Large-mode-number multimode group balanced optical fiber and design method thereof

Abstract

The invention discloses a large-mode-number multimode group balanced optical fiber and a design method thereof, and relates to the field of optical fiber communication. The multi-ring-core structure is composed of a preset number of layers of ring-shaped fiber cores; the multi-ring core structure is obtained through optimization of an NN-GA framework, and the NN-GA framework takes a mode distribution quality factor as a core optimization criterion; according to the large-mode-number multi-mode-group balanced optical fiber and the design method thereof, a multi-ring-core structure formed by the preset layer number of annular fiber cores is adopted, and optimization is carried out through a mode distribution quality factor comprehensively defined by the separation degree between mode groups, the consistency in the mode groups and the balance of the number of the mode groups, so that the effective refractive index distribution of each mode is accurately regulated and controlled; the cooperation of maximization of the effective refractive index difference among different mode groups and minimization of the effective refractive index difference in the same mode group is realized, finally the crosstalk among the mode groups is suppressed below-20 dB / km, and the average difference mode time delay is controlled within 10 ps / m.

Description

Technical Field

[0001] This invention relates to optical fiber communication technology, specifically to high mode number multimode group balanced optical fiber and its design method. Background Technology

[0002] With the rapid development of short-distance optical interconnect scenarios such as cloud computing and data centers, data traffic is exploding, placing higher demands on the capacity and cost of optical communication systems. Spatial division multiplexing (SDM) technology, by utilizing the spatial dimension of optical fibers to enhance transmission capacity, has become a key path to overcome the capacity bottleneck of traditional single-mode optical fibers. Among them, multimode optical fibers, due to their low cost and good compatibility with existing optical modules, have significant application potential in short-distance scenarios.

[0003] However, traditional multimode optical fibers suffer from two major problems: first, severe intermode crosstalk, with small differences in effective refractive index between different modes leading to energy coupling and signal interference; and second, significant differential mode delays, with large differences in propagation speeds between different modes within the same mode group, resulting in signal pulse broadening and distortion. Both of these factors reduce signal integrity and limit their application in high-speed systems. Traditional design methods mainly rely on empirical trial and error and numerical simulation, which are inefficient and struggle to find the optimal solution that simultaneously optimizes crosstalk, delay, and manufacturing tolerance within a complex multidimensional parameter space.

[0004] In recent years, machine learning and intelligent optimization algorithms have provided new pathways for optical device design: neural networks can quickly learn the complex nonlinear mapping between structural parameters and optical properties, while genetic algorithms excel at global optimization in large-scale parameter spaces. However, there is currently no solution that deeply integrates the two and designs a dedicated optimization framework for the multi-objective characteristics of multimode fibers, such as mode group separation and delay equalization. Summary of the Invention

[0005] This invention aims to overcome the aforementioned shortcomings of existing technologies and provide a high-mode-number multimode-group balanced optical fiber and its design method. This optical fiber can suppress inter-mode crosstalk, control differential mode delay, and possess good manufacturing tolerance. Its design method integrates the fast prediction capability of neural networks with the global search advantage of genetic algorithms, significantly improving design efficiency and performance.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a high mode number multimode group balanced fiber, including a multi-ring core structure;

[0007] The multi-ring core structure is composed of a predetermined number of ring-shaped fiber cores;

[0008] The multi-ring core structure is obtained by optimization through the NN-GA framework, which uses the pattern distribution quality factor as the core optimization criterion.

[0009] The model distribution quality factor comprehensively considers the separation degree between model groups, the consistency within model groups, and the balance of the number of models in each model group.

[0010] The high mode number multimode group equalized fiber supports multiple spatial modes, which are arranged in n... eff Divide into a predetermined number of modules, with a minimum Δn between modules. eff Designed to effectively suppress inter-group crosstalk;

[0011] Inter-group crosstalk is preset to be less than -20 dB / km;

[0012] The inter-group crosstalk of the high mode number multimode group balanced fiber is suppressed below a preset threshold, the average DMD is controlled within a preset range, and it has a preset manufacturing tolerance, that is, within the range of structural parameter fluctuations allowed by the manufacturing process, the performance degradation remains within an acceptable range.

[0013] The average DMD is controlled within less than 10 ps / m, and the manufacturing tolerance is 5%.

[0014] Furthermore, the ratio of the total cross-sectional area to the cladding area of ​​the multi-ring core structure is adapted to the interface specifications of the optical communication module.

