A windowing denoising method and system for live frequency domain reflection signals

Abstract

The present application belongs to the technical field of measuring electrical variables, and relates to a windowing noise reduction method and system for charged frequency domain reflection signals, comprising the following steps: obtaining an ideal initial guess sequence by analyzing a theoretical frequency modulation function; obtaining a time-state noise transmission channel model by acquiring a near-end strong reflection signal sequence and the ideal initial guess sequence; obtaining an in-situ actual injection signal and a convergent physical channel characteristic model by performing adaptive inverse inversion and iterative verification; obtaining a non-ideal synthetic noise signal by convolution mapping; obtaining a noise reduction time domain signal by operating the non-ideal synthetic noise signal and an original mixed frequency beat signal; and obtaining a physical positioning map by performing frequency domain conversion and coordinate conversion on the noise reduction time domain signal. The present application solves the problem that the prior art cannot effectively strip the short-range self-interference signal generated by the nonlinear coupling of the dynamically changing signal source and the hardware physical channel under the live operation scenario, so as to cause the cable far-end weak defect characteristics to be shielded.

Description

Technical Field

[0001] This invention belongs to the technical field of measuring electrical variables, and relates to a windowing noise reduction method and system for charged frequency domain reflected signals. Background Technology

[0002] Distribution network cables are the core channels for power transmission, and their insulation health directly affects the reliability of power supply. Frequency domain reflection (FDR), with its non-destructive and high-sensitivity characteristics, has become an important technical means for detecting latent defects such as localized aging and moisture absorption in cables. In a live-line testing environment, the physical interface between the signal coupling device of the testing system and the cable body typically generates strong near-end reflections. This strong energy signal introduced by the testing hardware itself is easily deeply superimposed on the weak reflection signals generated by minor defects at the far end of the cable, thus increasing the difficulty of effectively extracting defect information.

[0003] To reduce interference components in reflected signals, existing technologies have proposed several signal processing strategies based on time-frequency domain conversion. Chinese invention patent application CN116593831A discloses a cable defect location method, device, and medium. This method converts the incident signal and the received reflected signal (including the cable head and defect location) to the time-frequency domain. Then, it converts the reflected time-frequency domain signal into a pseudo-frequency domain signal for filtering, and finally converts it back to the time-frequency domain to obtain the filtered time-frequency domain signal. The defect location information is obtained based on the cross-correlation function of the incident and filtered time-frequency domain signals. This scheme attempts to improve location accuracy by reducing high-frequency interference in the signal amplitude region through pseudo-frequency domain filtering.

[0004] However, the above methods still have limitations when dealing with complex on-site live-line inspection physical scenarios. This filtering mechanism mainly relies on mathematical threshold truncation rules and does not consider the strong short-range self-interference characteristics generated by the nonlinear coupling between dynamic non-ideal signal sources and hardware physical channels under energized conditions. Near-end strong reflections caused by parasitic parameters of connectors and hardware circuits often contain complex tail oscillations, whose spectral characteristics highly overlap with the weak beat frequencies of real small defects. Conventional filtering processes struggle to effectively remove this broadband mask signal originating from hardware distortion, making it easy for the weak beat energy characteristics at the far end of the cable to be masked by the residual strong noise substrate, thus limiting the detection sensitivity of small, weak defects to some extent.

[0005] Based on the above problems, the present invention aims to solve the problem that the existing technology is unable to effectively remove the short-range self-interference signal generated by the nonlinear coupling between the dynamically changing signal source and the hardware physical channel in the live operation scenario, thereby causing the weak defect features of the far end of the cable to be masked. Summary of the Invention

[0006] In a first aspect, the present invention provides a windowing noise reduction method for charged frequency domain reflected signals, comprising the following steps:

[0007] S1. Obtain the original mixing beat signal, and extract the original mixing beat signal to obtain the near-end strong reflection signal sequence;

[0008] S2. Extract the theoretical frequency modulation function from the system's underlying configuration, analyze the theoretical frequency modulation function, and obtain the ideal initial guess sequence;

[0009] S3. Obtain the near-end strong reflection signal sequence and the ideal initial guess sequence, perform system identification processing, and obtain the temporal noise propagation channel model;

[0010] S4. The temporal noise propagation channel model and the near-end strong reflection signal sequence are subjected to adaptive inverse inversion and iterative verification to obtain the in-situ actual injected signal and the convergent physical channel characteristic model.

[0011] S5. The convergent physical channel feature model and the in-situ actual injected signal are subjected to simulation filtering and discrete time-domain convolution mapping to obtain a non-ideal synthetic noise signal.

[0012] S6. Perform coherent subtraction between the non-ideal synthesized noise signal and the original mixing beat signal to obtain the denoised time-domain signal;

[0013] S7. The noise-reduced time-domain signal undergoes frequency domain transformation and coordinate conversion to obtain a physical location map.

[0014] A further embodiment of the present invention, wherein obtaining the original mixing beat signal and extracting a sequence of near-end strong reflection signals from the original mixing beat signal includes the following steps:

[0015] Acquire the hybrid echo analog quantity and perform equidistant digital sampling processing;

[0016] The original mixed beat signal is generated by combining equidistant digital sample values, and the spatiotemporal propagation delay constant is retrieved.

[0017] The forced cutoff timestamp is determined based on the spatiotemporal propagation delay constant, and the data segment of the original mixing beat signal located before the forced cutoff timestamp is extracted to obtain the near-end strong reflection signal sequence.

[0018] A further aspect of the present invention, wherein determining the forced cutoff timestamp based on the spatiotemporal propagation delay constant includes the following steps:

[0019] Obtain the preset safety redundancy coefficient;

[0020] The forced cutoff timestamp is obtained by multiplying the spatiotemporal propagation delay constant with the safety redundancy coefficient.

[0021] A further aspect of this invention involves extracting a theoretical frequency modulation function from the system's underlying configuration, analyzing the theoretical frequency modulation function, and obtaining an ideal initial guess sequence, comprising the following steps:

[0022] Extract the theoretical frequency modulation function and record the starting frequency point parameters and sweep bandwidth parameters of the theoretical frequency modulation function;

[0023] Based on the starting frequency parameters and the sweep bandwidth parameters, plot the theoretical time-domain curve of sweep frequency transmission, and analyze the theoretical time-domain curve of sweep frequency transmission.

[0024] The digital array consisting of a large number of discrete amplitude points is extracted and encapsulated as a reference discrete waveform matrix. A bidirectional mapping data stack is established, and the ideal initial guess sequence is output.

[0025] A further aspect of the present invention involves obtaining the near-end strong reflection signal sequence and the ideal initial guess sequence, performing system identification processing, and obtaining a temporal noise propagation channel model, including the following steps:

[0026] Activate the iterative decoupling loop logic and set the ideal initial guess sequence as the first round of initial excitation signal sequence;

[0027] The first round of initial excitation signal sequence and the near-end strong reflection signal sequence are used together as input parameters to run the autoregressive moving average system identification algorithm.

[0028] Solve the system parameter transfer feature matrix that fits the current input-output relationship using the autoregressive moving average system identification algorithm, extract the system parameter transfer feature matrix, and establish the temporal noise propagation channel model.

[0029] A further aspect of this invention involves performing adaptive inverse inversion and iterative verification on the temporal noise propagation channel model and the near-end strong reflection signal sequence to obtain the in-situ actual injected signal and the convergent physical channel characteristic model, including the following steps:

[0030] Lock the system parameter transfer feature matrix contained in the temporal noise transmission channel model, and call the adaptive inverse solver module with regularization constraints to process the near-end strong reflection signal sequence;

[0031] The adaptive inverse solver module with regularization constraints is used to invert and reconstruct the sequence. The mean square error residual between the inverted and reconstructed sequence and the previous round of iterative guess sequence is calculated.

[0032] The mean square error residual value is compared with the system's preset judgment threshold. If the mean square error residual value meets the convergence criterion, the iterative decoupling loop logic is stopped, the in-situ actual injected signal is output, and the temporal noise propagation channel model generated in the current iteration is locked as the convergent physical channel feature model.

[0033] A further aspect of this invention involves converging the physical channel feature model with the in-situ actual injected signal, performing simulated filtering and discrete-time convolution mapping to obtain a non-ideal synthetic noise signal, including the following steps:

[0034] The pole array parameters and zero array parameters stored inside the convergent physical channel feature model are analyzed, the pole array parameters and zero array parameters are loaded, and an infinite impulse response simulation digital filter is configured.

[0035] The in-situ actual injected signal is converted into a discrete digital excitation sequence, and the discrete digital excitation sequence is injected into the pre-buffer node of the infinite impulse response simulation digital filter.

[0036] The infinite impulse response simulation digital filter is started to perform a discrete-time convolution mapping process for the discrete digital excitation sequence, record the continuous charged interference waveform, and generate the non-ideal synthetic noise signal.

[0037] A further aspect of this invention involves performing a coherent subtraction operation between the non-ideal synthesized noise signal and the original mixing beat signal to obtain a denoised time-domain signal, comprising the following steps:

[0038] Scan the peak timestamp of the main oscillation period of the non-ideal synthesized noise signal, and perform an overall translation operation on the time axis based on the peak timestamp of the main oscillation period to complete phase alignment;

[0039] After segmenting and aligning the phases of the dual-channel signal waveform grid, the ratio between the instantaneous peak amplitudes within the dual-channel waveform grid with the same time index coordinates is calculated to generate an adaptive cancellation amplitude alignment coefficient.

[0040] The adaptive cancellation amplitude alignment coefficient is mapped to the non-ideal synthesized noise signal, and a time-domain coherent subtraction operation is performed between it and the original mixing beat signal to output the denoised time-domain signal.

[0041] A further aspect of this invention involves performing frequency domain transformation and coordinate conversion on the denoised time-domain signal to obtain a physical location map, including the following steps:

[0042] Retrieve the time series grid contained in the denoised time-domain signal and apply a time-frequency transformation matrix operator;

[0043] Extract the frequency domain reflection beat peak value output by the time-frequency transformation matrix operator, and record the center frequency scale parameter occupied by the frequency domain reflection beat peak value;

[0044] The preset electromagnetic wave ranging formula is invoked to process the center frequency scaling parameter, convert it into the physical position coordinates of the distance detection start point, and generate the physical positioning map.

[0045] Secondly, the present invention provides a windowing noise reduction system for charged frequency domain reflected signals, comprising the following modules:

[0046] The near-end strong reflection signal interception module acquires the original mixing beat signal, and intercepts the original mixing beat signal to obtain the near-end strong reflection signal sequence;

[0047] The ideal initial guess sequence generation module extracts the theoretical frequency modulation function from the system's underlying configuration, parses the theoretical frequency modulation function, and obtains the ideal initial guess sequence.

[0048] The temporal noise propagation channel model establishment module acquires the near-end strong reflection signal sequence and the ideal initial guess sequence, performs system identification processing, and obtains the temporal noise propagation channel model.

[0049] The iterative decoupling and convergence determination module performs adaptive inverse inversion and iterative verification on the temporal noise propagation channel model and the near-end strong reflection signal sequence to obtain the in-situ actual injected signal and the converged physical channel feature model.

[0050] The non-ideal synthetic noise signal reconstruction module converges the physical channel feature model and performs simulated filtering and discrete-time convolution mapping with the in-situ actual injected signal to obtain the non-ideal synthetic noise signal.

[0051] The time-domain coherent cancellation noise reduction module performs a coherent subtraction operation between the non-ideal synthesized noise signal and the original mixing beat signal to obtain the noise-reduced time-domain signal;

[0052] The physical location map generation module performs frequency domain transformation and coordinate conversion on the denoised time domain signal to obtain the physical location map.

[0053] In summary, the present invention has the following beneficial technical effects:

[0054] 1. An autoregressive moving average system identification and adaptive inverse inversion with regularization constraints are performed on near-end strong reflection signal sequences and ideal initial guess sequences. Without prior knowledge of the precise physical parameters of the hardware, the system can dynamically invert the in-situ actual injected signal containing physical distortions and the matching temporal noise propagation channel model. This process places injection source distortion and channel aberrations within a closed-loop iterative solution framework, adapting to state changes caused by environmental temperature drift or equipment contact variations, and providing a mechanism for adaptive calibration of dynamically non-ideal hardware characteristics.

[0055] 2. The pole and zero array parameters of the convergent physical channel feature model are extracted, an infinite impulse response simulation digital filter is configured, and this filter is used to perform discrete-time convolution mapping on the discrete digital excitation sequence to reconstruct the non-ideal synthetic noise signal. This process reproduces the nonlinear superposition state of the signal source and the channel, enabling the generated noise signal to retain the trailing oscillations and secondary reflection characteristics caused by the parasitic parameters of the hardware circuit, providing an objective digital noise benchmark for subsequent removal of complex short-range interference signal components.

[0056] 3. The peak timestamp of the main oscillation period of the non-ideal synthesized noise signal and the original mixing beat signal are phase-aligned, and an adaptive cancellation amplitude alignment coefficient is calculated to perform time-domain coherent subtraction. This strategy, through two-dimensional matching and alignment of time and amplitude, directionally eliminates the near-end hardware transmit dead-segment energy that masks subsequent far-end signals in the original data. This processing effectively reduces the short-range strong reflection masking effect caused by the coupling between the signal source and the hardware channel, allowing the far-end weak defect features that were originally submerged under the interference substrate to be revealed. This improves the ability to extract the characteristics of small defects in cables under live operation scenarios and solves the technical problem of weak defects being easily masked in the background technology. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention.

[0058] Figure 1 A flowchart illustrating an embodiment of this application is disclosed.

[0059] Figure 2 The present application discloses a trend curve showing the relationship between the safety redundancy coefficient, noise residual rate, and actual defect signal retention rate in the embodiments of this application.

[0060] Figure 3 The time-domain mapping diagram of the ideal initial guess sequence waveform and instantaneous frequency characteristics in the embodiments of this application is disclosed.

[0061] Figure 4A schematic diagram of the system modules in an embodiment of this application is disclosed. Detailed Implementation

[0062] The following is in conjunction with the appendix Figure 1 - Figure 4 A preferred description of the present invention is provided below.

[0063] See attached document Figure 1 This invention proposes a windowing noise reduction method for charged frequency domain reflected signals, comprising the following steps:

[0064] S1. Obtain the original mixing beat signal, and extract the original mixing beat signal to obtain the near-end strong reflection signal sequence;

[0065] S2. Extract the theoretical frequency modulation function from the system's underlying configuration, analyze the theoretical frequency modulation function, and obtain the ideal initial guess sequence;

[0066] S3. Obtain the near-end strong reflection signal sequence and the ideal initial guess sequence, perform system identification processing, and obtain the temporal noise propagation channel model;

[0067] S4. The temporal noise propagation channel model and the near-end strong reflection signal sequence are subjected to adaptive inverse inversion and iterative verification to obtain the in-situ actual injected signal and the convergent physical channel characteristic model.

[0068] S5. The convergent physical channel feature model and the in-situ actual injected signal are subjected to simulation filtering and discrete time-domain convolution mapping to obtain a non-ideal synthetic noise signal.

[0069] S6. Perform coherent subtraction between the non-ideal synthesized noise signal and the original mixing beat signal to obtain the denoised time-domain signal;

[0070] S7. The noise-reduced time-domain signal undergoes frequency domain transformation and coordinate conversion to obtain a physical location map.

[0071] In one embodiment of the present invention, step S1 includes the following steps:

[0072] Acquire the hybrid echo analog quantity and perform equidistant digital sampling processing;

[0073] The original mixed beat signal is generated by combining equidistant digital sample values, and the spatiotemporal propagation delay constant is retrieved.

[0074] The forced cutoff timestamp is determined based on the spatiotemporal propagation delay constant, and the data segment of the original mixing beat signal located before the forced cutoff timestamp is extracted to obtain the near-end strong reflection signal sequence.

[0075] To achieve raw signal acquisition and preprocessing at broadband RF testing points in the distribution network, a frequency domain reflection detection terminal is activated to perform this process. The frequency domain reflection detection terminal controls its built-in ring-coupled capacitor sensor array to inject a preset sweep frequency detection signal into the target distribution network cable, which is currently energized. Simultaneously, it captures the mixed echo signals generated by impedance discontinuities in the target distribution network cable line, particularly the near-end test system connection points. The mixed echo signal, as a continuously varying analog voltage, is transmitted to the analog-to-digital converter integrated into the frequency domain reflection detection terminal.

[0076] The analog-to-digital converter module uses the sampling frequency set by the system firmware. Mixed echo signals at equal time intervals Perform digital sampling to generate discrete digital sample values, sampling frequency The 1 GHz sampling rate is set to satisfy the Nyquist sampling theorem while also considering the resolution of wideband signal detection. Since the sweep bandwidth of this application can reach more than 200 MHz, in order to prevent spectral aliasing in digital sampling, the sampling rate must be at least twice the highest frequency. The 1 GHz sampling rate not only retains sufficient redundant bandwidth, but its corresponding 1 ns time resolution can also capture the small near-end strong reflection distortion characteristics at the device interface.

[0077] The digital signal processor of the frequency domain reflection detection terminal combines the discrete digital sampled values ​​acquired in time sequence into a one-dimensional time domain digital sequence, which is defined as the original mixing beat signal. The original mixing beat signal represents the beat mixing voltage amplitude captured by the analog-to-digital conversion module at the target sampling time.

[0078] The digital signal processor retrieves the pre-calibrated spatiotemporal propagation delay constant from its internal non-volatile memory. Spatiotemporal propagation delay constant The theoretical time required for the swept-frequency detection signal to complete one round trip within the physical coupling hardware of the test system, such as the connector and the ring capacitor itself, is determined based on the physical dimensions, geometry, and electromagnetic wave propagation speed of the medium used in the test system. This time is determined by the digital signal processor based on the spatiotemporal propagation delay constant. Calculate the mandatory deadline timestamp The specific calculation formula is as follows:

[0079]

[0080] In the formula, To enforce the deadline timestamp; The spacetime propagation delay constant; the safety redundancy coefficient. It is a dimensionless preset constant, a safety redundancy factor used to ensure signal integrity and avoid introducing too many far-end signals. Empirical values ​​are typically between 1.2 and 1.5, for example, 1.3. This is based on the fact that non-ideal hardware parasitic capacitance and inductance at the connection between the testing instrument and the cable can cause parasitic broadening of the reflected pulse on the time axis. Verification has shown that a safety redundancy factor of 1.2 to 1.5 can encompass and eliminate the broadening blind spot caused by hardware distortion. Values ​​below this range will result in residual equipment noise, while values ​​above this range will mistakenly delete the actual defect signal near the cable end, ensuring that the captured time window covers the time broadening of the reflected signal caused by hardware manufacturing tolerances or differences in field installation.

[0081] See appendix Figure 2 ,when When the value is less than 1.2, the device noise residual rate is high because it fails to completely cover the widened blind zone caused by hardware distortion. When the value is greater than 1.5, although the device noise residual rate drops to close to 0, it will mistakenly delete the real defect signal at the near end, resulting in a significant decrease in the retention rate of the real defect signal. Figure 2 The shaded area in the figure defines the empirical range of [1.2, 1.5] for the safety redundancy coefficient, which can strike a balance between eliminating equipment noise and retaining real defect signals.

[0082] The digital signal processor iterates through the data points of the original mixer beat signal and extracts the timestamps. Meet the conditions subsequence, timestamp Meet the conditions The signal components of the subsequence originate from the beat transient components generated by hardware mixing of the strong impedance mismatch reflection at the interface between the test instrument and the target distribution network cable. It is defined as the near-end strong reflection signal sequence that characterizes the original physical response of the coupling path and is cached for later use.

[0083] In a specific embodiment, it is assumed that the analog-to-digital conversion module of the frequency domain reflection detection terminal has a sampling frequency of If set to 1GHz, the sampling time interval will be... The spatiotemporal propagation delay constant is 1 ns. The safety redundancy factor is 50 ns. The value is 1.3. The digital signal processor calculates the forced cutoff timestamp. The analog-to-digital converter continuously outputs digital sampled values ​​to form the original mixing beat signal. The digital signal processor extracts data points with timestamps between 0 ns and 64 ns, as the next timestamp, 65 ns, is the mandatory cutoff timestamp. The range was discarded. The digital signal processor extracted the first 65 data points from the head of the original mixer beat signal and stored them as a sequence of near-end strong reflection signals.

[0084] In one embodiment of the present invention, step S2 includes the following steps:

[0085] Extract the theoretical frequency modulation function and record the starting frequency point parameters and sweep bandwidth parameters of the theoretical frequency modulation function;

[0086] Based on the starting frequency parameters and the sweep bandwidth parameters, plot the theoretical time-domain curve of sweep frequency transmission, and analyze the theoretical time-domain curve of sweep frequency transmission.

[0087] The digital array consisting of a large number of discrete amplitude points is extracted and encapsulated as a reference discrete waveform matrix. A bidirectional mapping data stack is established, and the ideal initial guess sequence is output.

[0088] This step aims to generate an ideal initial guess sequence based on the system's underlying configuration, serving as the starting point for subsequent iterative decoupling algorithms. The process of generating the ideal initial guess sequence based on the system's underlying configuration is executed by the digital signal processor (DSP) inside the frequency domain reflection detection terminal. The DSP accesses the storage unit, extracts and loads the theoretical frequency modulation function used to define the rules for generating the sweep frequency signal. The key attributes of the theoretical frequency modulation function are determined by the starting frequency parameters. With sweep bandwidth parameters Definition, starting frequency parameter With sweep bandwidth parameters The values ​​are usually set according to the requirements of detection distance and resolution for defect detection in power distribution network cables, such as the starting frequency parameter. The sweep bandwidth parameter is set between 1 MHz and 10 MHz. The frequency range is set between 20 MHz and 200 MHz. The starting frequency is set above 1 MHz to avoid the 50 Hz power frequency and its harmonics, as well as transient interference from switching, and other power frequency and low-order harmonic frequency bands. The sweep bandwidth is set between 20 MHz and 200 MHz based on the radar ranging resolution formula. ,in, This bandwidth represents the propagation speed of electromagnetic waves in the cable medium and provides decimeter-level spatial resolution, meeting the location requirements of local defects in power distribution network cables.

[0089] The digital signal processor reads the starting frequency and sweep bandwidth parameters, and combines them with the sweep period set by the system. The system performs digital waveform generation calculations to plot the theoretical time-domain curve of the swept-frequency transmission. To convert this curve into a format suitable for digital algorithm processing, the digital signal processor (DSP) uses the same sampling frequency as the analog-to-digital converter in S1. Applying a discretization sampling rule, it analyzes and evaluates the swept-frequency transmission time-domain curve at equal time intervals, generating a digital array composed of numerous discrete amplitude points. The discretization waveform generation formula for this digital array is as follows:

[0090] In the formula, The first one in the generated digital waveform The discrete amplitude of each sampling point is a dimensionless normalized value; The preset normalized amplitude for the linear frequency modulated continuous wave signal is usually set to 1. For indexes of discrete time series; The sampling time interval is defined by the discretized waveform generation formula, which calculates a linear frequency modulated continuous wave signal at each discrete sampling time. The phase of the signal is determined, and the trigonometric function value of the phase is taken to generate a waveform sequence that conforms to the theoretical model. In this discretized waveform generation formula, the derivative of the phase represents the instantaneous frequency of the signal. Since the formula contains a quadratic sweep phase term, its instantaneous frequency mapping characteristic shows a linear increasing relationship with time, thus covering the set sweep bandwidth.

[0091] The digital signal processor encapsulates the digital array consisting of a large number of discrete amplitude points into a reference discrete waveform matrix and establishes a bidirectional mapped data stack that supports fast index lookup and bidirectional data read / write. The bidirectional mapped data stack outputs an ideal initial guess sequence representing the injected signal and loads the ideal initial guess sequence into a cache.

[0092] In a specific embodiment, the configuration read by the digital signal processor from the firmware is the start frequency parameter. Set to 5 MHz, sweep bandwidth parameters Set to 40 MHz. System-defined sweep period. 100 μs, preset normalization amplitude The value is 1.0. The digital signal processor generates a discrete-time series according to the above configuration. For the first point of the discrete-time series, i.e. The magnitude of the first point in the discrete time series is .

[0093] For the second point of the discrete time series, i.e. Substituting into the formula to calculate the phase, since the single sampling interval is extremely short, the phase term of the second frequency sweep is approximately The magnitude of the second point in the discrete-time series is negligible and can be approximated. For the third point of the discrete time series, i.e. The amplitude of the third point in the discrete time series is approximately The digital signal processor will continue to calculate according to this pattern until a complete sequence covering the entire sweep cycle is generated. This complete sequence, such as [1.0, 0.9995, 0.9980, ...], is defined as the ideal initial guess sequence and stored in the cache. Under this configuration, the instantaneous frequency of this sequence at the initial time (i.e., ...) The frequency at point 0 is initially 5 MHz and increases upwards over time. This linearly changing instantaneous frequency mapping characteristic provides a digital reference with uniform coverage across the entire frequency band for subsequent identification algorithms.

[0094] See appendix Figure 3 The figure shows the normalized amplitude waveform of the ideal initial guess sequence over time, and its corresponding instantaneous frequency mapping characteristics. The solid line in the figure represents the smooth oscillating waveform of the sweep transmission theory time-domain curve; the dashed line visually reflects the instantaneous frequency mapping characteristics of the sequence on the time axis. This figure demonstrates that the discretely generated waveform can serve as a distortion-free reference benchmark, providing standard input for the iterative identification of the feature matrix of subsequent system parameter transfer.

[0095] In one embodiment of the present invention, step S3 includes the following steps:

[0096] Activate the iterative decoupling loop logic and set the ideal initial guess sequence as the first round of initial excitation signal sequence;

[0097] The first round of initial excitation signal sequence and the near-end strong reflection signal sequence are used together as input parameters to run the autoregressive moving average system identification algorithm.

[0098] Solve the system parameter transfer feature matrix that fits the current input-output relationship using the autoregressive moving average system identification algorithm, extract the system parameter transfer feature matrix, and establish the temporal noise propagation channel model.

[0099] This step initially extracts a channel model from the acquired signal that characterizes the physical distortion characteristics of the near-end hardware. The entire process of creating this channel model is automatically executed by the digital signal processor (DSP) of the frequency domain reflection detection terminal according to preset software logic. The DSP activates the iterative decoupling loop logic defined in its internal firmware.

[0100] In the first round of calculation of the iterative decoupling loop logic, the digital signal processor loads the ideal initial guess sequence generated by S2 from the cache and extracts a data segment of the same length as the near-end strong reflection signal sequence extracted by S1. The extracted data segment is set as the first round initial excitation signal sequence of the current execution cycle at the algorithm level, that is, in the first iteration, it is assumed that the excitation source is a distortion-free signal.

[0101] The digital signal processor takes the initial excitation signal sequence and the near-end strong reflection signal sequence extracted by S1 as input parameters and feeds them into a preset autoregressive moving average (ARMA) system identification algorithm. The ARMA system identification module is a system identification technique used to estimate the parameters of a linear system when the input signal is partially unknown or contains noise.

[0102] When the ARMA system identification module is running, it treats the near-end strong reflection signal sequence as the output of a linear time-invariant system. The initial excitation signal sequence of the first round is regarded as the input of a linear time-invariant system. Therefore, the following system transfer function model is established in the Z-transform domain:

[0103] In the formula, Represents the system transfer function model; This represents the output of a linear time-invariant system, i.e., the representation of a near-end strong reflection signal sequence in the Z-domain; This represents the input of a linear time-invariant system, specifically the representation of the initial excitation signal sequence in the Z-domain. Represents discrete complex variables in the Z-transform domain. Indicates delay Operators for each sampling time interval; An index for the number of steps in a discrete summation calculation; and These are the autoregressive order and moving average order of the system transfer function model, respectively. The values ​​of the autoregressive and moving average orders are preset based on the complexity of the hardware system, typically between 2 and 5 to balance accuracy and computational power. The order is determined by the impedance discontinuities at the front end of the test system, such as internal instrument interfaces, test lead clamps, and cable shielding grounding points, which usually have 2 to 4 strong reflection interfaces. Setting a filter model of order 2 to 5 allows for a mathematical mapping to the number of these physical reflection surfaces, avoiding underfitting and preventing computational divergence and overfitting due to high-order noise introduced by higher-order models. and These are the target parameters for solving the system transfer function model, dimensionless real coefficients, namely the autoregressive coefficients and moving average coefficients.

[0104] The ARMA system identification module constructs an overdetermined system of equations for all available data points and solves it using the least squares method. After the calculation, the ARMA system identification module outputs a set of coefficients that can fit the input-output relationship with the least squares error, i.e., the autoregressive coefficients. With moving average coefficient The set of coefficients is defined as the system parameter transfer characteristic matrix.

[0105] The digital signal processor encapsulates and stores the system parameter transfer feature matrix, and establishes a temporal noise transfer channel model that describes the degree of hardware distortion of the signal in the coupling path. The temporal noise transfer channel model will be used for calculation in subsequent iteration steps.

[0106] In a specific embodiment, the digital signal processor (DSP) initiates the first round of the iterative decoupling loop logic. The DSP loads the near-end strong reflection signal sequence output by S1. This sequence has a length of 65 points, and the instance values ​​of the near-end strong reflection signal sequence are assumed to be Y_seq=[0.0, 0.8, -0.3, 0.1, 0.05,...]. Simultaneously, the first 65 points of the ideal initial guess sequence output by S2 are loaded as the initial excitation signal sequence X_seq=[1.0, 0.9995, 0.9980,...]. The DSP sets the autoregressive order to... The moving average order is After the ARMA system identification module is activated, it searches for a set of coefficients that minimizes the mean square error between the theoretical output calculated by the difference equation and the actual observed Y_seq.

[0107] Assume that the system parameter transfer feature matrix ultimately calculated and output by the ARMA system identification module is an autoregressive coefficient matrix. , and moving average coefficient , The digital signal processor then encapsulates and stores the coefficient set {a: [0.5, -0.2], b: [0.9, -0.4]} containing autoregressive coefficients and moving average coefficients, generating the temporal noise propagation channel model for the first round of iterations.

[0108] In one embodiment of the present invention, step S4 includes the following steps:

[0109] Lock the system parameter transfer feature matrix contained in the temporal noise transmission channel model, and call the adaptive inverse solver module with regularization constraints to process the near-end strong reflection signal sequence;

[0110] The adaptive inverse solver module with regularization constraints is used to invert and reconstruct the sequence. The mean square error residual between the inverted and reconstructed sequence and the previous round of iterative guess sequence is calculated.

[0111] The mean square error residual value is compared with the system's preset judgment threshold. If the mean square error residual value meets the convergence criterion, the iterative decoupling loop logic is stopped, the in-situ actual injected signal is output, and the temporal noise propagation channel model generated in the current iteration is locked as the convergent physical channel feature model.

[0112] The in-situ injected signal is inverted through a reverse solution process, and the convergence of the iterative decoupling loop logic process is determined. This inversion of the in-situ injected signal and determination of convergence of the iterative decoupling loop logic process follow immediately after the previous step, executed by the digital signal processor under the control of the iterative decoupling loop logic. The digital signal processor locks the temporal noise propagation channel model output by S3 and extracts the autoregressive coefficients and moving average coefficient sets.

[0113] The digital signal processor calls the adaptive inverse solver module with regularization constraints pre-integrated in the firmware. The adaptive inverse solver module with regularization constraints is an algorithm module that uses numerical optimization methods to solve the inverse problem of discrete-time difference systems. The adaptive inverse solver module with regularization constraints introduces amplitude limiting and smoothness regularization constraints during the inversion process to avoid the iteration getting trapped in the local optimum of small perturbations.

[0114] The digital signal processor takes the near-end strong reflection signal sequence output by S1 and the temporal noise propagation channel model with extracted system parameter transfer feature matrix as inputs to the adaptive inverse solver module with regularization constraints. When the adaptive inverse solver module with regularization constraints runs, it treats the temporal noise propagation channel model as a description of a discrete-time difference system and reverse-engineers the unknown input signal that can generate the near-end strong reflection signal sequence under the action of the discrete-time difference system.

[0115] After the adaptive inverse solver module with regularization constraints completes the calculation, it outputs a one-dimensional time-domain sequence of the same length as the near-end strong reflection signal sequence, which can theoretically generate the observed near-end reflection and the inverse reconstruction sequence after hardware link distortion.

[0116] After outputting the inversion and reconstruction sequence, the digital signal processor retrieves the guessed sequence used in the previous iteration. For the current first loop, this is the ideal initial guessed sequence. The processor then calculates the error between the guessed sequence and the newly generated inversion and reconstruction sequence, obtaining the mean square error residual value used to quantify the improvement effect of the iteration. The specific calculation formula is as follows:

[0117] In the formula, The length of the sequence is the total number of sampling points currently involved in the calculation. It is the inversion reconstruction sequence output by the adaptive inverse solver module with regularization constraints. The amplitude of each sampling point; It is the first guess in the sequence used in the previous iteration. The amplitude of each sampling point. The above formula for calculating the mean square error residual value evaluates the similarity between two signal waveforms by calculating the average squared difference between corresponding points in the inverted reconstructed sequence and the guessed sequence.

[0118] The digital signal processor calculates the mean square error residual value. Compared with the system's preset threshold for determining whether convergence has occurred, for example The numerical comparison is performed, and the threshold setting is based on the quantization noise floor of the system's analog-to-digital converter. When the fitting error decreases to... At this magnitude, it indicates that the residual has been submerged under the quantization noise level of the hardware itself, and further iteration has no physical meaning. This serves as a convergence stopping boundary that meets system requirements. If the mean square error residual value... If the result is less than or equal to the decision threshold, it indicates that the two guesses of the input signal are close enough, and the iteration has converged.

[0119] At this point, the digital signal processor (DSP) terminates the iterative decoupling loop logic and saves the currently calculated inversion reconstruction sequence as the result, defined as the in-situ actual injected signal. Simultaneously, the DSP also locks the temporal noise propagation channel model generated in this iteration and marks the locked temporal noise propagation channel model as the convergent physical channel characteristic model for use in subsequent steps. If the mean square error residual value... If the result exceeds the threshold, the currently generated inversion and reconstruction sequence will be used as the initial excitation signal sequence for the next iteration, and execution will return to step S3.

[0120] In a specific embodiment, the digital signal processor (DSP) loads the temporal noise propagation channel model generated by S3, whose coefficients are {a: [0.5, -0.2], b: [0.9, -0.4]}. Simultaneously, the DSP calls the near-end strong reflection signal sequence Y_seq. The adaptive inverse solver module with regularization constraints inverts the inversion and reconstruction sequence as s_distorted=[0.9, 0.98, 0.75, ...]. The DSP then obtains the previous guess sequence s_guess=[1.0, 0.9995, 0.9980, ...] and calculates the mean square error residual between the inversion and reconstruction sequence and the previous guess sequence. .

[0121] Assuming the mean squared error residual value calculated from the complete sequence Value The digital signal processor will calculate the mean square error residual value. Compared with the system's preset judgment threshold Comparison. Because The convergence condition was not met. Therefore, the digital signal processor returns the inverted reconstructed sequence s_distorted as the new guess sequence s_guess to S3 to continue the next iteration; otherwise, it outputs the in-situ actual injected signal and the converged physical channel characteristic model.

[0122] In one embodiment of the present invention, step S5 includes the following steps:

[0123] The pole array parameters and zero array parameters stored inside the convergent physical channel feature model are analyzed, the pole array parameters and zero array parameters are loaded, and an infinite impulse response simulation digital filter is configured.

[0124] The in-situ actual injected signal is converted into a discrete digital excitation sequence, and the discrete digital excitation sequence is injected into the pre-buffer node of the infinite impulse response simulation digital filter.

[0125] The infinite impulse response simulation digital filter is started to perform a discrete-time convolution mapping process for the discrete digital excitation sequence, record the continuous charged interference waveform, and generate the non-ideal synthetic noise signal.

[0126] The converged physical channel characteristic model, decoupled from the previous steps, is used to digitally reconstruct the complete non-ideal noise signal generated by the combined effects of the signal source and the hardware channel. The reconstruction process is performed by the digital signal processor (DSP) of the frequency domain reflection detection terminal. The DSP retrieves the converged physical channel characteristic model, locked after S4 iteration convergence, from memory. The DSP then analyzes the converged physical channel characteristic model, extracting the pole array parameters and zero array parameters stored within it.

[0127] The pole and zero array parameters correspond to the lag term (i.e., autoregressive coefficient) and lead term (i.e., moving average coefficient) of the transfer function in the temporal noise propagation channel model, respectively, defining the channel's response characteristics to each frequency component, such as phase delay and amplitude gain. Based on the extracted pole and zero array parameters, the digital signal processor dynamically configures and loads a software system for simulating the real physical signal propagation path, namely an infinite impulse response (IIR) simulation digital filter. The transfer function of the IIR simulation digital filter is consistent with the convergent physical channel characteristic model.

[0128] Simultaneously, the digital signal processor (DSP) acquires the in-situ actual injected signal output from S4 and converts its one-dimensional time-domain sequence format into a discrete digital excitation sequence. The DSP then uses this discrete digital excitation sequence as input and injects it into the pre-buffer node of the IIR simulation digital filter.

[0129] After injection, the digital signal processor (DSP) activates the IIR-simulated digital filter, performing a discrete-time convolution mapping process on the input discrete digital excitation sequence. This allows the sequence representing the actual injected signal to pass through a digital system representing the actual hardware channel. The difference equation calculation formula for the discrete-time convolution mapping process is as follows:

[0130] In the formula, It is the actual in-situ injected signal as input at time point The sampled values; Index for discrete delay steps; These are elements of the pole array parameters (i.e., the corresponding autoregressive coefficients mentioned above); These are elements of the zero-point array parameters (i.e., the moving average coefficients mentioned above); and These are the autoregressive order and the moving average order of the IIR simulated digital filter, respectively. It is an IIR simulation of a digital filter at discrete time points The output value is obtained by calculating point by point.

[0131] The digital signal processor records and acquires the continuous charged interference waveform at the output port of the IIR simulated digital filter in real time. Because the continuous charged interference waveform incorporates distortion characteristics from both the non-ideal signal source and the non-ideal hardware channel, it is defined as a non-ideal synthetic noise signal and is stored for use in the noise cancellation step.

[0132] In a specific embodiment, it is assumed that the digital signal processor has obtained a converged physical channel characteristic model, with pole array parameters a = [0.45, -0.18] and zero array parameters b = [0.92, -0.38]. The digital signal processor obtains instance values ​​x = [0.9, 0.98, 0.75, ...] of the in-situ actual injected signal. The digital signal processor configures a second-order IIR simulated digital filter based on the pole array parameters and zero array parameters.

[0133] To calculate the first point of the non-ideal synthesized noise signal Digital signal processor executes To calculate the second point Digital signal processor executes The digital signal processor continues to operate according to this pattern, generating a complete sequence y=[0.828, 0.187, 0.38249, ...]. The calculated sequence reproduces the non-ideal synthetic noise signal of the real near-end strong reflection noise waveform.

[0134] In one embodiment of the present invention, step S6 includes the following steps:

[0135] Scan the peak timestamp of the main oscillation period of the non-ideal synthesized noise signal, and perform an overall translation operation on the time axis based on the peak timestamp of the main oscillation period to complete phase alignment;

[0136] After segmenting and aligning the phases of the dual-channel signal waveform grid, the ratio between the instantaneous peak amplitudes within the dual-channel waveform grid with the same time index coordinates is calculated to generate an adaptive cancellation amplitude alignment coefficient.

[0137] The adaptive cancellation amplitude alignment coefficient is mapped to the non-ideal synthesized noise signal, and a time-domain coherent subtraction operation is performed between it and the original mixing beat signal to output the denoised time-domain signal.

[0138] The raw data acquired from the cable under test is aligned and coherently subtracted from the non-ideal synthetic noise signal generated in the previous step to remove the strongest near-range shielding component. This process of removing the strongest near-range shielding component is executed by the digital signal processor (DSP) of the frequency domain reflectance detection terminal. The DSP performs a traversal scan of the raw mixing beat signal acquired by S1, locating the main oscillation period with highly concentrated energy by finding the point where the absolute value of the amplitude of the strong near-end reflection segment of the raw mixing beat signal reaches its maximum, and recording the time coordinate corresponding to the maximum point, i.e., the peak timestamp of the main oscillation period.

[0139] The digital signal processor uses the peak timestamp of the main oscillation period as the absolute time reference and performs overall translation operations such as cyclic shifting or interpolation resampling on the entire time axis of the non-ideal synthetic noise signal generated by S5. This ensures that the main peak of the non-ideal synthetic noise signal is strictly aligned in time with the main peak caused by near-end reflection in the original mixing beat signal. Thus, phase alignment is completed without destroying the time flight reference of the original signal, generating the translated non-ideal synthetic noise signal.

[0140] The digital signal processor (DSP) synchronously meshes the original mixing beat signal and the shifted non-ideal synthetic noise signal, constructing a dual-path waveform mesh of uniform length at the same index position. At each identical time index coordinate point, the DSP extracts the instantaneous peak amplitude of both the original mixing beat signal and the shifted non-ideal synthetic noise signal. The DSP calculates the ratio between the amplitude of the original mixing beat signal and the amplitude of the shifted non-ideal synthetic noise signal within the corresponding mesh, and then applies this ratio through weighted averaging or smoothing filtering to generate an adaptive cancellation amplitude alignment coefficient for dynamically adjusting the noise signal amplitude. .

[0141] Subsequently, the digital signal processor will adaptively cancel the amplitude alignment factor. As an independent doubling operator, a point-by-sample time-domain coherent subtraction operation is performed. The formula for the time-domain coherent subtraction operation is as follows:

[0142] In the formula, It is the original mixing beat signal at time point The sampled values; It is a phase-aligned, shifted, non-ideal synthesized noise signal at a specific time point. The sampled values; The time-domain sequence output after coherent subtraction in the time domain at discrete time points The sampled values. The time-domain coherent subtraction operation can remove the near-end hardware emission blind spot energy generated by the connection of the test equipment itself in the original data and mask the subsequent far-end reflected signal. The generated noise-reduced time-domain signal eliminates the masking effect of strong near-range reflection, so that the far-end local defect signal can be revealed.

[0143] In a specific embodiment, the digital signal processor (DSP) scans the original mixing beat signal y_raw acquired by S1, assuming that the main peak of the near-end reflection of the original mixing beat signal is located at a timestamp of 120 ns. The DSP then examines the non-ideal synthesized noise signal y_synth generated by S5 and finds that its main peak is located at a timestamp of 100 ns. Therefore, the DSP shifts the entire non-ideal synthesized noise signal y_synth to the right by 20 sampling points, obtaining a phase-aligned shifted non-ideal synthesized noise signal y_synth_shifted.

[0144] Digital signal processor calculates adaptive cancellation amplitude alignment factor Assuming the main peak amplitude of y_raw is 0.9108 and the main peak amplitude of y_synth_shifted is 0.828, then the adaptive cancellation amplitude alignment coefficient... The digital signal processor multiplies each point of the shifted, non-ideal synthesized noise signal y_synth_shifted by 1.1 to obtain the amplitude-adjusted noise signal, and performs time-domain coherent subtraction operations, such as calculating the alignment point of the main peak. Noise reduction time domain signal By performing coherent subtraction in the time domain, the near-end main reflection peak caused by near-end reflection in the original signal is eliminated, while the true radio frequency echo time delay property is preserved.

[0145] In one embodiment of the present invention, step S7 includes the following steps:

[0146] Retrieve the time series grid contained in the denoised time-domain signal and apply a time-frequency transformation matrix operator;

[0147] Extract the frequency domain reflection beat peak value output by the time-frequency transformation matrix operator, and record the center frequency scale parameter occupied by the frequency domain reflection beat peak value;

[0148] The preset electromagnetic wave ranging formula is invoked to process the center frequency scaling parameter, convert it into the physical position coordinates of the distance detection start point, and generate the physical positioning map.

[0149] Spectral analysis and physical distance calculation are performed on the denoised time-domain signal to generate intuitive defect location results. The spectral analysis and physical distance calculation process is completed by the digital signal processor (DSP) of the frequency domain reflection detection terminal. The DSP retrieves the denoised time-domain signal output from the S6 from memory.

[0150] To convert the broadband radio frequency signal into a beat signal containing distance information, the digital signal processor (DSP) extracts the ideal initial guess sequence generated by S2, performs point-by-point multiplication of the denoised time-domain signal with the ideal initial guess sequence, and performs digital mixing. The DSP then passes the resulting multiplied sequence through a preset digital low-pass filter to remove the sum frequency component generated by mixing, retaining the beat frequency component containing fault distance information. Based on the Nyquist sampling theorem, the DSP performs digital downsampling on the filtered sequence, thereby significantly compressing the data dimensionality while preserving the beat information, generating a filtered beat time-domain sequence, and accessing the time series grid contained within the beat time-domain sequence.

[0151] The digital signal processor (DSP) configures and invokes the time-frequency transformation matrix operator based on the total number of sample points after downsampling. The DSP takes the beat time-domain sequence as input, applies the time-frequency transformation matrix operator to perform operations from the time domain to the frequency domain, and outputs a complex spectrum sequence.

[0152] The digital signal processor (DSP) traverses the output complex spectrum sequence and searches for isolated peaks with amplitudes higher than the background noise floor. These isolated peaks, generated by reflections from distant local defects, are the frequency domain reflection beat peaks. These peaks correspond to local impedance discontinuities in the cable line, such as latent aging or moisture defects. The DSP records the frequency coordinates of the identified frequency domain reflection beat peaks on the frequency axis; these coordinates are defined as the center frequency scale parameter. .

[0153] The digital signal processor (DSP) retrieves a physical formula from its internal physical parameter library that matches the type of cable being tested, such as a single-core cross-linked polyethylene insulated cable. This formula is a preset electromagnetic wave ranging formula, which describes a fixed mapping relationship between beat frequency and physical distance. The DSP then extracts the center frequency scaling parameter. The preset electromagnetic wave ranging formula is substituted into the input to calculate the final distance detection starting point's physical position coordinates, which are displayed in physical length units. The formula for calculating physical distance conversion is as follows:

[0154] In the formula, It is the propagation speed constant of electromagnetic waves in the target cable medium. For standard cross-linked polyethylene insulated cables, the propagation speed constant is usually pre-calibrated at [value missing]. m / s, this constant is derived from the electromagnetic wave propagation characteristics of cross-linked polyethylene insulation material, and the relative permittivity of cross-linked polyethylene insulation material. Approximately 2.3, according to the wave speed formula, , Using the speed of light as a constant, and correcting for the skin effect of the cable core, the actual electromagnetic wave propagation speed stabilizes at... The speed is around m / s, providing an accurate benchmark for distance conversion; It is the duration of a single frequency sweep; It is the sweep bandwidth. The coefficient 2 in the denominator of the formula is because the signal travels twice the path during its journey to the defect point and back.

[0155] By repeatedly performing frequency domain transformation and physical distance conversion on all identified frequency domain reflection beat peaks, the digital signal processor generates information that marks the physical location of each potential defect point, such as a list containing the distance of each defect or a graph marking the defect location on the distance axis, thus generating a map that marks the physical location of each potential defect point. This information is a physical location map that can intuitively show the local aging response of the line.

[0156] In a specific embodiment, the digital signal processor (DSP) performs point-by-point multiplication and low-pass filtering on the denoised time-domain signal output by S6 and the ideal initial guess sequence, and then performs a Fast Fourier Transform on the resulting beat time-domain sequence. Assuming that the spectral analysis results show frequency domain reflection beat peaks above the background noise, the DSP determines the center frequency scaling parameter of the frequency domain reflection beat peaks. The frequency is 20 kHz. The digital signal processor then calls the preset electromagnetic wave ranging formula, which includes a built-in electromagnetic wave propagation speed constant. for The duration of a single frequency sweep configured in the system. The sweep frequency is 100μs, and the bandwidth (BW) is 40MHz. The digital signal processor calculates... Calculation results show that at a distance of approximately [distance missing] from the start of the test... The location has potential defects. The digital signal processor records the location information of potential defects from the test start point. After all analyses are completed, a physical location map containing all defect locations is generated, providing a basis for line maintenance personnel to troubleshoot.

[0157] See appendix Figure 4 The present invention also proposes a windowing noise reduction system for charged frequency domain reflected signals, comprising the following modules:

[0158] The near-end strong reflection signal interception module acquires the original mixing beat signal, and intercepts the original mixing beat signal to obtain the near-end strong reflection signal sequence;

[0159] The ideal initial guess sequence generation module extracts the theoretical frequency modulation function from the system's underlying configuration, parses the theoretical frequency modulation function, and obtains the ideal initial guess sequence.

[0160] The temporal noise propagation channel model establishment module acquires the near-end strong reflection signal sequence and the ideal initial guess sequence, performs system identification processing, and obtains the temporal noise propagation channel model.

[0161] The iterative decoupling and convergence determination module performs adaptive inverse inversion and iterative verification on the temporal noise propagation channel model and the near-end strong reflection signal sequence to obtain the in-situ actual injected signal and the converged physical channel feature model.

[0162] The non-ideal synthetic noise signal reconstruction module converges the physical channel feature model and performs simulated filtering and discrete-time convolution mapping with the in-situ actual injected signal to obtain the non-ideal synthetic noise signal.

[0163] The time-domain coherent cancellation noise reduction module performs a coherent subtraction operation between the non-ideal synthesized noise signal and the original mixing beat signal to obtain the noise-reduced time-domain signal;

[0164] The physical location map generation module performs frequency domain transformation and coordinate conversion on the denoised time domain signal to obtain the physical location map.

[0165] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0166] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A windowing noise reduction method for charged frequency domain reflected signals, characterized in that, Includes the following steps: S1. Obtain the original mixing beat signal, and extract the original mixing beat signal to obtain the near-end strong reflection signal sequence; S2. Extract the theoretical frequency modulation function from the system's underlying configuration, analyze the theoretical frequency modulation function, and obtain the ideal initial guess sequence; S3. Obtain the near-end strong reflection signal sequence and the ideal initial guess sequence, perform system identification processing, and obtain the temporal noise propagation channel model; S4. The temporal noise propagation channel model and the near-end strong reflection signal sequence are subjected to adaptive inverse inversion and iterative verification to obtain the in-situ actual injected signal and the convergent physical channel characteristic model. S5. The convergent physical channel feature model and the in-situ actual injected signal are subjected to simulation filtering and discrete time-domain convolution mapping to obtain a non-ideal synthetic noise signal. S6. Perform coherent subtraction between the non-ideal synthesized noise signal and the original mixing beat signal to obtain the denoised time-domain signal; S7. The noise-reduced time-domain signal undergoes frequency domain transformation and coordinate conversion to obtain a physical location map.

2. The windowing noise reduction method for charged frequency domain reflected signals according to claim 1, characterized in that, The process of acquiring the original mixing beat signal and extracting a sequence of near-end strong reflection signals includes the following steps: Acquire the hybrid echo analog quantity and perform equidistant digital sampling processing; The original mixed beat signal is generated by combining equidistant digital sample values, and the spatiotemporal propagation delay constant is retrieved. The forced cutoff timestamp is determined based on the spatiotemporal propagation delay constant, and the data segment of the original mixing beat signal located before the forced cutoff timestamp is extracted to obtain the near-end strong reflection signal sequence.

3. The windowing noise reduction method for charged frequency domain reflection signals according to claim 2, characterized in that, Determining the forced cutoff timestamp based on the spatiotemporal propagation delay constant includes the following steps: Obtain the preset safety redundancy coefficient; The forced cutoff timestamp is obtained by multiplying the spatiotemporal propagation delay constant with the safety redundancy coefficient.

4. The windowing noise reduction method for charged frequency domain reflected signals according to claim 1, characterized in that, Extracting the theoretical frequency modulation function from the system's underlying configuration, analyzing the theoretical frequency modulation function, and obtaining the ideal initial guess sequence include the following steps: Extract the theoretical frequency modulation function and record the starting frequency point parameters and sweep bandwidth parameters of the theoretical frequency modulation function; Based on the starting frequency parameters and the sweep bandwidth parameters, plot the theoretical time-domain curve of sweep frequency transmission, and analyze the theoretical time-domain curve of sweep frequency transmission. The digital array consisting of a large number of discrete amplitude points is extracted and encapsulated as a reference discrete waveform matrix. A bidirectional mapping data stack is established, and the ideal initial guess sequence is output.

5. The windowing noise reduction method for charged frequency domain reflection signals according to claim 1, characterized in that, The near-end strong reflection signal sequence and the ideal initial guess sequence are obtained, and system identification processing is performed to obtain the temporal noise propagation channel model, including the following steps: Activate the iterative decoupling loop logic and set the ideal initial guess sequence as the first round of initial excitation signal sequence; The first round of initial excitation signal sequence and the near-end strong reflection signal sequence are used together as input parameters to run the autoregressive moving average system identification algorithm. Solve the system parameter transfer feature matrix that fits the current input-output relationship using the autoregressive moving average system identification algorithm, extract the system parameter transfer feature matrix, and establish the temporal noise propagation channel model.

6. The windowing noise reduction method for charged frequency domain reflection signals according to claim 5, characterized in that, The temporal noise propagation channel model and the near-end strong reflection signal sequence are subjected to adaptive inverse inversion and iterative verification to obtain the in-situ actual injected signal and the convergent physical channel characteristic model, including the following steps: Lock the system parameter transfer feature matrix contained in the temporal noise transmission channel model, and call the adaptive inverse solver module with regularization constraints to process the near-end strong reflection signal sequence; The adaptive inverse solver module with regularization constraints is used to invert and reconstruct the sequence. The mean square error residual between the inverted and reconstructed sequence and the previous round of iterative guess sequence is calculated. The mean square error residual value is compared with the system's preset judgment threshold. If the mean square error residual value meets the convergence criterion, the iterative decoupling loop logic is stopped, the in-situ actual injected signal is output, and the temporal noise propagation channel model generated in the current iteration is locked as the convergent physical channel feature model.

7. The windowing noise reduction method for charged frequency domain reflection signals according to claim 1, characterized in that, The convergent physical channel characteristic model is subjected to simulated filtering and discrete-time convolution mapping with the in-situ actual injected signal to obtain a non-ideal synthetic noise signal, including the following steps: The pole array parameters and zero array parameters stored inside the convergent physical channel feature model are analyzed, the pole array parameters and zero array parameters are loaded, and an infinite impulse response simulation digital filter is configured. The in-situ actual injected signal is converted into a discrete digital excitation sequence, and the discrete digital excitation sequence is injected into the pre-buffer node of the infinite impulse response simulation digital filter. The infinite impulse response simulation digital filter is started to perform a discrete-time convolution mapping process for the discrete digital excitation sequence, record the continuous charged interference waveform, and generate the non-ideal synthetic noise signal.

8. The windowing noise reduction method for charged frequency domain reflected signals according to claim 1, characterized in that, The non-ideal synthesized noise signal and the original mixing beat signal are coherently subtracted to obtain the denoised time-domain signal, including the following steps: Scan the peak timestamp of the main oscillation period of the non-ideal synthesized noise signal, and perform an overall translation operation on the time axis based on the peak timestamp of the main oscillation period to complete phase alignment; After segmenting and aligning the phases of the dual-channel signal waveform grid, the ratio between the instantaneous peak amplitudes within the dual-channel waveform grid with the same time index coordinates is calculated to generate an adaptive cancellation amplitude alignment coefficient. The adaptive cancellation amplitude alignment coefficient is mapped to the non-ideal synthesized noise signal, and a time-domain coherent subtraction operation is performed between it and the original mixing beat signal to output the denoised time-domain signal.

9. The windowing noise reduction method for charged frequency domain reflection signals according to claim 1, characterized in that, The denoised time-domain signal undergoes frequency-domain transformation and coordinate conversion to obtain a physical location map, including the following steps: Retrieve the time series grid contained in the denoised time-domain signal and apply a time-frequency transformation matrix operator; Extract the frequency domain reflection beat peak value output by the time-frequency transformation matrix operator, and record the center frequency scale parameter occupied by the frequency domain reflection beat peak value; The preset electromagnetic wave ranging formula is invoked to process the center frequency scaling parameter, convert it into the physical position coordinates of the distance detection start point, and generate the physical positioning map.

10. A windowing noise reduction system for charged frequency domain reflected signals, characterized in that, Includes the following modules: The near-end strong reflection signal interception module acquires the original mixing beat signal, and intercepts the original mixing beat signal to obtain the near-end strong reflection signal sequence; The ideal initial guess sequence generation module extracts the theoretical frequency modulation function from the system's underlying configuration, parses the theoretical frequency modulation function, and obtains the ideal initial guess sequence. The temporal noise propagation channel model establishment module acquires the near-end strong reflection signal sequence and the ideal initial guess sequence, performs system identification processing, and obtains the temporal noise propagation channel model. The iterative decoupling and convergence determination module performs adaptive inverse inversion and iterative verification on the temporal noise propagation channel model and the near-end strong reflection signal sequence to obtain the in-situ actual injected signal and the converged physical channel feature model. The non-ideal synthetic noise signal reconstruction module converges the physical channel feature model and performs simulated filtering and discrete-time convolution mapping with the in-situ actual injected signal to obtain the non-ideal synthetic noise signal. The time-domain coherent cancellation noise reduction module performs a coherent subtraction operation between the non-ideal synthesized noise signal and the original mixing beat signal to obtain the noise-reduced time-domain signal; The physical location map generation module performs frequency domain transformation and coordinate conversion on the denoised time domain signal to obtain the physical location map.

Images