Ultrasonic data compression sampling method of random information degree of freedom

Abstract

The invention discloses a random information freedom degree ultrasonic data compression sampling method, and relates to the field of data compression acquisition during ultrasonic detection of complex components. According to the method, information of component interface echoes and defect echoes can be reserved, the degree of freedom of detection echo signal information with randomness is obtained according to an effective peak value in a detection signal, and an effective peak value Gaussian pulse flow is obtained by utilizing a designed pulse flow extraction method; inputting the pulse flow into a designed adaptive bandwidth sampling core to modulate the pulse flow, performing sampling time compensation on the signal in combination with a time domain support interval of the sampling core, and obtaining sparse sampling data at a high innovation rate; a singular value decomposition method is adopted to intercept singular values conforming to the number of information freedom degrees and corresponding sparse data Fourier coefficients, feature parameters of the random information freedom degree ultrasonic detection signals are obtained through calculation, and signal recovery can be achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of ultrasonic nondestructive testing, and in particular relates to data compression and collection during ultrasonic testing of complex components. Background Art

[0002] Ultrasonic waves have the ability to penetrate a wide variety of materials, offering significant advantages in material integrity testing and assessment. When using ultrasound to inspect regular components, the test echoes typically consist of only echoes from the upper and lower surfaces and a defect. The degrees of freedom of information (DOF) of these echoes are relatively fixed. Using sparse sampling methods based on the DOF of the signal can significantly reduce the amount of collected data, enabling compressed sampling. Sparse sampling methods for this application are relatively mature. However, sparse sampling methods based on DOF are not universal sampling methods like Nyquist sampling, and they impose specific requirements on the sampled signals. When inspecting complex components such as high-speed rail tracks, the echo signal complexity increases significantly. Echoes include echoes from multiple interfaces and defect echoes, and these echoes exhibit similar time-domain waveforms. Therefore, in many cases, real-time defect detection is impossible. Therefore, the test echoes must be stored for subsequent defect identification and assessment. This results in a very large amount of data generated during long-distance track inspections, significantly complicating both data storage and subsequent processing. Furthermore, considering that the time of occurrence of the peak of the track detection echo is random and has mutual interference and superposition, the information freedom of the signal is uncertain, which brings difficulties to the sparse sampling of the information freedom, and it is impossible to directly use the existing ultrasonic signal sparse sampling method to sample and compress the data. For this reason, the present invention provides an ultrasonic signal data compression sampling method with random information freedom, which is used to solve the problems faced under the above-mentioned detection conditions.

[0003] Existing data compression technologies, such as wavelet transforms and compressed sensing, can reduce data volumes. However, wavelet transforms require the acquisition and recording of all raw data, making them a post-compression method that cannot meet the requirements of applications like track inspection. While compressed sensing falls under the category of front-end data compression, it requires a sparse transformation matrix of the same size as the acquired data volume and requires pre-storage, which effectively does not reduce storage pressure. Furthermore, as the compression resolution increases, the reconstructed signal accuracy decreases significantly, making it unsuitable for inspection requirements.

[0004] Vertteli first proposed the finite innovation rate sampling theory in 2002. Its characteristic is that the signal innovation rate is used as the minimum distortion-free sampling frequency. It is a sparse sampling method for non-bandlimited signals and is only applicable to non-bandlimited signals such as Dirac streams, derivatives of Dirac streams, nonuniform splines, and piecewise polynomials. In 2011, Eldar et al. first proposed a finite innovation rate sampling method for medical ultrasound signals, integrating finite innovation rate sampling theory with ultrasound signals. Jiangsu University has conducted research on sparse sampling of ultrasonic echo signals in the field of industrial ultrasonic testing, but this research primarily focuses on situations where the signal's degrees of freedom are relatively fixed.

