A navigation interference source monitoring data collection method

Abstract

The application discloses a navigation interference source monitoring data collection method, comprising the following steps: extracting an initial interference characteristic vector of an original radio frequency signal; generating an IRS phase encoding sequence according to the initial interference characteristic vector, and obtaining a differential characteristic signal according to the sequence, and then constructing a multi-view space fingerprint matrix; according to the fingerprint matrix, fusing path characteristics, generating an enhanced interference characteristic tensor; decomposing the characteristic tensor to obtain the position and mode parameters of the interference source, and generating a comprehensive characteristic report. Through the IRS active detection mechanism, the passive monitoring is converted into the incentive-response active detection, the technical problem that the traditional method is difficult to find weak interference and concealed interference is solved, and the accuracy and robustness of the interference source monitoring are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of data collection, and particularly relates to a navigation interference source monitoring data collection method. BACKGROUND

[0002] With the rapid development of emerging technologies such as unmanned aerial vehicles, autonomous driving and the Internet of Things, society is increasingly dependent on precise and continuous navigation positioning and timing services. However, GNSS signals are extremely susceptible to various types of radio signals due to their inherent characteristics of long transmission distance and weak landing power. These interference signals, whether unintentional adjacent band out-of-band radiation, or intentional malicious jamming or spoofing, will reduce or even block the performance of navigation services. Therefore, effective monitoring, accurate positioning and rapid identification of interference sources in the navigation band, timely detection and elimination of potential threats, have become an important research topic to ensure national information security and stable operation of the economy and society.

[0003] Currently, the monitoring technology for navigation interference sources mainly focuses on passive signal reception and analysis. These technologies usually use a single or multiple antenna receivers deployed on fixed sites or mobile platforms (such as vehicles, unmanned aerial vehicles) to continuously monitor electromagnetic signals in the target airspace. When the power or statistical characteristics of the received signal are abnormal, the system starts the analysis program. The mainstream technical methods include interference detection algorithms based on high-order cumulants, cyclostationarity and other statistical characteristics, and direction finding technology using antenna arrays for signal spatial spectrum estimation. For example, through classical algorithms such as multiple signal classification (MUSIC) and estimation of signal parameters via rotational invariance techniques (ESPRIT), the direction of arrival (DOA) of the interference signal can be estimated. After obtaining the direction finding results of multiple sites, the geographic location of the interference source is determined through intersection positioning algorithms. In addition, some research also focuses on analyzing the modulation mode, waveform parameters and other characteristics of the interference signal in order to preliminarily classify the interference type. These passive monitoring methods can play a certain role in dealing with high-power, continuous and simple interference sources, and constitute the basis of the current interference monitoring system.

[0004] However, the existing passive monitoring methods face two core and deep technical bottlenecks when dealing with increasingly complex electromagnetic environments and intelligent interference means. Firstly, the existing methods have serious deficiencies in detecting weak signals and covert interference. Secondly, the dimension of the interference source feature description is single, and it is difficult to form a comprehensive and accurate comprehensive cognition. SUMMARY

[0005] The application aims to provide a navigation interference source monitoring data collection method to solve the above problems existing in the prior art.

[0006] Technical solution: A method for collecting navigation interference source monitoring data, comprising:

[0007] Extracting the initial interference feature vector of the original RF signal;

[0008] Based on the initial interference feature vector, the IRS phase coding sequence of the smart reflective surface is generated, and the differential feature signal is generated using the IRS phase coding sequence to construct a multi-view spatial fingerprint matrix.

[0009] According to the multi-view spatial fingerprint matrix, the path features are fused to generate an enhanced interference feature tensor;

[0010] Decompose the enhanced interference feature tensor, obtain the location and pattern parameters of the interference source, and generate a comprehensive feature report of the interference source.

[0011] Beneficial effect: The present invention solves the problem that traditional methods are difficult to detect weak interference and hidden interference, and improves the accuracy and robustness of interference source monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flowchart of the steps of a navigation interference source monitoring data collection method provided in an embodiment of the present application.

[0013] Figure 2 A flowchart of the steps for constructing a multi-view spatial fingerprint matrix provided in an embodiment of the present application.

[0014] Figure 3 A flowchart of the steps for constructing an IRS phase coding sequence provided in an embodiment of the present application.

[0015] Figure 4 A flowchart of the steps for optimizing the design of the IRS phase coding sequence provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0017] It should be noted that the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or are inherent to these processes, methods, products or apparatus.

[0018] Research has found that passive monitoring relies on the energy intensity of the interference signal itself. When interference sources employ low-power, intermittent, or concealed transmission strategies similar to background noise, their signal energy is drowned out by complex urban multipath reflections and background noise. This makes traditional detection methods based on energy or statistical features highly susceptible to failure or high false alarms. Furthermore, passive monitoring systems lack the ability to actively interact with the environment or target and the ability to stimulate and capture interference source response characteristics from different perspectives. This results in extremely limited information dimensionality, making it difficult to distinguish multiple interference sources in close spatial locations or construct a unique fingerprint that can uniquely identify a specific interference source. Furthermore, traditional methods typically view multipath signals as harmful factors that need to be suppressed and lack mechanisms to effectively utilize multipath information to enhance positioning accuracy and scene perception. The extraction of parameters such as interference source location, signal pattern, and time-varying characteristics is often sequential and decoupled, ignoring the inherent physical connections between these features. This results in fragmented interference reports that are difficult to support in-depth analysis and prediction of interference source behavior and intentions.

[0019] like Figure 1 As shown, a method for collecting navigation interference source monitoring data is proposed, comprising the following steps:

[0020] Extract the initial interference feature vector of the original RF signal.

[0021] In this embodiment, this step is completed by the antenna array carried by the UAV. Specifically, the M-element antenna array is used to synchronously collect the original RF signals of multiple channels to form the original complex baseband signal matrix X(t), and at the same time record the global positioning system (GPS) timestamp sequence T gps and the drone's own position vector P uav (t). In order to eliminate the slight time delay introduced by the physical connection length and device differences between channels, an interpolation alignment algorithm is used to process the signals of each channel to generate a time-space aligned signal matrix X aligned At the same time, the current phase configuration matrix Φ(t) and physical parameters such as unit spacing and operating frequency are read from the IRS controller in real time to construct a spatial response mapping function h that can describe the response characteristics of the IRS to waves from different directions under a specific phase configuration. IRS(Φ, θ, δ), where θ is the elevation angle of the incident or reflected wave, and δ is the azimuth angle of the incident or reflected wave. On this basis, the time-space alignment signal matrix X aligned (t) Perform preprocessing, such as performing short-time Fourier transform (STFT) to obtain the time-frequency spectrum density matrix S tf (t, f), and then use the adaptive threshold detection algorithm based on local statistical characteristics to preliminarily determine whether there is interference and identify the potential interference frequency band set F jam and the corresponding temporal activity pattern A jam (t). All the above information, including the power spectrum density of the signal, the rough estimate of the angle of arrival (AOA) obtained by array signal processing, polarization characteristics, and time domain activity patterns, are integrated to construct a multi-dimensional initial interference feature vector V init This embodiment provides a reliable priori information basis for subsequent IRS active detection through the simultaneous collection and preliminary analysis of multi-dimensional and multi-source data, making subsequent detection more targeted and improving the efficiency and accuracy of the entire process.

[0022] Based on the initial interference eigenvector, the IRS phase coding sequence of the smart reflective surface is generated, and the differential characteristic signal is generated by the IRS phase coding sequence. Then, a multi-view spatial fingerprint matrix is ​​constructed according to the differential characteristic signal.

