A wavefront sensing-free phased hybrid optimization wavefront correction method

Abstract

The application discloses a wavefront-sensorless staged hybrid optimization wavefront correction method, and relates to the technical field of adaptive optical optimization control, in particular to a wavefront-sensorless staged hybrid optimization wavefront correction method.The method comprises the following steps: building a wavefront-sensorless adaptive optical system for laser communication; performing first-stage trust-region derivative-free optimization in a low-dimensional mode space composed of a plurality of Zernike mode coefficients; setting a joint switching mechanism; when the first-stage trust-region derivative-free optimization meets the joint switching mechanism, outputting first-stage optimal mode coefficients and corresponding optimal single-mode fiber coupling power; constructing an initial control vector for second-stage SPGD optimization based on the first-stage optimal mode coefficients; performing full-dimensional voltage space SPGD optimization based on the initial control vector, outputting a historical optimal drive voltage vector, and using the historical optimal drive voltage vector for wavefront correction.

Description

Technical Field

[0001] This invention relates to the field of adaptive optics optimization control technology, specifically to a wavefront correction method for staged hybrid optimization without wavefront sensing. Background Technology

[0002] Free-space laser communication, with its advantages of high communication speed, strong resistance to electromagnetic interference, abundant spectrum resources, and good confidentiality, has significant application value in scenarios such as space-to-ground communication, long-distance terrestrial communication, UAV relay communication, and mobile platform communication. However, when laser beams are transmitted in atmospheric channels, they are easily affected by atmospheric turbulence, resulting in phenomena such as wavefront phase distortion, spot expansion, drift, and flicker. This reduces the coupling efficiency of the single-mode fiber at the receiving end, increases the bit error rate, and may even cause communication interruption. Adaptive optics technology can dynamically compensate for distorted wavefronts through wavefront correctors, which is an important means to improve the coupling power and link stability of laser communication receivers. Traditional adaptive optics systems usually rely on wavefront sensors to obtain wavefront information. However, in miniaturized, low-cost, and highly integrated laser communication terminals, wavefront sensors increase the complexity of the system structure, the difficulty of assembly and adjustment, and the hardware cost.

[0003] Furthermore, under conditions of strong atmospheric turbulence and intense light scintillation, the rapid changes in the intensity distribution of the received light field lead to a deterioration in the signal quality of the sub-aperture of the wavefront sensor, thereby reducing the accuracy of wavefront measurement and the performance of closed-loop correction. Wavefront-sensorless adaptive optics systems directly acquire system performance indicators such as optical power and spot quality as feedback signals and use optimization algorithms to drive the wavefront corrector for closed-loop correction. This eliminates the need for a separate wavefront sensor, thus reducing the impact of wavefront measurement errors on system performance under complex light field conditions. Therefore, it shows promising application prospects in free-space laser communication.

[0004] Chinese invention patent application CN122172444A, entitled "A Fast Wavefront-Free Sensor-Free Adaptive Optical Correction Method, System, and Computer Program Product Based on a Hybrid of Staged Contraction and Quadratic Modeling," discloses a fast wavefront-free sensor-free adaptive optics correction method based on a hybrid of staged contraction and quadratic modeling. This method sets the control space of the wavefront corrector to the first M-order Zernike modes. First, it performs a linear scan of the Zernike mode coefficients to obtain a coarse search space. Then, it contracts the coarse search space and iteratively refines it within the contracted search interval using the SPGD algorithm. Finally, it establishes a quadratic model near the iterative optimum to solve for the optimum, and outputs the corresponding correction control quantity. This scheme improves the search efficiency and local optimization capability of the wavefront-free sensor-free adaptive optics correction process to a certain extent by performing linear search, interval contraction, random perturbation optimization, and local quadratic fitting within the Zernike mode space.

[0005] However, the optimization process of the existing methods mainly revolves around the coefficients of finite-order Zernike models, and their correction capability is limited by the selected model order, model orthogonality, and model representation range. When atmospheric turbulence is strong or wavefront distortion components are complex, finite-order Zernike models cannot fully characterize the higher-order components and local details in the actual distorted wavefront, and are prone to leaving uncompensated wavefront errors, resulting in problems such as energy diffusion, insufficient peak intensity, or limited coupling efficiency improvement in the corrected beam. Summary of the Invention

[0006] To address the problem that existing wavefront-sensorless adaptive optics correction methods still struggle to balance convergence efficiency and final correction accuracy under conditions of strong turbulence and high-precision correction, the present invention aims to propose a wavefront correction method with wavefront-sensorless phased hybrid optimization.

[0007] The method includes the following steps: S1. Build a wavefront-sensorless adaptive optics system for laser communication; S2. Initialize the Zernike coefficient vector, first-stage trust region derivative-free optimization parameters, and second-stage SPGD optimization parameters; S3. Perform the first-stage derivativeless optimization in the low-dimensional pattern space consisting of several Zernike pattern coefficients; A joint handover mechanism is set up. When the first-stage trust region derivative-free optimization satisfies the joint handover mechanism, the optimal mode coefficients for the first stage are output. and Corresponding objective function value ; S4, based on Construct the initial voltage control vector for the second-stage SPGD optimization. ; S5, based on Full-dimensional voltage space SPGD optimization is performed to output the historical best driving voltage vector, which is used as the optimal driving voltage vector for wavefront correction.

[0008] Furthermore, the wavefront-free adaptive optics system includes: a wavefront corrector, a focusing lens, a PD detector, a wavefront controller, and a single-mode fiber coupler. The incident distorted wavefront signal is incident on the wavefront corrector. The wavefront corrector generates a corresponding adjustable wavefront compensation amount according to the driving voltage signal output by the wavefront controller, and performs phase modulation on the incident distorted wavefront signal. The phase-modulated optical signal is focused by a focusing lens onto the incident end face of the single-mode fiber coupler to achieve single-mode fiber coupling. The coupled optical signal output from the single-mode fiber coupler is transmitted to the PD detector. The PD detector collects the output optical power of the coupled optical signal in real time and feeds the output optical power back to the wavefront controller. The wavefront controller uses the negative value of the output optical power as the objective function evaluation value, executes a phased hybrid optimization wavefront correction method, generates an updated driving voltage vector, and outputs it to the wavefront corrector.

[0009] Furthermore, in step S2, the Zernike coefficient vector is initialized as a zero vector: ; The parameters for the first-stage trust region derivative-free optimization include: Optimization problem parameters: number of low-dimensional patterns Mode response matrix Number of function evaluations in the first stage ; Trust region control parameters: Initial trust region radius Minimum trust region radius Trust domain trust threshold Increase in trust region radius and the reduction ratio of the trust region radius ; Stage switching parameters: Number of consecutive convergence checks in the first stage and stage switching threshold ; The second-stage SPGD optimization parameters include: the number of function evaluations in the second stage. Random perturbation amplitude δ, gradient descent step size Number of consecutive convergence checks in the second phase And optimize the termination threshold .

[0010] Furthermore, the first stage of derivative-free optimization in the trust region includes: S31. Taking the initial model coefficient vector as the center, select several initial model coefficient sample points in the n-dimensional Zernike model subspace, collect the objective function evaluation value corresponding to each initial model coefficient sample point, form an initial interpolation point set, and update the first stage function evaluation number. S32. Iteratively solve for the optimal mode coefficients in the first stage. Specifically: S321. Construct a local quadratic interpolation model based on the current interpolation point set, and optimize the local quadratic interpolation model to obtain the optimal solution for the current iteration prediction. ; S322, Collection The objective function evaluation value corresponding to the point is used to calculate the current iteration center point. and The ratio between the actual decrease and the predicted decrease at each point is used to update the number of evaluations for the first-stage function. S323, When the ratio is greater than the preset trust threshold of the trust region At that time, increase the trust region radius and set the current iteration center point. Updated to the current iteration's predicted optimal solution. ; When the ratio is not greater than the preset trust threshold, the trust region radius is reduced while maintaining the current iteration center point. constant; S324. Update the interpolation point set and iteratively execute steps S321 to S323 until the joint switching mechanism is satisfied, ending the first stage of the trust region derivative-free optimization and recording the optimal solution of the first stage. and the corresponding objective function value ,in The objective function is denoted as .

[0011] Furthermore, the joint handover mechanism is a condition-triggered mechanism, including a trust region shrinkage condition, a function evaluation budget constraint condition, and a performance gain convergence criterion condition. The joint handover mechanism is satisfied when any one of these conditions is met. The trust region shrinkage condition is that the trust region radius is not greater than the preset minimum trust region radius; The function evaluation budget constraint is that the number of function evaluations in the first stage is greater than or equal to... ; The performance gain convergence criterion condition is continuous In the next iteration .

[0012] Furthermore, the initial control vector for the second-stage SPGD optimization... The formula for calculation is: .

[0013] Furthermore, the full-dimensional voltage space SPGD optimization includes: S51. Initialize the historical best objective function value ; Initialize the historical best driving voltage vector ; Initialize the second-stage function to evaluate the counter count. ; Initialize the number of iterations of the second phase counter

[0014] S52. Generate a random perturbation vector in the full-dimensional driving voltage space of the wavefront corrector. ; S53, for the driving voltage vector of the current iteration respectively Apply positive and negative perturbations and collect the corresponding objective function evaluation values, denoted as . and And update the second-stage function evaluation counter count and iteration counter count; S54, according to and The difference is used to update the driving voltage vector for the current iteration; S55. Acquire the objective function evaluation value corresponding to the updated driving voltage vector after step S54; When the objective function evaluation value is less than the historical best objective function value, the updated driving voltage vector is updated to the historical best driving voltage vector, and the historical best objective function value is updated accordingly. S56. Iterate through steps S52 to S55 until the number of times the function evaluation counter in the second stage is greater than or equal to the specified number of times. or continuous In the next iteration ; After completing the second stage of full-dimensional voltage space SPGD optimization, the historical optimal driving voltage vector is used as the final driving voltage vector for wavefront correction.

[0015] An electronic device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the above method.

[0016] A computer-readable storage medium for storing computer instructions that, when executed by a processor, implement the steps of the above-described method.

[0017] The beneficial effects of the method described in this invention are as follows: (1) In the first stage, this invention confines the optimization process to a low-dimensional Zernike mode space and uses a trust region derivative-free optimization to construct a local quadratic interpolation model. The mode coefficients and trust region parameters are updated by combining model prediction with actual objective function evaluation. Compared with optimization methods that rely entirely on random perturbations or ergonomic search, this invention can quickly search for the dominant aberration correction direction using structural information in the low-dimensional mode space. Under conditions of low initial coupling power or fluctuating feedback signal, this invention improves the optimization search efficiency, reduces invalid iterations, and accelerates the initial convergence process of the system.

[0018] (2) This invention employs a joint switching mechanism comprised of the number of function evaluations, the trust region radius, and the continuous performance improvement. When the optimization gains in the first stage decrease, the trust region shrinks to a preset range, or the function evaluation overhead reaches its upper limit, the low-dimensional mode space optimization is terminated promptly. This avoids continuously performing redundant iterations with limited gains within a finite-order Zernike mode space, reduces the number of physical samplings, and improves the real-time performance of the wavefront-free adaptive optics system under dynamic turbulent conditions.

[0019] (3) In this invention, the optimal Zernike mode coefficients obtained in the first stage are converted into the driving voltage vector of the wavefront corrector through mode-voltage mapping, and this driving voltage vector is used as the hot start initial point for the second stage SPGD optimization. Thus, the second stage random perturbation optimization does not start from an uncorrected or random state, but from a higher coupling power state where low-order dominant aberration compensation has been completed. This improves the effectiveness of the performance feedback signal, reduces search fluctuations caused by random perturbations in the low coupling region, and enhances the continuity and stability of the two-stage optimization process.

[0020] (4) In the second stage of this invention, SPGD fine optimization is performed directly within the full-dimensional driving voltage space corresponding to all actuators of the wavefront compensator, instead of continuing to be limited to the finite-order Zernike mode coefficient space. As a result, all control degrees of freedom of the wavefront compensator can be further utilized to compensate for high-order wavefront components, local detail distortions and residual errors that are difficult to fully express by low-dimensional Zernike modes, reduce the impact of mode truncation error on the final correction effect, and improve the final wavefront correction accuracy under complex turbulent conditions.

[0021] (5) Through the synergistic effect of rapid initial optimization in low-dimensional mode space, hot start of mode-voltage mapping and fine optimization in full-dimensional voltage space, this invention can balance initial search efficiency and final correction accuracy without significantly increasing the cost of high-dimensional modeling. It improves the problems of decreased search efficiency, increased iteration number, fluctuation in optimization process and limited improvement of coupling efficiency that are easy to occur in existing methods under strong turbulence or low signal-to-noise ratio feedback conditions. This improves the optimization efficiency and dynamic correction performance of free-space laser communication adaptive optics system. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method described in this invention; Figure 2 This is a schematic diagram of the wavefront-free adaptive optics system described in this invention. Figure 3 Workflow diagram of the phased hybrid optimization scheme described in this invention; Among them, 1-wavefront corrector, 2-focusing lens, 3-PD detector, 4-wavefront controller and 5-single-mode fiber coupler. Detailed Implementation

[0023] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1 This embodiment provides a wavefront correction method with phased hybrid optimization and no wavefront sensing. The flowchart of the method is as follows: Figure 1 As shown, the method includes the following steps: S1. Build a wavefront-sensorless adaptive optics system for laser communication; The relevant operations in step S1 will be introduced with specific examples: like Figure 2 As shown, the wavefront-free adaptive optics system includes: a wavefront corrector 1, a focusing lens 2, a PD detector 3, a wavefront controller 4, and a single-mode fiber coupler 5. The distorted wavefront signal after atmospheric turbulence disturbance is incident on the wavefront corrector 1. The wavefront corrector 1 generates a corresponding adjustable wavefront compensation amount according to the driving voltage signal output by the wavefront controller 4, and performs phase modulation on the incident distorted wavefront signal. The phase-modulated optical signal is focused by the focusing lens 2 to the incident end face of the single-mode fiber coupler 5 to realize single-mode fiber coupling.

