Distorted vortex light correction system and method based on GS algorithm and deep learning

Abstract

The application discloses a distortion vortex light correction system and method based on a GS algorithm and deep learning. The vortex light distortion correction system comprises a vortex light emitting module, a turbulent transmission module, a GS algorithm iteration module, a convolutional neural network training module and a phase screen compensation module. Based on the GS algorithm, the intensity distribution of the initial light field and the distorted light field of the vortex light is used to calculate a compensation phase screen, and then the compensation phase screen and the distorted light field are used to train the convolutional neural network to form a training set. After training, the convolutional neural network learns the ability of the GS algorithm to generate the compensation phase screen, and can replace the GS algorithm to directly generate the compensation phase screen, so that the phase distortion of the vortex light beam can be corrected in real time. The application uses the trained convolutional neural network to replace the GS algorithm to repair the distorted vortex light, and the speed of generating the compensation phase screen is improved from several hundred milliseconds to tens of milliseconds.

Description

Technical Field

[0001] This invention relates to the fields of deep learning and optical communication, and is a distortion vortex optical correction system and method based on the GS algorithm and deep learning. Background Technology

[0002] A vortex beam is a type of beam with a unique wavefront structure; its cross-sectional intensity distribution is a hollow ring. A vortex beam carries orbital angular momentum, expressed in the formula as a helical phase term exp(ilφ), where l is called the topological charge. Theoretically, the topological charge can be any non-zero integer, and vortex beams with different topological charges are orthogonal to each other. This orthogonality of vortex beams has attracted considerable research interest—using vortex beam mode multiplexing for optical communication, thereby significantly expanding the channel capacity of optical communication systems.

[0003] However, due to the helical wavefront structure of vortex beams, they are prone to phase distortion when propagating through turbulent water, causing intersymbol interference (ISI) and significantly degrading system performance. Therefore, phase compensation and correction of distorted vortex beams are crucial.

[0004] However, traditional adaptive compensation methods all require wavefront analyzers, which are very expensive and impractical for real-world applications. While a single GS phase retrieval algorithm does not use a wavefront sensor, it can improve correction speed. This invention proposes using a deep learning model to learn the capabilities of the GS phase retrieval algorithm, thereby improving the speed of vortex beam phase correction. Summary of the Invention

[0005] To address the aforementioned shortcomings of existing technologies, this invention provides a distortion vortex light correction system and method based on the GS algorithm and deep learning, aiming to better solve the problem of severe crosstalk caused by turbulent media to vortex light transmission.

[0006] The technical solution adopted by this invention to solve the technical problem includes:

[0007] The vortex light emission module uses a laser and a spatial light modulator to emit initial vortex beams with different topological charges;

[0008] The turbulent transmission module uses a random turbulent phase screen to simulate a turbulent medium. The distorted vortex beam after passing through turbulence is used to obtain the vortex light intensity map via MAtlab.

[0009] The GS algorithm iteration module performs iterative calculations on the initial vortex beam and the distorted vortex beam to obtain the pre-corrected compensation phase screen C1(x,y).

[0010] The convolutional neural network training module takes the distorted vortex beam and its corresponding pre-correction compensation phase screen C1(x,y) as a group to form a training set, and performs mapping training on the convolutional neural network so that the convolutional neural network can learn the ability of the GS algorithm. The trained convolutional neural network directly generates the compensation phase screen C2(x,y) by inputting the distorted vortex light intensity map.

[0011] The phase screen compensation module applies the compensation phase screen C2(x,y) to the distorted vortex light intensity map to obtain the corresponding corrected vortex light intensity map.

[0012] The further iterative process of the GS algorithm of the present invention is as follows: it involves angular spectrum diffraction iteration calculation between the initial light field E(x,y) and the distorted light field E'(x,y), that is:

[0013] E(x,y)=E′(x,y)*H(f x ,f y );

[0014] In the formula, H(f) x ,f y ) represents the angular spectral transfer function, and * denotes convolution operation. The angular spectral inverse operation E(x,y) = E′(x,y) * H is required to transform the distorted light field E'(x,y) into the initial light field E(x,y). * (f x ,f y That is, the pre-correction compensation phase screen is C1(x,y)=H * (f x ,f y ).

