A reverse design method and system for backward multi-pump Raman fiber amplifier

Abstract

The present invention relates to a reverse design method and system for a backward multi-pumped Raman fiber amplifier, belonging to the field of Raman fiber amplifiers. An improved particle swarm optimization algorithm is first used to optimize the pump light parameter configuration. A neural network algorithm is then used to learn the nonlinear mapping relationship between output gain and pump light parameters to reversely design the Raman fiber amplifier. By determining the target output gain and generating matching pump light parameters, this method replaces the traditional method of numerically solving the Raman coupled wave differential equation. By combining the improved particle swarm optimization algorithm with the neural network, the accuracy of the neural network model is improved, resulting in a Raman fiber amplifier designed for C+L band signal light amplification. This improves computational efficiency and enhances the output gain and output gain flatness of the Raman amplifier.

Description

Technical Field

[0001] The present invention relates to the field of Raman fiber amplifiers, and in particular to a reverse design method and system for a backward multi-pump Raman fiber amplifier. Background Art

[0002] In the field of optical communication research, fiber amplifiers, as indispensable core components in optical fiber transmission systems, directly affect the transmission performance of the system. Raman fiber amplifiers (RFA) play an important role in future 6G transmission and communication systems due to their outstanding advantages such as large bandwidth, good gain flatness, low noise figure, and good compatibility. They have gradually become a hot topic in the field of optical amplifier research.

[0003] Theoretically, by selecting the appropriate pump light parameter configuration during RFA design, full-band Raman amplification can be achieved through stimulated Raman scattering, with the optical gain only dropping by 3dB as the amplified signal power approaches the pump power. The primary challenge in RFA design is selecting the pump power and wavelength to produce a specific gain profile. This process requires solving the Raman coupled wave differential equation that describes the nonlinear effects between the pump and signal light propagating along the fiber. Due to the complexity of this equation, only approximate solutions can be obtained. Current methods for numerically solving this equation include the Runge-Kutta method, the average power method, the shooting method, and the use of genetic algorithms to jointly describe the propagation equations for multi-channel light wave interactions. These methods are complex and cumbersome, and in some cases, face non-convergence.

[0004] In recent years, research on Raman amplifiers has largely focused on optimizing the configuration of pump power and wavelength. To address this issue, genetic algorithms, artificial bee colony algorithms, and differential evolution algorithms have been proposed. Although these algorithms can achieve the flat gain required for RFA, they require algorithm parameters to be adjusted and iterated based on the specific problem. Furthermore, during the design process, these algorithms are very time-consuming in solving the Raman coupled wave equation, and repeated cycles lead to unnecessary duplication of work and time loss.

[0005] To comprehensively improve the solution process of nonlinear differential equations, recent proposals have proposed using neural network algorithms in machine learning to learn the nonlinear mapping relationship between pump light and signal light, training neural network models to replace the solution process of the Raman coupled wave equation. In 2018, Chen Jing et al. from Fuzhou University combined an extreme learning machine (ELM) with a differential evolution (DE) algorithm. By leveraging the ELM's fast learning speed and high generalization, and the DE's powerful global search capabilities, they achieved a gain ripple of less than 0.5 dB. In 2020, Darko Zibar et al. from Denmark proposed a method for inverse design of Raman amplifiers using machine learning for refined optimization, accurately numerically predicting the pump design for arbitrary Raman gain curves. In 2021, Uiara C. de Moura et al. from Denmark experimentally characterized a machine learning framework for designing and modeling Raman amplifiers with arbitrary gain. Testing with different fiber types revealed an error of less than 0.5 dB. All of these studies demonstrate that inverse design of Raman amplifiers using neural networks is feasible and effective, but none of them address the need to improve the performance and amplification bandwidth of Raman fiber amplifiers while simultaneously increasing efficiency and reducing error. Summary of the Invention

[0006] The purpose of the present invention is to provide a reverse design method and system for a backward multi-pumped Raman fiber amplifier, so as to greatly improve the calculation efficiency and the output gain value and output gain flatness of the Raman amplifier.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A reverse design method for a backward multi-pump Raman fiber amplifier, the method comprising:

[0009] Determine the pump light parameters that affect the gain performance of backward multi-pump Raman fiber amplifiers;

[0010] According to the set pump light parameter value range, the improved particle swarm optimization algorithm is used to optimize the pump light parameters and obtain the optimal pump light parameters under different gain conditions;

[0011] According to the optimal pump light parameters under each gain condition, the gain curve under each gain condition is calculated using the nonlinear Raman coupling differential equation of the backward multi-pump Raman fiber amplifier;

[0012] The data set is constructed by taking the gain curves under different gain conditions as input and the optimal pump light parameters under different gain conditions as output.

[0013] Using the data set to train a BP neural network model to obtain a trained BP neural network model;

[0014] The target gain curve is input into the trained BP neural network model, and the optimal pump light parameters of the backward multi-pump Raman fiber amplifier are output.