[0023] Specifically, using the initial interference feature vector V init , especially the initial estimation of the interference arrival angle, to guide the design of a specific IRS phase coding sequence {Φ k The purpose of the design is not to enhance the signal as in traditional communication, but to intentionally create a controllable electromagnetic response disturbance in the estimated interference direction. According to the preset timing strategy, the coding sequence and the reference phase configuration Φ as the benchmark are configured. ref , and are loaded onto each unit of the IRS in turn. During the short duration of each coding and reference phase, the antenna array on the drone continuously collects signals and obtains several coded response signals {Y k (t)} and a reference response signal Y ref (t). Perform differential processing on each coded response signal and the reference response signal, that is, calculate ΔY k (t)=Y k (t)-Y ref (t), and obtain the differential characteristic signal. Differential processing can effectively separate the signal components actively excited by the IRS phase change from the complex, time-varying background noise and irrelevant signals, highlighting the weak features related to the interference source response. For each differential characteristic signal ΔY k (t) Perform spatial spectrum analysis, for example by calculating its covariance matrix Rk And perform eigendecomposition to extract the spatial pattern matrix U that can characterize the response pattern of the interference source under this specific coding k , and integrate all the spatial pattern matrices under encoding to generate a more informative multi-view spatial fingerprint matrix F spatial This matrix acts like a behavioral fingerprint of the interference source, uniquely recording its response characteristics when faced with a series of different electromagnetic environment probes.

[0024] According to the multi-view spatial fingerprint matrix, the path features are fused to generate an enhanced interference feature tensor.

[0025] In this embodiment, the received signal usually contains a direct path and multiple paths reflected by environmental objects (including IRS). These paths carry complementary information about the location and characteristics of the interference source. Effective fusion of them can improve the analysis performance. Specifically, based on the multi-view spatial fingerprint matrix F spatial and signal propagation model, using algorithms such as SAGE (Space-Alternating Generalized Expectation-maximization) to decompose the signal into direct path components Y direct and the reflected path component Y reflect，i , and extract their respective features. In order to quantify the correlation between these different paths, the mutual information between them is calculated. In particular, when calculating the mutual information, the phase configuration Φ of the IRS is k As a conditional variable, calculate the conditional mutual information I(Y i ; Y j |Φ k ), where I( ) is the conditional mutual information function, Y i is the i-th path component; this allows the analysis to reveal the potential correlation between paths generated by IRS active modulation. The path mutual information matrix I is obtained mutual Then, based on this matrix, the features of different paths are weighted and fused. For example, for path pairs with high mutual information values ​​and strong correlation (usually originating from the same interference source), coherent merging is used to enhance the signal-to-noise ratio; while for paths with low mutual information values, their independence is maintained to preserve spatial diversity information. Through intelligent fusion strategies, a fused enhanced feature set is obtained. This enhanced feature set is organized and arranged in multiple dimensions such as space, time, and frequency to construct a high-dimensional data structure, namely the enhanced interference feature tensor T. enhanced Optionally, in order to reduce the complexity of subsequent calculations, a tensor decomposition technique such as Tucker decomposition can be applied to moderately compress the tensor while retaining its main modal information.

[0026] Decompose the enhanced interference feature tensor to obtain the position and mode parameters of the interference source, and generate an interference source comprehensive feature report.

[0027] In this embodiment, this step aims to parse the interference source parameters of interest to the user with clear physical meaning from the highly concentrated feature tensor. Specifically, the enhanced interference feature tensor T enhanced Perform a tensor decomposition algorithm such as CP (CANDECOMP / PARAFAC) decomposition, which can uniquely decompose it into a product of a series of factor matrices, such as a spatial factor matrix A space , a time factor matrix A time , and a frequency factor matrix A freq . These factor matrices reveal the internal structure of the interference in different dimensions. Among them, the spatial feature factor A space is directly related to the spatial position information of the interference source; the time feature factor A time reveals the activity law and duration mode of the interference signal (such as continuous wave, pulse, frequency hopping, etc.); and the frequency feature factor A freq reflects the type and modulation method of the interference. By further analyzing these factor matrices, the position and mode parameters of the interference source can be jointly extracted. For example, the joint spatial feature factor and the known IRS geometric relationship can be used to solve the precise three-dimensional coordinates P jam of the interference source; and the combination of the time and frequency feature factors and the matching with the predefined interference mode library can identify the interference type L type . All the results obtained through analysis, including the precise position, position accuracy, interference type, signal strength, activity period, duty cycle, etc. Complete information is integrated into a structured interference source comprehensive feature report R final for output. The complex electromagnetic signal analysis problem is converted into a multi-dimensional data mining problem, and through tensor decomposition, the automatic, efficient, and accurate extraction of various key parameters of the interference source is realized.

[0028] As shown in Figure 2 , according to one aspect of the present application, a multi-view spatial fingerprint matrix is constructed, including:

[0029] According to the initial interference feature vector, an IRS phase encoding sequence is constructed to impose a controllable electromagnetic response disturbance on the estimated interference direction.

[0030] Further, as shown in Figure 3 , the IRS phase encoding sequence is constructed, including:

[0031] The initial estimate of the angle of arrival of the interference signal is extracted from the initial interference feature vector.

[0032] In this embodiment, the initial estimate of the angle of arrival is denoted as AOA rough=[θ0,δ0], is the initial interference feature vector V init The initial estimate is obtained from , where θ0 represents the preliminary estimate of the pitch angle and δ0 represents the preliminary estimate of the azimuth angle. Due to measurement errors and environmental influences, the initial estimate has a certain uncertainty.

[0033] Based on the initial estimation of the arrival angle, a set of interference direction sensitive angles is defined.

[0034] Specifically, around the initial estimate of the angle of arrival AOA rough Set the angle search range, for example, [θ0-Δθ, θ0+Δθ]×[δ0-Δδ, δ0+Δδ], where Δθ and Δδ are search tolerances set according to the uncertainty model, for example, both can be set to 5 degrees. Discretize this two-dimensional angle range into G×G grid points {(θ g , δ g For each grid point, calculate the IRS gain function G for that direction under a certain initial or default phase configuration Φ0 (e.g., all zero phase) IRS (θ g , δ g |Φ0), and further calculate the gradient of the gain function with respect to the angle |▽G IRS Those areas whose gradient values ​​exceed the preset threshold γ are identified as the sensitive angle set Ω sensitive Physically, these areas mean that a small change in the IRS phase will cause a drastic change in the far-field radiation pattern in that direction, so applying a perturbation to these areas is most likely to excite a response from the interferer.

[0035] The IRS phase coding sequence is optimized to enable the smart reflector to produce the maximum gain change within the interference direction sensitive angle set, thereby achieving controllable electromagnetic response disturbance.

[0036] Specifically, construct K basis Orthogonal phase encoding basis functions {Ψ k (m, n)}, used to modulate the phase of the (m, n)th unit on the IRS. These basis functions can include different forms such as linear phase gradient and Fresnel phase distribution, and are orthogonalized by the Gram-Schmidt method. Construct the optimization objective function, such as J(α)=Σ g∈Ωsensitive |G IRS (θ g , δ g |Φ0+∑ k α k ·Ψ k )-G IRS (θ g , δ g |Φ0)| 2; where α k is the weight coefficient of the kth coding basis function, Ψ k is the kth phase encoding basis function, J(α) represents the square sum of the total gain change caused by the phase perturbation generated by the linear combination of basis functions (with coefficient α) on the entire sensitive angle set. By using an optimization algorithm such as the gradient ascent method, the optimal coefficient vector α* that maximizes the objective function J(α) is solved. By selecting different coefficients, K code Different optimal codes Φ k =Φ0+∑ i=1 Kbasis α ik *·Ψ i , forming a preliminary set of coding sequences, where α ik * is the optimal coefficient of the i-th basis function under the k-th group encoding.

[0037] This embodiment achieves high-sensitivity detection of hidden interference sources by transforming the IRS from a passive reflecting surface into an active detection tool. Specifically, by designing a phase coding sequence that produces the maximum gain change within the sensitive angle set, interference signals that are originally weak in power or use low probability of intercept technology are stimulated to produce detectable response changes. Compared with traditional passive monitoring, the interference detection sensitivity is improved, especially for interference sources that use anti-detection technologies such as frequency hopping and direct spread. The nonlinear interaction between the electromagnetic disturbance field created in space by the IRS phase change and the interference signal, even if the interference signal itself is very weak, its response to the disturbance field can still be captured by the differential detection method.

