QD laser warning multi-target detection and positioning method based on spatial spectrum estimation

Abstract

This invention discloses a method for multi-target detection and localization of QD laser alarms based on spatial spectrum estimation. Specifically, it involves: establishing a precise mapping model between the output signal of a four-quadrant detector (QD) and the laser incident angle, including the signal superposition effect when multiple light sources illuminate simultaneously; based on the mapping model, employing a spatial spectrum estimation algorithm tailored to the limited spatial sampling characteristics of the QD detector to separate and estimate multiple simultaneously existing or temporally overlapping laser incident directions; and combining the pulse characteristics of the laser signal, fusing the temporal information with the spatial spectrum estimation results to achieve multi-target detection and localization. This invention overcomes the application limitations of traditional single-target localization models in multi-light source scenarios, providing an accurate input foundation for spatial spectrum estimation; and solves the technical bottleneck of traditional spatial spectrum estimation relying on multiple array elements, achieving direction separation and localization of simultaneously or temporally overlapping laser sources.

Description

Technical Field

[0001] This invention belongs to the field of laser detection and alarm technology, and particularly relates to a method for multi-target detection and localization of laser alarms using a four-quadrant detector based on spatial spectrum estimation. Background Technology

[0002] In modern optoelectronic countermeasures and security monitoring, rapid and accurate detection and localization of laser signals are crucial. Quadrant detectors (QDs), as a commonly used optoelectronic detection device, have advantages such as small size, low cost, and fast response speed, and are widely used in laser alarm systems.

[0003] Traditional four-quadrant detector localization technology is mainly based on single-target scenes, using single-target localization models such as the ratio method to determine the laser incident angle. However, in practical applications, there are often multi-target scenes where multiple laser sources illuminate simultaneously. In this case, the traditional single-target localization model cannot distinguish between signals superimposed from multiple light sources, resulting in significantly increased localization errors or even failure. The inability to differentiate between different laser sources greatly limits the application of four-quadrant detectors in multi-target scenes.

[0004] On the other hand, while existing spatial spectrum estimation methods can achieve multi-target detection and localization, these methods usually require a large number of array units to ensure detection performance. This makes the devices based on these methods large in size and expensive, making it difficult to meet the application requirements of miniaturized alarm devices.

[0005] Therefore, how to leverage the miniaturization and low cost advantages of four-quadrant detectors, combined with spatial spectrum estimation technology, to achieve effective detection and localization of multi-target laser signals has become an urgent problem to be solved in the field of laser detection and alarm technology. Summary of the Invention

[0006] To overcome the above-mentioned deficiencies of the prior art, the present invention provides a QD laser alarm multi-target detection and localization method based on spatial spectrum estimation.

[0007] The present invention provides a QD laser warning multi-target detection and localization method based on spatial spectrum estimation, comprising the following steps:

[0008] Step 1: Establish an accurate mapping model between the output signal of the four-quadrant detector (QD) and the incident angle of the laser. The model includes the signal superposition effect when multiple light sources illuminate simultaneously.

[0009] Step 2: Spatial spectrum estimation based on the mapping model: A spatial spectrum estimation algorithm adapted to the limited spatial sampling characteristics of the four-quadrant detector is adopted to separate and estimate multiple laser incident directions that exist simultaneously or overlap in time.

[0010] Step 3: Spatiotemporal joint processing: Combining the pulse characteristics of the laser signal, the temporal domain information and spatial spectrum estimation results are fused to optimize the performance of multi-target detection and localization.

[0011] Step 4: Multi-target energy decoupling and parameter estimation: Based on the mapping model in Step 1 and the target directions obtained in Step 2, decouple and estimate the intensity of each laser source.

[0012] Step 5: Enhance noise immunity, interference immunity, and robustness.

[0013] Step 6: Hardware architecture optimization: Optimize system performance through hardware design and algorithm collaboration.

[0014] Step 7: Output the angles and intensities of multiple targets to achieve multi-target detection and localization.

[0015] Furthermore, in step 1, the mapping model is a mathematical model of the voltage or current difference between the four channels of the four-quadrant detector and the laser incident angle. The model explicitly describes the nonlinear superposition physical characteristics of multiple light source signals in the four quadrants of QD.

