GNSS detection method and system based on machine learning

Abstract

The invention relates to the technical field of signal analysis, in particular to a GNSS (Global Navigation Satellite System) detection method and system based on machine learning, and the method comprises the steps: receiving a navigation radio-frequency signal containing reality and deception, and carrying out the down-conversion of an RF (Radio Frequency) front end to obtain intermediate-frequency data; gNSS signal capture is completed by searching correlation peaks in frequency and code phases based on intermediate frequency data; in the capturing stage, a motion state, a correlation peak number, a noise value, capturing parameters and peak + / -1 chip sampling features are extracted and input into a machine learning classification model to judge whether cheating exists or not; if so, separating the deception signal from the real signal, and respectively sending the deception signal and the real signal into independent tracking channels; measuring an array element carrier relative phase by an array antenna, establishing a phase center model, and correcting to obtain an effective phase center and a differential baseline; according to the method, a deception azimuth angle and a pitch angle are obtained, a beam weight is calculated, a deception direction is subjected to null setting, a real direction is subjected to main lobe protection, and a weight value is fed back in real time to improve a signal carrier-to-noise ratio and locking stability. According to the invention, the problem that the existing commercial GNSS signal is low in power and easy to be interfered by deception is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal analysis, and in particular to a GNSS detection method and system based on machine learning. BACKGROUND

[0002] The global navigation satellite system (GNSS) civilian signal has low power, open system and is easy to be deceived, which leads to the output of incorrect position and time results by the receiver, and is very harmful. Existing deception detection relies on tracking / positioning stage features or external information, and many direction finding schemes rely on special array antennas, which are high in cost and not conducive to large-scale deployment. In order to realize sensitive and high-precision direction finding of the deception source, deception detection needs to be completed in the acquisition stage, so that the true / false signals can be processed separately in the subsequent tracking stage.

[0003] On the other hand, commercial array antennas are often composed of microstrip / spiral elements, and the phase centers of the array elements are offset at different incident directions. The mutual coupling and edge effect can worsen the stability of the phase center, which significantly increases the short baseline direction finding error (the carrier measurement change introduced by the single antenna phase center can reach about 40°). The phase center correction model (PCO / PCV) proposed by the International GNSS Service (IGS) and its array differential processing can be used to reduce the direction measurement deviation, but it needs to be modeled and calibrated for array elements and incident directions. SUMMARY

[0004] In view of the above technical problems, the present application provides a GNSS detection method and system based on machine learning to solve the problems of low signal power of existing commercial GNSS, easy to be deceived, existing detection lagging behind tracking / positioning stage, difficult to separate true / false signals in time, and relying on special arrays leading to high cost, and commercial array phase center instability causing large short baseline direction finding error.

[0005] Other characteristics and advantages of the present application will become apparent from the following detailed description, or will be learned by practice of the present application.

[0006] According to an aspect of the present application, a GNSS detection method based on machine learning is provided, which comprises: receiving navigation radio frequency signals containing real signals and deception signals, and obtaining intermediate frequency data through RF front-end down-conversion processing; based on the intermediate frequency data, searching for correlation peaks in the frequency and code phase dimensions to acquire GNSS satellite signals; Extract a plurality of feature parameters generated at the time of capture as the basis for spoof detection, including the motion state of the receiver, the number of correlation peaks captured, the average noise value, the capture parameters used to determine satellite capture, and the sampling correlation peak values within ±1 chip range around the maximum correlation peak value; input the extracted plurality of feature parameters into a pre-trained machine learning classification model for analysis to determine whether spoof interference exists in the GNSS satellite signal; If the determination result exists spoof signal interference, separate the spoof signal from the real signal, and send the spoof signal and the real signal into independent tracking channels respectively; Perform direction measurement on the determined spoof signal using an array antenna, measure the relative phase difference between the spoof signal carriers received by each antenna element in the array, perform differential phase center correction on the measured relative phase difference, establish a phase center model for each antenna element in the array, which includes the fixed phase center offset of the antenna element and the phase center deviation varying with the incident direction, correct the geometric installation coordinates of each antenna element to effective phase center coordinates according to the phase center model, and calculate the differential phase center baseline vector between the reference antenna element and other antenna elements; Based on the differential phase center baseline vector and the relative phase difference, calculate the incident direction angle of the spoof signal, obtain the azimuth and elevation angles of the spoof signal, and use the incident azimuth angle and the elevation angle as control parameters for spatial suppression, solve and apply the beamforming weights of the array antenna, so that a null is formed in the incident direction of the spoof signal and a main lobe is formed in the incident direction of the real signal, and the updated weights are fed back to the real signal tracking channel in real time to improve the carrier-to-noise ratio and locking stability.

