Fiber laser polarization locking system and method based on genetic algorithm and proportional integral differential cooperative control

Abstract

The invention discloses a fiber laser polarization locking system and method based on genetic algorithm and proportional integral differential cooperative control, and belongs to the technical field of fiber laser control. The invention provides a cooperative control architecture of global searching, gradient sensing and local locking for solving the problems of difficult starting and easy lock losing caused by multiple extreme values of an unlocking space and non-uniform sensitivity distribution of a nonlinear polarization rotation mode-locked laser. When the system starts or loses lock, firstly, the global traversal capability of a genetic algorithm is utilized to quickly pull a polarization state to a quasi-steady-state mode locking interval; before switching to closed-loop locking, a local sensitivity calibration mechanism is introduced, the response gradient of a current working point to control voltage is detected in real time, and gain parameters of a PID controller are adaptively configured according to the response gradient, so that the non-uniform characteristic of a solution space is matched, and overmodulation oscillation is prevented; and then smoothly switching to a PID control mode based on negative feedback to realize continuous fine tuning and compensation of environmental disturbance. Meanwhile, a local backtracking self-healing mechanism based on gradient is built in the system, and quick recovery can be achieved when lock losing is caused by disturbance. According to the invention, through seamless cooperation of the genetic algorithm and PID control, rapid self-starting, high-precision locking and strong anti-interference capabilities are effectively considered, and the long-term operation stability of the fiber laser is significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of fiber laser control technology, specifically relating to a polarization state closed-loop control system and method for mode-locked fiber lasers. In particular, this invention relates to a polarization locking scheme based on a genetic algorithm and proportional-integral-derivative (PID) collaborative control architecture, achieving parameter optimization through local sensitivity adaptation. It is mainly used to solve the difficulties in automatic mode-locking startup and long-term steady-state maintenance in nonlinear polarization rotation (NPR) mode-locked lasers caused by the multiple extrema, non-convexity of the solution space, and uneven sensitivity distribution. Background Technology

[0002] Passively mode-locked fiber lasers, especially those based on the nonlinear polarization rotation (NPR) mechanism, have become key light sources in cutting-edge fields such as precision spectroscopy and micro / nano fabrication due to their ability to generate femtosecond-level ultrashort pulses and their compact structure. However, the core bottleneck in their transition from laboratory to industrial applications lies in their insufficient long-term operational stability: slight disturbances caused by environmental temperature fluctuations or mechanical vibrations can easily lead to polarization state drift within the cavity, resulting in mode-lock instability or even complete loss of lock.

[0003] The physical root of this stability problem lies in the NPR mode-locking mechanism itself. This mechanism essentially relies on the coupling between the nonlinear Kerr effect and the intracavity polarization evolution, making the resonant cavity equivalent to a nonlinear interferometer that is extremely sensitive to its environment. The inherent random birefringence of single-mode fiber causes the position of the intracavity polarization state on the Poincaré sphere to evolve randomly with environmental factors. More importantly, the system's effective solution space exhibits high nonconvexity and multiple extrema, and the response sensitivity (i.e., gradient) of the mode-locked state to control parameters (such as the polarization controller voltage) at different operating points is extremely non-uniform. This "non-uniform sensitivity" characteristic poses a fundamental challenge to traditional linear control strategies.

[0004] To address the above challenges, two main types of automatic mold-locking control schemes have been developed, but both have inherent limitations: The first category is global search strategies based on evolutionary algorithms (such as genetic algorithms and particle swarm optimization). While these schemes can achieve self-starting mode-locking, they rely on random mutation and crossover search mechanisms, which continuously inject random probe noise into the system when approaching or maintaining a steady state. This is equivalent to introducing additional phase perturbations at the controller end, leading to timing jitter and phase noise degradation of the output pulses, failing to meet the stringent requirements of high-precision applications for signal purity and long-term stability.

[0005] The second category is intelligent control schemes based on machine learning (especially deep learning). These schemes directly establish the mapping relationship between control parameters and mode-locked states by training neural networks, resulting in fast response times. However, their performance heavily relies on a large number of high-quality training samples, leading to high model training costs and requiring significant computing hardware support. This makes it difficult to achieve reliable online learning and real-time control on low-cost, low-power embedded control platforms, raising questions about their engineering practicality and robustness.

[0006] In summary, existing technologies present a dilemma: while evolutionary algorithm-based solutions offer strong adaptability, they suffer from poor steady-state accuracy and introduce noise; conversely, deterministic or data-driven solutions, while potentially offering high steady-state accuracy, either lack global search capabilities (e.g., fixed-parameter PID) or rely on stringent prior conditions and computational power (e.g., neural networks). Therefore, there is an urgent need in this field for a novel collaborative control architecture that can balance rapid global search, high-precision steady-state locking, strong anti-interference capabilities, and high engineering practicality. This is not only a key technology for improving the environmental adaptability of NPR mode-locked lasers but also a core prerequisite for their industrial application. Summary of the Invention

[0007] This invention aims to overcome the technical contradiction revealed in the background technology: namely, the contradiction between the inability of global search algorithms relying on stochastic mechanisms to ensure steady-state accuracy and the inability of deterministic control relying on fixed parameters to adapt to nonlinear and non-uniform solution spaces. To this end, this invention proposes a new paradigm of collaborative intelligent control: "global coarse search, gradient sensing, and local fine locking." Its core lies in introducing a dynamic "sensitivity adapter," seamlessly integrating the global exploration capability of genetic algorithms with the local locking capability of proportional-integral-derivative (PID) control. This allows for the simultaneous realization of rapid self-starting, high-precision steady-state maintenance, and strong anti-interference self-healing of mode-locked lasers under practical engineering conditions.

[0008] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: On the one hand, a fiber laser polarization-locking system based on genetic algorithm and proportional-integral-derivative (PID) collaborative control is provided. This system constructs a closed-loop intelligent control architecture of "perception-decision-execution," specifically including: The optical path execution unit, as the controlled object, adopts a ring laser resonator structure containing an electronically controlled polarization controller (EPC) to generate and maintain mode-locked laser pulses based on nonlinear polarization rotation effects; The signal sensing unit, acting as a feedback channel, is coupled to the output of the laser resonant cavity. It is used to acquire the time-domain waveform of the laser pulse at high speed and extract a set of multi-dimensional mode-locked quality feature vectors in real time through spectrum analysis and statistical processing. The collaborative control unit, acting as the control center, connects the two units mentioned above. Its innovation lies in its configuration to execute a multimodal, adaptive evolutionary collaborative control strategy, specifically comprising the following sequentially executed logical stages: Phase 1 (Global Exploration Mode): Upon system power-on initialization or when the system is determined to be completely unlocked, the genetic algorithm is initiated. This algorithm aims to maximize the fitness function constructed from the multidimensional mode-locking quality feature vectors. It performs population evolution and global parallel search within the multidimensional voltage parameter space of the electrically controlled polarizer until it outputs a set of "coarse-tuning" parameters that allows the laser to enter the quasi-steady-state mode-locking range.

[0009] The second stage (sensitivity sensing and parameter adaptation mode): After obtaining the quasi-steady-state operating point, the system pauses random search and instead performs local sensitivity calibration. Using the "coarse-tuning" parameter set obtained in the previous stage as a reference point, small test disturbance signals are sequentially injected into each control voltage channel, simultaneously monitoring the rate of change of the mode-locking quality characteristic vector, thereby calculating online the sensitivity gradient vector or Jacobian matrix of the mode-locking quality relative to each control parameter at the current operating point. Subsequently, based on this gradient information, the system uses a preset nonlinear negative correlation mapping function (e.g.: The system dynamically and adaptively calculates the optimal initial parameters of the PID controller (including proportional gain) that match the sensitivity characteristics of the local region. ), points time ( ) and differential time ( ).

[0010] The third stage (deterministic precision locking mode): After parameter adaptation is completed, the system control smoothly and seamlessly switches to the PID controller initialized with the above-mentioned optimized parameters. In this mode, the system applies a continuous and minute negative feedback adjustment voltage to the electronically controlled polarizer based entirely on the deterministic deviation between the mode-locking quality eigenvector and the preset target value, thereby accurately compensating for slow disturbances caused by ambient temperature drift, mechanical vibration, etc., and achieving long-term ultra-steady-state locking.

