A method and system for detecting atmospheric turbulence intensity based on photon diffraction neural networks

Abstract

This invention relates to a method and system for detecting atmospheric turbulence intensity based on a photon diffraction neural network. The method includes: establishing and training a photon diffraction neural network, which includes a forward propagation model and a backward propagation model, wherein the backward propagation model is an error backpropagation based on a loss function, used to train the photon diffraction neural network; acquiring light beams under different atmospheric intensities; inputting the light beams into the trained photon diffraction neural network; detecting the output light field intensity of the photon diffraction neural network based on a light intensity detector; and completing the atmospheric turbulence intensity detection according to the correspondence between the light field intensity and the atmospheric turbulence intensity, wherein the stronger the atmospheric turbulence intensity, the weaker the light field intensity. Compared with the prior art, this invention has the advantages of requiring no additional signal processing, fast detection speed, and high detection efficiency.

Description

Technical Field

[0001] This invention relates to the field of optical communication, and in particular to a method and system for detecting atmospheric turbulence intensity. Background Technology

[0002] With the rapid development of modern information society, people have increasingly higher requirements for communication speed and quality. However, traditional single-mode optical fiber is already saturated, and existing multiplexing technologies such as time-division multiplexing and code-division multiplexing are insufficient to help single-mode optical fiber overcome transmission capacity limitations. Spatial dimension is a degree of freedom in information multiplexing that can significantly improve network transmission capacity. In free-space mode-division multiplexing systems, atmospheric random disturbances are a major factor affecting system performance. Due to the influence of atmospheric turbulence, the propagation of a laser beam in the atmosphere causes additional beam spread, random drift of the instantaneous beam center, and random disturbances in light intensity, resulting in a decrease in laser beam quality and a reduction in the transmission capacity of the communication system. Therefore, determining the intensity of atmospheric turbulence is of great significance in improving communication efficiency.

[0003] Traditional methods typically use the refractive index structure constant to measure the intensity of atmospheric turbulence. Since the 1990s, methods for estimating atmospheric turbulence intensity using numerical weather prediction models have emerged. In recent years, machine learning algorithms have been widely applied in meteorology and other fields. The paper "Using an artificial neural network approach to estimate surface-layer optical turbulence at Mauna Loa, Hawaii" (Yao Wang, Sukanta Basu. Optics Letters, 2016, 41(10): 2334-2337) uses an artificial neural network, taking meteorological parameters such as temperature and relative humidity as inputs, to estimate the intensity of atmospheric turbulence at the sea surface near Mauna Loa. The results show that this method can estimate the turbulence intensity relatively accurately. The paper "Atmospheric turbulence intensity estimation based on deep convolutional neural networks" (Ma Shengjie, Hao Shiqi, et al. Chinese Journal of Lasers, 2021, 48(4): 277-286) uses a convolutional neural network to train images of Gaussian beam spots affected by atmospheric turbulence. Leveraging the powerful data fitting and image processing capabilities of deep convolutional neural networks, it completes the estimation from image to numerical value. The paper "Joint atmospheric turbulence detection and adaptive demodulation technique using the CNN for the OAM-FSO communication" (Li J, Zhang M, Wang DS, et al. Optics Express, 2018, 26(8): 10494–10508.) designs a CNN-based adaptive demodulator to achieve atmospheric turbulence detection with high accuracy in strong turbulence.

[0004] The paper "All-optical machine learning using diffffractive deep neural networks" (X. Lin, Y. Rivenson, N. Yardimci, M. Veli, Y. Luo, M. Jarrahi, and A. Ozcan. Science (80-.) 361, 1004 (2018).) proposed an all-optical diffractive deep neural network. Lin et al. designed a 5-layer all-optical neural network, which achieved an accuracy of 91.75% on the MNIST dataset. Compared with convolutional neural networks, photonic diffractive neural networks can transmit data at the speed of light. This network uses photons as the signal transmission medium entirely, requiring no additional signal processing methods, and has broad application prospects.

[0005] None of the above methods for detecting atmospheric turbulence employ photonic diffraction neural networks, requiring additional signal processing of the optical signal, which is quite complex. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for detecting atmospheric turbulence intensity based on a photon diffraction neural network, which uses photons as the transmission medium, eliminates the need for additional signal processing, and improves detection speed and efficiency.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A method for detecting atmospheric turbulence intensity based on a photon diffraction neural network includes the following steps:

[0009] A photon diffraction neural network is established and trained. The photon diffraction neural network includes a forward propagation model and a backward propagation model. The backward propagation model is an error backward propagation based on a loss function, which is used to train the photon diffraction neural network.

[0010] Acquire light beams under the influence of different atmospheric intensities;

[0011] The beam is fed into the trained photon diffraction neural network;

[0012] The intensity of the output light field of a photon diffraction neural network is detected based on a light intensity detector.

[0013] Based on the correspondence between light field intensity and atmospheric turbulence intensity, atmospheric turbulence intensity is detected. The correspondence between light field intensity and atmospheric turbulence intensity is as follows: the stronger the atmospheric turbulence intensity, the weaker the light field intensity.

[0014] The photonic diffraction neural network includes k diffraction layers. Each node in each diffraction layer is connected to the next layer. The center of the diffraction layer is the origin of the coordinate system. The light beam propagates along the z-direction. Each node serves as a neuron in the neural network, and the amplitude and phase of the neuron are used as training parameters.

[0015] The light wave generated by a beam passing through a node in the diffraction layer can be considered as a secondary wave source:

[0016]

[0017] Where l represents the l-th diffraction layer, i represents the i-th node of the l-th diffraction layer, and (x, y, z) represents the beam coordinates, (x...y...z ... i y i , z i () represents the coordinates of the i-th node. λ represents the distance from the beam to the i-th node, and λ is the wavelength of the beam.

[0018] The amplitude and phase of the secondary wave source are determined by the input wave and the transmission coefficient t at that node. Therefore, the output of the i-th node in the l-th layer is expressed as:

[0019]

[0020] in, g represents the input wave of the i-th node in the l-th layer, and g represents the set of beams of all output waves in the (l-1)-th layer that propagate to node i through different paths. The transmission coefficient is determined by the amplitude. and phase composition

[0021] The forward propagation model is determined based on Fresnel diffraction theory, and when the input beam propagates to the l-th layer, it is represented as follows: After diffraction at the l-th layer, the light field After the interlayer distance d l The transmission of light is equivalent to passing through a transmission matrix H representing Fresnel diffraction. Therefore, the output light field of the l-th layer is expressed as:

[0022]

[0023] Among them, (f x ,f y () are frequency domain coordinates. n represents the wave number, n = 2π / λ.

[0024] The loss function is the difference between the light field intensity measured by the light intensity detector and the ideal light field intensity of the forward propagation output light field.

[0025]

[0026] Where N is the number of nodes in the output layer of the last layer of the photon diffraction neural network, U i Y is the output light field intensity of node i. i Let be the ideal light field intensity at node i.

[0027] An atmospheric turbulence intensity detection system based on a photon diffraction neural network includes:

[0028] A photonic diffraction neural network, comprising a forward propagation model and a backward propagation model, wherein the backward propagation model is an error backward propagation based on a loss function, used to train the photonic diffraction neural network;

[0029] The input module is used to acquire light beams under different atmospheric intensities and input the light beams into the trained photon diffraction neural network.

[0030] A light intensity detector is used to detect the intensity of the output light field of a photon diffraction neural network;

[0031] The output module is used to complete the atmospheric turbulence intensity detection based on the light field intensity detected by the light intensity detector and the correspondence between the light field intensity and the atmospheric turbulence intensity. The correspondence between the light field intensity and the atmospheric turbulence intensity is that the stronger the atmospheric turbulence intensity, the weaker the light field intensity.

[0032] The photonic diffraction neural network includes k diffraction layers. Each node in each diffraction layer is connected to the next layer. The center of the diffraction layer is the origin of the coordinate system. The light beam propagates along the z-direction. Each node serves as a neuron in the neural network, and the amplitude and phase of the neuron are used as training parameters.

[0033] The light wave generated by a beam passing through a node in the diffraction layer can be considered as a secondary wave source:

[0034]

[0035] Where l represents the l-th diffraction layer, i represents the i-th node of the l-th diffraction layer, and (x, y, z) represents the beam coordinates, (x...y...z ... i y i , z i () represents the coordinates of the i-th node. λ represents the distance from the beam to the i-th node, and λ is the wavelength of the beam.

[0036] The amplitude and phase of the secondary wave source are determined by the input wave and the transmission coefficient t at that node. Therefore, the output of the i-th node in the l-th layer is expressed as:

[0037]

[0038] in, g represents the input wave of the i-th node in the l-th layer, and g represents the set of beams of all output waves in the (l-1)-th layer that propagate to node i through different paths. The transmission coefficient is determined by the amplitude. and phase composition

[0039] The forward propagation model is determined based on Fresnel diffraction theory, and when the input beam propagates to the l-th layer, it is represented as follows: After diffraction at the l-th layer, the light field After the interlayer distance d l The transmission of light is equivalent to passing through a transmission matrix H representing Fresnel diffraction. Therefore, the output light field of the l-th layer is expressed as:

[0040]

[0041] Among them, (f x ,f y () are frequency domain coordinates. n represents the wave number, n = 2π / λ.

[0042] The loss function is the difference between the light field intensity measured by the light intensity detector and the ideal light field intensity of the forward propagation output light field.

[0043]

[0044] Where N is the number of nodes in the output layer of the last layer of the photon diffraction neural network, U i Y is the output light field intensity of node i. i Let be the ideal light field intensity at node i.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] (1) This invention realizes the detection of atmospheric turbulence intensity based on photon diffraction neural network. It uses photons as the signal transmission medium, which has fast data transmission speed and fast processing speed, and greatly improves the detection speed.

[0047] (2) This invention does not require additional signal processing, which can greatly simplify the atmospheric intensity detection method and achieve high detection efficiency.

[0048] (3) This invention has high accuracy in atmospheric intensity detection and has broad application prospects in atmospheric detection. Attached Figure Description

[0049] Figure 1 This is a flowchart of the method of the present invention;

[0050] Figure 2 This is a schematic diagram of a photon diffraction neural network.

[0051] Figure 3 The following are simulation diagrams of the OAM beam under different atmospheric turbulence environments, where (a) represents (b) represents (c) represents (d) represents

[0052] Figure 4 This is a training result diagram of atmospheric intensity detection achieved through a photonic diffraction neural network in this embodiment;

[0053] Figure 5 This is a graph showing the training results after parameter modification. Detailed Implementation

[0054] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0055] Example 1

[0056] This embodiment provides a method for detecting atmospheric turbulence intensity based on a photon diffraction neural network, such as... Figure 1 As shown, it includes the following steps:

[0057] 1) Establish and train a photon diffraction neural network

[0058] The photon diffraction neural network, such as Figure 2 As shown, the network is trained using light beams affected by different atmospheric intensities. The trained photon diffraction neural network can then be used for atmospheric intensity detection.

[0059] The photonic diffraction neural network includes k diffraction layers. Each node in each diffraction layer is connected to the next layer. The center of the diffraction layer is the origin of the coordinate system. The light beam propagates along the z-direction. Each node serves as a neuron in the neural network, and the amplitude and phase of the neuron are used as training parameters.

[0060] The photon diffraction neural network includes a forward propagation model and a backward propagation model, wherein the backward propagation model is an error backpropagation based on a loss function, used to train the photon diffraction neural network.