[0025] The coupled optical signal output by the single-mode fiber coupler 5 is transmitted to the PD detector 3. The PD detector 3 collects the output optical power of the coupled optical signal in real time and feeds the output optical power back to the wavefront controller 4. The wavefront controller 4 uses the negative value of the output optical power as the objective function evaluation value, executes a phased hybrid optimization wavefront correction method, generates an updated driving voltage vector and outputs it to the wavefront corrector 1, thereby constructing a closed-loop optimization control process to achieve adaptive correction of wavefront distortion caused by atmospheric turbulence.

[0026] The wavefront corrector 1 is preferably a deformable mirror, used to achieve continuously adjustable wavefront phase modulation; in other embodiments, a liquid crystal spatial light modulator can also be used as an alternative.

[0027] Specifically, in this embodiment, the wavefront corrector 1 uses a Soleber DMH40 series deformable mirror with 40 actuators, and the wavefront controller 4 uses a high-performance computer and integrates the phased hybrid optimization wavefront correction method.

[0028] S2. Initialize the Zernike coefficient vector, the first-stage trust region derivative-free optimization parameters, and the second-stage SPGD optimization parameters; The relevant operations in step S2 will be introduced with specific examples: In this embodiment, the coupling power of the single-mode fiber coupler 5 is selected as the system performance index, and its negative value is defined as the optimization objective function. It is used to evaluate the system performance under the current wavefront correction state.

[0029] The coupling power of the single-mode fiber coupler 5 is acquired and output by the PD detector 3.

[0030] It should be noted that in other embodiments, the peak light intensity of the far-field spot, Strell ratio, power in the barrel, beam quality evaluation index, or other feedback quantities that can reflect the wavefront correction performance of the system can also be selected as system performance indicators according to the specific application scenario. This embodiment does not limit this.

[0031] The Zernike coefficient vector is initialized to a zero vector: .

[0032] The first-stage trust region derivative-free optimization parameters include: (1) Optimization problem parameters: number of low-dimensional patterns Mode response matrix Number of function evaluations in the first stage ; (2) Trust region control parameters: initial trust region radius Minimum trust region radius Trust domain trust threshold Increase in trust region radius and the reduction ratio of the trust region radius ; (3) Stage switching parameters: Number of consecutive convergence determinations in the first stage and stage switching threshold ; Wherein, the mode response matrix This is a control matrix used to characterize the mapping relationship between Zernike mode coefficients and wavefront compensator driving voltage, and is used to realize the transformation from low-dimensional mode space to high-dimensional voltage control space. Specifically, the response relationship of each actuator of the wavefront compensator to the Zernike mode is obtained through pre-calibration, and the mode response matrix is ​​constructed: ; in, B is the influence matrix of the wavefront corrector, used to describe the wavefront response generated by each actuator under a unit driving voltage; B is the target wavefront matrix composed of selected Zernike modes. This represents the pseudo-inverse of a matrix.

[0033] The number of consecutive convergence determinations The convergence decision window length is used to characterize the first-stage optimization. When continuous... The objective function in each iteration decreases by no more than a preset threshold. If the current trust domain optimization is considered to have reached the near-optimal of the pattern space or a local optimum, and further iteration is unlikely to achieve significant performance improvement, the first stage can be terminated and the process can switch to the second stage of optimization.

[0034] Specifically, in this embodiment, to achieve rapid initial correction of low-order wavefront aberrations, a low-dimensional Zernike mode number is set. (Corresponding to typical low-order astigmatism, coma, and defocus, etc., the upper limit of the number of function evaluations) Initial trust region radius Minimum trust region radius Number of consecutive convergence determinations Phase switching threshold .

[0035] In particular, This is used to determine the predictive reliability of the current secondary proxy model. In this embodiment, η = 0.75. =2, =0.5. The above parameter settings are only a preferred implementation of this embodiment. Under different turbulence intensities or system scales, they can be adaptively adjusted according to the convergence performance requirements. This invention does not limit this.

