A method for correcting nonlinear errors in an optically pumped laser

Abstract

The application discloses a kind of optical pumping laser nonlinear error correction methods, it is related to laser spectroscopy detection technical field, comprising: every frequency sweep cycle synchronously acquires measurement detector absorption signal and reference structure a, reference structure b reference signal, reference signal is preprocessed as reference event sequence and generates quality index, for joint estimation and gated decision-making.Based on two-way reference event joint estimation frequency sweep time to wave number / frequency mapping function parameter and reference drift parameter, obtain candidate spectral axis.To candidate spectral axis implementation monotonicity, multiple reference consistency and residual threshold verifiable constraint gate, freeze update and output health degree score when abnormal.In accordance with final spectral axis resampling absorption signal and fitting demodulation output concentration / content, write back or keep parameter according to health degree.The method inhibits spectral axis slow drift under temperature drift vibration, reduces quantitative deviation, reduces calibration frequency, and outputs status word to support process analysis.

Description

Technical Field

[0001] This invention relates to the field of laser spectral detection technology, specifically to a method for correcting nonlinear errors in optically pumped lasers. Background Technology

[0002] In industrial settings such as natural gas / biomethane quality control, chemical tail gas treatment, and semiconductor process gases, continuous monitoring of trace impurities like NH3 is required. The mainstream approach utilizes TDLAS laser spectroscopy and its modulation variants, sweeping the laser frequency near the target spectral line and using time-wavelength / wavenumber mapping to fit the spectral line and invert concentration. Due to variations in temperature, vibration, and pressure / flow rate at the site, in addition to measurability, long-term spectral axis calibration and traceable records are necessary to support comparable results across devices and time periods.

[0003] US Patent document US12322926B2 (publication date: June 3, 2025) discloses a tunable laser assembly: a tunable semiconductor laser, a pump laser, an optical amplifier, and a detector are assembled into a package, coupled via a PLC waveguide, and a wavelength monitoring circuit (WMC) is connected to the PLC. During frequency sweep, pulses are generated, and a closed-loop controller controls the absolute wavelength and scanning bandwidth based on the pulse timing / number.

[0004] In the aforementioned technical forms, spectral axis calibration is typically performed using a single reference link (WMC internal filter / interference structure, external interferometer cavity, or MZI). Temperature variations, vibrations, and packaging stress during industrial processes can alter the refractive index and geometry of the reference device, causing equivalent optical path difference or free spectral range drift, introducing phase perturbations, and resulting in slow changes in the reference pulse timing or interference fringe phase. For reference pulse closed-loop control schemes, when the reference scale itself drifts, the reference loop may still maintain a small reference residual, but the actual spectral axis drifts slowly with the reference coordinate system. For single-path MZI fringe resampling schemes, the introduction of path difference changes, along with wavenumber mapping, can lead to spectral axis errors over long-term operation. This slow drift can cause spectral line center deviation and, coupled with changes in linearity / baseline due to altered operating conditions, make concentration bias difficult to detect; in natural gas metering or online monitoring of toxic gases, this can cause misjudgments and increase calibration and maintenance costs.

[0005] The resulting technical problems are: the long-term stability and traceability of the spectral axis calibration of the swept frequency absorption spectroscopy system under industrial disturbances and changes in operating conditions, and the hidden system errors caused by reference link drift. Summary of the Invention

[0006] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method for correcting nonlinear errors in optically pumped lasers. Based on the joint estimation of the mapping function parameters from sweep time to wavenumber / frequency and reference drift parameters using two reference events, candidate spectral axes are obtained. Verifiable constraint gating based on monotonicity, multi-reference consistency, and residual thresholds is applied to the candidate spectral axes. In case of anomalies, the update is frozen and a health score is output. The absorption signal is resampled based on the final spectral axis, and the demodulated output concentration / content is fitted and written back or retained according to the health score. Slow spectral axis drift is suppressed under temperature drift vibration, reducing quantitative bias, and a status word is output to support process analysis; thus solving the technical problems described in the background art.

[0007] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: A method for correcting nonlinear errors in an optically pumped laser includes: synchronously acquiring the absorption signal of a measurement detector and the reference signals of reference structure A and reference structure B at the beginning of a frequency sweep cycle, and preprocessing the reference signals to obtain a reference event sequence; Candidate sweep frequency mapping functions are obtained by jointly estimating the sweep frequency mapping function parameters from sweep time to wavenumber / frequency based on the reference event sequence and the reference drift parameters of reference structure A and reference structure B. Verifiable constraint gating is performed on the candidate sweep frequency mapping function according to monotonicity constraints, multi-reference consistency constraints and residual statistical thresholds. The final sweep frequency mapping function, reference drift parameters and health score are output. The update is frozen when the gating fails. The calibration spectral axis is generated based on the final sweep frequency mapping function. The absorption signal is resampled and the absorption spectrum is fitted / demodulated to output the concentration / content results. The parameters of the final sweep frequency mapping function are written back or frozen in response to the health score.

[0008] Furthermore, before synchronous acquisition, the swept laser output is coupled into a measurement optical path, a reference optical path A, and a reference optical path B via a beam splitter. The measurement optical path is coupled to the measurement detector to form an absorption signal, the reference optical path A is coupled to the reference structure A, and the reference optical path B is coupled to the reference structure B to form a reference signal.

[0009] Furthermore, synchronous acquisition includes synchronously sampling the absorption signal and the reference signal by a multi-channel analog-to-digital converter, locking the sampling time at the start of the frequency sweep cycle, writing the sampled values ​​of the absorption signal and the reference signal at the same sampling time into the same buffer index, and storing the buffer index and the reference event sequence together within the same frequency sweep cycle.

[0010] Furthermore, synchronous acquisition includes inputting the reference signal into the comparator to generate flip pulses, and recording the flip time of the flip pulse as an event timestamp by a timestamp structure. The event timestamps are arranged in ascending order to form a reference event sequence, which shares the start time of the frequency sweep cycle with the absorption signal, and duplicate event timestamps are deleted.

[0011] Furthermore, the preprocessing includes sequentially performing DC removal, amplitude limiting, bandpass filtering, and jitter reduction on the reference signal, and performing interpolation at the sign changes of adjacent sampling points to determine the zero-crossing time. The zero-crossing times are arranged in ascending order to form a reference event sequence and are aligned with the absorption signal in the same frequency sweep period.

[0012] Furthermore, the preprocessing also includes counting the number of events in the reference event sequence within the frequency sweep period to obtain the event density, obtaining the event interval consistency based on the time difference between adjacent events, obtaining the noise level indication based on the differential amplitude of the reference signal sample values, and saving the event density, event interval consistency, noise level indication, ambient temperature, and substrate temperature together.

[0013] Furthermore, when the event density of reference structure A is lower than the minimum threshold, reference structure A is marked as a reference failure state and a freeze update is triggered; when the event density of reference structure B is lower than the minimum threshold, reference structure B is marked as a reference failure state and a freeze update is triggered; when reference structure A and reference structure B are in a reference failure state, the system enters a degradation mode and maintains the final sweep frequency mapping function of the previous sweep frequency cycle.

[0014] Furthermore, the joint estimation adopts a recursive estimation process, updating the sweep frequency mapping function parameters in each sweep frequency cycle and updating the reference drift parameters in every predetermined number of sweep frequency cycles. The sweep frequency mapping function parameters include the segment boundaries and polynomial coefficients of the piecewise polynomial, and the predetermined number is a fixed number determined by the factory calibration.

[0015] Furthermore, it can be verified that the threshold values ​​of the multi-reference consistency constraints and the residual statistics threshold in the constraint gating are determined based on the noise level indication and ambient temperature at the beginning of the frequency sweep cycle, and remain unchanged during the frequency sweep cycle and are prohibited from being updated. The range of the threshold values ​​is determined by the factory calibration.

[0016] Furthermore, in response to the health score meeting the write-back condition, the final sweep frequency mapping function parameters are written back for resampling correction in backend data processing, and the pre-correction update for the sweep frequency driving waveform is also written back; in response to the health score being lower than the write-back condition, the write-back is frozen and the final sweep frequency mapping function parameters written back in the previous sweep frequency cycle are retained.

[0017] (III) Beneficial Effects This invention provides a method for correcting nonlinear errors in optically pumped lasers, which has the following advantages: In each sweep cycle, the absorbed signal and reference signals from reference structures A and B are simultaneously acquired. The reference signals are used as reference event sequences to calculate event density, event interval consistency, and noise level indicators. In the event of fringe loss, saturation, or light outage, the update or degradation mode is frozen to prevent abnormal references from accessing the correction link. Based on the two reference event sequences, the sweep mapping function parameters and reference drift parameters are jointly estimated. Differential constraints are established by utilizing the different sensitivities of reference structures A and B to the same sweep mapping to separate the sweep nonlinear error from the reference drift. The estimated values ​​are displayed explicitly to avoid the single-reference closed loop absorbing the reference drift into the spectral axis.

[0018] The candidate sweep frequency mapping function is subject to monotonicity constraints, multi-reference consistency constraints, and residual statistical thresholds. The threshold is calculated based on the noise level indication and ambient temperature. When the threshold is passed, the mapping function and reference drift parameters are frozen, so that the closed-loop update is unconstrained and error correction is written, which can adapt to temperature cycling, vibration and packaging stress disturbances.

[0019] The final sweep frequency mapping function generates the correction spectrum axis, and after resampling the absorption signal, the absorption spectrum is fitted or demodulated to output the concentration or content. The output health score is used as the trigger condition for write-back or freeze, so that the back-end resampling correction and the sweep frequency drive waveform pre-correction share the same mapping parameter set.