[0061] According to Rayleigh-Soumfield diffraction theory, the light wave generated by a beam passing through a node in the diffraction layer is considered a secondary wave source:

[0062]

[0063] Where l represents the l-th diffraction layer, i represents the i-th node of the l-th diffraction layer, and (x, y, z) represents the beam coordinates, (x...y...z ... i y i , z i () represents the coordinates of the i-th node. λ represents the distance from the beam to the i-th node, and λ is the wavelength of the beam.

[0064] The amplitude and phase of the secondary wave source are determined by the input wave and the transmission coefficient t at that node. Therefore, the output of the i-th node in the l-th layer is expressed as:

[0065]

[0066] in, g represents the input wave of the i-th node in the l-th layer, and g represents the set of beams of all output waves in the (l-1)-th layer that propagate to node i through different paths. The transmission coefficient is determined by the amplitude. and phase composition Therefore, when the size of the diffraction layer and the interlayer distance are fixed, the light propagation mode is mainly determined by the amplitude and phase values ​​of each node.

[0067] In order to accurately control the light field using a multi-layer diffraction screen to detect turbulence intensity, we must predetermine the modulation parameters of the diffraction screen, treat each node as a neuron in a neural network, and train the neuron's amplitude and phase values ​​as learnable parameters.

[0068] The forward propagation model is determined based on Fresnel diffraction theory, and when the input beam propagates to the l-th layer, it is represented as follows: After diffraction at the l-th layer, the light field After the interlayer distance d l The transmission of light is equivalent to passing through a transmission matrix H representing Fresnel diffraction. Therefore, the output light field of the l-th layer is expressed as:

[0069]

[0070] Among them, (f x ,f y () are frequency domain coordinates. n represents the wave number, n = 2π / λ.

[0071] At the network's output, a light intensity detector is used to detect the network's output light field intensity, and the difference between the forward propagation output light field intensity U and the ideal light field intensity Y is defined as the loss function:

[0072]

[0073] Where N is the number of nodes in the output layer of the last layer of the photon diffraction neural network, U i Y is the output light field intensity of node i. i Let be the ideal light field intensity of node i. The Adam optimizer is used to optimize the network, and better prediction results are achieved by continuously changing the phase and amplitude of each neuron.

[0074] In this embodiment, an OAM beam affected by atmospheric turbulence is used as the training set for the photon diffraction neural network.

[0075] This embodiment first uses the split-step Fourier method to simulate the intensity of atmospheric disturbances, generating a training set of OAM beams affected by atmospheric turbulence. By adding different disturbance phase screens, the OAM beams in C... n 2 0, 1*10 -16 1*10 -15 1*10 -14 The images represent the ideal situation with no atmospheric turbulence, and the situations with weak atmospheric turbulence, moderate atmospheric turbulence, and strong atmospheric turbulence, respectively. That is, in this embodiment, the intensity of atmospheric turbulence is divided into four levels, and the characteristics of the light field intensity under these four levels are determined to establish the correspondence between the light field intensity and the intensity of atmospheric turbulence. For example... Figure 3 As shown, under weak atmospheric turbulence, the light spot shows no significant change, and the light intensity does not show a significant decrease; under medium atmospheric turbulence, the light spot distortion is obvious; under strong atmospheric turbulence, the light spot is not formed, and the light intensity decreases severely. That is, the relationship between light field intensity and atmospheric turbulence intensity is: the stronger the atmospheric turbulence intensity, the weaker the light field intensity.

[0076] 100 256*256 pixel light spot images were generated for each of the four scenarios to create a training set. For each training set, the average light intensity was taken as the ideal light intensity for that training set and compared with the undisturbed light field to serve as the training label.

[0077] This embodiment establishes a 5-layer photonic diffraction neural network, with each layer containing 128*128 neurons. Each neuron uses only its phase value as a training parameter, with the initial phase value randomly assigned. At the input, downsampling is used to adjust the image format to 128*128 before transmitting it into the photonic network. An ideal light field intensity ratio of 1:0.6132:0.9454:0.9941 is set as the ideal light intensity for four atmospheric environments. A CCD is used to detect the light field intensity of the network's output light field. The Adam optimizer is used to optimize the loss function. The learning step size of the Adam optimizer is set to 0.01, β1 = 0.9, and β2 = 0.999. A total of 1600 training rounds are conducted, with 20 datasets per round. The training results are as follows: Figure 4As shown, after 400 rounds of training, the accuracy rate of atmospheric intensity recognition reached 70%. This accuracy rate needs further improvement.

[0078] The network structure was further optimized by setting the learning step size to 0.03 and resetting the ideal light field intensity ratio to 1:0.62:0.94:0.99. The results after retraining are as follows. Figure 5 As shown, after 200 training rounds, the accuracy reached 70%, and after another 200 training rounds, the accuracy reached 90%. It can be seen that as the intensity of the ideal light field changes, the photon diffraction neural network can further improve the accuracy of atmospheric intensity recognition, and the training time decreases as the training step size increases.

[0079] 2) Obtain beams under the influence of different atmospheric intensities;

[0080] In this embodiment, 20 OAM beam images affected by the atmosphere are randomly generated again as beams to be detected.

[0081] 3) Input the beam into the trained photon diffraction neural network;

[0082] 4) Detect the output light field intensity of a photon diffraction neural network based on a light intensity detector;

[0083] 5) Based on the correspondence between light field intensity and atmospheric turbulence intensity, complete the atmospheric turbulence intensity detection.

[0084] In this embodiment, the improved network described above is used to detect atmospheric turbulence intensity. The final detection accuracy is 90%, which is consistent with the training results, demonstrating that the method described in this invention has the ability to correctly detect atmospheric turbulence intensity.

[0085] Example 2

[0086] This embodiment provides an atmospheric turbulence intensity detection system based on a photon diffraction neural network, used to implement the method shown in Embodiment 1, including:

[0087] A photonic diffraction neural network, comprising a forward propagation model and a backward propagation model, wherein the backward propagation model is an error backward propagation based on a loss function, used to train the photonic diffraction neural network;

[0088] The input module is used to acquire light beams under different atmospheric intensities and input the light beams into the trained photon diffraction neural network.

[0089] A light intensity detector is used to detect the intensity of the output light field of a photon diffraction neural network;

[0090] The output module is used to complete the atmospheric turbulence intensity detection based on the light field intensity detected by the light intensity detector and the correspondence between the light field intensity and the atmospheric turbulence intensity. The correspondence between the light field intensity and the atmospheric turbulence intensity is that the stronger the atmospheric turbulence intensity, the weaker the light field intensity.

[0091] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for detecting atmospheric turbulence intensity based on a photon diffraction neural network, characterized in that, Includes the following steps: A photon diffraction neural network is established and trained, the photon diffraction neural network including a forward propagation model and a backward propagation model; Acquire light beams under the influence of different atmospheric intensities; The beam is fed into the trained photon diffraction neural network; The intensity of the output light field of a photon diffraction neural network is detected based on a light intensity detector. Based on the correspondence between light field intensity and atmospheric turbulence intensity, atmospheric turbulence intensity is detected. The correspondence between light field intensity and atmospheric turbulence intensity is as follows: the stronger the atmospheric turbulence intensity, the weaker the light field intensity. The photon diffraction neural network includes k The diffraction layer consists of multiple diffraction layers, each node in which is connected to the next layer. The center of the diffraction layer is the origin of the coordinate system. The light beam propagates along the z-direction. Each node is a neuron in the neural network, and the amplitude and phase of the neuron are used as training parameters. The light wave generated by a beam passing through a node in the diffraction layer can be considered as a secondary wave source: in, Representing the diffraction layer, Representing the The first layer of diffraction layer 1 node, ( x , y , z ) represents the beam coordinates, ( x i , y i , z i ) indicates the first The coordinates of each node, , representing the beam of light to the th The distance between nodes, where λ is the wavelength of the beam; The amplitude and phase of the secondary wave source are determined by the input wave and transmission coefficient at that node. t The decision is made in the first... The first layer The output of each node is represented as: in, Representing the l The first layer The input wave at the nth node, g represents the nth node. All output waves from the layer propagate to the nodes via different paths. i Beamset; The transmission coefficient is determined by the amplitude. and phase composition ; The forward propagation model is determined based on Fresnel diffraction theory. When the input beam propagates to the first... Layer time is represented as After the first After the diffraction of the layers, the light field ; through interlayer distance The transmission of light is equivalent to the light field undergoing a transmission matrix representing Fresnel diffraction. H Then: the first The output light field of the layer is represented as: in, These are frequency domain coordinates. , n Indicates wave number, ; The backpropagation model is an error backpropagation based on a loss function, used to train a photon diffraction neural network.

