High dynamic range multi-channel laser interference control method and related equipment

Abstract

The invention relates to the technical field of laser control, and provides a high-dynamic-range multi-channel laser interference control method and related equipment. Channel signals of a plurality of target channels are obtained in real time according to a plurality of preset photoelectric detectors, time domain-frequency domain-space three-dimensional sampling is performed on the channel signals to construct an original tensor, joint optimization processing is performed on the original tensor to obtain a noise reduction tensor, phase unwrapping is performed on the noise reduction tensor to obtain an absolute phase matrix, and the absolute phase matrix is subjected to noise reduction processing to obtain an absolute phase matrix. Performing dynamic range expansion on the absolute phase matrix to obtain an enhanced phase matrix, performing multi-wavelength calculation and environment compensation on the enhanced phase matrix to obtain a distance matrix, and performing vibration compensation on the distance matrix to obtain a piezoelectric driving signal. According to the channel laser interference control method, through a series of advanced algorithms and processing means such as multi-dimensional signal acquisition, combined noise reduction optimization, precise phase unwrapping, dynamic range expansion and vibration compensation, comprehensive improvement of laser interference measurement signals is realized.

Description

Technical Field

[0001] This application relates to the field of laser control technology, and particularly to a high-dynamic-range multi-channel laser interference control method and related equipment. Background Art

[0002] Laser interference control technology is an important means to achieve ultra-high-precision displacement, vibration, and deformation measurement, and is widely used in fields such as micro-nano manufacturing, precision machining, optical detection, and space measurement. Its control accuracy directly affects the overall performance of the system. Traditional laser interference control methods mainly include single-channel interference technology and interference technology based on heterodyne and phase modulation.

[0003] Single-channel interference technology usually samples the laser interference signal in the time or frequency domain, and then uses algorithms such as Fourier transform and phase unwrapping to perform signal processing. High measurement accuracy can be obtained under the conditions of low noise and stable environment, but it essentially relies on a single sampling channel and cannot fully capture the spatial distribution characteristics of the signal. The interference technology based on heterodyne and phase modulation realizes multi-channel acquisition and phase demodulation of the signal by introducing frequency offset. The interference signal is processed by using algorithms such as fast Fourier transform and digital filtering to extract the phase information. However, since the acquisition means still focus on a single channel or a limited number of channels, it is difficult to separate spatially correlated noise (such as environmental vibration) from uncorrelated noise (such as photoelectric shot noise) when processing the signal, resulting in the effective signal being easily confused or weakened in a high-noise background.

[0004] Therefore, traditional laser interference control methods have limited dynamic range in aspects such as noise suppression, signal extraction, and measurement range extension, and cannot simultaneously meet the measurement requirements from single-photon level to high-intensity level. Summary of the Invention

[0005] In view of this, this application provides a high-dynamic-range multi-channel laser interference control method and related equipment to solve the problem of limited dynamic range of laser interference control.

[0006] The first aspect of this application provides a high-dynamic-range multi-channel laser interference control method, and the method includes: Real-time obtain channel signals of multiple target channels according to a preset plurality of photodetectors, and perform three-dimensional sampling in the time-frequency-space domain on the channel signals to construct an original tensor; Perform joint optimization processing on the original tensor to obtain a noise-reduced tensor; Perform phase unwrapping on the noise-reduced tensor to obtain an absolute phase matrix; Perform dynamic range expansion on the absolute phase matrix to obtain an enhanced phase matrix; Perform multi-wavelength resolution and environmental compensation on the enhanced phase matrix to obtain a distance matrix; Perform vibration compensation on the distance matrix to obtain a piezoelectric drive signal.

[0007] In an optional embodiment, the performing three-dimensional sampling of the channel signal in the time-frequency-space domain to construct the original tensor includes: Perform time-domain synchronization processing on the channel signal to obtain a time-aligned optimized channel signal set; Perform short-time Fourier transform processing on the time-aligned optimized channel signal set according to a preset adaptive window to obtain the time-frequency matrix of each target channel; Perform three-dimensional fusion module construction processing on the time-frequency matrix to obtain the original tensor.

[0008] In an optional embodiment, the performing joint optimization processing on the original tensor to obtain a denoised tensor includes: Step S21: Perform Tucker decomposition processing on the original tensor to obtain an initial low-rank signal; Step S22: Perform calculation processing of a preset soft-threshold function on the initial low-rank signal to obtain an initial noise signal; Step S23: Calculate a denoising residual tensor according to the original tensor and the initial noise signal, and perform Tucker decomposition processing on the denoising residual tensor to update the initial low-rank signal; Step S24: Calculate a reconstruction error according to the original tensor, the initial noise signal, and the updated initial low-rank signal to obtain a reconstruction error value, and compare the reconstruction error value with a preset error threshold; Repeat the execution of Step S22 to Step S24 until the reconstruction error value is less than the error threshold, and determine the updated initial low-rank signal as the denoised tensor.

[0009] In an optional embodiment, the performing phase unwrapping on the denoised tensor to obtain an absolute phase matrix includes: Perform frequency amplitude spectrum peak search on the denoised tensor to extract the main frequency phase value of each preset space-time unit and generate an original phase vector; Perform multi-frequency phase synthesis on the original phase vector according to three sets of preset wavelength laser parameters to calculate the synthesized wavelength parameter; Perform phase difference coupling operation on the original phase vector according to the synthesized wavelength parameter to generate an unwrapped absolute phase matrix; Perform non-linear refractive index compensation on the unwrapped absolute phase matrix according to the real-time acquired environmental parameter set to obtain a refractive index compensated phase matrix; Perform phase continuity constraint processing on the refractive index compensation phase matrix to obtain a noise reduction compensation phase matrix; Perform phase unwrapping on the noise reduction compensation phase matrix through a preset phase unwrapping depth network model to obtain the absolute phase matrix.

[0010] In an optional embodiment, the performing dynamic range expansion on the absolute phase matrix to obtain an enhanced phase matrix includes: Perform spatio-temporal gradient calculation on the absolute phase matrix to generate a phase gradient tensor; Construct a multi-dimensional state vector according to the phase gradient tensor, the environmental parameter set, and a preset laser wavelength drift amount; Perform parameter adaptive adjustment on the multi-dimensional state vector through a preset deep reinforcement learning network model to obtain a dynamic control parameter set; Perform parametric adjustment on the absolute phase matrix according to the dynamic control parameter set to obtain a frequency domain band-pass filtered phase matrix; Perform non-linear compression mapping on the frequency domain band-pass filtered phase matrix to obtain a compression mapping phase matrix; Perform tensor splicing on the compression mapping phase matrix and the phase gradient tensor to generate the enhanced phase matrix.

[0011] In an optional embodiment, the performing multi-wavelength resolution and environmental compensation on the enhanced phase matrix to obtain a distance matrix includes: Perform phase difference synthesis calculation on the enhanced phase matrix according to the three groups of wavelength laser parameters to generate a synthesized wavelength parameter; Perform equivalent synthesized wavelength calculation on the synthesized wavelength parameter through a preset synthesized wavelength formula to generate a multi-level unambiguous distance interval; Perform real-time refractive index correction on the multi-level unambiguous distance interval according to the environmental parameter set to obtain a refractive index corrected distance interval; Perform multi-source error coupling analysis on the refractive index corrected distance interval to obtain a wavelength stability correction factor and a geometric distortion compensation matrix; Perform tensor fusion on the refractive index corrected distance interval, the wavelength stability correction factor, and the geometric distortion compensation matrix according to a preset tensor fusion method to obtain an initial distance tensor; Perform residual compensation iterative optimization on the initial distance tensor to obtain the distance matrix.

[0012] In an optional embodiment, the performing vibration compensation on the distance matrix to obtain a piezoelectric drive signal includes: Perform a multi-channel joint fast Fourier transform on the distance matrix along the time dimension to generate a vibration spectrum tensor; Perform an inverse transfer function calculation process on the vibration spectrum tensor to obtain a feedforward compensation signal; Perform a spatio-temporal differential operation on the distance matrix to generate a real-time vibration error vector; Perform a control simulation on the feedforward compensation signal and the real-time vibration error vector through a preset hybrid controller simulation model to obtain a feedback control signal; Perform a gain scheduling on the feedback control signal according to a preset environmental stiffness parameter to obtain an optimized feedback control signal; Perform a pulse width modulation encoding on the optimized feedback control signal to generate a multi-level piezoelectric drive waveform; Perform a waveform adjustment on the multi-level piezoelectric drive waveform through a preset waveform shaping model to obtain the piezoelectric drive signal.

