Non-intrusive high-frequency magnetic field sensing data acquisition method based on high-bandwidth acquisition range

Abstract

According to the non-intrusive high-frequency magnetic field sensing data acquisition method based on the high-bandwidth acquisition range, the integrity of signals in the transmission process is ensured through the broadband matching network optimization technology, reflection and attenuation are reduced through impedance matching and loss compensation, the dynamic gain control technology is used for coping with the characteristic that the amplitude change of broadband signals is large, and the high-frequency magnetic field sensing data is acquired. Real-time gain adjustment is realized by using a series variable gain amplifier and an intelligent control algorithm, analog signals are acquired by a digital signal processing enhancement technology and then are converted into digital signals through a high-speed ADC, and the digital signals are processed in real time by using an FPGA; the self-adaptive filtering algorithm further improves the signal quality, filter parameters can be dynamically adjusted to adapt to noise characteristics of different frequency bands, the multi-sensor data fusion technology integrates data of a plurality of sensors, and high-precision non-intrusive measurement can be achieved in a wider frequency range.

Description

Technical Field

[0001] The invention relates to the field of acquisition of wide-spectrum transient signals in power systems, and in particular to a non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range. Background Art

[0002] Non-invasive high-frequency magnetic field sensors work based on Faraday's law of electromagnetic induction. When a high-frequency current flows through a conductor, a changing magnetic field is generated around it. The sensor senses these changing magnetic fields, generates an induced electromotive force, and thus detects high-frequency signals. The design of the sensor needs to take into account the high-frequency response characteristics to ensure that it can accurately sense changes in the high-frequency magnetic field. Non-invasive means that the sensor does not need to directly contact the conductor being measured, and only measures high-frequency signals by sensing changes in the magnetic field. This non-invasive design avoids interference with the system being measured. The existing technology specifically includes Rogowski coil technology, magnetoelectric sensor technology, and fiber optic current sensor technology.

[0003] Rogowski coil technology has been widely used in the field of non-invasive high-frequency magnetic field sensing. Its working principle is based on Faraday's law of electromagnetic induction, and the current is indirectly measured by measuring the change in magnetic flux passing through the coil. The advantages of Rogowski coils are good linearity and wide frequency band, which can theoretically cover the range from a few Hz to a few MHz. At the same time, it has the characteristics of non-invasive installation and does not require disconnecting the circuit under test. However, Rogowski coil technology still faces some significant challenges and limitations in practical applications. First, the sensitivity of Rogowski coils is relatively low, especially when measuring small currents, the signal quality is often not ideal, which limits its application in weak signal detection. Secondly, Rogowski coils are easily interfered by external magnetic fields, and the measurement accuracy may be significantly affected in complex electromagnetic environments. In addition, the temperature drift problem of Rogowski coils is more prominent, and frequent calibration is required to ensure the accuracy of long-term measurements. In high-frequency applications, the parasitic capacitance of the Rogowski coil itself affects its high-frequency characteristics, resulting in limited bandwidth. Finally, it is difficult for Rogowski coils to take into account both wide bandwidth and high sensitivity at the same time. When improving one indicator, the other indicator is often sacrificed, which limits it in some extremely demanding application scenarios.

[0004] Magnetoelectric sensor technology is an emerging non-invasive high-frequency magnetic field sensing method. Its working principle is based on the coupling of magnetostrictive effect and piezoelectric effect of magnetoelectric composite materials. When an external magnetic field acts on the magnetoelectric composite material, the magnetic phase will produce strain, which is transmitted to the piezoelectric phase through interface coupling, and finally generates an electrical signal in the piezoelectric phase, thereby realizing the direct conversion of magnetic field to electrical signal. In theory, magnetoelectric sensors can achieve high sensitivity and wide-band measurement, with the characteristics of passive operation and low power consumption, and show great potential in the fields of power system monitoring and non-destructive testing. However, as a new technology still in the research stage, magnetoelectric sensors face many challenges. First, the preparation and device packaging processes of magnetoelectric composite materials are complex, making it difficult to achieve large-scale industrialization, resulting in high costs. Secondly, the linearity and temperature stability of magnetoelectric sensors still need to be further improved, especially maintaining stability over a wide temperature range is a huge challenge. In addition, in a strong magnetic field environment, magnetoelectric sensors may have saturation problems, which limits their use in some high-current applications. Due to the immaturity of the technology, there are few actual application cases, lack of long-term reliability data and standardized product specifications, which to a certain extent hinders their widespread application in the industry. Finally, although the frequency band of magnetoelectric sensors can be very wide in theory, in practical applications, especially at the high-frequency end, they still face problems such as signal attenuation and phase distortion, which require further technological breakthroughs.

[0005] Fiber optic current sensor technology is a non-invasive high-frequency magnetic field sensing method based on the Faraday effect. Its working principle is to use the influence of magnetic field on polarized light to indirectly measure current. When polarized light propagates in a magnetic field, its polarization plane rotates. The rotation angle is proportional to the magnetic field strength. By measuring this rotation angle, the current can be inferred. The application of fiber optic current sensors in high-voltage and ultra-high-voltage systems is gradually increasing. Its advantages are excellent insulation performance and anti-electromagnetic interference ability. In theory, it can achieve wide-band measurement from DC to MHz. However, fiber optic current sensors still have some significant defects and limitations. First, the cost of fiber optic current sensors is relatively high, especially in medium and low voltage systems. Its economy is difficult to compare with traditional current transformers, which limits its promotion in some cost-sensitive application scenarios. Secondly, fiber optic current sensors are very sensitive to temperature changes and require complex temperature compensation mechanisms to ensure measurement accuracy, which increases the complexity and cost of the system. In addition, fiber optic current sensors are also sensitive to vibration and have high requirements for installation environment and use conditions, which may become a problem in applications in some harsh environments. The signal processing of fiber optic current sensors is also relatively complex, requiring specialized photoelectric conversion and signal conditioning circuits, which not only increases the complexity of the system, but may also introduce additional error sources. Finally, although fiber optic current sensors have a certain application history, their long-term stability and reliability still need to be further verified, especially in harsh environments. There is still some uncertainty in their long-term performance.

[0006] In summary, current high-frequency magnetic field sensors are usually limited by narrow bandwidths and are difficult to meet the acquisition requirements of wide-spectrum transient signals in power systems, especially when capturing high-frequency traveling wave signals. Although non-invasive installation is easy to implement, it often leads to signal strength attenuation and reduced signal-to-noise ratio, affecting the acquisition accuracy and reliability of high-frequency magnetic field signals. There are a large number of electromagnetic interference sources in power systems. How to effectively suppress various types of interference while ensuring broadband acquisition is a key challenge to improving signal acquisition quality. Summary of the invention

[0007] In view of this, the present invention provides a non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range, which can achieve high-precision non-invasive measurement in a wider frequency range.

[0008] A non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range comprises the following steps:

[0009] S1. Capture the magnetic field signal around the conductor under test by using a high-frequency magnetic field sensor array;

[0010] S2. Adjust the broadband matching network to ensure the best impedance matching in the whole frequency band;

[0011] S3. Achieve wide dynamic range gain adjustment from -18dB to +51dB through a three-stage VGA structure;

[0012] S4. Use ADC and implement interleaved sampling technology to perform high-resolution signal sampling;

[0013] S5. Preprocess the sampled signal using FPGA;

[0014] S6. Perform FFT spectrum analysis on the preprocessed sampled signal and convert the collected time domain signal into frequency domain;

[0015] S7. Comprehensively use LMS, FLMS and RLS algorithms to perform adaptive signal filtering and signal enhancement on the signal after frequency domain conversion;

[0016] S8. Use weighted average and DS evidence theory to fuse the sampling data of multiple sensors;

[0017] S9. Reconstruct the fused sampled data into the original signal based on the least squares method.

[0018] Further, step S1 includes:

[0019] S11. Arrange the high-frequency magnetic field sensor array reasonably according to the geometric characteristics of the conductor under test and the expected magnetic field distribution, taking into account the sensor spacing, angle and height factors during arrangement to ensure the best spatial sampling coverage;

[0020] S12. Before actual measurement, each sensor is accurately calibrated, including sensitivity calibration, frequency response calibration and phase calibration. A standard magnetic field source and precision measuring equipment are used to establish the calibration curve and calibration factor of each sensor to ensure the accuracy and consistency of the measurement results.

[0021] S13. Before starting the formal measurement, measure and record the environmental noise baseline, and record the environmental factors that may affect the measurement results, including background electromagnetic noise, temperature changes, and vibration;

[0022] S14. Set up a synchronization trigger mechanism, use a time synchronization device to ensure that all sensors start data collection at the same time, the time synchronization device includes a GPS clock or an atomic clock;

[0023] S15. performing an initial gain setting for each sensor channel based on the estimated signal strength, taking into account the expected current strength of the conductor being measured and the distance from the sensor to the conductor to ensure that the signal falls within the optimal dynamic range of the sensor;

[0024] S16. Start the data acquisition system to capture the magnetic field signal around the conductor under test in real time. During the process of capturing the magnetic field signal, continuously monitor the data quality, including signal amplitude and signal-to-noise ratio parameters, to ensure that the collected data is valid and reliable.

[0025] Further, step S2 includes:

[0026] S21. Perform a frequency sweep analysis on the entire measurement system using a network analyzer, wherein the measurement system includes a sensor, a transmission line, and a front-end amplifier;

[0027] S22. According to the network analysis results, an improved broadband matching network is selected, the core of which is to add a parallel resonant branch on the basis of the traditional π-type network;

[0028] S23. Use a vector network analyzer to test the matching network to verify its matching performance within the actual frequency band. The test indicators include standing wave ratio, return loss, and insertion loss.

[0029] S24. Fine-tune the matching network according to the test results;

[0030] S25. Considering the temperature changes in the actual working environment, design and implement a temperature compensation mechanism, use a combination of components with complementary temperature coefficients or add a temperature control circuit to ensure that the matching network maintains stable performance at different temperatures.

[0031] Further, step S3 includes:

[0032] S31. Design a three-stage cascaded variable gain amplifier (VGA) structure to achieve an overall dynamic range of -18dB to +51dB while ensuring linearity and noise performance;

[0033] S32. Deploy an intelligent gain control algorithm that can automatically adjust the gain of each VGA level according to the strength of the input signal;

[0034] S33. Optimize the noise performance of the VGA for different gain settings to ensure a good signal-to-noise ratio over the full dynamic range.

[0035] Further, step S4 includes:

[0036] S41. Select a suitable high-speed ADC based on the bandwidth requirements, resolution requirements, and dynamic range of the system;

[0037] S42. Design a multi-channel interleaved sampling architecture to increase the overall sampling rate. Each ADC samples with a phase-staggered clock, thereby achieving a higher equivalent sampling rate.

[0038] S43. Deploy a precise clock distribution network to provide each ADC with a sampling clock with a precise phase to ensure that the sampling time between channels is precisely staggered;

[0039] S44. Transmit the digital data output by the ADC to the subsequent FPGA processing unit.

[0040] Further, step S5 includes:

[0041] S51.FPGA receives digital sampled data from ADC through high-speed interface. To ensure the continuity and integrity of data, FPGA uses its internal BlockRAM to build a multi-level FIFO buffer to effectively prevent data loss and overflow.

[0042] S52. The received raw data is format converted to meet the needs of subsequent processing. The format conversion includes bit width adjustment and fixed-point to floating-point conversion operations to ensure that the data format is compatible with the processing algorithm;

[0043] S53. Use digital filters to remove high frequency noise and unwanted frequency bands;

[0044] S54. Conditioning the filtered signal, including DC offset removal and amplitude standardization;

[0045] S55. Divide the continuous data stream into data segments of appropriate length to facilitate subsequent FFT analysis and other algorithm processing;

[0046] S56. Apply an appropriate window function to each data segment to reduce spectrum leakage effect and improve the accuracy of FFT analysis. The window function includes a Hanning window and a Hamming window.

[0047] Further, step S6 includes:

[0048] S61. According to the characteristics of the preprocessed data and the analysis requirements, configure the FFT parameters, including the number of FFT points and the sampling rate;

[0049] S62. Load the pre-processed data segment from the BlockRAM of the FPGA to the FFT processing unit;

[0050] S63. Implementing radix 4 FFT algorithm in FPGA;

[0051] S64. Perform amplitude calculation on the FFT result, and convert the complex FFT output into a real amplitude spectrum;

[0052] S66. Calculate the phase information of the FFT result;

[0053] S67. Averaging the FFT results of multiple consecutive data segments to reduce the influence of random noise and improve the reliability of spectrum estimation, the averaging method is linear averaging or exponential averaging;

[0054] S68. When the frequency resolution needs to be improved, the FFT result is interpolated, and the interpolation is implemented by zero padding or a more complex interpolation algorithm to obtain a finer spectrum representation;

[0055] S69. Store the calculated spectrum data in the BlockRAM of the FPGA or the external DDR3SDRAM, and transmit it to a subsequent processing unit or a host computer.

[0056] Further, step S7 includes:

[0057] S71. According to the signal characteristics and processing requirements, select an appropriate adaptive algorithm, the adaptive algorithm includes LMS, FLMS or RLS, initialize the algorithm parameters, including filter order, step size or forgetting factor;

[0058] S72. Generate or select a suitable reference signal for the adaptive filtering algorithm, the reference signal is based on prior knowledge, a signal model or extracted from the original signal;

[0059] S73. Comprehensively utilize the advantages of LMS, FLMS and RLS algorithms to realize the fusion operation of LMS algorithm, FLMS algorithm or RLS algorithm, use RLS to converge quickly in the initial stage, and then switch to LMS maintenance with less calculation amount;

[0060] S74. Monitor the improvement indicators of the adaptive algorithm in real time, including mean square error (MSE) and signal-to-noise ratio, and dynamically adjust algorithm parameters or switch to different algorithms based on these improvement indicators.

[0061] Further, step S8 includes:

[0062] S81. First, the data from different sensors are synchronized in time, and interpolation, resampling or timestamp alignment technology is used to ensure that the data from different sensors correspond accurately in time;

[0063] S82. Preprocessing the synchronized data, including outlier detection and removal;

[0064] S83. Evaluate the performance indicators of each sensor, including accuracy, reliability, and response time;

[0065] S84. Implement a weighted average algorithm to assign weights according to the performance indicators of each sensor and calculate the fusion results. The weighting scheme is static or dynamically adjusted according to real-time performance;

[0066] S85. Establish a fusion framework based on DS evidence theory, define the recognition framework, basic probability distribution function BPA and combination rules;

[0067] S86. Assigning a basic probability to the measurement result of each sensor based on the sensor data and the performance indicator;

[0068] S87. Apply DS combination rules to synthesize evidence from multiple sensors;

[0069] S88. Based on the synthesized evidence, use the maximum confidence rule or other decision criteria to obtain the final fusion result, which is a judgment of a certain state or the best estimate of the measurement value.

[0070] Further, step S9 includes:

[0071] S91. According to the signal characteristics and application requirements, select polynomial fitting or Fourier series expansion as the signal reconstruction model;

[0072] S92. Initializing parameters for the selected reconstruction model, for polynomial fitting, determining the order of the polynomial; for Fourier series, determining the number of harmonics;

[0073] S93. Formulate the signal reconstruction problem as a least squares problem, including constructing a measurement matrix and a target vector, wherein the measurement matrix is ​​composed of basis functions of the selected model, and the target vector is the fused sampled data;

[0074] S94. Solve least squares problems using QR decomposition or Cholesky decomposition methods;

[0075] S95. Signal interpolation and smoothing: Interpolate the reconstructed discrete points to generate a continuous signal curve.

[0076] The present invention has the following beneficial effects:

[0077] 1. The broadband matching network optimization technology effectively reduces the reflection and attenuation of the signal during transmission through impedance matching and loss compensation mechanisms, ensuring the integrity of the signal. In particular, it improves the accuracy and reliability of signal acquisition by targeting the signal strength attenuation problem caused by non-invasive installation.

[0078] 2. Dynamic gain control technology uses a series variable gain amplifier and intelligent control algorithm to achieve a wide dynamic range gain adjustment from -18dB to +51dB. It can adapt to large changes in broadband signal amplitude in real time, ensuring that signals with appropriate gain can be obtained throughout the entire frequency band, effectively solving the limitations of traditional fixed gain systems when processing broadband signals.

[0079] 3. In the digital domain, the application of high-speed ADC and FPGA realizes high-resolution, high-speed signal sampling and real-time processing. The use of interleaved sampling technology further improves the effective sampling rate, while the parallel processing capability of FPGA ensures the real-time execution of complex algorithms such as data preprocessing and FFT spectrum analysis. These digital signal processing technologies provide a high-quality digital signal foundation for subsequent signal analysis and feature extraction.

[0080] 4. The application of adaptive filtering algorithms, especially the comprehensive application of LMS, FLMS and RLS algorithms, can dynamically adjust the filter parameters according to the signal characteristics to achieve the best noise suppression and signal extraction. It directly targets various types of electromagnetic interference in the power system and improves the system's anti-interference ability.

[0081] 5. Multi-sensor data fusion technology integrates data from multiple sensors through high-precision time synchronization and advanced data fusion algorithms, further improving the accuracy and reliability of measurement. It not only improves the accuracy of single-point measurement, but also enhances the system's resistance to local interference. Finally, advanced signal processing technologies such as signal reconstruction based on least squares method and wavelet transform further enhance the system's analysis capabilities.

[0082] From the front-end sensor design to the back-end signal processing and analysis, each link of this invention specifically solves the problems faced by the existing technology. The multi-stage core design and broadband matching network solve the basic problem of signal acquisition. The dynamic gain control and high-speed digital processing ensure the effective capture and processing of the signal within the entire bandwidth, while the adaptive filtering and data fusion further enhance the system's anti-interference ability and measurement accuracy. Through this comprehensive technical integration, the present invention successfully realizes broadband, high-precision, and highly anti-interference non-invasive high-frequency magnetic field signal acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 It is a flow chart of a non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range of the present invention;

[0084] Figure 2 It is a schematic diagram of the sensitivity change of the sensor at different frequencies;

[0085] Figure 3 is a schematic diagram of the peak detection circuit response;

[0086] Figure 4 It is a schematic diagram of the FFT performance simulation test results. DETAILED DESCRIPTION

[0087] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0088] See also Figure 1 The embodiment of the present invention provides a non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range, comprising the following steps:

[0089] S1. Using a high-frequency magnetic field sensor array to capture the magnetic field signal around the conductor under test, step S1 specifically includes:

[0090] S11. Arrange the high-frequency magnetic field sensor array reasonably according to the geometric characteristics of the conductor being measured and the expected magnetic field distribution. Consider factors such as sensor spacing, angle, and height during arrangement to ensure optimal spatial sampling coverage.

[0091] S12. Before actual measurement, each sensor is accurately calibrated, including sensitivity calibration, frequency response calibration, and phase calibration. Use a standard magnetic field source and precision measuring equipment to establish a calibration curve and calibration factor for each sensor to ensure the accuracy and consistency of the measurement results.

[0092] S13. Before starting the formal measurement, measure and record the environmental noise baseline. This step will help with subsequent signal processing and noise elimination. Record environmental factors that may affect the measurement results, including background electromagnetic noise, temperature changes, vibrations, etc.

[0093] S14. To ensure the synchronization of data collection time of multiple sensors, set up an accurate synchronization trigger mechanism. You can use GPS clock, atomic clock or other high-precision time synchronization equipment to ensure that all sensors start data collection at the same time.

[0094] S15. Perform initial gain setting for each sensor channel based on the estimated signal strength. This step needs to take into account factors such as the expected current strength of the conductor being measured and the distance from the sensor to the conductor to ensure that the signal falls within the optimal dynamic range of the sensor.

[0095] S16. Start the data acquisition system and start capturing the magnetic field signal around the conductor under test in real time. During this process, the data quality needs to be continuously monitored, including parameters such as signal amplitude and signal-to-noise ratio, to ensure that the collected data is valid and reliable.

[0096] The above steps complete the preliminary acquisition of high-frequency magnetic field signals. The output of this stage will serve as the input for broadband matching network optimization. High-quality raw data acquisition is critical to the performance of the entire system, so special attention should be paid to the precise placement and calibration of sensors and the control of environmental factors at this stage.

[0097] Preferably, the high-frequency magnetic field sensor can adopt a multi-level structure magnetic core design, the core of which is to utilize the advantages of different materials in different frequency bands to achieve high-performance magnetic field sensing within a wide frequency band.

[0098] The multi-level core structure is as follows:

[0099] 1) The nanocrystalline material Fe73.5Cu1Nb3Si13.5B9 used in the inner layer has extremely high initial magnetic permeability (μi>100,000) and low coercivity (Hc<1A / m), which makes it have excellent magnetic field sensing ability in the low frequency band (0.1kHz-100kHz). The grain size of nanocrystalline materials is usually in the range of 10-20nm. This extremely fine grain structure effectively inhibits the movement of magnetic domain walls, thereby reducing eddy current loss and hysteresis loss.

[0100] 2) The Mn-Zn ferrite used in the middle layer shows good performance in the medium frequency band (100kHz-1MHz), with a typical initial permeability between 2000-5000 and a saturation flux density of about 0.4-0.5 T. The high resistivity of Mn-Zn ferrite (about 1-10Ω·m) helps to further reduce eddy current losses.

[0101] 3) The Ni-Zn ferrite used in the outer layer is particularly suitable for high-frequency applications (1MHz-10MHz). Although its initial magnetic permeability is low (usually between 100-1000), it has a very high resistivity (10^3-10^6Ω·m), which allows it to maintain low losses at high frequencies. In actual design, Ni-Zn ferrite materials with an initial magnetic permeability of 500 and a Curie temperature of 130°C can be selected to ensure stability in high-frequency and high-temperature environments.

[0102] The concentric cylindrical three-layer structure of the core is designed to maximize the magnetic field induction efficiency and achieve a smooth transition of the frequency band. The selection of the inner layer diameter d needs to consider the size of the conductor to be measured and the expected measurement range. For example, for applications with a measurement range of 1A-1000A, d=30mm can be selected. The diameter ratio of the middle layer and the outer layer (1.2d and 1.5d) is the result of optimization. This ratio can ensure that each layer has sufficient cross-sectional area to carry the magnetic flux and maintain the compactness of the overall structure. The special magnetic resistance material filled between the layers can be a silicone rubber-based soft magnetic composite material, whose magnetic permeability is usually between 5-50 and the resistivity is as high as 10^9Ω·m. This material can not only effectively isolate the layers of the core and reduce the eddy current loss between the layers, but also optimize the magnetic flux path through its moderate magnetic permeability, thereby improving the overall magnetic field induction efficiency. The conical structure design at both ends of the core is based on the principle of flux focusing. By gradually reducing the cross-sectional area of ​​the core and increasing the magnetic flux density, the magnetic field induction efficiency in the high frequency band is improved. The choice of cone angle requires a balance between focusing effect and processing difficulty. Experiments show that a cone angle of 20° can achieve good results in most applications. The focusing structure can increase the upper frequency limit of the sensor from the original 5MHz to about 10MHz, significantly expanding the available frequency band.

[0103] In practical applications, the segmented winding technology can further optimize the performance of the sensor in different frequency bands.

[0104] 1) In the low frequency band (0.1kHz-100kHz), 500-1000 turns of winding can be wound on the inner nanocrystalline core, using enameled wire with a diameter of 0.1mm. This multi-turn winding can generate a higher induced voltage and improve the signal-to-noise ratio of the low-frequency signal.

[0105] 2) In the medium frequency band (100kHz-1MHz), 100-200 turns can be wound on the middle layer of Mn-Zn ferrite, using enameled wire with a diameter of 0.05mm.

[0106] 3) For high frequency band (1MHz-10MHz), only 10-20 turns are wound on the outer layer of Ni-Zn ferrite, using enameled wire with a diameter of 0.03mm. The few turns of winding can effectively reduce the distributed capacitance and improve the frequency response. All frequency bands use Litz wire structure, that is, multiple thin wires are twisted together after being connected in parallel. The design of Litz wire can significantly reduce the loss caused by skin effect and proximity effect, especially in the high frequency band. The skin depth formula is as follows:

[0107]

[0108] Among them, ρ is the conductor resistivity, f is the frequency, and μ is the conductor magnetic permeability. It can be calculated that the skin depth of the copper conductor is about 20μm at 10MHz. Therefore, choosing a thin wire with a diameter less than 40μm can effectively utilize the entire cross-sectional area of ​​the conductor.

[0109] During the test, through this multi-stage structure design and segmented winding technology, the sensor can maintain high sensitivity and linearity in the wide frequency range of 0.1kHz-10MHz. Specifically, the frequency response characteristics of the sensor can be measured using a network analyzer. The actual test results are as follows: Figure 2 As shown, from Figure 2 It can be seen that the sensor maintains a relatively flat response in the entire frequency band, and the sensitivity variation does not exceed ±10%, which fully proves the effectiveness of the multi-level structure core design.

[0110] S2. Adjust the broadband matching network to ensure the best impedance matching over the entire frequency band:

[0111] S21. Use a network analyzer to perform a frequency sweep analysis on the entire measurement system, including the sensor, transmission line, and front-end amplifier. This step is to obtain the impedance characteristics of the system over the entire frequency band.

[0112] S22. According to the network analysis results, an improved broadband matching network is selected. The core of the improved broadband matching network is to add parallel resonant branches to the traditional π-type network to form a multi-resonant structure.

[0113] The traditional π-type network consists of two parallel capacitors and a series inductor, while the improved structure adds one or more parallel LC resonant branches on this basis. The principle of this design is to use multiple resonance points to expand the bandwidth of the network. Each resonant branch presents high impedance at a specific frequency, so as not to affect the main signal path, and participates in impedance matching at other frequencies. The following parameters can be used for specific design:

[0114] The frequency band to be matched is 0.1MHz to 10MHz. The series inductance L1 of the main π-type network is selected to be 0.5μH, and the parallel capacitors C1 and C2 at both ends are 100pF each. Two groups of parallel resonant branches can be designed. The first group L2 = 1μH, C3 = 25pF, and the resonant frequency is about 1MHz; the second group L3 = 0.2μH, C4 = 12.5pF, and the resonant frequency is about 5MHz. These parameters can be preliminarily calculated by the following formula:

[0115]

[0116] Among them, f ris the resonant frequency, L is the inductance value, and C is the capacitance value. This improved π-type network structure can achieve a VSWR (voltage standing wave ratio) of less than 1.5:1 and an insertion loss of less than 0.5dB in the frequency band of 0.1MHz to 10MHz.

[0117] The multi-stage impedance transformation technology can gradually transform the complex impedance of the sensor into a 50Ω standard impedance through multiple stages. The transformation ratio of each stage is controlled between 1:2 and 1:3, because a smaller transformation ratio can ensure a wider bandwidth. In specific implementation, a cascaded L-type network or a stepped transmission line structure can be used. Usually, a three-stage impedance transformation is used, and the transformation process refers to the following formula:

[0118]

[0119] Among them, Z 0 is the characteristic impedance (usually 50Ω), ω is the angular frequency, R 1 and R 2 are the input and output impedances, respectively. In order to achieve dynamic adjustment of impedance, a controllable varactor diode is introduced. The capacitance value of the controllable varactor diode can be adjusted by changing the reverse bias voltage. Specifically, the BB833 variable capacitance diode can be selected, and its capacitance value can be adjusted between 1.8pF and 9.1pF (when the reverse bias voltage changes from 28V to 1V). By connecting the variable capacitance diode in parallel to the key nodes of the matching network, real-time adjustment of impedance can be achieved. The generation of control voltage can be achieved by digital potentiometers and operational amplifiers. Using the 10-bit digital potentiometer AD5280 with the operational amplifier OPA354, a finely adjustable voltage of 0-30V can be generated. According to the current operating frequency band, the microcontroller (such as STM32F103) can dynamically adjust the resistance value of the digital potentiometer through the SPI interface, thereby changing the capacitance value of the variable capacitance diode and realizing automatic adjustment of the matching network parameters.

[0120] S23. Use a vector network analyzer to test the matching network and verify its matching performance within the actual frequency band. Test indicators include standing wave ratio, return loss, insertion loss, etc.

[0121] S24. Fine-tune the matching network based on the test results.

[0122] S25. Consider the temperature changes in the actual working environment and design and implement a temperature compensation mechanism. Use a combination of components with complementary temperature coefficients or add a temperature control circuit to ensure that the matching network maintains stable performance at different temperatures.

[0123] The above steps complete the optimization of the broadband matching network. The optimized matching network ensures that the high-frequency magnetic field signal can be transmitted to the subsequent amplification and processing circuit with minimal loss and distortion. At the same time, the dynamic tuning mechanism at this stage provides the hardware foundation for the subsequent dynamic gain control, enabling the system to better adapt to changes in signal strength.

[0124] S3. Achieve wide dynamic range gain adjustment from -18dB to +51dB through a three-level VGA structure:

[0125] S31. Design a three-stage cascaded variable gain amplifier (VGA) structure to achieve an overall -18dB to +51dB dynamic range while ensuring linearity and noise performance.

