Ultrahigh frequency sensor denoising method based on auditory bionic model

Abstract

The invention provides an ultrahigh frequency sensor denoising method based on an auditory bionic model, and belongs to the technical field of partial discharge verification. Aiming at the problems that S parameter ultrahigh frequency signal (300MHz-1.5 GHz) processing is limited by frequency spectrum resolution and non-linear frequency perception of a biological auditory model is insufficient in a traditional method, the invention provides a verification method fusing a down-conversion technology and bionic signal enhancement. Comprising the following steps: down-converting a high-frequency signal to a baseband through digital frequency conversion; a 128-channel Gammatone filter bank is used for carrying out nonlinear frequency band division, the distribution characteristic of the basement membrane frequency is simulated, and frequency band coverage with dense low frequency and sparse high frequency is achieved; extracting a channel signal envelope through Hilbert transform; a phase-amplitude coupling mechanism is simulated in combination with a nerve distribution rate model, and dynamic gain is applied to a partial discharge characteristic frequency band (300MHz-1.5 GHz) based on an auditory attention mechanism to suppress irrelevant noise. According to the invention, through adaptive gain control of the frequency band importance weight function and the signal-to-noise ratio, suppression of irrelevant noise is realized, the key frequency band precision and the anti-interference capability are improved, and ultrahigh frequency sensor fault diagnosis and partial discharge verification can be efficiently realized.

Description

Technical Field

[0001] The present invention relates to the technical field of partial discharge verification, and in particular to a method for verifying noise signals of an ultra-high frequency sensor based on a Gammatone filter group. Background Art

[0002] Return loss S 21 As an important metric for measuring antenna performance, analysis often uses FFT or fixed-bandwidth filters, which are limited by the length and sampling rate of the time-domain signal. There is no standardized verification method in China. For S-parameters, a verification method based on ultra-high frequency sensor noise signals using a gammatone filter bank can be used. However, S-parameters are ultra-high frequency signals between 300 MHz and 1.5 GHz, while auditory bionic models are based on the human audible frequency range, with high resolution for low-frequency signals and low resolution for high-frequency signals. Therefore, traditional auditory bionic models cannot simulate the nonlinear frequency perception characteristics of the biological auditory system. Therefore, signal preprocessing is required, using frequency conversion technology to convert the signal to baseband or intermediate frequency.

[0003] This method can realize efficient processing and enhancement of the original high-frequency S parameters, improve the accuracy of key frequency bands, suppress irrelevant noise, and further improve the calibration accuracy. Summary of the Invention

[0004] The present invention proposes a UHF sensor denoising method based on an auditory bionic model, which divides the S parameter spectrum (such as S 21 The system performs three steps: (1) mapping the instantaneous amplitude / phase of the PD signal to 128 Gammatone filter channels; (2) time-frequency analysis (performing Hilbert transform on each channel signal to extract the instantaneous amplitude (similar to the cochlear nerve firing rate)); and (3) feature enhancement (applying gain to the PD characteristic frequency band (such as 300MHz-1.5GHz) and suppressing other frequency bands (similar to the auditory attention mechanism)) to achieve efficient processing and enhancement of high-frequency signals.

[0005] First, S-parameter UHF signal down-conversion

[0006] Using digital frequency conversion technology, the S parameter ultra-high frequency signal (300MHz-1.5GHz) is down-converted to baseband through a mixer, and the low-frequency signal is input into a low-pass filter, and then a low sampling rate (F s =44.1kHz). The S-parameter signal (frequency or time domain) is treated as an "auditory signal" and decomposed using a bionic filter bank, highlighting the PD's characteristic frequency band (e.g., 300MHz-1.5GHz) while suppressing irrelevant frequency bands such as power frequency noise and radio interference.