[0013] A second aspect of the present application provides a high-dynamic range multi-channel laser interference control device, the device includes: A feature acquisition module, configured to obtain channel signals of multiple target channels in real time according to a preset plurality of photodetectors, and perform three-dimensional sampling in the time domain-frequency domain-space on the channel signals to construct an original tensor; A noise reduction and optimization module, configured to perform a joint optimization process on the original tensor to obtain a noise reduction tensor; A phase unwrapping module, configured to perform phase unwrapping on the noise reduction tensor to obtain an absolute phase matrix; A dynamic expansion module, configured to perform a dynamic range expansion on the absolute phase matrix to obtain an enhanced phase matrix; A distance compensation module, configured to perform multi-wavelength resolution and environmental compensation on the enhanced phase matrix to obtain a distance matrix; A vibration compensation module, configured to perform vibration compensation on the distance matrix to obtain a piezoelectric drive signal.

[0014] A third aspect of the present application provides an electronic device, the electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the computer program, the steps of the high-dynamic range multi-channel laser interference control method as described above are implemented.

[0015] A fourth aspect of the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the high-dynamic range multi-channel laser interference control method as described above are implemented.

[0016] In summary, the present application includes at least the following beneficial technical effects: 1. Through spatio-temporal gradient calculation, multi-dimensional state vector construction, and parameter adaptive adjustment of the deep reinforcement learning network model, the dynamic range of the absolute phase matrix is expanded. It can adapt to large-scale signal changes and maintain high-precision measurement under different environmental conditions, improving the applicability and sensitivity of laser interference control.

[0017] 2. A high-precision distance matrix is obtained through tensor fusion. This multi-dimensional and multi-stage processing strategy can effectively compensate for environmental changes and system errors, ensuring the accuracy and reliability of the final measurement results. 3. A multi-level piezoelectric drive waveform is generated through pulse width modulation coding, realizing real-time and fine compensation for vibration. This not only reduces the interference of environmental vibration on measurement but also improves the dynamic response and control accuracy of the overall system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 is a flowchart of a high-dynamic range multi-channel laser interference control method provided by an embodiment of the present application; Figure 2 is a functional module diagram of a high-dynamic range multi-channel laser interference control device provided by an embodiment of the present application; Figure 3 is a schematic structural diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0021] As Figure 1 shown, it is a flowchart of a high-dynamic range multi-channel laser interference control method provided by an embodiment of the present application. The high-dynamic range multi-channel laser interference control method provided by an embodiment of the present application includes the following steps.

[0022] Step S1: Obtain the channel signals of multiple target channels in real time according to a plurality of preset photodetectors, and perform three-dimensional sampling in the time domain-frequency domain-space on the channel signals to construct an original tensor.

[0023] In the embodiment of the present application, to reduce the time deviation of the acquisition of each channel signal, a hardware trigger pin is configured for each photodetector, and the same external clock source (for example, a 50 MHz square wave signal output by an FPGA) is received as the sampling trigger reference. At the same time, each detector is equipped with two ADC buffer areas (i.e., Buffer A and Buffer B). When Buffer A performs analog-to-digital conversion, Buffer B transfers the acquired data to the memory through DMA to ensure no sampling gap. Further, the following cross-correlation algorithm is used to calculate the initial time deviation between channels: Among them, are the original sampling signals of two channels respectively, is the time offset, the search range is ±10 sampling points, and T is the calibration signal length (usually 1 ms of data is taken, that is, 1000 sampling points). Through the cross-correlation function, the maximum time offset between two channels is obtained, which is the inherent delay between channels.

[0024] The following cubic spline interpolation compensation is performed on the communication signal with delay: Among them, t i is the original sampling time point, a i , b i , c i , d i are the coefficients determined by the data of adjacent sampling points, satisfying the continuity of the second derivative. Through cubic spline interpolation compensation, the channel signals are aligned to the same time reference, and the alignment error is less than 0.1 ns, thereby eliminating the minute clock offset (usually at the ns level) caused by hardware differences between multiple channels and ensuring the spatio-temporal consistency of subsequent time-frequency analysis.

[0025] Further, a time-varying Gaussian modulation cosine window shown in the following formula is constructed to balance the time domain resolution and the frequency domain localization accuracy in a strong noise environment, thereby dynamically adjusting the time-frequency resolution.

[0026] Among them, is used to represent the time width adjustment coefficient, and the window is reduced when the signal change rate is large to capture transient characteristics. is used to represent the window shape correction coefficient to compensate for the frequency offset caused by the laser wavelength drift. For representing the frequency modulation factor and matching the spatial frequency of the interference fringes.

[0027] Thus, an adaptive window is applied to the time-aligned signal frame by frame to segment it into overlapping frames (e.g., frame length of 256 points and overlap rate of 75%). Frame windowing can be expressed by the following formula: where, is the channel number (1 to N), is the frame index M / H, H = frame shift = 64 points), is the in-frame sampling point index (0 to 255), is the sampling interval (1 μs).

[0028] Furthermore, the short-time Fourier transform calculation of each frame signal is performed according to the following formula to generate a time-frequency matrix. Thus, the time-domain signal is converted into the time-frequency domain energy distribution to identify the instantaneous frequency components of the interference fringes.

[0029] where f is the frequency index ( , corresponding to ), and j is the imaginary unit. Through the dynamically adjusted window function, the time-frequency characteristics of the interference signal are accurately extracted under complex working conditions (such as rapid movement or noise interference).

[0030] Finally, the time-frequency matrices of each channel are stacked along the channel dimension (n = 1 to N) to form a three-dimensional complex tensor (i.e., the original tensor). The three-dimensional complex tensor is as follows: where n is the spatial dimension (i.e., the channel number), m is the time dimension (i.e., the frame index), and f is the frequency dimension (i.e., the frequency band number). And the amplitude of each frequency band in the three-dimensional complex tensor is normalized between channels to eliminate the difference in detector sensitivity, ensure the comparability of the time-frequency energy of different channels, and avoid subsequent analysis deviation.

[0031] In an alternative embodiment, the standard deviation of the phase difference between channels is calculated to mark the abnormal channels, thereby facilitating the identification of abnormal channels caused by detector failures or optical misalignments, and facilitating their elimination in subsequent steps. The formula for the phase consistency index is as follows: where arg() is to take the complex phase angle, is the average phase between channels. By constructing a unified multi-dimensional data representation, it provides a structured input for subsequent processing such as noise suppression and phase unwrapping.

[0032] Step S2. Perform joint optimization processing on the original tensor to obtain a denoised tensor.

[0033] For the original tensor ( = number of channels, = number of time frames, K = number of frequency bands), perform dimension decomposition rank selection to obtain the spatial, time domain, and frequency domain dimension ranks as shown below: Channel dimension rank R = [N / 4] (for example, when N = 64, R = 16); Time dimension rank S = [M / 4] (for example, when M = 1024, S = 256); Frequency dimension rank T = [K / 4] (for example, when K = 128, T = 32).

[0034] Furthermore, perform Tucker decomposition using the following high-order singular value decomposition: where is the core tensor, used to describe the interaction strength of each dimension. is the channel factor matrix, and the column vectors are the channel mode bases. is the time factor matrix, and the column vectors are the time dynamic bases. is the frequency factor matrix, and the column vectors are the spectral distribution bases. Extract the structured components in the signal through low-rank approximation (i.e., the initial low-rank signal, where the initial low-rank signal is ). Initially separate the low-rank effective signal (e.g., interference fringes) and sparse noise (e.g., random interference) in the original tensor through Tucker decomposition.

[0035] After obtaining the initial low-rank signal, update the sparse noise (i.e., the initial noise signal) through the following soft threshold function: where is the residual tensor after removing the low-rank part. sign(x) is the sign function, which retains the positive and negative nature of the data. max(a, b) is to take the larger value of a and b. is the threshold parameter, which controls the noise sparsity (typical value = 0.1 × maximum residual amplitude). By suppressing the tiny fluctuations (noise) in the residual, retain the significant interference components (i.e., the initial noise signal).

[0036] It should be understood that the larger it is, the more noise components are set to zero (sparsity increases, but weak signals may be lost); the smaller it is, the more details are retained (sparsity decreases, but noise residue increases). represents threshold shrinkage of When , output 0, which is used to indicate that the noise is suppressed; when , output , which is used to indicate that significant noise components are retained. During the iterative process of joint optimization the following update strategy can be adopted: Preliminarily separate the sparse noise through threshold shrinkage to provide a starting point for iterative optimization.