[0015] Furthermore, the spatial pattern includes multiple conventional patterns;

[0016] The intensity and phase distributions of each conventional mode satisfy the condition that, within the same mode group in a multi-ring core structure, n... eff The difference is controlled within a preset small range to ensure the consistency of mode delay within the module group;

[0017] n in group mode eff The difference is less than 3×10 -4 ;

[0018] n of different mode groups eff The difference is designed to be large enough to effectively suppress crosstalk between mode groups.

[0019] n of different mode groups eff The difference is greater than 1.5 × 10 -3 ;

[0020] Furthermore, the verification method for the manufacturing tolerance is as follows:

[0021] Fluctuations within the manufacturing process allowable range are applied to the radius of each annular fiber core and the refractive index difference between it and the cladding. The mode distribution quality factor and key performance indicators after fluctuation are calculated by numerical simulation to ensure that the fiber performance can still meet the preset application requirements within the fluctuation range.

[0022] The key performance indicators include at least inter-group crosstalk and average differential mode delay (DMD).

[0023] The high mode number multimode group balanced fiber includes a multi-ring core structure composed of a predetermined number of ring cores. The fiber supports multiple spatial modes, each mode being divided into a predetermined number of mode groups based on its effective refractive index. Through optimized design, the minimum effective refractive index difference between modes in different mode groups is set to a minimum effective refractive index difference (Δn) sufficient to effectively suppress intermode crosstalk. eff The effective refractive index difference between modes within the same mode group is controlled within the maximum permissible effective refractive index difference (Δn) to ensure time delay consistency. eff Within the range of _intra-max), the inter-mode crosstalk of the optical fiber is lower than the preset value, the average differential mode delay is controlled within the preset range, and its performance degradation remains within an acceptable range under the structural parameter fluctuations allowed by the manufacturing process.

[0024] The mode distribution quality factor is used to quantify the mode group separation performance, delay equalization characteristics, and mode distribution consistency of an optical fiber. Its physical meaning is:

[0025] Taking into account the inter-group n eff The degree of separation of differences and n within the module group eff The "stability" of the difference indicates that the larger the value, the smaller the crosstalk between groups and the more uniform the DMD within the group; the "balance" that characterizes the number of modes in each mode group indicates that the smaller the value, the better the system scalability.

[0026] When the mode distribution quality factor reaches or exceeds the preset optimization threshold, the optical fiber meets the preset core design requirements for inter-group crosstalk and average DMD.

[0027] A method for designing a large mode number multimode group balanced fiber as described above includes the following steps:

[0028] Step 1: Define the mode distribution quality factor as a comprehensive evaluation criterion for fiber performance. The mode distribution quality factor is used to quantify the separation between mode groups, the consistency of delay within a mode group, and the balance of the number of modes between mode groups. Set an optimization threshold. When the mode distribution quality factor reaches or exceeds the optimization threshold, it is determined that the fiber meets the preset design requirements for inter-group crosstalk and average differential mode delay (DMD).

[0029] The mode distribution quality factor is defined as the core optimization criterion. This quality factor comprehensively quantifies the separation between mode groups, the consistency within mode groups, and the balance of the number of modes in each mode group. When its value reaches or exceeds a preset threshold, the optical fiber is considered to meet the crosstalk and delay design requirements.

[0030] Step 2: Construct the training dataset: Using the radius (r1-r) of each layer of the multi-ring core structure n ) and the refractive index difference between each layer and the cladding (Δn1-Δn) nUsing ) as input parameters, the effective refractive index (n) of each mode of the optical fiber under corresponding parameters is calculated through numerical simulation. eff ), generate structure parameters -n eff Mapping datasets;

[0031] Construct a neural network training dataset. Using the radius of each layer of the multi-ring core structure and the refractive index difference with the cladding as input parameters, calculate the effective refractive index of all guided modes under the corresponding structure using finite element simulation software, and generate a "structural parameters-effective refractive index" dataset.

[0032] Step 3: Train the Neural Network Prediction Model (NN Model): Construct a neural network model, which includes an input layer, hidden layers, and an output layer. The number of neurons in the input layer corresponds to the dimension of the structural parameters. The hidden layers are set to be multi-layered, with the number of neurons in each layer adapted to the nonlinear mapping requirements. The output layer is n for each pattern. eff Predicted value;

[0033] The NN model is trained using the dataset, by minimizing the prediction n. eff With numerical simulation n eff The MSE optimization model is continued until the model converges and reaches the prediction accuracy.

