PON optical performance monitoring method and system

Abstract

The invention provides a PON optical performance monitoring method and system, and the method comprises the steps: obtaining a receiving signal which is transmitted to a receiving end through a PON network and is subjected to digital signal processing, and obtaining system parameters related to a PON network communication system; performing fractional order Fourier transform on the received signal to construct a three-dimensional spectrogram, and collecting a two-dimensional projection image of the three-dimensional spectrogram on an amplitude-order plane; the two-dimensional projection image and the system parameters serve as input of a pre-trained multi-modal fusion convolutional neural network-full connection neural network CNN-FCNN, a convolutional neural network branch performs feature extraction on the input two-dimensional projection image, and a full connection network branch performs feature extraction on the input system parameters; and the fusion layer fuses the extracted features of different modes, and outputs an optical signal-to-noise ratio estimation value and a receiving optical power estimation value at the output layer.

Description

Technical Field

[0001] The present invention relates to the technical field of optical fiber communication, and in particular to a PON optical performance monitoring method and system. Background Art

[0002] With the explosive growth of data traffic and the rapid development of cloud computing and big data applications, the demand for high-speed data center optical interconnects is increasing. Data center Passive Optical Interconnect (POI), due to its unique advantages, has become a key development direction for current data center network architectures. Within the POI architecture, Passive Optical Network (PON), due to its unique advantages, has become a key development direction for current data center network architectures. As an efficient fiber-optic access technology, PON can meet the demand for high-speed, low-cost broadband access services. It has been widely used in scenarios such as home broadband and enterprise access, becoming a key technology driving the upgrade of communications infrastructure and supporting the development of smart cities and the Internet of Things.

[0003] However, due to the passive nature of PON and the dense links within data centers, signals are inevitably affected by loss, dispersion, and nonlinear effects during transmission. After long-distance transmission and passive splitting, the optical signal-to-noise ratio (OSNR) of optical signals can be severely affected, further impacting the reliability of the communication system. Therefore, real-time monitoring and evaluation of link quality is crucial to ensuring stable operation of data centers.

[0004] Optical Signal-to-Noise Ratio (OSNR) is a key indicator in optical performance monitoring (OPM). Real-time monitoring of OSNR is crucial for optimizing PON network resource allocation and improving communication system transmission performance.

[0005] Researchers have proposed a variety of methods for monitoring the optical signal-to-noise ratio (OSNR) in optical communication systems, including statistical moment-based (SMB), error vector magnitude (EVM), asynchronous amplitude histogram (AAH), differential pilot, spectrum analysis, and deep learning-based methods. Statistical moment-based methods rely solely on the amplitude envelope of the received signal, effectively extracting high-dimensional signal features and efficiently separating the signal from noise, and have garnered widespread attention from researchers both domestically and internationally. However, these methods fail to fully account for the impact of nonlinear phase noise on OSNR monitoring, resulting in significant OSNR monitoring errors for systems with higher-order modulation formats such as quadrature amplitude modulation (QAM). Furthermore, existing statistical moment-based methods typically employ fitting methods to obtain closed-form solutions, severely limiting their robustness for OSNR monitoring under varying PON system parameters.

[0006] In recent years, deep learning has been widely applied in many fields due to its powerful feature extraction capabilities. Many deep learning-based methods for optical signal-to-noise ratio monitoring have been proposed by scholars at home and abroad. However, the lack of interpretability of neural networks and their low accuracy on out-of-sample data have sparked widespread controversy within the academic community regarding their scientific validity and reliability.

[0007] Therefore, how to design a PON optical performance monitoring method and system to overcome the problems of poor interpretability and low accuracy on out-of-sample data of traditional neural networks is a technical problem that needs to be solved urgently. Summary of the Invention

[0008] In view of this, an embodiment of the present invention provides a PON optical performance monitoring method to eliminate or improve one or more defects in the prior art.

[0009] One aspect of the present invention provides a PON optical performance monitoring method, which includes the following steps: obtaining a received signal that has been transmitted through the PON network and reached a receiving end and has undergone digital signal processing, and obtaining system parameters related to the PON network communication system; performing a fractional-order Fourier transform on the received signal to construct a three-dimensional spectrum, and collecting a two-dimensional projection image of the three-dimensional spectrum on the "amplitude-order" plane; using the two-dimensional projection image and the system parameters as inputs of a pre-trained multimodal fusion convolutional neural network-fully connected neural network CNN-FCNN, the convolutional neural network branch of the CNN-FCNN extracting features from the input two-dimensional projection image, and the fully connected network branch of the CNN-FCNN extracting features from the input system parameters; the fusion layer of the CNN-FCNN extracts The features of different modalities extracted by the convolutional neural network branch and the fully connected network branch are fused, and the optical signal-to-noise ratio (OSNR) and received optical power (RXOP) estimates are output at the output layer. During the CNN-FCNN training process, the OSNR reference value measured by the optical spectrum analyzer at the receiving end and the RXOP reference value measured by the optical power meter are used as labels to construct a training set. Furthermore, the loss in PON network signal transmission is calculated as the equation loss using the nonlinear Schrödinger equation. The OSNR estimation error is calculated based on the OSNR estimate and the OSNR reference value, and the RXOP estimation error is calculated based on the RXOP estimate and the RXOP reference value. The weighted sum of the equation loss, OSNR estimation error, and RXOP estimation error is used as the loss function for CNN-FCNN training.

[0010] In some embodiments of the present invention, the system parameters include baud rate, transmit power and transmission distance.

[0011] In some embodiments of the present invention, the convolutional neural network branch includes a preset number of groups of convolution-maximum pooling units, each group of convolution-maximum pooling units includes a preset number of convolution layers and maximum pooling layers, the preset number of convolution layers are used to extract features of the input two-dimensional projection image, and the maximum pooling layer is used for downsampling; the fully connected network branch includes a single fully connected layer containing a preset number of neurons.

[0012] In some embodiments of the present invention, a fractional Fourier transform is performed on the received signal to construct a three-dimensional spectrogram, and a two-dimensional projection image of the three-dimensional spectrogram on the "amplitude-order" plane is collected, including: performing fractional Fourier transforms of different orders on the received signal to construct a three-dimensional spectrogram; for the three-dimensional spectrograms constructed by fractional Fourier transforms of different orders, taking the maximum amplitude corresponding to the order, thereby obtaining a two-dimensional projection image of the three-dimensional spectrogram on the "amplitude-order" plane.

[0013] In some embodiments of the present invention, the method further comprises: performing a subtraction operation on the estimated value of the received optical power and the estimated value of the optical signal-to-noise ratio to obtain the estimated value of the noise optical power.

[0014] In some embodiments of the present invention, the expression for the equation loss is: ; in, and represent the real and imaginary parts of a given complex number, respectively. The expression is: ; in, represents the loss coefficient, represents the nonlinear coefficient, Indicates the distance that the optical signal propagates along the axial direction in the optical fiber. Indicates the reference value of received optical power. Indicates the reference value of noise optical power, Represents an imaginary unit.

[0015] In some embodiments of the present invention, the loss function used for CNN-FCNN training is mathematically expressed as: ; in, are the weight coefficients of equation loss, optical signal-to-noise ratio estimation error, and received optical power estimation error, respectively. represents the equation loss, represents the optical signal-to-noise ratio estimation error, Indicates the received optical power estimation error.

[0016] Corresponding to the above method, the present invention also provides a PON optical performance monitoring system, including a processor, a memory, and a computer program / instruction stored in the memory, located at the receiving end of the PON network, the processor being used to execute the computer program / instruction. When the computer program / instruction is executed, the system implements the steps of the method described in any one of the above embodiments.

[0017] Corresponding to the above method, the present invention also provides a computer-readable storage medium having a computer program / instruction stored thereon, which implements the steps of the method described in any one of the above embodiments when the computer program / instruction is executed by a processor.

[0018] Corresponding to the above method, the present invention also provides a computer program product, including a computer program / instruction, which implements the steps of the method described in any one of the above embodiments when executed by a processor.

[0019] The PON optical performance monitoring method and system proposed in the present invention construct a loss function based on the nonlinear Schrödinger equation, thereby introducing physical information into the neural network model, which is conducive to overcoming the poor interpretability of neural networks in the prior art. By collecting the two-dimensional projection image of the three-dimensional spectrum on the "amplitude-order" plane as the input of the CNN-FCNN network model, it is conducive to reducing the complexity of the model without losing key information. In addition, by introducing multimodal data containing system parameters and two-dimensional projection images to extract multimodal feature maps, it is conducive to solving the problem of low accuracy of neural networks on out-of-sample data.

[0020] Additional advantages, objects, and features of the present invention will be set forth in part in the following description and will become apparent to those skilled in the art upon examination of the following or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained by the structures particularly pointed out in the description and drawings.

[0021] Those skilled in the art will understand that the purposes and advantages that can be achieved by the present invention are not limited to the above specific descriptions, and the above and other purposes that can be achieved by the present invention will be more clearly understood based on the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are intended to provide a further understanding of the present invention, constitute a part of this application, and do not constitute a limitation of the present invention. The components in the drawings are not drawn to scale, but are merely for the purpose of illustrating the principles of the present invention. To facilitate the illustration and description of certain portions of the present invention, corresponding portions in the drawings may be exaggerated, that is, may be larger than other components in an exemplary device actually manufactured according to the present invention. In the drawings: Figure 1 FIG. 4 is a flow chart of a method for monitoring PON optical performance according to an embodiment of the present invention.

[0023] Figure 2 The flowchart of implementing optical performance monitoring in a coherent optical fiber communication system according to one embodiment of the present invention is shown.

[0024] Figure 3 This is a CNN-FCNN neural network model structure for implementing PON optical performance monitoring in one embodiment of the present invention.

[0025] FIG4( a ) is a schematic diagram of a received signal after being transmitted through a PON network and arriving at a receiving end and undergoing digital signal processing in one embodiment of the present invention.

[0026] FIG4( b ) is a schematic diagram of a three-dimensional spectrum constructed after a received signal is subjected to fractional Fourier transform in one embodiment of the present invention.

[0027] FIG4( c ) is a schematic diagram of a two-dimensional projection image of a three-dimensional spectrum acquired on the “amplitude-order” plane according to an embodiment of the present invention.

[0028] Figure 5 This is a loss function variation curve of PON optical performance monitoring in a coherent optical fiber communication system in one embodiment of the present invention.

[0029] FIG6( a ) shows the optical signal-to-noise ratio estimation error and received optical power estimation error of PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 25 Gbaud and a modulation mode of DP-QSPK in one embodiment of the present invention.

[0030] FIG6( b ) shows the optical signal-to-noise ratio estimation error and received optical power estimation error of PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 30 Gbaud and a modulation mode of DP-QSPK in one embodiment of the present invention.

[0031] FIG6( c ) shows the optical signal-to-noise ratio estimation error and received optical power estimation error of PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 40 Gbaud and a modulation mode of DP-QSPK in one embodiment of the present invention.

[0032] FIG6( d ) shows the optical signal-to-noise ratio estimation error and received optical power estimation error of PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 50 Gbaud and a modulation mode of DP-QSPK according to an embodiment of the present invention.

[0033] FIG7( a ) is a comparison curve of optical signal-to-noise ratio monitoring performance of PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 25 Gbaud according to an embodiment of the present invention and that of a traditional algorithm.

[0034] FIG7( b ) is a comparison curve of optical signal-to-noise ratio monitoring performance of PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 30 Gbaud according to an embodiment of the present invention and that of a traditional algorithm.

[0035] FIG7( c ) is a comparison curve of optical signal-to-noise ratio monitoring performance of PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 40 Gbaud according to an embodiment of the present invention and that of a traditional algorithm.

[0036] FIG7( d ) is a comparison curve of the optical signal-to-noise ratio monitoring performance of the PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 50 Gbaud according to an embodiment of the present invention and that of the traditional algorithm.

[0037] FIG8( a ) shows the optical signal-to-noise ratio estimation error and received optical power estimation error of PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 25 Gbaud and a modulation mode of DP-16QAM in one embodiment of the present invention.

[0038] FIG8( b ) shows the optical signal-to-noise ratio estimation error and received optical power estimation error of PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 30 Gbaud and a modulation mode of DP-16QAM in one embodiment of the present invention.

[0039] FIG8( c ) shows the optical signal-to-noise ratio estimation error and received optical power estimation error of PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 40 Gbaud and a modulation mode of DP-16QAM in one embodiment of the present invention.

[0040] FIG8( d ) shows the optical signal-to-noise ratio estimation error and received optical power estimation error of PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 50 Gbaud and a modulation mode of DP-16QAM in one embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0042] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, the accompanying drawings only show structures and / or processing steps closely related to the solutions according to the present invention, while other details that are not closely related to the present invention are omitted.

[0043] It should be emphasized that the term "include / comprises" when used herein refers to the existence of features, elements, steps or components, but does not exclude the existence or addition of one or more other features, elements, steps or components.

[0044] It should also be noted that, unless otherwise specified, the term "connection" herein may refer not only to a direct connection but also to an indirect connection involving an intermediate.

[0045] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the accompanying drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0046] To address the existing challenges of deep learning network-based optical signal-to-noise ratio (OSNR) monitoring methods, such as the lack of interpretability and low accuracy on out-of-sample data, and the large OSNR monitoring errors in high-order modulation formats caused by statistical moment-based methods ignoring phase factors, a physics-informed OSNR monitoring model can be constructed. Using a physics-informed neural network (PINN), physical constraints can be incorporated into the neural network model, enabling intelligent optical performance diagnosis. This PINN approach, which incorporates physical constraints into the neural network model, has broad applications in various disciplines, including solving the Navier-Stokes equations in fluid mechanics, solving Maxwell's equations in electromagnetics, and modeling optical fiber communication system channels based on the nonlinear Schrödinger equation. This physics-informed OSNR monitoring model significantly enhances the model's interpretability and enables accurate prediction of out-of-sample data, providing an efficient, reliable, and physically interpretable intelligent optical performance diagnosis solution for PON networks.

[0047] Figure 1 This is a flow chart of a method for monitoring PON optical performance according to an embodiment of the present invention. The method comprises the following steps: Step S110: obtaining a received signal transmitted through the PON network to a receiving end and subjected to digital signal processing, and obtaining system parameters related to the PON network communication system.

[0048] Step S120: performing fractional Fourier transform on the received signal to construct a three-dimensional spectrum, and collecting a two-dimensional projection image of the three-dimensional spectrum on the "amplitude-order" plane.

[0049] Step S130: The two-dimensional projection image and the system parameters are used as inputs of a pre-trained multimodal fusion convolutional neural network-fully connected neural network (abbreviated as CNN-FCNN network in the specification). The convolutional neural network branch of the CNN-FCNN performs feature extraction on the input two-dimensional projection image, and the fully connected network branch of the CNN-FCNN performs feature extraction on the input system parameters.

[0050] Step S140: The fusion layer of the CNN-FCNN fuses the features of different modalities extracted by the convolutional neural network branch and the fully connected network branch, and outputs the estimated optical signal-to-noise ratio (OSNR) and the received optical power (ROP) at the output layer.

[0051] During the training process of CNN-FCNN, the optical signal-to-noise ratio reference value measured by the optical spectrum analyzer at the receiving end and the received optical power reference value measured by the optical power meter are used as labels to construct a training set. In addition, the loss in PON network signal transmission is calculated as the equation loss using the nonlinear Schrödinger equation. The optical signal-to-noise ratio estimation error is calculated based on the optical signal-to-noise ratio estimation value and the optical signal-to-noise ratio reference value. The received optical power estimation error is calculated based on the received optical power estimation value and the received optical power reference value. The weighted sum of the equation loss, optical signal-to-noise ratio estimation error, and received optical power estimation error is used as the loss function for CNN-FCNN training.

[0052] In a specific implementation process, the execution subject of this solution is located at the receiving end, for example, it can be a server end, more specifically, it can be a server-side optical network unit (ONU) device located at the server end.

[0053] The PON optical performance monitoring method and system proposed in the present invention construct a loss function based on the nonlinear Schrödinger equation, thereby introducing physical information into the neural network model, which is conducive to overcoming the poor interpretability of neural networks in the prior art. By collecting the two-dimensional projection image of the three-dimensional spectrum on the "amplitude-order" plane as the input of the CNN-FCNN network model, it is conducive to reducing the complexity of the model without losing key information. In addition, by introducing multimodal data containing system parameters and two-dimensional projection images to extract multimodal feature maps, it is conducive to solving the problem of low accuracy of neural networks on out-of-sample data.

[0054] The key point of this solution is that in the neural network of the CNN-FCNN architecture, the loss function is constructed based on the nonlinear Schrödinger equation, and the nonlinear Schrödinger equation is constructed based on physical information, thereby introducing physical information into the neural network model.

