Generative adversarial network-based atmospheric turbulence effect simulation method

Abstract

The invention belongs to the technical field of atmospheric optics, and particularly relates to an atmospheric turbulence effect simulation method based on a generative adversarial network, which comprises the following steps of: S1, obtaining atmospheric turbulence phase screens with different intensities and a light intensity distribution diagram under the influence of turbulence and without turbulence by adopting numerical simulation and a Fresnel diffraction theory, and generating a grey-scale map data set; and S2, constructing a generative adversarial network and defining a loss function according to the internal relation between the light beam and the phase screen under the turbulence and no turbulence influence in the S1. According to the method, the beam intensity step diagram and the atmospheric turbulence phase screen before and after the Gaussian beam propagation are obtained by utilizing the Fresnel diffraction theory and numerical simulation, the phase screen is predicted by utilizing the data set and the training adversarial network, the atmospheric turbulence effect can be simulated more accurately and quickly, and the method has the advantages of relatively high generalization, convenience, flexibility, repeatability and the like, and is suitable for being popularized and applied. The influence of atmospheric turbulence on laser communication can be more accurately researched indoors, and the communication quality is optimized.

Description

Technical Field

[0001] The present invention relates to the field of atmospheric optics technology, and in particular to a method for simulating atmospheric turbulence effects based on a generative adversarial network. Background Art

[0002] The random fluctuations of climate factors such as temperature and air pressure in the atmosphere will lead to random fluctuations of the atmospheric refractive index, so that the optical signal will produce wavefront distortion during propagation, that is, the turbulence effect. The existence of atmospheric turbulence will inevitably affect the propagation of optical signals and reduce the transmission quality. Therefore, the influence of turbulence on the quality of light beam transmission can be improved to a certain extent by simulating the atmospheric turbulence effect. The influence of atmospheric turbulence on the quality of light beam transmission is mainly reflected in the phase. The simulation process of atmospheric turbulence can be equivalent to the generation process of the phase screen. At present, most traditional numerical simulation methods are used, but these methods still have limitations such as insufficient low-frequency information and periodicity, which leads to low accuracy of the phase screen and it is difficult to improve the influence of atmospheric turbulence on the quality of light beam transmission. Therefore, we propose a simulation method of atmospheric turbulence effect based on generative adversarial network to solve the above problems. Summary of the invention

[0003] 1. Technical issues to be solved

[0004] In view of the shortcomings of the prior art, the present invention provides a simulation method for the atmospheric turbulence effects based on a generative adversarial network, which solves the limitations of the current traditional numerical simulation methods, such as insufficient low-frequency information and periodicity, which leads to low accuracy of the phase screen and makes it difficult to improve the influence of atmospheric turbulence on the quality of light beam transmission.

[0005] (II) Technical solution

[0006] In order to achieve the above-mentioned purpose, the present invention specifically adopts the following technical solutions:

[0007] A method for simulating atmospheric turbulence effects based on a generative adversarial network comprises the following steps:

[0008] S1: Use numerical simulation and Fresnel diffraction theory to obtain atmospheric turbulence phase screens of different intensities and light intensity distribution maps under the influence of turbulence and without turbulence, and generate grayscale image data sets;

[0009] S2: Construct a generative adversarial network and define a loss function based on the intrinsic connection between the beam under turbulence and without turbulence and the phase screen in S1;

[0010] S3: Train the designed adversarial network model, fix the model parameters after continuous optimization and adjustment, and the trained model can predict the atmospheric phase screen.