[0038] Alternatively, as Figure 4 As shown, the optimization design of the IRS phase coding sequence also includes: calculating the cross-correlation value between different phase codes in the IRS phase coding sequence; if the cross-correlation value exceeds a preset threshold, adjusting at least one phase coding to reduce the cross-correlation value, thereby ensuring that each phase code constituting the IRS phase coding sequence meets the detection independence requirement.

[0039] Specifically, calculate any two codes Φ i and Φ j The cross-correlation matrix R code (i, j) = |〈vec(Φ i ), vec(Φ j )〉| / (||Φ i ||·||Φ j ||), where vec(·) means to vectorize the matrix, 〈 〉 means vector inner product, ||Φ i || represents the encoding Φ iIf the calculated cross-correlation value exceeds a preset low correlation threshold, such as 0.1, it is considered that the detection effects of the two codes are too similar. In this case, one of the codes needs to be adjusted, for example, by orthogonal projection, such as Φ' j =Φ j -∑ i=1 j-1 ( j , Φ i 〉 / ||Φ i || 2 )·Φ i , Φ j In the existing code Φ i The relevant component projection is removed. After orthogonalization, the phase value needs to be renormalized to the range of [0, 2π] to ensure the feasibility of IRS hardware. This process needs to be iterated until the cross-correlation value R between all code pairs is code (i, j) are both less than a preset threshold, such as 0.1, so that each code can independently stimulate a different response mode of the interference source. This embodiment makes the detection performed by each code relatively independent, avoiding information redundancy.

[0040] In a specific embodiment, the initial code number K can be set code =8. After independence verification, 2-3 highly correlated codes may be eliminated, ultimately resulting in 5-6 valid independent codes. These codes are designed not only to maximize gain variation within the sensitive angle set but also to ensure detection independence from each other, enabling observation of the interference source's response characteristics from multiple perspectives.

[0041] The IRS phase coding sequence and reference phase configuration are loaded using a preset timing to capture the corresponding coded response signals and reference response signals respectively;

[0042] performing differential processing on the coded response signal and the reference response signal to obtain a differential characteristic signal;

[0043] The spatial response pattern of the differential feature signal is analyzed to construct a multi-view spatial fingerprint matrix.

[0044] The design goal of the encoding in this embodiment is not to optimize communication performance, but to maximize the electromagnetic perturbation in a specific spatial region; through the linear combination of orthogonal basis functions, phase distributions with different spatial characteristics can be flexibly constructed; the cross-correlation constraint ensures the information gain of multiple detections. In practical applications, the generated IRS phase coding sequence {Φ k} k=1 KcodeIt will be stored in the control system and loaded onto the IRS in a specific timing sequence in the subsequent sequential configuration step. After each code is loaded, the system will collect the corresponding response signal and extract the signal changes caused by the code through differential processing, and finally construct a multi-perspective spatial fingerprint matrix that can fully characterize the characteristics of the interference source. In addition, the code design method in this embodiment is adaptive. When the initial interference feature vector V init When the arrival angle estimation accuracy in is low, the angle search range Δθ and Δδ can be increased accordingly to ensure that the true interference direction is included in the sensitive angle set. At the same time, the specific form of the optimization objective function J(α) can also be adjusted according to actual needs, such as adding weighting for specific frequency components, or considering the situation where multiple interference sources exist at the same time. Through the adaptive phase coding sequence design method, active detection of the estimated interference direction can be achieved, which is a capability that traditional passive monitoring methods do not have. By applying controllable electromagnetic disturbances in space, the interference source can be stimulated to produce detectable response changes, thus laying the foundation for subsequent high-precision positioning and feature extraction.

[0045] This embodiment ensures the information independence of multiple detections by calculating and controlling the cross-correlation values ​​between different phase codes, thereby achieving all-round extraction of interference source features. When the cross-correlation value exceeds the threshold of 0.1, the code is adjusted by orthogonal projection so that each code excites a different response mode of the interference source. This allows only 5-8 independent codes to construct a complete interference source spatial feature, which shortens the data acquisition time compared to traditional methods that require dozens of observations. The orthogonal coding sequence creates different electromagnetic field distribution patterns in the spatial domain, and each pattern detects the interference source from a different angle, similar to the multi-angle projection in computed tomography, ultimately achieving three-dimensional feature extraction of the interference source.

[0046] According to one aspect of the present application, determining the preset timing includes:

[0047] The interference burst duration distribution is extracted from the interference time-domain activity pattern contained in the initial interference feature vector. The coding loading duration is determined based on the interference burst duration distribution. The coding loading duration is used as a key parameter of the preset timing to achieve dynamic adaptation of the loading of the IRS phase coding sequence to the interference time-domain activity pattern.

[0048] In this embodiment, the interference time domain activity pattern is the time domain activity pattern vector A jam (t), which contains the statistical description of the occurrence, duration, and period of the interference signal. Specifically, from A jam (t) to extract the key interference burst duration distribution p(τ burst). This distribution describes the possible duration of the interference signal each time it appears. In order to ensure that the complete interference characteristics can be captured during each code loading period and avoid feature truncation due to a too short measurement window, the code loading duration T code It needs to match the duration of the interference. A preferred implementation is to set T code Set to T code =max(T min ,α×E[τ burst ]), where E[τ burst ] is based on the distribution p(τ burst ) calculated the average burst duration, T min is a minimum duration, such as 10 milliseconds, to ensure a basic measurement signal-to-noise ratio. α is a scaling factor, such as 0.1, which strikes a balance between capturing features and maintaining high acquisition efficiency. This allows data acquisition timing parameters to no longer be fixed but to be adaptively adjusted based on the behavioral characteristics of the interference, demonstrating both intelligence and efficiency.

[0049] This embodiment dynamically matches detection timing with jammer behavior by extracting burst duration distributions from jammer time-domain activity patterns and determining code loading times accordingly. For typical military frequency-hopping jammers with burst durations of 10-100ms, the system automatically sets the code duration to 10% of the burst duration, ensuring at least one complete phase perturbation-response cycle within a single burst. This improves the detection success rate for time-varying jammers, particularly for intelligent jammers employing random hopping strategies.

[0050] According to one aspect of the present application, obtaining a differential characteristic signal includes:

[0051] Using a preset timing, the loading of a reference phase configuration is inserted into the loading gap of each phase code in the IRS phase coding sequence; the signals during the loading of each phase code and the reference phase configuration are respectively collected to obtain several coded response signals and reference response signals; the difference between the coded response signal and the reference response signal is calculated to generate a differential characteristic signal, thereby separating the signal changes caused by the phase change of the IRS phase coding sequence from the background environment.

[0052] In this embodiment, the reference phase configuration Φ ref It is the basis for comparison. ref The selection of can be dynamic. For example, when the interference signal is strong (such as SNR>20dB), the phase configuration that minimizes the gain in the estimated interference direction can be selected as a reference, that is, Φ ref =argmin Φ G IRS (θ jam , δjam |Φ), which helps to maximize the amount of change in the interference signal in the difference, where θ jam is the vertical angle of the interference signal relative to the normal of the IRS array plane, and δ jam is the horizontal angle of the interference signal in the array plane; in the case of weak or unknown interference, a neutral, e.g., all-zero, phase configuration can be chosen as the reference. The specific acquisition procedure is to control the IRS according to an alternating sequence such as {Φ1, Φ ref , Φ2, Φ ref , …, Φ k , Φ ref}. During the duration T k of loading each code Φ code , the coded response signal Y k (t) is acquired; during the duration of loading the reference phase Φ ref , the reference response signal Y ref (t) is acquired. By calculating the difference ΔY k (t) = Y k (t) - Y ref (t), the differential feature signal is generated. The differential processing takes advantage of the slowly varying characteristics of background noise and most environmental reflection signals in a short period of time. Since the code switching is very fast, the background environment can be considered to be essentially unchanged, so after the differential calculation, these common-mode, non-IRS phase-varying signal components are effectively canceled out, leaving only the part of the signal that is actively excited by the IRS phase change, i.e., the signal feature related to the interaction between the interference source and the IRS, which improves the signal-to-noise ratio of the target feature. Optionally, in order to further ensure the quality of the differential signal, a differential signal quality evaluation mechanism can also be included. Specifically, after the differential feature signal ΔY k (t) is generated, the differential quality index Q diff = ||ΔY k || 2 / ||Y ref || 2 is calculated, which quantifies the proportion of signal energy change caused by IRS coding relative to the reference signal energy. A threshold value, e.g., 0.01, can be set, and if the calculated Q diff is lower than the threshold value, it indicates that the current code Φ k has failed to produce sufficient and effective disturbance to the interference source, and the corresponding differential signal contains little information. Such a signal can be marked as invalid and excluded or given a very low weight in subsequent fingerprint matrix construction. This is equivalent to setting a data filter, which can effectively prevent low-quality data caused by improper IRS code selection or temporary disappearance of the interference signal from polluting the final analysis results.

[0053] The embodiment realizes accurate extraction of weak interference characteristics in a complex electromagnetic environment by inserting a reference phase configuration in the coding loading gap and performing differential processing. Specifically, by calculating the difference between the coding response signal and the reference response signal, the signal change caused by the IRS phase change is separated from the background noise, improving the signal-to-noise ratio. Especially in the case of a large amount of background signal interference in the urban environment, the traditional method often hides the real interference in the noise, while the differential mechanism of the embodiment can effectively eliminate the common-mode noise and only retain the difference signal caused by the IRS active disturbance. By sharing the same propagation environment between the time-adjacent reference signal and the coding signal, it is ensured that the differential result accurately reflects the response characteristics of the interference source to the IRS phase change.

[0054] According to an aspect of the present application, the spatial response mode of the differential characteristic signal is analyzed and a multi-view spatial fingerprint matrix is constructed, including:

[0055] Based on the differential characteristic signal, a corresponding differential covariance matrix is constructed for each phase code in the IRS phase code sequence; a spatial mode matrix is extracted from the differential covariance matrix to represent the response characteristics of the interference source to the phase code; and the spatial mode matrices obtained under different phase codes are integrated to generate a multi-view spatial fingerprint matrix in a structured manner.

[0056] Specifically, for each effective differential characteristic signal ΔY k (t) corresponding to each code k, its second-order statistical characteristics, i.e., the differential covariance matrix R k =E[ΔY k (t)·ΔY k (t) H ] need to be estimated, where E[·] represents the expectation operation, (·) H represents the conjugate transpose. In actual application, since there may be burst noise or outliers in the signal, direct estimation using sample average may lead to biased results. Therefore, a preferred implementation is to use a robust covariance estimation method. Traditional sample covariance estimation is very sensitive to outliers in the data, and a single outlier can cause a large deviation in the estimation result. The robust estimation method can effectively overcome this problem. For example, the initial sample covariance matrix and the data center can be calculated, and the potential outlier sample points can be identified by performing eigenvalue decomposition on it. According to the Mahalanobis distance of each sample point to the data center, the sample is reweighted using the Huber weight function w(r)=min(1,c / |r|), where r is a certain distance measure of the sample from the center, and c is a tuning parameter; this function can reduce the weight of outliers far from the center. Through iterative weighting, a more robust differential covariance matrix R kThis procedure is iterated several times until convergence. The resulting covariance matrix can more truly reflect the main spatial structure of the signal without being disturbed by a few abnormal data points.

[0057] The spatial pattern matrix is extracted from the differential covariance matrix. The covariance matrix R k contains the spatial distribution information of the differential signal. Its main eigenvectors (i.e. the eigenvectors corresponding to the larger eigenvalues) constitute the signal subspace, which represents the main response direction or pattern of the interference source under the current code Φ k . Therefore, the robust differential covariance matrix R k is subjected to eigenvalue decomposition: R k = U k λ k U k H , where U k is the eigenvector matrix and λ k is the diagonal matrix of eigenvalues. In order to adaptively determine how many main eigenvectors should be retained, an information theory criterion such as the minimum description length (MDL) criterion can be used to determine the effective rank p k of the matrix. The eigenvectors corresponding to the first p k largest eigenvalues in U k are extracted to form an Mxp k matrix, which is the spatial pattern matrix of the kth phase code, denoted as U k，main . This matrix U k，main characterizes the unique spatial response characteristics of the interference source to the kth phase code in a compact form.

[0058] All spatial pattern matrices are integrated to structurally generate a multi-angle spatial fingerprint matrix F spatial . This final fingerprint matrix is not simply a side-by-side arrangement of all U k，main . In order to contain more abundant information, its structure can be designed to be more sophisticated. For example, the spatial fingerprint matrix F spatial can be constructed as F spatial = [U1|ΔU 1，2 |…|ΔU K-1，K |M CR ]; where U k is the spatial pattern matrix under the kth code; ΔU k，j =∣∣U k -U j ∣∝ F is the difference measure (e.g. calculated using the Frobenius norm) between the spatial pattern matrices under different codes, representing the degree of change in the response pattern of the interference source; and M CRThe encoding-response mapping matrix describes the overall functional relationship between the characteristics of the IRS encoding sequence and the resulting spatial response pattern. k ), which also includes the relative change pattern (ΔU k，j ), also includes systematic mapping relationships (M CR A comprehensive matrix of 2D and 3D images forms a highly condensed, multi-perspective spatial fingerprint that uniquely identifies the interference source and its behavioral characteristics. This provides a solid data foundation for subsequent high-precision positioning and pattern recognition.

[0059] This embodiment constructs a multi-dimensional feature characterization system for interference sources by integrating spatial pattern matrices under different phase encodings. The differential covariance matrix excited by each phase encoding contains the response fingerprint of the interference source under a specific spatial disturbance, similar to electromagnetic photographs at different angles. The multi-perspective spatial fingerprint matrix constructed by independent encoding improves the accuracy of uniquely identifying the type of interference source. Different encodings create orthogonal disturbance patterns in space, and each pattern reveals different physical properties of the interference source, such as directionality, polarization characteristics, frequency selectivity, etc., ultimately forming a unique identifier similar to electromagnetic DNA.

[0060] According to one aspect of the present application, generating an enhanced interference feature tensor includes:

[0061] Direct path features and reflected path features are separated from the multi-view spatial fingerprint matrix.

[0062] Specifically, it can be based on the multi-view spatial fingerprint matrix F spatial and signal propagation model, using advanced parameter estimation algorithms in the field of signal processing, such as the SAGE (Space-Alternating Generalized Expectation-maximization) algorithm. This algorithm iteratively estimates and optimizes the parameters of the direct path and each reflected path (such as arrival angle, delay, amplitude), and ultimately decomposes the mixed received signal into a pure direct path signal component Y. direct and a set of reflected path signal components {Y reflect，i}, and extract their corresponding feature vectors.

[0063] The path mutual information between the direct path characteristics and the reflected path characteristics is calculated to quantify the correlation between different paths.

[0064] Path mutual information is not traditional mutual information, but rather conditional mutual information that better reflects the technical characteristics of the present invention. The steps for calculating path mutual information include: performing dimensionality reduction on the different phase configurations loaded on the smart reflector to generate corresponding conditional vectors; combining the direct path characteristics and the reflected path characteristics with the conditional vectors to construct augmented vectors; and estimating the conditional mutual information between paths based on the k-nearest neighbor distances in the augmented vector space to serve as the path mutual information.

[0065] Specifically, since the phase configuration matrix Φ of IRS k The dimension is very high (for example, a 32×32 IRS has 1024 phase units), and directly using it as a conditional variable will lead to the curse of dimensionality. Therefore, a method such as principal component analysis (PCA) is used to analyze each phase matrix vec(Φ k ) to reduce the dimension, extract its main components, and generate a low-dimensional conditional vector δ k For any pair of path signals (e.g. Y i and Y j ), and the corresponding conditional vector δ k Combine to form the augmented vector z=[Y i ; Y j ; δ k ]. A non-parametric method based on k-nearest neighbor (k-NN) is used to estimate the conditional mutual information I(Y i ; Y j ∣Φ k ), the calculation formula is I(Y i ; Y j ∣Φ k )=ψ(k)-〈ψ(n i +1)+ψ(n j +1)-ψ(n z +1)〉, where Ψ is the digamma function, k is the number of neighbors, and n i , n j , n z are the number of neighbor points in the projection of the corresponding space within the range determined by the k-nearest neighbor distance in the augmented vector space. This can accurately quantify the nonlinear correlation between the two path signals under a specific IRS phase configuration.

[0066] Based on the path mutual information, the direct path features and the reflected path features are weightedly fused to generate the fused features.

[0067] Furthermore, based on the path mutual information, the direct path and the reflected path are clustered to identify path pairs with high correlation; the direct path features and the reflected path features corresponding to the path pairs with high correlation are coherently merged to enhance the signal features, thereby generating fused features.

[0068] Specifically, based on the calculated path mutual information matrix, a similarity matrix is ​​constructed, and all paths are automatically divided into several clusters through algorithms such as spectral clustering. Generally, paths belonging to the same cluster have very high mutual information, which strongly suggests that they originate from the same physical interference source. For these identified high-correlation path pairs, coherent merging is performed, that is, their signals are added after compensating for the phase difference, which can effectively improve the signal-to-noise ratio and enhance weak interference characteristics. For paths between different clusters with low mutual information, their independence is maintained to retain the information brought by spatial diversity. Through this adaptive weighted fusion strategy based on mutual information, the fused feature F is generated. enhanced .

[0069] The fused features are organized in at least one dimension to construct an enhanced interference feature tensor.

[0070] Optionally, the fused features are organized in multiple dimensions such as space, time, and frequency to construct the final enhanced interference feature tensor T enhanced .

[0071] This embodiment achieves intelligent fusion and utilization of multipath information by calculating the conditional mutual information between direct and reflected paths. In navigation signal monitoring scenarios, building reflections are often considered harmful multipath. Mutual information analysis reveals that highly correlated path pairs (mutual information > 0.7) often originate from the same interference source. Coherently combining these paths can improve signal strength. This is particularly true in urban canyon environments, where direct paths may be obscured. By integrating information from strongly reflected paths, meter-level positioning accuracy can still be achieved.

[0072] According to one aspect of the present application, the step of decomposing the enhanced interference feature tensor includes: performing tensor decomposition on the enhanced interference feature tensor to obtain a factor matrix that reveals the intrinsic characteristics in different dimensions; from the factor matrix, jointly extracting the position information and pattern information of the interference source as the position and pattern parameters of the interference source.

[0073] In this embodiment, common tensor decompositions include CP (CANDECOMP / PARAFAC) decomposition or Tucker decomposition in multilinear algebra. For example, for the three-dimensional enhanced interference feature tensor T enhanced ∈R Dspace × Dtime × Dfreq Perform CP decomposition, where R is the rank of the CP decomposition, Dspace is the size of the spatial dimension, Dtime is the size of the time dimension, and Dfreq is the size of the frequency dimension; it can be expressed as the sum of multiple rank components and obtain three factor matrices: spatial factor matrix A space , time factor matrix A timeand the frequency factor matrix A freq The uniqueness of these factor matrices is that they capture the intrinsic structure and variation of the interference signal in the spatial, temporal and frequency dimensions in a decoupled manner. For example, A space The column vectors of A correspond to the spatial signatures of different interference sources. time The column vector of A describes its time-varying pattern (such as periodicity, duty cycle, etc.), and freq The column vector of reflects its spectral characteristics.

[0074] Furthermore, the steps of jointly extracting the location information of the interference source include: extracting spatial characteristic factors from the factor matrix; constructing a geometric relationship model including the reflection path of the intelligent reflection surface; combining the spatial characteristic factors with the geometric relationship model to establish and solve the positioning equation group, thereby obtaining the location information of the interference source.

[0075] Specifically, the spatial feature factor A extracted from the tensor decomposition result space , each column vector contains the spatial information related to a specific interference source, such as the angle information of the direct wave (DOA) and the reflected wave through the IRS. At the same time, a geometric relationship model is established, which accurately describes the spatial information of the interference source from the unknown interference source position P jam , passing through a reflection point P on the IRS v , arrive at the known UAV receiver position P uav The path length relationship, that is, L reflect =|P jam -P v |+|P v -P uav |, and the law of reflection must also be satisfied (the angle of incidence equals the angle of reflection). By combining the angle measurements extracted from the spatial characteristic factors with the path length constraints derived from the geometric model, an overdetermined nonlinear system of equations can be established for each interference source. For example, the direct path provides the direction constraint, while the reflected path provides the delay measurement τ reflect，k The converted path length constraint c·τ reflect，k By solving the equations using numerical optimization algorithms such as weighted least squares (WLS) or iterative extended Kalman filter (IEKF), the high-precision three-dimensional position coordinates P of the interference source can be obtained. jam .

[0076] Optionally, in order to improve the reliability of positioning, the positioning results under different IRS configurations can be used for consistency verification, or the motion continuity of the UAV can be introduced as a constraint to eliminate possible positioning ambiguity (such as the mirror point problem), thereby outputting the final unambiguous position estimate P jam，final .

[0077] According to one aspect of the present application, constructing an IRS phase coding sequence also includes: defining an interference observability index based on a Fisher information matrix; and generating an IRS phase coding sequence by optimizing the IRS phase coding sequence to maximize the interference observability index.

[0078] In this embodiment, the Fisher Information Matrix (denoted as F) is used to measure the information carried by an observable random variable (here, the received signal Y) about an unknown parameter (here, the location, direction, etc. of the interference source) jam ) is a measure of the amount of information contained in the IRS. Specifically, the elements of the Fisher information matrix characterize the second-order derivatives of the likelihood function with respect to the unknown parameters, and its inverse matrix provides the Cramer-Rao lower bound on the variance of the parameter estimates, which is the best variance achievable by any unbiased estimation method. Based on the Fisher information matrix, a scalar interference observability metric, O(Φ), can be defined. This metric is a function of the IRS phase configuration, Φ, and quantifies the theoretically optimal performance of estimating the interference source parameters under that phase configuration.

[0079] A preferred indicator is the determinant of the Fisher information matrix, that is, O(Φ)=det(F), where det represents the determinant of the matrix. This indicator is also called the D-optimality criterion. Maximizing this indicator means minimizing the confidence ellipsoid volume of the estimated parameters, thereby obtaining the highest comprehensive estimation accuracy. After defining the observability indicator, the design of the coding sequence is transformed into an optimization problem: finding a set of phase configurations {Φ k}, making the observability index O(Φ k ) is maximized. This optimization process can be solved using a variational method or a gradient optimization algorithm. For example, the optimal phase distribution gradient can be obtained by solving for ΔO / ΔΦ = 0. This ensures that the signal obtained in each measurement contains the maximum amount of information about the parameters of the interference source under test.

[0080] As a specific implementation method, the optimization process can generate elementary codes with specific physical meanings. For example, spiral phase coding can be designed to excite the orbital angular momentum pattern of the interference signal; checkerboard phase coding can be designed to generate specific spatial frequency modulation; and gradient phase coding can be designed to form highly directional disturbances. By combining these codes with different physical detection functions into a sequence, comprehensive detection of the multi-dimensional characteristics of the interference source can be achieved. This embodiment elevates the coding design problem from a signal-level optimization (such as gain change) to an information-level optimization, making the entire active detection process more targeted and with a more solid theoretical foundation, and is expected to achieve more accurate interference parameter estimation at a lower signal-to-noise ratio.