[0016] Furthermore, in step 2, the spatial spectrum estimation algorithm is an improved MUSIC algorithm, ESPRIT algorithm, Capon beamformer or a variant thereof. The algorithm estimates the laser incident direction by constructing a covariance matrix and spatial spectrum function adapted to the four-quadrant detector.

[0017] Furthermore, the time-domain information in step 3 includes pulse arrival time (TOA), pulse width (PulseWidth), and repetition frequency; the fusion process includes:

[0018] The pulse arrival time difference is used to help distinguish different targets in the spatial spectrum estimation results.

[0019] By performing coherent or incoherent accumulation of continuous sampling points within a time window, the signal-to-noise ratio and resolution of spatial spectrum estimation can be improved.

[0020] Pulse width characteristics can be used to assist in target identification or interference removal.

[0021] Furthermore, the decoupling process in step 4 is achieved by constructing a set of equations about the intensity of each laser source and solving them using the least squares method or iterative method.

[0022] Furthermore, the hardware architecture optimization steps in step 6 are as follows:

[0023] A microlens array, diffuser, or grating structure is placed in front of the four-quadrant detector to modulate the distribution of the incident light spot on the QD.

[0024] Configure high-speed signal acquisition and processing circuits that meet the real-time requirements of the algorithm.

[0025] Furthermore, the mapping model is based on the single-target localization model, and a nonlinear superposition model is constructed by introducing a multi-source energy superposition coefficient.

[0026] Furthermore, the spatial spectrum estimation algorithm in step 2 incorporates robust design features, including:

[0027] Robust covariance matrix estimation methods using diagonal loading or M-estimation are employed.

[0028] Design adaptive noise suppression or signal preprocessing modules.

[0029] Prior information or machine learning methods can be used to correct or filter the estimation results.

[0030] Furthermore, the peak search of the spatial spectral function is achieved by combining fast Fourier transform with local search.

[0031] The beneficial technical effects of this invention compared to the prior art are as follows:

[0032] This invention overcomes the limitations of traditional single-target localization models in multi-source scenarios by establishing a precise mapping model between the output signal of a four-quadrant detector (QD) and the incident angles of multiple laser sources, providing an accurate input basis for spatial spectrum estimation. By designing a spatial spectrum estimation algorithm adapted to the characteristics of QD finite elements, it innovatively applies array signal processing theory to a miniature optical array with only four detection units, solving the technical bottleneck of traditional spatial spectrum estimation relying on multiple array elements and achieving directional separation and localization of simultaneously or temporally overlapping laser sources. Combined with spatiotemporal processing of laser pulses, the fusion of time-domain information (TOA, pulse width) and spatial spectrum results significantly improves resolution and detection probability in complex scenarios such as low signal-to-noise ratio and close target angles. Through a multi-target energy decoupling algorithm, the intensity of each light source is separated from the mixed signal, providing key parameters for threat type judgment (such as rangefinders and laser designators). Embedded robust mechanisms for noise and interference resistance ensure stable operation of the algorithm in real-world complex optoelectronic environments. Furthermore, hardware architecture optimization (such as microlens array modulation of spot distribution and high-speed processing circuitry) further enhances the system's spatial information utilization and real-time performance.

[0033] Its core advantages are: First, it is the first time that multi-target laser alarm has been achieved on the miniaturized, low-cost hardware platform of QD, solving the fundamental defect of multi-target failure in traditional QD systems; second, through spatiotemporal fusion and robust design, it balances positioning accuracy (angle resolution is better than the traditional ratio method) and environmental adaptability; third, it not only achieves directional positioning, but also outputs energy information of each target, providing a more comprehensive basis for threat assessment; fourth, through software and hardware co-optimization, it maintains the miniaturization and low cost characteristics of the device while ensuring performance, making it easy to promote and apply in various laser alarm scenarios. Attached Figure Description

[0034] Figure 1 This is a flowchart of the QD laser alarm multi-target detection and localization method based on spatial spectrum estimation of the present invention.

[0035] Figure 2 The three-dimensional spatial spectrum is obtained by using an improved MUSIC algorithm when laser sources at two angles ([3.5, -2.8], [-4.2, 4.5]) are incident on a four-quadrant detector. Detailed Implementation

[0036] The present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0037] The flowchart of a QD laser alarm multi-target detection and localization method based on spatial spectrum estimation of the present invention is as follows: Figure 1 As shown, it includes the following steps:

[0038] Step 1: Establish an accurate mapping model between the output signal of the four-quadrant detector (QD) and the incident angle of the laser. The model includes the signal superposition effect when multiple light sources illuminate simultaneously.

