A radio frequency fingerprint extraction method based on cross-power spectrum

Abstract

The present invention discloses a radio frequency fingerprint extraction method based on mutual power spectrum, comprising the following steps: capturing one or more frames of signals from a received signal; performing symbol synchronization and frame synchronization on all symbols in each frame of the signal; performing phase offset correction and frequency offset correction on each frame of the signal; calculating the mutual power spectrum density of each symbol in a frame of the signal and the local symbol, and calculating the mean thereof to obtain a radio frequency fingerprint; performing feature extraction on the transformed radio frequency fingerprint to identify the device. By using the physical layer characteristics of the signal as the radio frequency fingerprint of the device, the present invention can realize the identity authentication and device identification of the device, is suitable for signals with cyclic shift characteristics, realizes the data independence of the radio frequency fingerprint, is suitable for low signal-to-noise ratio environments, and meets the requirements of long-distance communication.

Description

Technical Field

[0001] The present invention relates to the field of communication and physical layer security technology, and in particular to a radio frequency fingerprint extraction method based on cross-power spectrum. Background Art

[0002] The authentication access of wireless devices is the focus of IoT security issues. The device RF fingerprint feature is an important means to solve the device identity authentication and identification in wireless networks. RF fingerprint refers to the information with device characteristics in the signal emitted by the RF device. The power amplifier, mixer and local crystal oscillator at the transmitter of the RF device will have nonlinear deviations. These deviations are caused by the physical characteristics of the device itself, the composition of electronic components and the internal wiring of the circuit board. They are also related to the degree of use and aging of the device. Even if the same process is used for manufacturing, there will be differences in RF fingerprints due to the tolerance of production standards. Each device has a unique RF fingerprint feature. Its uniqueness makes it part of the physical layer security and difficult to be tampered with or cloned by attackers. Therefore, the RF fingerprint of the access device can be identified and verified to improve the security of communication.

[0003] The connection of IoT depends on different IoT devices, and there are different connection protocols between different devices. When a large number of IoT devices are connected to the system in a short period of time, the growing number of node devices will bring greater pressure to the communication network. The use of traditional one-to-one authentication mechanism will not only make the system network busy, but also occupy a lot of network resources. In addition, computing power will also limit the design of IoT security strategies. Attackers can launch DDOS attacks on devices by increasing communication data and computing power. Physical layer security is a new concept proposed in recent years. It uses unique features extracted from transmitter signals to protect the identity of wireless devices, also known as radio frequency fingerprinting.

[0004] The steady-state RF fingerprint feature is generated by the steady-state part of the signal. The steady-state signal refers to the effective part of the signal, which is mainly composed of the modulated signal. At present, there are many mature extraction schemes at home and abroad, including carrier offset, IQ offset, differential constellation trajectory diagram, etc., which have achieved good performance results in device classification and identification. However, there are also some problems. For example, the carrier offset method is easily affected by environmental changes and equipment aging, and the IQ offset method is susceptible to distance limitations and noise interference. In addition, some methods are easily affected by the content of the data sent by the device, or only use part of the signal segment such as the preamble signal segment for feature extraction. The amount and dimension of the feature data are small, which causes the instability of the RF fingerprint and reduces the accuracy of identity authentication. Summary of the invention

[0005] Technical problem: In order to overcome the deficiencies in the prior art, the present invention provides a method for extracting radio frequency fingerprints based on the cross-power spectrum. The invention uses the physical layer feature cross-power spectrum as the radio frequency fingerprint of the device, which can increase the dimension and data volume of the signal fingerprint feature, and utilizes the same cross-power spectrum characteristics between cyclic shift signals to achieve data independence of the radio frequency fingerprint. The stability of the cross-power spectrum is used to resist environmental noise, which is suitable for low signal-to-noise ratio environments and meets the requirements of long-distance communication.

[0006] Technical solution: To achieve the above purpose, the present invention adopts a radio frequency fingerprint extraction method based on cross-power spectrum, which includes the following steps:

[0007] Step 1, data preprocessing: After reading the signal and normalizing it, perform energy retrieval, remove noise segments and no-signal segments, and obtain valid signal segments;

[0008] Step 2, signal frame capture: perform cross-correlation operation on the valid signal segment and the local reference preamble signal, obtain the data point with the largest peak value as the preamble start data point, and consider that a frame of signal is successfully captured;

[0009] Step 3, signal symbol synchronization: respectively calculate the sliding cross-correlation function between each symbol contained in a frame signal and the local reference signal to obtain the starting point of each symbol;

[0010] Step 4, signal frame synchronization: perform symbol synchronization operation on each captured frame signal;

[0011] Step 5, carrier synchronization: calculate the phase difference between the corresponding points of two adjacent symbols, obtain the average phase deviation, perform frequency offset estimation, and implement frequency offset correction;

[0012] Step 6, superposition noise reduction: superimpose all symbols in a frame signal, remove additive noise through cross-correlation operation, and obtain the cross-correlation coefficient;

[0013] Step 7, feature extraction: Use Fourier transform to transform the mutual correlation coefficient into cross-power spectrum density, sum and divide the cross-power spectrum density of all symbols in a frame signal to obtain the RF fingerprint.

[0014] in:

[0015] In step 2, the complex conjugate correlation value r(n) between the received signal sent to the processor each time and the local reference preamble signal segment is assumed to be:

[0016]

[0017] Where n = 0, 1, ..., M p -1,M p The number of sampling points contained in each preamble signal segment;r (n) is the length of each input to the processor, which is M p The received signal of sample points; p (n) is the local reference preamble signal segment, For x p (n); according to the result, a threshold value TH is set. When r(n)>TH, it is considered that the leading code of a frame signal is captured; then a supplementary amount M is set c , front and back expansion M c Sampling points are slidingly correlated with the local reference preamble code, and the sampling point with the largest cross-correlation value is selected as the starting point of the preamble code, that is, the starting point of this frame signal.

[0018] In step 3, a supplement amount M is set. s , add M before and after each symbol s The sampling points are used for synchronization calculation, and the sampling point with the largest synchronization value is found out, so as to deduce the precise starting time of each symbol and realize accurate signal symbol synchronization.

[0019] In step 5, the average phase difference Δθ between the corresponding points m of adjacent symbols before and after a frame of signal is calculated. l for:

[0020]

[0021]

[0022] Wherein, l = 0, 1, ..., L-1, L is the number of symbols contained in a frame signal; m = 0, 1, ..., M-1, M is the number of sampling points of a symbol, and Δθ is the average phase deviation of this frame signal;

[0023] Therefore, the average frequency deviation Δf is:

[0024]

[0025] Among them, f s is the sampling frequency, M is the number of sampling points of one symbol;

[0026] For the synchronized signal y s (n) performs frequency offset correction, and the corrected signal y(n) is:

[0027]

[0028] Wherein, n=0, 1, ..., M-1, and M is the number of sampling points of one symbol.

[0029] In step 7, the cyclic cross-correlation function r between each symbol y(n) and the local reference symbol x(n) is calculated. c(n), and then use Fourier transform to find the cross power spectrum density R(k) of a symbol:

[0030]

[0031] R(k)=FFT(r c (n))

[0032] Where n = 0, 1, ..., M-1, k = 0, 1, ..., M-1, M is the number of sampling points of a symbol;

[0033] The cross power spectrum density of all symbols in a frame of signal is summed and evenly divided to obtain the radio frequency fingerprint RFF(k):

[0034]

[0035] Wherein, l=0, 1, ..., L-1, and L is the number of symbols contained in a frame signal.

[0036] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0037] (1) The method of identity authentication and device identification based on physical layer characteristics has better security. Using the cross-power spectrum density as the device's RF fingerprint can greatly increase the dimension and data volume of the signal fingerprint feature;

[0038] (2) The cyclically shifted signals have the same cross-power spectrum density, which can achieve data independence of RF fingerprints and use the complete signal frame for fingerprint extraction;

[0039] (3) The cross-power spectrum of the equipment has good stability and can resist additive noise. It is suitable for low signal-to-noise ratio environments and meets the requirements of long-distance communication. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a schematic flow chart of the radio frequency fingerprint extraction method based on cross-power spectrum according to the present invention;

[0041] Figure 2 This is a schematic diagram of the frame structure format of the physical layer signal frame of the LoRa device in the present invention;

[0042] Figure 3 It is a schematic diagram of signal symbol synchronization in the present invention;

[0043] Figure 4 A schematic diagram of the cross-power spectral density difference between the signal in the present invention and the local reference signal.

