Resonance sensor inherent frequency detection method based on adaptive convergence discrimination

Abstract

The invention discloses a resonant sensor inherent frequency detection method based on adaptive convergence discrimination, and belongs to the technical field of digital signal processing and sensors. The technical problem to be solved is to overcome the problem of difficulty in amplitude extraction in a low signal-to-noise ratio environment in a traditional orthogonal phase-locking amplitude extraction method. The core of the technical scheme is as follows: after orthogonal phase locking and narrow-band filtering are carried out on a resonance sensor vibration pickup signal, an output sequence is input to a digital characteristic processing module; self-adaptive amplitude extraction is carried out according to the designed convergence discrimination scheme; and the resonant frequency of the system is determined by comparing historical amplitude data. The accuracy and the reliability of amplitude extraction in a strong noise environment are remarkably improved, and the algorithm is friendly in hardware and easy to efficiently implement in an FPGA (Field Programmable Gate Array). The method is especially suitable for occasions with high requirements on frequency detection precision and real-time performance, such as a resonance tube densitometer.

Description

Technical Field

[0001] This invention relates to the fields of digital signal processing and sensor measurement technology, and in particular to a resonant sensor frequency detection method and system based on adaptive convergence discrimination, which is applicable to applications such as resonant tube density meters, tuning fork resonant sensors, quartz microbalances, and MEMS resonators where high accuracy and stability of frequency measurement are required. Background Technology

[0002] Resonant sensors typically generate mechanical vibrations at a specific frequency through an excitation structure. The natural frequency of this sensor has a definite functional relationship with the measured medium or physical quantity (such as density, mass, or concentration). Therefore, high-precision physical quantity detection can be achieved by accurately extracting the resonant frequency.

[0003] Existing frequency detection systems often employ quadrature phase-locked loop (I / Q demodulation) technology to acquire the amplitude and phase of the picked-up signal, and then find the point of maximum amplitude, i.e., the resonant frequency, by scanning different excitation frequencies. However, this approach faces significant challenges in practical applications:

[0004] 1. Amplitude fluctuations in low signal-to-noise ratio (SNR) environments: In actual industrial settings, sensor signals are highly susceptible to environmental electromagnetic interference, circuit noise, and fluid flow noise, resulting in typically low SNRs. When quadrature phase-locked loop (QLL) technology processes signals in low SNR scenarios, the output I and Q signals, as well as the calculated instantaneous amplitude A, exhibit significant random fluctuations. Directly using a single instantaneous value or a simple average value from these fluctuations as the amplitude introduces substantial errors, leading to inaccurate frequency positioning.

[0005] 2. Traditional filtering methods have limitations in digital circuit implementation. Existing methods typically use FIR low-pass filters based on rectangular windows to perform secondary processing on the phase-locked loop output. However, this method has an inherent contradiction: if the window length or filter order is too short, the filtering effect is poor and the amplitude estimation is still unstable; if the window length or order is too long, it will cause fixed-point bit width expansion in digital circuits and cause excessive consumption of computational resources, resulting in problems such as degraded timing performance.

[0006] 3. Under strong noise interference, the output of the quadrature phase-locked loop (QPL) algorithm will produce residual fluctuations due to the bandwidth limitation of the low-pass filter. A common method is to perform secondary processing on the output of the QPL algorithm based on the sliding window averaging scheme and set the extraction time according to engineering experience, that is, output the processing result at the time point after the secondary processing. However, this method relies heavily on empirical parameters, lacks adaptive ability, and has poor universality in different working environments.

[0007] Therefore, there is an urgent need in this field for a new scheme for amplitude extraction and frequency detection that can ensure high accuracy and achieve fast convergence in low signal-to-noise ratio and mixed noise environments. Summary of the Invention

[0008] (I) Purpose of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of existing resonant frequency detection methods and provide a signal amplitude extraction method and frequency detection system to resolve the contradiction between accuracy and speed in traditional amplitude extraction methods under low signal-to-noise ratio environments. Another objective of this invention is to provide an amplitude extraction scheme that is robust to both additive and multiplicative noise. A further objective of this invention is to provide a real-time processing architecture that consumes fewer hardware resources, has high computational efficiency, and is particularly suitable for FPGA implementation.

[0010] (II) Technical Solution

[0011] To achieve the above objectives, the technical solution adopted in this invention is as follows: a signal amplitude extraction method based on adaptive digital feature convergence discrimination, comprising the following steps: First, the input signal is acquired and amplitude feature extraction is completed in step one. Specifically, the vibration pickup signal output by the system under test is mixed with two mutually orthogonal reference signals. Based on this, a narrow-bandwidth low-pass structure composed of a cascaded CIC decimation filter and an FIR filter is used to filter the mixing result. The high decimation ratio of CIC and the bandwidth shaping capability of FIR effectively suppress out-of-band noise while reducing the data rate, thereby obtaining an extremely narrow effective bandwidth under limited hardware resources. The filtered I-channel and Q-channel DC components are used to calculate the instantaneous amplitude sequence, which serves as the input for subsequent digital feature analysis.

[0012] In step two, this invention treats the instantaneous amplitude sequence as a stochastic process that updates continuously over time, and updates the mean and variance of the sequence in real time using the Welford recursive algorithm. When each new data point arrives, the recursive algorithm only relies on the mean and variance state variables from the previous time step to calculate the new statistics, without needing to store all historical data within the window. This greatly reduces storage pressure and computational overhead, enabling stable operation in a real-time FPGA processing environment. This recursive method also features high numerical stability and is less prone to overflow, making it particularly important in high dynamic range sensor signal processing.

[0013] In step three, this invention constructs an adaptive convergence discrimination mechanism based on the evolution trend of sequence variance to determine whether the current amplitude has reached a statistically stable state. To improve the algorithm's adaptability to different noise environments, this invention provides two convergence strategies that can be used independently or in combination. One is a sliding window threshold method based on the rate of change of adjacent variances. By calculating the relative changes in continuous variances within a fixed window, if all changes are below a set threshold, the sequence is considered to have converged. This method has a fast response speed and simple hardware logic. The other is a sliding window discrimination method based on the normalized range. By comparing the normalized range of the maximum, minimum, and average values ​​within the window, if it is less than a preset threshold, the variance is considered to be in a stable range. This method has stronger robustness to both additive and multiplicative noise. The two schemes can be selected or used in combination according to the system noise characteristics, thereby ensuring stable and reliable convergence judgment capabilities over a wide noise range.

[0014] In step four, once the convergence discrimination module confirms that the current variance has reached a stable state, the present invention uses the mean output obtained from the current recursive estimation as the final amplitude result. Since this result is only output when it is statistically sufficiently converged, the present invention avoids the amplitude jitter and misjudgment problems caused by the direct use of filtered data in traditional methods, which can significantly improve the reliability of amplitude extraction and provide a stable amplitude input for subsequent resonant frequency detection.

[0015] When the above amplitude extraction method is applied to the frequency detection of a resonant sensor, the present invention further includes steps five to seven. In step five, the system applies excitation signals of different frequencies to the resonant sensor according to a preset frequency scanning strategy and acquires the vibration pickup signal as input in real time. In step six, the aforementioned amplitude extraction process is repeated for each excitation frequency point to obtain the corresponding stable amplitude estimate. In step seven, by comparing the stable amplitudes corresponding to each frequency point, the excitation frequency corresponding to the largest amplitude is selected as the natural frequency of the current system. For a resonant tube density meter, this natural frequency can be further substituted into the calibration model to solve for the medium density, realizing a complete measurement link from the vibration pickup signal to the physical property parameters.

[0016] (III) Beneficial Effects

[0017] This invention achieves a comprehensive performance improvement in amplitude extraction and resonant frequency detection by treating the amplitude sequence as a random process and introducing an adaptive convergence discrimination mechanism based on digital features. Compared with existing traditional methods that rely on stable filter outputs, this invention first utilizes a narrowband cascaded structure to effectively suppress out-of-band noise. Then, it updates the mean and variance in real time using the Welford recursive algorithm, ensuring high computational efficiency and numerical stability for amplitude estimation even in high-speed streaming data environments. More importantly, this invention no longer directly uses the filter output as the final amplitude. Instead, it uses variance convergence as the reliability criterion, judging the stability of the amplitude sequence through statistical features such as adjacent change rates or normalized ranges. The amplitude is output only when the sequence converges statistically, significantly reducing the probability of misjudgment caused by noise fluctuations. This structure achieves rapid adaptive convergence, significantly reducing the number of samples required for amplitude estimation, significantly improving response speed, and maintaining excellent performance even in environments with strong additive noise, multiplicative noise, and even mixed noise. Meanwhile, the computational logic of this invention consists of simple addition, subtraction, multiplication, division, and comparison operations, which are easy to implement in a pipelined manner in FPGA, with low hardware resource consumption, making it suitable for embedded scenarios with high real-time requirements. Thanks to the above improvements, in resonant sensor frequency detection applications, this invention can obtain smoother, more stable, and more reliable amplitude-frequency curves, significantly improve peak identification accuracy, and increase frequency scanning speed, thereby effectively improving the measurement accuracy, stability, and applicability of devices such as resonant tube density meters. Attached Figure Description

[0018] Figure 1 This is a simplified structural diagram of a resonant tube density meter.

[0019] Figure 2 This is a simplified structural diagram of a phase-sensitive detector unit.

[0020] Figure 3 This is a simplified structural diagram of an orthogonal phase-locked loop system.

[0021] Figure 4 This is a cascaded structure diagram of a narrow bandwidth low-pass filter (CIC+FIR) for quadrature phase-locked loop output in a preferred embodiment of the present invention.

[0022] Figure 5 This is a diagram of the Welford algorithm framework. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to examples. It should be understood that the specific examples described herein are merely illustrative and not intended to limit the invention.

[0024] Hardware Platform and Overall System Structure. This embodiment uses a resonant tube density meter as a specific application. Its core digital processing system is integrated into a single Xilinx Artix-7 series FPGA chip. The system hardware structure block diagram corresponds to... Figure 1 (Simplified diagram of a resonant tube density meter) mainly includes: Core sensing element: quartz glass resonant tube. Excitation and pickup elements: piezoelectric ceramic sheets attached to specific locations on the tube wall (determined through SIMO modal analysis, preferably excitation at 1 / 2L and pickup at 1 / 4L and 3 / 4L). Signal processing core: FPGA, used to implement all digital algorithms such as quadrature phase-locked loop, narrowband filtering, digital feature extraction, and convergence discrimination. Data converter: high-precision ADC for acquiring the pickup signal, and DAC for outputting the excitation signal.

[0025] Example: Detailed procedure for frequency detection using a resonant tube density meter

[0026] 1. Signal excitation and quadrature phase-locked loop processing

[0027] The frequency scanning and control module generates a frequency of f within the FPGA. i The digital excitation signal drives the piezoelectric ceramic excitation plate via a DAC, exciting the resonant tube to vibrate. The vibration pickup signal is sampled by an ADC and then sent to the quadrature phase-locked loop system within the FPGA.

[0028] The structure of this quadrature phase-locked loop system corresponds to Figure 3 (Simplified diagram of orthogonal phase-locked loop system structure), its core is two parallel phase-sensitive detector units. Figure 2 (Simplified diagram of phase-sensitive detector unit structure) shows the detailed composition of each unit:

[0029] Multiplication mixing: The input signal S_in(t) is multiplied by the orthogonal reference signal S_ref(t) (sin(ωt) or cos(ωt)) generated by the numerically controlled oscillator to obtain the modulation signal S_mod(t).

