A small-signal noise reduction processing method

Abstract

This invention proposes a small-signal denoising method, comprising: Step 1, receiving a collected analog signal and performing frequency detection to obtain the frequency of the analog signal; Step 2, determining the sampling method based on the frequency of the analog signal, and sampling the analog signal according to the sampling method to obtain a sampled signal; Step 3, performing an FFT transform on the sampled signal, taking the modulus of the FFT transform output and squaring it to obtain the value of the autocorrelation function in the frequency domain; Step 4, performing an IFFT transform on the value of the autocorrelation function in the frequency domain to obtain the autocorrelation function; Step 5, extracting the effective periodic signal from the sampled signal based on the autocorrelation function. This small-signal denoising method can significantly shorten the computation cycle and improve the speed of signal processing, while also reducing resource consumption during computation and improving the utilization rate of hardware resources.

Description

Technical Field

[0001] This invention belongs to the field of electrical variable measurement technology, and more specifically, relates to a small signal noise reduction processing method. Background Technology

[0002] In the context of the widespread application of modern sensor technology, when sensors accurately sense physical quantities and collect corresponding periodic electrical signals, if the amplitude of these electrical signals is relatively small, signals with amplitudes less than a certain value are generally referred to as small signals. Small signals are widely present in many practical applications, such as monitoring bioelectrical signals (e.g., weak electrical signals in electrocardiograms and electroencephalograms) in the biomedical field, and monitoring weak vibration electrical signals of equipment operating status on industrial automated production lines.

[0003] Small signals have a significant characteristic: their amplitude is usually close to or lower than the noise level. This means that the signal is easily masked by noise, making it difficult to accurately obtain key parameter values, such as the RMS value, in practical signal processing. The RMS value, as an important indicator of signal energy, is crucial for understanding signal characteristics and evaluating the information it carries. When small signals are masked by noise, the RMS value obtained using conventional methods often contains significant errors and fails to accurately reflect the actual state of the signal.

[0004] To improve the accuracy of small-signal parameter calculations, specialized data processing is often required for small signals with low signal-to-noise ratios in practical applications. In the field of signal processing, there are numerous methods for small-signal denoising, such as filtering, modulation amplification and demodulation, and feedback compensation. However, these methods typically suffer from drawbacks such as limited applicability, difficulty in parameter selection, and a tendency to distort effective periodic signals.

[0005] Among numerous data processing methods, the autocorrelation algorithm is a very useful tool. The core principle of the autocorrelation algorithm lies in its ability to cleverly separate zero-mean noise from periodic signals. In signals, noise typically exhibits random and irregular characteristics, while effective periodic signals possess a certain periodicity and regularity. By calculating and analyzing the autocorrelation function of a signal, the autocorrelation algorithm can effectively extract the periodic components in the signal, thereby separating the signal from the noise. Currently, the application of the autocorrelation algorithm for signal denoising is mainly performed in the time domain. However, because this process requires repeated traversal and calculation of the signal, the computational load increases significantly with the signal length. This massive computational load means that the algorithm consumes a large amount of computing resources and time during operation. Its practical computation places high demands on computing resources and performance. Computing platforms capable of meeting these high computational requirements are mainly based on ARM or x86 architectures, but even so, their computational cycle remains relatively long. Summary of the Invention

[0006] To address the problems of high computational complexity, high computational resource consumption, and long computation time associated with existing small signal denoising algorithms in the time domain, this invention proposes a small signal denoising method that can solve the aforementioned problems.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A small-signal noise reduction method includes: Step 1: Receive the acquired analog signal and perform frequency detection to obtain the frequency of the analog signal. ; Step 2: Determine the sampling method based on the frequency of the analog signal, and sample the analog signal according to the sampling method to obtain the sampled signal; Step 3: Perform FFT transformation on the sampled signal, take the modulus of the FFT output and square it to obtain the value of the autocorrelation function in the frequency domain; Step 4: Perform an IFFT transform on the value of the autocorrelation function in the frequency domain to obtain the autocorrelation function; Step 5: Extract the effective periodic signal from the sampled signal based on the autocorrelation function.

[0008] In some embodiments, the method for determining the sampling method in step two is as follows: Determine the type of the analog signal: high frequency, medium frequency, or low frequency. If the analog signal is medium frequency or low frequency, perform non-uniform sampling on the analog signal. If the analog signal is high frequency, sample according to the ADC's inherent sampling frequency. Perform sampling.

[0009] In some embodiments, the non-uniform sampling method includes: (21) Divide the number of sampling points M required to contain the whole-cycle signal by the number of points n of the FFT transformation to obtain the quotient q and the remainder r; (22) Initialize the number of sampling interval points D=q and the remainder variable R=r, and sample the first two points according to the current number of sampling interval points D; (23) Determine whether R+r is greater than n. If R+r is greater than n, update D=q+1; otherwise, D=q and sample the next point according to the current sampling interval number D. (24) Calculation The remainder is used to update the R value; (25) Determine if R is equal to 0. If R is not equal to 0, return to step (23) until R is equal to 0.

[0010] In some embodiments, the method for determining the number of sampling points M required to include an integer period signal is as follows: .

[0011] In some embodiments, for low-frequency analog signals, the number of sampling points M satisfies the condition of piecing together two complete signal cycles, i.e., m=1.

[0012] In some embodiments, for intermediate frequency analog signals, the number of sampling points M satisfies the requirement of piecing together... A complete signal cycle, where m is an integer greater than 1.

[0013] In some embodiments, a frequency greater than 200kHz is considered high frequency when the frequency of the analog signal is high frequency.

[0014] In some embodiments, the boundary value between the intermediate frequency and low frequency of the analog signal is... ,in is the inherent sampling frequency of the ADC, and n is the number of points in the FFT transform.

[0015] In some embodiments, the small signal noise reduction processing method is implemented based on FPGA.

[0016] Compared with existing technologies, the advantages and positive effects of this invention are as follows: The small-signal noise reduction method of this invention, based on the Fourier transform characteristics of signals, transforms the signal from the time domain to the frequency domain and utilizes the characteristics of the signal in the frequency domain to calculate the autocorrelation function. According to the properties of the Fourier transform, convolution operations in the time domain correspond to multiplication operations in the frequency domain, which greatly reduces the computational load of calculating the autocorrelation function in the frequency domain. This not only significantly shortens the calculation cycle and improves the speed of signal processing, but also reduces resource consumption during the calculation process and improves the utilization rate of hardware resources.

[0017] Other features and advantages of the present invention will become clearer after reading the detailed description of the embodiments of the present invention in conjunction with the accompanying drawings. Attached Figure Description

[0018] Figure 1 This is a flowchart of an embodiment of the small signal noise reduction processing method proposed in this invention; Figure 2 This is a schematic diagram of a circuit implementing the small signal noise reduction processing method proposed in this invention; Figure 3 This is a flowchart of a non-uniform sampling method in one embodiment of the small signal noise reduction processing method proposed in this invention; Figure 4 This is a schematic diagram of an analog signal in one embodiment of the small signal noise reduction processing method proposed in this invention; Figure 5This is a schematic diagram of the processed analog signal in one embodiment of the small signal noise reduction method proposed in this invention. Detailed Implementation

[0019] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1, see Figure 1 As shown, this embodiment proposes a small-signal noise reduction processing method, including: Step 1: Receive the acquired analog signal and perform frequency detection to obtain the frequency of the analog signal. .

[0022] Frequency detection can be achieved using existing detection technologies.

[0023] In some embodiments, the sensor is used to sense physical quantities and convert them into corresponding periodic electrical signals, which are collected analog signals. The reciprocal of the period of the electrical signal is the frequency of the analog signal.

[0024] Step two: Determine the sampling method based on the frequency of the analog signal, and sample the analog signal according to the sampling method to obtain the sampled signal.

[0025] Step 3: Perform an FFT transform on the sampled signal, take the modulus of the FFT (Fast Fourier Transform) output and square it to obtain the value of the autocorrelation function in the frequency domain.

[0026] Step four: Perform an IFFT (Inverse Fast Fourier Transform) on the value of the autocorrelation function in the frequency domain to obtain the autocorrelation function.

[0027] Step 5: Extract the effective periodic signal from the sampled signal based on the autocorrelation function.

[0028] The small-signal noise reduction method of this invention is based on the Fourier transform characteristics of signals. It transforms the signal from the time domain to the frequency domain and utilizes the characteristics of the signal in the frequency domain to calculate the autocorrelation function. According to the properties of the Fourier transform, convolution operations in the time domain correspond to multiplication operations in the frequency domain, which significantly reduces the computational load of calculating the autocorrelation function in the frequency domain. This not only significantly shortens the computation cycle and improves the speed of signal processing, but also reduces resource consumption during the computation process and improves the utilization rate of hardware resources.

[0029] Suppose the small signal input to the system is x(t), and its composition is as follows: .

[0030] Where s(t) is a periodic signal and n(t) is zero-mean noise. Its autocorrelation function is: .

[0031] Normally, a periodic signal s(t) is uncorrelated with zero-mean noise n(t), then: .

[0032] The power spectrum of zero-mean noise is uniformly distributed across the entire frequency range, and its autocorrelation function R0 n (τ) is mainly reflected near τ=0. When τ is large, the autocorrelation function R of x(t) is... x (τ) basically only reflects R s In the case of (τ), this achieves the separation of effective periodic signals and noise.

[0033] As can be seen from the above formula, the time-domain autocorrelation algorithm involves integral operations, and even in the discrete form, it requires solving a large number of multiplications and sums. The correlation theorem can be used to convert from the time domain to the frequency domain and reduce the computational load. Taking the Fourier transform of the autocorrelation function of x(t) yields: .

[0034] In this way, the complex multiplication and summation operations can be transformed into multiplication operations. Then, by taking the inverse Fourier transform, the autocorrelation function of x(t) can be obtained.

[0035] In some embodiments, the method for determining the sampling method in step two is as follows: Determine the type of the analog signal: high frequency, medium frequency, or low frequency. If the analog signal is medium frequency or low frequency, perform non-uniform sampling. If the analog signal is high frequency, use the ADC's inherent sampling frequency. Perform sampling.

[0036] The small signal to be denoised enters the acquisition circuit, which acquires the signal and converts the analog signal into a digital signal. The signal then enters the autocorrelation algorithm module to solve the autocorrelation function, sampling the signal differently according to the input signal frequency to avoid spectral leakage.

[0037] In this scheme, analog signals are classified according to their frequency. For high-frequency signals, the sampling frequency of the ADC is used. Sampling is sufficient. Since the signal involved in the FFT has a large number of periods, the spectral leakage can be ignored, so direct sampling is performed.

[0038] When a medium frequency signal is input, non-uniform sampling is used to piece the signal together into 2. m (m>1) complete signal cycles.

[0039] When a low-frequency signal is input, non-uniform sampling is used to piece the signal together into two complete signal cycles.

[0040] In some embodiments, a frequency of 200kHz or higher is considered high frequency when the frequency of the analog signal is not less than 200kHz, while the frequencies of intermediate and high frequency signals are less than 200kHz.

[0041] In some embodiments, the boundary value between the intermediate frequency and low frequency of the analog signal is... ,in Here, n is the inherent sampling frequency of the ADC, and n is the number of points in the FFT transform. That is, the frequency is... The frequency range is from 200kHz to 100kHz, which is less than the intermediate frequency. It is a low-frequency signal.

[0042] In some embodiments, the non-uniform sampling method includes: (21) Divide the number of sampling points M required to contain the whole-cycle signal by the number of points n of the FFT transform to obtain the quotient q and the remainder r. That is, M÷n=q…r.

[0043] (22) Initialize the number of sampling interval points D=q and the remainder variable R=r, and sample the first two points according to the current sampling interval point number D. That is, first sample the first point, and then sample the second point according to the current sampling interval point number D.

[0044] (23) Determine whether R+r is greater than n. If R+r is greater than n, update D=q+1. Otherwise, D=q and sample the next point according to the updated current sampling interval point number D.

[0045] (24) Calculation The remainder is used to update the R value. That is, the R value is updated to... The remainder.

[0046] (25) Determine if R is equal to 0. If R is not equal to 0, return to step (23) until R is equal to 0.

