High-performance temperature robust Wi-Fi equipment identification method based on artificial radio frequency fingerprints

Abstract

The invention relates to a high-performance temperature robust Wi-Fi equipment identification method based on artificial radio frequency fingerprints. The method comprises the following steps: S1, minimizing the overall misjudgment probability of artificial radio frequency fingerprints of different Wi-Fi equipment; s2, injecting an artificial radio frequency fingerprint into a digital baseband signal of the Wi-Fi equipment; s3, a receiving end captures an original radio frequency signal and extracts an artificial radio frequency fingerprint; and S4, identifying the Wi-Fi device corresponding to the extracted artificial radio frequency fingerprint. The method has the characteristic of equipment temperature fluctuation resistance, and the stability of the equipment identification method is remarkably improved.

Description

Technical Field

[0001] The present invention relates to a high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprint, belonging to the technical field of device identification. Background Art

[0002] With the continuous development of wireless communication technology, the number of Wi-Fi devices has increased annually, and the security of their authentication has also received increasing attention. Radio frequency fingerprinting, as an emerging method for Wi-Fi device authentication, has gained widespread attention in recent years due to its inherent anti-counterfeiting properties, derived from subtle physical differences in the hardware manufacturing process. Compared to traditional device identifiers (such as MAC addresses and IP addresses) and encrypted authentication mechanisms (such as WPA2 and WPA3), radio frequency fingerprinting is more difficult for attackers to forge and offers higher security.

[0003] Existing RF fingerprinting methods can be primarily categorized as deep learning-based and feature engineering-based. Regarding the former, Kong R, Chen H, et al., in their paper "DeepCRF: Deep Learning-Enhanced CSI-Based RF Fingerprinting for Channel-Resilient WiFi Device Identification" (IEEE Transactions on Information Forensics and Security, 2024), proposed a convolutional neural network-based feature extractor. This feature extractor, incorporating advanced techniques such as data augmentation, contrastive learning, and decision fusion, effectively improved the identification performance of Wi-Fi devices. Wang Q, Sun Y, et al., in their paper "Model-Based RF Fingerprint Extraction Approach for Robust IoT Device Identification" (IEEE Internet of Things Journal, 2025), designed an enhanced CNN architecture that combines group convolution and channel fusion modules to further improve the efficiency and robustness of RF fingerprint extraction. However, deep learning-based identification methods still face challenges in practical applications, such as weak generalization across diverse physical environments and insufficient robustness to environmental noise. In contrast, methods based on feature engineering generally perform better in terms of generalization. In their paper "Radio Frequency Fingerprinting for WiFi Devices Using Oscillator Drifts" (published in IEEE Transactions on Instrumentation and Measurement, 2024), Xue C, Li T, et al. proposed a cyclic similarity radio frequency feature based on oscillator drift and used a K-nearest neighbor classifier to perform device identification.

[0004] However, with the continuous improvement of wireless device manufacturing processes, the differences in RF characteristics between devices are becoming increasingly smaller, which can easily lead to misjudgment between devices. At the same time, noise from the device itself and its environment will inevitably be superimposed on the RF signal, significantly interfering with the extraction of RF characteristics and further weakening the distinguishability of fingerprint characteristics between devices. In addition, due to the unstable temperature of the device itself, traditional RF characteristics are also prone to fluctuations. Therefore, it is necessary to reduce the probability of RF characteristics in the device's transmitted signal being misjudged and enhance the temperature robustness of RF characteristics to improve the performance of device identification methods. Summary of the Invention

[0005] The present invention addresses the technical problems existing in the prior art and provides a high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprints. This method is resistant to device temperature fluctuations and significantly improves the stability of the device identification method.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: a high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprint, the method comprising the following steps:

[0007] S1,minimize the overall misjudgment probability of artificial RF fingerprints of different Wi-Fi devices;

[0008] S2, injecting artificial RF fingerprints into the digital baseband signal of the Wi-Fi device;

[0009] S3, the receiving end captures the original RF signal and extracts the artificial RF fingerprint;

[0010] S4. Identify the Wi-Fi device corresponding to the extracted artificial radio frequency fingerprint.

