A laser acceleration watchmaking method

Abstract

The application discloses a laser accelerator watch manufacturing method, belongs to the technical field of optical fiber sensing and signal processing, and is used for constructing a laser wavelength current lookup table, and comprises the following steps: laser initialization, configuration of five-way current for two-dimensional nested loop scanning, measurement of the center wavelength and the side mode suppression ratio of an optical signal, construction of a center wavelength sequence and division of a continuous longitudinal mode area; construction of a theoretical sampling grid and a theoretical sampling point set, calculation of the actual sampling point number by using a Gaussian distribution rule, and construction of a sparse measurement training data set; construction of a K singular value decomposition model and training; input of a to-be-measured target wavelength list into the K singular value decomposition model after training, and obtaining of the wavelength current lookup table. The K singular value decomposition algorithm is integrated into the laser wavelength current lookup table construction process, time cost is compressed, measurement load is reduced, deployment flexibility is enhanced, and the construction efficiency of the laser wavelength current lookup table is greatly improved.

Description

Technical Field

[0001] This invention discloses a laser-accelerated tabulation method, belonging to the field of fiber optic sensing and signal processing technology. Background Technology

[0002] Fiber Bragg grating (FBG) sensing technology has been widely used in industrial monitoring, aerospace, and energy sectors due to its outstanding advantages such as resistance to electromagnetic interference, corrosion resistance, and ease of multiplexing. As the core component of a FBG sensing system, the performance of the tunable laser directly determines the measurement accuracy, response speed, and operational reliability of the entire system.

[0003] Currently, the construction of laser wavelength-current lookup tables mainly involves the following technical solutions: The full measurement method based on fine current scanning, proposed by Zheng Shengyu and Yang Yuanhong et al., constructs the lookup table using a full-range fine current scanning approach. The main drawbacks of this method are: firstly, extremely high time cost. Scanning a 35mA range with a 0.1mA step size requires measuring 122,500 points just for the combination of the currents in the left and right grating reflector regions. Adding the settling and measurement time for each point, a single scan can take tens of hours. Secondly, the sampling strategy is indiscriminate, using a uniform grid scan, which cannot be optimized for the physical characteristics of the laser. Oversampling occurs in regions with gentle wavelength changes, while undersampling may occur in regions with drastic changes. Thirdly, poor boundary stability; unstable operating points are easily acquired in the longitudinal mode boundary region, and these points are prone to wavelength jumps during subsequent use, affecting system reliability. The quasi-continuous tuning method based on path planning identifies the quasi-continuous tuning path of the laser and performs current scanning only on smooth paths. This method first identifies multiple quasi-continuous tuning regions through coarse scanning, then performs fine scanning within each region along a smooth path, and finally achieves fine-grained wavelength retrieval through spline interpolation. However, it still has significant shortcomings. First, the sampling efficiency still has room for improvement. Although unstable regions are avoided, uniform and dense sampling is still used within the smooth path, failing to fully utilize the continuity and predictability of wavelength changes. Second, it lacks an intelligent optimization mechanism. The sampling strategy relies on empirical path selection and fails to establish a mathematical optimization model to automatically balance sampling efficiency and reconstruction accuracy. Third, its adaptability is limited. For lasers of different models or with individual differences, path planning needs to be re-planned, lacking adaptive capability.

[0004] In summary, existing technical solutions have failed to effectively resolve the contradiction between "sampling efficiency" and "measurement accuracy" during the construction of laser wavelength-current lookup tables, especially in achieving intelligent optimization sampling strategies based on full consideration of the physical characteristics of the laser. Summary of the Invention

[0005] The purpose of this invention is to provide a laser accelerated table-making method to solve the problems in the prior art, such as the significant time consumption of full-range fine current scanning operation, the instability of wavelength switching affecting tuning accuracy, the relatively complex power calibration process, and the exponential growth trend in the construction time of the lookup table when higher precision requirements are put forward for wavelength intervals, which makes it difficult to adapt to the needs of high-efficiency application scenarios.

[0006] A laser-accelerated tabulation method, comprising:

[0007] S1. Initialize the laser by configuring three fixed-value currents, the left grating reflector current, and the right grating reflector current. The three fixed-value currents include the gain region current, the semiconductor optical amplifier current, and the phase region current. Then, perform a two-dimensional nested cyclic scan, including using the left grating reflector current and the right grating reflector current to construct a two-dimensional nested cyclic scan grid and generate a current matrix for each grid. Output the optical signal based on the current matrix and the three fixed-value currents, measure the center wavelength and side-mode suppression ratio of the optical signal, construct the center wavelength sequence according to the order of the two-dimensional nested cyclic scan, and calculate the longitudinal mode boundary and boundary current. Divide the current plane into continuous longitudinal mode regions according to the longitudinal mode boundary.

[0008] S2. Construct a theoretical sampling grid and a set of theoretical sampling points based on two-dimensional nested cyclic scanning and boundary current. Using the Gaussian distribution rule, calculate the number of actual sampling points for the theoretical sampling points closest to the center of the longitudinal mode region in each longitudinal mode region. Calculate the left and right grating reflection zone currents of the actual sampling points. Measure the wavelength and real-time power of the optical signal at the actual sampling points. Construct a sparse measurement training dataset based on the left and right grating reflection zone currents and wavelengths of the actual sampling points.

[0009] S3. Construct a K-singular value decomposition model using the K-singular value decomposition algorithm. Train the model using a sparse measurement training dataset to obtain the trained K-singular value decomposition model. The K-singular value decomposition model includes constructing input features and predicting target values, initializing an overcomplete dictionary, sparse coding, and dictionary updating. Input the list of target wavelengths to be measured into the trained K-singular value decomposition model, and use the gradient descent optimization algorithm to inversely adjust the target current until the preset wavelength accuracy index and sampling efficiency index are met. Output the final wavelength current lookup table.

[0010] S1 includes S1.1, initializing the laser and configuring three fixed-value currents, including setting the gain region current. Semiconductor optical amplifier current and phase region current It is a fixed value;

[0011] S1 includes S1.2, and the two-dimensional nested cyclic scanning includes the current in the left grating reflection region. and the current in the right grating reflection region The current direction is used to construct a two-dimensional nested cyclic scanning grid, which includes an outer loop and an inner loop nesting.