[0015] Optionally, determining pump light parameters that affect the gain performance of the backward multi-pump Raman fiber amplifier specifically includes:

[0016] The simplified nonlinear Raman coupling differential equation of the backward multi-pump Raman fiber amplifier is established as

[0017] Where, ± corresponds to the forward or reverse injection of the pump light source, P i 、P j 、P k Represents the optical power of the i, j, and k channels respectively, v i 、v j 、v k Represent the optical frequencies of the i, j, and k channels respectively, g R (v i -v j ) represents the Raman gain coefficient between the two channels of light of the i-th and j-th channels, g R (v j -v k ) represents the Raman gain coefficient between the jth and kth channels, K eff represents the polarization factor, A eff Represents the effective core area of ​​the optical fiber, α j represents the attenuation coefficient of the light wave of the jth channel transmitted in the optical fiber, γ j represents the Rayleigh scattering coefficient, K and h represent the Boltzmann constant and Planck constant respectively, is the Bose-Einstein factor, T is the absolute temperature of the optical fiber;

[0018] According to the nonlinear Raman coupling differential equation, pump light parameters affecting the gain performance of the backward multi-pump Raman amplifier are determined; the pump light parameters include the power and wavelength of the pump light source.

[0019] Optionally, the weights in the improved particle swarm optimization algorithm are iteratively updated according to the following formula:

[0020] w=(w1+w2)·(T'-In) / T'+w2

[0021] Where w represents the weight factor at the current iteration number, w1 and w2 represent the initial value and final value of the weight factor respectively, In represents the current iteration number, In≤T', and T' represents the total number of iterations.

[0022] Optionally, using the data set to train a BP neural network model to obtain a trained BP neural network model specifically includes:

[0023] The BP neural network model is set to include input layer, hidden layer and output layer;

[0024] Under the premise that the number of hidden layers is preset to be 3-8 and the number of neurons in each layer is 1-100, a total of 600 BP neural network models are constructed;

[0025] The dataset was subjected to missing data filling, outlier processing and normalization, and the processed dataset was randomly divided into a training set, a test set and a validation set in a ratio of 7:1.5:1.5;

[0026] Based on the training set and validation set, TRAINSCG algorithm was used to train 600 BP neural network models;

[0027] Using the test set, the performance of 600 trained BP neural network models was compared with the regression value R value and mean square error MSE value as evaluation indicators. It was determined that the structure of the BP neural network model with the best performance was 6 hidden layers and 33 neurons in each layer.

[0028] The data set is used to train a BP neural network model with optimal performance to obtain a trained BP neural network model.

[0029] Optionally, the pump light parameter value range is: the wavelength range of each pump light is 1400-1500 nm, and the power range of each pump light source is 0-2 W.

[0030] A reverse design system for a backward multi-pumped Raman fiber amplifier, comprising:

[0031] An influencing factor determination module, used to determine pump light parameters that affect the gain performance of the backward multi-pump Raman fiber amplifier;

[0032] The optimization module is used to optimize the pump light parameters according to the set pump light parameter value range using an improved particle swarm optimization algorithm to obtain the optimal pump light parameters under different gain conditions;

[0033] A gain calculation module, configured to calculate a gain curve under each gain condition based on the optimal pump light parameters under each gain condition and using the nonlinear Raman coupling differential equation of the backward multi-pump Raman fiber amplifier;

[0034] A data set construction module is used to construct a data set by taking the gain curves under different gain conditions as input and the optimal pump light parameters under different gain conditions as labels;

[0035] A training module is used to train the BP neural network model using the data set to obtain a trained BP neural network model;

[0036] The application module is used to input the target gain curve into the trained BP neural network model and output the optimal pump light parameters of the backward multi-pump Raman fiber amplifier.

[0037] Optionally, the influencing factor determination module specifically includes:

[0038] The differential equation building submodule is used to build the simplified nonlinear Raman coupling differential equation of the backward multi-pump Raman fiber amplifier:

[0039] Where, ± corresponds to the forward or reverse injection of the pump light source, P i 、P j 、P k Represents the optical power of the i, j, and k channels respectively, v i 、v j 、v k Represent the optical frequencies of the i, j, and k channels respectively, g R (v i -v j ) represents the Raman gain coefficient between the two channels of light of the i-th and j-th channels, g R (v j -v k ) represents the Raman gain coefficient between the jth and kth channels, K eff represents the polarization factor, A eff Represents the effective core area of ​​the optical fiber, α j represents the attenuation coefficient of the light wave of the jth channel transmitted in the optical fiber, γ j represents the Rayleigh scattering coefficient, K and h represent the Boltzmann constant and Planck constant respectively, is the Bose-Einstein factor, T is the absolute temperature of the optical fiber;

[0040] The pump light parameter determination submodule is used to determine the pump light parameters that affect the gain performance of the backward multi-pump Raman amplifier according to the nonlinear Raman coupling differential equation; the pump light parameters include the power and wavelength of the pump light source.

[0041] Optionally, the weights in the improved particle swarm optimization algorithm are iteratively updated according to the following formula:

[0042] w=(w1+w2)·(T'-In) / T'+w2

[0043] Where w represents the weight factor at the current iteration number, w1 and w2 represent the initial value and final value of the weight factor respectively, In represents the current iteration number, In≤T', and T' represents the total number of iterations.

[0044] Optionally, the training module specifically includes:

[0045] The model setting submodule is used to set the BP neural network model including the input layer, hidden layer and output layer;

[0046] The model construction submodule is used to construct a total of 600 BP neural network models under the premise that the number of hidden layers is preset to be 3-8 layers and the number of neurons in each layer is 1-100;

[0047] A partitioning submodule is used to perform missing fill, outlier processing and normalization on the dataset, and randomly divide the processed dataset into a training set, a test set and a validation set in a ratio of 7:1.5:1.5;

[0048] The training submodule is used to train 600 BP neural network models using the TRAINSCG algorithm based on the training set and the validation set;

[0049] The test submodule is used to compare the performance of 600 trained BP neural network models using the test set, with the regression value R value and the mean square error MSE value as evaluation indicators, and to determine that the structure of the BP neural network model with the best performance is 6 hidden layers and each layer contains 33 neurons;

[0050] The model training submodule is used to train the BP neural network model with the best performance using the data set to obtain the trained BP neural network model.

