A large-capacity optical communication adaptive damage compensation and demodulation system

Abstract

The application discloses a large-capacity optical communication adaptive damage compensation and demodulation system, and is applied to the technical field of optical communication.The system comprises a beam splitter, an XFTC-Net model, a wavefront corrector and an OAM demultiplexer.The beam splitter divides the received distorted OAM light beam into transmitted light and reflected light, and the transmitted light and the reflected light are both distorted OAM information light beams; the transmitted light is input into the wavefront corrector, and the reflected light is input into the XFTC-Net model; the wavefront corrector completes adaptive damage compensation of the distorted OAM light beam by using the first output result of the XFTC-Net model; and the OAM demultiplexer completes demodulation of the OAM light beam by using the second output result of the XFTC-Net model.The application can realize mode recognition of the OAM light beam on the basis of completing atmospheric turbulence distortion phase reconstruction, does not need to additionally set a probe light beam channel, effectively reduces the system complexity while ensuring the correction accuracy, and realizes the OAM integrating miniaturization and high precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optical communication technology, more particularly to a large-capacity optical communication adaptive damage compensation and demodulation system. BACKGROUND

[0002] Orbital Angular Momentum (OAM) as a new spatial dimension of light beams, whether it is used for coding or as a carrier multiplexing, has obvious advantages compared to traditional dimensions such as frequency and polarization, and can theoretically achieve unlimited improvement of communication capacity. It is a new and popular field of current wireless optical communication technology. Spatial division multiplexing wireless optical communication technology based on OAM is an important technical direction to break through the capacity limit of current wireless optical communication systems and solve the bandwidth crisis.

[0003] However, in actual OAM wireless optical communication links, the turbulence effect caused by low-altitude atmospheric channels is the main challenge faced by OAM wireless optical links. When OAM light beams pass through atmospheric channels, atmospheric turbulence will cause beam drift, flicker and dynamic distortion of OAM spatial multiplexing beam structure phase, resulting in crosstalk between OAM modes, increased bit error rate and degraded transmission performance. Therefore, compensating for the effects of atmospheric turbulence on OAM light beams is crucial to improving the performance of OAM wireless optical communication systems.

[0004] Currently, signal damage suppression techniques for OAM wireless optical communication can be divided into two categories: Digital Signal Processing (DSP) and Adaptive Optics (AO) technology, which has gradually attracted attention in recent years. When using time-domain or frequency-domain multiple-input multiple-output DSP technology to reduce the influence of channel disturbance, the complexity and computational complexity of the structure will increase rapidly with the increase in the number of channels, making it difficult to balance performance improvement and system cost. Unlike DSP technology, adaptive optics technology directly processes optical signals in the optical domain, which can compensate for beam distortion caused by atmospheric turbulence and other factors in real time, and the scheme structure is less affected by the number of communication system channels. Therefore, compared with DSP technology, adaptive optics technology is an effective means to achieve flexible and controllable damage suppression of OAM light beams.

[0005] Due to the influence of OAM beam phase singularity and other factors, the traditional AO technology, including the method based on Shack-Hartmann wavefront sensor (SH-WFS), the scheme based on phase retrieval algorithm (PRA) and the scheme based on stochastic parallel gradient descent (SPGD) algorithm, needs multiple iterations and highly complex system, and it is difficult to balance the factors such as compensation accuracy and speed, system complexity and other factors, which increases the instability of the compensation system. Machine Learning (ML), especially Deep Learning (DL), brings new solutions to the AO scheme design of OAM wireless optical communication system. However, due to the limited receptive field of Convolutional Neural Networks (CNN), it is challenging to capture global information, resulting in poor performance in extracting detailed features such as texture and shape. And from the perspective of system structure, the above-mentioned AO technology is mainly suitable for AO system assisted by Gaussian or OAM probe beam, that is, a probe branch is designed, the phase difference caused by turbulence is calculated by using the probe beam, and then the OAM information beam is corrected. In this way, the existence of the probe branch increases the complexity and cost of the system. Therefore, how to provide a large-capacity optical communication adaptive damage compensation and demodulation system without a probe branch is a problem that those skilled in the art need to solve. SUMMARY

[0006] Therefore, the present application provides a large-capacity optical communication adaptive damage compensation and demodulation system, which improves the correction accuracy of the adaptive optics (AO) system while significantly reducing the system complexity without a probe branch.