[0011] The fourth stage (online self-healing mode): As a guarantee for steady-state locking, the system continuously monitors characteristic parameters. Once a loss of lock due to a sudden strong disturbance is detected, a local backtracking self-healing procedure is immediately triggered: starting from the historically optimal parameters stored just before the loss of lock, a rapid local search is performed in the adjacent parameter space using deterministic methods such as the steepest gradient ascent method, aiming to restore mode lock within milliseconds. Only when local self-healing fails will the system reset and re-trigger the global search of the first stage, thus constructing a two-level robust defense system of "global-local".

[0012] On the other hand, a fiber laser polarization locking method that is completely corresponding to the above system is provided, which specifically includes the following steps: S1. Signal perception and feature extraction: The laser output pulse signal is sampled in real time, and after analog-to-digital conversion, the spectral features are obtained through fast Fourier transform, and the stability features are obtained through time-domain statistical analysis. Together, they form a feature vector reflecting the mode-locking quality. S2. Global Parameter Optimization Based on Genetic Algorithm: Constructing a Multi-Objective Weighted Fitness Function Including the Mode-Locked Quality Feature Vectors from the Above: Global Search and Quasi-Steady-State Capture

[0013] in , , These are the theoretical limits or historical statistical maximum values ​​of each eigenvector, used for data normalization. If the mode-locking quality eigenvector does not meet the mode-locking threshold, a genetic algorithm is run with the goal of maximizing the comprehensive fitness function, iteratively searching the entire space of polarization control parameters, and finally outputting an optimal set of individual parameters to guide the system into the quasi-steady-state mode-locking region. S3. Local Sensitivity Online Calibration and Gradient Extraction: At the quasi-steady-state parameter set, the genetic algorithm is frozen. By applying a set of orthogonal, small-amplitude probe voltages to each control channel and observing the response change of the mode-locking quality evaluation function, the local sensitivity gradient information of the current operating point is calculated; S4. Adaptive tuning of PID controller parameters: Substitute the gradient magnitude obtained in step S3 into the preset nonlinear gain scheduling function to calculate the optimal proportional gain of the PID controller suitable for the local dynamic characteristics of the current operating point in real time, and derive the integral and derivative time constants accordingly. S5. Switch to adaptive PID steady-state feedback lockout: Load the PID parameters tuned in step S4 and switch the control system to classic PID negative feedback closed loop. The controller continuously outputs fine-tuning voltage based on the real-time error between the characteristic vector and the target value, so that the system operates stably in the optimal mode-locked state; S6. Hierarchical Fault Monitoring and Rapid Self-Healing: During steady-state operation, the mode-locking status is monitored in parallel. Once a lock loss is detected, the local gradient search and parameter retuning logic in steps S3-S4 is invoked first to attempt rapid recovery within a small neighborhood of the historical optimal point. If local recovery times out, it is determined to be a global lock loss, and the process returns to step S2 to start a completely new global optimization process.

[0014] Compared with existing technologies, this invention integrates the advantages of random search and deterministic control, and solves the key interface problem of their integration through an original "online sensitivity calibration-parameter adaptation" mechanism, thereby producing the following synergistic gain effects: 1. Achieved the unity of "fast start" and "stable locking" in the time dimension: The genetic algorithm completes global exploration in seconds to find the quasi-steady-state mode-locking interval; then switches to PID control without random noise for long-term precise locking, completely eliminating the phase noise introduced by continuous mutation in the steady-state stage of traditional evolutionary algorithms, thereby improving the signal-to-noise ratio and long-term time-domain stability of the output pulse.

[0015] 2. Intelligent matching between the controller and the complex solution space is achieved in the spatial dimension: By sensing and adapting to the sensitivity differences in different regions of the solution space in real time, the linear PID controller is able to handle highly nonlinear controlled objects. That is, it can automatically decelerate and prevent oscillations in the high-sensitivity "steep slope" region, and accelerate to catch up and eliminate errors in the low-sensitivity "plain" region, thereby greatly expanding the robust operating range of stable mode-locking.

[0016] 3. Achieving a balance between high intelligence and high engineering feasibility at the system level: The entire solution consists of classic and mature algorithm modules, requiring no reliance on big data or black-box models. The algorithm logic is transparent, with low computational load, and can be deployed on low-cost embedded hardware. Combined with a hierarchical self-healing mechanism, the system is not only intelligent but also robust and reliable, truly meeting the stringent requirements of industrial sites for "fully automatic, maintenance-free, and long-life" operation of light sources. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the polarization locking system for a ring cavity fiber laser provided in an embodiment of the present invention.

[0018] Figure 2 This is a flowchart illustrating the genetic algorithm and gradient adaptive PID collaborative control method provided in an embodiment of the present invention.

[0019] Figure 3 This is the output laser spectrum after steady-state locking according to an embodiment of the present invention.

[0020] Figure 4 This is a time-domain waveform diagram of the output pulse after steady-state locking according to an embodiment of the present invention.

[0021] Figure 5 This is a pulse radio frequency spectrum diagram after steady-state locking according to an embodiment of the present invention.

[0022] Figure 6 This is a graph showing the recovery process of the mode-locked characteristic parameters over time after interference is injected in the simulation according to an embodiment of the present invention.

[0023] Figure 7 This is a schematic diagram of the state logic of the collaborative control unit switching between different working modes in an embodiment of the present invention.

[0024] Figure 8 This is a graph showing the variation of the four control voltages of the electrically controlled polarization controller throughout the entire control process in an embodiment of the present invention.

[0025] Figure 9 This is a schematic diagram of the frequency domain characteristic parameters (dominant frequency amplitude A and harmonic suppression ratio H) used to evaluate mode-locking quality in this invention.

[0026] Figure 10 This is a schematic diagram of the time-domain characteristic parameter (pulse amplitude jitter variance S) used to evaluate the mold-locking quality in this invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. This embodiment takes an erbium-doped fiber ring cavity mode-locked laser based on a nonlinear polarization rotation mechanism as an example, but the invention is not limited thereto.

[0028] 1. Hardware System Architecture Construction The fiber laser polarization locking system provided in this embodiment has the following hardware connection relationship: Figure 1 As shown, it specifically includes: (1) Fiber laser resonator The controlled object employs a fiber optic ring cavity structure with a center wavelength of 1550 nm. This resonant cavity, along the optical propagation direction, sequentially comprises: Pump source: A 980nm semiconductor laser is used to provide pump energy for the system.

[0029] Wavelength division multiplexer (WDM): Couples 980nm pump light into the resonant cavity.

[0030] Gain medium: A 3-meter-long erbium-doped fiber (EDF) is used to amplify and lasing optical signals.

[0031] Electrically Controlled Polarization Controller (EPC): A four-channel electrically controlled polarization controller (EPC) is placed inside the cavity. This controller receives four independent 0-5V DC control voltages (V1, V2, V3, V4). Each voltage changes the stress birefringence of the fiber by extruding it, thereby achieving full degree of freedom control of the polarization state within the cavity on the Poincaré sphere.

[0032] Optical isolator (ISO): Ensures unidirectional light transmission and suppresses backscattered light.

[0033] Output Coupler (OC): Couples 10% of the intracavity oscillating laser to the output, while 90% remains in the cavity for circulation.

[0034] (2) Signal acquisition and conversion module Photodetector unit: In this embodiment, a high-speed InGaAs photodetector with a bandwidth of 10GHz is used and connected to the beam splitting output of the output coupler to convert optical pulses into analog electrical signals.

[0035] Analog-to-digital conversion unit: In this embodiment, an ADC chip with a sampling rate of 1GS / s and a resolution of 16 bits is selected to discretize the analog signal and provide a mode-locked quality feature vector for subsequent algorithms.

[0036] (3) Cooperative control unit As the physical carrier of the collaborative control unit described in claim 1, this embodiment uses a high-performance microcontroller (STM32H7 series) based on the ARM Cortex-M7 core.

[0037] The program logic stored and executed internally by the microprocessor is configured to perform the hierarchical control architecture described in claim 1. Specifically, the internal logic of the MCU is divided into a global optimization task, a gradient calculation task, and a PID control task.