[0012] S1 includes S1.3, setting the outer loop, including setting... loop variable , , In order to be in The number of scan points in the direction; set according to the laser's current range. step size , The outer loop increments sequentially;

[0013] calculate real-time value :

[0014] ;

[0015] S1 includes S1.4, setting inner loop nesting, including setting... loop variable , , In order to be in The number of scan points in the direction; set according to the laser's current range. step size , Incrementing sequentially based on the inner nested loop;

[0016] calculate real-time value :

[0017] ;

[0018] S1 includes S1.5, in each loop node. Set the current matrix :

[0019] ;

[0020] according to , , and Output optical signal and measure the center wavelength of the optical signal. Main mode optical power and maximum side-mode optical power Calculate the side mode suppression ratio :

[0021] .

[0022] S1 includes, S1.6, scanning in the order of two-dimensional nested loops. Serialization yields the center wavelength sequence. , Let be the theoretical total number of sampling points for a two-dimensional nested cyclic scanning grid. For the index of the center wavelength, ,calculate difference :

[0023] ;

[0024] In the formula, It is the absolute value;

[0025] Set the jump threshold ,like ,Will Set as longitudinal boundary;

[0026] according to Divide the current plane into A continuous longitudinal modulus region ,set up For the index of the longitudinal region, The current plane is based on and In a two-dimensional parameter space defined by coordinate axes, the minimum current of the right grating reflection region in each longitudinal mode region is recorded. Maximum current in the right grating reflection region Minimum current in the left grating reflection zone Maximum current in the left grating reflection region Center wavelength and number of scan points .

[0027] S2 includes S2.1, generating the theoretical sampling mesh, which includes setting the target number of theoretical points for each longitudinal model region. , The range of current variation within the longitudinal mode region is calculated using boundary currents:

[0028] ;

[0029] ;

[0030] In the formula, for Range of directional changes for The range of changes in direction;

[0031] according to and Constructing the theoretical sampling grid:

[0032] ;

[0033] ;

[0034] In the formula, The rounding up symbol, for Number of grids in the direction, for Number of grid cells in the direction;

[0035] Determine the theoretical point set based on the theoretical sampling grid. , , For the theoretical sampling point index, for The Middle One theoretical sampling point, for Two-dimensional current coordinates, For the first Secondary theoretical sampling points , For the first Secondary theoretical sampling points ;

[0036] S2 includes S2.2, which calculates the standard deviation and current scan step size according to the matching principle. Standard deviation and Standard deviation:

[0037] ;

[0038] ;

[0039] In the formula, for standard deviation for standard deviation is the diffusion coefficient.

[0040] S2 includes, S2.3, and the actual sampling points follow the order of... A two-dimensional Gaussian distribution centered at the sampling amplification factor is set. and target sampling rate ,calculate The actual number of sampling points to be collected :

[0041] ;

[0042] In the formula, It is a rounding function;

[0043] Set the loop counter variable Perform cyclic sampling iterations. In each iteration, the actual sampling points are generated according to a two-dimensional Gaussian distribution rule:

[0044] ;

[0045] ;

[0046] In the formula, for No. actual sampling points , for No. actual sampling points , for No. Each actual sampling point relative to exist Random perturbation value in direction, for No. Each actual sampling point relative to exist Random perturbation value in direction, To conform to the distribution sign, It follows a Gaussian distribution;

[0047] S2 includes S2.4, for each actual sampling point, if , Located within the current longitudinal modulus region boundary, and with a distance greater than the safety threshold from the longitudinal modulus region boundary, according to , , , and Output No. The optical signal at each actual sampling point is collected by a wavelength meter to obtain the corresponding actual wavelength. and real-time power .

[0048] S2 includes S2.5, constructing a sparse measurement training dataset, including when Finish After the first actual sampling, All actual sampling points form a subset of training data. , ;

[0049] Set the effective power range according to the laser model. ,set up threshold ,like Not in or If the actual sampling point is determined to be a bad point, then the bad point is removed from... Delete; if In and Determine the actual sampling point as a good point and retain it. In the end, a subset of training data after removing bad pixels is obtained. , for No. actual sampling points ;

[0050] like The total number of sampling points is less than Re-examine Perform cyclic sampling iterations and set a threshold for the number of iterations to count the loop. If the actual number of loop iterations is greater than and The total number of actual sampling points is less than , recorded as The number of actual sampling points within the range of the centered two-dimensional Gaussian distribution is insufficient, so the corresponding points are deleted. ;

[0051] For all Summarize and construct a sparse measurement training dataset. .

[0052] S3 includes, S3.1, extraction middle , Based on input features, To predict the target value, the basic input features are expanded into a high-dimensional feature vector. , All actual sampling points Stacking to construct feature matrices , ,set up This is the index of the actual sampling point. The actual number of sampling points. For the first actual sampling points , for each actual sampling point As elements, construct the target value vector , , For the first actual sampling points ;

[0053] S3 includes S3.2, constructing a K-singular value decomposition model using the K-singular value decomposition algorithm. The K-singular value decomposition model includes a randomly initialized, overcomplete dictionary matrix. , , , For the set of real numbers, The number of atoms in the dictionary matrix. The dimension of the feature vector. for OK Let be the set of real matrices of columns. for The first in One atom, For the index of atoms, ;Will denoted as measurement sample , , for 3D real vector space;

[0054] S3 includes S3.3, which uses the fit method in the K singular value decomposition algorithm to start iterative optimization. The fit method includes sparse coding and dictionary update.

[0055] S3 includes S3.4, the sparse coding stage, which uses the orthogonal matching pursuit algorithm to... Perform iterations, calculating the current residual vector and... The absolute value of the inner product of each atom is used to select the atom with the largest absolute value of the inner product as the most relevant atom. After iteration, the result is obtained. The most relevant atom;

[0056] Fixed in each iteration ,right Solve for the corresponding sparse coefficient vector :

[0057] ;

[0058] ;

[0059] In the formula, The sparsity coefficient is . For the first The sparsity coefficients of each actual sampling point The number of non-zero elements. The sign is less than the order of magnitude;

[0060] S3 includes S3.5, the dictionary update phase, which fixes all... And form a sparse coefficient matrix:

[0061] ;

[0062] ;

[0063] In the formula, It is the L0 norm;

[0064] Construct a target function that minimizes the global reconstruction error:

[0065] ;

[0066] In the formula, It is the Frobenius norm;

[0067] according to Find all users The measurement samples constitute an index set. ,right For each measurement sample, calculate the removal The reconstructed residuals will The reconstructed residuals of all measured samples are combined into a residual matrix. ,right Perform singular value decomposition and take the largest left singular vector as the updated atom. Update using the product of the maximum right singular vector and the maximum singular value. ; Traverse and update all atoms to complete the process. A new round of updates.

[0068] S3 includes S3.6, which iteratively executes steps S3.4 and S3.5. When the objective function of minimizing the global reconstruction error is lower than a preset tolerance threshold, or when the number of model iterations reaches a set model iteration threshold, training is terminated, and the optimized dictionary is obtained. And the trained K-singular value decomposition model;

[0069] S3 includes, S3.7, obtaining the wavelength range of the laser under test based on a two-dimensional nested cyclic scan, with a generation interval of... Global target wavelength list ,set up For the index of the wavelength of the target to be measured, , The number of wavelengths of the target to be measured;

[0070] Will Input the trained K-singular value decomposition model, and use And gradient descent optimization algorithm, inversely adjust the first Current in the right grating reflection region corresponding to the wavelength of the target to be measured and the The current in the left grating reflection region corresponding to the wavelength of the target to be measured. until the predicted wavelength and To minimize the error, then set Adjustment threshold , Adjustment threshold , Adjustment threshold and number of anomaly detection scan points Random selection Each sampling point, based on adjusting the threshold and... , and Perform anomaly detection, if The sampling points satisfy , and The difference between the measured value and the fixed value is less than , and Finally determined The corresponding five-dimensional current vector , To be optimal , To be optimal As the first in the wavelength current lookup table The data is processed by traversing the list of target wavelengths to obtain a preliminary wavelength current lookup table.

[0071] S3 includes S3.8, preset wavelength accuracy indicators, including root mean square error threshold. and maximum absolute error threshold Calculate the root mean square error of the preliminary wavelength current lookup table. and maximum absolute error :

[0072] ;

[0073] ;

[0074] In the formula, This represents the total number of all actual sampling points.

[0075] Preset sampling efficiency metrics, including sampling rate thresholds and time saving rate threshold Calculate the sampling rate of the preliminary wavelength current lookup table. and time saving rate :

[0076] ;

[0077] ;

[0078] ;

[0079] In the formula, In a two-dimensional nested cyclic scan mesh, with , The total number of sampling points obtained by scanning point by point at equal intervals in the direction;

[0080] If the current preliminary wavelength current lookup table simultaneously satisfies , , and Based on four conditions, the current preliminary wavelength current lookup table will be output as the final wavelength current lookup table.

[0081] Compared with existing technologies, this invention has the following advantages: By incorporating the K-singular value decomposition algorithm into the laser wavelength current lookup table construction process, this invention reduces time costs, lowers measurement load, and enhances deployment flexibility. It not only effectively reduces the frequency and intensity of manual intervention but also significantly shortens the occupancy cycle of core equipment such as fiber optic demodulators, reducing equipment wear and maintenance costs. In terms of scenario adaptability, the efficient table-building process makes this invention particularly suitable for industrial and scientific research applications involving frequent laser calibration and rapid equipment replacement. It eliminates the need for lengthy waiting times for the lookup table to be completed before system operation can resume, significantly improving the availability and maintenance efficiency of the entire laser tuning system and providing strong support for the continuous and stable operation of the equipment. Attached Figure Description

[0082] Figure 1 This is a flowchart of the method of the present invention;

[0083] Figure 2 This is a schematic diagram of the longitudinal mode region of the laser;

[0084] Figure 3 This is a flowchart of a sampling strategy based on Gaussian distribution;

[0085] Figure 4 This is a structural diagram of the K-singular value decomposition model;

[0086] Figure 5 This is the original sampling wavelength-current diagram of the test system;

[0087] Figure 6 This is a scan image obtained using the "group sampling" method;

[0088] Figure 7 This is the reconstructed wavelength-current graph. Detailed Implementation

[0089] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention are described clearly and completely below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0090] A laser-accelerated tabulation method, comprising:

[0091] S1. Initialize the laser by configuring three fixed-value currents, the left grating reflector current, and the right grating reflector current. The three fixed-value currents include the gain region current, the semiconductor optical amplifier current, and the phase region current. Then, perform a two-dimensional nested cyclic scan, including using the left grating reflector current and the right grating reflector current to construct a two-dimensional nested cyclic scan grid and generate a current matrix for each grid. Output the optical signal based on the current matrix and the three fixed-value currents, measure the center wavelength and side-mode suppression ratio of the optical signal, construct the center wavelength sequence according to the order of the two-dimensional nested cyclic scan, and calculate the longitudinal mode boundary and boundary current. Divide the current plane into continuous longitudinal mode regions according to the longitudinal mode boundary.

[0092] S2. Construct a theoretical sampling grid and a set of theoretical sampling points based on two-dimensional nested cyclic scanning and boundary current. Using the Gaussian distribution rule, calculate the number of actual sampling points for the theoretical sampling points closest to the center of the longitudinal mode region in each longitudinal mode region. Calculate the left and right grating reflection zone currents of the actual sampling points. Measure the wavelength and real-time power of the optical signal at the actual sampling points. Construct a sparse measurement training dataset based on the left and right grating reflection zone currents and wavelengths of the actual sampling points.

[0093] S3. Construct a K-singular value decomposition model using the K-singular value decomposition algorithm. Train the model using a sparse measurement training dataset to obtain the trained K-singular value decomposition model. The K-singular value decomposition model includes constructing input features and predicting target values, initializing an overcomplete dictionary, sparse coding, and dictionary updating. Input the list of target wavelengths to be measured into the trained K-singular value decomposition model, and use the gradient descent optimization algorithm to inversely adjust the target current until the preset wavelength accuracy index and sampling efficiency index are met. Output the final wavelength current lookup table.

[0094] S1 includes S1.1, initializing the laser and configuring three fixed-value currents, including setting the gain region current. Semiconductor optical amplifier current and phase region current It is a fixed value;

[0095] S1 includes S1.2, and the two-dimensional nested cyclic scanning includes the current in the left grating reflection region. and the current in the right grating reflection region The current direction is used to construct a two-dimensional nested cyclic scanning grid, which includes an outer loop and an inner loop nesting.

[0096] S1 includes S1.3, setting the outer loop, including setting... loop variable , , In order to be in The number of scan points in the direction; set according to the laser's current range. step size , The outer loop increments sequentially;

[0097] calculate real-time value :

[0098] ;

[0099] S1 includes S1.4, setting inner loop nesting, including setting... loop variable , , In order to be in The number of scan points in the direction; set according to the laser's current range. step size , Incrementing sequentially based on the inner nested loop;

[0100] calculate real-time value :

[0101] ;

[0102] S1 includes S1.5, in each loop node. Set the current matrix :

[0103] ;

[0104] according to , , and Output optical signal and measure the center wavelength of the optical signal. Main mode optical power and maximum side-mode optical power Calculate the side mode suppression ratio :

[0105] .

[0106] S1 includes, S1.6, scanning in the order of two-dimensional nested loops. Serialization yields the center wavelength sequence. , Let be the theoretical total number of sampling points for a two-dimensional nested cyclic scanning grid. For the index of the center wavelength, ,calculate difference :

[0107] ;

[0108] In the formula, It is the absolute value;

[0109] Set the jump threshold ,like ,Will Set as longitudinal boundary;

[0110] according to Divide the current plane into A continuous longitudinal modulus region ,set up For the index of the longitudinal region, The current plane is based on and In a two-dimensional parameter space defined by coordinate axes, the minimum current of the right grating reflection region in each longitudinal mode region is recorded. Maximum current in the right grating reflection region Minimum current in the left grating reflection zone Maximum current in the left grating reflection region Center wavelength and number of scan points .

[0111] S2 includes S2.1, generating the theoretical sampling mesh, which includes setting the target number of theoretical points for each longitudinal model region. , The range of current variation within the longitudinal mode region is calculated using boundary currents:

[0112] ;

[0113] ;

[0114] In the formula, for Range of directional changes for The range of changes in direction;

[0115] according to and Constructing the theoretical sampling grid:

[0116] ;

[0117] ;

[0118] In the formula, The rounding up symbol, for Number of grids in the direction, for Number of grid cells in the direction;

[0119] Determine the theoretical point set based on the theoretical sampling grid. , , For the theoretical sampling point index, for The Middle One theoretical sampling point, for Two-dimensional current coordinates, For the first Secondary theoretical sampling points , For the first Secondary theoretical sampling points ;

[0120] S2 includes S2.2, which calculates the standard deviation and current scan step size according to the matching principle. Standard deviation and Standard deviation:

[0121] ;

[0122] ;

[0123] In the formula, for standard deviation for standard deviation is the diffusion coefficient.

[0124] S2 includes, S2.3, and the actual sampling points follow the order of... A two-dimensional Gaussian distribution centered at the sampling amplification factor is set. and target sampling rate ,calculate The actual number of sampling points to be collected :

[0125] ;

[0126] In the formula, It is a rounding function;

[0127] Set the loop counter variable Perform cyclic sampling iterations. In each iteration, the actual sampling points are generated according to a two-dimensional Gaussian distribution rule:

[0128] ;

[0129] ;

[0130] In the formula, for No. actual sampling points , for No. actual sampling points , for No. Each actual sampling point relative to exist Random perturbation value in direction, for No. Each actual sampling point relative to exist Random perturbation value in direction, To conform to the distribution sign, It follows a Gaussian distribution;

[0131] S2 includes S2.4, for each actual sampling point, if , Located within the current longitudinal modulus region boundary, and with a distance greater than the safety threshold from the longitudinal modulus region boundary, according to , , , and Output No. The optical signal at each actual sampling point is collected by a wavelength meter to obtain the corresponding actual wavelength. and real-time power .

[0132] S2 includes S2.5, constructing a sparse measurement training dataset, including when Finish After the first actual sampling, All actual sampling points form a subset of training data. , ;

[0133] Set the effective power range according to the laser model. ,set up threshold ,like Not in or If the actual sampling point is determined to be a bad point, then the bad point is removed from... Delete; if In and Determine the actual sampling point as a good point and retain it. In the end, a subset of training data after removing bad pixels is obtained. , for No. actual sampling points ;

[0134] like The total number of sampling points is less than Re-examine Perform cyclic sampling iterations and set a threshold for the number of iterations to count the loop. If the actual number of loop iterations is greater than and The total number of actual sampling points is less than , recorded as The number of actual sampling points within the range of the centered two-dimensional Gaussian distribution is insufficient, so the corresponding points are deleted. ;

[0135] For all Summarize and construct a sparse measurement training dataset. .

[0136] S3 includes, S3.1, extraction middle , Based on input features, To predict the target value, the basic input features are expanded into a high-dimensional feature vector. , All actual sampling points Stacking to construct feature matrices , ,set up This is the index of the actual sampling point. The actual number of sampling points. For the first actual sampling points , for each actual sampling point As elements, construct the target value vector , , For the first actual sampling points ;

[0137] S3 includes S3.2, constructing a K-singular value decomposition model using the K-singular value decomposition algorithm. The K-singular value decomposition model includes a randomly initialized, overcomplete dictionary matrix. , , , For the set of real numbers, The number of atoms in the dictionary matrix. The dimension of the feature vector. for OK Let be the set of real matrices of columns. for The first in One atom, For the index of atoms, ;Will denoted as measurement sample , , for 3D real vector space;

[0138] S3 includes S3.3, which uses the fit method in the K singular value decomposition algorithm to start iterative optimization. The fit method includes sparse coding and dictionary update.

[0139] S3 includes S3.4, the sparse coding stage, which uses the orthogonal matching pursuit algorithm to... Perform iterations, calculating the current residual vector and... The absolute value of the inner product of each atom is used to select the atom with the largest absolute value of the inner product as the most relevant atom. After iteration, the result is obtained. The most relevant atom;

[0140] Fixed in each iteration ,right Solve for the corresponding sparse coefficient vector :

[0141] ;

[0142] ;

[0143] In the formula, The sparsity coefficient is . For the first The sparsity coefficients of each actual sampling point The number of non-zero elements. The sign is less than the order of magnitude;

[0144] S3 includes S3.5, the dictionary update phase, which fixes all... And form a sparse coefficient matrix:

[0145] ;

[0146] ;

[0147] In the formula, It is the L0 norm;

[0148] Construct a target function that minimizes the global reconstruction error:

[0149] ;

[0150] In the formula, It is the Frobenius norm;

[0151] according to Find all users The measurement samples constitute an index set. ,right For each measurement sample, calculate the removal The reconstructed residuals will The reconstructed residuals of all measured samples are combined into a residual matrix. ,right Perform singular value decomposition and take the largest left singular vector as the updated atom. Update using the product of the maximum right singular vector and the maximum singular value. ; Traverse and update all atoms to complete the process. A new round of updates.