[0007] Second, the frequency band division of bionic filters

[0008] After the external speech signal enters the cochlea, it propagates along the basilar membrane. Signals of different frequencies will generate traveling waves of different frequencies, forming different resonance amplitudes at different locations of the basilar membrane. A 128-channel Gammatone filter bank is used to achieve nonlinear frequency band division, such as Figure 2 As shown. The impulse response formula of the gammatone filter is: Where h(t) is the filter impulse response, f center is the filter center frequency.

[0009] In the Gammatone filter bank, each filter is responsible for a specific frequency range. The center frequency of the filter determines the frequency it mainly processes, which conforms to the equivalent rectangular bandwidth (ERB) scale distribution, with low frequency bands spaced more closely and high frequency bands spaced less closely. The center frequency calculation formula is: According to the first preset formula:

[0010] ERB k =ERB low +k·127ERB high -ERB low ,k=0,1,...,127,

[0011] The ERB rate interval [ERB low ,ERB high ] is evenly divided into 128 points, the ERB rate of each channel is obtained, and the ERB rate is converted into the corresponding center frequency.

[0012] Third, neural firing rate simulation

[0013] 1. Extract the signal envelope after performing Hilbert transform on each channel output

[0014] The Hilbert transform formula is Where p·v· is the Cauchy principal value integral, and then the signal envelope detection is performed. To extract the key characteristic information of the signal, thereby simplifying subsequent signal analysis and processing.

[0015] 2. Introducing the neural firing rate model

[0016] The neural firing rate model is based on the phase-amplitude coupling (PAC) mechanism, which simulates the response of neurons to signal intensity and provides a biological explanation for signal processing. Where k is the steepness parameter, k = 10s -1 , is the threshold parameter, θ=0.5·max(x env (t)); convert the input signal into a biological neural signal.

[0017] Fourth, auditory attention is enhanced

[0018] 1. Establish frequency band importance weight function

[0019] The frequency band importance weight function is based on the sensitivity of the human ear to different frequency bands, and the formula is:

[0020] in: (focus on baseband high frequency band), α=0.2kHz -1 (Increase the steepness of high-frequency band weights), w(f) is the frequency band weight, and α is the steepness parameter of the weight function.

[0021] 2Dynamic gain control

[0022] Adaptively adjust the gain G(n) according to the signal-to-noise ratio (SNR), Among them: G min =6dB,G max =20dB, c is the gain adjustment steepness parameter, c=0.1dB, d is the gain adjustment center SNR value, where d=10dB,

[0023] P signal (n)=∑{t}x evn (n,t)^2*w(n);P noise (n) = medium (P signal (n-Δ:n+△)); △=5, realizing dynamic gain of the signal, thereby achieving noise reduction effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings are only used to illustrate the embodiments and are not to be considered as limiting the present invention. In addition, the same reference symbols are used to represent the same components throughout the drawings. In the drawings:

[0025] Figure 1 Schematic diagram of the process of the UHF sensor calibration and denoising method of the present invention;

[0026] Figure 2 Schematic diagram of nonlinear frequency band division for Gammatone filter bank;

[0027] Figure 3 This is a schematic structural diagram of an embodiment of a GIS UHF built-in sensor performance verification system of the present invention; DETAILED DESCRIPTION

[0028] The following is a further explanation of the present invention in conjunction with a specific implementation process. It should be emphasized that the specific implementation cases described herein are only used to explain the present invention and are not intended to limit the scope of the present invention and its claims.

[0029] Implementation Case 1

[0030] The first embodiment of the present invention discloses a method for denoising a UHF sensor based on an auditory bionic model. The verification method can be applied in actual field. The method of the embodiment of the present invention comprises the following steps: Step S1: frequency conversion processing of the S signal

[0031] Specifically, this step includes the following processes:

[0032] The S-parameter ultra-high frequency signal (300MHz-1.5GHz) is down-converted to baseband through a mixer (310), and a low-frequency signal is generated and input into a low-pass filter, and then a low sampling rate (F s =44.1KHz) ADC conversion.