[0038] 2. Configuration of Hybrid Adaptive Control Algorithm The core of this invention lies in the "genetic algorithm and gradient-adaptive PID collaborative strategy" running in the collaborative control unit. Its specific parameter configuration and implementation logic are as follows: (1) Signal quality characteristics and threshold settings The collaborative control unit performs Fast Fourier Transform (FFT) and statistical analysis on the digital signal, extracts the following features and sets criteria, and combines them with... Figure 9 and Figure 10 The mode-locking quality feature vector is described in detail below: Dominant frequency amplitude (A): As shown by the double-headed arrow marked A in the figure, it is defined as the height difference between the peak of the main spectral peak and the noise floor level. This indicator characterizes the energy intensity of the main signal in the current operating state of the laser. In this invention, the larger the value of A, the stronger the lasing energy in the cavity, which is the primary criterion for determining whether the laser has started oscillating.

[0039] Harmonic suppression ratio (H): As shown by the double-headed arrow marked H in the figure, it is defined as the intensity difference between the peak of the primary peak and the peak of the secondary side peak. This index characterizes the purity of the mode-locked pulse in the frequency domain. In this invention, a larger H value (i.e., a lower side peak) indicates less multi-pulse splitting or higher-order soliton interference, and higher mode-locking quality.

[0040] As shown in the text description, S is defined as the amplitude deviation of a number of consecutive pulses (e.g., 1024) within the sampling window. The statistical variance or root mean square error of the laser is used to characterize the temporal stability of the laser during long-term operation. In this invention, a smaller S value indicates better amplitude consistency of the pulse sequence, effectively eliminating unstable states such as Q-mode-locked (QML).

[0041] Based on the above definitions, the fitness function F constructed in this invention maximizes A and H and minimizes the weighted combination of S, enabling the control system to automatically find the optimal operating point with the strongest energy, purest spectrum, and most stable time domain, thereby achieving a necessary and sufficient characterization of the steady-state mode-locked state. This embodiment sets the following criteria: Dominant frequency amplitude (A): The first preset threshold is set to 0.8 (normalized value), which is used to characterize whether the laser starts to oscillate.

[0042] Harmonic suppression ratio (H): The second preset threshold is set to 60dB to characterize whether multipulse splitting occurs.

[0043] Pulse time-domain amplitude jitter variance (S): The third preset threshold (pulse peak normalized variance) is set to 0.05 to characterize whether it is in a Q-switching unstable state.

[0044] Fitness function construction: Employing a multi-objective weighting function:

[0045] in , , These are the theoretical limits or historical statistical maximum values ​​of each eigenvector.

[0046] Weighting coefficient configuration principle: This embodiment sets =0.5, =0.3, =0.2. This configuration is based on the physical evolution of laser mode-locking: in the initial search phase, the primary task is to accumulate energy within the cavity to overcome the oscillation threshold, therefore, the energy index A is assigned the highest weight. Secondly, it is necessary to suppress multi-pulse splitting to ensure fundamental frequency mode locking; therefore, the purity index H has a certain weight. Secondly; finally, in the quasi-steady state, we focus on time-domain jitter, hence the stability index S has a weight. As a detailed evaluation item.

[0047] (2) Configuration of global optimizer for genetic algorithm Population size: 50.

[0048] Maximum number of iterations: 100.

[0049] Crossover probability: 0.8; mutation probability: 0.05.

[0050] (3) Local sensitivity calibration and PID parameter adaptation To address the problem that traditional fixed-parameter PID controllers cannot adapt to nonlinear polarization rotation and nonlinear solution spaces, this invention employs a gradient adaptive tuning mechanism. The specific implementation is as follows: Triggering timing: When the genetic algorithm finishes its search and the system enters a quasi-steady state, the system switches to a stable state.

[0051] Calibration action: Optimal voltage vector output by a genetic algorithm. Centered on the four voltage channels, small perturbation voltages are applied sequentially. (e.g., 10mV); Gradient calculation: Real-time measurement of the change in the fitness function after applying a perturbation, and calculation of the sensitivity gradient vector at the current operating point. .

