Variable coefficient NLSE multi-soliton transmission control system and method based on structure-preserving micropropagator

Abstract

The invention discloses a variable coefficient NLSE multi-soliton transmission control system and a variable coefficient NLSE multi-soliton transmission control method based on a guaranteed structure micro-propagator. The system comprises a link parameter sensing unit, a variable coefficient NLSE digital twinning unit, an analysis multi-soliton seed generation unit, a neural operator precondition unit, a conservation quantity projection and device constraint unit, a transmitting end waveform synthesis and loading unit, a receiving end and a relay feedback unit, the variable coefficient NLSE digital twinning unit is used for constructing a variable coefficient NLSE propagation model, and the other units are used for generating and loading a transmitted waveform under device constraint and conservation constraint, and updating a control strategy according to receiving end / relay feedback. According to the method, a structure-preserving micro-propagator and a conservation quantity projection operator are introduced, inverse problem solving is accelerated by analyzing multi-soliton seed and neural operator Jacobian preconditions, and rapid self-adaptive generation and stable transmission of multi-soliton waveforms are achieved under engineering constraints.

Description

Technical Field

[0001] This invention belongs to the field of optical fiber communication technology and relates to a digital processing method for optical signals, specifically a closed-loop multi-soliton synthesis and transmission control method based on "sensing-modeling-inverse calculation-projection-driving". Background Technology

[0002] To obtain multi-soliton structures, CN103117806A constructs an exact solution to the nonlinear Schrödinger equation through topological and nonlinear transformations and variable separation, and then constructs nonlinear functions to obtain spatial multi-optical solitons of varying numbers. CN103199932A constructs an exact solution to the Schrödinger equation using the Riccati differential equation mapping to realize multi-optical solitons. These schemes lean towards analytical construction and structure development, but do not address the issues of "realizable waveform synthesis and online control" under parameter uncertainties, noise disturbances, device bandwidth, or quantization constraints in the engineering process.

[0003] To address nonlinear compensation in coherent optical systems, such as digital backpropagation, Volterra, and machine learning equalization, numerous patents and solutions exist. US11777612B2 uses digital backpropagation as an example, employing the Fourier method to solve for the Nonlinear Sequence of Arrays (NLSE) for backpropagation compensation of the receiver signal. However, the complexity increases rapidly with transmission distance and the number of WDM channels, making real-time implementation difficult. WO2023243545A1 / US20230409877A1 / EP4385177A1 / B1 utilize "Physical Information Neural Network (PINN) / Differential Digital Twin" for nonlinear compensation and parameter adaptive training, using Fourier operator methods to accelerate fiber optic channel modeling. These solutions primarily focus on "receiver compensation or channel modeling," failing to provide an integrated engineering implementation path for "controllable generation of target multi-soliton waveforms in variable-coefficient links, parameter closed-loop estimation, and stable backpropagation under conservation constraints."

[0004] In summary, there is a need for a transmitter adaptive synthesis and online control, pre-compensation and link parameter closed-loop control method and system stability scheme that can quickly and stably synthesize and maintain a specified multi-soliton waveform even when the link parameters change with temperature, stress and amplifier gain drift. Summary of the Invention

[0005] This invention provides a structure-preserving differentiable propagator-based variable-coefficient NLSE multi-soliton transmission control system and method. Based on the variable-coefficient NLSE digital model, a structure-preserving differentiable propagator and a conserved projection operator are introduced. The inverse problem is solved by "analytical multi-soliton seed and neural operator Jacobi precondition", thereby realizing the rapid adaptive generation and stable transmission of multi-soliton waveforms under engineering constraints.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] A structure-preserving differentiable propagator-based variable-coefficient NLSE multi-soliton transmission control system includes a link parameter sensing unit, a variable-coefficient NLSE digital twin unit, an analytical multi-soliton seed generation unit, a neural operator preconditioning unit, a conserved quantity projection and device constraint unit, a transmitter waveform synthesis and loading unit, a receiver, and a relay feedback unit, wherein:

[0008] The link parameter sensing unit is used to acquire and estimate the dispersion coefficient. Nonlinear coefficients ,loss and amplifier gain Spatial cross-section;

[0009] The variable coefficient NLSE digital twin unit is used to construct a variable coefficient NLSE propagation model. This model takes VC-NLSE as its core, adopts the structure-preserving adaptive modified split-step Fourier method (SSFM) as a high-precision propagator, and provides a differentiable backpropagation interface to calculate the gradient of the objective function.

