Underwater propeller modulation feature extraction method based on periodic sparse envelope spectrum

Abstract

The invention discloses an underwater thruster modulation feature extraction method based on a periodic sparse envelope spectrum. The method comprises the following steps: (1) collecting signal noise of an underwater thruster as a monitoring signal and calculating a spectrum coherence function; (2) the spectral coherence is sliced in the carrier frequency direction, the correlation kurtosis of different slices is calculated, and correlation kurtosis vectors are obtained; (3) performing traversal calculation in the candidate characteristic frequency range to obtain a related kurtosis matrix; (4) when the sum of elements of a certain vector in the correlation kurtosis matrix is maximum, the frequency corresponding to the vector is the target characteristic frequency; (5) performing square normalization operation on the related kurtosis vector to obtain a periodic sparse weighting function; and (6) weighting spectral coherence by using the related kurtosis vector to calculate periodic sparse spectral coherence, and integrating absolute values to obtain a periodic sparse envelope spectrum. The method can enhance and extract the modulation signal features of the underwater propeller under complex noise interference under the condition of weakening dependence on feature frequency priori knowledge.

Description

Technical Field

[0001] The invention belongs to the field of signal processing, and in particular relates to a method for extracting modulation features of an underwater thruster based on a periodic sparse envelope spectrum. Background Art

[0002] Surface ships and submarines are widely used in various fields such as commercial transportation and military industry, and propeller noise is the main source of their noise. Collecting underwater noise and extracting its characteristic frequency can help detect the key information of underwater targets without exposing themselves.

[0003] For example, a Chinese patent document with publication number CN117591910A discloses a target classification and recognition method based on noise signal feature extraction, including: using a hydrophone to collect propeller noise signals or a noise vibration meter to collect automobile engine noise and vibration signals, and collecting real noise from different ships or different engine failures; performing low-frequency extraction on the collected real noise data to remove high-frequency noise; performing power spectrum analysis on the real noise data based on short-time Fourier transform STFT to extract amplitude information of harmonic frequencies; selecting corresponding harmonic frequency combinations according to propellers with different numbers of blades, calculating the amplitudes of each harmonic, and performing normalization processing to obtain normalized harmonic amplitudes.

[0004] The characteristic frequency extraction methods of underwater noise mainly include spectrum analysis, time-frequency analysis and demodulation analysis. Spectrum analysis mainly converts time domain signals to frequency domain through Fourier transform and its derived power spectrum density, so as to analyze the frequency, amplitude and phase information of key information in the signal. However, it is difficult to analyze random signals, and random signals are more common in practical applications, so the spectrum analysis method may be limited in some occasions.

[0005] Time-frequency analysis analyzes signals from both the time domain and the frequency domain, such as short-time Fourier transform and wavelet transform, and can obtain richer time-frequency texture and its change information than spectrum analysis. However, this method depends on parameter settings, and for non-stationary signals, linear time-frequency analysis methods such as short-time Fourier transform may not reveal their nonlinear characteristics.

[0006] The mainstream methods of demodulation analysis include narrowband envelope demodulation and cyclostationary analysis, which can extract the characteristic frequency of underwater thrusters from modulated signals and are still effective for non-stationary signals such as cyclostationary signals. The narrowband envelope demodulation method mainly divides the original signal into multiple sub-band signals, and calculates the characteristic index that can characterize the characteristic frequency for each sub-band signal, thereby selecting the best demodulation frequency band to enhance the frequency characteristics; after the selected demodulation frequency band is Hilbert transformed to obtain the envelope of the sub-band signal, Fourier transform is performed to obtain the envelope spectrum, and finally the target characteristic frequency is identified. However, this method relies on the selection of characterization indicators, and the selection process is susceptible to noise interference.

[0007] Cyclostationary analysis is based on the cyclostationary theory. After performing a quadratic Fourier transform on the autocorrelation function of the signal, a two-dimensional spectral correlation matrix can be obtained, thereby separating the cyclostationary noise signal of the underwater thruster in two dimensions: carrier frequency and cyclic frequency. However, other mechanical or electrical equipment will produce cyclostationary noise, which cannot be effectively suppressed by traditional cyclostationary analysis methods, which limits the application of this method in certain scenarios. Summary of the invention

[0008] The present invention provides a method for extracting underwater thruster modulation features based on periodic sparse envelope spectrum, which can enhance and extract the modulation signal features of underwater thrusters under complex noise interference while weakening the dependence on prior knowledge of characteristic frequencies, and can be applied in fields such as underwater target recognition.

[0009] A method for extracting modulation features of an underwater thruster based on a periodic sparse envelope spectrum comprises the following steps:

[0010] (1) Collecting noise data of the underwater propeller as a monitoring signal and calculating the spectral coherence function of the monitoring signal;

[0011] (2) Determine the candidate characteristic frequency range based on prior information; slice the spectral coherence function along the carrier frequency direction, take a certain frequency in the candidate characteristic frequency distribution range, calculate the relevant kurtosis of each slice, and obtain the relevant kurtosis vector;

[0012] (3) Traverse the given characteristic frequency distribution range and repeat the above calculation of correlation kurtosis to obtain the correlation kurtosis matrix;

[0013] (4) When the sum of the correlation kurtosis of a certain correlation kurtosis vector in the correlation kurtosis matrix reaches the maximum, the corresponding characteristic frequency is taken as the characteristic frequency of the target thruster propeller;

[0014] (5) taking the relevant kurtosis vector corresponding to the above characteristic frequency and calculating its square, and normalizing the result to obtain the periodic sparse weighting function;

[0015] (6) Performing periodic sparse weighted processing on the spectral coherence function to obtain periodic sparse spectral coherence, and further performing an absolute value integration operation along the carrier frequency direction to obtain a periodic sparse envelope spectrum.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] 1. The present invention proposes a method for constructing a correlation kurtosis matrix based on spectral coherence along carrier frequency slices, which can be used as a characterization indicator of periodic sparsity at different spectral frequencies, and can adaptively select the optimal characteristic frequency from the candidate characteristic frequency range.

[0018] 2. The present invention proposes a periodic sparse weighted function, which performs weighted processing according to periodic sparsity, taking a higher value where the periodic sparsity is stronger, and taking a lower value where the periodic sparsity is weaker.

[0019] 3. The periodic sparse envelope spectrum proposed in the present invention uses periodic sparse weighting of spectral coherence functions as the main means to perform signal demodulation. While weakening the need for prior knowledge of characteristic frequencies, it can also effectively enhance and extract features of the modulated signal and eliminate the influence of complex background noise signals. It can be used in fields such as underwater target recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of a method for extracting modulation features of an underwater thruster based on periodic sparse envelope spectrum of the present invention;

[0021] Figure 2 It is a time domain diagram of a simulated signal containing Gaussian noise in an embodiment of the present invention;

[0022] Figure 3 is the enhanced envelope spectrum result of the simulation signal in the embodiment of the present invention;

[0023] Figure 4 is the kurtosis spectrum demodulation result of the simulation signal in the embodiment of the present invention;

[0024] Figure 5 is the periodic sparse envelope spectrum result of the simulation signal in the embodiment of the present invention;

[0025] Figure 6 The enhanced envelope spectrum result of a propeller noise signal of a merchant ship in an embodiment of the present invention is shown in FIG.

[0026] Figure 7 is the kurtosis spectrum demodulation result of a propeller noise signal of a merchant ship in an embodiment of the present invention;

[0027] Figure 8 This is the periodic sparse envelope spectrum result of a merchant ship propeller noise signal in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be pointed out that the embodiments described below are intended to facilitate the understanding of the present invention and do not have any limiting effect on the present invention.

[0029] like Figure 1 As shown, a method for extracting modulation features of an underwater thruster based on a periodic sparse envelope spectrum comprises the following steps:

[0030] S01, collecting noise data of the underwater propeller as a monitoring signal, and calculating the spectral coherence function of the monitoring signal.

[0031] (1-1) The sampling frequency is F S , in Hz, collects the noise data of the underwater propeller as the monitoring signal x(t n ), t n Refers to the sampling frequency F S The time at which the noise data is obtained, where t n =n / F s , the monitoring signal sampling time is T, in seconds;

[0032] (1-2) Calculate the monitoring signal x(t n )'s short-time Fourier transform X STFT (i,f k ):

[0033]

[0034] Where N w is the window width, R is the moving step length, w[n] is the window function, x[n] is x(t n ), f k is the discrete frequency, f k =kΔf,k=0,...,N w-1 , frequency resolution Δf=F s / N w .

[0035] (1-3) For the monitoring signal x(t n )'s short-time Fourier transform X STFT (i,f k ) for phase correction:

[0036]

[0037] Where, X w (i,f k ) is the signal x(t n ) in iR / F s time, with f kThe complex envelope with the center and bandwidth Δf, |X w (i,f k )| 2 Represents the energy flow within a frequency band.

[0038] (1-4) Calculate the monitoring signal x(t n )’s average cycle spectrum is related to:

[0039]

[0040] In the formula, K = (LN w +R) / R, is the total number of windows moving with a step size of R in a signal of length L, and the length of these windows is N w , α is the cycle frequency, f is the carrier frequency, where

[0041] (1-5) Calculate the monitoring signal x(t n )'s spectral correlation function S x (α,f), the calculation formula is as follows:

[0042]

[0043] (1-6) Calculate the monitoring signal x(t n )'s spectral coherence function γ x (α,f), the calculation formula is as follows:

[0044]

[0045] S02, determine the candidate feature frequency range [f L ,f U ], and thus the cycle frequency period range [T L ,T U ]; Slice the spectral coherence function along the carrier frequency direction, take a certain frequency in the candidate characteristic frequency distribution range, calculate the relevant kurtosis of each slice, and obtain the relevant kurtosis vector.

[0046] (2-1) Spectral correlation function γ x The discrete form of (α,f) is γ x (α m ,f n ). Its cyclic frequency includes α m (m=1,2,...,M) has a total of M values, and the spectrum frequency includes f n (n=1,2,...,N) There are N values ​​in total.

[0047] (2-2) Slice the spectral coherence function along the carrier frequency direction to obtain the spectral coherence function at a specific spectral frequency f nSlice at γ x (α m ,f n );

[0048] (2-3) Assume that the candidate feature frequency distribution range [f L ,f U ] includes a frequency f i , then the corresponding cycle frequency period T i =f i / Δα, where Δα is the resolution of the cyclic frequency α. The frequency f is calculated from this n The associated kurtosis of the corresponding slices is as follows:

[0049]

[0050] In the formula, γ x (α m-kT ,f n ) represents γ x (α m ,f n ) is shifted by T along the cyclic frequency α, where K is the displacement order.

[0051] (2-4) Calculate the relevant kurtosis of each slice and obtain the relevant kurtosis vector:

[0052] CK i ={CK i,n |n=1,2,3,...,N}

[0053] S03, traverse the given characteristic frequency distribution range, repeat the above calculation, and obtain the relevant kurtosis matrix.

[0054] Traverse T i ∈[T L ,T U ], calculate its related kurtosis vector CK according to step S2 i Finally, the relevant kurtosis matrix CK = {CK i |i=1,2,3,...,I}, where I is the total number of value points within the cycle frequency period, that is, the number of traversals.

[0055] S04, when the sum of the correlation kurtosis of a certain correlation kurtosis vector in the correlation kurtosis matrix reaches a maximum, the corresponding characteristic frequency is taken as the characteristic frequency of the target thruster propeller.

[0056] For the relevant kurtosis matrix CK, calculate each relevant kurtosis vector CK i The correlation kurtosis and Find the value of i that maximizes the above sum. The calculation formula is as follows:

[0057] R=[TL ,T U ]

[0058]

[0059] The target thruster characteristic frequency f is obtained from this i .

[0060] S05, taking the relevant kurtosis vector corresponding to the above characteristic frequency and calculating its square, and normalizing the result to obtain a periodic sparse weighted function.

[0061] (5-1) According to the correlation kurtosis vector with the maximum correlation kurtosis sum in step (4), calculate the weighting function:

[0062]

[0063] (5-2) Normalize the weighting function w(f) to obtain the periodic sparse weighting function:

[0064]

[0065] S06, performing periodic sparse weighted processing on the spectral coherence function to obtain periodic sparse spectral coherence, and further performing an absolute value integration operation along the carrier frequency direction to obtain a periodic sparse envelope spectrum.

[0066] (6-1) Spectral coherence function γ x (α, f) is weighted to obtain the periodic sparse spectral coherence The calculation formula is as follows:

[0067]

[0068] (6-2) The periodic sparse spectrum coherence γ x (α, f) calculates the absolute value integral along the carrier frequency α direction to obtain the periodic sparse envelope spectrum. The calculation formula is as follows:

[0069]

[0070] In the formula, F S is the sampling frequency of the monitoring signal.

[0071] In order to verify the effectiveness of the present invention, the above method is used to analyze the simulated signal containing Gaussian noise, as follows:

[0072]

[0073] Where n(t) represents Gaussian white noise. The parameter settings of the simulation signal are shown in Table 1 below:

[0074] Table 1

[0075]

[0076] Figure 2 It is the time domain signal of the simulation signal, which includes the modulation signal and the simulation signal. It can be seen from the figure that the modulation signal is completely submerged by the background noise. Figure 3 This is the enhanced envelope spectrum demodulation result of the simulated signal, and the basic modulation frequency f can be identified. SF and its harmonics, but the second to fifth harmonics are almost completely drowned out by the background noise (see Figure 3 The enhanced envelope spectrum demodulation result is not ideal for extracting signal features. Figure 4 This is the demodulation result of the kurtosis spectrum narrowband envelope of the simulated signal. The basic modulation frequency f can also be observed. SF and its harmonics (see Figure 4 The red circle in the middle), and it is less interfered by the background noise, indicating that its demodulation effect is better than the enhanced envelope spectrum demodulation result, but there are still some interfering spectral lines. Figure 5 is the periodic sparse envelope spectrum of the simulated signal, and we know that the basic modulation frequency f SF and its harmonics are significantly enhanced compared with the previous two methods (see Figure 5 The above results show that under extremely low signal-to-noise ratio conditions, the periodic sparse envelope spectrum proposed in the present invention can accurately extract the simulated modulation frequency characteristics.

[0077] The above three methods are used to process the noise signal of a five-blade propeller of a merchant ship. The results are as follows: Figure 6-8 As shown. Figure 6 In the demodulation results of the enhanced envelope spectrum, in addition to the double fundamental frequency, the first to fifth frequencies can be observed (see Figure 6 The fifth frequency is the most significant, while the other frequencies are basically submerged in the background noise; Figure 7 The demodulation results of the narrowband envelope of the kurtosis spectrum are similar to those of the enhanced envelope spectrum. In addition to the double frequency, the first to fifth frequencies can also be observed (see Figure 7 red circle); finally Figure 8 In the periodic sparse spectrum, the first to fifth harmonics can be clearly observed, and the target characteristic frequency, i.e. the fifth harmonic, is significantly enhanced, indicating that the periodic sparse envelope spectrum proposed in the present invention can accurately extract the characteristic frequency of the merchant ship propeller noise signal.

[0078] The embodiments described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for extracting modulation features of underwater thrusters based on periodic sparse envelope spectrum, characterized in that: The following steps are involved: (1) Collecting noise data of the underwater propeller as a monitoring signal and calculating the spectral coherence function of the monitoring signal; (2) Determine the candidate feature frequency range based on prior information; Slice the spectral coherence function along the carrier frequency direction, take a certain frequency in the candidate characteristic frequency distribution range, calculate the relevant kurtosis of each slice, and obtain the relevant kurtosis vector; (3) Traverse the given characteristic frequency distribution range and repeat the above calculation of correlation kurtosis to obtain the correlation kurtosis matrix; (4) When the sum of the correlation kurtosis of a certain correlation kurtosis vector in the correlation kurtosis matrix reaches the maximum, the corresponding characteristic frequency is taken as the characteristic frequency of the target thruster; (5) taking the relevant kurtosis vector corresponding to the above characteristic frequency and calculating its square, and normalizing the result to obtain the periodic sparse weighting function; (6) Performing periodic sparse weighted processing on the spectral coherence function to obtain periodic sparse spectral coherence, and further performing an absolute value integration operation along the carrier frequency direction to obtain a periodic sparse envelope spectrum.

2. The underwater propeller modulation feature extraction method based on periodic sparse envelope spectrum according to claim 1 is characterized in that: In step (1), the specific process of calculating the spectral coherence function of the underwater thruster noise signal is as follows: (1-1) Set the sampling frequency to F S , collect the noise data of the underwater propeller as the monitoring signal x(t n ), t n Refers to the sampling frequency F S The time at which the noise data is obtained, where t n =n / F s , the monitoring signal sampling duration is T; (1-2) Calculate the monitoring signal x(t n )'s short-time Fourier transform X STFT (i,f k ): Where N w is the window width, R is the moving step length, w[n] is the window function, x[n] is x(t n ), f k is the discrete frequency, f k =kΔf, k=0, ..., N w-1 , frequency resolution Δf=F s / N w ; (1-3) For the monitoring signal x(t n )'s short-time Fourier transform X STFT (i,f k ) for phase correction: Where, X w (i,f k ) is the signal x(t n ) in iR / F s time, with f k The complex envelope with the center and bandwidth Δf, |X w (i,f k )| 2 Indicates the energy flow within a frequency band; (1-4) Calculate the monitoring signal x(t n )’s average cycle spectrum is related to: In the formula, K = (LN w +R) / R, is the total number of windows moving with a step size of R in a signal of length L, and the length of these windows is N w , α is the cycle frequency, f is the carrier frequency, where (1-5) Calculate the monitoring signal x(t n )'s spectral correlation function S x (α, f), the calculation formula is as follows: (1-6) Calculate the monitoring signal x(t n )'s spectral coherence function γ x (α, f), the calculation formula is as follows:

3. The underwater propeller modulation feature extraction method based on periodic sparse envelope spectrum according to claim 1 is characterized in that: In step (2), the candidate feature frequency range [f L , f U ], and thus the cycle frequency period range [T L , T U ].

4. The underwater propeller modulation feature extraction method based on periodic sparse envelope spectrum according to claim 3 is characterized in that: In step (2), the specific process of calculating the relevant kurtosis vector is: (2-1) Spectral correlation function γ x The discrete form of (α, f) is γ x (α m , f n ); m=1, 2,..., M, n=1, 2,..., N; (2-2) Slice the spectral coherence function along the carrier frequency direction to obtain the spectral coherence function at a specific spectral frequency f n Slice at γ x (α m , f n ); (2-3) Assume that the candidate feature frequency distribution range [f L , f U ] includes a frequency f i , then the corresponding cycle frequency period T i =f i / △α, where △α is the resolution of the cycle frequency α, from which the frequency f is calculated n The associated kurtosis of the corresponding slices is as follows: In the formula, γ x (α m-kT , f n ) represents γ x (α m , f n ) is shifted by T along the cyclic frequency α, where K is the displacement order; (2-4) Calculate the relevant kurtosis of each slice and obtain the relevant kurtosis vector: CK i ={CK i，n |n=1,2,3,...,N}.

5. The underwater propeller modulation feature extraction method based on periodic sparse envelope spectrum according to claim 4 is characterized in that: In step (3), the specific process of calculating the correlation kurtosis matrix is ​​as follows: Traverse T i ∈[T L , T U ], calculate its related kurtosis vector CK according to step (2) i Finally, the relevant kurtosis matrix CK = {CK i |i=1,2,3,...,I}, where I is the total number of value points within the cycle frequency period, that is, the number of traversals.

6. The underwater propeller modulation feature extraction method based on periodic sparse envelope spectrum according to claim 5 is characterized in that: In step (4), the specific process of calculating the characteristic frequency of the target thruster is as follows: For the relevant kurtosis matrix CK, calculate each relevant kurtosis vector CK i The relevant kurtosis and Find the value of i that maximizes the above sum. The calculation formula is as follows: R=[T L ,T U ] The target thruster characteristic frequency f is obtained from this i .

7. The method for extracting modulation features of underwater propellers based on periodic sparse envelope spectrum according to claim 6, characterized in that: In step (5), the specific process of calculating the periodic sparse weighted function is as follows: (5-1) According to the relevant kurtosis vector with the maximum relevant kurtosis sum in step (4), calculate the weighting function: (5-2) Normalize the weighting function w(f) to obtain the periodic sparse weighting function:

8. The method for extracting modulation features of underwater propellers based on periodic sparse envelope spectrum according to claim 7, characterized in that: In step (6), the specific process of calculating the periodic sparse envelope spectrum is as follows: (6-1) Spectral coherence function γ x (α, f) is weighted to obtain the periodic sparse spectrum coherence The calculation formula is as follows: (6-2) The periodic sparse spectrum coherence γ x (α, f) calculates the absolute value integral along the carrier frequency α direction to obtain the periodic sparse envelope spectrum. The calculation formula is as follows: In the formula, F S is the sampling frequency of the monitoring signal.

Images