CCNN algorithm-based AO system Shack-Hartmann sensor optimization method

By introducing CCNN algorithm into the Shackhartmann sensor, combining conformal mapping and convolutional neural network, the problem of low spot center of mass positioning accuracy is solved, and the mixing efficiency and communication performance of the FSOC system are improved.

CN118971981APending Publication Date: 2024-11-15JILIN UNIVERSITY
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410987441.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2024-07-09
Filing Date
2024-07-23
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing Shackhartmann sensor spot centroid positioning method has the problems of low positioning accuracy and poor reconstruction effect, especially in the case of spots with uneven energy distribution.

Method used

The Xiake Hartmann sensor optimization method of AO system based on CCNN algorithm is adopted to extract image features through convolutional layers and predict spot center of mass coordinates in combination with full connection layers. The spot map is preprocessed using conformal mapping to enhance the learning efficiency of neural networks. , thereby improving the positioning accuracy of the center of mass of the spot.

Benefits of technology

The wavefront reconstruction capability is improved, the mixing efficiency of the FSOC system is improved, the bit error rate is reduced, and the communication performance of the FSOC system is significantly improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118971981A_ABST
    Figure CN118971981A_ABST
Patent Text Reader

Abstract

The invention relates to an AO system Shack-Hartmann sensor optimization method based on a CCNN algorithm, and belongs to the technical field of free space optical communication. The objective of the invention is to solve the problems of low centroid positioning precision and poor reconstruction effect of an existing Shack-Hartmann sensor light spot centroid positioning method. The method comprises the following steps: step 1, taking a CCNN algorithm as an SHWFS light spot centroid prediction algorithm; and step 2, applying the SHWFS after performance optimization to an AO system to correct wavefront aberration generated by atmospheric turbulence interference of an FSOC system. According to the CCNN algorithm-based AO system Shack-Hartmann sensor optimization method, the CCNN algorithm with higher light spot mass center position prediction precision is applied to the FSOC system. Simulation results show that the algorithm enables the Shack-Hartmann sensor to have better wavefront reconstruction capability, the mixing efficiency of the FSOC system is improved, and the bit error rate is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of free space optical communication, and in particular to an optimization method of a Shack-Hartmann sensor of an AO system based on a CCNN algorithm. Background Art

[0002] Free Space Optical Communication (FSOC) is a wireless communication method that uses laser as a carrier and the atmosphere as a channel. FSOC has the advantages of high confidentiality, strong anti-electromagnetic interference ability, higher channel utilization and faster transmission rate, and is one of the important research hotspots in the field of wireless communication. FSOC system is widely used in military and civil fields such as satellite-to-ground communication, inter-satellite communication, border and coastal defense, "the last mile", emergency relief communication, etc.

[0003] At present, in the FSOC system, compared with the traditional FSOC system using intensity modulation / direct detection, the coherent FSOC (Coherent Free Space Optical Communication, CFSOC) system has become one of the technologies widely studied and applied in the FSOC field due to its advantages such as flexible modulation, high detection sensitivity and good channel selectivity. However, when the laser signal is transmitted through the atmospheric channel, it will be affected by the atmospheric turbulence and cause the light beam to produce wavefront aberration, divergence, drift and other phenomena, which ultimately leads to a decrease in the mixing efficiency (ME) of the FSOC, an increase in the bit error rate (BER), and a significant decrease in communication performance.

[0004] Adaptive optics (AO) systems are used at the receiving end of the FSOC system to correct the effects of atmospheric turbulence in real time. A typical AO system consists of three parts: a wavefront sensor, a wavefront controller, and a wavefront corrector. Among them, the wavefront sensor is responsible for real-time detection of the dynamic incident light wavefront distortion information; the wavefront controller is responsible for quantitatively calculating the control signal based on the wavefront distortion information, and driving the controlled optical active element to produce a conjugate phase to compensate for the wavefront distortion; the wavefront corrector is an actuator, which is responsible for real-time dynamic correction of the wavefront distortion according to the control signal. After correction by the AO system, a plane wave can be obtained, the wavefront error at the receiving end will be greatly reduced, and the beam quality will be significantly improved. Since the wavefront recovery process of the AO system without a wavefront sensor is poorly stable, and there are certain limitations in its practical application in complex systems. Therefore, in actual communications, the AO system based on a wavefront sensor is mainly used to compensate for aberrations. The Shack-Hartmann wavefront sensor (SHWFS) is a traditional pupil plane wavefront sensor. Its simple structure and strong adaptability make it widely used in AO systems. SHWFS is mainly composed of a microlens array and a camera. The light beam passing through the microlens array will be focused on the camera to form a focus point array. Due to wavefront distortion, the focus in each sub-aperture will deviate from the ideal position. The offset between the centroid position of the light spot in the sub-aperture and the reference centroid position in the horizontal and vertical directions is measured, and then the wavefront is reconstructed. Therefore, the accuracy of SHWFS in locating the centroid position of the light spot directly determines the working performance of the entire FSOC system. The wavefront distortion information measured by SHWFS is fed back to the wavefront correction system to compensate for the distortion of the wavefront to be measured, thereby effectively reducing the influence of turbulence effects and compensating for the light wave phase disturbance caused by atmospheric turbulence. However, the existing centroid method has low positioning accuracy for the centroid of the light spot, poor effect for spots with uneven energy distribution, and poor applicability. Summary of the invention

[0005] The present invention aims to solve the technical problems of low centroid positioning accuracy and poor reconstruction effect in the existing Shack-Hartmann sensor spot centroid positioning method, and provides an AO system Shack-Hartmann sensor optimization method based on CCNN algorithm.

[0006] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0007] A Shack-Hartmann sensor optimization method for an AO system based on a CCNN algorithm comprises the following steps:

[0008] Step 1: Use the CCNN algorithm as the SHWFS spot centroid prediction algorithm;

[0009] Step 2: Apply the performance-optimized SHWFS to the AO system to correct the wavefront aberration of the FSOC system caused by atmospheric turbulence.

[0010] In the above technical solution, step 1 is specifically: using the convolution layer to extract image features, the learned features are used as input to the fully connected layer, and the network output predicts the center of mass coordinates of the light spot.

[0011] In the above technical solution, step 1 specifically includes: in the angle-preserving mapping part, the traditional spot I tra Upsampling followed by conformal mapping generates I con ; In the convolutional neural network part, the generated I con Input into the network to predict the center of mass coordinates of the light spot.

[0012] In the above technical solution, step 1 is specifically as follows:

[0013] The grid (x 1 ,y 1 ) is mapped to the grid (x) on the plane Γ2 by the function F 2 ,y 2 ):

[0014] x 2 +jy 2 =F(x 1 +jy 1 )

[0015] In order to conformally map the intensity field of the spot I 0 (x 1 ,y 1 ), the mapping function F is defined as:

[0016]

[0017] F(·) maps the arc AB to the straight line A'B'; (x 1 ,y 1 ) is the light spot coordinate with the optical axis as the origin,

[0018] (x 2 ,y 2 ) is the mapping coordinate; the spot diagram of conformal mapping is expressed as follows:

[0019] I con (x 2 ,y 2 )=I 0 (exp(x 2 )cosy 2 ,exp(x 2 )siny 2 )

[0020] In order to make the conformal mapped spot image computer readable, con Discrete into I con:

[0021]

[0022] I con =(m,n)=I 0 (ξ' m,n ,η' m,n ),m=1,...,M,n=1,...,N.

[0023] Where FOV is the imaging field of view of the focusing objective, (ξ' m,n ,η' m,n ) represents the grid sampling of conformal mapping, ξ' m,n represents the real part after mapping, η' m,n represents the imaginary part after mapping, M and N represent the maximum sampling points of the sampled image length and height respectively, m represents the sampling points from 1 to M (m=1,2,...M), n represents the sampling points from 1 to N (n=1,2,...N), Re(·) represents the real part of the complex number, Im(·) represents the imaginary part of the complex number, j is the mark when the complex number is represented by an exponential function, and has no actual reference meaning; the traditional spot I tra Directly from I 0 sampling:

[0024]

[0025] I tra (m,n)=I 0 (ξ m,n ,η m,n ),m=1,...,M,n=1,...,N.