[0007] In order to achieve the above purpose, the present application provides the following technical scheme:

[0008] A large-capacity optical communication adaptive damage compensation and demodulation system, comprising: a beam splitter, an XFTC-Net model, a wavefront corrector, an OAM demultiplexer; the received distorted OAM beam is divided into transmitted light and reflected light by the beam splitter, both the transmitted light and the reflected light are distorted OAM information beams, the transmitted light is input to the wavefront corrector, and the reflected light is input to the XFTC-Net model; the wavefront corrector completes adaptive damage compensation of the distorted OAM beam by using the first output result of the XFTC-Net model; and the OAM demultiplexer completes OAM beam demodulation by using the second output result of the XFTC-Net model.

[0009] Optionally, the XFTC-Net model contains a CCD camera for capturing the distorted light intensity of the reflected light, reconstructing the distorted turbulent phase, and identifying the OAM beam mode. The XFTC-Net model uses the CCD camera to capture the distorted light intensity pattern of the reflected light, completes the distorted turbulent phase reconstruction and OAM beam mode identification, and the wavefront corrector uses the distorted turbulent phase reconstruction result of the XFTC-Net model to complete the adaptive compensation of the distorted OAM beam. The OAM demultiplexer uses the OAM beam mode identification result of the XFTC-Net model to complete the OAM beam demodulation, and completes the efficient correction and demodulation of the distorted OAM beam in the wireless optical communication system without the need for a probe beam.

[0010] Optionally, the XFTC-Net model includes a downsampling path module, an upsampling path module, a lower path module, and an upper path module. The downsampling path module extracts features from the input distorted light intensity pattern, the upsampling path module reconstructs and restores the features extracted by the downsampling path module, the lower path module reconstructs the atmospheric turbulent phase screen, and the upper path module completes OAM mode identification.

[0011] Optionally, the downsampling path module includes an encoding block, a Transformer unit, and a pooling layer. The encoding block preliminarily extracts features from the input distorted light intensity pattern, the pooling layer reduces the spatial dimension through a max-pooling operation, and the Transformer unit captures global features through a self-attention mechanism after three encoding block processing and pooling layer processing. After the Transformer unit processes the global features, the feature map processed by the pooling layer is input into the Transformer unit to further mine the subtle changes and mutual relationships of the light beam in the overall structure.

[0012] Optionally, the upsampling path module includes an encoding block, a Transformer unit, and an upsampling layer. The features extracted by the downsampling path module are expanded through an upsampling operation of the upsampling layer and then input into the Transformer unit to fuse multi-level feature information and capture global context relationships. The Transformer unit outputs a feature map, which is expanded in size by the upsampling layer and then input into the encoding block for further extraction and optimization of feature information. The feature map output by the encoding block is expanded in size by the upsampling layer and then input into the encoding block again to continue to strengthen the feature representation. Finally, the feature map is expanded in size by the upsampling layer and input into the encoding block for final optimization and enhancement of high-resolution features. The feature map in the upsampling path module transmits information to the lower path module through a skip connection to help the reconstruction task of the atmospheric turbulent phase screen.

[0013] Optionally, the lower path module comprises a jump connection, an up-sampling layer and a convolution layer, the jump connection fuses the feature maps of various sizes extracted by the down-sampling path module and the feature maps of various sizes expanded by the up-sampling path module, so that the lower path module can simultaneously utilize the deep features in the down-sampling path module and the feature maps gradually reconstructed in the up-sampling path module, and the fused feature maps are expanded in size by the up-sampling layer, and all the fused feature maps after up-sampling are output as the first output result by the jump connection.

[0014] Optionally, the upper path module comprises an encoding block, a pooling layer and a fully connected layer, the upper path module performs multiple convolution and pooling operations on the features extracted by the down-sampling path module, and outputs the OAM mode recognition result as the second output result through the fully connected layer.