[0010] The analytical multi-soliton seed generation unit is used to construct a seed containing pulse numbers by utilizing the separation of variables and a known family of exact solutions. The initial multi-soliton parameterized expressions for position, amplitude, phase, and group velocity are used as the initial values ​​for iterative solution of the inverse problem;

[0011] The neural operator preconditioning unit is used to train the neural operator network to realize the mapping of "input waveform → short-range propagation increment / gradient approximation", and uses it as a precondition for the Jacobian approximation and step size and direction.

[0012] The conserved quantity projection and device constraint unit is used to perform conserved quantity projection on the waveform after each update and apply device realizability constraints to finally generate an I / Q drive waveform that meets the requirements.

[0013] The transmitter waveform synthesis and loading unit is used to load the I / Q driving waveform onto the high-speed DAC and IQ modulator / optical pulse shaper to realize multi-soliton optical field emission;

[0014] The receiver and relay feedback unit are used to construct feedback quantities based on the receiver bit error rate, EVM, pilot mismatch, and link sensor output to drive the system to perform online updates and adjustments.

[0015] A method for controlling the transmission of multiple solitons based on a structure-preserving differentiable propagator with variable coefficients in NLSE includes the following steps:

[0016] Step S1, Target Definition: Input the target multi-soliton waveform target or target feature vector ;

[0017] Step S2, Link Coefficient Profile Estimation: Estimated by fusion of pilot and sensor data. This forms a discrete profile or a piecewise constant model;

[0018] Step S3, Seed initialization: Based on the target feature vector Generate multi-soliton initial parameters The initial transmit waveform is generated by parameterization. ,in It is time;

[0019] Step S4, Structure-Preserving Differentiable Propagation: Adaptive correction of SSFM is used for iterative... The function is obtained once. ,right Forward propagation is obtained Calculate the mismatch function with the target. Simultaneously, computation is performed through a differentiable interface. ,in It is a multi-soliton waveform function. It is a function Find the partial derivative;

[0020] Step S5, Neural Operator Precondition Update: Use neural operators to provide gradient precondition directions. The step size is determined using a line search trust region method. Received temporary update ;

[0021] Step S6, Conservative Quantity Projection and Device Constraints: Temporarily update the results Projecting onto the feasible region that satisfies the conservation of quantities and device constraints, we obtain... ;

[0022] Step S7, Convergence Judgment: If If the threshold condition is met or the maximum number of iterations is reached, then output. The transmitted waveform is used; otherwise, return to step S4 to continue iteration.

[0023] Step S8, Online closed loop: Repeat steps S2 to S7 periodically or by event triggering (link state change) to achieve online self-adaptation of the system.

[0024] Compared with the prior art, the present invention has the following advantages:

[0025] 1. Feasibility: By using bandwidth, peak-to-average power ratio, and quantization constraint projection, the solution of the nonlinear Schrödinger equation is directly transformed into a modulator-loadable driving waveform, reducing the error of "simulation feasible but hardware infeasible".

[0026] 2. Adaptability: The link coefficient profile is updated online, which can maintain the phase stability of the target multi-soliton structure and performance indicators after environmental disturbances or network reconstruction.

[0027] 3. Real-time performance: Compared with pure SSFM iterative inverse calculation or transmitter DBP, the number of iterations and propagation steps can be significantly reduced by using the pre-set conditions of parsing and neural operators, making it possible to achieve near real-time implementation on FPGA / GPU platforms.

[0028] 4. Scalability: Supports multi-pulse structures and supports the synthesis of specified channels or specified multiple solitons in single-channel or WDM scenarios.