[0007] Further, the machine learning classification model is constructed using the gradient boosting decision tree algorithm and is trained offline based on historical data of GNSS signal spoofing interference scenarios, so that the machine learning classification model can classify and determine the capture result according to a plurality of feature parameters to detect the existence of spoof signals.

[0008] Further, the classification evaluation of the machine learning model includes accuracy, precision, recall, and F1 score, wherein the accuracy is the sum of true positives and true negatives divided by the sum of true positives, true negatives, false positives, and false negatives; the precision is the true positives divided by the sum of true positives and false positives; the recall is the true positives divided by the sum of true positives and false negatives; and the F1 score is twice the harmonic mean of the precision and the recall.

[0009] Further, during the direction measurement of the spoofing signal, the fixed phase difference and time delay difference between each antenna unit of the array are calibrated and compensated, and the random phase fluctuation introduced by thermal noise is reduced by reducing the carrier tracking loop bandwidth and smoothing the phase measurement data, so that the jitter of the measured relative phase difference is reduced.

[0010] Further, the fixed phase center offset is a three-dimensional constant vector of each antenna unit in the array reference coordinate system, which is used to compensate for manufacturing and installation errors and channel average time delay difference; the phase center deviation is a direction-related vector function of incident azimuth and elevation angle, which is used to represent the transient deviation caused by mutual coupling and reflection edge effect of the array element; the geometric installation coordinates of each antenna unit are corrected to effective phase center coordinates according to the fixed phase center offset and the phase center deviation, and the differential phase center baseline vector is constructed accordingly.

[0011] Further, the fixed phase center offset and the phase center deviation are obtained by calibration, and in the calibration, a calibration source with known direction or a navigation satellite with known spatial position is used as a natural calibration source, and the phase observation of each antenna unit is obtained by traversing multiple sampling points. After removing the channel fixed time delay, the fixed phase center offset and the phase center deviation are fitted; the phase center model sets an index according to the working frequency band, and stores and calls the corresponding fixed phase center offset and phase center deviation for correction in different frequency bands.

[0012] Further, the calculation of the differential phase center baseline vector specifically includes: The geometric installation coordinates of each antenna unit in the array are added to the phase center offset value of the antenna unit, and the corresponding phase center change value is superimposed according to the incident direction of the spoofing signal, to obtain the effective phase center coordinates of each antenna unit; taking a certain antenna unit in the array as a reference, the differential phase center baseline vector corresponding to the reference antenna unit and any other antenna unit is calculated according to the difference between the effective phase center coordinates of the reference antenna unit and the other antenna unit.

[0013] According to another aspect of the present application, a GNSS detection system based on machine learning is provided, comprising: The acquisition module is used for receiving navigation radio frequency signals containing real signals and spoofing signals, and obtaining intermediate frequency data through RF front-end down-conversion processing; based on the intermediate frequency data, searching for correlation peaks in frequency and code phase dimensions to acquire GNSS satellite signals; A feature analysis module is configured to extract a plurality of feature parameters generated during the capturing as the basis for spoofing detection, the feature parameters including the motion state of the receiver, the number of correlation peaks captured, the average noise value, the capture parameters for determining satellite capture, and the sampling correlation peak values within a range of ±1 chip around the maximum correlation peak value; the plurality of extracted feature parameters are input into a pre-trained machine learning classification model for analysis to determine whether spoofing interference exists in the GNSS satellite signal; A signal separation module is configured to separate the spoofing signal from the real signal when the determination result indicates that spoofing signal interference exists, and send the spoofing signal and the real signal into independent tracking channels respectively. A direction finding observation module is configured to perform direction measurement on the determined spoofing signal by using an array antenna, measure the relative phase difference between the spoofing signal carriers received by each antenna element of the array, perform differential phase center correction on the measured relative phase difference, establish a phase center model of each antenna element in the array, the phase center model including a fixed phase center offset of the antenna element and a phase center deviation varying with the incident direction, correct the geometric installation coordinates of each antenna element to effective phase center coordinates according to the phase center model, and calculate the differential phase center baseline vector between the reference antenna element and other antenna elements. A DOA solving and spatial domain suppression module is configured to calculate the incident direction angle of the spoofing signal based on the differential phase center baseline vector and the relative phase difference, obtain the incident azimuth angle and the incident elevation angle of the spoofing signal, and use the incident azimuth angle and the incident elevation angle as control parameters for spatial domain suppression, solve and apply the beamforming weight values of the array antenna, so that a null is formed in the incident direction of the spoofing signal, a main lobe is formed in the incident direction of the real signal, and the updated weight values are fed back to the real signal tracking channel in real time to improve the carrier-to-noise ratio and the locking stability.