[0051] Optionally, the pump light parameter value range is: the wavelength range of each pump light is 1400-1500 nm, and the power range of each pump light source is 0-2 W.

[0052] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0053] The present invention discloses a reverse design method and system for a backward multi-pumped Raman fiber amplifier. This method first optimizes the pump light parameter configuration using an improved particle swarm optimization algorithm. A neural network algorithm is then used to learn the nonlinear mapping relationship between output gain and pump light parameters to reversely design the Raman fiber amplifier. By determining the target output gain and generating matching pump light parameters, this method replaces the traditional method of numerically solving the Raman coupled wave differential equation. By combining the improved particle swarm optimization algorithm with the neural network, the accuracy of the neural network model is enhanced, resulting in a Raman fiber amplifier designed for C+L band signal light amplification. This improves computational efficiency and enhances the output gain and output gain flatness of the Raman amplifier. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 A flowchart of a reverse design method for a backward multi-pumped Raman fiber amplifier provided by an embodiment of the present invention;

[0056] Figure 2 A structural diagram of a backward multi-pump Raman amplifier provided in an embodiment of the present invention;

[0057] Figure 3 A flowchart of an improved particle swarm optimization algorithm provided by an embodiment of the present invention;

[0058] Figure 4 A schematic diagram of a neural network construction principle provided by an embodiment of the present invention;

[0059] Figure 5 A topological diagram of a neural network model provided by an embodiment of the present invention;

[0060] Figure 6 The RFA output gain diagram before and after optimization provided by the embodiment of the present invention;

[0061] Figure 7 The distribution diagram of R values ​​provided by the embodiment of the present invention under different hidden layers and different numbers of neurons; Figure 7 (a) is the distribution diagram of R value when the number of hidden layers is 3. Figure 7 (b) is the distribution diagram of R value when the number of hidden layers is 4. Figure 7 (c) in the figure is the distribution diagram of R value when the number of hidden layers is 5. Figure 7 (d) is the distribution diagram of R value when the number of hidden layers is 6. Figure 7 (e) in the figure is the distribution diagram of R value when the number of hidden layers is 7. Figure 7 (f) is the distribution diagram of R value when the number of hidden layers is 8;

[0062] Figure 8 The distribution diagram of the MSE values ​​provided by the embodiment of the present invention under different hidden layers and different numbers of neurons; Figure 8 (a) is the distribution diagram of MSE value when the number of hidden layers is 3. Figure 8 (b) is the distribution diagram of MSE value when the number of hidden layers is 4. Figure 8 (c) is the distribution diagram of MSE value when the number of hidden layers is 5. Figure 8(d) is the distribution diagram of MSE value when the number of hidden layers is 6. Figure 8 (e) is the distribution diagram of MSE value when the number of hidden layers is 7. Figure 8 (f) is the distribution diagram of MSE value when the number of hidden layers is 8;

[0063] Figure 9 A graph showing how the R value and MSE value of the optimal model provided by an embodiment of the present invention change with the number of hidden layers; Figure 9 (a) in the figure shows the change of R value with the number of hidden layers. Figure 9 (b) shows the change of MSE value with the number of hidden layers;

[0064] Figure 10 The R value and error result diagram under the optimal neural network model structure provided by the embodiment of the present invention; Figure 10 (a) is the R value and error training result diagram under the optimal neural network model structure. Figure 10 (b) is the R value and error verification result diagram under the optimal neural network model structure. Figure 10 (c) is the R value and error test result diagram under the optimal neural network model structure;

[0065] Figure 11 This is a diagram of target gain and predicted gain under different gain values ​​provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0067] The purpose of the present invention is to provide a reverse design method and system for a backward multi-pumped Raman fiber amplifier, so as to greatly improve the calculation efficiency and the output gain value and output gain flatness of the Raman amplifier.

[0068] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0069] The embodiment of the present invention provides a reverse design method for a backward multi-pump Raman fiber amplifier, such as Figure 1 As shown, the method includes the following steps:

[0070] Step S1, determining pump light parameters that affect the gain performance of a backward multi-pump Raman fiber amplifier.

[0071] The backward multi-pump RFA structure used in the present invention is shown in FIG. Figure 2 As shown, m beams of pump light (Pump light) and n beams of signal light (Signal light) are respectively coupled into the optical fiber through the optical multiplexer unit (OMU) for transmission, and the transmission direction of the pump light is opposite to that of the signal light. The backward pumping method can effectively reduce the time and distance of the interaction between the signal light and the pump light, and reduce the system noise. In order to achieve high-gain output of the RFA, tellurite fiber is used as the gain medium, and the amplified signal light is output after filtering out the excess pump light through a filter, thereby completing the amplification of the C+L band signal light. The present invention takes four pump lights as an example to amplify 125 signal lights, that is, m=4, n=125. Figure 2 In the input, transmission, and output, the wavelengths of the n signal beams are λ1, λ2, λ3, ..., λ n , the wavelengths of the m pump beams are λ p1 ,λ p2 ,λ p3 ,…,λ pm .

[0072] For RFA, its amplification process is to use the stimulated Raman scattering effect generated between different wavelength light beams in the optical fiber to achieve signal light amplification during transmission. By controlling the wavelength of the signal light and the pump light during transmission, the two are within a specific frequency difference range, allowing the signal light to fully absorb the energy of the pump light to achieve its own amplification. The mathematical model of the Raman fiber amplifier is a simplified nonlinear Raman coupling differential equation, as shown in formula (1). In this invention, the main considerations are the loss of the signal light itself during transmission, the stimulated Raman scattering effect between different wavelength light beams (signal light and signal light, signal light and pump light, pump light and pump light), and the influence of spontaneous radiation and Rayleigh scattering on the RFA.

[0073]

[0074] Where, ± corresponds to the forward or reverse injection of the pump light source, P i 、P j 、P k Represents the optical power of the i, j, and k channels respectively, v i 、v j 、v k Represent the optical frequencies of the i, j, and k channels respectively, g R (v i -v j) represents the Raman gain coefficient between the two channels of light of the i-th and j-th channels, g R (v j -v k ) represents the Raman gain coefficient between the jth and kth channels, K eff represents the polarization factor, A eff Represents the effective core area of ​​the optical fiber, α j represents the attenuation coefficient of the light wave of the jth channel transmitted in the optical fiber, γ j represents the Rayleigh scattering coefficient, K and h represent the Boltzmann constant and Planck constant respectively, is the Bose-Einstein factor, and T is the absolute temperature of the optical fiber.

[0075] The main parameters for evaluating the performance of Raman amplifiers are two aspects, namely the output gain and gain fluctuation of the RFA. The gain and gain flatness are defined as equations (2) and (3) respectively:

[0076]

[0077] Δ=max(G)-min(G) (3)

[0078] In formula (2), P j (0), P j Where (L) is the initial input optical power of the jth signal light and the optical power after transmission distance L. Equation (3) is the difference between the maximum and minimum gain values, expressed as gain flatness Δ. The smaller the difference, the more consistent the output gain values ​​of the amplified signal light, and the flatter the output gain curve. These two indicators are important indicators for evaluating RFA performance.

[0079] According to the nonlinear Raman coupling differential equation, the pump light parameters affecting the gain performance of the backward multi-pump Raman amplifier are determined; the pump light parameters include the power and wavelength of the pump light source.

[0080] Step S2 : optimizing the pump light parameters using an improved particle swarm optimization algorithm according to the set pump light parameter value range to obtain the optimal pump light parameters under different gain conditions.

[0081] Based on the traditional particle swarm algorithm, the weights are restricted, which improves the global search capability while avoiding the regionalization of search particles. The optimal solution is obtained by searching within a given pump wavelength and power range.

[0082] Specific improvements to the particle swarm optimization algorithm:

[0083] The optimization of RFA mainly focuses on improving the system's output gain G and reducing the gain flatness Δ, so that the gain output is a smooth curve. In this paper, an improved particle swarm algorithm is used to optimize G and Δ.

[0084] The improved particle swarm optimization algorithm (IPSO algorithm) is also derived from the simulation of bird foraging behavior. The difference from the traditional particle swarm optimization algorithm is that the weight factor is improved in the present invention, so that the particles can optimize with variable inertia within the search range. Figure 3 ,The specific process of the algorithm includes the following steps:

[0085] First, set the search dimension to D, the total number of particles to M, and the total number of iterations to T.

[0086] 1) In the D-dimensional search space, the position information of each particle is X i =(x i1 ,x i2 ,…,x iD ), the speed information is V i =(v i1 ,v i2 ,…,v iD ). Where i represents the i-th particle in the search space, x i1 ,x i2 ,…,x iD are the position information of each component, v i1 ,v i2 ,…,v iD The particle fitness function is fit(·) = 1 / Δ. For the design of RFA, the particle velocity and position information in the algorithm corresponds to the wavelength and power information of the pump light. The particle optimization process is a process of finding extreme values, which includes two extreme values. One is the individual extreme value of each particle, p id , the other is the global extreme value p after comparison of each particle in the entire search space gd .

[0087] 2) According to formula (4), the particle weight factor, velocity information and position information are updated at different iteration times.

[0088]

[0089] Where w1 and w2 are the initial and final values ​​of the weight factor w, respectively, and In≤T' is the current iteration number. The last two formulas represent the velocity v and position x of particle i at iteration d (k+1). c1 and c2 are learning factors, and r1 and r2 are random numbers in the range [0,1].

[0090] Continuously update the number of iterations and determine whether the maximum number of iterations is reached. If In = T', the iteration ends and the optimal position and velocity information are output. If In < T', the previous step is repeated.

[0091] Step S3: According to the optimal pump light parameters under each gain condition, use the nonlinear Raman coupling differential equation of the backward multi-pump Raman fiber amplifier to calculate the gain curve under each gain condition.

[0092] Step S4: Use the gain curves under different gain conditions as input quantities and the optimal pump light parameters under different gain conditions as output quantities to form a data set.

[0093] The IPSO algorithm is used to optimize the power and wavelength of the pump light. After 3000 iterations, a collection of 3000 groups of optimal pump light parameter configurations is formed to compose a data set, and the size of the data set is 9 * 3000.

[0094] Step S5: Use the data set to train the BP neural network model to obtain a trained BP neural network model.

[0095] Generally speaking, directly integrating the Raman coupled wave differential equation is a very complex process. Generally, numerical solutions are adopted, such as the Runge-Kutta method, the average power method, etc. For the backward-pumped RFA, the shooting method needs to be combined with the Runge-Kutta method or the average power method to obtain its analytical solution. In some cases, the results face the situation of non-convergence. In this paper, the neural network reverse design of RFA is adopted. By training multiple times to construct a neural network model, it replaces the traditional numerical analysis process, greatly saving the time for solving the nonlinear equation and improving the calculation efficiency.

[0096] Simplify Equation (1) into the Y = f(X) function, where X represents the input data of Equation (1), that is, the power and wavelength information of the pump light, and Y represents the output value of Equation (1), that is, the output gain of the RFA. The function f(·) is regarded as the complex nonlinear mapping relationship from the whole X to Y. The goal of the reverse design of RFA is to obtain the matching X value according to the given Y value, that is, it is necessary to determine the mapping relationship of the f -1 (·) function. To solve this problem, a neural network model NN(·) is constructed to learn this complex mapping. Theoretically, when NN(·) is successfully trained, the optimal pump light parameter configuration under any gain can be obtained in only seconds.

[0097] Take the gain value as the input variable of the neural network model and the pump light parameter as the output variable of the neural network model. After filling in the missing values, processing the outliers and normalizing the 3000 groups of samples, they are randomly divided into a training set, a test set and a validation set according to the ratio of 7:1.5:1.5.

[0098] The neural network model consists of three parts: an input layer, hidden layers, and an output layer. Based on the dataset, the input layer was assigned one variable, and the output layer was assigned eight variables. The experiment set the number of hidden layers to 3-8, with each layer containing 1-100 neurons, for a total of 600 models. The models were trained using the TRAINSCG algorithm, using regression R and mean squared error (MSE) as evaluation metrics to compare the performance of different models. The final neural network structure was determined to be six hidden layers with 33 neurons per layer. After the neural network model was determined, it was trained using the dataset and verified to have reasonable errors in predicting the pump light parameters. The neural network model could then be directly called.

[0099] The establishment of the neural network model NN(·) is divided into three steps: establishment, training and verification. Figure 4 As shown, the improved particle swarm optimization algorithm (IPSO) is first used to optimize the pump light parameter K. The IPSO algorithm can be found in the reference. This is used as the input of the equation solver. The gain spectrum G is obtained by solving the Raman coupled wave differential equation. G and K together construct a data set. The nonlinear mapping from G to K is learned by constructing a neural network NN(·). Once the neural network model NN(·) is successfully trained, the target gain curve is input to obtain the predicted pump light parameter configuration. This pump light parameter configuration is used as input to obtain the actual gain curve through the equation solver. The accuracy of the prediction using the model is determined by comparing the error between the target gain curve and the actual gain curve. Figure 4 In the figure, Actual gain spectrum represents the actual gain spectrum, Equation solver represents the equation solver, Ramancoupled wave equation represents the solution of the coupled wave equation, Gain represents the gain, Frequency represents the frequency, Dataset represents the dataset, IPSO algorithm represents the IPSO algorithm, Target gain spectrum represents the target gain spectrum, Neural Networks represents the neural network, and Predicted gain spectrum represents the predicted gain peak.

[0100] The closer the absolute value difference between the two is to 0, the higher the accuracy of the network model is, that is, it can accurately replace the traditional method to solve the Raman coupled wave equation. If the absolute value difference between the two is too large, it will continue to be optimized until the error is within a feasible range.

[0101] The present invention adopts BP neural network model, and its topology diagram is as follows Figure 5As shown in Figure 2, the neural network model is divided into three parts: input layer, hidden layer and output layer. n ] T is the input of the neural network, which represents the output gain value of n signal lights of different wavelengths after amplification, K=[λ1,…,λ m ,p1,…,p m ] T is the output of the neural network, and is the wavelength λ and power p of m pump lights. x Indicates the number of hidden layers contained in the structure, R y Represents the number of neurons contained in each hidden layer unit. Different neural network structure algorithms will lead to different training results. The process of determining the optimal neural network model structure is as follows:

[0102] The signal light power is set to 0.01W, the signal light interval is set to 0.8nm, and a total of 125 channels of signal light in the C+L band of 1530nm-1630nm are amplified. Four pump light sources are used to jointly achieve signal light amplification. The present invention first illustrates the optimization effectiveness of the IPSO algorithm.

[0103] Set the parameters and optimization range of the IPSO algorithm, and output the optimal pump light parameter configuration after the iteration within the specified search range. Figure 6 As shown in the figure, the RFA output gain spectrum comparison before and after optimization is shown. The pump light parameters before optimization take the empirical values ​​of 1400nm, 1430nm, 1440nm and 1470nm, and the powers are 0.01W, 0.04W, 0.14W and 0.8W respectively. The obtained output gain is 15.82dB and the flatness is 1.51dB. Compared with the optimized output gain value of 15.93dB and the gain fluctuation range of 0.36dB, it can be seen that the flatness of the RFA is greatly improved after using the IPSO algorithm, that is, the optimization algorithm has a high adaptability to the RFA pump light parameter optimization, and a data set is constructed based on this for training.