[0126] The dynamic gain control adopts a series three-stage VGA structure. The core is to disperse the gain control into three independent amplifier stages, and the gain range of each stage is -6dB to +20dB. The advantage of this design is that it can achieve fine gain adjustment in a wide dynamic range while maintaining good linearity and noise performance. The core of each VGA is a high-linearity analog multiplier. The working principle of this multiplier is to use the transconductance characteristics of the differential pair to achieve linear modulation of the signal amplitude. In specific implementation, the amplitude range of the input signal is 1mV to 10V. The gain range of the first stage VGA is set to -6dB to +14dB, the second stage is -6dB to +17dB, and the third stage is -6dB to +20dB. This setting can cover a total gain range of -18dB to +51dB, equivalent to a dynamic range of about 70dB. The gain control of each VGA uses a 10-bit digital potentiometer, such as AD5280, which can achieve a fine adjustment step of 0.5dB. The relationship between the gain and the resistance of the digital potentiometer can be approximated by the following formula:

[0127]

[0128] Among them, R max and R minare the maximum and minimum resistance values ​​of the digital potentiometer, respectively, and D is the digital control word (0-1023). In order to achieve fast gain adjustment, a broadband peak detection circuit with a response time of less than 100ns is adopted. A dual-path parallel structure is adopted to detect positive and negative peaks respectively, which can effectively process asymmetric signal waveforms. The core of the peak detection circuit includes a high-speed comparator and a sample-and-hold circuit. The simplified principle of the positive peak detection circuit is as follows: the input signal charges the capacitor (typical value 10pF) after passing through the diode, and when the signal drops, the capacitor maintains the peak voltage. The negative peak detection circuit is similar, but uses a reverse-connected diode. The hold time of the sample-and-hold circuit can be adjusted by the RC time constant, with a typical value of 1ms, which allows the signal to change slowly while maintaining the peak value. The detected peak signal is converted into a digital signal by a high-speed ADC for FPGA processing. Figure 3 Figure 3 shows the response of the peak detection circuit to a 10MHz sinusoidal signal. It can be clearly seen that the detection circuit is able to quickly track the peak value of the signal.

[0129] S32. Deploy an intelligent gain control algorithm that can automatically adjust the gain of each VGA level according to the strength of the input signal.

[0130] In the dynamic gain control process, intelligent dynamic gain control is implemented based on FPGA. The Xilinx Artix-7 series FPGA can provide sufficient processing power and flexibility. The control algorithm adopts a prediction-correction strategy to predict the signal change trend based on historical data. Specifically, the Kalman filter is used to estimate the amplitude and change rate of the signal. The state equation and observation equation of the Kalman filter can be expressed as:

[0131] Equation of state: x k =Fx k-1 +w k

[0132] Observation equation: z k =Hx k +v k

[0133] Among them, x k is the state vector (including signal amplitude and rate of change), F is the state transfer matrix, H is the observation matrix, and w k and v k are process noise and observation noise respectively.

[0134] The smoothness of gain adjustment is optimized by introducing fuzzy control theory. The input variables of the fuzzy controller are the signal amplitude error and the error change rate, and the output variable is the gain adjustment amount. On the other hand, the frequency-related gain change can be compensated by establishing a gain-frequency characteristic curve and storing it in the Block RAM of the FPGA as a lookup table. For example, for a frequency band of 0.1MHz to 10MHz, the frequency axis can be divided into 100 points, and the gain compensation value corresponding to each point is represented by a 16-bit fixed-point number. According to the current operating frequency band, the accurate gain compensation value is calculated by linear interpolation.

[0135] The above dynamic gain control technology enables the sensor system to maintain good linearity within a dynamic range of 100dB. Specific performance indicators include: gain adjustment range -18dB to +51dB, gain step accuracy 0.5dB, gain adjustment speed <1μs (small signal), <10μs (large signal), input signal range 1mV to 10V, output signal range ±1V, total harmonic distortion <-60dB (1kHz, 1Vpp input).

[0136] S33. Optimize the noise performance of the VGA for different gain settings to ensure a good signal-to-noise ratio over the full dynamic range.

[0137] The above steps achieve a wide dynamic range gain adjustment capability. The output of this stage ensures that the signals from S1 and S2, regardless of strength, can be adjusted to the appropriate amplitude range, providing an ideal input signal for subsequent ADC sampling. Dynamic gain control directly affects the system's ability to handle signals of different strengths and is the key to achieving broadband, high dynamic range measurements. At the same time, the gain information at this stage will also be passed as an important parameter to the subsequent digital signal processing and spectrum analysis stages for accurate recovery and analysis of signal amplitude.

[0138] S4. Use ADC and implement interleaved sampling technology for high-resolution signal sampling:

[0139] S41. Select a suitable high-speed ADC based on the system's bandwidth requirements, resolution requirements, and dynamic range.

[0140] In order to further improve the signal quality and information extraction capability, the present invention integrates advanced digital signal processing technology. A 16-bit, 100MSPS ADC is used to ensure high-resolution sampling. The working principle of this ADC is based on a pipeline structure, and high-speed and high-precision conversion is achieved through multi-stage sub-ADCs and digital correction technology.

[0141] S42. Design a multi-channel interleaved sampling architecture to increase the overall sampling rate. Each ADC samples with a phase-staggered clock, thereby achieving a higher equivalent sampling rate.

[0142] 16-bit resolution means that 2^16=65536 voltage levels can be distinguished, and theoretically a signal-to-noise ratio of 98dB can be achieved. However, in practical applications, due to the influence of noise and nonlinear factors, the effective number of bits (ENOB) is usually around 13-14 bits. In order to further increase the sampling rate, interleaved sampling technology is used. In specific implementation, two 100MSPS ADCs are used, and the two ADCs work alternately through precisely controlled clock phases, thereby increasing the total sampling rate to 200MSPS. The key to this technology is to ensure the time and amplitude matching between the two ADCs.

[0143] S43. Deploy a precise clock distribution network to provide each ADC with a sampling clock with a precise phase to ensure that the sampling time between channels is precisely staggered.

[0144] Time matching is achieved through a precision clock distribution network, and a dedicated clock buffer such as the AD9520 can be used. Amplitude matching requires calibration in the analog front end, and an adjustable gain amplifier can be used to compensate for small differences between the two channels.

[0145] One challenge of interleaved sampling is the spurious components caused by mismatch between channels. Assume that there is a gain mismatch g and a phase mismatch between the two ADCs. Then an image component will be generated in the spectrum, and its amplitude can be estimated by the following formula:

[0146]

[0147] Where SFDR is the spurious free dynamic range.

[0148] A differential input buffer amplifier is designed at the front end of the ADC to improve the ability to resist common-mode interference. The differential structure can not only suppress common-mode noise, but also improve the dynamic range. Assuming that the peak-to-peak voltage of the single-ended signal is Vpp, the peak-to-peak voltage of the differential signal is 2Vpp, which can theoretically improve the signal-to-noise ratio by 6dB. The gain of the differential amplifier can be set by the following formula:

[0149]

[0150] Among them, R f and R g They are the feedback resistor and the gain setting resistor respectively.

[0151] High-performance FPGA is used as the core processing unit to implement complex real-time signal processing algorithms. Xilinx Ultrascale+ series FPGA is selected, the number of DSPSlices can reach 6840, and the number of logic units exceeds 2.1 million, which is enough to support multi-channel parallel FFT and other complex algorithms. The use of radix 4 FFT algorithm can reduce the computational complexity. 1024-point FFT requires log4(1024) = 5 levels of butterfly operations. Each radix 4 butterfly unit requires 3 complex multiplications and 8 complex additions. Assuming fixed decimal point operations, each complex multiplication requires 4 DSPSlices and each complex addition requires 2 DSPSlices. Then, a 1024-point FFT requires approximately 3*4*256+8*2*256=7168 DSPSlices. This exceeds the capacity of a single FPGA, so a parallel processing strategy is needed. The 1024-point FFT can be decomposed into 4 256-point FFTs, and then an additional merging stage is used to obtain the final result. This approach can reduce the use of DSPSlices to about 2000, which is well within the capabilities of XCVU9P.

[0152] The pipeline structure is used to optimize the processing delay. Registers are inserted between each FFT stage so that the entire FFT can be completed within multiple clock cycles. Assuming the FPGA operates at a clock frequency of 500MHz, the theoretical minimum delay of a 256-point FFT is: log4(256)*4=32 clock cycles, or 64ns. Considering the additional overhead in the actual implementation, the total delay can be expected to be around 100-200ns. The performance simulation test results of a single FFT are shown in Figure 2. Figure 4 shown.

[0153] S44. Transmit the digital data output by the ADC to the subsequent FPGA processing unit.

[0154] Through the above steps, high-resolution and wide-bandwidth signal sampling is completed, providing a high-quality digital data foundation for the entire system.

[0155] S5. Preprocess the sampled signal using FPGA:

[0156] S51.FPGA first receives the digital sampled data from the ADC through a high-speed interface. To ensure the continuity and integrity of the data, the FPGA uses its internal BlockRAM to build a multi-level FIFO buffer to effectively prevent data loss and overflow.

[0157] S52. The received raw data needs to be converted to meet the needs of subsequent processing, including bit width adjustment, fixed-point number conversion to floating-point number, etc., to ensure that the data format is compatible with the processing algorithm.

[0158] S53. Use digital filters to remove high-frequency noise and unwanted frequency bands. The filter coefficients can be dynamically adjusted through configuration to adapt to different signal characteristics.

[0159] S54. Condition the filtered signal, including DC offset removal, amplitude normalization, etc.

[0160] S55. Split the continuous data stream into data segments of appropriate length to facilitate subsequent FFT analysis and other algorithm processing. The segment length is configurable to balance processing delay and frequency resolution.

[0161] S56. Apply appropriate window functions (such as Hanning window, Hamming window, etc.) to each data segment to reduce spectrum leakage effects and improve the accuracy of FFT analysis.

[0162] The above steps lay the foundation for subsequent FFT analysis and signal processing, ensuring data quality and processing efficiency. The pre-processed data will be passed to the next step for FFT spectrum analysis.

[0163] S6. Perform FFT spectrum analysis on the sampled signal and convert the sampled time domain signal into frequency domain:

[0164] S61. According to the characteristics of the preprocessed data and the analysis requirements, configure the FFT parameters, including the number of FFT points, sampling rate, etc.

[0165] S62. Load the preprocessed data segment from the BlockRAM of the FPGA to the FFT processing unit.

[0166] S63. Implement the radix-4 FFT algorithm in FPGA. Compared with radix-2 FFT, radix-4 FFT can reduce the computational complexity and improve the processing speed.

[0167] S64. Perform amplitude calculation on the FFT result and convert the complex FFT output into a real amplitude spectrum.

[0168] S66. Calculate the phase information of the FFT result.

[0169] S67. Average the FFT results of multiple consecutive data segments to reduce the influence of random noise and improve the reliability of spectrum estimation. The averaging method can be linear averaging or exponential averaging, which is selected according to actual needs.

[0170] S68. When the frequency resolution needs to be improved, the FFT result is interpolated, which can be achieved by zero padding or a more complex interpolation algorithm to obtain a more refined spectrum representation.

[0171] S69. Store the calculated spectrum data in the BlockRAM of the FPGA or the external DDR3SDRAM, and prepare to transmit it to the subsequent processing unit or the host computer.

[0172] The above steps convert the time domain signal into frequency domain representation, which provides a basis for the frequency characteristic analysis of the signal. The results of FFT analysis directly affect the subsequent adaptive filtering and signal enhancement process, so its accuracy and efficiency are crucial.

[0173] S7. Comprehensively use LMS, FLMS and RLS algorithms for adaptive signal filtering and signal enhancement:

[0174] S71. Select an appropriate adaptive algorithm (LMS, FLMS or RLS) according to the signal characteristics and processing requirements. Initialize the algorithm parameters, including the filter order, step size (for LMS and FLMS) or forgetting factor (for RLS).

[0175] The LMS (Least Mean Square) algorithm is the basis of adaptive filtering. The advantages of the LMS algorithm are simple calculation and easy implementation, but the convergence speed is slow. In order to improve the performance, the LMS algorithm is extended to the frequency domain to form a frequency domain adaptive filtering algorithm. The basic principle of the frequency domain LMS (FLMS) algorithm is to convert the time domain convolution into frequency domain multiplication, and achieve fast calculation through FFT and IFFT. The specific implementation steps are as follows:

[0176] First, the input signal x(n) and the desired signal d(n) are divided into frames, each frame length is N; then N-point FFT is performed on each frame signal to obtain X(k) and D(k); then the filter coefficients are updated in the frequency domain:

[0177] W(k+1)=W(k)+μ*X*(k)*E(k)

[0178] Where W(k) is the filter coefficient, μ is the step size factor, X*(k) is the conjugate of X(k), and E(k) is the frequency domain error signal. Finally, the time domain filtering result is obtained through IFFT. The computational complexity of the FLMS algorithm is O(NlogN), which is significantly lower than the O(N^2) of the time domain LMS.

[0179] In practical applications, the overlap preservation method (OLS) is selected to process the transition between frames to reduce edge effects. For 8192-point FFT, 50% overlap can be selected, that is, 4096 new samples are processed each time. The selection of the step size factor μ is crucial to the performance of the algorithm. The general value range is 0<μ<2 / (λmax), where λmax is the maximum eigenvalue of the autocorrelation matrix of the input signal.

[0180] The FLMS algorithm performs well when processing broadband signals and is particularly suitable for applications such as channel equalization and echo cancellation in communication systems. However, in non-stationary environments, the convergence speed of the FLMS algorithm may not be ideal. In order to further improve the performance, it is possible to consider combining other advanced technologies, such as a variable step size mechanism or introducing the idea of ​​the RLS algorithm to achieve a better balance between convergence speed and computational complexity. The RLS (Recursive Least Squares) algorithm is another important adaptive filtering algorithm that has faster convergence speed and better tracking performance than the LMS algorithm. The core idea of ​​the RLS algorithm is to minimize the weighted least squares error, in which newer data is given a greater weight. The basic update formula of the RLS algorithm is as follows:

[0181] k(n)=λ (-1) *P(n-1)*x(n) / (1+λ (-1) *x T (n)*P(n-1)*x(n))

[0182] α(n)=d(n)-w T (n-1)*x(n)

[0183] w(n)=w(n-1)+k(n)*α(n)

[0184] P(n)=λ (-1) *P(n-1)-λ (-1) *k(n)*x T (n)*P(n-1)

[0185] Where w(n) is the filter coefficient vector, x(n) is the input signal vector, d(n) is the desired signal, λ is the forgetting factor (usually close to 1, such as 0.99), and P(n) is an estimate of the inverse correlation matrix.

[0186] A key advantage of the RLS algorithm is that it converges about 10 times faster than the LMS algorithm. However, the computational complexity of the RLS algorithm is relatively high, O(N^2), where N is the filter order. In practical applications, the fast RLS (FRLS) algorithm can be used to reduce the computational complexity. In practical systems, it is possible to consider combining the advantages of LMS and RLS to design a hybrid algorithm, which can achieve a good balance between performance and computational efficiency, especially suitable for resource-constrained embedded systems.

[0187] S72. Generate or select a suitable reference signal for the adaptive filtering algorithm. The reference signal is based on prior knowledge, a signal model, or is extracted from the original signal.

[0188] S73. Comprehensively utilize the advantages of LMS, FLMS and RLS algorithms to realize the fusion operation of LMS algorithm, FLMS algorithm or RLS algorithm, for example, use RLS to converge quickly in the initial stage, and then switch to LMS with less computational effort to maintain.

