Multi-band Radio Measurement Calibration Method and System

Abstract

The present invention relates to the field of radio metrology technology, and discloses a multi-band radio metrology calibration method and system. The method includes: S1: Sampling and preprocessing a multi-band radio signal to obtain a preprocessed signal; S2: Performing signal classification to obtain a mixed signal; S3: Reconstructing the mixed signal to obtain a pure target signal; S4: Dynamically dividing the pure target signal into sub-bands, introducing a regularization factor in each sub-band, and iteratively calculating the minimum mean square error until the residual is less than a first target value to obtain initial calibration parameters; S5: Establishing a frequency point distribution with logarithmic spacing based on the initial calibration parameters to obtain compensation calibration coefficients; S6: Verifying the compensation calibration coefficients using a standard calibration signal. When the root mean square error exceeds a second target value, repeat steps S4-S6 until the accuracy requirement is met to obtain a target calibration result. The present invention realizes effective suppression of interference signals and improves the calibration accuracy.

Description

Technical Field

[0001] The present invention relates to the field of radio metrology technology, and particularly to a multi-band radio metrology calibration method and system. Background Art

[0002] With the rapid development of radio communication technology, multi-band radio devices are widely used in various fields, which puts forward higher requirements for calibration accuracy and efficiency. However, in the actual calibration environment, external electromagnetic interference will significantly affect the quality of the measurement signal, resulting in a decrease in the accuracy of the calibration result.

[0003] Traditional radio metrology calibration methods mainly adopt a single-band calibration strategy, which is difficult to meet the calibration requirements of multi-band devices. At the same time, due to the use of fixed signal processing parameters and simple interference suppression algorithms, in an environment with complex electromagnetic interference, the calibration accuracy is easily affected. Especially when performing high-precision calibration, external interference will cause a large deviation in the measurement result. Summary of the Invention

[0004] The present invention provides a multi-band radio metrology calibration method and system, which effectively suppresses interference signals and improves calibration accuracy.

[0005] In the first aspect, the present invention provides a multi-band radio metrology calibration method, and the multi-band radio metrology calibration method includes:

[0006] S1: Sampling and preprocessing a multi-band radio signal to obtain a preprocessed signal;

[0007] S2: Performing a fast Fourier transform and signal classification on the preprocessed signal to obtain a mixed signal;

[0008] S3: Constructing the mixed signal into a Hankel matrix and performing signal reconstruction in combination with a correlation coefficient threshold to obtain a pure target signal;

[0009] S4: Performing dynamic sub-band division on the pure target signal, introducing a regularization factor with adaptive signal-to-noise ratio in each sub-band, and iteratively calculating the minimum mean square error until the residual is less than a first target value to obtain initial calibration parameters;

[0010] S5: Establishing a logarithmic-spacing frequency point distribution based on the initial calibration parameters and performing linear weighted averaging within the overlapping bandwidth of adjacent sub-bands to obtain a compensation calibration coefficient;

[0011] S6: Verifying the compensation calibration coefficient with a standard calibration signal. When the root mean square error exceeds a second target value, repeat steps S4 - S6 until the accuracy requirement is met to obtain a target calibration result.

[0012] In a second aspect, the present invention provides a multi-band radio metrology calibration system, and the multi-band radio metrology calibration system includes:

[0013] A sampling module, configured to sample and preprocess a multi-band radio signal to obtain a preprocessed signal;

[0014] A classification module, configured to perform a fast Fourier transform and signal classification on the preprocessed signal to obtain a mixed signal;

[0015] A reconstruction module, configured to construct the mixed signal into a Hankel matrix, and perform signal reconstruction in combination with a correlation coefficient threshold to obtain a pure target signal;

[0016] A dynamic sub-band division module, configured to perform dynamic sub-band division on the pure target signal, introduce a regularization factor with SNR adaptability in each sub-band, and iteratively calculate the minimum mean square error until the residual is less than a first target value to obtain an initial calibration parameter;

[0017] A linear weight averaging module, configured to establish a logarithmic-spacing frequency point distribution based on the initial calibration parameter, and perform linear weight averaging within an overlapping bandwidth of adjacent sub-bands to obtain a compensation calibration coefficient;

[0018] A verification module, configured to verify the compensation calibration coefficient by using a standard calibration signal. When the root mean square error exceeds a second target value, repeat the steps from the dynamic sub-band division module to the verification module until the accuracy requirement is met to obtain a target calibration result.

[0019] In the technical solution provided by the present invention, by adopting a dynamic sampling and preprocessing strategy, in combination with a Hanning window function and a band-pass filter bank, the quality of signal acquisition is effectively improved. Based on a signal classification method using a fast Fourier transform and a support vector machine classifier, the effective components and interference components in the mixed signal are accurately identified, significantly improving the accuracy of signal classification; by using Hankel matrix decomposition and singular value reconstruction techniques, in combination with correlation coefficient threshold judgment, effective suppression of interference signals is achieved, ensuring the purity of the target signal; by introducing dynamic sub-band division and an adaptive regularization factor, and iteratively optimizing the minimum mean square error, the computational complexity is effectively reduced, and the calibration efficiency is improved; through a logarithmic-spacing frequency point distribution and a linear weight averaging strategy, smooth transition between adjacent sub-bands is achieved, ensuring the continuity and stability of the calibration result; a complete calibration verification and optimization mechanism is established. Through root mean square error control and iterative optimization, the reliability and accuracy of the calibration result are ensured, significantly improving the overall calibration accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0021] Figure 1 It is a schematic diagram of the steps of the multi-band radio metrology calibration method in the embodiments of the present invention;

[0022] Figure 2 It is a schematic diagram of the structure of the multi-band radio metrology calibration system in the embodiments of the present invention. Specific embodiments

[0023] The embodiments of the present invention provide a multi-band radio metrology calibration method and system. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned accompanying drawings of the present invention are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the term "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0024] For ease of understanding, the following describes the specific process of the embodiments of the present invention. Please refer to Figure 1 An embodiment of the multi-band radio metrology calibration method in the embodiments of the present invention includes:

[0025] Step S1: Sample and preprocess the multi-band radio signal to obtain a preprocessed signal;

[0026] It can be understood that the execution subject of the present invention can be a multi-band radio metrology calibration system, or a terminal or a server. Specifically, it is not limited here. The embodiments of the present invention are described by taking the server as the execution subject as an example.

[0027] Specifically, the sampling frequency is set according to the highest frequency of the signal to be measured to meet the requirements of the Nyquist sampling theorem, ensure the integrity of the signal, and avoid the generation of aliasing effects. On this basis, a high-precision analog-to-digital converter (ADC) is used to digitally sample the multi-band radio signal to obtain a digital sampling signal with high time resolution. The digital sampling signal is framed based on a preset overlap rate and the number of sampling points to obtain a framed signal. The frame length is reasonably selected so that it can cover sufficient signal information and avoid a reduction in spectral resolution caused by an overly short time window. At the same time, an appropriate overlap rate is used so that there is a certain degree of redundancy between adjacent frames to reduce the boundary effects that occur in subsequent signal processing processes such as the short-time Fourier transform. The framed signal is windowed with a Hanning window function. The Hanning window function has the characteristic of smooth transition in the time domain, thereby effectively reducing the energy of the spectral sidelobes and improving the frequency resolution of the signal. By applying the Hanning window to the framed signal, the signal edge is smoothly transitioned to avoid spectral distortion problems caused by truncation effects. The windowed signal is input into an analog-to-digital converter (ADC) for quantization processing to obtain a quantization signal that better meets the requirements of digital signal processing. The quantization signal is corrected for DC bias using a high-pass filter or adaptive mean filtering method to eliminate low-frequency components and restore the true waveform of the signal to obtain a signal corrected for DC bias. The corrected signal is band-pass filtered. Five groups of band-pass filters are used, and the center frequency of each filter corresponds to each frequency band of the signal to be measured, thereby ensuring that signals in different frequency bands are extracted and analyzed separately. The design of the band-pass filter needs to take into account the bandwidth characteristics of the signal, ensure that its passband range matches the spectral characteristics of the target signal, and at the same time have a relatively steep cut-off characteristic to reduce the influence of out-of-band interference signals. Through this process, filtered signals corresponding to different frequency bands are obtained. The filtered signals are stored in a signal buffer to obtain a preprocessed signal for subsequent processing modules to sequentially read and perform corresponding calculations. The design of the signal buffer meets the requirements of low latency and high throughput to ensure that the entire system can efficiently process multi-band radio signals and avoid affecting the calibration accuracy due to buffer overflow or read latency.

[0028] Step S2: Perform a fast Fourier transform and signal classification on the preprocessed signal to obtain a mixed signal;

[0029] Specifically, perform a fast Fourier transform on the preprocessed signal to convert the time-domain signal into a frequency-domain signal spectrum, which contains the amplitude and phase information of different frequency components. The amplitude spectrum is used for signal classification and feature extraction, while the phase spectrum is used for subsequent signal reconstruction and coherence analysis. To analyze the energy distribution of the signal, normalize the frequency-domain signal spectrum and calculate the ratio of the energy in the target frequency band to the energy in the full frequency band to obtain the spectral energy ratio feature parameter, which reflects the energy proportion of a specific frequency band in the entire spectrum and is further used to judge the relative intensity of the target signal in the overall spectrum. Calculate the ratio of the spectral centroid to the spectral bandwidth based on the frequency-domain signal spectrum to obtain the spectral concentration feature parameter. The spectral centroid represents the position of the "center of gravity" of the signal in the spectrum and measures whether the frequency distribution of the signal is concentrated in a specific frequency band, while the spectral bandwidth reflects the spectral expansion degree of the signal, that is, the width of the signal energy distribution. By calculating the ratio of the spectral centroid to the spectral bandwidth, the concentration degree of the signal is characterized to distinguish narrowband signals from broadband signals. At the same time, extract the main peak and secondary peak of the frequency-domain signal spectrum and calculate the energy difference between them. The main peak is the strongest frequency component in the signal, while the secondary peak is the secondary frequency component caused by modulation, interference, or other signal characteristics. Calculate the energy difference between the main peak and the secondary peak and perform normalization processing to obtain the main-secondary peak energy difference feature parameter, which effectively reflects the modulation characteristics of the signal and the possibility of the existence of mixed signals. Construct a three-dimensional feature vector based on the spectral energy ratio feature parameter, the spectral concentration feature parameter, and the main-secondary peak energy difference feature parameter. To ensure the efficiency and accuracy of classification, use a support vector machine (SVM) classifier based on the Gaussian kernel function to classify the three-dimensional feature vector. SVM is a machine learning method suitable for data classification tasks in high-dimensional feature spaces. By using the Gaussian kernel function, different categories of signals can be effectively identified in the non-linear space, improving the classification accuracy. In the training stage, use a set of signal samples with known categories to train the SVM to determine the optimal classification hyperplane. In the actual classification stage, the SVM projects the input three-dimensional feature vector into the high-dimensional feature space and uses the support vectors to determine the class label of the signal. After the SVM classifier completes the classification, extract the data segments determined to be mixed signals from the preprocessed signal based on the classification results and record the corresponding time positions and frequency characteristics to obtain the complete mixed signal information.

[0030] Step S3: Construct the mixed signal into a Hankel matrix and perform signal reconstruction in combination with the correlation coefficient threshold to obtain a pure target signal;

[0031] Specifically, perform appropriate matrix transformation on the mixed signal to decompose the signal structure using linear algebra methods. Construct a Hankel matrix with a specific structure from the mixed signal according to one-third of the number of sampling points as the number of rows of the Hankel matrix, so as to provide a suitable matrix form for subsequent singular value decomposition. The characteristic of the Hankel matrix is that the elements on its diagonal are the same. This structure can effectively retain the timing information of the signal and highlight the main signal components in the subsequent decomposition process, while suppressing random noise and interference components, resulting in a signal Hankel matrix. Perform singular value decomposition on the Hankel matrix to obtain the main components of the signal. Singular value decomposition disassembles the matrix into left and right singular vectors and their corresponding singular value sequences, where the singular values represent the energy distribution of the signal, and the singular vectors correspond to different characteristic modes of the signal. Through this decomposition method, the main components of the signal are effectively extracted, and the low-energy noise or interference components are identified. Further analyze the obtained singular value sequence, that is, arrange the singular values in descending order according to the energy contribution rate and calculate the cumulative energy contribution rate. When the cumulative energy contribution rate reaches 95%, it is considered that the main signal characteristics have been fully captured. Therefore, extract these singular values and their corresponding singular vectors to obtain the signal subspace characteristics. Based on the signal subspace characteristics, reconstruct the mixed signal to remove noise and irrelevant components, obtaining a preliminary reconstructed signal. Use the interference feature library to perform correlation analysis on the preliminary reconstructed signal to identify the existing interference components. By calculating the correlation coefficient between the preliminary reconstructed signal and the known interference patterns in the interference feature library and setting a reasonable correlation coefficient threshold, the signal is decomposed into target signal components and interference signal components. A higher correlation coefficient indicates that this part of the signal is affected by the known interference pattern and should therefore be classified as an interference signal component, while the signal part with a lower correlation is retained as the target signal component. Perform time-frequency analysis on the target signal components to ensure their purity. Time-frequency analysis uses methods such as short-time Fourier transform to accurately describe the distribution of the signal at different times and frequencies. Through this analysis process, the remaining interference signal components can be more clearly identified and removed by appropriate filtering or non-linear signal decomposition methods, finally obtaining a pure target signal.

[0032] Step S4: Dynamically divide the pure target signal into subbands, introduce a regularization factor with SNR adaptability in each subband, and iteratively calculate the minimum mean square error until the residual is less than the first target value to obtain the initial calibration parameter;

[0033] Specifically, perform spectral analysis on the pure target signal to determine an appropriate sub-band division scheme. The sub-band division should be dynamic, that is, adaptively divide into N sub-bands according to the power spectral density characteristics and energy concentration regions of the signal, so as to ensure that each sub-band fully covers the main feature information of the signal, while avoiding the loss of calibration accuracy caused by unnecessary spectral segmentation. After the division is completed, an N-sub-band signal sequence is obtained. Each sub-band corresponds to a certain frequency range and contains the corresponding target signal components. Perform statistical analysis on each sub-band signal in the sub-band signal sequence, calculate the autocorrelation matrix and cross-correlation vector to extract the relevant characteristic quantities of the signal. The autocorrelation matrix reflects the time correlation of the signal, while the cross-correlation vector describes the correlation characteristics between different sub-bands. Calculate the signal-to-noise ratio of each sub-band based on the relevant characteristic quantities, and dynamically generate a regularization factor according to the signal-to-noise ratio value. The regularization factor plays a role in balancing the stability and accuracy of the solution in the optimization calculation. For sub-bands with high signal-to-noise ratio, a smaller regularization factor is used to improve the accuracy of parameter estimation, while for sub-bands with low signal-to-noise ratio, a larger regularization factor is required to suppress the influence of noise, thereby improving the robustness of the system. The finally obtained optimization parameters will provide necessary adjustment terms for the iterative calculation to ensure the convergence and accuracy of the solution. Substitute the optimization parameters into the Wiener-Hopf equation and use the conjugate gradient method to solve it to obtain the iterative calculation sequence. The Wiener-Hopf equation is an optimal filtering equation, which has high computational efficiency and theoretical optimality when solving the minimum mean square error problem. Since the equation involves matrix inversion calculation, direct solution leads to too high computational complexity, so the conjugate gradient method is used for numerical iterative solution. The conjugate gradient method is an efficient optimization algorithm suitable for solving large-scale linear equations. By gradually approaching the optimal solution, it converges to a stable solution within fewer iterations, thereby improving the computational efficiency. During the iterative calculation process, continuously monitor the change of the residual, and calculate the residual of the current solution after each iteration. When the residual is less than the preset first target value, it is considered that the solution has reached the convergence state, and the final sub-band calibration parameters are obtained. Perform local iterative optimization on the frequency points in the sub-band calibration parameters whose errors exceed the threshold to ensure that the final calibration accuracy meets the expected target. During the local optimization process, the optimization step size should be gradually decreased to avoid numerical oscillation or unstable optimization effect caused by too large a step size. The optimization step size decreases by 0.01 times the initial step size, so as to achieve more refined error correction and make the final optimization result more accurately approach the true calibration parameters. Combine the optimization results of each sub-band to ensure the continuity and consistency of the calibration parameters in the entire frequency band range, and finally obtain accurate initial calibration parameters.

[0034] Step S5: Based on the initial calibration parameters, establish a logarithmically spaced frequency point distribution, and perform linear weighted averaging within the overlapping bandwidth of adjacent sub-frequency bands to obtain a compensation calibration coefficient;

[0035] Specifically, the signal frequency band to be measured is divided into M sub - frequency bands, and 3 calibration frequency points are distributed at logarithmic intervals within each sub - frequency band to construct a frequency point distribution method that better conforms to the physical characteristics of the signal. The logarithmic - interval calibration frequency point distribution is more adaptable to the uneven characteristics of the radio signal frequency distribution compared to the linear equal - interval distribution. At the same time, it ensures that the sampling density in the high - frequency region is more reasonable than that in the low - frequency region, thereby improving the calibration accuracy. Through this step, a calibration frequency point distribution covering the entire test frequency band is obtained. Based on the calibration frequency point distribution, interpolation processing is performed on the initial calibration parameters to construct the frequency response function of each sub - frequency band. For the interpolation calculation of each sub - frequency band, interpolation processing is not only performed on the amplitude response but also on the phase response to ensure the integrity of the calibration parameters. During the interpolation process of amplitude and phase, the piece - wise cubic spline interpolation method is used. This method ensures interpolation smoothness while maintaining a high fitting accuracy and avoiding over - fitting problems. Through this process, a frequency response curve is obtained, enabling the initial calibration parameters to be extended to the entire calibration frequency point distribution and forming a relatively accurate frequency response model. A certain overlapping bandwidth is introduced between adjacent sub - frequency bands, and linear weighted averaging of the frequency response is performed within this overlapping bandwidth to obtain the calibration value of the transition region, thereby achieving a continuous frequency response transition at the junction of sub - frequency bands and avoiding the problem of spectral discontinuity caused by independent calculations between sub - frequency bands. The weight assignment within the overlapping bandwidth adopts a linear function form, that is, at the starting position of the overlapping bandwidth, the weight of the lower - frequency band is larger, while the weight of the higher - frequency band is smaller. As the frequency increases, the weights are gradually reversed, thus forming a smooth transition calibration value within the entire overlapping region. Through linear weighted averaging calculation, the mutation during spectrum splicing is effectively eliminated, enabling the entire frequency response curve to maintain good continuity between different sub - frequency bands. Based on the frequency response curve and the calibration value of the transition region, the calibration coefficient of each frequency point is calculated to construct a compensation sequence. The calculation method of the compensation sequence is based on the reciprocal of the frequency response or the inverse fitting model to ensure that the calibration coefficient can effectively compensate for the system error and keep the measured signal highly accurate across the entire frequency spectrum range. During the calculation process, the non - linear effects and measurement errors of the system are considered to optimize the stability and applicability of the compensation coefficient. The compensation sequence is iteratively optimized, that is, the updated compensation sequence is calculated in each iteration cycle and compared with the result of the previous cycle until the change rate between adjacent iteration cycles is less than the preset target percentage value, at which point the optimization process is terminated and the stable compensation calibration coefficient is finally obtained. The core of iterative optimization lies in dynamically adjusting the compensation coefficient to continuously approach the optimal calibration solution. During the iterative process, an adaptive step - size adjustment strategy is adopted, enabling the optimization process to converge quickly with a large step - size in the initial stage and gradually reduce the step - size in the later optimization stage to ensure that the accuracy of the final calibration coefficient reaches the optimal level.

[0036] Step S6: Verify the compensation calibration coefficients using a standard calibration signal. When the root mean square error exceeds the second target value, repeat Steps S4 - S6 until the accuracy requirement is met to obtain the target calibration result.

[0037] Specifically, a standard signal source is used to generate calibration signals to ensure that the reference signal for the entire measurement process has sufficient stability and accuracy. The standard signal source needs to cover the entire frequency band to be measured and generate a series of equally spaced standard calibration signals at a nominal frequency interval of 20 kHz. This interval selection ensures that there are sufficiently dense measurement points across the entire frequency band without causing an excessive computational burden, thus guaranteeing high precision and operability for subsequent calibration error calculation and optimization adjustment. After the signals generated by the standard signal source cover the entire frequency band to be measured, standard calibration signals are obtained. The standard calibration signals are compared point by point with the compensation calibration coefficients in terms of frequency and amplitude to accurately calculate the measurement error at each frequency point. The amplitude and frequency of the standard calibration signals are precisely measured and compared with the measurement results corrected by the compensation calibration coefficients to obtain the error values at each frequency point. An error distribution histogram is generated based on the measurement errors at each frequency point to visually display the distribution characteristics of the errors, and the root mean square error and maximum error of each sub-band are calculated to obtain the error statistical results. The root mean square error is an important indicator for measuring measurement accuracy, reflecting the overall level of the errors, while the maximum error is used to evaluate the measurement deviation in the most extreme cases. By calculating the error statistical indicators, the performance of the current compensation calibration coefficients is reflected. An optimization determination is made on the error statistical results to decide whether further calibration parameters need to be adjusted. The calculated root mean square error is compared with the second target value to determine whether the current calibration accuracy meets the requirements. If the root mean square error is lower than the second target value, it indicates that the optimization of the compensation calibration coefficients has achieved the expected effect, and the final target calibration result is obtained. Based on the optimization determination result, the frequency bands with root mean square errors exceeding the second target value are marked, and the key parameters during the calibration process are recorded to form frequency band optimization marks. These key parameters include the current calibration coefficients, the trend of measurement errors, the signal-to-noise ratio distribution, and the convergence of the previous few rounds of optimization iterations. By comprehensively analyzing this information, it is determined which factors affect the frequency bands with larger errors. After determining the frequency bands that need to be optimized, return to step S4 according to the frequency band optimization marks, and repeat steps S4 - S6 to specifically optimize the calibration parameters of the marked frequency bands. During the optimization process, the calibration accuracy is improved by adjusting the signal-to-noise ratio adaptive regularization factor, optimizing the sub-band division strategy, and improving the minimum mean square error calculation method. At the same time, during the iterative optimization process, a dynamic step size adjustment strategy is adopted to ensure that the optimization process can converge quickly without causing numerical oscillations due to an overly large step size. After each iterative optimization, the root mean square error is recalculated and compared with the second target value to determine whether further optimization is required. When the root mean square error finally converges to be lower than the second target value, it is considered that the calibration process has achieved the expected accuracy requirement, and the final target calibration result is obtained.

[0038] In the embodiments of the present invention, by adopting a dynamic sampling and preprocessing strategy, combining with the Hanning window function and a bank of band-pass filters, the quality of signal acquisition is effectively improved. Based on the signal classification method using the fast Fourier transform and the support vector machine classifier, the effective components and interference components in the mixed signal are accurately identified, significantly improving the accuracy of signal classification; by using the Hankel matrix decomposition and singular value reconstruction technology, combined with the correlation coefficient threshold judgment, the effective suppression of interference signals is achieved, ensuring the purity of the target signal; by introducing dynamic sub-band division and an adaptive regularization factor, and iteratively optimizing the least mean square error, the computational complexity is effectively reduced and the calibration efficiency is improved; through the logarithmic-spaced frequency point distribution and the linear weight averaging strategy, the smooth transition between adjacent sub-bands is achieved, ensuring the continuity and stability of the calibration result; a complete calibration verification and optimization mechanism is established. Through the root mean square error control and iterative optimization, the reliability and accuracy of the calibration result are ensured, significantly improving the overall calibration accuracy.

[0039] In a specific embodiment, the process of executing step S1 may specifically include the following steps:

[0040] Set the sampling frequency according to the highest frequency of the signal to be measured, and digitally sample the multi-band radio signal based on the sampling frequency to obtain a digital sampling signal;

[0041] Based on a preset overlap rate and the number of sampling points, perform frame division processing on the digital sampling signal to obtain a framed signal;

[0042] Perform windowing processing on the framed signal with the Hanning window function and quantize it through an analog-to-digital converter to obtain a quantized signal, and perform DC bias correction on the quantized signal to obtain a corrected signal;

[0043] Filter the corrected signal through five groups of band-pass filters to obtain a filtered signal, and store the filtered signal in the signal buffer to obtain a preprocessed signal, where the center frequencies of each band-pass filter respectively correspond to the respective frequency bands of the signal to be measured.

[0044] Specifically, set the sampling frequency according to the highest frequency of the signal to be measured to ensure the integrity of the signal and avoid the influence of the aliasing effect. According to the Nyquist sampling theorem, that is, the sampling frequency must be at least twice the highest frequency of the signal to be measured , that is:

[0045]

[0046] where represents the highest frequency of the signal to be measured, is the set sampling frequency. If the signal bandwidth is relatively wide, appropriately increase the sampling rate to ensure that there is no spectral aliasing phenomenon in the subsequent signal processing process. Based on this sampling frequency, the multi-band radio signal is digitally sampled to obtain a digital sampling signal. The process of digital sampling involves analog-to-digital conversion, that is, converting a continuous analog signal into a discrete digital signal. Let the number of sampling points be , then the discrete sampling signal is expressed as:

[0047]

[0048] where, represents the discrete signal value of the th sampling point, is the original analog signal, is the sampling interval, that is:

[0049]

[0050] Every seconds, a signal value is collected to form a digital time-domain signal sequence. The sampled digital signal is framed to improve the computational efficiency of subsequent analysis and reduce the impact of short-time signal mutations on the overall spectrum calculation. The framing method is based on a preset overlap rate and the number of sampling points . The overlap rate is set between 50% and 75% to ensure sufficient information sharing between adjacent frames and reduce the impact of signal mutations. If the length of each frame is , then the step size of adjacent frames is:

[0051]

[0052] For example, if the frame length is 1024 points and the overlap rate is set to 50%, then the step size of adjacent frames is 512 points, obtaining a series of overlapping framed signal sequences. In order to reduce the spectral leakage effect and improve the spectral resolution during framing, the framed signal is windowed, using the Hanning window function, and its mathematical expression is:

[0053]

[0054] where, is the window value of the Hanning window function, is the current sample point, is the frame length. The windowed signal is expressed as:

[0055]

[0056] The Hanning window can effectively reduce the sidelobes and improve the accuracy of signal analysis. After windowing, the signal is quantized by an analog-to-digital converter to convert it into a digital value with a fixed number of bits. For example, if the ADC has a 12-bit resolution, the numerical range of the signal is quantized to different discrete levels, thereby improving the accuracy of subsequent signal processing. The quantized signal is corrected for DC bias. The DC bias is calculated by taking the average value of the signal , that is:

[0057]

[0058] Then the signal is debiased with this value, that is:

[0059]

[0060] where is the corrected signal. After correcting the signal, the signal is filtered by five groups of band-pass filters, and the center frequencies of these filters correspond to the respective frequency bands of the signal to be measured. Assume that the frequency bands of the signal to be measured include , then the transfer function of each band-pass filter is expressed as:

[0061]

[0062] where is the center frequency of the th band-pass filter, is the filter order, taking 4 or 6 to ensure good passband characteristics and steep cut-off characteristics. The function of each filter is to extract the signal in the corresponding frequency band and suppress other frequency components. The filtered signal is stored in the signal buffer for further signal analysis and processing later. The design of the signal buffer needs to ensure sufficient storage space to avoid data overflow, and the read rate must match the sampling rate to ensure data integrity.

[0063] In a specific embodiment, the process of executing step S2 may specifically include the following steps:

[0064] Perform a fast Fourier transform on the preprocessed signal to obtain the frequency-domain signal spectrum;

[0065] Calculate the ratio of the energy of the target frequency band to the energy of the full frequency band in the frequency-domain signal spectrum to obtain the spectral energy ratio characteristic parameter;

[0066] Calculate the ratio of the spectral centroid to the spectral bandwidth based on the frequency-domain signal spectrum to obtain the spectral concentration characteristic parameter;

[0067] Extract the energy difference between the main peak and the secondary peak from the frequency-domain signal spectrum and perform normalization processing to obtain the main-secondary peak energy difference characteristic parameter;

[0068] Construct a three-dimensional feature vector based on the spectral energy ratio characteristic parameter, the spectral concentration characteristic parameter, and the main-secondary peak energy difference characteristic parameter, and input the three-dimensional feature vector into a support vector machine classifier based on the Gaussian kernel function to obtain a classification result;

[0069] Extract the data segment determined to be a mixed signal, its corresponding time position, and frequency characteristics from the preprocessed signal based on the classification result to obtain the mixed signal.

[0070] Specifically, perform a fast Fourier transform on the preprocessed signal to convert the signal from the time domain to the frequency domain to obtain spectral information. The mathematical expression of the fast Fourier transform is:

[0071]

[0072] where, is the complex representation of the th frequency component, is the time-domain signal of the th sampling point, is the number of points of the Fourier transform, is the imaginary unit, represents the frequency index. In actual calculations, an efficient fast Fourier transform algorithm is used to implement the transform with a time complexity of to accelerate the spectral analysis process. The obtained frequency-domain signal spectrum contains the amplitude information and phase information of the signal, and the amplitude spectrum is used for feature extraction. After obtaining the frequency-domain signal spectrum, calculate the spectral energy ratio characteristic parameter, that is, the ratio of the energy of the target frequency band to the energy of the full frequency band. The energy of the target frequency band is expressed as:

[0073]

[0074] where, and respectively represent the starting frequency index and the ending frequency index of the target frequency band, represents the power of the frequency-domain signal spectrum at the th frequency point. The energy of the full frequency band is the total energy within the entire frequency spectrum range, and its calculation method is:

[0075]

[0076] The spectral energy ratio characteristic parameter is expressed as:

[0077]

[0078] Among them, reflects the proportion of the energy in the target frequency band to the total energy of the entire signal. This parameter is used to measure the significance of the target signal in signal classification. Calculate the spectral concentration characteristic parameter, which is represented by the ratio of the spectral centroid to the spectral bandwidth . The calculation method of the spectral centroid is as follows:

[0079]

[0080] Among them, is the actual frequency corresponding to the th frequency point. The numerator part represents the weighted sum of the signal energy on the frequency axis, and the denominator is the total signal energy, ensuring that the calculated reflects the position of the "center of gravity" of the signal. The spectral bandwidth is calculated based on the extensibility of the signal energy distribution and is calculated in the way of the second moment, that is:

[0081]

[0082] The spectral concentration characteristic parameter is expressed as:

[0083]

[0084] Among them, reflects the degree of concentration of the signal energy in the spectrum. A larger value indicates that the signal energy is more concentrated at a certain specific frequency, while a smaller value indicates that the signal energy is more dispersed. In order to further extract the modulation characteristics of the signal, calculate the main and secondary peak energy difference characteristic parameter, that is, the amplitude difference between the main peak and the secondary peak in the spectrum, and perform normalization processing. The main peak and the secondary peak are identified by using the local maximum search algorithm, where:

[0085]

[0086]

[0087] Among them, is the index where the main peak is located, and the secondary peak is the second largest peak after removing the main peak. The main and secondary peak energy difference characteristic parameter is calculated as

[0088]

[0089] This parameter is used to measure the modulation characteristics of the signal. A larger value indicates that the signal has an obvious single main peak, while a smaller value means that the signal has a strong modulation component or interference component. After extracting the spectral energy ratio and spectral concentration and the energy difference between the main and secondary peaks Construct a three-dimensional feature vector, that is,

[0090]

[0091] The three-dimensional feature vector is input into the support vector machine (SVM) classifier based on Gaussian kernel function for signal classification. The decision boundary of the SVM classifier is solved by the following optimization problem:

[0092]

[0093] in, is the classification hyperplane normal vector, is the bias term, is the regularization parameter, It is The class labels of samples, is the input feature vector. Using the Gaussian kernel function:

[0094]

[0095] Map the data to a high-dimensional space to enhance the classification capability. Based on the classification results, extract the data segments determined to be mixed signals from the preprocessed signals, and record their corresponding time positions and frequency features to construct a complete mixed signal data set.

[0096] In a specific embodiment, the process of executing step S3 may specifically include the following steps:

[0097] The mixed signal is constructed with 1 / 3 of the number of sampling points as the number of rows to obtain the signal Hankel matrix;

[0098] Perform singular value decomposition on the signal Hankel matrix to obtain the singular value sequence and its corresponding left and right singular vectors;

[0099] Arrange the singular value sequence in descending order according to the energy contribution rate, extract the singular values ​​and their corresponding singular vectors with a cumulative energy contribution rate of 95%, and obtain the signal subspace features;

[0100] Reconstruct the signal according to the signal subspace characteristics to obtain a preliminary reconstructed signal;

[0101] Perform correlation analysis on the preliminary reconstructed signal and the interference feature library, and divide the signal components based on the correlation coefficient threshold to obtain the target signal component and the interference signal component;

[0102] Perform time-frequency analysis on the target signal components and remove the interference signal components to obtain a pure target signal.

[0103] Specifically, the mixed signal is used to construct a Hankel matrix with the number of rows being 1 / 3 of the number of sampling points, forming a structured representation of the signal, so as to use matrix decomposition methods for noise reduction and signal decomposition in the subsequent process. The Hankel matrix is a special structured matrix, where the elements on each anti-diagonal are the same. Therefore, for a discrete mixed signal with a length of a Hankel matrix of the following form is constructed: For

[0104]

[0105] where the number of rows of the matrix is set to and the number of columns is to ensure that the time series information of the signal is better retained in the matrix. The constructed Hankel matrix can provide redundant information in signal processing and help remove random noise and interference signals. The singular value decomposition (SVD) is performed on the Hankel matrix of the signal to extract the main components of the signal and remove noise. The mathematical expression form of the singular value decomposition is as follows:

[0106]

[0107] where is the left singular vector matrix, is the right singular vector matrix, and is the diagonal matrix, where the diagonal elements are the singular values of the Hankel matrix. These singular values represent the energy distribution of the signal in different modes. Larger singular values correspond to the main components of the signal, while smaller singular values usually correspond to noise and interference. The singular value sequence is sorted in descending order according to the energy contribution rate, and the cumulative energy contribution rate is calculated to select the most important signal components. The calculation method of the energy contribution rate is:

[0108]

[0109] where represents the cumulative energy proportion of the first singular values. The goal is to find the smallest such that:

[0110]

[0111] That is, to ensure that the total energy contributed by the selected singular values reaches at least 95%. The corresponding singular vectors are extracted from and to construct the signal subspace features, so as to retain the main components of the signal and eliminate noise. After obtaining the signal subspace features, the signal is reconstructed to obtain the preliminary reconstructed signal. The reconstruction process is expressed as:

[0112]

[0113] Among them, and only contain the first singular vectors, while only retains the first singular values. Perform a correlation analysis on the preliminary reconstructed signal and the interference feature library to distinguish the target signal components and the interference signal components. Let the reference signal in the interference feature library be where represents different types of interference signals, and calculate the correlation coefficient between the reconstructed signal and each interference signal:

[0114]

[0115] Among them, reflects the correlation between the current signal and a specific interference signal. If exceeds the set correlation coefficient threshold , it is considered that this signal component is mainly affected by this interference signal. Therefore, this interference signal component needs to be removed to obtain the division result of the target signal component and the interference signal component. For the extracted target signal components, perform time-frequency analysis to ensure that the finally obtained signal is pure and reliable. The time-frequency analysis uses the short-time Fourier transform (STFT) method, and its mathematical expression is as follows:

[0116]

[0117] Among them, is the time-frequency spectrum at time and frequency , is the window function, is the frame shift parameter. Analyze the frequency components of the signal at different time positions through STFT to identify the remaining interference components and further remove them to obtain a pure target signal.

[0118] In a specific embodiment, the process of executing step S4 may specifically include the following steps:

[0119] Perform N sub-band divisions on the pure target signal to obtain a sub-band signal sequence;

[0120] Calculate the autocorrelation matrix and cross-correlation vector for each sub-band signal in the sub-band signal sequence to obtain correlation feature quantities;

[0121] Calculate the signal-to-noise ratio of each sub-band based on the correlation feature quantities and dynamically generate a regularization factor according to the signal-to-noise ratio value to obtain optimization parameters;

[0122] Substitute the optimized parameters into the Wiener-Hopf equation and solve it using the conjugate gradient method to obtain an iterative calculation sequence;

[0123] Perform residual calculation on the iterative calculation sequence, stop the iteration when the residual is less than the first target value, and obtain the sub-band calibration parameters;

[0124] Perform local iterative optimization on the frequency points in the sub-band calibration parameters whose errors exceed the threshold. The optimization step size decreases in accordance with 0.01 times the initial step size, and merge the optimization results of each sub-band to obtain the initial calibration parameters.

[0125] Specifically, perform sub-band divisions on the pure target signal to obtain a sub-band signal sequence. Analyze the characteristics of the signal independently in different frequency bands, so that the calibration process is optimized according to the characteristics of different frequency bands. Assume that the total bandwidth of the pure target signal is , then the bandwidth of each sub-band is defined as:

[0126]

[0127] where represents the bandwidth of a single sub-band, represents the total number of sub-bands. On this basis, decompose the target signal into sub-band signals through a band-pass filter, and each sub-band signal only contains the frequency components corresponding to the corresponding frequency band. Calculate the autocorrelation matrix and cross-correlation vector for each sub-band signal in the sub-band signal sequence to extract relevant characteristic quantities. The autocorrelation matrix is used to describe the correlation of the signal at different time delays, and its calculation method is:

[0128]

[0129] where represents the sub-band signal, represents its conjugate transpose, represents the mathematical expectation operation. The cross-correlation vector is then used to measure the correlation between the sub-band signal and the desired signal , and is defined as follows:

[0130]

[0131] where is the reference signal corresponding to this sub-band. Calculate the signal-to-noise ratio of each sub-band based on the relevant characteristic quantities, and dynamically generate a regularization factor according to the signal-to-noise ratio value to obtain the optimized parameters. The calculation method of the signal-to-noise ratio is:

[0132]

[0133] Among them, is the noise component in the sub-band signal, while is the signal component. According to the calculated , an adaptive regularization strategy is adopted to set the regularization factor , for example, using the inverse relationship:

[0134]

[0135] Among them, is a tuning parameter, is a small constant to avoid the denominator being zero. The role of the regularization factor is to prevent numerical instability problems during the matrix solution process and improve the convergence of the optimization process. Substitute the optimization parameters into the Wiener-Hopf equation and use the conjugate gradient method for solution to obtain the iterative calculation sequence. The basic form of the Wiener-Hopf equation is as follows:

[0136]

[0137] Among them, is the sub-band calibration parameter to be solved. Since the direct solution has a high computational complexity, the conjugate gradient method is used for iterative solution. The iterative update formula of the conjugate gradient method is

[0138]

[0139]

[0140] Among them, is the search direction, is the residual, and are the step size parameters. Initially, let:

[0141]

[0142]

[0143] And the step size parameter update method is as follows:

[0144]

[0145]

[0146] The advantage of this algorithm is that it avoids the computational complexity of directly solving the matrix inverse operation, enabling the calibration parameters to converge within a limited number of steps. During the conjugate gradient method iteration, continuously monitor the residual and perform calculations. When the residual satisfies:

[0147]

[0148] When it reaches this point, the iteration stops and the sub-band calibration parameters are obtained. Among them, is the first target value, which is used to judge the convergence. However, in some cases, there are still frequency points with relatively large errors in some sub-band calibration parameters, and local iteration optimization is performed on these frequency points. For the frequency points whose errors exceed the threshold, let their errors be If:

[0149]

[0150] Then the calibration parameters need to be further adjusted. The optimization step size Adopts a gradually decreasing method, according to:

[0151]

[0152] Among them, Is the initial step size, and the optimization fineness is ensured by continuously reducing the step size. The local optimization process is corrected using the gradient descent method, that is

[0153]

[0154] Among them, Is the gradient of the error with respect to the parameter. After the optimization of all sub-band calibration parameters is completed, the optimization results of each sub-band are merged to form the complete initial calibration parameters. The merging method adopts a weighted average strategy, that is:

[0155]

[0156] Among them, Is the normalized weight coefficient, set as:

[0157]

[0158] To ensure that the sub-bands with high signal-to-noise ratio occupy a greater weight in the final calibration parameters.

[0159] Among them, the signal-to-noise ratio of each sub-band is calculated based on the relevant characteristic quantities, and the regularization factor is dynamically generated according to the signal-to-noise ratio value to obtain the optimization parameters, including: performing eigenvalue decomposition on the autocorrelation matrix R in the relevant characteristic quantities to obtain the eigenvalue sequence and eigenvectors; calculating the ratio of the main component to the secondary component based on the eigenvalue sequence to obtain the sub-band signal-to-noise ratio; performing logarithmic transformation and normalization processing on the sub-band signal-to-noise ratio to obtain the normalized signal-to-noise ratio; constructing an exponential decay function according to the normalized signal-to-noise ratio and substituting the signal-to-noise ratio into this function to obtain the initial regularization factor; performing iterative update on the initial regularization factor, and the update step size for each iteration is 0.1 times the current value until the difference between two adjacent iterations is less than 10 -8, the final regularization factor is obtained; the final regularization factor is combined with relevant characteristic quantities to obtain the optimization parameters.

[0160] Among them, substituting the optimization parameters into the Wiener-Hopf equation and using the conjugate gradient method to solve it, an iterative calculation sequence is obtained, including: substituting the optimization parameters into the Wiener-Hopf equation Rx = P to construct the initial system equation, where R is the autocorrelation matrix and P is the cross-correlation vector, to obtain the system of equations to be solved; calculating the initial residual vector r0 and the initial search direction p0 for the system of equations to be solved to obtain the initial iteration values; calculating the step size factor α based on the initial iteration values and updating the current solution vector xk and the residual vector rk to obtain the updated parameters; calculating the conjugate coefficient β for the updated parameters and updating the search direction pk to obtain the new iteration direction; calculating the residual norm based on the new iteration direction to determine whether it is less than 10 -6 , to obtain the convergence criterion; determine whether to continue the iteration according to the convergence criterion, and when the convergence condition is met, output the current solution sequence as the iterative calculation sequence.

[0161] In a specific embodiment, the process of executing step S5 may specifically include the following steps:

[0162] Divide the frequency band of the signal to be measured into M sub-bands, and set 3 calibration frequency points in each sub-band according to the logarithmic relationship to obtain the calibration frequency point distribution;

[0163] Perform interpolation processing on the initial calibration parameters based on the calibration frequency point distribution to obtain the frequency response function of each sub-band, and perform piecewise cubic spline interpolation on the amplitude and phase of the frequency response function to obtain the frequency response curve;

[0164] Set an overlapping bandwidth between adjacent sub-bands, and perform linear weighted averaging on the frequency response within the overlapping bandwidth to obtain the calibration value of the transition region;

[0165] Calculate the calibration coefficient of each frequency point based on the frequency response curve and the calibration value of the transition region to obtain the compensation sequence;

[0166] Iteratively optimize the compensation sequence until the change rate of the results of adjacent iteration cycles is less than the target percentage value to obtain the compensation calibration coefficient.

[0167] Specifically, divide the frequency band of the signal to be measured into sub-bands, and set 3 calibration frequency points in each sub-band according to the logarithmic relationship to ensure the reasonable distribution of the calibration frequency points in the entire frequency range. The calibration frequency point setting method of the logarithmic relationship is expressed as:

[0168]

[0169] Among them, represents the th One calibration frequency point, representing three calibration frequency points, is the minimum frequency of the entire measurement range, and is the maximum frequency. The calibrated frequency points thus divided can provide a higher frequency resolution at lower frequencies and a more reasonable coverage in the high-frequency part. After setting the calibration frequency points, interpolation processing is performed on the initial calibration parameters based on the calibration frequency points to obtain the complete frequency response function. Since the initial calibration parameters are discrete, an interpolation method is used to extend them to all calibration frequency points. Suppose the initial calibration parameter is , where

[0170]

[0171] where, represents the interpolation function within the sub-band , are the interpolation coefficients to be solved, is the center frequency of the current sub-band. By constructing continuity and smoothness constraint equations, these interpolation coefficients are solved so that the obtained frequency response curve has high smoothness and accuracy within different sub-bands. After obtaining the complete frequency response curve, in order to ensure smooth transition of the calibration parameters between adjacent sub-bands, an overlapping bandwidth is set between adjacent sub-bands, and linear weighted averaging is performed on the frequency response within this overlapping region to obtain the calibration value of the transition region. Let the frequency responses of two adjacent sub-bands be and , respectively, then the transition calibration value within the overlapping bandwidth is expressed as:

[0172]

[0173] where the weight functions and are defined in a linearly increasing and decreasing manner, that is,

[0174]

[0175] where, is the starting frequency of the overlapping bandwidth. Through linear weighted averaging, it is ensured that the frequency responses of adjacent sub-bands smoothly transition within the overlapping region, avoiding frequency discontinuity problems caused by sub-band division. After obtaining the complete frequency response curve and the calibration value of the transition region, the calibration coefficients at each frequency point are calculated to obtain the compensation sequence. The compensation coefficient is calculated and corrected based on the measurement error of the reference signal, that is:

[0176]

[0177] Among them, is the ideal frequency response function, while is the currently calculated frequency response. This compensation coefficient is used to adjust the non-ideal characteristics of the measurement system to make the measurement results more accurate. In order to optimize the compensation sequence, an iterative optimization method is adopted so that the calibration coefficient can gradually approach the optimal value. Let the compensation coefficient of the th iteration be , then the following update formula is used for iterative optimization:

[0178]

[0179] Among them, is the learning rate, which takes a smaller value to ensure convergence stability. After each iteration, calculate the change rate of the results of adjacent iteration periods to judge whether the optimization converges. The change rate is expressed as:

[0180]

[0181] If is less than the set target percentage value , that is:

[0182]

[0183] then terminate the iteration and obtain the final compensation calibration coefficient.

[0184] Among them, after obtaining the compensation calibration coefficient and before verifying with the standard signal, it also includes: setting a time delay threshold based on the compensation calibration coefficient, performing periodic sampling on the calibration coefficients of each frequency band to obtain a sampling sequence; performing a steady-state analysis on the sampling sequence, calculating the oscillation characteristics of the calibration coefficients of each frequency band to obtain oscillation parameters; establishing a coupling relationship model between frequency bands according to the oscillation parameters, and extracting key coupling characteristics to obtain a coupling characteristic sequence; performing time-domain decomposition on the coupling characteristic sequence to determine the phase difference and delay characteristics between each frequency band to obtain a delay characteristic quantity; constructing an adaptive adjustment model of the compensation calibration coefficient based on the delay characteristic quantity to obtain a dynamic compensation function; performing periodic correction on the compensation calibration coefficient according to the dynamic compensation function to obtain a correction coefficient sequence; performing stability evaluation based on the correction coefficient sequence, and calculating the system response characteristics to obtain response parameters; performing optimization screening on the response parameters, and when the system reaches a stable oscillation state, output the compensation calibration coefficient.

[0185] In a specific embodiment, the process of executing step S6 may specifically include the following steps:

[0186] Generate a calibration signal with a nominal frequency interval of 20 kHz using a standard signal source, covering the entire frequency band to be measured, and obtain a standard calibration signal;

[0187] Compare the frequency and amplitude of the standard calibration signal with the compensation calibration coefficient to obtain the measurement error at each frequency point;

[0188] Generate an error distribution histogram based on the measurement errors at each frequency point, and calculate the root mean square error and maximum error of each frequency band to obtain the error statistical result;

[0189] Compare the root mean square error in the error statistical result with the second target value to obtain an optimization determination result;

[0190] Based on the optimization determination result, mark the frequency bands where the root mean square error exceeds the second target value, and record the key parameters of the calibration process to obtain a frequency band optimization mark;

[0191] According to the frequency band optimization mark, return to step S4, and repeat steps S4 - S6 to optimize the calibration parameters of the marked frequency bands until the root mean square error is less than the second target value to obtain the target calibration result.

[0192] Specifically, use a standard signal source to generate a calibration signal with a nominal frequency interval of 20 kHz, and ensure that the signal can cover the entire frequency band to be measured, so as to verify the calibration accuracy within the entire measurement range. The frequency distribution of the standard calibration signal is expressed as:

[0193]

[0194] Among them, represents the frequency of the th calibration frequency point, is the lowest frequency, 20 kHz is the frequency interval, is the calibration frequency point index, and the maximum depends on the highest frequency of the measurement system, that is:

[0195]

[0196] This means that the entire measurement frequency band is equally divided to ensure that each frequency point can be accurately covered by the standard signal and provide high-precision reference data. Compare the frequency and amplitude of the standard calibration signal with the compensation calibration coefficient to obtain the measurement error at each frequency point. Let the response of the measurement system at frequency be , and the ideal response of the standard calibration signal be , then the measurement error at this frequency point is defined as:

[0197]

[0198] Meanwhile, the normalized measurement error is defined as:

[0199]

[0200] where represents the absolute error of the measurement system, while represents the relative error. The latter representation more intuitively reflects the relative magnitude of the errors at different frequency points. After obtaining the measurement errors at all frequency points, statistical analysis is performed on the error data. An error distribution histogram is generated based on the error values at all frequency points to observe the overall distribution of the errors. The construction method of the error distribution histogram is to divide the measurement error into several error intervals and count the number of frequency points in each interval to obtain the probability distribution of the errors. Calculate the root mean square error (RMSE) and the maximum error for each sub-band to quantify the overall characteristics of the errors. Among them, the formula for calculating the root mean square error is:

[0201]

[0202] where represents the root mean square error of the th sub-band, is the number of calibration frequency points in this sub-band, is the set of all frequency points included in this sub-band, and the maximum error is defined as:

[0203]

[0204] where represents the maximum measurement error within this sub-band. Through these error statistical indicators, evaluate the effectiveness of the current compensation calibration coefficient and determine whether further optimization is needed. After obtaining the error statistical results, compare the root mean square error of each sub-band with the second target value to determine whether the current calibration accuracy meets the requirements. If the root mean square error of a certain sub-band is greater than the second target value , that is:

[0205]

[0206] it indicates that there are still relatively large errors in the compensation calibration of this sub-band and further optimization is needed. Otherwise, the calibration parameters of this sub-band can be accepted and no further optimization is required. For the sub-bands with root mean square error exceeding the second target value, mark them based on the optimization determination results and record the key parameters of this sub-band during the calibration process for use in subsequent optimization processes. The key parameters include: the initial calibration parameters of this sub-band, the current error statistical value , the error convergence trend and signal-to-noise ratio information in the previous rounds of optimization. Based on this recorded information, more targeted optimization is carried out on those frequency points with larger errors. According to the frequency band optimization flag, return to step S4, and repeat steps S4 - S6 to optimize the calibration parameters of the marked frequency band. Recalculate the calibration parameters for the sub-bands with larger root mean square errors and adjust them through the least mean square error optimization method. During the optimization process, an adaptive step size strategy is adopted to avoid oscillation caused by too large an optimization step size or slow convergence caused by too small an optimization step size. The step size adjustment strategy is defined as:

[0207]

[0208] where is the optimization step size of this sub-band, is the initial step size. In this way, larger step sizes are used for the frequency bands with larger errors to accelerate optimization, while smaller step sizes are used for the frequency bands with smaller errors to maintain stable convergence. After each optimization iteration, recalculate the root mean square error and check whether it satisfies:

[0209]

[0210] If it is satisfied, the optimization process of this sub-band ends; otherwise, continue the optimization until the error converges within the target range.

[0211] The multi-band radio metrology calibration method in the embodiment of the present invention has been described above. Next, the multi-band radio metrology calibration system in the embodiment of the present invention will be described. Please refer to Figure 2 , an embodiment of the multi-band radio metrology calibration system in the embodiment of the present invention includes:

[0212] A sampling module, configured to sample and preprocess the multi-band radio signal to obtain a preprocessed signal;

[0213] A classification module, configured to perform a fast Fourier transform and signal classification on the preprocessed signal to obtain a mixed signal;

[0214] A reconstruction module, configured to construct the mixed signal into a Hankel matrix and perform signal reconstruction in combination with a correlation coefficient threshold to obtain a pure target signal;

[0215] A dynamic sub-band division module, configured to perform dynamic sub-band division on the pure target signal, introduce a regularization factor adaptive to the signal-to-noise ratio in each sub-band, and iteratively calculate the least mean square error until the residual is less than a first target value to obtain initial calibration parameters;

[0216] A linear weight averaging module, configured to establish a logarithmic interval frequency point distribution based on the initial calibration parameters and perform linear weight averaging within the overlapping bandwidth of adjacent sub-bands to obtain a compensated calibration coefficient;

[0217] A verification module is used to verify the compensation calibration coefficients using a standard calibration signal. When the root mean square error exceeds the second target value, the steps from the dynamic sub-band division module to the verification module are repeatedly executed until the accuracy requirement is met, and a target calibration result is obtained.

[0218] Through the collaborative cooperation of the above-mentioned various components, by adopting dynamic sampling and preprocessing strategies, combining the Hanning window function and the band-pass filter bank, the quality of signal acquisition is effectively improved. Based on the signal classification method of the fast Fourier transform and the support vector machine classifier, the effective components and interference components in the mixed signal are accurately identified, significantly improving the accuracy of signal classification; by using Hankel matrix decomposition and singular value reconstruction technology, combined with the judgment of the correlation coefficient threshold, the effective suppression of interference signals is achieved, ensuring the purity of the target signal; by introducing dynamic sub-band division and adaptive regularization factors, and iteratively optimizing the least mean square error, the computational complexity is effectively reduced, and the calibration efficiency is improved; through the logarithmic-spaced frequency point distribution and the linear weight averaging strategy, the smooth transition between adjacent sub-frequency bands is achieved, ensuring the continuity and stability of the calibration result; a complete calibration verification and optimization mechanism is established, and through root mean square error control and iterative optimization, the reliability and accuracy of the calibration result are ensured, significantly improving the overall calibration accuracy.

[0219] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0220] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0221] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the present invention in each embodiment.

Claims

1. A multi-band radio measurement calibration method, characterized in that: The method comprises: S1: sampling and preprocessing the multi-band radio signal to obtain a preprocessed signal; S2: performing fast Fourier transform and signal classification on the preprocessed signal to obtain a mixed signal; S3: constructing the mixed signal into a Hankel matrix, and reconstructing the signal in combination with a correlation coefficient threshold to obtain a pure target signal; S4: dynamically divide the pure target signal into sub-bands, introduce a signal-to-noise ratio adaptive regularization factor in each sub-band, iteratively calculate the minimum mean square error and calculate the error residual in the iterative process until the residual is less than the first target value, and obtain the initial calibration parameters; S5: establishing a logarithmically spaced frequency point distribution based on the initial calibration parameters, and performing linear weighted averaging within the overlapping bandwidth of adjacent sub-frequency bands to obtain a compensation calibration coefficient; S6: Use a standard calibration signal to verify the compensation calibration coefficient. When the root mean square error exceeds the second target value, repeat steps S4-S6 until the accuracy requirement is met to obtain a target calibration result.

2. The multi-band radio measurement calibration method according to claim 1, characterized in that: The sampling and preprocessing of the multi-band radio signal to obtain the preprocessed signal includes: Setting a sampling frequency according to the highest frequency of the signal to be tested, and digitally sampling the multi-band radio signal based on the sampling frequency to obtain a digital sampling signal; Based on a preset overlap rate and sampling points, the digital sampling signal is subjected to frame processing to obtain a frame signal; Performing windowing processing using a Hanning window function on the framed signal and quantizing it through an analog-to-digital converter to obtain a quantized signal, and performing direct current offset correction on the quantized signal to obtain a corrected signal; The correction signal is filtered through five groups of bandpass filters to obtain a filtered signal, and the filtered signal is stored in a signal buffer to obtain a preprocessed signal, wherein the center frequency of each bandpass filter corresponds to each frequency band of the signal to be measured.

3. The multi-band radio measurement calibration method according to claim 1, characterized in that: The step of performing fast Fourier transform and signal classification on the preprocessed signal to obtain a mixed signal comprises: Performing a fast Fourier transform on the preprocessed signal to obtain a frequency domain signal spectrum; Calculating the ratio of the target frequency band energy to the full frequency band energy in the frequency domain signal spectrum to obtain a spectrum energy ratio characteristic parameter; Calculating the ratio of the spectrum centroid to the spectrum bandwidth based on the frequency domain signal spectrum to obtain a spectrum concentration characteristic parameter; Extracting the energy difference between the main peak and the secondary peak from the frequency domain signal spectrum and performing normalization processing to obtain a main-secondary peak energy difference characteristic parameter; Constructing a three-dimensional feature vector based on the spectrum energy ratio feature parameter, the spectrum concentration feature parameter and the primary and secondary peak energy difference feature parameter, and inputting the three-dimensional feature vector into a support vector machine classifier based on a Gaussian kernel function to obtain a classification result; Based on the classification result, the data segments determined to be mixed signals and their corresponding time positions and frequency characteristics are extracted from the preprocessed signals to obtain mixed signals.

4. The multi-band radio measurement calibration method according to claim 1, characterized in that: The mixed signal is constructed as a Hankel matrix, and the signal is reconstructed in combination with a correlation coefficient threshold to obtain a pure target signal, including: Constructing a Hankel matrix of the mixed signal according to 1 / 3 of the number of sampling points as the number of rows to obtain a signal Hankel matrix; Performing singular value decomposition on the signal Hankel matrix to obtain a singular value sequence and its corresponding left and right singular vectors; Arrange the singular value sequence in descending order according to the energy contribution rate, extract the singular values ​​and their corresponding singular vectors whose cumulative energy contribution rate reaches 95%, and obtain the signal subspace feature; Reconstructing the signal according to the signal subspace characteristics to obtain a preliminary reconstructed signal; Performing correlation analysis on the preliminary reconstructed signal and the interference feature library, and dividing the signal components based on a correlation coefficient threshold to obtain a target signal component and an interference signal component; The target signal components are subjected to time-frequency analysis and interference signal components are eliminated to obtain a pure target signal.

5. The multi-band radio measurement calibration method according to claim 1, characterized in that: The method of dynamically dividing the pure target signal into sub-bands, introducing a signal-to-noise ratio adaptive regularization factor in each sub-band, iteratively calculating the minimum mean square error and calculating the error residual in the iterative process until the residual is less than the first target value, and obtaining the initial calibration parameters includes: Dividing the pure target signal into N sub-bands to obtain a sub-band signal sequence; Calculating an autocorrelation matrix and a cross-correlation vector for each subband signal in the subband signal sequence to obtain a related feature quantity; Calculating the signal-to-noise ratio of each subband based on the relevant feature quantity, and dynamically generating a regularization factor according to the signal-to-noise ratio value to obtain an optimization parameter; Substituting the optimization parameters into the Wiener-Hopper equation to perform minimum mean square error calculation, and solving it using the conjugate gradient method to obtain an iterative calculation sequence; Performing residual calculation on the iterative calculation sequence, stopping iteration when the residual is less than a first target value, and obtaining a subband calibration parameter; The frequency points in the sub-band calibration parameters where the errors exceed the threshold are locally iteratively optimized, the optimization step size is decreased by 0.01 times of the initial step size, and the optimization results of each sub-band are combined to obtain the initial calibration parameters.

6. The multi-band radio measurement calibration method according to claim 1, characterized in that: The step of establishing a logarithmically spaced frequency point distribution based on the initial calibration parameters and performing linear weighted averaging within the overlapping bandwidth of adjacent sub-bands to obtain a compensation calibration coefficient includes: The frequency band of the signal to be measured is divided into M sub-frequency bands, and three calibration frequency points are set in each sub-frequency band according to a logarithmic relationship to obtain a calibration frequency point distribution; Performing interpolation processing on the initial calibration parameters based on the calibration frequency point distribution to obtain a frequency response function of each sub-band, and performing piecewise cubic spline interpolation on the amplitude and phase of the frequency response function to obtain a frequency response curve; An overlapping bandwidth is set between adjacent sub-bands, and a linear weighted average is performed on the frequency response within the overlapping bandwidth to obtain a transition region calibration value; Calculate the calibration coefficient of each frequency point based on the frequency response curve and the transition region calibration value to obtain a compensation sequence; The compensation sequence is iteratively optimized until the result change rate of adjacent iteration cycles is less than the target percentage value, thereby obtaining the compensation calibration coefficient.

7. The multi-band radio measurement calibration method according to claim 1, characterized in that: The compensation calibration coefficient is verified by using a standard calibration signal, and when the root mean square error exceeds a second target value, steps S4-S6 are repeated until the accuracy requirement is met to obtain a target calibration result, including: A standard signal source is used to generate a calibration signal with a nominal frequency interval of 20kHz, covering all the frequency bands to be tested, to obtain a standard calibration signal; Comparing the standard calibration signal with the compensation calibration coefficient in frequency and amplitude to obtain the measurement error of each frequency point; Generate an error distribution histogram based on the measurement errors of each frequency point, and calculate the root mean square error and maximum error of each frequency band to obtain error statistics results; Comparing the root mean square error in the error statistical result with the second target value to obtain an optimization determination result; Mark the frequency band whose root mean square error exceeds the second target value based on the optimization determination result, and record the key parameters of the calibration process to obtain a frequency band optimization mark; Return to step S4 according to the frequency band optimization mark, and repeat steps S4-S6 to optimize the calibration parameters of the marked frequency band until the root mean square error is less than the second target value, thereby obtaining a target calibration result.

8. A multi-band radio metrology calibration system, characterized in that: Used to perform the multi-band radio metrology calibration method according to any one of claims 1 to 7, the multi-band radio metrology calibration system comprising: A sampling module, used for sampling and preprocessing multi-band radio signals to obtain preprocessed signals; A classification module, used for performing fast Fourier transform and signal classification on the preprocessed signal to obtain a mixed signal; A reconstruction module, used to construct the mixed signal into a Hankel matrix, and perform signal reconstruction in combination with a correlation coefficient threshold to obtain a pure target signal; A dynamic sub-band division module is used to dynamically divide the pure target signal into sub-bands, introduce a signal-to-noise ratio adaptive regularization factor in each sub-band, iteratively calculate the minimum mean square error and calculate the error residual in the iterative process until the residual is less than the first target value, and obtain the initial calibration parameters; A linear weighted averaging module, used to establish a logarithmically spaced frequency point distribution based on the initial calibration parameters, and perform linear weighted averaging within the overlapping bandwidth of adjacent sub-bands to obtain a compensation calibration coefficient; The verification module is used to verify the compensation calibration coefficient using a standard calibration signal. When the root mean square error exceeds the second target value, the steps from the dynamic sub-band division module to the verification module are repeatedly executed until the accuracy requirement is met to obtain a target calibration result.

Images