[0026] Among them, (ξ m,n ,η m,n ) represents the traditional uniform grid sampling, ξ m,n and η m,n They refer to the horizontal and vertical coordinates of the sampling respectively. M and N represent the maximum values ​​of the sampling points of the length and height of the sampled image respectively. m represents the sampling points from 1 to M (m=1,2,...M), and n represents the sampling points from 1 to N (n=1,2,...N).

[0027] In the above technical solution, after step 2, there is a further step of verifying the feasibility of the algorithm through simulation experiments.

[0028] In the above technical solution, the steps of verifying the feasibility of the algorithm through simulation experiments are specifically as follows:

[0029] The beam quality of an optical system is described using the wavefront root mean square (RMS):

[0030]

[0031] in, represents the phase of the light beam, and <·> represents the mathematical expectation;

[0032] The expression of mixing efficiency ME is:

[0033]

[0034] In the formula, A s and A o represent the amplitudes of the signal light and the local oscillator light respectively, represents the phase difference between the signal light and the local oscillator light, ∫ s ·ds is the integral symbol for performing integral operations, and cos(·) is a trigonometric function operation;

[0035] The expression of bit error rate BER is:

[0036]

[0037] In the formula, δ represents the quantum efficiency of the photon detector at the receiving end, N P represents the number of single-bit photons, and ME represents the mixing efficiency.

[0038] The present invention has the following beneficial effects:

[0039] The CCNN algorithm-based AO system Shack-Hartmann sensor optimization method of the present invention uses a CCNN algorithm with higher prediction accuracy of the spot centroid position to be applied to the FSOC system. Simulation results show that the algorithm enables the Shack-Hartmann sensor to have better wavefront reconstruction capability, improves the mixing efficiency of the FSOC system and reduces the bit error rate.

[0040] The Shack-Hartmann sensor optimization method for AO system based on CCNN algorithm of the present invention introduces CCNN algorithm as the centroid prediction algorithm of SHWFS, combines conformal mapping with CNN network by using CCNN algorithm, and predicts the centroid coordinate position of the light spot by learning a large number of light spot features efficiently; SHWFS based on CCNN algorithm is applied to AO system to measure wavefront aberration to meet the application requirements of FSOC in various scenarios. Simulation results show that CCNN algorithm can predict the centroid position more accurately, and AO system of SHWFS based on CCNN algorithm can also effectively improve the performance of FSOC system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0042] Figure 1 Schematic diagram of the FSCO system structure based on AO technology.

[0043] Figure 2 This is a schematic diagram of the working principle of SHWFS.

[0044] Figure 3 Schematic diagram of the conformal mapping principle.

[0045] Figure 4 Schematic diagram of the CCNN structure.

[0046] Figure 5 Schematic diagram of some spot images in the dataset.

[0047] Figure 6 This is a schematic diagram of the loss curve.

[0048] Figure 7 This is a schematic diagram of the rmse curve.

[0049] Figure 8 Schematic diagram of the spot centroid position error.

[0050] Fig. 9 Schematic diagram of wavefront reconstruction results; (a) is the original wavefront; (b) is the wavefront reconstructed using the centroid method; (c) is the wavefront reconstructed using CCNN.

[0051] Fig.10 Schematic diagram of the comparison of reconstructed Zenike coefficients; (a) is the comparison diagram of Zenike coefficients reconstructed using the centroid method; (b) is the comparison diagram of Zenike coefficients reconstructed using CCNN.

[0052] Fig.11 Schematic diagram of the correction results; (a) is the original wavefront; (b) is the residual wavefront obtained using the centroid method; (c) is the residual wavefront obtained using CCNN. DETAILED DESCRIPTION

[0053] The English abbreviations involved in the present invention have the following Chinese meanings:

[0054] CCNN: Conformal Convolutional Neural Network for Conformal Mapping;

[0055] AO: Ambient Occlusion, ambient light absorption;

[0056] BER: Bit Error Rate, bit error rate;

[0057] ME: Mixing Efficiency, mixing efficiency;

[0058] FSOC: Free Space Optical Communication, free space optical communication;

[0059] SHWFS: Shack-Hartmann Wavefront Sensor, Hartmann wavefront detector;

[0060] RMS: Root Mean Square, wavefront root mean square;

[0061] FOV: Field Of View, field of view angle, imaging field of view of the focusing objective lens;

[0062] CCD: Charge Coupled Device, charge coupled device;

[0063] Zernike: Zernike polynomials;

[0064] LO: Local Oscillator, local oscillator, local oscillator;

[0065] MATLAB: Matrix&Laboratory, matrix factory (matrix laboratory), commercial mathematical software produced by MathWorks, an American company.

[0066] The inventive concept of the present invention is:

[0067] The CCNN algorithm-based Shack-Hartmann sensor optimization method for AO system of the present invention introduces the CCNN algorithm to predict the spot position of SHWFS, and applies the improved SHWFS in the AO system to measure the wavefront aberration caused by atmospheric turbulence, so as to further improve the performance of the FSOC system. The CCNN algorithm uses conformal mapping to preprocess the spot image, which can effectively learn more abundant spot features after preprocessing and enhance the learning efficiency of the neural network. The BER and ME performance in the FSOC design process are studied. The simulation results show that compared with the SHWFS using the centroid method, the SHWFS using the CCNN algorithm has a better effect of measuring wavefront distortion, can improve the correction ability of AO, and make the FSOC system perform better.

[0068] The CCNN algorithm-based AO system Shack-Hartmann sensor optimization method of the present invention comprises the following steps:

[0069] Step 1: Use the CCNN algorithm as the SHWFS spot centroid prediction algorithm;

[0070] Step 2: Apply the performance-optimized SHWFS to the AO system to correct the wavefront aberration of the FSOC system caused by atmospheric turbulence.

[0071] In step 1, the image features are extracted by using the convolution layer, and the learned features are used as input to the fully connected layer, and the network outputs the predicted center of mass coordinates of the light spot. tra Upsampling followed by conformal mapping generates I con ; In the convolutional neural network part, the generated I con Input into the network to predict the center of mass coordinates of the light spot.

[0072] Step 1 is as follows:

[0073] The grid (x 1 ,y 1 ) is mapped to the grid (x) on the plane Γ2 by the function F 2 ,y 2 ):

[0074] x 2 +jy 2 =F(x 1 +jy 1 )

[0075] In order to conformally map the intensity field of the spot I 0 (x 1 ,y 1 ), the mapping function F is defined as:

[0076]

[0077] F(·) can map the arc AB into the straight line A'B'; (x 1 ,y 1 ) is the light spot coordinate with the optical axis as the origin, (x 2 ,y 2 ) is the mapping coordinate; the spot diagram of conformal mapping is expressed as follows:

[0078] I con (x 2 ,y 2 )=I 0 (exp(x 2 )cosy 2 ,exp(x 2 )siny 2 )

[0079] In order to make the conformal mapped spot image computer readable, con Discrete into I con :

[0080]

[0081] I con =(m,n)=I 0 (ξ'm,n ,η' m,n ),m=1,...,M,n=1,...,N.

[0082] Where FOV is the imaging field of view of the focusing objective, (ξ' m,n ,η' m,n ) represents the grid sampling of conformal mapping, ξ' m,n represents the real part after mapping, η' m,n Represents the imaginary part after mapping, M and N represent the maximum sampling points of the sampled image length and height respectively, m represents the sampling points from 1 to M (m=1,2,...M), n represents the sampling points from 1 to N (n=1,2,...N), Re(·) represents the real part of the complex number, Im(·) represents the imaginary part of the complex number, j is the mark when the complex number is represented by an exponential function, and has no actual reference meaning. Traditional spot I tra Directly from I 0 sampling:

[0083]

[0084] Among them, (ξ m,n ,η m,n ) represents the traditional uniform grid sampling, ξ m,n and η m,n They refer to the horizontal and vertical coordinates of the sampling respectively. M and N represent the maximum values ​​of the sampling points of the length and height of the sampled image respectively. m represents the sampling points from 1 to M (m=1,2,...M), and n represents the sampling points from 1 to N (n=1,2,...N).

