A method, system and device for extracting and identifying radio frequency (RF) fingerprint features of a frequency shift keying (FSK) signal
By employing a method for extracting radio frequency fingerprint features from FSK signals based on instantaneous frequency wavelet transform and feature dimensionality reduction, combined with lightweight gradient boosting machine learning, the problems of low feature extraction capability and poor discriminability under low signal-to-noise ratio of FSK signals are solved, achieving efficient identification of individual radiation sources.
Patent Information
- Application Number
- CN202310949742.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-07-31
- Publication Date
- 2026-02-10
- Estimated Expiration
- 2043-07-31
AI Technical Summary
Existing technologies have low capabilities for extracting radio frequency fingerprint features from FSK signals, and poor distinguishability of FSK signal fingerprint features under low signal-to-noise ratios, making it difficult to achieve effective individual identification of radiation sources.
The radio frequency fingerprint features of FSK signals are extracted using instantaneous frequency wavelet transform, and then identified by a machine learning classifier that combines feature dimensionality reduction and lightweight gradient boosting. The process includes signal preprocessing, feature extraction, dimensionality reduction, and classification.
It improved the accuracy of individual radiation source identification, especially under low signal-to-noise ratio conditions, achieving an average identification accuracy of 94.2%, and enhanced the fingerprint feature discriminability and robustness of FSK signals.
Smart Images
Figure CN116975599B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of radiation source individual identification technology, specifically a method, identification method and system for extracting radio frequency fingerprint features of FSK signals. Background Technology
[0002] The characteristics of individual radiation source identification originate from the non-ideal characteristics of the transmitter hardware, namely, unintentional errors in the transmitter design and manufacturing process. Subtle differences exist in the analog circuit components of transmitters from different radiation source devices, thus forming the uniqueness of the radio frequency fingerprint (RF fingerprint). Individual radiation source identification extracts a stable and distinguishable RF fingerprint to identify each radiation source device. Therefore, RF fingerprints can serve as a unique identifier for radiation source devices, associating them with target platforms and enabling individual identification of target platforms. This has significant application prospects in fields such as wireless network security and communication reconnaissance.
[0003] FSK signals are a commonly used communication method. From the perspective of FSK signal radio frequency fingerprint feature extraction methods, they can be mainly divided into feature extraction based on signal parameters, feature extraction based on domain transformation, feature extraction based on mechanism modeling, and feature extraction based on deep learning.
[0004] (1) Feature extraction based on signal parameters mainly uses the parameter error of the signal as fingerprint features, and some features have good stability. The modulation frequency, frequency shift interval, symbol rate and symbol transition time are mainly used as fingerprint features. However, since it is difficult to accurately measure these parameters, the feature extraction capability is generally poor.
[0005] (2) Feature extraction based on domain transform mainly uses signal transformation tools to describe the signal, maximizing the display of the subtle features contained in the signal and obtaining the hidden fingerprint features in the signal. The main signal transforms used include: frequency domain, time-frequency domain, wavelet transform, and higher-order transforms, etc. Some signal transforms result in significant information loss and poor recognition stability.
[0006] (3) Feature extraction based on mechanism modeling involves establishing a mechanism model of the analog device and estimating the distortion parameters in the model using the received signal and demodulation results to obtain features. This method is simple to implement, has low algorithm complexity, and the features are interpretable, but mechanism modeling requires extensive professional knowledge and a large number of experiments.
[0007] (4) Deep learning-based feature extraction delegates the understanding and extraction of features to neural networks, eliminating the need for manual feature engineering and target modeling. However, model training requires a large number of samples. The main deep learning methods used include: Convolutional Neural Network (CNN), Deep Residual Network (ResNet), and Dilated Causal Convolution (DCC).
[0008] In summary, the existing technology has the following problems:
[0009] (1) FSK signal has low RF fingerprint feature extraction capability: At present, FSK signals usually extract modulation frequency, frequency shift interval, symbol rate and symbol transition time as fingerprint features. Since it is difficult to accurately measure these parameters, the feature extraction capability is generally low.
[0010] (2) Poor distinguishability of FSK signal RF fingerprint features under low signal-to-noise ratio: Under low signal-to-noise ratio, due to the influence of noise, the fingerprint features of FSK signal are easily confused, and the distinguishability of fingerprint features becomes poor. Summary of the Invention
[0011] To overcome the shortcomings of the prior art, the present invention provides a method, identification method and system for extracting radio frequency fingerprint features of FSK signals, which solves the problems of low FSK signal radio frequency fingerprint feature extraction capability and poor fingerprint feature distinguishability of FSK signals under low signal-to-noise ratio.
[0012] The technical solution adopted by the present invention to solve the above problems is:
[0013] A method for extracting radio frequency fingerprint features from FSK signals, characterized by comprising the following steps:
[0014] S1, Feature Extraction: Based on the instantaneous frequency wavelet transform, fingerprint features that can describe the radiation source are extracted, and the wavelet coefficients of the instantaneous frequency of the FSK signal are used as fingerprint features.
[0015] As a preferred technical solution, step S1 includes the following steps:
[0016] S11, classify the N symbols of each sample in the FSK signal:
[0017] The N symbols are divided into M classes according to the M-ary system. The classified FSK signals are as follows:
[0018]
[0019] Among them, s iThis represents the classified FSK signal, where i represents the FSK signal frequency number, i = 1, 2, ..., M, M represents the total number of FSK signal frequencies, j represents the number of symbols in the i-th class of signal, and P i,j,K This represents the sample point of the j-th symbol in the i-th type of signal;
[0020] S12, average the M types of FSK signals after classification:
[0021] Each FSK signal i The j signals are averaged according to their column vectors to obtain the averaged FSK signal as follows:
[0022]
[0023] in, This represents the averaged FSK signal. s i The average of the Mth column signal;
[0024] S13, averaging the M signals The instantaneous frequencies were calculated separately, resulting in M sets of instantaneous frequency vectors as follows:
[0025] f i =[W i,1 W i,2 ...W i,D ];
[0026] Among them, f i W represents the instantaneous frequency vector group. i,D This represents the D-th instantaneous frequency vector, where D represents the total number of extracted instantaneous frequencies;
[0027] S14, for the instantaneous frequency f of group M i Performing wavelet transforms on each group yields the following M sets of wavelet coefficient vectors:
[0028] F i =[C i,1 C i,2 ...C i,L ];
[0029] Among them, F i Let L represent the wavelet coefficient vector group, and C represent the total number of wavelet coefficients. i,L Represents the Lth wavelet coefficient;
[0030] S15, F i The amplitudes of the wavelet coefficients form the eigenvector, F = [|F1| |F2|...|F...]. M |], F is the extracted radio frequency fingerprint feature;
[0031] Where F represents the eigenvector, |F M | represents the magnitude of the wavelet coefficient vector of the Mth group.
[0032] As a preferred technical solution, the following steps are also included:
[0033] S2, Feature Dimensionality Reduction: Reduces high-dimensional feature vectors to low-dimensional feature vectors.
[0034] As a preferred technical solution, linear discriminant analysis is used to reduce high-dimensional feature vectors to low-dimensional feature vectors and remove redundant information.
[0035] As a preferred technical solution, M=2.
[0036] An FSK signal fingerprint recognition method, comprising the aforementioned FSK signal radio frequency fingerprint feature extraction method, further comprising the following steps:
[0037] S3, Classification and Recognition: Using a classifier to classify and recognize feature vectors.
[0038] As a preferred technical solution, in step S3, a lightweight gradient boosting machine learning classifier is used to divide the fingerprint features of all samples into a training set and a test set. The feature vectors in the training set are used to train the LightGBM classifier, and the feature vectors in the test set are input into the trained LightGBM classifier to obtain the test results. The average recognition accuracy is obtained by comparing the test results with the actual category results.
[0039] As a preferred technical solution, the following steps are also included before step S1:
[0040] B. Signal preprocessing: The received signal is preprocessed using digital signal processing techniques; the preprocessing includes one or more of the following: filtering, sample synchronization, frequency offset estimation / correction, and normalization.
[0041] As a preferred technical solution, the following steps are also included before step B:
[0042] A. Signal reception: The signal received by the antenna is amplified, filtered, frequency-converted, and analog-to-digital converted.
[0043] An FSK signal fingerprint recognition system for implementing the aforementioned FSK signal fingerprint recognition method includes the following modules connected in sequence:
[0044] Feature extraction module: used to extract fingerprint features that can describe the radiation source based on instantaneous frequency wavelet transform, with the wavelet coefficients of the instantaneous frequency of the FSK signal as fingerprint features;
[0045] Feature dimensionality reduction module: used to reduce high-dimensional feature vectors to low-dimensional feature vectors;
[0046] Classification and recognition module: Used to classify and recognize feature vectors using a classifier.
