A real-time high-precision wavefront distortion phase compensation system

Abstract

The present invention discloses a real-time high-precision wavefront distortion phase compensation system, comprising a deformable mirror, a beam splitter, a first focusing lens, a second focusing lens, a receiving module, a CCD camera, a U‑Net convolutional neural network processing module and a wavefront reconstruction module; the light intensity information of a light beam with a distorted phase is collected by the CCD camera, the U‑Net network processing module performs real-time high-precision sensing of the wavefront distortion phase, the wavefront reconstruction module generates a corresponding driving voltage according to the wavefront distortion phase information to deform the deformable mirror, thereby compensating the wavefront distortion phase. The present invention can realize real-time high-precision compensation of the wavefront distortion phase.

Description

Technical Field

[0001] The invention relates to a real-time high-precision wavefront distortion phase compensation system, and belongs to the field of adaptive optics without wavefront sensing. Background Art

[0002] With the rapid development of science and technology, traditional radio frequency communication systems can no longer meet the rapid demand for high-speed communication services in terms of communication rate or data capacity. Free space coherent optical communication is considered to be an important technical means to break through the existing high-speed communication bottleneck due to its outstanding technical advantages of fast communication rate and large information capacity. However, in practical applications, optical signals will be affected by negative effects such as light intensity flicker and beam drift caused by atmospheric turbulence, which seriously affects the communication quality of free space coherent optical communication systems.

[0003] Compared with conventional adaptive optics, the biggest feature of non-wavefront sensing adaptive optics is that it does not use wavefront sensors to detect wavefront distortion, but directly controls the wavefront corrector through optimization algorithms to optimize the signal light. It has become one of the main research directions in recent years to suppress atmospheric turbulence and improve the communication quality of free-space coherent optical communication systems. SW Paine and JRFienup of the University of Rochester in the United States used a non-wavefront sensing adaptive optics system based on the traditional InceptionV3 network architecture to effectively predict wavefront aberrations. Y. Nishizaki and others from Osaka University in Japan used a CNN-based neural network to accurately calculate the first 32 Zernike coefficients. Cao Jingtai and others in China used the Stochastic Parallelism Gradient Descent (SPGD) algorithm to correct wavefront aberrations in free-space coherent optical communication systems. The results of simulation and experiments show that the SPGD algorithm can effectively compensate for wavefront distortion and improve the communication quality of free-space coherent optical communication systems. Ma Huimin et al. proposed a wavefront-free sensing algorithm based on convolutional neural network. The simulation results show that after distortion compensation, the Strehl ratio of the spot has been significantly improved. The above methods have the following problems:

[0004] 1. Most of the atmospheric turbulence models responsible for providing training samples are simulated using a method based on Zernike polynomials. The turbulence samples generated by this method have insufficient high-frequency information, which can easily lead to low sensing accuracy of the network model trained using these samples, resulting in an inability to accurately compensate for the wavefront distortion phase.

[0005] 2. The optimization algorithms responsible for controlling the wavefront reconstruction module mostly use traditional convolutional neural networks. These network models have the problem of weak data processing capabilities and cannot achieve real-time compensation of wavefront distortion phase.

[0006] Therefore, designing a system that uses a wavefront-free sensing algorithm to perform real-time and accurate correction of wavefront distortion phase is a technical problem that urgently needs to be solved. Summary of the invention

[0007] In order to solve the above technical problems, the present invention provides a real-time high-precision wavefront distortion phase compensation system, which has a simple system structure, low cost, strong data processing capability of the distortion phase correction algorithm, and high sample accuracy used in network training, which is conducive to achieving real-time high-precision compensation of the wavefront distortion phase.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0009] A real-time high-precision wavefront distortion phase compensation system, comprising a deformable mirror, a beam splitter, a first focusing lens, a second focusing lens, a receiving module, a CCD camera, a U-Net convolutional neural network processing module and a wavefront reconstruction module;

[0010] The wavefront reconstruction module generates a corresponding driving voltage according to the wavefront distortion phase information extracted by the U-Net network processing module. The deformable mirror deforms under the control of the driving voltage to correct the distortion wave and output a modulated light signal. The modulated light signal corrected by the deformable mirror passes through the beam splitter, one way through the first focusing lens to reach the receiving module, and the other way through the second focusing lens to be collected by the CCD camera. The light intensity image data collected by the CCD camera is input into the U-Net network processing module. The U-Net network processing module uses the sample error loss function composed of the predicted distortion phase information and the actual distortion phase information to iteratively update the network parameters until the set maximum number of iterations K is reached and the iterative update is stopped.

[0011] The sample error loss function of the iterative update of the U-Net network processing module is:

[0012]

[0013]

[0014]

[0015] Where the evaluation index is j, Relu(x)=max(0,x); p (j) Indicates the actual distortion phase; represents the predicted distortion phase;

[0016] Predicted distortion phase The specific expression is:

[0017]

[0018] The smaller the error, the closer the predicted distortion phase is to the actual distortion phase. Finally, the average error of all samples in the training set is used to measure the quality of the network model prediction. The specific expression is:

[0019]

[0020] Where m is the number of samples.

