Method and system for optimizing multi-frequency antenna signals based on data processing
By adaptively determining the filter order and constructing a multi-objective cost function, iteratively solving for the optimal zeros and poles, and generating a discrete-domain digital transfer function, the problem of fixed filter parameters is solved, achieving efficient interference suppression and phase fidelity for multi-frequency antenna signals, thus improving communication quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHAANXI TURN ELECTRONICS TECH
- Filing Date
- 2026-06-11
- Publication Date
- 2026-07-10
AI Technical Summary
In existing technologies, filter parameters are fixed and lack adaptive capabilities, making it impossible to effectively balance interference suppression and phase fidelity, resulting in poor optimization of multi-frequency antenna signals.
By acquiring the digital intermediate frequency signal characteristic data of a multi-frequency antenna, the filter order is adaptively determined, a multi-objective cost function is constructed, the optimal zeros and poles are iteratively solved, a discrete-domain digital transfer function is generated, and signal filtering is performed.
While effectively filtering out interference from adjacent frequency bands, it ensures high-fidelity transmission of passband signals, significantly improving the communication quality of multi-frequency antenna signals.
Smart Images
Figure CN122372011A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of antenna signal processing technology, specifically relating to a method and system for optimizing multi-frequency antenna signals based on data processing. Background Technology
[0002] The signals received by the multi-frequency antenna are located in a dense and variable spectral space. The multi-frequency signal components within the target frequency band exhibit frequency cross-correlation, leading to aliasing and amplitude / phase distortion in key frequency ranges within the passband. Spectral congestion causes high-power random interference to permeate adjacent frequency bands, leaking strong interference into the target frequency band, compressing the signal range and reducing the signal-to-noise ratio. Digital signal processing techniques are combined, and filter models are used to select and optimize the digital intermediate frequency signal. Among these, the Chebyshev filter exhibits an extremely fast roll-off rate and a steep attenuation characteristic in the transition band.
[0003] Chinese invention patent CN115425998B discloses an adaptive method for a multi-band, multi-stage anti-interference intelligent ultra-wideband antenna, comprising: acquiring the frequency band of the interference source, which includes single, dual, and triple frequency bands; when the interference source is a single frequency band, activating a single-band first-order interference filtering and suppression scheme; when the single-band first-order interference filtering and suppression scheme fails to meet the standard, activating a single-band second-order interference filtering and suppression scheme; when the single-band second-order interference filtering and suppression scheme fails to meet the standard, activating a single-band third-order interference filtering and suppression scheme; when the interference source is a dual frequency band, activating a dual-band first-order interference filtering and suppression scheme; when the dual-band first-order interference filtering and suppression scheme fails to meet the standard, activating a dual-band second-order interference filtering and suppression scheme.
[0004] However, traditional methods typically employ fixed filter orders and standard construction parameters, failing to adjust the order based on the spectral entropy of the target frequency band's power spectral density. They also lack the ability to identify key frequency ranges and high-interference frequency points and perform frequency pre-distortion processing by incorporating frequency cross-correlation matrices and adjacent-band interference distribution. Chebyshev filters focus solely on amplitude-frequency attenuation as their pole and zero points, lacking a multi-objective joint optimization mechanism that integrates passband ripple, stopband attenuation, and group delay flatness, easily leading to phase distortion in multi-frequency signals after filtering. Existing filter solution processes lack customized penalty weights for passband ripple deviation and insufficient attenuation at specific high-interference frequency points. This results in the generated analog transfer function and the discretized digital transfer function failing to achieve an optimal balance between passband flatness and deep stopband attenuation, reducing the optimization effect of multi-frequency antenna signals and failing to meet the high-quality signal transmission requirements of modern communication systems. Summary of the Invention
[0005] This invention provides a multi-frequency antenna signal optimization method and system based on data processing to solve the technical problems in the prior art, such as fixed filter parameters, lack of adaptive capability, difficulty in balancing interference suppression and phase fidelity, and poor signal optimization effect.
[0006] In a first aspect, the present invention provides a multi-frequency antenna signal optimization method based on data processing, comprising the following steps: S1, acquire the digital intermediate frequency signal of the multi-frequency antenna, calculate the target frequency band power spectral density and frequency cross-correlation matrix of the digital intermediate frequency signal, and calculate the interference power distribution data of the adjacent frequency band; S2, determine the filter order of the discrete domain digital transfer function based on the spectral entropy value of the target frequency band power spectral density, identify the key frequency range in the passband according to the frequency cross-correlation matrix, set strict passband ripple constraints, set high attenuation target values for high interference frequency points in the stopband according to the interference power distribution data of adjacent frequency bands, and perform frequency pre-distortion processing on the key frequency range and high interference frequency points. S3. Establish a multi-objective cost function with the positions of complex plane poles and transmission zeros corresponding to the filter order as variables. In the multi-objective cost function, the passband ripple parameter, group delay flatness parameter, and stopband attenuation parameter are integrated. Penalty weights are applied to the ripple deviation in the key frequency range and the insufficient attenuation at high interference frequency points. The poles of the standard Chebyshev filter are used as the initial pole values and the initial transmission zero coordinates are configured. The multi-objective cost function is solved iteratively to obtain the optimal set of complex plane poles and the optimal set of transmission zeros. S4 constructs a continuous-domain analog transfer function from the optimal set of complex plane poles and the optimal set of transmission zeros. The continuous-domain analog transfer function is then discretized into a discrete-domain digital transfer function. The discrete-domain digital transfer function is used to filter the digital intermediate frequency signal, and the optimized multi-frequency antenna signal is output.