[0015] According to the convolutional neural network training module of the present invention, 20,000 distorted vortex light intensity distribution maps and pre-correction compensation phase screen C1(x,y) are set as training data, the training model learns their mapping relationship, and 2,000 distorted vortex beams are used as verification data. During the verification process, the parameters are continuously adjusted and updated so that the parameters of the model reach the optimization.

[0016] The beneficial effects of this invention are as follows:

[0017] This invention, based on the GS algorithm, calculates a compensating phase screen by analyzing the intensity distributions of the initial and distorted light fields of the vortex beam. The compensating phase screen and the distorted light field are then used as a training set to train a convolutional neural network. The trained convolutional neural network learns the ability of the GS algorithm to generate the compensating phase screen and can directly generate it, thus directly correcting the phase distortion of the vortex beam in real time. This invention trains a convolutional neural network to replace the GS algorithm for repairing distorted vortex beams, and reduces the speed of generating the compensating phase screen from hundreds of milliseconds to tens of milliseconds. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the distortion vortex light real-time compensation system based on GS algorithm and deep learning according to the present invention.

[0019] Figure 2 This is a flowchart of the process for calculating the correction and compensation phase screen based on the GS algorithm and convolutional neural network in this invention.

[0020] Figure 3 A comparison of the intensity distribution of the vortex light field with a topological charge of 3 before and after correction.

[0021] Figure 4 This is a comparison diagram of the intensity of each orbital angular momentum component of the vortex beam before and after correction. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0023] like Figure 1 As shown, the present invention includes a vortex light emission module, a turbulent transmission module, a GS algorithm iteration module, a convolutional neural network training module, and a phase screen compensation module.

[0024] The vortex light emission module uses a laser and a spatial light modulator to emit initial vortex beams with different topological charges.

[0025] The turbulence transmission module uses a random turbulence phase screen to simulate a turbulent medium. The distorted vortex beam after passing through turbulence is used to obtain the vortex beam intensity map via MATLAB. In the turbulence transmission module, the effect of turbulence on the vortex beam can be simulated using a phase-type diffraction grating, that is, using an atmospheric turbulence phase screen to modulate the vortex beam.

[0026] The GS algorithm iteration module performs iterative calculations on the initial vortex beam and the distorted vortex beam to obtain the pre-corrected compensation phase screen C1(x,y).

[0027] The convolutional neural network training module takes the distorted vortex beam and its corresponding pre-correction compensation phase screen C1(x,y) as a group to form a training set, and performs mapping training on the convolutional neural network so that the convolutional neural network can learn the ability of the GS algorithm. The trained convolutional neural network directly generates the compensation phase screen C2(x,y) by inputting the distorted vortex light intensity map.

[0028] The phase screen compensation module applies the compensation phase screen C2(x,y) to the distorted vortex light intensity map to obtain the corresponding corrected vortex light intensity map.

[0029] like Figure 2 As shown, the initial vortex beam and the vortex beam distorted by turbulence are iteratively processed by the GS algorithm to obtain the pre-corrected compensation phase screen C1(x,y). The distorted vortex beam and the pre-corrected compensation phase screen C1(x,y) are then used as a training set to enable the convolutional neural network to learn the capabilities of the GS algorithm. It can directly generate the compensation phase screen C2(x,y) from the input distorted vortex beam. Finally, the intensity map of the vortex light distorted by turbulence is input into the trained convolutional neural network to obtain the corresponding corrected intensity map of the vortex light.

[0030] The key feature of the GS algorithm is that it calculates the phase distribution changes of the vortex beam based on the initial vortex beam and the intensity distribution of the light field after turbulence distortion. The phase distribution calculated by the GS algorithm serves the same purpose as a flow phase screen. Therefore, by introducing the phase inversion calculated by the GS algorithm, the effect of turbulence can be counteracted, thereby achieving pre-compensation for the distorted beam. This is the basic principle of the GS phase retrieval algorithm.

[0031] The further iterative process of the GS algorithm of the present invention is as follows: it involves angular spectrum diffraction iteration calculation between the initial light field E(x,y) and the distorted light field E'(x,y), that is:

[0032] E(x,y)=E′(x,y)*H(f x ,f y );

[0033] In the formula, H(f) x ,f y ) represents the angular spectral transfer function, and * denotes convolution operation. The angular spectral inverse operation E(x,y) = E′(x,y) * H is required to transform the distorted light field E'(x,y) into the initial light field E(x,y). * (f x ,f y That is, the pre-correction compensation phase screen is C1(x,y)=H * (f x ,f y ).

[0034] The implementation method of the distortion vortex light correction system based on the GS algorithm and deep learning includes the following steps:

[0035] Step 1: Use a laser and a spatial light modulator to emit initial vortex beams with different topological charges;

[0036] Step 2: Use a random turbulent phase screen to simulate the turbulent medium. The distorted vortex beam after turbulence is used to obtain the vortex light intensity map through MAtlab.

[0037] Step 3: The initial vortex beam and the distorted vortex beam are subjected to iterative calculation using the GS algorithm to obtain the pre-corrected compensation phase screen C1(x,y).

[0038] Step 4: Take the distorted vortex beam and its corresponding pre-correction compensation phase screen C1(x,y) as a group to form a training set, and perform mapping training on the convolutional neural network so that the convolutional neural network learns the ability of the GS algorithm. The trained convolutional neural network directly generates the compensation phase screen C2(x,y) by inputting the distorted vortex light intensity map.

[0039] Step 5: Apply the compensated phase screen C2(x,y) to the distorted vortex light intensity map to obtain the corresponding corrected vortex light intensity map.

[0040] In this embodiment, 20,000 distorted vortex light intensity distribution maps and a pre-corrected compensation phase screen C1(x,y) are randomly set as training data to train the model to learn their mapping relationship. 2,000 distorted vortex beams are used as verification data. During the verification process, the parameters are continuously adjusted and updated to optimize the various parameters of the model.

[0041] In this embodiment, the key factors affecting the generalization ability of the convolutional neural network model are the loss function and the number of iterations. The loss function is used to estimate the degree of error between the predicted and actual values. As the number of iterations increases, the loss value gradually decreases and tends to stabilize. When the number of iterations increases to 1000, the loss value is close to 0, and the corrected vortex beam intensity map is almost the same as the initial beam, indicating that our proposed convolutional neural network model has high accuracy.

[0042] In this embodiment, the convolutional neural network structure includes 12 convolutional layers and three deconvolutional layers: the first convolutional layer uses a 5×5 filter, and the other convolutional and deconvolutional layers all use 3×3 filters, which are used to extract features from the input image.

[0043] In this embodiment, each image is resized to 64×64 and normalized using the maximum grayscale value. The network output is a corrected vortex intensity map of size 64×64.

[0044] In this embodiment, the purity of the vortex beam mode is greater than 90% after compensation under different turbulence intensities; especially under strong turbulence conditions, the mode purity is improved from 36.54% to 97.81%, and the compensation speed is improved from hundreds of milliseconds in the traditional GS algorithm to tens of milliseconds.

[0045] This embodiment uses a single-mode vortex beam for verification. When the topological charge is 3, a generalized anisotropic von-Karman spectrum is used as the atmospheric turbulence phase screen, and the refractive index structure constant of the turbulence intensity is set. That is, moderate turbulence intensity. Experimental results are as follows: Figure 3 As shown in the figure, from left to right, the vortex light intensity distributions are for the following conditions: no turbulence, turbulence without correction, and turbulence with pre-correction. It can be seen from the figure that the light intensity distribution is significantly improved before and after correction.

[0046] Meanwhile, in order to better evaluate the phase recovery of the vortex beam before and after pre-correction, we analyzed the orbital angular momentum spectrum of the multimode multiplexed vortex beam received before and after pre-correction. Figure 4 The orbital angular momentum spectra before and after pre-correction are presented for the topological charge numbers of the vortex beams: -3, -2, -1, 0, 1, 2, and 3. It can be seen that the dispersion of the orbital angular momentum spectrum is significantly improved after pre-correction.

[0047] In summary, the present invention proposes a distortion vortex light correction system based on the GS algorithm and deep learning. Compared with previous systems, this system utilizes the ability of a convolutional neural network to learn the GS algorithm to generate a pre-corrected phase screen. After training, it can directly correct the input distortion vortex light, thereby improving the system's correction speed and accuracy.

[0048] While the present invention has been described in detail above, those skilled in the art should understand that the present invention is not limited to the above embodiments. Any modifications, equivalent substitutions, improvements, etc., made without departing from the concept of the present invention should be included within the protection scope of the present invention.

Claims

1. A distortion vortex light correction system based on GS algorithm and deep learning, characterized in that... It includes a vortex light emission module, a turbulent transport module, a GS algorithm iteration module, a convolutional neural network training module, and a phase screen compensation module; The vortex light emission module uses a laser and a spatial light modulator to emit initial vortex beams with different topological charges. The turbulent transmission module uses a random turbulent phase screen to simulate a turbulent medium. The distorted vortex beam after passing through turbulence is used to obtain the vortex light intensity map via MAtlab. The GS algorithm iteration module performs iterative calculations on the initial vortex beam and the distorted vortex beam to obtain the pre-corrected compensation phase screen C1(x,y). The convolutional neural network training module takes the distorted vortex beam and its corresponding pre-correction compensation phase screen C1(x,y) as a group to form a training set, and performs mapping training on the convolutional neural network so that the convolutional neural network can learn the ability of the GS algorithm. The trained convolutional neural network directly generates the compensation phase screen C2(x,y) by inputting the distorted vortex light intensity map. The phase screen compensation module applies the compensation phase screen C2(x,y) to the distorted vortex light intensity map to obtain the corresponding corrected vortex light intensity map.

2. The distortion vortex light correction system based on GS algorithm and deep learning according to claim 1, characterized in that... The specific iterative process of the GS algorithm iterative module is as follows: The calculation involves iterative angular spectrum diffraction between the initial light field E(x,y) and the distorted light field E'(x,y), namely: E′(x,y)=E(x,y)*H(f x ,f y ); In the formula, H(f) x ,f y ) represents the angular spectral transfer function, and * represents the convolution operation; the angular spectral inverse operation E(x,y) = E′(x,y) * H is required to transfer the distorted light field E'(x,y) to the initial light field E(x,y). * (f x ,f y That is, the pre-correction compensation phase screen is C1(x,y)=H * (f x ,f y ).

3. The distortion vortex light correction system based on GS algorithm and deep learning according to claim 1, characterized in that... 20,000 distorted vortex light intensity maps and a pre-corrected compensation phase screen C1(x,y) were randomly set as training data to train the model to learn their mapping relationship. 2,000 distorted vortex beams were used as validation data. During the validation process, the parameters were continuously adjusted and updated to optimize the model's parameters.

4. The distortion vortex light correction system based on GS algorithm and deep learning according to claim 1, characterized in that... The convolutional neural network structure contains 12 convolutional layers and three deconvolutional layers: the first convolutional layer uses a 5×5 filter, and the other convolutional and deconvolutional layers use 3×3 filters. Its function is to extract features from the input image.

5. The distortion vortex light correction system based on GS algorithm and deep learning according to claim 4, characterized in that... Each image is resized to 64×64 and normalized using the maximum grayscale value. The network output is a corrected vortex light intensity map of size 64×64.

6. A distortion vortex light correction method based on GS algorithm and deep learning, characterized in that... The method includes the following steps: Step 1, using a laser and a spatial light modulator to emit initial vortex beams with different topological charges; Step 2: Use a random turbulent phase screen to simulate the turbulent medium, and obtain the vortex light intensity map of the distorted vortex beam after turbulence through MAtlab; Step 3: The initial vortex beam and the distorted vortex beam are subjected to iterative calculation using the GS algorithm to obtain the pre-corrected compensation phase screen C1(x,y); Step 4: Take the distorted vortex beam and its corresponding pre-correction compensation phase screen C1(x,y) as a group to form a training set, and perform mapping training on the convolutional neural network so that the convolutional neural network learns the ability of the GS algorithm. The trained convolutional neural network directly generates the compensation phase screen C2(x,y) by inputting the distorted vortex light intensity map. Step 5: Apply the compensated phase screen C2(x,y) to the distorted vortex light intensity map to obtain the corresponding corrected vortex light intensity map.

Images