[0044] Figure 5 It is a two-dimensional schematic diagram of the comparison of characteristic values ​​of the signal in the present invention and the local reference signal. DETAILED DESCRIPTION

[0045] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings:

[0046] The present invention may be implemented in many different forms and should not be considered limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0047] like Figure 1 FIG. 1 is a schematic diagram of the overall process of a radio frequency fingerprint extraction method based on a cross power spectrum proposed by the present invention, and the method specifically comprises the following steps:

[0048] Step 1, data preprocessing: read the demodulated binary baseband signal in the format of Complex64, that is, the real part and the imaginary part are 32 bits each. It can be obtained that the LoRa signal is composed of Chirp signals, and its frame structure contains three parts: preamble, header and data payload. The preamble part contains 14 Up-Chirp signals and 2.25 Down-Chirp signals, and the rest is the result of the standard Up-Chirp signal after cyclic shift. Then, after the data is normalized, the sliding window method is used to retrieve the energy of the data segment. The total energy in a window is calculated to eliminate the influence of occasional high-intensity noise on the total energy of the signal in the window. The energy of a sliding window with a length of 200 is:

[0049]

[0050] Among them, r n is the amplitude value of the nth received signal point, E n is the sum of the energy values ​​of all signal points starting from the nth received signal point in a sliding window of length 200. The window is slid by adding the previous point of the window and removing the last point of the window, so the sum of the energy values ​​of the next window is:

[0051] E n+1 =E n +|r n+1 | 2 -|r n-199 | 2

[0052] When the energy exceeds the set threshold of 0.1, it is considered that the noise segment is successfully removed and the valid signal segment is collected.

[0053] Step 2, signal frame capture: perform cross-correlation operation on the valid signal segment and the local single standard Up-Chirp signal. A standard preamble contains 14 standard Up-Chirp signals, each Chirp signal contains 1024 sampling points. Suppose the received signal with a length of 1024 sample points sent to the processor each time is y r (n), the local single standard Up-Chirp signal is x u (n), the complex conjugate correlation value between the two is calculated as:

[0054]

[0055] in, is x u The conjugate of (n). Set the threshold value TH to 280. When r(n)>TH, it is considered that an Up-Chirp signal is captured. After 14 Up-Chirp signals are captured continuously, it can be considered that the preamble of a frame signal is captured.

[0056] Then, 14 Up-Chirp signals are combined, the supplement amount is set to 40, and 20 sampling points are expanded before and after, and then sliding correlation is performed with the local standard preamble code. The sampling point with the largest cross-correlation value is selected as the starting point of the preamble code, that is, the starting point of this frame signal;

[0057] Step 3, signal symbol synchronization: Calculate the cyclic correlation function between each Chirp symbol in a frame signal and a local single standard Up-Chirp signal using the cyclic shift characteristic to obtain the starting point of each symbol;

[0058] Since the total number of sampling points for each Chirp symbol is fixed at 1024 points, the theoretical starting point of all Chirp symbols can be calculated from the starting point of the preamble code. Due to interference, frequency deviation and other influences in the actual environment, the starting position of each symbol will have a certain offset, so the captured symbols are corrected one by one. Set the supplement amount to 200, add 100 sampling points before and after each symbol, and intercept a total of 1224 sampling points for synchronization calculation. Correspondingly, the Down-Chirp signal part in the received signal is cross-correlated with the standard Down-Chirp signal, and the rest is cross-correlated with the standard Up-Chirp signal.

[0059] Assume that the signal received by 1224 points is y r (n), the local single standard Up-Chirp signal is x u (n), the local single standard Down-Chirp signal is x d (n), its corresponding complex correlation value is:

[0060]

[0061]

[0062] Where, m = 0, 1, ..., 199, x u * (n), x d * (n) are x u (n) and x d (n) is the conjugate. Find r u (m) and r d (m) takes the value of m when it is at its maximum value, and uses this to deduce the start time of each symbol to achieve accurate signal symbol synchronization;

[0063] Step 4, signal frame synchronization: perform symbol synchronization operation on each frame signal in the valid data segment;

[0064] Step 5, carrier synchronization: Assume that a frame signal contains 40 symbols, and calculate the average phase difference between the corresponding points m of the adjacent symbols:

[0065]

[0066] Where l = 0, 1, ..., 39, the average phase deviation of the signal in this frame is:

[0067]

[0068] Then the average frequency deviation is expressed as:

[0069]

[0070] Among them, the sampling frequency f s is 2MHz. For the synchronized signal x s (n) Perform frequency offset correction, and obtain the corrected signal as follows:

[0071]

[0072] Wherein, n = 0, 1, ..., 1023. All signals are corrected according to the results of phase offset estimation and frequency offset estimation to achieve carrier synchronization;

[0073] Step 6, superposition noise reduction: When the LoRa signal is transmitted through the channel, the signal received by the receiver will be affected by additive noise, so the i-th Chirp symbol can be expressed as:

[0074] y i (n) = s i (n)*h(n)+v i (n)

[0075] Where n = 0, 1, ..., 1023, y(n) is the received signal, s(n) is the transmitted signal, assuming that the channel experienced by each symbol is h(n), and v(n) is the additive noise. Then the cross-correlation function between the i-th symbol and the adjacent symbol is:

[0076]