[0055] In some embodiments of the present invention, system parameters include baud rate, transmit power, and transmission distance. It should be noted that the present solution is not limited thereto, and the above system parameters refer to system parameters that can be used to characterize the signal transmission characteristics of the PON network. For example, the system parameters may also include channel bandwidth, etc.

[0056] By adopting the embodiment of the invention, multimodal data can be introduced to extract multimodal features. The introduction of multimodal features is conducive to solving the problem of low accuracy of neural networks on out-of-sample data.

[0057] In some embodiments of the present invention, the convolutional neural network branch includes a preset number of groups of convolution-max pooling units, each group of convolution-max pooling units includes a preset number of convolution layers and maximum pooling layers, the preset number of convolution layers are used to extract features of the input two-dimensional projection image, and the maximum pooling layer is used for downsampling.

[0058] In some embodiments of the present invention, the fully connected network branch includes a single fully connected layer including a preset number of neurons.

[0059] In a specific implementation, the convolutional neural network branch includes two sets of convolution-max pooling units, each of which contains 64 convolutional layers with 3×3 convolution kernels and a maximum pooling layer with a stride of 2×2. The fully connected network branch can use a single fully connected layer containing 32 neurons.

[0060] The present invention is not limited thereto, and both the number of the two groups of convolution-maximum pooling units and the scale and size of the maximum pooling units can be flexibly adjusted as needed.

[0061] The above two embodiments of the invention can be freely combined to provide a specific implementation method of the CNN-FCNN network.

[0062] In some embodiments of the present invention, in step S120, fractional Fourier transform is performed on the received signal to construct a three-dimensional spectrum, and a two-dimensional projection image of the three-dimensional spectrum on the "amplitude-order" plane is collected, including: (1) performing fractional Fourier transforms of different orders on the received signal to construct a three-dimensional spectrum; (2) for the three-dimensional spectrum constructed by fractional Fourier transforms of different orders, the maximum amplitude corresponding to the order is taken, thereby obtaining a two-dimensional projection image of the three-dimensional spectrum on the "amplitude-order" plane.

[0063] By adopting this embodiment of the invention, by collecting the two-dimensional projection image of the three-dimensional spectrum on the "amplitude-order" plane as the input of the CNN-FCNN network model, it is beneficial to reduce the complexity of the model without losing key information.

[0064] In some embodiments of the present invention, the method further comprises: performing a subtraction operation on the received optical power estimation value and the optical signal-to-noise ratio estimation value to obtain the noise optical power estimation value.

[0065] Based on the above embodiments of the invention, it can be known that the noise optical power estimation value can also be used to characterize the optical performance monitoring of the PON network. The parameters used to characterize the optical performance can be any one or a combination of optical signal-to-noise ratio, received optical power and noise optical power.

[0066] Furthermore, in order to better embody the present invention, a mathematical expression of the equation loss is given as follows: ; in, and represent the real and imaginary parts of a given complex number, respectively. The expression is: ; in, represents the loss coefficient, represents the nonlinear coefficient, Indicates the distance that the optical signal propagates along the axial direction in the optical fiber. Indicates the reference value of received optical power. Indicates the reference value of noise optical power, Represents an imaginary unit.

[0067] In addition, the loss function for CNN-FCNN training is given, which is mathematically expressed as: ; in, are the weight coefficients of equation loss, optical signal-to-noise ratio estimation error, and received optical power estimation error, respectively. represents the equation loss, represents the optical signal-to-noise ratio estimation error, Indicates the received optical power estimation error.

[0068] In order to better verify the present invention, a simulation experiment is performed to verify the optical performance monitoring method proposed in the present invention.

[0069] First, it is necessary to build a fiber optic communication system and collect the raw data required for training the CNN-FCNN network.

[0070] Generally speaking, an optical fiber communication system includes a transmitter, an optical fiber transmission link, and a receiver. Their functions are as follows: (1) The transmitter mainly includes a bit sequence generation module, a digital-to-analog conversion module, and an optoelectronic modulator. (2) The optical fiber transmission link includes optical fiber, an optical power amplifier, and an adjustable optical attenuator; (3) The receiver mainly includes an optical spectrum analyzer, an optical power meter, a receiver, and a digital signal processing module. The process of signal propagation through an optical fiber communication system includes: (1) At the transmitter, a series of binary bit sequences randomly generated by the bit sequence generation module are first modulated into digital symbols of a specific modulation format. Subsequently, the digital-to-analog conversion module converts these digital symbol sequences into analog signals and modulates them onto an optical carrier using an optoelectronic modulator to achieve optoelectronic conversion; (2) In the optical fiber transmission link, after the optical signal has been transmitted over a certain span of optical fiber, it is necessary to use an optical power amplifier to pre-amplify the optical signal, and use an adjustable optical attenuator to adjust the signal optical power. (3) At the receiving end, an optical spectrum analyzer is used to collect the optical signal-to-noise ratio (OSNR) of the signal light, and an optical power meter is used to collect the received optical power of the signal light. The received optical signal is then restored to an electrical signal through the receiver and digital signal processing is performed through the digital signal processing module; the digital signal processing module includes steps such as IQ imbalance compensation and orthogonal normalization, dispersion compensation, clock recovery, channel dynamic equalization, and polarization demultiplexing.

[0071] A coherent fiber optic communication system is a specific implementation of a fiber optic communication system. Its structure includes core switching nodes, fiber optic transmission links, and server-side optical network units (ONUs). Their functions are as follows: (1) The core switching node is responsible for processing data traffic and transmitting data to the ONU device through a passive optical network; (2) The PON transmission link distributes signals to multiple server racks through fiber optic transmission links and passive optical splitters; (3) The server-side ONU device is responsible for receiving downlink optical signals, completing optical-to-electrical conversion, and performing signal demodulation. It mainly includes a receiver, a digital signal processing module, an optical power measurement module, and an OSNR monitoring module.