[0081] According to one aspect of the present application, the step of calculating the path mutual information also includes: defining different phase configurations loaded on the smart reflection surface as processing variables; estimating the causal effect of the processing variables on the correlation between the direct path characteristics and the reflected path characteristics through a causal inference method; and using the causal effect as the path mutual information.

[0082] In this embodiment, the signal analysis problem is transformed into a causal analysis problem similar to that in medical or social science research. Specifically, the treatment variable is defined as the phase configuration Φ applied to the IRS, which is the variable that is actively controlled and changed. The outcome variable is defined as the target to be observed, that is, the direct path signal Y direct and the reflected path signal Y reflect The correlation between them can be expressed as mutual information I(Y direct ; Y reflect ) to measure. Causal inference methods aim to estimate the net effect of processing on the results. In signal processing scenarios, the signals of the direct path and the reflected path may be affected by an unobserved environmental factor (for example, a moving scatterer) at the same time, which may cause a statistical correlation between them, but this correlation is not caused by the regulation of IRS. Causal inference methods can help eliminate the interference of such confounding factors. A feasible causal inference method is propensity score matching. Specifically, for each observation (i.e., at a certain phase Φ k The propensity score is calculated based on a series of covariates (such as time, drone location, etc.) under the condition that the observation is assigned to the current treatment (i.e., phase Φ k ). Find the control group, which is the observation that is very similar in all other covariates but has the IRS phase configured as the reference phase Φ0. By comparing the treatment group (with phase Φ k The difference between the outcome variables of the control group (phase Φ0) and the matched control group (phase Φ0) can be used to estimate the causal effect of the treatment. This causal effect, the average treatment effect (ATE), can be calculated as ATE=E[I(Y direct ; Y reflect )∣Φ=Φ k ]-E[I(Y direct ; Y reflect )|Φ=Φ0]. This ATE value cleanly quantifies the IRS phase change from Φ0 to Φ kThis event has a net impact on the mutual information between the two paths. The causal effect is taken as the path mutual information. In other words, the calculated ATE value is used to fill the path mutual information matrix I mutual Instead of focusing on the total correlation between paths, we focus only on the correlation caused by the active regulation of the IRS. This mutual information better reflects the true physical connection between paths (for example, whether they have actually been reflected by the IRS), providing a more reliable and physically interpretable basis for subsequent path clustering and feature fusion, and avoiding incorrect fusion decisions due to spurious correlations.

[0083] According to one aspect of the present application, constructing a multi-perspective spatial fingerprint matrix can also be as follows: treating each encoded response of the differential feature signal as a data point in a high-dimensional space; applying a diffusion mapping algorithm to discover the intrinsic manifold structure between the data points; extracting the low-dimensional manifold coordinates of the manifold structure as a spatial response pattern, and integrating the low-dimensional manifold coordinates to construct a multi-perspective spatial fingerprint matrix.

[0084] Specifically, for each IRS code Φ k The generated differential characteristic signal ΔY k (t), a high-dimensional feature vector can be extracted from it as its representation in high-dimensional space. This feature vector can be the difference covariance matrix R k The vector obtained after vectorization can also be the multi-dimensional differential feature tensor T mentioned in the alternative embodiment. diff Regardless of the form adopted, each encoded response is regarded as an independent data point in a high-dimensional Euclidean space. Diffusion Maps algorithm is applied to discover the intrinsic manifold structure between data points. Traditional data dimensionality reduction methods such as principal component analysis (PCA) can only discover linear structures in the data, while manifold learning methods such as diffusion mapping can reveal complex nonlinear relationships between data points. Its specific implementation method includes: calculating the similarity between all pairs of data points, constructing a similarity kernel matrix K, where the element K(T i , T j )=exp(-||T i -T j ∣∣ 2 / Ξ), where T i and T j represents the data points of the i-th and j-th encoded responses, and Ξ is the scale parameter that determines the neighborhood. By normalizing the kernel matrix, a Markov transition probability matrix P is constructed. This matrix can be understood as describing the probability of a random walk on a graph with data points as nodes and similarity as edge weights. The transition probability matrix P is subjected to eigenvalue decomposition, and its first k largest eigenvectors Ψ1, ..., Ψ kThis creates a new set of coordinates that can describe the diffusion distance between data points, or the ease of randomly walking from one point to another on the graph, in the least dimensional space. These eigenvectors are called low-dimensional manifold coordinates. For each original high-dimensional data point (i.e., each encoded response), its corresponding coordinate Ψ can be found in this new low-dimensional coordinate system. i =[Ψ1(i),…,Ψ k (i)] Low-dimensional coordinate vector Ψ i It is considered as the spatial response pattern of the coded response. Integrate the low-dimensional manifold coordinates of all coded responses, for example, all vectors Ψ i Arrange them in order to construct the final multi-view spatial fingerprint matrix.

[0085] In this embodiment, although the interference source's responses to different IRS codes manifest in complex and diverse forms, their inherent patterns of variation may follow a low-dimensional nonlinear manifold. For example, all response points may be roughly distributed on a curved surface. The diffusion mapping algorithm flattens this surface and finds its intrinsic coordinate system. Using this intrinsic coordinate system as a feature is more resistant to noise interference than using raw high-dimensional coordinates, and better captures the essential differences and connections between different responses, thereby forming a more accurate and robust interference fingerprint.

[0086] According to one aspect of the present application, the fused features can also be generated by: considering the feature distributions of the direct path features and the reflected path features as a set of probability measures; solving the optimal transmission mapping for mapping the probability measures through the optimal transmission theory to minimize the preset transmission cost; and applying the optimal transmission mapping to generate the fused features.

[0087] Specifically, the features extracted from each path (including the direct path and each reflected path) are no longer regarded as isolated feature vectors. Instead, the set of all feature sample points on each path is regarded as an empirical probability distribution, which is mathematically called the probability measure μ. i Intuitively, each measure μ can be i Imagine a pile of sand distributed in the feature space. The optimal transmission mapping is solved by the optimal transmission theory. The goal of feature fusion can be redefined as: finding the best way to combine the sand piles from different paths (i.e., the probability measure μ i ) are transported and merged into a single, fused sand pile μ fusionOptimal transport theory is a mathematical tool to solve such problems. It aims to find a transport map T that moves each sand grain x in a measure to a target location T(x) with the lowest total transport cost. This transport cost is pre-defined and can be designed according to the specific problem. For example, a simple cost function is the square of Euclidean distance c(x, y) = ||x - y||2 2 Minimizing it means hoping for the shortest geometric path of the transport. Alternatively, more complex cost functions can also be designed, for example, c(x, y) = ||x - y||2 2 + λ · L info (x, y), where λ is a trade-off parameter, and L info (x, y) is an information loss term to penalize the mapping that causes information loss during transport. By solving this constrained optimization problem (for example, using the efficient Sinkhorn algorithm), the optimal transport matrix T opt is obtained. With the optimal transport mapping T opt , it can be applied to the original path features to realize feature fusion. This process is not a simple weighted average, but a geometric deformation and reshaping of the features according to the optimal transport scheme, so that the feature distributions of different paths can be aligned and merged in a way that preserves their internal structure. The final fused feature is the best combination of all path information in the sense of optimal transport.

[0088] The embodiment provides a geometric and distribution-based feature fusion framework. Compared with the traditional fusion method which only considers the first or second order statistics of the feature vector, optimal transport can process and utilize the complete distribution information of the feature, and is particularly effective for fusing feature distributions with various shapes and complex structures. The found is not only a fused value, but also a geometric path of fusion, so that the fusion result can better preserve the internal structure of the original feature, thereby generating an enhanced feature with higher quality and less information loss.