[0039] This step is used to establish the mathematical relationship between the output signal of the four-quadrant detector (QD) and the incident angle of the laser. The model must include the signal superposition effect when multiple light sources illuminate simultaneously. The specific process is as follows:

[0040] 1.1 Construction of a single-objective basic model.

[0041] Taking the four quadrants Q1, Q2, Q3, and Q4 of a four-quadrant detector as the research object, a turntable is used to control the laser source to continuously scan within a preset horizontal and vertical angle range, and the output voltage (or current) of each quadrant is collected. Based on the single-target localization principle, a curve fitting method is used to establish the mapping relationship between the output signal and the incident angle under single-target conditions. For example, for quadrant Q1, the output signal can be expressed as the product of the response coefficient of that quadrant and the angle function. Similarly, the single-target models of Q2, Q3, and Q4 can be obtained, where the angle function is used to describe the correlation between the laser incident angle and the quadrant energy distribution.

[0042] The specific implementation method is as follows:

[0043] Experimental setup: A Si-PIN type four-quadrant detector (photosensitive surface size 5mm×5mm) was selected, a continuous laser with a wavelength of 1064nm was used as the laser source, a two-dimensional motorized turntable (angular resolution 0.01°) was selected as the turntable, and a data acquisition card with 16-bit resolution and a sampling rate of 100kHz was used.

[0044] Data acquisition: The turntable is controlled to make the laser source scan in 0.1° steps within the horizontal angle [-5°, 5°] and the pitch angle [-5°, 5°]. 100 data points are collected at each position and the average value is taken to obtain the four-quadrant output voltage. .

[0045] Model fitting: A quadratic polynomial model was used. The parameters were fitted using MATLAB's CurveFitting Toolbox to obtain the coefficients. Similarly, fit the models for Q2, Q3, and Q4.

[0046] 1.2 Extension of the multi-objective superposition model.

[0047] Based on the single-objective model, a multi-source energy superposition coefficient is introduced. Assuming M laser sources are incident simultaneously, the energy contribution of the m-th source in quadrant Q1 is the product of its intensity, its angle function, and its quadrant response coefficient. The total output of Q1 is then the sum of the energy contributions of all sources in that quadrant plus the noise component. Similarly, multi-objective output models for Q2, Q3, and Q4 are established, forming a matrix mapping relationship. The matrix elements contain the product of the angle function and quadrant response coefficient of each source, thus describing the nonlinear superposition characteristics of multi-source signals.

[0048] The specific implementation method is as follows:

[0049] Superposition model: For two objectives (M=2), the Q1 quadrant output is represented as:

[0050] ;

[0051] in The response coefficient is Q1. It is an angle function. Light source intensity, It is noise.

[0052] Matrix form: Represent the four-quadrant output as a matrix equation:

[0053] ;

[0054] in It is a column vector of the mapping matrix, and its elements are determined by the angle function and the response coefficient.

[0055] Parameter calibration: Two laser sources with known angles are set by controlling the turntable, and the output in four quadrants is collected. The response coefficients are solved using the least squares method. and angle function parameters.

[0056] 1.3 Model Validation and Optimization.

[0057] The Monte Carlo simulation method was used to randomly generate multi-target incident scenes within a preset angle range, and the error between the model calculation results and the theoretical values ​​was compared. At the same time, the model was verified by experimental measurement: multiple laser sources with known angles were set on a turntable, the model estimation results were compared with the actual angles, and the model parameters (such as response coefficients and angle function coefficients) were adjusted until the error converged, ensuring the accuracy of the model in multi-target scenes.

[0058] The specific implementation method is as follows:

[0059] Simulation verification: Using MATLAB's Monte Carlo Simulation Toolbox, 1000 sets of dual-target angle combinations were randomly generated within the range of θ∈[-5°,5°] and φ∈[-5°,5°]. The mean square error (MSE) between the theoretical output and the model estimate was calculated, and the requirement was that MSE<0.01.

[0060] Experimental verification: Two laser sources with known angles are set on the turntable (e.g., ; The four quadrant outputs are collected, substituted into the model to calculate the angle, and compared with the actual angle. The error should be <0.1°.

[0061] Model optimization: If the error exceeds the threshold, adjust the form of the angle function (e.g., add higher-order terms) or introduce cross-coupling coefficients, and refit the parameters until the error converges.

[0062] Step 2: Spatial spectrum estimation based on the mapping model: A spatial spectrum estimation algorithm adapted to the limited spatial sampling characteristics of the four-quadrant detector is adopted to separate and estimate multiple laser incident directions that exist simultaneously or overlap in time.

[0063] This step employs a spatial spectrum estimation algorithm adapted to the limited spatial sampling characteristics of a four-quadrant detector to achieve the separation and estimation of multiple laser incident directions. The specific process is as follows:

[0064] 2.1 Signal preprocessing and covariance matrix construction.

[0065] The four-quadrant output signal is preprocessed: DC drift and low-frequency noise are removed by filtering, and the signal is converted into a digital signal by AD conversion. Taking advantage of the four-quadrant detector having only four elements, the sample size is expanded by time series analysis. Multiple sets of signals are continuously acquired to construct a signal matrix (e.g., a 4×N matrix, where N is the number of sampling points). Based on this matrix, the covariance matrix of the signal is calculated to reflect the correlation and power information between signals.

[0066] The specific implementation method is as follows:

[0067] Hardware filtering: A second-order RC low-pass filter (cutoff frequency 10kHz) is used to remove high-frequency noise, and an operational amplifier (such as OPA2333) is used to form a voltage follower to isolate the signal source.

[0068] AD Conversion: The analog signal is converted into a digital signal using the ADS8320 chip (16-bit resolution, 200kHz sampling rate), and communicates with the FPGA via the SPI interface.

[0069] Data caching: Implement a FIFO cache in an FPGA (such as Xilinx Spartan-6) to continuously store 1024 sets of four-quadrant sampled data (4×1024 matrix).

[0070] Covariance calculation: In FPGA, the covariance matrix is ​​calculated using parallel multipliers and accumulators. The formula is as follows: , where X is a 4×N data matrix.

[0071] 2.2 Definition of spatial spectral function and eigenvalue decomposition.

[0072] A spatial spectrum estimation algorithm optimized for four-quadrant detectors (such as the improved MUSIC algorithm, ESPRIT algorithm, etc.) is employed: eigenvalue decomposition is performed on the covariance matrix to obtain the eigenvalues. and the corresponding feature vector; through the feature value difference threshold (such as The signal subspace dimension (i.e., the number of targets M) is determined by the preset value, and the signal subspace and noise subspace are divided. Based on the mapping model, the steering vector (which includes the mapping relationship between angle and four-quadrant response) is defined, and then the spatial spectrum function is constructed. This function is used to characterize the signal energy distribution under different incident angles.

[0073] The specific implementation method is as follows:

[0074] Eigenvalue decomposition: Implement the Jacobi iterative algorithm in a DSP (such as TITMS320C6748) to perform eigenvalue decomposition on the 4×4 covariance matrix. After 10 iterations, the eigenvalue error is < .

[0075] Subspace partitioning: Calculating the ratio of eigenvalues If the ratio is greater than 3, then the number of targets M is determined to be 2, and the two feature vectors with the smallest ratio are used to form the noise subspace. .

[0076] Guide vector: Construct the guide vector based on the mapping model in step 1. ,in Let be the angle function in the i-th quadrant.

[0077] Spectral function calculation: Defining the spatial spectral function The calculation of this function is implemented in the DSP.

[0078] 2.3 Peak search and direction estimation.

[0079] Within a preset laser incident angle range, a dual-resolution search strategy is employed: first, a coarse search is performed using Fast Fourier Transform to identify potential spectral peak regions; then, a fine search is conducted within these regions, with the step size dynamically adjusted according to the angular resolution requirements. Simultaneously, the symmetry constraint of the four-quadrant response (e.g., symmetrical energy distribution of the laser spot in the symmetrical quadrant) is utilized to eliminate false spectral peaks and resolve the angular ambiguity problem. The peak positions of the spectral function are extracted, with each peak corresponding to the incident direction of a laser source, thereby achieving separation and estimation of multiple target directions. (Appendix) Figure 2 The three-dimensional spatial spectrum calculated by the method described in step 2.3 of this invention is shown in a dual-target incident scenario.

[0080] The specific implementation method is as follows:

[0081] Coarse Search: Implement the 2D-FFT algorithm in a DSP, sampling the ranges θ∈[-5°, 5°] and φ∈[-5°, 5°] (step size 0.5°), and calculate the spectral function value for each sampling point. (See attached...) Figure 2 The approximate areas of these two energy bulges can be quickly identified across the entire surface shown.

[0082] Fine-grained search: Centered on the peak angle obtained from the coarse search, a second search is performed within a range of ±0.5° with a step size of 0.05°. The CORDIC algorithm is used to calculate the trigonometric function values. This process is described in the appendix. Figure 2 This corresponds to a fine scan of the two spectral peak regions, ultimately locking the peak position to the aforementioned high-precision coordinates.

[0083] Peak selection: Calculate the half-width at half-maximum (FWHM) of the spectral peaks, remove spurious peaks with FWHM > 1°, and retain the peaks that meet the criteria as the target angle estimation results. (Appendix) Figure 2 The two spectral peaks in the middle are sharp in shape and have a clean substrate, with an FWHM much less than 1°, which matches the characteristics of the real target. Therefore, they were successfully retained as valid estimation results. False broad peaks caused by noise or interference will be removed.

[0084] Step 3: Spatiotemporal joint processing.

[0085] By combining the pulse characteristics of laser signals and fusing time-domain information with spatial spectrum estimation results, the performance of multi-target detection and localization is optimized. The specific process is as follows:

[0086] 3.1 Extraction of time domain information.

[0087] The characteristic parameters of the laser pulse are identified by the pulse detection circuit, including pulse arrival time (TOA), pulse width, repetition frequency, etc. The arrival time is determined by detecting the leading edge of the pulse, and the pulse width is the duration for which the pulse exceeds a preset threshold.

[0088] The specific implementation method is as follows:

[0089] Pulse detection circuit: A high-speed comparator (such as LM311) is used to compare the four-quadrant output signal with the threshold voltage (3 times the root mean square of the noise) and output a square wave signal.

[0090] TOA measurement: A 100MHz clock counter is used. The counting is triggered on the rising edge of the comparator output. The counter value is converted into time, which is the TOA. The measurement accuracy is 10ns.

[0091] Pulse width calculation: Continuously count during the high level output of the comparator. The duration of the high level is the pulse width. The pulse width calculation is implemented through FPGA logic.

[0092] Repetition frequency estimation: Record the TOA of 10 consecutive pulses, calculate the average of the differences between adjacent TOA, and take the reciprocal to obtain the repetition frequency.

[0093] 3.2 Signal accumulation within the time window.

[0094] The time window width is set according to the repetition frequency of the laser pulse to ensure that each time window contains the complete laser pulse. Within the time window, the continuously acquired signals are coherently or incoherently accumulated to obtain the accumulated covariance matrix, thereby improving the signal-to-noise ratio and resolution of the spatial spectrum estimation.

[0095] The specific implementation method is as follows:

[0096] Time window generation: Implement a programmable timer in the FPGA and dynamically set the time window width (e.g., 1 / 1.2 times the repetition frequency) according to the repetition frequency.

[0097] Coherent accumulation: The covariance matrix is ​​accumulated within a time window, as shown in the formula. , where R i Let N be the covariance matrix within the i-th time window, and N be the number of accumulations (e.g., N=32).

[0098] Incoherent accumulation: Accumulating the spectral function values ​​within a time window, as shown in the formula. This increases the probability of detecting weak signals.

[0099] 3.3 Spatiotemporal information fusion.

[0100] Using pulse arrival time difference to help distinguish targets: If two spatial spectral peaks are close in angle, but the arrival time difference of the corresponding pulses exceeds a preset threshold (such as 1.5 times the pulse width), they are determined to be different targets; combined with pulse width feature screening results: the measured pulse width is compared with the preset threat pulse width range, and interference spectral peaks that exceed the range are eliminated.