[0005] Given the high degree of freedom and randomness of information in ultrasonic echo signals, a signal time compensation method and a pulse stream extraction method were designed to ensure that all pulses do not exceed the time domain support interval of the sampling kernel, thereby ensuring their complete preservation. An adaptive bandwidth SoS sampling kernel was designed based on the signal's time and frequency domain characteristics and information freedom. The parameters were adaptively and dynamically adjusted according to the bandwidth of the pulse stream signal to achieve optimal matching and ensure the signal's effective information richness. Sampling at a high innovation rate and appropriately increasing the number of sampling points were proposed to improve the accuracy of parameter estimation and signal reconstruction. A Fourier coefficient truncation method based on singular value decomposition was designed to improve the efficiency of parameter estimation. By combining known information with calculated parameter results, the effective information of the ultrasonic echo signal can be accurately recovered, significantly reducing the amount of acquired data. Summary of the Invention

[0006] The present invention proposes a data compression sampling method for ultrasonic signals based on random information freedom. According to the effective peak value in the ultrasonic detection signal, the information freedom of the detection echo signal with randomness is obtained, and the designed pulse stream extraction method is used to obtain the effective peak Gaussian pulse stream. The pulse stream is then input into the designed adaptive bandwidth sampling kernel for modulation, and sparse sampling data is obtained at a high innovation rate. The singular value decomposition method is used to intercept the singular values ​​that meet the number of information freedom degrees and the corresponding sparse data Fourier coefficients, and the characteristic parameters of the ultrasonic detection signal with random information freedom degrees are calculated, so that the ultrasonic echo signal that retains the complete interface and defect information can be restored using a data amount far less than that of the Nyquist sampling method.

[0007] In order to achieve the above-mentioned object of the invention, the technical solution provided by the present invention includes the following steps:

[0008] Step 1. According to the received ultrasonic echo signal x(t), determine the time length of the signal as T, determine the number of effective peaks L, calculate the information freedom Ω of the effective peaks, and complete the time compensation of the signal x(t). The time length after compensation is T P ;

[0009] Step 2. Perform pulse flow extraction on the ultrasonic echo signal x(t) to obtain the pulse flow signal g c (t), g c (t) is set to half-peak pulse width The pulse width factor is Gaussian pulse stream signal;

[0010] Step 3. According to the Gaussian pulse flow signal g c (t) is used to design the SoS sampling kernel with adaptive bandwidth based on the time domain and frequency domain characteristics. The sampling kernel bandwidth ω is set. w , establish ω w The relationship between the maximum value M of the sampling kernel index point m determines the number of index points N M ;

[0011] Step 4. Gaussian pulse stream signal g c (t) is input into the SoS sampling core for modulation to obtain the signal y(t);

[0012] Step 5. According to the information freedom Ω of the effective peak and the number of SoS sampling kernel index points N M Determine the number of signal sparse sampling points N s , calculate the sampling rate ρ of the high innovation rate s And the sampling interval T s The discrete sequence y[n] is obtained by sampling the signal at equal intervals at a high innovation rate, where n = 0, 1, 2, ..., N s -1;

[0013] Step 6. Use singular value decomposition (SVD) to extract the most important 2L+1 components of the Fourier coefficients Y[r] of the discrete sequence y[n] obtained by sparse sampling, where r = 0, 1, 2, ..., N s -1, solve to get the zeroing filter coefficient P[l];

[0014] Step 7. Obtain the signal parameter estimate from the zeroing filter coefficient P[l] using the spectrum estimation method t l is the delay of the Gaussian pulse, e l is the amplitude of the Gaussian pulse, combined with the Gaussian pulse function and the pulse width factor Recovering ultrasonic echo signals

[0015] Furthermore, the step 1 specifically includes the following process:

[0016] The received ultrasonic echo signal x(t) contains multiple echoes of the measured object interface, defect echoes and noise. Different from the conventional ultrasonic detection echo signal, the echo complexity is significantly increased after interference superposition, and the signal information freedom is random. The effective detection echo signal threshold X is obtained by statistics. T , the peak value in the statistical echo signal is greater than X T The number of effective peaks is L, and the information freedom of the signal is Ω = 2L. The compensation of the beginning and end of the received ultrasonic echo signal x(t) is The time length of the compensated signal is , which ensures that the initial wave and the bottom wall echo are complete and not lost in the subsequent process, avoiding the reduction of reconstruction accuracy.

[0017] Furthermore, the step 2 specifically includes the following process:

[0018] According to the center frequency f0 and bandwidth B of the ultrasonic echo signal x(t) w Determine the passband cutoff frequency of a bandpass filter and Stopband cutoff frequency f s and f p , passband ripple A p and the stopband attenuation A s , use a bandpass filter to filter x(t), retain the effective frequency components, and suppress noise interference. Then perform Hilbert transform to get g h (t), can be expressed as:

[0019]

[0020] Where τ is the integration variable.

[0021] The time domain signal g h (t) is converted into an instantaneous power signal to further improve the dynamic range of the signal and obtain a pulse flow signal g c (t) is expressed as:

[0022] g c (t)=|g h (t)| 2 (2)

[0023] The expression of Gaussian pulse function G(t) is known Where q is the pulse width factor of the Gaussian pulse. According to the prior knowledge, the pulse stream signal g c (t) is set to the half-peak pulse width of the effective peak signal. Gaussian pulse stream signal, the pulse width factor of the Gaussian pulse is correspondingly expressed as The relationship between the half-peak pulse width and the pulse width factor of the Gaussian pulse is established as follows: Gaussian pulse stream signal g c (t) can be expressed as:

[0024]

[0025] Where, e l is the amplitude of the Gaussian pulse, t l is the delay of the Gaussian pulse.

[0026] Furthermore, the step 3 specifically includes the following process:

[0027] According to the Gaussian pulse flow signal g c (t) Get the bandwidth of the signal ω g When designing the SoS sampling core, the bandwidth of the sampling core ω w With ω g To match, the frequency domain expression of the SoS sampling kernel is:

[0028]

[0029] Where ω is the angular frequency, m is the index point, c m is the weight.

[0030] The time domain support length of the SoS sampling kernel is T P , weight c m is the Hamming window, and the value of m in the frequency domain expression affects the function Pan left or right Units. The bandwidth of the superimposed S(ω) is affected by the value of m, taking m∈[-M,M], where M∈N + , let N M is the number of index points m, that is, N M =2M+1, establish the SoS sampling core bandwidth ω w The relationship with the maximum value M of the SoS sampling kernel index point m is:

[0031]

[0032] Where, This is a round-up operation.

[0033] Furthermore, the step 4 specifically includes the following process:

[0034] The Gaussian pulse stream signal g c (t) is modulated by the input sampling kernel s(t) to obtain the signal y(t), which can be expressed as:

[0035] y(t)=g c (t)*s(t) (6)

[0036] Furthermore, the step 5 specifically includes the following process:

[0037] According to the determined signal information freedom Ω, the innovation rate is obtained as:

[0038]

[0039] The number of sparse sampling points N s The minimum number of sampling points 2L+1 required by the information freedom Ω and the number of SoS sampling core index points N M Jointly determine the number of sparse sampling points N s for:

[0040]

[0041] Get the equally spaced sampling rate ρ s for:

[0042]

[0043] Since the equally spaced sampling rate is always higher than the innovation rate, the equally spaced sampling rate ρ is called s is the high innovation rate. According to the sampling rate ρ of the signal s The sparse sampling interval T can be calculated s for:

[0044]

[0045] After sampling y(t) at equal intervals, the discrete sequence y[n] is obtained, where n = 0, 1, 2, ..., N s -1.

[0046] Furthermore, the step 6 specifically includes the following process:

[0047] Perform DFT on the discrete sequence y[n] to obtain the Fourier coefficient Y[r], which is in the form of:

[0048]

[0049] Where r = 0, 1, 2, ..., N s -1.

[0050] It is known that the number of discrete sequences and the number of Fourier coefficients obtained exceed the information degrees of freedom. Let the Fourier coefficients Y[r] form a Toeplitz matrix, which is in the form of:

[0051]

[0052] Use singular value decomposition (SVD) to decompose the matrix A T Decompose it into the form:

[0053] A T =U T B(V T ) T (13)

[0054] Where U T Is an orthogonal matrix, containing the matrix A T The left singular vector of ; Β is a diagonal matrix, the elements on the diagonal are the matrix A T The singular values ​​of V in descending order; T Is an orthogonal matrix, containing the matrix A T The right singular vectors of .

[0055] According to the size of the elements in B, select the Fourier coefficients Y[l] corresponding to the largest 2L+1 singular values ​​to reconstruct the L-dimensional Toeplitz matrix A L For the subsequent parameter estimation process, the matrix A L Expressed as:

[0056]

[0057] Construct a zeroing filter with L coefficients, assuming that the filter is If there is a zero point, the Z domain representation of the filter is:

[0058]

[0059] The Fourier coefficient Y[l] is filtered using an annihilation filter to satisfy P[l]*Y[l]=0. The Yule-Walker equation is established using the L-dimensional Toeplitz matrix to solve the annihilation filter coefficient P[l].

[0060]

[0061] Furthermore, the step 7 specifically includes the following process:

[0062] Using the zeroing filter coefficient P[l] and the calculated u l Delay parameters for the signal To solve,

[0063]

[0064] Using the known Fourier coefficients Y[l] and u l Establish the Vandermonde equation for the amplitude parameter of the signal To solve,

[0065]

[0066] Combined calculation results of delay and amplitude parameters Gaussian pulse function G(t) and pulse width factor of Gaussian pulse Recovered Gaussian pulse stream signal Expressed as:

[0067]

[0068] The present invention has the following beneficial effects:

[0069] An adaptive bandwidth sampling kernel is designed to dynamically match the sampling kernel bandwidth to the signal bandwidth, preventing performance degradation due to the randomness of the information degrees of freedom. This avoids spectral aliasing caused by insufficient bandwidth or interference introduced by excessive bandwidth, thereby improving signal reconstruction accuracy. Sampling at a high innovation rate and appropriately increasing the number of sampling points can improve the accuracy of parameter estimation and signal reconstruction, preserving more detailed signal information. A Fourier coefficient truncation method based on singular value decomposition is designed to remove redundant information from the perspective of singular values, retaining the most characteristic components of the data, reducing the amount of data and matrix size involved in the reconstruction calculation, and improving the efficiency of the parameter calculation and signal reconstruction process.

[0070] The present invention addresses the problem of increased echo signal complexity, increased echo information degrees of freedom, and random information degrees of freedom when multiple interface echoes and defect echoes are superimposed in the echo when using ultrasonic detection of complex components. The invention invents an ultrasonic data compression sampling method with random information degrees of freedom, which can greatly compress the amount of collected data while retaining valid information. A sampling signal duration self-compensation method is proposed to improve reconstruction accuracy. The sampling kernel can be adaptively and dynamically matched according to the signal bandwidth, the SoS sampling kernel parameter configuration is optimized, and the sampling interval and the calculation method of the number of sampling data points are redefined. Taking into account the problem that the detection method of the present invention cannot interpret defect information in real time during actual detection due to the multiple and random characteristics of the echo peaks, sampling is performed at a high innovation rate that is more than twice that of the traditional method. This can retain both the defect echo information in the detection signal and the multiple interface echo information, and can retain more complete information for subsequent information interpretation. A Fourier coefficient truncation method based on singular value decomposition was designed to eliminate redundant information from the perspective of singular values, retain the main components that best reflect the characteristics of the data, reduce the amount of data and the size of the matrix involved in the reconstruction calculation, and improve the efficiency of parameter calculation and signal reconstruction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0072] Figure 1 is a flow chart of the method of the present invention;

[0073] Figure 2 Schematic diagram of the ultrasonic detection system layout and standard defects in an embodiment of the present invention;

[0074] Figure 3 1 is a diagram of an ultrasonic echo signal and a restored ultrasonic echo signal in Example 1 of the present invention;

[0075] Figure 4 1 is a diagram of an ultrasonic echo signal and a restored ultrasonic echo signal in Example 2 of the present invention;

[0076] Figure 5 1 is a diagram of an ultrasonic echo signal and a restored ultrasonic echo signal in Example 3 of the present invention. DETAILED DESCRIPTION

[0077] The specific forms of the present invention are further described below in conjunction with the accompanying drawings and embodiments. Figure 1 As shown. The present invention can also be applied through other equivalent embodiments. The embodiments and accompanying drawings provided in the following embodiments only illustrate the basic method of the present invention by way of example. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0078] The technical solution of the present invention is further described below in conjunction with specific embodiments. Specific embodiment 1:

[0080] In the embodiment of the present invention, the ultrasonic detection system is arranged as follows: Figure 2 As shown in the figure, the center frequency of the ultrasonic sensor probe is 5MHz, the probe diameter is 13mm, and the model is I2-5P13-H (IGI0736). A 60kg / m rail was selected as a test block. The test block contained three sections: no defects, rail head through hole defects, and rail waist through hole defects. The defect size and location are consistent with the national standard, as shown in the figure. Figure 2 As shown in the figure, the longitudinal velocity of ultrasonic waves in rails is 5900 m / s. The ultrasonic wave is incident vertically downward. The center of the ultrasonic sensor probe is positioned at the center of the top of the test block and pressed firmly. Water is used as the coupling medium. The sensor is tested in self-transmitting and self-receiving mode, with a sampling time of 63 μs and a Nyquist sampling frequency of 14 MHz.

[0081] The ultrasonic echo signal obtained by ultrasonic sensor when detecting a rail without defects is as follows: Figure 3As shown, the signal threshold is set to 20% of the maximum amplitude, resulting in a valid peak number of 4 and an information degree of freedom of 8. The time length is increased by 7.9 μs before the start of the signal and by 7.9 μs after the end, and the time length of the signal is 78.8 μs.

[0082] The ultrasonic echo signal was filtered by a bandpass filter with a passband of 2-6 MHz, stopband frequencies of 1 MHz and 7 MHz, a passband ripple of 1 dB, and a stopband attenuation of 40 dB. The instantaneous power signal was extracted as a Gaussian pulse stream signal after Hilbert transform. The half-peak pulse widths of the four effective peaks were 0.4 μs, 0.4 μs, 0.7 μs, and 0.44 μs, respectively. The corresponding Gaussian pulse width factors were 12.5, 12.5, 4.1, and 10.3, respectively.

[0083] The bandwidth of the Gaussian pulse stream signal is determined to be 0.33 MHz. The Hamming window is selected as the weight of the SoS sampling kernel, and the bandwidth of the sampling kernel is set to 0.36 MHz. It can be deduced that the number of SoS sampling kernel index points is 57.

[0084] The Gaussian pulse stream signal is input into the SoS sampling kernel for convolution.

[0085] With 57 sampling points and a high innovation rate of 0.72 MHz as the sampling rate, the signal modulated by the sampling kernel is sparsely sampled at equal intervals to obtain a sparse sequence.

[0086] The sparse sequence is subjected to DFT to obtain the corresponding 57 Fourier coefficients. Singular value decomposition (SVD) is used to retain the Fourier coefficients corresponding to the first 9 largest singular values ​​to construct a 9-dimensional Toeplitz matrix to solve the annihilation filter coefficients.

[0087] From the known results, we get {0.141, 2.320, 15.750, 59.625} as the delay parameter estimation result, and {0.550, 0.578, 0.423, 1.000} as the normalized result of the amplitude parameter estimation. Combining the Gaussian pulse function and the pulse width factor of the Gaussian pulse to restore the ultrasonic echo signal, the amount of sampling data is reduced to 5.14% of the traditional sampling method, and the maximum absolute error of the delay is 0.21μs. The recovered ultrasonic echo signal is shown in Figure 3 shown. Specific embodiment 2:

[0089] In the embodiment of the present invention, the test block and the defect geometric distribution are as follows: Figure 2 As shown, the diameter of the circular through hole is 3 mm, and the distance between the center of the through hole and the rail vertex is 18 mm. The ultrasonic detection system layout, ultrasonic sensor, and detection method used are the same as those in Example 1.

[0090] The ultrasonic echo signal obtained by ultrasonic sensor when detecting rail with rail head through hole defect is as follows: Figure 4 As shown, the signal threshold is 20% of the maximum amplitude, resulting in 6 effective peaks and 12 degrees of freedom of information. The time length is increased by 5.3 μs before the start of the signal and 5.3 μs after the end, and the time length of the signal is 73.6 μs.

[0091] The ultrasonic echo signal was filtered by a bandpass filter with a passband of 2-6 MHz, stopband frequencies of 1 MHz and 7 MHz, a passband ripple of 1 dB, and a stopband attenuation of 40 dB. The instantaneous power signal was extracted as a Gaussian pulse stream signal after Hilbert transform. The half-peak pulse widths of the six effective peaks were 0.4 μs, 0.44 μs, 0.7 μs, 0.4 μs, 0.68 μs, and 0.6 μs, respectively. The corresponding Gaussian pulse width factors were 12.5, 10.3, 4.1, 12.5, 4.3, and 5.6, respectively.

[0092] The bandwidth of the Gaussian pulse stream signal is determined to be 0.33 MHz. The Hamming window is selected as the weight of the SoS sampling kernel, and the bandwidth of the sampling kernel is set to 0.36 MHz. It can be deduced that the number of SoS sampling kernel index points is 53.

[0093] The Gaussian pulse stream signal is input into the SoS sampling kernel for convolution.

[0094] With 53 sampling points and a high innovation rate of 0.72 MHz as the sampling rate, the signal modulated by the sampling kernel is sparsely sampled at equal intervals to obtain a sparse sequence.

[0095] The sparse sequence is subjected to DFT to obtain the corresponding 53 Fourier coefficients. Singular value decomposition (SVD) is used to retain the Fourier coefficients corresponding to the first 13 largest singular values ​​to construct a 13-dimensional Toeplitz matrix to solve the annihilation filter coefficients.

[0096] From the known results, we get {0.422, 3.375, 5.977, 11.320, 15.820, 59.700} as the delay parameter estimation result, and {0.741, 0.126, 1.000, 0.142, 0.098, 0.955} as the normalized result of the amplitude parameter estimation. Combining the Gaussian pulse function with the pulse width factor of the Gaussian pulse to restore the ultrasonic echo signal, the amount of sampling data is reduced to 5.14% of the traditional sampling method, and the maximum absolute error of the delay is 0.35μs. The recovered ultrasonic echo signal is shown in Figure 4 shown. Specific embodiment 3:

[0098] In the embodiment of the present invention, the test block and the defect geometric distribution are as follows: Figure 2As shown, the diameter of the circular through hole is 4 mm, and the distance between the center of the through hole and the rail vertex is 97 mm. The ultrasonic detection system layout, ultrasonic sensor, and detection method used are the same as those in Example 1.

[0099] The ultrasonic echo signal obtained by the ultrasonic sensor when detecting the rail with the rail waist through hole defect is as follows: Figure 5 As shown, the signal threshold is 20% of the maximum amplitude, resulting in a valid peak number of 5 and an information degree of freedom of 10. The time length is increased by 6.3 μs before the start of the signal and by 6.3 μs after the end, and the time length of the signal is 75.6 μs.

[0100] The ultrasonic echo signal was filtered by a bandpass filter with a passband of 2-6 MHz, stopband frequencies of 1 MHz and 7 MHz, a passband ripple of 1 dB, and a stopband attenuation of 40 dB. The instantaneous power signal was extracted as a Gaussian pulse stream signal after Hilbert transform. The half-peak pulse widths of the five effective peaks were 0.4 μs, 0.6 μs, 0.7 μs, 0.44 μs, and 0.4 μs, respectively. The corresponding Gaussian pulse width factors were 12.5, 5.6, 4.1, 10.3, and 12.5, respectively.

[0101] The bandwidth of the Gaussian pulse stream signal is determined to be 0.32 MHz. The Hamming window is selected as the weight of the SoS sampling kernel, and the bandwidth of the sampling kernel is set to 0.35 MHz. It can be deduced that the number of SoS sampling kernel index points is 53.

[0102] The Gaussian pulse stream signal is input into the SoS sampling kernel for convolution.

[0103] With 53 sampling points and a high innovation rate of 0.70 MHz as the sampling rate, the signal modulated by the sampling kernel is sparsely sampled at equal intervals to obtain a sparse sequence.

[0104] The sparse sequence is subjected to DFT to obtain the corresponding 53 Fourier coefficients. Singular value decomposition (SVD) is used to retain the Fourier coefficients corresponding to the first 11 largest singular values ​​to construct an 11-dimensional Toeplitz matrix to solve the annihilation filter coefficients.

[0105] From the known results, we obtain {0.352, 2.813, 15.820, 32.700, 59.625} as the delay parameter estimation results, and {0.949, 0.120, 0.935, 0.790, 1.000} as the normalized amplitude parameter estimation results. Combining the Gaussian pulse function with the pulse width factor of the Gaussian pulse to restore the ultrasonic echo signal, the sampling data volume is reduced to 5.01% of the traditional sampling method, and the maximum absolute error of the delay is 0.35μs. The recovered ultrasonic echo signal is shown in Figure 5 shown.

Claims

1. A random information degree of freedom ultrasonic data compression sampling method, characterized in that: The steps include: Step 1. According to the received ultrasonic echo signal x(t), determine the time length of the signal as T, determine the number of effective peaks L, calculate the information freedom Ω of the effective peaks, and complete the time compensation of the signal x(t). The time length after compensation is T P ; step 2. Perform pulse flow extraction on the ultrasonic echo signal x(t) to obtain the pulse flow signal g c (t), g c (t) is set to half-peak pulse width The pulse width factor is Gaussian pulse stream signal, where l is the valid peak number; Step 3. According to the Gaussian pulse flow signal g c (t) is used to design the SoS sampling kernel with adaptive bandwidth based on the time domain and frequency domain characteristics. The sampling kernel bandwidth ω is set. w , establish ω w The relationship between the maximum value M of the sampling kernel index point m determines the number of index points N M ; Step 4. Gaussian pulse stream signal g c (t) is input into the SoS sampling core for modulation to obtain the signal y(t); Step 5. According to the information freedom Ω of the effective peak and the number of SoS sampling kernel index points N M Determine the number of signal sparse sampling points N s , calculate the sampling rate ρ of the high innovation rate s And the sampling interval T s The discrete sequence y[n] is obtained by sampling the signal at equal intervals at a high innovation rate, where n = 0, 1, 2, ..., N s -1; Step 6. Use singular value decomposition (SVD) to extract the most important 2L+1 components of the Fourier coefficients Y[r] of the discrete sequence y[n] obtained by sparse sampling, where r = 0, 1, 2, ..., N s -1, solve to get the zeroing filter coefficient P[l]; Step 7. Obtain the signal parameter estimate from the zeroing filter coefficient P[l] using the spectrum estimation method t l is the delay of the Gaussian pulse, e l is the amplitude of the Gaussian pulse, combined with the Gaussian pulse function and the pulse width factor Recovering ultrasonic echo signals 2. A method according to claim 1, characterized in that: In step 1, the received ultrasonic echo signal x(t) contains multiple echoes of the measured body interface, defect echoes and noise. Different from the conventional ultrasonic detection echo signal, the echo complexity is significantly increased after interference superposition, and the information freedom of the signal is random. The effective detection echo signal threshold X is obtained by statistics. T , the peak value in the statistical echo signal is greater than X T The number of effective peaks is L, and the information freedom of the signal is Ω=2L.

3. A method according to claim 1, characterized in that: In step 1, the first and last ends of the received ultrasonic echo signal x(t) are compensated The time length of the compensated signal is , which ensures that the initial wave and the bottom wall echo are complete and not lost in the subsequent process, avoiding the reduction of reconstruction accuracy.

4. A method according to claim 1, characterized in that: In step 2, according to the center frequency f0 and bandwidth B of the ultrasonic echo signal x(t), w Determine the passband cutoff frequency of a bandpass filter and Stopband cutoff frequency f s and f p , passband ripple A p and the stopband attenuation A s , use a bandpass filter to filter the ultrasonic echo signal x(t), retain the effective frequency components, suppress noise interference, and perform Hilbert transform on the filtered signal to obtain the signal g h (t), and then converted into an instantaneous power signal to improve the dynamic range of the signal, and obtain the pulse flow signal g c (t).

5. A method according to claim 1, characterized in that: In step 2, the pulse function type is obtained by prior knowledge and is Gaussian pulse by default. The pulse flow signal g c (t) is set to the half-peak pulse width of the effective peak signal. Gaussian pulse stream signal, where is the pulse width factor of the Gaussian pulse, and establishes the effective peak signal half-peak pulse width Pulse width factor of Gaussian pulse The relationship, that is 6. A method according to claim 1, characterized in that: In step 3, the Gaussian pulse flow signal g is calculated. c The bandwidth ω of (t) g and time length T P As a condition for designing the sampling kernel; In step 3, the frequency domain expression of the SoS sampling kernel is in the form of weighted superposition of sinc functions, and the time domain support length is the Gaussian pulse stream signal g c (t) the length of time T P , the weight is Hamming window, sampling kernel index point m∈[-M,M], where M∈N + , let N M is the number of index points m, that is, N M =2M+1, establish the SoS sampling core bandwidth ω w The relationship with the maximum value M of the SoS sampling kernel index point m is: Where, This is a round-up operation.

7. A method according to claim 1, characterized in that: In step 4, the Gaussian pulse stream signal g c (t) Input the designed SoS sampling kernel and obtain the output signal y(t).

8. A method according to claim 1, characterized in that: In step 5, the number of sparse sampling points N s The effective peak information freedom Ω, the minimum number of sampling points 2L+1 and the number of SoS sampling core index points N M Decision, expressed as: The sampling rate is expressed as And because ρ s Greater than the innovation rate of the signal, called high innovation rate, the sampling interval is determined as After sampling at high innovation rate, the discrete sequence y[n] is obtained, where n = 0, 1, 2, ..., N s -1.

9. A method according to claim 1, characterized in that: In step 6, DFT is performed on the sampled discrete sequence y[n] to calculate the Fourier coefficient Y[r], where r=0, 1, 2, ..., N s -1; In step 6, the Fourier coefficient Y[r] is formed into N s dimensional Toeplitz matrix A T ; Use singular value decomposition (SVD) to N M dimensional matrix A T Perform the transformation and let A T =U T B(V T ) T , where U T Is an orthogonal matrix, containing the matrix A T The left singular vector of ; B is a diagonal matrix, the elements on the diagonal are the matrix A T The singular values ​​of V in descending order; T Is an orthogonal matrix, containing the matrix A T The right singular vector of B; according to the size of the elements in B, select the Fourier coefficients corresponding to the largest 2L+1 singular values ​​to reconstruct the L-dimensional Toeplitz matrix A L ; Using the matrix A through spectral estimation method L Complete the solution of the zeroing filter coefficient P[l].

10. A method according to claim 1, characterized in that: In step 7, the calculated annihilation filter coefficient P[l] and the L-dimensional Toeplitz matrix A L Complete the delay and amplitude parameters Calculation of Combined with the known Gaussian pulse function G(t) and the pulse width factor of the Gaussian pulse As well as delay and amplitude parameters Recovering ultrasonic echo signals

Images