A dual-comb photoacoustic signal adaptive filtering system

Abstract

This invention relates to the field of filtering technology, specifically to a dual-comb photoacoustic signal adaptive filtering system, comprising a pulse calibration module, a phase calculation module, a trend recognition module, a filtering control module, and a result output module. In this invention, accurate identification of periodic components is achieved by acquiring the timing information of the main pulse and extracting symmetrical segments of the signal; dynamic monitoring of signal structural stability is achieved through phase angle calculation and abrupt change recognition; precise control of the filter weight adjustment rhythm is achieved by determining the status label based on the changing trend of the phase difference and setting an update rate factor; and the filter weight command output is controlled by logical judgment based on historical step size states, and the output continuity is verified, effectively improving the adaptability and stability of the filtering process. In complex backgrounds, it enhances the ability to preserve the characteristics of the target signal and suppresses the influence of non-stationary noise, thereby improving the overall signal-to-noise ratio and the reliability of the filter.

Description

Technical Field

[0001] This invention relates to the field of filtering technology, and in particular to an adaptive filtering system for dual-comb photoacoustic signals. Background Technology

[0002] The field of filtering technology involves the identification and suppression of useless or interfering components in signals, including signal feature extraction, spectrum analysis, noise cancellation, and adaptive filtering strategies. It is widely used in the real-time processing and analysis of various types of signals, such as images, audio, biomedical signals, and photoacoustic signals, playing a crucial role, especially in improving the signal-to-noise ratio and signal reconstruction accuracy in complex backgrounds. With the improvement of optical detection accuracy, dual-comb spectroscopy technology has been gradually introduced into the field of filtering, becoming an important means of high-resolution signal acquisition. Traditional photoacoustic adaptive filtering systems refer to processing devices that dynamically filter weak acoustic signals generated by optical excitation to suppress environmental and system noise. They typically rely on moving average algorithms, least mean square adaptive algorithms, or threshold denoising methods based on time-frequency analysis. These methods include adjusting filter weights based on historical sample mean, calculating error signals in real time and updating filter parameters, and segmenting the signal spectrum by setting energy thresholds based on short-time Fourier transforms to separate target and interfering components in the signal.

[0003] Traditional photoacoustic adaptive filtering systems rely on moving average, least mean square algorithms, or energy threshold methods based on the time and frequency domain. These methods suffer from unstable time references when processing periodic signals, making it difficult to accurately locate the characteristics of the main pulse signal. This leads to delayed or ineffective filter weight updates and difficulty in timely identifying signal structure changes during abrupt phase perturbations, resulting in increased fluctuations in the filter output. Furthermore, the historical sample mean adjustment method is slow to respond to changes in the current signal, which can easily lead to response lag or misjudgment. Spectrum segmentation processing relies on fixed energy thresholds, which are difficult to adapt to dynamic signal background fluctuations and affect signal restoration quality and filter stability under complex interference conditions. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a dual-comb photoacoustic signal adaptive filtering system.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a dual-comb photoacoustic signal adaptive filtering system includes:

[0006] The pulse calibration module acquires the detector output response signal, identifies the periodic peak position points, verifies the interval between adjacent main peaks at equal intervals, records the time positioning information of all main pulses, and generates a main pulse time distribution record.

[0007] The phase calculation module extracts symmetrical signal segments aligned with the center of each main pulse based on the main pulse time distribution record, extracts the analytical form of each signal segment, obtains the phase angle value corresponding to each main pulse, and generates main pulse phase change data.

[0008] The trend recognition module calculates the phase angle difference between two adjacent main pulses based on the main pulse phase change data, compares the current phase difference value with the phase difference value of the previous cycle, marks abrupt change points and stable state points, counts and compares the number of stable state points with the number of abrupt change points, and constructs a stable distribution result of phase trend.

[0009] The filter control module, based on the phase trend stable distribution result and the updated rate factor set by the marked state, drives the step size factor state selector for filter weight update, and performs logical judgment based on the historical step size state of the previous cycle to generate filter weight update instructions.

[0010] The output module verifies the current filter state based on the filter weight update command, records the ratio of the number of command responses to the number of weight-locked segments throughout the entire cycle, and outputs the adaptive filtering state of the dual-comb photoacoustic signal.

[0011] As a further aspect of the present invention, the mutation state point specifically refers to the point when the sign is reversed or the change in the current phase difference value compared with the phase difference value of the previous cycle exceeds the set phase difference mutation threshold; otherwise, it is marked as the stable state point.

[0012] As a further aspect of the present invention, the main pulse time distribution record includes repetition period positioning information, main peak time interval sequence, time base correspondence, and main peak time sequence; the main pulse phase change data includes phase angle sequence, main pulse corresponding time point, and complex domain projection mapping value; the phase trend stable distribution result includes stable segment duration statistics, abrupt segment location distribution, trend stability ratio, and abrupt frequency index; and the filter weight update instruction includes step size factor setting value, weight update start signal, historical state reference identifier, and filter adaptive selection state.

[0013] As a further aspect of the present invention, the pulse calibration module includes:

[0014] The data stream receiving submodule acquires the detector output response signal, detects the external trigger clock signal and the original acoustic response signal superimposed in the response signal, performs frame division and buffering processing on the signal data according to the sampling period and timing calibration parameters, extracts the repetition frequency information of the external trigger clock signal and the time domain waveform sequence of the original acoustic response signal, and generates the response signal alignment data group.