[0037] The iterative process of joint optimization is as follows: First, calculate the denoising residual tensor through the following formula: where, is the denoising residual tensor, is the original tensor, is the initial noise signal.

[0038] Secondly, perform a new round of Tucker decomposition on the denoising residual to optimize the core tensor and factor matrices. The Tucker decomposition operation can be expressed as the following formula: where, the initial decomposition ranks R, S, T are retained, but the factor matrices and the core tensor are allowed to be recalculated.

[0039] Thus, the updated initial low-rank signal is obtained. Gradually optimize the low-rank signal estimation through iterative decomposition to reduce noise pollution.

[0040] After each iteration is completed, calculate the total error of the current low-rank signal, noise signal, and original tensor according to the original tensor, initial noise signal, and updated initial low-rank signal through the following formula.

[0041] where, is the Frobenius norm, that is, the square root of the sum of the squares of the tensor elements is the reconstruction error after the first iteration. Thus, judge whether to complete the iterative operation through the preset error threshold , if , terminate the iteration and output as the denoising tensor; otherwise, continue the iteration. Among them, the error threshold is set according to the data scale and noise level (for example, ).

[0042] In an alternative embodiment, the threshold parameter Update is performed using the dynamic adjustment method shown below: Through the dynamic adjustment method, the value gradually decreases as the error decreases, enabling refined noise separation.

[0043] Step S3: Perform phase unwrapping on the denoised tensor to obtain an absolute phase matrix.

[0044] Divide the denoised tensor into spatiotemporal units, where each unit corresponds to the full-band data of a specific channel and time frame. Label the spatiotemporal units through the preset channel number and time frame number for distinction between each spatiotemporal unit. Among them, is the channel number (for example, in a 64-channel interferometer, n = 1~64); is the time frame number (for example, 1024 time frames). Perform frequency-domain amplitude maximum detection on each spatiotemporal unit using the formula shown below: where where, is the frequency band index K, such as K = 128) is the complex phase angle (unit: radian), is the frequency band index with the maximum amplitude.

[0045] Select the frequency component with the strongest energy as the main frequency to avoid noise interference. Arrange the main frequency phase values of all spatiotemporal units in the channel and time dimensions to generate an original phase vector. Extract the most significant phase information from the denoised signal to provide a data basis for subsequent unwrapping.

[0046] Furthermore, according to the three preset laser wavelengths (in the embodiments of the present application, nm, nm, nm), form a parameter group . Combine the wavelengths within the parameter group in pairs, and generate an equivalent composite wavelength for the wavelengths in the combination through the following formula.

[0047] where, is the composite wavelength of nm and 633 nm, approximately 3.2 μm; The synthetic wavelength of nm and 780 nm is approximately 4.7 μm. The unambiguous measurement range is extended through the long synthetic wavelength. Thus, the original phase vectors are weighted and fused according to the wavelength ratio, and the weighted fusion process can be expressed by the following formula: where, The original phase value corresponding to the i-th wavelength. By performing ratio-weighted fusion of the synthetic wavelength on the original phase vectors, the 2π ambiguity problem in single-wavelength measurement is eliminated. The obtained synthetic wavelength parameters break through the periodic limitation of single-wavelength measurement and achieve large-range absolute distance calculation.

[0048] Furthermore, the phase difference between adjacent channels is calculated along the spatial dimension (i.e., the channel direction) through the following formula: Meanwhile, the phase change rate is calculated along the time dimension through the following formula: where, Time frame interval (e.g., 1 ms) Finally, the spatial difference and the time derivative are combined through the following formula to generate the unwrapped phase matrix: where, is the unwrapped absolute phase matrix. Phase jumps are eliminated through spatio-temporal gradient integration, and a continuous phase distribution is initially generated.

[0049] Meanwhile, temperature, air pressure, and humidity data are obtained in real time by the sensor, and the air refractive index is calculated based on the following modified Edlén equation: where, is the ambient temperature, atmospheric pressure, is the relative humidity.

[0050] Furthermore, through refractive index calculation, the influence of air refractive index change on the optical path is clarified, and thus the phase value is adjusted according to the refractive index change. The compensation for the influence of air refractive index change on light can be expressed by the following formula: where, Reference refractive index (i.e., the value under the calibration environment, e.g., 1.000273), is the refractive index calculated in real time, is the refractive index compensation phase matrix. Measurement drift caused by temperature and humidity changes is eliminated through phase compensation, and the absolute accuracy is improved.

[0051] To eliminate the phase discontinuity caused by residual noise, noise reduction compensation is performed on the refractive index compensation phase matrix through gradient and filtering constraints. First, the second-order gradient of the phase difference between adjacent channels is calculated for spatial consistency verification to suppress mutations. The process of spatial gradient constraint can be expressed by the following formula: where, Smoothing coefficient, controlling the gradient suppression intensity.

[0052] Furthermore, adaptive weighted median filtering is performed on the time series of each channel through the following formula: The noise reduction compensation phase matrix is obtained through time-domain filtering.

[0053] To further solve the problem of residual phase ambiguity under complex field bursts, the present application adopts a preset phase unwrapping depth network model. Among them, the phase unwrapping depth network model is a pre-trained U-Net type network. The U-Net type network includes an encoder composed of 5 layers of convolution, a decoder composed of 5 layers of transposed convolution, and a skip connection arranged between the encoder and the decoder. The phase unwrapping depth network model receives an input tensor spliced by the noise reduction compensation phase matrix, the environmental parameter set, and the synthetic wavelength parameter. The encoder extracts multi-scale phase features from the input tensor, the decoder gradually restores the high-resolution phase, and finally the low-level details and high-level semantic information are fused through the skip connection. The absolute phase matrix shown below is generated through the forward propagation of the phase unwrapping depth network model: Step S4: Perform dynamic range expansion on the absolute phase matrix to obtain an enhanced phase matrix.

[0054] The first-order difference of the phase difference between adjacent channels is calculated along the channel dimension (i.e., the spatial axis) through the following formula to capture the spatial change rate of surface deformation: where, Channel number, Time frame number Spatial gradient operator. The first-order difference is used to detect the phase mutation caused by alignment error or local deformation between channels.

[0055] At the same time, the second-order derivative of the phase change rate is calculated along the time dimension through the following formula to identify dynamic vibration characteristics: where, Sampling time interval (e.g., 1 ms), is the second-order time derivative operator. By taking the second-order time derivative, the vibration acceleration is quantified to distinguish steady-state and transient motions.

[0056] Finally, the obtained spatial gradient and time gradient are combined into a three-dimensional tensor Among them, the first two channels in the phase gradient tensor are spatial gradient components, and the third channel in the phase gradient tensor is the time gradient component. Through spatio-temporal gradient analysis, the local characteristics and dynamic properties of phase changes are quantified, providing a physical basis for parameter adjustment.

[0057] The temperature, air pressure, and humidity data collected in real time in step S3 are normalized to zero mean and unit variance by the following formula: where are the mean and standard deviation of the historical data, respectively.

[0058] Furthermore, the wavelength drift of the laser is estimated according to the temperature change by the following formula: where is the nominal wavelength (e.g., 532 nm), is the thermal expansion coefficient (e.g., 11×10×-6 / °C), is the reference temperature (e.g., 25°C).

[0059] Finally, the gradient tensor, environmental parameters, and wavelength drift amount are fused into a multi-dimensional vector by the following formula: Among them, each state vector contains gradient components (3D) + environmental parameters (3D) + wavelength drift (14D) = 7D.

[0060] By splicing the state vectors, the comprehensive environment, device characteristics, and signal characteristics are combined to comprehensively describe the dynamic state of the system.

[0061] To balance noise suppression and signal fidelity and achieve real-time optimization of detector parameters according to the system state. In the embodiments of the present application, a preset deep reinforcement learning network model is used to adaptively adjust the parameters of the multi-dimensional state vector to obtain a dynamic control parameter set. It should be understood that the deep reinforcement learning network model uses a deep deterministic policy gradient network. The deep deterministic policy gradient network includes an Actor network and a Critic network. Among them, the Actor network can output continuous control parameters according to the input state vector. The Critic network is used to evaluate the Q value of the state-action pair. The deep reinforcement learning network model outputs three-dimensional control parameters according to the input 7-dimensional state vector (i.e., the multi-dimensional state vector). (i.e., gain, integration time, cut-off frequency).

