For example, if it is found that the energy of a certain frequency band is... The time rose sharply, and with If the timing of the switching events (defined in step 1.1) closely matches, it can be determined that the frequency band is driven by common-mode transient voltage. The result of this step directly determines the frequency range that needs to be protected in subsequent coding strategies, and is a frequency domain deepening of the correlation features in step 1.2.

[0076] 1.4 Generate the reference noise spectrum vector and store it in local non-volatile memory:

[0077] Finally, the dominant interference frequency bands and their corresponding energy distribution characteristics identified in step 1.3 are integrated into a standardized reference noise spectrum vector. This vector contains not only the center frequencies of each frequency band. and relative energy weight It also recorded the statistical patterns of this frequency band's changes over time, such as the mean. and variance .vector The specific components are as follows:

[0078] ;

[0079] Where n is the number of identified dominant interference frequency bands. (Generated) The vector is written to local non-volatile memory as a reference for subsequent system operation. The generation of this vector depends entirely on the analysis results of the first three sub-steps, particularly the frequency domain energy distribution data provided in step 1.3. Once the system enters normal operating mode, new input signals will continuously interact with... By comparison, any abnormal fluctuations that deviate from the spectrum will be considered as potential interference or fault signals.

[0080] Step 2: Real-time interference fingerprint matching based on dynamic feature extraction:

[0081] This step builds upon the baseline noise spectrum established in the first step, focusing on real-time monitoring of the coupling channel status during system operation. It extracts real-time interference feature fingerprints and dynamically matches them with the baseline spectrum. This process aims to capture subtle changes in environmental noise, determine whether the system is in a high-risk transient interference period, and thus decide whether to trigger dynamic coding adjustments. Relying solely on a static baseline would render the system sluggish in the face of sudden environmental changes, while real-time fingerprint matching provides the system with a heightened sensitivity, specifically including:

[0082] 2.1. Capture real-time signal segments using a sliding window and perform short-time energy-weighted statistics:

[0083] During the continuous operation phase of the system, a sliding window signal segment x(t) of length L is extracted from the real-time output stream of the coupled channel. To highlight the impact of transient interference, short-time energy weighting statistics are performed on this segment, with a focus on increasing the weight of high-frequency components, since common-mode transients typically exhibit high-frequency oscillations. The weighted energy statistics are shown below. The calculation formula is:

[0084] ;

[0085] in, The exponentially decaying weighting coefficient is defined as follows: , This is an attenuation factor used to emphasize the contribution of recent samples. This method allows for more sensitive detection of sudden voltage jumps. The calculation of this statistic directly utilizes the frequency band of interest defined in the baseline spectrum in step 1.4, ensuring the targeted nature of the extracted features. The extraction of real-time signal segments relies on the sampling rate and clock synchronization mechanism determined in the first step, guaranteeing the continuity of the time axis.

[0086] 2.2 Calculate the real-time feature fingerprint vector and perform Euclidean distance measurement with the benchmark spectrum:

[0087] Based on the real-time energy statistics obtained in step 2.1, the current real-time feature fingerprint vector is constructed. This vector also consists of energy characteristics from multiple frequency bands, and its structure is similar to the reference noise spectrum vector in step 1.4. Maintain consistency. Then, calculate... and The Euclidean distance d(t) between the two states is used to quantify the deviation of the current environmental state from the baseline state. The distance calculation formula is as follows:

[0088] ;

[0089] In the formula, Let be the real-time normalized energy of the j-th frequency band at time t. This represents the standard energy weight for the corresponding frequency band in the baseline spectrum. When d(t) is less than a preset safety threshold... When the environment is considered stable, no special treatment is required; when d(t) exceeds However, the critical value was not reached. At that time, it indicates the presence of slight interference; if If the condition is not met, it is determined to be a strong interference environment, and dynamic coding protection must be activated immediately. This measurement process directly uses the real-time features extracted in step 2.1 and compares them with the benchmark data generated in step 1.4, realizing the leap from static benchmark to dynamic evaluation.