[0011] Furthermore, the phase screen under different atmospheric turbulence intensities is obtained by using the fast Fourier algorithm in S1, namely:

[0012]

[0013] Among them, m,n,m′,n′=-M / 2,…,M / 2-1 are integer indices, represents the imaginary unit, Δ′=2π / D and Δ=D / M are the sampling interval of spatial frequency and spatial domain respectively, B theory (m′Δ′,n′Δ′) is the discrete power spectrum of turbulence, which can be expressed as:

[0014]

[0015] Among them, L 0 is the outer scale of turbulence, r 0 is the Fried parameter;

[0016] Using Gaussian beam as signal carrier, its light field can be expressed as:

[0017]

[0018] where w 0 is the central beam waist radius; k = 2π / λ is the wave number; λ is the beam wavelength; z R is the confocal parameter of the Gaussian beam; represents the radius of curvature of the isophase surface; ψ is the phase factor;

[0019] The effect of atmospheric turbulence on optical signals is equivalent to adding a random phase to the optical signals, which mainly affects the phase factor. Therefore, the Gaussian light field affected by turbulence can be expressed as:

[0020] E 2 (x,y,z+Δz)=E 1 (x,y,z)×FFT(exp[iφ FFT ])×FFT -1 [exp(iAΔz)]

[0021] Where exp(iAΔz) is the Fresnel diffraction transfer function, Δz is the equivalent width of the turbulent region, and z+Δz is the total link length.

[0022] Furthermore, the generative adversarial network in S2 consists of two parts: a generator and a discriminator. In the generator, the grayscale images of the beam distribution under the influence of turbulence and without turbulence are first encoded into a low-dimensional space through an encoder, and the phase features are extracted. Then, the features extracted from the transmission module are decoded by the decoder; the discriminator includes 4 convolutional layers, and the output layer of the discriminator adopts a fully connected layer, which generates a scalar value to indicate whether the input image is real or generated. The generator uses ReLU activation function for all layers, and the discriminator uses Leaky ReLU as the activation function.

[0023] Furthermore, in the encoder, a Gaussian light intensity grayscale image with an input size of 1×64×64 is passed through a series of 4×4 convolutional layers to obtain an output feature vector with a size of 512×4×4; in the decoder, a deconvolution layer is used to restore the low-level features, and the low-level features are transformed to obtain a generated image with a size of 1×64×64.

[0024] Furthermore, the loss function defined in S2 is composed of a generator loss function and a discriminator loss function, and the generator loss function is:

[0025]

[0026] Where m is the batch size, z is generated by the encoder, G(z) is the output of the generator, and D(G(z)) is the probability that the data G(z) generated by the generator is a real sample;

[0027] The discriminator loss function is:

[0028]

[0029] Among them, z′ is a sample from the true prior;

[0030] The total loss function is:

[0031]

[0032] Furthermore, in S3, the established adversarial network model is trained and the parameters are adjusted to minimize the loss function value through several iterations. After the parameters of the model are fixed, the corresponding model is generated, and the atmospheric turbulence phase screens of different intensities can be directly predicted according to the different input beam intensity distribution diagrams.

[0033] (III) Beneficial effects

[0034] Compared with the prior art, the present invention provides a method for simulating atmospheric turbulence effects based on a generative adversarial network, which has the following beneficial effects:

[0035] The present invention obtains the beam intensity step-by-step diagram and the atmospheric turbulence phase screen before and after the propagation of a Gaussian beam by utilizing the Fresnel diffraction theory and numerical simulation. By using this data set, an adversarial network is trained to predict the phase screen, which can more accurately and quickly simulate the atmospheric turbulence effect. It has the advantages of strong generalization, convenience, flexibility, and repeatability, and is convenient for more accurate indoor research on the impact of atmospheric turbulence on laser communication and optimization of communication quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of the method of the present invention;

[0037] Figure 2 This is a structural diagram of the adversarial network generator used in the present invention;

[0038] Figure 3 This is a structural diagram of the adversarial network discriminator used in the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Example

[0041] like Figure 1-3 As shown, a method for simulating atmospheric turbulence effects based on a generative adversarial network is proposed in one embodiment of the present invention, comprising the following steps:

[0042] S1: Use the fast Fourier algorithm numerical simulation and Fresnel diffraction theory to obtain the atmospheric turbulence phase screen of different intensities and the intensity distribution diagram of the propagating beam under the influence of turbulence and without turbulence, and generate a grayscale image data set;

[0043] The phase screen under different atmospheric turbulence intensities is obtained by using the fast Fourier algorithm, namely:

[0044]

[0045] The above formula can be realized by FFT. One FFT can produce two independent phase screens, one for the real part and one for the imaginary part.

[0046] Among them, m,n,m′,n′=-M / 2,...,M / 2-1 are integer indices, represents the imaginary unit, Δ′=2π / D and Δ=D / M are the sampling interval of spatial frequency and spatial domain respectively, B theory(m′Δ′,n′Δ′) is the discrete power spectrum of turbulence, which can be expressed as:

[0047]

[0048] Among them, L 0 is the outer scale of turbulence, r 0 is the Fried parameter.

[0049] Using a Gaussian beam as a signal carrier, its light field can be expressed as:

[0050]

[0051] where w 0 is the central beam waist radius; k = 2π / λ is the wave number; λ is the beam wavelength; z R is the confocal parameter of the Gaussian beam; represents the radius of curvature of the isophase surface; ψ is the phase factor;

[0052] The effect of atmospheric turbulence on optical signals is equivalent to adding a random phase to the optical signals. After propagating for a certain distance, the accumulation of phase fluctuations caused by turbulence will cause changes in the complex amplitude, resulting in distortion of the light intensity, which mainly affects the phase factor. Therefore, the Gaussian light field affected by turbulence can be expressed as:

[0053] E 2 (x,y,z+Δz)=E 1 (x,y,z)×FFT(exp[iφ FFT ])×FFT -1 [exp(iAΔz)]

[0054] Where, exp(iAΔz) is the Fresnel diffraction transfer function, Δz is the equivalent width of the turbulent region, and z+Δz is the total link length;

[0055] S2: Construct a generative adversarial network and loss function based on the intrinsic connection between the beam and the phase screen under the influence of turbulence and without turbulence in S1;

[0056] Generative adversarial network is an unsupervised learning method that can gradually learn the distribution characteristics of raw data and generate new data based on these rules. Generative adversarial network consists of two parts: generator and discriminator. In the generator, the grayscale image of the beam distribution under the influence of turbulence and without turbulence is first encoded into a low-dimensional space through an encoder, the phase features are extracted, and then the features extracted by the transmission module are decoded by the decoder. In the encoder, the input Gaussian light intensity grayscale image of size 1×64×64 is passed through a series of 4×4 convolutional layers to obtain an output feature vector with a size of 512×4×4; in the decoder, the deconvolution layer is used to restore the low-level features, and the low-level features are transformed to obtain a generated image of size 1×64×64; the discriminator contains 4 convolutional layers, and the output layer of the discriminator uses a fully connected layer, which produces a scalar value to indicate whether the input image is real or generated. The generator uses the ReLU activation function for all layers, which helps to introduce nonlinearity, allowing the model to learn more complex mapping relationships. The discriminator Leaky ReLU is used as the activation function, which can retain a certain negative gradient during the back propagation process, help deal with the gradient disappearance problem, and make the model more stable during training.

[0057] The loss function is defined to consist of the generator loss function and the discriminator loss function; the generator loss function is:

[0058]

[0059] Where m is the batch size, z is generated by the encoder, G(z) is the output of the generator, and D(G(z)) is the probability that the data G(z) generated by the generator is a real sample;

[0060] The discriminator loss function is:

[0061]

[0062] Among them, z′ is a sample from the true prior;

[0063] The total loss function is:

[0064]

[0065] S3: Train the designed adversarial network model, fix the model parameters after continuous optimization and adjustment, and the trained model can predict the atmospheric phase screen;

[0066] The established adversarial network model is trained, and the parameters are adjusted to minimize the loss function value through several iterations. After the parameters of the model are fixed, the corresponding model is generated, which can directly predict the atmospheric turbulence phase screen of different intensities according to the input different beam intensity distribution diagrams.