[0062] At the same time, the deep reinforcement learning network model updates the network weights according to the real-time interaction data, and the policy gradient update can be expressed by the following formula: Wherein, are the Actor network parameters, are the Critic network parameters, is the state-action value function.

[0063] The Actor network generates dynamic control parameters through forward propagation by the following formula: Wherein, the dynamic control parameter set .

[0064] To suppress out-of-band noise and enhance the effective signal components, the absolute phase matrix is dynamically adjusted according to the dynamic control parameter set.

[0065] First, the absolute phase matrix is adaptively amplified by the following formula: Wherein, the weak signal region (i.e., ( ; the strong signal region .

[0066] Furthermore, the time-domain integration length is adjusted according to the integration time parameter by the following formula: At the same time, a variable cut-off frequency filter is designed by the following formula to retain the effective frequency band: Wherein, the band-pass filtering formula is as follows: The filter response can be expressed by the following formula: Among them, is the passband width.

[0067] While retaining the details of weak signals, it is necessary to prevent the strong signals in the frequency-domain band-pass filtering phase matrix from being oversaturated. The high-dynamic-range phase values in the frequency-domain band-pass filtering phase matrix are nonlinearly compressed by the following formula: Among them, is the saturation voltage (for example, 1V), which controls the compression intensity. Hyperbolic tangent saturation suppression of the high dynamic range is performed through the above hyperbolic tangent compression calculation.

[0068] It should be understood that the phase signal may cover a very large dynamic range during the measurement process. For example, for weak signals (such as long-distance detection or low-reflectivity surfaces), the phase values are very small (close to the noise level); for strong signals (such as high-reflection targets at close range), the phase values may reach the saturation threshold. If uniform quantization is directly used (that is, the same number of bits is used in all regions), weak signals will be drowned out by noise due to insufficient resolution, resulting in loss of details, and strong signals will occupy too much storage resources and may introduce computational redundancy. Therefore, it is necessary to set a quantization table to dynamically allocate quantization bits according to the local signal-to-noise ratio, balance the precision requirements of different regions, and achieve high precision for weak signals and high efficiency for strong signals.

[0069] For each spatio-temporal unit the local signal-to-noise ratio of the phase data is calculated through the following formula: Among them, is the square of the current phase value, is the variance of the phase difference of adjacent spatio-temporal units. According to the local signal-to-noise ratio, the present application adopts the following quantization table for dynamic allocation of quantization bits: It should be understood that the design of the quantization table is based on the dynamic evaluation of the signal quality (SNR). In the high SNR region (SNR ≥ 40dB), the signal quality is high and the influence of noise is small, and high-bit quantization (such as 12bit) is used to retain more phase details; in the medium SNR region (20dB ≤ SNR < 40dB), the signal and noise coexist, and medium-bit quantization (such as 8bit) is used to balance precision and noise suppression; in the low SNR region (SNR < 20dB), the noise dominates, and low-bit quantization (such as 6bit) is used to reduce the interference of noise on the quantization result.

[0070] Furthermore, the following quantization operation is performed according to the quantization table to obtain the compressed mapping phase matrix: By replicating the phase gradient tensor along the channel dimension to match the size of the phase matrix, the dimension of the gradient tensor is expanded. Then, the compressed phase matrix and the gradient tensor are concatenated along the feature dimension for multimodal data concatenation to obtain an enhanced phase matrix. The tensor concatenation operation can be expressed by the following formula: Among them, is , is the compressed mapped phase matrix, is the phase gradient tensor. The output has a dimension of (i.e., 1 phase channel plus 3 gradient channels).

[0071] Step S5: Perform multi-wavelength resolution and environmental compensation on the enhanced phase matrix to obtain a distance matrix.

[0072] To achieve large-range absolute distance measurement, an equivalent "virtual ruler" is constructed using wavelength differences. According to the preset three groups of wavelength laser parameters, phase difference synthesis calculations are performed on the enhanced phase matrix to generate synthetic wavelength parameters. Since the process of multi-wavelength phase difference synthesis calculation is the same as the equivalent synthetic wavelength generation process in step S3, it will not be elaborated here. For details, please refer to the equivalent synthetic wavelength generation process in step S3.

[0073] To balance millimeter-level measurement range and nanometer-level resolution, wavelength grading is used to achieve distance resolution across orders of magnitude.

[0074] First, the synthetic wavelength parameters are divided into different wavelength combinations through the following formula to generate multi-level unambiguous intervals: Among them, is the first-level synthetic wavelength (3.2μm measurement range), is the second-level synthetic wavelength (15.6μm measurement range).

[0075] Secondly, according to the following formula, the distance interval levels are defined based on the synthetic wavelength to obtain multi-level unambiguous distance intervals: Among them, the obtained multi-level unambiguous distance intervals .

[0076] To eliminate the optical path drift caused by changes in air temperature and humidity and improve the absolute measurement accuracy. First, based on the real-time collected environmental parameters, the air refractive index is calculated using the modified Edlén equation. Since the calculation process of the air refractive index is the same as that in step S3 when calculating the air refractive index using the Edlén equation, it will not be elaborated further here. For details, please refer to the calculation process of calculating the air refractive index using the Edlén equation in step S3. Further, the actual optical path is adjusted according to the calculated refractive index change to obtain the refractive index correction distance interval. The refractive index compensation formula is as follows: Where, is the reference refractive index (i.e., the calibrated environmental value, such as 1.000273), is the real-time refractive index. The obtained refractive index correction distance interval .

[0077] To separate the systematic error and random noise, so as to achieve targeted compensation. By combining the real-time measured laser wavelength and historical data through the following formula, the Kalman filter is used to estimate the laser wavelength drift amount (i.e., the time dynamic error): Where, is the wavelength drift amount at the k-th iteration, which is used to represent the deviation between the actual wavelength of the laser and the nominal value is the state transition coefficient (i.e., dimensionless), usually taking , indicating the time correlation of the drift amount (for example, A = 0.98 means that the current drift amount is affected by the previous moment with a weight of 98%). is the process noise, which follows a Gaussian distribution , reflecting the random drift caused by factors such as temperature fluctuations. is the observed value, the actual wavelength value measured by a wavelength meter or an interferometer. is the observation matrix, usually H = 1 (directly observing the drift amount). is the observation noise, which follows a Gaussian distribution , used to reflect the precision error of the measuring device.

[0078] At the same time, through the following geometric distortion calculation formula, the mean absolute deviation of the distance difference between adjacent channels in the refractive index correction distance interval is calculated to reflect the spatial distribution of the alignment error (i.e., the geometric distortion compensation matrix). The geometric distortion compensation matrix reflects the difference between channels caused by imperfect optical alignment (i.e., the spatial static error). The geometric distortion calculation can be expressed as the following formula: Where, is the geometric distortion compensation coefficient at time m for channel n, which is used to reflect the inconsistency of distance measurement caused by alignment errors between multiple channels. is the refractive index corrected distance at time m for channel i. N is the total number of channels, for example, N = 64.

[0079] To integrate multiple factors such as environmental compensation and error correction, a unified solution framework is constructed. Tensor expansion is performed through the following formula to make the corrected distance interval, wavelength stability correction factor, and geometric distortion compensation matrix to the same dimension.

[0080] Among them, is the geometric distortion compensation matrix. is the wavelength stability correction factor, which is generated according to the wavelength drift amount through a preset non-linear model. Further, the multi-dimensional data is fused along the feature dimension through the following formula to obtain the initial distance tensor: Among them, is the expanded refractive index corrected distance interval tensor (dimension N×M×2), which is generated by copying the original data to the third dimension. is the expanded wavelength stability correction factor (dimension N×M), and the same correction factor is copied for each channel. C(n,m) is the geometric distortion compensation matrix (dimension N×M). : represents concatenation along the third dimension, and the final tensor dimension is (i.e., 2 distance levels + 1 wavelength correction + 1 geometric compensation + 1 reserved dimension).

[0081] By performing residual compensation iterative operations on the obtained initial distance tensor, systematic errors and random noises are eliminated through joint optimization to achieve sub-nanometer accuracy. In the embodiments of the present application, the alternating direction multiplier method (ADMM) is used for residual compensation iterative optimization. ADMM is an iterative algorithm for solving constrained optimization problems, and its core is to decompose the original problem into multiple sub-problems and solve them alternately. ADMM can be expressed as the following formula: Among them, is the nuclear norm, which enforces to be low-rank, and is used to characterize the global correlation of systematic errors. is the L1 norm, which enforces to be sparse, and is used to characterize the local characteristics of random noises. is the regularization coefficient, which is used to balance the weights of the low-rank and sparse terms, and usually takes .