[0015] Via the above technical solution, compared with the prior art, the present application provides a large-capacity optical communication adaptive damage compensation and demodulation system, which has the following beneficial effects:

[0016] 1. The present application fully utilizes the advantages of CNN fine-grained feature extraction and Transformer global feature extraction, realizes the mode recognition of OAM light beams on the basis of completing the reconstruction of the phase of atmospheric turbulence distortion, does not need to additionally set a probe light beam path, effectively reduces the system complexity while ensuring the correction accuracy, and realizes the integration of miniaturization and high precision of OAM.

[0017] 2. Compared with the prior art, the present application can more quickly and accurately complete the prediction of the turbulence phase screen and the identification of the OAM mode number, and then realize the OAM light beam damage compensation with high correction accuracy, and improve the adaptive compensation efficiency and transmission performance of the large-capacity wireless optical communication system. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only embodiments of the present application, and those skilled in the art can obtain other drawings according to the provided drawings without any creative labor.

[0019] Figure 1 It is a structural schematic diagram of the optical communication adaptive damage compensation and demodulation system of the present application.

[0020] Figure 2 It is a structural schematic diagram of the XFTC-Net model of the present application.

[0021] Figure 3 It is a XFTC-Net model training iteration loss curve diagram in the embodiment of the present application.

[0022] Figure 4 This is a diagram showing the OAM beam intensity distribution before and after compensation in an embodiment of the present invention;

[0023] Figure 5 This is a schematic diagram of the system bit error rate performance before and after compensation in an embodiment of the present invention;

[0024] In the diagram: 101-beam splitter, 102-XFTC-Net model, 103-wavefront corrector, 104-OAM demultiplexer, 1021-first output result, 1022-second output result, 201-downsampling path module, 202-upsampling path module, 203-down path module, 204-up path module. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] This invention discloses a high-capacity optical communication adaptive impairment compensation and demodulation system, such as... Figure 1 As shown, it includes: a beam splitter 101, an XFTC-Net model 102, a wavefront corrector 103, and an OAM demultiplexer 104. The beam splitter 101 splits the received distorted OAM beam into transmitted light and reflected light. Both transmitted light and reflected light are distorted OAM information beams. The transmitted light is input to the wavefront corrector 103, and the reflected light is input to the XFTC-Net model 102. The wavefront corrector 103 uses the first output result 1021 of the XFTC-Net model 102 to complete adaptive damage compensation of the distorted OAM beam. The OAM demultiplexer 104 uses the second output result 1022 of the XFTC-Net model 102 to complete OAM beam demodulation.

[0027] Further, the XFTC-Net model 102 contains a CCD camera for capturing the distorted light intensity of the light beam, reconstructing the distorted turbulence phase, and calculating the OAM light beam mode recognition. The XFTC-Net model 102 uses the CCD camera to capture the distorted light intensity of the reflected light, complete the distorted turbulence phase reconstruction and OAM light beam mode recognition, and the wavefront corrector 103 uses the distorted turbulence phase reconstruction result of the XFTC-Net model 102 to complete the adaptive compensation of the distorted OAM light beam. The OAM demultiplexer 104 uses the OAM light beam mode recognition result of the XFTC-Net model 102 to complete the OAM light beam demodulation, and completes the efficient correction and demodulation of the distorted OAM light beam in the wireless optical communication system without the need for a probe light beam.

[0028] Further, as shown in FIG. 2, the XFTC-Net model 102 includes a down-sampling path module 201, an up-sampling path module 202, a lower path module 203, and an upper path module 204. The down-sampling path module 201 extracts features from the input distorted light intensity pattern, the up-sampling path module 202 reconstructs and restores the features extracted by the down-sampling path module 201, the lower path module 203 reconstructs the atmospheric turbulence phase screen, and the upper path module 204 completes the OAM mode recognition. Figure 2

[0029] Further, the down-sampling path module 201 includes an encoding block, a Transformer unit, and a pooling layer. The encoding block preliminarily extracts features from the input distorted light intensity pattern, the pooling layer reduces the spatial dimension through a max-pooling operation, and the Transformer unit captures global features through a self-attention mechanism after three encoding block processing and pooling layer processing. After the Transformer unit processes the global features, the pooling layer performs a max-pooling operation, and the feature map processed by the pooling layer is input to the Transformer unit to further mine the subtle changes and mutual relationships of the light beam in the overall structure.