[0007] Its effects are as follows: by acquiring the frequency domain characteristic data of the digital intermediate frequency signal of the multi-frequency antenna, the filter order, passband ripple constraint and stopband attenuation target are adaptively determined, a multi-objective cost function with multiple parameters is constructed and the optimal zeros and poles are iteratively solved, and finally a discrete domain digital transfer function is generated to complete the signal filtering. It can effectively filter out interference from adjacent frequency bands while ensuring high-fidelity transmission of the passband signal, significantly improving the overall signal communication quality of the multi-frequency antenna, and solving the technical problems of fixed filter parameters and difficulty in balancing interference suppression and phase fidelity in traditional filters.
[0008] Furthermore, the filter order of the discrete-domain digital transfer function is determined based on the spectral entropy value of the target frequency band power spectral density, including: The target frequency band is divided into several equidistant frequency grids, the power spectral density amplitude in each frequency grid is obtained, and a discrete amplitude sequence is constructed. After normalizing the discrete amplitude sequence, the probability distribution is calculated, and the spectral entropy value of the target frequency band is calculated using the Shannon entropy formula. Calculate the reciprocal of the spectral entropy value, multiply the reciprocal by a preset scaling factor to amplify it, and round the amplified value up to obtain an integer that meets the signal complexity requirements. Use the integer as the filter order.
[0009] Its effect is as follows: by dividing the target frequency band into equidistant frequency grids, constructing discrete amplitude sequences based on power spectral density, calculating spectral entropy values through normalization and Shannon entropy, and then determining the filter order by converting the reciprocal of the spectral entropy value and rounding it up, it can accurately match the signal complexity to allocate the filter order, avoiding the waste of computing power caused by an excessively high order or the insufficient filtering accuracy caused by an excessively low order. It provides an order basis for subsequent filter design that is adapted to the signal characteristics, ensuring the efficiency and adaptability of filtering processing.
[0010] Furthermore, key frequency ranges within the passband are identified based on the frequency cross-correlation matrix, and strict passband ripple constraints are set, including: Extract frequency pairs with correlation coefficients greater than a preset threshold from the frequency cross-correlation matrix, and merge consecutive high-correlation frequency pairs with frequency spacing less than a preset interval to obtain at least one continuous frequency interval as the key frequency interval. Set the upper limit of passband ripple in the key frequency range as the first ripple tolerance value; The upper limit of the passband ripple in the frequency region outside the critical frequency range within the passband is set as the second ripple tolerance value, wherein the first ripple tolerance value is less than the second ripple tolerance value.
[0011] Its effects are as follows: by extracting high-correlation frequency pairs through the frequency cross-correlation matrix and merging them into key frequency intervals, and setting differentiated passband ripple tolerance values for key and non-key intervals, the core signal frequency band within the passband can be accurately locked and strict ripple constraints can be implemented. At the same time, the constraints of non-key frequency bands can be relaxed. While ensuring the flatness of the core signal amplitude and avoiding amplitude and phase distortion, the filter design difficulty is reduced, and signal fidelity and engineering feasibility are taken into account.
[0012] Furthermore, based on the interference power distribution data of adjacent frequency bands, high attenuation target values are set for high interference frequency points within the stopband, including: The amplitude of interference power is collected in adjacent frequency bands, and the average interference power of all collection points is calculated as the judgment threshold. Frequency points with interference power greater than the judgment threshold are extracted as high interference frequency points; A first stopband attenuation target value is set for high interference frequency points, and a second stopband attenuation target value is set for the remaining non-high interference frequency points within the stopband, wherein the attenuation degree of the first stopband attenuation target value is higher than the attenuation degree of the second stopband attenuation target value.
[0013] Its effects are as follows: by calculating the average value of interference power in adjacent frequency bands to determine the judgment threshold, accurately extracting high interference frequency points and setting differentiated stopband attenuation targets, it can focus limited filtering resources on strong interference frequency points to achieve deep attenuation, avoid blindly increasing the global attenuation index leading to filter order expansion, effectively suppress strong interference leakage in adjacent bands, improve the signal-to-noise ratio of the target signal, and adapt to the interference suppression requirements in a congested spectrum environment.
[0014] Furthermore, a multi-objective cost function is established with the positions of complex plane poles and transmission zeros corresponding to the filter order as variables. This includes: using the real and imaginary coordinates of the complex plane poles and transmission zeros corresponding to the filter order under conjugate symmetry constraints as optimization variable vectors; constructing a first cost term representing passband ripple error, a second cost term representing group delay flatness, and a third cost term representing stopband attenuation; assigning a ripple penalty weight higher than other passband frequencies to the key frequency range after frequency pre-distortion processing in the first cost term, and assigning an attenuation penalty weight higher than other stopband frequencies to the high interference frequency points after frequency pre-distortion processing in the third cost term; and then weighting and summing the first, second, and third cost terms after dimensionless processing to construct the multi-objective cost function.
[0015] Its effect is as follows: by setting the coordinates of zeros and poles as optimization variables, constructing a multi-objective cost term including passband ripple, group delay flatness, and stopband attenuation, applying differentiated penalty weights to key frequency ranges and high interference frequency points and performing dimensionless weighted summation, it is possible to achieve multi-objective collaborative optimization of passband fidelity, phase linearity, and stopband suppression, prioritizing the strict constraints of core frequency bands and strong interference frequency points, and ensuring that the optimization direction aligns with the core needs of signal processing.
[0016] Furthermore, using the poles of the standard Chebyshev filter as initial pole values and configuring initial transmission zero coordinates, the multi-objective cost function is iteratively solved to obtain the optimal set of complex plane poles and the optimal set of transmission zeros. This includes: calculating the coordinates of the complex plane poles of the standard Chebyshev filter at the filter order as initial pole values, configuring initial transmission zero coordinates, and using both types of coordinates as initial values for the optimization variable vector; using a numerical optimization algorithm to iteratively update the positions of the complex plane poles and transmission zeros, calculating the changes in the positions of the complex plane poles and transmission zeros, as well as the changes in the values of the multi-objective cost function in each iteration; stopping the iteration when the changes in the values of the multi-objective cost function are less than a preset convergence threshold or when the maximum number of iterations is reached, and outputting the current positions of the complex plane poles and transmission zeros, thus forming the optimal set of complex plane poles and the optimal set of transmission zeros.
[0017] Furthermore, a continuous-domain analog transfer function is constructed using the optimal set of complex plane poles and the optimal set of transmission zeros. This continuous-domain analog transfer function is then discretized into a discrete-domain digital transfer function. The discrete-domain digital transfer function is used to filter the digital intermediate frequency signal, outputting an optimized multi-frequency antenna signal, including: Calculate the polynomial coefficients of the continuous-domain simulated transfer function based on the optimal set of complex plane poles and the optimal set of transport zeros; By employing the bilinear transform method, the continuous-domain analog transfer function is mapped to the discrete domain, thus obtaining the digital filter coefficients of the discrete-domain digital transfer function. The digital intermediate frequency signal is used as the input sequence, and discrete filtering is performed using digital filter coefficients to generate and output the filtered and optimized multi-frequency antenna signal.
[0018] Furthermore, the method for acquiring the digital intermediate frequency signal of a multi-frequency antenna includes: performing a fixed sampling rate analog-to-digital conversion on the analog radio frequency signal received by the multi-frequency antenna using an analog-to-digital converter to obtain a discrete time series as the digital intermediate frequency signal.
[0019] Furthermore, the method for obtaining the spectral entropy value includes: summing and normalizing the distribution sequence of the target frequency band power spectral density to convert it into a probability distribution sequence, and using the scipy.stats.entropy function to calculate the Shannon information entropy of the probability distribution sequence as the spectral entropy value.
[0020] Secondly, the present invention provides a multi-frequency antenna signal optimization system based on data processing, including a memory and a processor. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned multi-frequency antenna signal optimization method based on data processing is implemented.
[0021] The beneficial effects are as follows: This invention obtains the target frequency band power spectral density, frequency cross-correlation matrix, and adjacent frequency band interference power distribution data of the digital intermediate frequency signal, thereby mastering the frequency domain characteristics of the signal and interference. It rationally determines the filter order using spectral entropy values, sets strict ripple constraints on key frequency ranges, and sets high attenuation targets for high-interference frequency points, combined with frequency pre-distortion processing. A multi-objective cost function integrating passband ripple, group delay flatness, and stopband attenuation parameters is constructed, and penalty weights are applied to ripple deviation in key ranges and insufficient attenuation at high-interference points. The optimal set of complex plane poles and transmission zeros is obtained through iterative solution. The generated discrete-domain digital transfer function can filter multi-frequency antenna signals, ensuring high-fidelity transmission of the passband signal while filtering out adjacent band interference, thus improving the overall signal communication quality of the multi-frequency antenna. Attached Figure Description
[0022] Figure 1 This is a flowchart of a multi-frequency antenna signal optimization method based on data processing.
[0023] Figure 2 This is a schematic diagram of the power spectral density.
[0024] Figure 3 This is a schematic diagram of the zero-pole distribution.
[0025] Figure 4 This is a schematic diagram of group delay characteristics. Detailed Implementation
[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0027] An embodiment of the multi-frequency antenna signal optimization method based on data processing provided by this invention: like Figure 1 As shown, the multi-frequency antenna signal optimization method based on data processing includes the following steps: S1, acquire the intermediate frequency signal and calculate the spectral density and interference distribution.
[0028] Acquire the digital intermediate frequency (IF) signal from the multi-frequency antenna, calculate the target frequency band power spectral density and frequency cross-correlation matrix of the digital IF signal, and calculate the interference power distribution data of the adjacent frequency band.
[0029] The analog radio frequency signal received by the multi-frequency antenna is converted to digital signal at a fixed sampling rate using an analog-to-digital converter (ADC) to obtain a discrete-time series as the digital intermediate frequency (IF) signal. The `scipy.signal.welch` function in the Python environment is called to segment the discrete-time series using the Welch method, apply a Hamming window, and then perform a Fast Fourier Transform (FFT) to calculate the average, yielding the power spectral density distribution sequence of the target frequency band. A filter bank based on bandpass filters is used to divide the digital IF signal into multiple sub-bands. The `numpy.cov` function is called to calculate the covariance of the complex spectral vectors of each sub-band signal, obtaining a frequency cross-correlation matrix representing the correlation of different frequency components. The frequency bands offset by a preset megahertz on both sides of the center frequency are extracted as the neighboring frequency bands. The power spectral density distribution sequence is then accumulated using discrete Riemann integrals within the neighboring frequency bands to obtain the interference power distribution data of the neighboring frequency bands.
[0030] S2, determine the order, set constraints, and complete the frequency pre-distortion processing.
[0031] The filter order of the discrete domain digital transfer function is determined based on the spectral entropy value of the target frequency band power spectral density. The key frequency range in the passband is identified based on the frequency cross-correlation matrix, and strict passband ripple constraints are set. High attenuation target values are set for high interference frequency points in the stopband based on the interference power distribution data of adjacent frequency bands. Frequency pre-distortion processing is performed on the key frequency range and high interference frequency points.
[0032] The distribution sequence of the target frequency band power spectral density is summed and normalized to transform it into a probability distribution sequence. The Shannon information entropy of this probability distribution sequence is calculated using the `scipy.stats.entropy` function, and this entropy value is used as the spectral entropy value. The reciprocal of this spectral entropy value is calculated, multiplied by a preset fundamental order constant, and rounded up to determine the filter order of the discrete-domain digital transfer function. The off-diagonal elements of the frequency cross-correlation matrix are traversed, and the row and column frequency intervals corresponding to elements with a normalized correlation coefficient absolute value greater than 0.8 are defined as the critical frequency intervals within the passband. The maximum allowable peak-to-peak value of the gain fluctuation in this critical frequency interval is set to 0.1 dB as a strict passband ripple constraint. The `scipy.signal.find_peaks` function is used to perform peak finding operations on the adjacent frequency band interference power distribution data to identify local maxima as high interference frequencies. The target stopband attenuation value at the high interference frequency points is set to -60 dB. The frequency pre-distortion algorithm in the bilinear transform method is used to calculate the tangent of half the product of the analog angular frequency of the boundary points of the key frequency range and the high interference frequency points and the sampling period. Then, it is multiplied by 2 and divided by the sampling period to convert it into the pre-distorted digital angular frequency.