[0014] The technical solution of the present application has the following beneficial effects: The present application extracts multiple source features in the capturing stage and uses machine learning classification, so that whether spoofing interference exists can be determined without relying on the tracking / positioning link; the features used can effectively cover the peak shape change and noise rise caused by large / small delay spoofing, so that the real / fake signal can be separated before entering the tracking, mutual pollution is reduced, and pure observation is provided for subsequent direction finding and suppression.

[0015] In the aspect of direction finding, the application proposes a differential phase center correction for commercial arrays: compensating for both average and instantaneous phase center deviations, significantly reducing system errors caused by element mutual coupling and structural edge effects, improving the direction measurement accuracy of deception signals, and forming an integrated process with the receiver's acquisition / detection / tracking / finding modules, which can further drive beamforming to achieve deception direction nulling and true star direction main lobe. Compared with the scheme relying on special arrays, it has the engineering advantages of lower cost and scalable deployment. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flowchart of a machine learning-based GNSS detection method in an embodiment of the present specification; Figure 2 A structural block diagram of a machine learning-based GNSS detection system in an embodiment of the present specification. DETAILED DESCRIPTION

[0017] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these implementations are provided so that this disclosure will be thorough and complete, and will fully convey the scope of example implementations to those skilled in the art. The described features, structures, or characteristics can be combined in one or more implementations. In the following description, numerous specific details are provided to give a thorough understanding of implementations of the application. One skilled in the relevant art will recognize, however, that the implementations can be practiced without one or more of the specific details, or with other methods, components, materials, and so forth. In other instances, well-known structures have not been described in detail to avoid obscuring aspects of the application.

[0018] Furthermore, the accompanying drawings are only schematic and are non-limiting. Like references signs denote like parts, and thus corresponding specifics are omitted. Some of the blocks in the drawings are functional entities that do not necessarily have to be physically or logically independent entities. These functional entities can be implemented in software, or in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0019] The present application provides a product-based machine learning GNSS detection method. Referring to Figure 1The diagram shown illustrates a flowchart of a machine learning-based GNSS detection method for a product according to an embodiment of the present invention. This method can be applied to electronic devices such as personal computers and servers. The method can be executed by a device, which can be implemented by software and / or hardware. Specifically, the method may include the following steps S101-S103: In step S101, a navigation radio frequency signal containing real and deceptive signals is received, and intermediate frequency data is obtained through RF front-end downconversion processing; based on the intermediate frequency data, correlation peaks are searched in the frequency and code phase dimensions to capture GNSS satellite signals.

[0020] In this step, the receiver first receives navigation radio frequency (RF) signals, which include real GNSS signals and spoofing signals, in the form of an array or a single antenna. Then, it performs conventional analog / digital-analog processing such as down-conversion and sampling on the RF signals and outputs intermediate frequency (IF) data. Based on the IF data, it scans in two dimensions, Doppler frequency and code phase, to identify and lock the correlation peaks of the GNSS satellite signals. The acquisition module scans the frequency and code phase to identify and lock the GNSS signals.

[0021] Specifically, in an environment where deception exists, the signal received by the receiver is a combination of the real signal and the deception. Deceptive signals With additive white Gaussian noise The superposition is expressed as:

[0022]

[0023] ; N is the number of visible satellites. It is the first Index of satellites, It is the first The actual GNSS signal power of the satellite, It is the propagation delay of the signal from the nth satellite. It is the first The pseudocode sequence of satellite signals, It is the first Data code information modulated by satellite signals, It is the first The intermediate frequency of satellite signals, It is the first Doppler shift of satellite signals It is the first The initial carrier phase of the satellite signal, where m is the number of spoofing signals. It is the signal power of the k-th spoofing signal. is the code sequence of the kth spoofing signal, is the propagation delay of the kth spoofing signal, is the data code information modulated by the kth spoofing signal, is the intermediate frequency of the kth spoofing signal, is the Doppler shift of the kth spoofing signal, is the initial carrier phase of the kth spoofing signal.

[0024] In the acquisition stage, the interference integrator output is affected by the Doppler frequency of the local carrier and the code phase of the local C / A code. The interference integrator output of the kth satellite channel is represented as:

[0025]

[0026] where, and are the frequency difference and phase difference between the real signal of the kth satellite and the local carrier, respectively; and are the frequency difference and phase difference between the spoofing signal corresponding to the kth satellite and the local carrier, respectively. is the interference integration time; is the normalized autocorrelation function of the C / A code, represented as: represents the correlation noise, which is composed of the interference integration results of the noise, the real signals from other satellites, the spoofing signals and the locally generated signals, and can be represented as:

[0027] where, and represent the real signal and the spoofing signal of the jth satellite, respectively. By traversing the (f, τ) two-dimensional grid, a two-dimensional correlation energy map can be formed and the correlation peak can be found accordingly.

