Fast and accurate modeling method and system for distributed optical fiber channels based on feature decoupling

Abstract

The present invention provides a method and system for fast and accurate modeling of distributed optical fiber channels based on feature decoupling, including: step S1: collecting input and output data under different optical fiber channel conditions as training data sets; step S2: compensating the output data of channels including nonlinear features by linear compensation algorithm; step S3: performing data preprocessing and arrangement to obtain a nonlinear data set suitable for neural network training; step S4: modeling the nonlinear channel features based on the neural network structure and deep learning algorithm, and obtaining a nonlinear channel model through training; step S5: in the transmission process, modeling the nonlinear model, the linear model, and the amplifier model in turn, and then performing multiple iterations to finally realize the long-distance transmission of the optical fiber channel model. The present invention does not require too many iterative steps to achieve low-complexity, high-precision, and strong generalization optical fiber channel modeling.

Description

Technical Field

[0001] The present invention relates to the technical field of optical fiber channel modeling and neural network application, and in particular to a method and system for fast and accurate modeling of distributed optical fiber channels based on feature decoupling. Background Art

[0002] Fiber channel modeling is of great significance for communication system design, performance prediction and simulation. The traditional channel modeling method is based on the split-step Fourier method (SSFM), which is implemented by approximately solving the nonlinear Schrödinger equation. It has been proven in engineering that it can accurately express the linear and nonlinear damage suffered by the signal during transmission. However, the excessive number of repeated iterative steps in SSFM brings very high computational complexity.

[0003] Patent document CN102857389A (application number: CN201210279777.7) discloses a frequency domain distortion prediction channel modeling method for a 1553B bus, comprising the following steps: firstly decomposing a typical Manchester code as an input signal into a group of sinusoidal signals, and taking each sinusoidal signal as input, inputting it into the 1553B bus system, solving the transfer function between any nodes, and obtaining each output sinusoidal signal, superimposing the group of signals to predict the degree of signal distortion, thereby excluding nodes with a large degree of signal distortion, that is, unavailable nodes.

[0004] In order to improve the computing speed of the fiber channel model, a fiber channel modeling scheme based on deep learning has been proposed. In 2020, the team from Beijing University of Posts and Telecommunications published a paper titled "Data-driven Optical Fiber Channel Modeling: A Deep Learning Approach" in the Journal of Lightwave Technology. The paper used a bidirectional long short-term memory model (BiLSTM) to model the fiber channel of single-channel transmission. It proved that BiLSTM has a good fitting ability for the characteristics of the fiber channel, but their work only focused on modeling within a short distance of 80 km and ignored the modeling of channel ASE noise. In 2021, the team from Shanghai Jiao Tong University published a paper titled "Fast and Accurate Optical Fiber Channel Modeling Using GenerativeAdversarial Network" in the Journal of Lightwave Technology. The paper proposed the use of conditional generative adversarial networks (CGAN) for long-distance single-channel fiber channel modeling. These fast fiber channel modeling methods are a holistic effect modeling scheme, that is, a neural network is used to model all effects of the fiber channel. For a neural network, it is impossible to accurately model complex fiber channel effects, such as cross-phase modulation and four-wave mixing in wavelength division multiplexing fiber channels. Summary of the invention

[0005] In view of the defects in the prior art, the object of the present invention is to provide a method and system for fast and accurate modeling of distributed optical fiber channels based on feature decoupling.

[0006] The method for fast and accurate modeling of a distributed optical fiber channel based on feature decoupling provided by the present invention comprises:

[0007] Step S1: collecting input and output data under different fiber channel conditions as a training data set;

[0008] Step S2: compensating the output data of the channel including nonlinear characteristics by a linear compensation algorithm;

[0009] Step S3: preprocessing and arranging the data to obtain a nonlinear data set suitable for neural network training;

[0010] Step S4: Modeling the nonlinear channel characteristics based on the neural network structure and the deep learning algorithm, and obtaining the nonlinear channel model through training;

[0011] Step S5: During the transmission process, the nonlinear model, the linear model, and the amplifier model are modeled in sequence, and then multiple iterations are performed to ultimately achieve long-distance transmission of the optical fiber channel model.

