A method for analyzing the formation mechanism of filter radio frequency fingerprint

Abstract

The present invention discloses a method for analyzing the formation mechanism of radio frequency fingerprint of a filter, which systematically analyzes the generation mechanism of radio frequency fingerprint. Radio frequency fingerprint is formed due to errors in the production process of the filter and various differences between filters, which leads to differences in the output of each filter. Therefore, the formation mechanism of radio frequency fingerprint of the filter is analyzed by establishing a connection between the size error and the frequency response curve of the output. The specific idea is as follows: first, the output frequency response curve of the filter with different size errors is obtained, and then the frequency response curve value of the filter is interpolated and fitted with the size error of the filter to obtain the interpolation function; then, the frequency response curve value of each size error filter is predicted by the interpolation function, and the frequency response curve data is trained; finally, the actual filter is classified by the trained classifier to verify whether the fingerprint characteristics obtained by the interpolation function and the fingerprint characteristics of the actual filter of the size can match.

Description

Technical Field

[0001] The invention belongs to the field of radiation source identification, and specifically proposes a method for analyzing the radio frequency fingerprint formation mechanism of a filter. Background Art

[0002] In today's connected world, sensors and communication radiation sources are widely used across various industries. Filter circuits are crucial components of sensors and radiation sources. The fingerprint of filters often influences the overall fingerprint of the sensor and radiation source. Therefore, research on the RF fingerprint of filters is of great significance. RF fingerprints have been a research hotspot in radiation source identification in recent years, primarily used to extract radiation source characteristics for classification and identification. RF fingerprints are caused by hardware variations during the manufacturing process, resulting in errors. Currently, in the field of radiation source identification, much research focuses on extracting RF fingerprints from radiation sources through signal processing to identify different radiation source devices. However, research on the mechanisms underlying RF fingerprint formation is limited. Therefore, a method is needed to analyze and study the mechanisms and principles underlying the formation of RF fingerprints in filters.

[0003] RF fingerprint extraction methods are primarily categorized into transient signal feature extraction, steady-state signal feature extraction, and deep learning-based feature extraction. Transient signals are generated when a radiator transitions from off to on. They are independent of the data portion of the signal and are only affected by internal hardware and manufacturing errors. However, due to their short duration and difficulty in acquisition, transient signals place high demands on signal extraction equipment, significantly increasing acquisition costs. Steady-state signal feature extraction methods primarily use the signal's preamble and data portions to extract features such as carrier frequency deviation, phase deviation, and amplitude gain error. Furthermore, signal processing is used to extract frequency domain and spectral features. Steady-state signal feature extraction targets specific signal segments, which can result in information loss, leading to poor feature extraction performance and poor recognition results. Deep learning feature extraction, which can input the entire signal segment, addresses this information loss issue in steady-state signal feature extraction. However, most features extracted using deep learning are statistically derived and lack interpretability. Whether it is transient signal feature extraction, steady-state signal feature extraction or feature extraction through deep learning, they only utilize RF fingerprints and do not conduct in-depth research and analysis on the generation mechanism of RF fingerprints.

[0004] This paper proposes a method for analyzing the formation mechanism of a radiation source's RF fingerprint. It systematically analyzes the causes of the RF fingerprint of a single component, the filter. By varying the dimensional tolerances of the filter module and observing the filter's output response curve, the relationship between filter production tolerances and the nonlinearity of the output response curve is explored. Summary of the Invention

[0005] The purpose of the present invention is to provide an analysis method for the formation mechanism of the radio frequency fingerprint of a radiation source, which can clearly and intuitively explain the formation mechanism of the radio frequency fingerprint of a reaction filter module.

[0006] The present invention provides the following technical solutions:

[0007] A method for analyzing the formation mechanism of filter radio frequency fingerprints, the technical solution includes:

[0008] The RF fingerprint mechanism of the filter circuit is analyzed. The output frequency response curves of filters with different size errors are obtained respectively. The size error value of the filter and the numerical value of the output frequency response curve are interpolated to obtain the interpolation function. The frequency response curves of filters with different size errors are predicted by the interpolation function for machine learning. Then, the classifier is used to classify and identify the actual filters.

[0009] A method for analyzing the formation mechanism of radio frequency fingerprint of a radiation source comprises the following steps:

[0010] Step 1: Select a filter size as a standard, artificially add errors based on the size of the standard filter, use HFSS software to simulate 10 2.4GHz bandpass filters with different size errors, and then process the actual products.

[0011] Step 2, obtain the |S of these 10 bandpass filters 21 | curve, select 5 filters, and compare the filter size error and |S 21 The |values ​​are interpolated using cubic spline interpolation to observe the size error and |S 21 |The law of change of value.

[0012] Step 3: Based on the interpolation function obtained in step 2, the filter |S under different error sizes is predicted. 21 | values ​​as training data, and different filters as training labels.

[0013] Step 4: Based on the training data and training labels of step 3, use the KNN algorithm to train, and use the |S based on the remaining filter of step 2 21 | value as test data, and then use the KNN algorithm for classification and recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Flowchart of the cubic spline interpolation algorithm

[0015] Figure 2 Flowchart for filter RF fingerprint mechanism analysis DETAILED DESCRIPTION

[0016] The present invention will be further described below in conjunction with the accompanying drawings:

[0017] The technical solution adopted by the present invention is: a method for analyzing the formation mechanism of filter radio frequency fingerprint, which specifically includes the following steps:

[0018] Step 1: Select a filter size as a standard, artificially add errors based on the size of the standard filter, use HFSS software to simulate 10 2.4GHz bandpass filters with different size errors, and then process the actual products.

[0019] Step 2, obtain the |S of 10 bandpass filters 21 | curve, select 5 filters, and compare the filter size error and |S 21 The |values ​​are interpolated using cubic spline interpolation to observe the size error and |S 21 The specific steps of the change of the value are as follows:

[0020] 2.1 The coupler width w1 in the filter is used as a variable, and filters with w1 errors of +0.001mm, -0.001mm, +0.003mm, -0.003mm, and +0.005mm are selected. In the 2GHz-2.7GHz frequency band, the error value Δw1 of the filter at each frequency point and the frequency response curve of the filter |S 21 |The interpolation function f1 is obtained by interpolating the value:

[0021] |S 21 |=f1(Δw1)

[0022] 2.2 The specific steps of cubic spline interpolation are as follows: (1) The size error Δw1 of the selected filter is used as the known interpolation variable x i , calculate the step length h between the known interpolation point and the unknown interpolation point x i , calculate the cubic spline coefficient M according to the three bending moment equations i , where the three bending moment equations are:

[0023] γ i M i-1 +2M i +α i M i+1 =β i

[0024] in γ i =1-α i ,

[0025] (2) The calculated M i(i=1,2,...,n), substitute into the formula to obtain the interpolation formula f1

[0026]

[0027] Step 3: By substituting the error value of the coupler in the filter within the range of -0.005mm to +0.005mm into the interpolation function f1, the corresponding |S 21 The values ​​are used as training data and trained using the KNN algorithm.

[0028] Step 4: Based on step 2, the unselected filters are measured to obtain the actual |S 21 The data value is used as the test data, and the KNN algorithm is used for classification and recognition to verify whether the size error of the filter corresponding to the label label obtained by the KNN algorithm is consistent with the size error of the actual filter.

Claims

1. A method for analyzing the formation mechanism of filter radio frequency fingerprint, comprising the following steps: Step 1: Select a filter size as a standard, artificially add errors based on the size of the standard filter, simulate 10 2.4GHz bandpass filters with different size errors using HFSS software, and then process the actual products; Step 2, obtain the |S of 10 bandpass filters 21 | curve, select 5 filters, and compare the filter size error and |S 21 The |values ​​are interpolated using cubic spline interpolation to observe the size error and |S 21 The specific steps of the change of the value are as follows: 2.1 The width w1 of the coupler in the filter is used as a variable, and filters with w1 errors of +0.001mm, -0.001mm, +0.003mm, -0.003mm, and +0.005mm are selected; in the 2GHz-2.7GHz frequency band, the error value Δw1 of the filter at each frequency point and the frequency response curve of the filter |S 21 |The interpolation function f1 is obtained by interpolating the value: |S 21 |=f1(Δw1) 2.2 The specific steps of cubic spline interpolation are as follows: (1) The size error Δw1 of the selected filter is used as the known interpolation variable x i , calculate the step length h between the known interpolation point and the unknown interpolation point x i , calculate the cubic spline coefficient M according to the three bending moment equations i , where the three bending moment equations are: c i M i-1 +2M i +a i M i+1 =b i among them c i =1-a i , (2) The calculated M i (i=1,2,...,n), where we substitute into the formula to obtain the interpolation formula f1 Step 3: By substituting the error value of the coupler in the filter within the range of -0.005mm to +0.005mm into the interpolation function f1, the corresponding |S 21 |values ​​are used as training data and trained using the KNN algorithm; Step 4: Based on step 2, the unselected filters are measured to obtain the actual |S 21 The data value is used as the test data, and the KNN algorithm is used for classification and recognition to verify whether the size error of the filter corresponding to the label label obtained by the KNN algorithm is consistent with the size error of the actual filter.

Images