[0082] The specific process of the residual compensation iterative operation is as follows: Initialization: Set the penalty parameter , the maximum number of iterations (e.g., 100) and the tolerance error (e.g., 1e-6). And set the initial low-rank term as , the initial sparse term as , and the initial Lagrange multiplier as . Among them, performs Tucker decomposition on the initial distance tensor through the following formula to obtain: .

[0083] Iterative update: Update the low-rank term through the singular value threshold shown below. The singular value threshold performs singular value decomposition on the matrix expansion and truncates the small singular values.

[0084] Update through the soft threshold function shown below: sign Among them, controls the constraint satisfaction speed. A larger value (e.g., 10) accelerates convergence but may oscillate, while a smaller value (e.g., 1) has high stability but slow convergence. determines the noise suppression intensity and needs to be adjusted according to the signal-to-noise ratio (e.g., λ is smaller at high SNR, retaining more details).

[0085] At the same time, update sub through the formula shown below: Among them controls the constraint satisfaction speed.

[0086] After each iterative update, calculate the residual through the following formula, and judge whether the residual compensation iterative operation is completed by comparing the residual with the tolerance error.

[0087] When or reaches , terminate the iteration; otherwise continue. After the iteration terminates, use the low-rank term as the distance matrix.

[0088] Step S6: Perform vibration compensation on the distance matrix to obtain the piezoelectric drive signal.

[0089] The distance matrix is segmented into overlapping frames (frame length 256 points, overlap rate 75%) in the time dimension to form a three-dimensional data block. And a Hanning window as shown below is applied to each frame of data to reduce spectral leakage, thereby smoothing the frame edges and suppressing the subsequent sidelobe effect of the fast Fourier transform.

[0090] Furthermore, the fast Fourier transform as shown below is performed on each frame of data to extract frequency-domain features: where, is the frequency index (i.e., , corresponding to Hz vibration frequency band), is the imaginary unit.

[0091] Finally, by combining the amplitude, phase, and harmonic distortion rate into a vibration spectrum tensor, the frequency distribution, phase relationship, and harmonic distortion characteristics of the vibration energy are quantified, providing the required input data for feedforward compensation. The construction of the spectrum tensor can be expressed by the formula as shown below: The obtained vibration spectrum tensor .

[0092] Furthermore, the inverse system response processing is performed on the vibration spectrum tensor through the inverse transfer function to cancel the vibration energy and achieve predistortion compensation. It should be understood that the frequency response function of the piezoelectric actuator measured by the sweep frequency method is as shown below: where, is the static gain, , is the natural frequency.

[0093] Therefore, the inverse transfer function of the inverse filter designed according to the frequency response function to cancel the system response is as shown below: Furthermore, by multiplying the vibration spectrum by the inverse function, frequency-domain compensation can be achieved to generate an initial feedforward compensation signal. The frequency-domain compensation can be expressed by the formula as shown below: where, is the vibration spectrum, is the inverse function, is the initial feedforward compensation signal.

[0094] It should be understood that the initial feedforward compensation signal obtained after frequency-domain compensation is a signal represented in the frequency domain, while the actual control system (e.g., driving a piezoelectric actuator) requires the input to be a time-domain signal, that is, there is a corresponding compensation amount at each moment. Therefore, the complex compensation signal obtained in the frequency domain must be converted into a time-domain signal. The inverse FFT (IFFT) is exactly the mathematical tool for converting a frequency-domain signal into a time-domain signal. It reconstructs a time-related signal based on the amplitude and phase information of each frequency component in the frequency domain, so that the compensation signal can be directly used for physical control and driving. The inverse FFT conversion formula is as follows: where the obtained feedforward compensation signal .

[0095] At the same time, by performing spatio-temporal differential processing on the distance matrix data, vector information representing vibration errors is obtained, so as to provide quantified input data for the hybrid controller simulation model. By successively performing spatial gradient calculation, time derivative calculation, and error vector synthesis on the distance matrix data, error information generated due to environmental vibration, geometric distortion, or dynamic changes is extracted and quantified into a numerical vector, thereby providing a basis for subsequent correction for the hybrid controller.

[0096] Specifically, first, a spatial gradient calculation is performed on the distance matrix. The second-order difference between adjacent channel data is calculated along the spatial (i.e., channel) dimension to detect spatial distortion caused by alignment or geometric arrangement errors between multiple channels. Specifically, the following formula is used to calculate the spatial second-order gradient: where represents the distance measurement value at the th channel and at the th moment; and respectively represent the measurement values at the same moment of the previous and the next channels adjacent to the th channel; is the discrete approximation of the second-order derivative in space, and its value reflects the curvature or bending of the spatial data. The spatial non-uniformity caused by uneven equipment arrangement or external interference is detected through the spatial second-order difference calculation, and information on local geometric changes is provided for subsequent feedback adjustment.

[0097] Furthermore, the derivative of the distance matrix is calculated along the time dimension to extract velocity and acceleration information reflecting dynamic changes. Time derivative processing can reveal the instantaneous change rate and trend of the measured values during vibration, thus reflecting the dynamic characteristics during the vibration process. Time differential processing is divided into two parts: velocity calculation and acceleration calculation. The central difference method is used to obtain a more accurate discrete approximation. The specific formulas are as follows: where, denotes the instantaneous velocity at the -th channel and the -th moment; denotes the instantaneous acceleration under the same conditions; and respectively represent the distance values of the sampling points immediately before and after the current moment; denotes the time interval between adjacent sampling points; the denominator is used for the central difference method to calculate the first-order derivative, while the denominator is used for calculating the second-order derivative. Through time differential processing, the dynamic behavior during vibration is quantitatively described to obtain velocity and acceleration information. These two types of information can reflect the response of the system when instantaneously affected by external disturbances, providing a key parameter basis for subsequent feedback control design.

[0098] Finally, the obtained spatial gradient, velocity, and acceleration information are synthesized into a comprehensive error vector, which comprehensively expresses the error information caused by vibration, geometric deformation, and dynamic deviation. The process of synthesizing the error vector uses the vector splicing method to combine the processing results of the first two parts into a multi-dimensional vector, and each component corresponds to the spatial, velocity, and acceleration errors respectively. The specific formula description is as follows: where, e denotes the error vector formed at the -th channel and the -th moment. The construction of the error vector is used to comprehensively describe the error information generated by factors such as vibration at each spatio-temporal point. These information will serve as the basis for the feedback signal in the hybrid controller, enabling the control system to specifically correct the vibration deviation, thereby improving the response speed and accuracy of the overall system.

[0099] Based on the obtained feedforward compensation signal and the real-time vibration error vector generated by spatio-temporal differentiation, a feedback control signal is calculated through a hybrid control strategy to perform real-time correction on the dynamic vibration of the system. First, the input data is decomposed, where the input data includes the feedforward compensation signal and the real-time vibration error vector. The input data respectively reflects the estimated part of the system vibration and the dynamic error information measured in real time. The preset hybrid controller simulation model aims to organically combine these two parts of information to generate a comprehensive feedback signal, which can effectively cancel the vibration influence in a short time and ensure the stability of the system response.

[0100] It should be understood that the hybrid controller simulation model in the embodiments of the present application adopts the classical proportional-integral-differential (PID) control method and adds a feedforward compensation term. The specific hybrid control law can be expressed as the following formula: Wherein, represents the feedback control signal at the channel and the moment, and are respectively the proportional, integral and differential control gain coefficients, all of which are preset constant parameters; is defined as the proportional error signal, which is the second-order spatial difference is the integral error, where e i uses the velocity component v for accumulation, is the sampling time interval, which is used to scale-normalize the integral result; is the differential error, usually using the acceleration component is the feedforward weight coefficient, and the value is fixed at 0.8, which is used to balance the contribution of the feedforward signal to the overall control output. By weighted combination of each component in the error vector, it not only reflects the instantaneous deviation of the current system (i.e., the proportional term), the historical cumulative deviation (i.e., the integral term) and the future trend (i.e., the differential term), but also combines the pre-compensated feedforward information, so as to achieve comprehensive compensation for vibration disturbances.

[0101] To prevent the integral action from causing overshoot or oscillation when the error is too large, conditional control is performed on the integral gain. The integral separation method determines whether to apply the integral action by judging the error amplitude, so as to limit the unstable factors brought by the integral action. The specific integral separation condition is as follows: Wherein, represents at the channel and the The absolute value of the proportional error at a moment. When the absolute value of the proportional error is less than 0.1, the integral gain is set to 0.5; when the proportional error is large, the integral action is not taken into account, so it is set to 0. By judging the magnitude of the current proportional error, integral compensation is adopted when the error is small, avoiding introducing additional unstable factors during drastic error fluctuations, and ensuring that the overall control output is more stable and reliable.

[0102] Aiming at the problem that the stiffness of the actuator in the control system changes due to environmental factors such as temperature and humidity, by constructing an environmental stiffness model and performing gain scheduling, the proportional and differential gains in the feedback control signal are adjusted, so that when the environmental parameters change, the control system can automatically correct the compensation strategy, thereby ensuring the stability and accuracy of the overall system response.

[0103] First, obtain the relationship between the environmental temperature and the actuator stiffness through experimental data or calibration. Temperature changes will cause the attenuation or enhancement of the stiffness of piezoelectric ceramics or other actuators. Therefore, a stiffness attenuation model is constructed, and the stiffness attenuation model can be expressed by the following formula: where, represents the actuator stiffness at temperature ; is the stiffness at room temperature (i.e., the reference temperature ), and the unit is usually N / μm; is the temperature coefficient, and its unit is , which is used to describe the attenuation rate of the stiffness caused by temperature changes; is the base of the natural logarithm, and the exponential function is used to express the exponential attenuation effect of temperature changes on the stiffness. By establishing the mathematical relationship between temperature and actuator stiffness, it is used to quantify the impact of environmental changes on the control system parameters, providing a numerical basis for subsequent gain scheduling.

[0104] Based on the value calculated according to the stiffness model, the adaptive adjustment of the feedback control parameters is carried out. Specifically, according to the stiffness change situation, the proportional gain and differential gain in the controller are scheduled to ensure that when the environmental temperature changes, the feedback signal output by the control system can be corrected accordingly. The gain scheduling rule is shown in the following formula: where, represents the adjusted proportional gain; represents the adjusted differential gain; and are the original proportional and differential control gains, usually fixed values; the fraction It represents the ratio between the current stiffness and the stiffness at room temperature, and this ratio reflects the relative influence of environmental temperature on the actuator stiffness. The change in environmental temperature is mapped through the stiffness model to the adjustment value of the feedback control parameter, enabling the control system to maintain an appropriate control force and response speed under different temperature conditions, thereby ensuring the stability and accuracy of the system output.

[0105] The optimized feedback control signal after environmental stiffness parameter gain scheduling is converted into a discrete Pulse-Width Modulation (PWM) waveform. The PWM conversion process mainly includes two sub-steps: level quantization and duty cycle calculation, which convert the continuously varying control signal into a digital signal that can directly drive the hardware circuit while retaining the amplitude information reflecting vibration compensation in the signal, thus ensuring high response accuracy when the control system is actually executed.

[0106] First, level quantization processing is performed on the continuous control signal. The process of level quantization is to discretize the amplitude of the continuous signal into several fixed level values, and 7 levels are selected here. The following piecewise function is specifically used to describe each level interval, and signals with different amplitude ranges are respectively mapped to the predetermined voltage values.

[0107] = Among them, is the feedback control signal after stiffness parameter scheduling at time t. Discretizing the continuous signal into a fixed-level PWM waveform facilitates subsequent digital signal processing and matching with the hardware drive interface, ensuring that the output signal numerically meets the hardware control requirements.

[0108] Furthermore, duty cycle calculation is performed on the discretized signal. The duty cycle is an index describing the ratio of the high-level duration to the total cycle time in the PWM waveform, used to determine the width of each PWM pulse, thereby converting the amplitude information of the continuous control signal into the pulse width information of the PWM waveform. The duty cycle formula is as follows: Among them, represents at the channel, at the moment, the duty cycle, with the unit of percentage; represents the absolute value of the feedback control signal, ensuring that the duty cycle is non-negative; the denominator "3V" is the maximum voltage amplitude, corresponding to the highest voltage value in level quantization. Converting the optimized continuous control signal into the duty cycle data corresponding to the PWM waveform provides a numerical basis for generating the specific PWM signal subsequently.

[0109] The PWM signal obtained by pulse-width modulation encoding is subjected to waveform shaping to correct the distortion caused by the nonlinearity and hysteresis characteristics of the drive circuit, thereby generating the final drive signal suitable for the piezoelectric actuator. This process is divided into two parts: First, a drive nonlinear model is established for distortion modeling, and then a pre-trained deep learning model is used for waveform correction to ensure that the final output signal meets the requirements of the piezoelectric actuator in terms of amplitude and time-domain shape. (That is, the waveform shaping model includes a nonlinear model and a deep learning model). Specifically, first, the response curve of the piezoelectric drive system is obtained through experimental measurement, and a nonlinear hysteresis model is constructed based on this data to describe the nonlinear relationship between the input signal and the output signal. This model takes into account the hysteresis phenomenon existing in the piezoelectric material when the input voltage changes, as well as the nonlinear amplitude response caused by physical characteristics. The hysteresis model can be expressed by the following formula: where represents the actual voltage signal output by the piezoelectric actuator; represents the voltage of the PWM signal input to the actuator; is the hyperbolic tangent function used to describe the nonlinear saturation effect; is the gain coefficient used to adjust the output amplitude; is the nonlinear adjustment parameter used to control the slope of the hyperbolic tangent function; is the differential gain parameter used to describe the influence of the input signal change rate on the output signal; represents the time derivative of the input signal, reflecting the signal change rate. The nonlinear hysteresis model describes the nonlinear relationship between the input and output signals due to the physical characteristics of the piezoelectric material itself and the hysteresis effect in the drive circuit. This model provides a theoretical basis for explaining why waveform distortion occurs after the input PWM signal is directly driven and subsequent shaping processing is required.

[0110] Furthermore, in order to correct the nonlinear distortion that cannot be completely eliminated in the above model, a deep learning method is used to perform waveform shaping. Using a pre-trained convolutional neural network model, the signal is nonlinearly corrected according to the input PWM signal, the current ambient temperature, and the vibration spectrum information, and a drive signal matching the actual piezoelectric drive requirements is output. The forward propagation process of the deep learning model can be expressed by the following formula: where represents the piezoelectric drive signal after shaping; represents the forward propagation function of the pre-trained convolutional neural network model, which is used to perform nonlinear mapping on the input data; is a discrete PWM signal after pulse-width modulation encoding; represents the current ambient temperature parameter, which is used to describe the influence of temperature on the driver response; is a vibration spectrum tensor that provides frequency-domain information related to vibration. The deep learning model corrects the PWM signal, uses the ambient temperature and vibration spectrum information to assist in judgment, and realizes the compensation of nonlinear distortion, thereby generating a final voltage signal that meets the requirements of the piezoelectric driver. This method can overcome the limitations of traditional models in dealing with complex nonlinear characteristics and ensure that the output signal has higher linearity and stability in form.

[0111] This application is applied to the field of laser control technology. By real-time acquiring channel signals of multiple target channels according to a plurality of preset photodetectors, performing three-dimensional sampling in the time domain-frequency domain-space on the channel signals to construct an original tensor, performing joint optimization processing on the original tensor to obtain a denoised tensor, performing phase unwrapping on the denoised tensor to obtain an absolute phase matrix, performing dynamic range expansion on the absolute phase matrix to obtain an enhanced phase matrix, performing multi-wavelength resolution and environmental compensation on the enhanced phase matrix to obtain a distance matrix, and performing vibration compensation on the distance matrix to obtain a piezoelectric drive signal. The channel laser interference control method of this application realizes the comprehensive improvement of laser interference measurement signals through a series of advanced algorithms and processing means such as multi-dimensional signal acquisition, joint denoising optimization, precise phase unwrapping, dynamic range expansion, and vibration compensation.

[0112] As Figure 2 shown, it is a functional module diagram of a high-dynamic-range multi-channel laser interference control device provided by an embodiment of this application.

[0113] In some embodiments, the high-dynamic-range multi-channel laser interference control device 2 may include a plurality of functional modules composed of computer program segments. The computer programs of each program segment in the high-dynamic-range multi-channel laser interference control device 2 can be stored in the memory of the server and executed by at least one processor to execute (see details in Figure 1 description) the functions of the high-dynamic-range multi-channel laser interference control method.

[0114] In this embodiment, according to the functions it executes, the high-dynamic-range multi-channel laser interference control device 2 can be divided into a plurality of functional modules. The functional modules may include: a feature acquisition module 21, a denoising optimization module 22, a phase unwrapping module 23, a dynamic expansion module 24, a distance compensation module 25, and a vibration compensation module 26. The module referred to in the present invention means a series of computer program segments that can be executed by at least one processor and can complete fixed functions, and are stored in the memory. In this embodiment, the functions of each module will be detailed in subsequent embodiments.

[0115] The feature acquisition module 21 is configured to obtain channel signals of multiple target channels in real time according to a plurality of preset photodetectors, and perform three-dimensional sampling in the time domain-frequency domain-space on the channel signals to construct an original tensor.

[0116] In an optional embodiment, the feature acquisition module 21 is specifically configured to: Perform time domain synchronization processing on the channel signals to obtain a time-aligned optimized channel signal set; Perform short-time Fourier transform processing on the time-aligned optimized channel signal set according to a preset adaptive window to obtain a time-frequency matrix of each target channel; Perform three-dimensional fusion module construction processing on the time-frequency matrix to obtain the original tensor.

[0117] The noise reduction and optimization module 22 is configured to perform joint optimization processing on the original tensor to obtain a noise reduction tensor.

[0118] In an optional embodiment, the noise reduction and optimization module 22 is specifically configured to: Step S21, perform Tucker decomposition processing on the original tensor to obtain an initial low-rank signal; Step S22, perform calculation processing of a preset soft threshold function on the initial low-rank signal to obtain an initial noise signal; Step S23, calculate a denoising residual tensor according to the original tensor and the initial noise signal, and perform Tucker decomposition processing on the denoising residual tensor to update the initial low-rank signal; Step S24, calculate a reconstruction error according to the original tensor, the initial noise signal, and the updated initial low-rank signal to obtain a reconstruction error value, and compare the reconstruction error value with a preset error threshold; Repeat the execution of step S22 to step S24 until the reconstruction error value is less than the error threshold, and determine the updated initial low-rank signal as the noise reduction tensor.

[0119] The phase unwrapping module 23 is configured to perform phase unwrapping on the noise reduction tensor to obtain an absolute phase matrix.

[0120] In an optional embodiment, the phase unwrapping module 23 is specifically configured to: Perform frequency amplitude spectrum peak search on the noise reduction tensor to extract the main frequency phase value of each preset spatio-temporal unit and generate an original phase vector; Perform multi-frequency phase synthesis on the original phase vector according to three groups of preset wavelength laser parameters to calculate a synthesized wavelength parameter; Perform a phase difference coupling operation on the original phase vector according to the synthesized wavelength parameter to generate an unwrapped absolute phase matrix; Perform a non - linear refractive index compensation on the unwrapped absolute phase matrix according to the environmentally - parameter set obtained in real - time to obtain a refractive - index - compensated phase matrix; Perform a phase continuity constraint process on the refractive - index - compensated phase matrix to obtain a noise - reduction - compensated phase matrix; Perform phase unwrapping on the noise - reduction - compensated phase matrix through a preset phase - unwrapping depth network model to obtain the absolute phase matrix.

[0121] The dynamic expansion module 24 is used to perform a dynamic range expansion on the absolute phase matrix to obtain an enhanced phase matrix.

[0122] In an optional implementation manner, the dynamic expansion module 24 is specifically used for: Perform a spatio - temporal gradient calculation on the absolute phase matrix to generate a phase gradient tensor; Construct a multi - dimensional state vector according to the phase gradient tensor, the environmentally - parameter set, and a preset laser wavelength drift amount; Perform a parameter adaptive adjustment on the multi - dimensional state vector through a preset deep reinforcement learning network model to obtain a dynamic control parameter set; Perform a parameterized adjustment on the absolute phase matrix according to the dynamic control parameter set to obtain a frequency - domain band - pass filtered phase matrix; Perform a non - linear compression mapping on the frequency - domain band - pass filtered phase matrix to obtain a compression - mapped phase matrix; Perform a tensor splicing of the compression - mapped phase matrix and the phase gradient tensor to generate the enhanced phase matrix.

[0123] The distance compensation module 25 is used to perform a multi - wavelength solution and environmental compensation on the enhanced phase matrix to obtain a distance matrix.

[0124] In an optional implementation manner, the distance compensation module 25 is specifically used for: Perform a phase - difference synthesis calculation on the enhanced phase matrix according to the three - group wavelength laser parameters to generate a synthesized wavelength parameter; Perform an equivalent synthesized wavelength calculation on the synthesized wavelength parameter through a preset synthesized wavelength formula to generate a multi - level non - ambiguous distance interval; Perform a real - time refractive index correction on the multi - level non - ambiguous distance interval according to the environmentally - parameter set to obtain a refractive - index - corrected distance interval; Perform a multi - source error coupling analysis on the refractive - index - corrected distance interval to obtain a wavelength stability correction factor and a geometric distortion compensation matrix; Perform tensor fusion on the refractive index correction distance interval, the wavelength stability correction factor, and the geometric distortion compensation matrix according to a preset tensor fusion method to obtain an initial distance tensor; Perform residual compensation iterative optimization on the initial distance tensor to obtain the distance matrix.

[0125] The vibration compensation module 26 is configured to perform vibration compensation on the distance matrix to obtain a piezoelectric drive signal.

[0126] In an optional embodiment, the vibration compensation module 26 is specifically configured to: Perform multi-channel joint fast Fourier transform on the distance matrix along the time dimension to generate a vibration spectrum tensor; Perform inverse transfer function calculation processing on the vibration spectrum tensor to obtain a feedforward compensation signal; Perform spatio-temporal differential operation on the distance matrix to generate a real-time vibration error vector; Perform control simulation on the feedforward compensation signal and the real-time vibration error vector through a preset hybrid controller simulation model to obtain a feedback control signal; Perform gain scheduling on the feedback control signal according to a preset environmental stiffness parameter to obtain an optimized feedback control signal; Perform pulse width modulation encoding on the optimized feedback control signal to generate a multi-level piezoelectric drive waveform; Perform waveform adjustment on the multi-level piezoelectric drive waveform through a preset waveform shaping model to obtain the piezoelectric drive signal.

[0127] It should be understood that the various change modes and specific embodiments in the methods provided in the above embodiments are equally applicable to the high-dynamic range multi-channel laser interference control device in this embodiment. Through the foregoing detailed description of the high-dynamic range multi-channel laser interference control method, those skilled in the art can clearly know the implementation method of the high-dynamic range multi-channel laser interference control device in this embodiment. For the sake of brevity of the specification, it will not be described in detail here.

[0128] As Figure 3 shown, it is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

[0129] In a preferred embodiment of the present invention, the electronic device 3 may include, but is not limited to: a memory 31, at least one processor 32, and at least one communication bus 33.

[0130] Those skilled in the art should understand, Figure 3The structure of the illustrated electronic device 3 does not constitute a limitation of the embodiments of the present invention. The electronic device 3 may further include more or fewer other hardware or software than shown in the figure, or different component arrangements.

[0131] In some embodiments, the electronic device 3 is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits, programmable gate arrays, digital signal processors, and embedded devices, etc.

[0132] It should be noted that the electronic device 3 is only an example. Other existing or future electronic products that can be adapted to this application should also be included within the protection scope of this application and are incorporated herein by reference.

[0133] In some embodiments, a computer program is stored in the memory 31. When the computer program is executed by the at least one processor 32, all or part of the steps in the high-dynamic-range multi-channel laser interference control method as described are implemented. The memory 31 includes read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM), or other optical disc memories, magnetic disk memories, tape memories, or any other computer-readable medium capable of carrying or storing data. Further, the computer-readable storage medium mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, etc.

[0134] In some embodiments, the at least one processor 32 is the control core (Control Unit) of the electronic device 3, connecting various components of the entire electronic device 3 through various interfaces and circuits, and by running or executing programs or modules stored in the memory 31, and calling data stored in the memory 31, to perform various functions of the electronic device 3 and process data. For example, when the at least one processor 32 executes the computer program stored in the memory 31, all or part of the steps of the high-dynamic-range multi-channel laser interference control method described in the embodiments of the present application are implemented; or all or part of the functions of the high-dynamic-range multi-channel laser interference control device are implemented. The at least one processor 32 may be composed of integrated circuits. For example, it may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same or different functions packaged, including a combination of one or more central processing units (Central Processing Unit, CPU), microprocessors, digital processing chips, graphics processors, and various control chips, etc.

[0135] In some embodiments, the at least one communication bus 33 is configured to implement connection communication between the memory 31 and the at least one processor 32, etc. Although not shown, the electronic device 3 may further include a power supply (such as a battery) for powering each component. Preferably, the power supply can be logically connected to the at least one processor 32 through a power management device, so as to implement functions such as management of charging, discharging, and power consumption management through the power management device. The power supply may also include any components such as one or more DC or AC power supplies, a recharge device, a power failure detection circuit, a power converter or inverter, and a power status indicator. The electronic device 3 may also include various sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be elaborated here.

[0136] The above-mentioned integrated unit implemented in the form of a software functional module can be stored in a computer-readable storage medium. The above-mentioned software functional module is stored in a storage medium, including several instructions for causing an electronic device (which may be a personal computer, an electronic device, or a network device, etc.) or a processor to execute part of the methods described in the various embodiments of the present application.

[0137] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0138] The module described as a separation component may or may not be physically separated. The component shown as a module may or may not be a physical unit, and it may be located in one place or distributed across multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0139] The above are all preferred embodiments of this application. The protection scope of this application is not limited accordingly. Therefore, all equivalent changes made according to the structure, shape, and principle of this application shall be covered within the protection scope of this application.

Claims

1. A high dynamic range multi-channel laser interferometer control method, characterized in that: The method comprises: Acquire channel signals of multiple target channels in real time according to multiple preset photoelectric detectors, and perform three-dimensional sampling of the channel signals in time domain, frequency domain and space to construct an original tensor; Performing joint optimization processing on the original tensor to obtain a denoised tensor; performing phase unwrapping on the denoised tensor to obtain an absolute phase matrix; Performing dynamic range expansion on the absolute phase matrix to obtain an enhanced phase matrix; Performing multi-wavelength solution and environmental compensation on the enhanced phase matrix to obtain a distance matrix; The distance matrix is ​​vibration compensated to obtain a piezoelectric driving signal.

2. The high dynamic range multi-channel laser interferometer control method according to claim 1, characterized in that: The performing time-domain-frequency-space three-dimensional sampling on the channel signal to construct the original tensor comprises: Performing time domain synchronization processing on the channel signals to obtain a time-aligned optimized channel signal set; Performing short-time Fourier transform processing on the time-aligned optimized channel signal set according to a preset adaptive window to obtain a time-frequency matrix of each target channel; The time-frequency matrix is ​​processed by three-dimensional fusion module construction to obtain the original tensor.

3. The high dynamic range multi-channel laser interferometer control method according to claim 1, characterized in that: The performing joint optimization processing on the original tensor to obtain the denoised tensor comprises: Step S21, performing Tucker decomposition processing on the original tensor to obtain an initial low-rank signal; Step S22, performing a preset soft threshold function calculation process on the initial low-rank signal to obtain an initial noise signal; Step S23, calculating a denoised residual tensor according to the original tensor and the initial noise signal, and performing Tucker decomposition processing on the denoised residual tensor to update the initial low-rank signal; Step S24, performing reconstruction error calculation according to the original tensor, the initial noise signal and the updated initial low-rank signal to obtain a reconstruction error value, and comparing the reconstruction error value with a preset error threshold; Repeat step S22 to step S24 until the reconstruction error value is less than the error threshold, and determine the updated initial low-rank signal as the denoising tensor.

4. The high dynamic range multi-channel laser interferometer control method according to claim 1, characterized in that: The performing phase unwrapping on the denoised tensor to obtain an absolute phase matrix comprises: Performing a frequency amplitude spectrum peak search on the noise reduction tensor to extract the main frequency phase value of each preset space-time unit and generate an original phase vector; According to the preset three sets of wavelength laser parameters, multi-frequency phase synthesis is performed on the original phase vector to calculate the synthetic wavelength parameters; According to the synthetic wavelength parameter, performing a phase difference coupling operation on the original phase vector to generate an un-unwrapped absolute phase matrix; Performing nonlinear refractive index compensation on the unwrapped absolute phase matrix according to the environmental parameter set acquired in real time to obtain a refractive index compensated phase matrix; Performing phase continuity constraint processing on the refractive index compensation phase matrix to obtain a noise reduction compensation phase matrix; The noise reduction compensation phase matrix is ​​phase unwrapped by a preset phase unwrapping deep network model to obtain the absolute phase matrix.

5. The high dynamic range multi-channel laser interferometer control method according to claim 4, characterized in that: The performing dynamic range expansion on the absolute phase matrix to obtain an enhanced phase matrix comprises: Performing spatiotemporal gradient calculation on the absolute phase matrix to generate a phase gradient tensor; Constructing a multidimensional state vector according to the phase gradient tensor, the environmental parameter set and a preset laser wavelength drift; Adaptively adjusting the parameters of the multidimensional state vector through a preset deep reinforcement learning network model to obtain a dynamic control parameter set; Performing parameterized adjustment on the absolute phase matrix according to the dynamic control parameter set to obtain a frequency domain bandpass filter phase matrix; Performing nonlinear compression mapping on the frequency domain bandpass filtering phase matrix to obtain a compression mapping phase matrix; The compressed mapping phase matrix is ​​tensor-concatenated with the phase gradient tensor to generate the enhanced phase matrix.

6. The high dynamic range multi-channel laser interferometer control method according to claim 4, characterized in that: The performing multi-wavelength solution and environmental compensation on the enhanced phase matrix to obtain a distance matrix comprises: Performing phase difference synthesis calculation on the enhanced phase matrix according to the three sets of wavelength laser parameters to generate synthetic wavelength parameters; Performing equivalent synthetic wavelength calculation on the synthetic wavelength parameters by using a preset synthetic wavelength formula to generate multi-level unambiguous distance intervals; Performing real-time refractive index correction on the multi-level unambiguous distance intervals according to the environmental parameter set to obtain a refractive index corrected distance interval; Performing multi-source error coupling analysis on the refractive index correction distance interval to obtain a wavelength stability correction factor and a geometric distortion compensation matrix; Performing tensor fusion on the refractive index correction distance interval, the wavelength stability correction factor and the geometric distortion compensation matrix according to a preset tensor fusion method to obtain an initial distance tensor; The initial distance tensor is subjected to residual compensation iterative optimization to obtain the distance matrix.

7. The high dynamic range multi-channel laser interferometer control method according to claim 1, characterized in that: The performing vibration compensation on the distance matrix to obtain a piezoelectric driving signal comprises: Performing a multi-channel joint fast Fourier transform on the distance matrix along the time dimension to generate a vibration spectrum tensor; Performing inverse transfer function calculation processing on the vibration spectrum tensor to obtain a feedforward compensation signal; Performing a spatiotemporal differentiation operation on the distance matrix to generate a real-time vibration error vector; Performing control simulation on the feedforward compensation signal and the real-time vibration error vector through a preset hybrid controller simulation model to obtain a feedback control signal; Performing gain scheduling on the feedback control signal according to a preset environmental stiffness parameter to obtain an optimized feedback control signal; Performing pulse width modulation encoding on the optimized feedback control signal to generate a multi-level piezoelectric drive waveform; The multi-level piezoelectric driving waveform is waveform-adjusted by a preset waveform shaping model to obtain the piezoelectric driving signal.

8. A high dynamic range multi-channel laser interferometer control device, characterized in that: The device comprises: A feature acquisition module is used to acquire channel signals of multiple target channels in real time according to multiple preset photoelectric detectors, and perform three-dimensional sampling of the channel signals in time domain, frequency domain and space to construct an original tensor; A denoising optimization module, used for performing joint optimization processing on the original tensor to obtain a denoised tensor; A phase unwrapping module, used for performing phase unwrapping on the noise reduction tensor to obtain an absolute phase matrix; A dynamic expansion module, used for expanding the dynamic range of the absolute phase matrix to obtain an enhanced phase matrix; A distance compensation module, used for performing multi-wavelength solution and environmental compensation on the enhanced phase matrix to obtain a distance matrix; The vibration compensation module is used to perform vibration compensation on the distance matrix to obtain a piezoelectric driving signal.

9. An electronic device, characterized in that: The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the high dynamic range multi-channel laser interference control method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the high dynamic range multi-channel laser interferometer control method according to any one of claims 1 to 7 are implemented.

Images