[0104] The IPSO algorithm was used to optimize the pump power and wavelength. After 3000 iterations, a collection of 3000 optimal pump parameter configurations was constructed to form a dataset. The dataset was randomly divided into training, validation, and test sets with a ratio of 70%, 15%, and 15%. The network model was trained using the time-saving TRAINSCG function. Mean Square Error (MSE) and regression R value were used as network performance evaluation metrics. MSE represents the mean square error between the predicted and actual values; closer to 0, the smaller the error. R value indicates the closeness between the predicted and actual values; closer to 1, the higher the prediction accuracy of the neural network model. The neural network architecture, such as the number of hidden layers and the number of neurons in each hidden layer, must be determined during training. To simplify the simulation, the number of neurons in each hidden layer was set to be the same, within the range of (0, 100). Excessive or excessive numbers of hidden layers were eliminated, and the number of hidden layers was set within the range of [3, 8]. 600 neural network models with different architectures were constructed.

[0105] like Figure 7 and Figure 8 The following plots show the distribution of R and MSE values ​​for the 600-model model as a function of the number of hidden layers and neurons. When the number of hidden layers increases to 5 and 6, the aliasing phenomenon is reduced, with the overall R value distributed between [0.94, 1] and the MSE value distributed within the range of (0, 1). The performance is even better when the number of neurons is within the range of [20, 80].

[0106] Compare the optimal R value and MSE value of each layer, such as Figure 9 As shown in Figure 1, (a) and (b) show the maximum R and minimum MSE values ​​for each hidden layer in the range [3, 8]. As the number of hidden layers increases, the overall R value gradually increases, while the MSE value gradually decreases. However, when the number of hidden layers increases to 7 and 8, the overall performance decreases, indicating that as the neural network model becomes increasingly complex, its learning ability deviates. It can be clearly seen that the overall performance of the neural network model is better when the hidden layer is 6.

[0107] The optimal neural network structure for reverse engineering RFA was finally determined to contain 6 hidden layers, with 33 neurons in each layer. Based on this structure, 3000 sets of data were trained, and the results are as follows: Figure 10 (a), (b), and (c) are the training, validation, and test results, respectively. The regression R values ​​are 0.9964, 0.99513, and 0.99541, respectively. It can be seen that all data points are concentrated on a straight line, the linear fitting ability is high, and the relationship between the target and the output is approximately linear.

[0108] In order to check the error between the target value and the predicted value, the gain is arbitrarily specified, and the pump light wavelength and power value that matches it is predicted by the NN(·) model. It is used as input and compared with the actual gain value obtained by using the numerical solver, as shown in Figure 11 The figure shows the error distribution between the target and the actual value under different gains. The dotted line in the figure is the target gain value, and the solid line is the gain value predicted by the neural network model. The predicted values ​​are set to 4dB, 7.65dB, 10.75dB, and 13.55dB, respectively. It can be seen that when the target gain value is smaller, the error between the target value and the predicted value is smaller, and the degree of overlap between the two lines is higher. This is because when the power value of the pump light is low, the degree of stimulated Raman scattering effect between the pump light and the signal light is small, the complexity is low, and the error is small, and the error does not exceed 0.47dB.

[0109] Step S6: input the target gain curve into the trained BP neural network model, and output the optimal pump light parameters of the backward multi-pump Raman fiber amplifier.

[0110] To address the slow response and narrow bandwidth shortcomings of next-generation ultra-large-capacity, ultra-high-speed optical transmission systems, this application discloses an effective method for rapidly predicting and optimizing the performance of backward multi-pump Raman fiber amplifiers. First, an improved particle swarm optimization algorithm is used to optimize the pump light parameter configuration. A neural network algorithm is then used to learn the nonlinear mapping relationship between output gain and pump light parameters to inversely design the Raman fiber amplifier. By determining the target output gain, a matching pump wavelength and power are generated, replacing the traditional method of numerically solving the Raman coupled wave differential equation. Combining these two methods improves the accuracy of the neural network model while yielding a Raman fiber amplifier designed for C+L band signal light amplification. Experimental results show that the error between the target and predicted values ​​of the designed Raman fiber amplifier during prediction does not exceed 0.47 dB. This provides valuable insights into the flexible design of future Raman fiber amplifiers.

[0111] The present invention uses an improved particle swarm optimization algorithm to optimize the wavelength and gain of the pump light. On the basis of the traditional particle swarm algorithm, the weight is restricted. While improving the global search capability, it avoids the search particle tendency to regionalization. By finding the optimal solution within a given pump wavelength and power range, this is used as a data set. The backward multi-pump RFA is reverse-designed through a neural network algorithm. The neural network is used to learn the nonlinear mapping relationship between the RFA target gain curve and the pump light wavelength and power, effectively solving the complex calculation process of integrating the Raman coupled wave equation when designing the Raman amplifier. While greatly improving the calculation efficiency, it can also improve the output gain value of the Raman amplifier and slow down the fluctuation of the output gain. By combining the two algorithms, a discrete Raman fiber amplifier for the C+L band is designed. By determining the parameters of the optimal neural network model and the optimization algorithm, the designed Raman amplifier can provide arbitrary gain with high accuracy and improve the output gain flatness. It provides an effective way to design an RFA with a specific gain. At the same time, this method provides a theoretical basis and technical support for the research of optical transmission methods for 6G communication networks.

[0112] An embodiment of the present invention further provides a reverse design system for a backward multi-pumped Raman fiber amplifier, the system comprising:

[0113] An influencing factor determination module, used to determine pump light parameters that affect the gain performance of the backward multi-pump Raman fiber amplifier;

[0114] The optimization module is used to optimize the pump light parameters according to the set pump light parameter value range using an improved particle swarm optimization algorithm to obtain the optimal pump light parameters under different gain conditions;

[0115] A gain calculation module, configured to calculate a gain curve under each gain condition based on the optimal pump light parameters under each gain condition and using the nonlinear Raman coupling differential equation of the backward multi-pump Raman fiber amplifier;

[0116] A data set construction module is used to construct a data set by taking the gain curves under different gain conditions as input and the optimal pump light parameters under different gain conditions as labels;

[0117] A training module is used to train the BP neural network model using the data set to obtain a trained BP neural network model;

[0118] The application module is used to input the target gain curve into the trained BP neural network model and output the optimal pump light parameters of the backward multi-pump Raman fiber amplifier.

[0119] The influencing factor determination module specifically includes:

[0120] The differential equation building submodule is used to build the simplified nonlinear Raman coupling differential equation of the backward multi-pump Raman fiber amplifier: Where, ± corresponds to the forward or reverse injection of the pump light source, P i 、P j 、P k Represents the optical power of the i, j, and k channels respectively, v i 、v j 、v k Represent the optical frequencies of the i, j, and k channels respectively, g R (v i -v j ) represents the Raman gain coefficient between the two channels of light of the i-th and j-th channels, g R (v j -v k ) represents the Raman gain coefficient between the jth and kth channels, K eff represents the polarization factor, A eff Represents the effective core area of ​​the optical fiber, α j represents the attenuation coefficient of the light wave of the jth channel transmitted in the optical fiber, γ j represents the Rayleigh scattering coefficient, K and h represent the Boltzmann constant and Planck constant respectively, is the Bose-Einstein factor, T is the absolute temperature of the optical fiber;

[0121] The pump light parameter determination submodule is used to determine the pump light parameters that affect the gain performance of the backward multi-pump Raman amplifier according to the nonlinear Raman coupling differential equation; the pump light parameters include the power and wavelength of the pump light source.

[0122] The weights in the improved particle swarm optimization algorithm are iteratively updated according to the following formula:

[0123] w=(w1+w2)·(T'-In) / T'+w2

[0124] Where w represents the weight factor at the current iteration number, w1 and w2 represent the initial value and final value of the weight factor respectively, In represents the current iteration number, In≤T', and T' represents the total number of iterations.

[0125] Training modules include:

[0126] The model setting submodule is used to set the BP neural network model including the input layer, hidden layer and output layer;

[0127] The model construction submodule is used to construct a total of 600 BP neural network models under the premise that the number of hidden layers is preset to be 3-8 layers and the number of neurons in each layer is 1-100;

[0128] A partitioning submodule is used to perform missing fill, outlier processing and normalization on the dataset, and randomly divide the processed dataset into a training set, a test set and a validation set in a ratio of 7:1.5:1.5;

[0129] The training submodule is used to train 600 BP neural network models using the TRAINSCG algorithm based on the training set and the validation set;

[0130] The test submodule is used to compare the performance of 600 trained BP neural network models using the test set, with the regression value R value and the mean square error MSE value as evaluation indicators, and to determine that the structure of the BP neural network model with the best performance is 6 hidden layers and each layer contains 33 neurons;

[0131] The model training submodule is used to train the BP neural network model with the best performance using the data set to obtain the trained BP neural network model.

[0132] The pump light parameter value range is: the wavelength range of each pump light is 1400-1500nm, and the power range of each pump light source is 0-2W.

[0133] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0134] The present invention uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A reverse design method for a backward multi-pumped Raman fiber amplifier, characterized in that: The method comprises: Determine the pump light parameters that affect the gain performance of backward multi-pump Raman fiber amplifiers; According to the set pump light parameter value range, the improved particle swarm optimization algorithm is used to optimize the pump light parameters and obtain the optimal pump light parameters under different gain conditions; According to the optimal pump light parameters under each gain condition, the gain curve under each gain condition is calculated using the nonlinear Raman coupling differential equation of the backward multi-pump Raman fiber amplifier; The data set is constructed by taking the gain curves under different gain conditions as input and the optimal pump light parameters under different gain conditions as output. Using the data set to train a BP neural network model to obtain a trained BP neural network model; The target gain curve is input into the trained BP neural network model, and the optimal pump light parameters of the backward multi-pump Raman fiber amplifier are output.

2. The method according to claim 1, characterized in that The determining of pump light parameters affecting the gain performance of the backward multi-pump Raman fiber amplifier specifically includes: The simplified nonlinear Raman coupling differential equation of the backward multi-pump Raman fiber amplifier is established as Where, ± corresponds to the forward or reverse injection of the pump light source, P i 、P j 、P k Represents the optical power of the i, j, and k channels respectively, v i 、v j 、v k Represent the optical frequencies of the i, j, and k channels respectively, g R (v i -v j ) represents the Raman gain coefficient between the two channels of light of the i-th and j-th channels, g R (v j -v k ) represents the Raman gain coefficient between the jth and kth channels, K eff represents the polarization factor, A eff Represents the effective core area of ​​the optical fiber, α j represents the attenuation coefficient of the light wave of the jth channel transmitted in the optical fiber, γ j represents the Rayleigh scattering coefficient, K and h represent the Boltzmann constant and Planck constant respectively, is the Bose-Einstein factor, T is the absolute temperature of the optical fiber; According to the nonlinear Raman coupling differential equation, pump light parameters affecting the gain performance of the backward multi-pump Raman amplifier are determined; the pump light parameters include the power and wavelength of the pump light source.

3. The method according to claim 1, characterized in that The weights in the improved particle swarm optimization algorithm are iteratively updated according to the following formula: w=(w1+w2)·(T'-In) / T'+w2 Where w represents the weight factor at the current iteration number, w1 and w2 represent the initial value and final value of the weight factor, In represents the current iteration number, In≤T', and T' represents the total number of iterations.

4. The method according to claim 1, wherein The BP neural network model is trained using the data set to obtain a trained BP neural network model, specifically including: The BP neural network model is set to include input layer, hidden layer and output layer; Under the premise that the number of hidden layers is preset to be 3-8 and the number of neurons in each layer is 1-100, a total of 600 BP neural network models are constructed; The dataset was subjected to missing data filling, outlier processing and normalization, and the processed dataset was randomly divided into a training set, a test set and a validation set in a ratio of 7:1.5:1.5; Based on the training set and validation set, TRAINSCG algorithm was used to train 600 BP neural network models; Using the test set, the performance of 600 trained BP neural network models was compared with the regression value R value and mean square error MSE value as evaluation indicators. It was determined that the structure of the BP neural network model with the best performance was 6 hidden layers and 33 neurons in each layer. The data set is used to train a BP neural network model with optimal performance to obtain a trained BP neural network model.

5. The method according to claim 1, characterized in that The pump light parameter value ranges are: the wavelength range of each pump light is 1400-1500nm, and the power range of each pump light source is 0-2W.

6. A backward multi-pump Raman fiber amplifier reverse design system, characterized in that: The system comprises: An influencing factor determination module, used to determine pump light parameters that affect the gain performance of the backward multi-pump Raman fiber amplifier; The optimization module is used to optimize the pump light parameters according to the set pump light parameter value range using an improved particle swarm optimization algorithm to obtain the optimal pump light parameters under different gain conditions; A gain calculation module, configured to calculate a gain curve under each gain condition based on the optimal pump light parameters under each gain condition and using the nonlinear Raman coupling differential equation of the backward multi-pump Raman fiber amplifier; A data set construction module is used to construct a data set by taking the gain curves under different gain conditions as input and the optimal pump light parameters under different gain conditions as labels; A training module is used to train a BP neural network model using the data set to obtain a trained BP neural network model; The application module is used to input the target gain curve into the trained BP neural network model and output the optimal pump light parameters of the backward multi-pump Raman fiber amplifier.

7. The system according to claim 6, characterized in that The influencing factor determination module specifically includes: The differential equation building submodule is used to build the simplified nonlinear Raman coupling differential equation of the backward multi-pump Raman fiber amplifier: Where, ± corresponds to the forward or reverse injection of the pump light source, P i 、P j 、P k Represents the optical power of the i, j, and k channels respectively, v i 、v j 、v k Represent the optical frequencies of the i, j, and k channels respectively, g R (v i -v j ) represents the Raman gain coefficient between the two channels of light of the i-th and j-th channels, g R (v j -v k ) represents the Raman gain coefficient between the jth and kth channels, K eff represents the polarization factor, A eff Represents the effective core area of ​​the optical fiber, α j represents the attenuation coefficient of the light wave of the jth channel transmitted in the optical fiber, γ j represents the Rayleigh scattering coefficient, K and h represent the Boltzmann constant and Planck constant respectively, is the Bose-Einstein factor, T is the absolute temperature of the optical fiber; The pump light parameter determination submodule is used to determine the pump light parameters that affect the gain performance of the backward multi-pump Raman amplifier according to the nonlinear Raman coupling differential equation; the pump light parameters include the power and wavelength of the pump light source.

8. The system according to claim 6, characterized in that The weights in the improved particle swarm optimization algorithm are iteratively updated according to the following formula: w=(w1+w2)·(T'-In) / T'+w2 Where w represents the weight factor at the current iteration number, w1 and w2 represent the initial value and final value of the weight factor, In represents the current iteration number, In≤T', and T' represents the total number of iterations.

9. The system according to claim 6, wherein: The training module specifically includes: The model setting submodule is used to set the BP neural network model including the input layer, hidden layer and output layer; The model construction submodule is used to construct a total of 600 BP neural network models under the premise that the number of hidden layers is preset to be 3-8 layers and the number of neurons in each layer is 1-100; A partitioning submodule is used to perform missing fill, outlier processing and normalization on the dataset, and randomly divide the processed dataset into a training set, a test set and a validation set in a ratio of 7:1.5:1.5; The training submodule is used to train 600 BP neural network models using the TRAINSCG algorithm based on the training set and the validation set; The test submodule is used to compare the performance of 600 trained BP neural network models using the test set, with the regression value R value and the mean square error MSE value as evaluation indicators, and to determine that the structure of the BP neural network model with the best performance is 6 hidden layers and each layer contains 33 neurons; The model training submodule is used to train the BP neural network model with the best performance using the data set to obtain the trained BP neural network model.

10. The system according to claim 6, wherein: The pump light parameter value ranges are: the wavelength range of each pump light is 1400-1500nm, and the power range of each pump light source is 0-2W.

Images