A dynamic dual-adaptive composite filtering method suitable for environments with strong electromagnetic interference.

Abstract

The application discloses a dynamic double self-adaptive composite filtering method and system suitable for a strong electromagnetic interference environment, and belongs to the technical field of signal processing and data acquisition. In view of the problem that signals in a strong electromagnetic interference environment such as a nuclear fusion device are superimposed by wideband random noise and high-frequency stray noise, a composite filtering method based on the combination of a dynamic step length normalized least mean square (NLMS) adaptive filtering and a finite impulse response (FIR) low-pass filtering is provided. The method realizes the cooperative dynamic adjustment of the NLMS filtering step length parameter and the FIR filtering cutoff frequency by constructing a double adaptive mechanism, wherein the step length parameter is adaptively updated according to the error signal energy to balance the convergence speed and the steady-state error, and the cutoff frequency is dynamically adjusted according to the input signal spectrum characteristics to suppress the high-frequency noise. Compared with a traditional fixed parameter filtering method, the application can effectively improve the signal-to-noise ratio in a time-varying noise environment and maintain the signal amplitude and phase characteristics, and is suitable for real-time signal processing scenes in a high-speed data acquisition system.

Description

Technical Field

[0001] This invention relates to the field of signal processing and high-speed data acquisition technology, and to an adaptive composite filtering method and system suitable for strong electromagnetic interference environments, which can adaptively adjust the step size attenuation coefficient and cutoff frequency according to the estimated noise intensity and signal main frequency. Background Technology

[0002] In nuclear fusion devices, power system transient monitoring, and strong electromagnetic industrial environments, the voltage, current, and other signals acquired by sensors are often accompanied by complex electromagnetic interference. This type of interference typically manifests as a superposition of broadband random noise and high-frequency spurious noise, characterized by a wide spectral distribution, drastic amplitude variations, and strong time-varying properties, severely affecting the signal-to-noise ratio and the accuracy of subsequent feature extraction.

[0003] Commonly used filtering methods in existing technologies mainly include two categories: fixed-coefficient filtering and adaptive filtering. Fixed-coefficient filtering (such as FIR or IIR filters) suppresses specific frequency bands by pre-designing filter parameters. It has the advantages of simple implementation and high stability. However, its filter parameters are fixed, making it difficult to adapt to the changes in noise spectrum over time in strong electromagnetic environments. When the filter bandwidth is not properly selected, signal distortion or insufficient noise suppression can easily occur.

[0004] Adaptive filtering methods (such as LMS or NLMS algorithms) can continuously adjust the filter weights based on the error signal, thereby achieving dynamic tracking of time-varying noise. However, traditional adaptive filtering methods typically use a fixed step size parameter, which makes it difficult to balance convergence speed and steady-state error in noisy environments: while a larger step size can improve the convergence speed, it can easily lead to increased fluctuations in the output signal; while a smaller step size can reduce the steady-state error, it has a weaker response to noise changes.

[0005] Furthermore, a single filter structure is often insufficient to effectively suppress both broadband random noise and high-frequency spurious noise simultaneously. For example, adaptive filters have good suppression effects on broadband noise, but their ability to filter out high-frequency narrowband interference is limited; while low-pass filters can suppress high-frequency interference, if the cutoff frequency is fixed, they can easily cause attenuation of useful signal components under different signal scenarios.

[0006] In summary, existing technologies have the following shortcomings: fixed filtering parameters make it difficult to adapt to time-varying noise environments; a single filtering method cannot meet the needs of suppressing multiple types of noise; and in environments with strong electromagnetic interference, it is difficult to simultaneously ensure signal fidelity and filtering effect.

[0007] Therefore, it is necessary to propose a composite filtering method that can dynamically adjust both the filtering step size and the filtering bandwidth simultaneously to improve signal processing performance in complex electromagnetic environments. Summary of the Invention

[0008] The purpose of this invention is to overcome the above-mentioned technical deficiencies and provide a composite filtering scheme of dynamic step size NLMS adaptive filtering + zero phase FIR low-pass filtering, as well as a dual adaptive mechanism. It effectively suppresses noise in the context of high-frequency noise characteristics and complex time-varying noise environments under strong electromagnetic conditions, and solves the problems of large steady-state error after fixed parameter convergence, low signal-to-noise ratio of output signal, and waveform distortion.

