Detection signal anti-multipath interference method and system based on multi-modal feature fusion

Abstract

The invention relates to a detection signal anti-multipath interference method and system based on multi-modal feature fusion, and belongs to the technical field of radio detection, and the method comprises the following steps: 1, unmanned plane signal fingerprint feature extraction: analyzing physical layer parameters of a signal; step 2, unmanned aerial vehicle signal airspace DOA estimation and error analysis: acquiring signal spatial spectrum information by using an array antenna; 3, physical modeling of a ship environment: constructing a multi-path reflection path prediction model in combination with the ship three-dimensional model; 4, dynamic DOA error correction, wherein DOA estimation deviation caused by multiple paths is suppressed through iterative optimization of signal features and a physical model; 5, multi-modal feature fusion decision making: constructing a feature fusion mechanism based on Bayesian probability, and distinguishing a real signal from a false alarm signal; the method has the beneficial effects that suppression of the multipath effect is realized, and unmanned aerial vehicle signal detection in a dynamic complex scene is facilitated.

Description

Technical Field

[0001] The present invention belongs to the technical field of radio detection, and particularly relates to a method and system for suppressing multipath interference of detection signals based on multi-modal feature fusion. Background Art

[0002] With the wide application of unmanned aerial vehicle (UAV) technology, the signal detection and positioning and direction-finding capabilities of UAVs have become the core requirements for airspace safety supervision. The current technical system is mainly built based on traditional radio detection: one category relies on signal physical layer features (such as: sudden change in spectrum energy, time-domain waveform matching) to achieve target recognition, and the other category depends on geometric positioning principles (such as: TDOA time difference of arrival positioning, DOA direction of arrival estimation) to complete spatial positioning and direction-finding. These methods perform well under ideal conditions of line-of-sight propagation and static scenarios, but in strong reflection scenarios such as urban building complexes and sea ships, the signal aliasing problem caused by multipath effects severely restricts the detection efficiency. Among them, the multipath components generated by the reflection and diffraction of electromagnetic waves through complex environments not only cause signal feature distortion, but also form false signal sources, resulting in problems such as an increased false alarm rate and deteriorated positioning accuracy of traditional detection systems.

[0003] In terms of suppressing signal multipath effects, there are currently the following three main problems:

[0004] 1) The signal characterization dimension is single. Traditional methods often independently analyze time-domain, frequency-domain, or spatial-domain features, lacking the ability to jointly model multi-dimensional features and making it difficult to effectively separate direct signals from multipath components;

[0005] 2) The environmental perception ability is insufficient. The existing static channel assumptions do not fully consider the time-varying characteristics of the propagation medium, resulting in poor adaptability to dynamic environments;

[0006] 3) The signal processing dimension is fragmented. The front-end detection and back-end positioning links usually adopt a cascaded processing architecture, and cascading errors will accumulate and amplify step by step on the processing link. Summary of the Invention

[0007] The present invention provides a method and system for suppressing multipath interference of detection signals based on multi-modal feature fusion, which is used to solve the technical problem of suppressing the multipath transmission effect of UAV signals. By establishing a three-dimensional joint analysis framework of "signal fingerprint - spatial direction of arrival - environmental physical constraint", the suppression of multipath effects is achieved, and it is convenient for UAV signal detection in dynamic and complex scenarios.

[0008] To achieve the above object, the present invention is realized through the following technical solutions:

[0009] A method for suppressing multipath interference of detection signals based on multi-modal feature fusion includes the following steps:

[0010] Step 1: UAV signal fingerprint feature extraction: Analyze the physical layer parameters of the signal;

[0011] Step 2: UAV signal spatial DOA estimation and error analysis: Use an array antenna to obtain the signal spatial spectrum information;

[0012] Step 3: Physical modeling of the ship environment: Combine the ship's 3D model to construct a multipath reflection path prediction model;

[0013] Step 4: Dynamic DOA error correction: Suppress the DOA estimation bias caused by multipath through iterative optimization of signal features and physical models;

[0014] Step 5: Multimodal feature fusion decision-making: Construct a feature fusion mechanism based on Bayesian probability to distinguish real signals from false alarm signals.

[0015] Optionally, in Step 1, the analysis of the physical layer parameters of the signal includes: analysis of the center frequency point and bandwidth, extraction of the frequency hopping sequence, and identification of the modulation mode.

[0016] Optionally, extract the data of the center frequency point and bandwidth through short-time Fourier transform, detect the frequency hopping period and frequency hopping sequence of the signal, and then identify the modulation mode of the UAV signal based on high-order cumulants.

[0017] Optionally, in Step 2, the array antenna consists of multiple antenna elements. When the signal is incident on the array antenna from different directions, due to the different distances between each antenna element and the signal source, different phase delays will be generated. By measuring the phase delay and signal amplitude, an array signal model is established to evaluate the signal source information.

[0018] Optionally, the signal spatial spectrum information includes: constructing an array signal model and multi-signal classification spatial spectrum estimation.

[0019] Optionally, in Step 3, regard the wireless signal as light and analyze the propagation path of the wireless signal in the environment according to the principle of geometric optics.

[0020] Optionally, in Step 2, the multipath reflection path prediction model adopts three-dimensional coordinate transformation and multipath reflection path prediction.

[0021] Optionally, in Step 4, suppressing the DOA estimation bias caused by multipath adopts joint estimation of multipath parameters and dynamic compensation of the ship platform attitude perturbation.

[0022] Optionally, in Step 5, dynamic DOA error correction adopts feature likelihood function modeling and adaptive decision-making.

[0023] The detection signal anti-multipath interference system based on multimodal feature fusion includes:

[0024] An extraction module for extracting UAV signal features;

[0025] An analysis module for UAV signal spatial DOA estimation and error analysis;

[0026] A physical modeling module for physical modeling of ship environments;

[0027] A correction module for dynamic DOA error correction;

[0028] A decision-making module for modal feature fusion decision-making;

[0029] The extraction module is connected to the analysis module, the analysis module is connected to the physical modeling module, the physical modeling module is connected to the correction module, and the correction module is connected to the decision-making module.

[0030] Advantages of the present invention:

[0031] 1. By constructing a multi-dimensional fusion architecture of "signal fingerprint - spatial direction of arrival - environmental physical constraints", the present invention deeply couples the signal features processed independently in the traditional way with the spatial geometric information and dynamic environmental parameters. This architecture not only uses the signal fingerprint features to provide physical layer identity authentication, but also converts the multipath reflection path prediction into a constraint condition for DOA estimation through the ship's three-dimensional model, enabling the system to actively identify and separate the direct signal and multipath interference. Compared with the traditional single-dimensional suppression method, the present invention reduces the false alarm rate while maintaining the detection sensitivity of low SNR signals.

[0032] 2. By combining the dynamic environment coupling mechanism and the adaptive decision-making framework, the present invention solves the time-varying problem of multipath effects on mobile platforms. Through real-time fusion of ship attitude data and array manifold correction, it can accurately compensate for the wavefront distortion caused by the movement of the carrier. The multi-modal decision-making strategy can dynamically adjust the classification threshold according to the SNR and environmental stability parameters. This closed-loop optimization mechanism enables the system to maintain stable signal detection performance in complex electromagnetic environments and dynamic motion scenarios, significantly improving the applicability of the existing UAV detection system on ship platforms in strong reflection scenarios.

[0033] 3. The modular technical architecture proposed by the present invention has both engineering practicality and technical extensibility. Each module in the signal processing link adopts a loose coupling design, which can not only be quickly deployed relying on the existing TDOA / DOA hardware platform, but also be compatible with the extended access of future new signal features, making the system have a wide range of application prospects in the fields of port security and urban low-altitude supervision, and providing key technical support for building an all-weather and all-airspace UAV monitoring network. Description of the Drawings

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0035] Figure 1 It is the system structure diagram of the present invention;

[0036] Figure 2 It is the working flow chart of the present invention. Detailed implementation manners

[0037] The following will describe the embodiments of the present application in detail with reference to the drawings.

[0038] Multipath effect: The phenomenon that electromagnetic waves propagate through different paths (such as reflection and diffraction) to reach the receiving end, resulting in signal delay spread, phase distortion, and energy superposition. The interference between the multipath components and the direct signal will cause fluctuations in the amplitude of the received signal, waveform distortion, and spatial positioning deviation.

[0039] Signal fingerprint feature extraction: A technology that establishes a unique identifier by extracting the inherent features of the physical layer of the signal (such as spectral envelope, modulation parameters, or frequency hopping sequence) to achieve the authentication of the transmitting source and the identification of interference signals.

[0040] Frequency hopping sequence extraction: Extracting the spread-spectrum communication signal whose carrier frequency of the signal jumps periodically according to a predetermined random sequence. The instantaneous bandwidth of the spread-spectrum communication signal is narrow, but the total bandwidth remains unchanged. By frequency hopping, interference can be avoided and the anti-interception ability of communication can be enhanced.

[0041] Support Vector Machine (SVM) classifier: A supervised learning model based on structural risk minimization. It maps low-dimensional inseparable data to a high-dimensional feature space through a kernel function, constructs an optimal hyperplane to achieve pattern classification, and in pattern classification recognition, uses the principle of maximizing the classification margin to improve the generalization ability under the condition of small samples.

[0042] MUSIC (Multiple Signal Classification), multi-signal classification: A high-resolution spatial direction of arrival combined with the spatial DOA estimation algorithm based on subspace decomposition. By analyzing the covariance matrix of the received signal, the signal subspace and the noise subspace are separated, and a spatial spectrum function is constructed using the orthogonality between the signal subspace and the noise subspace.

[0043] Embodiment 1;

[0044] As Figure 1 shown, the multi-modal feature fusion-based detection signal anti-multipath interference system includes:

[0045] An extraction module for extracting the signal features of an unmanned aerial vehicle (UAV).

[0046] An analysis module for estimating the direction of arrival (DOA) in the airspace of the UAV signal and analyzing the error.

[0047] A physical modeling module for physically modeling the ship environment.

[0048] A correction module for correcting the dynamic DOA error.

[0049] A decision-making module for making decisions on modal feature fusion.

[0050] The extraction module is connected to the analysis module, the analysis module is connected to the physical modeling module, the physical modeling module is connected to the correction module, and the correction module is connected to the decision-making module.

[0051] Work is carried out in sequence according to the extraction module, the analysis module, the physical modeling module, the correction module, and the decision-making module.

[0052] Embodiment 2

[0053] Based on Embodiment 1, as Figure 2 shown, this embodiment provides a method for detecting signal multipath interference resistance based on multimodal feature fusion, including the following steps:

[0054] Step 1: Extracting the fingerprint features of the UAV signal: Analyzing the physical layer parameters of the signal; among them, the analysis of the physical layer parameters of the signal includes: analysis of the center frequency point and bandwidth, extraction of the frequency hopping sequence, and identification of the modulation method.

[0055] Extracting the UAV signal features:

[0056] The center frequency point and bandwidth data are extracted through the short-time Fourier transform (STFT), the frequency hopping period and frequency hopping sequence of the signal are detected, and then the modulation method of the UAV signal is identified based on the high-order cumulant (HOC).

[0057] (1) Analysis of the center frequency point and bandwidth;

[0058] Assume that the discrete-time signal sequence received by the radio detection device is , and the short-time Fourier transform calculation formula:

[0059] ;

[0060] Among them, is the discrete-time index, representing the sampling points of the signal at different times; is the continuous-time variable, that is, different times, and is also the center position of the window function. By changing By varying the value, the window function can be shifted along the time axis, enabling spectral analysis of different time segments of the signal. It is usually an integer. is the window function, and the discrete-time signal sequence is There is overlap. The step size of each window function shift can be adjusted as needed. If the step size is 1, the time resolution of the analysis is the highest, but the computational complexity is also the greatest. If the step size is greater than 1, the computational complexity will be reduced, but the time resolution will decrease.

[0061] is a complex exponential function. is the imaginary unit, used for spectral analysis. is the number of points of the discrete Fourier transform (DFT).

[0062] Specifically, the Hanning window can reduce spectral leakage. The Hanning Window is used to weight each frame of the signal to balance the main lobe width and sidelobe suppression ability. The window size can be dynamically adjusted according to the hopping speed, and the typical value is 5 ms, which is suitable for the analysis of UAV signals with a hopping period ≥ 10 ms.

[0063] is the frequency index. By calculating the spectral amplitude , a time-frequency diagram can be generated. In the time-frequency diagram, for each time slice (corresponding to a frame of the signal, and a frame of the signal is ), spectral peak detection is performed, and the frequency points with amplitudes exceeding the detection threshold are retained as candidate frequency modulation points. Through spectral peak detection, the center frequency and bandwidth of the signal are obtained.

[0064] The center frequency refers to the center position of each frequency interval. For the STFT of a discrete-time signal, the center frequency corresponding to the frequency index can be calculated by the formula , where is the sampling frequency of a frame of the signal being , and is the number of points for the discrete Fourier transform (DFT).

[0065] The bandwidth refers to the frequency range covered by each frequency interval. For the STFT of the window function, its bandwidth is related to the spectral characteristics of the window function and the number of DFT points. Find the minimum frequency point and the maximum frequency point among the candidate frequency modulation points, then the signal bandwidth can be expressed as: .

[0066] The center frequency determines the position of the bandwidth in the entire spectrum. The selection of the center frequency and bandwidth will directly affect the performance of the communication system. A higher center frequency can provide a wider available bandwidth.

[0067] (2) Hopping sequence extraction;

[0068] If there are multiple candidate hopping frequencies within a single frame, the highest amplitude point is preferentially selected as the current hopping frequency. To eliminate false hopping caused by transient interference or noise, the current hopping frequency can be smoothed by combining adjacent candidate hopping frequencies through a sliding window.

[0069] Hopping period estimation:

[0070] Statistical moments of the hopping frequency switching , and then calculate the autocorrelation function of the time interval sequence to find the period value corresponding to the periodic peak , and the mode is statistically counted as the hopping period :

[0071] ;

[0072] where is a statistical distribution function mainly used to return the value with the highest frequency of occurrence in the dataset, that is, the mode.

[0073] Frequency-channel mapping :

[0074] ;

[0075] where is the lower frequency limit, is the channel interval, is the frequency occupied by each channel, represents rounding down.

[0076] (3) Signal modulation mode recognition;

[0077] The recognition of the UAV signal modulation mode is based on the high-order cumulant (HOC) and includes the recognition of the modulation types of BPSK, QPSK, and OFDM.

[0078] Intercept the signal segment (i.e., the discrete-time signal sequence), calculate the fourth-order cumulants and :

[0079] ;

[0080] ;

[0081] Among them, is the joint cumulant function, is the complex conjugate of is the fourth-order origin moment of is the second-order central moment variance of is the second-order origin moment variance of

[0082] In practical applications, finite sample data is usually selected to estimate the fourth-order cumulant and :

[0083] , where is the fourth-order origin of is the second-order center of is the total number of finite sample data;

[0084] , where is the second-order origin of

[0085] For common signal modulation methods, the values of the fourth-order cumulant are:

[0086] QPSK modulation: ;

[0087] BPSK modulation: ;

[0088] 16-QAM modulation: ;

[0089] OFDM modulation: .

[0090] According to the high-order cumulant characteristics of different modulation methods, the discrimination of modulation methods can be realized through a support vector machine (SVM) classifier.

[0091] Step 2: UAV signal spatial DOA estimation and error analysis: Use an array antenna to obtain signal spatial spectrum information; among them, the signal spatial spectrum information includes: constructing an array signal model and multiple signal classification spatial spectrum estimation;

[0092] The array antenna is composed of multiple antenna elements. When signals are incident on the array antenna from different directions, due to the different distances between each antenna element and the signal source, different phase delays will be generated. By measuring these phase delays and signal amplitudes, an array signal model can be established to evaluate the signal source information of the signals.

[0093] DOA Estimation and Error Analysis of UAV Signals in the Airspace:

[0094] (1) Construct an array signal model;

[0095] Suppose the radio signal receiving device receives signal sources, then the array output signal is:

[0096] ;

[0097] where is the signal received at time in continuous time, is the array manifold matrix, that is, is the signal response from direction , the signal response from direction , is a source signal in the far field, is the additive noise on the array element, is at time , is the direction of arrival of the th signal, usually the angle relative to the array normal direction, is the wavelength of the signal, is the spacing between adjacent array elements, is the imaginary unit, , is used to describe the geometric relationship between the direction of arrival of the th signal and the array, that is, calculate the wave path difference between array elements caused by different signal incident angles, and then obtain the phase difference, is the conjugate transpose.

[0098] (2) MUSIC (Multiple Signal Classification) spatial spectrum estimation;

[0099] Calculate the covariance matrix of the received signal as:

[0100] ;

[0101] where the covariance matrix is a symmetric matrix. Physically speaking, the signal and The covariance and and The covariances of are the same, reflecting the correlation between the two signal elements, so the covariance matrix is ​​symmetric about the main diagonal. For signal, is the conjugate transpose of the signal; represents the mathematical expectation, It's a signal The mean vector of ;

[0102] is the covariance matrix For any non-zero vector , both ≥0, this property can be understood from the definition of covariance, As a signal After linear transformation A measure of the variance after , the variance is always non-negative, so a≥0. From the perspective of the matrix, semi-positive definiteness means that the eigenvalues ​​of the covariance matrix are all greater than or equal to zero, which ensures the stability and solvability of the covariance matrix in mathematical operations and signal processing applications.

[0103] Diagonal elements of the covariance matrix It's a signal Variance , which indicates the signal The degree of fluctuation around its mean reflects the signal The larger the variance of energy distribution in a certain dimension, the more drastic the change of the signal in a certain dimension, and the more energy it carries. for The received signal The number of samples.

[0104] Through eigenvalue decomposition, the noise subspace is obtained and the spatial spectrum function is constructed as follows:

[0105] ;

[0106] in, Reflected in different angles The distribution of signal energy on From a certain angle When incident on the array, if the angle matches the actual arrival angle of the signal source, a peak value will appear in the spatial spectrum function at this angle.

[0107] Denominator of the spatial spectrum function Represents an array manifold vector In the noise subspace matrix The square of the projection length on is the array manifold vector 's conjugate transpose, is the noise subspace matrix 's conjugate transpose, is the vector projection.

[0108] When it is 0, tends to infinity and forms a peak, while when is not the arrival angle of the signal source, is not completely orthogonal to the noise subspace, when it is greater than 0, is a finite value and relatively small. Therefore, by searching for 's peak position, the arrival angle of the signal source can be estimated.

[0109] The spatial spectrum function divides the observation space into a signal subspace and a noise subspace based on eigenvalue decomposition, and uses the projection characteristics of the array manifold vector on the noise subspace to estimate the arrival angle of the signal source through the peak of the spatial spectrum function.