[0011] Wherein, step S1 is specifically as follows:

[0012] S11. Assume that the number of Wi-Fi devices to be identified is M, the number of assignable I / Q imbalance amplitude features is N1, and the number of assignable differential carrier frequency offset (DCFO) features is N2. For the I / Q imbalance amplitude feature, the i-th assigned value is expressed as follows:

[0013]

[0014] Among them, a i Expressed as the i-th I / Q imbalance amplitude, g′ l and g′ r are the minimum and maximum values of the I / Q imbalance amplitude respectively. For the DCFO feature, the i-th allocation value is expressed as follows:

[0015]

[0016] Among them, b i Expressed as the i-th DCFO value, v′ l 、v′ r are the minimum and maximum values of DCFO respectively;

[0017] S12, increase N1 from 1 to M in steps of 1, and N2 is Form multiple combinations of N1 and N2, then calculate the following misjudgment probability objective function, and select N1 and N2 that minimize the objective function as the optimal allocation method:

[0018]

[0019] Where σ1 and σ2 are the standard deviations of the I / Q imbalance amplitude and DCFO, respectively, and erf is the error function. In addition, let N2 be This value ensures that the number of fingerprints required for all devices is met while minimizing the objective function of false positive probability when N1 is a fixed value. This method provides an effective quantitative indicator for the similarity between two-dimensional fingerprints, thereby reducing the false positive probability between assigned RF fingerprints, ultimately enhancing the significance of each fingerprint compared to the rest and improving the performance of subsequent device recognition.

[0020] Among them, S2 is as follows:

[0021] S21. Insert the I / Q imbalance amplitude ε into the short training symbols from index 5 to index 8 in the short training field x of the Wi-Fi device. The insertion method is as follows:

[0022] s[n]=x I (n)+j(1+ε)x Q (n)

[0023] Among them, x I (n) and x Q (n) represents the real and imaginary parts of the signal x[n], j is the imaginary unit, s[n] represents the signal after the I / Q imbalance amplitude is inserted into x[n],

[0024] S22, DCFO characteristic Δ dcfo Insert the short training symbols indexed 5 to 9 into another message. The insertion formula is as follows:

[0025]

[0026] Where k represents the index number of the short training symbol, s[k] is the signal corresponding to index k in the short training field, and s ′ [k] is the signal after s[k] is inserted into DCFO, j is the imaginary unit, f s is the sampling rate of the Wi-Fi device. The DCFO signature has greater temperature stability than the traditional Carrier Frequency Offset (CFO) signature, helping to improve subsequent device identification performance. Furthermore, actively inserting a value into the digital baseband signal to adjust the RF fingerprint eliminates the need for hardware modification, effectively reducing deployment costs.

[0027] Among them, S3, the receiving end captures the original RF signal and extracts the artificial RF fingerprint, as follows:

[0028] S31, the receiving end uses traditional signal processing technologies such as signal detection and time synchronization to accurately locate the exact position of the short training field in all received I / Q signals and extract it.

[0029] S32, I / Q imbalance amplitude estimation is to use the received short training symbols from index 1 to index 4 as reference OFDM symbol O, and the received short training symbols from index 5 to index 8 as inserted OFDM symbol O ′ , the estimated value of the I / Q imbalance amplitude ε ′ as follows:

[0030]

[0031] Among them, R and R ′ are OFDM symbols O and O respectively. ′ After Fourier transform, the frequency domain representation is C -1 is the inverse function of the inherent distortion function of the transmitting device,

[0032] S33. Calculate the CFO corresponding to the first five short training symbols (index 0 to 4) and the last five short training symbols (index 5 to 9), wherein the CFO radio frequency feature f1 corresponding to the first five short training symbols is:

[0033]

[0034] Among them, r s Represents the received short training field, the ∠ operator represents the complex argument operation, m represents the relative position of the signal in the short training symbol, and k represents the index number of the short training symbol;

[0035] The CFO radio frequency characteristics f2 in the last 5 short training code elements are as follows:

[0036]

[0037] DCFO value Δ dcfo The calculation method of ′ is as follows:

[0038] Δ dcfo ′=f2-f1

[0039] By subtracting the CFO of the last five short training symbols from the CFO of the first five short training symbols, the temperature-related CFO component can be eliminated, thereby obtaining a temperature-independent DCFO feature added at the Wi-Fi device.

[0040] Wherein, the step S4 is specifically as follows:

[0041] Generate an I / Q imbalance amplitude set according to the steps in S1 and DCFO collection Corresponding I / Q imbalance amplitude index set and DCFO index collection When the I / Q imbalance amplitude and DCFO corresponding index are y a,i and y b,j When , the extracted I / Q imbalance amplitude and DCFO probability of t and g are respectively expressed as follows: I / Q imbalance amplitude:

[0042]

[0043] DCFO features:

[0044]

[0045] Among them, σ a and σ b are the standard deviations of the I / Q imbalance amplitude RF characteristics and the DCFO RF characteristics, respectively. The exp operator is an exponential operator.