[0034] A neural network model is constructed and trained, taking the aforementioned structural parameters as input and the predicted effective refractive index of each mode as output. The model is trained using a dataset, and convergence is achieved by minimizing the mean square error between the predicted and simulated values, ultimately obtaining a surrogate model that can quickly and accurately predict optical properties.

[0035] Step 4: Global optimization based on genetic algorithm (GA): Using the quality factor of the pattern distribution as the fitness function, the parameters of the multi-ring core structure are binary encoded. The parameter population is iteratively updated through selection, crossover, and mutation operations. In each generation, the individual optimized parameters are input into the NN model trained in Step 3 to predict n. eff The pattern distribution quality factor is calculated until the fitness function converges or the preset number of iterations is reached, and the optimal combination of structure parameters is output.

[0036] Global optimization is performed based on a genetic algorithm. The structural parameters are encoded, and the fitness function is the aforementioned pattern distribution quality factor. In each generation of optimization, the individual parameters of the population are input into the trained neural network model to quickly obtain the effective refractive index and calculate the fitness. Iterative evolution is performed through selection, crossover, and mutation operations until the fitness converges or the preset number of iterations is reached, and the optimal structural parameters are output.

[0037] Step 5, Performance Verification and Determination: Substitute the optimal combination of structural parameters into numerical simulation to calculate its Δn. effIf all performance indicators meet the preset design requirements, such as inter-group crosstalk, DMD, and manufacturing tolerance performance, then the structure corresponding to this set of parameters is determined to be the multimode group balanced fiber; otherwise, return to step 4 to adjust the optimization parameters and search again.

[0038] Performance verification and determination. The optimal parameters output by the genetic algorithm are substituted into simulation software for precise verification, confirming that the mode group separation, crosstalk, delay, and manufacturing tolerance performance all meet the design requirements, thereby determining the final multimode group balanced fiber structure.

[0039] Furthermore, the parameter range of the multi-ring core structure in step 2 satisfies:

[0040] The values ​​of the radii of each layer are chosen to make the total diameter of the fiber core suitable for the packaging requirements of short-distance optical communication. The refractive index difference between each layer and the cladding layer meets the material properties and optical field constraints, avoiding abnormal optical field distribution caused by excessively narrow layer width or excessive refractive index difference.

[0041] In step 2, the range of values ​​for each parameter of the multi-ring core structure must ensure that the total diameter of the optical fiber is compatible with the standard optical communication packaging interface, and that the refractive index difference of each layer is within the material's achievable range, so as to avoid optical field distortion or manufacturing difficulties caused by structural extremes. The ratio of the total cross-sectional area of ​​the multi-ring core structure of the optical fiber to the cladding area must be compatible with the interface specifications of mainstream optical communication modules.

[0042] Furthermore, the convergence criterion for training the NN model in step 3 is: when the number of iterations reaches a certain number, the MSE fluctuation amplitude is less than the preset value, and the MSE does not decrease significantly after multiple consecutive iterations.

[0043] The convergence criterion for NN training is: after more than 100 iterations, if the fluctuation range of MSE is less than 1×10⁻⁶ for more than 20 consecutive iterations. -5 If the model converges, then the model is considered to have converged.

[0044] In step 3, the convergence criterion for the neural network model is that the training loss decreases by less than a preset value in consecutive iterations.

[0045] Furthermore, the parameters of GA in step 4 are set as follows:

[0046] The population size, crossover probability, and mutation probability are preset based on optimization efficiency and global search capability. The iteration termination condition is that the improvement of the pattern distribution quality factor is less than the preset value for several consecutive generations or the number of iterations reaches the preset upper limit.

[0047] In step 4, the population size, crossover probability, and mutation probability of the genetic algorithm are preset according to the complexity of the optimization problem, and the iteration termination condition is that the fitness improvement is less than a threshold for multiple consecutive generations.

[0048] When the optimal fitness value (pattern distribution quality factor) of the population no longer shows significant improvement over multiple generations (i.e., the relative improvement rate remains below a small threshold), the algorithm can be determined to have converged to the optimal region, and the iteration can be terminated.

[0049] Furthermore, the performance verification in step 5 also includes:

[0050] Calculate n of multimode group-equalized fiber in the commonly used optical communication bands. eff With Δn eff The variation ensures that intergroup crosstalk and DMD within this wavelength range are always maintained at a preset level of less than 10 ps / m.

[0051] The performance verification also includes scanning wavelengths within the target communication band to verify the stability of the fiber's mode group characteristics, crosstalk, and time delay performance throughout the entire band.

[0052] Furthermore, the numerical simulation in step 2 is set as follows: commonly used optical communication bands, and boundary conditions and mesh generation accuracy are adopted to ensure n eff The calculation error is within an acceptable range;

[0053] When structural parameters fluctuate within the allowable range of the manufacturing process, performance degradation remains within an acceptable range, making it suitable for industrial-grade manufacturing precision.

[0054] Compared with the prior art, the large mode number multimode group balanced fiber and its design method provided by the present invention adopt a multi-ring core structure composed of a ring core with a preset number of layers, and optimizes the mode distribution quality factor by comprehensively defining the inter-mode group separation, intra-mode group consistency and mode group quantity balance, thereby precisely controlling the effective refractive index distribution of each mode. This achieves the synergy of maximizing the effective refractive index difference between different mode groups and minimizing the effective refractive index difference within the same mode group, and finally suppresses inter-mode group crosstalk to below -20 dB / km and controls the average differential mode delay to within 10 ps / m.

[0055] An intelligent design framework integrating neural networks and genetic algorithms was constructed. First, the neural network was trained to establish a fast and accurate mapping from structural parameters to effective refractive index, replacing time-consuming numerical simulation. Then, the quality factor of the pattern distribution was used as the fitness function of the genetic algorithm to automatically iterate and search in the global parameter space. This framework transforms the multi-objective and multi-constraint optical fiber design process from manual trial and error that relies on experience to efficient automatic global optimization, shortening the design cycle by orders of magnitude.

[0056] The optimization objective incorporates considerations of manufacturing process fluctuations. By performing tolerance sensitivity analysis on key structural parameters (such as the radius and refractive index difference of each layer) within the preset process error range and feeding it back to the optimization loop, it ensures that the optimal combination of structural parameters can still meet system requirements under typical manufacturing fluctuations of ±3%, thereby improving the robustness of the design, the manufacturability and yield of the product, and making it suitable for scenarios such as data center interconnection and cloud storage optical links. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0058] Figure 1 This is a flowchart of the genetic algorithm and neural network collaborative optimization provided in an embodiment of the present invention;

[0059] Figure 2 A scatter plot of the neural network training dataset provided in this embodiment of the invention;

[0060] Figure 3 The convergence curve of the pattern distribution quality factor in the iterative optimization of the genetic algorithm provided in the embodiments of the present invention;

[0061] Figure 4 In the embodiments of the present invention, 'a' represents different module groups 'n'. eff The characteristic curves of wavelength variation are shown in Figure b, where Δn is the value of different mode groups provided in the embodiments of the present invention. eff Characteristic curves as a function of wavelength. Detailed Implementation

[0062] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0063] As attached Figure 1 To be continued Figure 4 As shown:

[0064] Example 1:

[0065] This embodiment discloses a high mode number multimode group balanced fiber, including a multi-ring core structure;

[0066] The multi-ring core structure consists of a predetermined number of ring-shaped fiber cores;

[0067] The multi-ring core structure was obtained through optimization using the NN-GA framework, which takes the pattern distribution quality factor as the core optimization criterion.

[0068] The model distribution quality factor comprehensively considers the separation degree between model groups, the consistency within model groups, and the balance of the number of models in each model group.

[0069] High mode number multimode group equalized fiber supports multiple spatial modes, with spatial modes arranged in n... eff Divide into a predetermined number of modules, with a minimum Δn between modules. eff Designed to effectively suppress inter-group crosstalk;

[0070] Inter-group crosstalk in high mode number multimode group balanced fiber is suppressed below a preset threshold, the average DMD is controlled within a preset range, and it has a preset manufacturing tolerance, that is, within the range of structural parameter fluctuations allowed by the manufacturing process, the performance attenuation remains within an acceptable range.

[0071] The ratio of the total cross-sectional area to the cladding area of ​​the multi-ring core structure is adapted to the interface specifications of the optical communication module.

[0072] Spatial patterns include several conventional patterns;

[0073] The intensity and phase distributions of each conventional mode satisfy the condition that, within the same mode group in a multi-ring core structure, n... eff The difference is controlled within a preset small range to ensure the consistency of mode delay within the module group;

[0074] n of different mode groups eff The difference is designed to be large enough to effectively suppress crosstalk between mode groups.

[0075] The verification method for manufacturing tolerance is as follows:

[0076] Fluctuations within the manufacturing process allowable range are applied to the radius of each annular fiber core and the refractive index difference between it and the cladding. The mode distribution quality factor and key performance indicators after fluctuation are calculated by numerical simulation to ensure that the fiber performance can still meet the preset application requirements within the fluctuation range.

[0077] Key performance indicators include at least inter-group crosstalk and average differential mode delay (DMD).

[0078] A design method for high mode number multimode group balanced optical fiber includes the following steps:

[0079] Step 1: Define the mode distribution quality factor as a comprehensive evaluation criterion for fiber performance. The mode distribution quality factor is used to quantify the separation between mode groups, the consistency of delay within a mode group, and the balance of the number of modes between mode groups. Set an optimization threshold. When the mode distribution quality factor reaches or exceeds the optimization threshold, it is determined that the fiber meets the preset design requirements for inter-group crosstalk and average differential mode delay (DMD).

[0080] Step 2: Construct the training dataset: Using the radius (r1-r) of each layer of the multi-ring core structuren ) and the refractive index difference between each layer and the cladding (Δn1-Δn) n Using ) as input parameters, the effective refractive index (n) of each mode of the optical fiber under corresponding parameters is calculated through numerical simulation. eff ), generate structure parameters -n eff Mapping datasets;

[0081] Step 3: Train the Neural Network Prediction Model (NN Model): Construct a neural network model, which includes an input layer, hidden layers, and an output layer. The number of neurons in the input layer corresponds to the dimension of the structural parameters. The hidden layers are set to be multi-layered, with the number of neurons in each layer adapting to the non-linear mapping requirements. The output layer is n for each pattern. eff Predicted value;

[0082] Train the NN model using the dataset, and minimize the prediction n eff With numerical simulation n eff The MSE optimization model is continued until the model converges and reaches the prediction accuracy.

[0083] Step 4: Global optimization based on genetic algorithm (GA): Using the pattern distribution quality factor as the fitness function, the parameters of the multi-ring core structure are binary encoded. The parameter population is iteratively updated through selection, crossover, and mutation operations. In each generation, the individual optimized parameters are input into the NN model trained in Step 3 to predict n. eff The pattern distribution quality factor is calculated until the fitness function converges or the preset number of iterations is reached, and the optimal combination of structure parameters is output.

[0084] Step 5, Performance Verification and Determination: Substitute the optimal combination of structural parameters into the numerical simulation and calculate its Δn. eff If all performance indicators meet the preset design requirements, such as inter-group crosstalk, DMD, and manufacturing tolerance performance, then the structure corresponding to this set of parameters is determined to be a multimode group balanced fiber; otherwise, return to step 4 to adjust and optimize the parameters and search again.

[0085] The parameter range of the multi-ring core structure in step 2 satisfies:

[0086] The values ​​of the radii of each layer are chosen to make the total diameter of the fiber core suitable for the packaging requirements of short-distance optical communication. The refractive index difference between each layer and the cladding layer meets the material properties and optical field constraints, and avoids abnormal optical field distribution caused by excessively narrow layer width or excessive refractive index difference.

[0087] The convergence criterion for training the NN model in step 3 is: when the number of iterations reaches a certain number, the MSE fluctuation amplitude is less than the preset value, and the MSE does not decrease significantly after multiple consecutive iterations.

[0088] In step 4, the parameters of GA are set as follows:

[0089] The population size, crossover probability, and mutation probability are preset based on optimization efficiency and global search capability. The iteration termination condition is that the improvement of the pattern distribution quality factor is less than the preset value for several consecutive generations or the number of iterations reaches the preset upper limit.

[0090] Step 5 performance verification also includes:

[0091] Calculate n of multimode group-equalized fiber in the commonly used optical communication bands. eff With Δn eff The variation ensures that intergroup crosstalk and DMD within this wavelength range are always maintained at a preset level of less than 10 ps / m.

[0092] The numerical simulation in step 2 is set as follows: commonly used optical communication bands, and boundary conditions and mesh generation accuracy are adopted to ensure n eff The calculation error is within an acceptable range.