[0085] After step 2, the method further includes the following steps: verifying the feasibility of the algorithm through simulation experiments, specifically:

[0086] The beam quality of an optical system is described using the wavefront root mean square (RMS):

[0087]

[0088] in, represents the phase of the light beam, and <·> represents the mathematical expectation;

[0089] The expression of mixing efficiency ME is:

[0090]

[0091] In the formula, A S and A O represent the amplitudes of the signal light and the local oscillator light respectively, represents the phase difference between the signal light and the local oscillator light, ∫ s ·ds is the integral symbol for performing integral operations, and cos(·) is a trigonometric function operation.

[0092] The expression of bit error rate BER is:

[0093]

[0094] In the formula, δ represents the quantum efficiency of the photon detector at the receiving end, N P represents the number of single-bit photons, and ME represents the mixing efficiency.

[0095] The Shack-Hartmann sensor optimization method for an AO system based on a CCNN algorithm of the present invention is described in detail below in conjunction with the accompanying drawings.

[0096] 1.FSOC system

[0097] Figure 1 Schematic diagram of a free-space optical communication system with a closed-loop AO system. The information waveform is modulated onto an optical carrier by amplitude, frequency or phase, and the optical signal is transmitted to the receiving end through the atmospheric channel. At the receiving end, the AO system corrects the wavefront distortion caused by atmospheric turbulence. The corrected optical signal is mixed with the local oscillator (LO) light, and then converted into an electrical signal through a photodetector for output.

[0098] Among them, the AO system is mainly composed of a wavefront sensor, a wavefront controller and a wavefront corrector. The wavefront sensor detects wavefront distortion in real time, and the wavefront controller processes the detection signal fed back by the wavefront sensor and generates a control signal. Finally, the wavefront corrector generates the corresponding wavefront compensation amount according to the control signal to correct the wavefront aberration in real time. After correction by the AO system, the performance of the FSOC system will be significantly improved. In the AO system, the measurement accuracy of the wavefront sensor on the wavefront aberration will affect the final correction effect, so the wavefront sensor is crucial in this process.

[0099] 2. SHWFS working principle

[0100] The most widely used wavefront sensor with the best engineering effect is SHWFS. Its working principle is as follows: Figure 2As shown. SHWFS consists of a set of two-dimensional microlens arrays and an image sensor located on the focal plane of the microlens. The microlens array can divide the incident wavefront into several sub-wavefront arrays. For an ideal plane wave, its sub-wavefront spot imaging will be focused on the center position of the sub-aperture, while the actual focusing position of the sub-wavefront spot of the incident distorted wave will deviate from the ideal center position. Therefore, by measuring the deviation information of the spot array and the center position of each sub-aperture, the relationship between the sub-spots of all sub-apertures and the wavefront gradient is calculated jointly, and the least squares method is used to solve it to achieve the restoration of the complete incident wavefront phase information. Therefore, the positioning accuracy of the spot centroid position affects the measurement accuracy of the wavefront distortion information, which also determines the wavefront correction capability of the AO system.

[0101] 3. Performance evaluation indicators

[0102] In the AO system, the wavefront root mean square (RMS) can be used to describe the beam quality of the optical system. RMS is an indicator that calculates the deviation from the ideal wavefront based on the current wavefront information. Its formula is:

[0103]

[0104] in, represents the phase of the light beam, and <·> represents the mathematical expectation.

[0105] The FSOC system performance mainly uses two indicators, Mixing Efficiency (ME) and Bit Error Rate (BER), to describe the quality of optical communication.

[0106] The expression of mixing efficiency ME is:

[0107]

[0108] In the formula, A s and A o represent the amplitudes of the signal light and the local oscillator light respectively, represents the phase difference between the signal light and the local oscillator light, ∫ s ·ds is the integral symbol for integration operation, and cos(·) is a trigonometric function operation. The mixing efficiency can reflect the matching degree between the signal light and the local oscillator light. The larger the ME, the smaller the wavefront phase distortion of the signal light, and the higher the FSOC communication quality, and vice versa.

