A scanning probe microscope signal adaptive filtering method and system based on multi-rate spectral contrast

Abstract

The application discloses a kind of scanning probe microscope signal adaptive filtering method and system based on multi-rate spectrum contrast, the method includes signal acquisition and parameter record, multi-rate pre-scanning and feature extraction, spatial frequency alignment, invariance test and classification, adaptive digital filter design, real-time filtering and imaging and online incremental learning, mainly according to the characteristics that real topography signal frequency is proportionally offset by scanning parameter change, but system / environment noise frequency remains absolutely constant to automatically separate signal and noise;The system includes signal acquisition module, feature analysis and classification module, adaptive filter module and signal reconstruction module.The application does not depend on preset noise frequency, and does not need to increase additional reference sensor hardware such as accelerometer, but by extracting the multi-rate characteristics of detection signal itself, noise can be automatically and accurately identified and separated, which greatly reduces system complexity and hardware cost.

Description

Technical Field

[0001] This invention relates to the fields of scanning probe microscope signal processing, digital filtering, and microscopic imaging feedback control, and particularly to an adaptive filtering method and system for scanning probe microscope signals based on multi-rate spectral comparison. Background Technology

[0002] Scanning probe microscopy (STM) images the atomic-level morphology of a surface by detecting minute physical quantities (such as tunneling current) between the probe and the sample. Taking STM as an example, tunneling current... Where z is the tip-sample distance and κ is the attenuation constant. The detection signal simultaneously contains real surface morphology information (atomic arrangement, vacancy defects, step edges, etc.) and environmental and instrument noise (mechanical vibration, power frequency electromagnetic interference and harmonics, preamplifier thermal noise, etc.). Noise severely affects imaging quality, leading to decreased atomic resolution, difficulty in defect identification, and interference with the stability of the feedback control system. Existing noise reduction techniques mainly include hardware shielding and isolation, analog filtering and lock-in amplification, image post-processing, and external reference sensor filtering.

[0003] The common drawbacks of the aforementioned methods are that they all require prior knowledge of noise or external reference signals, and cannot automatically identify noise frequency components from the microscopic detection signal itself; nor do they utilize the physical property that "changes in scanning parameters (such as rate) will cause a proportional shift in the frequency of the true morphology signal, but the system / environmental noise frequency remains absolutely constant" to automatically separate the signal from the noise. Therefore, it is necessary to propose an adaptive filtering method that requires no prior knowledge of noise, no additional reference sensors, and can automatically identify and filter out noise frequency components from the detection signal of a scanning probe microscope, fully preserving atomic-level surface features, and achieving real-time, adaptive high-quality imaging and high-precision feedback control. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide an adaptive filtering method and system for scanning probe microscope signals based on multi-rate spectral comparison.

[0005] Technical solution: The adaptive filtering method for scanning probe microscope signals based on multi-rate spectral contrast described in this invention is characterized by the following steps:

[0006] S1: Signal Acquisition and Parameter Recording. Continuously acquires SPM detection signals and transmits them via an analog-to-digital converter at a sampling rate f. s Discretize into a time series, and synchronously record the scan rate v and scan range L;

[0007] S2: Multi-rate pre-scanning and feature extraction, acquiring detection signals at at least two different scan rates v1 and v2, and performing frequency domain transformation to obtain power spectral density (PSD);

[0008] S3: Spatial frequency alignment, mapping the power spectral density at different rates to a spatial frequency coordinate system.

[0009] S4: Invariance test and classification, calculate the coefficient of variation (CV) for each frequency point of the spectrum after spatial frequency alignment, construct the invariance score, and set the first and second preset thresholds.

[0010] S5: Adaptive digital filter design, which automatically designs the corresponding digital filter based on the identified noise frequency location and introduces a filter safety protection mechanism;

[0011] S6: Real-time filtering and imaging. During the formal scanning process, a digital filter is applied to the real-time acquired detection signal. The filtered signal is synchronously output to the imaging display module and the Z-axis piezoelectric ceramic feedback control loop.

[0012] S7: Online incremental learning, continuously extracting and classifying frequency features during the scanning process.

[0013] Furthermore, step S1 applies to different types of SPMs, with a sampling rate f. s Adaptive adjustment: For STM, f s At the 10kHz level; for AFM tap mode, f s It is much higher than the resonant frequency of the cantilever beam, and usually needs to reach the MHz level.

[0014] Furthermore, in step S2, the two different scan rates v1 and v2 satisfy v2 ≥ 2v1.

[0015] Furthermore, step S2 utilizes multi-rate scanning to assemble time-domain signals at different scan rates into an observation matrix:

[0016]

[0017] The FastICA algorithm is used to decompose the original signal into multiple independent source components. By analyzing the spectral stability of each source component as the scanning rate changes, the noise source components whose frequency position does not change with the rate are automatically identified. After setting them to zero, the clean morphology signal can be obtained by reconstruction.

[0018] Further, in step S3, the mapped power spectral amplitude is multiplied by the corresponding scan rate scaling factor v. i Mapped to the spatial frequency coordinate system.

[0019] Furthermore, in step S4, when the invariance test and classification is higher than the first preset threshold, it is determined to be a noise component; when it is lower than the second preset threshold, it is determined to be a signal component; and when it is in between, it is marked as uncertain and will be determined after more frames of data are accumulated.

[0020] Furthermore, step S4 also includes adding a phase consistency criterion to calculate the phase consistency of the same frequency component in M ​​consecutive frames:

[0021]

[0022] If PC(f) is low and its amplitude remains constant, it is identified as noise; if PC(f) is high and its spatial frequency amplitude is consistent, it is identified as a real signal.

[0023] Furthermore, the filter safety protection mechanism in step S5 includes real-time monitoring of the total energy attenuation caused by filtering. If the filtered energy exceeds a preset proportional threshold, or if the correlation between the signals before and after filtering decreases significantly, the filter attenuation depth is automatically reduced or the stopband width is adjusted.

[0024] Furthermore, step S7 includes dynamically adding newly emerging frequency peaks to the filter list after confirmation by consecutive frames; and performing real-time frequency tracking on the confirmed noise.

[0025] The adaptive filtering system for scanning probe microscope signals based on multi-rate spectral contrast described in this invention includes:

[0026] Signal acquisition module: used to acquire raw detection signals and record scanning parameters at different scan rates;

[0027] Feature analysis and classification module: used to compare the frequency correlation features of signals at different rates, and based on the degree of invariance of features with scanning rate, to identify the signal as a signal component or a noise component;

[0028] Adaptive filter module: used to dynamically construct and adjust digital filters based on the identified noise component characteristics;

[0029] Signal reconstruction module: used to output the filtered signal and connect it to the imaging unit or feedback control unit of the microscope.

[0030] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0031] (1) No prior noise knowledge or external sensors required: It does not rely on preset noise frequencies, nor does it require additional reference sensor hardware such as accelerometers. It can automatically and accurately identify and separate noise by extracting the multi-rate characteristics of the detection signal itself, which greatly reduces system complexity and hardware cost.

[0032] (2) Lossless preservation of atomic and nanoscale real morphology: It breaks through the bottleneck that traditional low-pass and high-pass analog filters are prone to damaging the real signal. It uses "multi-rate invariance" to accurately locate noise. Combined with adaptive narrowband filtering or independent source separation, it can accurately remove noise components and completely preserve the real signal frequency band, avoiding problems such as blurred step edges and masked atomic defects.

[0033] (3) Highly adaptive to different scanning conditions: The algorithm can automatically calculate the spatial frequency mapping and dynamically adjust the filtering strategy and filter parameters according to the actual scanning rate, scanning range and other parameters. When switching different scanning tasks or changing scanning parameters, there is no need to manually readjust the parameters, realizing true "foolproof" adaptation.

[0034] (4) Enhanced real-time online processing and control loop: It adopts an efficient feature extraction and computing architecture with low processing latency, and can output clear images in real time during the formal scanning process. More importantly, it can provide clean and noise-free control signals for the Z-axis piezoelectric feedback control system of the microscope in real time, thereby further improving the closed-loop control accuracy of the instrument from the physical source.

[0035] (5) It has continuous incremental learning capability: the algorithm continuously updates and iterates during the scanning process. The longer the scanning time, the more accurate the classification of complex frequency bands. It can also track the slow frequency drift of environmental noise sources in real time, ensuring extremely high stability under long-term continuous scanning.

[0036] (6) Extremely strong cross-platform versatility: It is not limited to the probe current of STM, but also perfectly applicable to the signal processing of various devices such as AFM deflection signal, STS spectral signal and MFM, and has extremely high instrumentation and commercialization prospects. Attached Figure Description

[0037] Figure 1 This is the overall system architecture of the adaptive filtering method for scanning probe microscope signals based on multi-rate spectral comparison described in this invention;

[0038] Figure 2 A comparison of the power spectral density of the SPM detection signal at the normal scan rate and the doubled scan rate;

[0039] Figure 3 A classification result image showing PSD data at different rates mapped and aligned to the spatial frequency coordinate system;

[0040] Figure 4 The amplitude-frequency response curve of the adaptive digital filter bank;

[0041] Figure 5 A visual comparison of real-time image quality before and after filtering;

[0042] Figure 6 The decision logic flowchart for determining component attributes. Detailed Implementation

[0043] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0044] like Figure 1 As shown, the adaptive filtering method for scanning probe microscope signals based on multi-rate spectral contrast according to the present invention includes the following steps:

[0045] S1: Signal acquisition and parameter recording, continuously acquiring SPM detection signals (such as STM probe current or AFM deflection signal), and transmitting them via an analog-to-digital converter at a sampling rate f. s Discretize into a time series. Simultaneously record the scan rate v and scan range L. For different types of SPM, the sampling rate f... s It needs to be adjusted adaptively. For STM, f s Typically at the 10kHz level; for AFM tap mode, f s The frequency needs to be much higher than the resonant frequency of the cantilever beam (usually reaching the MHz level) to meet the high-frequency carrier sampling requirements.

[0046] S2: Multi-rate pre-scanning and feature extraction. In the pre-scanning stage, the detection signal is acquired at at least two different scan rates v1 and v2 (v2 ≥ 2v1 is recommended to ensure discriminative power), and a frequency domain transformation is performed to obtain the power spectral density (PSD). Figure 2 As shown, the power spectral density (PSD) of the SPM detection signal is displayed at (a) a normal scan rate v1 and (b) a doubled scan rate v2 = 2v1, respectively. It is clearly shown that the positions of the ambient noise peaks (50Hz, 100Hz, 5Hz) remain constant, while the sample morphology signal peaks (f...)... s The physical property of moving proportionally with speed. Since this method compares the statistical characteristics of the spectrum rather than the absolute correspondence of spatial points, slight thermal drift or piezoelectric ceramic creep during the pre-scanning process will not interfere with the accuracy of identification.

[0047] S3: Spatial frequency alignment maps the power spectral density at different rates to the spatial frequency coordinate system σ = f / v. To ensure the total power is conserved before and after the frequency coordinate transformation (i.e., satisfying ∫S(f)df = ∫S(σ·v)·vdσ), the mapped power spectral amplitude needs to be multiplied by the corresponding scan rate scaling factor v. i ;like Figure 3As shown, the classification results are displayed after mapping and aligning PSD data at different rates to a spatial frequency coordinate system (σ = f / v). The real signal peaks are perfectly aligned in the spatial frequency domain, while the noise peaks show significant deviations due to alignment failure, thus achieving accurate distinction between signal and noise.

[0048] S4: Invariance test and classification. Calculate the coefficient of variation (CV) (the ratio of standard deviation to mean) for each frequency point of the spatially frequency-aligned spectrum. Construct the invariance score Inv = 1 / (1 + α·CV). Set a first preset threshold (e.g., 0.8-0.9) and a second preset threshold (e.g., 0.3-0.4). Components above the first preset threshold are identified as noise components; components below the second preset threshold are identified as signal components; components between the two are marked as uncertain and will be determined after accumulating more frames of data.

[0049] S5: Adaptive digital filter design, automatically designing corresponding digital filters (such as IIR notch filter banks or band-stop filters) based on the identified noise frequency locations. Figure 4 The figure shows the amplitude-frequency response curve of an adaptive digital filter bank (such as a Trap filter bank). Its stopband center frequency is automatically and accurately aligned with the noise frequencies (e.g., 5Hz, 50Hz, 100Hz) identified through multi-rate feature comparison. The figure also illustrates the design principles of the adaptive filter configuration transfer function H(z). A filter safety protection mechanism is introduced: real-time monitoring of the total energy attenuation caused by filtering. If the filtered energy exceeds a preset percentage threshold (e.g., 20%-40% adaptively set according to sample roughness), or if the signal correlation before and after filtering decreases significantly, the filter attenuation depth is automatically reduced or the stopband width is adjusted to prevent signal distortion.