[0029] 5. Stability: Conservative projection and structure-preserving propagator can suppress numerical energy drift, reduce the divergence probability in multi-soliton collision scenarios, and improve convergence robustness. Attached Figure Description

[0030] Figure 1 The block diagram of a structure-preserving differentiable propagator-based variable-coefficient NLSE multi-soliton transport control system is shown below.

[0031] Figure 2 This is a flowchart of a multi-soliton transport control method based on a structure-preserving differentiable propagator with variable coefficients (NLSE). Detailed Implementation

[0032] The technical solution of the present invention will be further described below with reference to the accompanying drawings, but it is not limited thereto. Any modifications or equivalent substitutions to the technical solution of the present invention that do not depart from the spirit and scope of the technical solution of the present invention should be covered within the protection scope of the present invention.

[0033] This invention provides a structure-preserving differentiable propagator-based variable-coefficient NLSE multi-soliton transmission control system, such as... Figure 1 As shown, the system includes a link parameter sensing unit, a variable coefficient NLSE digital twin unit, an analytical multi-soliton seed generation unit, a neural operator preconditioning unit, a conserved quantity projection and device constraint unit, a transmitter waveform synthesis and loading unit, a receiver unit, and a relay feedback unit. The link parameter sensing unit provides link state parameters, the variable coefficient NLSE digital twin unit constructs a variable coefficient NLSE propagation model, and the remaining units generate and load the transmitted waveform under device and conserved quantity constraints, and update the control strategy based on receiver / relay feedback. Specific functions are as follows:

[0034] The link parameter sensing unit is used to acquire and estimate the dispersion coefficient. Nonlinear coefficients ,loss and amplifier gain The spatial profile can be obtained by OSNR monitoring, pilot training sequence estimation, and fusion of distributed fiber optic sensor outputs.

[0035] The variable coefficient NLSE digital twin unit uses VC-NLSE as its core, adopts structure-preserving adaptive correction SSFM as a high-precision propagator, and provides a differentiable backpropagation interface to calculate the gradient of the objective function.

[0036] The analytical multi-soliton seed generation unit utilizes separation of variables and a known family of exact solutions to construct a structure containing the number of pulses. The initial multi-soliton parameterized expressions for parameters such as position, amplitude, phase, and group velocity are used as the initial values ​​for iterative solution of the inverse problem;

[0037] The neural operator preconditioning unit is used to train Fourier neural operators or other types of operator networks. This network can realize the mapping of "input waveform → short-range propagation increment / gradient approximation", which is used as a precondition for Jacobian approximation and step size and direction, thereby improving iteration efficiency.

[0038] The conserved quantity projection and device constraint unit is used to perform conserved quantity projection (such as energy, momentum, center frequency shift, etc.) on the waveform after each update, and apply device realizability constraints (such as bandwidth, peak-to-average power ratio, quantization limit) to finally generate an I / Q drive waveform that meets the requirements.

[0039] The transmitter waveform synthesis and loading unit is used to load the I / Q driving waveform onto the high-speed DAC and IQ modulator / optical pulse shaper to realize multi-soliton optical field emission;

[0040] The receiver and relay feedback unit are used to construct feedback quantities based on the receiver bit error rate, EVM, pilot mismatch, and link sensor output to drive the system to perform online updates and adjustments.

[0041] This invention also provides a structure-preserving differentiable propagator-based variable-coefficient NLSE multi-soliton transport control method, such as... Figure 2 As shown, the method includes the following steps:

[0042] Step S1, Target Definition: Input the target multi-soliton waveform function or target feature vector ,For example The peak position, peak power, relative phase, pulse width, frequency shift, etc. of each pulse.

[0043] Step S2, Link Coefficient Profile Estimation: Estimated by fusion of pilot and sensor data. This forms a discrete profile or piecewise constant model.

[0044] Step S3, Seed initialization: Based on the target feature vector Generate multi-soliton initial parameters The initial transmit waveform function is generated by parameterization. ,in These are the initial parameters. It's time.

[0045] Step S4, Structure-Preserving Differentiable Propagation: Adaptive correction of SSFM is used for iterative... The function is obtained once. ,right Forward propagation is obtained Calculate the mismatch function with the target. Simultaneously, computation is performed through a differentiable interface. ,in Multi-soliton waveform function, It is a function Find the partial derivative.

[0046] Step S5, Neural Operator Precondition Update: Use neural operators to provide gradient precondition directions. And use line search trust region to determine step size Received temporary update .

[0047] Step S6, Conserved Quantity Projection and Device Constraints: [The sentence is incomplete and requires further context to Projecting onto the feasible region that satisfies the conservation of quantities and device constraints, we obtain... .

[0048] Step S7, Convergence Judgment: If If the threshold condition is met or the maximum number of iterations is reached, then output. This is the transmitted waveform; otherwise, return to step S4 to continue iterating.

[0049] Step S8, Online closed loop: Repeat steps S2 to S7 periodically or by event triggering (link state change) to achieve online self-adaptation of the system.

[0050] Key inventive points and innovations of this invention:

[0051] 1. Combining "analytic multi-soliton operators and differentiable solutions": Utilizing analytical solution families to quickly generate approximate feasible initial values, avoiding local minima and divergence caused by directly inverting from random initial values.

[0052] 2. Introduce conservation projection operators and device constraint operators: map mathematically feasible solutions to engineering-realizable waveforms, improving the robustness of online deployment.

[0053] 3. Deep integration with link parameter awareness: Enables online updates of VC-NLSE coefficient profiles to adapt to time-varying channels caused by temperature, stress, amplifier drift, etc.

[0054] 4. Neural operators are used for "Jacobi / gradient preconditioning" rather than directly replacing the physical model: physical interpretability is preserved while significantly reducing the number of iterations and total propagation steps.

[0055] 5. Using “structure-preserving adaptive SSFM” as a differentiable propagator: Based on the traditional SSFM, local error control and time window adaptation are introduced, and the conserved quantities are kept approximately stable through structured splitting, so that the numerical accuracy of multi-soliton interactions can be maintained at a large step size.

[0056] Example:

[0057] This embodiment takes a "400G coherent single-channel + cross-segment amplification" link as an example. The specific steps are as follows:

[0058] I. Offline Training and Online Deployment:

[0059] 1. Offline Phase: On a representative set of link parameters, high-precision adaptive SSFM is used to generate training pairs (input... Short-range propagation increment (and gradient approximation), Fourier neural operator As a preconditioner; simultaneously constructing a parse multi-soliton seed library (according to...) (Pulse width, phase difference, and interval binning).

[0060] 2. Online Phase: The receiver periodically transmits pilot signals, which, combined with OSNR or dispersion estimation and amplifier gain telemetry, form... , , , Piecewise constant profiles (segment lengths, e.g., 0.5–5 km). If the network supports distributed fiber optic sensing, these profiles can be further refined using temperature and strain estimates. .

[0061] II. Target waveform and loss function:

[0062] The goal is =8 multiple solitaries, time window =2 14 Sampling points, sampling rate =64GSa / s. Target feature vector Includes: 8 peak positions Peak power Relative phase With tolerance ,in .

[0063] Possible loss functions:

[0064]

[0065] in, This refers to peak detection of a signal or physical quantity. Here is a formula for calculating the angle of arrival using the phase difference. This is the bandwidth / peak-to-average ratio regularization term. For optimal constant parameters, Let be the objective function for the multi-soliton waveform, and be... Sobolev space Norm.

[0066] III. Iterative Inverse Calculation and Projection:

[0067] 1. Initial value of iteration It is generated by the analytic function in each iteration.

[0068] 2. Forward Propagation: Obtained by using structure-preserving adaptive correction of SSFM propagation. .

[0069] 3. Gradient calculation: Obtained through backpropagation using a differentiable transducer. .

[0070] 4. Precondition update: Online search results .

[0071] 5. Conserved quantity projection: Calculating energy ,momentum And, Projected to ( , The feasible region is determined by the center frequency shift ≈ 0; then bandwidth truncation and phase quantization are performed (DAC bit width, such as 8~12 bits).

[0072] 6. Output: After convergence, Load the DAC / IQ modulator at the transmitter; if the link parameters change and trigger the threshold, then re-enter the iterative update.

[0073] IV. Example of the effect:

[0074] When link parameters experience slight drift (e.g., dispersion changes due to temperature variations), this invention can update... It performs a small number of iterations (e.g., 1 to 5 times) to recover the target multi-soliton feature error to within the threshold; compared with pure gradient inversion without projection, it can significantly reduce divergence and waveform unrealizable situations.

Claims

1. A structure-preserving differentiable propagator-based variable-coefficient NLSE multi-soliton transmission control system, characterized in that... The system includes a link parameter sensing unit, a variable coefficient NLSE digital twin unit, an analytical multi-soliton seed generation unit, a neural operator preconditioning unit, a conserved quantity projection and device constraint unit, a transmitter waveform synthesis and loading unit, a receiver, and a relay feedback unit, wherein: The link parameter sensing unit is used to acquire and estimate the dispersion coefficient. Nonlinear coefficients ,loss and amplifier gain Spatial cross-section; The variable coefficient NLSE digital twin unit is used to construct the variable coefficient NLSE propagation model; The analytical multi-soliton seed generation unit is used to construct a sequence containing pulse numbers by utilizing the separation of variables and a known family of exact solutions. The initial multi-soliton parameterized expressions for position, amplitude, phase, and group velocity are used as the initial values ​​for iterative solution of the inverse problem; The neural operator preconditioning unit is used to train the neural operator network to realize the mapping of "input waveform → short-range propagation increment / gradient approximation", and uses it as a precondition for the Jacobian approximation and step size and direction. The conserved quantity projection and device constraint unit is used to perform conserved quantity projection on the waveform after each update and apply device realizability constraints to finally generate an I / Q drive waveform that meets the requirements. The transmitter waveform synthesis and loading unit is used to load the I / Q driving waveform onto the high-speed DAC and IQ modulator / optical pulse shaper to realize multi-soliton optical field emission; The receiver and relay feedback unit are used to construct feedback quantities based on the receiver bit error rate, EVM, pilot mismatch, and link sensor output to drive the system to perform online updates and adjustments.

2. The structure-preserving differentiable propagator-based variable-coefficient NLSE multi-soliton transmission control system according to claim 1, characterized in that... The variable coefficient NLSE propagation model is based on VC-NLSE, uses the structure-preserving adaptive correction step-fourth-eighths method as a high-precision propagator, and provides a differentiable backpropagation interface to calculate the gradient of the objective function.

3. A method for multi-soliton transport control based on a structure-preserving differentiable propagator with variable coefficients (NLSE) using the system described in any one of claims 1-2, characterized in that... The method includes the following steps: Step S1, Target Definition: Input the target multi-soliton waveform target or target feature vector ; Step S2, Link Coefficient Profile Estimation: Estimated by fusion of pilot and sensor data. This forms a discrete profile or a piecewise constant model; Step S3, Seed initialization: Based on the target feature vector Generate multi-soliton initial parameters The initial transmit waveform is generated by parameterization. ,in It is time; Step S4, Structure-Preserving Differentiable Propagation: An adaptive modified step-by-step Fourier method is used for iterative propagation. The function is obtained once. ,right Forward propagation is obtained Calculate the mismatch function with the target. Simultaneously, computation is performed through a differentiable interface. ,in It is a multi-soliton waveform function. It is a function Find the partial derivative; Step S5, Neural Operator Precondition Update: Use neural operators to provide gradient precondition directions. The step size is determined using a line search trust region method. Received temporary update ; Step S6, Conservative Quantity Projection and Device Constraints: Temporarily update the results Projecting onto the feasible region that satisfies the conservation of quantities and device constraints, we obtain... ; Step S7, Convergence Judgment: If If the threshold condition is met or the maximum number of iterations is reached, then output. As the transmitted waveform; Otherwise, return to step S4 and continue the iteration; Step S8, Online Closed Loop: Repeat steps S2 to S7 periodically or through event triggering to achieve online self-adaptation of the system.

Images