Real-time lightning location method based on FPGA signal acquisition preprocessing

Abstract

The application discloses a real-time lightning positioning method, a plurality of stations collect lightning signals synchronously through GNSS to obtain waveform data; the lightning signal waveform data are subjected to FFT band-pass filtering and denoising processing; a sliding window dynamic threshold is adopted to extract peak value characteristics, peak value characteristics of the waveform in different time windows are obtained, t p , A p , W p ), a local noise standard deviation is calculated, only effective peak values are reserved, lightning data real-time feedback is realized while ensuring positioning accuracy. A calculation center reconstructs the waveform based on the peak value characteristics, t p , A p , W p ), the reconstructed waveform more accurately reflects the characteristics of the lightning signal, and reduces time delay measurement errors caused by signal distortion. The reconstructed lightning waveform data of every two stations are calculated to obtain a maximum correlation time delay, a weighted nonlinear least square method is adopted to solve a TDOA equation set according to signal time delay information, and three-dimensional lightning positioning results are output. The positioning result reliability is evaluated through residual analysis and a geometric dilution of precision.

Description

Technical Field

[0001] The present invention relates to the fields of meteorology, atmospheric electricity and space science and technology, and in particular to a real-time lightning location method based on peak waveform reconstruction cross-correlation. Background Art

[0002] Lightning is a key phenomenon in atmospheric electromagnetic activity, holding significant research implications for meteorology, atmospheric electricity, and space science. Lightning location, which involves detecting the charge structure and density distribution within thunderstorm clouds and their evolution as the thunderstorm progresses, holds significant scientific significance for uncovering the physical mechanisms of lightning generation and development. Numerous existing lightning location methods exist, such as time-of-arrival (TOA) and time-difference-of-arrival (TDOA), which are currently common. Due to the high sampling rate of lightning, lightning signals must be sampled at a rate of ≥10 MS / s. The amount of sampled data from each lightning event is enormous, typically reaching gigabytes. This data cannot be transmitted in real time via conventional communication networks, and typically relies on local storage, reprocessing, and transmission. This results in location delays and is unable to meet the requirements of scenarios requiring sub-second response times. Existing real-time methods simply extract peak time and amplitude for matching. However, due to the overlapping peaks of lightning waveforms, mismatch rates can reach as high as 15%-20%, resulting in low peak matching efficiency. Signals received by detection stations often contain noise and other interference. Low-frequency (VLF / LF) signals are susceptible to interference from power-frequency noise and surface charges, which can obscure the peak characteristics of the signal. Lightning signals can be affected by noise, terrain obstruction, and electromagnetic interference during propagation, leading to signal distortion. Therefore, developing a real-time lightning location method is essential. Summary of the Invention

[0003] The purpose of this invention is to provide a real-time lightning location method, which adopts FPGA signal acquisition preprocessing, multi-station GNSS synchronous acquisition of lightning signals and FFT bandpass filtering to denoise, uses sliding window dynamic threshold to extract peak features, calculates local noise standard deviation, retains only effective peaks, and greatly compresses the original signal data volume, realizing real-time return of lightning signals. Through waveform feature compression and reconstruction technology, based on peak parameters ( t p , A p , W p ) reconstruct the waveform. In | t - t p |≤ W p The waveform is fitted within the range and the rectangular function waveform is used for reconstruction. The cross-correlation time difference optimization algorithm is used to accelerate the calculation of the cross-correlation function through FFT, and the time delay between measuring stations is accurately calculated through cross-correlation analysis, with the time delay resolution increased to 0.1 ΔtThe TDOA equations are solved using a weighted nonlinear least squares method to obtain lightning location results. Residual screening and GDOP evaluation are then used to improve location reliability.

[0004] The technical solution adopted by the present invention is to provide a real-time lightning location method based on FPGA signal acquisition and preprocessing, comprising the following steps:

[0005] S1, N measuring stations obtain lightning signal waveform data;

[0006] S2, performing denoising processing on the lightning signal waveform data;

[0007] S3. Extract the peak features of the waveform signal to obtain the positive and negative peak amplitudes, absolute time, and half-peak width of the waveform in different time windows, and transmit the information back to the computing center;

[0008] S4, the computing center reconstructs the waveform based on the peak characteristic information;

[0009] S5. Calculate the cross-correlation of the reconstructed lightning waveform data between every two measuring stations and extract the maximum correlation time delay;

[0010] S6. Solve the TDOA equations using a weighted nonlinear least squares method based on the signal time delay information and output a three-dimensional lightning location result.

[0011] S7. Evaluate the reliability of positioning results through residual analysis and geometric precision coefficient.

[0012] In steps S1, S2, and S3, the collected lightning data is pre-processed using FPGA logic gate circuits at the measuring station, peak information is extracted, and the data is transmitted back to the computing center in real time via the 5G network. In steps S4, S5, S6, and S7, waveform reconstruction, cross-correlation calculation, lightning location, and result evaluation are performed in the computing center.

[0013] The specific method of S1 obtaining multiple lightning signal waveform data is as follows: i The lightning signal waveform collected by each measuring station is s i ( t ),in t For time, i = 1, 2, …, N , N is the number of stations, and the sampling time interval of each station is Δ t , the number of sampling points is M , baseline correction eliminates DC offset to a baseline of 0; the signal can be expressed in discrete form:

[0014] ,

[0015] GNSS timing modules are used between each measuring station to achieve time synchronization, with a synchronization error of less than 25n s , sampling rate f s ≥10MS / s.

[0016] The S2 denoising process adopts the Fourier transform denoising method, which specifically includes: S i ( t ) Perform fast Fourier transform (FFT) to obtain the frequency domain signal S i ( f ), identify the frequency range of the noise, and pass it through a bandpass filter H BP ( f ) removes high-frequency noise and low-frequency trends, converts the processed frequency domain signal back to the time domain through inverse Fourier transform (IFFT), and further performs baseline correction on the denoised signal to eliminate DC offset to obtain the denoised lightning signal. The Fourier transform denoising formula is as follows:

[0017] ,

[0018] After removing the high-frequency noise, the filtered frequency domain signal is subjected to inverse Fourier transform (IFFT) to restore the time domain signal and obtain the denoised signal. s i de ( t ):

[0019] ,

[0020] in, s i de ( t ) is the i A denoised signal, IFFT is the inverse fast Fourier transform, which is used to convert the frequency domain signal back to the time domain. S i ( f ) is the i The frequency domain of a signal, H BP ( f ) is the frequency domain response of the filter.

[0021] The S3 peak feature extraction method is as follows: s i de ( t ), the peak feature extraction uses a sliding time window to detect positive and negative peaks, and the amplitude threshold is set to 3 σ i( σ i The absolute time of the peak value is recorded as the absolute time of the positive and negative peak values A p The absolute time of the peak value is recorded as the absolute time of the positive and negative peak values σ i The peak value is an effective peak value, and the absolute time of the positive and negative peak values is recorded t p The amplitude A p and half-peak width W p of the peak value are recorded

[0022] ,

[0023] wherein, t p + and t p − are the times of the positive and negative peak values, respectively A p + and A p − are the amplitudes of the positive and negative peak values, respectively

[0024] The amplitude A p , absolute time t p and half-peak width W p of the peak value are recorded W p The calculation formula of the half-peak width

[0025] ,

[0026] wherein, t left and t right are the time points at which the peak value amplitude is half.

[0027] The S4 waveform reconstruction is based on the extracted peak feature points t p , A p , W p A single peak value is fitted with a rectangular function, and the data outside the peak value is zero. The rectangular function is:

[0028] ,

[0029] All positive and negative peak values are superimposed to generate a reconstructed waveform:

[0030] ,

[0031] where: s i re ( t ) is the reconstructed signal of the i th station, ∑ p∈Peaki+ and ∑ p ∈Peaki− are the summations of the positive and negative peaks, respectively, A p + and A p − are the amplitudes of the positive and negative peaks, respectively, t p + and t p − are the time positions of the positive and negative peaks, respectively, W p + and W p − are the widths of the positive and negative peaks, respectively.

[0032] The S5 cross-correlation uses linear cross-correlation with zero padding, for the reconstructed waveforms i and j of station S and station i re ( t ) and S j re ( t ), the waveforms need to be aligned in baseline before cross-correlation, the cross-correlation function is calculated as:

[0033] ,

[0034] R ij ( τ ) is the cross-correlation function of the lightning waveform of the i th station and the j th station, which describes the similarity of the two waveforms at different time delays τ ; s i re ( t ) and s j re ( t ): are the cross-correlation functions of thei the first station and the second station reconstruct the signal, in discrete form: j

[0035] (zero padding when out of range),

[0036] find the delay time that maximizes the function value in the cross-correlation function R ij τ and extract the maximum correlation time delay τ ij

[0037]

[0038] wherein, τ ij represents the maximum correlation time delay between the first station and the second station of the lightning waveform i j extract the maximum correlation coefficient and the corresponding time delay .

[0039] The S6 positioning algorithm TDOA equation set uses the following formula:

[0040]

[0041] wherein, ( x , y , z ) is the lightning position, ( x i , y i , z i ) and ( x j , y j , z j ) are the positions of the station i and the station j , c is the speed of light, ε ij is the observation residual, i , j =1,2,..., N and i ≠ j , N ≥5;

[0042] The TDOA algorithm is based on the initial estimate of the least square, and calculates the initial value​​​​​​​x 0 ,y 0 ,z 0); construct the objective function and solve it iteratively:

[0043] ,

[0044] in, , weight ;

[0045] right z The ≥0 constraint is processed by projection method. When the iteration step ||Δx||<1m or the function change rate Or terminate when the number of iterations ≥ 100.

[0046] The S7 reliability evaluation includes: if the time delay residuals of multiple station pairs The root mean square (RMS) is greater than 0.1 μs , then remove the station pairs with amplitude correlation lower than 0.8 and recalculate. τ ij me is the measured value, τ ij ca is the calculated value;

[0047] Calculate the geometric dilution of precision (GDOP) for error assessment:

[0048] ,

[0049] in: J T is the observation matrix J The transpose of J T J is a m × m The matrix of is called the normal equation matrix, ( J T J ) −1 is the inverse matrix of the normal equation matrix, trace(( J T J ) −1 ) is the trace of the inverse matrix, that is, the sum of the diagonal elements of the inverse matrix;

[0050] When GDOP>5, the residual screening mechanism is automatically triggered, including:

[0051] a) Residual analysis, calculate the time difference residual of each station pair If it is greater than 0.2 μ s is marked as an outlier;

[0052] b) Geometric Dilution of Precision (GDOP) error evaluation, when GDOP>3, automatically remove the largest 20% of the residual station to recalculate;

[0053] c) Result output, generate a positioning result containing three-dimensional coordinates, elevation, confidence (95% elliptical area) and signal-to-noise ratio based weight.

[0054] The present application has the beneficial effects that: through lightning data preprocessing of the station, only the peak characteristics (time, amplitude, half-peak width) are transmitted, the single lightning data volume is reduced to the MB level, the data compression rate is >90%, 5G real-time backhaul is supported, real-time lightning positioning is carried out, and the problem that the prior art cannot meet the requirement of second-level response scene is solved. Key features are extracted from the original signal, and the reconstructed waveform can more accurately reflect the characteristics of the lightning signal through peak waveform reconstruction, reducing the time delay measurement error caused by signal distortion. The lightning signals received by different stations may have amplitude and phase differences, and waveform reconstruction can unify the signal characteristics, facilitating subsequent cross-correlation analysis. The cross-correlation function can effectively calculate the time delay between two signals, even in the case of signal noise or distortion, the relative time difference of the signal can be determined by finding the maximum value of the cross-correlation function. Lightning signals are usually complex electromagnetic pulses, and the cross-correlation method can process signals of different frequency components, and the time delay calculated by the cross-correlation method, combined with the TDOA algorithm, can significantly improve the accuracy of lightning positioning. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 The lightning positioning method of the present application is a step block diagram;

[0056] Figure 2 The principle block diagram of the positioning method of the present application;

[0057] Figure 3 Comparison diagram of original waveform and reconstructed waveform of lightning signal;

[0058] Figure 4 The partial enlarged view of Figure 3 ;

[0059] Figure 5 Comparison diagram of cross-correlation results of lightning signal waveform. DETAILED DESCRIPTION

[0060] In order for those skilled in the art to better understand the present application, the specific embodiments of the present application are described in conjunction with Figures 1 to 5 The present application is further described, and the content mentioned in the embodiments is not a limitation of the present application.

[0061] The real-time lightning positioning method based on FPGA signal acquisition and preprocessing of the application comprises the following steps: N stations acquire lightning signal waveform data; the lightning signal waveform data is denoised; the peak value characteristics of the waveform signal are extracted to obtain the positive and negative peak value amplitudes, absolute time and half-peak width information of the waveform in different time windows, which are returned to the calculation center; the calculation center reconstructs the waveform according to the peak value characteristic information; the reconstructed lightning waveform data of each two stations are correlated to extract the maximum correlation time delay; the TDOA equation set is solved by using the weighted nonlinear least square method according to the signal time delay information, and the three-dimensional lightning positioning result is output; and the positioning result reliability is evaluated through residual analysis and geometric dilution of precision. Figure 1 .

[0062] The specific method for acquiring N lightning signal waveform data is as follows: let the lightning signal waveform collected by the i-th station be i , s i ( t ), wherein t is time, i = 1, 2, …, N , N is the number of stations, the sampling time interval of each station is Δ t , the number of sampling points is M , and the direct current offset is eliminated to the baseline of 0. The signal can be expressed in a discrete form:

[0063] ,

[0064] The GNSS timing module is used to realize time synchronization between stations, and the synchronization error is less than 25n s , the sampling rate f s ≥10MS / s.

[0065] The TDOA lightning positioning needs a master station and three sub-stations to establish an equation set, and at least one redundant sub-station is needed to remove the ambiguous solution, so the number of stations N should be ≥5. The TDOA positioning of the application adopts the CHAN algorithm, which is a non-iterative algorithm for solving positioning equations and an approximate realization method of maximum likelihood estimation in a line-of-sight environment, and has the characteristics of small calculation amount and accurate result. However, when the signal propagation path is non-line-of-sight, the performance of the CHAN algorithm deteriorates rapidly. Therefore, the setting of the station should avoid the shielding of high-rise buildings and mountains.

[0066] The station uses an FPGA lightning signal acquisition system, and the sampling frequency f sShould be ≥ 10 MS / s, meet the Nyquist sampling requirement of the high-frequency component (main frequency ≤ 5 MHz) of the lightning pulse, the sampling point number M=10000, cover the time window T w =1 ms. The original signal baseline correction eliminates the DC offset. The high-speed A / D conversion module of the acquisition system converts the analog signal output by the lightning detection sensor into a digital signal. The GNSS time unit outputs a high-precision clock frequency as the sampling reference clock of the high-speed A / D conversion module, ensuring the accuracy of the signal conversion time, and the GNSS time unit uses a BDS / GPS receiver. The FPGA logic gate circuit realizes analog-to-digital conversion control, lightning detection data acquisition, timestamp acquisition, denoising processing, and peak feature extraction.

[0067] The purpose of lightning signal waveform data denoising processing is to improve the quality and usability of the signal so as to more accurately perform subsequent analysis and processing. Lightning signals are often disturbed by various noises during acquisition and transmission, which may come from natural environment, equipment itself or other external factors. Denoising processing can effectively reduce or eliminate these noise components and highlight the effective information of the signal. The lightning signal waveform data denoising processing of the present application adopts a Fourier transform denoising method, which specifically includes: performing fast Fourier transform (FFT) on the signal S i ( t ) to obtain a frequency domain signal S i ( f ), identifying the frequency range where the noise is located, removing high-frequency noise and low-frequency trend through a band-pass filter H BP ( f ), converting the processed frequency domain signal back to the time domain through inverse Fourier transform (IFFT), further performing baseline correction on the denoised signal to eliminate the DC offset, and obtaining the denoised lightning signal. The Fourier transform denoising formula is as follows:

[0068] ,

[0069] After removing the high-frequency noise, inverse Fourier transform (IFFT) is performed on the filtered frequency domain signal to restore the time domain signal, and moving average filtering (window length 1 ms) is adopted for baseline correction to eliminate the offset, and the denoised signal s i de ( t ) is obtained:

[0070] ,

[0071] Among them, s ide ( t ) is the i A denoised signal, IFFT is the inverse fast Fourier transform, which is used to convert the frequency domain signal back to the time domain. S i ( f ) is the i The frequency domain of a signal, H BP ( f ) is the frequency domain response of the filter.

