Signal processing device, signal processing program, and signal processing method
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- OKI ELECTRIC INDUSTRY CO LTD
- Filing Date
- 2025-01-17
- Publication Date
- 2026-07-30
AI Technical Summary
【0009】 本発明によれば、入力信号から固有モード関数を抽出する際に、抽出した固有モードに関するゆらぎを抑制することができる。
Smart Images
Figure 2026123392000001_ABST
Abstract
Description
[Technical Field]
[0001] This invention relates to a signal processing device, a signal processing program, and a signal processing method, and can be applied, for example, to a system that decomposes an input signal to calculate an eigenmode function. [Background technology]
[0002] Conventionally, vibration analysis for anomaly diagnosis of machinery including rotating mechanisms (hereinafter referred to as "rotating machinery") has employed methods using signal decomposition. Furthermore, conventionally, signal decomposition has been performed on the vibration signal of rotating machinery, and the intrinsic mode function (hereinafter referred to as "IMF") has been extracted to calculate the instantaneous frequency. However, in rotating machinery in actual factory operations, noise is unavoidable in the vibration signal. Patent Document 1 describes a technology that addresses such situations. The method described in Patent Document 1 determines a filtering boundary for the signal time spectrum where energy is more concentrated, and calculates the IMF by performing an empirical wavelet transform based on the filtering boundary. This makes it possible to clearly separate the frequency bands of the extracted peaks and noise using the method described in Patent Document 1. [Prior art documents] [Patent Documents]
[0003] [Patent Document 1] Japanese Patent Publication No. 2022-87010 [Overview of the project] [Problems that the invention aims to solve]
[0004] However, if the noise and the component to be extracted are in the same frequency band, the method described in Patent Document 1 will result in noise being introduced into the IMF. As a result, the method described in Patent Document 1 will cause problems such as fluctuations in the instantaneous frequency estimation results due to the influence of noise.
[0005] In light of the above problems, there is a need for a signal processing device, a signal processing program, and a signal processing method that can suppress fluctuations related to the extracted eigenmode functions when extracting them from an input signal. [Means for solving the problem]
[0006] The first aspect of the present invention is a signal processing device comprising: noise estimation means for estimating noise components included in the input power spectrum obtained as a result of frequency analysis of an input signal and acquiring estimated noise; noise removal means for performing noise removal processing to remove the estimated noise from the input power spectrum and acquiring a noise-removed power spectrum; and signal decomposition means for performing signal decomposition on a signal waveform based on the noise-removed power spectrum and extracting an eigenmode function.
[0007] The second signal processing program of the present invention is characterized in that a computer functions as a noise estimation means for estimating noise components included in the input power spectrum obtained as a result of frequency analysis of the input signal and acquiring estimated noise; a noise removal means for performing a noise removal process to remove the estimated noise from the input power spectrum and acquiring a noise-removed power spectrum; and a signal decomposition means for performing signal decomposition on the signal waveform based on the noise-removed power spectrum and extracting eigenmode functions.
[0008] The third aspect of the present invention relates to a signal processing method performed by a signal processing device, wherein the signal processing device includes noise estimation means, noise removal means, and signal decomposition means, wherein the noise estimation means estimates noise components included in the input power spectrum obtained as a result of frequency analysis of the input signal and obtains estimated noise, the noise removal means performs noise removal processing to remove the estimated noise from the input power spectrum and obtains a denoised power spectrum, and the signal decomposition means performs signal decomposition on the signal waveform based on the denoised power spectrum and extracts eigenmode functions. [Effects of the Invention]
[0009] According to the present invention, when extracting the intrinsic mode function from an input signal, it is possible to suppress the fluctuation regarding the extracted intrinsic mode.
Brief Description of the Drawings
[0010] [Figure 1] It is a block diagram showing an example of the functional configuration of the signal processing apparatus according to the embodiment. [Figure 2] It is a block diagram showing an example of the hardware configuration of the signal processing apparatus according to the embodiment.
Modes for Carrying Out the Invention
[0011] (A) Main Embodiment Hereinafter, an embodiment of a signal processing apparatus, a signal processing program, and a signal processing method according to the present invention will be described in detail while referring to the drawings.
[0012] (A-1) Configuration of the Embodiment FIG. 1 is a block diagram showing the functional configuration of the signal processing apparatus 1 of this embodiment.
[0013] The signal processing apparatus 1 includes an IMF extraction unit 10 and an instantaneous frequency / power determination unit 20.
[0014] The IMF extraction unit 10 performs a process of extracting N (N is an integer of 1 or more) IMFs (intrinsic mode functions) for the input signal.