[0030] Low-pass filtering: S_mod(t) is low-pass filtered to extract the DC component. This invention makes key improvements to the low-pass filter at this stage, as detailed in the next section.

[0031] 2. Implementation of a narrow-bandwidth low-pass filter

[0032] To address the difficulties of implementing traditional filters in narrow bandwidth, this invention employs... Figure 4 The diagram (showing a cascaded structure of a narrow-bandwidth low-pass filter) illustrates a cascaded CIC decimation filter and an FIR compensation filter. This module replaces... Figure 2 The conventional low-pass filter in the process.

[0033] 3. Recursive Update and Convergence Criterion of Digital Features

[0034] After narrowband filtering and amplitude calculation, the instantaneous amplitude sequence A(k) is obtained. This invention abandons the traditional fixed-window-length averaging method and introduces... Figure 5 The digital feature processing flow is shown in the (Welford algorithm framework diagram).

[0035] Welford recursive update: This module receives the amplitude sequence A(k) and updates the mean μ of the sequence in real time according to the Welford recursive formula. k and variance

[0036]

[0037] The advantage of this algorithm is that it does not require storing the entire historical data sequence, has low computational cost, and is very suitable for pipelined implementation on FPGAs. Convergence determination: The convergence determination module receives the variance sequence output by the Welford module. This invention proposes two preferred discrimination schemes, which can be selected either one in practical applications:

[0038] Option 1 (Sliding Window Threshold Method for Relative Change Rates of Adjacent Elements): Within a sliding window of length L, calculate the relative change rates of the latest L variance values. Convergence is determined when all change rates within the window are less than the threshold.

[0039] Option 2 (Normalized Range Sliding Window Threshold Method): Within a sliding window of length L, calculate the normalized range of the variance values. Convergence is determined when the normalized range is less than a threshold. Assume the sliding window Win(x1, x2, ..., x...). n If the length of the given array is n, then the normalized range is calculated as follows:

[0040]

[0041] The preferred embodiment is Scheme 2, because it exhibits lower absolute and relative errors and more stable statistical characteristics in subsequent simulations.

[0042] 4. Amplitude Output and Frequency Determination

[0043] When the convergence discrimination module sends a "convergence" signal, the output control module will set the mean value μ at the current time. k This serves as the final effective stable amplitude output at that excitation frequency.

[0044] The system then switches the excitation frequency and repeats the above process to obtain a stable amplitude curve across the entire frequency sweep range.

[0045] The frequency determination module finds the peak amplitude by comparing all historical amplitude data, and the corresponding frequency is the natural frequency of the system. Finally, the medium density is calculated using the calibration formula.

[0046] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A signal amplitude extraction method based on adaptive convergence discrimination, characterized in that, Includes the following steps: Step 1: Acquire the input signal generated by or associated with a resonant sensor; Step 2: Demodulate the amplitude features of the input signal to obtain an amplitude sequence composed of multiple sampling points; Step 3: Update the digital features of the amplitude sequence in real time using an online recursive method. The digital features include at least the mean and variance. Step 4: During the update process, the stability of the variance or its related statistics is determined in real time based on a preset convergence discrimination strategy. Step 5: When the convergence condition is met, output the mean obtained by the current recursion as the effective magnitude estimation result; The effective amplitude is used to determine the resonant characteristic frequency of the sensor's vibration state.

2. The method according to claim 1, characterized in that: The online recursive method is an incremental update algorithm for the mean and variance. This incremental update algorithm does not require storing all historical data points, and the mean μ at any given time is... k With variance Based on the current input sampling point x k It is calculated from the previous update result.

3. The method according to claim 2, characterized in that: The incremental update algorithm includes a statistical update method based on Welford recursion, and its update rule is as follows: Where x k (k = 1, 2, ..., n) is a discrete sequence of continuous input μ k and As x k The mean and variance are updated in real time, and the initial condition μ0 = 1 is satisfied.

4. The method according to any one of claims 1 to 3, characterized in that... The convergence discrimination strategy includes a discrimination method based on the rate of change of adjacent statistics, namely: setting a sliding window to store the relative rate of change of several continuously updated variance statistics, and determining that convergence has been achieved when all rates of change within the window are lower than a threshold.

5. The method according to any one of claims 1 to 3, characterized in that... The convergence discrimination strategy includes a discrimination method based on the stability of statistical distribution, namely: calculating the normalized range of variance statistic in a sliding window, and determining that convergence has been achieved when the statistic is below a threshold.

6. The method according to claim 1, characterized in that... The amplitude feature demodulation includes an orthogonal phase-locked loop step, namely: the input signal is transformed with two mutually orthogonal reference signals respectively, and the amplitude of the fundamental frequency component is extracted from the transformed signal to form the amplitude sequence.

7. The method according to claim 6, characterized in that... The extraction of the fundamental frequency component is achieved through a narrow-bandwidth low-pass filter, which consists of a cascaded integrator comb filter and a finite-length impulse response filter. The decimation rate and filter order can be adaptively configured according to the target bandwidth and the implementation platform.

8. The method according to claim 7, characterized in that The cascaded integrator-comb filter includes an integrator and a comb section. Its output, after gain compensation, is used as the input of the FIR filter to achieve a balance between narrow bandwidth characteristics and resource overhead.

9. A frequency detection method for a resonant sensor, characterized in that, include: (1) Acquire sensor vibration pickup signals at multiple excitation frequencies; (2) Perform the method of any one of claims 1 to 8 on the pickup signal at each excitation frequency to obtain the corresponding effective amplitude; (3) The excitation frequency corresponding to the maximum effective amplitude is taken as the resonant characteristic frequency.

10. The method according to claim 9, characterized in that... The resonant sensor is a resonant tube density meter, and its natural frequency and the density of the medium inside the tube are mapped through a calibration function.

11. An adaptive amplitude extraction system, characterized in that, include: The amplitude demodulation module is used to extract the fundamental frequency amplitude of the input signal and output an amplitude sequence; The digital feature recursion module is used to update the mean and variance of the amplitude sequence in real time in an incremental manner; The convergence determination module is used to determine whether convergence has been achieved based on statistical change characteristics; The output control module is used to output the mean value as the effective amplitude when the convergence condition is met. All of these modules can be implemented in a programmable digital logic device.

12. The system according to claim 11, characterized in that... The amplitude demodulation module includes an orthogonal reference signal generation unit, a signal transformation unit, and an amplitude calculation unit.

13. The system according to claim 12, characterized in that... The output of the signal conversion unit passes through a narrow-bandwidth low-pass filter unit composed of an integral comb filter and an FIR filter to suppress noise and harmonic interference.

14. A resonant sensor frequency detection system, characterized in that, include: The signal amplitude extraction system as described in any one of claims 11 to 13 is used to process the vibration pickup signal of a resonant sensor; The frequency scanning and control module is configured to generate the frequency-variable excitation signal and control the excitation frequency to switch between different frequency points. The frequency determination module is configured to receive the effective amplitude output by the signal amplitude extraction system at different excitation frequency points, and determine the frequency corresponding to the maximum effective amplitude as the inherent frequency by comparison.

15. The system according to claim 14, characterized in that: The signal amplitude extraction system, frequency scanning and control module, and frequency determination module are all integrated into the same field-programmable gate array (FPGA) chip.

16. A density measuring device, characterized in that, include: The resonant transistor serves as the core sensing element. A piezoelectric ceramic excitation plate and a piezoelectric ceramic pickup plate are attached to the wall of the resonant tube; the resonant sensor frequency detection system as described in claim 14 or 15, wherein its output terminal is connected to the piezoelectric ceramic excitation plate and its input terminal is connected to the piezoelectric ceramic pickup plate; The natural frequency determined by the resonant sensor frequency detection system is used to calculate the density of the medium inside the resonant tube.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 10.

Images