[0152] S3 includes S3.6, which iteratively executes steps S3.4 and S3.5. When the objective function of minimizing the global reconstruction error is lower than a preset tolerance threshold, or when the number of model iterations reaches a set model iteration threshold, training is terminated, and the optimized dictionary is obtained. And the trained K-singular value decomposition model;

[0153] S3 includes, S3.7, obtaining the wavelength range of the laser under test based on a two-dimensional nested cyclic scan, with a generation interval of... Global target wavelength list ,set up For the index of the wavelength of the target to be measured, , The number of wavelengths of the target to be measured;

[0154] Will Input the trained K-singular value decomposition model, and use And gradient descent optimization algorithm, inversely adjust the first Current in the right grating reflection region corresponding to the wavelength of the target to be measured and the The current in the left grating reflection region corresponding to the wavelength of the target to be measured. until the predicted wavelength and To minimize the error, then set Adjustment threshold , Adjustment threshold , Adjustment threshold and number of anomaly detection scan points Random selection Each sampling point, based on adjusting the threshold and... , and Perform anomaly detection, if The sampling points satisfy , and The difference between the measured value and the fixed value is less than , and Finally determined The corresponding five-dimensional current vector , To be optimal , To be optimal As the first in the wavelength current lookup table The data is processed by traversing the list of target wavelengths to obtain a preliminary wavelength current lookup table.

[0155] S3 includes S3.8, preset wavelength accuracy indicators, including root mean square error threshold. and maximum absolute error threshold Calculate the root mean square error of the preliminary wavelength current lookup table. and maximum absolute error :

[0156] ;

[0157] ;

[0158] In the formula, This represents the total number of all actual sampling points.

[0159] Preset sampling efficiency metrics, including sampling rate thresholds and time saving rate threshold Calculate the sampling rate of the preliminary wavelength current lookup table. and time saving rate :

[0160] ;

[0161] ;

[0162] ;

[0163] In the formula, In a two-dimensional nested cyclic scan mesh, with , The total number of sampling points obtained by scanning point by point at equal intervals in the direction;

[0164] If the current preliminary wavelength current lookup table simultaneously satisfies , , and Based on four conditions, the current preliminary wavelength current lookup table will be output as the final wavelength current lookup table.

[0165] The following description, in conjunction with the accompanying drawings, further illustrates the method flow of this invention. Figure 1 As shown, after starting the tabulation process, the laser system is initialized, including determining parameters such as current and temperature; then, longitudinal mode region identification is performed, including identification... The third step involves executing a "group sampling" strategy, which is a sampling strategy based on Gaussian distribution, and determining the sampling rate index. The fourth step involves training a K-SVD (K-Singular Value Decomposition) dictionary, including determining the accuracy index. Then, the lookup table coefficients are reconstructed, power calibration is optimized, and accuracy is verified. Finally, a multi-dimensional evaluation is performed. If the preset conditions are not met, parameter optimization is performed, and the process returns to the third step to execute the "group sampling" strategy. If the preset conditions are met, the lookup table is output.

[0166] In terms of time savings, traditional full-range fine current scanning methods require approximately 25 minutes to construct a lookup table with a 40nm range and 8pm intervals. However, this invention, relying on the group sampling strategy to accurately remove redundant data and the efficient feature reconstruction capability of the K-SVD algorithm, can complete the construction of a lookup table with the same precision in just 5 to 7 minutes, achieving a stable time saving rate of 75% to 80%. This significant time compression stems from the optimization of the amount of sampled data and the synergistic improvement of the algorithm's computational efficiency. In terms of optimizing the measurement workload, this invention significantly reduces the sampling rate for lookup table construction from 100% in traditional methods to 15% to 25%, corresponding to a sharp reduction in the number of actual measurement points from 5000 to 750 to 1250.

[0167] The embodiments of the present invention are verified and demonstrated through the following coherent experimental procedure, which is consistent with... Figure 1The algorithm system flow shown is completely consistent.

[0168] The first step is experiment startup and system initialization. After the experiment begins, the hardware and software systems are initialized first. The host computer establishes communication with the laser driver board and performs the initialization according to the preset protocol. , and Set to a fixed value. , , The high-precision temperature control module is then activated to suppress wavelength drift. This step establishes a stable and repeatable reference operating point for all subsequent measurements.

[0169] The second step is coarse scanning and longitudinal region identification; the system performs... , Perform a two-dimensional nested loop coarse scan to measure the output wavelength under each current combination. Compared with edge mode suppression ratio .like Figure 2 As shown, after the scan is completed, the wavelength data exhibits a clear blocky distribution on the current plane, with each "block" corresponding to a laser longitudinal mode. This is achieved by calculating the difference in the wavelength sequence and setting... The extreme values ​​of the longitudinal modulus region are respectively , , , Obtain by adding a counter to the scanning program This divides the entire current plane into multiple independent continuous longitudinal mode regions. And record the boundary current and center wavelength of each region.

[0170] The third step is adaptive "group sampling". Based on the predefined longitudinal mode regions, the system enters the "group sampling" stage, and its strategy is as follows: Figure 3 As shown, according to , Longitudinal modulus region identification is performed within each region, including generating a uniform theoretical sampling grid and generating theoretical sampling points based on the grid. Then, Gaussian group sampling is performed around each theoretical point, including setting... (Values ​​range from 1.2 to 1.5) and (Values ​​range from 0.15 to 0.25), and then actual measurement points are taken, including a group of actual points randomly generated according to a two-dimensional Gaussian distribution. This "dense at the center and sparse at the periphery" sampling method can effectively avoid the unstable region at the longitudinal mode boundary, while capturing the local gradient and curvature information of the current-wavelength mapping with the fewest sampling points. A safety threshold is set for each actual sampling point. The system determines whether to output an optical signal based on a safety threshold; during the sampling process, it sets... for arrive ,set up Real-time detection of each point and Remove bad pixels and keep good pixels to generate ,like The total number of sampling points is less than Set a threshold for the number of iterations in the loop count. (Values ​​range from 3 to 5 times) perform point replenishment; if after point replenishment... The total number of sampling points is still less than This theoretical point The small region to which it belongs is marked as a low-confidence sampling area. The number of theoretical points corresponding to the marked points is insufficient. It will not be included in the final sparse measurement training dataset. In this process, data that is not representative enough or may contain hidden instabilities is excluded from subsequent K-SVD model training to avoid negatively impacting the overall model accuracy. Simultaneously, a log is generated, providing data support for subsequent analysis of the reasons for laser instability in specific current regions, which can be used for device performance evaluation or process improvement. Finally, all data is fused. Construct a high-quality sparse measurement training dataset .

[0171] The fourth step is K-SVD (K-Singular Value Decomposition) dictionary training and model building, where the sparse dataset obtained in the previous step is input into the K-SVD algorithm for training. The training process is as follows: Figure 4 As shown, after starting dictionary training, data preparation and initialization are performed first, followed by iterative optimization using the `fit` method. The main iteration loop achieves collaborative optimization of the dictionary and sparse representation by alternating between two phases: sparse encoding and dictionary update. The specific process includes initializing the iteration counter dictionary matrix. Sparse constraints are set for overcomplete matrices that are randomly generated or constructed based on theoretically sampled grids. =5. Maximum number of iterations Reconstruction error tolerance threshold Calculate the initial reconstruction error :

[0172] ;

[0173] In the formula, This forms the initial coefficient matrix; then the coefficient encoding stage begins.