[0033] Step S2: Perform bionic calculation on the S signal to achieve anti-interference processing

[0034] Specifically, this step includes the following processes:

[0035] 1. Apply the S parameter input to a 128-channel Gammatone filter bank to implement nonlinear frequency band division (330).

[0036] 2. Perform Hilbert transform on each channel output to extract the instantaneous value.

[0037] 3. Apply gain to the UHF characteristic frequency band and suppress other frequency bands.

[0038] Step S3: Calculate the average attenuation S of the attenuation S after simulation calculation. av

[0039]

[0040] S 21 (i) is the amplitude of the i-th sampling point, i is the sampling point corresponding to 300M, j is the sampling point corresponding to 1.5GHz, and N is the total number of sampling points between 300M-2GHz, which is equal to ji.

[0041] Step S4: using a standard sensor to inject a discharge signal at a sensor adjacent to the sensor to be tested, the amplitude of which is the minimum partial discharge.

[0042] Specifically, this step includes the following processes:

[0043] 1. Install a standard sensor of the same model as the sensor to be tested adjacent to the sensor to be tested.

[0044] 2. Inject an equivalent 5pc partial discharge signal through a standard sensor.

[0045] Step S5: Receive the UHF signal sent back by the sensor to be tested.

[0046] Step S6: Acquire and process the partial discharge signal data received by the sensor under test

[0047] Specifically, this step includes the following processes:

[0048] 1. The detection system reads the UHF signal (340) sent back by the sensor to be tested.

[0049] 2. The average attenuation S calculated in S3 av The sensor attenuation signal obtained in S6 is compared. If the signal attenuation is lower than the calculated value, the sensor is faulty and needs to be repaired.

Claims

1. A UHF sensor denoising method based on an auditory bionic model, characterized in that: The steps include: S1: Use digital frequency conversion technology to convert the S-parameter UHF signal to baseband; S2: Use 128-channel Gammatone filter bank to achieve nonlinear frequency band division; S3: After performing Hilbert transform on each channel output, the signal envelope is extracted and the neural firing rate model is introduced; S4: Enhanced auditory attention.

2. The S signal digital frequency conversion technology method according to claim 1, characterized in that: In step S1, the digital frequency conversion technology method is as follows: (1) Down-convert the S-parameter UHF signal (300MHz-1.5GHz) to baseband through a mixer; (2) Perform low sampling rate (F s =44.1KHz) ADC conversion.

3. The method for dividing the frequency band of a bionic filter according to claim 1, wherein: In step S2, the nonlinear frequency band center frequency division method is as follows: According to the first preset formula ERB n =ERB low +n·127ERB high -ERB low ,n=0,1,...,127, the equivalent rectangular bandwidth (ERB) interval [ERB low ,ERB high ] is evenly divided into 128 points, and the ERB rate of each channel is obtained, which is converted into the center frequency.

4. The method of performing Hilbert transform on each channel output and introducing a neural firing rate model as described in claim 1, wherein: In step S3, the Hilbert transform and the neural firing rate model introduction method are as follows: Then perform signal envelope detection, To extract the key feature information of the signal; The neural firing rate model is based on the phase-amplitude coupling (PAC) mechanism, which simulates the response of neurons to signal intensity.

5. The method for enhancing auditory attention according to claim 1, wherein: In step S4, the method is as follows: Based on the auditory attention mechanism, dynamic gain is applied to the local discharge characteristic frequency band (300MHz-1.5GHz), and according to the second preset formula in: G min =6dB,G max =20dB, c is the gain adjustment steepness parameter, c=0.1dB, d is the gain adjustment center SNR value, d=10dB; P signal (n) = ∑{t}x evn (n, t)^2*w(n);P noise (n) = medium (P signal (n-Δ:n+Δ)); Δ=5 achieves dynamic gain and suppresses irrelevant noise.

Images