[0072] The process of signal propagation through a coherent optical fiber communication system includes: (1) At the core switching node, a series of binary bit sequences randomly generated by a bit sequence generation module are first modulated into digital symbols of a specific modulation format. Subsequently, the digital-to-analog conversion module converts these digital symbol sequences into analog signals and modulates them onto an optical carrier using an optoelectronic modulator to achieve optoelectronic conversion; (2) On the optical fiber transmission link, the optical signal is distributed to multiple server ends through a passive optical network. When the optical signal reaches the server end, the optical performance may be attenuated (characterized by the optical signal-to-noise ratio). Therefore, it is necessary to monitor the optical performance of the received optical signal in the server-end ONU device to evaluate the transmission quality of the link; (3) At the server end, the optical power measurement module and the optical performance monitoring module are used to collect the received optical power and optical signal-to-noise ratio of the signal light, respectively. The received optical signal is then restored to an electrical signal through the receiver and digital signal processing is performed through the digital signal processing module (digital signal processing includes steps such as IQ imbalance compensation and orthogonal normalization, dispersion compensation, clock recovery, channel dynamic equalization, and polarization demultiplexing). The received signal after dynamic channel equalization and polarization demultiplexing and its corresponding system parameters such as baud rate, transmit power, and transmission distance are collected. During the training process, system parameters that also need to be collected include received optical power and optical signal-to-noise ratio to facilitate the subsequent construction of the data set.

[0073] Figure 2 This is a flowchart for implementing optical performance monitoring in a coherent optical fiber communication system according to one embodiment of the present invention. First, the coherent optical fiber communication system transmits and receives signals. At the receiving end, data is collected to obtain a received signal after dynamic channel equalization and polarization demultiplexing, along with corresponding system parameters such as baud rate, transmit power, transmission distance, received optical power, and optical signal-to-noise ratio (OSNR). Using the received optical power and OSNR as labels, other system parameters related to the PON network communication system and the received signal are obtained as input to a CNN-FCNN network. The received signal is processed using a Fourier transform of varying order to obtain a two-dimensional projection image. The labels, 2D projection image, and system parameters are then input into a multimodal fusion convolutional neural network (CNN-FCNN) that integrates physical information. The CNN-FCNN network is then trained using a loss function based on the nonlinear Schrödinger equation.

[0074] In the specific implementation process, in step S120, it is necessary to perform a fractional Fourier transform on the received signal and collect a two-dimensional projection image of its three-dimensional spectrum on the "amplitude-order" plane. The process specifically includes: The schematic diagram of fractional domain Fourier transform two-dimensional projection image generation is as follows Figure 3 Shown: The signal collected by the receiving end Perform fractional Fourier transform, which The order fractional Fourier transform is defined as: (1) in, is the kernel function of the fractional Fourier transform, defined as: (2) in, , through Perform different orders The fractional Fourier transform of is used to construct a three-dimensional spectrum: (3) For fractional Fourier transforms of different orders, take the maximum amplitude corresponding to the order, (4) That is, it is the two-dimensional projection image of the three-dimensional spectrum obtained by fractional Fourier transform on the "amplitude-order" plane.

[0075] In step S130, system parameters including baud rate, transmission power, and transmission distance, as well as the collected two-dimensional projection image, are used as input, and the optical signal-to-noise ratio measured by the spectrum analyzer and the received optical power measured by the optical power meter are used as labels to construct a dataset. The CNN-FCNN network is trained based on the constructed dataset to train the CNN-FCNN network's ability to predict estimated optical signal-to-noise ratio values and received optical power estimates.

[0076] In one embodiment of the present invention, the CNN-FCNN network architecture includes a convolutional neural network branch (CNN branch) and a fully connected network branch (FCNN branch). The CNN branch receives a captured 2D projection image, while the FCNN branch receives system parameters, including baud rate, transmit power, and transmission distance. Different branches train and extract features based on their respective inputs, and then fuse the training to achieve multimodal learning.

[0077] Figure 3 This is a CNN-FCNN neural network model structure for implementing PON optical performance monitoring in one embodiment of the present invention. Its specific structure includes: (1) For the CNN branch, a convolutional layer containing 64 convolutional kernels of size 3×3 is first used to perform preliminary feature extraction on the image input, that is, the two-dimensional image data. Then, a maximum pooling layer with a step size of 2×2 is used to downsample the feature map obtained by the preliminary feature extraction, so as to reduce the computational overhead as much as possible without losing key features. Subsequently, another set of convolution-maximum pooling units is used to further extract high-dimensional features. The extracted high-dimensional features are flattened by the hidden layer and input into the fusion layer for fusion processing with the high-dimensional features extracted by the FCNN branch.

[0078] (2) For the FCNN branch, a single fully connected layer containing 32 neurons is used to achieve efficient extraction of data input features. Where Rs, LP, and D represent the baud rate, transmission power, and transmission distance in the system parameters, respectively.

[0079] (3) The fusion layer (Dropout) adopts a fully connected structure to splice the different modal features extracted by the CNN branch and the FCNN branch, and fully mines and associates cross-modal information through linear transformation, and realizes multimodal learning through fusion training, and then obtains the OSNR estimation value and The model also introduces a dynamic learning rate and early stopping mechanism. During training, if the test set loss does not decrease for five consecutive training cycles, the learning rate is reduced to 50% of the current value. When the validation set loss falls below 0.2, the early stopping mechanism is triggered. If the test set loss does not decrease for eight consecutive training cycles, training is terminated early. The combination of a dynamic learning rate and early stopping mechanism effectively prevents model overfitting, reduces redundant training iterations, lowers computational overhead, and ensures optimal model performance.