[0072] The purpose of extracting peak features of lightning signal waveform is to identify and extract representative and key information from complex lightning signals, so as to facilitate subsequent signal analysis, processing and application. The peak feature extraction of waveform signal obtains the positive and negative peak amplitude, absolute time, and half-peak width of the waveform in different time windows and transmits it back to the computing center. The peak feature extraction method is: s i de ( t ), the peak feature extraction uses a sliding time window to detect positive and negative peaks. The amplitude threshold is set to 3 σ i ( σ i is the standard deviation), satisfying ∣ A p ∣>3 σ i The peak value is the effective peak value, and the absolute time of the positive and negative peak values ​​is recorded. t p , amplitude A p and half-peak width W p The set of positive and negative peaks in the detection waveform is:

[0073] ,

[0074] in, t p + and t p − are the times of the positive and negative peaks, respectively. A p + and A p − The amplitudes of the positive and negative peaks respectively; record their amplitudes A p , absolute time t p and half-peak widthW p When the amplitude of the positive peak is greater than 3 A p + , it is determined as a valid positive peak, and when the amplitude of the negative peak is greater than 3 σ i in absolute value, it is determined as a valid negative peak. A p − σ i σ i is the standard deviation of the signal, used to measure the degree of fluctuation of the signal. 3 σi is used as the threshold because in a normal distribution, about 99.7% of the data will fall within ±3 σ of the mean, so the peak with an amplitude greater than 3 σi is likely to be an outlier or noise. For each peak point determined to be valid t p , the half-peak width W p needs to be calculated. The specific steps are as follows: determine the peak center, find the center time point of the peak t p , i.e. the time position of the peak occurrence. Half amplitude, calculate half of the peak amplitude, i.e. A p / 2, where A p is the amplitude of the peak. Search time points, starting from t p , search left and right respectively, find the time points A p and t left when the signal amplitude first drops to t right / 2. The half-peak width W p is the difference between the two time points, and the formula for calculating the half-peak width W p is:

[0075] ,

[0076] where t left and t right are the time points at half the peak amplitude. A sliding window dynamic threshold algorithm is used to extract only the peaks that satisfy A p ​​The effective peak value parameter is 3 sigma. The single peak data is greatly compressed, and the real-time transmission demand of the 5G network is met.

[0077] The above steps are completed by an FPGA logic gate circuit embedded operating system at the station. After the lightning signal collected by the station is preprocessed, the speed of 80MHZ 16-bit data output from the FPGA is 160MB / s. The data transmission speed is limited by the network speed, and the lightning signal collection speed is much greater than the data transmission speed. Therefore, using DDR as a cache can effectively solve this problem and make up for the speed difference between lightning signal collection and data transmission. The data is temporarily stored in the DDR data storage unit, and the ARM processing unit embedded operating system extracts the peak features from the DDR data storage unit buffer after receiving the notification of the collected data, t p , A p , W p ), and transmits it to the computing center through the data transmission unit. The principle diagram of the positioning method of the application is shown in Figure 2 .

[0078] The computing center receives lightning data information, reconstructs the waveform according to the peak feature information, and reconstructs the peak of the lightning signal. The peak feature of the signal is recovered through the processing and analysis of the lightning signal, so as to more accurately identify and analyze the signal. The waveform reconstruction is based on the extracted peak feature point ( t p , A p , W p ), and the rising edge of the lightning pulse is approximately a quadratic function. The application uses a rectangular function to reconstruct, which can more strictly constrain the steep front of the lightning pulse and avoid the smoothing effect of the parabolic model on the rising edge. The waveform reconstruction is based on the extracted peak feature point ( t p , A p , W p ), and a single peak is fitted by a rectangular function, and the data outside the peak is zero. The rectangular function is:

[0079] ,

[0080] The rectangular function in the time domain is a square pulse with constant amplitude. The horizontal axis (time axis) defines the time window of the pulse, and the center point is t p (positive peak) or t p −(negative peak). The vertical axis (amplitude axis) defines the amplitude intensity of the pulse Ap + (positive peak) or A p - (negative peak). The pulse width is controlled by W p + + (positive peak) or W p − - (negative peak). The symmetric extension of the control window is controlled by t = t p At t = 0, the function reaches its maximum value A p and is zero at all other times.

[0081] The reconstructed waveform is generated by superimposing all positive and negative peaks, which are superimposed independently to avoid confusion of polarity:

[0082] ,

[0083] where: s i re ( t ) is the i th reconstructed signal, ∑ p∈Peaki+ and ∑ p ∈Peaki− are the summations over the positive and negative peaks, respectively, A p + and A p − are the amplitudes of the positive and negative peaks, respectively, t p + and t p − are the time positions of the positive and negative peaks, respectively, W p + and W p − are the widths of the positive and negative peaks, respectively. When processing the signal, the signal is set to zero in the region where ∣ t-t p ∣> W p The purpose is to reduce the sidelobe interference in the frequency domain, which is caused by the abrupt change of the time-domain signal and appears as other frequency components in addition to the main lobe in the frequency domain. By setting the signal to zero in the region where ∣ t-t p ∣> W pThe region of the signal is set to zero, which can effectively limit the time range of the signal and reduce the generation of sidelobes. Through the reconstruction of the rectangular function, the ability to preserve the key transient characteristics of lightning signals is enhanced under the premise of sacrificing the frequency domain smoothness, which is especially suitable for lightning location. The correlation coefficient of the reconstructed signal and the original denoised signal is calculated ρ ≥0.8, otherwise re-extract the peak value.

[0084] The reconstructed lightning waveform data of every two stations are calculated for cross-correlation, and the maximum correlation time delay is extracted. The cross-correlation uses linear cross-correlation with zero padding. For the reconstructed waveform i of station j and station S i re ( t ) and S j re ( t ), the waveforms need to be aligned before cross-correlation, and the s i re ( t ) and S j re ( t ) are zero-padded to length 2M, and the cross-correlation function is calculated as follows:

[0085] ,

[0086] R ij ( τ ) is the cross-correlation function of the lightning waveform of the i th station and the j th station, which describes the similarity of the two waveforms at different time delays τ . s i re ( t ) and s j re ( t ): the reconstructed signals of the i th station and the j th station, respectively, in discrete form are:

[0087] , (zero-padded when out of range);

[0088] The delay time that maximizes the function value is found in the cross-correlation function R ij ( τ ), and the maximum correlation time delay is extractedτ ij :

[0089] ,

[0090] τ ij Indicates the i Stations and j The maximum correlation time delay of lightning waveforms between the measuring stations is , extract the maximum correlation coefficient and the corresponding time delay .

[0091] According to the signal time delay information, the weighted nonlinear least squares method is used to solve the TDOA equations and output the three-dimensional lightning location results. The TDOA equations of the location algorithm use the following formula:

[0092] ,

[0093] in, ( x , y , z ) is the lightning position, ( x i , y i , z i ) and( x j , y j , z j ) are measuring stations i and measuring stations j location, c is the speed of light, ε ij is the observation residual, i , j =1,2,..., N and i ≠ j , N ≥5;

[0094] The TDOA algorithm is based on the initial estimate of least squares and calculates the initial value ( x 0 ,y 0 ,z 0); construct the objective function and solve it iteratively:

[0095] ,

[0096] in, , weight ;

[0097] For z ≥0 constraint is handled by projection method, when iteration step ||Δx||<1m or function change rate x ||f(x)||<0.001 z or iteration number ≥100, the optimization algorithm is terminated. After the optimization algorithm converges, the three-dimensional position of the signal source is obtained.

[0098] The reliability of the positioning result is evaluated by residual analysis and geometric dilution of precision. Residual is the difference between the observed value and the model predicted value. In TDOA positioning, the residual represents the difference between the measured time difference and the calculated time difference according to the solved position. If the root mean square (RMS) of the time delay residual of multiple station pairs is greater than 0.1 μs , after removing the station pairs with correlation less than 0.8, the calculation is restarted. τ ij me is the measured value, τ ij ca is the calculated value. Geometric dilution of precision (GDOP) is an index to evaluate the influence of station geometry layout on positioning accuracy. The smaller the GDOP value, the more ideal the geometry layout of the receiving point, and the higher the positioning accuracy. Calculate the geometric dilution of precision (GDOP) to evaluate the error:

[0099] ,

[0100] where: J T is the transpose of the observation matrix J , J T J is a m × m matrix, called the normal equation matrix, J T J ) −1 is the inverse matrix of the normal equation matrix, and trace( J T J ) −1 is the trace of the inverse matrix, which is the sum of the diagonal elements of the inverse matrix. When GDOP>5, the residual screening mechanism is automatically triggered, including:

[0101] a) residual analysis, calculate the time delay residual of each station pair , if greater than 0.2 μ s, it is marked as an outlier;

[0102] b) Geometric Dilution of Precision (GDOP) error evaluation, when GDOP>3, automatically remove the largest 20% of the residual station to recalculate;

[0103] c) Result output, generate a positioning result containing three-dimensional coordinates, elevation, confidence (95% elliptical area) and signal-to-noise ratio based weight.