[0015] The periodic peak identification submodule calculates the local extreme point information of the amplitude sequence based on the original acoustic response signal in the alignment data group of the response signal, filters the peak points in the amplitude sequence that exceed the set amplitude judgment benchmark value, and performs a spacing difference judgment operation on the peak point interval. After removing the peak points that do not meet the set periodic spacing error threshold, it constructs an equidistant peak point sequence to obtain an equidistant main peak position index group.

[0016] The pulse time recording submodule reads the position points in the equidistant main peak position index group, aligns the sampling time sequence of each index position with the response signal alignment data group, calculates the actual timestamp corresponding to the index position, arranges them in order of sampling time sequence and constructs an index time corresponding dictionary to generate the main pulse time distribution record.

[0017] As a further aspect of the present invention, the phase calculation module includes:

[0018] The symmetrical segment extraction submodule extracts a fixed-length sampling segment with equal number of sampling points on the left and right sides, centered on the time point, based on each main pulse time point in the main pulse time distribution record. It then retrieves the corresponding sampling points in the original acoustic response signal according to the time distribution index and constructs an equal-length segment sequence to generate a center-aligned signal block group.

[0019] The analytical signal extraction submodule performs a discrete Hilbert transform operation on each time-domain sampling sequence in the center-aligned signal block to construct the complex analytical signal form corresponding to each segment, and extracts the real and imaginary part data sequences to obtain the complex form signal parameter group.

[0020] The phase angle calculation submodule performs an arctangent function operation on each data point based on the imaginary and real part values ​​corresponding to each segment in the complex form signal parameter group, calculates the instantaneous phase angle sequence of the signal, outputs the phase angle distribution corresponding to each main pulse point according to the segment structure, and generates main pulse phase change data.

[0021] As a further aspect of the present invention, the trend recognition module includes:

[0022] The phase difference calculation submodule obtains the average phase angle sequence of continuous main pulses in the main pulse phase change data, performs a difference operation on the phase angle values ​​corresponding to adjacent main pulse numbers, constructs the difference between the current main pulse and the previous main pulse, and arranges them in order to generate a phase angle difference sequence.

[0023] The mutation state determination submodule performs a product sign judgment on the current difference and the previous period difference based on the difference items of two consecutive frames in the phase angle difference sequence and extracts the absolute value of the difference. Based on the judgment result, it marks the mutation state and calculates the current main pulse mutation state mark value. The mark value is 1 to indicate a mutation state and 0 to indicate a stable state, thus obtaining a phase mutation mark list.

[0024] The trend segment construction submodule divides the main pulse sequence intervals within a continuous time period according to the marker value sequence in the phase mutation marker list, counts the number of mutation state values ​​of 1 and the number of stable state values ​​of 0 in each interval, compares the number of the two types of state points in the interval, and obtains the phase trend stable distribution result.

[0025] As a further aspect of the present invention, the formula for calculating the current main pulse abrupt change state marker value is as follows:

[0026] ;

[0027] in, Indicates the first The phase angle difference between the main pulses, This is the difference between the previous main pulse and the current pulse. This is the phase change threshold. This is the current main pulse abrupt change state marker value.

[0028] As a further aspect of the present invention, the filtering control module includes:

[0029] The trend state extraction submodule obtains the time series trend labels in the phase trend stable distribution results, reads the trend classification value at the corresponding position of each main pulse period, determines whether the current label is a sudden change state, pairs the label value with the main pulse timestamp to perform structured trend mapping, and obtains the trend state determination sequence.

[0030] The update factor setting submodule performs condition judgment operations based on the marker value in the trend state determination sequence. If the current main pulse marker value is 1, the update rate factor is set to 0. If the marker value is 0, the update rate factor is set to the maximum value of 1, and the filter update control factor group is obtained.

[0031] The step size instruction generation submodule performs logical judgment operations to adjust the step size update status based on the filter update control factor group and the historical step size value stored in the previous cycle. It uses the current control factor status and the historical step size to jointly construct the step size adjustment instruction and obtain the filter weight update instruction.

[0032] As a further aspect of the present invention, the result output module includes:

[0033] The response status verification submodule, based on the execution structure of the filter weight update instruction, detects whether the weight update command corresponding to each main pulse number in the current period has been correctly loaded into the filter, records the filter output value after each weight change in sequence, compares the absolute difference between the continuous output values ​​before and after the current main pulse time, and determines whether it exceeds the set continuity error upper limit threshold. If it does not exceed the threshold, it is marked as continuous; otherwise, it is marked as interrupted, and a filter output continuity mark table is generated.

[0034] The weight behavior statistics submodule, based on the state flag value corresponding to each main pulse in the continuity flag table of the filter output, traverses all main pulse nodes in the entire cycle, counts the number of weight update instructions marked as responded, counts the total number of update instructions that trigger responses in the cycle, and counts the number of times the weight is maintained in the continuous segment due to state locking. It calculates the ratio of the total number of responses to the number of locked segments and generates instruction response to lock ratio data.

[0035] The adaptive state output submodule performs joint determination based on the instruction response and lock ratio data and the filter update record of the previous cycle, classifies the adaptive adjustment state of the current cycle into intervals, establishes a classification state table indexed by the cycle number, and obtains the adaptive filtering state of the dual optical comb photoacoustic signal.