[0052] Parameter generation: The collaborative control unit has a pre-set gain mapping model, which is based on the gradient magnitude. Inverse calculation of the proportional gain of the PID controller To prevent over-modulation oscillations in steep solution space regions (i.e., regions of extremely high sensitivity), this embodiment uses the following nonlinear decay mapping formula for calculation:

[0053] in, Set the system's default base gain (e.g., 1.5). This is the sensitivity inhibition factor (e.g., set to 0.5). This is a nonlinear exponent (usually 1 or 2). This formula ensures that: when the gradient... When I was very young, Maintain at In the vicinity, a rapid response, characterized by "big strides and fast runs," is achieved. This step is a key technical means to overcome the non-uniform solution space characteristics of nonlinear polarization rotation mode-locking.

[0054] (4) Loss of lock determination and recovery mechanism Loss of lock criterion: The main frequency amplitude A drops by more than 10% of the target value and lasts for 10ms.

[0055] Local backtracking procedure: Employs a multidimensional gradient hill-climbing method. The system reads the historical optimal voltage parameters stored before the lock-out and uses these as a starting point to perform a search within a preset local neighborhood. The local neighborhood is specifically defined as a multidimensional voltage space with a radius of centered at the historical optimal voltage point, where the preferred values ​​are the half-wave voltage of the electrically controlled polarizer. 5% to 10% (e.g.) Within this range, the system simultaneously adjusts the four control voltages based on the real-time calculated gradient direction, performing a rapid iterative search along the direction of steepest mode-locking quality improvement. Compared to a global random scan, this limitation... Gradient search within a small neighborhood can compress the lock recovery time from seconds to milliseconds.

[0056] 3. Detailed Explanation of System Workflow like Figure 2 The algorithm flowchart shown illustrates the system's operation in the following steps: Step S1: Power on the laser, turn on the pump source, and the co-control unit begins to collect optical signals.

[0057] Step S2: The microprocessor processes the acquired signal in real time and calculates the mode-locking quality feature vector (A, H, S).

[0058] Step S3: If the target mode-locking state is not reached, start the genetic algorithm to perform a global optimization search on the four voltages of the electronically controlled polarizer in order to maximize the fitness function F.

[0059] Step S4: When the genetic algorithm makes the fitness value converge to the quasi-steady-state mode-locked interval, the microprocessor pauses the random search, executes the local sensitivity calibration program, calculates the gradient characteristics of the current operating point, and generates adaptive PID control parameters accordingly.

[0060] Step S5: Switch to PID controller. Based on the parameters generated in step S4 and the optimal voltage point provided by the genetic algorithm, perform microsecond-level closed-loop fine-tuning of the four voltages according to the real-time error, effectively suppressing phase jitter caused by environmental noise.

[0061] Step S6: During the steady-state operation of the PID controller, continuously monitor the mode-locking status. If a lock loss occurs, prioritize initiating a gradient-based local backtracking procedure for millisecond-level rapid repair; only if the local repair fails will the genetic algorithm be re-triggered for global optimization.

[0062] To verify the actual performance of the "local sensitivity adaptive genetic algorithm and PID cooperative polarization locking system" described in this invention, a system was built in a laboratory environment as follows. Figure 1 The hardware test platform shown was used to conduct comprehensive testing and analysis on the system's steady-state locking characteristics, anti-interference capability, and control voltage evolution process.

[0063] (1) Steady-state locking model behavior analysis (e.g.) Figure 3 , Figure 4 , Figure 5 ) Spectral characteristics ( Figure 3 ):like Figure 3As shown, after the system entered steady-state lock, the output spectrum was measured using a spectrometer. The center wavelength of the spectrum was approximately 1550 nm, and the 3dB bandwidth was approximately 4.5 nm. The spectral shape exhibited a typical and symmetrical hyperbolic secant envelope, with clear and sharp Kelly flanks on both sides of the main peak. This characteristic strongly confirms that the laser is operating in the negative dispersion region and generates high-quality conventional soliton pulses.

[0064] Time-domain pulse sequence ( Figure 4 ):like Figure 4 As shown, the output pulse sequence of the laser was recorded using a high-speed oscilloscope. The waveform displays a series of pulse trains with uniform time intervals and highly consistent pulse amplitudes. Within a long observation window, the normalized variance S of the pulse peak value remained below 0.005, and no envelope fluctuations characteristic of Q-switched mode-locked (QML) were observed, proving that the system achieved true continuous-wave mode-locking.