[0047] Compared with the prior art, the present invention has the following advantages:
[0048] The present invention provides a radio frequency fingerprint feature for FSK signals based on instantaneous frequency wavelet transform, which has good discriminative power and robustness to noise. The signal-to-noise ratio is ≥15dB, and the average recognition accuracy of 10 radiation sources is ≥94.2%. Compared with the instantaneous frequency radio frequency fingerprint feature extraction method for FSK signals, the average recognition accuracy is improved. Attached Figure Description
[0049] Figure 1 Flowchart for individual identification of radiation sources;
[0050] Figure 2 The time-domain plot shows the average of the FSK signals from the two radiation sources.
[0051] Figure 3 Wavelet coefficient diagram of the instantaneous frequency of FSK signals from two radiation sources;
[0052] Figure 4 The flowchart shows the RF fingerprint feature extraction process for FSK signals based on instantaneous frequency wavelet transform.
[0053] Figure 5 Feature views for using the present invention;
[0054] Figure 6 This is an individual identification confusion matrix diagram using the present invention. Detailed Implementation
[0055] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.
[0056] Example 1
[0057] like Figures 1 to 6 As shown, the current low capability of radio frequency fingerprint feature extraction makes it impossible to meet the requirements for individual identification of radiation sources. This invention aims to solve the following technical problems:
[0058] 1. Solving the problem of low FSK signal RF fingerprint feature extraction capability: The present invention extracts FSK signal fingerprint features based on instantaneous frequency wavelet transform (in this embodiment, discrete wavelet transform is used), which extracts richer transmitter distortion information and can effectively improve the individual identification performance of radiation sources.
[0059] 2. Solving the problem of poor distinguishability of FSK signal radio frequency fingerprint features under low signal-to-noise ratio: This invention suppresses the influence of noise on fingerprint features by averaging multiple FSK signals of the same frequency, thereby improving the distinguishability of fingerprint features under low signal-to-noise ratio.
[0060] More specifically:
[0061] (1) Individual identification process for radiation sources:
[0062] The process of identifying individual radiation sources mainly includes signal preprocessing, feature extraction, feature dimensionality reduction, and classification, such as... Figure 1 As shown.
[0063] Step 1: The receiving device amplifies, filters, converts, and converts the signal received by the antenna to digital. The digital signal is then input into the signal preprocessing unit. The signal preprocessing unit uses digital signal processing techniques to filter, synchronize samples, estimate / correct frequency offset, and normalize the received signal, providing a signal with good consistency for feature extraction.
[0064] Step 2: Feature extraction refers to extracting fingerprint features that can describe a specific radiation source based on the signal or a certain transformation. Here, the FSK signal radio frequency fingerprint feature extraction method based on instantaneous frequency wavelet transform is adopted.
[0065] Step 3: Feature dimensionality reduction. LDA (Linear Discriminant Analysis) is used to reduce high-dimensional feature vectors to low-dimensional feature vectors, remove redundant information, improve feature discrimination ability, and reduce computational complexity.
[0066] Step 4: Classification and Recognition. The LightGBM (Light Gradient Boosting Machine) classifier is used. The fingerprint features of all samples are divided into training set and test set. The LightGBM classifier is trained with the feature vectors in the training set. The feature vectors in the test set are input into the trained LightGBM classifier to obtain the test results. The average recognition accuracy is obtained by comparing the test results with the actual class results.
[0067] (2) Extraction of RF fingerprint features of FSK signals based on instantaneous frequency wavelet transform:
[0068] Radio frequency (RF) fingerprint features originate from transmitter distortion. Due to manufacturing tolerances of transmitter components, distortions exist in the analog circuits such as modulators, frequency sources, mixers, filters, and power amplifiers. These distortions form different RF fingerprint features. Transmitter distortion manifests in the RF signal generated by the transmitter. The signal features extracted based on transmitter distortion are relatively stable and reliable RF fingerprint features, possessing the ability to identify individual radiation sources.
[0069] Transmitter distortion can cause frequency distortion in the FSK signal, such as... Figure 2 The time-domain plot of the averaged FSK signals from the two radiation sources is shown below. Figure 3 The wavelet coefficients of the instantaneous frequency of the FSK signals from the two radiation sources are shown.