[0030] ​In the embodiment of the present application, each encoding block contains two layers of convolutional layers, batch normalization layers and linear rectifier units, the convolutional kernels of the convolutional units are all 3x3, the stride is 1, the input of the down-sampling path module 201 is an intensity map of the distorted OAM beam with a size of 128x128, which contains the spatial distribution information of the beam, preliminary feature extraction is performed through the first encoding block, these convolutional operations can extract local low-level features (such as edges, textures, etc.) from the input image, and the output size remains a feature map with a size of 128x128; then the first max-pooling operation is performed, which down-samples the size of the feature map from 128x128 to 64x64, reduces the spatial dimension while retaining the most significant features, helps to reduce the amount of calculation and memory usage, so that the network is more efficient when processing high-dimensional data; after the second layer of encoding block operation, the size of the feature map remains 64x64, the extracted features are more abstract and complex at this time, and can capture larger scale structural information; the feature map size is down-sampled to 32x32 again through the pooling operation, further reducing the spatial resolution while maintaining the diversity and key information of the information; then the third layer of encoding block continues to extract deeper features, and the size of the feature map remains 32x32, but the features have become more abstract and can capture more complex beam patterns; the pooling operation down-samples the size of the feature map to 16x16, the spatial dimension of the feature map is significantly reduced at this stage, providing a more compact and efficient feature representation; after each convolution and pooling, the size of the feature map gradually decreases, but the abstractness of the features is constantly improved, the preliminary convolutional operation extracts low-level features, and with the increase of the network depth, the subsequent convolutional layers extract higher-level features;

[0031] The Transformer unit includes layer normalization, multi-head self-attention mechanism and feed-forward network. After the foregoing encoding and pooling operations, a 16x16 feature map is finally obtained, which is input into the Transformer unit. The Transformer captures global features through the self-attention mechanism, can effectively model the long-range dependencies between features, and extract the mutual relationship between different regions in the image. After the global features are processed in the Transformer unit, the size of the output feature map is 16x16, and then a max-pooling operation is performed to downsample the feature map size to 8x8, further reducing the spatial dimension and highlighting more condensed and abstract global features, which helps to retain the core information after the Transformer processing. The 8x8 feature map after the pooling operation is input into the Transformer unit again to further explore the subtle changes and mutual relationships of the light beams on the overall structure. This path combines the advantages of convolutional neural network (CNN) and Transformer. The convolutional layer is responsible for extracting local features (such as edges, textures and small-scale patterns), and the pooling operation reduces the spatial dimension and redundant information; while the Transformer unit takes over the extraction of global features, which can capture the global dependency relationship between different regions. Through the organic combination of local and global information, the network can more comprehensively understand the input data and more effectively reconstruct the spatial distribution information of atmospheric turbulence. This is particularly important for processing complex data such as distorted OAM light beams, because these light beams often have complex morphology and structure, and there may be strong mutual relationships between different regions.

[0032] Further, the up-sampling path module 202 includes an encoding block, a Transformer unit and an up-sampling layer. The features extracted by the down-sampling path module 201 are expanded by the up-sampling layer and then input into the Transformer unit to fuse multi-level feature information and capture global context relationships. The Transformer unit outputs a feature map, which is further expanded in size by the up-sampling layer and then input into the encoding block for further extraction and optimization of feature information. The feature map output by the encoding block is expanded in size by the up-sampling layer and then input into the encoding block again to continue to strengthen the feature representation. Finally, the feature map is expanded in size by the up-sampling layer and input into the encoding block for final optimization and enhancement of high-resolution features. The feature map in the up-sampling path module 202 transmits information to the lower path module 203 through a skip connection, which helps the reconstruction task of the atmospheric turbulence phase screen.

[0033] In the embodiment of the present application, the 8*8 feature map is expanded to 16*16 through the upsampling operation, and then is input to the Transformer unit to fuse multi-level feature information, capture global context relationship, and improve the reconstruction capability of the features of the down-sampling path. The Transformer unit outputs a 16*16 feature map, which is then enlarged to 32*32 in size through the upsampling operation, and is input to the encoding block for further extraction and optimization of feature information, and outputs a 32*32 feature map. Then, the feature map is expanded to 64*64 through the upsampling operation, and is input to the encoding block again to continue to strengthen the feature representation and ensure that the details of the atmospheric turbulence phase screen are effectively reconstructed. Finally, the feature map is expanded to 128*128 through the upsampling operation, and is input to the encoding block for final optimization and enhancement of the high-resolution feature. In this path, the feature map is not directly output, but the information is transmitted to the lower path through the skip connection to help the reconstruction task of the atmospheric turbulence phase screen; the Transformer unit is used to fuse multi-level feature information in this path to improve the reconstruction capability of the features of the down-sampling path; at the same time, the encoding block strengthens the abstraction and representation of the features to ensure that the useful information of the down-sampling path can be effectively used in the reconstruction process.

[0034] Further, the lower path module 203 includes a skip connection, an upsampling layer and a convolution layer. The skip connection fuses the feature maps of various sizes extracted by the down-sampling path module 201 with the feature maps of various sizes expanded by the up-sampling path module 202, so that the lower path module 203 can simultaneously use the deep features in the down-sampling path module 201 and the feature maps reconstructed step by step in the up-sampling path module 202. The fused feature maps are expanded in size by the upsampling layer, and the prediction result of the atmospheric turbulence phase screen output by the skip connection of all the fused feature maps after upsampling is the first output result 1021.

[0035] In the embodiment of the present application, at each layer of the skip connection, the feature maps of the same dimension are connected together to share information, thereby strengthening the understanding of the model to the local details and the global structure. In this path, the fused feature maps are respectively up-sampled to a size of 128*128 to ensure the high resolution of the feature maps and retain more key information.

[0036] Further, the upper path module 204 includes an encoding block, a pooling layer and a fully connected layer. The upper path module 204 performs multiple convolution and pooling operations on the features extracted by the down-sampling path module 201, and outputs the OAM mode recognition result through the fully connected layer as the second output result 1022.

[0037] In the embodiment of the present application, the XFTC-Net model 102 is compared with the Y-Net-based AO scheme, and the training loss is as follows: Figure 3It can be seen that, with the increase of the number of iterations, the loss function presents a downward trend and gradually converges, indicating that the XFTC-Net model 102 trained in the embodiment can effectively complete the turbulent phase screen prediction and OAM mode number identification. When the iteration is about 40 steps, the MSE value corresponding to the XFTC-Net trained in the embodiment is 0.12, and the MSE value of the Y-Net in the control group is 0.44. The final MSE value of the XFTC-Net converges to 0.07, and the final MSE value of the Y-Net in the control group converges to 0.37. In addition, in terms of the convergence of the BCE, the BCE curves of the XFTC-Net and the Y-Net can quickly converge, and the XFTC-Net exhibits a better BCE value. Therefore, whether from the MSE convergence (i.e. the turbulent phase screen prediction effect) or the BCE convergence (i.e. the OAM mode number identification effect), the XFTC-Net proposed in the present application has better convergence performance.

[0038] The comparison results of the OAM beam intensity distribution before and after correction are shown in Figure 4 It can be seen that, with the increase of the turbulence intensity, the OAM beam distortion caused by the turbulence is more serious. Compared with the Y-Net-AO scheme, the OAM beam compensated by the XFTC-Net AO scheme presents a more uniform ring-shaped intensity distribution.

[0039] The bit error rate performance of the OAM wireless optical communication system before and after correction is shown in Figure 5 It can be seen that, after the correction by the Y-Net-AO and the XFTC-Net AO scheme, the system bit error rate is effectively reduced. The bit error rate performance after the correction by the XFTC-Net AO scheme is obviously better than that after the correction by the Y-Net-AO scheme. This further confirms the effectiveness and superiority of the present application in the adaptive compensation of the OAM wireless optical communication link.