[0009] To achieve the above objectives, the technical solution of the present invention provides a dual adaptive composite filtering scheme, comprising the following steps:

[0010] S1: Acquire the input signal to be processed output by the data acquisition module, wherein the input signal includes broadband random noise and high-frequency electromagnetic interference noise;

[0011] S2: Input the input signal into the Normalized Least Mean Square (NLMS) adaptive filter, and iteratively update the filter weights based on the current error signal and the energy of the input signal to obtain the first filtered output signal;

[0012] Its weight update formula is:

[0013]

[0014] in and Here, μ represents the filter weights, and μ is a fixed step size. For error signals, The input signal is denoted as μ. The step size μ is dynamically adjusted to adaptively adapt to the input signal strength. Let the total number of iterations be N, and the initial step size be μ. 0, final step size is N attenuation coefficient for:

[0015] γ=exp[ln(μN / μ0) / N]

[0016] S3: The attenuation coefficient is dynamically adjusted based on the real-time estimated noise intensity, and the step size parameter of the NLMS filter is adaptively adjusted in real time, so that the NLMS algorithm can always maintain the optimal balance between convergence speed and steady-state accuracy in time-varying electromagnetic environments.

[0017] The aforementioned real-time noise power estimation method specifically employs a sliding variance estimation method based on the error signal. The sliding estimate of the noise power is as follows:

[0018]

[0019] in For error signals, This represents the length of the sliding window.

[0020] The formula for calculating the instantaneous signal-to-noise ratio is:

[0021]

[0022] It is the average power of the input signal over the most recent M sampling points, which reflects the signal strength.

[0023] The attenuation coefficient The signal-to-noise ratio is dynamically adjusted based on the instantaneous signal-to-noise ratio. The specific formula is as follows:

[0024]

[0025] in =15dB is the threshold. =0.9999, =0.999.

[0026] S4: Input the first filtered output signal into a Finite Impulse Response (FIR) low-pass filter for second-stage filtering;

[0027] The filtering module specifically works as follows: First, a linear phase output is obtained through forward FIR filtering. Let the input signal be... This is the output after NLMS filtering. Based on the principle of linear convolution, the output of the forward filter is:

[0028]

[0029] in Represents linear convolution. The delay term of the input signal. These are the weights of each delay term. The output at this point is... It has linear phase characteristics. To compensate for the phase delay, the positive output is first... Perform a reverse flip to obtain an inverted signal. Defined as:

[0030] Then The same FIR low-pass filter is input again for forward filtering to obtain the secondary filtered output. The formula is:

[0031]

[0032] Finally, the output is flipped in reverse again to obtain the final zero-phase output. .

[0033] S5: Based on the spectral characteristics, signal-to-noise ratio, or main frequency information of the input signal or the first filtered output signal, dynamically adjust the cutoff frequency of the FIR low-pass filter;

[0034] The aforementioned signal main frequency estimation method specifically involves: firstly, performing a windowed FFT transform on the input signal, the formula of which is:

[0035]

[0036] This is the standard formula for the Discrete Fourier Transform (DFT). It is a time-domain sampled signal. This is the Hanning window function, used to reduce spectral leakage. N is the number of FFT points, and k is the frequency index. It is a frequency domain representation.

[0037] Furthermore, the true peak position offset is calculated through parabolic fitting:

[0038]

[0039] in , , The amplitude values ​​of the main peak and its adjacent points on the left and right. The correction is calculated using quadratic interpolation. It is assumed that the area around the spectral peak can be fitted with a quadratic function. Indicates the true peak position relative to The offset. Finally, the accurate clock frequency estimate is obtained:

[0040]

[0041] Based on the estimated clock frequency The cutoff frequency of the FIR filter is dynamically calculated. The operating mode selection is determined by the currently estimated signal-to-noise ratio, specifically by the following formula:

[0042]

[0043] in The current estimated signal-to-noise ratio, The signal-to-noise ratio (SNR) threshold is used, and the operating modes are divided into strict mode and relaxed mode. Strict filtering mode is suitable for low SNR environments, and its cutoff frequency is calculated using the following formula: The relaxed filtering mode is suitable for high signal-to-noise ratio environments, and its cutoff frequency is calculated using the following formula: .

[0044] S6: Output the target signal after processing by the NLMS adaptive filtering and FIR low-pass filtering. Attached Figure Description

[0045] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0046] Figure 1 This is a framework diagram of the dynamic dual adaptive composite filtering method applicable to strong electromagnetic interference environments according to the present invention.

[0047] Figure 2 This is a diagram showing the result of dual adaptive composite filtering according to an embodiment of the present invention.

[0048] Figure 3 These are the curves showing the variation of the NLMS step size and the attenuation coefficient in the adaptive mechanism of this invention. Detailed Implementation

[0049] 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 for illustrative purposes only and are not intended to limit the invention.

[0050] Figure 1 The basic framework diagram of the dynamic dual adaptive composite filtering method for strong electromagnetic interference environments proposed in this invention is illustrated, including the following steps:

[0051] S1: Acquire the input signal to be processed output by the data acquisition module, wherein the input signal includes broadband random noise and high-frequency electromagnetic interference noise;

[0052] S2: Input the input signal into the Normalized Least Mean Square (NLMS) adaptive filter, and iteratively update the filter weights based on the current error signal and the energy of the input signal to obtain the first filtered output signal;

[0053] Its weight update formula is:

[0054]

[0055] in and Here, μ represents the filter weights, and μ is a fixed step size. For error signals, The input signal is denoted as μ. The step size μ is dynamically adjusted to adaptively adapt to the input signal strength. Let the total number of iterations be N, and the initial step size be μ. 0, final step size is N attenuation coefficient for:

[0056] γ=exp[ln(μN / μ0) / N]

[0057] S3: The attenuation coefficient is dynamically adjusted based on the real-time estimated noise intensity, and the step size parameter of the NLMS filter is adaptively adjusted in real time, so that the NLMS algorithm can always maintain the optimal balance between convergence speed and steady-state accuracy in time-varying electromagnetic environments.

[0058] The aforementioned real-time noise power estimation method specifically employs a sliding variance estimation method based on the error signal. The sliding estimate of the noise power is as follows:

[0059]

[0060] in For error signals, This represents the length of the sliding window.

[0061] The formula for calculating the instantaneous signal-to-noise ratio is:

[0062]

[0063] It is the average power of the input signal over the most recent M sampling points, which reflects the signal strength.

[0064] The attenuation coefficient The signal-to-noise ratio is dynamically adjusted based on the instantaneous signal-to-noise ratio. The specific formula is as follows:

[0065]

[0066] in =15dB is the threshold. =0.9999, =0.999.

[0067] S4: Input the first filtered output signal into a Finite Impulse Response (FIR) low-pass filter for second-stage filtering;

[0068] The filtering module specifically works as follows: First, a linear phase output is obtained through forward FIR filtering. Let the input signal be... This is the output after NLMS filtering. Based on the principle of linear convolution, the output of the forward filter is:

[0069]

[0070] in Represents linear convolution. The delay term of the input signal. These are the weights of each delay term. The output at this point is... It has linear phase characteristics. To compensate for the phase delay, the positive output is first... Perform a reverse flip to obtain an inverted signal. Defined as:

[0071] Then The same FIR low-pass filter is input again for forward filtering to obtain the secondary filtered output. The formula is:

[0072]

[0073] S5: Based on the spectral characteristics, signal-to-noise ratio, or main frequency information of the input signal or the first filtered output signal, dynamically adjust the cutoff frequency of the FIR low-pass filter;

[0074] The aforementioned signal main frequency estimation method specifically involves: firstly, performing a windowed FFT transform on the input signal, the formula of which is:

[0075]

[0076] This is the standard formula for the Discrete Fourier Transform (DFT). It is a time-domain sampled signal. This is the Hanning window function, used to reduce spectral leakage. N is the number of FFT points, and k is the frequency index. It is a frequency domain representation.

[0077] Furthermore, the true peak position offset is calculated through parabolic fitting:

[0078]

[0079] in , , The amplitude values ​​of the main peak and its adjacent points on the left and right. The correction is calculated using quadratic interpolation. It is assumed that the area around the spectral peak can be fitted with a quadratic function. Indicates the true peak position relative to The offset. Finally, the accurate clock frequency estimate is obtained:

[0080]

[0081] Based on the estimated clock frequency The cutoff frequency of the FIR filter is dynamically calculated. The operating mode selection is determined by the currently estimated signal-to-noise ratio, specifically by the following formula:

[0082]

[0083] in The current estimated signal-to-noise ratio, The signal-to-noise ratio (SNR) threshold is used, and the operating modes are divided into strict mode and relaxed mode. Strict filtering mode is suitable for low SNR environments, and its cutoff frequency is calculated using the following formula: The relaxed filtering mode is suitable for high signal-to-noise ratio environments, and its cutoff frequency is calculated using the following formula: .

[0084] S6: Output the target signal after processing by the NLMS adaptive filtering and FIR low-pass filtering.

[0085] In this case, the filtering method was tested using the Matlab platform. Specifically, the effectiveness of the proposed dual adaptive mechanism was verified by constructing a test signal. .in For time-varying main frequency, It is time-varying noise. The input is standard Gaussian white noise. In the simulation, the sampling rate Fs=10 MSPS, the signal duration T=1 ms, the input is a test signal with a sinusoidal frequency ranging from 0.5 to 1.5 MHz, and time-varying Gaussian white noise with intensity modulated by a 50Hz sinusoidal modulation is superimposed.

[0086] The result after dual adaptive composite filtering is shown in the figure below. Figure 2 As shown, it contains a total of 4 subgraphs, among which Figure 1 This is a time-domain comparison graph of a clean time-varying master frequency signal and a noisy input signal with added time-varying frequency noise. Figure 2 To obtain the time-domain signal of the signal after dual adaptive filtering, Figure 3 Figure 4 shows a comparison of the frequency domain analysis of the three signals, with the adaptive method and the fixed parameter method showing an error comparison. As can be seen from the figures, after dual adaptive filtering, the output signal's time-domain waveform closely matches the clean signal, and residual noise is effectively suppressed. Frequency domain analysis shows that the spectral peak at the signal's main frequency remains intact, while out-of-band noise is attenuated by approximately 20-25 dB. The initial signal-to-noise ratio (SNRin) is 2.3 dB, and the final output SNRout = 13.8 dB, an improvement of 11.5 dB over the input, and superior to the fixed parameter scheme (fixed attenuation coefficient γ = 0.9995, fixed cutoff frequency f). c The SNR fixed at 1.8 MHz is 7.95 dB, representing a relative improvement of 5.85 dB. This demonstrates that the dual adaptive mechanism of this invention is significantly more effective than fixed-parameter filtering methods in handling complex time-varying noise environments.

[0087] Furthermore, the variation curves of the NLMS step size and the decay coefficient in the adaptive mechanism are as follows: Figure 3As shown in the figure, the NLMS step size decay coefficient γ(n) is dynamically adjusted within the range of 0.9990 to 0.99990 based on the real-time estimated noise power. The figure shows the curves of the NLMS step size decay coefficient versus the step size. It can be seen from the figure that in the period before 0.3ms, which is a weak noise period, γ(n) remains above 0.9998, and the step size decays slowly, maintaining a relatively fast convergence speed. Afterwards, in the mostly strong noise period, γ(n) decays to below 0.9995, and the step size decays rapidly, allowing the algorithm to quickly enter a small step size state to reduce steady-state error. The step size μ(n) adaptively changes with γ(n), with an initial step size μ0 = 0.01, eventually converging to around μmin = 0.0005, achieving a dynamic balance between convergence speed and steady-state accuracy.

[0088] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A dynamic dual-adaptive composite filtering method suitable for environments with strong electromagnetic interference, characterized in that, Includes the following steps: S1: Acquire the input signal to be processed output by the data acquisition module, wherein the input signal includes broadband random noise and high-frequency electromagnetic interference noise; S2: Input the input signal into the Normalized Least Mean Square (NLMS) adaptive filter, and iteratively update the filter weights based on the current error signal and the energy of the input signal to obtain the first filtered output signal; S3: The attenuation coefficient is dynamically adjusted based on the real-time estimated noise intensity, and the step size parameter of the NLMS filter is adaptively adjusted in real time, so that the NLMS algorithm can always maintain the optimal balance between convergence speed and steady-state accuracy in time-varying electromagnetic environments. S4: Input the first filtered output signal into a Finite Impulse Response (FIR) low-pass filter for second-stage filtering; S5: Based on the spectral characteristics, signal-to-noise ratio, or main frequency information of the input signal or the first filtered output signal, dynamically adjust the cutoff frequency of the FIR low-pass filter; S6: Output the target signal after processing by the NLMS adaptive filtering and FIR low-pass filtering.

2. According to claim 1, the input signal is input into the normalized least mean square (NLMS) adaptive filter, and its weight update formula is as follows: in and Here, μ represents the filter weights, and μ is a fixed step size. For error signals, The input signal is denoted as μ. The step size μ is dynamically adjusted to adaptively adapt to the input signal strength. Let the total number of iterations be N, and the initial step size be μ. 0, final step size is N attenuation coefficient for: γ=exp[ln(μN / μ0) / N].

3. According to claim 1, the attenuation coefficient is dynamically adjusted based on the real-time estimated noise intensity, and the step size parameter of the NLMS filter is adaptively adjusted in real time. The aforementioned real-time noise power estimation method specifically employs a sliding variance estimation method based on the error signal. The sliding estimate of the noise power is as follows: in For error signals, This represents the length of the sliding window. The formula for calculating the instantaneous signal-to-noise ratio is: It is the average power of the input signal over the most recent M sampling points, which reflects the signal strength. The attenuation coefficient The signal-to-noise ratio is dynamically adjusted based on the instantaneous signal-to-noise ratio. The specific formula is as follows: in =15dB is the threshold. =0.9999, =0.

999.

4. The first filtered output signal is input into a finite impulse response (FIR) low-pass filter for second-stage filtering processing as claimed in claim 1. First, a linear phase output is obtained through forward FIR filtering. Let the input signal be... This is the output after NLMS filtering. Based on the principle of linear convolution, the output of the forward filter is: in Represents linear convolution. The delay term of the input signal. These are the weights of each delay term. The output at this point is... It has linear phase characteristics. To compensate for the phase delay, the positive output is first... Perform a reverse flip to obtain an inverted signal. Defined as: Then The same FIR low-pass filter is input again for forward filtering to obtain the secondary filtered output. The formula is: Finally, the output is flipped in reverse again to obtain the final zero-phase output. .

5. The cutoff frequency of the FIR low-pass filter is dynamically adjusted according to the spectral characteristics, signal-to-noise ratio, or main frequency information of the input signal or the first filtered output signal as claimed in claim 1. First, perform a windowed FFT transform on the input signal: This is the standard formula for the Discrete Fourier Transform (DFT). It is a time-domain sampled signal. This is the Hanning window function, used to reduce spectral leakage. N is the number of FFT points, and k is the frequency index. It is a frequency domain representation. Furthermore, the true peak position offset is calculated through parabolic fitting: in , , The amplitude values ​​of the main peak and its adjacent points on the left and right. The correction is calculated using quadratic interpolation. It is assumed that the area around the spectral peak can be fitted with a quadratic function. Indicates the true peak position relative to The offset. Finally, the accurate clock frequency estimate is obtained: Based on the estimated clock frequency The cutoff frequency of the FIR filter is dynamically calculated. The operating mode selection is determined by the currently estimated signal-to-noise ratio, specifically by the following formula: in The current estimated signal-to-noise ratio, The signal-to-noise ratio (SNR) threshold is used, and the operating modes are divided into strict mode and relaxed mode. Strict filtering mode is suitable for low SNR environments, and its cutoff frequency is calculated using the following formula: The relaxed filtering mode is suitable for high signal-to-noise ratio environments, and its cutoff frequency is calculated using the following formula: .

Images