[0015] The instantaneous frequency / power determination unit 20 calculates the instantaneous frequency and the instantaneous power of the IMFs extracted by the IMF extraction unit 10.
[0016] Next, the internal configuration of the IMF extraction unit 10 will be described.
[0017] The IMF extraction unit 10 includes a spectrum calculation unit 11, a noise removal unit 12, a waveform calculation unit 13, a signal decomposition unit 14, a power calculation unit 15, a gain calculation unit 16, and a noise estimation unit 17. Details of each element of the IMF extraction unit 10 will be described later.
[0018] The signal processing device 1 may be entirely configured hardware-wise (e.g., configured by a dedicated chip or the like), or partially or entirely configured software-wise.
[0019] Next, the hardware configuration of the signal processing device 1 will be described.
[0020] FIG. 2 is a block diagram showing an example of the hardware configuration of the signal processing device 1.
[0021] FIG. 2 shows an example of the hardware configuration when the signal processing device 1 is configured using software (a computer).
[0022] The signal processing device 1 shown in FIG. 2 has a computer 300 in which a program (including the signal processing program of the embodiment) is installed as a hardware component. Also, the computer 300 may be a computer dedicated to the signal processing program or may be configured to be shared with programs of other functions. Even when the signal processing device 1 is configured by the computer 300, the functional configuration of the signal processing device 1 (signal processing program) can be shown as in FIG. 2. In this embodiment, the signal processing device 1 (signal processing program) is described as having an instantaneous frequency / power determination unit 20 included therein, but the instantaneous frequency / power determination unit 20 may be configured as a separate device (separate hardware).
[0023] The computer 300 shown in Figure 2 includes a processor 301, a primary storage unit 302, and a secondary storage unit 303. The primary storage unit 302 is a storage means that functions as a working memory for the processor 301, and can be a high-speed memory such as DRAM (Dynamic Random Access Memory). The secondary storage unit 303 is a storage means that records various data such as the OS (Operating System) and program data (including data for the signal processing program according to the embodiment), and can be a non-volatile memory such as FLASH® memory, HDD (Hard Disk Drive), or SSD (Solid State Drive). In the computer 300 of this embodiment, when the processor 301 starts up, it reads the OS and program (including the signal processing program according to the embodiment) recorded in the secondary storage unit 303, loads them onto the primary storage unit 302, and executes them. Note that the specific configuration of the computer 300 is not limited to the configuration in Figure 2, and various configurations can be applied. For example, if the primary storage unit 302 is a non-volatile memory, the configuration may exclude the secondary storage unit 303.
[0024] (A-2) Operation of the embodiment Next, the operation of the signal processing device 1 of this embodiment, which has the configuration described above, will be explained.
[0025] The spectrum calculation unit 11 performs frequency analysis on the input signal X(t) to calculate the frequency spectrum S(t,f), which is then supplied to the power calculation unit 15 and the noise reduction unit 12.
[0026] The input signal X(t) can be an acoustic signal obtained by capturing sound emitted from the object to be analyzed (e.g., rotating machinery not shown, or living organisms (e.g., humans, animals, etc.)) using a microphone not shown. The objects to be analyzed above may include various objects that emit sounds that can be analyzed using IMF.
[0027] The spectrum calculation unit 11 can apply methods such as the Fast Fourier Transform (FFT), wavelet transform, or filter bank, but the FFT is preferred. In this case, the spectrum calculation unit 11 will perform frequency analysis of the input signal X(t) using the FFT.
[0028] The power calculation unit 15 takes the square of the absolute value of the frequency spectrum S(t,f) as input power spectrum |S(t,f)| 2 This value is calculated and provided to the noise estimation unit 17 and the gain calculation unit 16.
[0029] The noise estimation unit 17 is |S(t,f)| 2 The noise is estimated by smoothing the signal and then fed to the gain calculation unit 16. For example, the noise estimation unit 17 calculates the input power spectrum |S(t,f)| 2 The frequency-smoothed provisional noise |~S(t,f)| is obtained by applying a median filter with a window size of (2k+1) to the frequency direction of each time frame, as shown in equation (1) below. 2 It is also possible to calculate this in this manner. Furthermore, in the text portion of this specification, for the sake of notation, the symbol with a tilde (~) above the S in each formula will also be represented as "~S".
[0030] Next, the noise estimation unit 17 calculates the frequency smoothing provisional noise |~S(t,f)| 2 In contrast, the provisional estimated noise |^S(t,f)| is obtained by smoothing the signal by applying a median filter with a window size of (2h+1) in the time direction of each frequency, as shown in equation (2) below. 2 This calculates the following. Note that, for the sake of notation, in the text portion of this specification, the symbol with a hat (^) above the S in each formula will also be represented as "^S".
[0031] Then, the noise estimation unit 17 calculates the provisional estimated noise |^S(t,f)| 2 W|^S(t,f)| obtained by multiplying by the scaling factor W. 2It is calculated as the estimated noise. Here, it is desirable to select the window sizes (2k + 1) and (2h + 1) of the median filter so that the main peaks are removed from the estimated noise. For example, it is preferable that (2k + 1) is a frequency 31 times the frequency resolution of the FFT, and (2h + 1) is a time 11 times the time resolution of the FFT. Through the above processing, the estimated noise W|^S(t,f)| calculated by the noise estimation unit 17 2 will be composed of noise components excluding the main peaks.
Number
[0032] The gain calculation unit 16 subtracts the estimated noise W|^S(t,f)| 2 from the input power spectrum |S(t,f)| 2 and divides it by the input power spectrum |S(t,f)| 2 and gives it to the noise removal unit 12. Then, the gain calculation unit 16 calculates the gain H determined using the following equation (3). The gain H replaces negative values with 0 for all values.
Number
[0033] The noise removal unit 12 multiplies the gain H by the frequency spectrum S(t,f), calculates the spectrum Y(t,f) with noise removed using equation (4), and gives it to the waveform calculation unit 13. Y(t,f)=H·S(t,f)…(4)
[0034] The waveform calculation unit 13 calculates the signal Y(t) with noise removed by performing IFFT (Inverse FFT) on the spectrum Y(t,f) with noise removed, and gives it to the signal decomposition unit 14.
[0035] The signal decomposition unit 14 calculates (extracts) the IMF using the noise-removed signal Y(t) and provides it to the instantaneous frequency / power determination unit 20. In the signal decomposition unit 14, the signal decomposition method (method for calculating the IMF) is the decomposed signal IMF for the noise-removed signal Y(t). n (t) is the result, and any method that outputs N signals (eigenmode functions) is acceptable. For example, the signal decomposition unit 14 may obtain the IMF by applying iterative filtering, empirical mode decomposition, singular value decomposition, or derived methods thereof.
[0036] The instantaneous frequency / power determination unit 20 processes the decomposed signal IMF. n For the instantaneous time of N signals decomposed using (t), the frequency f n (t) and Power A n Calculate (t).
[0037] In the instantaneous frequency / power determination unit 20, any method that instantaneously determines the frequency and power with respect to the time waveform is acceptable for determining the instantaneous frequency and power. For example, the instantaneous frequency / power determination unit 20 may use an IMFogram or a Hilbert fan transform. The instantaneous frequency / power determination unit 20 generates a time-frequency representation in a format that matches the method used to determine the instantaneous frequency and instantaneous power.
[0038] (A-3) Effects of the Embodiment This embodiment can achieve the following effects.
[0039] In this embodiment of the signal processing device 1, noise estimation processing (acquisition of estimated noise) is performed on the power spectrum of the original input signal X(t) to remove the noise and return it to a signal waveform. Then, in this embodiment of the signal processing device 1, signal decomposition is performed on the obtained signal to extract the IMF. In this embodiment of the signal processing device 1, by removing noise before extracting the IMF, it is possible to obtain an IMF that does not contain noise. In other words, in this embodiment of the signal processing device 1, a time-frequency representation using instantaneous frequency and power that is not fluctuated by noise is possible. Specifically, for example, when determining the instantaneous frequency using IMFogram or Hilbert transform, fluctuations in the estimation result can be prevented and the accuracy of the instantaneous frequency can be improved.
[0040] (B) Other embodiments The present invention is not limited to the embodiments described above, and modified embodiments such as those exemplified below can also be cited.
[0041] (B-1) In the above embodiment, the noise estimation method (method for calculating estimated noise) in the noise estimation unit 17 is not limited to the above example and various processes can be applied. For example, the noise estimation unit 17 may apply a noise estimation method described in any of the following references 1 to 3.
[0042] The noise estimation method described in Reference 1 is based on the discovery that the time-direction peak of the input power represents the presence of the target sound, while the valleys can be used to estimate the smoothed noise power. Specifically, in Reference 1, the minimum value of the input power from the present to a predetermined time in the past is taken as the first estimated noise power. However, the first estimated noise power has a bias and tends to be smaller than the true noise power. Therefore, in the method described in Reference 1, this bias is estimated from the expected value of the first estimated noise power. Then, in the method described in Reference 1, the first estimated noise power is corrected using the obtained bias estimate to obtain the second estimated noise power (final estimate).
[0043] The noise estimation method described in Reference 2 involves multiplying the input power by an appropriate weighting coefficient, storing the resulting weighted input power for a predetermined time, and using the average of the stored weighted input powers as the estimated noise power. In the noise estimation method described in Reference 2, the appropriate weighting coefficient is calculated based on the aftermath SNR (Signal-to-Noise Ratio), which is obtained by dividing the current input power by the previously estimated noise power. Specifically, in the noise estimation method described in Reference 2, the weighting coefficient is set to 1.0 when the aftermath SNR is less than or equal to a predetermined value G1, the weighting coefficient is set to be inversely proportional to the aftermath SNR when the aftermath SNR is greater than or equal to G1, and the weighting coefficient is set to 0.0 when the aftermath SNR is greater than or equal to a predetermined value G2. Furthermore, in the noise estimation method described in Reference 2, the weighted input power is not stored when the weighting coefficient is 0.0.
[0044] The noise estimation method described in Reference 3 assumes that the complex spectral distributions of both the target speech and the noise follow a complex normal distribution with a mean of zero, and uses the maximum likelihood estimate of the variance of the noise's complex spectrum as the estimated noise power. Based on this assumption, the complex spectral distribution of the input speech follows a complex normal distribution with a mean of zero, where the variance is the sum of the variances of the speech's complex spectrum and the noise's complex spectrum. Therefore, the noise estimation method described in Reference 3 introduces a hidden variable regarding whether the current input is degraded speech or noise, and applies an online EM (Expectation Maximization) algorithm with a forgetting coefficient to calculate the maximum likelihood estimate of the noise's complex spectrum.
[0045] [Reference 1]: R. Martin, “Spectral Subtraction Based on Minimum Statistics”, in Proceedings of 7th European Signal Processing Conference, 1994, pp. 1182-1185”
[0046] [Reference 2]: Japanese Patent Publication No. 2002-204175
[0047] [Reference 3]: M. Souden, M. Delcroix, K. Kinsoshita, T. Yoshioka, and T. Nakatani, “Noise Power Spectral Density Tracking: A Maximum Likelihood Perspective”, IEEE Signal Processing Letters, Vol. 19, No. 8, 2012, pp. 495-498 [Explanation of Symbols]
[0048] 1...Signal processing unit, 10...IMF extraction unit, 11...Spectrum calculation unit, 12...Noise reduction unit, 13...Waveform calculation unit, 14...Signal decomposition unit, 15...Power calculation unit, 16...Gain calculation unit, 17...Noise estimation unit, 20...Power determination unit
Claims
1. A noise estimation means for estimating noise by estimating the noise component contained in the input power spectrum obtained as a result of frequency analysis of the input signal, A noise reduction means that performs a noise reduction process to remove the estimated noise from the input power spectrum and obtains a noise-reduced power spectrum, A signal decomposition means for extracting eigenmode functions by performing signal decomposition on the signal waveform based on the denoised power spectrum, and A signal processing device characterized by having
2. The signal processing apparatus according to claim 1, characterized in that the noise estimation means obtains a function as the estimated noise by applying a median filter in the frequency and time directions to the input power spectrum and scaling it.
3. The system further comprises a gain calculation means for calculating a noise reduction gain to remove the estimated noise from the input power spectrum, The noise reduction means removes the estimated noise from the input power spectrum using the noise reduction gain. The signal processing device according to claim 1.
4. The signal processing apparatus according to claim 3, characterized in that the gain calculation means calculates the noise reduction gain by subtracting the input power spectrum by the estimated noise, dividing by the input power spectrum, and replacing all negative values with zero.
5. The signal processing apparatus according to claim 4, characterized in that the noise reduction means obtains the noise-reduced power spectrum by multiplying the input power spectrum by the noise reduction gain.
6. Computers, A noise estimation means for estimating noise by estimating the noise component contained in the input power spectrum obtained as a result of frequency analysis of the input signal, A noise reduction means that performs a noise reduction process to remove the estimated noise from the input power spectrum and obtains a noise-reduced power spectrum, A signal decomposition means for extracting eigenmode functions by performing signal decomposition on the signal waveform based on the denoised power spectrum, and A signal processing program characterized by its ability to function in this way.
7. In a signal processing method performed by a signal processing device, The signal processing device includes noise estimation means, noise reduction means, and signal decomposition means. The noise estimation means estimates the noise component included in the input power spectrum obtained as a result of frequency analysis of the input signal and obtains estimated noise. The noise reduction means performs a noise reduction process to remove the estimated noise from the input power spectrum and obtains a noise-reduced power spectrum. The signal decomposition means performs signal decomposition on the signal waveform based on the denoised power spectrum to extract the eigenmode function. A signal processing method characterized by the following: