Low complexity frequency domain sparse adaptive line enhancement method and system

Abstract

The application provides a low-complexity frequency-domain sparse adaptive line spectrum enhancement method and system, wherein the method comprises the following steps: S1, reading a radiated noise signal generated by a water sound target; S2, dividing a decorrelation signal into snapshots as time-domain inputs of an adaptive filter; S3, obtaining frequency-domain snapshots; S4, taking the frequency-domain snapshots as frequency-domain inputs of the adaptive filter; S5, calculating filtered output signals and corresponding estimation errors; S6, constructing a cost function and introducing a norm of a frequency-domain tap weight vector as a regularization term; S7, obtaining an updated frequency-domain tap weight vector; S8, obtaining time-domain inputs at the next moment; and S9, repeating steps S3 to S8 to obtain an enhanced radiated noise signal. The application converts time-domain inputs into the frequency domain by using a sliding discrete Fourier transform, reduces the calculation complexity from O(L) to O(Llog2L), greatly reduces the calculation complexity of ALE, and lays a foundation for engineering application.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of sonar technology in underwater acoustic engineering, in particular, to a low-complexity frequency-domain sparse adaptive line enhancement method and system. BACKGROUND

[0002] In the frequency spectrum composition of underwater acoustic target radiated noise, the characteristics of continuous spectrum and line spectrum are superimposed. The main sources of line spectrum generation include the reciprocating operation of the mechanical parts of the underwater acoustic target, the periodic water striking of the propeller, and the resonance phenomenon of the target itself. Due to the relatively stable working condition and large inertia of the underwater acoustic target, the line spectrum often has high stability. This stability makes the line spectrum a key basis for underwater acoustic target detection and identification.

[0003] However, with the continuous progress and wide application of modern shock absorption and noise reduction technology, the underwater acoustic target radiated noise level shows a decreasing trend year by year. Under this background, the line spectrum is easily covered by strong interference and difficult to highlight. Therefore, breaking through the line spectrum enhancement technology under the condition of low signal-to-noise ratio has important practical significance and application value for improving the underwater acoustic target detection and identification capability. In the current field of underwater acoustic signal processing, the adaptive line enhancer (ALE) is a commonly used line spectrum enhancement method.

[0004] Through the search of patent documents, it is found that the patent document with application number 202311269192.1 discloses a frequency-domain sparse adaptive line spectrum enhancement method. Based on the sparse characteristics of the ALE frequency response, a sparse adaptive algorithm is used to iteratively update the frequency-domain tap weight vector of the ALE. In this way, it is promoted to be closer to the Wiener solution of the ALE, thereby effectively improving the processing gain and widening the signal-to-noise ratio range.

[0005] However, ALE still faces severe challenges in actual engineering applications. The root cause lies in the point-by-point sliding processing mode adopted by ALE. This means that at each time step, the input vector needs to be updated. That is, first, the inner product operation of the current input vector and the tap weight vector of ALE is performed to generate the output signal of ALE, and then the error is calculated based on the output signal and the tap weight vector is updated accordingly to optimize the performance of ALE. After completing the above operations, the input vector at the next time is obtained by sliding one sample point, and the above series of tedious operations are repeated. Due to these complex and frequent operations, the calculation complexity of ALE is high, which greatly limits its wide application and promotion in actual engineering.

[0006] In summary, in view of the problems of the prior art, a low-complexity frequency-domain sparse adaptive line enhancement method and system are researched to provide a more efficient and practical line enhancement solution for underwater acoustic signal processing, which is a key task to be solved at present. SUMMARY

[0007] In view of the defects in the prior art, the purpose of the present application is to provide a low-complexity frequency-domain sparse adaptive line enhancement method and system.

[0008] According to the low-complexity frequency-domain sparse adaptive line enhancement method provided by the present application, the following steps are included:

[0009] Step S1, read the radiated noise signal generated by the underwater acoustic target as the reference signal of the adaptive filter;

[0010] Step S2, delay the radiated noise signal to obtain a decorrelation signal, and then divide the decorrelation signal into snapshots according to the length of the tap weight vector of the adaptive filter, as the time-domain input of the adaptive filter;

[0011] Step S3, transform the time-domain input to the frequency domain by using a sliding discrete Fourier transform to obtain a frequency-domain snapshot;

[0012] Step S4, take the frequency-domain snapshot as the frequency-domain input of the adaptive filter;

[0013] Step S5, based on the reference signal and the frequency-domain input of the adaptive filter, calculate the filtered output signal and the corresponding estimation error;

[0014] Step S6, according to the estimation error, construct a cost function based on the maximum correlation entropy criterion, and introduce the norm of the frequency-domain tap weight vector as a regularization term;

[0015] Step S7, calculate the gradient of the cost function, derive the update formula of the frequency-domain tap weight vector of ALE, and obtain the updated frequency-domain tap weight vector;

[0016] Step S8, slide the decorrelation signal by one sample point to obtain the time-domain input at the next time;

[0017] Step S9, loop steps S3 to S8 until the processing of all radiated noise signals is completed, and obtain the enhanced radiated noise signal.

[0018] Preferably, step S2 includes the following sub-steps:

[0019] Step S2.1, delay the radiated noise signal s(n) to obtain a decorrelation signal s(n-Δ);

[0020] Step S2.2, according to the length of the tap weight vector of the adaptive filter, the decorrelated signal s(n-Δ) is divided into snapshots as the time-domain input of the adaptive filter at time n, the calculation formula is:

[0021] s(n-Δ) = [s(n-Δ),...,s(n-Δ-L+1)] T (1)

[0022] wherein, L is the length of the tap weight vector of the adaptive filter, Δ is the delay time, T represents the transpose.

[0023] Preferably, in step S2.2, L=1000, Δ=1.

[0024] Preferably, in step S3, the formula of the frequency-domain input of the adaptive filter at time n is:

[0025]

[0026] wherein, s F (n-Δ) is the discrete Fourier transform of the time-domain input, is each frequency component of the discrete Fourier transform, i=0,…,L-1;

[0027] The derivation formula is:

[0028]

[0029] wherein, j represents the imaginary unit, m and p represent the summation index.

[0030] Based on the derivation formula, the frequency-domain input s F (n-Δ) of the adaptive filter at time n is represented by the frequency-domain input s F (n-1-Δ) of the adaptive filter at time n-1.

[0031] Preferably, in step S5, for each time n, the calculation formula of the filtered output signal is as follows:

[0032]

[0033] wherein, is the frequency-domain tap weight vector of the adaptive filter at time n, the initial value is 0;

[0034] The corresponding estimation error calculation formula is:

[0035] e(n) = s(n)-y(n) (5)

[0036] wherein, s(n) is the input signal, (·) H represents the conjugate transpose.

[0037] Preferably, in step S6, the cost function is calculated as follows:

[0038]

[0039] wherein σ represents the kernel width of the maximum correlation entropy criterion; ||·||1 represents the L1 norm; ρ>0 represents a balance factor for balancing the influence of the regularization term; exp represents the exponential function.

[0040] Preferably, in step S6, σ=1; ρ=0.01.

[0041] Preferably, in step S7, the update formula of the frequency-domain tap weight vector is as follows:

[0042]

[0043] wherein μ represents the update step size of the frequency-domain tap weight vector; (·) * represents the conjugate operation; sgn(·) represents the sign function.

[0044] Preferably, in step S7, μ=10 -5 .

[0045] The application further provides a low-complexity frequency-domain sparse adaptive line enhancement system, comprising:

[0046] Module M1 reads the radiated noise signal generated by the underwater target as the reference signal of the adaptive filter;

[0047] Module M2 performs a delay operation on the radiated noise signal to obtain a decorrelation signal, and then divides the decorrelation signal into snapshots according to the length of the tap weight vector of the adaptive filter, as the time-domain input of the adaptive filter;

[0048] Module M3 transforms the time-domain input into the frequency domain by using a sliding discrete Fourier transform to obtain a frequency-domain snapshot;

[0049] Module M4 takes the frequency-domain snapshot as the frequency-domain input of the adaptive filter;

[0050] Module M5 calculates the filtered output signal and the corresponding estimation error based on the reference signal and the frequency-domain input of the adaptive filter;

[0051] Module M6 constructs a cost function based on the maximum correlation entropy criterion according to the estimation error, wherein the cost function introduces the norm of the frequency-domain tap weight vector as a regularization term;

[0052] Module M7 calculates the gradient of the cost function, derives the update formula of the frequency-domain tap weight vector of the ALE, and obtains the updated frequency-domain tap weight vector;

[0053] Module M8, sliding the decorrelation signal by one sample point to obtain the time domain input at the next moment;

[0054] Module M9, cyclically executing module M3, module M4, module M5, module M6, module M7 and module M8 until the processing of all radiation noise signals is completed, to obtain the enhanced radiation noise signal.