[0012] Preferably, step S1 includes: collecting channel input and output data from the optical fiber channel model, the channel data is set to data at multiple sampling rates, which has all the characteristics of analog signals, and the channel data characteristics are changed by adjusting signal parameters and channel parameters, so as to collect data sets under different conditions.

[0013] Preferably, the linear compensation algorithm of the signal includes chromatic dispersion compensation and polarization film dispersion compensation, and the linear compensation algorithm is a reversible process, and the linear compensation algorithm is derived from the nonlinear Schrödinger equation.

[0014] Preferably, step S3 comprises:

[0015] Step S3.1: normalizing the input data and the output data, and normalizing the power of the channel input and channel output data under different conditions to within a preset range;

[0016] Step S3.2: Connect the real part and the imaginary part of the dual polarization signal to form a one-dimensional vector as the input data and output data of the neural network, and the length of the arranged data is the length of the data generated by the neural network each time;

[0017] Step S3.3: Use the sliding window to select input data of required length in sequence, and the size of the sliding window is related to the time memory length of the channel.

[0018] Preferably, the parameters of the neural network are trained using a stochastic gradient descent method to minimize the loss function; the loss function is the mean square error between the channel output nonlinear data and the neural network output data;

[0019] The amplifier model includes amplification characteristics and random noise characteristics caused by the device.

[0020] The distributed optical fiber channel fast and accurate modeling system based on feature decoupling provided by the present invention comprises:

[0021] Module M1: collects input and output data under different fiber channel conditions as training data sets;

[0022] Module M2: compensating the output data of the channel including nonlinear characteristics by a linear compensation algorithm;

[0023] Module M3: Perform data preprocessing and arrangement to obtain a nonlinear data set suitable for neural network training;

[0024] Module M4: Model the nonlinear channel characteristics based on the neural network structure and deep learning algorithm, and obtain the nonlinear channel model through training;

[0025] Module M5: During the transmission process, the nonlinear model, linear model, and amplifier model are modeled in turn, and then multiple iterations are performed to finally realize the long-distance transmission of the optical fiber channel model.

[0026] Preferably, the module M1 includes: collecting channel input and output data from the optical fiber channel model, the channel data is set to data at multiple sampling rates, which has all the characteristics of analog signals, and the channel data characteristics are changed by adjusting signal parameters and channel parameters, so as to collect data sets under different conditions.

[0027] Preferably, the linear compensation algorithm of the signal includes chromatic dispersion compensation and polarization film dispersion compensation, and the linear compensation algorithm is a reversible process, and the linear compensation algorithm is derived from the nonlinear Schrödinger equation.

[0028] Preferably, the module M3 includes:

[0029] Module M3.1: Normalize the input data and output data, and normalize the power of channel input and channel output data under different conditions within a preset range;

[0030] Module M3.2: Connect the real and imaginary parts of the dual polarization signal to form a one-dimensional vector as the input data and output data of the neural network. The length of the data is the length of the data generated by the neural network each time.

[0031] Module M3.3: Use a sliding window to sequentially select input data of the required length. The size of the sliding window is related to the time memory length of the channel.

[0032] Preferably, the parameters of the neural network are trained using a stochastic gradient descent method to minimize the loss function; the loss function is the mean square error between the channel output nonlinear data and the neural network output data;

[0033] The amplifier model includes amplification characteristics and random noise characteristics caused by the device.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] (1) The present invention uses a one-step modeling approach to greatly reduce the number of model iterations, thereby reducing model complexity;

[0036] (2) The feature decoupling scheme of the present invention can enhance nonlinear features, which is conducive to accurate modeling of nonlinear features;

[0037] (3) The feature decoupling scheme of the present invention can reduce the time memory length in the data, thereby reducing the length of the sliding window, and ultimately enabling the model to use a neural network structure with lower complexity to achieve accurate nonlinear feature modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] 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:

[0039] Figure 1 A flowchart of a method for fast and accurate modeling of a distributed optical fiber channel based on feature decoupling according to an embodiment of the present invention;

[0040] Figure 2 It is a schematic diagram of a data acquisition scheme in step S1 of a method for fast and accurate modeling of a distributed optical fiber channel based on feature decoupling according to an embodiment of the present invention;