[0033] In some implementations, the filter order of the discrete-domain digital transfer function is determined based on the spectral entropy value of the target frequency band power spectral density, including: The target frequency band is divided into several equidistant frequency grids, the power spectral density amplitude in each frequency grid is obtained, and a discrete amplitude sequence is constructed. After normalizing the discrete amplitude sequence, the probability distribution is calculated, and the spectral entropy value of the target frequency band is calculated using the Shannon entropy formula. Calculate the reciprocal of the spectral entropy value, multiply the reciprocal by a preset scaling factor to amplify it, and round the amplified value up to obtain an integer that meets the signal complexity requirements. Use the integer as the filter order.
[0034] Hardware computing resources are allocated by analyzing the degree of disorder in the signal power distribution within the frequency band. The specific process is as follows: Determine the total bandwidth of the target frequency band to be processed by the system, for example, set it to 20MHz. Divide this 20MHz bandwidth evenly into N equidistant frequency grids, preferably N=1024, then the frequency step size of each grid is approximately 19.53kHz. Extract the power spectral density amplitude P(i) at the center frequency of each grid, i=1,2,...,1024, to construct a discrete amplitude sequence representing the energy distribution of the entire frequency band.
[0035] The sequence is then subjected to probability normalization, calculated using the following formula: This ensures that the sum of the probabilities of all frequency points equals 1. The Shannon entropy formula is then used to calculate the spectral entropy value H. When a wideband mixed signal exists within the frequency band, the energy distribution is relatively uniform, resulting in a larger calculated spectral entropy value H. The required transition band is lower, and a smaller filter order is sufficient. When only a single or a few discrete narrowband carriers exist, the energy distribution is highly concentrated, the spectral entropy value H is smaller, and the spectrum shape is steeper and more complex. A higher filter order is needed to provide sufficient pole degrees of freedom.
[0036] The spectral entropy value H is converted into the filter order. The specific calculation logic is as follows: Filter order Where K is a preset proportionality coefficient, preferably ranging from 15 to 30. This represents the floor function.
[0037] Assuming the signal acquired at the site contains some narrowband interference and the main carrier, with concentrated energy distribution, the calculated spectral entropy value is relatively small, H=3.2. The preset scaling factor K is set to 24. Substituting into the above formula, we obtain 7.5. Rounding up 7.5 determines the filter order to be 8th.
[0038] In some implementations, key frequency ranges within the passband are identified based on the frequency cross-correlation matrix, and strict passband ripple constraints are set, including: Extract frequency pairs with correlation coefficients greater than a preset threshold from the frequency cross-correlation matrix, and merge consecutive high-correlation frequency pairs with frequency spacing less than a preset interval to obtain at least one continuous frequency interval as the key frequency interval. Set the upper limit of passband ripple in the key frequency range as the first ripple tolerance value; The upper limit of the passband ripple in the frequency region outside the critical frequency range within the passband is set as the second ripple tolerance value, wherein the first ripple tolerance value is less than the second ripple tolerance value.
[0039] Observational data of the input signal at different frequency points are extracted, and a two-dimensional frequency cross-correlation matrix is calculated. This matrix is then traversed to extract frequency pairs with cross-correlation values greater than a preset threshold, such as 0.85, with a preferred range of 0.75-0.90. Highly correlated frequencies typically belong to the same modulated broadband signal.
[0040] Perform frequency domain adjacency detection: If the absolute value of the center frequency spacing between the extracted high correlation frequency points is less than the preset step size, such as 200kHz, then they are merged and concatenated into an array.
[0041] If every pair of frequency points from 2.411 GHz, 2.413 GHz to 2.419 GHz shows a high correlation of over 0.85, and the interval between each pair is less than 200 kHz, then these frequency points are merged to generate a continuous interval with a start and end frequency of [2.410 GHz, 2.420 GHz], which is marked as the critical frequency interval. The measured target frequency band power spectral density distribution in this embodiment is as follows: Figure 2 As shown, the power characteristics, peak distribution of the main carrier, and interference power distribution in the key frequency range of 2.410 GHz to 2.420 GHz are presented.
[0042] For the ripple constraint configuration, the first ripple tolerance parameter within the aforementioned critical frequency range of [2.410GHz, 2.420GHz] is locked at an upper limit of 0.1dB to ensure an absolutely flat amplitude response of the core carrier and avoid deterioration of the error vector amplitude. For the remaining non-critical frequency sub-regions within the passband, the corresponding second ripple tolerance value is relaxed to 0.5dB. This asymmetric dual-threshold mechanism balances signal fidelity with the feasibility of filter construction.
[0043] In some implementations, high attenuation target values are set for high interference frequency points within the stopband based on adjacent frequency band interference power distribution data, including: The amplitude of interference power is collected in adjacent frequency bands, and the average interference power of all collection points is calculated as the judgment threshold. Frequency points with interference power greater than the judgment threshold are extracted as high interference frequency points; A first stopband attenuation target value is set for high interference frequency points, and a second stopband attenuation target value is set for the remaining non-high interference frequency points within the stopband, wherein the attenuation degree of the first stopband attenuation target value is higher than the attenuation degree of the second stopband attenuation target value.
[0044] For electromagnetic interference environments at the frequency band edges, this embodiment uses a probe mechanism to generate non-uniform stopband attenuation constraints. The specific process is as follows: Within the transition and guard bands immediately adjacent to the target passband, interference power amplitudes were cyclically collected at 512 sample points with a fixed sampling step size of 50 kHz. The power values of the 512 data points were linearly summed and the arithmetic mean was taken, which was then used as the benchmark threshold.
[0045] By comparing 512 discrete sampling points in a loop, if the absolute power amplitude of a certain frequency point exceeds the above-mentioned judgment threshold, an interference set is specially established and marked as a high interference frequency point.
[0046] If the threshold is -90dBm, and strong radiation with a power of -60dBm is detected at 2.45GHz, then 2.45GHz is marked as a high-interference frequency. In this case, the optimization function forces a first stopband attenuation target value of -75dB for 2.45GHz; while for other frequency regions within the stopband that do not exceed the -90dBm threshold, only the most basic baseband signal isolation specifications need to be met, and a second stopband attenuation target value of the standard -50dB is configured. Through this strategy, with a limited total filter order, scarce transmission zeros can be directed to areas of strong interference, achieving targeted elimination and avoiding order inflation caused by blindly increasing the global attenuation index.
[0047] S3, construct the cost function and iteratively solve for the optimal zeros and poles.
[0048] A multi-objective cost function is established, with the positions of complex plane poles and transmission zeros corresponding to the filter order as variables. The passband ripple parameter, group delay flatness parameter, and stopband attenuation parameter are integrated into the multi-objective cost function. Penalty weights are applied to the ripple deviation in the key frequency range and the insufficient attenuation at high interference frequency points. The poles of the standard Chebyshev filter are used as the initial pole values and the initial transmission zero coordinates are configured. The multi-objective cost function is solved iteratively to obtain the optimal set of complex plane poles and the optimal set of transmission zeros.
[0049] A multi-objective cost function is constructed, with one-dimensional vectors of pole and zero coordinates (composed of real and imaginary parts) as independent variables. This function consists of a weighted sum of three terms: the first term is the root mean square error of the passband ripple difference between the current iteration's transfer function and 0.1 dB; the second term is the group delay variance obtained by differentiating the phase of the transfer function with respect to the angular frequency; and the third term is the sum of squares of the actual stopband attenuation difference from -60 dB. A piecewise Lagrange multiplier is introduced into the multi-objective cost function, triggering a penalty mechanism when the ripple in the critical frequency range exceeds 0.1 dB or the absolute value of the attenuation at high interference frequencies is less than 60 dB. This penalty is multiplied by a penalty weight coefficient of 1000. The `scipy.signal.cheb1ap` function is called to generate the complex pole coordinates of the first-type Chebyshev low-pass prototype filter based on the previously determined filter order and the 0.1 dB passband ripple parameter, serving as the initial pole values. Coordinate points are then evenly spaced on the imaginary axis outside the unit circle as the initial transmission zero coordinates. The `scipy.optimize.minimize` function is called to employ the Sequential Least Squares Programming (SLSQP) algorithm to reduce the total value of the multi-objective cost function, performing up to one thousand iterations to find the optimal direction. The output is an array of poles and zeros that converges the gradient norm of the cost function to the minimum threshold, serving as the optimal set of complex plane poles and the optimal set of transported zeros, respectively.
[0050] In some implementations, a multi-objective cost function is established, using the locations of complex plane poles and transmission zeros corresponding to the filter order as variables, including: The coordinates of the real and imaginary parts of the complex plane poles and transmission zeros corresponding to the filter order under the conjugate symmetry constraint are used as the optimization variable vector; A first cost term representing passband ripple error, a second cost term representing group delay flatness, and a third cost term representing stopband attenuation are constructed respectively. In the first cost term, the critical frequency range after frequency pre-distortion is assigned a ripple penalty weight higher than other passband frequencies. In the third cost term, the high interference frequency point after frequency pre-distortion is assigned an attenuation penalty weight higher than other stopband frequencies. The first, second, and third cost terms are dimensionless and then weighted and summed to construct a multi-objective cost function.
[0051] In practical implementation, taking a defined 8th-order filter as an example, an optimization variable with 8 complex poles and 8 transmission zeros is constructed. Since the physical system requires the coefficients to be real numbers, the poles and zeros must appear as conjugate complex pairs. Therefore, the optimization variable vector is actually composed of the coordinates of the real and imaginary parts of half of the poles and zeros.
[0052] Based on this variable vector, a three-dimensional multi-objective cost function J is constructed: First cost item Calculate the least squares variance between the actual amplitude-frequency response and the set 0.1dB / 0.5dB ripple tolerance. The calculation utilizes a non-uniform passband ripple penalty: in the critical frequency range, the weighting factor is set to 10.0; in the normal passband region, the weight is set to 1.0.
[0053] Second cost item Extract the passband group delay from the derivative of the complex phase frequency distribution model and calculate the range.
[0054] Third cost item Calculate the smoothed quadratic difference between the measured gain and the target attenuation value. Similarly, utilize the non-uniform stopband attenuation penalty: for the 2.45GHz high-interference frequency point requiring a -75dB deep attenuation, configure a high deviation suppression penalty coefficient, such as 15.0; in the ordinary stopband region, only associate the basic scalar with a value of 1.0.
[0055] Before calculating the total cost, first... , , Each value is divided by its initial error value for dimensionless normalization. The inner products are then summed using a preset three-dimensional weight coefficient ratio vector [0.45, 0.25, 0.30]. The total cost function is then... This configuration not only prioritizes passband ripple fidelity with a weight of 0.45, but also uses an internal penalty factor to ensure that the optimization direction prioritizes meeting the stringent constraints of critical regions and high-interference points.
[0056] In some implementations, the poles of the standard Chebyshev filter are used as the initial pole values, and the initial coordinates of the transmission zeros are configured. The multi-objective cost function is then iteratively solved to obtain the optimal set of complex plane poles and the optimal set of transmission zeros, including: The coordinates of the complex plane poles of the standard Chebyshev filter at the order of the filter are calculated as the initial values of the poles, and the coordinates of the initial transmission zeros are configured. Both types of coordinates are used as the initial values of the optimization variable vector. A numerical optimization algorithm is used to iteratively update the positions of poles and transmission zeros in the complex plane. In each iteration, the changes in the positions of poles and transmission zeros in the complex plane, as well as the changes in the value of the multi-objective cost function, are calculated. When the change in the value of the multi-objective cost function is less than the preset convergence threshold or the maximum number of iterations is reached, the iteration stops, and the current complex plane pole positions and transmission zero positions are output, forming the optimal complex plane pole set and the optimal transmission zero set.
[0057] The derivation formula for the equidistant semi-major and semi-minor axes of an 8th-order standard Chebyshev Type I filter is used analytically to obtain the x and y coordinates of eight complex poles on the continuous elliptical trajectory in the stable region of the left half-complex plane, serving as the starting point for pole convergence guidance. Simultaneously, pairs of initial zeros are arranged on the imaginary axis near the aforementioned high-interference frequency points, such as 2.45 GHz. These pole-zero coordinates are then combined to form the initial input vector for the numerical optimization algorithm.
[0058] After initiating iterative computation, the optimization engine monitors the total cost function in real time at each step of the deduction. Iterative reduction of net asset value .
[0059] A stringent stopping and monitoring mechanism has been established internally: the maximum allowed number of iterations is set to 1000, and the absolute precision dead zone for convergence is set to... During recursive optimization, if the total cost function is reduced by a certain amount in 5 consecutive iterations... All less than If the iteration count reaches the exhaustion limit of 1000, the kernel will forcibly terminate the loop. At this point, the register latches and outputs the current coordinate array, thus forming a fully customized set of optimal complex plane poles and optimal set of transport zeros. After iterative optimization using the multi-objective cost function, the optimal complex plane pole and zero distribution is obtained, as shown below. Figure 3 As shown.
[0060] In another embodiment, the search step size of the numerical optimization algorithm in the next iteration is dynamically adjusted according to the changes in the positions of the complex plane poles and the transmission zeros. When the position changes are greater than a preset oscillation threshold, the search step size is reduced to suppress optimization oscillation. When the value change of the multi-objective cost function is less than a preset cost convergence threshold, and the changes in the positions of the complex plane poles and the transmission zeros are both less than a preset spatial convergence threshold, or when the maximum number of iterations is reached, the iteration is stopped. Specifically, after each iteration, the Euclidean distance or norm of the pole and zero coordinate vectors compared to the previous round is calculated, and these are combined to form the current spatial position movement step. This step is then compared with a preset adjustment threshold: if the movement step is greater than the preset oscillation threshold, it indicates that the algorithm's search span is too large, which may lead to overshooting or repeated oscillations near the optimal solution. In this case, the search step size for the next iteration is multiplied by a decay coefficient less than 1 to forcibly reduce the search radius. Conversely, if the movement step is consistently too small, it indicates that the algorithm is progressing slowly in a flat region. In this case, it can be multiplied by an expansion coefficient greater than 1 to increase the step size. The dynamically scaled new step size is then directly substituted into the coordinate update formula of the numerical optimization algorithm to execute the next iteration.
[0061] S4, construct the transfer function and complete the signal filtering output.
[0062] A continuous-domain analog transfer function is constructed using the optimal set of complex plane poles and the optimal set of transmission zeros. This continuous-domain analog transfer function is then discretized into a discrete-domain digital transfer function. The discrete-domain digital transfer function is used to filter the digital intermediate frequency signal, resulting in an optimized multi-frequency antenna signal.
[0063] The optimal set of transmission zeros is used as the roots of the numerator polynomial of the transfer function, and the optimal set of complex plane poles is used as the roots of the denominator polynomial. The `scipy.signal.zpk2tf` function is called to expand the zero-pole-gain model and combine like terms into the numerator and denominator polynomial coefficient vectors of the continuous-domain analog transfer function of the complex variable `s`. Using the bilinear transformation mapping rule, the numerator and denominator coefficient vectors of the complex variable `s` domain are mapped to the complex variable `z` domain using the system sampling frequency, generating the discrete polynomial coefficients of the discrete-domain digital transfer function. The `scipy.signal.tf2sos` function is called to convert the digital transfer function into multiple second-order cascaded structures. This second-order cascaded structure serves as the system state parameter of the difference equation, acting sequentially on the time series of the original digital intermediate frequency signal to perform time-domain multiplication-addition convolution operations, eliminating adjacent frequency band interference and flattening the passband group delay characteristics, ultimately outputting a filtered and optimized multi-frequency antenna signal sequence.
[0064] In some implementations, a continuous-domain analog transfer function is constructed from the optimal set of complex plane poles and the optimal set of transmission zeros. This continuous-domain analog transfer function is then discretized into a discrete-domain digital transfer function. The discrete-domain digital transfer function is then used to filter the digital intermediate frequency signal, outputting an optimized multi-frequency antenna signal, including: Calculate the polynomial coefficients of the continuous-domain simulated transfer function based on the optimal set of complex plane poles and the optimal set of transport zeros; By employing the bilinear transform method, the continuous-domain analog transfer function is mapped to the discrete domain, thus obtaining the digital filter coefficients of the discrete-domain digital transfer function. The digital intermediate frequency signal is used as the input sequence, and discrete filtering is performed using digital filter coefficients to generate and output the filtered and optimized multi-frequency antenna signal.
[0065] The optimal complex plane pole and optimal transmission zero point Through continuous product expansion, a classic Laplace continuous-domain rational polynomial analogous transfer function is constructed. Furthermore, through algebraic expansion, the continuous state constant coefficient set of each power order is extracted.
[0066] Discretization is performed based on the sampling clock parameters of the actual communication baseband system. Assuming the underlying hardware clock bus sampling frequency is configured at 122.88MHz, the corresponding sampling period T is approximately 8.138ns. Substituting the classic bilinear transform operator, the analog s-domain is completely and equivalently mapped to the digital discrete z-domain. This process achieves seamless replacement to prevent frequency superposition distortion. After unpacking and merging, the discrete digital filter coefficients specifically used to drive the underlying pulse operation are extracted, i.e., the feedforward numerator difference coefficients. and feedback denominator suppression coefficient .
[0067] The above-extracted and The digital coefficient set is programmed and loaded into the high-speed multiply-accumulate register of the underlying FPGA. The raw multi-channel digital intermediate frequency stream containing various spurious signals and out-of-band aliasing, sent from the front-end antenna bus A / D converter, is injected as a discrete input sequence x[n]. The FPGA computing engine concurrently executes differential equation operations within one clock cycle. Through the above high-speed hardware pipeline differential iteration, interference at high interference points is suppressed, and perfect waveforms are preserved in key frequency regions, outputting the purified multi-frequency antenna digital signal to the subsequent baseband stage.
[0068] The experimental conditions were set using a real communication signal with a sampling frequency of 122.88MHz as the test data source, with a total signal bandwidth of 20MHz. The core critical frequency range was concentrated between 2.41GHz and 2.42GHz, and high spurious interference points were present in the adjacent frequency band. Based on spectral entropy characteristics, the proposed scheme calculated the initial filter order to be 8th order, configured the passband ripple tolerance constraint in the critical range to be 0.1dB, set the stopband target at high interference frequencies to an extremely deep attenuation of 75dB, and assigned a very high targeting penalty weight parameter to perform multi-objective cost iterative optimization. The control group experiment used the same 8th order benchmark, a conventional equal-ripple Chebyshev digital filter design architecture, with a globally uniform hard constraint of 0.1dB for the passband ripple index, and a globally fixed stopband attenuation defense line of 65dB, abandoning the regionally differentiated ripple broadening and frequency-point heavy-pressure notch filtering mechanism based on a multi-dimensional matrix.
[0069] Regarding amplitude-frequency characteristics and ripple performance, the proposed method achieves a measured extreme passband ripple level of 0.07 dB within the critical frequency band, with the ripple in non-critical information bands relaxed to 0.43 dB. In contrast, the traditional control group measures a global ripple lock at 0.09 dB. In the time-domain phase distortion assessment, the proposed method achieves a measured full-band group delay ripple range of only 5.4 ns, while the control group, due to the pressure of the global limit approaching the zero position, exhibits a measured group delay deviation range soaring to 22.6 ns. A comparison of the proposed method with the measured group delay characteristics of the traditional Chebyshev filter is shown below. Figure 4As shown in the figure, in the comparison of out-of-band noise floor and strong interference cancellation performance, the proposed scheme pinpointed the high interference target and the measured suppression notch depth reached 77.5dB. The background suppression in other non-high-risk frequency domains remained at 51.8dB. When the control group faced completely identical out-of-band high-risk interference radiation, the measured suppression attenuation could only barely be maintained at 65.4dB.
[0070] Compared to the fixed and rigid traditional standard approximation methods, the architecture of this embodiment, under the stringent precondition of not adding any additional hardware resource costs such as multiply-accumulator computing units and total memory cache area, not only ensures that the backbone sensitive carrier frequency band has completely flat signal fidelity characteristics, but also achieves a significant 76% optimization of the group delay flatness index of the entire communication chain. Based on the targeted weight penalty deep-dive mechanism, the measured point-based anti-compression isolation depth against externally strong interfering frequency sources is improved by more than 12dB. At the lowest level of the computing core architecture, it overcomes the technical challenge of simultaneously achieving a steep high-frequency roll-off rate for filters and linearity of in-band phase frequency characteristics, a bottleneck inherent in traditional digital signal processing technologies.
[0071] An embodiment of the multi-frequency antenna signal optimization system based on data processing provided by the present invention: The multi-frequency antenna signal optimization system based on data processing includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the aforementioned multi-frequency antenna signal optimization method based on data processing.
[0072] The multi-frequency antenna signal optimization system based on data processing also includes other components well known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art and will not be described in detail here.
[0073] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.
Claims
1. A multi-frequency antenna signal optimization method based on data processing, characterized in that, Includes the following steps: S1, acquire the digital intermediate frequency signal of the multi-frequency antenna, calculate the target frequency band power spectral density and frequency cross-correlation matrix of the digital intermediate frequency signal, and calculate the interference power distribution data of the adjacent frequency band; S2, determine the filter order of the discrete domain digital transfer function based on the spectral entropy value of the target frequency band power spectral density, identify the key frequency range in the passband according to the frequency cross-correlation matrix, set strict passband ripple constraints, set high attenuation target values for high interference frequency points in the stopband according to the interference power distribution data of adjacent frequency bands, and perform frequency pre-distortion processing on the key frequency range and high interference frequency points. S3. Establish a multi-objective cost function with the positions of complex plane poles and transmission zeros corresponding to the filter order as variables. In the multi-objective cost function, the passband ripple parameter, group delay flatness parameter, and stopband attenuation parameter are integrated. Penalty weights are applied to the ripple deviation in the key frequency range and the insufficient attenuation at high interference frequency points. The poles of the standard Chebyshev filter are used as the initial pole values and the initial transmission zero coordinates are configured. The multi-objective cost function is solved iteratively to obtain the optimal set of complex plane poles and the optimal set of transmission zeros. S4 constructs a continuous-domain analog transfer function from the optimal set of complex plane poles and the optimal set of transmission zeros. The continuous-domain analog transfer function is then discretized into a discrete-domain digital transfer function. The discrete-domain digital transfer function is used to filter the digital intermediate frequency signal, and the optimized multi-frequency antenna signal is output.