[0036] The second-stage SPGD optimization parameters include: the number of function evaluations in the second stage. Random disturbance amplitude Gradient descent step size Number of consecutive convergence checks in the second phase And optimize the termination threshold ; Among them, the number of consecutive convergence determinations The decision window length is used to characterize the convergence state of the second-stage algorithm. When continuous... The coupling power improvement in each iteration does not exceed the preset threshold. If the current optimization process is considered to have basically converged, and the performance improvement brought by continued iteration is limited, the second stage of optimization can be terminated.

[0037] In this embodiment, to achieve rapid and precise optimization of the high-dimensional voltage space, the following settings are made: =500, (in (The maximum amplitude of the deformable mirror driving voltage). , =5, =0.01.

[0038] The above parameters are used to further improve the accuracy of residual wavefront correction in the high-coupling-power operating region. Since the local gradient information of the objective function is of high quality, SPGD is significantly less sensitive to the disturbance amplitude and step size parameters compared to the initial optimization stage.

[0039] S3. Perform the first-stage derivativeless optimization in the low-dimensional pattern space consisting of several Zernike pattern coefficients; A joint handover mechanism is set up. When the first-stage trust region derivative-free optimization satisfies the joint handover mechanism, the optimal mode coefficients for the first stage are output. and Corresponding objective function value ; The relevant operations in step S3 will be introduced with specific examples: like Figure 3 As shown, the first stage of derivative-free optimization in the trust region includes: S31, Initialization function evaluates the counter count. Number of iterations counter .

[0040] exist In the Zernike pattern subspace, with Centered on, select An initial set of model coefficient sample points is generated, and the objective function evaluation value corresponding to each sample point is collected to form an initial interpolation point set: In this embodiment, the number of interpolation points is set to... When the pattern dimension n = 8, we take M = 17.

[0041] After the initial interpolation point set is formed, the number of objective function evaluations completed during the construction of the initial interpolation point set is included in the number of function evaluations in the first stage; wherein, the number of objective function evaluations completed is consistent with the number of initial model coefficient sample points in the initial interpolation point set, and each initial model coefficient sample point corresponds to one objective function evaluation.

[0042] S32, Based on the current interpolation point set Constructing a local quadratic interpolation model :

[0043] in For the current stage The next trust region iteration center point is defined as the point up to the [number]th iteration. The optimal control vector obtained at the end of the iteration, when hour , for The independent variable. For gradient estimation, To approximate the Hessian matrix, based on the current set of interpolation points. By satisfying the interpolation conditions The solution has been determined.

[0044] Based on local quadratic interpolation model Iteratively solve for the optimal mode coefficients in the first stage. Specifically: S321. Optimize the local quadratic interpolation model under the current trust region radius constraint to obtain the optimal solution for the current iteration prediction. , ,in The step size vector within the trust region. For the first The radius of the trust region in the next iteration, when hour . For the first The local quadratic initial interpolation model for the next iteration. Used to find the value of the independent variable that minimizes the objective function.

[0045] S322, Collection The objective function evaluation value corresponding to the point is used to calculate the current iteration center point. and The ratio between the actual decrease and the predicted decrease at the corresponding point ,

[0046] After completing the objective function evaluation of the current iteration's predicted optimal solution, update the number of first-stage function evaluations; wherein, the current iteration's predicted optimal solution corresponds to one actual objective function evaluation, so the number of first-stage function evaluations is increased by one based on the current count.

[0047] S323, when the ratio Greater than the preset trust threshold At that time, increase the trust region radius, i.e. and the current iteration center point Updated to the current iteration's predicted optimal solution. ; When the ratio is not greater than a preset trust threshold, the trust region radius is reduced, i.e. and maintain the current iteration center point. constant; S324, according to the above The interpolation point set is updated with the corresponding objective function value to obtain the next round of interpolation point set. Among them, when When accepted, it is used as the center point for the next round of trust region iteration and is used to update the interpolation point set; when If the candidate mode coefficient is not accepted, the current iteration center point remains unchanged, and the candidate mode coefficient points are used as evaluated sample points to update the interpolation point set.

[0048] After completing the current iteration's prediction of the optimal solution, objective function evaluation, trust region radius update, and interpolation point set update, the first-stage iteration counter is incremented once, and the next round of trust region iteration begins.

[0049] Iteratively execute steps S321 to S323 until the first-stage trust region derivative-free optimization satisfies the joint switching mechanism, then end the iteration and record the optimal solution of the first stage. and the corresponding objective function value ,in Let be the objective function. The optimal mode coefficients obtained in the low-dimensional stage have a clear physical meaning, representing the wavefront state where the dominant low-order aberrations have been effectively corrected. Using these coefficients as the initial hot start point for high-dimensional optimization allows the high-dimensional stochastic perturbation optimization process to start from the high-performance region, avoiding inefficient searches in the low-performance region, thereby improving the convergence efficiency of the high-dimensional stage.

[0050] To address the problem of the difficulty in adaptively determining the switching timing during the two-stage optimization process, this invention proposes a joint switching mechanism based on the trust region shrinkage state, function evaluation budget, and performance gain convergence criterion, which enables adaptive switching from low-dimensional mode space optimization to high-dimensional voltage space optimization.

[0051] Specifically, when the radius of the trust region shrinks to a preset lower limit during the derivative-free optimization process of the trust region. hour( This indicates that the current optimization process has entered a relatively small search region, and the update range of the mode coefficients is limited. Further narrowing the search range or increasing the number of iterations will only result in limited improvement of the objective function. At the same time, an upper limit on the number of function evaluations is introduced. As a computational resource constraint, it limits the maximum optimization overhead of the first stage. Once the trust region optimization has essentially completed the wavefront correction task in the mode space, further iterations will yield very limited performance improvements. By setting an upper limit on the number of function evaluations, redundant optimizations in the mode space can be avoided in the first stage, thus improving overall optimization efficiency.

[0052] Based on the above conditions, this invention further introduces a convergence criterion based on performance gain, namely, when continuous... In each iteration, the decrease in the objective function was less than the preset threshold. hour( The study concluded that the current optimization process has entered a state of local convergence or plateau, indicating that the benefits of further optimization in the low-dimensional model space have significantly decreased.

[0053] The optimization phase switch is triggered when any one of the following conditions is met: trust region shrinkage, function evaluation budget constraint, or performance gain convergence criterion. This joint switching mechanism comprehensively considers the convergence state of the surrogate model, computational resource consumption, and changes in system performance gain, achieving adaptive determination of the optimization phase switch.

[0054] Once the joint switching mechanism is satisfied, this invention uses a control space mapping mechanism to convert the low-dimensional optimal mode information obtained in the first stage into the initial control state of the deformable mirror voltage control space. This initial state serves as the hot start point for the SPGD algorithm, allowing stochastic optimization to iterate directly from the high-performance region. Under higher coupling power conditions, the signal-to-noise ratio of the objective function is significantly improved, and the stochastic gradient estimation is more stable. Furthermore, since the voltage control space can directly act on all deformable mirror actuators, it can fully utilize all control degrees of freedom of the system to further compensate for residual wavefront errors that are difficult to describe in the mode space, achieving higher final wavefront correction accuracy based on the optimization results of the first stage.

[0055] S4, based on Construct the initial control vector for the second stage of SPGD optimization. ; The relevant operations in step S4 will be introduced with specific examples: Construct the initial point for hot start. Find the low-dimensional optimal solution for the first stage. Mapping to the voltage control space, construct the initial control vector for the second-stage SPGD:

[0056] This mapping enables state inheritance between the low-dimensional pattern space and the high-dimensional voltage space, allowing the second-stage optimization to start from the high-performance region, thereby avoiding performance degradation caused by random initialization.

[0057] The hot-start switching mechanism proposed in this invention can achieve a smooth expansion of the control space dimension while maintaining the correction performance already obtained in the first stage. This avoids the introduction of additional disturbances due to random initialization in the initial stage of high-dimensional optimization, thereby improving the optimization stability and convergence continuity during the two-stage switching process.

[0058] S5, based on Full-dimensional voltage space SPGD optimization is performed to output the historical best driving voltage vector, which is then used as the final driving voltage vector for wavefront correction.

[0059] The relevant operations in step S5 will be introduced with specific examples: like Figure 3 As shown, the full-dimensional voltage space SPGD optimization includes: S51. Initialize the second-stage function evaluation counter. Iteration counter ; Initialize the historical best objective function value ; Initialize the historical best driving voltage vector ; S52, Generate within the full-dimensional driving voltage space of the wavefront corrector (1) 3D random perturbation vector ; N The number of actuators for the wavefront corrector. Each component independently originates from {+δ, It is randomly generated in the Bernoulli distribution of δ}.

[0060] S53, for the driving voltage vector of the current iteration respectively Apply positive and negative perturbations and collect the corresponding objective function evaluation values. and ; After completing the objective function evaluation under the current positive and negative perturbations of the driving voltage vector, the two actual objective function evaluations are counted in the second stage function evaluation count, so that the second stage function evaluation count is increased by two based on the current count.

[0061] S54. Update the driving voltage vector for the current iteration, specifically: Update the full-dimensional control vector according to the direction of stochastic gradient descent:

[0062] S55. Acquire the objective function evaluation value corresponding to the updated driving voltage vector after step S54; When the objective function evaluation value corresponding to the updated driving voltage vector is less than the historical best objective function value, that is... The updated driving voltage vector is then updated to the historical best driving voltage vector. ), and simultaneously update the historical best objective function value ( ); After completing the positive perturbation evaluation, negative perturbation evaluation, driving voltage vector update, and historical best driving voltage vector update judgment of the current driving voltage vector, the second-stage iteration counter is incremented once, and the next round of full-dimensional voltage space SPGD optimization is entered.

[0063] S56. Iterate through steps S52 to S55 until the second-stage function evaluation counter count is greater than or equal to the second-stage function evaluation count. or continuous In the next iteration .

[0064] After completing the second stage of full-dimensional voltage space SPGD optimization, the historical best driving voltage vector is determined as the optimal driving voltage vector, i.e. .

[0065] Wavefront corrector 1 generates the corresponding wavefront compensation amount according to the optimal driving voltage vector, and performs phase modulation on the incident distorted wavefront signal to achieve wavefront correction.

[0066] Based on the specific embodiments described above, this invention proposes a staged hybrid optimization wavefront correction method for wavefront-sensorless adaptive optics systems. This method decomposes the high-dimensional wavefront correction problem into two stages: low-dimensional fast convergence optimization and high-dimensional fine compensation optimization, thereby achieving the synergistic utilization of advantages in different control spaces.

[0067] In the first stage, a local quadratic approximation model of the objective function is constructed in the low-dimensional Zernike mode space using the trust region derivative-free optimization algorithm, thereby achieving rapid suppression of the dominant low-order wavefront aberrations. In the second stage, the SPGD algorithm is used in the high-dimensional voltage control space to finely optimize the residual high-order wavefront error, thereby further improving the final correction accuracy of the system.

[0068] Compared with the traditional single SPGD optimization method, this invention effectively overcomes the problems of slow convergence speed, sensitivity to perturbation parameters, and easy getting trapped in inefficient random search in strong turbulence high-dimensional scenarios. While significantly reducing the number of function evaluations required for effective convergence, it improves the system's fast response capability and steady-state tracking performance under different turbulence intensities, and has good engineering adaptability and application value.

[0069] The core innovation of this invention lies in proposing a phased optimization framework for the coordination of low-dimensional mode space and high-dimensional voltage space in wavefront-sensorless adaptive optics systems. By controlling the structured mapping of the control space, performance-driven adaptive switching, and a hot-start enhancement mechanism, information transmission and complementary advantages of different optimization strategies in different control spaces are realized, thereby significantly improving the overall convergence efficiency while ensuring the final correction accuracy.

[0070] It should be understood that the specific embodiments described herein are merely some examples of the present invention and are not intended to limit the present invention. Any modifications made in accordance with the technical concept proposed in this invention should be included within the scope of protection of this invention.

Claims

1. A wavefront correction method with phased hybrid optimization and no wavefront sensing, characterized in that, The method includes the following steps: S1. Build a wavefront-sensorless adaptive optics system for laser communication; S2. Initialize the Zernike coefficient vector, first-stage trust region derivative-free optimization parameters, and second-stage SPGD optimization parameters; S3. Perform the first-stage derivativeless optimization in the low-dimensional pattern space consisting of several Zernike pattern coefficients; A joint handover mechanism is set up. When the first-stage trust region derivative-free optimization satisfies the joint handover mechanism, the optimal mode coefficients for the first stage are output. and Corresponding objective function value ; S4, based on Construct the initial voltage control vector for the second-stage SPGD optimization. ; S5, based on Full-dimensional voltage space SPGD optimization is performed to output the historical best driving voltage vector, which is used as the optimal driving voltage vector for wavefront correction.

2. The wavefront correction method for staged hybrid optimization without wavefront sensing according to claim 1, characterized in that, The wavefront-free adaptive optics system includes: a wavefront corrector (1), a focusing lens (2), a PD detector (3), a wavefront controller (4), and a single-mode fiber coupler (5). The incident distorted wavefront signal is incident on the wavefront corrector (1). The wavefront corrector (1) generates a corresponding adjustable wavefront compensation amount according to the driving voltage signal output by the wavefront controller (4) and performs phase modulation on the incident distorted wavefront signal. The phase-modulated optical signal is focused by the focusing lens (2) to the incident end face of the single-mode fiber coupler (5) to realize single-mode fiber coupling. The coupled optical signal output by the single-mode fiber coupler (5) is transmitted to the PD detector (3). The PD detector (3) collects the output optical power of the coupled optical signal in real time and feeds the output optical power back to the wavefront controller (4). The wavefront controller (4) uses the negative value of the output optical power as the objective function evaluation value, executes the phased hybrid optimization wavefront correction method, generates the updated driving voltage vector and outputs it to the wavefront corrector (1).

3. The wavefront correction method for staged hybrid optimization without wavefront sensing according to claim 2, characterized in that, In step S2, the Zernike coefficient vector is initialized as a zero vector: ; The parameters for the first-stage trust region derivative-free optimization include: Optimization problem parameters: number of low-dimensional patterns Mode response matrix Number of function evaluations in the first stage ; Trust region control parameters: Initial trust region radius Minimum trust region radius Trust domain trust threshold Increase in trust region radius and the reduction ratio of the trust region radius ; Stage switching parameters: Number of consecutive convergence checks in the first stage and stage switching threshold ; The second-stage SPGD optimization parameters include: the number of function evaluations in the second stage. Random perturbation amplitude δ, gradient descent step size Number of consecutive convergence checks in the second phase And optimize the termination threshold .

4. The wavefront correction method for staged hybrid optimization without wavefront sensing according to claim 3, characterized in that, The first stage of derivative-free optimization in the trust region includes: S31. Taking the initial model coefficient vector as the center, select several initial model coefficient sample points in the n-dimensional Zernike model subspace, collect the objective function evaluation value corresponding to each initial model coefficient sample point, form an initial interpolation point set, and update the first stage function evaluation number. S32. Iteratively solve for the optimal mode coefficients in the first stage. Specifically: S321. Construct a local quadratic interpolation model based on the current interpolation point set, and optimize the local quadratic interpolation model to obtain the optimal solution for the current iteration prediction. ; S322, Collection The objective function evaluation value corresponding to the point is used to calculate the current iteration center point. and The ratio between the actual decrease and the predicted decrease at the corresponding point is used to update the number of evaluations for the first-stage function. S323, When the ratio is greater than the preset trust threshold of the trust region At that time, increase the trust region radius and set the current iteration center point. Updated to the current iteration's predicted optimal solution. ; When the ratio is not greater than the preset trust threshold, the trust region radius is reduced while maintaining the current iteration center point. constant; S324. Update the interpolation point set and iteratively execute steps S321 to S323 until the joint switching mechanism is satisfied, ending the first stage of trust region derivative-free optimization and recording the optimal solution of the first stage. and the corresponding objective function value ,in The objective function is denoted as .

5. The wavefront correction method for staged hybrid optimization without wavefront sensing according to claim 4, characterized in that, The joint handover mechanism is a condition-triggered mechanism, including a trust region shrinkage condition, a function evaluation budget constraint condition, and a performance gain convergence criterion condition. The joint handover mechanism is satisfied when any one of these three conditions is met. The trust region shrinkage condition is that the trust region radius is not greater than the preset minimum trust region radius; The function evaluation budget constraint is that the number of function evaluations in the first stage is greater than or equal to... ; The performance gain convergence criterion condition is continuous In the next iteration .

6. The wavefront correction method for staged hybrid optimization without wavefront sensing according to claim 5, characterized in that, The initial control vector for the second stage of SPGD optimization The formula for calculation is: .

7. The wavefront correction method for staged hybrid optimization without wavefront sensing according to claim 6, characterized in that, The full-dimensional voltage space SPGD optimization includes: S51. Initialize the historical best objective function value ; Initialize the historical best driving voltage vector ; Initialize the second-stage function to evaluate the counter count. ; Initialize the number of iterations of the second phase counter ; S52. Generate a random perturbation vector in the full-dimensional driving voltage space of the wavefront corrector (1). ; S53, for the driving voltage vector of the current iteration respectively Apply positive and negative perturbations and collect the corresponding objective function evaluation values, denoted as . and And update the second-stage function evaluation counter count and iteration counter count; S54, according to and The difference is used to update the driving voltage vector for the current iteration; S55. Acquire the objective function evaluation value corresponding to the updated driving voltage vector after step S54; When the objective function evaluation value is less than the historical best objective function value, the updated driving voltage vector is updated to the historical best driving voltage vector, and the historical best objective function value is updated accordingly. S56. Iterate through steps S52 to S55 until the number of times the function evaluation counter in the second stage is greater than or equal to the specified number of times. or continuous In the next iteration ; After completing the second stage of full-dimensional voltage space SPGD optimization, the historical optimal driving voltage vector is used as the final driving voltage vector for wavefront correction.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-7.

9. A computer-readable storage medium for storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method according to any one of claims 1-7.

Images