[0047] This ensures that the signal input to the FFT is a complete signal period. After the autocorrelation function is solved, the relevant parameters of the autocorrelation function are extracted and uploaded to the host computer for calculation to obtain the effective value of the denoised small signal.

[0048] In some embodiments, the method for determining the number of sampling points M required to include an integer period signal is as follows: .

[0049] In some embodiments, for low-frequency analog signals, the number of sampling points M needs to be able to be pieced together to form two complete signal cycles, i.e., m=1.

[0050] In some embodiments, for intermediate frequency analog signals, the number of sampling points M satisfies the requirement of piecing together... A complete signal cycle, where m is an integer greater than 1.

[0051] In some embodiments, the small signal noise reduction processing method is implemented based on FPGA.

[0052] like Figure 2 As shown, the circuit used in the small signal denoising method also includes a signal conditioning circuit, a data acquisition circuit, a host computer, and a display system. The small signal to be denoised is input from the front-end analog circuit, pre-processed by the signal conditioning circuit, and then enters the data acquisition circuit. The FPGA drives the ADC to acquire the signal, converting the analog signal into a digital signal. The signal then enters the autocorrelation algorithm module to solve the autocorrelation function, and the signal is sampled differently according to the input signal frequency to avoid spectral leakage.

[0053] The autocorrelation algorithm in this solution has a further reduced computational complexity, enabling high-speed pipelined computation with low resource occupancy when deployed on an FPGA. This represents a significant advantage over traditional autocorrelation algorithms and implementations.

[0054] Figure 4 The image shows a sinusoidal signal with a frequency of 20kHz and a signal-to-noise ratio of 0dB. Figure 5 The waveform after denoising using the method of this invention is shown. The signal-to-noise ratio (SNR) of the signal after denoising is significantly increased, and the autocorrelation of the noise is mainly reflected at index zero. Table 1 shows the error rates of the small signal RMS values ​​before and after denoising using the method of this invention, with SNRs of 10dB, 5dB, 0dB, and -5dB, respectively. As can be seen from the table, the method of this invention can significantly reduce the error rate of the small signal RMS values ​​and reduce the interference of noise on the effective periodic signal.

[0055] . Table 1

[0056] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A small-signal noise reduction processing method, characterized in that, include: Step 1: Receive the acquired analog signal and perform frequency detection to obtain the frequency of the analog signal. ; Step 2: Determine the sampling method based on the frequency of the analog signal, and sample the analog signal according to the sampling method to obtain the sampled signal; Step 3: Perform FFT transformation on the sampled signal, take the modulus of the FFT output and square it to obtain the value of the autocorrelation function in the frequency domain; Step 4: Perform an IFFT transform on the value of the autocorrelation function in the frequency domain to obtain the autocorrelation function; Step 5: Extract the effective periodic signal from the sampled signal based on the autocorrelation function; In step two, when the frequency f of the analog signal is less than 200kHz, the analog signal is sampled non-uniformly; when the frequency f of the analog signal is not less than 200kHz, it is sampled according to the inherent sampling frequency of the ADC. Perform sampling; The method for determining the number of sampling points M required to contain an integer period signal is as follows: ； Where m is an integer greater than 1; The non-uniform sampling method includes: (21) Divide the number of sampling points M required to contain the whole-cycle signal by the number of points n of the FFT transformation to obtain the quotient q and the remainder r; (22) Initialize the number of sampling interval points D=q and the remainder variable R=r, and sample the first two points according to the current number of sampling interval points D; (23) Determine whether R+r is greater than n. If R+r is greater than n, update D=q+1; otherwise, D=q and sample the next point according to the current sampling interval number D. (24) Calculation The remainder is used to update the R value; (25) Determine if R is equal to 0. If R is not equal to 0, return to step (23) until R is equal to 0.

2. The small-signal noise reduction method according to claim 1, characterized in that, For low-frequency analog signals, the number of sampling points M satisfies the condition of being pieced together to form two complete signal periods, i.e., m=1; for mid-frequency analog signals, the number of sampling points M satisfies the condition of being pieced together to form two complete signal periods. A complete signal cycle, where m is an integer greater than 1.

3. The small-signal noise reduction processing method according to claim 1 or 2, characterized in that, The small signal noise reduction method is implemented based on FPGA.

Images