[0174] The sparse coding stage aims to find the optimal sparse representation for each training sample given the current dictionary. The algorithm first fixes the current dictionary. Then, it iterates through each training sample, and for each input current point sample, uses the Orthogonal Matching Pursuit (OMP) algorithm to solve for the sparse coefficients, integrating them to obtain the coefficient matrix. This algorithm uses a greedy iterative strategy to successively select the current point from the dictionary that best matches the residual of the current sample—that is, the current point with the largest inner product—and includes it in the representation set. After each selection, the algorithm recalculates the optimal linear combination coefficients using the least squares method based on the selected set of current points and updates the residual. This process is repeated until the preset sparsity is reached, thus obtaining a highly concise sparse representation of the sample, consisting only of a few key current points and their corresponding coefficients. After traversing all samples, the complete sparse coefficient matrix of the current iteration step is obtained. Specifically, the current dictionary is fixed. , The iteration number is the number of iterations for the training sample matrix. Each column in The orthogonal matching pursuit algorithm is used to solve for the sparse coefficient vector. Solve The process includes, for a given sample and its current residual Calculate the dictionary The Middle Atoms and correlation measure :

[0175] ;

[0176] In the formula, For vector dot product, To take the absolute value;

[0177] Set atomic extraction conditions:

[0178] ;

[0179] In the formula, The index value of the atom selected for this iteration; The atoms selected for this iteration;

[0180] For a given sample Finding sparse coefficient vectors Set sparsity This makes in Under the constraints, it uses a dictionary. The linear reconstruction has the smallest error compared to the original sample, i.e., it satisfies... and . Representing vectors The Euclidean norm is used to measure reconstruction error. express The number of non-zero elements ( (Pseudonorm).

[0181] After obtaining the sparse representations of all sample current points, the algorithm enters the dictionary update phase. At this point, with the coefficient matrix fixed, the algorithm iterates through each atom, focusing on optimizing each current point in the dictionary to more effectively represent all sample current points that use it. Specifically, for each current point in the dictionary, the set of samples using that atom is identified; that is, the system first identifies a subset of all sample current points that use that current point based on the sparse coefficient matrix. Next, the residual matrix is ​​calculated and SVD is performed, including calculating the representation residual after removing the contribution of this current point, combining the residual vectors corresponding to these sample current points into a residual matrix, performing singular value decomposition on this residual matrix, and selecting its largest left singular vector as the updated new current point. This operation is mathematically equivalent to finding the optimal unit vector that best represents the common direction of the set of residuals, thus achieving directional optimization for that current point. Simultaneously, the corresponding sparse coefficients are also adjusted synchronously according to the results of the singular value decomposition to maintain representation consistency. This process iterates through all current points in the dictionary, completing the dictionary update and corresponding coefficient update. Specifically, the coefficient matrix is ​​fixed. Update the dictionary column by column. atoms Find out which atoms are used. sample index set Calculate the residual matrix after removing the contribution of this atom. Singular value decomposition is performed on the residual matrix, and the largest left singular vector is taken as the new atom. Synchronous updates The coefficient at the corresponding position in the middle.

[0182] The two stages described above constitute a complete iteration. The algorithm iteratively executes an alternating "encode-update" cycle, allowing the dictionary and sparse representation to adjust and improve each other during the iteration. The convergence of the entire process is controlled by the decrease in reconstruction error: the iteration terminates when the change in reconstruction error between two adjacent iterations is less than a preset tolerance threshold, or when the number of iterations reaches a preset upper limit. The final output is the fully trained optimal dictionary and its corresponding sparse representation model, which can reconstruct the current-wavelength mapping relationship across the entire laser domain from sparsely sampled data with extremely high efficiency and accuracy.

[0183] The fifth step involves reverse reconstruction and verification of the lookup table. Using the trained K-SVD model, the target wavelength list is solved in reverse. For each target wavelength... The gradient descent optimization algorithm is used to iteratively adjust... and By using forward prediction, the model learns the complex mapping relationship between current and wavelength, calculates and outputs a predicted wavelength value, then calculates the error, analyzes the sensitivity of the error to the two current variables (i.e., gradient calculation), and finally fine-tunes the model along the reverse direction of the calculated gradient. and The wavelength obtained in the coarse scanning stage is closest to the target value. The measured current is used as the initial starting point for iteration until the global reconstruction error objective function is minimized to a preset tolerance threshold (0.001 nm) or the number of iterations exceeds an iteration threshold (20 times). This ensures that the wavelength value predicted by the trained K-SVD model based on the current is consistent with the given target wavelength. The error between them was minimized. Subsequently, the fixed current settings for the other three circuits were adjusted. , , , Fine-tuning was performed using a "small-area rapid scan," as shown in Table 1:

[0184] Table 1. Definition of parameters for small-range fast scan

[0185] ;

[0186] After verification, the final five-dimensional current combination is obtained. After traversing all target wavelengths, a complete wavelength-current lookup table is generated. Finally, by calculating the root mean square error, maximum absolute error, and sampling rate of the lookup table and comparing them with preset thresholds, the overall performance of the invention is verified to meet expectations.

[0187] To verify the effectiveness and superiority of this invention, an SSGDBR tunable laser was used for experimental verification. The current range of the left and right grating reflection regions of the laser was selected to be 0mA to 35mA, and the scanning step size was... mA. According to the traditional full-scan method, measurements must be taken point-by-point within a two-dimensional current plane, resulting in a total of [number of points missing]. The calculation formula is as follows:

[0188] ;

[0189] Traditional methods use a step size that covers the entire current range while avoiding skipping longitudinal modes or losing key features due to excessively large intervals. A single full scan using traditional methods takes approximately 25 minutes. The method of this invention sets a target sampling rate. Amplification factor Actual number of measurement points The score was 24,640, and the time taken was approximately 5 minutes. Key performance comparisons are shown in Table 2.

[0190] Table 2. Performance Comparison between the Method of the Present Invention and Traditional Full Scan Methods

[0191] ;

[0192] As shown in Table 2, the method of this invention requires only 20% of the measurement points of the traditional method, reducing the construction time from 25 minutes to 5 minutes, achieving a time saving rate of 80%. Regarding wavelength accuracy, the root mean square error only slightly increases from 0.005 nm to 0.008 nm, and the maximum absolute error increases from 0.012 nm to 0.015 nm. Both indicators meet the typical requirements for wavelength accuracy in fiber optic demodulation systems (root mean square error ≤ 0.01 nm, maximum absolute error ≤ 0.02 nm). This indicates that the present invention significantly improves the construction efficiency of the lookup table while ensuring wavelength mapping accuracy.