[0065] Radio frequency spectrum characteristics ( Figure 5 ):like Figure 5 As shown, the fundamental frequency signal of the pulse was measured using an RF spectrum analyzer. The results show that the fundamental frequency peak is sharp, and the background noise is effectively suppressed to below -110 dBm. Calculations show that the rejection ratio (SNR) of this mode-locked pulse is as high as 70 dB or more. This extremely high signal-to-noise ratio indicates that the cooperative control strategy of this invention effectively suppresses phase noise and timing jitter, and its locking stability is superior to that of similar lasers controlled by traditional single algorithms.

[0066] (2) Dynamic disturbance resistance and self-healing ability analysis (e.g.) Figure 6 ) To verify the "local backtracking self-healing function" described in claims 2 and 7, strong physical vibrations were artificially introduced as external disturbances when the system was in steady-state operation (at time axis t=70).

[0067] Response process ( Figure 6 ):like Figure 6 As shown, at the instant of interference injection, the amplitude of the main frequency (black solid line A), which represents energy, drops rapidly, while the jitter variance (gray solid line S), which represents instability, rises sharply, and the system briefly loses lock.

[0068] Self-healing effect: The collaborative control unit immediately triggers a lock-out monitoring interrupt and initiates a local backtracking procedure. Within an extremely short timeframe of t=70 (simulation step number) to t=75, the amplitude A stops falling and rebounds, and S rapidly converges to zero. The system automatically recovers to a high-energy steady-state mode-locked state in only about 5 simulation steps (corresponding to approximately 50ms in actual time), without triggering a time-consuming global cold start. This fully demonstrates the excellent robustness of this system in the face of sudden environmental disturbances.

[0069] (3) Control voltage adaptive evolution analysis (e.g.) Figure 8 ) Figure 8 The system recorded the variation curves of the four control voltages (V1-V4) throughout the entire process, intuitively demonstrating the logic switching of "cooperative control": Phase I (Global Search): In the t=0~65 interval, the voltage exhibits a large step change, indicating that the genetic algorithm is performing a large-scale discrete jump search in the solution space to find the quasi-steady-state mode-locked interval.

[0070] Phase II (Sensitivity Calibration and Switching): Around t=65, the voltage exhibits slight quadrature jitter (calibration action), followed by a smooth transition.

[0071] Phase III (PID steady-state lockout): After t > 70 (simulation steps), the four voltages enter continuous fine-tuning mode, and the curves exhibit slow and smooth changes following environmental drift. This voltage evolution trajectory is a direct manifestation of the coarse-tuning + fine-tuning described in this invention.

[0072] The method and system described in this invention overcome the nonlinearity problem of the polarization solution space of fiber lasers through the above-mentioned hardware and software co-design and the use of gradient adaptive mechanism, realizing a seamless connection from global coarse adjustment to local fine adjustment, and significantly improving the self-starting speed and long-term operational robustness of mode-locked lasers.

Claims

1. An adaptive polarization locking system for a fiber laser, characterized in that, include: An optical path execution unit is used to generate laser pulses based on nonlinear polarization rotation effects; a signal sensing unit is used to acquire the time-domain signal of the laser pulse in real time and convert it into a mode-locked quality feature vector; a cooperative control unit is connected to the optical path execution unit and the signal sensing unit respectively. The collaborative control unit is configured as follows: In the start-up or unlocked state, the genetic algorithm is run to drive the optical path execution unit to perform a global search in the polarization control parameter space until the mode-locking quality determined based on the mode-locking quality feature vector meets the preset conditions, and the first control parameter set at this time is recorded. Using the first set of control parameters as a reference, a local sensitivity calibration operation is performed to obtain the gradient information of the mode-locking quality relative to the control parameters; Based on the gradient information, the initial control parameters of the proportional-integral-derivative controller are adaptively generated. Switch to the proportional-integral-derivative control mode based on the initial control parameters, and perform negative feedback fine-tuning on the optical path execution unit according to the deviation between the mode-locking quality feature vector and the target value.

2. The system according to claim 1, characterized in that, The cooperative control unit is configured to perform the local sensitivity calibration operation in the following manner: based on the first set of control parameters, it sequentially applies independent test perturbations to each control channel of the optical path execution unit, and calculates the partial derivative or gradient magnitude of the mode-locking quality evaluation function with respect to the parameters of each control channel based on the change of the mode-locking quality feature vector before and after applying the test perturbations.

3. The system according to claim 2, characterized in that, The initial control parameters of the adaptive proportional-integral-derivative (PID) controller include: calculating the proportional gain coefficient of the PID controller through a nonlinear mapping relationship based on the gradient magnitude.

4. The system according to claim 3, characterized in that, The nonlinear mapping relationship satisfies the following formula: in, For adaptive proportional gain, As the reference gain, The local sensitivity gradient magnitude output by the sensitivity calibration module. As a sensitivity inhibitor, It is a non-linear decay exponent.

5. The system according to claim 1, characterized in that, The collaborative control unit is also configured to: Under the proportional-integral-derivative control mode, the mode-locking quality feature vector is continuously monitored; When a loss of lock is detected, the historical optimal control parameters recorded in the memory are read, and a gradient-direction-based search is performed in a preset neighborhood centered on the historical optimal control parameters to restore the lock. If the model lock is not restored within a preset number of attempts or within a preset time, the genetic algorithm will be retried.

6. The system according to any one of claims 1 to 5, characterized in that, The mode-locked quality feature vector includes at least one of the following features: spectral main frequency amplitude (A), harmonic suppression ratio (H), and pulse time-domain amplitude jitter variance (S).

7. The system according to any one of claims 1 to 5, characterized in that, The optical path execution unit includes a ring cavity structure and an electrically controlled polarization controller (EPC) disposed therein; the electrically controlled polarization controller (EPC) is a squeeze-type fiber polarization controller or a waveplate polarization controller.

8. The system according to any one of claims 1 to 5, characterized in that, The collaborative control unit is implemented based on a field-programmable gate array, a digital signal processor, or an embedded microprocessor.

9. An adaptive polarization locking method for a fiber laser, characterized in that, Includes the following steps: Step S1: Real-time acquisition of the laser pulse signal output by the fiber laser and extraction of the mode-locking quality feature vector; Step S2: If the mode-locking quality feature vector does not meet the mode-locking condition, run the genetic algorithm, adjust the polarization control parameters to perform a global search until it enters the quasi-steady-state mode-locking interval, and obtain the corresponding first set of control parameters; Step S3: Using the first set of control parameters as a reference, perform local sensitivity calibration to obtain the gradient information of mode-locking quality with respect to control parameters; Step S4: Adaptively tune the initial control parameters of the proportional-integral-derivative controller based on the gradient information; Step S5: Switch to the proportional-integral-derivative control mode based on the initial control parameters after tuning, and perform negative feedback fine-tuning according to the deviation between the mode-locking quality feature vector and the target value.

10. The method according to claim 9, characterized in that, Step S3 includes: Based on the first set of control parameters, independent test perturbations are applied sequentially to each control channel. The gradient information of the mode-lock quality evaluation function is calculated based on the changes in the mode-lock quality feature vector before and after the application of the test perturbation.

11. The method according to claim 10, characterized in that, Step S4 includes: Based on the magnitude of the gradient information, the proportional gain coefficient of the proportional-integral-derivative controller is calculated through a nonlinear mapping relationship.

12. The method according to claim 11, characterized in that, The nonlinear mapping relationship satisfies the following formula: in, For adaptive proportional gain, As the reference gain, The local sensitivity gradient magnitude output by the sensitivity calibration module. As a sensitivity inhibitor, It is a non-linear decay exponent.

13. The method according to claim 9, characterized in that, Following step S5, step S6 is also included: Under the proportional-integral-derivative control mode, the mode-locking quality feature vector is continuously monitored; When a loss of lock is detected, the stored historical optimal control parameters are read, and a gradient-direction-based search is performed within a preset neighborhood centered on the historical optimal control parameters to restore lock mode. If the lock mode is not restored within the preset number of attempts or within the preset time, step S2 will be triggered again.

14. The method according to any one of claims 9 to 13, characterized in that, The mode-locked quality feature vector includes at least one of the following features: spectral main frequency amplitude (A), harmonic suppression ratio (H), and pulse time-domain amplitude jitter variance (S).

Images