[0110] (3)The DOA error caused by multipath effects;

[0111] When there are multipath components, the actual array manifold deformation is:

[0112] ;

[0113] Among them, , are respectively the attenuation coefficient and phase shift of the th path, is the single-path steering vector corresponding to the incident angle , is the imaginary unit. The traditional MUSIC algorithm will generate false peaks under this condition, resulting in a large deviation in angle estimation. represents the phase factor of the th multipath component, is the phase shift of this multipath signal relative to the reference phase;

[0114] The actually received array manifold vector is the result of the combined action of multipath components, and is the steering vector corresponding to each multipath component after amplitude weighting and phase adjustment It is formed by superposition. Each multipath component contributes to the total array manifold, and the contribution size is determined by the amplitude and phase.

[0115] Step 3: Physical modeling of the ship environment: Combine the three-dimensional ship model (including real-time attitude data for three-dimensional coordinate transformation) to construct a multipath reflection path prediction model; that is, regard the wireless signal as light, and analyze the propagation path of the signal in the environment according to the principles of geometric optics, such as reflection, refraction, and diffraction.

[0116] Physical modeling of the ship environment:

[0117] (1) Three-dimensional coordinate transformation;

[0118] Establish the ship body coordinate system and the global coordinate system conversion relationship:

[0119] ;

[0120] Among them, the rotation matrix of the ship global coordinate system is and the rotation matrix of the ship body coordinate system is , is the attitude rotation matrix, that is, real-time attitude data, is the roll, is the pitch, is the yaw, is the yaw angle of rotation around the z-axis, is the pitch angle of rotation around the y-axis, is the roll angle of rotation around the x-axis, is the translation vector.

[0121] When the ship is sailing at sea, its position and attitude are constantly changing. Through the conversion formula, the motion parameters in the ship body coordinate system, such as speed, acceleration, and attitude angle, can be converted into the global coordinate system, so as to accurately describe the motion state of the ship in a unified global reference frame, which is convenient for navigation, monitoring, and command.

[0122] (2) Multipath reflection path prediction;

[0123] Based on the geometric optical theory and Fermat's principle, solve the reflection path of the radio signal through the ship's rigid body structure by the ray tracing method.

[0124] The solution of the signal reflection path needs to meet dual conditions:

[0125] Path length extreme value condition: The reflection point makes the total transmission path obtain the minimum value, is the emission point, is the receiving point; is the modulus (i.e., length) of the path vector from point to point, is the modulus of the path vector from point to point.

[0126] The physical meaning of the above extreme value condition is that when light (or radio signals) propagates along a path that satisfies Fermat's principle, the path length takes an extreme value. Fermat's principle states that when light propagates in a medium, the actual path is the one that makes the optical path take an extreme value (minimum value, maximum value, or stable value). In the case of reflection problems, it means that the path of the reflected light makes the total path length satisfy a certain extreme value condition, thus determining the position of the reflection point.

[0127] Snell's law of reflection: The angle of reflection is equal to the angle of incidence, i.e., the surface normal vector at the reflection point satisfies:

[0128] ;

[0129] wherein, is the path from point to point, is the path from point to point;

[0130] , are all unit vectors after normalization, i.e., is the incident unit vector, is the reflected unit vector, satisfying , are both 1.

[0131] After normalization, the vector only retains the direction information, and the length is unified to 1. After normalizing the normal vector, it can be used for dot product operations in lighting calculations (e.g., calculating the cosine values of the incident angle and the reflection angle).

[0132] Define the reflection path optimization problem as:

[0133] ;

[0134] wherein, is the optimal reflection point, is the reflection point, is the set of surface points of the ship's 3D model, is the square of the modulus of the path vector from point to point, is the path from point The square of the modulus of the path vector of the point To calculate the optimal reflection point

[0135] By introducing the Lagrange operator (Lagrange multiplier), taking the law of reflection as a constraint condition :

[0136] ;

[0137] where is the incident direction vector is the reflection direction vector is the non - linear constraint is to minimize the optical path (i.e., minimize the path length).

[0138] The Lagrange multiplier method takes the law of reflection as a constraint condition. By constructing the Lagrange function and solving the extreme value condition, the path that satisfies the law of reflection can be found

[0139] Deriving with respect to the reflection point gives the stationary point condition:

[0140] ;

[0141] where is the optical path, that is, the sum of the incident unit vector and the reflection unit vector is the normal component, that is the component is in the opposite direction of the incident unit vector, and the component is in the opposite direction of the reflection unit vector

[0142] means that at the reflection point the incident unit vector is parallel to the component in the opposite direction of the incident unit vector, and the reflection unit vector is parallel to the component in the opposite direction of the reflection unit vector, which means that the stationary point condition automatically implies the law of reflection

[0143] Simplifying gives the vector form of the law of reflection:

[0144] ;

[0145] The left side of this formula is the sum of the incident unit vector and the reflection unit vector, and the right side is the scaling of the normal vector with the scaling coefficient being .