[0036] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0037] In this invention, the periodic components are accurately identified by acquiring the timing information of the main pulse and extracting the symmetrical segments of the signal. The stability of the signal structure is dynamically monitored by calculating the phase angle and identifying abrupt changes. The status label is determined by the changing trend of the phase difference and the update rate factor is set to achieve precise control of the adjustment rhythm of the filter weights. The filter weight command output is controlled by logical judgment based on the historical step size state and the continuity of the output is verified, which effectively improves the adaptability and stability of the filtering process. In complex backgrounds, the ability to preserve the characteristics of the target signal is enhanced and the influence of non-stationary noise is suppressed, thereby improving the overall signal-to-noise ratio and the reliability of the filter. Attached Figure Description

[0038] Figure 1 This is a system flowchart of the present invention;

[0039] Figure 2 This is a flowchart of the pulse calibration module of the present invention;

[0040] Figure 3 This is a flowchart of the phase calculation module of the present invention;

[0041] Figure 4 This is a flowchart of the trend recognition module of the present invention;

[0042] Figure 5 This is a flowchart of the filter control module of the present invention;

[0043] Figure 6 This is a flowchart of the result output module of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0045] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0046] Please see Figure 1 A dual-comb photoacoustic signal adaptive filtering system includes:

[0047] The pulse calibration module acquires the response signal output by the detector, extracts the external trigger clock signal with the repetition frequency and compares its amplitude with the original acoustic response signal, identifies the periodic peak position points, verifies the interval between adjacent main peaks at equal intervals, records the time positioning information of all main pulses, and generates a main pulse time distribution record.

[0048] The phase calculation module records the time distribution of the main pulse, extracts the symmetrical signal segment aligned with the center of each main pulse, extracts the analytical form of each signal segment, takes the imaginary and real parts of the analytical result as input, executes the arctangent function to obtain the phase angle value corresponding to each main pulse, and generates the main pulse phase change data.

[0049] The trend recognition module calculates the phase angle difference between two adjacent main pulses based on the main pulse phase change data. It compares the current phase difference value with the phase difference value of the previous cycle. When the sign is reversed or the change in the current phase difference value compared with the phase difference value of the previous cycle exceeds the set phase difference abrupt change threshold, the corresponding point is marked as an abrupt change state. Otherwise, it is marked as a stable state. The marking results are arranged in chronological order. The number of stable state points and the number of abrupt change state points in the continuous state segment are counted and compared to construct a stable distribution result of the phase trend.

[0050] The filter control module extracts the trend marker state of the current period based on the phase trend stable distribution result. When the marker is a sudden change state, the update rate factor is set to zero. When the marker is a stable state, the update rate factor is set to the maximum weight. The step size factor state selector of the filter weight update is driven. The module also performs logical judgment in combination with the historical step size state of the previous cycle to generate the filter weight update instruction.

[0051] The output module verifies the current filter state based on the filter weight update command, determines whether the filter output remains continuous before and after the command execution, records the ratio of the number of command responses to the number of weight-locked segments throughout the entire cycle, and outputs the adaptive filtering state of the dual-comb photoacoustic signal.

[0052] The main pulse time distribution record includes repetition period location information, main peak time interval sequence, time base correspondence, and main peak time sequence. The main pulse phase change data includes phase angle sequence, main pulse corresponding time point, and complex domain projection mapping value. The phase trend stable distribution results include stable segment duration statistics, abrupt segment location distribution, trend stability ratio, and abrupt frequency index. The filter weight update instructions include step size factor setting value, weight update start signal, historical state reference identifier, and filter adaptive selection state.

[0053] Please see Figure 2 The pulse calibration module includes:

[0054] The data stream receiving submodule acquires the detector output response signal, detects the external trigger clock signal and the original acoustic response signal superimposed in the response signal, performs frame division and buffering processing on the signal data according to the sampling period and timing calibration parameters, extracts the repetition frequency information of the external trigger clock signal and the time domain waveform sequence of the original acoustic response signal, and generates the response signal alignment data group.

[0055] The detector's output response signal mainly consists of two parts: an external trigger clock signal and the original acoustic response signal. After acquiring this response signal, the continuous analog electrical signal should first be digitized using an analog-to-digital converter, decomposing it into digital frame signals of fixed period length. With a sampling rate set to 250kHz, 250,000 data points can be obtained per second, suitable for detecting pulse responses on the order of 0.1ms in common industrial scenarios. Taking actual device detection data as an example, with the acquisition time set to 0.2 seconds, a total of 50,000 samples are obtained. The signal is divided into 100 frames, each containing 500 data points. Sampling points are collected and stored in a data buffer after frame buffering. Then, an edge detection function is used to mark the boundaries of any square wave pulse signals that may exist in the sampled data. Fast transition signals with frequencies higher than 2kHz are marked as external trigger clock signals, and the remaining signals are marked as the original acoustic response signals. The repetition frequency of the external trigger signal is extracted using the maximum cross-correlation coefficient. The cross-correlation coefficient of each signal segment within the time interval is extracted, and the number of interval points corresponding to the correlation peak is determined to calculate the repetition frequency. If the interval between two adjacent peak positions is 1000 sampling points, then the repetition frequency is 250Hz. A complete timing index matrix is ​​established by combining the frame header timestamp with the intra-frame sampling position. The repetition frequency information and the waveform sequence of the acoustic response signal are output in a structured manner, and finally the response signal alignment data group is formed.

[0056] The periodic peak identification submodule calculates the local extreme point information of the amplitude sequence based on the original acoustic response signal in the alignment data group of the response signal, filters the peak points in the amplitude sequence that exceed the set amplitude judgment benchmark value, and performs a spacing difference judgment operation on the peak point interval. After removing the peak points that do not meet the set periodic spacing error threshold, it constructs an equidistant peak point sequence to obtain an equidistant main peak position index group.

[0057] The original acoustic response signal in the alignment data set is read, and its amplitude sequence is processed point by point. First, the amplitude judgment benchmark is defined as the sum of the average and standard deviation of the entire signal amplitude sequence. For example, if the average of a certain segment of the acoustic signal amplitude sequence is 0.4V and the standard deviation is 0.1V, then the judgment benchmark is... Based on this, all sampling points with amplitudes greater than 0.5V are selected, and their locations are recorded as the initial candidate peak point set. For this point set, it is determined whether the positional distance between adjacent points is within the set periodic distance error threshold. This threshold is based on the theoretical interval of 1000 points corresponding to a repetition frequency of 250Hz, and the allowable error range is set to ±3%, that is, the allowable sampling point distance is between 970 and 1030 points. The candidate point set is traversed sequentially. If the distance between two adjacent points is within this range, the point is retained as the main peak. If it exceeds this range, the abnormal point is removed, thereby selecting the main peak point sequence that meets the periodic interval. The main peak point sequence index is constructed. Table 1 lists the typical candidate point distances and judgment results in the screening operation:

[0058] Table 1. Spacing between screening points on the main peak

[0059] ;

[0060] As shown in Table 1, a sequence of main peak positions with spacing within a reasonable threshold was constructed. This sequence is the equidistant main peak position index group.

[0061] The pulse time recording submodule reads the position points in the equidistant main peak position index group, aligns the sampling time sequence of each index position with the response signal alignment data group, calculates the actual timestamp corresponding to the index position, arranges them in order of sampling time sequence and constructs the index time corresponding dictionary to generate the main pulse time distribution record.

[0062] Based on the selected peak positions in the equidistant peak position index group, for each index point's corresponding time position in the response signal alignment data group, extract its frame number and intra-frame sampling point number. Multiply the frame number by the duration of each frame, add the intra-frame sampling points, and divide by the sampling rate to obtain the timestamp corresponding to the peak. For example, if a peak is located at the 250th sampling point in the 5th frame, with a sampling rate of 250kHz and 500 points per frame, then the frame duration is 2ms, and the timestamp of this peak is... This method calculates the time of all main peak locations, constructs a mapping structure containing index points and corresponding timestamps, forms a complete time series, and establishes an index-time dictionary to store the actual occurrence time of each main peak. The final main peak time distribution sequence is the main pulse time distribution record.

[0063] Please see Figure 3 The phase calculation module includes:

[0064] The symmetrical segment extraction submodule extracts a fixed-length sampling segment centered on each main pulse time point in the main pulse time distribution record. The total sampling length is set to 128 points, symmetrically distributed as 64 points before the center and 64 points after the center. Based on the time distribution index, the corresponding sampling points are retrieved in the original acoustic response signal and an equal-length segment sequence is constructed to generate a center-aligned signal block group.

[0065] Based on each time location point in the main pulse time distribution record, the sampling segment position of the main pulse is obtained and used as the center point for symmetrical segment extraction. A fixed-length symmetrical window structure is then set to extract the signal segment. The total window length is set to 128 sampling points, i.e., 64 points before and after the center point. If the sampling rate is 250kHz, the time span of the symmetrical segment is... This ensures the capture of the complete periodic signal content around the main pulse. Furthermore, the sampling point index corresponding to the extracted segment is located in the original acoustic response signal sequence. The voltage value or sound intensity amplitude value corresponding to each index is extracted to form a signal segment array. For example, if the main pulse time is 12.384ms and the corresponding sampling point number is 30960, then its symmetrical segment range is [30896, 31023], a total of 128 points. After extraction, a one-dimensional signal vector of length 128 is formed. Multiple main pulses will sequentially form several independent sampling sequence groups of 128 points. All sequences are named with the main pulse time index number and arranged sequentially to construct a unified signal segment set structure for subsequent analysis. This set structure is the center-aligned signal group.

[0066] The analytical signal extraction submodule performs a discrete Hilbert transform operation on each time-domain sampling sequence in the center-aligned signal block to construct the complex analytic signal form corresponding to each segment, and extracts the real and imaginary part data sequences to obtain the complex form signal parameter group.

[0067] Extract each signal sequence from the center-aligned signal block and perform analytic signal construction. For each 128-point sampled sequence, perform a discrete Hilbert transform to convert it from a real signal to a complex signal. This process first requires performing a Fourier transform on the original sequence to obtain the frequency domain coefficients. In the frequency domain, set the negative frequency components to zero, retain the DC component and multiply it by the positive frequency components, and then perform an inverse Fourier transform to reconstruct the analytic signal. The real part is the original signal, and the imaginary part is the instantaneous component after the Hilbert transform. For example, if the input signal sequence is... The output complex analytic signal is ,in Extract the output value of the Hilbert transform. and The corresponding sampling point values ​​form two 128-dimensional arrays, forming an imaginary part sequence and a real part sequence. A mapping structure between the imaginary and real components and the main pulse number is established. The corresponding main pulse index number is recorded to track the source data position. The complex signal parameters constructed from all segments are summarized and output uniformly to obtain a complex form signal parameter group.

[0068] The phase angle calculation submodule performs an arctangent function operation on each data point based on the imaginary and real part values ​​corresponding to each segment in the complex form signal parameter group, using numerical relationships. Calculate the instantaneous phase angle sequence of the signal, output the phase angle distribution corresponding to each main pulse point according to the segment structure, and generate the main pulse phase change data;

[0069] Based on each complex signal sequence in the complex form signal parameter group, perform the phase angle calculation operation corresponding to each sampling point. The calculation formula is as follows: ,in This represents the imaginary part of the nth sampling point. To obtain the real part value, the corresponding phase angle value needs to be calculated for each of the 128 sampling points in each group. The arctangent function in the numerical library is used to calculate the angle at a single point and the unit is retained in rad. At the same time, sampling points with a real part of zero are screened in the angle sequence, and the arctangent extension function arctan2 form with quadrant reflection mechanism is used at these points to avoid phase jump errors, for example, when , When, the phase angle is set to rad, which is 90 degrees, after processing in sequence, each main pulse segment will generate a 128-dimensional phase angle sequence. The angle sequences corresponding to all main pulse numbers are summarized in sequence to form a two-dimensional array arranged in time order. Table 2 lists the phase angle variation range corresponding to the three main pulses.

[0070] Table 2. Main Pulse Phase Angle Data Table

[0071] ;

[0072] As shown in Table 2, all main pulse sequences generated equal-length phase distribution arrays, thus obtaining the main pulse phase change data.

[0073] Please see Figure 4 The trend recognition module includes:

[0074] The phase difference calculation submodule obtains the average phase angle sequence of continuous main pulses in the main pulse phase change data, performs a difference operation on the phase angle values ​​corresponding to adjacent main pulse numbers, constructs the difference between the current main pulse and the previous main pulse, and arranges them in order to generate a phase angle difference sequence.

[0075] After acquiring the phase change data of the main pulse, the average phase angle of each main pulse is first extracted according to the main pulse time index. Assume the phase angle sequence of 128 points for each main pulse is as follows: Then its average phase angle is denoted as The calculation method is as follows This mean represents the concentrated trend of instantaneous phase change within the complete cycle of the main pulse. The average phase angle is calculated for all main pulses sequentially and recorded in the time series table. Then, adjacent mean values ​​in the time series are subtracted to construct a difference sequence. That is, for any... and The difference between the two main pulses is If the average phase angle values ​​of the main pulses P1, P2, and P3 in the aforementioned extended module are 0.07 rad, -0.02 rad, and 0.05 rad, respectively, then the corresponding difference is... rad, rad, and so on, to generate a complete difference sequence. To clearly express the phase fluctuation trend between each main pulse, the timestamp corresponding to the main pulse is also used as the horizontal axis during the construction of the difference sequence, so that... By binding the data with the main pulse time information, the subsequent mutation determination is made time-traceable. This operation generates time-ordered phase angle difference data, and finally obtains the phase angle difference sequence.

[0076] The mutation state determination submodule performs a sign judgment on the product of the current difference and the previous period's difference based on the difference between two consecutive frames in the phase angle difference sequence, extracts the absolute value of the difference, and marks the mutation state according to the judgment result, using the formula:

[0077] ;

[0078] The calculation obtains the current main pulse abrupt change state flag value, where a flag value of 1 indicates an abrupt change state and 0 indicates a stable state, resulting in a phase abrupt change flag list; where... Indicates the first The phase angle difference between the main pulses, This is the difference between the previous main pulse and the current pulse. This is the phase change threshold. This is the current main pulse abrupt change state marker value;

[0079] Combining the previously generated phase angle difference sequence, with Indicates the first The phase difference value of each main pulse, for the current difference value of each main pulse. Difference from the previous period To construct a comparison model, the product operation of the two is performed first. To determine if the signs are reversed, if the product is less than 0, the signs are different, indicating a reversal of the trend, further calculations are needed. and mutation threshold For comparison, the threshold setting is based on empirical values ​​of typical phase change thresholds in industrial vibration scenarios, set to 0.6 rad. This is derived from statistical analysis of phase changes in 10 sets of main pulse data during equipment commissioning, with a standard deviation mean ranging from 0.3 to 0.6 rad. Therefore, a threshold of 0.6 rad is more representative. If either of the above two conditions is met, the current main pulse is marked as a sudden change state 1; otherwise, it is marked as a stable state 0. The execution process is shown in Table 3.

[0080] Table 3. Examples of Main Pulse Phase Difference Abrupt Change Judgment Table

[0081] ;

[0082] As shown in Table 3, P3 is marked as a mutation because of sign reversal; although P5 is not reversed, it is also judged as a mutation because the absolute value of the difference is 1.05 rad, which is higher than the threshold of 0.6 rad. The marking rules are strictly based on logical judgment, forming a marking sequence point by point, and finally obtaining the phase mutation marking list.

[0083] The trend segment construction submodule divides the main pulse sequence intervals within a continuous time period according to the marker value sequence in the phase mutation marker list, and counts the number of mutation state values ​​of 1 and the number of stable state values ​​of 0 in each interval. It then compares the number of the two types of state points in the interval to obtain the phase trend stable distribution result.

[0084] Based on the state values ​​of each main pulse number in the phase transition marker list, segment identification and numbering are performed on consecutive points with the same marker value. Let the state sequence be {0, 0, 1, 1, 1, 0, 0, 0, 1, 1}, then it can be identified as 4 segments: S1={0, 0}, S2={1, 1, 1}, S3={0, 0, 0}, S4={1, 1}. The state values ​​of points within each segment are consistent. Each segment is defined as numbered as follows: ,in Assign a segment index number to each segment. Internal statistical number of mutation state points Number of steady state points To determine the phase change trend structure of the segment, the abrupt change ratio is further calculated. ,like If the trend is abnormal, the overall trend of the segment is considered to be dominated by sudden changes; otherwise, it is dominated by stability. All segments are classified and judged in this way, and a trend type label table and a trend mapping matrix are constructed to form multi-segment phase evolution trend distribution data, as shown in Table 4.

[0085] Table 4 Statistical Structure of Trend Sections

[0086] ;

[0087] Finally, all segment trend classification structures are encoded and output in chronological order to establish an overall trend data structure, forming a time-series structural feature dataset that reflects the dynamic evolution of the main pulse, and obtaining a stable distribution result of the phase trend.

[0088] Please see Figure 5 The filter control module includes:

[0089] The trend state extraction submodule obtains the time series trend labels from the phase trend stable distribution results, reads the trend classification value at the corresponding position of each main pulse period, determines whether the current label is a sudden change state, pairs the label value with the main pulse timestamp to perform structured trend mapping, and obtains the trend state determination sequence.

[0090] The time-series data structure of the phase trend stable distribution results is obtained, and the trend classification label information corresponding to each main pulse time point is extracted item by item. Based on the segment trend type determination table established in the previous module, each main pulse is mapped to the trend state of its segment. If the segment is determined to be "abrupt", the main pulse trend label is set to 1, otherwise it is 0, forming a one-to-one correspondence between the main pulse number and the trend state. The mapping table is indexed and reconstructed, and the unified output format is set to triplet form, that is, {main pulse number, timestamp, trend state value}. For example, the main pulse P10 timestamp is 10.24ms, and its segment trend is "abrupt", so the trend state value is 1, and the triplet is {P10, 10.24ms, 1}. After traversing all subsequent main pulse nodes, a complete trend state list is constructed. Table 5 shows some of the extracted sample results.

[0091] Table 5. Examples of Trend State Extraction

[0092] ;

[0093] As shown in Table 5, the trend status value is determined based on the trend nature of the segment. Subsequently, a structured mapping table is established to obtain the trend status determination sequence.

[0094] The update factor setting submodule performs condition judgment operations based on the marker value in the trend state judgment sequence. If the current main pulse marker value is 1, the update rate factor is set to 0. If the marker value is 0, the update rate factor is set to the maximum value of 1, and the filter update control factor group is obtained.

[0095] Read the main pulse marker value in the trend state determination sequence, read the trend state value item by item and set the corresponding filter update rate control factor. When the trend state value is 1, it means that the current main pulse is in a sudden change state, so the update rate factor is set to 0. If the trend state value is 0, it means that the state is stable, so the factor takes the maximum value of 1. The actual value range is limited to the closed interval [0, 1], forming a binary control factor sequence for filter control. Then, construct the triplet structure of main pulse number, timestamp, and control factor. For example, the main pulse P12 has a trend state value of 0 and a corresponding control factor of 1. The structure is {P12, 11.28ms, 1}. All main pulses are expanded in this way to form a complete control factor sequence. To verify the control factor distribution structure, the proportion of the two types of states is counted and some representative data are listed in Table 6.

[0096] Table 6. Example of Filter Update Factor Setting

[0097] ;

[0098] As shown in Table 6, the control factor values ​​correspond one-to-one with the trend status and are used as the basis for subsequent weight update judgment to obtain the filter update control factor group.

[0099] The step size instruction generation submodule performs logical judgment operations to adjust the step size update status based on the filter update control factor group and the historical step size value stored in the previous cycle. It uses the current control factor status and the historical step size to jointly construct the step size adjustment instruction and obtain the filter weight update instruction.

[0100] Based on the filter update control factor group and the historical step size value recorded in the previous cycle, the current filter step size status judgment operation is performed. First, the step size factor value of each main pulse node in the previous cycle is called and paired with the control factor of the current node to form a binary judgment pair. If the current control factor is 0, the step size of the previous cycle remains unchanged. If the control factor is 1 and the step size of the previous cycle is less than the maximum step size threshold of the system (set to 0.2), the current step size update value is incremented by 0.01 for progressive fine-tuning. If the threshold has been reached, it remains unchanged. For example, if the step size of the main pulse P13 in the previous cycle is 0.15 and the control factor is 1, the current step size update value is 0.16. Table 7 lists examples of the judgment process.

[0101] Table 7. Sample Table of Step Size Instruction Generation

[0102] ;

[0103] As shown in Table 7, the update operation is performed under the condition that the control factor is 1 and the current step size has not reached the threshold, forming a complete step size instruction data sequence and obtaining the filter weight update instruction.

[0104] Please see Figure 6 The output module includes:

[0105] The response status verification submodule is based on the execution structure of the filter weight update instruction. It detects whether the weight update command corresponding to each main pulse number in the current period has been correctly loaded into the filter. It records the filter output value after each weight change in sequence, compares the absolute difference between the continuous output values ​​before and after the current main pulse time, and determines whether it exceeds the set continuity error upper limit threshold. If it does not exceed the threshold, it is marked as continuous; otherwise, it is marked as interrupted. It generates a filter output continuity mark table.

[0106] Based on each loading record of the filter weight update instruction, the current filter execution response is verified. First, the output values ​​of the main pulse for two consecutive frames are read from the filter output buffer. Let the current frame number be... The previous frame number is The corresponding output values ​​are respectively , Perform absolute difference calculation, setting the upper limit threshold for continuity error to 0.05. If the condition is met... The current frame is considered continuous, otherwise it is considered discontinuous. This judgment operation is combined with the instruction flag bit for mapping and matching. If there is a weight update instruction in the current frame, it is necessary to additionally confirm whether the output value before and after the weight change meets the above conditions. If it meets the conditions, the instruction is judged to be executed continuously and effectively. Otherwise, it is considered to be discontinuous. The judgment result is statistically bound to the main pulse number and output to generate a marker sequence. Table 8 lists some response state examples.

[0107] Table 8 Filter Continuity Verification Table

[0108] ;

[0109] As shown in Table 8, the frame-by-frame comparison results based on the continuity error threshold form a state determination list, and finally obtain the filter output continuity label table.

[0110] The weight behavior statistics submodule is based on the state flag value corresponding to each main pulse in the continuous flag table of the filter output. It traverses all main pulse nodes in the whole cycle, counts the number of weight update instructions marked as responded, counts the total number of update instructions that trigger response in the cycle, and counts the number of times the weight is maintained in the continuous segment due to state lock. It calculates the ratio of the total number of responses to the number of locked segments and generates instruction response and lock ratio data.

