Method for Generating Atmospheric Turbulence Phase Screen Based on Deep Convolutional Generative-Adversarial Networks

Zeyang Wang; Yue Zhu; Yan An

doi:10.3788/LOP232738

Laser & Optoelectronics Progress, Volume. 61, Issue 21, 2101001(2024)

Method for Generating Atmospheric Turbulence Phase Screen Based on Deep Convolutional Generative-Adversarial Networks

Zeyang Wang^1,2, Yue Zhu³, and Yan An^1,2、*

Author Affiliations

¹School of Optoelectronic Engineering, Changchun University of Science and Technology, Changchun 130022, Jilin , China

²Institute of Space Optoelectronics Technology, Changchun University of Science and Technology, Changchun 130022, Jilin , China

³College of Exploration and Geomatics Engineering, Changchun Institute of Technology, Changchun 130021, Jilin , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(15)

Fig. 1. Random phase screens for each turbulence intensity. (a) (b) $C_{n}^{2} = 2.0 \times 10^{- 20}$ ; (c) (d) $C_{n}^{2} = 2.0 \times 10^{- 15}$ ; (e) (f) $C_{n}^{2} = 2.0 \times 10^{- 13}$

Download full size

Fig. 2. Phase screen images when atmospheric turbulence intensity of $2.0 \times 10^{- 20}$ . (a) Plane image; (b) 3D image

Download full size

Fig. 3. Network structure diagram

Download full size

Fig. 4. Generator structure diagram

Download full size

Fig. 5. Discriminator structure diagram

Download full size

Fig. 6. Loss function of generator and discriminator during training process

Download full size

Fig. 7. Random atmospheric turbulence phase screen images corresponding to training once, 15000 times, and 33000 times

Download full size

Fig. 8. Structure function curve of atmospheric turbulence random phase screen

Download full size

Fig. 9. Atmospheric turbulence phase screen images. (a) Phase screen plane image and (b) 3D image generated by DCGAN model; (c) phase screen plane image and (d) 3D image generated by numerical simulation method

Download full size

Fig. 10. RMSE variation of DCGAN corresponding model prediction values

Download full size

Fig. 11. Standard deviation line graph of grayscale value of turbulent phase screen at different training times. (a) 1 epoch; (b) 200 epochs; (c) 400 epochs

Download full size

Table 1. Specific parameters of random phase screen
View table
Table 1. Specific parameters of random phase screen
Parameter Simulation value
Number of grid elements 256
Length z /m 1000
Grid spacing About 0.0023
Waist radius ω₀ /cm 2
Width of phase screen L /m 0.6
Number of phase screens 1
$C_{n}^{2} / m^{- 2 / 3}$ $1.0 \times 10^{- 20} - 1.0 \times 10^{- 13}$

Table 2. Specific parameters of generator

View table

Table 2. Specific parameters of generator

Generator	Convolutional kernel size and stride	Output size
Input		100×1
Reshap and project		4×4×1024
ConvT2d	4×4×2	8×8×512
BatchNormal		8×8×512
ReLU		8×8×512
ConvT2d	4×4×2	16×16×256
BatchNormal		16×16×256
ReLU		16×16×256
ConvT2d	4×4×2	32×32×128
BatchNormal		32×32×128
ReLU		32×32×128
ConvT2d	4×4×2	64×64×3
BatchNormal		64×64×3
Tanh		64×64×3

Table 3. Specific parameters of discriminator

View table

Table 3. Specific parameters of discriminator

Discriminator	Convolutional kernel size and stride	Output size
Input		64×64×3
ConvT2d	4×4×2	32×32×128
BatchNormal		32×32×128
LeakyReLU		32×32×128
ConvT2d	4×4×2	16×16×256
BatchNormal		16×16×256
LeakyReLU		16×16×256
ConvT2d	4×4×2	8×8×512
BatchNormal		8×8×512
LeakyReLU		8×8×512
ConvT2d	4×4×2	4×4×1024
BatchNormal		4×4×1024
LeakyReLU		4×4×1024
FC	（4×4×1024，1）	1×1
Sigmoid		1×1

Table 4. Calculation time for generating atmospheric turbulence phase screens

View table

Table 4. Calculation time for generating atmospheric turbulence phase screens

Method	Calculation time /s
Method	Number of 1	Number of 50	Number of 100	Number of 250	Number of 500	Number of 750	Number of 1000	Number of 2000
Numerical simulation	0.0912	4.6451	5.9864	16.3848	29.8978	45.5753	63.1894	126.3965
DCGAN	0.2566	0.7580	1.8719	2.8301	12.1570	12.5023	17.5682	39.4204

Tools

Get Citation

Copy Citation Text

Zeyang Wang, Yue Zhu, Yan An. Method for Generating Atmospheric Turbulence Phase Screen Based on Deep Convolutional Generative-Adversarial Networks[J]. Laser & Optoelectronics Progress, 2024, 61(21): 2101001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites