Fiber channel waveform level modeling method and system based on nonlinear feature decoupling

Abstract

The invention provides an optical fiber channel waveform level modeling method and system based on nonlinear feature decoupling, and relates to the technical field of optical fiber channel modeling, and the method comprises the steps: S1, constructing a neural network group comprising a self-phase modulation model and a cross-phase modulation model; s2, through numerical simulation, collecting input and output waveforms of a single channel as a self-phase modulation data set; s3, training a self-phase modulation model through the self-phase modulation data set; s4, collecting input and output waveforms of multiple channels through numerical simulation, and constructing a cross phase modulation data set in combination with the self-phase modulation model; and S5, training the cross phase modulation model by using the data set of the cross phase modulation. According to the method, the multi-channel nonlinear effect can be decoupled, a design architecture of a plurality of sub-neural networks is adopted, the model training difficulty is reduced, and the prediction performance in a multi-channel scene is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of optical fiber channel modeling, and in particular to an optical fiber channel waveform level modeling method and system based on nonlinear characteristic decoupling. Background Art

[0002] With the rapid development of technologies such as artificial intelligence and large language models, optical fiber communication systems have become the infrastructure of modern communication systems, carrying more than 95% of the world's data traffic services. Optical communication networks are developing towards ultra-wideband, ultra-high speed, and ultra-large capacity. Optical communication simulation systems play an important role in the entire life cycle of optical communication, such as network planning, network parameter optimization, algorithm design, etc. Traditional simulations use the step-by-step Fourier method to solve through a large number of iterations. However, its complexity and signal bandwidth show a quadratic growth trend and cannot be applied to modern optical communication systems.

[0003] In recent years, deep learning technology has been introduced into the field of fiber channel modeling due to its powerful nonlinear fitting and parallel computing capabilities. However, fiber channel modeling solutions based on deep learning are mainly focused on low-rate single-channel and wavelength division multiplexing scenarios with a small number of channels, and cannot support the simulation requirements of high-speed and broadband systems. Summary of the invention

[0004] In view of the defects in the prior art, the present invention provides a fiber channel waveform level modeling method and system based on nonlinear characteristic decoupling.

[0005] According to a fiber channel waveform level modeling method and system based on nonlinear characteristic decoupling provided by the present invention, the scheme is as follows:

[0006] In a first aspect, a fiber channel waveform level modeling method based on nonlinear feature decoupling is provided, the method comprising:

[0007] Step S1: constructing a neural network group including a self-phase modulation model and a cross-phase modulation model;

[0008] Step S2: collecting the input and output waveforms of a single channel as a self-phase modulation data set through numerical simulation;

[0009] Step S3: training a self-phase modulation model using the self-phase modulation data set;

[0010] Step S4: collecting input and output waveforms of multiple channels through numerical simulation, and constructing a cross-phase modulation data set in combination with the self-phase modulation model;

[0011] Step S5: using the cross phase modulation data set to train the cross phase modulation model.

[0012] Preferably, the step S1 comprises:

[0013] Step S1.1: for a wavelength division multiplexing system with M channels, a neural network group consisting of 1 self-phase modulation model and 2 (M-1) cross-phase modulation models is constructed to model the optical fiber link of the wavelength division multiplexing system;

[0014] Step S1.2: Each neural network selects a bidirectional long short-term memory network; the input of the self-phase modulation network is the waveform of the current channel; the input of the cross-phase modulation network includes the waveform of the current channel and the concatenation of the waveforms of other channels, and the input dimension is twice that of the self-phase modulation model.

[0015] Preferably, step S2 comprises:

[0016] Step S2.1: using numerical simulation based on the split-step Fourier algorithm, collecting input and output waveforms in a single-channel case;

[0017] Step S2.2: performing data preprocessing on the collected waveform;

[0018] For the input waveform, a sliding window method is used to construct an input window containing relevant information of the previous and next symbols, which facilitates the neural network to model the timing correlation; for the output waveform, its linear characteristics are decoupled to obtain a data set containing only nonlinear effects; the data pair consisting of the input and output waveforms is used to construct a self-phase modulation data set.

[0019] Preferably, the step S4 comprises:

[0020] Step S4.1: using numerical simulation based on the split-step Fourier algorithm, collecting input and output waveforms in the case of M-channel wavelength division multiplexing; separating the collected original waveforms by wavelength division multiplexing, and down-sampling the separated signal waveforms to obtain a two-dimensional array containing M single-channel waveforms;

[0021] Step S4.2: respectively sending the signal waveforms of the M single channels into the trained self-phase modulation model to obtain M signal waveforms containing only the self-phase modulation effect;

[0022] Step S4.3: subtracting the M original multi-channel output waveforms from the M waveforms containing only the self-phase modulation effect, to obtain M waveforms containing only the cross-phase modulation effect;

[0023] Step S4.4: Take M single-channel signal waveforms as input and M waveforms containing only cross-phase modulation effects as output, and construct a cross-phase modulation data set.

[0024] Preferably, the step S5 comprises: training a cross-phase modulation network using the constructed cross-phase modulation data set;

[0025] In each batch of training, there are M cycles, which are used to model M different channels respectively. In the process of modeling any channel, the signal of the current channel will be spliced ​​with the signals of the M-1 adjacent channels around it, forming M-1 different inputs, which represent the cross-phase modulation caused by the M-1 channels. The M-1 inputs are sent to the corresponding M-1 cross-phase modulation models to obtain M-1 outputs. Finally, the M-1 outputs are added to obtain the final modeling value of the cross-phase modulation of the channel.

[0026] All M channels are cyclically modeled according to the above process, and the corresponding cross-phase modulation models are trained respectively.

[0027] In a second aspect, a fiber channel waveform level modeling system based on nonlinear feature decoupling is provided, the system comprising:

[0028] Module M1: Constructing a neural network group including a self-phase modulation model and a cross-phase modulation model;

[0029] Module M2: Through numerical simulation, the input and output waveforms of a single channel are collected as a self-phase modulation data set;

[0030] Module M3: training a self-phase modulation model using the self-phase modulation data set;

[0031] Module M4: collect input and output waveforms of multiple channels through numerical simulation, and construct a cross-phase modulation data set in combination with the self-phase modulation model;

[0032] Module M5: Using the cross-phase modulation data set, training the cross-phase modulation model.

[0033] Preferably, the module M1 comprises:

[0034] Module M1.1: For a wavelength division multiplexing system with M channels, a neural network group consisting of 1 self-phase modulation model and 2 (M-1) cross-phase modulation models is constructed to model the optical fiber link of the wavelength division multiplexing system;

[0035] Module M1.2: Each neural network selects a bidirectional long short-term memory network; the input of the self-phase modulation network is the waveform of the current channel; the input of the cross-phase modulation network includes the waveform of the current channel and the concatenation of the waveforms of other channels, and the input dimension is twice that of the self-phase modulation model.

[0036] Preferably, the module M2 comprises:

[0037] Module M2.1: Use numerical simulation based on the split-step Fourier algorithm to collect input and output waveforms in a single-channel case;

[0038] Module M2.2: perform data preprocessing on the collected waveform;

[0039] For the input waveform, a sliding window method is used to construct an input window containing relevant information of the previous and next symbols, which facilitates the neural network to model the timing correlation; for the output waveform, its linear characteristics are decoupled to obtain a data set containing only nonlinear effects; the data pair consisting of the input and output waveforms is used to construct a self-phase modulation data set.

[0040] Preferably, the module M4 comprises:

[0041] Module M4.1: Using numerical simulation based on the split-step Fourier algorithm, collect the input and output waveforms of M-channel wavelength division multiplexing; separate the collected original waveforms through wavelength division multiplexing, and downsample the separated signal waveforms to obtain a two-dimensional array containing M single-channel waveforms;

[0042] Module M4.2: respectively send the signal waveforms of M single channels into the trained self-phase modulation model to obtain M signal waveforms containing only the self-phase modulation effect;

[0043] Module M4.3: Subtract the M original multi-channel output waveforms from the M waveforms containing only the self-phase modulation effect to obtain M waveforms containing only the cross-phase modulation effect;

[0044] Module M4.4: Take M single-channel signal waveforms as input and M waveforms containing only cross-phase modulation effects as output to construct a cross-phase modulation data set.

[0045] Preferably, the module M5 comprises: training a cross-phase modulation network using the constructed cross-phase modulation data set;

[0046] In each batch of training, there are M cycles, which are used to model M different channels respectively. In the process of modeling any channel, the signal of the current channel will be spliced ​​with the signals of the M-1 adjacent channels around it, forming M-1 different inputs, which represent the cross-phase modulation caused by the M-1 channels. The M-1 inputs are sent to the corresponding M-1 cross-phase modulation models to obtain M-1 outputs. Finally, the M-1 outputs are added to obtain the final modeling value of the cross-phase modulation of the channel.

[0047] All M channels are cyclically modeled according to the above process, and the corresponding cross-phase modulation models are trained respectively.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. The present invention adopts the idea of ​​decoupling nonlinear characteristics and uses different neural networks to model different nonlinear effects respectively. It is more suitable for multi-channel wavelength division multiplexing systems and has higher modeling accuracy.

[0050] 2. The present invention directly reuses the self-phase modulation model trained in a single-channel case to a multi-channel system, thereby improving the flexibility of the solution;

[0051] 3. The multi-channel signal modeled by the present invention can ensure that the accuracy of signals from different channels is relatively consistent;

[0052] 4. The present invention can be more conveniently expanded to a wavelength division multiplexing system with a larger channel.

[0053] Other beneficial effects of the present invention will be explained in the specific implementation manner through the introduction of specific technical features and technical solutions. Through the introduction of these technical features and technical solutions, those skilled in the art should be able to understand the beneficial technical effects brought about by the technical features and technical solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0055] Figure 1 The figure is a flow chart of the method of the present invention.

[0056] Figure 2 It is a structural diagram of the neural network model of the present invention.

[0057] Figure 3 Schematic diagram of the cross-phase modulation model training and reasoning of the present invention.

[0058] Figure 4 Schematic diagram of the dataset construction and model training process of the present invention.

[0059] Figure 5 This is a display diagram comparing the center channel waveform error results with the linear feature decoupling method when the present invention is applied to multi-channel, high-speed optical transmission channel modeling.

[0060] Figure 6 This is a display diagram comparing the waveform error results of different channels with the linear feature decoupling method when the present invention is applied to multi-channel, high-speed optical transmission channel modeling. DETAILED DESCRIPTION

[0061] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0062] The embodiment of the present invention provides a fiber channel waveform-level modeling method based on nonlinear feature decoupling, which decouples the self-phase modulation from the channel and the cross-phase modulation from the channels, and uses multiple neural networks to model different nonlinear effects respectively, effectively improving the modeling accuracy of the neural network in multi-channel and high-speed scenarios, and has the ability to expand to wavelength division multiplexing scenarios with more channels. Figure 1 As shown, the method specifically includes:

[0063] Step S1: construct a neural network group including a self-phase modulation model and a cross-phase modulation model.

[0064] Reference Figure 2 and Figure 3 As shown, specifically, step S1 includes:

[0065] Step S1.1: for a wavelength division multiplexing system with M channels, a neural network group consisting of 1 self-phase modulation model and 2 (M-1) cross-phase modulation models is constructed to model the optical fiber link of the wavelength division multiplexing system;

[0066] Step S1.2: Each neural network selects a bidirectional long short-term memory network; the input of the self-phase modulation network is the waveform of the current channel; the input of the cross-phase modulation network includes the waveform of the current channel and the concatenation of the waveforms of other channels, and the input dimension is twice that of the self-phase modulation model.

[0067] Reference Figure 4 As shown, step S2: through numerical simulation, collecting the input and output waveforms of a single channel as a self-phase modulation data set.

[0068] Specifically, step S2 includes:

[0069] Step S2.1: using numerical simulation based on the split-step Fourier algorithm, collecting input and output waveforms in a single-channel case;

[0070] Step S2.2: performing data preprocessing on the collected waveform;

[0071] For the input waveform, a sliding window method is used to construct an input window containing relevant information of the previous and next symbols, which facilitates the neural network to model the timing correlation; for the output waveform, its linear characteristics are decoupled to obtain a data set containing only nonlinear effects; the data pair consisting of the input and output waveforms is used to construct a self-phase modulation data set.

[0072] Step S3: training a self-phase modulation network using the self-phase modulation data set to obtain a trained self-phase modulation model.

[0073] Step S4: by numerical simulation, collecting input and output waveforms of multiple channels, combining the self-phase modulation model, and constructing a cross-phase modulation data set.