[0093] Specifically, it demonstrates the entire process of designing a multimode group balanced optical fiber suitable for short-distance communication in the C-band (1530nm-1570nm) using the NN-GA framework:

[0094] S1: Dataset Construction and Preprocessing

[0095] To train a neural network model to accurately predict the optical properties of multi-ring fiber, a comprehensive and representative dataset is first needed. In this embodiment, key geometric and optical parameters constituting the multi-ring structure are selected as input variables. Specifically, the input parameters include the radius of each ring core (r1-r...). n Where n is the number of annular core layers and the refractive index difference (Δn1-Δn) of each annular core relative to the cladding. n These parameters take values ​​that cover the range feasible by actual manufacturing processes and ensure that the target number of spatial modes can be excited. The values ​​of the radii of each layer conform to the limitations of fiber drawing process and light field distribution, while the refractive index difference takes into account both mode constraint capability and material dispersion characteristics.

[0096] Numerical simulations of the fiber optic structure corresponding to each set of input parameters were performed using the mature commercial finite element method (FEM) simulation software (COMSOL). The simulation wavelength was set in the C-band (1530nm to 1570nm), commonly used in optical communication. This band has low transmission loss and abundant optoelectronic device resources. In the simulation settings, a perfectly matched layer (PML) was used as the boundary condition to effectively absorb electromagnetic energy at the boundary, reduce reflection interference, and ensure the accuracy of the simulation results. An adaptive meshing strategy was adopted, with finer meshing in the fiber core region and regions with drastic refractive index changes to ensure calculation accuracy. Through simulation, the n-values ​​of each guided mode in the fiber were extracted. effAs output labels, the final result is a "structural parameter-n" set containing thousands to tens of thousands of samples. eff "To avoid model overfitting and to evaluate its generalization ability, the dataset is randomly divided into a training set and a validation set in a preset ratio of 7:3."

[0097] Construction and training of S2 and NN models:

[0098] The number of neurons in the input layer of the neural network is consistent with the total dimension of the selected structural parameters. If n layers of toroidal fiber core are considered, the number of neurons in the input layer is 2n (n radius parameters and n refractive index difference parameters). The hidden layer of the network adopts a multi-layer design, with each layer containing multiple neurons to enhance the model's ability to learn complex nonlinear mapping relationships. The activation function of the hidden layer neurons is the Rectified Linear Unit (ReLU), because it can effectively alleviate the gradient vanishing problem and accelerate training convergence. The number of neurons in the output layer corresponds to the number of guided modes that the optical fiber may support in the target band. Each output neuron corresponds to an effective refractive index prediction value for a specific mode.

[0099] During training, the mean squared error (MSE) was used as the loss function to measure the deviation between the model's predicted values ​​and the FEM simulation values. The Adam optimizer was employed to iteratively update the network parameters. This optimizer combines momentum and adaptive learning rate strategies to efficiently find the optimal solution. During training, appropriate batch sizes and initial learning rates were set and dynamically adjusted based on changes in the validation set loss. Training iterations continued until the loss function converged to a preset, relatively small threshold, or the validation set loss no longer decreased significantly, to prevent overfitting. After thousands of iterations, the MSE of the training set decreased to the order of 10⁻⁷, and the MSE of the validation set reached a similar level. Simultaneously, the correlation coefficient (R) between the predicted effective refractive index and the FEM simulation values ​​was close to 0.99, indicating that the model has extremely high prediction accuracy and good generalization performance. (See also...) Figure 2 As shown, Figure 2 The results show a high degree of agreement between the NN predictions and the COMSOL simulations.

[0100] S3, GA optimization and optimal structure search:

[0101] The population size of the genetic algorithm is set to a moderate value to achieve a balance between search efficiency and diversity. Crossover probability and mutation probability are key parameters affecting the GA's search capability; in this embodiment, they are set to 0.7 and 0.02, respectively. The crossover probability controls the frequency of gene exchange; a higher crossover probability helps generate new individual combinations. The mutation probability ensures population diversity and prevents the algorithm from prematurely converging to a local optimum. The number of iterations in the optimization process is set to 100 generations to ensure the algorithm has sufficient time to explore the parameter space.

[0102] The optimization objective is to maximize the mode distribution quality factor defined earlier. Therefore, the fitness function of GA directly adopts the mode distribution quality factor. For each individual in the population (representing a set of fiber structure parameters), it is first decoded into actual structure parameter values, and then input into the trained NN model to quickly predict the effective refractive index of each mode corresponding to the structure. Then, the mode distribution quality factor is calculated based on these predicted effective refractive indices, which serves as the fitness of that individual.

[0103] The GA optimization process follows the classic steps of selection, crossover, and mutation in a cyclical manner. In each generation, superior individuals are selected based on their fitness values ​​to advance to the next generation, and new individuals are generated through crossover and mutation operations. Please refer to [link to relevant documentation]. Figure 3 The convergence curve shows that after multiple generations of iteration, the average fitness and optimal fitness of the population gradually increase and tend to stabilize. When the number of iterations reaches the preset number or the fitness value no longer increases significantly, the optimization process terminates.

[0104] Ultimately, the individual with the highest fitness in the GA output is the corresponding optimal combination of structural parameters for the multi-ring fiber.

[0105] S4. Optimal Structure Performance Verification and Analysis:

[0106] The optimal structural parameters obtained from GA optimization are input into COMSOL for detailed analysis:

[0107] Simulation results show that the fiber structure can stably support more than 100 spatial modes within the target communication band. Based on the effective refractive index of each mode, these modes can be clearly divided into 10 mode groups. The minimum Δn between mode groups... eff Greater than 1.5×10 -3 This provides a solid physical basis for effectively suppressing crosstalk between modules. Please refer to [link / reference]. Figure 4 , Figure 4 The paper shows the effective refractive index distribution and module division of different modes.

[0108] The inter-group crosstalk of the optimized optical fiber was evaluated through simulation calculations:

[0109] The results show that, at typical transmission distances, the crosstalk between any two different modules is less than -20 dB / km, and the crosstalk between some modules can even reach below -25 dB / km, ensuring the independence and integrity of signal transmission in different modules.

[0110] The propagation delay of different modes in each module of the optimized optical fiber was accurately calculated. The results show that the maximum DMD in each module and the average DMD of all modules are less than 10 ps / m. This means that different modes in the same module can be transmitted at similar speeds, effectively reducing inter-symbol interference and simplifying the DSP burden at the receiver.

[0111] To evaluate the robustness of the optimized design, this embodiment focuses on the key structural parameter r of the large mode number multimode group balanced fiber. n With Δn n A certain range of ±5% random fluctuation was applied, and its impact on the core performance of optical fiber (such as mode distribution quality factor, crosstalk, and DMD) was analyzed.

[0112] The results show that even within these manufacturing error ranges, although the core performance indicators of the optical fiber in this embodiment have slightly decreased, they can still meet the preset application requirements. This fully demonstrates that the optical fiber structure obtained by the design method of this invention has good manufacturing tolerance and practical application value.

[0113] This invention designs a high mode number multimode group balanced fiber by combining the mode distribution quality factor criterion with the NN-GA framework. This solves the crosstalk and DMD problems of traditional multimode fibers, while improving design efficiency and manufacturing compatibility. The high mode number multimode group balanced fiber has significant performance and cost advantages in short-distance, high-capacity optical communication scenarios, providing a key optical fiber solution for the practical application of SDM technology.

[0114] In the future, the generalization ability of the NN model can be further optimized to adapt to the transmission needs of more bands (such as L-band).

[0115] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A high mode number multimode group balanced fiber, characterized in that, Including multi-ring core structures; The multi-ring core structure is composed of a predetermined number of ring-shaped fiber cores; The multi-ring core structure is obtained by optimization through the NN-GA framework, which uses the pattern distribution quality factor as the core optimization criterion. The model distribution quality factor comprehensively considers the separation degree between model groups, the consistency within model groups, and the balance of the number of models in each model group. The high mode number multimode group equalized fiber supports multiple spatial modes, which are arranged in n... eff Divide into a predetermined number of modules, with a minimum Δn between modules. eff Designed to effectively suppress inter-group crosstalk; The inter-group crosstalk of the high mode number multimode group balanced fiber is suppressed below a preset threshold, the average DMD is controlled within a preset range, and it has a preset manufacturing tolerance, that is, within the range of structural parameter fluctuations allowed by the manufacturing process, the performance degradation remains within an acceptable range.

2. The high mode number multimode group balanced fiber according to claim 1, characterized in that, The ratio of the total cross-sectional area to the cladding area of ​​the multi-ring core structure is adapted to the interface specifications of the optical communication module.

3. The high mode number multimode group balanced fiber according to claim 1, characterized in that, The spatial pattern includes multiple conventional patterns; The intensity and phase distributions of each conventional mode satisfy the condition that, within the same mode group in a multi-ring core structure, n... eff The difference is controlled within a preset small range to ensure the consistency of mode delay within the module group; n of different mode groups eff The difference is designed to be large enough to effectively suppress crosstalk between mode groups.

4. The high mode number multimode group balanced fiber according to claim 1, characterized in that, The verification method for the manufacturing tolerance is as follows: Fluctuations within the manufacturing process allowable range are applied to the radius of each annular fiber core and the refractive index difference between it and the cladding. The mode distribution quality factor and key performance indicators after fluctuation are calculated by numerical simulation to ensure that the fiber performance can still meet the preset application requirements within the fluctuation range. The key performance indicators include at least inter-group crosstalk and average differential mode delay (DMD).

5. A method for designing a large mode number multimode group balanced optical fiber as described in any one of claims 1-4, characterized in that, Includes the following steps: Step 1: Define the mode distribution quality factor as a comprehensive evaluation criterion for fiber performance. The mode distribution quality factor is used to quantify the separation between mode groups, the consistency of delay within a mode group, and the balance of the number of modes between mode groups. An optimization threshold is set. When the mode distribution quality factor reaches or exceeds the optimization threshold, it is determined that the optical fiber meets the preset inter-group crosstalk and average differential mode delay (DMD) design requirements. The optimization threshold is between 0.70 and 0.90; Step 2: Construct the training dataset: Using the radius (r1-r) of each layer of the multi-ring core structure n ) and the refractive index difference between each layer and the cladding (Δn1-Δn) n Using ) as input parameters, the effective refractive index (n) of each mode of the optical fiber under corresponding parameters is calculated through numerical simulation. eff ), generate structure parameters -n eff Mapping datasets; Step 3: Train the Neural Network Prediction Model (NN Model): Construct a neural network model, which includes an input layer, hidden layers, and an output layer. The number of neurons in the input layer corresponds to the dimension of the structural parameters. The hidden layers are set to be multi-layered, with the number of neurons in each layer adapted to the nonlinear mapping requirements. The output layer is n for each pattern. eff Predicted value; The NN model is trained using the dataset, by minimizing the prediction n. eff With numerical simulation n eff The MSE optimization model is continued until the model converges and reaches the prediction accuracy. Step 4: Global optimization based on genetic algorithm (GA): Using the quality factor of the pattern distribution as the fitness function, the parameters of the multi-ring core structure are binary encoded. The parameter population is iteratively updated through selection, crossover, and mutation operations. In each generation, the individual optimized parameters are input into the NN model trained in Step 3 to predict n. eff The pattern distribution quality factor is calculated until the fitness function converges or the preset number of iterations is reached, and the optimal combination of structure parameters is output. Step 5, Performance Verification and Determination: Substitute the optimal combination of structural parameters into numerical simulation to calculate its Δn. eff Inter-group crosstalk, DMD and manufacturing tolerance performance; if all performance indicators meet the preset design requirements, then the structure corresponding to this set of parameters is determined to be the multimode group balanced fiber. Otherwise, return to step 4 to adjust and optimize the parameters and search again.

6. The design method for high mode number multimode group balanced fiber according to claim 5, characterized in that, The parameter range of the multi-ring core structure in step 2 satisfies: The values ​​of the radii of each layer are chosen to make the total diameter of the fiber core suitable for the packaging requirements of short-distance optical communication, and the refractive index difference between each layer and the cladding satisfies the material properties and optical field constraints.

7. The design method for high mode number multimode group balanced fiber according to claim 5, characterized in that, The convergence criterion for training the NN model in step 3 is: when the number of iterations reaches a certain number, the MSE fluctuation amplitude is less than the preset value, and the MSE does not decrease significantly after multiple consecutive iterations.

8. The design method for high mode number multimode group balanced fiber according to claim 5, characterized in that, In step 4, the parameters of GA are set as follows: The population size, crossover probability, and mutation probability are preset based on optimization efficiency and global search capability. The iteration termination condition is that the improvement of the pattern distribution quality factor is less than the preset value for several consecutive generations or the number of iterations reaches the preset upper limit.

9. The design method for high mode number multimode group balanced fiber according to claim 6, characterized in that, The performance verification in step 5 also includes: Calculate n of multimode group-equalized fiber in the commonly used optical communication bands. eff With Δn eff The variation ensures that intergroup crosstalk and DMD within this wavelength range are always maintained at a preset level of less than 10 ps / m.

10. The design method for high mode number multimode group balanced fiber according to claim 9, characterized in that, The numerical simulation in step 2 is set as follows: commonly used optical communication bands, and boundary conditions and mesh generation accuracy are adopted to ensure n eff The calculation error is within an acceptable range.

Images