[0077]

[0078] Among them, m = -1023, -1022, ..., -1, 0, 1, 1023, and the additive noise v (n) can be regarded as a stationary random process with a mean of zero, which is uncorrelated with the Chirp signal. Therefore, the cross-correlation coefficient between different symbols can be expressed as:

[0079]

[0080] That is, adjacent Chirp symbols in a frame of LoRa signal are superimposed, and the additive noise v(n) can be removed by cross-correlation operation;

[0081] Step 7, feature extraction: Calculate the cyclic cross-correlation function of each symbol y(n) and the local reference symbol x(n) as:

[0082]

[0083] Where n = 0, 1, ..., 1023. Then the cross power spectrum density of a symbol is obtained by Fourier transform:

[0084] R(k)=FFT(r c (n))

[0085] Where k = 0, 1, ..., 1023. The cross power spectrum density of the forty symbols in a frame of signal is summed and evenly divided to obtain the RF fingerprint:

[0086]

[0087] Among them, l=0,1,…,39.

[0088] It should be noted that the above-described embodiments only express some implementation methods of the present invention, and the description thereof cannot be understood as limiting the scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements can be made without departing from the concept of the present invention, and these should all fall within the scope of protection of the present invention.

Claims

1. A radio frequency fingerprint extraction method based on cross power spectrum, characterized in that: The method comprises the following steps, Step 1, data preprocessing: After reading the signal and normalizing it, perform energy retrieval, remove noise segments and no-signal segments, and obtain valid signal segments; Step 2, signal frame capture: perform cross-correlation operation on the valid signal segment and the local reference preamble signal, obtain the data point with the largest peak value as the preamble start data point, and consider that a frame of signal is successfully captured; Step 3, signal symbol synchronization: respectively calculate the sliding cross-correlation function between each symbol contained in a frame signal and the local reference signal to obtain the starting point of each symbol; Step 4, signal frame synchronization: perform symbol synchronization operation on each captured frame signal; Step 5, carrier synchronization: calculate the phase difference between the corresponding points of two adjacent symbols, obtain the average phase deviation, perform frequency offset estimation, and implement frequency offset correction; Step 6, superposition noise reduction: superimpose all symbols in a frame signal, remove additive noise through cross-correlation operation, and obtain the cross-correlation coefficient; Step 7, feature extraction: transform the mutual correlation coefficient into cross-power spectrum density using Fourier transform, sum and divide the cross-power spectrum density of all symbols in a frame signal to obtain the RF fingerprint; in, In step 2, the complex conjugate correlation value r(n) between the received signal sent to the processor each time and the local reference preamble signal segment is assumed to be: Where n = 0, 1, ..., M p -1,M p The number of sampling points contained in each preamble signal segment; r (n) is the length of each input to the processor, M p The received signal of sample points; x p (n) is the local reference preamble signal segment, For x p (n); according to the result, a threshold value TH is set. When r(n)>fH, it is considered that the leading code of a frame signal is captured; then a supplementary amount M is set c , front and back expansion M c Sampling points are slidingly correlated with the local reference preamble code, and the sampling point with the largest cross-correlation value is selected as the starting point of the preamble code, that is, the starting point of the current frame signal; In step 3, a supplement amount M is set. s , add M before and after each symbol s The sampling points are used for synchronization calculation, and the sampling point with the largest synchronization value is found to deduce the precise starting time of each symbol, thus achieving accurate signal symbol synchronization; In step 5, the average phase difference Δθ between the corresponding points m of adjacent symbols before and after a frame of signal is calculated. l for: Wherein, l = 0, 1, ..., L-1, L is the number of symbols contained in a frame signal; m = 0, 1, ..., M-1, M is the number of sampling points of a symbol, and Δθ is the average phase deviation of this frame signal; Therefore, the average frequency deviation Δf is: Among them, f s is the sampling frequency, M is the number of sampling points of one symbol; For the synchronized signal y s (n) performs frequency offset correction, and the corrected signal y(n) is: Wherein, n=0, 1, ..., M-1, M is the number of sampling points of a symbol; In step 7, the cyclic cross-correlation function r between each symbol y(n) and the local reference symbol x(n) is calculated. c (n), and then use Fourier transform to find the cross power spectrum density R(k) of a symbol: R(k)=FFT(r c (n)) Wherein, n=0, 1, ..., M-1, k=0, 1, ..., M-1, M is the number of sampling points of one symbol; The cross power spectrum density of all symbols in a frame of signal is summed and evenly divided to obtain the radio frequency fingerprint RFF(k): Wherein, l=0, 1, ..., L-1, and L is the number of symbols contained in a frame signal.

Images