[0074] The step S4 comprises:

[0075] Step S4.1: using numerical simulation based on the split-step Fourier algorithm, collecting input and output waveforms in the case of M-channel wavelength division multiplexing; separating the collected original waveforms by wavelength division multiplexing, and down-sampling the separated signal waveforms to obtain a two-dimensional array containing M single-channel waveforms;

[0076] Step S4.2: respectively sending the signal waveforms of the M single channels into the trained self-phase modulation model to obtain M signal waveforms containing only the self-phase modulation effect;

[0077] Step S4.3: subtracting the M original multi-channel output waveforms from the M waveforms containing only the self-phase modulation effect, to obtain M waveforms containing only the cross-phase modulation effect;

[0078] Step S4.4: Take M single-channel signal waveforms as input and M waveforms containing only cross-phase modulation effects as output, and construct a cross-phase modulation data set.

[0079] Step S5: using the cross phase modulation data set to train the cross phase modulation model.

[0080] The step S5 includes: training the cross-phase modulation network using the constructed cross-phase modulation data set;

[0081] In each batch of training, there are M cycles, which are used to model M different channels respectively. In the process of modeling any channel, the signal of the current channel will be spliced ​​with the signals of the M-1 adjacent channels around it, forming M-1 different inputs, which represent the cross-phase modulation caused by the M-1 channels. The M-1 inputs are sent to the corresponding M-1 cross-phase modulation models to obtain M-1 outputs. Finally, the M-1 outputs are added to obtain the final modeling value of the cross-phase modulation of the channel.

[0082] All M channels are cyclically modeled according to the above process, and the corresponding cross-phase modulation models are trained respectively.

[0083] The present invention also provides a fiber channel waveform level modeling system based on nonlinear feature decoupling, which can be implemented by executing the process steps of the fiber channel waveform level modeling method based on nonlinear feature decoupling, that is, those skilled in the art can understand the fiber channel waveform level modeling method based on nonlinear feature decoupling as a preferred implementation of the fiber channel waveform level modeling system based on nonlinear feature decoupling. The system specifically includes:

[0084] Module M1: Construct a neural network group including a self-phase modulation model and a cross-phase modulation model.

[0085] Specifically, the module M1 includes:

[0086] Module M1.1: For a wavelength division multiplexing system with M channels, a neural network group consisting of 1 self-phase modulation model and 2 (M-1) cross-phase modulation models is constructed to model the optical fiber link of the wavelength division multiplexing system;

[0087] Module M1.2: Each neural network selects a bidirectional long short-term memory network; the input of the self-phase modulation network is the waveform of the current channel; the input of the cross-phase modulation network includes the waveform of the current channel and the concatenation of the waveforms of other channels, and the input dimension is twice that of the self-phase modulation model.

[0088] Module M2: Through numerical simulation, the input and output waveforms of a single channel are collected as a self-phase modulation data set.

[0089] Specifically, the module M2 includes:

[0090] Module M2.1: Use numerical simulation based on the split-step Fourier algorithm to collect input and output waveforms in a single-channel case;

[0091] Module M2.2: perform data preprocessing on the collected waveform;

[0092] For the input waveform, a sliding window method is used to construct an input window containing relevant information of the previous and next symbols, which facilitates the neural network to model the timing correlation; for the output waveform, its linear characteristics are decoupled to obtain a data set containing only nonlinear effects; the data pair consisting of the input and output waveforms is used to construct a self-phase modulation data set.

[0093] Module M3: training a self-phase modulation network using the self-phase modulation data set to obtain a trained self-phase modulation model.

[0094] Module M4: Through numerical simulation, collect the input and output waveforms of multiple channels, combine the self-phase modulation model, and build a cross-phase modulation data set.

[0095] The M4 module includes:

[0096] Module M4.1: Using numerical simulation based on the split-step Fourier algorithm, collect the input and output waveforms of M-channel wavelength division multiplexing; separate the collected original waveforms through wavelength division multiplexing, and downsample the separated signal waveforms to obtain a two-dimensional array containing M single-channel waveforms;

[0097] Module M4.2: respectively send the signal waveforms of M single channels into the trained self-phase modulation model to obtain M signal waveforms containing only the self-phase modulation effect;

[0098] Module M4.3: Subtract the M original multi-channel output waveforms from the M waveforms containing only the self-phase modulation effect to obtain M waveforms containing only the cross-phase modulation effect;

[0099] Module M4.4: Take M single-channel signal waveforms as input and M waveforms containing only cross-phase modulation effects as output to construct a cross-phase modulation data set.