[0046] Since P(Y=y a,i )、P(X a =t), P(Y=y b,j ) and P(X b =g) are fixed constants, and the index number corresponding to the I / Q imbalance amplitude t and the DCFO feature g is jointly estimated to obtain the corresponding index number estimate value and The specific method is shown below:

[0047]

[0048] where y a,i ∈Y a Represents index y a,i Belongs to set Y a ,y b,j ∈Y b Represents index y b,j Belongs to set X b , the argmax operator means finding the category y a,i ∈Y a and y b,j ∈Y b The combination of P(X a =t|Y=y a,i )+P(X b =g|Y=y b,j) is the largest, and the index number of the device finally identified is

[0049] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprint is implemented.

[0050] A computer-readable storage medium stores computer instructions, which, when executed by a processor, implement the high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprints.

[0051] The beneficial effects of the present invention are:

[0052] (1) The DCFO feature inserted by the present invention eliminates the components of the traditional CFO related to temperature fluctuations and has stability under temperature fluctuations, thereby significantly improving the accuracy of the device identification method.

[0053] (2) The present invention proposes a similarity quantification index between the two-dimensional RF fingerprint formed by the I / Q imbalance amplitude RF feature and the DCFO RF feature, and designs a two-dimensional feature joint optimal allocation method with low computational cost, so as to minimize the probability of the device's RF fingerprint being misjudged.

[0054] (3) This invention uses a device identification method based on a naive Bayesian classifier, which can effectively reduce the training and inference time of device models compared to deep learning models. In addition, this method has a small parameter size and low memory usage, and can be deployed on devices with limited computing resources.

[0055] (4) The two types of artificial RF fingerprints in the present invention are inserted only in the short training field, without destroying the channel state required in the signal demodulation process, thereby ensuring that the Wi-Fi device inserted with the artificial RF fingerprint can still communicate normally with the existing Wi-Fi device and maintain its original communication rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flowchart of a high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprint proposed by the present invention.

[0057] Figure 2 The present invention extracts the differential carrier frequency offset block diagram at the receiving end,

[0058] Figure 3 This is a performance diagram of the device identification under different temperatures of the present invention.

[0059] Figure 4 This is a performance diagram of device identification in different time periods of the present invention.

[0060] Figure 5 This is a schematic diagram of the device recognition performance at different distances according to the present invention. DETAILED DESCRIPTION

[0061] In order to deepen the understanding of the present invention, this embodiment is described in detail below with reference to the accompanying drawings.

[0062] Example: Figure 1 As shown in FIG, a high-performance temperature-robust Wi-Fi device identification method based on artificial RF fingerprint includes the following specific steps:

[0063] S1,minimize the overall misjudgment probability of artificial RF fingerprints of different Wi-Fi devices;

[0064] S2, injecting artificial RF fingerprints into the digital baseband signal of the Wi-Fi device;

[0065] S3, the receiving end captures the original RF signal and extracts the artificial RF fingerprint;

[0066] S4, identifying the Wi-Fi device corresponding to the extracted artificial RF fingerprint;

[0067] The step S1 is specifically as follows:

[0068] S11. Assume that the number of Wi-Fi devices to be identified is M, the number of I / Q imbalance amplitude features that can be assigned is N1, and the number of differential carrier frequency offset (DCFO) features that can be assigned is N2. To minimize the probability of misjudgment between single-class features, for the I / Q imbalance amplitude feature, the i-th assigned value is expressed as follows:

[0069]

[0070] Among them, a i Expressed as the i-th I / Q imbalance amplitude, g′ l and g′ r are the minimum and maximum values of the I / Q imbalance amplitude respectively. Meanwhile, for the DCFO feature, the i-th allocation value is expressed as follows:

[0071]

[0072] Among them, b i Expressed as the i-th DCFO value, v′ l 、v′ r are the minimum and maximum values of DCFO respectively.

[0073] S12, increase N1 from 1 to M in steps of 1. At the same time, N2 is Multiple combinations of N1 and N2 can be formed. Then calculate the following misjudgment probability objective function and select N1 and N2 that minimize the objective function as the optimal allocation method:

[0074]

[0075] Where σ1 and σ2 are the standard deviations of the I / Q imbalance amplitude and DCFO, respectively, and erf is the error function.

[0076] The step S2 is specifically as follows:

[0077] S21. Insert the I / Q imbalance amplitude ε into the short training symbols from index 5 to index 8 in the short training field x of the Wi-Fi device. The insertion method is as follows:

[0078] s[n]=x I (n)+j(1+ε)x Q (n)

[0079] Among them, x I (n) and x Q (n) represents the real and imaginary parts of the signal x[n], j is the imaginary unit, and s[n] represents the signal after the I / Q imbalance amplitude is inserted into x[n].

[0080] S22, DCFO characteristic Δ dcfo Insert the short training symbols indexed from 5 to 9 into another message. The insertion formula is as follows:

[0081]

[0082] Where k represents the index number of the short training symbol, s[k] is the signal corresponding to index k in the short training field, s′[k] is the signal after s[k] is inserted into DCFO, j is the imaginary unit, and f s is the sampling rate of the Wi-Fi device.

[0083] The step S3 is as follows Figure 2 As shown:

[0084] S31. The receiving end uses traditional signal processing technologies such as signal detection and time synchronization to accurately locate the exact position of the short training field in all received I / Q signals and extract it.

[0085] S32, I / Q imbalance amplitude estimation is to use the received short training symbols from index 1 to index 4 as reference OFDM symbol O, and the received short training symbols from index 5 to index 8 as inserted OFDM symbol O ′The estimated value of the I / Q imbalance magnitude ε ′ as follows:

[0086]

[0087] Among them, R and R ′ are OFDM symbols O and O respectively. ′ After Fourier transform, the frequency domain representation is C -1 It is the inverse function of the inherent distortion function of the transmitting device.

[0088] S33. Calculate the carrier frequency offset (CFO) corresponding to the first five short training symbols (indexes 0 to 4) and the last five short training symbols (indexes 5 to 9), where the CFO radio frequency feature f1 corresponding to the first five short training symbols is:

[0089]

[0090] Among them, r s Represents the received short training field, the ∠ operator represents the complex argument operation, m represents the relative position of the signal in the short training codeword, and k represents the index number of the short training codeword.

[0091] Similar to the calculation of f1, the CFO RF feature f2 in the last 5 short training symbols is as follows:

[0092]

[0093] DCFO value Δ dcfo The calculation method of ′ is as follows:

[0094] Δ dcfo ′=f2-f1

[0095] The step S4 is specifically as follows:

[0096] Generate an I / Q imbalance amplitude set according to the steps in S1 and DCFO collection Corresponding I / Q imbalance amplitude index set and DCFO index collection When the I / Q imbalance amplitude and DCFO corresponding index are y a,i and y b,j When , the extracted I / Q imbalance amplitude and DCFO probability of t and g are respectively expressed as follows: I / Q imbalance amplitude:

[0097]

[0098] DCFO features:

[0099]

[0100] Among them, σ a and σ b are the standard deviations of the I / Q imbalance amplitude RF characteristics and the DCFO RF characteristics, respectively. The exp operator is an exponential operator.

[0101] Since P(Y=y a,i )、P(X a =t), P(Y=y b,j ) and P(X b =g) are fixed constants, and the index number corresponding to the I / Q imbalance amplitude t and the DCFO feature g can be jointly estimated to obtain the corresponding index number estimate value and The specific method is shown below:

[0102]

[0103] where y a,i ∈Y a Represents index y a,i Belongs to set Y a ,y b,j ∈Y b Represents index y b,j Belongs to set Y b , the argmax operator means finding the category y a,i ∈Y a and y b,j ∈Y b The combination of P(X a =t|Y=y a,i )+P(X b =g|Y=y b,j )maximum.

[0104] The index number of the device finally identified is like Figure 3 、 4 As shown in Figures 5 and 5, the device identification method proposed in the present invention can achieve a precision rate, a recall rate, and an F1-score of 98% under the conditions of different temperatures, different time periods, and different propagation distances, and has good device identification performance.

[0105] It should be noted that the above embodiments are not intended to limit the scope of protection of the present invention, and equivalent changes or substitutions made on the basis of the above technical solutions fall within the scope of protection of the claims of the present invention.

Claims

1. A high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprint, characterized in that: The method comprises the following steps: S1,minimize the overall misjudgment probability of artificial RF fingerprints of different Wi-Fi devices; S2, injecting artificial RF fingerprints into the digital baseband signal of the Wi-Fi device; S3, the receiving end captures the original RF signal and extracts the artificial RF fingerprint; S4. Identify the Wi-Fi device corresponding to the extracted artificial radio frequency fingerprint.