[0089] According to an aspect of the present application, the step of jointly extracting the position information of the interference source comprises: modeling the positioning problem as a geometric constraint graph, wherein the nodes include the interference source, the receiver and the virtual anchor point on the intelligent reflecting surface, and the edges represent the distance constraints between the nodes; and using a graph neural network to learn and reason on the geometric constraint graph to optimize and solve the position information of the interference source.

[0090] In this embodiment, the node set V of the geometric constraint graph G = (V, E) represents all key entities in the positioning scenario. Specifically, the nodes include: the interference source node whose position is unknown and to be solved; the receiver node (i.e., drone) whose position is known; and a set of virtual anchor nodes whose positions are known set on the IRS surface. These virtual anchor points are not real sensors that exist physically, but are computing nodes introduced to represent the reflection path in the graph. The edge set E of the graph represents the geometric relationship and constraints between these nodes. For example, the attribute of an edge connecting a receiver node and an interference source node can be the arrival time difference (Td) of the signal. DOA ) or arrival time (T OA ) distance constraints obtained from the measurement. The total length of a path connecting an interference source node, a virtual anchor node, and a receiver node is constrained by the reflection path delay. The edge weight can be used to represent the confidence or accuracy of the corresponding measurement value.

[0091] Graph neural networks are used for learning and reasoning. Graph neural networks operate through a message-passing mechanism. Within each layer, each node aggregates information from its neighboring nodes to update its own feature representation (also known as node embedding). In this embodiment, a geometric constraint graph is used as input to the GNN. The GNN performs end-to-end learning and reasoning on the graph, aiming to adjust the position coordinates of unknown nodes (i.e., interference source nodes) in the graph to maximize the satisfaction of all geometric constraints represented by edges (such as distances and angles). In other words, by iteratively passing and aggregating constraint information between nodes, the GNN learns the complex internal connections of the entire geometric constraint system and ultimately infers the solution that best satisfies the global constraints.

[0092] The output of a GNN can directly be the estimated location coordinates of the interference source node. When well-trained, the GNN can directly provide a solution that is very close to the true location. Alternatively, the GNN output can be used as a high-quality initial value and fed into a traditional nonlinear optimizer (such as a trust region algorithm) for the final few steps of fine-tuning to obtain the final optimal location solution.

[0093] This embodiment transforms the problem of solving highly coupled nonlinear equations into a pattern recognition and graph reasoning problem. Compared to traditional solvers, which are prone to local optima and are sensitive to initial values, GNNs can learn universal patterns for solving such problems from data. They typically demonstrate greater robustness and global optimization capabilities in complex real-world scenarios with noise, measurement loss, and multipath effects.

[0094] According to one aspect of the present application, an abnormal response detection and processing mechanism may also be included. k (t) after which the instantaneous power P of the signal can beinst (t)=|ΔY k (t)| 2 Monitor by calculating its time derivative |dP inst / dt| can detect whether there are power mutation points exceeding the preset threshold η. These mutation points may indicate a sudden change in the behavior of the interference source (such as power on / off, mode switching, rapid movement, etc.), or an abnormality in the system itself. Unlike simply treating them as noise or bad points, this embodiment can perform detailed feature extraction on these abnormal segments, such as extracting the mutation amplitude, duration, spectral morphology changes, etc., to form the abnormal feature vector F anomaly This abnormal feature vector can be appended to the differential feature signal to provide additional important information for subsequent interference pattern recognition, allowing the system to not only locate the interference source but also perceive the dynamic changes in its behavior.

[0095] According to one aspect of the present application, in order to make the entire detection process more cognitively intelligent, an intelligent scheduling strategy based on information gain can be adopted in the step of determining the code loading timing. This strategy is different from the simple adaptation based on the duration of the interference burst, and introduces a feedback loop. Specifically, the system can calculate the unresolved information entropy H based on the detection results of the previous stage. residual , which quantifies the current uncertainty about the characteristics of the interference source. The loading duration of the next code is T k It can be dynamically calculated based on the information entropy, for example, T k =T base ·(1+log(1+H residual / H0)); where T base is the basic duration, and H0 is the reference entropy. The principle of this formula is that when the system is not clear about the interference source (H residual When the interference source characteristics have been basically mastered (H residual If the detection accuracy is low, the observation time is shortened and the code is quickly switched to detect other features or track the target. This cognitive-driven timing control enables detection resources to be intelligently allocated where they are most needed.

[0096] Optionally, in order to make the analysis model more consistent with physical reality, physical constraints can be introduced for regularization in the process of constructing the encoding-response mapping relationship. A mapping model learned purely by data-driven learning may violate the basic laws of electromagnetic physics, resulting in poor generalization ability. Therefore, a physical regularization term R can be added to the learning objective function. physics (f CR ). The regularization term can contain multiple physical constraints. For example, based on the reciprocity principle of electromagnetic wave propagation, the mapping function can be constrained to satisfy fCR (Φ) = f CR T (-Φ), where T is the transpose, f CR (Φ) is a mapping function from the encoding matrix Φ to the response vector. Another example is based on the law of conservation of energy, which can constrain that the signal energy generated by IRS reflection cannot exceed the total energy of the incident signal, i.e., ||f CR (Φ)|| ≤ ||Y 2 ||. input 2 where Y input is the total energy of the incident signal. By introducing these constraints in the optimization process, it can be ensured that the learned mapping model not only fits the observed data, but is also physically self-consistent, resulting in a more accurate and robust encoding-response model.

[0097] In addition, in order to improve the reliability of the final output position, a mechanism for eliminating positioning ambiguity can be introduced. In a complex environment, solving based on a single geometric constraint can result in multiple plausible position solutions (e.g., mirror points with respect to a certain plane). To eliminate this ambiguity, multiple independent positioning measurements can be made by changing the phase configuration of the IRS. If the dispersion ∑ jam，k of the set of position estimates {P pos} obtained by multiple measurements exceeds a preset threshold, it indicates that there may be ambiguity. At this time, other dimensional information can be used to assist in judgment, for example, the motion continuity constraint contained in the time-domain activity pattern vector A jam (t) of the interference source is used to select the solution that is most consistent with the historical trajectory as the final result. Another is to use a sequential test based on Bayesian inference. For each possible position solution P i , a hypothesis H i is established, and its posterior probability P(Pi|Y) is calculated according to the measurement result Y. If the maximum posterior probability does not exceed the confidence threshold, the system will automatically design and request a new IRS phase configuration that can maximize the expected information gain for a new measurement, until the posterior probability of a certain hypothesis wins by a decisive advantage.

[0098] According to an aspect of the present application, in the difference processing step, multi-dimensional difference processing can be performed, not just direct subtraction of the signal vector. It aims to reveal the response caused by IRS disturbance from different domains. Specifically, a multi-dimensional difference feature tensor T diff can be constructed. Different slices of this tensor can include: amplitude difference ΔA = |Y k |-|Y ref |, where Y k ​received signal under current coding configuration; directly reveals the power variation of interference signal under specific coding; phase difference ΔΦ = unwrap(∠Y k -∠Y ref ), where unwrap() is the phase unwrap function, and ∠ is the element-wise phase; reveals the disturbance of IRS on the phase of interference signal, which is particularly effective for analyzing coherent interference sources; and spectrum difference ΔS = FFT(Y k )- FFT(Y ref ), where FFT is the fast Fourier transform; reveals the response characteristics of the interference signal in the frequency domain, which helps to identify the type of interference. By integrating these different domain difference features into a tensor structure, we can provide richer input than a single signal vector for subsequent manifold learning or covariance analysis, thereby constructing a wider information dimension interference fingerprint.

[0099] Optionally, in the path correlation calculation, multi-scale correlation analysis can be performed to mine the correlation patterns between paths under different time scales. The behavior of the interference source and the change of the channel environment often exhibit different characteristics at different time scales. Therefore, a multi-scale analysis tool such as wavelet transform can be used to decompose the mutual information sequence I(t) between the two path signals Y i and Y j over time, obtaining the wavelet coefficients WI(s, t) at different scales s and times t. By analyzing these coefficients, a multi-scale correlation tensor I scale can be constructed. Different scale slices of this tensor have different physical meanings: strong correlation (I short ) at short time scales may reflect the transient response characteristics caused by IRS rapid switching; correlation (I medium ) at medium time scales may reveal the periodic behavior patterns of the interference source itself; and correlation (I long ) at long time scales may reflect the trend characteristics caused by the movement of the UAV or the slow change of the environment. This enables the system to separate and identify path correlation patterns driven by different physical mechanisms from dynamically changing data, deepening the understanding of channel and interference source behavior.

[0100] In a specific numerical calculation case, specific scenarios and initial parameters are set. Initial parameters: Intelligent Reflecting Surface (IRS): Assume a KxL = 8x8 = 64 element uniform planar array. Signal wavelength: λ = 0.05 meters (corresponding to a 6 GHz frequency band). IRS element spacing: d = λ / 2 = 0.025 meters. Initial interference feature vector provides information: the preliminary estimate of the angle of arrival (AOA) from V init is AOArough =[θ0=30°, δ0=45°]. Angle uncertainty and search range: assuming the angle uncertainty is ±5°, the search range of the sensitive angle is set as θ ∈ [25°, 35°] and δ ∈ [40°, 50°]. Search grid discretization: to simplify the calculation, the above search range is discretized into a G x G = 3 x 3 grid, i.e., θ takes {25°, 30°, 35°} and δ takes {40°, 45°, 50°}, for a total of 9 angle directions.

[0101] At the above 9 discrete angle directions, the influence of the IRS phase change on the far-field gain is evaluated in terms of its severity. Specifically, the IRS gain function G(θ, δ) is calculated at a certain reference phase (e.g., Φ0is all-zero phase) and the gradient of the gain function is calculated as IRS . IRS The gradient of the gain function is calculated as sensitive .

[0102] The phase encoding Φ1is constructed such that it produces the maximum total gain variation in the three directions within the set Ω sensitive .

[0103] The phase encoding Φ2 may be constructed to produce another disturbance pattern in the sensitive direction, such as a change in the sidelobe shape. Example results: Assume that the second phase encoding matrix Φ2 is obtained by a similar method. Now it is necessary to verify whether Φ1 and Φ2 meet the requirements of detection independence. Vectorize these two 8×8 matrices into vectors of length 64, respectively, and record them as vec(Φ1) and vec(Φ2). Calculate the cross-correlation value between them: ρ(Φ1, Φ2) = |〈vec(Φ1), vec(Φ2)〉| / (||vec(Φ1)||·|| vec(Φ2)||); where 〈 〉 represents the vector inner product, and ||·|| represents the Euclidean norm of the vector. The specific numerical calculation is: Assume that after calculation, the vector inner product 〈vec(Φ1), vec(Φ2)〉=1.2π 2 , ||vec(Φ1)||=4.1π, ||vec(Φ2)||=3.8π. Then the cross-correlation value is: ρ(Φ1, Φ2)= ||1.2π 2 | / 4.1π·3.8π≈0.077. This calculated result is compared with a preset threshold (e.g., 0.1). Since 0.077 < 0.1, the two phase codes Φ1 and Φ2 are determined to be sufficiently independent, meeting the detection independence requirement. If the calculated result is greater than 0.1, orthogonal projection adjustment is performed on one of the codes (e.g., Φ2), and the cross-correlation value is recalculated until the requirement is met.

[0104] By repeating the above steps, a complete sequence {Φ opt Each code in this sequence is optimized for the sensitive direction of the interference source, and the codes are highly independent of each other, laying a solid foundation for subsequent high-quality differential signal acquisition and spatial fingerprint construction.

[0105] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the scope of protection of the present invention.

Claims

1. A method for collecting navigation interference source monitoring data, characterized in that: include: Extracting the initial interference feature vector of the original RF signal; Based on the initial interference feature vector, the phase coding sequence of the intelligent reflector (IRS) is generated, and the differential feature signal is generated using the phase coding sequence of the IRS. Based on this, a multi-view spatial fingerprint matrix is ​​constructed. According to the multi-view spatial fingerprint matrix, the path features are fused to generate an enhanced interference feature tensor; Decompose the enhanced interference feature tensor, obtain the location and pattern parameters of the interference source, and generate a comprehensive feature report of the interference source; Among them, constructing a multi-view spatial fingerprint matrix includes: Based on the initial interference eigenvector, the IRS phase coding sequence is constructed to apply a controllable electromagnetic response disturbance to the estimated interference direction; The phase coding sequence and reference phase configuration of the IRS are loaded using a preset timing to capture the corresponding coded response signal and reference response signal respectively; Performing differential processing on the coded response signal and the reference response signal to obtain a differential characteristic signal; The spatial response pattern of the differential characteristic signal is analyzed to construct a multi-perspective spatial fingerprint matrix; this includes: constructing a corresponding differential covariance matrix for each phase code in the IRS phase coding sequence based on the differential characteristic signal; extracting a spatial pattern matrix from the differential covariance matrix to characterize the response characteristics of the interference source to the phase code; integrating the spatial pattern matrices obtained under different phase codes to structuredly generate a multi-perspective spatial fingerprint matrix.

2. The method according to claim 1, characterized in that Construct the phase encoding sequence of IRS, including: Extracting an initial estimate of the arrival angle of the interference signal from the initial interference feature vector; Based on the initial estimation of the angle of arrival, the interference direction sensitive angle set is defined; The phase coding sequence of the IRS is optimized and designed to enable the smart reflector to produce the maximum gain change within the sensitive angle set of the interference direction, thereby achieving controllable electromagnetic response disturbance.

3. The method according to claim 2, characterized in that Optimizing the design of the IRS phase coding sequence also includes: Calculate the cross-correlation values ​​between different phase codes in the phase code sequence of IRS; If the cross-correlation value exceeds a preset threshold, at least one phase code is adjusted to reduce the cross-correlation value, thereby ensuring that each phase code constituting the phase code sequence of the IRS meets the detection independence requirement.

4. The method according to claim 1, wherein Obtain differential characteristic signals, including: Using a preset timing, inserting a reference phase configuration loading into the loading gap of each phase code in the phase code sequence of the IRS; respectively collecting signals during the loading of each phase encoding and reference phase configuration to obtain encoding response signals and reference response signals; The difference between the coded response signal and the reference response signal is calculated to generate a differential characteristic signal.

5. The method according to claim 1, wherein Determination of preset timing includes: Extracting the interference burst duration distribution from the interference time-domain activity pattern contained in the initial interference feature vector; Determining the coding loading duration according to the interference burst duration distribution; The encoding loading duration is used as a key parameter for preset timing.

6. The method according to claim 1, characterized in that Generate enhanced interference feature tensor, including: Separate direct path features and reflected path features from the multi-view spatial fingerprint matrix; Calculate the path mutual information between the direct path characteristics and the reflected path characteristics to quantify the correlation between different paths; Based on the path mutual information, the direct path features and the reflected path features are weightedly fused to generate the fused features; The fused features are organized in at least one dimension to construct an enhanced interference feature tensor.

7. The method according to claim 6, characterized in that Calculate path mutual information, including: Perform dimensionality reduction processing on different phase configurations loaded on the smart reflector to generate corresponding conditional vectors; Combine the direct path features, the reflected path features and the conditional vector to construct an augmented vector; Based on the k-nearest neighbor distances in the augmented vector space, the conditional mutual information between paths is estimated as the path mutual information, where k is the preset number of neighbors.

8. The method according to claim 6, characterized in that Generate fused features, including: Based on the mutual information of paths, the direct paths and reflected paths are clustered to identify the highly correlated path pairs. Coherent merging is performed on the direct path features and reflected path features corresponding to the highly correlated path pairs to generate fused features.

Images