[0146] This equation accurately describes the geometric relationship at the reflection point. By numerical calculation (such as: Newton - Raphson method) on the point set of the ship's three - dimensional model, the effective reflection point set is obtained

[0147] Step 4: Dynamic DOA Error Correction: Through the iterative optimization of signal features and physical models, suppress the DOA estimation bias caused by multipath;

[0148] Dynamic DOA Error Correction:

[0149] (1) Joint estimation of multipath parameters;

[0150] In the radio detection array antenna received signal model, assume there are multipath transmission components, then the output signal of the receiving antenna array is expanded as:

[0151] ;

[0152] where, is the signal received at time in continuous time, corresponds to the direct path of the signal when it is 0, is the amplitude coefficient of the th signal component, which controls the intensity of the corresponding signal component and is a real number, is the complex attenuation coefficient, is a vector, called the array manifold vector, is the parameter related to the th signal component. For example, in array signal processing, it often represents the direction of arrival (DOA, Direction-Of-Arrival) of the signal. The vector reflects the spatial characteristics of the signal on the array.

[0153] is a complex exponential term, is the imaginary unit, is the phase offset, that is, is the phase of the th signal component. This complex exponential term is used to represent the phase information of the signal. In signal processing, the phase has an important impact on the synthesis and interference characteristics of the signal.

[0154] is the source signal (or transmitted signal), which is a function of time and describes the variation law of the signal with time.

[0155] is the additive noise, which is also a function of time . In the actual signal reception process, noise is an inevitable interference factor. It will be superimposed on the useful signal and affect the detection and processing of the signal.

[0156] Establish the parameter estimation optimization problem:

[0157] ;

[0158] Among them, is a signal. This formula is an optimization method that minimizes an expression by selecting an appropriate set of parameters . Here, min represents finding the minimum value, represents the norm of a vector (Euclidean norm), which is used to measure and The "distance" between them, that is, the degree of difference between the two. The goal of optimization is to minimize this difference, is a summation operation, indicating that Multipath transmission components are accumulated. By adjusting , and , a desired signal form can be synthesized.

[0159] Limited by the physical constraints of the ship platform environment:

[0160] ;

[0161] represents a set, is a set defined by and as parameters. The constraint set is jointly determined by the surface point set of the ship's three-dimensional model and the real-time attitude matrix . is the symbol for belonging to a set, belongs to . By parameterizing the feasible solution space, physically unreachable reflection paths are excluded, indicating that the th signal component must be within the range of the surface point set of the three-dimensional model and the real-time attitude matrix .

[0162] (2) Dynamic compensation for ship platform attitude disturbances;

[0163] The change in the ship platform attitude will cause a time-varying distortion of the received signal manifold of the array antenna. Define the array position disturbance model in the body coordinate system:

[0164] ;

[0165] Among them, is the Jacobian matrix of the ship platform attitude angular velocity, is the Jacobian matrix, is the attitude angular velocity vector of the ship platform, represents the matrix which is a 3-by-3 real matrix. In the ship attitude problem, it usually corresponds to the conversion relationship of the angular velocities related to the three attitude angles of roll, pitch, and yaw;

[0166] is the attitude angle perturbation, is the roll angle, is the pitch angle, is the yaw angle, is the transpose of the vector. The attitude angle perturbation refers to the small change in the attitude angles (such as pitch angle, yaw angle, and roll angle) of an object in space relative to a certain reference attitude, and it is used to accurately describe the deviation of the object's attitude from the ideal or nominal attitude.

[0167] The array manifold correction term is:

[0168] ;

[0169] where the array manifold correction term is used to describe the response error or perturbation of the array antenna in the direction ; represents the correction amount of the array manifold (i.e., the ideal response of the array in the direction caused by the element position error . The position deviation of each element will introduce a phase error, and the correction term is obtained by superimposing the contributions of all elements.

[0170] is the linear phase perturbation of the element position error in the direction ; is the element at the ideal position with the original phase delay, is the imaginary unit, is the summation of the contributions of all elements to obtain the total correction amount.

[0171] is the unit vector of the signal beam arrival direction, is the transpose of the vector, is the wavelength, combined with the wavelength represents the propagation direction of the signal in space, is the projection component of the incident direction on the y-axis, is the projection component of the incident direction on the x-axis.

[0172] The array manifold after real-time compensation is:

[0173] ;

[0174] is achieved by adding the error term to the ideal manifold to dynamically correct the array response, thereby improving the robustness and accuracy of algorithms such as beamforming and direction finding;

[0175] represents the time-varying or fixed error introduced by factors such as hardware defects, temperature changes, and signal interference in the actual environment;

[0176] represents the response of the array antenna to the direction under error-free conditions, which is usually a complex vector containing the amplitude and phase information of each array element;

[0177] By estimating the attitude disturbance amount through an online Kalman filter, the dynamic calibration of the array response is achieved. Through real-time compensation, the system can more accurately reflect the response characteristics of the actual array and improve its performance in complex environments.