[0080] (4) Input the constructed data set into the network for training. The trained model can output the estimated value of optical signal-to-noise ratio and the estimated value of received optical power. Calculate the noise power estimate .

[0081] like Figure 3 As shown, the input data collected in the coherent optical fiber communication system includes two-dimensional projection images and system parameters. The input data is input into a multimodal fusion convolutional neural network-fully connected neural network that integrates physical information. A loss function is constructed based on the nonlinear Schrödinger equation, and the loss feedback is used to train the multimodal fusion convolutional neural network-fully connected neural network.

[0082] In the specific implementation process, the derivation process of the equation loss in the loss function based on the nonlinear Schrödinger equation is as follows: First, the estimated value of the optical signal-to-noise ratio and the estimated value of the received optical power output by the CNN-FCNN model can be used to obtain the estimated value of the noise power: (5) in, are the estimated values of optical signal-to-noise ratio, received optical power, and noise optical power, respectively.

[0083] The nonlinear Schrödinger equation (NLSE) provides a theoretical basis for studying the loss, dispersion, and nonlinear effects of optical signals during optical fiber transmission. It is crucial for understanding signal evolution in optical fibers and has become an indispensable analytical tool in optical fiber communication systems. Its mathematical expression is: (6) Since the collected received signal has been dispersion compensated, the nonlinear Schrödinger equation can be rewritten as: (7) in, represent the loss coefficient, dispersion coefficient and nonlinear coefficient respectively, Indicates the distance that the optical signal propagates along the axial direction in the optical fiber. is the complex envelope of the signal light field, The expression for group delay is: (8) in, is the ratio of signal light power to noise power, and are the standard normalized light field amplitudes The real and imaginary parts of the standard normalized light field amplitude Defined as: (9) in, and are the linear and nonlinear parts of phase noise, respectively. When spontaneous emission noise (Amplifier Spontaneous Emission Noise, ASE) propagates independently in an optical fiber, the nonlinear effect does not change the statistical characteristics of the ASE noise. However, when the signal and ASE noise are transmitted together in an optical fiber link, the nonlinear distortion generated by the interaction between the signal and ASE noise causes the statistical characteristics of the signal to deviate from the Gaussian distribution. At the same time, when the optical fiber is modeled as a zero-memory nonlinear (ZMNL) system, it can be assumed that the ASE noise is uniformly distributed in the optical fiber link, and the signal and noise are statistically independent of each other. Therefore, the phase noise caused by the signal and ASE noise can be separated by a stochastic differential equation: (10) Among them, the first represents the deterministic non-phase noise caused by the Kerr effect, and the second term represents the random phase noise introduced by ASE noise, including the linear and nonlinear phase noise introduced by the Kerr effect expressed in formula (11) and the linear and nonlinear phase noise introduced by ASE noise expressed in formula (12): (11) (12) Among them, γ represents the nonlinear coefficient, Indicates the reference value of received optical power, see formula (2) for details , Indicates the distance that the optical signal propagates along the axial direction in the optical fiber. represents the linear phase noise introduced by ASE noise, represents the nonlinear phase noise introduced by ASE noise.

[0084] Formula (12) is the Itô Integral, which cannot be directly solved by closed-form calculation. The traditional solution is to use Monte Carlo simulation to generate a large number of random samples to numerically calculate the statistical properties of the Itô integral. This method has high computational overhead and slow convergence in high-dimensional scenarios, and may be affected by insufficient sample size or pseudo-randomness. Without loss of generality, a deterministic approximation of the Itô integral is performed, and Formula (12) can be rewritten as: ; (13) Then the complex envelope expression of the signal light field in formula (8) can be rewritten as: ; (14) Substituting the above formula into formula (7), the expression of the equation loss is: (15) in, and represent the real and imaginary parts of a given complex number, respectively. The expression is: (16) Neural network loss includes equation loss , optical signal-to-noise ratio estimation error and received optical power estimation error The three parts are defined as: (17) are the weight coefficients of the equation loss, optical signal-to-noise ratio estimation error, and received optical power estimation error, respectively. The optical signal-to-noise ratio estimation error and received optical power estimation error are defined as: (18) in, and The reference value of the optical signal-to-noise ratio measured by the spectrum analyzer and the reference value of the received optical power measured by the optical power meter are and They represent the estimated value of optical signal-to-noise ratio and the estimated value of received optical power respectively. By calculating the model loss, iterating repeatedly, and continuously optimizing the network parameters, the ideal estimated value of optical signal-to-noise ratio and the estimated value of received optical power can be obtained.

[0085] For the method of acquiring two-dimensional projection images proposed in step S120, please refer to Figures 4(a)-(c) for details, wherein Figure 4(a) is a schematic diagram of a received signal after PON network transmission reaches the receiving end and undergoes digital signal processing in one embodiment of the present invention, Figure 4(b) is a schematic diagram of a three-dimensional spectrum constructed after the received signal undergoes fractional Fourier transform in one embodiment of the present invention, and Figure 4(c) is a schematic diagram of a two-dimensional projection image of the three-dimensional spectrum acquired in one embodiment of the present invention on the "amplitude-order" plane.

[0086] First, Figure 4(a) illustrates the signal forms of a PON communication system under two modulation formats (16QAM and QPSK). Figure 4(b) then shows the fractional-order Fourier transform of the signal collected at the receiver, revealing the three-dimensional structure in the fractional-order p, fractional-domain u, and amplitude dimensions. Furthermore, Figure 4(c) shows the projection result onto the "fractional-order p-amplitude" plane.