[0090] 2.3 Determine the interference level and update the interference confidence score based on the distance metric results:

[0091] Based on the Euclidean distance d(t) calculated in step 2.2, the current interference state is divided into three levels: low risk, medium risk, and high risk. For each level, the system maintains an interference confidence index. This indicator reflects the reliability of the system's judgment on the current interference. The update rule for the confidence level is as follows:

[0092] ;

[0093] in, This is a forgetting factor used to retain the inertia of historical judgments. This is the boundary threshold for the medium-risk level. is a smoothing coefficient used to control the steepness of the sigmoid function. If d(t) shows a high-risk state for multiple consecutive time points, then... The confidence level will rapidly approach 1, ensuring accurate identification of persistent strong interference. This step relies on the distance metric result from step 2.2, and by introducing the cumulative effect of the time dimension, it avoids misjudgments caused by instantaneous noise spikes. The establishment of the confidence index provides a quantitative basis for the selection of subsequent coding strategies.

[0094] 2.4 Generate a dynamic interference fingerprint matching report and transmit it to the coding decision unit:

[0095] Finally, the interference level and interference confidence level determined in step 2.3 will be... The system, along with relevant feature deviation data, is packaged into a dynamic interference fingerprint matching report. This report not only includes the current state label but also lists the main frequency band contributors causing state changes, i.e., which frequency band energy fluctuations contribute most to the distance d(t). The report is transmitted to the next-level encoding decision unit via the internal bus. The report's generation relies entirely on the data flow of the first three sub-steps, particularly the decision result of step 2.3. This report will serve as the direct input for the initial encoding mapping in the third step, ensuring that the encoding strategy can be customized for specific interference types, rather than blindly applying general rules. Through this progressive logic, the system achieves accurate perception and rapid response to environmental interference.

[0096] Step 3: Adaptive initial encoding mapping based on interference fingerprint matching:

[0097] This step, based on the dynamic interference fingerprint matching report output in step two, performs initial encoding mapping on the original digital logic signal at the transmitting end. This process is not a simple linear transformation, but rather dynamically adjusts the redundancy, hopping code mode, and duty cycle of the encoding according to the strength and frequency band distribution of the interference fingerprint, in order to minimize the impact of common-mode interference at the physical level. By changing the time-frequency characteristics of the signal, the originally easily interfered encoding form is transformed into a new form with stronger anti-interference capabilities, thereby suppressing the generation of spurious pulses at the source. Specifically, this includes:

[0098] 3.1. Analyze the interference fingerprint matching report and extract the optimal encoding template under the current operating conditions:

[0099] After receiving the dynamic interference fingerprint matching report from the second step, the interference level and confidence information are first analyzed to extract the optimal coding template for the current operating condition. If the interference level is low-risk, the default baseline coding template is used to maintain communication efficiency; if it is medium-risk, a coding template with medium redundancy is used, and a check bit is added; if it is high-risk, a highly redundant dynamic hopping code template is switched to significantly reduce the probability of misjudgment caused by single-bit errors. The template selection is based on the following mapping relationship:

[0100] ;

[0101] in, These represent baseline, medium, and high redundancy coding templates, respectively. and This is the confidence level threshold. This step directly relies on the report content generated in step 2.4, transforming the abstract interference assessment into a specific coding strategy selection. The template selection ensures that the coding scheme always matches the current electromagnetic environment, avoiding resource waste or insufficient protection.

[0102] 3.2. Preprocess the original logic signals by grouping and polarity reversal according to the optimal encoding template:

[0103] Select the optimal encoding template Then, the original logic signal sequence to be sent... The signal is divided according to the grouping length G specified in the template. For each group of signals, preprocessing is performed according to the polarity reversal rule in the template. In strong interference environments, frequent signal transitions can easily induce common-mode transients; therefore, a specific polarity reversal strategy is needed to reduce the occurrence of long strings of 0s or 1s and balance the DC component. The preprocessed signal... Represented as:

[0104] ;

[0105] in, This represents the XOR operation, where P[i] is the value generated by the template. A pseudo-random polarity inversion sequence is generated. The design principle of this sequence is to maximize the flipping probability of adjacent bits, thereby disrupting signal patterns that might induce resonance. This step relies on the template selected in step 3.1, which defines the block length and the seed for generating the inversion sequence. Through this preprocessing, the characteristics of the original signal are reshaped to better suit the current high-voltage transient environment.

[0106] 3.3 Introducing a dynamic Manchester coding variant and calculating the instantaneous coded symbol sequence:

[0107] Building upon preprocessing, a dynamic Manchester coding variant is introduced to convert binary data into a self-synchronizing biphase code. Unlike traditional Manchester coding, this variant fine-tunes the clock edge position based on the dominant frequency band in the interference fingerprint, avoiding known noise-dense regions. Instantaneous coded symbol sequence. The calculation formula is:

[0108] ;

[0109] Here, b[k] is the k-th bit after preprocessing, and freq_shift is the frequency adjustment factor calculated based on the dominant frequency band offset identified in step 2.3. This factor ensures that the main energy of the coded signal is concentrated in the frequency band region with weaker interference. The generation of the coded symbol sequence directly inherits the preprocessing results of step 3.2 and closely combines the coding strategy with frequency domain features, realizing joint optimization in the time and frequency domains.

[0110] 3.4. Generate the initial encoded mapping data packet and inject it into the send buffer queue:

[0111] Finally, the instantaneous coded symbol sequence generated in step 3.3 is... The data is encapsulated into a complete initial encoded mapping data packet. The packet header includes an encoding template identifier, a timestamp, and a checksum to ensure correct decoding at the receiving end. The generated data packet is injected into the transmit buffer queue, awaiting modulation and transmission by the physical layer. The injection of the data packet marks the end of the encoding mapping process and also prepares for the real-time feedback correction in the fourth step. By transforming complex encoding logic into a standard packet format, the system achieves a smooth transition from logical signals to physical transmission signals, laying a solid foundation for subsequent stable transmission.

[0112] Step 4: Burst suppression and logic reconstruction based on receiver dynamic threshold calibration:

[0113] This step follows the initial encoded mapping data packet generated in step three, focusing on the receiving end processing of the signal after transmission through the capacitively coupled channel. Since parasitic pulse trains caused by transient voltage (dv / dt) on the high-voltage side often exhibit high-frequency, low-amplitude spurious transitions, traditional fixed threshold decisions are prone to failure. Therefore, the core of this step lies in constructing a mechanism capable of real-time sensing of the channel state and dynamically adjusting the decision threshold. Through a filtering-then-decision strategy, the distorted analog waveform is reconstructed into a clean digital logic level. This process strictly relies on the reference spectrum and encoding template established in the preceding steps to ensure that the decoding strategy remains consistent with the encoding characteristics of the transmitting end, specifically including:

[0114] 4.1 Extract the original waveform from the receiving end and perform adaptive filtering based on a frequency band mask:

[0115] First, the raw analog waveform y(t) after physical transmission is obtained from the receiving pin of the capacitive isolator. This waveform contains not only the coded and modulated effective signal components but also parasitic interference noise n(t) generated by common-mode transient induction. To suppress these interferences, the reference noise spectrum vector constructed in the first step is used. An adaptive notch filter is designed as a frequency domain mask. The transfer function H(f) of this filter is defined as:

[0116] ;

[0117] in, The center frequency of the j-th dominant interference band identified in step 1.3 is... This is the attenuation gain coefficient for this frequency band, and its magnitude depends on the Euclidean distance d(t) calculated in the second step, i.e. k is a proportionality constant. This is the notch bandwidth parameter. By applying this filter, the pre-filtered waveform is obtained. The formula is expressed as:

[0118] ;

[0119] In the formula, Y(f) is the Fourier transform of the original waveform y(t). This represents the inverse Fourier transform. This step directly uses the reference spectrum generated in step 1.4 and the distance metric calculated in step 2.2, ensuring that the filter parameters can be dynamically adjusted according to changes in environmental interference, thereby filtering out parasitic pulses in a specific frequency band to the greatest extent possible without damaging the effective signal.

[0120] 4.2 Calculate the instantaneous signal-to-noise ratio distribution and determine the dynamic decision threshold sequence:

[0121] Obtain the filtered waveform The next step is to address the issue that fixed thresholds cannot adapt to signal amplitude fluctuations. The system... Perform sliding window analysis and calculate the local mean within each window. and standard deviation This allows for the construction of an instantaneous signal-to-noise ratio (SNR) distribution map. Based on this distribution, a dynamically changing decision threshold sequence Th(t) is generated. This sequence is not a constant value but rather follows the signal baseline drift. The dynamic threshold calculation formula is as follows:

[0122] ;

[0123] in, This is a safety factor used to balance the false alarm rate and the false negative rate. If the interference level is high (such as the high-risk assessment in step 2.3),... The value automatically increases to raise the decision threshold and prevent low-amplitude interference from being misinterpreted as a high level; if the environment is stable, The value is reduced to improve sensitivity. The generation of this threshold sequence depends on the statistical characteristics of the filtered waveform obtained in step 4.1, and also implies a reverse understanding of the polarity reversal rule of the encoding template in step 3, because an effective logic flip must be accompanied by a significant jump exceeding the dynamic threshold. Through this dynamic adjustment, the system can effectively distinguish between true logic edges and residual minor interference glitches.

[0124] 4.3 Execute dual-edge triggered logic reconstruction and eliminate single-cycle abnormal pulses:

[0125] Based on the dynamic decision threshold sequence Th(t) determined in step 4.2, the waveform is... Make digital decisions and generate preliminary binary sequences. However, because parasitic bursts can be extremely short (e.g., lasting only a few nanoseconds), a single threshold comparison can still produce isolated erroneous bits. To address this, a dual-edge triggered logic reconstruction mechanism is introduced, requiring that a change in logic level must be maintained for at least two consecutive sampling periods on the time axis to be acknowledged. The reconstructed logic sequence... The judgment rules are as follows:

[0126] ;

[0127] This rule mandates that the state must be flipped between the two consecutive time points before updating the current state, effectively eliminating spurious pulses with a width less than twice the sampling period. This logic reconstruction process directly utilizes the binary sequence output from step 4.2 and further purifies the data stream through timing constraints. Essentially, it compensates for the physical characteristics of the capacitively coupled channel, using redundancy in the time dimension to combat interference in the spatial dimension, ensuring extremely high stability of the final output logic level.

[0128] 4.4. Encapsulate the reconstructed logical frame and verify the consistency of the encoding template:

[0129] Finally, the pure logic sequence generated in step 4.3 is... According to the encoding template used in step three The data is segmented and reassembled to form a complete logical frame. During this process, the system again calls the encoding template identifier parsed in step 3.1 to verify whether the format of the currently received frame is consistent with that of the sender. If a frame header error or checksum mismatch is detected, the frame is marked as invalid and discarded, and a retransmission request is triggered or the system enters protection mode. The encapsulated logical frame data will be output to the subsequent error correction unit. This step is the final step in the receiver's processing. It converts the filtering results in the analog domain and the decision results in the digital domain into a standard logical data stream and completes the closed-loop verification with the sender's encoding strategy. The entire process closely revolves around the output of the preceding steps, realizing a complete conversion from analog interference to a reliable digital signal.

[0130] Step 5: Dynamic error correction and integrity assurance based on redundancy check and error correction codes:

[0131] This step follows the reconstructed logic frame output from step four, focusing on data integrity and error correction. Although step four has eliminated most obvious parasitic pulses through filtering and logic reconstruction, random errors that are difficult to completely eliminate may still exist in extremely harsh electromagnetic environments. This step utilizes the redundant coding information injected in step three and an independent verification algorithm to perform deep cleaning of the data, ensuring the accuracy of transmitted data. Specifically, this includes:

[0132] 5.1 Parse the redundancy check bits in the logical frame and calculate the Hamming distance matrix:

[0133] The received logical frame contains redundant check bits (such as parity bits or cyclic redundancy check (CRC) generated in the third step. The system first extracts these check bits and recalculates the check value for the data payload according to the encoding rules agreed upon by the sender. Then, the calculated check value is compared bit-by-bit with the check bits carried in the frame to construct the Hamming distance matrix. The formula for calculating Hamming distance is:

[0134] ;

[0135] in, For the calculated check bits, The received parity bits are M, and M is the total length of the parity bits. This distance value quantifies the number of bit flips that occurred during data transmission. If This indicates that the data is correct; if If the data is corrupted, an error exists. This step directly relies on the logical frame output from step four, especially its redundant information. By calculating the Hamming distance, the system can quantitatively assess the degree of data corruption, providing a precise basis for selecting subsequent error correction strategies.

[0136] 5.2. Select an adaptive error correction strategy based on the Hamming distance value and perform correction calculations:

[0137] Hamming distance calculated based on step 5.1 The system dynamically selects the error correction strategy. If If the error correction capability t is small (e.g., less than or equal to the error correction code's correction capability t), then linear block codes such as Hamming codes are used for single-bit or multi-bit correction; if If the error is large but still within the detectable range, an interleaving decoding strategy is initiated to attempt to disperse the impact of the sudden error by rearranging the bit order; if If the error exceeds the correction limit, it is marked as an unrecoverable error. The specific process of the correction operation can be described as follows:

[0138] ;

[0139] in, The received raw data sequence, This is the error pattern vector derived from the error correction algorithm. This is the corrected data sequence. Error mode vector. The generation depends on a specific error-correcting code algebra structure, and its purpose is to find a way to make the code more accurate and efficient. Minimize the most probable transmission sequence. This step is entirely based on the evaluation results of step 5.1, ensuring that error correction resources are allocated reasonably, avoiding misoperations caused by over-correction, and guaranteeing the data's self-healing capability under slight interference.

[0140] 5.3 Perform multi-source cross-validation and generate data credibility scores:

[0141] To further improve the reliability of error correction results, the system introduces a multi-source cross-validation mechanism. In addition to relying on the current check bit, the system also performs a logical consistency check between the corrected data and historical data from the previous few clock cycles. For example, for control command data, actions in two consecutive cycles must meet specific physical constraints (e.g., cannot instantly change from fully on to fully off). Reliability Score The calculation formula is as follows:

[0142] ;

[0143] in, and These are the check bit weight and the logical consistency weight, respectively. This is a logical matching indicator variable (1 for compliance with the constraint, 0 for otherwise). This score reflects the reliability of the data in the current context. If the data falls below the preset safety threshold, even if error correction is successful, the system will consider the data unreliable and discard it. This step combines the correction results from step 5.2 with historical context information, eliminating potential blind spots in single-verification through multi-dimensional verification and enhancing the system's robustness.

[0144] 5.4 Output the final clean data stream and record the error correction log:

[0145] After verification in step 5.3, the system outputs the finally confirmed correct data stream to the upper-layer application interface. Simultaneously, the system automatically records error correction logs during this transmission process, including the error type, bit differences before and after correction, Hamming distance values, and the final confidence score. This log data is not only used for internal system performance monitoring but also serves as feedback to the first step, updating the baseline noise spectrum. The generation of log records depends on the correction process in step 5.2 and the scoring results in step 5.3, forming a complete data quality traceability chain. Thus, the data undergoes a complete contamination-cleaning-verification cycle from encoding at the sending end to error correction at the receiving end, ensuring reliable transmission with zero bit error rate even under high-voltage transient interference.

[0146] Step 6: Self-evolutionary parameter optimization and system state archiving based on end-to-end feedback:

[0147] This step aims to achieve the system's self-evolution and long-term adaptability, moving beyond the processing of single data transmissions to focus on the performance evolution of the entire isolated system over extended periods. By collecting various intermediate data and result metrics generated in the first five steps, the system can automatically adjust internal parameters, optimize coding strategies, and archive critical states to address potential new interference patterns in the future. Specifically, this includes:

[0148] 6.1 Aggregate full-process performance indicators and construct multi-dimensional feature evolution curves:

[0149] First, the system aggregates all key performance indicators generated in steps one through five, including the energy changes of the baseline noise spectrum, the Euclidean distance fluctuations of the real-time interference fingerprint, the adjustment range of the dynamic threshold, the number of successful error corrections, and the distribution of data credibility scores. These discrete data points are integrated into a multi-dimensional feature vector and accumulated over time to construct the system's end-to-end performance evolution curve. The mathematical expression of the curve can be summarized as follows:

[0150] ;

[0151] in, Indicates the amount of change in the baseline spectrum. ErrRate represents the bit error rate, indicating the adjustment amount of the dynamic threshold. These are the weighting coefficients for each indicator. This evolution curve visually demonstrates the trend of the system's health status and anti-interference capability over a period of time.

[0152] 6.2 Predicting future disturbance modes based on evolution curve trends and pre-setting parameter adjustment schemes:

[0153] Using the evolution curve constructed in step 6.1, the system employs time series analysis techniques (such as moving average or exponential smoothing) to predict interference trends within a short future time window. If the prediction results indicate that the interference intensity will continue to rise, or that new frequency band characteristics will emerge, the system will pre-set a parameter adjustment scheme before the interference actually worsens. The update rules for the pre-set parameters are as follows:

[0154] ;

[0155] in, and These are the new and old parameter sets (such as the filter cutoff frequency, the redundancy of the coding template, etc.). To learn step length, The directional derivative based on the gradient of the evolution curve indicates the optimal direction for parameter optimization. This predictive parameter tuning enables the system to be proactive, preparing for defenses before disturbances occur, rather than reacting passively.

[0156] 6.3. Perform online fine-tuning of parameters and update local non-volatile storage configuration:

[0157] Based on the preset scheme generated in step 6.2, the system performs online fine-tuning of key operating parameters. This process does not require a system restart but is completed silently in the background. The adjusted parameters take effect immediately and are synchronously updated to the local non-volatile memory to ensure that the system inherits the current optimal configuration upon the next power-on. The configuration file update process is described as follows:

[0158] ;

[0159] The Merge operation retains stable parameters from the original configuration that are independent of the current environment, replacing only dynamic parameters affected by disturbances. This step ensures the continuity and persistence of the system state, preventing the loss of learning results due to power outages. The fine-tuning of parameters is entirely based on the prediction results from step 6.2, demonstrating the system's keen insight into environmental changes and its rapid adaptability. Through this continuous iterative optimization, the system gradually converges to an optimal operating point for the current conditions.

[0160] 6.4. Archive system snapshots and generate adaptive reports for external auditing:

[0161] Finally, the system packages the current system status, the latest parameter configuration, recent performance evolution curves, and key error correction examples into a complete runtime snapshot. This snapshot is archived to the system's long-term storage and an adaptive report is generated. The report details the system's response strategies, adjustment processes, and final effects when facing specific disturbances, providing a comprehensive basis for subsequent system maintenance, troubleshooting, or upgrades.

[0162] The report generation relies on the update results of step 6.3 and the historical data summary of step 6.1. Through this archiving mechanism, the system not only solves the current problem but also accumulates valuable data assets for future technological evolution. Thus, the entire capacitive coupling dynamic coding isolation method and system form a complete closed loop from sensing, processing, error correction to evolution, effectively solving the problem in existing technologies where fixed thresholds and simple filtering cannot cope with dynamic common-mode transient voltage interference, leading to misjudgment of logic levels at the receiver.

[0163] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A capacitively coupled dynamic coding isolation method, characterized in that, include: Simultaneously acquire the common-mode transient voltage waveform and the output response of the coupling channel on the high-voltage side, calculate the cross-correlation function to extract the spatiotemporal correlation features, construct a multi-band noise energy distribution map and identify the dominant interference frequency band, and generate a reference noise spectrum vector; Extract real-time signal segments and perform short-time energy weighted statistics, calculate real-time feature fingerprint vectors and perform Euclidean distance measurement with the benchmark noise spectrum vector, determine the interference level based on the distance measurement results and update the interference confidence, and generate a dynamic interference fingerprint matching report. The dynamic interference fingerprint matching report is analyzed and the optimal coding template under the current working condition is extracted. Based on the optimal coding template, the original logic signal is preprocessed by grouping and polarity reversal. A dynamic Manchester coding variant is introduced and the instantaneous coding symbol sequence is calculated to generate the initial coding mapping data packet. Extract the original waveform from the receiving end and perform adaptive filtering based on the frequency band mask, calculate the instantaneous signal-to-noise ratio distribution map and determine the dynamic decision threshold sequence, perform dual-edge triggered logic reconstruction and remove single-cycle abnormal pulses, encapsulate the reconstructed logic frame and verify the consistency of the encoding template; Parse the redundant check bits in the logical frame and calculate the Hamming distance matrix. Select an adaptive error correction strategy based on the Hamming distance value and perform correction operations. Perform multi-source cross-validation and generate a data credibility score. Output the final clean data stream. Aggregate full-process performance indicators and construct multi-dimensional feature evolution curves. Based on the trends of the evolution curves, predict future interference patterns and pre-set parameter adjustment schemes. Perform online fine-tuning of execution parameters and update local non-volatile storage configurations. Archive system runtime snapshots.

2. The capacitive coupling dynamic coding isolation method according to claim 1, characterized in that, In the synchronous acquisition of the common-mode transient voltage waveform and the output response of the coupling channel on the high-voltage side, a high-bandwidth differential probe and a high-speed oscilloscope interface are used to synchronously capture the common-mode transient voltage waveform generated by the high-voltage side power device at the moment of turn-on and turn-off. At the same time, the original output response of the capacitive coupling channel within the same time window is recorded to ensure that the rising edge of the sampling clock and the high-voltage side drive signal are strictly time-aligned.

3. The capacitive coupling dynamic coding isolation method according to claim 1, characterized in that, The cross-correlation function is calculated to extract spatiotemporal correlation features. The cross-correlation function between the common-mode transient voltage waveform and the original output response is calculated. By performing peak detection on the cross-correlation function, the delay time and corresponding peak intensity of the main interference component are determined. The peak intensity reflects the degree to which the common-mode interference weakens the channel signal-to-noise ratio.

4. The capacitive coupling dynamic coding isolation method according to claim 1, characterized in that, In the process of extracting real-time signal segments and performing short-time energy weighted statistics, a sliding window signal segment of length L is extracted from the real-time output stream of the coupled channel. Short-time energy weighted statistics are then performed on this segment, with a focus on enhancing the weight of high-frequency components and highlighting the impact of transient interference. The weighting coefficients are defined in an exponential decay form to emphasize the contribution of recent samples.

5. The capacitive coupling dynamic coding isolation method according to claim 1, characterized in that, In the process of calculating the real-time feature fingerprint vector and measuring the Euclidean distance with the reference noise spectrum vector, a real-time feature fingerprint vector composed of energy features of multiple frequency bands is constructed. The Euclidean distance between the real-time feature fingerprint vector and the reference noise spectrum vector is calculated. When the Euclidean distance exceeds the preset safety threshold but does not reach the critical value, a slight interference is indicated.

6. The capacitive coupling dynamic coding isolation method according to claim 1, characterized in that, In the preprocessing of grouping and polarity reversal of the original logic signal according to the optimal coding template, the original logic signal sequence to be sent is divided according to the grouping length specified by the template. For each group of signals, the preprocessing operation is performed according to the polarity reversal rule in the template. The polarity reversal strategy is used to reduce the occurrence of long 0s or long 1s and balance the DC component.

7. The capacitive coupling dynamic coding isolation method according to claim 1, characterized in that, In the process of extracting the original waveform at the receiving end and performing adaptive filtering based on the frequency band mask, an adaptive notch filter is designed using the reference noise spectrum vector as the frequency domain mask. The transfer function of this filter includes a Gaussian attenuation term for the center frequency of the dominant interference band, and the attenuation gain coefficient depends on the Euclidean distance calculated in real time.

8. The capacitive coupling dynamic coding isolation method according to claim 1, characterized in that, In the process of selecting an adaptive error correction strategy based on the Hamming distance value and performing correction operations, if the Hamming distance is small, linear block codes are used for single-bit or multi-bit correction; if the Hamming distance is large but still within the detectable range, an interleaving decoding strategy is initiated; if the Hamming distance exceeds the error correction limit, it is marked as an unrecoverable error.

9. The capacitive coupling dynamic coding isolation method according to claim 1, characterized in that, In the proposed method of predicting future interference patterns based on evolution curve trends and setting up parameter adjustment schemes, time series analysis is used to predict the interference trend within a short time window. If the prediction results show that the interference intensity will continue to rise, a parameter adjustment scheme is set up in advance. The new parameter set consists of the old parameter set plus the directional derivative term based on the gradient of the evolution curve.

10. A capacitively coupled dynamic coding isolation system for implementing the method as described in any one of claims 1-9, characterized in that, include: The synchronous acquisition module is used to synchronously acquire the common-mode transient voltage waveform and the output response of the coupling channel on the high-voltage side, calculate the cross-correlation function to extract the spatiotemporal correlation features, construct a multi-band noise energy distribution map and identify the dominant interference frequency band, and generate a reference noise spectrum vector. The real-time feature extraction module is used to extract real-time signal segments and perform short-time energy weighted statistics, calculate real-time feature fingerprint vectors and perform Euclidean distance measurement with the reference noise spectrum vector, determine the interference level based on the distance measurement result and update the interference confidence, and generate a dynamic interference fingerprint matching report. The adaptive coding mapping module is used to parse the dynamic interference fingerprint matching report and extract the optimal coding template under the current working condition. Based on the optimal coding template, the original logic signal is preprocessed by grouping and polarity reversal. Dynamic Manchester coding variants are introduced and instantaneous coding symbol sequences are calculated to generate the initial coding mapping data packet. The pulse suppression and reconstruction module is used to extract the original waveform at the receiving end and perform adaptive filtering based on the frequency band mask, calculate the instantaneous signal-to-noise ratio distribution map and determine the dynamic decision threshold sequence, perform dual-edge triggered logic reconstruction and remove single-cycle abnormal pulses, encapsulate the reconstructed logic frame and verify the consistency of the encoding template. The error correction module is used to parse the redundant check bits in the logical frame and calculate the Hamming distance matrix. Based on the Hamming distance value, it selects an adaptive error correction strategy and performs correction operations. It performs multi-source cross-validation and generates a data credibility score, and outputs the final clean data stream. The self-evolutionary optimization module is used to aggregate full-process performance indicators and construct multi-dimensional feature evolution curves. Based on the trend of the evolution curves, it predicts future interference patterns and presets parameter adjustment schemes. It performs online fine-tuning of parameters and updates local non-volatile storage configurations, and archives system running snapshots.

Images