[0178] Step 5: Multi-modal feature fusion decision-making: Construct a feature fusion mechanism based on Bayesian probability to distinguish real signals from false alarm signals.

[0179] Multi-modal feature fusion decision-making:

[0180] (1) Feature likelihood function modeling;

[0181] Define the hypothesis space as:

[0182] : The observed signal comes from a real unmanned aerial vehicle;

[0183] : The observed signal is a multipath false alarm;

[0184] Based on the conditional independence assumption of the signal spectral characteristics , modulation characteristics and DOA estimation quantity, establish a joint feature likelihood function model:

[0185] ;

[0186] where is categories, is the estimated parameter, and are the features respectively and features under the assumption of the likelihood;

[0187] For parameter estimation in the distribution under, reflecting the uncertainty or prior knowledge of the parameter.

[0188] Indicates that under the assumption the features , and the estimated parameter of the joint probability distribution.

[0189] Signal spectrum feature likelihood function:

[0190] ;

[0191] where represents the mean of the signal spectrum feature under the assumption , which reflects the average characteristics of the signal spectrum feature when the assumption holds, is the inverse matrix of the covariance matrix of the signal spectrum feature under the assumption , calculates half of the square of the "Mahalanobis distance" between the signal spectrum feature and the mean , measures the distance from a point to a distribution, represents a proportional relationship.

[0192] This formula is based on the probability density function form of the multivariate normal distribution. In signal processing, it is often assumed that the signal spectrum feature follows a normal distribution, and the likelihood function is used to measure the likelihood of observing the signal spectrum feature when the assumption holds.

[0193] Signal modulation feature likelihood function:

[0194] ;

[0195] where is the product symbol, indicating a product operation on the terms from to , meaning that the feature consists of multiple sub-features It consists of multiplying each sub - feature's contribution to the likelihood probability comprehensively.

[0196] is the output of the Support Vector Machine (SVM). SVM is a machine - learning classification algorithm that takes the sub - features as input and processes them under the hypothesis to output a numerical value related to the sub - feature under a specific hypothesis, which can be understood as a measure of the sub - feature belonging to a certain class (corresponding to a specific modulation method) under the hypothesis .

[0197] is the weight coefficient, which is used to measure the relative importance of each sub - feature in the calculation of the overall likelihood probability. Different values can be obtained according to prior knowledge or training to highlight the role of some more discriminative features.

[0198] Likelihood function for DOA spatial estimation:

[0199] ;

[0200] where is the normalization constant of the Gaussian distribution, whose role is to ensure that the integral of the entire probability density function over the domain is equal to 1, making the probability have reasonable statistical significance.

[0201] is the predicted value of the direction - of - arrival parameter, which is the direction of arrival predicted by the model under the hypothesis ;

[0202] is the variance of the estimated parameter , which measures the degree of dispersion of the estimated value. The smaller the variance, the more concentrated the estimated parameter is around ; the larger the variance, the more dispersed the distribution of the estimated parameter. It reflects the impact of the deviation degree of the estimated parameter from on the probability. The smaller the deviation, the closer the value is to 1 and the larger the probability density.

[0203] is the probability density function of the Gaussian distribution, which reflects the impact of the degree of deviation of the data from the mean on the probability density. The greater the degree of deviation, the smaller the value of the estimated parameter and the corresponding probability density.

[0204] This formula is used to describe that under the hypothesis , the estimated parameter The likelihood function measures the probability of observing the current estimate under a given model and parameter assumptions. The most likely direction of arrival estimate can be found by maximizing the likelihood function.

[0205] (2) Adaptive decision making;

[0206] The dynamic threshold function is constructed as:

[0207] ;

[0208] in: is the dynamic threshold, It is an initial, fixed parameter, usually a constant, that acts as a baseline scaling factor, determining the approximate magnitude of the entire function range.

[0209] : Time-varying noise power estimation, expressed at time The power of the noise;

[0210] : Signal power moving average, representing the time The power sliding average of the signal;

[0211] It is the ratio of time-varying noise power to signal power. When the time-varying noise power is larger than the signal power, it means that the signal is seriously interfered by noise. At this time, the threshold needs to be increased to reduce the impact of noise on subsequent processing.

[0212] When the signal power is larger than the time-varying noise power, it indicates that the signal quality is good and the threshold is appropriately lowered so that the signal changes can be detected more sensitively.

[0213] : The norm of the ship platform attitude change rate, which measures The rate of change in space (if spatial dimensions are involved) or other relevant dimensions, is the gradient, is the real-time posture matrix;

[0214] Denominator along with increases with the increase of The overall decrease will be When the rate of change is large, the signal can be detected more sensitively in areas of drastic change.