[0067] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for simulating atmospheric turbulence effects based on a generative adversarial network, characterized in that: The following steps are involved: S1: Use numerical simulation and Fresnel diffraction theory to obtain atmospheric turbulence phase screens of different intensities and light intensity distribution maps under the influence of turbulence and without turbulence, and generate grayscale image data sets; S2: Construct a generative adversarial network and define a loss function based on the intrinsic connection between the beam under turbulence and without turbulence and the phase screen in S1; S3: Train the designed adversarial network model, fix the model parameters after continuous optimization and adjustment, and the trained model can predict the atmospheric phase screen.

2. The method for simulating atmospheric turbulence effects based on a generative adversarial network according to claim 1, characterized in that: In S1, the fast Fourier algorithm is used to obtain the phase screen under different atmospheric turbulence intensities, namely: Among them, m,n,m′,n′=-M / 2,…,M / 2-1 are integer indices, represents the imaginary unit, Δ′=2π / D and Δ=D / M are the sampling interval of spatial frequency and spatial domain respectively, B theory (m′Δ′,n′Δ′) is the discrete power spectrum of turbulence, which can be expressed as: Among them, L0 is the outer scale of turbulence, r0 is the Fried parameter; Using Gaussian beam as signal carrier, its light field can be expressed as: where w0 is the central beam waist radius; k = 2π / λ is the wave number; λ is the beam wavelength; z R is the confocal parameter of the Gaussian beam; represents the radius of curvature of the isophase surface; ψ is the phase factor; The effect of atmospheric turbulence on optical signals is equivalent to adding a random phase to the optical signals, which mainly affects the phase factor. Therefore, the Gaussian light field affected by turbulence can be expressed as: E2(x,y,z+Δz)=E1(x,y,z)×FFT(exp[iφ FFT ])×FFT -1 [exp(iAΔz)] Where exp(iAΔz) is the Fresnel diffraction transfer function, Δz is the equivalent width of the turbulent region, and z+Δz is the total link length.

3. The method for simulating atmospheric turbulence effects based on a generative adversarial network according to claim 1, characterized in that: The generative adversarial network in S2 consists of two parts: a generator and a discriminator. In the generator, the grayscale images of the beam distribution under the influence of turbulence and without turbulence are first encoded into a low-dimensional space through an encoder, and the phase features are extracted. Then, the features extracted from the transmission module are decoded by the decoder; the discriminator includes 4 convolutional layers, and the output layer of the discriminator adopts a fully connected layer, which generates a scalar value to indicate whether the input image is real or generated. The generator uses the ReLU activation function for all layers, and the discriminator uses Leaky ReLU as the activation function.

4. The method for simulating atmospheric turbulence effects based on a generative adversarial network according to claim 3, characterized in that: In the encoder, a Gaussian light intensity grayscale image of input size 1×64×64 is passed through a series of 4×4 convolutional layers to obtain an output feature vector with a size of 512×4×4; in the decoder, a deconvolution layer is used to restore low-level features, and the low-level features are transformed to obtain a generated image of size 1×64×64.

5. The method for simulating atmospheric turbulence effects based on a generative adversarial network according to claim 3, characterized in that: The loss function defined in S2 is composed of a generator loss function and a discriminator loss function. The generator loss function is: Where m is the batch size, z is generated by the encoder, G(z) is the output of the generator, and D(G(z)) is the probability that the data G(z) generated by the generator is a real sample; The discriminator loss function is: Among them, z′ is a sample from the true prior; The total loss function is:

6. The method for simulating atmospheric turbulence effects based on a generative adversarial network according to claim 1, characterized in that: In the S3, the established adversarial network model is trained and the parameters are adjusted to minimize the loss function value through several iterations. After the parameters of the model are fixed, the corresponding model is generated, and the atmospheric turbulence phase screens of different intensities can be directly predicted according to the different input beam intensity distribution diagrams.

Images