Micromirror control method and system based on intelligent driving circuit

Abstract

The invention discloses a micro-mirror control method and system based on an intelligent driving circuit. The method comprises the following steps: acquiring target pose data and micro-mirror motion characteristics and generating a smooth track; measuring an amplification characteristic of the amplification chain to generate a steady / unsteady identity; stopping sampling and consistency calculation in an unsteady state, turning to a steady state, and opening a stable sampling window to collect a quality sample after a track is terminated; and based on track and attitude optical model prediction and sample consistency verification, a correction instruction is completed or issued according to a verification result, and soft degradation is carried out when a threshold value is reached. The technical scheme provided by the embodiment of the invention not only improves the accuracy of judgment and the reliability of data, but also reduces frequent compensation, jitter or misoperation caused by misjudgment, so that the action of the micromirror in a complex multi-link environment is more controllable and more stable.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optical communication technology, and in particular to a micro-mirror control method and system based on an intelligent driving circuit. BACKGROUND

[0002] In the application of modern optical links and high-precision actuators, micro reflective elements are widely used to complete the rapid switching, scanning or modulation of light beams. Since such elements are usually coupled with external optical signal transmission devices, their working state is not only controlled by their own driving circuit, but also affected by multiple links such as optical amplification and regeneration. In a complex working environment, the power, signal quality and dynamic response in the link often exhibit multi-time scale and multi-state characteristics, which brings uncertainty and delay problems to the control and verification of the actuator.

[0003] Traditional control strategies often rely on real-time feedback to directly drive the actuator to act and make state judgments. However, when there are transient fluctuations or feedback lags, the signals obtained by direct collection and calculation may be mixed with unstable components, leading to unreliable judgment and adjustment, and even frequent compensation, oscillation or misoperation. Therefore, in a complex optical link and multi-link feedback environment, there is an urgent need for a solution that can still guarantee the reliability of control and judgment under feedback delay and transient fluctuations, in order to improve the stability and robustness of micro optical actuators in dynamic scenarios. SUMMARY

[0004] The present application provides a micro-mirror control method and system based on an intelligent driving circuit, which can effectively improve the stability and robustness of micro optical actuators in dynamic scenarios.

[0005] According to an aspect of the present application, a micro-mirror control method based on an intelligent driving circuit is provided, comprising:

[0006] acquiring target pose data; measuring the motion characteristics of the micro-mirror to obtain first measurement data, the first measurement data being used to represent the current initial state of the micro-mirror, and generating a smooth trajectory based on the target pose data and the first measurement data; measuring the amplification characteristics of the optical signal through the amplification chain to obtain second measurement data, and generating an amplification chain state identifier based on the second measurement data, the amplification chain state identifier including a steady state identifier and a non-steady state identifier;

[0007] in response to the amplification chain state identifier being a non-steady state identifier, not starting a stable sampling window and not performing consistency calculation until the amplification chain state identifier changes to a steady state identifier; in response to the amplification chain state identifier being a steady state identifier and the smooth trajectory reaching a termination condition, starting a stable sampling window, obtaining quality measurement results in the stable sampling window, and selecting target sub-window samples;

[0008] calculating a first predicted value based on the smooth trajectory and a model of pose to optical response, generating a first check value based on the target sub-window sample, and calculating a consistency result based on the first predicted value and the first check value;

[0009] in response to the consistency result satisfying a consistency condition, generating a completion flag; in response to the consistency result not satisfying the consistency condition, generating a correction instruction and executing the correction instruction; and generating a soft degradation instruction when a correction number reaches a preset threshold.

[0010] Optionally, the generating a smooth trajectory based on the target pose data and the first measurement data comprises:

[0011] constructing an initial trajectory with limited acceleration change rate based on the target pose data and the first measurement data;

[0012] generating a power change trend indication based on the target pose data and the first measurement data through a mapping of pose to optical response, and determining a change rate limit parameter and a terminal approaching speed based on the power change trend indication; the power change trend indication comprises slope and inflection point information;

[0013] performing a slope limit reconstruction on a tail segment of the initial trajectory based on the change rate limit parameter and the terminal approaching speed, and setting an alignment segment such that a tail segment arrival time corresponds to a start condition of a stable sampling window, to obtain the smooth trajectory.

[0014] Optionally, in response to the amplifier chain state identifier being a non-stable state identifier, the method further comprises:

[0015] stopping execution of the smooth trajectory, latching a current pose, and applying a change rate limit to a remaining segment of the smooth trajectory until the amplifier chain state identifier changes to the stable state identifier.

[0016] Optionally, the generating a power change trend indication comprises:

[0017] combining the target pose data and the first measurement data in time sequence to form a pose sampling sequence, and selecting a candidate dictionary entry matching the pose sampling sequence in a preset response dictionary;

[0018] obtaining a mapping coefficient on the candidate dictionary entry by sparse constraint regression, and applying a non-negative constraint and a monotonicity constraint to obtain a local mapping of pose to optical response;

[0019] mapping and time sequence smoothing the pose sampling sequence based on the local mapping to generate a power change trend indication.

[0020] Optionally, the obtaining a local mapping of pose to optical response comprises:

[0021] constructing a sparse regression problem with outlier suppression term, solving a first coefficient set satisfying non-negative constraint for the candidate dictionary entries;

[0022] imposing support preserving constraint and total variation constraint on the first coefficient set based on time sequence of the pose sampling sequence, obtaining a second coefficient set satisfying time consistency;

[0023] grouping the candidate dictionary entries and setting group sparse weight based on the second coefficient set and according to the amplifier chain state identifier, and in the solving process, preferentially selecting the dictionary entries corresponding to steady state identifier, obtaining a third coefficient set;

[0024] defining the local mapping by dictionary weighted combination corresponding to the third coefficient set, imposing monotonicity constraint on the local mapping with respect to pose variable and slope limiting given by the change rate limiting parameter, and outputting the local mapping.

[0025] Optionally, the determining of the change rate limiting parameter and the terminal approaching speed comprises:

[0026] generating a transient sensitivity index representing the sensitivity degree of the amplifier chain to incident power change based on the slope and inflection point information and the amplifier chain state identifier;

[0027] constructing a power change safety boundary based on the transient sensitivity index and the starting condition of the stable sampling window, and generating a parameter set including candidate change rate limiting parameter and candidate terminal approaching speed according to the power change safety boundary;

[0028] performing transient prediction on the parameter set based on the model of the pose to optical response and calculating power change envelope, and selecting a target change rate limiting parameter and a target terminal approaching speed that make the power change envelope satisfy the power change safety boundary and the ending time consistent with the starting condition of the stable sampling window.

[0029] Optionally, the generating of the amplifier chain state identifier based on the second measurement data comprises:

[0030] calculating gain estimate, output power change rate, short window variance and kurtosis of output power, and noise sideband energy ratio from the second measurement data, forming a state discrimination feature sequence;

[0031] matching the state discrimination feature sequence based on a pre-set transient response template library, and performing sequential update using a semi-Markov state discriminator with residence time constraint, obtaining a steady state confidence sequence;

[0032] Based on the steady-state confidence sequence and the candidate stable interval detected by the change point detection, a minimum residence time test, a variance upper bound test and a monotonic convergence test are performed, a steady-state identifier is generated and a steady-state timestamp is recorded when the test is satisfied, and a non-steady-state identifier is generated when the test is not satisfied.

[0033] Optionally, the obtaining of the steady-state confidence sequence comprises:

[0034] Time scale normalization and multi-resolution shape coding are performed on the state discrimination feature sequence, the shape coding comprises wavelet scattering coefficients and Hankel subspace angles, and a coding vector is matched with each template in the transient response template library through dynamic time warping and scale-invariant regularization to obtain a template likelihood vector sequence;

[0035] A hidden semi-Markov discriminator with residence time constraints is constructed with the template likelihood vector sequence as an observation, a non-parametric kernel density is used for the residence time distribution, minimum residence time constraints and monotonic hazard rate constraints are imposed, and a state posterior probability sequence is obtained through forward-backward sequential updating;

[0036] Optimal transport regularization is introduced in the sequential updating to minimize the observation distribution distance of adjacent time slices and impose a transition sparsity penalty to suppress high-frequency state switching, and the updated state posterior probability sequence is taken as the steady-state confidence sequence.

[0037] Optionally, the calculating of the first prediction value based on the smooth trajectory and the posture-to-optical response model, the generating of the first check value based on the target sub-window sample, and the calculating of the consistency result based on the first prediction value and the first check value comprise:

[0038] The smooth trajectory is mapped to an observable feature space, and a prediction sequence with uncertainty is established based on the posture-to-optical response model, a Koopman operator is used to obtain a first prediction value and its confidence interval aligned with the stable sampling window;

[0039] Robust denoising and abnormal clipping are performed on the target sub-window sample, and the sample is projected to the feature space to generate a first check value and its variance estimate;

[0040] An optimal transport divergence with entropy regularization is calculated based on the prediction distribution of the first prediction value and the empirical distribution of the target sub-window sample, and a consistency score is obtained by combining the confidence interval overlap rate, and the consistency score is compared with a threshold corresponding to the consistency condition to form the consistency result.

[0041] According to another aspect of the present application, a micro-mirror control system based on an intelligent driving circuit is provided, comprising:

[0042] The collection module is configured to acquire target pose data, measure motion characteristics of the micro mirror to obtain first measurement data, wherein the first measurement data is used to represent a current initial state of the micro mirror, generate a smooth trajectory based on the target pose data and the first measurement data, measure amplification characteristics of an optical signal through an amplification chain to obtain second measurement data, and generate an amplification chain state identifier based on the second measurement data, wherein the amplification chain state identifier includes a steady state identifier and a non-steady state identifier.

[0043] The judgment module is configured to, in response to the amplification chain state identifier being the non-steady state identifier, not start a stable sampling window and not perform consistency calculation until the amplification chain state identifier changes to the steady state identifier, and in response to the amplification chain state identifier being the steady state identifier and the smooth trajectory reaching a termination condition, start the stable sampling window, acquire a quality measurement result in the stable sampling window, and select a target sub-window sample.

[0044] The processing module is configured to calculate a first predicted value based on the smooth trajectory and a model of an attitude to an optical response, generate a first check value based on the target sub-window sample, and calculate a consistency result based on the first predicted value and the first check value.

[0045] The generation module is configured to, in response to the consistency result satisfying a consistency condition, generate a completion flag, in response to the consistency result not satisfying the consistency condition, generate a correction instruction and execute the correction instruction, and when a correction number reaches a preset threshold, generate a soft degradation instruction.

[0046] The embodiment of the present application provides a kind of micro mirror control method and system based on intelligent drive circuit.The method comprises: obtaining target pose data;Measure the motion characteristics of micro mirror, obtain first measurement data, first measurement data is used to characterize the current initial state of micro mirror, and generate smooth trajectory based on target pose data and first measurement data;Measure the amplification characteristics of optical signal through amplification chain, obtain second measurement data, and generate amplification chain state identifier based on second measurement data, amplification chain state identifier includes steady-state identifier and non-steady-state identifier;In response to the amplification chain state identifier being non-steady-state identifier, do not open stable sampling window and do not perform consistency calculation until the amplification chain state identifier changes to steady-state identifier;In response to the amplification chain state identifier being steady-state identifier and the smooth trajectory reaching termination condition, open stable sampling window, obtain quality measurement result in stable sampling window, and select target sub-window sample;Calculate first predicted value based on smooth trajectory and model of pose to optical response, generate first check value based on target sub-window sample, and calculate consistency result based on first predicted value and first check value;In response to the consistency result meeting consistency condition, generate completion flag;In response to the consistency result not meeting consistency condition, generate correction instruction and execute correction instruction;When the correction number reaches the preset threshold, generate soft degradation instruction.The technical scheme provided by the embodiment of the present application compared with prior art, the embodiment of the present application establishes more close dynamic association between actuator and optical link.Through introducing state determination, data screening and model-based checking mechanism in control process, it can avoid unstable data directly entering control and checking link under the condition that optical link is in dynamic fluctuation or feedback lag.This not only improves the accuracy of determination and the reliability of data, but also reduces the frequent compensation, jitter or misoperation caused by misjudgment, so that the action of micro mirror is more controllable and stable in complex multi-link environment.

[0047] The technical scheme provided by the embodiment of the present application also provides adaptive correction and safety degradation capability under the condition of feedback exception or continuous fluctuation, so that the micro mirror control can still maintain a predictable and manageable state when facing extreme working conditions or long-term transient state, avoiding the problems of dead loop, excessive regulation and control failure in traditional schemes.This cooperative application of dynamic state recognition, reliable sample extraction, model prediction verification and limited correction significantly improves the robustness and fault tolerance of micro mirror control and optical link interaction, thereby improving the overall performance and stability of the system in actual operation.

[0048] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present application, nor is it used to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to make the technical solution in the embodiments of the present application clearer, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0050] Figure 1 A flow chart of a micro-mirror control method based on an intelligent driving circuit provided by the embodiment of the present application is shown in the figure.

[0051] Figure 2 A flow chart of a smooth trajectory generation method provided by the embodiment of the present application is shown in the figure.

[0052] Figure 3 A flow chart of a power change trend indication method provided by the embodiment of the present application is shown in the figure.

[0053] Figure 4 A structural schematic diagram of a micro-mirror control system based on an intelligent driving circuit provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0054] In order to make the technical solution in the embodiments of the present application clearer, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0055] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily have to include only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product or device.

[0056] Figure 1A flow chart of a micro-mirror control method based on an intelligent driving circuit is provided for an embodiment of the present application. The embodiment can be applied in a complex optical link and multi-link feedback environment, and can guarantee control and decision reliability under feedback delay and transient fluctuation conditions. The method can be executed by a micro-mirror control system based on an intelligent driving circuit. The system can be implemented in the form of hardware and / or software, and can be configured in any electronic device with communication function. Figure 1 The method comprises the following steps.

[0057] In S101, target pose data is acquired. The motion characteristics of the micro-mirror are measured to obtain first measurement data. The first measurement data is used to represent the current initial state of the micro-mirror. A smooth trajectory is generated based on the target pose data and the first measurement data. The amplification characteristics of the optical signal through the amplification chain are measured to obtain second measurement data. An amplification chain state identifier is generated based on the second measurement data. The amplification chain state identifier includes a steady state identifier and a non-steady state identifier.

[0058] In a specific implementation, the method can be executed on some hardware units, such as a processing unit for executing the method, an intelligent driving circuit electrically connected to the processing unit, a motion characteristic measurement channel coupled to the micro-mirror mechanical structure, and a monitoring channel for acquiring optical amplification chain related measurement quantities. The processing unit time-stamps the sampling results of the above channels, and aligns and caches each item of data with a local unified time base.

[0059] In a specific implementation, the target pose data is first acquired. The target pose data is provided by an upper control logic or triggered to be generated by a preset task table at a specified time. The data structure at least includes the angle setting value of the target axis, and for a two-dimensional micro-mirror, can also include the angle setting values of two orthogonal rotation axes. The target pose data is identified with a task number and a time label. The processing unit can perform validity check after receiving the target pose data and write it into a to-be-executed queue.

[0060] In an embodiment, the motion characteristics of the micro-mirror are measured to obtain the first measurement data. The measurement of the motion characteristics can be realized by at least one of a capacitive angle sensor, a Hall position sensor, an optical encoder, or a reflective optical readout unit. The processing unit acquires the original readings at a sampling rate not lower than the control loop frequency, and obtains a parameter set for representing the current state of the micro-mirror through digital filtering and robust statistical operation. The parameter set at least includes the current angle and the current angular velocity. In order to reduce the influence of instantaneous disturbance on the initial condition, the processing unit can aggregate a plurality of sampling results within a short time window before executing the trajectory planning, determine the current initial state by using the method of median and quantile scale estimation, and complete the calibration of the zero point and the proportional coefficient in combination with a temperature compensation table.

[0061] In an embodiment, the processing unit generates a smooth trajectory based on the target pose data and the first measurement data. The generation of the smooth trajectory follows bounded constraints on velocity, acceleration and jerk to avoid exciting mechanical resonances and to guarantee continuity of the driving current / voltage variations. A S-curve planning with jerk constraints or an interpolation planning based on multi-segment time stamping can be adopted: the processing unit takes the current initial state as the starting point and the target pose as the end point, and calculates a sequence of discrete sampling points according to the constraints, which can be directly executed by the smart driving circuit sample by sample. To facilitate task scheduling, the smooth trajectory carries timing information such as start time and estimated end time for subsequent coordination with the optical side sampling actions.

[0062] In an embodiment, the amplification characteristics of the optical signal amplification chain are measured to obtain the second measurement data. The amplification characteristics can include at least one of output optical power, input optical power, a gain approximation derived from the ratio of the two, a pump indicator related to the gain, or a sideband indicator used to characterize the noise level; the monitoring channel reports the above quantities to the processing unit at a fixed period or in an event-triggered manner. The processing unit performs time alignment and windowed statistics on the second measurement data to reduce the uncertainty caused by instantaneous reading jitter.

[0063] In an embodiment, the processing unit generates an amplification chain state identification based on the second measurement data. To ensure the objectivity of the judgment, the state identification is at least distinguished into two categories: steady state identification and non-steady state identification. The processing unit calculates the absolute value of the output optical power change rate and the short window variance within a sliding time window, and compares the above statistical quantities with preset thresholds: when the change rate is lower than the threshold and the short window variance is lower than the threshold, and the time that this state is continuously maintained is not less than the minimum residence time, the output steady state identification is output and the transition to steady state time label is recorded; when any condition is not met, the non-steady state identification is output. The thresholds and residence time can be adjusted according to the device model, the cascade structure of the amplification chain and the environmental conditions, and the setting principle is to exclude transient fluctuations without delaying the normal steady state judgment under typical working conditions.

[0064] In the above embodiments, the acquisition method of the target pose data, the specific sensing path of the motion characteristics, and the specific measurement quantity of the amplification characteristics are not limited to the foregoing examples. Those skilled in the art can select alternative implementation methods according to the specific device configuration and interface conditions, as long as they can form input data that can be aligned and analyzed by the processing unit, and complete the generation of the smooth trajectory and the output of the state identification, which falls within the scope of the present application.

[0065] S102, in response to the amplifier chain state identifier being the non-stable identifier, not starting the stable sampling window and not performing the consistency calculation until the amplifier chain state identifier changes to the stable identifier; in response to the amplifier chain state identifier being the stable identifier and the smooth trajectory reaching the termination condition, starting the stable sampling window, obtaining the quality measurement result in the stable sampling window, and selecting the target sub-window sample.

[0066] In a specific implementation, the processing unit continuously receives and updates the amplifier chain state identifier. When the state identifier is non-stable, the processing unit enters a gate mode waiting for stability: in this mode, the creation of the stable sampling window is not triggered, the consistency calculation task is not scheduled, and the related tasks queued are paused to the waiting queue; at the same time, the time tag of the start of the non-stable state is recorded, and the state identifier is updated according to the set polling period or event subscription mode. When it is detected that the state identifier has been continuously kept stable and reaches the preset minimum residence time, the processing unit determines that the amplifier chain has reached the sampleable condition.

[0067] Further, the processing unit determines whether the smooth trajectory reaches the termination condition. Exemplarily, the termination condition can include: the trajectory index reaches the last sampling point, the absolute values of the last segment velocity and the last segment acceleration are not higher than the preset convergence threshold, respectively, or the time completion flag is triggered. The above conditions can be used alone or in combination; when used in combination, the time completion flag is given priority, followed by the last sampling point determination, and then the convergence threshold determination. If any condition cannot be met within the limited time limit, the processing unit executes an abnormal bottom-up strategy, records the exception and turns into a safe waiting state to avoid a race condition with the amplifier chain state.

[0068] In an embodiment, the processing unit starts the stable sampling window when both the amplifier chain is stable and the smooth trajectory reaches the termination condition. The starting time of the stable sampling window is aligned with the later time when the above two conditions are met, the window duration is set to the interval of milliseconds to tens of milliseconds according to the device and link characteristics, and the sampling frequency in the window is not lower than the update frequency of the quality measurement quantity. The quality measurement quantity can include at least one of the output optical power, the error code statistical quantity, and the signal-to-noise index; to ensure the timing consistency, the processing unit adds a time stamp to the collected sample and organizes it in the local buffer according to the time sequence.

[0069] In an embodiment, the processing unit performs selection of target sub-window samples within the stable sampling window. To this end, a number of candidate sub-windows of equal or variable length are generated, and a stability index and a continuity index are calculated for each candidate sub-window; the stability index is used to constrain the fluctuation amplitude and change rate within the sub-window, and the continuity index is used to constrain the proportion of abnormal points and data missing. The processing unit preferentially selects a candidate sub-window that meets both the stability and continuity indexes, and if there are multiple candidates, preferentially selects a candidate that is close to the tail end of the window and has a longer residence time; if there is no candidate that meets the conditions, the current window is marked as undeterminable, and the processing unit can extend the waiting or restart the window. Once the target sub-window samples are determined, they are used as data input for the subsequent verification stage and the corresponding time labels are retained.

[0070] Further, when the stable sampling window is opened, if the amplifier chain state identifier reverts to a non-stable state, or the quality measurement quantity exhibits a significant abnormal jump, the processing unit immediately terminates the current window and records the event, and enters the stable state gating mode; the window is restarted after the next stable state and termination condition is met. To reduce resource occupation, the processing unit can cancel or suspend the unfinished consistency calculation task, and reschedule it after the condition is restored.

[0071] In this way, through the synergy of gating, alignment and screening, the data participating in the subsequent verification has a clear source, time definition and stability guarantee; at the same time, through abnormal and boundary processing, it is ensured that the process can still be predictable and the record can still be traceable under the condition of state fluctuation and data abnormality, thereby providing reliable input for the subsequent consistency calculation and correction process.

[0072] S103, calculate a first predicted value based on the smooth trajectory and the attitude-to-optical response model, generate a first verification value based on the target sub-window samples, and calculate a consistency result based on the first predicted value and the first verification value.

[0073] In an embodiment, after obtaining the target sub-window samples within the stable sampling window, the processing unit performs consistency calculation. Specifically, the processing unit first obtains a predicted optical response value for subsequent consistency determination based on the pre-constructed and stored attitude-to-optical response model and the determined smooth trajectory.

[0074] In the embodiment, the model of pose to optical response can be established by historical experiments or calibration data, and the implementation includes: first, taking the light signal response characteristics of the micro-mirror at different poses as input, and recording the pose data and the corresponding optical power response data by accumulating historical data in the laboratory calibration or actual operation; second, constructing the mapping relationship between the micro-mirror pose and the optical response by regression analysis or interpolation fitting to obtain a numerical prediction table or a parameterized function of the optical response; the processing unit discretizes the smooth trajectory into a plurality of continuous pose sampling points, and performs model prediction on each sampling point to obtain a series of predicted optical power values corresponding to the pose sequence, forming a first prediction value sequence.

[0075] In addition, to improve the prediction reliability, the model can further provide uncertainty information of the prediction value, for example, representing the prediction fluctuation range by statistical variance or percentage position confidence interval of the prediction history.

[0076] In an embodiment, the target sub-window samples collected by the processing unit in the stable sampling window include at least one of the output optical power, the error code statistics, or the signal-to-noise indicator. The processing unit can pre-process the original sample data in the window by removing outliers and robust denoising, for example, performing median filtering on abnormal measurement points; then calculate the check value of the measurement data in the window by a robust statistical method such as median, truncated mean, etc., to obtain a representative value as the first check value; to be compatible with consistency calculation, the processing unit can also calculate the variance or confidence interval of the first check value to reflect the fluctuation degree of the actual measurement data.

[0077] In an embodiment, the processing unit subsequently performs consistency calculation on the first prediction value and the first check value. To ensure the comparability between the two data, the processing unit first aligns the prediction value and the check value in time sequence according to the unified time stamp to obtain a value pair with the same time reference; second, the processing unit quantifies the consistency degree of the two by calculating the difference between the prediction value and the check value, for example, calculating the absolute difference or relative error between the two; then, compare the difference degree with the pre-set tolerance threshold, if the calculated difference degree is less than or equal to the pre-set tolerance threshold, it is determined that the consistency is satisfied; otherwise, it is determined that the consistency is not satisfied; to clearly represent the determination result, the processing unit can output a binary consistency result, such as "satisfied" or "not satisfied", or output a continuous score value or percentage to reflect the consistency degree of the current prediction and check; the specific setting of the tolerance threshold can be adjusted according to the accuracy index and the running environment requirements of the actual device.

[0078] In this way, the reliability of the micro-mirror control action determination can be ensured under the condition of feedback lag and transient fluctuation, and reliable data basis is provided for subsequent action confirmation and execution.

[0079] S104, in response to the consistency result satisfying the consistency condition, generating a completion flag; in response to the consistency result not satisfying the consistency condition, generating a correction instruction and executing the correction instruction; when the number of corrections reaches a preset threshold, generating a soft degradation instruction.

[0080] In a specific implementation, after completing the consistency calculation, the processing unit executes subsequent action logic according to the calculated consistency result. When the consistency result satisfies a preset consistency condition, the processing unit generates a completion flag to indicate that the target control action of the micro mirror has been successfully completed.

[0081] Specifically, the processing unit compares the consistency result with a tolerance threshold or an allowable error range stored in advance. If the error amplitude, difference value or consistency score represented by the consistency result satisfies the condition, the processing unit writes a task completion flag or status code into an internal state register or a task control table, and can further notify the upper control unit or the related management system through the interface unit that the current control task has been completed, so that the upper system schedules subsequent actions or releases resources accordingly.

[0082] When the consistency result does not satisfy the preset consistency condition, the processing unit generates a corresponding correction instruction and executes the correction. The specific generation method of the correction instruction includes but is not limited to: the processing unit determines the correction direction and correction amplitude of the micro mirror according to the difference or error direction between the predicted value and the verification value; for example, when the measured verification value is significantly higher than the predicted value, the processing unit can determine the direction of reducing the attitude angle of the micro mirror; on the contrary, when the verification value is significantly lower than the predicted value, the direction of increasing the attitude angle is determined; the specific value of the correction amplitude can be determined according to a pre-set fixed step or in proportion to the measurement error. Subsequently, the processing unit transmits the correction instruction to the intelligent drive circuit for execution, and the intelligent drive circuit adjusts the driving voltage or current according to the received correction instruction, so as to drive the micro mirror to perform fine tuning action to reduce the difference between the predicted value and the verification value. After the fine tuning action is performed, the processing unit re-evaluates the attitude and the measurement quantity to prepare for the next round of consistency calculation.

[0083] The processing unit records the current number of corrections in the memory or the register after each implementation of the correction action. When the number of corrections reaches a pre-set threshold, for example, a reasonable value between 3 and 10 times, the processing unit determines that the current action cannot achieve the ideal consistency target through a limited number of corrections, and then triggers and generates a soft degradation instruction.

[0084] For example, the specific implementation of the soft degradation instruction includes but is not limited to: the processing unit sends an instruction to the intelligent driving circuit, so that the micro-mirror returns to a preset safe or default position immediately to avoid further adjustment induced oscillation or control instability; at the same time, the processing unit can send a state abnormality notification to the upper control unit through the interface unit, indicating that the current task has control difficulties, and record the soft degradation event in the log file or state management module for subsequent analysis and disposal by maintenance personnel.

[0085] In actual operation, the processing unit can set additional exception and boundary processing mechanisms, for example, when the state identifier of the amplification chain returns to the non-steady state again during the execution of the correction process, or the posture of the micro-mirror appears an abnormal condition that cannot act according to the instruction, the processing unit should suspend the current correction action in time, record the abnormal event and perform corresponding safety degradation measures or alarm notification, to ensure the safety and robustness of the control flow.

[0086] In this way, in the complex scene of feedback lag and transient fluctuation, adaptive control adjustment and stable completion determination can be implemented, and a soft degradation strategy and an abnormal safety mechanism are provided to effectively prevent control error from continuously expanding and waste of resources, and the reliability and safety of the micro-mirror control action are significantly improved.

[0087] Compared with the prior art, the technical scheme provided by the embodiment of the present application establishes a closer dynamic association between the actuator and the optical link. By introducing the state determination, data filtering and model-based verification mechanism in the control flow, unstable data can be avoided from directly entering the control and verification link when the optical link is in dynamic fluctuation or feedback lag. In this way, the accuracy of the determination and the reliability of the data are improved, and the frequent compensation, jitter or misoperation caused by misjudgment are reduced, so that the micro-mirror action is more controllable and stable in a complex multi-link environment. The technical scheme provided by the embodiment of the present application also provides adaptive correction and safety degradation capability in the case of feedback abnormality or continuous fluctuation, so that the micro-mirror control can still maintain a predictable and manageable state when facing extreme working conditions or long-term transients, avoiding the problems of dead loop, excessive adjustment and control failure in the traditional scheme. The cooperative application of dynamic state recognition, reliable sample extraction, model prediction verification and limited correction significantly improves the robustness and fault tolerance of the micro-mirror control and optical link interaction, thereby improving the overall performance and stability of the system in actual operation.

[0088] Optionally, Figure 2 A flowchart of a method for generating a smooth trajectory provided by the embodiment of the present application is further refined on the basis of the above-mentioned embodiment. Referring to Figure 2 The method comprises the following steps:

[0089] S201: constructing an initial trajectory with limited acceleration change rate based on target pose data and first measurement data.

[0090] S202: generating a power change trend indication through mapping of pose to optical response based on target pose data and first measurement data, and determining a change rate limit parameter and terminal approaching speed based on the power change trend indication; the power change trend indication includes slope and inflection point information.

[0091] S203: performing slope limit reconstruction on the end segment of the initial trajectory based on the change rate limit parameter and the terminal approaching speed, and setting an alignment segment so that the end segment reaches the starting condition of the stable sampling window, to obtain a smooth trajectory.

[0092] In the terminal approaching process of the above micro-mirror motion trajectory, a slight change in the motion pose of the micro-mirror will cause a significant transient disturbance to the automatic gain control (AGC) or automatic power control (APC) loop of the downstream optical amplification chain, which may cause a temporary unstable state of the optical signal. This condition not only prolongs the waiting time for system stability, but also may cause uncertainty in subsequent control and measurement tasks, significantly reducing control efficiency and reliability.

[0093] Therefore, the embodiments of the present application further provide an intelligent trajectory generation scheme capable of predicting and actively avoiding sensitive transient regions in advance, to reduce the influence of trajectory terminal approaching stage on the transient disturbance of optical link, thereby improving the stability of micro-mirror control action and the reliability of system determination.

[0094] Specifically, the processing unit first constructs an initial trajectory with limited acceleration change rate according to target pose data and the current initial state of the micro-mirror. For example, the processing unit can start from the current angle and current angular velocity of the micro-mirror, and according to the target pose data, such as the set target angle or target scanning range, use a trajectory generation method with explicit boundary constraints, such as S-curve planning, polynomial interpolation or piecewise polynomial method, to calculate a preliminary trajectory. This preliminary trajectory can ensure that the motion of the micro-mirror has continuous acceleration change and smooth speed throughout the process, thereby effectively avoiding sudden changes in driving circuit current and mechanical resonance.

[0095] For example, the maximum angular acceleration change rate can be set not to exceed a certain value, such as several degrees per second squared per millisecond, to avoid circuit transient impact or optical signal jitter.

[0096] Then, the processing unit further predicts the optical response of each discrete pose point on the preliminary trajectory through a pre-established mapping model of pose to optical response, and generates a corresponding power change trend indication.

[0097] In a specific implementation, the processing unit utilizes the stored pose-optical response calibration data or historical operation data to establish a mapping relationship between the pose and the optical signal power through regression analysis, curve fitting or interpolation, etc.

[0098] For example, the processing unit samples a plurality of discrete micro-mirror poses, maps each pose to a corresponding sequence of optical power prediction values, and further analyzes the trend of these prediction values, particularly focusing on the slope and inflection point information in the trend, to identify high-risk regions in the micro-mirror motion pose change process that may trigger sensitive transient response of the amplification chain.

[0099] According to the obtained power trend indication, the processing unit subsequently determines the change rate limit parameter and the terminal approach speed to actively avoid triggering the sensitive transient response region.

[0100] For example, the processing unit marks the key inflection point positions in the power trend and sets the corresponding change rate limit parameter, such as forcing the trajectory change rate not to exceed a certain threshold, e.g. 50% of the original design change rate, and limiting the terminal approach speed within a certain numerical range, e.g. between 30%-60% of the original terminal speed, in the last segment of the trajectory approaching these inflection points, thereby effectively slowing down the terminal approach speed of the micro-mirror pose and reducing the disturbance to the transient response of the optical link.

[0101] Further, the processing unit reconstructs the last segment of the original trajectory based on the determined change rate limit parameter and terminal approach speed, and sets a trajectory alignment segment.

[0102] In a specific implementation, the processing unit first adjusts the sampling point density of the last segment of the trajectory to ensure that the last segment of the trajectory can approach the target pose in a more fine and gentle manner; then the processing unit recalculates the pose increment between the trajectory sampling points according to the aforementioned change rate limit parameter to obtain a reconstructed trajectory that satisfies the change rate limit condition; at the same time, the processing unit appropriately shifts the timing of trajectory execution in combination with the state switching time of the amplification chain steady state identifier, so that the trajectory termination time can be accurately aligned with the start time of the subsequent stable sampling window.

[0103] For example, if the original trajectory terminal is expected to end at time T, and the opening time of the steady state sampling window is at T+Δt, the processing unit can appropriately stretch the execution time length of the last segment of the trajectory, e.g. from T to T+Δt, to achieve accurate timing matching between the trajectory terminal and the opening of the stable window.

[0104] Thus, the problem of the sensitive transient response of the optical link in the approaching process of the micro-mirror terminal is effectively solved, the attitude change of the micro-mirror can be cooperated with the steady-state determination and the stable sampling window of the amplification chain with high precision, and thus the stability and reliability of the micro-mirror control in the actual optical link operation are significantly improved.

[0105] Optionally, in the actual control process of the micro-mirror, the state of the amplification chain may be returned from the steady state to the non-steady state due to various reasons, such as signal fluctuation of the optical link, transient response of the amplifier gain, or the amplification chain is in a non-steady state when the initial control action is performed. At this time, if the micro-mirror continues to act according to the original planned trajectory, the transient state of the downstream optical link amplifier may be further intensified, and even the amplification chain cannot recover to the steady state for a long time, and in severe cases, the system may be unstable or deteriorated. In this case, if there is no corresponding processing mechanism, the micro-mirror motion itself may become an obstacle to the recovery of the amplification chain, thereby prolonging the control process time, reducing the control efficiency and the determination accuracy.

[0106] Therefore, the embodiment of the present application further actively suspends the execution of the micro-mirror trajectory, latches the current attitude, and implements the change rate limitation mechanism for the remaining segment of the trajectory when the amplification chain is in a non-steady state, so as to actively reduce the excitation amplitude of the transient disturbance of the downstream amplification chain. Through the coordination of the suspension of the micro-mirror action and the steady-state state, on the one hand, the amplification chain can recover to the steady state more quickly under the relatively static or slowly changing incident conditions; on the other hand, the reliability of the subsequent sampling, checking and other control processes is also ensured. Therefore, the mechanism can significantly reduce the uncertainty in the micro-mirror control process and the transient response of the amplification chain, and improve the overall control performance and robustness.

[0107] In a specific implementation, when the processing unit detects that the state of the amplification chain is identified as a non-steady state identifier during the execution of the trajectory, the processing unit will immediately suspend the execution of the current micro-mirror smooth trajectory to avoid further excitation of the transient response of the amplification chain by the continuous trajectory action. The specific implementation is that: after detecting the non-steady state identifier, the processing unit sends a trajectory execution suspension instruction to the intelligent driving circuit, the intelligent driving circuit receives the instruction and immediately terminates the execution of the current motion trajectory of the micro-mirror, and the current attitude state of the micro-mirror is latched and maintained. For example, the intelligent driving circuit can collect the current micro-mirror attitude data in real time and feedback control as a closed-loop reference target value, so as to ensure that the micro-mirror attitude is stable at the state at the time when the trajectory is suspended, and avoid further action during the non-steady state to cause the system to deteriorate.

[0108] To further ensure the stability of the subsequent trajectory restart, the processing unit then performs a rate of change limiting process on the remaining segment of the original trajectory, that is, re-planning is performed on the part of the original trajectory that has not been executed. Specifically, the processing unit first sets the rate of change limiting parameter of the remaining segment of the original trajectory. For example, according to the preset trajectory safety limiting threshold, a relatively low attitude change rate, for example, between 20% and 50% of the original design value, can be selected as the upper limit of the trajectory re-planning. Then the processing unit implements re-interpolation or parameter adjustment on the remaining segment of the trajectory with the selected rate of change limiting parameter, so that the attitude change rate is smooth and slow when the trajectory is restarted subsequently, avoiding triggering the sensitive transient response of the amplifier chain state again.

[0109] During the pause of the micro-mirror action, the processing unit continuously monitors the amplifier chain state identifier until the amplifier chain state returns to the steady state. In a specific implementation, when the amplifier chain identifier changes from the non-steady state to the steady state, the processing unit further judges whether the steady state is stably maintained for more than a preset minimum dwell time,

[0110] For example, the minimum dwell time can be set to the order of milliseconds to hundreds of milliseconds, such as 50 ms to 200 ms, to ensure that the amplifier chain truly reaches the stable state rather than a short pseudo-steady state. When the processing unit determines that the steady state identifier has been maintained for more than the dwell time, the processing unit notifies the intelligent drive circuit through the interface to safely restart the trajectory execution, and uses the trajectory position corresponding to the switching time when the steady state is stable and the trajectory termination time as the restart point, to ensure the continuity and smoothness of the action execution.

[0111] In actual operation, the processing unit also provides processing measures for boundary abnormal conditions such as the amplifier chain failing to recover to the steady state for a long time or the state identifier fluctuating frequently. For example, when the non-steady state lasts for more than a few seconds, such as 2 s to 5 s, and still does not recover to the steady state, the processing unit will actively trigger a soft degradation protection mechanism to force the micro-mirror to return to the safe default attitude, avoiding serious control instability.

[0112] Optionally, Figure 3 A flowchart of a method for generating a power change trend indication provided by an embodiment of the present application is provided. Based on the above embodiments, the foregoing embodiments are further refined. Referring to Figure 3 The method comprises:

[0113] S301: Combining the target pose data and the first measurement data in time sequence to form a pose sampling sequence, and selecting a candidate dictionary entry matching the pose sampling sequence in a preset response dictionary.

[0114] S302: sparse constraint regression is used to solve mapping coefficients on the candidate dictionary entries, and non-negative constraint and monotonicity constraint are applied to obtain local mapping of the attitude to the optical response.

[0115] S303: mapping and timing smoothing are performed on the attitude sampling sequence based on the local mapping, to generate a power change trend indication.

[0116] In the micro-mirror control process, especially for the optical link terminal approximation stage, the amplification chain is extremely sensitive to the input attitude change, and even a slight change in the micro-mirror attitude can cause a significant optical response transient. In this case, how to accurately and reliably predict the optical response corresponding to the attitude change becomes a key prerequisite for trajectory refinement planning and amplification chain sensitive transient avoidance.

[0117] Embodiments of the present application select candidate entries matching the attitude sampling sequence in the preset response dictionary, use sparse constraint regression and non-negative and monotonic constraints for local mapping, and then obtain an accurate, smooth and time-continuous power change trend indication.

[0118] In one embodiment, the processing unit forms an attitude sampling sequence based on the aforementioned target attitude data and first measurement data at a millisecond level or higher frequency within a current task window. The attitude sampling sequence is arranged in chronological order and includes at least a micro-mirror angle; when necessary, angular velocity can be attached as an auxiliary dimension. The length of the sampling sequence can be adaptively determined according to the task duration, and is usually tens to hundreds of sampling points.

[0119] Further, the response dictionary is constructed by offline calibration and online operation log accumulation. Each dictionary entry contains a short-time attitude segment and its aligned optical power response segment, and is attached with meta-information of the acquisition working condition, such as temperature interval and path configuration identifier. In the matching stage, the processing unit uses the attitude sampling sequence as a query, which can be retrieved in the dictionary using the similarity measure commonly used in the art, and the first several, for example, 10-50, candidate dictionary entries most similar to the current sequence are selected. When entries under different working conditions are mixed, the entry set consistent or similar to the current working condition can be filtered based on the meta-information.

[0120] In one embodiment, the processing unit performs sparse constraint regression based on the candidate dictionary entries to solve the combined weights of each entry. Two types of physical priors are introduced in the regression process:

[0121] Non-negative constraint: the weight is not negative, avoiding non-physical interpretation caused by mutual cancellation.

[0122] Monotonicity constraint: the monotonic direction (rising or falling) of the combined post-mapping is limited within the current attitude working segment, which is consistent with the known coupling relationship or local calibration trend.

[0123] The constraint strength can be tuned by a parameter table, and stable solutions can be found in small step increments. When candidate entries are redundant or highly correlated, the processing unit can first do a brief screening or regularization suppression to obtain a sparse and stable local mapping. The local mapping is output in the form of "input posture point-predicted power value", which can be stored as a discrete lookup table or saved as a parameterized segment for subsequent fast call.

[0124] Further, the processing unit maps the posture sampling sequence point by point with the local mapping to obtain the estimated power response sequence. To improve the trend readability and suppress small noise, the processing unit performs time series smoothing processing on the sequence, such as sliding average or local regression smoothing, and the window length is set according to the sampling frequency and the expected response speed.

[0125] Then, the slope and inflection point information is extracted: the rising, falling and platform segments of the continuous interval are taken as the trend label, and the inflection point is marked at the turning time from rising to falling or from falling to rising. The final output power change trend indication includes at least: the start and end indexes of the trend interval, the representative slope in the interval, the key inflection point position and the confidence score, which are used to drive the selection of subsequent change rate limiting parameters and terminal approaching speed.

[0126] In one embodiment, when the dictionary matching score is overall low or the sparse solution does not converge, the processing unit records the quality identifier and enters the degradation path: preferentially narrowing the candidate set, relaxing the similarity threshold, or falling back to the coarse-grained mapping calibrated by the device at the factory; if the quality requirement is still not met, only the conservative trend label is output, such as "suspected sensitive interval", for the subsequent trajectory end to adopt a more stringent speed limiting strategy. This degradation does not affect the usability of the main process.

[0127] In this way, without adding additional sensors and complex linkage, the reusable response dictionary and constraint regression are used to construct a local mapping and trend indication that meets the physical prior, which not only guarantees the accuracy and stability of local prediction, but also provides the key structured information required for trajectory optimization at the end, suitable for optical communication switching, optical detection and other scenarios that require stability at the end.

[0128] Optionally, obtaining the local mapping of the posture to the optical response includes:

[0129] A sparse regression problem containing an outlier suppression term is constructed, and a first coefficient set satisfying the non-negative constraint of the coefficient is solved for the candidate dictionary entry.

[0130] Based on the time sequence of the posture sampling sequence, the support preservation constraint and the total variation constraint are imposed on the first coefficient set to obtain a second coefficient set satisfying the time sequence consistency.

[0131] The candidate dictionary entries are grouped and sparse weights of the groups are set based on the second set of coefficients and the state identifier of the amplification chain, and in the solving process, the dictionary entries corresponding to the steady state identifier are preferentially selected to obtain a third set of coefficients.

[0132] The local mapping is defined by the dictionary entries corresponding to the third set of coefficients, and the local mapping is subjected to a monotonicity constraint with respect to the pose variable and a slope limiting given by a rate of change limiting parameter, and the local mapping is outputted.

[0133] In an embodiment, to ensure that the prediction of the optical response in the process of controlling the micro-mirror can accurately reflect the actual relationship between the pose change and the optical power, the processing unit further optimizes the local mapping of the pose to the optical response. Specifically, the processing unit first constructs a sparse regression problem with an outlier suppression function to eliminate the adverse effects that may be caused by abnormal measurement values or noise interference.

[0134] For example, the processing unit calculates the degree of difference between each candidate dictionary entry and the current to-be-measured pose sample sequence based on the candidate dictionary entries obtained in the foregoing steps, and identifies potential outliers based on points or segments with greater difference. Subsequently, the processing unit sets sample weights based on the degree of difference when calculating the weights corresponding to each candidate dictionary entry, and the weights corresponding to the dictionary entries with significantly greater difference are significantly reduced. The processing unit further repeats the foregoing process in an iterative manner, that is, the weights of the dictionary entries are updated after each round of calculation, until the change amplitude of the weights does not exceed a predetermined threshold, for example, the change amplitude is less than 0.001, so as to obtain a first set of coefficients satisfying the non-negative constraint of the coefficients. This set of coefficients preliminarily determines the contribution degree of each candidate dictionary entry in the pose-to-optical power mapping, eliminates the influence of obvious abnormal points, and ensures the stability of the preliminary regression result.

[0135] To further enhance the time continuity of the mapping coefficients, the processing unit imposes a support preserving constraint and a total variation constraint on the first set of coefficients based on the natural continuous characteristics of the pose sample sequence in time, to obtain a second set of coefficients that are smooth and support continuous over time.

[0136] For example, the processing unit first checks the continuity of the coefficients at adjacent sampling times, and if the difference between the coefficients of a dictionary entry at adjacent times exceeds a predetermined threshold, for example, the difference between the coefficients at two adjacent times exceeds 30%, the coefficients of the entry are adjusted to avoid sudden changes in the coefficients. In addition, the processing unit further adopts a total variation constraint mechanism, that is, the fluctuation amplitude of the coefficients of each entry over time is limited.

[0137] For example, the processing unit calculates the average value of each dictionary entry coefficient over the time series, and limits the variation of each coefficient at each time point to a certain proportion of the average value, for example, not more than 20% of the average value. Through such constraints, the variation of the coefficient sequence in the time dimension is ensured to be smooth and continuous, avoiding unreasonable coefficient mutations caused by instantaneous interference.

[0138] In addition, in order to better reflect the influence of the amplification chain state on the prediction of the optical response, the processing unit further combines the aforementioned second coefficient set and, according to the amplification chain state identifier, performs state grouping on the candidate dictionary entries and corresponding group sparse weight setting to obtain a third coefficient set. In specific implementation, the processing unit first divides the candidate dictionary entries into two main groups, i.e., a "steady state group" and a "non-steady state group", based on the amplification chain state identifier. Subsequently, the processing unit gives different weight preferences to different groups in the sparse regression process, for example, when the amplification chain state is a steady state, the weight of the steady state group dictionary entries is set to be several times that of the non-steady state group entries, for example, the weight of the steady state group is set to be 3 times that of the non-steady state group; and when in a non-steady state, the weight of the steady state group entries is correspondingly reduced. Through this state weighting method, it is ensured that the dictionary entries that are most matched with the state can be more effectively selected and combined under the current amplification chain state, thereby improving the sensitivity and accuracy of the local mapping to the actual working conditions.

[0139] Finally, the processing unit performs weighted combination on the corresponding candidate dictionary entries with the third coefficient set to obtain a local mapping function of the pose to the optical response. Before outputting the local mapping, in order to ensure that the mapping result conforms to the actual device operation rule, the processing unit further applies a monotonicity constraint to the mapping function. For example, if the historical calibration shows that the optical power in a specific pose interval changes monotonically with the pose, the processing unit performs monotonicity correction on the output prediction result: if it is found that the mapping result appears non-monotonic in a local section, it is made to restore the monotonicity feature by means of smooth connection or local correction.

[0140] At the same time, the processing unit also limits the slope of the local mapping function on the pose variable, so as to avoid that the change rate of the predicted optical power exceeds the range that can be tolerated by the actual device or stably tracked by the amplification chain. For example, the processing unit checks the change rate between adjacent prediction points according to the change rate limit parameter determined in advance, and if the change rate exceeds a preset threshold, such as an optical power change of not more than 2% to 5% per millisecond, the change rate of the predicted optical power is made to return to the safe range by means of interpolation or re-adjusting the pose sampling point density of the local section, so as to ensure that the output local mapping result is more stable and controllable in time sequence.

[0141] In this way, the outlying data interference can be effectively inhibited, the continuity and smoothness of the mapping coefficients in the time dimension can be ensured, and the coefficient distribution can be dynamically adjusted according to the actual amplification chain state and the most suitable dictionary entry can be preferentially selected.

[0142] Optionally, determining the change rate limit parameter and the terminal approaching speed comprises:

[0143] Based on the slope and inflection point information and the amplification chain state identifier, a transient sensitivity index representing the sensitivity of the amplification chain to the change in incident power is generated; based on the transient sensitivity index and the starting condition of the stable sampling window, a power change safety boundary is constructed, and a parameter set including the candidate change rate limit parameter and the candidate terminal approaching speed is generated accordingly; based on the model of the attitude to the optical response, the parameter set is transiently predicted and the power change envelope is calculated, and the target change rate limit parameter and the target terminal approaching speed that make the power change envelope meet the power change safety boundary and the end time consistent with the starting condition of the stable sampling window are selected.

[0144] During the micro-mirror terminal approaching stage, a small attitude change can cause a significant power change of the amplification chain, thereby triggering a sensitive transient response of the amplification chain. This sensitive transient response can not only prolong the time for the system to reach a steady state, but also reduce the stability and accuracy of the control system. Therefore, the embodiment further provides a refined method for determining the change rate limit parameter and the terminal approaching speed of the trajectory end segment, so as to effectively avoid the sensitive transient excitation of the amplification chain and achieve accurate alignment of the trajectory end point and the stable sampling window.

[0145] In specific implementation, first, the power change trend indication obtained in the previous step is used, which includes slope information and inflection point information, and the current amplification chain state identifier is combined to construct a transient sensitivity index that can quantify the sensitivity of the amplification chain to the change in incident light power.

[0146] For example, the processing unit identifies the key sensitive segments according to the power change trend indication, such as positions where the slope significantly increases or inflection points are present, and records these positions and amplitudes as regions with high sensitivity; then, the sensitivity index is adjusted according to the amplification chain state identifier, for example, when the amplification chain is in a non-steady state, the sensitivity index is correspondingly increased to indicate that the amplification chain is more susceptible to attitude changes; and in a steady state, the index can be appropriately reduced. The transient sensitivity index obtained in this way can objectively and clearly reflect the sensitivity of the amplification chain under different attitude change modes.

[0147] Next, to ensure that no significant optical power transient fluctuation is triggered during the trajectory approaching stage, a power change safety boundary is further constructed based on the above transient sensitivity index and the starting condition of the stable sampling window.

[0148] Specifically, according to the transient sensitivity index, a specific safety upper limit is set for the attitude change rate, for example, in the area with high sensitivity, a lower allowed change rate is set, such as 30%-60% of the original change rate; and in the area with relatively low sensitivity, the limit can be appropriately relaxed, for example, 60%-90% of the original change rate is allowed. At the same time, the processing unit sets a reasonable candidate interval for the terminal approximation speed of the micro-mirror, for example, 0.2-1.0 degree / millisecond, by taking the starting time of the stable sampling window as the accurate target time of the trajectory terminal approximation, combined with actual engineering experience, to ensure that the end attitude is consistent in time with the starting condition of the stable sampling window.

[0149] On the basis of the above power change safety boundary, the processing unit generates a series of parameter sets containing candidate change rate limit parameters and candidate terminal approximation speeds according to predetermined rules.

[0150] For example, a plurality of candidate parameter pairs can be formed for different change rate limit parameters and terminal approximation speed combinations, each of which is used as a test object for subsequent evaluation.

[0151] Typically, a plurality of parameter combinations can be set for testing, for example, the change rate limit is 30%, 50%, and 70% of the original change rate, and the terminal approximation speed is 0.3, 0.5, and 0.8 degree / millisecond, respectively, to form a parameter set with a reasonable number, for example, no more than 10 groups, for subsequent fine selection.

[0152] Subsequently, the processing unit uses the pre-constructed and stored attitude-to-optical response mapping model to perform transient prediction of the attitude-to-power response for each candidate parameter combination in the above parameter set. In specific implementation, the processing unit inputs the trajectory end segment sampling sequence corresponding to each candidate parameter combination into the aforementioned attitude-to-optical response model to generate a corresponding power response prediction sequence. Then, the power change envelope (i.e., the maximum, minimum, and fluctuation range of the power prediction sequence) is calculated to evaluate whether the optical power change meets the above determined power change safety boundary under the candidate parameter combination. For example, the processing unit can check whether the maximum fluctuation amplitude in the power change envelope is less than the maximum change value allowed by the safety boundary, and if not, the parameter combination is excluded; if so, it is recorded as an alternative.

[0153] After completing the prediction evaluation of all candidate parameter combinations, the processing unit further screens the optimal target parameter combination from the alternative solutions that meet the power change safety boundary as the parameter for actual trajectory execution. Specifically, the processing unit further selects the parameter combination that is most accurately aligned in time with the stable sampling window starting condition according to the terminal attitude arrival time of the alternative solution, for example, the parameter combination with the smallest difference from the window starting time. If there are multiple parameter combinations that can meet the alignment accuracy requirement, the processing unit can further select the solution that has the lowest or most stable power change amplitude as the final solution to further improve the safety and stability of the trajectory end execution.

[0154] The above method can ensure that the attitude of the micro mirror can accurately control the power change within the safety boundary range when approaching the target endpoint, actively avoid the amplification chain sensitive transient region, thereby realizing accurate alignment of the trajectory terminal and the stable sampling window, and effectively improving the stability and reliability of the overall micro mirror control action.

[0155] In addition, those skilled in the art can make appropriate adjustments or optimizations to the above parameter selection and evaluation method according to the actual equipment and control scene, as long as it is within the technical solution and protection scope of the claims of the present application, it belongs to the implementation range of the present application.

[0156] Optionally, generating the amplification chain state identifier based on the second measurement data comprises:

[0157] The gain estimate, the output optical power change rate, the short window variance and kurtosis of the output optical power, and the noise sideband energy ratio are calculated from the second measurement data to form a state discrimination feature sequence; the state discrimination feature sequence is matched based on a preset transient response template library, and a semi-Markov state discriminator with residence time constraint is used for sequential updating to obtain a steady state confidence sequence; based on the steady state confidence sequence and the candidate stable interval detected by the change point detection, a minimum residence time test, a variance upper bound test, and a monotonic convergence test are performed, a steady state identifier is generated and a steady state time stamp is recorded when the test is satisfied, and a non-steady state identifier is generated when the test is not satisfied.

[0158] In the actual application of micro mirror control, the running state of the amplification chain directly affects the system control decision and the accurate regulation of the micro mirror attitude. However, the working state of the amplification chain is often affected by optical path fluctuations, external interference or device characteristics, which manifests as significant transient changes in gain or output optical power and other optical characteristics. The traditional simple single-threshold state determination method often cannot accurately distinguish between true steady state and transient state, which easily leads to misjudgment or misaction. Therefore, the present embodiment comprehensively utilizes multi-dimensional state discrimination features and a steady state template library, and combines the high reliability state discrimination of the semi-Markov state model to accurately generate the amplification chain state identifier, significantly improving the accuracy and robustness of state judgment.

[0159] In one embodiment, the processing unit collects second measurement data in real time or periodically, such as the amplified chain output optical power, the input-output optical power ratio (gain estimate), the statistical features of the output optical power in a short window, such as the short window variance and kurtosis, and the sideband energy ratio related to the optical signal noise level.

[0160] For example, the processing unit continuously acquires these measurement quantities at a fixed sampling frequency, performs basic data preprocessing for each measurement quantity, and then arranges the processed measurement quantities in chronological order to form a state discrimination feature sequence represented by a multi-dimensional vector, which truly reflects the dynamic changes of the amplified chain operating state.

[0161] Further, a transient response template library is established using historical data or offline experimental data, each template being a historical sequence of state discrimination features in a typical transient or steady state, covering typical patterns of different amplified chain states such as steady-state stability, transient rise, transient fall, fluctuation jitter, etc.

[0162] Then, the current real-time obtained state discrimination feature sequence is compared with each template sequence in the template library for similarity, such as using Euclidean distance, dynamic time warping (DTW) distance or cosine similarity to calculate the difference between sequences, to identify a number of candidate templates from the template library that best match the current state feature sequence, forming a preliminary result of template matching.

[0163] To accurately determine whether the amplified chain is in a steady state, a semi-Markov state discriminator with residence time constraints is further used for sequential updating of the state.

[0164] In a specific implementation, first, the amplified chain state space is defined, such as including only two basic states of "steady state" and "non-steady state", or further subdividing the transient phase, such as "steady state", "transient rise", "transient fall", etc. The processing unit then takes the preliminary result of template matching as an observation input, and uses a semi-Markov model (Semi-Markov) or a hidden semi-Markov model (Hidden Markov Model, HSMM) to calculate the steady-state confidence of each state in real time.

[0165] For example, the residence time constraint is based on the statistical features of the respective durations of the steady state and the non-steady state, such as the typical duration of the steady state being statistically calculated from historical data, and the processing unit explicitly adding a constraint on this duration in the semi-Markov model to improve the stability of state estimation and avoid temporary fluctuations being incorrectly determined as state changes.

[0166] After obtaining the steady-state confidence sequence, further change point detection is performed to identify possible turning points between steady states and non-steady states, thereby obtaining candidate stable intervals. In specific implementation, the processing unit automatically identifies positions where the steady-state confidence changes significantly as candidate change points based on the steady-state confidence being higher than a certain threshold (such as 0.8) by using common change point detection methods such as sliding window threshold detection or local mean change detection, and then extracts intervals between adjacent change points as candidate stable intervals for subsequent verification.

[0167] Next, a series of strict steady-state tests are performed on the candidate stable intervals to ensure high reliability of steady-state determination. Specifically, first, a minimum residence time test is performed, that is, whether the duration of the candidate interval exceeds a predetermined minimum steady-state residence time (for example, at least 50 milliseconds) is checked; second, a variance upper bound test is performed, that is, the variance value of the optical power change in the candidate interval is calculated, and it is confirmed that it is lower than a certain preset upper limit, for example, the optical power change variance is lower than a small value; and finally, a monotonic convergence test is performed, that is, it is confirmed that the optical power, gain estimate value or other features in the candidate interval show a stable trend and do not appear obvious repeated fluctuations or non-monotonic changes. For example, the processing unit can check whether the optical power sequence in the candidate interval gradually stabilizes to a platform over time and does not appear obvious reversal.

[0168] When the candidate interval passes all the above steady-state tests, it is determined that the current amplification chain enters a steady state, and a steady-state identifier is generated; at the same time, the starting time point of entering the steady state is recorded as a "steady-state transition timestamp" for subsequent accurate alignment of stable sampling window and trajectory terminal timing. If the candidate interval fails to meet any of the above tests, it is determined to be a non-steady state, a corresponding non-steady-state identifier is generated, and the non-steady-state gating process is continued or entered, and the amplification chain is re-evaluated after re-entering the steady state.

[0169] In this way, accurate discrimination of the running state of the amplification chain is achieved, the reliability and accuracy of state identifier generation are significantly improved, false positives or false actions that are prone to occur in traditional methods are effectively avoided, a reliable state basis is provided for subsequent micromirror control decisions, and the stability and robustness of micromirror control actions are overall improved.

[0170] Optionally, to obtain steady-state confidence sequence for subsequent gating and timing alignment, the processing unit first performs time and amplitude normalization on the state discrimination feature sequence derived from the second measurement data. Specifically, the feature sequence is resampled at uniform intervals, e.g. 0.5 ms to 2 ms, linear or nearest-neighbor interpolation is applied to segments with short-time sampling irregularities, e.g. no more than 2 to 3 consecutive missing samples, and amplitude normalization or robust normalization is performed on each feature, e.g. based on median and quantile difference, and short-window detrending, e.g. using sliding mean or equivalent methods to remove slow drift, then time-stamp alignment is performed to form a multi-dimensional sequence of equal length, equal frequency, and same phase, as input for subsequent encoding.

[0171] After pre-processing is completed, the processing unit performs multi-resolution shape encoding on the above multi-dimensional sequence to enhance robustness against time scale stretching and small perturbations. In one embodiment, wavelet scattering coefficients at several scales, one or two levels, are used, e.g. scale number can be 3 to 6, level preferably 1 or 2, step size and bandwidth are set according to device noise and dynamic range, while Hankel fragments are constructed within a sliding window and principal subspace angles or similarity measures are extracted, with window length of 8 to 32 samples, step size of 1 to 4 samples. The two types of encoded vectors are concatenated in channel dimension and attached with time index to form a continuous encoded sequence, which is used as input for matching against the template library.

[0172] The processing unit maintains a transient response template library constructed from offline calibration and historical operation data, covering typical patterns such as steady state, transient rise, transient fall, dithering plateau, etc. The encoded sequence and template trajectories are matched through dynamic time warping, with scale-invariant regularization applied to allow time axis stretching within 5% to 15% range and amplitude linear scaling within 0.8 to 1.2 range. At each time instant, the processing unit aggregates matching scores against several most similar templates, normalizes the scores to 0 to 1 using weighted or voting methods, to obtain a template likelihood vector at that time instant; concatenating over time yields a template likelihood vector sequence, which provides observation input for subsequent sequential discrimination.

[0173] To suppress false positives caused by pseudo-steady state and short-time dithering, the processing unit constructs a hidden semi-Markov discriminator with residence time constraint using the template likelihood vector sequence as observation. The state set contains at least “steady state” and “non-steady state”, and can be further subdivided into transient types if needed; residence time distributions of each state are estimated from historical data through non-parametric kernel density, with minimum residence time lower bound set, e.g. 50 ms to 200 ms for steady state, and 20 ms to 100 ms for non-steady state. Hazard rate curves of residence time are processed through monotonicization to avoid unreasonable decrease over time. The discrimination process uses forward-backward sequential update to calculate state posterior probability at each time instant, and introduces normalization or logarithmic domain accumulation in the calculation to ensure numerical stability.

[0174] To reduce non-physical jumps between adjacent time slices and suppress high-frequency state switching, the processing unit adds regularization at the distribution level on top of the sequential update. One feasible way is to introduce an optimal transport type of alignment penalty on the observation distribution or posterior distribution of adjacent time slices, with the weight coefficient set in the range of 0.01 to 0.2; at the same time, implement sparsification constraints on state transitions, reduce multiple switching in a short time by reducing the upper limit of the non-diagonal elements of the transition matrix, enabling the secondary gating of the minimum residence time, or pruning low-confidence transitions. The state posterior probability sequence obtained after this processing is smooth and has higher physical consistency.

[0175] The processing unit outputs the posterior probability of the "steady state" state over time as a steady-state confidence sequence (with a value of 0 to 1). When the steady-state confidence is higher than the threshold for a long time and meets the minimum residence time, the time is recorded as the steady-state timestamp, which is used for the alignment of the trigger of the stable sampling window and the trajectory termination condition.

[0176] If the template matching quality is low or the sequential update does not converge, the matching threshold can be temporarily lowered, the template set can be expanded, or the residence time lower limit can be increased; if it still cannot be stabilized, it will fall back to the conservative judgment based on the rate of change and variance, and record the event log for maintenance analysis.

[0177] Optionally, in an embodiment, the processing unit takes the stable sampling window as the time reference to complete the consistency judgment after aligning the data on the prediction side and the verification side in the same domain and phase. First, the attitude sequence of the smoothed trajectory in the terminal approximation stage is mapped to the observable feature space. The feature space at least includes the output optical power, and if necessary, the error code statistics or signal-to-noise index can be incorporated. The attitude to optical response model used for mapping is the device-level model trained by the aforementioned calibration and running data. The processing unit takes millisecond-level discrete samples of the trajectory and infers point by point to obtain a prediction response sequence consistent with the period covered by the stable sampling window.

[0178] To reflect the prediction uncertainty, the processing unit generates a confidence interval based on historical residuals, or generates a set of parallel prediction trajectories through multi-model voting and resampling, and then gives a range-based prediction at each time. Considering the stability of short-term extrapolation, the processing unit trains a data-driven linear propagation operator in the observable feature space, and propagates step by step within the window length, so that the first prediction value is strictly aligned with the window boundary and the internal sampling time, while the corresponding confidence interval or equivalent variance estimate is retained. The number of propagation steps can be equal to the window length, such as 10 to 15 milliseconds, and an equal number of steps are used; the propagation operator can be updated on a weekly or daily scale, and numerically normalized and truncated to avoid divergence.

[0179] On the verification side, the processing unit reads the selected target sub-window samples within the stable sampling window, performs robust denoising and outlier clipping on the original sequence. The common practice is to perform percentile clipping on single-point mutations and smooth them with short-window median or local regression, and then project the processed samples into the feature space consistent with the prediction. To facilitate statistical comparison with the prediction, the processing unit generates the first verification value and its variance estimate within the sub-window. The variance can be obtained from the robust dispersion measure of the sub-window samples, or by resampling the samples. In the case of insufficient sample size or continuous missing, the processing unit preferentially extends the sub-window or marks the verification as undetermined, and records the reason for subsequent maintenance.

[0180] To evaluate the consistency of the prediction side and the verification side under the same metric, the processing unit assembles the multi-point information of the time-aligned two sides into two comparable distributions. The prediction side can regard the set of sampling points corresponding to parallel prediction or confidence interval as the prediction distribution, and the verification side can regard the set of sub-window samples as the empirical distribution; if necessary, first do bucketing or kernel density estimation on both sides of the data to obtain a stable discrete representation.

[0181] Subsequently, the processing unit adopts the optimal transport divergence with entropy regularization as the main criterion, and obtains the distance measure between the two distributions through the iterative scaling solution process; the regularization weight is taken to a small to moderate range to balance numerical stability and sensitivity to differences, and the upper limit of iteration and convergence tolerance are set according to the device computing power and latency requirements.

[0182] To take into account the prediction uncertainty, the processing unit also calculates the overlap rate between the confidence interval of the first prediction value and the variance of the first verification value, and takes this overlap rate as an auxiliary factor to be fused with the optimal transport divergence to form a single consistency score. A robust approach is to weight the two with fixed weights, and the weights can be adjusted according to the station environment and historical misjudgment rate, and configured and managed in the system version.

[0183] The threshold value corresponding to the consistency condition comes from the device accuracy and the field calibration result. When the consistency score is better than the threshold value, the processing unit judges that the consistency result meets the requirements; when the score is worse than the threshold value, it is judged as not meeting the requirements; for the boundary case close to the threshold value, additional short-window resampling or increasing the overlap rate weight can be performed for review to reduce the false rejection caused by incidental noise.

[0184] The judgment result is output in the form of binary label or continuous score, and is accompanied by time label, confidence interval summary and key diagnostic quantities such as sample size, clipping ratio, and solution iteration rounds. When the numerical value is unstable or the solution does not converge, the processing unit falls back to a simple bucketing distance or energy-type measure according to the established degradation path, and simultaneously records the log for subsequent audit.

[0185] In one example, the stable sampling window length is 12 milliseconds and the sampling interval is 1 millisecond; the prediction side generates a prediction distribution with 10 parallel trajectories and gives a confidence interval, and the verification side retains 10 to 12 valid samples after median filtering and percentile clipping; the optimal transport regularization weight is set to 0.05, the upper limit of the number of iterations is set to 200, and the convergence tolerance is set to a small relative change amount; the consistency threshold is determined to be a medium level through field calibration.

[0186] Under this setting, when the amplification chain is in a steady state and the slope reconstruction at the end of the trajectory is effective, the consistency score is usually better than the threshold and gives a high overlap rate; if the sub-window samples still have transient fluctuations, the optimal transport divergence increases and the overlap rate decreases, and the consistency result is determined to be unsatisfactory and triggers the subsequent correction process.

[0187] Through the above implementation path, the system realizes reproducible and adjustable determination of the consistency between prediction and measurement without changing the existing measurement hardware, with a unified feature space, aligned time reference, controllable distribution metric, and explicit threshold system, meeting the common requirements of real-time and interpretability in engineering.

[0188] Optionally, the first prediction value is calculated based on a smooth trajectory and a model of pose to optical response, the first verification value is generated based on the target sub-window samples, and the consistency result is calculated based on the first prediction value and the first verification value, including:

[0189] The smooth trajectory is mapped to an observable feature space, and a prediction sequence with uncertainty is established based on a model of pose to optical response, and a Koopman operator is used to obtain the first prediction value and its confidence interval aligned with the stable sampling window;

[0190] Robust denoising and abnormal clipping are performed on the target sub-window samples and projected to the feature space to generate the first verification value and its variance estimate;

[0191] The optimal transport divergence with entropy regularization is calculated based on the prediction distribution of the first prediction value and the empirical distribution of the target sub-window samples, and the consistency score is obtained by combining the confidence interval overlap rate, and the consistency score is compared with the threshold corresponding to the consistency condition to form the consistency result.

[0192] In one embodiment, to achieve the consistency determination of the predicted value and the check value, the processing unit first maps the generated smooth trajectory to a predefined observable feature space, and constructs a prediction sequence in combination with a model of the pose to optical response. In specific operation, the processing unit inputs the discrete pose points on the trajectory into the model point by point to obtain the corresponding optical power prediction value, and by introducing historical residual statistics or model integration method, attaches an uncertainty description to each prediction point, which can be in the form of confidence interval or variance estimate. In order to ensure the time alignment of the prediction value and the sampling window, the processing unit further adopts a pushing method based on Koopman operator to establish an approximate linear evolution model in the observable feature space, and pushes the prediction sequence to the start and internal time of the stable sampling window, so as to obtain the first prediction value and its confidence interval strictly aligned with the sampling window. Exemplarily, in the optical communication switching scene, the processing unit can use the historical running data in the range of hundreds of milliseconds to fit the Koopman operator, and obtain the window-aligned prediction sequence by a few steps of pushing.

[0193] After obtaining the prediction value, the processing unit performs preprocessing on the target sub-window sample to form the first check value which can be directly compared. Specifically, the processing unit first performs robust denoising on the sample sequence, and the common way includes short window median filtering or local regression smoothing to eliminate transient peak noise; then the abnormal points exceeding the preset threshold are cut or the weight is lowered to ensure that the check value is not disturbed by a few abnormalities. The processed sample sequence is projected into the same feature space as the prediction value to obtain a set of representative check statistics. In order to characterize the fluctuation of the data in the sub-window, the processing unit also calculates the variance of the sample or estimates its distribution dispersion by resampling method, thereby generating the first check value containing the numerical value and the variance estimate.

[0194] Exemplarily, in the laboratory optical path interference strong working condition, the processing unit will enable a more stringent abnormal cutting ratio to ensure the robustness of the check data.

[0195] In the comparison link, the processing unit measures the difference between the prediction distribution of the first prediction value and the empirical distribution formed by the target sub-window sample. In order to obtain an index sensitive to both the distribution shape and the confidence interval, the processing unit adopts the optimal transport divergence with entropy regularization as the main measurement method.

[0196] In implementation, the prediction distribution is constituted by the aforementioned confidence interval or the sample result, and the empirical distribution is generated by the calibration sample and its resampling. The processing unit calls the iterative approximation solver to solve the optimal transport divergence, and introduces an entropy regularization term in the objective function to ensure the stability and interpretability of the solution. At the same time, the processing unit calculates the overlap rate of the prediction confidence interval and the calibration variance interval, and combines the index with the optimal transport divergence to form a single consistency score. Finally, the processing unit compares the consistency score with the threshold corresponding to the preset consistency condition, and determines that the consistency is satisfied if the score is better than the threshold, and otherwise determines that the consistency is not satisfied.

[0197] For example, in a field deployment, the processing unit can compare the consistency score with a 95% confidence threshold obtained by historical experimental statistics to determine whether to enter the correction or degradation process. Through this process, the processing unit realizes reliable consistency determination under the conditions of feedback lag and transient fluctuation, and provides a clear basis for subsequent completion flag generation, correction instruction execution or soft degradation strategy triggering.

[0198] Based on the same inventive concept, the embodiment of the present application also provides a micro-mirror control system based on an intelligent driving circuit corresponding to a micro-mirror control method based on an intelligent driving circuit. Since the principle of the system in the embodiment of the present application solves the problem similar to the above-mentioned micro-mirror control method based on an intelligent driving circuit, the implementation of the system can be referred to the implementation of the method, and the repeated parts will not be described again.

[0199] Figure 4 A structural schematic diagram of a micro-mirror control system based on an intelligent driving circuit provided by the embodiment of the present application is shown in Figure 4 The system comprises a collection module 410, a judgment module 420, a processing module 430 and a generation module 440.

[0200] The collection module 410 is configured to obtain target pose data; measure the motion characteristics of the micro-mirror to obtain first measurement data, the first measurement data being used to represent the current initial state of the micro-mirror, and generate a smooth trajectory based on the target pose data and the first measurement data; measure the amplification characteristics of the optical signal through the amplification chain to obtain second measurement data, and generate an amplification chain state identifier based on the second measurement data, the amplification chain state identifier comprising a steady state identifier and a non-steady state identifier.

[0201] The judgment module 420 is configured to, in response to the amplification chain state identifier being the non-steady state identifier, not start the stable sampling window and not perform the consistency calculation until the amplification chain state identifier changes to the steady state identifier; and in response to the amplification chain state identifier being the steady state identifier and the smooth trajectory reaching a termination condition, start the stable sampling window, obtain quality measurement results in the stable sampling window, and select target sub-window samples.

[0202] The processing module 430 is configured to calculate a first prediction value based on the smoothed trajectory and the model of the pose to optical response, generate a first check value based on the target sub-window sample, and calculate a consistency result based on the first prediction value and the first check value.

[0203] The generating module 440 is configured to generate a completion flag in response to the consistency result satisfying a consistency condition, generate a correction instruction and execute the correction instruction in response to the consistency result not satisfying the consistency condition, and generate a soft degradation instruction when a correction number reaches a preset threshold.

[0204] The micro-mirror control system based on the intelligent driving circuit provided by the embodiment of the present application can execute the micro-mirror control method based on the intelligent driving circuit provided by any embodiment of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0205] It should be understood that the steps can be reordered, added, or deleted using the various forms of flow shown above. For example, each step described in the present application can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions of the present application can be achieved, and the present application is not limited herein.

[0206] The specific embodiments described above do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for controlling a micro-mirror based on an intelligent driving circuit, characterized in that, The method comprises: acquiring target pose data; measuring the motion characteristics of the micromirror to obtain first measurement data, the first measurement data being used to represent the current initial state of the micromirror, and generating a smooth trajectory based on the target pose data and the first measurement data; measuring the amplification characteristics of the optical signal amplification chain to obtain second measurement data, and generating an amplification chain state identifier based on the second measurement data, the amplification chain state identifier comprising a steady state identifier and a non-steady state identifier; in response to the amplification chain state identifier being a non-steady state identifier, not starting a stable sampling window and not performing consistency calculation until the amplification chain state identifier changes to the steady state identifier; in response to the amplification chain state identifier being a steady state identifier and the smooth trajectory reaching a termination condition, starting a stable sampling window, acquiring a quality measurement result in the stable sampling window, and selecting a target sub-window sample; calculating a first predicted value based on the smooth trajectory and a model of pose to optical response, generating a first check value based on the target sub-window sample, and calculating a consistency result based on the first predicted value and the first check value; in response to the consistency result satisfying a consistency condition, generating a completion flag; in response to the consistency result not satisfying the consistency condition, generating a correction instruction and executing the correction instruction; when the number of corrections reaches a preset threshold, generating a soft degradation instruction.

2. The method of claim 1, wherein the method is based on an intelligent driving circuit. The method of generating a smooth trajectory based on the target pose data and the first measurement data comprises: constructing an initial trajectory with limited acceleration change rate based on the target pose data and the first measurement data; generating a power change trend indicator through mapping of pose to optical response based on the target pose data and the first measurement data, and determining a change rate limit parameter and a terminal approaching speed based on the power change trend indicator; the power change trend indicator comprises slope and inflection point information; performing slope limit reconstruction on the last segment of the initial trajectory based on the change rate limit parameter and the terminal approaching speed, and setting an alignment segment so that the last segment reaches a time corresponding to the starting condition of the stable sampling window, to obtain the smooth trajectory.

3. The method of claim 1, wherein the method is based on an intelligent driving circuit. The method of responding to the amplification chain state identifier being a non-steady state identifier further comprises: stopping execution of the smooth trajectory, latching the current pose, and applying a change rate limit to the remaining segment of the smooth trajectory until the amplification chain state identifier changes to the steady state identifier.

4. The method of claim 2, wherein the method is based on an intelligent driving circuit. The method of generating a power change trend indicator comprises: combining the target pose data and the first measurement data in time sequence to form a pose sampling sequence, and selecting a candidate dictionary entry matching the pose sampling sequence in a preset response dictionary; solving mapping coefficients on the candidate dictionary entry by sparse constraint regression, and applying non-negative constraint and monotonicity constraint to obtain a local mapping of pose to optical response; mapping and time smoothing the pose sampling sequence based on the local mapping to generate a power change trend indicator.

5. The method of claim 4, wherein the method further comprises: The method of obtaining a local mapping of pose to optical response comprises: constructing a sparse regression problem containing an outlier suppression term, and solving a first coefficient set satisfying a non-negative constraint for the candidate dictionary entry; impose a support preserving constraint and a total variation constraint on the first coefficient set based on time order of the pose sampling sequence, to obtain a second coefficient set satisfying time consistency; group and set group sparse weight for the candidate dictionary entries based on the second coefficient set and according to the amplification chain state identifier, to obtain a third coefficient set by preferentially selecting a dictionary entry corresponding to a steady state identifier in solving; define the local mapping by dictionary weighted combination corresponding to the third coefficient set, impose a monotonicity constraint on the local mapping with respect to a pose variable and a slope limit given by the change rate limiting parameter, and output the local mapping.

6. The method of claim 2, wherein the method is based on an intelligent driving circuit. The determination of the change rate limiting parameter and the terminal approaching speed includes: generate a transient sensitivity index representing a degree of sensitivity of the amplification chain to incident power change based on the slope and the inflection point information and the amplification chain state identifier; construct a power change safety boundary based on the transient sensitivity index and a starting condition of the stable sampling window, and generate a parameter set including a candidate change rate limiting parameter and a candidate terminal approaching speed according to the power change safety boundary; perform transient prediction on the parameter set based on the model of the pose to optical response and calculate a power change envelope, and select a target change rate limiting parameter and a target terminal approaching speed that make the power change envelope satisfy the power change safety boundary and the end time consistent with the starting condition of the stable sampling window.

7. The method of claim 1, wherein the method is based on an intelligent driving circuit. The generation of the amplification chain state identifier based on the second measurement data includes: calculate a gain estimate, an output optical power change rate, a short window variance and kurtosis of the output optical power, and a noise sideband energy ratio from the second measurement data to form a state discrimination feature sequence; match the state discrimination feature sequence based on a preset transient response template library, and perform sequential update using a semi-Markov state discriminator with a residence time constraint to obtain a steady state confidence sequence; based on the steady state confidence sequence and the candidate stable interval detected by the change point, perform minimum residence time test, variance upper bound test and monotonic convergence test, generate a steady state identifier and record the steady state time stamp when the test is satisfied, generate a non-steady state identifier when the test is not satisfied.

8. The method of claim 7, wherein the method further comprises: The steady state confidence sequence includes: perform time scale normalization and multi-resolution shape coding on the state discrimination feature sequence, the shape coding includes wavelet scattering coefficients and Hankel subspace angles, and match the coding vector with each template in the transient response template library through dynamic time warping and scale invariant regularization to obtain a template likelihood vector sequence; construct a hidden semi-Markov discriminator with a residence time constraint using the template likelihood vector sequence as an observation, the residence time distribution uses a non-parametric kernel density and imposes a minimum residence time constraint and a hazard rate monotonic constraint, and performs forward-backward sequential update to obtain a state posterior probability sequence; introduce an optimal transport regularization in the sequential update to minimize the observation distribution distance of adjacent time slices and impose a transition sparse penalty to suppress high frequency state switching, and use the updated state posterior probability sequence as the steady state confidence sequence.

9. A micromirror control method based on an intelligent driving circuit according to claim 1, characterized in that, The calculating a first prediction value based on the smooth trajectory and the model of pose to optical response, generating a first check value based on the target sub-window sample, and calculating a consistency result based on the first prediction value and the first check value includes: mapping the smooth trajectory to an observable feature space and establishing a prediction sequence with uncertainty based on the model of pose to optical response, and using a Koopman operator to obtain a first prediction value and a confidence interval thereof aligned with the stable sampling window; performing robust denoising and abnormal clipping on the target sub-window sample and projecting to the feature space to generate a first check value and a variance estimate thereof; calculating an optimal transport divergence with entropy regularization based on a prediction distribution of the first prediction value and an empirical distribution of the target sub-window sample, and combining a confidence interval overlap rate to obtain a consistency score, and comparing the consistency score with a threshold corresponding to the consistency condition to form the consistency result.

10. A micro-mirror control system based on an intelligent driver circuit, characterized by, comprises: The acquisition module is configured to acquire target pose data, measure motion characteristics of a micromirror to obtain first measurement data, wherein the first measurement data is used to represent a current initial state of the micromirror, and generate a smooth trajectory based on the target pose data and the first measurement data; measure amplification characteristics of an optical signal through an amplification chain to obtain second measurement data, and generate an amplification chain state identifier based on the second measurement data, wherein the amplification chain state identifier comprises a steady state identifier and a non-steady state identifier; The judgment module is configured to, in response to the amplification chain state identifier being the non-steady state identifier, not start a stable sampling window and not perform consistency calculation until the amplification chain state identifier changes to the steady state identifier; and in response to the amplification chain state identifier being the steady state identifier and the smooth trajectory reaching a termination condition, start the stable sampling window, acquire quality measurement results in the stable sampling window, and select a target sub-window sample. The processing module is configured to calculate a first prediction value based on the smooth trajectory and the model of pose to optical response, generate a first check value based on the target sub-window sample, and calculate a consistency result based on the first prediction value and the first check value. The generation module is configured to, in response to the consistency result satisfying a consistency condition, generate a completion flag. In response to the consistency result not satisfying the consistency condition, generate a correction instruction and execute the correction instruction; and when a correction number reaches a preset threshold, generate a soft degradation instruction.

Images