[0055] Compared with the prior art, the present application has the following beneficial effects:

[0056] 1. The present application uses sliding discrete Fourier transform to convert the time domain input to the frequency domain, reduces the computational complexity from O(L) to O(Llog2L), greatly reduces the computational complexity of ALE, and lays a foundation for engineering application.

[0057] 2. The present application designs the cost function of ALE based on the maximum correlation entropy criterion, so that ALE still has good applicability in non-Gaussian noise environment, and effectively expands the application scene.

[0058] 3. The present application introduces a regularization term in the cost function of ALE, encourages the sparsity of the frequency domain tap weight vector, makes the tap weight vector of ALE closer to the Wiener solution, and improves the processing gain of ALE. BRIEF DESCRIPTION OF DRAWINGS

[0059] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments, made with reference to the accompanying drawings:

[0060] Figure 1 Flow chart of a low-complexity frequency-domain sparse adaptive line spectrum enhancement method in an embodiment of the present application;

[0061] Figure 2 Time domain waveform diagram of non-Gaussian noise in an embodiment of the present application;

[0062] Figure 3 LOFAR spectrum of the input signal in an embodiment of the present application;

[0063] Figure 4 Frequency domain tap weight vector of the conventional ALE in an embodiment of the present application;

[0064] Figure 5 Frequency domain tap weight vector of the method of the present application in an embodiment of the present application;

[0065] Figure 6 LOFAR spectrum of the output signal of the conventional ALE in an embodiment of the present application;

[0066] Figure 7 LOFAR spectrum of the output signal of the method of the present application in an embodiment of the present application. DETAILED DESCRIPTION

[0067] The application will be described in detail below with specific examples. The following examples will help those skilled in the art to further understand the application, but do not limit the application in any form. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the application. These are within the scope of the application.

[0068] The technical problem to be solved by the application is to provide a low-complexity frequency-domain sparse adaptive line spectrum enhancement method to solve the problem that the frequency-domain sparse adaptive line spectrum enhancement method has high complexity and is difficult to apply in engineering.

[0069] Embodiment 1

[0070] Figure 1 A flowchart of a low-complexity frequency-domain sparse adaptive line spectrum enhancement method in the embodiment of the application.

[0071] As shown in Figure 1 The embodiment provides a low-complexity frequency-domain sparse adaptive line spectrum enhancement method, which comprises the following steps:

[0072] Step S1, reading the radiated noise signal s(n) generated by the underwater acoustic target as the reference signal of the adaptive filter.

[0073] In the embodiment, the input signal is x(n) = A cos(2πfn / f s ) + noise(n), the line spectrum frequency f = 100 Hz, the sampling rate f s = 2000 Hz, the signal length is 20 s, the signal-to-noise ratio is -27 dB, and the input noise is non-Gaussian noise obeying a mixed Gaussian distribution.

[0074] Figure 2 A time-domain waveform diagram of non-Gaussian noise in the embodiment of the application; Figure 3 The LOFAR spectrum of the input signal in the embodiment of the application.

[0075] The time-domain waveform diagram of non-Gaussian noise is shown in Figure 2 Compared with Gaussian noise, non-Gaussian noise has a "heavy tail" characteristic, resulting in a large dynamic range of non-Gaussian noise, which is not conducive to line spectrum detection; the LOFAR spectrum of the input signal is shown in Figure 3 It can be seen that the signal-to-noise ratio of the input signal is extremely low, and the line spectrum is completely submerged by noise.

[0076] Step S2, delay operation is performed on the radiated noise signal s(n) to obtain an decorrelation signal, and then the decorrelation signal is divided into snapshots according to the length of the tap weight vector of the adaptive filter, as the time domain input of the adaptive filter.