[0040] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts of each embodiment can be referred to each other.

[0041] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A high-capacity optical communication adaptive impairment compensation and demodulation system, characterized in that, include: The components include a beam splitter, an XFTC-Net model, a wavefront corrector, and an OAM demultiplexer. The beam splitter splits the received distorted OAM beam into transmitted and reflected beams, both of which are distorted OAM information beams. The transmitted beam is input to the wavefront corrector, and the reflected beam is input to the XFTC-Net model. The wavefront corrector uses the first output of the XFTC-Net model to perform adaptive damage compensation for the distorted OAM beam, and the OAM demultiplexer uses the second output of the XFTC-Net model to perform OAM beam demodulation. The XFTC-Net model incorporates a CCD camera for capturing distorted beam intensity, reconstructing distorted turbulence phase, and calculating OAM beam pattern recognition. The XFTC-Net model uses the CCD camera to acquire distorted beam intensity patterns of reflected light, completing distorted turbulence phase reconstruction and OAM beam pattern recognition. The wavefront corrector uses the distorted turbulence phase reconstruction results from the XFTC-Net model to perform adaptive damage compensation for the distorted OAM beam. The OAM demultiplexer uses the OAM beam pattern recognition results from the XFTC-Net model to demodulate the OAM beam, achieving efficient correction and demodulation of distorted OAM beams in wireless optical communication systems without the need for a probe beam. The XFTC-Net model includes: a downsampling path module, an upsampling path module, a downsampling path module, and an upsampling path module. The downsampling path module extracts features from the input distorted beam intensity pattern, the upsampling path module reconstructs and restores the features extracted by the downsampling path module, the downsampling path module reconstructs the atmospheric turbulence phase screen, and the upsampling path module completes OAM pattern recognition. The down-path module includes skip connections, upsampling layers, and convolutional layers. Skip connections fuse feature maps of various sizes extracted by the downsampling path module with feature maps of various sizes expanded by the upsampling path module. This allows the down-path module to utilize both deep features from the downsampling path module and feature maps reconstructed step by step from the upsampling path module. The fused feature maps are expanded in size by the upsampling layer. All the fused feature maps after upsampling are combined through skip connections to output the prediction result of the atmospheric turbulence phase screen as the first output result. The upper path module includes an encoding block, a pooling layer, and a fully connected layer. The upper path module performs multiple convolution and pooling operations on the features extracted by the downsampling path module, and outputs the OAM pattern recognition result through the fully connected layer as the second output result.

2. The high-capacity optical communication adaptive impairment compensation and demodulation system according to claim 1, characterized in that, The downsampling path module includes an encoding block, a Transformer unit, and a pooling layer. The encoding block extracts preliminary features from the intensity pattern of the input distorted beam. The pooling layer performs downsampling through max pooling to reduce spatial dimensions. After three processing steps by the encoding block and the pooling layer, the Transformer unit captures global features through a self-attention mechanism. After the Transformer unit processes the global features, max pooling is performed through the pooling layer. The feature map processed by the pooling layer is then input into the Transformer unit to further explore the subtle changes and interrelationships of the beam in its overall structure.

3. The high-capacity optical communication adaptive impairment compensation and demodulation system according to claim 1, characterized in that, The upsampling path module includes an encoding block, a Transformer unit, and an upsampling layer. The features extracted by the downsampling path module are expanded by the upsampling layer and then input into the Transformer unit to fuse multi-level feature information and capture global contextual relationships. The Transformer unit outputs a feature map, which is further expanded by the upsampling layer and then input into the encoding block for further extraction and optimization of feature information. The feature map output by the encoding block is expanded by the upsampling layer and input into the encoding block again to further enhance the feature representation. Finally, the feature map is expanded by the upsampling layer and input into the encoding block for final optimization and enhancement of high-resolution features. The feature maps in the upsampling path module pass information to the downsampling module through skip connections, which helps in the reconstruction task of the atmospheric turbulence phase screen.

Images