[0101] The specific implementation method is as follows:

[0102] TOA difference decision: For two peaks obtained from spatial spectrum estimation, compare the TOA difference of the corresponding pulses. If the difference is greater than 1.5 times the pulse width, they are determined to be different targets; otherwise, they are merged into one target.

[0103] Pulse width screening: Preset the pulse width range of the threat laser (e.g., 10ns-100ns), calculate the pulse width of the pulse corresponding to each spectral peak, and remove spectral peaks that are outside the range.

[0104] Fusion logic: The state machine is implemented in the FPGA. It performs weighted decision based on the TOA difference, pulse width and spectral peak intensity, and outputs the final target angle and intensity estimation results.

[0105] Step 4: Multi-objective energy decoupling and parameter estimation.

[0106] Based on the mapping model in step 1 and the target direction obtained in step 2, the intensity of each laser source is decoupled and estimated. The specific process is as follows:

[0107] 4.1 Modeling the inverse problem.

[0108] Given the incident angles of each target, based on the multi-target mapping model established in step 1, a set of equations is constructed regarding the intensity of each laser source. The left side of the equations represents the measured output signals in the four quadrants, and the right side represents the superposition of the product of the intensity of each light source and the corresponding mapping matrix element.

[0109] The specific implementation method is as follows:

[0110] Equation system construction: for two objectives ( )and( Based on the mapping model, establish a system of equations:

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] Written in matrix form as , where V is the four-quadrant output vector, K is a 4×2 mapping matrix, and A is the intensity vector.

[0116] 4.2 Strength solution.

[0117] The above equation system can be solved using either the least squares method or an iterative method: By constructing a weighting matrix (such as the inverse of the noise covariance matrix), the overdetermined equation system can be solved; when the equation system has a high degree of nonlinearity, an iterative solution method is used: the initial value is set as the least squares solution, and the iterative formula is as follows: (α is the step size), and the optimal solution is gradually approximated by gradient descent. The iteration terminates when the relative error between two adjacent iterations is less than a preset threshold, or when the maximum number of iterations is reached, and finally the intensity estimate of each laser source is obtained, thus achieving decoupling of the energy of multiple targets.

[0118] The specific implementation method is as follows:

[0119] Least Squares Method: Implementing the Cholesky decomposition algorithm in a DSP to solve the weighted least squares problem. , where W is the noise covariance matrix (estimated from data during periods without signal).

[0120] Iterative solution: When the system of equations is highly nonlinear, the gradient descent method is used for iterative solution. The iterative formula is as follows: ,in Let be the cost function, α be the step size (e.g., α = 0.01), and let the gradient ∇J(A) be calculated by numerical differentiation.

[0121] Termination condition: Set the iteration termination condition to... (like (or reach the maximum number of iterations (e.g., 100 times)).

[0122] Step 5: Enhance noise immunity, interference immunity, and robustness.

[0123] The following mechanisms can be used to improve the stability of the algorithm in complex environments:

[0124] 5.1 Robust covariance matrix estimation.

[0125] The covariance matrix is ​​subjected to diagonal loading, i.e., a preset loading amount is added. (I is the identity matrix), where λ is dynamically adjusted according to the noise level:

[0126] When the signal-to-noise ratio is detected to be lower than the preset threshold, the loading amount is automatically increased (e.g., adjusted from 0.1×max(diag(R)) to 0.2×max(diag(R))) to suppress eigenvalue fluctuations caused by noise and enhance the stability of eigenvalue decomposition.

[0127] The specific implementation method is as follows:

[0128] Noise estimation: Collect 1000 samples during periods without signal, calculate the variance of the four quadrant outputs, and take the average value as the noise variance σ².

[0129] Load calculation: The load is dynamically calculated based on the noise variance and signal power. , where tr(R) is the trace of the covariance matrix.

[0130] Diagonal loading: Implement matrix addition in the FPGA by loading the calculated λ onto the diagonal of the covariance matrix, i.e. .

[0131] 5.2 Adaptive noise suppression.

[0132] The four-quadrant output signal is preprocessed using methods such as wavelet thresholding: the signal is decomposed into wavelets, and high-frequency coefficients are thresholded. The threshold is updated in real time according to the noise level (e.g., the noise standard deviation σ is estimated based on the statistical characteristics of the no-signal period, and the threshold is set to a certain value). (where N is the number of sampling points), to reduce the impact of noise on subsequent processing.

[0133] The specific implementation method is as follows:

[0134] Wavelet decomposition: Implement a 3-level discrete wavelet transform (DWT) in a DSP, using the db4 wavelet basis to decompose the four-quadrant output signal.

[0135] Threshold calculation: The threshold is calculated based on the noise standard deviation σ and the number of sampling points N. , where σ is estimated from data during periods without signal.

[0136] Thresholding: Soft thresholding is applied to the high-frequency coefficients after wavelet decomposition. The formula is as follows:

[0137]

[0138] Where c is the original coefficient and c' is the processed coefficient.

[0139] Signal reconstruction: Implement inverse discrete wavelet transform (IDWT) in DSP to reconstruct the processed coefficients into a denoised signal.

[0140] 5.3 Machine learning-assisted screening.

[0141] A classification model (such as a support vector machine) is established based on training samples. The training samples include typical real targets (angle error within a preset range) and typical interferences (such as light spots concentrated in a single quadrant, sudden changes in background light, etc.). The spatial spectrum estimation results are screened based on features such as peak intensity, width, and symmetry, eliminating interference peaks and retaining valid targets.

[0142] The specific implementation method is as follows:

[0143] Training samples: Collect 500 sets of real target data (angle error <0.5°) and 500 sets of interference data (such as background light abrupt change, single quadrant strong light), and extract 10 features such as spectral peak intensity, half width at half maximum, and four quadrant energy ratio.

[0144] Model training: The SVM classifier was trained using Python's Scikit-learn library, employing the RBF kernel function and optimizing parameters through 5-fold cross-validation.

[0145] Model deployment: The trained SVM model is converted into C language code and classified and predicted in DSP. The probability of each spectral peak belonging to the real target is calculated, and spectral peaks with a probability <0.8 are removed.

[0146] Step 6: Hardware architecture optimization.

[0147] System performance is optimized through hardware design and algorithm collaboration.

[0148] 6.1 Optical front-end optimization.

[0149] By placing a microlens array, diffuser, or grating structure in front of the four-quadrant detector, the diversity of spatial information can be increased by modulating the distribution of the incident light spot on the detector (such as forming a majority light spot), thus helping to improve the resolution of spatial spectrum estimation.

[0150] The specific implementation method is as follows:

[0151] Microlens array design: A 4×4 microlens array is fabricated using photolithography. Each lens has a diameter of 0.5mm and a focal length of 3mm. PMMA is used as the lens material. The lenses are aligned with the QD (error <0.1mm) using precision positioning pins.

[0152] Diffuser selection: Select an optical diffuser with a scattering angle of 20° and place it 5mm in front of QD. Adjust the thickness of the diffuser (e.g., 0.5mm) through experimental testing to make the light spot evenly cover the four quadrants.

[0153] Grating structure: A one-dimensional blazed grating (period 10μm, blaze angle 30°) is designed to split the incident laser into multiple sub-spots through optical diffraction, thereby increasing the number of spatial sampling points.

[0154] 6.2 High-speed processing circuit configuration.

[0155] High-speed signal acquisition and processing circuits adapted to the real-time requirements of the algorithm are adopted: the sampling rate of the AD converter is matched with the characteristics of the laser pulse (e.g., ≥10 times the pulse bandwidth), and the processing unit (e.g., FPGA) adopts a parallel architecture to realize the rapid operation of modules such as covariance matrix calculation and spectral peak search, ensuring the real-time response of the system.

[0156] The specific implementation method is as follows:

[0157] AD conversion module: The ADS5463 chip (14-bit resolution, 250MSPS sampling rate) is selected. It acquires four-quadrant signals through differential input interface, and the sampled data is transmitted to FPGA through JESD204B protocol.

[0158] FPGA selection: Xilinx Kintex UltraScale+KU060 was selected. Internal resource allocation: 40% for data reception, 30% for covariance matrix calculation, 20% for spectral peak search, and 10% for control logic.

[0159] Parallel processing architecture: Four covariance calculation engines are implemented in parallel in the FPGA, each engine processes 256 sets of data, and continuous data stream processing is achieved through pipeline architecture, with a processing latency of <5μs.

[0160] Data storage: DDR4 SDRAM (4GB capacity, 34.1GB / s bandwidth) is used to store intermediate results, and efficient data read and write is achieved through the MIG core. The burst length is set to 32.