[0021] A further improvement of the technical solution of the present invention is that the initial driving voltage vector output by the wavefront reconstruction module is set to The driving voltage vector is iteratively updated using formula (6):

[0022] v (k+1) =v (k) +γΔv (k) ΔJ (k) (6)

[0023] Among them, v (k+1) 、v (k) are the driving voltage vectors obtained from the k+1th iteration and the kth iteration respectively; γ is the positive gain coefficient; is the disturbance voltage vector generated at the kth iteration, Δv (k) Each element in follows a Bernoulli distribution, and the absolute value of each element is fixed, and the probability of taking a positive or negative sign is 1 / 2; ΔJ (k) is the change of performance index J, which is calculated according to formula (7):

[0024]

[0025] are the performance index changes in the positive and negative directions, respectively, calculated according to formulas (8) and (9):

[0026]

[0027]

[0028] Where J[v (k) +Δv (k) ] and J[v (k) -Δv (k) ] are the performance index objective functions when the voltage changes in the positive and negative directions respectively; J[v (k) ] is the performance indicator objective function of the driving voltage vector of the kth iteration; the deformable mirror deforms according to the driving voltage after the iteration, thereby compensating for the distortion phase of the wavefront.

[0029] A further improvement of the technical solution of the present invention is that the deformable mirror adopts a deformable mirror with 19 units, 21 units, 32 units or 45 units.

[0030] A further improvement of the technical solution of the present invention is that: the number of samples m≥70000.

[0031] A further improvement of the technical solution of the present invention is that: the network parameters are the convolution kernel parameters and scalar deviation of the network.

[0032] A further improvement of the technical solution of the present invention is that: the maximum number of iterations K≥4000.

[0033] A further improvement of the technical solution of the present invention is that the U-Net network processing module is pre-trained using a large number of precise samples and optimized using a residual block, wherein the large number of precise samples come from a precise atmospheric turbulence simulation model simulated based on a power spectrum inversion method.

[0034] A further improvement of the technical solution of the present invention is that: 1V-1.5V.

[0035] A further improvement of the technical solution of the present invention is that the gain coefficient γ is 1.2-1.6.

[0036] A further improvement of the technical solution of the present invention is that: It is 0.2V-0.3V.

[0037] Due to the adoption of the above technical solution, the technical progress achieved by the present invention is:

[0038] The present invention uses a U-Net network-based closed-loop correction system without wavefront sensors to accurately compensate for the wavefront distortion phase in real time; the CCD camera is responsible for collecting light intensity information (usually the light spot brightness value or the light spot energy distribution diagram), which is easier to obtain than the wavefront aberration; the U-Net network and the wavefront reconstruction module generate corresponding driving voltages according to the collected light intensity information to control the deformation of the deformable mirror and generate corresponding compensation phase; in the process of distortion phase compensation, there is no need to use a wavefront sensor, but a U-Net network trained with massive precise samples is used to invert the phase distortion of the light beam through the light intensity information and realize distortion phase compensation, which has the advantages of small size, low cost and simple structure; at the same time, because the U-Net network has powerful real-time data processing capabilities, the present invention can realize real-time and high-precision compensation of the wavefront distortion phase. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is the overall system block diagram of the present invention;

[0040] Figure 2 It is the distribution diagram of the 32-unit deformable mirror driver of the present invention;

[0041] Figure 3Schematic diagram of a U-Net network in which a residual block is added in the present invention;

[0042] Among them, 1. deformable mirror, 2. beam splitter, 3. first focusing lens, 4. second focusing lens, 5. receiving module, 6. CCD camera, 7. U-Net convolutional neural network processing module, 8. wavefront reconstruction module. DETAILED DESCRIPTION

[0043] In order to make the technical solution, advantages and purposes of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and a detailed implementation method and a detailed operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

[0044] Combination Figure 1 to Figure 3 The real-time high-precision wavefront distortion phase compensation system of the present invention comprises a deformable mirror 1, a beam splitter 2, focusing lenses 3 and 4, a receiving module 5, a CCD camera 6, a U-Net convolutional neural network processing module 7, and a wavefront reconstruction module 8, wherein the CCD camera 6 is a CCD scientific research-level camera. The wavefront reconstruction module 8 generates a corresponding driving voltage according to the wavefront distortion phase information extracted by the U-Net network processing module 7, and the deformable mirror 1 deforms under the control of the driving voltage to correct the distortion wave and output a modulated light signal; the modulated light signal corrected by the deformable mirror passes through the beam splitter 2 through the first focusing lens 3 to reach the receiving module 5, and passes through the second focusing lens 4 to be collected by the CCD camera 6; the light intensity image data collected by the CCD camera 6 is input into the U-Net network processing module 7; the U-Net network processing module 7 uses the sample error loss function composed of the predicted distortion phase information and the actual distortion phase information to iteratively update the network parameters until the set maximum number of iterations K is reached, and the iterative update is stopped, K≥4000.

[0045] The deformable mirror 1 can be a 19-unit, 21-unit, 32-unit, 45-unit, etc. deformable mirror. Figure 2 The 32-unit deformable mirror shown.

[0046] Figure 3 The U-Net convolutional neural network shown in the figure inputs the wavefront distortion phase information collected by a scientific research-grade CCD camera. After the input distortion phase information passes through a series of downsampling and convolution encoding and a series of upsampling and deconvolution decoding processes, the turbulence distortion phase information extracted by the network is finally output. The U-Net network with the addition of residual blocks can increase the number of network layers, and the middle copy jump connection layer is used to improve the ability to extract network features. The present invention uses a large number of accurate samples provided by the atmospheric turbulence model simulated based on the power spectrum inversion method to train the network, thereby improving the accuracy of distortion phase compensation.

[0047] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A real-time high-precision wavefront distortion phase compensation system, characterized in that: It comprises a deformable mirror (1), a beam splitter (2), a first focusing lens (3), a second focusing lens (4), a receiving module (5), a CCD camera (6), a U-Net convolutional neural network processing module (7) and a wavefront reconstruction module (8); The wavefront reconstruction module (8) generates a corresponding driving voltage according to the wavefront distortion phase information extracted by the U-Net convolutional neural network processing module (7); the deformable mirror (1) deforms under the control of the driving voltage to correct the distortion wave and output a modulated light signal; the modulated light signal corrected by the deformable mirror (1) passes through the beam splitter (2), one path passes through the first focusing lens (3) to reach the receiving module (5), and the other path passes through the second focusing lens (4) and is collected by the CCD camera (6); the light intensity image data collected by the CCD camera (6) is input into the U-Net convolutional neural network processing module (7); the U-Net convolutional neural network processing module (7) uses the sample error loss function composed of the predicted distortion phase information and the actual distortion phase information to iteratively update the network parameters until the set maximum number of iterations K is reached and the iterative update is stopped; The sample error loss function of the iterative update of the U-Net convolutional neural network processing module (7) is: Where the evaluation index is j, Relu(x)=max(0,x); p (j) Indicates the actual distortion phase; represents the predicted distortion phase; Predicted distortion phase The specific expression is: The smaller the error, the closer the predicted distortion phase is to the actual distortion phase. Finally, the average error of all samples in the training set is used to measure the quality of the network model prediction. The specific expression is: Where m is the number of samples.

2. A real-time high-precision wavefront distortion phase compensation system according to claim 1, characterized in that: The initial driving voltage vector output by the wavefront reconstruction module (8) is set to The driving voltage vector is iteratively updated using formula (6): v (k+1) =v (k) +γΔv (k) ΔJ (k) (6), Among them, v (k+1) 、v (k) are the driving voltage vectors obtained from the k+1th iteration and the kth iteration respectively; γ is the positive gain coefficient; is the disturbance voltage vector generated at the kth iteration, Δv (k) Each element in follows a Bernoulli distribution, and the absolute value of each element is fixed, and the probability of taking a positive or negative sign is 1 / 2; ΔJ (k) is the change of performance index J, which is calculated according to formula (7): are the performance index changes in the positive and negative directions, respectively, calculated according to formulas (8) and (9): Where J[v (k) +Δv (k) ] and J[v (k) -Δv (k) ] are the performance index objective functions when the voltage changes in the positive and negative directions respectively; J[v (k) ] is the performance indicator objective function of the driving voltage vector of the kth iteration; the deformable mirror (1) deforms according to the driving voltage after the iteration, thereby compensating for the distortion phase of the wavefront.

3. A real-time high-precision wavefront distortion phase compensation system according to claim 1, characterized in that: The deformable mirror (1) is a deformable mirror with 19 units, 21 units, 32 units or 45 units.

4. The real-time high-precision wavefront distortion phase compensation system according to claim 1, characterized in that: The sample number m≥70000.

5. The real-time high-precision wavefront distortion phase compensation system according to claim 1, characterized in that: The network parameters are the convolution kernel parameters and scalar deviation of the network.

6. The real-time high-precision wavefront distortion phase compensation system according to claim 1, characterized in that: The maximum number of iterations K≥4000.

7. The real-time high-precision wavefront distortion phase compensation system according to claim 1, characterized in that: The U-Net convolutional neural network processing module is pre-trained using a large number of precise samples and optimized using a residual block. The large number of precise samples comes from a precise atmospheric turbulence simulation model simulated based on a power spectrum inversion method.

8. The real-time high-precision wavefront distortion phase compensation system according to claim 2, characterized in that: Said 1V-1.5V.

9. The real-time high-precision wavefront distortion phase compensation system according to claim 2, characterized in that: The gain coefficient γ is 1.2-1.

6.

10. The real-time high-precision wavefront distortion phase compensation system according to claim 2, characterized in that: Said It is 0.2V-0.3V.

Images