[0041] Figure 3 A schematic diagram of data preprocessing and arrangement in step S3 of a method for fast and accurate modeling of a distributed optical fiber channel based on feature decoupling according to an embodiment of the present invention;

[0042] Figure 4 This is a schematic diagram of the neural network model structure of step S4 in a method for fast and accurate modeling of a distributed optical fiber channel based on feature decoupling according to an embodiment of the present invention (taking a bidirectional long short-term memory neural network as an example);

[0043] Figure 5 This is a schematic diagram of a result of rapid and accurate modeling of a distributed optical fiber channel based on feature decoupling according to an embodiment of the present invention. Taking a multi-channel multiplexing scenario as an example, the number of channels is 9, 21, and 41, respectively, and the distance is 80 km;

[0044] Figure 6 A fast and accurate modeling result of a distributed optical fiber channel based on feature decoupling according to an embodiment of the present invention, and a constellation diagram at different generalization distances;

[0045] Figure 7 This is a result of fast and accurate modeling of a distributed optical fiber channel based on feature decoupling according to an embodiment of the present invention, and a comparison chart of the running time of two models under the same conditions. DETAILED DESCRIPTION

[0046] 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.

[0047] Example:

[0048] like Figure 1 As shown, the present invention proposes a method for fast and accurate modeling of a distributed optical fiber channel based on feature decoupling, taking a wavelength division multiplexing optical fiber channel model as an example, comprising the following steps:

[0049] S1: Figure 2 As shown, input and output data under different fiber channel conditions are collected as training data sets. The channel data input data is the data after wavelength division multiplexing, and the channel output data is the data after passing through a section of optical fiber. Both the input and output data have all the characteristics of analog signals.

[0050] S2: Compensate the output data of the channel through a linear compensation algorithm so that the output data mainly contains nonlinear characteristics.

[0051] The linear compensation algorithm of the signal mainly includes chromatic dispersion compensation and polarization film dispersion compensation, and the linear compensation algorithm is required to be a reversible process. The linear compensation algorithm is derived from the nonlinear Schrödinger equation. For dual-polarization signals, the dispersion compensation expression can be simplified to:

[0052]

[0053]

[0054] Wherein, A represents the light field on two arbitrary orthogonal polarization modes, β2 is the group velocity dispersion parameter, Δβ1 is the differential group delay caused by PMD, L represents the distance, a minus sign before L indicates a linear compensation process, and no minus sign indicates a linear modeling process. Modeling and compensation are inverse processes of each other.

[0055] S3: Figure 3 As shown, data preprocessing and arrangement are performed to obtain a nonlinear data set suitable for neural network training. This step specifically includes the following sub-steps:

[0056] S3.1: Normalize the input and output data. Normalize the power of the channel input and channel output data under different conditions to an appropriate range. For a data set composed of multiple input optical powers, the normalization method is: N p Indicates the number of different power values, P i is the power of the i-th group of data.

[0057] S3.2: Connect the real and imaginary parts of the dual polarization signal to form a one-dimensional vector as the input and output data of the neural network. If the length of the data generated by the neural network each time is n, such as Figure 3As shown, the real and imaginary parts of n dual-polarization sample points are connected and arranged as a one-dimensional vector of length 4n, which is used as the input data vector at a moment. The arrangement of the output data completely corresponds to the input vector, and 4n-length data is generated each time.

[0058] S3.3: Use the sliding window to select the input data of the required length in sequence. The size of the sliding window is related to the time memory length of the channel. Figure 3 As shown, data from several adjacent moments are selected each time. For the modeling of 9 channels, the window length is 145 when feature decoupling is not performed. After feature decoupling, the window length can be reduced to 13.

[0059] S4: Model the nonlinear channel characteristics based on the neural network structure and deep learning algorithm, and obtain the nonlinear channel model through training. Figure 4 As shown, a model with excellent performance is constructed by taking a bidirectional long short-term memory neural network (BiLSTM) as an example. In the input layer, data at different times are input into different time units of BiLSTM, which is the same as the sliding window length in S33. The data of each time unit is the one-dimensional vector constructed in step S32.