[0077] Specifically, step S2 includes the following sub-steps:

[0078] Step S2.1, delay operation is performed on the radiated noise signal s(n) to obtain an decorrelation signal s(n-Δ).

[0079] Step S2.2, the decorrelation signal s(n-Δ) is divided into snapshots according to the length of the tap weight vector of the adaptive filter, as the time domain input of the adaptive filter at time n, and the calculation formula is:

[0080] s(n-Δ) = [s(n-Δ), …, s(n-Δ-L+1)] T (1)

[0081] Wherein, L is the length of the tap weight vector of the adaptive filter, Δ is the delay time, and T represents transposition.

[0082] In this embodiment, L = 1000 and Δ = 1.

[0083] Step S3, the time domain input is transformed to the frequency domain by using the sliding discrete Fourier transform to obtain the frequency domain snapshot.

[0084] Specifically, the formula of the frequency domain input of the adaptive filter at time n is:

[0085]

[0086] Wherein, s F (n-Δ) is the discrete Fourier transform of the time domain input, is each frequency component of the discrete Fourier transform, and i = 0, …, L-1.

[0087] Specifically,

[0088]

[0089] Wherein, j represents the imaginary unit, and m and p represent the summation index.

[0090] It can be seen that the frequency domain input s F (n-Δ) of the adaptive filter at time n is represented by the frequency domain input s F (n-1-Δ) of the adaptive filter at time n-1.

[0091] Compared with the discrete Fourier transform and the fast discrete Fourier transform, the sliding discrete Fourier transform has significant advantages in transforming the time-domain input into the frequency domain. First, the sliding discrete Fourier transform and the discrete Fourier transform have the same effect and do not cause loss of frequency-domain information. Second, the sliding discrete Fourier transform only needs one complex multiplication operation and two real addition operations to transform the time-domain input into the frequency domain, and the computational complexity is O(L), while the computational complexity of the discrete Fourier transform and the fast discrete Fourier transform is O(L 2 ) and O(Llog2L), respectively. Therefore, the complexity of the sliding discrete Fourier transform is lower, and its advantage is more obvious when the tap weight vector L of the adaptive filter is large.

[0092] Step S4, the frequency-domain snapshot is taken as the frequency-domain input of the adaptive filter.

[0093] Step S5, based on the reference signal and the frequency-domain input of the adaptive filter, the filtered output signal and the corresponding estimation error are calculated.

[0094] Specifically, for each time n, the calculation formula of the filtered output signal is as follows:

[0095]

[0096] wherein, is the frequency-domain tap weight vector of the adaptive filter at time n, and the initial value is 0.

[0097] The calculation formula of the corresponding estimation error is as follows:

[0098] e(n)=s(n)-y(n) (5)

[0099] wherein, s(n) is the input signal, (·) H represents the conjugate transpose.

[0100] Step S6, according to the estimation error, a cost function based on the maximum correlation entropy criterion is constructed, and the L1 norm of the frequency-domain tap weight vector is introduced as a regularization term in the cost function to encourage the sparsity of the frequency-domain tap weight vector.

[0101] Specifically, the calculation formula of the cost function is as follows:

[0102]

[0103] wherein, σ represents the kernel width of the maximum correlation entropy criterion; ||·||1 represents the L1 norm; ρ>0 represents a balance factor for balancing the influence of the regularization term; exp represents the exponential function.

[0104] In this embodiment, σ=1; ρ=0.01.

[0105] Step S7: Calculate the gradient of the cost function, derive the update formula for the frequency domain tap weight vector of ALE, and obtain the updated frequency domain tap weight vector.

[0106] Specifically, the update formula for the frequency domain tap weight vector is as follows:

[0107]

[0108] Where μ represents the update step size of the frequency domain tap weight vector; (·) * represents the conjugate operation; sgn(·) represents the symbolic function.

[0109] In this embodiment, μ = 10 -5 .

[0110] Step S8: Slide the decorrelation signal by one sampling point to obtain the time domain input at the next moment;

[0111] Step S9: Repeat steps S3, S4, S5, S6, S7 and S8 until all radiated noise signals have been processed and the enhanced radiated noise signal is obtained.

[0112] Figure 4 The frequency domain tap weight vector of a conventional ALE in this embodiment of the invention; Figure 5 The frequency domain tap weight vector of the method of the present invention in this embodiment; Figure 6 The LOFAR spectrum of a conventional ALE output signal in this embodiment of the invention; Figure 7 The LOFAR spectrum of the output signal of the method of the present invention in this embodiment.

[0113] like Figure 4 , 5 As shown, due to the influence of the regularization term, the frequency domain tap weight vector of the method of the present invention is closer to the Wiener solution, which can better suppress broadband noise and achieve higher processing gain.

[0114] like Figure 6 , 7 As shown, the enhanced line spectrum produced by the method of the present invention is clearly visible and significantly superior to conventional ALE.

[0115] Example 2:

[0116] The present invention also provides a low-complexity frequency domain sparse adaptive line spectrum enhancement system. The low-complexity frequency domain sparse adaptive line spectrum enhancement system can be implemented by executing the process steps of the low-complexity frequency domain sparse adaptive line spectrum enhancement method. That is, those skilled in the art can understand the low-complexity frequency domain sparse adaptive line spectrum enhancement method as a preferred embodiment of the low-complexity frequency domain sparse adaptive line spectrum enhancement system.

[0117] The frequency-domain sparse adaptive line enhancement system comprises:

[0118] Module M1 reads the radiated noise signal generated by the underwater target as a reference signal of the adaptive filter;

[0119] Module M2 performs delay resolution operation on the radiated noise signal to obtain a decorrelation signal, and divides the decorrelation signal into snapshots according to the length of the tap weight vector of the adaptive filter as time-domain inputs of the adaptive filter;

[0120] Module M3 transforms the time-domain inputs into the frequency domain by using a sliding discrete Fourier transform to obtain frequency-domain snapshots;

[0121] Module M4 takes the frequency-domain snapshots as frequency-domain inputs of the adaptive filter;

[0122] Module M5 calculates a filtered output signal and a corresponding estimation error based on the reference signal and the frequency-domain inputs of the adaptive filter;

[0123] Module M6 constructs a cost function based on the maximum correlation entropy criterion according to the estimation error, and introduces the norm of the frequency-domain tap weight vector as a regularization term;

[0124] Module M7 calculates the gradient of the cost function, derives an update formula of the frequency-domain tap weight vector of the ALE, and obtains an updated frequency-domain tap weight vector;

[0125] Module M8 slides the decorrelation signal by one sample point to obtain time-domain inputs at the next moment;

[0126] Module M9 cyclically executes modules M3, M4, M5, M6, M7 and M8 until the processing of all radiated noise signals is completed to obtain enhanced radiated noise signals.

[0127] Those skilled in the art know that, in addition to implementing the system provided by the present application and each device, module and unit thereof in a pure computer-readable program code manner, the system provided by the present application and each device, module and unit thereof can also be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps to achieve the same functions. Therefore, the system provided by the present application and each device, module and unit thereof can be considered as a hardware component, and the devices, modules and units included therein for achieving various functions can also be considered as structures within the hardware component; the devices, modules and units for achieving various functions can also be considered as both software modules implementing methods and structures within the hardware component.

[0128] The specific embodiments of the present application are described above. It needs to be understood that the present application is not limited to the specific embodiments described above, and various changes or modifications can be made by those skilled in the art within the scope of the claims, which does not affect the essential content of the present application. The embodiments of the present application and the features in the embodiments can be combined with each other at will without conflict.