[0050] S6: Real-time filtering and imaging. During the formal scanning process, a digital filter is applied to the real-time acquired detection signal. The filtered signal is synchronously output to the imaging display module and the Z-axis piezoelectric ceramic feedback control loop.

[0051] S7: Online incremental learning, continuously extracting and classifying frequency features during the scanning process. Newly emerging frequency peaks are dynamically added to the filter list after being confirmed in consecutive frames; confirmed noise is tracked in real time (e.g., using exponential moving average) to cope with the slow frequency drift of environmental noise.

[0052] Example:

[0053] Taking an STM scanning range of 100nm×100nm as an example, the setpoint current is 1nA and the bias voltage is 0.5V. Pre-scanning is performed at v1=100nm / s and v2=200nm / s to scan the same local region of the sample. The two power spectra are mapped to the spatial frequency domain and the invariance fraction (α = 10) is calculated. If Inv > 0.85 at 50Hz and 100Hz, it is determined to be power frequency and its second harmonic noise; if Inv > 0.85 at 5Hz, it is determined to be mechanical vibration; while at 33Hz (which becomes 66Hz under v2), Inv < 0.3, it is determined to be the real atomic morphology signal doubling with the rate. Subsequently, an adaptive IIR notch filter array is designed, with the center frequency automatically aligned to 5Hz, 50Hz, and 100Hz. The transfer function is designed as follows:

[0054]

[0055] The parameter 'r' is automatically configured based on noise spectrum broadening. The formal scan is performed at 100 nm / s, eliminating the aforementioned noise in real time, ensuring a clean control signal for the STM feedback loop, and significantly improving the imaging clarity at step edges and defect locations. For example... Figure 5 The image shows a visual comparison of the real-time imaging quality before and after filtering. (a) is the topography of the original detection signal acquired in real time, which contains severe mechanical and electrical periodic stripe noise; (b) is the topography after the multi-rate feature comparison adaptive filtering of this invention, where the noise is completely eliminated, while the atomic-level grid, step edges and defect structures on the sample surface are completely and undamaged.

[0056] Phase enhancement recognition assists in judgment

[0057] Add a phase consistency criterion: calculate the phase consistency of the same frequency component in M ​​consecutive frames.

[0058]

[0059] If PC(f) is low and its amplitude remains constant, it is identified as noise; if PC(f) is high and its spatial frequency amplitude is consistent, it is identified as a real signal. By combining the comparison of forward and reverse scan spectrum differences, the classification accuracy of complex frequency band overlapping regions can be further improved.

[0060] Alternative classification schemes based on blind source separation (ICA)

[0061] Multi-rate scanning is used to break the statistical independence assumption between the signal and certain correlated noise under single-rate conditions. An observation matrix is ​​constructed from time-domain signals at different scan rates.

[0062]

[0063] The FastICA algorithm is used to decompose the original signal into multiple independent source components. By analyzing the spectral stability of each source component as the scan rate changes, noise source components whose frequency position does not change with the rate are automatically identified. After setting these noise source components to zero, the clean topography signal can be obtained by reconstruction.

[0064] Deep learning-assisted feature fusion

[0065] A lightweight 1D convolutional neural network is trained for classification. The training dataset is generated in large quantities by simulating topographic signals at different scan rates and using a random environment / instrument noise model. The network input is a spatially frequency-aligned multi-channel spectral vector, and the output is a predicted probability. This probability is then fused with the invariance score of the physical model using Bayesian methods to improve the robustness of the decision in extreme low signal-to-noise ratio environments.

[0066] The adaptive filtering system for scanning probe microscope signals based on multi-rate spectral contrast described in this invention includes:

[0067] Signal acquisition module: used to acquire raw detection signals and record scanning parameters at different scan rates;

[0068] Feature analysis and classification module: used to compare the frequency correlation features of signals at different rates, and based on the degree of invariance of features with scanning rate, to identify the signal as a signal component or a noise component;

[0069] Adaptive filter module: used to dynamically construct and adjust digital filters based on the identified noise component characteristics;

[0070] Signal reconstruction module: used to output the filtered signal and connect it to the imaging unit or feedback control unit of the microscope.

Claims

1. An adaptive filtering method for scanning probe microscope signals based on multi-rate spectral contrast, characterized in that, Includes the following steps: S1: Signal Acquisition and Parameter Recording. Continuously acquires SPM detection signals and transmits them via an analog-to-digital converter at a sampling rate f. s Discretize into a time series, and synchronously record the scan rate v and scan range L; S2: Multi-rate pre-scanning and feature extraction, acquiring detection signals at at least two different scan rates v1 and v2, and performing frequency domain transformation to obtain power spectral density (PSD); S3: Spatial frequency alignment, mapping the power spectral density at different rates to a spatial frequency coordinate system. S4: Invariance test and classification, calculate the coefficient of variation (CV) for each frequency point of the spectrum after spatial frequency alignment, construct the invariance score, and set the first and second preset thresholds. S5: Adaptive digital filter design, which automatically designs the corresponding digital filter based on the identified noise frequency location and introduces a filter safety protection mechanism; S6: Real-time filtering and imaging. During the formal scanning process, a digital filter is applied to the real-time acquired detection signal. The filtered signal is synchronously output to the imaging display module and the Z-axis piezoelectric ceramic feedback control loop. S7: Online incremental learning, continuously extracting and classifying frequency features during the scanning process.

2. The adaptive filtering method for scanning probe microscope signals based on multi-rate spectral contrast according to claim 1, characterized in that, Step S1 applies to different types of SPMs, with a sampling rate f. s Adaptive adjustment: For STM, f s At the 10kHz level; for AFM tap mode, f s It is much higher than the resonant frequency of the cantilever beam, reaching the MHz level.

3. The adaptive filtering method for scanning probe microscope signals based on multi-rate spectral contrast according to claim 1, characterized in that, In step S2, the two different scan rates v1 and v2 satisfy v2 ≥ 2v1.

4. The adaptive filtering method for scanning probe microscope signals based on multi-rate spectral contrast according to claim 1, characterized in that, Step S2 uses multi-rate scanning to form an observation matrix from time-domain signals at different scanning rates: The FastICA algorithm is used to decompose the original signal into multiple independent source components. By analyzing the spectral stability of each source component as the scanning rate changes, the noise source components whose frequency position does not change with the rate are automatically identified. After setting them to zero, the clean morphology signal can be obtained by reconstruction.

5. The adaptive filtering method for scanning probe microscope signals based on multi-rate spectral contrast according to claim 1, characterized in that, Step S3 multiplies the mapped power spectrum amplitude by the corresponding scan rate scaling factor v. i Mapped to the spatial frequency coordinate system.

6. The adaptive filtering method for scanning probe microscope signals based on multi-rate spectral contrast according to claim 1, characterized in that, In step S4, when the invariance score is higher than the first preset threshold, it is determined to be a noise component; when it is lower than the second preset threshold, it is determined to be a signal component; when it is between the two, it is marked as uncertain and will be determined after more frames of data are accumulated.

7. The adaptive filtering method for scanning probe microscope signals based on multi-rate spectral contrast according to claim 1, characterized in that, Step S4 further includes adding a phase consistency criterion to calculate the phase consistency of the same frequency component in M ​​consecutive frames: If PC(f) is low and its amplitude remains constant, it is identified as noise; if PC(f) is high and its spatial frequency amplitude is consistent, it is identified as a real signal.

8. The adaptive filtering method for scanning probe microscope signals based on multi-rate spectral contrast according to claim 1, characterized in that, The filter safety protection mechanism in step S5 includes real-time monitoring of the total energy attenuation caused by filtering. If the filtered energy exceeds a preset ratio threshold, or if the correlation between the signals before and after filtering decreases significantly, the filter attenuation depth is automatically reduced or the stopband width is adjusted.

9. The adaptive filtering method for scanning probe microscope signals based on multi-rate spectral contrast according to claim 1, characterized in that, Step S7 includes dynamically adding newly emerging frequency peaks to the filter list after confirmation by consecutive frames; and performing real-time frequency tracking on the confirmed noise.

10. An adaptive filtering system for scanning probe microscope signals based on multi-rate spectral contrast, characterized in that, include: Signal acquisition module: used to acquire raw detection signals and record scanning parameters at different scan rates; Feature analysis and classification module: used to compare the frequency correlation features of signals at different rates, and based on the degree of invariance of features with scanning rate, to identify the signal as a signal component or a noise component; Adaptive filter module: Used to dynamically construct and adjust digital filters based on the identified noise component characteristics; Signal reconstruction module: used to output the filtered signal and connect it to the imaging unit or feedback control unit of the microscope.

Images