[0161] Through the above steps, the four-quadrant detector can detect, locate, and estimate the parameters of multiple target laser sources. The technical solution is universal and scalable, and the relevant parameters can be adjusted according to the actual scenario, all of which fall within the protection scope of this invention.

Claims

1. A QD laser warning multi-target detection and localization method based on spatial spectrum estimation, characterized in that, Includes the following steps: Step 1: Establish an accurate mapping model between the output signal of the four-quadrant detector (QD) and the incident angle of the laser. The model includes the signal superposition effect when multiple light sources illuminate simultaneously. Step 2: Spatial spectrum estimation based on the mapping model: A spatial spectrum estimation algorithm adapted to the limited spatial sampling characteristics of the four-quadrant detector is adopted to separate and estimate multiple simultaneous or temporally overlapping laser incident directions; Step 3: Spatiotemporal joint processing: Combining the pulse characteristics of the laser signal, the temporal domain information and spatial spectrum estimation results are fused to optimize the performance of multi-target detection and localization; Step 4: Multi-target energy decoupling and parameter estimation: Based on the mapping model in Step 1 and the target directions obtained in Step 2, decouple and estimate the intensity of each laser source; Step 5: Enhance noise immunity, interference immunity, and robustness; Step 6: Hardware Architecture Optimization: Optimize system performance through collaborative hardware design and algorithms; Step 7: Output the angles and intensities of multiple targets to achieve multi-target detection and localization.

2. The QD laser warning multi-target detection and localization method based on spatial spectrum estimation according to claim 1, characterized in that, In step 1, the mapping model is a mathematical model between the voltage or current difference of the four channels of the four-quadrant detector and the laser incident angle. The model explicitly describes the nonlinear superposition physical characteristics of multiple light source signals in the four quadrants of the QD.

3. The QD laser warning multi-target detection and localization method based on spatial spectrum estimation according to claim 1, characterized in that, In step 2, the spatial spectrum estimation algorithm is an improved MUSIC algorithm, ESPRIT algorithm, Capon beamformer or a variant thereof. The algorithm estimates the laser incident direction by constructing a covariance matrix and spatial spectrum function adapted to the four-quadrant detector.

4. The QD laser warning multi-target detection and localization method based on spatial spectrum estimation according to claim 1, characterized in that, The time-domain information in step 3 includes pulse arrival time (TOA), pulse width (PulseWidth), and repetition frequency; the fusion process includes: The pulse arrival time difference is used to help distinguish different targets in the spatial spectrum estimation results; Coherent or incoherent accumulation of continuous sampling points within a time window improves the signal-to-noise ratio and resolution of spatial spectrum estimation. Pulse width characteristics can be used to assist in target identification or interference removal.

5. The QD laser warning multi-target detection and localization method based on spatial spectrum estimation according to claim 1, characterized in that, The decoupling process in step 4 is achieved by constructing a set of equations about the intensity of each laser source and solving them using the least squares method or iterative method.

6. The QD laser alarm multi-target detection and localization method based on spatial spectrum estimation according to claim 1, characterized in that, The hardware architecture optimization steps in step 6 are as follows: A microlens array, diffuser, or grating structure is placed in front of the four-quadrant detector to modulate the distribution of the incident light spot on the QD. Configure high-speed signal acquisition and processing circuits that meet the real-time requirements of the algorithm.

7. The QD laser warning multi-target detection and localization method based on spatial spectrum estimation according to claim 2, characterized in that, The mapping model is based on the single-target localization model, and a nonlinear superposition model is constructed by introducing a multi-source energy superposition coefficient.

8. The QD laser warning multi-target detection and localization method based on spatial spectrum estimation according to claim 3, characterized in that, The robustness design embedded in the spatial spectrum estimation algorithm in step 2 includes: Robust covariance matrix estimation methods using diagonal loading or M-estimation are employed. Design an adaptive noise suppression or signal preprocessing module; Prior information or machine learning methods can be used to correct or filter the estimation results.

9. The QD laser alarm multi-target detection and localization method based on spatial spectrum estimation according to claim 3, characterized in that, The peak search of the spatial spectral function is achieved by combining fast Fourier transform with local search.

Images