[0215] when When larger, it indicates The changes were drastic. The value of is small, the dynamic threshold It will decrease accordingly, so that signal changes can be detected more sensitively in areas with drastic changes, avoiding missing important information.

[0216] When is small, it indicates that changes gently. At this time, has a large value, and the dynamic threshold will increase accordingly to reduce false detections in stable areas.

[0217] The final decision rule is defined as:

[0218] ;

[0219] is the probability that hypothesis , and the estimated parameter hold under the given features . is the probability that hypothesis , and the estimated parameter hold under the given features .

[0220] is a conceptual expression of the relative status of the two hypotheses, helping to measure the preference for supporting a certain hypothesis;

[0221] is the decision operation. When the value of is greater than the threshold , it is determined that the condition holds, that is, the observed signal comes from a real drone; if the value of is less than the threshold , it is determined that the condition holds, that is, the observed signal is a multipath false alarm.

[0222] can be a function related to the time , which affects the trade-off of the probability ratio of the two hypotheses.

[0223] By online estimating the environmental parameters of the ship platform, the discriminant dynamic threshold is dynamically adjusted to effectively control the false alarm rate caused by the signal multipath transmission effect while ensuring the detection probability.

[0224] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope recorded by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims described above.

Claims

1. A detection signal anti-multipath interference method based on multimodal feature fusion, characterized in that: The steps include: Step 1: Extracting drone signal fingerprint features: Analyzing the physical layer parameters of the signal; Step 2: UAV signal airspace DOA estimation and error analysis: Use array antenna to obtain signal spatial spectrum information; Step 3: Physical modeling of the ship environment: Combined with the three-dimensional model of the ship, a multipath reflection path prediction model is constructed; Step 4: Dynamic DOA error correction: Suppress the DOA estimation deviation caused by multipath through iterative optimization of signal characteristics and physical models; Step 5: Multimodal feature fusion decision: Construct a feature fusion mechanism based on Bayesian probability to distinguish real signals from false alarm signals.

2. The detection signal anti-multipath interference method based on multimodal feature fusion according to claim 1 is characterized in that: In the step 1, the physical layer parameter analysis of the signal includes: center frequency and bandwidth analysis, frequency hopping sequence extraction and modulation mode identification.

3. The detection signal anti-multipath interference method based on multimodal feature fusion according to claim 2 is characterized in that: The data of the center frequency and the bandwidth are extracted by short-time Fourier transform, and the frequency hopping period and the frequency hopping sequence of the signal are detected, and then the modulation mode of the drone signal is identified based on the high-order cumulant.

4. The detection signal anti-multipath interference method based on multimodal feature fusion according to claim 1 is characterized in that: In step 2, the array antenna is composed of multiple antenna units. When the signal is incident on the array antenna from different directions, different phase delays will be generated due to the different distances between each antenna unit and the signal source. By measuring the phase delay and signal amplitude, an array signal model is established to evaluate the signal source information.

5. The detection signal anti-multipath interference method based on multimodal feature fusion according to claim 4 is characterized in that: The signal spatial spectrum information includes: constructing an array signal model and multi-signal classification spatial spectrum estimation.

6. The detection signal anti-multipath interference method based on multimodal feature fusion according to claim 1 is characterized in that: In step 3, the wireless signal is regarded as a light ray, and the propagation path of the wireless signal in the environment is analyzed according to the principle of geometric optics.

7. The detection signal anti-multipath interference method based on multimodal feature fusion according to claim 6 is characterized in that: In the step 2, the multipath reflection path prediction model uses three-dimensional coordinate transformation and multipath reflection path prediction.

8. The detection signal anti-multipath interference method based on multimodal feature fusion according to claim 1 is characterized in that: In step 4, the DOA estimation deviation caused by multipath is suppressed by joint estimation of multipath parameters and dynamic compensation of ship platform attitude disturbance.

9. The detection signal anti-multipath interference method based on multimodal feature fusion according to claim 1 is characterized in that: In step 5, the dynamic DOA error correction adopts characteristic likelihood function modeling and adaptive decision making.

10. A detection signal anti-multipath interference system based on multimodal feature fusion, used to execute the detection signal anti-multipath interference method based on multimodal feature fusion according to any one of claims 1 to 9, characterized in that: include: Extraction module, used for extracting features of drone signals; Analysis module, used for airspace DOA estimation and error analysis of drone signals; Physical modeling module, used for physical modeling of ship environment; Correction module, used for dynamic DOA error correction; Decision module, used for modal feature fusion decision; The extraction module is connected to the analysis module, the analysis module is connected to the physical modeling module, the physical modeling module is connected to the correction module, and the correction module is connected to the decision module.

Images