[0087] Figure 5 - Figure 8 shows the experimental comparison results of the optical performance monitoring method proposed by the present invention and other methods. The specific description is as follows: Figure 5 This figure shows the loss function curve for PON optical performance monitoring in a coherent optical fiber communication system in one embodiment of the present invention, including three curves: training set loss, test set loss, and equation loss. It can be seen that after the 68th round of training, the validation set loss failed to decrease for five consecutive training cycles. The dynamic learning rate mechanism was triggered, and the learning rate was reduced to 50% of the original value, effectively achieving a further reduction in the validation set loss. When the validation set loss fell below the 0.2 threshold, the early stopping mechanism was activated. The model terminated training early if the loss did not improve for eight consecutive training cycles. After the early stopping mechanism was activated, the optimal model was saved. Furthermore, the equation loss curve shows that in the initial stages of training, the model focused more on reducing the equation loss. However, as training progressed, the optical signal-to-noise ratio estimation error continued to decrease, causing the equation loss to increase to a certain extent before stabilizing. Overall, however, the model's equation loss is very low. Considering the inevitable errors caused by factors such as equation modeling and data acquisition, it can be considered that the modeling of the complex envelope of the signal light field of the nonlinear Schrödinger equation is reasonable and relatively accurate.

[0088] It should be noted that 400km, 600km, 800km, 1000km and 1200km in Figures 6 to 8 refer to coherent optical fiber communication systems with different system parameters, and are distinguished here by transmission distance.

[0089] Figures 6(a)-(d) are respectively the optical signal-to-noise ratio estimation error (OSNR estimation error) and the received optical power estimation error (OSNR estimation error) of the PON optical performance monitoring system under the coherent optical fiber communication system with the baud rates of 25 Gbaud, 30 Gbaud, 40 Gbaud and 50 Gbaud and the modulation mode of DP-QSPK in one embodiment of the present invention. Estimation error), the unit of estimation error is Db, and the value is the absolute value. As shown in Figure 6, it can be seen that the model achieves ideal estimation performance at 25, 30, and 40 Gbaud, and the average OSNR estimation errors are 0.11 dB, 0.13 dB, and 0.14 dB, respectively. Furthermore, 50 Gbaud data was collected to test the OSNR monitoring performance of the PON optical performance monitoring method based on physical information neural network disclosed in the present invention for out-of-sample data. The experimental data show that the average OSNR estimation error is 0.32 dB. The PON optical performance monitoring method based on physical information neural network disclosed in the present invention embeds the nonlinear Schrödinger equation into the loss function and guides the neural network training with physical constraints, thereby enhancing the interpretability of the neural network and improving the accuracy of optical performance monitoring of out-of-sample data.

[0090] Figures 7(a)-(d) show comparison curves of optical signal-to-noise ratio (OSNR) monitoring performance of PON optical performance monitoring in coherent optical fiber communication systems with baud rates of 25 Gbaud, 30 Gbaud, 40 Gbaud, and 50 Gbaud, respectively, compared to conventional algorithms. The horizontal axis represents the OSNR reference value, and the vertical axis represents the OSNR estimation error. The optical performance monitoring method disclosed in the present invention is compared with conventional statistical moment-based (SMB) methods and baud rate-tolerant OSNR monitoring methods (BRTM). Different PINN models are used to distinguish coherent optical fiber communication systems with different system parameters (400 km, 600 km, 800 km, 1000 km, and 1200 km). It can be seen that the optical performance monitoring method disclosed in the present invention, by considering the impact of phase noise caused by nonlinear effects on OSNR monitoring and introducing physical constraints during the training process, demonstrates higher OSNR estimation accuracy than conventional methods in scenarios with strong nonlinearity and different baud rates.

[0091] Figures 8(a)-(d) show the optical signal-to-noise ratio estimation error (OSNR estimation error) and received optical power estimation error (OSNR estimation error) of PON optical performance monitoring in a coherent optical fiber communication system with a baud rate of 25 Gbaud, 30 Gbaud, 40 Gbaud, and 50 Gbaud and a modulation mode of DP-16QAM in one embodiment of the present invention. Estimation error). It is used to test the OSNR estimation error and the optical performance monitoring method disclosed in the present invention under different baud rates and system parameters of the DP-16QAM system. Estimation error. It can be seen that the model also shows ideal estimation performance under the 16QAM system, and the average OSNR estimation errors of 25, 30, 40, and 50Gbaud are 0.21 dB, 0.18 dB, 0.19 dB, and 0.22 dB, respectively. Experimental results show that the PON optical performance monitoring method based on physical information neural network disclosed in the present invention embeds the nonlinear Schrödinger equation into the loss function, analyzes the impact of nonlinear phase noise on OSNR monitoring, and solves the problem of large OSNR monitoring errors for high-order modulation systems such as QAM caused by traditional methods that only consider the signal amplitude envelope and ignore the phase factor. It shows ideal OSNR monitoring performance in PON networks based on DP-QPSK and high-order DP-16QAM.

[0092] The PON optical performance monitoring method and system proposed in the present invention construct a loss function based on the nonlinear Schrödinger equation, thereby introducing physical information into the neural network model, which is conducive to overcoming the poor interpretability of neural networks in the prior art. By collecting the two-dimensional projection image of the three-dimensional spectrum on the "amplitude-order" plane as the input of the CNN-FCNN network model, it is conducive to reducing the complexity of the model without losing key information. In addition, by introducing multimodal data containing system parameters and two-dimensional projection images to extract multimodal feature maps, it is conducive to solving the problem of low accuracy of neural networks on out-of-sample data.

[0093] By embedding the nonlinear Schrödinger equation into the loss function, the present invention analyzes the impact of nonlinear phase noise on OSNR monitoring. This solves the problem of large OSNR monitoring errors for high-order modulation systems such as QAM caused by traditional methods that only consider the signal amplitude envelope and ignore the phase factor. At the same time, physical constraints are used to guide neural network training, thereby enhancing the interpretability of the model and improving the OSNR monitoring accuracy of out-of-sample data. In addition, by performing a fractional Fourier transform on the input signal and then using the two-dimensional projection image of the obtained three-dimensional spectrum as the model input, the model can achieve multi-scale analysis of the signal through joint time-frequency analysis, further improving the OSNR monitoring accuracy of the PON network.

[0094] Experiments have confirmed that the PON optical performance monitoring method disclosed in the present invention demonstrates high OSNR estimation accuracy in dual-polarization quadrature phase shift keying (DP-QPSK) and dual-polarization 16 quadrature amplitude modulation (DP-16QAM) systems with various baud rates and system parameters.

[0095] Corresponding to the above method, the present invention also provides a PON optical performance monitoring system, which includes a computer device, wherein the computer device includes a processor and a memory, and is located at the receiving end of the PON network. The memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps of the method described above.

[0096] Corresponding to the above method, the present invention further provides a computer-readable storage medium having a computer program / instructions stored thereon. When the computer program / instructions are executed by a processor, the steps of the method described in any of the above embodiments are implemented. The computer-readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the art.

[0097] Corresponding to the above method, the present invention also provides a computer program product, including a computer program / instruction, which implements the steps of the method described in any one of the above embodiments when executed by a processor.

[0098] It should be understood by those skilled in the art that the various exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether to implement the system in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention. When implemented in hardware, it may be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via a data signal carried in a carrier wave.

[0099] It should be understood that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted. In the above embodiments, several specific steps are described and illustrated as examples. However, the method of the present invention is not limited to the specific steps described and illustrated. Those skilled in the art may make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present invention.

[0100] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or replace features of other embodiments.

[0101] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations to the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A PON optical performance monitoring method, characterized in that: include: Obtaining a received signal that has been transmitted through the PON network and reached the receiving end and has undergone digital signal processing, and obtaining system parameters related to the PON network communication system; Performing a fractional Fourier transform on the received signal to construct a three-dimensional spectrum, and collecting a two-dimensional projection image of the three-dimensional spectrum on an "amplitude-order" plane; The two-dimensional projection image and the system parameters are used as inputs of a pre-trained multimodal fusion convolutional neural network-fully connected neural network CNN-FCNN, the convolutional neural network branch of the CNN-FCNN performs feature extraction on the input two-dimensional projection image, and the fully connected network branch of the CNN-FCNN performs feature extraction on the input system parameters; The fusion layer of CNN-FCNN fuses the features of different modalities extracted by the convolutional neural network branch and the fully connected network branch, and outputs the estimated optical signal-to-noise ratio and received optical power at the output layer; During the training process of CNN-FCNN, the optical signal-to-noise ratio reference value measured by the optical spectrum analyzer at the receiving end and the received optical power reference value measured by the optical power meter are used as labels to construct a training set. In addition, the loss in PON network signal transmission is calculated as the equation loss using the nonlinear Schrödinger equation. The optical signal-to-noise ratio estimation error is calculated based on the optical signal-to-noise ratio estimation value and the optical signal-to-noise ratio reference value. The received optical power estimation error is calculated based on the received optical power estimation value and the received optical power reference value. The weighted sum of the equation loss, optical signal-to-noise ratio estimation error, and received optical power estimation error is used as the loss function for CNN-FCNN training.

2. The method according to claim 1, characterized in that The system parameters include baud rate, transmission power and transmission distance.

3. The method according to claim 1, characterized in that The convolutional neural network branch includes a preset number of convolution-maximum pooling units, each group of convolution-maximum pooling units includes a preset number of convolution layers and maximum pooling layers, the preset number of convolution layers are used to extract features from the input two-dimensional projection image, and the maximum pooling layer is used for downsampling; The fully connected network branch includes a single fully connected layer containing a preset number of neurons.

4. The method according to claim 1, wherein Performing a fractional Fourier transform on the received signal to construct a three-dimensional spectrum, and collecting a two-dimensional projection image of the three-dimensional spectrum on an "amplitude-order" plane, including: Performing fractional Fourier transform of different orders on the received signal to construct a three-dimensional spectrogram; For the three-dimensional spectra constructed by fractional Fourier transform of different orders, the maximum amplitude corresponding to the order is taken to obtain a two-dimensional projection image of the three-dimensional spectra on the "amplitude-order" plane.

5. The method according to claim 1, wherein The method further comprises: performing a subtraction operation on the received optical power estimation value and the optical signal-to-noise ratio estimation value to obtain a noise optical power estimation value.

6. The method according to claim 5, characterized in that The loss expression of the equation is: ； in, and represent the real and imaginary parts of a given complex number, respectively. The expression is: ； in, represents the loss coefficient, represents the nonlinear coefficient, Indicates the distance that the optical signal propagates along the axial direction in the optical fiber. Indicates the reference value of received optical power. Indicates the reference value of noise optical power, Represents an imaginary unit.

7. The method according to claim 1, characterized in that The loss function used for CNN-FCNN training is mathematically expressed as: ； in, are the weight coefficients of equation loss, optical signal-to-noise ratio estimation error, and received optical power estimation error, respectively. represents the equation loss, represents the optical signal-to-noise ratio estimation error, Indicates the received optical power estimation error.

8. A PON optical performance monitoring system comprising a processor, a memory, and a computer program / instruction stored in the memory, characterized in that: Located at the receiving end of the PON network, the processor is used to execute the computer program / instructions. When the computer program / instructions are executed, the system implements the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images