[0020] Dual-reference event-based observation, joint estimation, verifiable constraint gating, and conditional write-back form a closed loop, outputting health scores and trigger cause codes to output operation records. On-site location of reference link failures and pump link disturbances reduces manual calibration and maintenance intervention and increases the comparability of process analysis results. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a schematic diagram of the optical path and acquisition hardware composition of the present invention; Figure 3 This is a timing diagram illustrating the simultaneous acquisition of measurement signals and two reference signals, and the extraction of reference event sequences, according to the present invention. Figure 4 This is a schematic diagram illustrating the principle of joint estimation of frequency sweep mapping and reference drift based on dual reference increment constraints in this invention. Figure 5 This is a schematic diagram illustrating the verifiable constraint gating and freeze state machine determination process of the present invention; Figure 6 This is a schematic diagram showing the comparison of the spectral axis alignment effect and absorption curves before and after the correction of this invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figures 1-6 This invention provides a method for correcting nonlinear errors in an optically pumped laser, comprising: Step 1: Generate two different reference event sequences A with the same timing in each frequency sweep cycle. and reference event sequence B At the same time, the reference validity quality parameter is obtained, so that the subsequent joint estimation of frequency sweep mapping and reference drift has an observable boundary.

[0024] In this process, frequency sweep absorption measurement maps time-domain sampling points onto the wavenumber domain spectral axis. The scale information provided by the reference device causes its refractive index and geometry to change over time. Variations in industrial field temperature, vibration, and packaging stress can alter the refractive index and geometry of the reference device, causing the phase zero and fringe spacing of the reference fringes to change over time. If the acquisition link maps and estimates the amplitude variation of the reference waveform, detector saturation, and electromagnetic noise input, subsequent steps continuously update the residuals, absorbing the drift onto the spectral axis.

[0025] Step one uses a collaborative chain of physical spectral splitting, synchronous acquisition, event-based extraction, and quality labeling to deconstruct reference information from amplitude noise and slowly varying baselines. This information is then transformed into timestamped observations that can be directly constrained by subsequent estimators. The usability of the reference signal is then converted into quality parameters and freeze trigger conditions. The simultaneous existence of two references ensures that the same sweep wavenumber variation generates different event densities and spacing structures on the two links, allowing subsequent estimators to determine changes in the sweep map shape and the scale of the reference device, thus preserving observational redundancy.

[0026] In this system, the beam splitter, reference structure A, reference structure B, and three detectors form an optoelectronic link, while the data acquisition board and processor form an electrical signal link. The data acquisition board uses the start trigger signal of the frequency sweep cycle as a unified time reference, and the processor performs event extraction and quality marking on this time reference. In one embodiment, each of the three detectors is a photodiode and is connected to a transimpedance amplifier. The feedback resistor and compensation capacitor of the transimpedance amplifier are selected to maintain stable gain within the reference fringe frequency band, avoiding misinterpreting the self-oscillation of the analog front end as the reference fringe.

[0027] Field disturbances simultaneously affect the swept-frequency light source, reference device, and connection structure, causing coupled changes in the amplitude, phase, and arrival time of the reference signal. If the acquisition links are not synchronized in time, or if the differences between reference optical paths are insufficient, subsequent joint estimation will treat acquisition delay errors as spectral axis nonlinearity or reference drift, leading to confusion about the source of drift and introducing the risk of parameter write-back. On the other hand, if the coupling point between the reference optical path and the measurement optical path shifts slightly with vibration, the reference event sequence may experience sudden gaps or clustering. Therefore, the physical connection relationship and data flow relationship must be clearly defined during the acquisition stage.

[0028] A unified time reference is established at the start trigger point of the frequency sweep cycle. The frequency sweep output is simultaneously sent to the measurement optical path and two reference optical paths through a beam splitter structure, and the sampling or timestamp recording of three signals is completed under the same time reference. Reference structure A and reference structure B adopt different optical path differences or different free spectral ranges, so that they have different sensitivities to frequency sweep wavenumber changes and external disturbances. This difference is realized through device geometry and packaging boundary conditions.

[0029] In order to provide two non-overlapping scale information for reference structure A and reference structure B within the same sweep frequency cycle, the beam splitter optically divides the sweep frequency laser output into three paths: the measurement beam, reference beam A, and reference beam B, and mechanically limits the coupling position and stress path of the three beams.

[0030] In one embodiment, the beam splitter uses an optical fiber coupler, with the input end and the output end of the swept frequency light source connected by fusion splicing to reduce insertion loss fluctuations caused by end-face contamination. In another embodiment, an integrated waveguide coupler is used, with the coupling region and the reference structure A formed on the same substrate, shortening the length of the external optical fiber to reduce microbending loss introduced by vibration.

[0031] In a preferred embodiment, the beam splitting structure employs a single 1-to-3 fiber coupler, with the measurement optical path accounting for a larger proportion than the two reference optical paths. The two reference optical paths have equal proportions, ensuring that the operating points of the reference detectors are consistent. The three output fibers are preferably single-mode fibers, wound with a fixed radius and then constrained by a flexible clamp to keep the force point away from the coupler body. If polarization-maintaining fibers are used, the beam splitting structure and the input end of the reference structure are assembled along the same principal axis direction, reducing the fringe visibility changes caused by polarization state fluctuations.

[0032] The reference difference is manifested through the difference in optical path or free spectrum between reference structure A and reference structure B. For example, reference structure A implements a Mach-Zehnder interferometer with a planar optical wave circuit, whose waveguide is formed by a silicon dioxide cladding and an enthalpy-doped silica core layer, placed on a silicon substrate; reference structure B implements a fiber Mach-Zehnder interferometer, with the two arms forming optical paths of different lengths from the same single-mode fiber, fixed on a metal support, and the two arms have different thermal and stress boundaries. In another parallel implementation method, reference structure B uses a Fabry-Perot interferometer cavity housed in a metal sleeve, with a temperature sensor added to the outer wall of the sleeve for the sensor's state quantity.

[0033] In use, the beam splitting structure simultaneously sends the same sweep frequency output into the measurement optical path and two reference optical paths, making the measurement signal and the two reference signals from the same source; the reference difference causes the two references to present different event structures, providing an information basis for subsequent drift separation; mechanical limit and assembly direction control reduce the sudden changes in the reference signal caused by the drift of the coupling point.

[0034] To ensure that subsequent steps align the reference event and the measurement sampling point to the same timeline, the data acquisition board uses the sweep cycle start trigger signal as a unified time reference and initiates sampling or timestamp capture of the three electrical signals at the moment the trigger arrives. The trigger signal is provided by the synchronization pulse output from the sweep drive board. The pulse edge simultaneously enters the sweep drive circuit and the trigger input terminal of the data acquisition board, binding the start of drive with the start of acquisition.

[0035] The data acquisition board has an internal clock source. The timestamp counter is cleared or latched when the trigger arrives, so that all events within the same sweep cycle are represented by the same counting base. When the timestamp counter reaches the maximum count value, the overflow flag is set. After the processor reads the overflow flag, it marks the cycle as invalid and sets the freeze trigger condition.

[0036] In a preferred embodiment, the data acquisition board uses a multi-channel analog-to-digital converter to synchronously sample the outputs of the measurement detector, the first reference detector, and the second reference detector. The three analog front-ends use the same transimpedance amplifier topology to maintain consistent phase delay. In another alternative embodiment, the two reference signals are converted into square waves by a Schmitt trigger comparator, and the data acquisition board only timestamps the flip edges of the square waves. The measurement signal is sampled by the analog-to-digital converter and written to a circular buffer. The circular buffer is set with a full flag and an overwrite flag. When the overwrite flag is set, the processor marks the period as incomplete data and outputs a freeze trigger condition.

[0037] Taking a rack-mounted natural gas moisture analyzer as an example, the beam splitter and two reference detectors are fixed on the same metal base plate, and the measuring chamber is connected to the beam splitter via optical fiber. After the control board sends a sweep cycle start trigger signal, the data acquisition board begins sampling the detector voltage and writes the flip edges of the two reference comparators into the event buffer. At the end of the sweep cycle, the processor reads out the event buffer, forming reference event sequence A. With reference event sequence B And check if the event list is complete. Is the cover flag set? Whether the overflow flag is set is written to the operation log; maintenance personnel can confirm whether the reference channel link is connected through the panel indicators.

[0038] As a supplement: when using an analog-to-digital converter for synchronous sampling, the sampling time... It can be determined by the sampling frequency With sampling point number Here it is: ; Where: sampling time : Measurement signal or reference signal The time of each sampling point, the value range is Sampling frequency The sampling frequency of the analog-to-digital converter, with a value range of [value range missing]. The configuration register of the data acquisition board is set and can be read back; the sampling point sequence number. : Sampling point index, which is a positive integer and satisfies Frequency sweep cycle duration The duration of one frequency sweep cycle, with a value ranging from 1 to 2. ; Meanwhile, the synchronized time base aligns the reference event with the measurement sampling point, avoiding the misinterpretation of spectral axis errors due to later sampling delays; synchronized sampling or timestamp capture fixes the source of inter-channel drift as circuit delay, and the relative relationship is observed to be stable within a week; buffer integrity and overflow flags directly provide triggering conditions for freeze updates.