2. The method for detecting atmospheric turbulence intensity based on a photon diffraction neural network according to claim 1, characterized in that, The loss function is the difference between the light field intensity measured by the light intensity detector and the ideal light field intensity of the forward propagation output light field. in, N This represents the number of nodes in the output layer of the last layer of the photon diffraction neural network. For nodes i The output light field intensity, For nodes i The ideal light field intensity.

3. An atmospheric turbulence intensity detection system based on a photon diffraction neural network, characterized in that, include: A photon diffraction neural network, comprising a forward propagation model and a backward propagation model; The input module is used to acquire light beams under different atmospheric intensities and input the light beams into the trained photon diffraction neural network. A light intensity detector is used to detect the intensity of the output light field of a photon diffraction neural network; The output module is used to complete the atmospheric turbulence intensity detection based on the light field intensity detected by the light intensity detector and the correspondence between the light field intensity and the atmospheric turbulence intensity. The correspondence between the light field intensity and the atmospheric turbulence intensity is: the stronger the atmospheric turbulence intensity, the weaker the light field intensity. The photon diffraction neural network includes k The diffraction layer consists of multiple diffraction layers, each node in which is connected to the next layer. The center of the diffraction layer is the origin of the coordinate system. The light beam propagates along the z-direction. Each node is a neuron in the neural network, and the amplitude and phase of the neuron are used as training parameters. The light wave generated by a beam passing through a node in the diffraction layer can be considered as a secondary wave source: in, Representing the diffraction layer, Representing the The first layer of diffraction layer 1 node, ( x , y , z ) represents the beam coordinates, ( x i , y i , z i ) indicates the first The coordinates of each node, , representing the beam of light to the th The distance between nodes, where λ is the wavelength of the beam; The amplitude and phase of the secondary wave source are determined by the input wave and transmission coefficient at that node. t The decision is made in the first... The first layer The output of each node is represented as: in, Representing the l The first layer The input wave at the nth node, g represents the nth node. All output waves from the layer propagate to the nodes via different paths. i Beamset; The transmission coefficient is determined by the amplitude. and phase composition ; The forward propagation model is determined based on Fresnel diffraction theory. When the input beam propagates to the first... Layer time is represented as After the first After the diffraction of the layers, the light field ; through interlayer distance The transmission of light is equivalent to the light field undergoing a transmission matrix representing Fresnel diffraction. H Then: the first The output light field of the layer is represented as: in, These are frequency domain coordinates. , n Indicates wave number, ; The backpropagation model is an error backpropagation based on a loss function, used to train a photon diffraction neural network.

4. The atmospheric turbulence intensity detection system based on a photon diffraction neural network according to claim 3, characterized in that, The loss function is the difference between the light field intensity measured by the light intensity detector and the ideal light field intensity of the forward propagation output light field. in, N This represents the number of nodes in the output layer of the last layer of the photon diffraction neural network. For nodes i The output light field intensity, For nodes i The ideal light field intensity.

Images