2. The high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprint according to claim 1 is characterized in that: Step S1: Minimize the overall misjudgment probability of artificial RF fingerprints of different Wi-Fi devices, as follows: S11. Assume that the number of Wi-Fi devices to be identified is M, the number of assignable I / Q imbalance amplitude features is N1, and the number of assignable differential carrier frequency offset (DCFO) features is N2. For the I / Q imbalance amplitude feature, the i-th assigned value is expressed as follows: Among them, a i Expressed as the i-th I / Q imbalance amplitude, g′ l and g′ r are the minimum and maximum values of the I / Q imbalance amplitude respectively. For the DCFO feature, the i-th allocation value is expressed as follows: Among them, b i Expressed as the i-th DCFO value, v′ l 、v′ r are the minimum and maximum values of DCFO respectively; S12, increase N1 from 1 to M in steps of 1, and N2 is Form multiple combinations of N1 and N2, then calculate the following misjudgment probability objective function, and select N1 and N2 that minimize the objective function as the optimal allocation method: Where σ1 and σ2 are the standard deviations of the I / Q imbalance amplitude and DCFO, respectively, and erf is the error function.

3. The high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprint according to claim 2 is characterized in that: S2. Inject an artificial RF fingerprint into the digital baseband signal of the Wi-Fi device as follows: S21. Insert the I / Q imbalance amplitude ε into the short training symbols from index 5 to index 8 in the short training field x of the Wi-Fi device. The insertion method is as follows: s[n]=x I (n)+j(1+ε)x Q (n) Among them, x I (n) and x Q (n) represents the real and imaginary parts of the signal x[n], j is the imaginary unit, s[n] represents the signal after the I / Q imbalance amplitude is inserted into x[n], S22, DCFO characteristic Δ dcfo Insert the short training symbols indexed 5 to 9 into another message. The insertion formula is as follows: Where k represents the index number of the short training symbol, s[k] is the signal corresponding to index k in the short training field, s′[k] is the signal after s[k] is inserted into DCFO, j is the imaginary unit, and f s is the sampling rate of the Wi-Fi device.

4. The high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprint according to claim 2 is characterized in that: S3, the receiving end captures the original RF signal and extracts the artificial RF fingerprint, as follows: S31, the receiving end uses traditional signal processing technologies such as signal detection and time synchronization to accurately locate the exact position of the short training field in all received I / Q signals and extract it. S32, I / Q imbalance amplitude estimation is to use the received short training symbols from index 1 to index 4 as reference OFDM symbol O, and the received short training symbols from index 5 to index 8 as inserted OFDM symbol O′. The estimated value ε′ of the I / Q imbalance amplitude is as follows: Among them, R and R′ are the frequency domain representations of OFDM symbols O and O′ after Fourier transform, C -1 is the inverse function of the inherent distortion function of the transmitting device, S33. Calculate the carrier frequency offset (CFO) corresponding to the first five short training symbols, i.e., indexes 0 to 4, and the last five short training symbols, i.e., indexes 5 to 9. The CFO radio frequency feature f1 corresponding to the first five short training symbols is: Among them, r s Represents the received short training field, the ∠ operator represents the complex argument operation, m represents the relative position of the signal in the short training symbol, and k represents the index number of the short training symbol; The CFO radio frequency characteristics f2 in the last 5 short training code elements are as follows: DCFO value Δ dcfo The calculation method of ′ is as follows: Δ dcfo ′=f2-f1。 5. The high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprint according to claim 2 is characterized in that: The step S4 is specifically as follows: Generate an I / Q imbalance amplitude set according to the steps in S1 and DCFO collection Corresponding I / Q imbalance amplitude index set and DCFO index collection When the I / Q imbalance amplitude and DCFO corresponding index are y a,i and y b,j When , the extracted I / Q imbalance amplitude and DCFO probability of t and g are respectively as follows: I / Q imbalance magnitude: DCFO features: Among them, σ a and σ b are the standard deviations of the I / Q imbalance amplitude RF characteristics and the DCFO RF characteristics, respectively. The exp operator is an exponential operator. Since P(Y=y a,i )、P(X a =t), P(Y=y b,j ) and P(X b =g) are fixed constants, and the index number corresponding to the I / Q imbalance amplitude t and the DCFO feature g is jointly estimated to obtain the corresponding index number estimate value and The specific method is shown below: Among them, y a,i ∈Y a Represents index y a,i Belongs to set Y a ,y b,j ∈Y b Represents index y b,j Belongs to set Y b , the argmax operator means finding the category y a,i ∈Y a and y b,j ∈Y b The combination of P(X a =t|Y=y a,i )+P(X b =g|Y=y b,j ) is the largest, and the index number of the device finally identified is 6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprint is implemented as described in any one of claims 1 to 5.

7. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, a high-performance temperature-robust Wi-Fi device identification method based on artificial radio frequency fingerprint is implemented as described in any one of claims 1 to 5.

Images