[0109] The expression of bit error rate BER is:

[0110]

[0111] In the formula, δ represents the quantum efficiency of the photon detector at the receiving end, N P Represents the number of single-bit photons. BER is one of the important indicators for evaluating communication transmission efficiency. Due to the influence of atmospheric turbulence, the laser will produce beam tilt and phase distortion during transmission. This will increase the bit error rate BER of the FSOC system and reduce the reliability of communication. Therefore, the lower the BER, the higher the communication quality of the system.

[0112] 4.CCNN Algorithm

[0113] The main process of CNN-based spot centroid prediction is to use the convolution layer to extract image features, and the learned features are used as input to the fully connected layer, and finally the network outputs the predicted coordinates. The effective extraction of features is crucial to the overall performance of the network. Since the light spot has a circular feature, combining the conformal mapping algorithm with the CNN neural network to form a conformal convolutional neural network (CCNN) can rectangularize the circular features of the light spot, so that the convolution filter can learn the features of the light spot more effectively.

[0114] A conformal mapping can be considered as a complex projection between two subsets of the complex plane. 1 ,y 1 ) is mapped to the grid (x) on the plane Γ2 by the function F 2 ,y 2 ), the formula is:

[0115] x 2 +jy 2 =F(x 1 +jy 1 ) (4)

[0116] In order to conformally map the intensity field of the spot I 0 (x 1 ,y 1 ), the mapping function F is defined as:

[0117]

[0118] F(·) can map the arc AB to the straight line A'B'. The mapping process is as follows: Figure 3 As shown. (x 1 ,y 1 ) is the light spot coordinate with the optical axis as the origin, (x 2 ,y 2 ) is the mapping coordinate. The spot diagram of conformal mapping is shown as follows:

[0119] I con (x 2 ,y 2 )=I 0 (exp(x 2 )cosy2 ,exp(x 2 )siny 2 ) (6)

[0120] In order to make the conformal mapped spot image computer readable, con Discrete into I con , the formula is as follows:

[0121]

[0122] Where FOV is the imaging field of view of the focusing objective, (ξ' m,n ,η' m,n ) represents the grid sampling of conformal mapping, ξ' m,n represents the real part after mapping, η' m,n Represents the imaginary part after mapping, M and N represent the maximum values ​​of the sampling points of the sampled image length and height respectively, m represents the sampling points from 1 to M (m=1,2,...M), n represents the sampling points from 1 to N (n=1,2,...N), Re(·) represents the real part of the complex number, Im(·) represents the imaginary part of the complex number, j is the identifier when the complex number is represented by an exponential function, and has no actual reference meaning.

[0123] In contrast, the traditional spot I tra is directly obtained from I 0 The sampling formula is as follows:

[0124]

[0125] Among them, (ξ m,n ,η m,n ) represents the traditional uniform grid sampling, ξ m,n and η m,n They refer to the horizontal and vertical coordinates of the sampling respectively. M and N represent the maximum values ​​of the sampling points of the length and height of the sampled image respectively. m represents the sampling points from 1 to M (m=1,2,...M), and n represents the sampling points from 1 to N (n=1,2,...N).

[0126] The working principle of conformal convolutional neural network (CCNN) based on conformal mapping is as follows Figure 4 In the conformal mapping part, firstly tra Upsampling followed by conformal mapping generates I con In the convolutional neural network part, the generated I con Input into the network to predict the center of mass coordinates of the light spot.

[0127] 5. Create a Dataset

[0128] In the FSOC system, Zernike polynomials are currently commonly used to represent wavefront aberrations. Zernike polynomials are polynomials with a root mean square value of 1 and defined on the unit circle. Different orders show different phase characteristics and can be represented in polar coordinates or rectangular coordinates. Table 1 shows the first 15 Zernike polynomials.

[0129] Table 1 Coordinate representation and physical meaning of the first 15 Zernike polynomials

[0130]

[0131] Under different turbulence conditions, the light beam is affected differently during the transmission of the system. According to the Shack-Hartmann working principle, the light spot formed by the disturbed light beam through the CCD sub-aperture is also different. Therefore, in the present invention, 4-36 order Zernike coefficients are used to simulate different turbulence conditions, and MATLAB is used to collect the light spot diagrams under different turbulence conditions to form the required data set. The data set consists of 17040 light spot diagrams, of which 14200 are used as training sets and 2840 are used as test sets. Figure 5 A portion of the dataset is shown.

[0132] 6. Simulation Experiment

[0133] In order to verify the feasibility of the CCNN algorithm, a series of simulation experiments were carried out. Figure 6 and Figure 7 The loss curve and rmse curve of the CCNN network are respectively. In order to verify the accuracy of the CCNN algorithm in predicting the center of mass position of the light spot, 20 groups of pictures were randomly selected from the test set to predict the center of the light spot. By comparing the errors of the center of mass position of the light spot of the centroid method and the CCNN method, it can be seen that the CCNN method has better prediction effect and higher center of mass positioning accuracy. The comparison results are shown in Figure 8 shown.

[0134] In order to verify the improvement of SHWFS reconstruction effect by CCNN algorithm, we randomly selected a set of wavefront aberrations in the test set. The initial aberration map is as follows: Fig. 9 As shown in (a), the result of SHWFS reconstructing the wavefront using the centroid method is as follows Fig. 9 As shown in (b), the result of SHWFS reconstructing the wavefront using the CCNN method is as follows Fig. 9 As shown in (c), it can be seen that the CCNN method has a certain improvement in the reconstruction ability of SHWFS.

[0135] To further demonstrate the reconstruction effect, Fig.10The Zenike coefficients reconstructed using different methods are compared with the original Zenike coefficients. The Zenike coefficients reconstructed by the CCNN method have a smaller error than the initial comparison, which shows that the CCNN method has a better reconstruction effect than the centroid method and is more suitable for application in SHWFS.

[0136] The wavefront correction performance of the AO system and the communication performance of the FSOC system after using the CCNN algorithm are further tested. Fig.11 (a) is the original wavefront, and the wavefront RMS value is 1.63; Fig.11 (b) is the residual wavefront obtained after correction of the AO system using the centroid method. The RMS value of the wavefront is 0.41. Fig.11 (c) is the residual wavefront obtained after correction of the AO system using the CCNN algorithm, and the wavefront RMS value is 0.04. RMS can represent the index of the offset of the current wavefront information calculation relative to the ideal wavefront. After using the CCNN algorithm, the RMS of the residual wavefront obtained by the AO system after correction is lower, so the system correction ability is better for the centroid method. Table 2 shows the test results of the communication performance indicators of the FSOC system. After correction using the CCNN algorithm, the ME is increased from 0.07 of the original wavefront to 0.998, while after correction using the centroid method, the ME can only be increased to 0.845. By comparing the BER of the system, it can be seen that the BER of the uncorrected original wavefront is far from meeting the communication requirements. After correction, the BER has been improved to a certain extent. Compared with the centroid method, the CCNN algorithm reduces the BER of the system by 1e orders of magnitude.

[0137] The simulation results show that the CCNN algorithm can predict the center of mass position more accurately, which improves the reconstruction capability of SHWFS and the wavefront correction capability of the AO system to a certain extent, and effectively ensures the communication quality of the FSOC system.

[0138] Table 2 Communication performance evaluation indicators

[0139]

[0140] The CCNN algorithm-based AO system Shack-Hartmann sensor optimization method of the present invention uses a CCNN algorithm with higher prediction accuracy of the spot centroid position to be applied to the FSOC system. Simulation results show that the algorithm enables the Shack-Hartmann sensor to have better wavefront reconstruction capability, improves the mixing efficiency of the FSOC system and reduces the bit error rate.

[0141] The Shack-Hartmann sensor optimization method for AO system based on CCNN algorithm of the present invention introduces CCNN algorithm as the centroid prediction algorithm of SHWFS, combines conformal mapping with CNN network by using CCNN algorithm, and predicts the centroid coordinate position of the light spot by learning a large number of light spot features efficiently; SHWFS based on CCNN algorithm is applied to AO system to measure wavefront aberration to meet the application requirements of FSOC in various scenarios. Simulation results show that CCNN algorithm can predict the centroid position more accurately, and AO system of SHWFS based on CCNN algorithm can also effectively improve the performance of FSOC system.