[0028] ​​​​​In step S102, multiple feature parameters generated during acquisition are extracted as the basis for spoofing detection. The feature parameters include the receiver's motion state, the number of captured correlation peaks, the average noise value, the acquisition parameters used to determine satellite acquisition, and the sampled correlation peaks within ±1 chip range around the maximum correlation peak. The extracted multiple feature parameters are input into a pre-trained machine learning classification model for analysis to determine whether spoofing interference exists in the GNSS satellite signal.

[0029] In feature extraction, the receiver's state, whether stationary or mobile, affects the frequency domain search during the acquisition phase; therefore, the receiver's state is considered a feature. Without deception, the receiver's acquisition result shows only a single correlation peak. Specifically, the correlation peak distribution obtained from the two-dimensional search directly characterizes whether the target satellite and its approximate code phase and frequency position have been acquired: Without deception interference, the acquisition result presents a single correlation peak; when the time delay difference between the deception signal and the real signal exceeds one chip, a "double peak" phenomenon appears during the acquisition phase; while when the time delay difference is less than one chip, although no double peak appears, the correlation shape around the maximum peak (within ±1 chip range) will exhibit observable distortion; therefore, the number of correlation peaks is selected as one of the features. Furthermore, deception affects the receiver's noise characteristics. Therefore, the noise mean over a period of time is selected as one of the features.

[0030] To determine whether a satellite has been acquired, we can assess whether the acquisition parameters exceed a certain threshold. The acquisition parameters can be expressed as: ; Here, MaxCorPeak is the maximum correlation peak value within a chip range, MeanCor is the correlation mean within the same chip after removing the maximum value, and StdCor is the correlation standard deviation within the same chip. When this parameter exceeds a threshold, the corresponding (f,τ) grid point is considered to have completed capture. Since this indicator is entirely based on the correlation results of the capture phase, it is used as one of the features.

[0031] Under normal, deception-free conditions, the capture output should exhibit an ideal autocorrelation function along the code phase direction. It is a single sharp main peak with a shape; but when the time delay difference between the deceptive signal and the real signal... At the 1-chip stage, the coherent integral output of the acquisition correlator becomes a "vector superposition of two closely related terms." Its peak shape is no longer equivalent to a standard triangular autocorrelation curve, but exhibits morphological changes such as a broadened peak, raised "shoulders," curve asymmetry, and a slight shift in the main peak position. This change can be quantified using statistics from the acquisition stage; therefore, the sampled correlation peak value within ±1 chip range around the maximum correlation peak can be used as one of the features. Specifically, in order to achieve... The nine correlation peaks near the maximum correlation peak are selected, and the chip differences of the nine correlation peaks are: -1, -0.75, -0.5, -0.25, 0, 0.25, 0.5, 0.75, and 1, and the nine correlation peaks are taken as nine features.

[0032] The machine learning classification model is constructed by using a gradient boosting decision tree algorithm, and is trained offline based on historical data of GNSS signal spoofing interference scenes, so that the machine learning classification model can classify and judge the capture result according to the plurality of feature parameters to detect the presence of a spoofing signal. The machine learning classification model can be constructed by using a gradient boosting decision tree implemented by LightGBM, and the real signal is marked as 1 and the spoofing signal is marked as 2. The offline training is completed by using the above multi-dimensional features and the labeled historical data.

[0033] In step S103, if the discrimination result exists spoofing signal interference, the spoofing signal is separated from the real signal, and the spoofing signal and the real signal are respectively sent into independent tracking channels.

[0034] The discrimination result of the capture stage is used as a basis for distinguishing the real / spoifing signals, and the observation containing the spoofing component and the real component are processed separately in the tracking stage. Specifically, the detection in the capture stage enables the real GNSS signal and the spoofing signal to be processed separately in the tracking stage, so that independent tracking channels are established and initialized according to the categories. The tracking module for the spoofing signal and the tracking module for the real signal can be set correspondingly to ensure the continuous and accurate synchronization of the two types of signals. After the two types of signals are identified in the capture stage, the real and spoofing signals can be tracked simultaneously to realize physical separation and parallel tracking at the channel level, so as to avoid mutual interference of the two types of signals in the same channel and leave independent measurement and state quantity output for subsequent processing.

[0035] In step S104, the spoofing signal determined in step S103 is subjected to direction measurement by using an array antenna. The relative phase difference between the spoofing signal carriers received by each antenna element of the array is measured, the measured relative phase difference is subjected to differential phase center correction, a phase center model of each antenna element in the array is established, the phase center model includes a fixed phase center offset of the antenna element and a phase center deviation varying with the incident direction, the geometric installation coordinates of each antenna element are corrected to effective phase center coordinates according to the phase center model, and a differential phase center baseline vector between the reference antenna element and other antenna elements is calculated.

[0036] In the direction measurement of the spoofing signal, the fixed phase difference and time delay difference between the antenna elements of the array are calibrated and compensated, and the random phase fluctuation caused by thermal noise is reduced by reducing the carrier tracking loop bandwidth and smoothing the phase measurement data, so that the jitter of the measured relative phase difference is reduced.

[0037] Specifically, for the aforementioned determined spoofing signal, the receiver uses the array antenna to perform direction measurement, first differentiates the carrier phase observation between each array element and the reference array element to form a relative phase difference sequence These relative phase differences not only contain geometric phases determined by the incident direction, azimuth angle and elevation angle, but also superimpose three main errors: fixed phase difference caused by inconsistent channel group delay, thermal noise phase difference affected by carrier tracking filter bandwidth, and antenna phase center deviation. Commercial anti-jamming arrays often use microstrip or spiral elements, and the stability of the phase center is usually only on the order of λ / 10 (about 2 cm). The change of the phase center of a single antenna with the incident direction can cause a carrier phase measurement fluctuation of about 0.1 rad, which significantly affects the accuracy of short baseline direction measurement. Therefore, before performing direction measurement, the channel inconsistency needs to be calibrated, the thermal noise needs to be suppressed by bandwidth contraction and data smoothing, and the phase center offset / change needs to be included in the unified array phase center model to correct the measured relative phase difference.

[0038] Specifically, the phase center correction idea of the International GNSS Service is used to model the fixed phase center offset PCO of each array element and the direction-dependent phase center variation PCV, and it is pointed out that the mutual coupling between array elements and the edge effect of array elements and reflectors will further degrade PCV. The direction measurement deviation caused by the relative phase difference sequence needs to be eliminated at different incident directions to improve the accuracy of spoofing signal direction finding.

[0039] The fixed phase center offset is a three-dimensional constant vector of each antenna element in the array reference coordinate system, which is used to compensate for manufacturing and installation errors and channel average time delay difference; the phase center deviation is a direction-dependent vector function of incident azimuth angle and elevation angle, which is used to represent the transient deviation caused by mutual coupling and reflection edge effect of array elements; the geometric installation coordinates of each antenna element are corrected to effective phase center coordinates according to the fixed phase center offset and the phase center deviation, and the difference phase center baseline vector is constructed accordingly. The effective phase center coordinates can be written as:

[0040] Accordingly, the OA baseline space vector in the sense of difference phase center is:

[0041] wherein, and is the geometric installation coordinate of array element A and O, PCO is the average phase center offset, PCV is the phase center deviation varying with the incident direction. The phase center model is used to correct the geometric installation coordinates of each array element to effective phase center coordinates, and then construct the differential phase center baseline vector between the reference array element, which is called in the subsequent direction finding solution and suppression steps with the relative phase difference as the observation.

[0042] More specifically, the fixed phase center offset and the phase center deviation are obtained by calibration, and in calibration, a calibration source with a known direction or a natural calibration source using the known spatial position of a navigation satellite is used to obtain phase observations of each antenna element at multiple sampling points. After removing the channel fixed delay, the fixed phase center offset and the phase center deviation are fitted.

[0043] The calculation of the differential phase center baseline vector specifically includes: adding the geometric installation coordinates of each antenna element in the array to the phase center offset value of the antenna element, and superimposing the corresponding phase center variation according to the incident direction of the spoofing signal to obtain the effective phase center coordinates of each antenna element; taking a certain antenna element in the array as a reference, the differential phase center baseline vector corresponding to the reference antenna element and any other antenna element is calculated according to the difference between the effective phase center coordinates of the reference antenna element and the other antenna element.

[0044] Among them, in the measurement and correction process of the relative phase difference, the following phase differences exist: The phase difference caused by the inconsistency of the channel group delay, which is independent of the spoofing direction and is a constant in stable working conditions, can be effectively calibrated by a signal source; The thermal noise phase difference, which is affected by the receiver carrier tracking filter bandwidth, can be significantly reduced by limiting the noise bandwidth; The phase center deviation between the array elements, which varies with the incident direction, is the dominant error term of the array direction measurement accuracy.

[0045] By calibrating and compensating these phase differences, using bandwidth contraction and smoothing noise suppression, and correcting the influence on the coordinate level by the phase center model, the measured relative phase difference can be restored to nearly ideal geometric magnitude, significantly reducing phase jitter and improving direction finding accuracy under short baseline conditions.

[0046] In step S105, the incident direction angle of the spoofing signal is calculated based on the differential phase center baseline vector and the relative phase difference, the incident azimuth angle and the elevation angle of the spoofing signal are obtained, and the incident azimuth angle and the elevation angle are taken as control parameters of spatial domain suppression, the beam forming weight of the array antenna is solved and applied, so that the incident direction of the spoofing signal forms a null, the main lobe is formed in the incident direction of the real signal, and the updated weight is fed back to the real signal tracking channel in real time to improve the carrier-to-noise ratio and the locking stability.

[0047] Wherein, based on the differential phase center baseline vector and the carrier phase difference of each array element relative to the reference array element O obtained in step S104, the incident direction angle (azimuth angle φ and elevation angle θ) of the spoofing signal is first solved. In the scenario of seven-element (OABCDFG) array and spoofing signal incident from OE direction, the carrier phase difference observation model of each baseline is given. The phase difference of OA, OB, OC, OD, OF and OG six baselines satisfies:

[0048] and is limited in The main value domain. Wherein, is the fixed phase difference introduced by the inconsistent channel group delay, which is independent of the spoofing direction and can be regarded as a constant in the stable working condition; n is the thermal noise phase difference affected by the receiver tracking filter bandwidth, which can be effectively reduced by limiting the noise bandwidth; is the carrier phase center deviation between array elements, which depends on the incident azimuth and elevation and dominates the direction finding accuracy. Exemplarily, the short baseline group O-A-C satisfies the inverse tangent relationship with the long baseline group O-B-D, for example: ; When the channel group delay difference, antenna phase center error and thermal noise are all zero, the direction finding results of AOC and BOD two groups of baselines are consistent; under the condition that the above errors are non-zero but each channel is statistically consistent, the direction finding error of BOD group is about 1 / 2 of that of AOC group, and the joint baseline can be used to stably solve ; The incident azimuth angle and the elevation angle of the spoofing signal obtained above are taken as the control parameters of spatial domain suppression (beam forming), and the direction finding method can be connected with the existing "null anti-jamming receiver", so as to form a null in the incident direction of the spoofing signal and maintain the effective reception of the real GNSS signal direction; the spoofing detection in the acquisition stage helps to process the real and spoofing separately in the tracking and subsequent stages, so that the direction finding result can be directly used for back-end array processing and service output. Thus, in this step, the incident azimuth angle and the elevation angle of the spoofing signal are solved based on Solve and apply array beamforming weights to form deep nulls for spoofing incident directions, maintain main lobe gain for real signal incident directions, and apply updated weights to real signal tracking channels in real time to enhance synchronization robustness and overall performance.

[0049] In an embodiment, after the machine learning model is trained, its classification ability needs to be evaluated, the classification evaluation of the machine learning model includes accuracy, precision, recall and F1 score, wherein the accuracy is the sum of true positives and true negatives divided by the sum of true positives, true negatives, false positives and false negatives; the precision is the true positives divided by the sum of true positives and false positives; the recall is the true positives divided by the sum of true positives and false negatives; and the F1 score is twice the harmonic mean of the precision and the recall.

[0050] Based on the same idea, as shown in Figure 2 a machine learning-based GNSS detection system is provided, comprising: The acquisition module 201 is configured to receive navigation radio frequency signals containing real signals and spoofing signals, and obtain intermediate frequency data through RF front-end down-conversion processing; based on the intermediate frequency data, search for correlation peaks in the frequency and code phase dimensions to acquire GNSS satellite signals; The feature analysis module 202 is configured to extract a plurality of feature parameters generated during acquisition as the basis for spoofing detection, including the motion state of the receiver, the number of correlation peaks acquired, the average noise value, the acquisition parameters for determining satellite acquisition, and the sampling correlation peak values within ±1 chip range around the maximum correlation peak value; input the extracted plurality of feature parameters into a pre-trained machine learning classification model for analysis to determine whether there is spoofing interference in the GNSS satellite signal; The signal separation module 203 is configured to separate the spoofing signal from the real signal when the determination result exists spoofing signal interference, and send the spoofing signal and the real signal into independent tracking channels respectively; The direction finding observation module 204 is configured to perform direction measurement on the determined spoofing signal using an array antenna, measure the relative phase difference between the spoofing signal carriers received by each antenna element in the array, perform differential phase center correction on the measured relative phase difference, establish a phase center model of each antenna element in the array, the phase center model includes a fixed phase center offset of the antenna element and a phase center deviation varying with the incident direction, correct the geometric installation coordinates of each antenna element to effective phase center coordinates according to the phase center model, and calculate the differential phase center baseline vector between the reference antenna element and other antenna elements; The DOA solving and spatial domain suppression module 205 is used for calculating the incident direction angle of the spoofing signal based on the differential phase center baseline vector and the relative phase difference, obtaining the incident azimuth angle and the incident elevation angle of the spoofing signal, taking the incident azimuth angle and the incident elevation angle as control parameters of spatial domain suppression, solving and applying the beam forming weight value of the array antenna, so that the incident direction of the spoofing signal forms a null, the incident direction of the real signal forms a main lobe, and the updated weight value is fed back to the real signal tracking channel in real time to improve the carrier-to-noise ratio and the locking stability.

[0051] From the above system, the embodiment has the following beneficial effects: The system extracts multi-source features in the capture stage and uses machine learning classification, so that whether there is spoofing interference can be distinguished without relying on the tracking / positioning link; the features used can effectively cover the peak shape change and noise rise caused by large / small delay spoofing, so that the true / false signal can be separated before entering tracking, reducing mutual pollution and providing pure observation for subsequent direction finding and suppression.

[0052] In the direction finding aspect, the application proposes a differential phase center correction for commercial array: simultaneously compensating for the average and instantaneous phase center deviation, significantly reducing the system error caused by element mutual coupling and structural edge effect, improving the direction measurement accuracy of the spoofing signal, and forming an integrated process with the receiver's acquisition / detection / tracking / direction finding module, which can further drive beam forming to realize spoofing direction nulling and true star direction main lobe. Compared with the scheme relying on a dedicated array, it has the engineering advantages of lower cost and scalable deployment.

[0053] The specific details of each module in the above system have been described in detail in the method part embodiment, and the undisclosed details can be referred to the embodiment content of the method part, so no further description is given.

[0054] Through the above description of the embodiments, those skilled in the art can easily understand that the example embodiments described herein can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solutions according to the embodiments of the application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a U disk, a mobile hard disk, etc.) or a network, and includes a plurality of instructions to make a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) execute the method according to the example embodiments of the application.

[0055] Moreover, the above-described figures are only a schematic representation of the processes comprised in the method according to the exemplary embodiments of the application, and are not intended to limit purposes. It is readily understood that the processes shown in the above-described figures do not indicate or limit the chronological order of these processes. In addition, it is readily understood that these processes can be executed, for example, in a synchronous or asynchronous manner in a plurality of modules.

[0056] It should be noted that, although in the detailed description above several modules or units of the device for action execution are mentioned, such a division is not mandatory. Indeed, according to the exemplary embodiments of the application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into embodied by a plurality of modules or units.

[0057] Other embodiments of the application will be readily apparent to those skilled in the art upon considering the description hereof in light of the drawings and practice of the application disclosed herein. The application is intended to cover any variations, uses, or adaptations of the application following, in general, the principles of the application and including such further modifications as come within the true scope of the application. The application is to be considered as illustrative only and the true scope of the application is indicated by the appended claims.

[0058] It should be understood that the application is not limited to the precise structures herein described and illustrated above, and that various modifications and changes can be made without departing from the scope thereof. The scope of the application is limited only by the claims that follow.