[0104] Figure 3 For the comparison chart of the original waveform and the reconstructed waveform, the upper graph is the waveform of the station i , and the lower graph is the waveform of the station j , Figure 3 is a superimposed graph of the original waveform and the reconstructed waveform of 1ms, the original waveform is curved and fluctuates, and the reconstructed waveform is a rectangle with straight up and straight down. Figure 4 is a zoomed-in view of the local part of 0~0.1ms. Figure 3 From Figure 3 , Figure 4 , it can be seen that the data amount of the original waveform is large, and only the peak value feature is extracted through the reconstruction of the rectangular function, the retention capability of the key transient feature of the lightning signal is enhanced on the premise of sacrificing the frequency domain smoothness, and the data amount of the back transmission is greatly reduced. Figure 5 is a comparison chart of the lightning signal waveform cross-correlation results of the station i and the station j , the upper graph is the original waveform cross-correlation result, and the lower graph is the reconstructed waveform cross-correlation result. Figure 5 From Figure 5 , it can be seen that the original waveform directly reflects the instantaneous change of the lightning current, contains high-frequency details and steep rising edges, and through the reconstructed waveform cross-correlation, the instantaneous matching degree of the lightning signal can be accurately captured by comparing the phase similarity of the original waveform. The application is used for lightning positioning, and the original waveform cross-correlation is not needed, only the similarity comparison of the two.

[0105] The specification and drawings of the present application are only specific to one embodiment, and are not limiting, and those skilled in the art can make many forms under the inspiration of the present application without departing from the scope of the present application and the scope protected by the claims, and all are within the scope of the present application.

Claims

1. A real-time lightning location method based on FPGA signal acquisition and preprocessing, characterized in that: The following steps are involved: S1, N Each measuring station obtains lightning signal waveform data; S2, performing denoising processing on the lightning signal waveform data; S3. Extract the peak features of the waveform signal to obtain the positive and negative peak amplitudes, absolute time, and half-peak width of the waveform in different time windows, and transmit the information back to the computing center; S4, the computing center reconstructs the waveform based on the peak characteristic information; S5. Calculate the cross-correlation of the reconstructed lightning waveform data between every two measuring stations and extract the maximum correlation time delay; S6. Solve the TDOA equations using a weighted nonlinear least squares method based on the signal time delay information and output a three-dimensional lightning location result. S7. Evaluate the reliability of positioning results through residual analysis and geometric precision dilution; In steps S1, S2, and S3, the collected lightning data is pre-processed using FPGA logic gate circuits at the measuring station, peak information is extracted, and the data is transmitted back to the computing center in real time via the 5G network. In steps S4, S5, S6, and S7, waveform reconstruction, cross-correlation calculation, lightning location, and result evaluation are performed in the computing center.

2. The real-time lightning location method based on FPGA signal acquisition and preprocessing according to claim 1 is characterized in that: The specific method of S1 obtaining multiple lightning signal waveform data is as follows: i The lightning signal waveform collected by each measuring station is s i ( t ),in t For time, i = 1, 2, …, N , N is the number of stations, and the sampling time interval of each station is Δ t , the number of sampling points is M , baseline correction eliminates DC offset to a baseline of 0; the signal can be expressed in discrete form: ， GNSS timing modules are used between each measuring station to achieve time synchronization, with a synchronization error of less than 25n s , sampling rate f s ≥10MS / s.

3. The real-time lightning location method based on FPGA signal acquisition and preprocessing according to claim 1 is characterized in that: The S2 denoising process adopts the Fourier transform denoising method, which specifically includes: S i ( t ) Perform fast Fourier transform (FFT) to obtain the frequency domain signal S i ( f ), identify the frequency range of the noise, and pass it through a bandpass filter H BP ( f ) remove high-frequency noise and low-frequency trends, convert the processed frequency domain signal back to the time domain through inverse Fourier transform (IFFT), further perform baseline correction on the denoised signal, eliminate DC offset, and obtain the denoised lightning signal. The Fourier transform denoising formula is as follows: ， After removing the high-frequency noise, the filtered frequency domain signal is subjected to inverse Fourier transform (IFFT) to restore the time domain signal and obtain the denoised signal. s i de ( t ): ， in, s i de ( t ) is the i A denoised signal, IFFT is the inverse fast Fourier transform, which is used to convert the frequency domain signal back to the time domain. S i ( f ) is the i The frequency domain of a signal, H BP ( f ) is the frequency domain response of the filter.

4. The real-time lightning location method based on FPGA signal acquisition and preprocessing according to claim 1 is characterized in that: The S3 peak feature extraction method is as follows: s i de ( t ), the peak feature extraction uses a sliding time window to detect positive and negative peaks, and the amplitude threshold is set to 3 σ i ,in σ i is the standard deviation, satisfying ∣ A p ∣>3 σ i The peak value is the effective peak value, and the absolute time of the positive and negative peak values ​​is recorded. t p , amplitude A p and half-peak width W p ; The set of positive and negative peaks in the detection waveform is: ， in, t p + and t p − are the times of the positive and negative peaks, respectively. A p + and A p − are the amplitudes of the positive and negative peaks, respectively; Record its amplitude A p , absolute time t p and half-peak width W p ; Half-peak width W p The calculation formula is: ， in, t left and t right are the time points at half the peak amplitude.

5. The real-time lightning location method based on FPGA signal acquisition and preprocessing according to claim 1 is characterized in that: The S4 waveform reconstruction is based on the extracted peak feature points ( t p , A p , W p ), a single peak is fitted with a rectangle, and the data outside the peak are zero. The rectangular function is: ， Add all positive and negative peaks to generate the reconstructed waveform: ， in: s i re ( t ) is the i reconstructed signal, ∑ p∈Peaki+ and ∑ p ∈Peaki− are the sum of positive and negative peaks, respectively. A p + and A p − are the amplitudes of the positive and negative peaks, respectively, t p + and t p − are the time positions of the positive and negative peaks, respectively. W p + and W p − are the widths of the positive and negative peaks, respectively.

6. The real-time lightning location method based on FPGA signal acquisition and preprocessing according to claim 1 is characterized in that: The S5 cross-correlation adopts linear cross-correlation with zero filling. i and measuring stations j The reconstructed waveform S i re ( t )and S j re ( t ), before cross-correlation, the waveform baseline needs to be aligned and the cross-correlation function calculated: ， R ij ( τ ) is the i Stations and j The cross-correlation function of the lightning waveforms at the observation station describes the two waveforms at different time delays. τ the degree of similarity under s i re ( t ) and s j re ( t ): respectively i Stations and j The signal reconstructed by each station is in discrete form: , where zero is added when out of range. In the cross-correlation function R ij ( τ ) to find the delay time that maximizes the function value and extract the maximum relevant time delay τ ij : ， τ ij Indicates the i Stations and j The maximum correlation time delay of the lightning waveform between the measuring stations is obtained, and the maximum correlation coefficient and the corresponding time delay are extracted.

7. The real-time lightning location method based on FPGA signal acquisition and preprocessing according to claim 1, characterized in that: The S6 positioning algorithm TDOA equation group adopts the following formula: ， in, ( x , y , z ) is the lightning position, ( x i , y i , z i )and( x j , y j , z j ) are measuring stations i and measuring stations j location, c is the speed of light, ε ij is the observation residual, i , j =1,2,..., N and i ≠ j , N ≥5; The TDOA algorithm is based on the initial estimate of least squares and calculates the initial value ( x 0 , y 0 , z 0); construct the objective function and solve it iteratively: ， in, , weight ; right z The ≥0 constraint is processed using the projection method and terminated when the iteration step ||Δx|| is less than 1m or the function change rate or the number of iterations is ≥100.

8. The real-time lightning location method based on FPGA signal acquisition and preprocessing according to claim 1, characterized in that: The S7 reliability evaluation includes: if the time delay residuals of multiple station pairs The root mean square (RMS) is greater than 0.1 μs , then remove the station pairs with amplitude correlation lower than 0.8 and recalculate. τ ij me is the measured value, τ ij ca is the calculated value; Calculate the geometric dilution of precision (GDOP) for error assessment: ， in: J T is the observation matrix J The transpose of J T J is a m × m The matrix of is called the normal equation matrix, ( J T J ) −1 is the inverse matrix of the normal equation matrix, trace(( J T J ) −1 ) is the trace of the inverse matrix, that is, the sum of the diagonal elements of the inverse matrix; When GDOP>5, the residual screening mechanism is automatically triggered, including: a) Residual analysis: calculate the time delay residual of each station pair. If it is greater than 0.2 μ s is marked as an outlier; b) Geometric Dilution of Precision (GDOP) error assessment: when GDOP>3, the 20% of station pairs with the largest residual errors are automatically eliminated and recalculated; c) Result output: Generates positioning results including three-dimensional coordinates, elevation, 95% confidence ellipse area and signal-to-noise ratio weight.

Images