[0189] S74. Monitor the performance of the adaptive algorithm in real time, including indicators such as mean square error (MSE) and signal-to-noise ratio improvement. Dynamically adjust algorithm parameters or switch to different algorithms based on these indicators.

[0190] The above steps significantly improve the signal quality and suppress various interferences and noises. The result of adaptive filtering provides high-quality input for subsequent multi-sensor data fusion, ensuring the performance of the entire system.

[0191] S8. Use weighted average and DS evidence theory to fuse the sampling data of multiple sensors:

[0192] S81. First, the data from different sensors are synchronized in time, which involves techniques such as interpolation, resampling or timestamp alignment to ensure that the data from different sensors correspond accurately in time.

[0193] S82. Preprocess the synchronized data, including outlier detection and removal.

[0194] S83. Evaluate the performance indicators of each sensor, such as accuracy, reliability, response time, etc. These indicators will be used for subsequent weighted averaging and weight allocation in DS evidence theory.

[0195] S84. Implement a weighted average algorithm to assign weights according to the performance indicators of each sensor and calculate the fusion result. The weighting scheme can be static or dynamically adjusted according to real-time performance.

[0196] S85. Establish a fusion framework based on DS evidence theory, define the recognition framework, basic probability assignment function (BPA) and combination rules. This step provides a theoretical basis for dealing with uncertainty and conflicting information.

[0197] S86. Assign a basic probability to each sensor's measurement based on the sensor data and performance indicators. This step quantifies the credibility of the information provided by each sensor.

[0198] S87. Apply the DS combination rule to synthesize the evidence from multiple sensors. This process takes into account the mutual support and conflict between sensors and generates a comprehensive confidence assignment.

[0199] S88. Based on the synthesized evidence, use the maximum confidence rule or other decision criteria to obtain the final fusion result, which is a judgment of a certain state or the best estimate of the measurement value.

[0200] Through the implementation of the above steps, the data of multiple sensors are effectively integrated, which improves the overall accuracy and reliability of the measurement. The fused data provides a more reliable basis for the final signal reconstruction and ensures the high quality of the system output.

[0201] S9. Reconstruct the fused sampled data into the original signal based on the least squares method:

[0202] S91. According to the signal characteristics and application requirements, select polynomial fitting or Fourier series expansion as the signal reconstruction model.

[0203] S92. Initialize parameters for the selected reconstruction model. For polynomial fitting, the order of the polynomial needs to be determined; for Fourier series, the number of harmonics needs to be determined.

[0204] S93. The signal reconstruction problem is expressed as a least squares problem, including constructing a measurement matrix and a target vector, wherein the measurement matrix is ​​composed of the basis functions of the selected model, and the target vector is the fused sampled data.

[0205] S94. Solve least squares problems using numerical methods such as QR decomposition or Cholesky decomposition.

[0206] S95. Signal interpolation and smoothing: Interpolate the reconstructed discrete points to generate a continuous signal curve.

[0207] The present invention has the following characteristics:

[0208] 1. The present invention achieves high-precision, wide-bandwidth, and large-dynamic-range magnetic field signal measurement and reconstruction through systematic multi-stage processing. First, the accuracy and spatial coverage of the original signal acquisition are ensured by carefully arranged and calibrated high-frequency magnetic field sensor arrays. The reasonable layout of the sensors takes into account the geometric characteristics of the conductor being measured and the expected magnetic field distribution. At the same time, the initial quality of the measurement is significantly improved through precise calibration and environmental noise baseline measurement.

[0209] 2. The innovative broadband matching network design, especially the introduction of the parallel resonant branch, significantly improves the impedance matching performance of the system within a wide bandwidth and reduces signal transmission loss. This design not only optimizes the performance of the traditional π-type network, but also ensures the stability of the matching network under different environmental conditions through the temperature compensation mechanism. This innovation greatly improves the signal transmission efficiency and signal-to-noise ratio of the system, providing high-quality input for subsequent signal processing. The implementation of the three-level VGA structure not only expands the dynamic range of the system to -18dB to +51dB, but also ensures the flexibility and adaptability of signal processing through an intelligent gain control algorithm. This design can effectively cope with input signals of different intensities, ensuring that the optimal signal level is always maintained, while optimizing noise performance and providing ideal input conditions for subsequent ADC sampling.

[0210] 3. The application of high-speed ADC combined with interleaved sampling technology has greatly improved the sampling rate and resolution of the system, laying a solid foundation for subsequent digital signal processing. This technology not only improves the bandwidth of the system, but also ensures the precise synchronization of multi-channel sampling through a precise clock distribution network, effectively avoiding time deviation between channels. The introduction of FPGA enables complex digital signal preprocessing and FFT analysis to be carried out in real time and efficiently, especially the implementation of the radix 4 FFT algorithm, which significantly improves the speed and efficiency of spectrum analysis. FPGA not only realizes the reception and caching of high-speed data, but also provides high-quality data input for subsequent spectrum analysis through flexible digital filtering, signal conditioning and segmented processing. In particular, the application of window functions and multi-segment data averaging effectively reduce spectrum leakage and improve the reliability of spectrum estimation.

[0211] 4. The present invention innovatively integrates multiple adaptive filtering algorithms such as LMS, FLMS and RLS, and realizes effective suppression of different types of noise and interference through real-time performance monitoring and dynamic algorithm switching. This multi-algorithm fusion strategy not only improves the system's adaptability to various interferences, but also balances the algorithm performance and computational complexity through an intelligent switching mechanism, achieving efficient real-time processing. The multi-sensor data fusion strategy, especially the combined application of weighted average and DS evidence theory, not only improves the overall accuracy of the measurement, but also enhances the system's ability to handle uncertainty and conflicting information. This fusion method fully considers the characteristics and reliability of each sensor, effectively integrates multi-source information through scientific weight allocation and evidence synthesis, and significantly improves the robustness and measurement accuracy of the system.

[0212] 5. Finally, the signal reconstruction process based on the least squares method achieves high-precision recovery of the original signal by flexibly selecting polynomial fitting or Fourier series expansion models. This stage not only takes into account the characteristics of the signal, but also ensures the efficiency and accuracy of the reconstruction process through parameter optimization and the selection of numerical methods. Interpolation and smoothing further improve the continuity and availability of the reconstructed signal. Overall, the present invention significantly improves the performance indicators of the magnetic field signal measurement system through multi-level and multi-angle signal processing and optimization strategies. It has not only made breakthroughs in measurement accuracy, bandwidth and dynamic range, but also demonstrated innovation in anti-interference ability, adaptive processing and multi-source data fusion.

[0213] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range, characterized in that: The steps include: S1. Capture the magnetic field signal around the conductor under test by using a high-frequency magnetic field sensor array; S2. Adjust the broadband matching network to ensure the best impedance matching in the whole frequency band; S3. Achieve wide dynamic range gain adjustment from -18dB to +51dB through a three-stage VGA structure; S4. Use ADC and implement interleaved sampling technology to perform high-resolution signal sampling; S5. Preprocess the sampled signal using FPGA; S6. Perform FFT spectrum analysis on the preprocessed sampled signal and convert the collected time domain signal into frequency domain; S7. Comprehensively use LMS, FLMS and RLS algorithms to perform adaptive signal filtering and signal enhancement on the signal after frequency domain conversion; S8. Use weighted average and DS evidence theory to fuse the sampling data of multiple sensors; S9. Reconstruct the fused sampled data into the original signal based on the least squares method.

2. The non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range as claimed in claim 1, characterized in that: Step S1 includes: S11. Arrange the high-frequency magnetic field sensor array reasonably according to the geometric characteristics of the conductor under test and the expected magnetic field distribution, taking into account the sensor spacing, angle and height factors during arrangement to ensure the best spatial sampling coverage; S12. Before actual measurement, each sensor is accurately calibrated, including sensitivity calibration, frequency response calibration and phase calibration. A standard magnetic field source and precision measuring equipment are used to establish the calibration curve and calibration factor of each sensor to ensure the accuracy and consistency of the measurement results. S13. Before starting the formal measurement, measure and record the environmental noise baseline, and record the environmental factors that may affect the measurement results, including background electromagnetic noise, temperature changes, and vibration; S14. Set up a synchronization trigger mechanism, use a time synchronization device to ensure that all sensors start data collection at the same time, the time synchronization device includes a GPS clock or an atomic clock; S15. performing an initial gain setting for each sensor channel based on the estimated signal strength, taking into account the expected current strength of the conductor being measured and the distance from the sensor to the conductor to ensure that the signal falls within the optimal dynamic range of the sensor; S16. Start the data acquisition system to capture the magnetic field signal around the conductor under test in real time. During the process of capturing the magnetic field signal, continuously monitor the data quality, including signal amplitude and signal-to-noise ratio parameters, to ensure that the collected data is valid and reliable.

3. The non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range as claimed in claim 1, characterized in that: Step S2 includes: S21. Perform a frequency sweep analysis on the entire measurement system using a network analyzer, wherein the measurement system includes a sensor, a transmission line, and a front-end amplifier; S22. According to the network analysis results, an improved broadband matching network is selected, the core of which is to add a parallel resonant branch on the basis of the traditional π-type network; S23. Use a vector network analyzer to test the matching network to verify its matching performance within the actual frequency band. The test indicators include standing wave ratio, return loss, and insertion loss. S24. Fine-tune the matching network according to the test results; S25. Considering the temperature changes in the actual working environment, design and implement a temperature compensation mechanism, use a combination of components with complementary temperature coefficients or add a temperature control circuit to ensure that the matching network maintains stable performance at different temperatures.

4. The non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range as claimed in claim 1, characterized in that: Step S3 includes: S31. Design a three-stage cascaded variable gain amplifier (VGA) structure to achieve an overall dynamic range of -18dB to +51dB while ensuring linearity and noise performance; S32. Deploy an intelligent gain control algorithm that can automatically adjust the gain of each VGA level according to the strength of the input signal; S33. Optimize the noise performance of the VGA for different gain settings to ensure a good signal-to-noise ratio over the full dynamic range.

5. The non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range as claimed in claim 1, characterized in that: Step S4 includes: S41. Select a suitable high-speed ADC based on the bandwidth requirements, resolution requirements, and dynamic range of the system; S42. Design a multi-channel interleaved sampling architecture to increase the overall sampling rate. Each ADC samples with a phase-staggered clock, thereby achieving a higher equivalent sampling rate. S43. Deploy a precise clock distribution network to provide each ADC with a sampling clock with a precise phase to ensure that the sampling time between channels is precisely staggered; S44. Transmit the digital data output by the ADC to the subsequent FPGA processing unit.

6. The non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range as claimed in claim 5, characterized in that: Step S5 includes: S51.FPGA receives digital sampled data from ADC through high-speed interface. To ensure the continuity and integrity of data, FPGA uses its internal BlockRAM to build a multi-level FIFO buffer to effectively prevent data loss and overflow. S52. The received raw data is format converted to meet the needs of subsequent processing. The format conversion includes bit width adjustment and fixed-point to floating-point conversion operations to ensure that the data format is compatible with the processing algorithm; S53. Use digital filters to remove high frequency noise and unwanted frequency bands; S54. Conditioning the filtered signal, including DC offset removal and amplitude standardization; S55. Divide the continuous data stream into data segments of appropriate length to facilitate subsequent FFT analysis and other algorithm processing; S56. Apply an appropriate window function to each data segment to reduce spectrum leakage effect and improve the accuracy of FFT analysis. The window function includes a Hanning window and a Hamming window.

7. The non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range as claimed in claim 6, characterized in that: Step S6 includes: S61. According to the characteristics of the preprocessed data and the analysis requirements, configure the FFT parameters, including the number of FFT points and the sampling rate; S62. Load the pre-processed data segment from the BlockRAM of the FPGA to the FFT processing unit; S63. Implementing radix 4 FFT algorithm in FPGA; S64. Perform amplitude calculation on the FFT result, and convert the complex FFT output into a real amplitude spectrum; S66. Calculate the phase information of the FFT result; S67. Averaging the FFT results of multiple consecutive data segments to reduce the influence of random noise and improve the reliability of spectrum estimation, the averaging method is linear averaging or exponential averaging; S68. When the frequency resolution needs to be improved, the FFT result is interpolated, and the interpolation is implemented by zero padding or a more complex interpolation algorithm to obtain a finer spectrum representation; S69. Store the calculated spectrum data in the BlockRAM of the FPGA or the external DDR3SDRAM, and transmit it to a subsequent processing unit or a host computer.

8. The non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range as claimed in claim 1, characterized in that: Step S7 includes: S71. According to the signal characteristics and processing requirements, select an appropriate adaptive algorithm, the adaptive algorithm includes LMS, FLMS or RLS, initialize the algorithm parameters, including filter order, step size or forgetting factor; S72. Generate or select a suitable reference signal for the adaptive filtering algorithm, the reference signal is based on prior knowledge, a signal model or extracted from the original signal; S73. Comprehensively utilize the advantages of LMS, FLMS and RLS algorithms to realize the fusion operation of LMS algorithm, FLMS algorithm or RLS algorithm, use RLS to converge quickly in the initial stage, and then switch to LMS maintenance with less calculation amount; S74. Monitor the improvement indicators of the adaptive algorithm in real time, including mean square error (MSE) and signal-to-noise ratio, and dynamically adjust algorithm parameters or switch to different algorithms based on these improvement indicators.

9. The non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range as claimed in claim 1, characterized in that: Step S8 includes: S81. First, the data from different sensors are synchronized in time, and interpolation, resampling or timestamp alignment technology is used to ensure that the data from different sensors correspond accurately in time; S82. Preprocessing the synchronized data, including outlier detection and removal; S83. Evaluate the performance indicators of each sensor, including accuracy, reliability, and response time; S84. Implement a weighted average algorithm to assign weights according to the performance indicators of each sensor and calculate the fusion results. The weighting scheme is static or dynamically adjusted according to real-time performance; S85. Establish a fusion framework based on DS evidence theory, define the recognition framework, basic probability distribution function BPA and combination rules; S86. Assigning a basic probability to the measurement result of each sensor based on the sensor data and the performance indicator; S87. Apply DS combination rules to synthesize evidence from multiple sensors; S88. Based on the synthesized evidence, use the maximum confidence rule or other decision criteria to obtain the final fusion result, which is a judgment of a certain state or the best estimate of the measurement value.

10. The non-invasive high-frequency magnetic field sensing data acquisition method based on a high-bandwidth acquisition range as claimed in claim 1, characterized in that: Step S9 includes: S91. According to the signal characteristics and application requirements, select polynomial fitting or Fourier series expansion as the signal reconstruction model; S92. Initializing parameters for the selected reconstruction model, for polynomial fitting, determining the order of the polynomial; for Fourier series, determining the number of harmonics; S93. Formulate the signal reconstruction problem as a least squares problem, including constructing a measurement matrix and a target vector, wherein the measurement matrix is ​​composed of basis functions of the selected model, and the target vector is the fused sampled data; S94. Solve least squares problems using QR decomposition or Cholesky decomposition methods; S95. Signal interpolation and smoothing: Interpolate the reconstructed discrete points to generate a continuous signal curve.

Images