[0100] Module M5: Using the cross-phase modulation data set, training the cross-phase modulation model.

[0101] The module M5 includes: training a cross-phase modulation network using a constructed cross-phase modulation data set;

[0102] In each batch of training, there are M cycles, which are used to model M different channels respectively. In the process of modeling any channel, the signal of the current channel will be spliced ​​with the signals of the M-1 adjacent channels around it, forming M-1 different inputs, which represent the cross-phase modulation caused by the M-1 channels. The M-1 inputs are sent to the corresponding M-1 cross-phase modulation models to obtain M-1 outputs. Finally, the M-1 outputs are added to obtain the final modeling value of the cross-phase modulation of the channel.

[0103] All M channels are cyclically modeled according to the above process, and the corresponding cross-phase modulation models are trained respectively.

[0104] Next, the present invention will be described in more detail.

[0105] In this embodiment, the simulation system parameters used are: single-carrier dual-polarization coherent optical transmission, symbol rate 140GBaud, number of channels 5, modulation format 16QAM, single-span transmission distance 80km, and maximum number of transmission spans 15 (1200km).

[0106] The present invention provides a fiber channel waveform level modeling method based on nonlinear characteristic decoupling, comprising:

[0107] Step S1: constructing a neural network group including a self-phase modulation model and a cross-phase modulation model;

[0108] Step S2: collecting the input and output of a single channel as a self-phase modulation data set through numerical simulation;

[0109] Step S3: training a self-phase modulation model using the sub-self-phase modulation data set;

[0110] Step S4: by numerical simulation, collecting the input and output of multiple channels, combining the self-phase modulation model, and constructing a cross-phase modulation data set;

[0111] Step S5: using the cross phase modulation data set to train the cross phase modulation model.

[0112] Specifically, step S1 includes:

[0113] Step 1.1: For a wavelength division multiplexing system with 5 channels, a neural network group consisting of 1 self-phase modulation model and 8 cross-phase modulation models is constructed to model the optical fiber link of the wavelength division multiplexing system.

[0114] Step 1.2: Each neural network selects a bidirectional long short-term memory network. The input of the self-phase modulation network is the waveform of the current channel; the input of the cross-phase modulation network includes the concatenation of the waveform of the current channel and the waveforms of other channels, and the input dimension is twice that of the self-phase modulation model.

[0115] Specifically, step S2 includes:

[0116] Step 2.1: Use numerical simulation based on the split-step Fourier algorithm to collect input and output waveforms in the single-channel case.

[0117] Step 2.2: Perform data preprocessing on the collected original waveform. For the input waveform, a sliding window method is used to construct an input window containing relevant information of the 40 symbols before and after, so that the neural network can model the timing correlation. For the output waveform, its linear features are decoupled to obtain a data set containing only nonlinear effects. The data pair consisting of the input and output waveforms is used to construct a data set of self-phase modulation.

[0118] Specifically, step S3 includes:

[0119] The self-phase modulation network is trained using the self-phase modulation dataset.

[0120] Specifically, step S4 includes:

[0121] Step S4.1: Using numerical simulation based on the split-step Fourier algorithm, collect the input and output waveforms under the 5-channel wavelength division multiplexing condition. Separate the collected original waveforms through wavelength division multiplexing, and downsample the separated signal waveforms to obtain a two-dimensional array containing 5 single channel waveforms.

[0122] Step S4.2: Send the five single-channel signal waveforms to the self-phase modulation network trained in S3 respectively to obtain five signal waveforms containing only the self-phase modulation effect.

[0123] Step S4.3: Subtract the five original multi-channel output waveforms from the five waveforms containing only the self-phase modulation effect to obtain five waveforms containing only the cross-phase modulation effect.

[0124] Step S4.4: Take 5 single-channel signal waveforms as input and 5 waveforms containing only cross-phase modulation effects as output to construct a cross-phase modulation data set.

[0125] Specifically, step S5 includes:

[0126] The cross-phase modulation network is trained using the constructed cross-phase modulation dataset. Specifically, each batch of training includes 5 cycles, which are used to model 5 different channels. In the process of modeling any channel, the signal of the current channel will be spliced ​​with the signals of the 4 adjacent channels around it to form 5 different inputs, representing the cross-phase modulation caused by the 5 channels. The 5 inputs are sent to the corresponding 4 cross-phase modulation models to obtain 5 outputs. Finally, the 5 outputs are added to obtain the final modeling value of the cross-phase modulation of the channel. All 5 channels are cyclically modeled according to the above process, and the corresponding cross-phase modulation models are trained respectively.

[0127] Figure 5 The waveform error level of the fiber channel waveform modeling method based on nonlinear feature decoupling is demonstrated in the modeling of multi-channel and high-speed fiber channels. The waveform error is quantified by the normalized mean square error, and the calculation formula is:

[0128]

[0129] Where y is the SSFM waveform, is the waveform of the AI ​​model, N data is the total amount of data. From the perspective of normalized mean square error, the fiber channel waveform-level modeling method based on nonlinear feature decoupling is very accurate in both 3-channel and 5-channel scenarios.

[0130] Figure 6 The error levels of different channels in multi-channel, high-speed fiber channel modeling are demonstrated by using a fiber channel waveform-level modeling method based on nonlinear feature decoupling.

[0131] The embodiment of the present invention provides a fiber channel waveform-level modeling method and system based on nonlinear feature decoupling, which decouples multi-channel nonlinear effects and adopts a design architecture of multiple sub-neural networks, thereby reducing the difficulty of model training and improving the prediction performance in multi-channel scenarios.

[0132] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0133] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A fiber channel waveform level modeling method based on nonlinear feature decoupling, characterized in that: include: Step S1: constructing a neural network group including a self-phase modulation model and a cross-phase modulation model; Step S2: collecting the input and output waveforms of a single channel as a self-phase modulation data set through numerical simulation; Step S3: training a self-phase modulation model using the self-phase modulation data set; Step S4: collecting input and output waveforms of multiple channels through numerical simulation, and constructing a cross-phase modulation data set in combination with the self-phase modulation model; Step S5: using the cross phase modulation data set to train the cross phase modulation model.

2. The fiber channel waveform level modeling method based on nonlinear feature decoupling according to claim 1 is characterized in that: The step S1 comprises: Step S1.1: for a wavelength division multiplexing system with M channels, a neural network group consisting of 1 self-phase modulation model and 2 (M-1) cross-phase modulation models is constructed to model the optical fiber link of the wavelength division multiplexing system; Step S1.2: Each neural network selects a bidirectional long short-term memory network; the input of the self-phase modulation network is the waveform of the current channel; the input of the cross-phase modulation network includes the waveform of the current channel and the concatenation of the waveforms of other channels, and the input dimension is twice that of the self-phase modulation model.

3. The fiber channel waveform level modeling method based on nonlinear feature decoupling according to claim 1, characterized in that: The step S2 comprises: Step S2.1: using numerical simulation based on the split-step Fourier algorithm, collecting input and output waveforms in a single-channel case; Step S2.2: performing data preprocessing on the collected waveform; For the input waveform, a sliding window method is used to construct an input window containing relevant information of the previous and next symbols, which facilitates the neural network to model the timing correlation; for the output waveform, its linear characteristics are decoupled to obtain a data set containing only nonlinear effects; the data pair consisting of the input and output waveforms is used to construct a self-phase modulation data set.

4. The fiber channel waveform level modeling method based on nonlinear feature decoupling according to claim 1, characterized in that: The step S4 comprises: Step S4.1: using numerical simulation based on the split-step Fourier algorithm, collecting input and output waveforms in the case of M-channel wavelength division multiplexing; separating the collected original waveforms by wavelength division multiplexing, and down-sampling the separated signal waveforms to obtain a two-dimensional array containing M single-channel waveforms; Step S4.2: respectively sending the signal waveforms of the M single channels into the trained self-phase modulation model to obtain M signal waveforms containing only the self-phase modulation effect; Step S4.3: subtracting the M original multi-channel output waveforms from the M waveforms containing only the self-phase modulation effect, to obtain M waveforms containing only the cross-phase modulation effect; Step S4.4: Take M single-channel signal waveforms as input and M waveforms containing only cross-phase modulation effects as output, and construct a cross-phase modulation data set.

5. The fiber channel waveform level modeling method based on nonlinear feature decoupling according to claim 1, characterized in that: The step S5 comprises: training a cross-phase modulation network using the constructed cross-phase modulation data set; In each batch of training, there are M cycles, which are used to model M different channels respectively. In the process of modeling any channel, the signal of the current channel will be spliced ​​with the signals of the M-1 adjacent channels around it, forming M-1 different inputs, which represent the cross-phase modulation caused by the M-1 channels. The M-1 inputs are sent to the corresponding M-1 cross-phase modulation models to obtain M-1 outputs. Finally, the M-1 outputs are added to obtain the final modeling value of the cross-phase modulation of the channel. All M channels are cyclically modeled according to the above process, and the corresponding cross-phase modulation models are trained respectively.

6. A fiber channel waveform level modeling system based on nonlinear feature decoupling, characterized in that: include: Module M1: Constructing a neural network group including a self-phase modulation model and a cross-phase modulation model; Module M2: Through numerical simulation, the input and output waveforms of a single channel are collected as a self-phase modulation data set; Module M3: training a self-phase modulation model using the self-phase modulation data set; Module M4: collect input and output waveforms of multiple channels through numerical simulation, and construct a cross-phase modulation data set in combination with the self-phase modulation model; Module M5: Using the cross-phase modulation data set, training the cross-phase modulation model.

7. The fiber channel waveform level modeling system based on nonlinear characteristic decoupling according to claim 6, characterized in that: The module M1 comprises: Module M1.1: For a wavelength division multiplexing system with M channels, a neural network group consisting of 1 self-phase modulation model and 2 (M-1) cross-phase modulation models is constructed to model the optical fiber link of the wavelength division multiplexing system; Module M1.2: Each neural network selects a bidirectional long short-term memory network; the input of the self-phase modulation network is the waveform of the current channel; the input of the cross-phase modulation network includes the waveform of the current channel and the concatenation of the waveforms of other channels, and the input dimension is twice that of the self-phase modulation model.

8. The fiber channel waveform level modeling system based on nonlinear characteristic decoupling according to claim 6, characterized in that: The module M2 comprises: Module M2.1: Use numerical simulation based on the split-step Fourier algorithm to collect input and output waveforms in a single-channel case; Module M2.2: perform data preprocessing on the collected waveform; For the input waveform, a sliding window method is used to construct an input window containing relevant information of the previous and next symbols, which facilitates the neural network to model the timing correlation; for the output waveform, its linear characteristics are decoupled to obtain a data set containing only nonlinear effects; the data pair consisting of the input and output waveforms is used to construct a self-phase modulation data set.

9. The fiber channel waveform level modeling system based on nonlinear characteristic decoupling according to claim 6, characterized in that: The module M4 comprises: Module M4.1: Using numerical simulation based on the split-step Fourier algorithm, collect the input and output waveforms of M-channel wavelength division multiplexing; separate the collected original waveforms through wavelength division multiplexing, and downsample the separated signal waveforms to obtain a two-dimensional array containing M single-channel waveforms; Module M4.2: respectively send the signal waveforms of M single channels into the trained self-phase modulation model to obtain M signal waveforms containing only the self-phase modulation effect; Module M4.3: Subtract the M original multi-channel output waveforms from the M waveforms containing only the self-phase modulation effect to obtain M waveforms containing only the cross-phase modulation effect; Module M4.4: Take M single-channel signal waveforms as input and M waveforms containing only cross-phase modulation effects as output to construct a cross-phase modulation data set.

10. The fiber channel waveform level modeling system based on nonlinear characteristic decoupling according to claim 6, characterized in that: The module M5 includes: training a cross-phase modulation network using a constructed cross-phase modulation data set; In each batch of training, there are M cycles, which are used to model M different channels respectively. In the process of modeling any channel, the signal of the current channel will be spliced ​​with the signals of the M-1 adjacent channels around it, forming M-1 different inputs, which represent the cross-phase modulation caused by the M-1 channels. The M-1 inputs are sent to the corresponding M-1 cross-phase modulation models to obtain M-1 outputs. Finally, the M-1 outputs are added to obtain the final modeling value of the cross-phase modulation of the channel. All M channels are cyclically modeled according to the above process, and the corresponding cross-phase modulation models are trained respectively.

Images