An atmospheric turbulence phase extraction method based on a U-Net network
By constructing a model using the U-Net network combined with the power spectrum inversion method and the phase-intensity formula, the problem of inaccurate atmospheric turbulence phase extraction in existing technologies is solved, realizing real-time and accurate extraction of atmospheric turbulence phase, and improving the compensation efficiency and communication quality of free-space coherent optical communication systems.
Patent Information
- Application Number
- CN202211418287.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-14
- Publication Date
- 2025-11-21
- Estimated Expiration
- 2042-11-14
AI Technical Summary
Existing technologies struggle to extract the atmospheric turbulence phase accurately in real time, resulting in low wavefront distortion compensation efficiency in wavefront-less adaptive optics for free-space coherent optical communication systems, which affects communication quality.
A U-Net network is used in conjunction with the power spectrum inversion method and the phase-intensity formula to construct a signal light wavefront phase and intensity distribution model. A large number of accurate samples are provided to train the U-Net network, and a high-precision atmospheric turbulence phase extraction method is obtained through iterative optimization.
It enables real-time and accurate extraction of atmospheric turbulence phase, improves the compensation efficiency of wavefront-free adaptive optics in free-space coherent optical communication systems, and enhances communication quality.
Smart Images

Figure QLYQS_1 
Figure QLYQS_2 
Figure QLYQS_4
Abstract
Description
TECHNICAL FIELD
[0001] The application relates to an atmospheric turbulence phase extraction method based on a U-Net network and belongs to the field of wavefront-sensor-free adaptive optics. BACKGROUND
[0002] Free space coherent optical communication technology is a wireless communication mode taking light as an information carrier and taking the atmosphere as a transmission channel. The technology has outstanding technical advantages such as large data capacity and high communication rate and is considered as an important technical means for breaking through the existing high-speed communication bottleneck. However, when the optical signal is transmitted in the atmosphere, the optical signal will be affected by gas molecules and aerosol particles, such as optical power attenuation caused by absorption and scattering of the optical signal and beam flickering and bending caused by atmospheric turbulence. Among them, the influence of atmospheric turbulence on optical signal transmission is the most significant. When the optical signal passes through the turbulent atmosphere, the wavefront distortion and phase fluctuation of the optical signal are caused, negative effects such as optical intensity flickering, speckle and angle of arrival fluctuation are caused, the quality of the transmission beam is seriously affected, and the performance of the free space coherent optical communication system is reduced.
[0003] At present, wavefront-sensor-free adaptive optics has become one of the main research directions for atmospheric turbulence suppression and improvement of the communication quality of the free space coherent optical communication system in recent years due to its low cost and simple structure. The biggest feature of the wavefront-sensor-free adaptive optics is to directly control the wavefront corrector to compensate for the wavefront distortion through an optimization algorithm. However, the processing time and precision of the signal light wavefront are mainly controlled by the optimization algorithm, so an optimization algorithm capable of real-time and accurate extraction of the atmospheric turbulence phase is urgently needed to realize phase compensation of the wavefront distortion. Xu Qiwei et al. of Shenzhen University of Technology successfully extracted the atmospheric turbulence phase under different turbulence intensities using a 17-layer (not including input and output) CNN network. In the Chinese patent CN113225130A, a CNN network with 12 convolution layers and 3 deconvolution layers is used to predict the atmospheric turbulence equivalent phase screen grayscale image. The optimization network used in the above methods has a simple structure and a long iteration time, and cannot realize real-time prediction and extraction of the atmospheric turbulence phase.
[0004] Therefore, designing an optimization algorithm capable of real-time and accurate extraction of the atmospheric turbulence phase is an urgent technical problem to be solved. SUMMARY
[0005] The technical problem to be solved by the application is to provide an atmospheric turbulence phase extraction method based on a U-Net network, effectively suppress the negative effects such as wavefront phase distortion caused by the atmospheric turbulence on the free space coherent optical communication system, and realize an adaptive compensation method for the signal light wavefront.
[0006] To solve the above technical problem, the technical scheme adopted by the application is:
[0007] A method for extracting atmospheric turbulence phase based on U-Net network, specifically comprising the following steps:
[0008] Step 1: Based on the power spectrum inversion method and the phase-intensity formula, a quantitative model of the wavefront phase and the intensity distribution of the signal light wave after the atmospheric turbulence transmission is established, so as to provide a large number of accurate samples for the U-Net network training;
[0009] Step 2: The U-Net network is supervisedly trained according to the large number of accurate samples provided in step 1;
[0010] Step 3: Based on the trained network obtained in step 2, an atmospheric turbulence phase extraction model based on the U-Net network is constructed, so as to form a real-time high-precision atmospheric turbulence phase extraction method based on the U-Net network.
[0011] Further improvement of the technical scheme of the application is that the specific process of step 1 is:
[0012] The atmospheric turbulence random phase screen is generated based on the power spectrum inversion method: first, a Hermitian complex Gaussian random number matrix R(k x ,k y ) with zero mean and unit variance in the frequency domain is generated, then it is filtered by the power spectrum density function F φ (k x ,k y ) conforming to the Kolmogorov spectrum of atmospheric turbulence, and then inverse Fourier transform is performed to obtain the atmospheric turbulence random phase screen φ(x,y), that is,
[0013]
[0014] Wherein, x and y are n times of the sampling interval in the spatial domain, n is an integer; k x and k y are m times of the sampling interval in the wave number domain, m is an integer; the constant C comes from the scale factor (Δk x Δk y ) 1 / 2 , which is an adjustment constant for controlling the variance of the phase screen;
[0015] After the atmospheric turbulence random phase screen is simulated, the atmospheric transmission of the signal light is processed by using the multi-layer phase screen model, that is, the continuous random medium is divided into a series of parallel slices with a thickness of Δz, and the phase modulation caused by each slice can be regarded as a wireless thin phase screen located at the center of the slice; the signal light is modulated by the phase screen z i , and then propagates in free space to the next phase screen z i+1The position of each phase screen is changed only by changing the phase of the light wave without affecting the amplitude of the light wave; the signal light transmission path Z is divided into N segments, and each phase screen is divided into N*N grids, and the width of each grid is Δx, then according to the phase-intensity formula Z i+1 The light field of the phase screen is:
[0016]
[0017] Wherein, f and f -1 Respectively represent the Fourier transform and the inverse Fourier transform; S(r, z i ) represents the phase change of the signal light after the phase screen z i .
[0018] In combination with the above process, the quantitative model of the wavefront phase and the intensity distribution of the signal light wave after atmospheric turbulence transmission is constructed by using Matlab simulation software, and a large number of accurate samples are provided for U-Net network training.
[0019] The further improvement of the technical scheme of the application is that: the large number of accurate samples include n atmospheric turbulence equivalent phase screen gray scale images, and the intensity distribution diagrams of the propagation signal under the influence of turbulence and without the influence of turbulence respectively, wherein n≥70000.
[0020] The further improvement of the technical scheme of the application is that: the specific process of step 2 is:
[0021] Step 2.1, constructing a U-Net network;
[0022] Step 2.2, defining a loss function according to the network input, iteratively training the U-Net network with the large number of accurate sample data sets generated by the model in step 1, adjusting the weight w and the bias b of the network, and minimizing the loss function value after K iterations, so as to obtain the optimal U-Net network model.
[0023] The further improvement of the technical scheme of the application is that: in step 2.1, the U-Net network is improved by adding a residual block, so as to deepen the network layers.
[0024] The further improvement of the technical scheme of the application is that: in step 2.2, the average loss function is defined as:
[0025]
[0026] Wherein, Relu represents a rectified linear unit function, that is, an activation function; And The error loss function of the sample with index j is evaluated; p (j) Indicates the actual atmospheric turbulence phase; Indicates the turbulence phase predicted and extracted by the U-Net network.
[0027] Further improvement of the technical scheme of the present application is that in step 2.2, the iteration number K is greater than or equal to 4000.
[0028] Further improvement of the technical scheme of the present application is that the specific process of step 3 is:
[0029] The U-Net network model obtained in step 2.2 is input with an arbitrary signal light intensity distribution image affected by turbulence to realize atmospheric turbulence phase extraction.
[0030] Due to the adoption of the above technical scheme, the present application has achieved the following technical progress:
[0031] The present application uses power spectrum inversion method and phase-intensity formula to obtain intensity distribution graph before and after signal light propagation and atmospheric turbulence equivalent phase screen gray graph, uses the sample data set to train U-Net network to quickly and accurately extract atmospheric turbulence phase under the condition that arbitrary signal light intensity distribution graph is input, overcomes the problem of long prediction time of ordinary simple CNN network, speeds up network iteration operation speed, and improves communication system performance. DETAILED DESCRIPTION
[0032] The present application will be further described in detail below in combination with embodiments:
[0033] A kind of atmospheric turbulence phase extraction method based on U-Net network, specifically including the following steps:
[0034] Step 1, based on power spectrum inversion method and intensity-phase formula, construct the quantitative model of wavefront phase and light intensity distribution of signal light wave after atmospheric turbulence transmission, obtain massive accurate sample data set, the sample data set includes: initial signal light intensity distribution image not affected by turbulence, signal light intensity distribution image under the influence of different atmospheric turbulence intensity, atmospheric turbulence equivalent phase screen gray image of different turbulence intensity;
[0035] Based on power spectrum inversion method, generate atmospheric turbulence random phase screen: first generate a Hermitian complex Gaussian random number matrix R (k x ,k y ) with zero mean and unit variance in frequency domain, then filter it with power spectrum density function F φ (k x ,k y ) conforming to atmospheric turbulence Kolmogorov spectrum, and then perform inverse Fourier transform to obtain atmospheric turbulence random phase screen φ (x, y), i.e.
[0036]
[0037] where x and y are n times of the sampling interval in the spatial domain, n is an integer; k x and k y are m times of the sampling interval in the wave number domain, m is an integer; the constant C comes from the scaling factor (Δk x Δk y ) 1 / 2 is the adjustment constant of the phase screen variance.
[0038] After simulating the atmospheric turbulence random phase screen, the signal light atmospheric transmission is processed by using the multi-layer phase screen model, that is, the continuous random medium is divided into a series of parallel slices with a thickness of Δz, and the phase modulation caused by each slice can be regarded as a wireless thin phase screen located at the center of the slice. The signal light is modulated by the phase screen z i , and then propagates in free space to the position of the next phase screen z i+1 , and each transmission only changes the phase of the light wave without affecting its amplitude. The signal light transmission path Z is divided into N segments, and each phase screen is divided into N×N grids, and the width of each grid is Δx, then the light field of Z i+1 is:
[0039]
[0040] where f and f -1 represent the Fourier transform and the inverse Fourier transform respectively; S(r, z i ) represents the phase change of the signal light after passing through the phase screen z i .
[0041] In combination with the above process, the quantitative model of the wavefront phase and light intensity distribution of the signal light wave after atmospheric turbulence transmission is constructed by using Matlab simulation software, so as to provide a large number of accurate samples for U-Net network training.
[0042] Step 2, according to the large number of accurate samples provided in step 1, the U-Net network is trained in a supervised manner;
[0043] The U-Net network is constructed to learn the sample data, the loss function is defined according to the network input, the weights w and the bias b of the network are adjusted, and after K iterations, the loss function value is minimized, so as to obtain the optimal U-Net network model.
[0044] The average loss function is defined as:
[0045]
[0046] where Relu represents the rectified linear unit function, that is, the activation function; and are the error loss functions of the samples with index j; p (j)representing the actual atmospheric turbulence phase; representing the turbulence phase predicted by the U-Net network.
[0047] Step 3, inputting the signal intensity distribution image of any turbulence-affected signal into the U-Net network model obtained in step 2.2 to realize atmospheric turbulence phase extraction.
Claims
1. A method for extracting atmospheric turbulence phase based on a U-Net network, characterized in that: Specifically comprising the following steps: Step 1, based on the power spectrum inversion method and the phase-intensity formula, a quantitative model of wavefront phase and intensity distribution of signal light wave after atmospheric turbulence transmission is established, and a large number of accurate samples are provided for U-Net network training; the specific process of step 1 is as follows: Generation of atmospheric turbulence random phase screen based on power spectrum inversion method: first generate a Hermitian complex Gaussian random number matrix with zero mean and unit variance in frequency domain Then filter it with the power spectrum density function consistent with the Kolmogorov spectrum of atmospheric turbulence And then get the atmospheric turbulence random phase screen by inverse Fourier transform That is (1), wherein and is n times the sampling interval in the spatial domain, n being an integer; and is m times the sampling interval in the wave number domain, m being an integer; the constant comes from the scaling factor is a regulation constant that controls the phase screen variance; After simulating the atmospheric turbulence random phase screen, the atmospheric transmission of the signal light is processed using a multi-layer phase screen model, which involves dividing the continuous random medium into a series of layers with a thickness of [missing information]. The phase modulation caused by each of the parallel thin sheets can be regarded as a wirelessly thin phase screen located at the center of the sheet; the signal light passes through the phase screen. Modulation, and then propagation in free space to the next phase screen. The position of the light wave is such that each transmission only changes the phase of the light wave without affecting its amplitude; the signal light transmission path is changed. Divided into Segments, and divide each phase screen into There are 1 grid, and the width of each grid is 1. According to the phase-intensity formula The light field is: (2), wherein and denote the Fourier transform and the inverse Fourier transform, respectively; denotes the phase change of the signal light after passing through the phase plate ; In combination with the above process, a quantitative model of wavefront phase and intensity distribution of signal light wave after atmospheric turbulence transmission is constructed by using Matlab simulation software, and a large number of accurate samples are provided for U-Net network training Step 2, according to the large number of accurate samples provided in step 1, the U-Net network is trained in a supervised manner; the specific process of step 2 is as follows: Step 2.1, constructing the U-Net network; Step 2.2, defining the loss function according to the network input, using the large number of accurate sample data sets generated by the model in step 1 to iteratively train the U-Net network, adjusting the weight w and bias b of the network, and minimizing the loss function value after K iterations, thereby obtaining the optimal U-Net network model; Step 3, based on the trained network obtained in step 2, an atmospheric turbulence phase extraction model based on U-Net network is constructed, forming a real-time high-precision atmospheric turbulence phase extraction method based on U-Net network.
2. The method of claim 1, wherein the U-Net network is used to extract the atmospheric turbulence phase. The large number of accurate samples include n atmospheric turbulence equivalent phase screen gray scale images, and signal light intensity distribution images affected by turbulence and not affected by turbulence respectively, wherein n≥70000.
3. The method of claim 1, wherein the U-Net network is used to extract the atmospheric turbulence phase. In step 2.1, the U-Net network is improved by adding a residual block, thereby deepening the network layers.
4. The method of claim 1, wherein the U-Net network is used to extract the atmospheric turbulence phase. In step 2.2, the average loss function is defined as: (3), wherein, represents a rectified linear unit function, i.e. an activation function; , and is an error loss function for evaluating the sample indexed . denotes the actual atmospheric turbulence phase; denotes the turbulence phase predicted by the U-Net network.
5. The method of claim 1, wherein the U-Net network is used to extract the atmospheric turbulence phase. In step 2.2, the iteration number K≥4000.
6. The method of claim 1, wherein the U-Net network is used to extract the atmospheric turbulence phase. The specific process of step 3 is as follows: Input any signal light intensity distribution image affected by turbulence into the U-Net network model obtained in step 2.2 to realize atmospheric turbulence phase extraction.
Citation Information
Patent Citations
Atmospheric turbulence equivalent phase screen prediction method based on machine learning
CN113225130A
Atmospheric turbulence real-time compensation method and system based on deep learning
CN110332990A
Vortex beam phase distortion correction method and system and communication device
CN114499660A