[0070] from Figure 2 , Figure 3 As can be seen, the two radiation sources have different frequencies. Characterizing and extracting this difference using wavelet coefficients of the instantaneous frequency gives us the fingerprint features of the radiation sources. Since the two radiation sources exhibit different distortions, and these distortions do not change randomly, the wavelet coefficients of the instantaneous frequency of the FSK signal can serve as effective and stable radio frequency fingerprint features.
[0071] The MFSK signal sample consists of N symbols, with a sampling factor of K. A method for extracting RF fingerprint features of the FSK signal based on instantaneous frequency wavelet transform is as follows: Figure 4 As shown.
[0072] Step 1: Classify the N symbols of each sample in the FSK signal. Divide the samples corresponding to the current symbol into M classes according to the M-ary system. The classified signal is as follows:
[0073] Here, j represents the number of FSK signal frequencies, and j represents the number of frequencies in that class.
[0074] M represents the total number of FSK signal frequencies (i.e., the total number of categories);
[0075] Step 2: Average the M categories of FSK signals after classification. Average the FSK signals s of each category. i The j signals are averaged according to their column vectors to obtain the averaged FSK signal as follows:
[0076]
[0077] Step 3: Average the M signals The instantaneous frequencies were calculated separately, resulting in M sets of instantaneous frequency vectors as follows:
[0078] f i =[W i,1 W i,2 ...W i,D [, i = 1, 2, ..., M is the number of FSK signal frequencies, D is the number of extracted instantaneous frequencies]
[0079] Step 4: For the instantaneous frequencies f of group M i Performing wavelet transforms on each group yields the following M sets of wavelet coefficient vectors:
[0080] F i =[Ci,1 C i,2 ...C i,L ], i = 1, 2, ..., M is the number of FSK signal frequencies, and L is the number of wavelet coefficients;
[0081] Step 5: Place F i The amplitudes of the wavelet coefficients form the eigenvector, F = [|F1| |F2|....|F M |], F represents the extracted radio frequency fingerprint feature.
[0082] (3) Experimental verification:
[0083] The experimental parameters were set as follows: 10 radiation sources, 2FSK signal modulation, symbol rate of 20KBaud / s, sampling rate of 1.6MSps, 150 symbols per sample, and signal-to-noise ratio of 15dB to 16dB.
[0084] The test procedure is as follows:
[0085] Step 1: Preprocess the acquired signal samples, including filtering, sample synchronization, frequency offset estimation / correction, normalization, etc., to filter out out-of-band noise, eliminate random interference in the channel, and provide a signal with better consistency for feature extraction.
[0086] Step 2: Extracting RF fingerprint features of FSK signals based on instantaneous frequency wavelet transform. The symbols of the samples are divided into two classes. The FSK signals of each class are averaged separately. The instantaneous frequency and wavelet transform of the averaged FSK signals are calculated separately. The amplitude of the wavelet coefficients of the two transformed signals are combined into a vector as fingerprint features.
[0087] Step 3: Use LDA to reduce the dimensionality of fingerprint features from 120 dimensions to 10 dimensions, remove redundant information, improve feature discrimination ability and reduce computational complexity;
[0088] Step 4: Use the LightGBM classifier to classify, train, and test fingerprint features. Divide the fingerprint features of all samples into training and test sets. Train the LightGBM classifier with the feature vectors in the training set, and input the feature vectors in the test set into the trained LightGBM classifier to obtain the test results. By comparing the test results with the actual category results, the average recognition accuracy is obtained.
[0089] Experimental results: The FSK signal RF fingerprint feature extraction method based on instantaneous frequency wavelet transform achieved an average recognition accuracy of 94.2% for 10 radiation sources; the FSK signal instantaneous frequency RF fingerprint feature extraction method achieved an average recognition accuracy of 80.5% for 10 radiation sources.
[0090] The fingerprint feature vector diagram using the present invention is as follows: Figure 5As shown:
[0091] The individual identification confusion matrix of the present invention is as follows: Figure 6 As shown:
[0092] The present invention provides a radio frequency fingerprint feature for FSK signals based on instantaneous frequency wavelet transform, which has good discriminative power and robustness to noise. The signal-to-noise ratio is ≥15dB, and the average recognition accuracy of 10 radiation sources is ≥94.2%. Compared with the instantaneous frequency radio frequency fingerprint feature extraction method for FSK signals, the average recognition accuracy is improved.
[0093] As described above, the present invention can be implemented well.