Claims

1. A GNSS detection method based on machine learning, characterized in that, The method comprises: receiving navigation radio frequency signals containing real signals and spoof signals, and obtaining intermediate frequency data through RF front-end down-conversion processing; based on the intermediate frequency data, searching for correlation peaks in the frequency and code phase dimensions to capture GNSS satellite signals; extracting multiple feature parameters generated during capture as the basis for spoof detection, the feature parameters including the motion state of the receiver, the number of correlation peaks captured, the average noise value, the capture parameters used to determine satellite capture, and the sampling correlation peak values within ±1 chip around the maximum correlation peak value; inputting the extracted multiple feature parameters into a pre-trained machine learning classification model for analysis to determine whether spoof interference exists in the GNSS satellite signals; if the determination result indicates that spoof signal interference exists, separating the spoof signals from the real signals, and sending the spoof signals and the real signals into independent tracking channels respectively; performing direction measurement on the determined spoof signals using an array antenna, measuring the relative phase difference between the spoof signal carriers received by each antenna element of the array, performing differential phase center correction on the measured relative phase difference, establishing a phase center model for each antenna element in the array, the phase center model including a fixed phase center offset of the antenna element and a phase center deviation that varies with the incident direction, correcting the geometric installation coordinates of each antenna element to effective phase center coordinates according to the phase center model, and calculating a differential phase center baseline vector between a reference antenna element and other antenna elements; based on the differential phase center baseline vector and the relative phase difference, calculating the incident direction angle of the spoof signals, obtaining the azimuth and elevation angles of the spoof signals, and using the incident azimuth angle and the elevation angle as control parameters for spatial domain suppression to solve and apply beamforming weights of the array antenna, so that a null is formed in the incident direction of the spoof signals and a main lobe is formed in the incident direction of the real signals, and the updated weights are fed back to the real signal tracking channel in real time to improve the carrier-to-noise ratio and locking stability. 2.The machine learning based GNSS detection method according to claim 1, characterized in that, The machine learning classification model is constructed using a gradient boosting decision tree algorithm and is trained offline based on historical data of GNSS signal spoof interference scenarios, so that the machine learning classification model can classify and determine the capture result according to multiple feature parameters to detect the presence of spoof signals. 3.The GNSS detection method based on machine learning according to claim 1, characterized in that, The classification evaluation of the machine learning model includes accuracy, precision, recall, and F1 score, wherein the accuracy is the sum of true positives and true negatives divided by the sum of true positives, true negatives, false positives, and false negatives; the precision is the true positives divided by the sum of true positives and false positives; the recall is the true positives divided by the sum of true positives and false negatives; the F1 score is twice the harmonic mean of the precision and the recall. 4.The machine learning based GNSS detection method of claim 1, wherein, During the direction measurement of the spoof signals, the fixed phase difference and time delay difference between each antenna element of the array are calibrated and compensated, and the random phase fluctuations introduced by thermal noise are reduced by reducing the carrier tracking loop bandwidth and smoothing the phase measurement data, so that the jitter of the measured relative phase difference is reduced. 5.The machine learning based GNSS detection method of claim 1, wherein, The fixed phase center offset is a three-dimensional constant vector of each antenna unit in an array reference coordinate system, used to compensate for manufacturing and installation errors and channel average time delay differences; the phase center deviation is a direction-dependent vector function of incident azimuth and elevation angles, used to represent transient deviations caused by array element mutual coupling and reflection edge effects; the geometric installation coordinates of each antenna unit are corrected to effective phase center coordinates according to the fixed phase center offset and the phase center deviation, and the differential phase center baseline vector is constructed accordingly. 6.The machine learning based GNSS detection method of claim 1, wherein, The fixed phase center offset and the phase center deviation are obtained through calibration, in which a calibration source with a known direction or a navigation satellite with a known spatial position is used as a natural calibration source, phase observations of each antenna unit are obtained by traversing multiple sampling points, and the fixed phase center offset and the phase center deviation are fitted after removing the channel fixed time delay; the phase center model is indexed according to the working frequency band, and the corresponding fixed phase center offset and phase center deviation are stored and called for correction under different frequency bands. 7.The machine learning based GNSS detection method according to claim 1, wherein, The calculation of the differential phase center baseline vector specifically includes: The geometric installation coordinates of each antenna unit in the array are added to the phase center offset value of the antenna unit, and the corresponding phase center variation is superimposed according to the incident direction of the spoofing signal, to obtain the effective phase center coordinates of each antenna unit; taking a certain antenna unit in the array as a reference, the differential phase center baseline vector corresponding to the difference between the effective phase center coordinates of the reference antenna unit and any other antenna unit is calculated.

8. A machine learning based GNSS detection system, characterized in that, It includes: The acquisition module is used to receive navigation radio frequency signals containing real signals and spoofing signals, and obtain intermediate frequency data through RF front-end down-conversion processing; Based on the intermediate frequency data, search for correlation peaks in frequency and code phase dimensions to capture GNSS satellite signals; The feature analysis module is used to extract a plurality of feature parameters generated during acquisition as spoofing detection basis, the feature parameters including the motion state of the receiver, the number of correlation peaks captured, the average noise value, the capture parameters for judging satellite capture, and the sampling correlation peak values within ±1 chip range around the maximum correlation peak value; The extracted plurality of feature parameters are input into a pre-trained machine learning classification model for analysis to determine whether there is spoofing interference in the GNSS satellite signal; The signal separation module is used to separate the spoofing signal from the real signal when the determination result exists spoofing signal interference, and send the spoofing signal and the real signal into independent tracking channels respectively. The direction finding observation module is configured to perform direction measurement on the determined spoofing signal by using the array antenna, measure relative phase difference between the spoofing signal carriers received by each antenna unit of the array, perform differential phase center correction on the measured relative phase difference, establish a phase center model of each antenna unit in the array, the phase center model including a fixed phase center offset of the antenna unit and a phase center deviation varying with the incident direction, correct the geometric installation coordinates of each antenna unit to effective phase center coordinates according to the phase center model, and calculate a differential phase center baseline vector between the reference antenna unit and other antenna units; The DOA solving and spatial domain suppression module is configured to calculate the incident direction angle of the spoofing signal based on the differential phase center baseline vector and the relative phase difference, obtain the incident azimuth angle and the incident elevation angle of the spoofing signal, take the incident azimuth angle and the incident elevation angle as control parameters of spatial domain suppression, solve and apply beamforming weights of the array antenna, so that a null is formed in the incident direction of the spoofing signal and a main lobe is formed in the incident direction of the real signal, and the updated weights are fed back to the real signal tracking channel in real time to improve the carrier-to-noise ratio and the locking stability.

Images