[0193] Furthermore, to illustrate the adaptive sampling effect of the present invention in different longitudinal modulus regions, Table 3 lists the sampling statistics for three random longitudinal modulus regions:

[0194] Table 3. Distribution statistics of sampling points in typical longitudinal model areas

[0195] ;

[0196] As shown in Table 3, the method of this invention achieves stable and uniform sparse sampling in different longitudinal modulus regions, with the sampling rate in each region maintained at a set value of approximately 20%. This is due to the region division based on longitudinal modulus boundary recognition and the "group sampling" strategy based on Gaussian distribution, which enables sampling points to adaptively avoid unstable boundaries and uniformly cover the stable regions within each longitudinal modulus, thereby providing high-quality and highly representative training data for subsequent K-SVD dictionary learning.

[0197] Simultaneously, illustrations of the actual modeling effect of the present invention are provided. Figure 5 The distribution of the original sampling points of the experimental system with current ranging from 0mA to 35mA is shown. It can be observed that the current-wavelength relationship presents a clear longitudinal mode region grouping characteristic based on the original measurement data. Figure 6 This shows the sparse sampling point distribution guided by the "group sampling" method, with a significantly smaller number of points than... Figure 5 This demonstrates the efficiency of this method in the data acquisition phase. Figure 7 The final current-wavelength relationship image obtained after reconstruction through the complete K-SVD process is compared with... Figure 5 As can be seen from the comparison, the reconstructed image is highly consistent with the original measurement results in terms of trend and detail, with minimal error. This directly verifies that the present invention can maintain excellent wavelength mapping reconstruction accuracy while significantly reducing the sampling burden.

[0198] In summary, the experimental results show that the present invention, through the deep integration of the "group sampling" strategy and K-SVD dictionary learning, reduces the lookup table construction time by 80% while ensuring the accuracy of wavelength-current mapping, thus significantly improving the deployment efficiency and practicality of laser wavelength scanning control.

[0199] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A laser-accelerated tabulation method, characterized in that, include: S1. Initialize the laser by configuring three fixed-value currents, the left grating reflector current, and the right grating reflector current. The three fixed-value currents include the gain region current, the semiconductor optical amplifier current, and the phase region current. Then, perform a two-dimensional nested cyclic scan, including using the left grating reflector current and the right grating reflector current to construct a two-dimensional nested cyclic scan grid and generate a current matrix for each grid. Output the optical signal based on the current matrix and the three fixed-value currents, measure the center wavelength and side-mode suppression ratio of the optical signal, construct the center wavelength sequence according to the order of the two-dimensional nested cyclic scan, and calculate the longitudinal mode boundary and boundary current. Divide the current plane into continuous longitudinal mode regions according to the longitudinal mode boundary. S2. Construct a theoretical sampling grid and a set of theoretical sampling points based on two-dimensional nested cyclic scanning and boundary current. Using the Gaussian distribution rule, calculate the number of actual sampling points for the theoretical sampling points closest to the center of the longitudinal mode region in each longitudinal mode region. Calculate the left and right grating reflection zone currents of the actual sampling points. Measure the wavelength and real-time power of the optical signal at the actual sampling points. Construct a sparse measurement training dataset based on the left and right grating reflection zone currents and wavelengths of the actual sampling points. S3. Construct a K-singular value decomposition model using the K-singular value decomposition algorithm. Train the model using a sparse measurement training dataset to obtain the trained K-singular value decomposition model. The K-singular value decomposition model includes constructing input features and predicting target values, initializing an overcomplete dictionary, sparse coding, and dictionary updating. Input the list of target wavelengths to be measured into the trained K-singular value decomposition model, and use the gradient descent optimization algorithm to inversely adjust the target current until the preset wavelength accuracy index and sampling efficiency index are met. Output the final wavelength current lookup table.

2. The laser-accelerated tabulation method according to claim 1, characterized in that, S1 includes S1.1, initializing the laser and configuring three fixed-value currents, including setting the gain region current. Semiconductor optical amplifier current and phase region current It is a fixed value; S1 includes S1.2, and the two-dimensional nested cyclic scanning includes the current in the left grating reflection region. and the current in the right grating reflection region The current direction is used to construct a two-dimensional nested cyclic scanning grid, which includes an outer loop and an inner loop nesting. S1 includes S1.3, setting the outer loop, including setting... loop variable , , In order to be in The number of scan points in the direction; set according to the laser's current range. step size , The outer loop increments sequentially; calculate real-time value : ； S1 includes S1.4, setting inner loop nesting, including setting... loop variable , , In order to be in The number of scan points in the direction; set according to the laser's current range. step size , Incrementing sequentially based on the inner nested loop; calculate real-time value : ； S1 includes S1.5, in each loop node. Set the current matrix : ； according to , , and Output optical signal and measure the center wavelength of the optical signal. Main mode optical power and maximum side-mode optical power Calculate the side mode suppression ratio : 。 3. The laser-accelerated tabulation method according to claim 2, characterized in that, S1 includes, S1.6, scanning in the order of two-dimensional nested loops. Serialization yields the center wavelength sequence. , Let be the theoretical total number of sampling points for a two-dimensional nested cyclic scanning grid. For the index of the center wavelength, ,calculate difference : ； In the formula, It is the absolute value; Set the jump threshold ,like ,Will Set as longitudinal boundary; according to Divide the current plane into A continuous longitudinal modulus region ,set up For the index of the longitudinal region, The current plane is based on and In a two-dimensional parameter space defined by coordinate axes, the minimum current of the right grating reflection region in each longitudinal mode region is recorded. Maximum current in the right grating reflection region Minimum current in the left grating reflection zone Maximum current in the left grating reflection region Center wavelength and number of scan points .

4. The laser-accelerated tabulation method according to claim 3, characterized in that, S2 includes S2.1, generating the theoretical sampling mesh, which includes setting the target number of theoretical points for each longitudinal model region. , The range of current variation within the longitudinal mode region is calculated using boundary currents: ； ； In the formula, for Range of directional changes for The range of changes in direction; according to and Constructing the theoretical sampling grid: ； ； In the formula, The rounding up symbol, for Number of grids in the direction, for Number of grid cells in the direction; Determine the theoretical point set based on the theoretical sampling grid. , , For the theoretical sampling point index, for The Middle One theoretical sampling point, for Two-dimensional current coordinates, For the first Secondary theoretical sampling points , For the first Secondary theoretical sampling points ; S2 Including S2.2, calculating based on the matching principle of standard deviation and current scan step size. Standard deviation and Standard deviation: ； ； In the formula, for standard deviation for standard deviation is the diffusion coefficient.

5. The laser-accelerated tabulation method according to claim 4, characterized in that, S2 includes, S2.3, and the actual sampling points follow the order of... A two-dimensional Gaussian distribution centered at the sampling amplification factor is set. and target sampling rate ,calculate The actual number of sampling points to be collected : ； In the formula, It is a rounding function; Set the loop counter variable Perform cyclic sampling iterations. In each iteration, the actual sampling points are generated according to a two-dimensional Gaussian distribution rule: ； ； In the formula, for No. actual sampling points , for No. actual sampling points , for No. Each actual sampling point relative to exist Random perturbation value in direction, for No. Each actual sampling point relative to exist Random perturbation value in direction, To conform to the distribution sign, It follows a Gaussian distribution; S2 includes S2.4, for each actual sampling point, if , Located within the current longitudinal modulus region boundary, and with a distance greater than the safety threshold from the longitudinal modulus region boundary, according to , , , and Output No. The optical signal at each actual sampling point is collected by a wavelength meter to obtain the corresponding actual wavelength. and real-time power .

6. The laser-accelerated tabulation method according to claim 5, characterized in that, S2 includes S2.5, constructing a sparse measurement training dataset, including when... Finish After the first actual sampling, All actual sampling points form a subset of training data. , ; Set the effective power range according to the laser model. ,set up threshold ,like Not in or If the actual sampling point is determined to be a bad point, then the bad point is removed from... Delete; if In and Determine the actual sampling point as a good point and retain it. In the end, a subset of training data after removing bad pixels is obtained. , for No. actual sampling points ; like The total number of sampling points is less than Re-examine Perform cyclic sampling iterations and set a threshold for the number of iterations to count the loop. If the actual number of loop iterations is greater than and The total number of actual sampling points is less than , recorded as The number of actual sampling points within the range of the centered two-dimensional Gaussian distribution is insufficient, so the corresponding points are deleted. ; For all Summarize and construct a sparse measurement training dataset. .

7. The laser-accelerated tabulation method according to claim 6, characterized in that, S3 includes, S3.1, extraction middle , Based on input features, To predict the target value, the basic input features are expanded into a high-dimensional feature vector. , All actual sampling points Stacking to construct feature matrices , ,set up This is the index of the actual sampling point. The actual number of sampling points. For the first actual sampling points , for each actual sampling point As elements, construct the target value vector , , For the first actual sampling points ; S3 includes S3.2, constructing a K-singular value decomposition model using the K-singular value decomposition algorithm. The K-singular value decomposition model includes a randomly initialized, overcomplete dictionary matrix. , , , For the set of real numbers, The number of atoms in the dictionary matrix. The dimension of the feature vector. for OK Let be the set of real matrices of columns. for The first in One atom, For the index of atoms, ;Will denoted as measurement sample , , for 3D real vector space; S3 includes S3.3, which uses the fit method in the K singular value decomposition algorithm to start iterative optimization. The fit method includes sparse coding and dictionary update.

8. The laser-accelerated tabulation method according to claim 7, characterized in that, S3 includes S3.4, the sparse coding stage, which uses the orthogonal matching pursuit algorithm to... Perform iterations, calculating the current residual vector and... The absolute value of the inner product of each atom is used to select the atom with the largest absolute value of the inner product as the most relevant atom. After iteration, the result is obtained. The most relevant atom; Fixed in each iteration ,right Solve for the corresponding sparse coefficient vector : ； ； In the formula, The sparsity coefficient is . For the first The sparsity coefficients of each actual sampling point The number of non-zero elements. The sign is less than the order of magnitude; S3 includes S3.5, the dictionary update phase, which fixes all... And form a sparse coefficient matrix: ； ； In the formula, It is the L0 norm; Construct a target function that minimizes the global reconstruction error: ； In the formula, It is the Frobenius norm; according to Find all users The measurement samples constitute an index set. ,right For each measurement sample, calculate the removal The reconstructed residuals will The reconstructed residuals of all measured samples are combined into a residual matrix. ,right Perform singular value decomposition and take the largest left singular vector as the updated atom. Update using the product of the maximum right singular vector and the maximum singular value. ; Traverse and update all atoms to complete the process. A new round of updates.

9. A laser-accelerated tabulation method according to claim 8, characterized in that, S3 includes S3.6, which iteratively executes steps S3.4 and S3.

5. When the objective function of minimizing the global reconstruction error is lower than a preset tolerance threshold, or when the number of model iterations reaches a set model iteration threshold, training is terminated, and the optimized dictionary is obtained. And the trained K-singular value decomposition model; S3 includes, S3.7, obtaining the wavelength range of the laser under test based on a two-dimensional nested cyclic scan, with a generation interval of... Global target wavelength list ,set up For the index of the wavelength of the target to be measured, , The number of wavelengths of the target to be measured; Will Input the trained K-singular value decomposition model, and use And gradient descent optimization algorithm, inversely adjust the first Current in the right grating reflection region corresponding to the wavelength of the target to be measured and the The current in the left grating reflection region corresponding to the wavelength of the target to be measured. until the predicted wavelength and To minimize the error, then set Adjustment threshold , Adjustment threshold , Adjustment threshold and number of anomaly detection scan points Random selection Each sampling point, based on adjusting the threshold and... , and Perform anomaly detection, if The sampling points satisfy , and The difference between the measured value and the fixed value is less than , and Finally determined The corresponding five-dimensional current vector , To be optimal , To be optimal As the first in the wavelength current lookup table The data is processed by traversing the list of target wavelengths to obtain a preliminary wavelength current lookup table.

10. A laser-accelerated tabulation method according to claim 9, characterized in that, S3 includes S3.8, preset wavelength accuracy indicators, including root mean square error threshold. and maximum absolute error threshold Calculate the root mean square error of the preliminary wavelength current lookup table. and maximum absolute error : ； ； In the formula, This represents the total number of all actual sampling points. Preset sampling efficiency metrics, including sampling rate thresholds and time saving rate threshold Calculate the sampling rate of the preliminary wavelength current lookup table. and time saving rate : ； ； ； In the formula, In a two-dimensional nested cyclic scan mesh, with , The total number of sampling points obtained by scanning point by point at equal intervals in the direction; If the current preliminary wavelength current lookup table simultaneously satisfies , , and Based on four conditions, the current preliminary wavelength current lookup table will be output as the final wavelength current lookup table.

Images