[0111] Based on the status flag values ​​of each main pulse in the filter output continuity flag table, and in conjunction with the update command records issued in previous cycles, the total number of commands issued throughout the entire cycle is calculated. The number of instruction responses that are determined to be in a continuous state is denoted as . Additionally, in the identification filter state, segments where multiple consecutive frames do not trigger weight update instructions are identified. The weights within these segments remain unchanged; these segments are called weight-locked segments, and their number is counted. The final output includes two statistics to calculate the command response rate. Response-lock ratio Table 9 shows examples of statistical data for a certain period.

[0112] Table 9. Weighted Behavior Statistics Table

[0113] ;

[0114] As shown in Table 9, the statistical results directly reflect the relationship between the filter command execution activity and the lock ratio, generating command response and lock ratio data.

[0115] The adaptive state output submodule makes a joint judgment based on the command response and lock ratio data and the filter update record of the previous cycle, and classifies the adaptive adjustment state of the current cycle into intervals. If the response command ratio is greater than 0.6, the current cycle is marked as "dynamic adjustment state"; if it is less than 0.4, it is marked as "weight frozen state"; and the remaining intervals are set as "semi-locked state". A classification state table is established with the cycle number as the index to obtain the adaptive filtering state of the dual optical comb photoacoustic signal.

[0116] Response rate calculated based on command response to lock ratio data and ratio The call cycle number is used as an index to classify the response mode within each cycle. A classification threshold is set (based on the system's tolerance range for filter frequency domain stability and response dynamics; the upper limit of the response rate threshold of 0.6 stems from the maximum allowable weight update frequency of the filter under a continuous main pulse sequence; if the frequency is higher, it will lead to increased output response oscillation and weight instability; the lower limit of the response rate of 0.4 corresponds to the allowable lower limit of filter output energy fluctuation; below this value, most instructions are in a failed state, unable to complete adaptive adjustment to sudden trends. The response-lock ratio threshold of 1.0 is set because if the number of weight responses is less than the number of locks within a cycle, it indicates that the adjustment mechanism is suppressed, and the overall performance is non-adaptive. Therefore, this ratio limit directly affects the state classification. The fluctuation range of this value varies with the average amplitude of the main pulse; the larger the main pulse amplitude, the higher the system's expected number of filter responses, thus pushing the ratio towards a higher range). The threshold is: If... If the current cycle is determined to be in a "dynamic adjustment state", then... and If the response rate is 0.60, it is close to the upper limit boundary. Combined with the ratio of 2.0, it is judged to be in the "dynamic adjustment state". The response rate of period C1 is 0.32 and the ratio is 0.8, both of which are lower than the set boundary value. Therefore, it is judged to be in the "weight frozen state". The statistical results are used to construct a mapping table with the period number as the primary key. Finally, the period-state classification structure is output. Table 10 shows an example of the output state structure.

[0117] Table 10 Adaptive Filter Status Output Table

[0118] ;

[0119] As shown in Table 10, the output table constructs a filter state distribution data structure based on two quantization indicators to obtain the adaptive filtering state of the dual-comb photoacoustic signal.

[0120] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A dual-comb photoacoustic signal adaptive filtering system, characterized in that, include: The pulse calibration module acquires the detector output response signal, identifies the periodic peak position points, verifies the interval between adjacent main peaks at equal intervals, records the time positioning information of all main pulses, and generates a main pulse time distribution record. The phase calculation module extracts symmetrical signal segments aligned with the center of each main pulse based on the main pulse time distribution record, extracts the analytical form of each signal segment, obtains the phase angle value corresponding to each main pulse, and generates main pulse phase change data. The trend recognition module calculates the phase angle difference between two adjacent main pulses based on the main pulse phase change data, compares the current phase difference value with the phase difference value of the previous cycle, marks abrupt change points and stable state points, counts and compares the number of stable state points with the number of abrupt change points, and constructs a stable distribution result of phase trend. The filter control module reads the previous cycle step size history state based on the phase trend stable distribution result, sets the update rate factor in combination with the marked state, drives the step size factor state selector for filter weight update, and performs logical judgment in combination with the previous cycle step size history state to generate filter weight update instructions. The output module verifies the current filter state based on the filter weight update command, records the ratio of the number of command responses to the number of weight-locked segments throughout the entire cycle, and outputs the adaptive filtering state of the dual-comb photoacoustic signal.

2. The dual-comb photoacoustic signal adaptive filtering system according to claim 1, characterized in that: The abrupt change point specifically refers to the point when the sign is reversed or the change in the current phase difference value compared to the previous cycle exceeds the set phase difference abrupt change threshold; otherwise, it is marked as the stable state point.

3. The dual-comb photoacoustic signal adaptive filtering system according to claim 1, characterized in that: The main pulse time distribution record includes repetition period positioning information, main peak time interval sequence, time base correspondence, and main peak time sequence. The main pulse phase change data includes phase angle sequence, main pulse corresponding time point, and complex domain projection mapping value. The phase trend stable distribution result includes stable segment duration statistics, abrupt segment location distribution, trend stability ratio, and abrupt frequency index. The filter weight update instruction includes step size factor setting value, weight update start signal, historical state reference identifier, and filter adaptive selection state.

4. The dual-comb photoacoustic signal adaptive filtering system according to claim 1, characterized in that, The pulse calibration module includes: The data stream receiving submodule acquires the detector output response signal, detects the external trigger clock signal and the original acoustic response signal superimposed in the response signal, performs frame division and buffering processing on the signal data according to the sampling period and timing calibration parameters, extracts the repetition frequency information of the external trigger clock signal and the time domain waveform sequence of the original acoustic response signal, and generates the response signal alignment data group. The periodic peak identification submodule calculates the local extreme point information of the amplitude sequence based on the original acoustic response signal in the alignment data group of the response signal, filters the peak points in the amplitude sequence that exceed the set amplitude judgment benchmark value, and performs a spacing difference judgment operation on the peak point interval. After removing the peak points that do not meet the set periodic spacing error threshold, it constructs an equidistant peak point sequence to obtain an equidistant main peak position index group. The pulse time recording submodule reads the position points in the equidistant main peak position index group, aligns the sampling time sequence of each index position with the response signal alignment data group, calculates the actual timestamp corresponding to the index position, arranges them in order of sampling time sequence and constructs an index time corresponding dictionary to generate the main pulse time distribution record.

5. The dual-comb photoacoustic signal adaptive filtering system according to claim 1, characterized in that, The phase calculation module includes: The symmetrical segment extraction submodule extracts a fixed-length sampling segment with equal number of sampling points on the left and right sides, centered on the time point, based on each main pulse time point in the main pulse time distribution record. It then retrieves the corresponding sampling points in the original acoustic response signal according to the time distribution index and constructs an equal-length segment sequence to generate a center-aligned signal block group. The analytical signal extraction submodule performs a discrete Hilbert transform operation on each time-domain sampling sequence in the center-aligned signal block to construct the complex analytical signal form corresponding to each segment, and extracts the real and imaginary part data sequences to obtain the complex form signal parameter group. The phase angle calculation submodule performs an arctangent function operation on each data point based on the imaginary and real part values ​​corresponding to each segment in the complex form signal parameter group, calculates the instantaneous phase angle sequence of the signal, outputs the phase angle distribution corresponding to each main pulse point according to the segment structure, and generates main pulse phase change data.

6. The dual-comb photoacoustic signal adaptive filtering system according to claim 1, characterized in that, The trend recognition module includes: The phase difference calculation submodule obtains the average phase angle sequence of continuous main pulses in the main pulse phase change data, performs a difference operation on the phase angle values ​​corresponding to adjacent main pulse numbers, constructs the difference between the current main pulse and the previous main pulse, and arranges them in order to generate a phase angle difference sequence. The mutation state determination submodule performs a product sign judgment on the current difference and the previous period difference based on the difference items of two consecutive frames in the phase angle difference sequence and extracts the absolute value of the difference. Based on the judgment result, it marks the mutation state and calculates the current main pulse mutation state mark value. The mark value is 1 to indicate a mutation state and 0 to indicate a stable state, thus obtaining a phase mutation mark list. The trend segment construction submodule divides the main pulse sequence intervals within a continuous time period according to the marker value sequence in the phase mutation marker list, counts the number of mutation state values ​​of 1 and the number of stable state values ​​of 0 in each interval, compares the number of the two types of state points in the interval, and obtains the phase trend stable distribution result.

7. The dual-comb photoacoustic signal adaptive filtering system according to claim 6, characterized in that, The formula for calculating the current main pulse mutation state marker value is as follows: ； in, Indicates the first The phase angle difference between the main pulses, This is the difference between the previous main pulse and the current pulse. This is the phase change threshold. This is the current main pulse abrupt change state marker value.

8. The dual-comb photoacoustic signal adaptive filtering system according to claim 1, characterized in that, The filtering control module includes: The trend state extraction submodule obtains the time series trend labels in the phase trend stable distribution results, reads the trend classification value at the corresponding position of each main pulse period, determines whether the current label is a sudden change state, pairs the label value with the main pulse timestamp to perform structured trend mapping, and obtains the trend state determination sequence. The update factor setting submodule performs condition judgment operations based on the marker value in the trend state determination sequence. If the current main pulse marker value is 1, the update rate factor is set to 0. If the marker value is 0, the update rate factor is set to the maximum value of 1, and the filter update control factor group is obtained. The step size instruction generation submodule performs logical judgment operations to adjust the step size update status based on the filter update control factor group and the historical step size value stored in the previous cycle. It uses the current control factor status and the historical step size to jointly construct the step size adjustment instruction and obtain the filter weight update instruction.

9. The dual-comb photoacoustic signal adaptive filtering system according to claim 1, characterized in that, The result output module includes: The response status verification submodule, based on the execution structure of the filter weight update instruction, detects whether the weight update command corresponding to each main pulse number in the current period has been correctly loaded into the filter, records the filter output value after each weight change in sequence, compares the absolute difference between the continuous output values ​​before and after the current main pulse time, and determines whether it exceeds the set continuity error upper limit threshold. If it does not exceed the threshold, it is marked as continuous; otherwise, it is marked as interrupted, and a filter output continuity mark table is generated. The weight behavior statistics submodule, based on the state flag value corresponding to each main pulse in the continuity flag table of the filter output, traverses all main pulse nodes in the entire cycle, counts the number of weight update instructions marked as responded, counts the total number of update instructions that trigger responses in the cycle, and counts the number of times the weight is maintained in the continuous segment due to state locking. It calculates the ratio of the total number of responses to the number of locked segments and generates instruction response to lock ratio data. The adaptive state output submodule performs joint determination based on the instruction response and lock ratio data and the filter update record of the previous cycle, classifies the adaptive adjustment state of the current cycle into intervals, establishes a classification state table indexed by the cycle number, and obtains the adaptive filtering state of the dual optical comb photoacoustic signal.

Images