[0039] Reference signals are often simultaneously affected by amplitude fluctuations, detector saturation, scattered light interference, and polarization drift in the field. Directly using the reference waveform for fitting introduces amplitude noise into the wavenumber mapping, making subsequent estimations sensitive to slowly varying baselines. Only by converting the reference waveform into event moments that are insensitive to amplitude, and marking the integrity and consistency of the event sequence, can subsequent steps freeze updates when reference anomalies occur. Since the reference device is sensitive to temperature perturbations, reference anomalies can manifest as abrupt changes in event intervals or short-term clustering; therefore, quality marking is preferentially constructed from the event time sequence structure.

[0040] The first and second reference signals are conditionally processed to meet the event extraction prerequisites in terms of level, bandwidth, and edge jitter; then, the reference event sequence A is extracted from the conditionalized signals. With reference event sequence B Then, from reference event sequence A... With reference event sequence B Construct quality parameters and output them along with the ambient temperature status; if the quality parameters do not meet the minimum requirements, set the freeze trigger flag and output it to the next step.

[0041] The reference signal conditionalization process is mainly based on DC removal, amplitude adjustment, bandwidth adjustment, and edge stabilization. The reference signal is DC removed by capacitor blocking or digital de-averaging, then linearized by limiting or automatic gain control, then bandwidth-limited by a high-pass and low-pass filter, and finally enters a Schmitt trigger comparator.

[0042] The upper and lower thresholds of the Schmitt trigger comparator are determined by the self-updating peak-to-peak value of the reference signal within one sweep cycle, so that small noise near the zero crossover does not cause multiple flips.

[0043] When the reference signal is sampled using the full waveform, the processor performs sign change detection on the sampling sequence and performs linear interpolation on adjacent sampling points that cross zero levels to obtain the zero-crossing time. Let the adjacent sampling time be the sampling time. With sampling time The corresponding sampled value is the sampled value. With sampled values Zero crossover time Satisfy the following formula: ; Where: zero crossover time : Indicates the moment when the reference signal crosses zero level, used to form a reference event sequence, and its value range satisfies Sampling time : indicates the first The sampling time, being a real number and falling within one frequency sweep period, is used to provide interpolation endpoints; the sampling time... : indicates the first The time of each sampling point is a real number and satisfies the sampling time difference. A positive real number, used to provide interpolation endpoints; Sample value : Indicates the sampling time The corresponding reference amplitude, a real number, is used to determine the zero-crossing direction and interpolation ratio; sampled value : Indicates the sampling time The corresponding reference amplitude is a real number and is related to the sampled value. Interpolation is triggered when the signs are opposite, used to determine the interpolation ratio; index. : Represents the sampling point index, which is a positive integer and satisfies , used to identify adjacent sampling points within the same sweep frequency period; When the denominator is close to zero or the sampled values ​​have the same sign, the processor does not calculate the zero-crossing moment. This segment is marked as invalid to avoid numerical amplification; when the reference signal uses a comparator plus a timestamp, the data acquisition board directly records the flip-edge time. To allow the conditionalization effect to be utilized by subsequent gating, the processor constructs a variance-independent noise level indicator. This is used to describe the relative amplitude of rapid fluctuations of the reference waveform within the window, and its calculation formula is: ; Where: Noise level indication : Represents the proportion of the relative amplitude of rapid fluctuations in the reference waveform, used to determine whether electromagnetic interference or comparator jitter is significant, with a value range of . ; Sample value : indicates the first [number]th ... The amplitude of each reference sampling point is a real number, and its range is determined by the analog-to-digital converter's range. It is used to calculate the difference and accumulation; the number of samples... : Represents the number of sampling points within the window, which is a positive integer and satisfies The window covers multiple stripe periods to avoid random points dominating; index : Represents the sampling point index within the window, which is a positive integer and satisfies , used to traverse the summation range and construct the difference; After the event extraction is complete, refer to event sequence A. With reference event sequence B The events are sorted in ascending order by time and accompanied by an event number. If an event time is not incremented or a duplicate time is detected, the processor marks the cycle as an event sequence abnormality and sets the freeze trigger flag.

[0044] In practice, the conditional chain separates saturation clipping, slowly varying baselines, and high-frequency interference from the reference information, making event extraction primarily dependent on phase flips; linear interpolation identifies zero-crossing moments. Released from the sampling grid, allowing for finer temporal resolution of event moments; noise level indication. The event sequence anomaly flag provides input for subsequent threshold determination.

[0045] Quality parameters are obtained by referencing event sequence A. and reference event sequence B The sequence is represented as follows. The event count for each reference channel within one sweep cycle is multiplied by the sweep cycle duration to obtain the event density. This density is used to determine if stripes are lost or saturated, resulting in a decrease in the number of flips. The calculation formula is: ; Where: event density The frequency of reference events occurring per unit time is used to measure the integrity of the reference fringe, and its value is [value missing]. Event count The number of events detected within one frequency sweep cycle, which is a non-negative integer and satisfies ,when The reference link is determined to be out of power or severely saturated at any time; the frequency sweep period duration is also considered. The duration of one sweep cycle is a positive real number and satisfies The frequency sweep trigger interval or driver board parameters are given. To identify event clustering caused by vibration spikes or interval lengthening caused by event loss, the processor calculates the event interval and constructs an interval consistency coefficient, which is calculated as follows: ; Where: Consistency coefficient The event interval uniformity index is used to identify crowding and jumps, and its value is [value missing]. Event interval The time difference between two consecutive events is a positive real number and satisfies the following conditions: ,index The range of values ​​is ; The processor will handle event density Consistency coefficient Noise level indication The ambient temperature status is written into the runtime log, and a freeze trigger flag is generated: event density. Below the minimum threshold, or consistency coefficient Below the minimum threshold, or noise level indication When the value exceeds the maximum threshold, the freeze trigger flag is set; when both references have the freeze trigger flag set, the degradation state is set and output to the next step. The ambient temperature status is taken from the base plate temperature sensor or a temperature sensor near the reference structure, and is used to distinguish between gradual temperature changes and sudden disturbances when calculating the threshold or selecting weights in the next step.

[0046] When used, event density Converting stripe loss, light cutoff, and saturation clipping into determineable quantities triggers freeze; consistency coefficient Clustering and jumps are converted into single coefficients for rapid identification of anomalous references; noise level indication. The output is synchronized with the ambient temperature status to ensure that the subsequent gating caliber does not change with the operating conditions.

[0047] Two references are used to obtain reference event sequence A based on rule-based conditionalization. With reference event sequence B The reference information is transformed from the amplitude domain to the time domain; event density Consistency coefficient With noise level indication Merge them into quality tags and enable freeze preset to prevent abnormal observations from entering the update parameters; output the two reference quality tags in parallel, and the next step is to distinguish between single-path and dual-path anomalies.

[0048] Step 2: Within a unified time reference of a single frequency sweep cycle, simultaneously determine the candidate frequency sweep mapping function. The candidate reference drift parameters enable the frequency sweep nonlinearity error to be distinguished from the reference drift error during the solution process.

[0049] The fringe events generated by reference structures A and B provide a wavenumber increment scale. Industrial disturbances can alter the equivalent optical path difference of the reference devices, causing the scale to shift over time; if only a single-path reference is used, this scale shift will be hidden in the mapping result. Step two uses the drift coefficient of reference structure A to represent the reference scale change. Drift coefficient with reference structure B The expression is used, and by utilizing the different sensitivities of the two references to the same frequency sweep map, two sets of incremental constraints are constructed, thus obtaining candidate values ​​for the shape and scale changes of the map within the same solution framework. The event density given in step one... Consistency coefficient With noise level indication This step is used as a basis for weighting and ill-conditioning, thereby enabling candidate solutions to have a controllable backoff path for fringe loss, clustering, and electromagnetic interference.

[0050] The reference event sequence has been eventified in step one. Furthermore, the wavenumber increment of adjacent events is used as the constraint unit to avoid directly bringing the zero-point drift of the reference absolute phase into the spectral axis mapping, thus leaving degrees of freedom for the explicit solution of the reference drift parameters.

[0051] The processor first scans the frequency cycle duration. Internally establish sweep frequency time Local coordinates and parameterized sweep frequency mapping function Then refer to event sequence A. With reference event sequence B The event time sequence is converted into a constraint equation using the incremental relationship between two adjacent events, which determines the drift coefficient of the reference structure A. Drift coefficient with reference structure B Enter the same constraint system.

[0052] The processor will trigger the frequency sweep time at the start of each frequency sweep cycle. Set to zero, and during the frequency sweep period Internal incremental timing. To ensure the mapping shape is continuous and easy to differentiate, this implementation uses a cubic polynomial to represent the frequency sweep mapping function. And the mapping coefficients are used as variables to be estimated: ; Where: wavenumber mapping function Time-to-wavenumber mapping, used to generate spectral axis coordinates; frequency sweep time. The time from the start of the cycle to zero, with a value range of [value missing]. Frequency sweep cycle duration : Duration of the period, with a range of values. , used to limit Domain; mapping coefficients : Constant term coefficients, used to indicate the initial wavenumber bias; mapping coefficients : Linear term coefficient, used to give the principal slope; mapping coefficient : Quadratic coefficients, used to describe nonlinear bending; mapping coefficients : Cubic coefficients, used to describe higher-order nonlinear shapes; To avoid abrupt changes during the cycle, the processor inherits the initial values ​​of the mapping coefficients for the current cycle from the gated mapping coefficients of the previous cycle and writes the inheritance flag into the runtime record; during the startup phase, the initial values ​​of the mapping coefficients are taken from the factory calibration values. This endpoint inheritance is reflected in the solution as the mapping coefficients... With mapping coefficients The steady-state constraints ensure that candidate solutions under strong vibrations do not deviate from the historical trajectory.

[0053] In practice, the cubic polynomial constrains the mapping shape to a continuous and differentiable family, facilitating the stable application of event increment constraints; endpoint inheritance ensures continuous mapping during the cycle, reducing parameter jumps caused by sudden disturbances. Mapping coefficients With mapping coefficients The updates are driven by event constraints, reducing the amount of reference noise written into higher-order nonlinearities.

[0054] The processor will refer to event sequence A The reference event time A is analyzed as increasing. Refer to event sequence B The reference event time B is analyzed as increasing. For a Mach-Zehnder interferometer that uses zero-crossing as the event, the phase increases for adjacent events. Therefore, the wavenumber increment of adjacent events is equal to the phase increment divided by the equivalent optical path difference.

[0055] Reference structure A has a nominal optical path difference. Drift coefficient with reference structure A This represents the equivalent optical path difference, while reference structure B uses the nominal optical path difference of reference structure B. Drift coefficient with reference structure B This represents the equivalent optical path difference, thus yielding two sets of incremental constraints: ; Where: Event phase increment : The phase increment corresponding to adjacent events of reference structure A, with a value range of . When the event is defined as zero crossover, take... When the event is defined as a peak, take Event phase increment The phase increment corresponding to adjacent events in reference structure B has a range of values. The rules for determining the value are the same as above.

[0056] Wavenumber mapping function Same as the previous formula, used to calculate the wavenumber difference corresponding to the event; refer to event time A. Reference Structure A At any given moment, satisfying Reference event time A Reference Structure A Each event moment is used to form adjacent increments; Reference structure A nominal optical path difference The factory-calibrated equivalent optical path difference has a range of values. ; Drift coefficient of reference structure A The relative scale change, with a preferred range of... Event number : Event index, a positive integer, used to traverse adjacent event pairs; And reference structure B satisfies: ; Where: wavenumber mapping function Same as above; refer to event time B. Reference Structure B At any given moment, satisfying Reference event time B Reference Structure B At each event moment; nominal optical path difference of reference structure B The factory-calibrated equivalent optical path difference has a range of values. ; Drift coefficient of reference structure B The relative scale change, with a preferred range of... Pi Same as above; event number : Event index, a positive integer, used to traverse adjacent event pairs; The nominal optical path difference is determined by the geometry and refractive index of the reference device and written into non-volatile memory after factory calibration; the drift coefficient obtained during operation reflects the scale changes caused by temperature and stress. Preferably, the nominal optical path difference of reference structure B is... Nominal optical path difference with reference structure A The ratio falls between 2 and 10, ensuring that the event densities of the two paths are different yet not too sparse, thus providing separable sensitivity for subsequent solutions. Since the constraints are in the form of adjacent event increments, the absolute phase zero-point drift of the reference device does not enter the right-hand side of the equation, thereby reducing the direct coupling of zero-point drift to the mapped shape.

[0057] In use, two sets of incremental equations simultaneously constrain a mapping shape and two reference scales. The reference drift is explicitly represented and hidden in the mapping. The incremental form removes the phase zero-point drift, which facilitates simultaneous solution. With both constraints in effect, the solution space for the mapping shape and scale drift is finite, which facilitates the formation of candidate parameter sets.

[0058] In addition, event density As stripe visibility and occlusion vary, vibration peaks exhibit clustering and gaps. The processor solves all parameters simultaneously once a week, writing mapping coefficients when values ​​become unstable. A cost function is constructed at the cost of incremental error, and the impact of anomalous events is limited by weights and backtracking paths, ensuring physical consistency of candidate results.

[0059] The processor calculates the wavenumber increment for each adjacent event pair based on the incremental equation and constructs a cost function with the accumulation of absolute error as its core. Then, it adopts an alternating update strategy, first updating the mapping coefficient under the condition of fixed drift coefficient, and then updating the drift coefficient under the condition of fixed mapping coefficient. After each round of update, the cost function and consistency deviation are calculated. When the ill-conditioned condition is triggered, the processor outputs the hold value and records the reason for direct judgment by the gating in step three.

[0060] Specifically, the processor denotes the coefficient matrix composed of event differences as the coefficient matrix. When the coefficient matrix condition number Greater than the condition number threshold Timely determination of pathological conditions and placement of backoff flags; condition number thresholds The device parameters are written to non-volatile memory. Press The norm is preferably chosen as the second norm. When it is inconvenient to invert the platform, the ratio of the maximum singular value to the minimum singular value can be obtained by singular value decomposition as the condition number.

[0061] The processor records the wavenumber increment of adjacent events of reference structure A as wavenumber increment A. The wavenumber increment of adjacent events in reference structure B is denoted as wavenumber increment B. To make the noise level indication... High or consistency coefficient The impact of lower-order segments on the solution is controlled; the processor assigns a weighting factor to each incremental term and uses the cumulative weighted absolute error as the cost function. ; Where: cost function : Error weighted cumulative amount, with a value of Used to drive parameter updates; weight factor A Channel A, No. Each incremental term has a weight, and its value is... ,Depend on and Mapping yields weight factor B. Channel B, No. The weights of each incremental term, with a value range of [value range missing]. ,Depend on and Mapped to obtain; Wavenumber increment A Channel A: Wavenumber difference between adjacent events, used for comparison with theoretical increment; Wavenumber increment B : Wavenumber difference between adjacent events in channel B, used for comparison with theoretical increment; Nominal optical path difference in reference structure A. Reference structure B nominal optical path difference , Drift coefficient of reference structure A , Drift coefficient of reference structure B Pi : Consistent with the previous statement; Event number Event number : Index of adjacent event pairs, which is a positive integer; ; Where: weighting factor Channel A, No. The weights of each incremental term, with a value range of [value range missing]. Consistency coefficient Same as above; lower limit of consistency coefficient : The weight used to suppress clustering / jumping segments, with a value of Upper limit of consistency coefficient : Used to determine the upper bound of sufficiently uniform intervals, with a value of Noise level indication Same as above; lower limit of noise level Used to determine low-noise sections, with a value range of [value range missing]. ; upper limit of noise level Used to identify high-noise sections, with a value range of [value range missing]. ;function , : Used for lower bound clipping and upper bound clipping respectively, so that the weight factor falls within Inner; Channel B weighting factor Using the same mapping format, only the index is replaced. ; The processor binds the weighting factor to the quality label: when the event density... If the coefficient is below the threshold, all weight factors are set to zero and the system immediately enters a backoff state; when the consistency coefficient... Below the threshold, only weighting factors for consecutive intervals are retained; when the noise level indicator... When the value exceeds the threshold, the upper limit is set at a preset value, as this segment contributes little to the updating of the mapping coefficients. Alternating updates employ a linear solution method with a fixed number of iterations and a fixed step size. The mapping coefficients are updated within a sliding window using a recursive least squares solver. To mapping coefficients The drift coefficient update is based on a one-dimensional search using the monotonicity of the cost function with respect to the drift coefficient, and the drift coefficient of reference structure A is updated. and the drift coefficient of reference structure B Cut off to the preferred range.

[0062] Taking a rack-mounted natural gas hydrogen sulfide analyzer as an example, the processor reads reference event sequence A after each frequency sweep cycle. With reference event sequence B Then calculate wavenumber increment A With wavenumber increment B And update the mapping coefficients To mapping coefficients When on-site vibrations increase event density When the frequency decreases, the processor does not write new coefficients, but maintains the coefficients of the previous cycle and outputs a hold status word on the interface; when the vibration disappears and the event density recovers, the processor starts writing candidate coefficients again, and the measurement signal buffer is then resampled according to the new spectral axis.

[0063] When used, the cost function suppresses outlier events by accumulating absolute error, and the weight mapping embeds quality labels into the solution to control the contribution of outlier segments; alternating updates decompose the coupled problem into low-dimensional updates, which is convenient for deployment on platforms with limited computing power.

[0064] To ensure that step three has a direct criterion, the processor generates a consistency deviation and an ill-conditioning flag after each update. The consistency deviation is obtained by comparing the theoretical increment ratio of the two references with the factory calibration ratio, where the factory calibration ratio is the nominal optical path difference of reference structure A. Nominal optical path difference with reference structure B The ratio is calculated and stored in non-volatile memory; the processor simultaneously calculates the current increment ratio during operation, and the difference between the two is the consistency deviation. Ill-conditioned rollback follows a causal chain: if the event density... Consistency coefficient Or noise level indication At any trigger threshold, the processor immediately outputs the mapping coefficient and drift coefficient passed through the gate in the previous cycle as the hold value, and sets the backoff flag; if the cost function If the value does not decrease within the limited number of iterations, the processor will shorten the sliding window and recalculate once. If the condition is still not met, it will back off.

[0065] In one implementation, candidate outputs are written into shared memory as fixed-length data frames. These data frames contain sweep cycle numbers, mapping coefficients, drift coefficients, and a cost function. Fields, consistency deviation field, and backoff flag field; in another implementation, these are provided to the subsequent unit for reading in the form of a register set, and latched when the frequency sweep is triggered.

[0066] In use, the consistency deviation transforms the consistency of the two reference scales into a determinable quantity, facilitating direct gating in step three; the backoff output uses the held value as a safety baseline, ensuring that the spectral axis does not drift with abnormal observations during strong disturbances; the two encapsulation methods, data frame and register group, provide equivalent implementation paths for different hardware platforms. (Based on the cost function...) The alternating update process generates candidate mapping coefficients and drift coefficients, and backs down the output to maintain the values ​​when the quality flag is triggered. The consistency deviation and back-down flag provide verifiable inputs for step three, giving the gating rejection of updates a clear basis.

[0067] Step 3: Map the candidate frequency sweep function within each frequency sweep cycle. Physical consistency and dual-reference consistency constraints are imposed on candidate reference drift parameters, so that the update action only occurs when the observation is reliable and the solution converges, thereby avoiding abnormal reference events from taking the parameter set away from the implementable boundary.

[0068] Step two has already set the drift coefficient of reference structure A. Drift coefficient with reference structure B Explicitly incorporate it into the solution and output the cost function. With backoff flags; however, reference fringe loss, event clustering, comparator jitter, and analog front-end saturation caused by industrial disturbances can occur concentratedly in certain sweep cycles, making the cost function Smaller values ​​can still be obtained within a short window, but the corresponding mapping shape does not satisfy physical monotonicity, or the ratio of the two reference increments deviates from the factory calibration relationship.

[0069] In step three, the monotonicity constraint, the joint constraint of residuals and quality, the multi-reference consistency constraint, and the single-chain action of the frozen state machine nest the quality parameters of step one and the solution output of step two into a single decision logic, thereby obtaining a reproducible update criterion. Its characteristics are that all decision indicators are obtained from the acquired event times and mapping coefficients, and each decision has clear boundary conditions and output flags.

[0070] The data acquisition board generates the reference event sequence A in steps one and two. Reference event sequence B Event density Consistency coefficient With noise level indication This is considered an input in this step and will not be generated again in this step.

[0071] Candidate mapping coefficients To mapping coefficients Derived from the event increment constraint, if the reference event experiences clustering or absence within a short period, the mapping coefficients will tend to compensate for local errors with higher-order terms, thus producing non-physical backtracking; simultaneously, if the backtracking flag is not set but the cost function The stagnation during the iteration process indicates that the candidate parameter set is driven by noise rather than constraints, and continuing to write will push the endpoint inheritance of subsequent cycles in the wrong direction.

[0072] First verify the monotonicity of the mapping shape, then apply the cost function. It is used in conjunction with quality parameters for boundary screening to ensure that candidate solutions entering the multi-reference consistency determination are in the feasible domain.

[0073] The processor reads the candidate mapping coefficients for the current cycle. To mapping coefficients Cost function Set a rollback flag and simultaneously read the event density. Consistency coefficient With noise level indication First calculate the frequency sweep mapping function. Frequency sweep cycle duration The slope sign within the function is given, and the monotonicity determination result is provided. Then, based on the backoff flag and the cost function... Construct a valid determination result; if either of the two determinations is not satisfied, set the freeze trigger flag and mark the candidate parameter group for this cycle as the source of the retained value.

[0074] Candidate sweep frequency mapping function When using the cubic polynomial form, whether the mapping shape shows a reversal can be directly determined by the sign of the derivative. The processor's frequency sweep cycle duration... Calculate the derivative within the domain of the expression and use the sign of the derivative as a criterion for monotonicity. Mapped derivative. Given by the following formula: ; Where: wavenumber mapping derivative : Represents the rate of change of wavenumber caused by changes in sweep time, used to determine whether the mapping is monotonic, and its value range is real numbers; mapping coefficient : The coefficient of the first term, used to determine the principal slope, is a real number; the mapping coefficient : Quadratic coefficient, used to determine the degree of curvature, is a real number; mapping coefficient : The coefficient of the cubic term, used to determine higher-order bending, takes the value of a real number; Sweep time The time from the start of the cycle to zero, with a value range of [value missing]. Frequency sweep cycle duration : Duration of the period, with a range of values. , used to limit Scope; The processor will exist The internal discretization is divided into a set of coverage points, with the interval between the coverage points consistent with the sampling interval of the measurement signal, ensuring that the monotonicity determination is consistent with the subsequent resampling grid; the derivative is calculated at the coverage points using the Horner method. If the wavenumber mapping derivative is at any coverage point... If the value is negative, it indicates that the candidate mapping function in this cycle has a backtracking behavior. The freeze trigger flag is set, and the monotonicity failure flag is written to the status register. To prevent boundary noise from causing misjudgments of isolated points, the processor jointly judges the derivative signs of three consecutive covered points. The monotonicity failure flag is only set when negative derivatives appear consecutively, thereby distinguishing between occasional interpolation errors and structural backtracking.

[0075] In practice, derivative discrimination transforms the mapping reversal from a fitting residual phenomenon into a computable physical consistency condition, aligning it with the measurement sampling grid to provide a reproducible basis for the freeze trigger flag. Discrete coverage and continuous determination suppress misjudgments of isolated points, making monotonicity constraints more robust to vibration spikes.

[0076] Monotonicity passing does not necessarily mean the candidate parameter set is reliable, because the reference event sequence A... With reference event sequence B Even with missing fringes, a small number of events still occur, affecting the cost function. The processor therefore obtains smaller values ​​for a few incremental terms. Together with the quality parameters from step one, they are used for boundary screening, and the screening sequence follows a causal chain of observation followed by solution.

[0077] The processor first reads the event density. When event density When the value is below the minimum threshold, it is considered that the observations in this cycle are insufficient to support parameter updates. The freeze trigger flag is set directly, and the rollback flag is used as the source of the hold value. Subsequently, the noise level indicator is read. When the noise level indicator When the value exceeds the maximum threshold, it is assumed that comparator jitter or electromagnetic interference is dominant, and the candidate cost function for this period is determined. The decrease does not reflect physical consistency; therefore, the freeze trigger flag is set and the noise failure flag is written to the status register; finally, the consistency coefficient is read. When the consistency coefficient When the threshold is lower than the minimum threshold, it is considered that the event interval has clustered or jumped, the risk of higher-order compensation of candidate mapping coefficients increases, and the freeze trigger flag is also set.

[0078] The processor only uses the cost function when all the above quality parameters meet the threshold. The validity of the solution is determined. A two-stage rule is used for determining validity: the first stage compares the cost function of the current cycle. Cost function of gating from the previous cycle If the cost function of this period If the value increases significantly, the candidate solution is determined to deviate from the historical constraints, and the freeze trigger flag is set; the second stage reads the backoff flag. If the backoff flag is set, then regardless of the cost function... Both the size and the candidate parameter set are considered as sources of hold values. To make boundary screening more implementable, the processor uses the cost function passed through the gate in the previous cycle. Saved in the workspace and overwritten and updated each time a gate is accessed.

[0079] For significant increases: replace the ratio criterion with: ; Where: cost function ratio : Measures the ratio of the current period's cost function to the reference cost function, with a value range of . Cost function The cumulative error output in step two has a range of values. Reference cost function The cost function passed through the gating in the previous cycle has a range of values. It is stored in the status field of the valid parameter area; The gating rule is defined as follows: when the cost function ratio... Greater than the cost function threshold The freeze trigger flag is set at time, where the cost function threshold is... The device parameters are then written to non-volatile memory. This transforms gating from subjective judgment into executable rules.

[0080] When using quality parameters for priority screening, periods with insufficient observations and dominant interference are eliminated to prevent low-information updates from misleading updates. Cost function The backoff flag explicitly defines solution stagnation and ill-conditioned backoff, while the freeze trigger flag has the state source and the direction of endpoint inheritance during the cycle. A first-layer gating system is constructed through monotonicity constraints and priority screening of quality parameters. Candidate parameter groups entering the next layer of multi-reference consistency determination satisfy mapping shape consistency and sufficient observation information, and the freeze trigger flag is used as input to the state machine. Even if the candidate mapping function satisfies monotonicity and the quality parameters meet the threshold, there is still a situation where single-path reference drift is absorbed into the drift coefficient, causing the scale relationship given by the two references for the same mapping to deviate from the factory calibration relationship. If mapping coefficients and drift coefficients are still written in this case, the endpoint inheritance in step two will continuously carry the deviation into subsequent cycles, causing the spectral axis to shift during long-term operation.

[0081] With multi-reference consistency as the core constraint, and a frozen state machine to convert the gating results into explicit update and hold actions, step four always receives a set of parameters from a definite source.

[0082] The processor calculates the candidate drift coefficients without setting the freeze trigger flag. With candidate drift coefficients A predetermined scale ratio is determined, and this ratio is compared with the factory calibration ratio to obtain a consistency judgment result. If the consistency judgment fails, the freeze trigger flag is set and the system enters the freeze state machine. The freeze state machine takes the freeze trigger flag, the rollback flag, and the consecutive failure count as inputs, and outputs an update enable flag. Maintain the allow flag and health score, and write the mapping coefficient and drift coefficient into the corresponding register group for reading in step four.

[0083] The incremental theoretical values ​​for the two references are given by the nominal optical path difference and the drift coefficient, which have already been used in step two of the incremental equation. and Expressing equivalent optical path difference.

[0084] Furthermore, this expression is further refined into a scale consistency ratio, used to determine whether the scale relationship between the two references is consistent with the factory calibration. The processor calculates the scale consistency ratio using the following formula. : ; Where: Scale consistency ratio : Represents the ratio of the equivalent optical path differences between two reference paths, used to determine the relationship between the two reference scales, with a value range of . Reference structure B nominal optical path difference Factory calibration value, the range of which is: ; Drift coefficient of reference structure B The relative scale change, with a preferred range of... ; Reference structure A nominal optical path difference Factory calibration value, the range of which is: ; Drift coefficient of reference structure A The relative scale change, with a preferred range of... ; calibrating scale ratio The factory-calibrated scale consistency ratio has a range of values. The nominal optical path difference of reference structure B Nominal optical path difference with reference structure A Calculated; During the factory calibration phase, the output of the swept frequency light source is coupled to an absolute wavenumber measurement device (e.g., a wavelength meter) or to a reference gas cell with known spectral line positions; the swept frequency light source is then instructed to perform several swept frequency cycles, and reference event sequence A is acquired in each cycle. With reference event sequence B Get the event count and The start and end wavenumbers of the sweep frequency were obtained from the absolute wavenumber measurement device. and Calculate the periodic average wavenumber span ; For a Mach-Zehnder interferometer employing zero-crossing events, the event phase increment Pick Calculate the nominal optical path difference: The nominal optical path difference of reference structure A is ,in The nominal optical path difference of reference structure B is... ,in .Will and Write to non-volatile memory and load at runtime.