[0060] S5: During the transmission process, the nonlinear model, the linear model, and the amplifier model are modeled in sequence, and then multiple iterations are performed to finally realize the long-distance transmission of the optical fiber channel model.

[0061] This embodiment also provides a schematic diagram of the implementation results of the channel modeling method on a multi-channel multiplexed optical fiber channel. The results of the feature decoupling distributed modeling corresponding to different channel numbers are shown in Figure 2. Figure 5 As shown, the step-by-step Fourier transform is a traditional modeling method with high accuracy but high complexity, and the distributed feature decoupling represents a fast and accurate modeling method proposed in the present invention. Figure 6 is the constellation diagram under different generalization distances, Figure 7 It is the computation time required for the distributed Fourier and distributed decoupling schemes under the same conditions.

[0062] In the above technical solution, the spectrum results show that the distributed modeling method based on feature decoupling is highly consistent with the traditional method, and the constellation diagrams at different distances are completely consistent, which realizes the generalization of distance and requires much less time than the traditional modeling method.

[0063] The distributed optical fiber channel fast and accurate modeling system based on feature decoupling provided by the present invention comprises: module M1: collecting input and output data under different optical fiber channel conditions as a training data set; module M2: compensating the output data of the channel including nonlinear features through a linear compensation algorithm; module M3: performing data preprocessing and arrangement to obtain a nonlinear data set suitable for neural network training; module M4: modeling the nonlinear channel features based on the neural network structure and the deep learning algorithm, and obtaining a nonlinear channel model through training; module M5: in the transmission process, modeling the nonlinear model, the linear model, and the amplifier model in turn, and then performing multiple iterations to finally realize the long-distance transmission of the optical fiber channel model.

[0064] The module M1 includes: collecting channel input and output data from the optical fiber channel model, the channel data is set as data at multiple sampling rates, which has all the characteristics of analog signals, and the channel data characteristics are changed by adjusting the signal parameters and channel parameters, so as to collect data sets under different conditions. The linear compensation algorithm of the signal includes chromatic dispersion compensation and polarization film dispersion compensation, and the linear compensation algorithm is a reversible process, which is derived from the nonlinear Schrödinger equation. The module M3 includes: module M3.1: normalizing the input data and output data, and normalizing the power of the channel input and channel output data under different conditions within a preset range; module M3.2: connecting the real and imaginary parts of the dual polarization signal to form a one-dimensional vector as the input data and output data of the neural network, and the length of the arranged data is the length of the data generated by the neural network each time; module M3.3: using a sliding window to select the input data of the required length in turn, and the size of the sliding window is related to the time memory length of the channel. The parameters of the neural network are trained by using the stochastic gradient descent method to minimize the loss function; the loss function is the mean square error between the channel output nonlinear data and the neural network output data; the amplifier model includes the amplification characteristics and the random noise characteristics caused by the device.

[0065] The method for fast and accurate modeling of a distributed optical fiber channel based on feature decoupling described in this embodiment does not require too many iterative steps and realizes optical fiber channel modeling with very low complexity. The feature decoupling scheme is used to enhance nonlinear features and realize accurate modeling of nonlinear features. The feature decoupling scheme can weaken the time memory length in the data, thereby reducing the length of the sliding window, and ultimately enables the model to use a neural network structure with lower complexity to realize accurate nonlinear feature modeling. Through the distributed modeling scheme, flexible generalization of distance is realized.

[0066] Those skilled in the art know that, in addition to implementing the system, device and its various modules provided by the present invention in a purely computer-readable program code, it is entirely possible to implement the same program 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, device and its various modules provided by the present invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing the method and structures within the hardware component.

[0067] 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 method for fast and accurate modeling of distributed optical fiber channels based on feature decoupling, characterized in that: include: Step S1: collecting input and output data under different fiber channel conditions as a training data set; Step S2: compensating the output data of the channel including nonlinear characteristics by a linear compensation algorithm; Step S3: preprocessing and arranging the data to obtain a nonlinear data set suitable for neural network training; Step S4: Modeling the nonlinear channel characteristics based on the neural network structure and the deep learning algorithm, and obtaining the nonlinear channel model through training; Step S5: During the transmission process, the nonlinear model, the linear model, and the amplifier model are modeled in sequence, and then multiple iterations are performed to ultimately achieve long-distance transmission of the optical fiber channel model.

2. The method for fast and accurate modeling of distributed optical fiber channels based on feature decoupling according to claim 1 is characterized in that: The step S1 includes: collecting channel input and output data from the optical fiber channel model, the channel data is set to data at multiple sampling rates, which has all the characteristics of analog signals, and the channel data characteristics are changed by adjusting signal parameters and channel parameters, so as to collect data sets under different conditions.

3. The method for fast and accurate modeling of distributed optical fiber channels based on feature decoupling according to claim 1 is characterized in that: The linear compensation algorithm of the signal includes chromatic dispersion compensation and polarization film dispersion compensation, and the linear compensation algorithm is a reversible process, and the linear compensation algorithm is derived from the nonlinear Schrodinger equation.

4. The method for fast and accurate modeling of distributed optical fiber channels based on feature decoupling according to claim 1, characterized in that: The step S3 comprises: Step S3.1: normalizing the input data and the output data, and normalizing the power of the channel input and channel output data under different conditions to within a preset range; Step S3.2: Connect the real part and the imaginary part of the dual polarization signal to form a one-dimensional vector as the input data and output data of the neural network, and the length of the arranged data is the length of the data generated by the neural network each time; Step S3.3: Use the sliding window to select input data of required length in sequence, and the size of the sliding window is related to the time memory length of the channel.

5. The method for fast and accurate modeling of distributed optical fiber channels based on feature decoupling according to claim 1, characterized in that: The parameters of the neural network are trained using the stochastic gradient descent method to minimize the loss function; the loss function is the mean square error between the channel output nonlinear data and the neural network output data; The amplifier model includes amplification characteristics and random noise characteristics caused by the device.

6. A distributed optical fiber channel fast and accurate modeling system based on feature decoupling, characterized in that: include: Module M1: collects input and output data under different fiber channel conditions as training data sets; Module M2: compensating the output data of the channel including nonlinear characteristics by a linear compensation algorithm; Module M3: Perform data preprocessing and arrangement to obtain a nonlinear data set suitable for neural network training; Module M4: Model the nonlinear channel characteristics based on the neural network structure and deep learning algorithm, and obtain the nonlinear channel model through training; Module M5: During the transmission process, the nonlinear model, linear model, and amplifier model are modeled in turn, and then multiple iterations are performed to finally realize the long-distance transmission of the optical fiber channel model.

7. The distributed optical fiber channel fast and accurate modeling system based on feature decoupling according to claim 6 is characterized in that: The module M1 includes: collecting channel input and output data from the optical fiber channel model, the channel data is set to data at multiple sampling rates, which has all the characteristics of analog signals, and the channel data characteristics are changed by adjusting signal parameters and channel parameters, so as to collect data sets under different conditions.

8. The distributed optical fiber channel fast and accurate modeling system based on feature decoupling according to claim 6 is characterized in that: The linear compensation algorithm of the signal includes chromatic dispersion compensation and polarization film dispersion compensation, and the linear compensation algorithm is a reversible process, and the linear compensation algorithm is derived from the nonlinear Schrodinger equation.

9. The distributed optical fiber channel fast and accurate modeling system based on feature decoupling according to claim 6, characterized in that: The module M3 comprises: Module M3.1: Normalize the input data and output data, and normalize the power of channel input and channel output data under different conditions within a preset range; Module M3.2: Connect the real and imaginary parts of the dual polarization signal to form a one-dimensional vector as the input data and output data of the neural network. The length of the data is the length of the data generated by the neural network each time. Module M3.3: Use a sliding window to sequentially select input data of the required length. The size of the sliding window is related to the time memory length of the channel.

10. The distributed optical fiber channel fast and accurate modeling system based on feature decoupling according to claim 6, characterized in that: The parameters of the neural network are trained using the stochastic gradient descent method to minimize the loss function; the loss function is the mean square error between the channel output nonlinear data and the neural network output data; The amplifier model includes amplification characteristics and random noise characteristics caused by the device.

Images