[0142] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A Shack-Hartmann sensor optimization method for AO system based on CCNN algorithm, characterized in that: The following steps are involved: Step 1: Use the CCNN algorithm as the SHWFS spot centroid prediction algorithm; Step 2: Apply the performance-optimized SHWFS to the AO system to correct the wavefront aberration of the FSOC system caused by atmospheric turbulence.

2. The AO system Shack-Hartmann sensor optimization method based on CCNN algorithm according to claim 1 is characterized in that: Step 1 is as follows: use the convolutional layer to extract image features, the learned features are used as input into the fully connected layer, and the network outputs the predicted center of mass coordinates of the light spot.

3. The AO system Shack-Hartmann sensor optimization method based on CCNN algorithm according to claim 1 is characterized in that: Step 1 is as follows: In the conformal mapping part, the traditional spot I tra Upsampling followed by conformal mapping generates I con ; In the convolutional neural network part, the generated I con Input into the network to predict the center of mass coordinates of the light spot.

4. The AO system Shack-Hartmann sensor optimization method based on CCNN algorithm according to claim 1 is characterized in that: Step 1 is as follows: Map the grid (x1, y1) on plane Γ1 to the grid (x2, y2) on plane Γ2 through function F: x2+jy2=F(x1+jy1) In order to conformally map the intensity field I0(x1,y1) of the light spot, the mapping function F is defined as: F(·) maps the arc AB into a straight line A'B'; (x1, y1) is the coordinate of the light spot with the optical axis as the origin, (x2, y2) is the mapping coordinate; the spot diagram of conformal mapping is expressed as follows: I con (x2,y2)=I0(exp(x2)cosy2,exp(x2)siny2) In order to make the conformal mapped spot image computer readable, con Discrete into I con : I con =(m,n)=I0(ξ' m,n ,η' m,n ),m=1,...,M,n=1,...,N. Where FOV is the imaging field of view of the focusing objective, (ξ' m,n ,η' m,n ) represents the grid sampling of conformal mapping, ξ' m,n represents the real part after mapping, η' m,n represents the imaginary part after mapping, M and N represent the maximum sampling points of the sampled image length and height respectively, m represents the sampling points from 1 to M, n represents the sampling points from 1 to N, Re(·) represents the real part of the complex number, Im(·) represents the imaginary part of the complex number, and j is the identifier when the complex number is represented by an exponential function; the traditional spot I tra Sampling directly from I0 via a uniform grid: I tra (m,n)=I0(ξ m,n ,η m,n ),m=1,...,M,n=1,...,N. Among them, (ξ m,n ,η m,n ) represents the traditional uniform grid sampling, ξ m,n and η m,n They refer to the horizontal and vertical coordinates of the sampling respectively. M and N represent the maximum values ​​of the sampling points of the length and height of the sampled graph respectively. m represents the sampling points from 1 to M, and n represents the sampling points from 1 to N.

5. The Shack-Hartmann sensor optimization method for AO system based on CCNN algorithm according to any one of claims 1 to 4, characterized in that: After step 2, there is another step: verifying the feasibility of the algorithm through simulation experiments.

6. The AO system Shack-Hartmann sensor optimization method based on CCNN algorithm according to claim 5 is characterized in that: The steps to verify the feasibility of the algorithm through simulation experiments are as follows: The beam quality of an optical system is described using the wavefront root mean square (RMS): in, represents the phase of the light beam, and <·> represents the mathematical expectation; The expression of mixing efficiency ME is: In the formula, A s and A o represent the amplitudes of the signal light and the local oscillator light respectively, represents the phase difference between the signal light and the local oscillator light, ∫ s ·ds is the integral symbol for performing integral operations, and cos(·) is a trigonometric function operation; The expression of bit error rate BER is: In the formula, δ represents the quantum efficiency of the photon detector at the receiving end, N P represents the number of single-bit photons, and ME represents the mixing efficiency.

Citation Information

Cited By

  • Adaptive optical remote control interaction method for quantum communication

    CN119544087A

  • An adaptive optical remote control interaction method for quantum communication

    CN119544087B