[0094] All features disclosed in all embodiments of this specification, or steps in all methods or processes implied in the disclosure, may be combined and / or extended or replaced in any way, except for mutually exclusive features and / or steps.
[0095] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Based on the technical essence of the present invention, any simple modifications, equivalent substitutions, and improvements made to the above embodiments within the spirit and principles of the present invention shall still fall within the protection scope of the present invention.
Claims
1. A method for extracting radio frequency fingerprint features from FSK signals, characterized in that, Its characteristic is that it includes the following steps: S1, Feature Extraction: Based on the instantaneous frequency wavelet transform, fingerprint features that can describe the radiation source are extracted, and the wavelet coefficients of the instantaneous frequency of the FSK signal are used as fingerprint features; Step S1 includes the following steps: S11, each sample of the FSK signal Classify the symbols: according to base The symbols are divided into The classified FSK signals are as follows: ； in, This represents the classified FSK signal. Indicates the FSK signal frequency number. , Indicates the total number of FSK signal frequencies. Indicates the first The number of symbols in a signal class Indicates the first In the class of signals, the first Sample points of each symbol; S12, for the categorized... The FSK signals were averaged separately: Each FSK signal of The signals are averaged according to their column vectors to obtain the averaged FSK signal as follows: ； in, This represents the averaged FSK signal. express The The average of the column signals; S13, after averaging One signal Calculate the instantaneous frequency separately, and obtain The instantaneous frequency vectors of the group are as follows: ； in, Represents the instantaneous frequency vector group. Indicates the first A vector of instantaneous frequencies. This represents the total number of extracted instantaneous frequencies; S14, for Group instantaneous frequency Perform wavelet transforms separately to obtain The group of wavelet coefficient vectors are as follows: ； in, Represents the wavelet coefficient vector group. This represents the total number of wavelet coefficients. Indicates the first Wavelet coefficients; S15, The amplitudes of the wavelet coefficients form the eigenvectors. , This refers to the extracted radio frequency fingerprint features; in, Represents the eigenvector. Indicates the first The amplitude of the group of wavelet coefficient vectors.
2. The method for extracting radio frequency fingerprint features from FSK signals according to claim 1, characterized in that, It also includes the following steps: S2, Feature Dimensionality Reduction: Reduces high-dimensional feature vectors to low-dimensional feature vectors.
3. The method for extracting radio frequency fingerprint features from FSK signals according to claim 2, characterized in that, Linear discriminant analysis is used to reduce high-dimensional feature vectors to low-dimensional feature vectors and remove redundant information.
4. A method for extracting radio frequency fingerprint features from an FSK signal according to claim 2 or 3, characterized in that, 。 5. A method for fingerprint recognition using FSK signals, characterized in that, The method for extracting radio frequency fingerprint features from an FSK signal according to any one of claims 2 to 4 further includes the following steps: S3, Classification and Recognition: Using a classifier to classify and recognize feature vectors.
6. The FSK signal fingerprint recognition method according to claim 5, characterized in that, In step S3, a lightweight gradient boosting machine learning classifier is used to divide the fingerprint features of all samples into training and test sets. The LightGBM classifier is trained using the feature vectors in the training set, and the feature vectors in the test set are input into the trained LightGBM classifier to obtain the test results. The average recognition accuracy is obtained by comparing the test results with the actual class results.
7. The FSK signal fingerprint recognition method according to claim 5 or 6, characterized in that, It also includes the following steps preceding step S1: B. Signal preprocessing: The received signal is preprocessed using digital signal processing techniques; the preprocessing includes one or more of the following: filtering, sample synchronization, frequency offset estimation / correction, and normalization.
8. The FSK signal fingerprint recognition method according to claim 7, characterized in that, It also includes the following steps preceding step B: A. Signal reception: The signal received by the antenna is amplified, filtered, frequency-converted, and analog-to-digital converted.
9. An FSK signal fingerprint recognition system, characterized in that, A method for implementing an FSK signal fingerprint recognition method according to any one of claims 5 to 8 includes the following modules connected in sequence: Feature extraction module: used to extract fingerprint features that can describe the radiation source based on instantaneous frequency wavelet transform, with the wavelet coefficients of the instantaneous frequency of the FSK signal as fingerprint features; Feature dimensionality reduction module: used to reduce high-dimensional feature vectors to low-dimensional feature vectors; Classification and recognition module: Used to classify and recognize feature vectors using a classifier.