[0085] The processor calculates the calibration scale ratio during the factory calibration phase. The data is written to non-volatile memory and loaded into the register set upon power-on. Consistency determination uses absolute difference comparison: the processor calculates the scale consistency ratio. Compared with the calibration scale The difference is calculated and compared with a consistency threshold. If the difference exceeds the threshold, consistency is considered to have failed. To couple the consistency threshold with the ambient noise, the processor uses the noise level indicator... It is divided into three levels, each corresponding to a set of consistency thresholds; when the noise level indication... The strictest threshold is used at the lowest setting, when the noise level indicator is... The most lenient threshold is used when the highest threshold is reached, thus avoiding false triggering of consistency determination when noise is dominant. This three-level threshold table is written to non-volatile memory during the factory calibration stage, and only table lookup and loading are performed during field operation.

[0086] When in use, the scale consistency ratio Transforming the dual-reference relationship into a single computable quantity makes the consistency constraints independent of additional sensors. Calibration scale ratio. The three-level threshold table solidifies the boundary conditions for consistency determination into device parameters, and the processor only needs to perform arithmetic comparisons to give reproducible gating results.

[0087] Consistency determination failure, monotonicity failure, quality parameter failure, and setting the rollback flag will all cause the freeze trigger flag to be set, but the subsequent actions of these failures should not be exactly the same: refer to event density. During short-term declines, it is advisable to maintain the parameters of the previous cycle; oscillation spikes lead to a decrease in the consistency coefficient. During a short-term drop, it is suitable to wait for recovery before updating again. A set rollback flag indicates that step two has rolled back the held value, and step three requires encapsulating the held value into a stable output. Therefore, the processor sets up a frozen state machine, using the freeze trigger flag as a state transition condition, and using a consecutive failure count to determine whether to enter a degraded output.

[0088] The frozen state machine includes three states: update state, frozen state, and degraded state. In the update state, the update enable flag is set, and the processor processes the candidate mapping coefficients for the current cycle. To mapping coefficients With candidate drift coefficients , Write the values ​​to the mapping register set and drift register set, and clear the consecutive failure count to zero. In the frozen state, keep the enable flag set, and the processor does not overwrite the mapping register set and drift register set, only updating the failure source bit in the status register. In the degraded state, the processor sets the degrade flag and outputs a fixed source of the hold parameter. The degrade flag and hold parameter are sent to step four, so that step four takes a fixed path during absorption calculation. The consecutive failure count is updated at the end of each frequency sweep cycle: the count is incremented when the freeze trigger flag is set, and cleared when the freeze trigger flag is not set. When the consecutive failure count reaches the preset upper limit, the state machine transitions from the frozen state to the degraded state. The preset upper limit is written to non-volatile memory during the factory calibration stage and loaded by the processor during field operation.

[0089] The health score is based on the state machine output and uses discrete levels instead of floating-point weighting. The processor maps updated states to the highest level, frozen states to intermediate levels, and degraded states to the lowest level. Monotonicity failure, quality parameter failure, and consistency failure are each mapped to a three-bit flag. These three flags and the level together constitute the health score field. This field is written to the status register and sent to the host computer interface along with the output of step four, enabling the operations and maintenance team to distinguish when parameter updates have occurred. Parameter retention occurs and degraded output occurs.

[0090] Taking a fixed chemical tail gas ammonia monitoring cabinet as an example, the processor inside the cabinet reads the cost function at the end of each frequency sweep cycle. Consistency ratio with scale The comparison results were used to update the frozen state machine. After a filter replacement during a cabinet opening, the on-site personnel failed to tighten the fiber optic clamp, resulting in a slight bend in the reference optical path B, and an event density of [missing information]. The event density decreases over several cycles, and the processor enters a frozen state in the frozen state machine, while the mapped register set retains the value of the previous cycle; when the personnel re-fix the tablet, the event density... Upon recovery, the processor returns to the update state and overwrites the mapped register set in the next frequency sweep cycle. The cabinet's status indicator lights are driven by a degradation flag, illuminating when entering the degradation state, allowing personnel to locate the reference link anomaly without needing to analyze the waveform.

[0091] To prevent competitors from circumventing this by changing their hardware platforms, two implementation methods are parallelly implemented: one involves a microcontroller executing a frozen state machine and outputting the state through a register set; the other involves a field-programmable gate array (FPGA) executing the frozen state machine and outputting the state word, with frequency sweep triggering rising edge latching. The health score field remains consistent in both methods, without altering the parsing rules.

[0092] In use, the frozen state machine transforms the gating judgment result into definite write and hold actions, ensuring that the parameter set read in step four has a definite source, discrete health scores do not require floating-point calculations, and status output remains consistent across platforms. In an example, when the reference link is affected by mechanical assembly, the frozen state machine blocks the writing of abnormal observations to the mapping register set, maintaining continuous operation of the measurement link.

[0093] When applying, the scale consistency ratio is used. The second layer of gating is formed with the frozen state machine. When an error occurs in the dual reference scale relationship, the parameters are not updated. The gating result is converted into a state output through the health field and the degradation flag. Step four selects the absorption solution path and ensures that the spectral axis generates consistent input.

[0094] Step 4: Within the update region of the freeze trigger flag, utilize the gated wavenumber mapping function. The measurement signal is converted from the time axis to the wavenumber axis, and the absorption calculation of the target gas content is completed on the wavenumber axis. The parameter set in this cycle is written into the effective parameter area of ​​the next cycle according to the action closed loop of updating or maintaining.

[0095] In step one, the fringe waveforms of reference structure A and reference structure B are converted into reference event sequence A. With reference event sequence B And give the event density Consistency coefficient With noise level indication Step two involves mapping coefficients. To mapping coefficients Drift coefficient with reference structure A , Drift coefficient of reference structure B Solve the equations simultaneously to obtain and output the cost function. Step 3 compares the monotonicity constraint with the scale consistency ratio. This data is superimposed onto the frozen state machine to ensure that the parameter set entering step four has a definite source. Step four then converts the time-aligned sampling data into wavenumber-aligned sampling data and binds the measurement output and health score to the same output frame, enabling the host computer in the industrial field to synchronously read the status word when reading the measurement value. Simultaneously, step four writes the gated mapping coefficient and drift coefficient into the valid parameter area as the source of the initial value for the endpoint inheritance of the next cycle, avoiding drive mismatch caused by sudden parameter jumps.

[0096] As a supplement: lower limit of event density Used to determine stripe loss / light interruption; lower limit of consistency coefficient. Used to determine event clustering / jumping; upper limit of noise level. Used to determine whether jitter or electromagnetic interference is dominant.

[0097] Threshold determination method: During the factory calibration phase, maintain the sweep cycle duration. With sampling frequency Unchanged, multiple scan cycles are collected under mechanical vibration isolation environment. , , The distribution is determined, and a threshold that covers the normal operating fluctuation range is found. No changes are made to the field operation. In maintenance mode, the threshold register is updated through authorized instructions.

[0098] All steps are executed by the processor within the same measuring device; the measuring detector, measuring gas chamber, first reference detector and second reference detector have formed an acquisition link in step one, and the processor in step four only reads the measurement signal buffer index saved in step one, without changing the sampling trigger and optical path connection relationship.

[0099] The absorption signal output by the measurement detector is uniformly distributed at the sampling time, but the stepping of the frequency-sweeping light source on the wavenumber axis is not uniform. If the sampling sequence number is directly used as the wavenumber coordinate, systematic interpolation errors will occur at the steep edges of the spectral lines. Step three has already ensured the wavenumber mapping derivative. To provide a positive and updated or maintained action boundary, this boundary needs to be implemented as a set of wavenumber coordinates aligned point-by-point with the measurement signal. Stable resampling is then performed on the wavenumber coordinates so that subsequent absorption calculations are only affected by measurement noise and operating condition changes, and no longer subject to sampling distortion caused by frequency sweep nonlinearity.

[0100] After the frequency sweep cycle is completed, the processor reads the mapping coefficients from the mapping register set. To mapping coefficients The processor reads the update enable flag and hold enable flag output by the frozen state machine; when the hold enable flag is set, the processor retrieves the mapping coefficients passed through the gating in the previous cycle from the valid parameter area as the values ​​used in the current cycle.

[0101] The processor then uses the sampling time from step one as an index to calculate the wavenumber mapping function for each sampling time within the measurement signal buffer. The wavenumber values ​​are used to form a calibration spectral axis that corresponds one-to-one with the measurement sampling points. Finally, the processor resamples the measurement signal on a fixed wavenumber grid based on the calibration spectral axis and outputs the resampled absorption signal sequence, the corresponding wavenumber grid sequence, and the resampled valid segment flag.

[0102] The processor uses the measurement signal buffer written in step one as a sample queue arranged hierarchically at each sampling time, and latches the buffer's starting index and sweep cycle number to prevent read / write contention caused by the next cycle's acquisition overwriting. Next, the processor retrieves the wavenumber mapping function for each sampling time. To reduce the number of multiplications and fixed-point overflow, the Horner method is selected to obtain the wave value. When the mapping coefficients are stored in floating-point format, the floating-point format is converted to fixed-point format and the scaling factor is latched to keep the scaling factor unchanged throughout the entire period.

[0103] The wavenumber coordinates are generated using a cubic polynomial, similar to step two, i.e.: ; Where: wavenumber mapping function Sampling time The corresponding wavenumber coordinates are used to construct the correction spectral axis, and their values ​​are in the real number range; sampling time : Measurement signal buffer number The time of each sampling point, the value range is ; Mapping coefficients : The constant term of the wavenumber mapping function, taking values ​​in the real number range; mapping coefficients : The coefficient of the first-order term of the wavenumber mapping function, taking values ​​in the real number range; mapping coefficient : Coefficients of the quadratic term in the wavenumber mapping function, taking values ​​in the real number range; mapping coefficients The coefficients of the cubic term in the wavenumber mapping function take values ​​in the real number range. Frequency sweep cycle duration The duration of one frequency sweep cycle, with a value ranging from [value range missing]. ;index : Sampling point number, which is a positive integer and satisfies ; The processor generates a validity mask simultaneously when generating wavenumber coordinates: when a sampling time exceeds the limit, the analog-to-digital converter saturation flag is set, or the measurement signal buffer overlay flag is set, the corresponding sampling point is marked as invalid; the invalid mark is used to skip the point during resampling and form a valid segment flag at the output. To ensure strict alignment between the wavenumber coordinates and the measurement signal, the processor uses the same index for the wavenumber coordinate array and the measurement signal array. And write the sweep cycle number and the starting index of the latched buffer to the beginning of the array.

[0104] In practice, the wavenumber coordinates calculated at each point establish a one-to-one correspondence between the calibrated spectral axis and the sampling points of the measured signal, separating the effects of frequency sweep nonlinearity from subsequent fitting. Buffer latching and validity masking represent coverage, saturation, and out-of-bounds as explicit coverage, saturation, and out-of-bounds, giving the resampled input boundaries and reducing the propagation of erroneous samples.

[0105] Given that the monotonicity constraint in step three is met, the wavenumber coordinate sequence obtained increases with sampling time. Therefore, a monotonic scanning fixed wavenumber grid resampling method is selected. A wavenumber window is selected according to the measurement task. The wavenumber window consists of the area near the center of the target spectral line and the baseline regions on both sides. The boundaries are clipped using the minimum and maximum values ​​of the wavenumber coordinates during the sweep period to avoid excessive resampling.

[0106] The processor then generates a fixed wavenumber grid in the wavenumber window. In a preferred embodiment, the spacing between the fixed wavenumber grid points is selected to be 1-4 times the wavenumber span corresponding to the measurement sampling interval, so that the same grid point is sandwiched between two sampling points, reducing interpolation amplification.

[0107] Fixed wavenumber grid point spacing - preferred implementation method based on event density Segmentation settings: If event density If the event density is high and the noise level indication is low, then a small point spacing should be selected. Low, and noise level indication If the noise level is high, a larger dot spacing should be selected to avoid amplifying the noise during interpolation.

[0108] During resampling, the processor sets two pointers: the first pointer points to the current index in the measurement sampling point sequence, and the second pointer points to the current index in the fixed wavenumber grid. The processor incrementally scans along the fixed wavenumber grid and moves the measurement pointer until the wavenumber of a grid point falls between the wavenumbers of two adjacent sampling points. When the clamping relationship is established, the processor performs linear interpolation on the two clamped sampling points to obtain the resampled value of the grid point and marks the grid point as valid. If consecutive invalid sampling points occur, causing the clamping relationship to fail to be established, the processor marks the corresponding grid segment as a gap and encapsulates it at the output with the gap start and end indices, so that sub-step 402 can exclude this segment during fitting.

[0109] Another parallel implementation does not directly generate wavenumber coordinate sequences at sampling points, but instead performs a reverse solution for each fixed wavenumber grid point: the processor performs the frequency sweep during the frequency sweep time. From 0 to the duration of the sweep frequency cycle Wavenumber mapping function within the interval A binary search is performed until the wavenumber difference between the beginning and end of the search interval is less than a preset threshold. Then, neighboring samples are taken from the measurement signal buffer at the searched time point for interpolation. This reverse solution utilizes the monotonicity condition verified in step three to prevent multiple solutions from being generated during the search process. To reduce the shape compression introduced by linear interpolation in steep spectral segments, another parallel implementation uses monotonically cubic interpolation: the processor constructs piecewise cubic polynomials at four adjacent effective sampling points and applies a constraint to the polynomial slope to prevent overshoot, ensuring the interpolation curve remains monotonic on the wavenumber axis.

[0110] In practice, monotonic scan resampling limits the computation to a single traversal, enabling the processor to complete full-cycle resampling even with limited computing power. Gap encapsulation explicitly propagates invalid segments to subsequent fitting, preventing the fitter from misinterpreting gaps as absorption peak shape changes, thereby reducing solution drift under operating conditions.

[0111] Furthermore, the resampled absorption signal has been aligned to the wavenumber grid, but industrial environments still exhibit slow fluctuations in optical power, baseline changes caused by contamination of the measurement chamber, and linear variations due to pressure and temperature fluctuations. If the calculation process mixes these variations with spectral axis errors, content bias will occur during long-term operation and will be difficult to locate. The health score and update permission flag provided in step three provide state boundaries for the calculation process, enabling step four to both output measured values ​​and take hold actions when the state is not met, thereby avoiding parameter write-back that solidifies abnormal periods as endpoint inheritance for the next period.

[0112] The processor reads the output resampled absorbed signal sequence and wavenumber grid sequence, and reads the health score, update allow flag, maintain allow flag and degrade flag output in step three.

[0113] The processor first performs intensity normalization and baseline elimination on the wavenumber grid, then performs spectral model fitting within the target spectral window to obtain the target gas content and the fitting quality indicator. If the fitting quality indicator does not meet the threshold or the degradation flag is set, the processor marks the measurement value of this cycle as the source of the output and maintains the effective parameter range of the previous cycle unchanged. Subsequently, the processor encapsulates the measurement output, status word, and effective parameter update action into the same output frame and sends it to the host computer via fieldbus or serial interface. At the same time, it sends a parameter update completion or parameter hold handshake signal to the sweep frequency driver board, so that the driver selects the correct precalibration table when triggering the next cycle.

[0114] The fit quality indicator is defined as the cumulative absolute value of the residuals within the resampling window. : ; Where: Fitting quality indicator The cumulative absolute value of the residuals is used to determine the reliability of the fit, and its value range is... Resampled absorption signal : No. Normalized absorbed signal values ​​at each wavenumber grid point; model prediction values The first calculation based on the spectral line model Predicted value per wavenumber grid point; number of grid points The number of wavenumber grid points within the spectral window is a positive integer and satisfies the following conditions: ; The rollback rule is defined as follows: when Greater than the fitting threshold The time-set bit retains the output source bit and keeps the effective parameter range unchanged; fitting threshold The device parameters are set and written to non-volatile memory.

[0115] When performing baseline elimination on the wavenumber grid, the processor first selects the baseline regions on both sides of the target spectral line as reference segments, and performs a polynomial fitting on the reference segments to obtain the baseline curve. The fitting process uses the absolute error accumulation criterion and limits the polynomial order to first or second order to ensure that the baseline curve does not absorb the main peak of the spectral line. Subsequently, the processor uses this baseline curve to normalize the full-window signal, so that the slow fluctuations in optical power at different sweep cycles are eliminated in a multiplicative manner.

[0116] As a supplement: the baseline curve is taken in second-order polynomial form. ; in The baseline coefficients to be estimated are shown. Baseline fitting is performed only on the grid point set of the baseline segments on both sides of the target spectral line. This set is defined and written into the configuration at the factory with a fixed offset on both sides of the center of the target spectral line. The fitting criterion can be to minimize the cumulative absolute value of the residuals to avoid baseline absorption peaks.

[0117] After normalization, the processor establishes a spectral line model within the target spectral line window. The spectral line model preferably uses a Voigt or Lorentz shape, and the initial value of the line center position is inherited from the gated solution result of the previous cycle. When the allow flag is set, the processor only updates the line amplitude and not the line center position, thus limiting the sensitivity of the spectral line center to spectral axis perturbations within the update cycle. The spectral line fitting solver preferably uses a Levenberg-Marquardt solver. The processor sets upper and lower limits for each parameter to be estimated and writes the upper limit of the iteration count into non-volatile memory as a device parameter.

[0118] To ensure the fallback path is feasible, the processor calculates the cumulative absolute value of the fitting residuals after fitting is complete as a fitting quality indicator and compares it with a threshold. When the fitting quality indicator exceeds the threshold, the notch encapsulation covers the central segment of the target spectral line, or the degradation flag is set, the processor does not update the target gas content output. Instead, it outputs the target gas content that passed the gate control in the previous cycle and sets the hold output source bit in the status word for the current cycle. This hold output source bit is used on the host computer side to distinguish between the new solution value for the current cycle and the value reused from the previous cycle.

[0119] As a supplement: Normalized transmittance : Measurement intensity after resampling With baseline intensity Absorption for ; Target gas content In optical path length When the spectral line is known, it is obtained by fitting a spectral line model. An example model is as follows: ; Where: transmission intensity Measurement intensity after wavenumber mapping and resampling; baseline intensity : Background intensity without absorption obtained from baseline fitting; absorbance : The negative logarithm of transmittance, used to eliminate multiplicative power drift; target gas content The concentration or volume fraction to be estimated, with a range of values. ; Optical path length : Measure the effective optical path length of the air cell, with a value range of . The linear strength coefficient is determined by the device structure and written into the configuration file. The intensity coefficient of the target spectral line, with a value range of [value missing]. This is given by the device configuration file or written by the maintenance calibration; linear function : A normalized linear function that varies with wavenumber, at least determined by the wavenumber at the center of the line. With linear parameters The decision is made that the specific implementation of the linear function may be by lookup table interpolation or numerical approximation, and at least one implementation (e.g., a lookup table implementation of the Lorentz linear or Voigt linear) shall be disclosed in the specification. In practice, baseline elimination separates the slow fluctuations in optical power and the multiplicative changes caused by window contamination from the spectral shape, making the spectral fitting primarily respond to absorption characteristics rather than baseline drift. Inheriting the initial line center value and freezing the line center in a hold state reduces abnormal periods by writing spectral center drift into the solution path, thus making the long-term content output more stable. Failure rollback explicitly maps gaps and degradation states to hold output source bits, enabling the host computer to identify whether hold has occurred in the solution path.

[0120] The processor encapsulates the target gas content, health score, update permission flag, hold permission flag, degradation flag, and hold output source bits for the current cycle into a fixed-length output frame. It also writes the sweep cycle number into the output frame header to ensure the host computer can still determine the frame order during communication retransmissions. When the output frame is sent via the fieldbus, the processor appends a cyclic redundancy check field to the end of the frame and removes the current cycle's output frame from the transmission queue after receiving an acknowledgment frame. If the acknowledgment frame is not returned within a preset time, the processor retransmits the output frame while maintaining the driver handshake signal in hold mode to prevent communication anomalies from causing incorrect updates at the driver end.

[0121] The parameter write-back adopts a double-buffered structure: the processor sets up two storage areas, a valid parameter area and a candidate parameter area, and the candidate parameter area is written with the gating mapping coefficients in step three. To mapping coefficients Drift coefficient with reference structure A , Drift coefficient of reference structure B The valid parameter area stores the parameter set inherited by the endpoint in the next cycle. When the update enable flag is set, the processor copies the candidate parameter area to the valid parameter area after the frequency sweep cycle ends, and completes the copying before the start of the next frequency sweep cycle. When the enable flag is kept set, the processor does not overwrite the valid parameter area, but only updates the status word. To prevent the data from being read by step two during the copying process, the processor sets the mutex flag before copying and clears the mutex flag after copying. When step two reads the mutex flag setting, it directly uses the mapping coefficients latched in the previous cycle as the initial values.

[0122] When the device is configured for driver-side pre-calibration, the processor sends a new pre-calibration table to the frequency sweep driver board after the update enable flag is set for multiple consecutive cycles. The pre-calibration table is arranged by time index, with each time index corresponding to a drive voltage sample value; the processor uses the mapping coefficients of the valid parameter area when generating the pre-calibration table. To mapping coefficients The wavenumber coordinates are calculated, and the desired wavenumber sequence is mapped to the time index, thereby correcting the nonlinearity of the sweep drive waveform. After receiving the pre-calibration table, the driver board latches the table at the start of the next sweep cycle to avoid waveform breakpoints caused by switching within the cycle.

[0123] Taking a semiconductor process gas ammonia monitoring cabinet as an example, at the end of each frequency sweep cycle, the processor inside the cabinet writes the target gas content and status word into the output frame and sends it to the plant management system via Ethernet. The plant management system interface displays the content value and health rating level on the same line. After maintenance personnel replace the optical window, the baseline of the measuring chamber briefly changes, triggering the fitting quality indicator threshold. The processor will maintain the output source position and keep the effective parameter area unchanged, and the status word on the interface will change to "hold". After the maintenance personnel fix the window, the baseline is restored. In subsequent cycles, the processor restores and updates the allow flag and writes it into the effective parameter area. When the driver board triggers at the beginning of the next cycle, it latches the new pre-calibration table, the frequency sweep waveform remains continuous, and the measurement link can return to normal output without stopping.

[0124] During use, the measured value and status word are fixed into a single field. Combined with cyclic redundancy check and acknowledgment frames, the driver remains active even in the event of communication anomalies, preventing erroneous updates. Double-buffered write-back separates candidate and valid parameters, and the switching time between endpoint inheritance and pre-calibration table updates is fixed, avoiding changes in periodic parameters. The driver pre-calibration and back-end resampling use the same set of mapping coefficients, ensuring waveform correction and data processing are performed in the same coordinate system, avoiding implicit biases caused by differences between the driver and processing ends. Baseline elimination and spectral fitting are performed on the calibration spectral axis, and in cases of fitting failure or degradation, the output source bit and parameter hold actions are executed, preventing abnormal periods from being fixed as endpoint inheritance for the next period. Simultaneously, double-buffered write-back and driver handshake signals limit the update action to the sweep cycle boundary, allowing the driver waveform and back-end processing to share the same set of valid parameter areas.

[0125] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0126] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0127] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0128] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0129] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for correcting nonlinear errors in an optically pumped laser, characterized in that: include, At the beginning of the frequency sweep cycle, the absorption signal of the measurement detector and the reference signals of reference structure A and reference structure B are synchronously acquired and the reference signals are preprocessed to obtain the reference event sequence. Candidate sweep frequency mapping functions are obtained by jointly estimating the sweep frequency mapping function parameters from sweep time to wavenumber / frequency based on the reference event sequence and the reference drift parameters of reference structure A and reference structure B. Verifiable constraint gating is performed on the candidate sweep frequency mapping function according to monotonicity constraints, multi-reference consistency constraints and residual statistical thresholds. The final sweep frequency mapping function, reference drift parameters and health score are output. The update is frozen when the gating fails. The calibration spectrum axis is generated based on the final sweep frequency mapping function. The absorption signal is resampled and the absorption spectrum is fitted / demodulated to output the concentration / content results. The parameters of the final sweep frequency mapping function are written back or frozen in response to the health score. Preprocessing includes sequentially performing DC removal, amplitude limiting, bandpass filtering, and jitter reduction on the reference signal, and performing interpolation at the sign change points of adjacent sampling points to determine the zero-crossing time. The zero-crossing times are arranged in ascending order to form a reference event sequence and are aligned with the absorption signal in the same frequency sweep period. The preprocessing also includes counting the number of events in the reference event sequence within the frequency sweep period to obtain the event density, obtaining the event interval consistency based on the time difference between adjacent events, obtaining the noise level indication based on the differential amplitude of the reference signal sample values, and saving the event density, event interval consistency, noise level indication, ambient temperature, and substrate temperature together.

2. The method for correcting nonlinear errors in an optically pumped laser according to claim 1, characterized in that: Before synchronous acquisition, the swept laser output is coupled into a measurement optical path, a reference optical path A, and a reference optical path B via a beam splitter. The measurement optical path is coupled to the measurement detector to form an absorption signal, the reference optical path A is coupled to the reference structure A, and the reference optical path B is coupled to the reference structure B to form a reference signal.

3. The method for correcting nonlinear errors in an optically pumped laser according to claim 2, characterized in that: Synchronous acquisition includes synchronously sampling the absorption signal and the reference signal by a multi-channel analog-to-digital converter, locking the sampling time at the start of the frequency sweep cycle, writing the sampled values ​​of the absorption signal and the reference signal at the same sampling time into the same buffer index, and storing the buffer index and the reference event sequence together within the same frequency sweep cycle.

4. The method for correcting nonlinear errors in an optically pumped laser according to claim 2, characterized in that: Synchronous acquisition involves inputting the reference signal into a comparator to generate a flip pulse, and recording the flip time of the flip pulse as an event timestamp using a timestamp structure. The event timestamps are arranged in ascending order to form a reference event sequence, which shares the start time of the frequency sweep cycle with the absorption signal, and duplicate event timestamps are deleted.

5. The method for correcting nonlinear errors in an optically pumped laser according to claim 4, characterized in that: When the event density of reference structure A is lower than the minimum threshold, reference structure A is marked as a reference failure state and a freeze update is triggered; when the event density of reference structure B is lower than the minimum threshold, reference structure B is marked as a reference failure state and a freeze update is triggered; when reference structures A and B are in a reference failure state, the system enters a degradation mode and maintains the final sweep frequency mapping function of the previous sweep frequency cycle.

6. The method for correcting nonlinear errors in an optically pumped laser according to claim 5, characterized in that: The joint estimation adopts a recursive estimation process, updating the sweep frequency mapping function parameters in each sweep frequency cycle and updating the reference drift parameters in every predetermined number of sweep frequency cycles. The sweep frequency mapping function parameters include the segment boundaries and polynomial coefficients of the piecewise polynomial, and the predetermined number is a fixed number determined by the factory calibration.

7. The method for correcting nonlinear errors in an optically pumped laser according to claim 6, characterized in that: The threshold values ​​of the multi-reference consistency constraints and the residual statistics threshold in the verifiable constraint gating are determined at the beginning of the frequency sweep cycle based on the noise level indication and the ambient temperature, and remain unchanged during the frequency sweep cycle and are prohibited from being updated. The range of the threshold values ​​is determined by the factory calibration.

8. The method for correcting nonlinear errors in an optically pumped laser according to claim 7, characterized in that: In response to the health score meeting the write-back condition, the final sweep frequency mapping function parameters are written back for resampling correction in backend data processing, and the pre-correction update for sweep frequency driving waveform is also written back; in response to the health score falling below the write-back condition, the write-back is frozen and the final sweep frequency mapping function parameters written back in the previous sweep frequency cycle are retained.

Images