2. The multi-frequency antenna signal optimization method based on data processing according to claim 1, characterized in that, Determining the filter order of the discrete-domain digital transfer function based on the spectral entropy value of the target frequency band power spectral density includes: The target frequency band is divided into several equidistant frequency grids, the power spectral density amplitude in each frequency grid is obtained, and a discrete amplitude sequence is constructed. After normalizing the discrete amplitude sequence, the probability distribution is calculated, and the spectral entropy value of the target frequency band is calculated using the Shannon entropy formula. Calculate the reciprocal of the spectral entropy value, multiply the reciprocal by a preset scaling factor to amplify it, and round the amplified value up to obtain an integer that meets the signal complexity requirements. Use the integer as the filter order.
3. The multi-frequency antenna signal optimization method based on data processing according to claim 1, characterized in that, The key frequency range within the passband is identified based on the frequency cross-correlation matrix, and strict passband ripple constraints are set, including: Extract frequency pairs with correlation coefficients greater than a preset threshold from the frequency cross-correlation matrix, and merge consecutive high-correlation frequency pairs with frequency spacing less than a preset interval to obtain at least one continuous frequency interval as the key frequency interval. Set the upper limit of passband ripple in the key frequency range as the first ripple tolerance value; The upper limit of the passband ripple in the frequency region outside the critical frequency range within the passband is set as the second ripple tolerance value, wherein the first ripple tolerance value is less than the second ripple tolerance value.
4. The multi-frequency antenna signal optimization method based on data processing according to claim 1, characterized in that, Based on the interference power distribution data of adjacent frequency bands, high attenuation target values are set for high interference frequency points within the stopband, including: The amplitude of interference power is collected in adjacent frequency bands, and the average interference power of all collection points is calculated as the judgment threshold. Frequency points with interference power greater than the judgment threshold are extracted as high interference frequency points; A first stopband attenuation target value is set for high interference frequency points, and a second stopband attenuation target value is set for the remaining non-high interference frequency points within the stopband, wherein the attenuation degree of the first stopband attenuation target value is higher than the attenuation degree of the second stopband attenuation target value.
5. The multi-frequency antenna signal optimization method based on data processing according to claim 1, characterized in that, A multi-objective cost function is established, using the positions of complex plane poles and transmission zeros corresponding to the filter order as variables. This includes: using the real and imaginary coordinates of the complex plane poles and transmission zeros corresponding to the filter order under conjugate symmetry constraints as optimization variable vectors; constructing a first cost term representing passband ripple error, a second cost term representing group delay flatness, and a third cost term representing stopband attenuation; assigning a ripple penalty weight higher than other passband frequencies to the key frequency range after frequency pre-distortion processing in the first cost term, and assigning an attenuation penalty weight higher than other stopband frequencies to the high interference frequency points after frequency pre-distortion processing in the third cost term; and then weighting and summing the first, second, and third cost terms after dimensionless processing to obtain the multi-objective cost function.
6. The multi-frequency antenna signal optimization method based on data processing according to claim 1, characterized in that, Using the poles of a standard Chebyshev filter as initial pole values and configuring initial transmission zero coordinates, the multi-objective cost function is iteratively solved to obtain the optimal set of complex plane poles and the optimal set of transmission zeros. This includes: calculating the coordinates of the complex plane poles of the standard Chebyshev filter at the filter order as initial pole values, configuring initial transmission zero coordinates, and using both types of coordinates as initial values for the optimization variable vector; using a numerical optimization algorithm to iteratively update the positions of the complex plane poles and transmission zeros, calculating the changes in the positions of the complex plane poles and transmission zeros, as well as the changes in the value of the multi-objective cost function in each iteration; stopping the iteration when the change in the value of the multi-objective cost function is less than a preset convergence threshold or the maximum number of iterations is reached, and outputting the current positions of the complex plane poles and transmission zeros, thus forming the optimal set of complex plane poles and the optimal set of transmission zeros.
7. The multi-frequency antenna signal optimization method based on data processing according to claim 1, characterized in that, A continuous-domain analog transfer function is constructed using the optimal set of complex plane poles and the optimal set of transmission zeros. This continuous-domain analog transfer function is then discretized into a discrete-domain digital transfer function. The discrete-domain digital transfer function is used to filter the digital intermediate frequency signal, outputting an optimized multi-frequency antenna signal, including: Calculate the polynomial coefficients of the continuous-domain simulated transfer function based on the optimal set of complex plane poles and the optimal set of transport zeros; By employing the bilinear transform method, the continuous-domain analog transfer function is mapped to the discrete domain, thus obtaining the digital filter coefficients of the discrete-domain digital transfer function. The digital intermediate frequency signal is used as the input sequence, and discrete filtering is performed using digital filter coefficients to generate and output the filtered and optimized multi-frequency antenna signal.
8. The multi-frequency antenna signal optimization method based on data processing according to claim 1, characterized in that, The method for acquiring digital intermediate frequency (IF) signals from a multi-frequency antenna includes: performing a fixed sampling rate analog-to-digital conversion on the analog radio frequency signal received by the multi-frequency antenna using an analog-to-digital converter to obtain a discrete time series as the digital IF signal.
9. The multi-frequency antenna signal optimization method based on data processing according to claim 1, characterized in that, The method for obtaining spectral entropy includes: summing and normalizing the distribution sequence of the target frequency band power spectral density to convert it into a probability distribution sequence, and using the scipy.stats.entropy function to calculate the Shannon information entropy of the probability distribution sequence as the spectral entropy value.
10. A multi-frequency antenna signal optimization system based on data processing, characterized in that, It includes a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the multi-frequency antenna signal optimization method based on data processing as described in any one of claims 1-9 is implemented.