Claims

1. A low complexity frequency domain sparse adaptive line enhancement method, characterized in that, The method comprises the following steps: Step S1, reading a radiated noise signal generated by a water acoustic target as a reference signal of an adaptive filter; Step S2, performing a delay de-correlation operation on the radiated noise signal to obtain a de-correlated signal, and dividing the de-correlated signal into snapshots according to a length of a tap weight vector of the adaptive filter as time domain inputs of the adaptive filter; Step S3, transforming the time domain inputs into a frequency domain by using a sliding discrete Fourier transform to obtain frequency domain snapshots; Step S4, taking the frequency domain snapshots as frequency domain inputs of the adaptive filter; Step S5, calculating a filtered output signal and a corresponding estimation error based on the reference signal and the frequency domain inputs of the adaptive filter; Step S6, constructing a cost function based on a maximum correlation entropy criterion according to the estimation error, and introducing a norm of a frequency domain tap weight vector as a regularization term in the cost function; Step S7, calculating a gradient of the cost function, deriving an update formula of the frequency domain tap weight vector of the ALE, and obtaining an updated frequency domain tap weight vector; Step S8, sliding the de-correlated signal by one sample point to obtain time domain inputs at a next time point; Step S9, repeatedly performing Step S3, Step S4, Step S5, Step S6, Step S7 and Step S8 until processing of all the radiated noise signals is completed to obtain enhanced radiated noise signals; In Step S3, a formula of the frequency domain inputs of the adaptive filter at a time point n is: （2） wherein is the discrete Fourier transform of the time domain input, is the individual frequency component of the discrete Fourier transform, ; The derivation formula is: （3） Where j represents an imaginary unit, and m and p represent summation indices; Based on the derived formula, the frequency domain input of the adaptive filter at time n from the frequency domain input of the adaptive filter at time n-1 denotes; In Step S6, a calculation formula of the cost function is: （6） wherein denotes the kernel width of the maximum correlation entropy criterion; denotes norm; denotes a balancing factor to balance the influence of the regularization term; exp denotes the exponential function; In Step S7, an update formula of the frequency domain tap weight vector is as follows: （7） wherein denotes an update step size of the frequency domain tap weight vector; denotes a conjugate operation; denotes a sign function.

2. The low complexity frequency domain sparse adaptive pitch enhancement method according to claim 1, wherein, Step S2 comprises the following sub-steps: Step S2.

1. performing a delay operation on the radiated noise signal to obtain a decorrelated signal ; Step S2.2, depending on the length of the tap weight vector of the adaptive filter, the decorrelated signal is divided into snapshots as time domain input of the adaptive filter at time n, the calculation formula is: （1） where L is the length of the tap weight vector of the adaptive filter, is the delay time, and T denotes the transpose.

3. The low complexity frequency domain sparse adaptive pitch enhancement method according to claim 2, wherein, In said step S2.2, L = 1000, .

4. The low complexity frequency-domain sparse adaptive pitch enhancement method according to claim 1, wherein, In Step S5, for each time point n, a calculation formula of the filtered output signal is as follows: （4） wherein, is the frequency-domain tap weight vector of the adaptive filter at time n, with an initial value of 0; A calculation formula of the corresponding estimation error is as follows: （5） where s(n) is the input signal, denotes the conjugate transpose.

5. The low complexity frequency domain sparse adaptive pitch enhancement method according to claim 1, wherein, In the step S6, ; .

6. The low complexity frequency-domain sparse adaptive pitch enhancement method according to claim 1, wherein, In the step S7, .

7. A low complexity frequency domain sparse adaptive line enhancement system employing a low complexity frequency domain sparse adaptive line enhancement method according to any one of claims 1 to 6, characterized in that, The method comprises: Module M1, reading a radiated noise signal generated by a water acoustic target as a reference signal of an adaptive filter; Module M2, performing a delay de-correlation operation on the radiated noise signal to obtain a de-correlated signal, and dividing the de-correlated signal into snapshots according to a length of a tap weight vector of the adaptive filter as time domain inputs of the adaptive filter; Module M3, transforming the time domain inputs into a frequency domain by using a sliding discrete Fourier transform to obtain frequency domain snapshots; Module M4, taking the frequency domain snapshots as frequency domain inputs of the adaptive filter; Module M5, calculating a filtered output signal and a corresponding estimation error based on the reference signal and the frequency domain inputs of the adaptive filter; Module M6, constructing a cost function based on a maximum correlation entropy criterion according to the estimation error, and introducing a norm of a frequency domain tap weight vector as a regularization term in the cost function; Module M7, calculating a gradient of the cost function, deriving an update formula of the frequency domain tap weight vector of the ALE, and obtaining an updated frequency domain tap weight vector; Module M8, sliding the de-correlated signal by one sample point to obtain time domain inputs at a next time point; Module M9, cyclically executing module M3, module M4, module M5, module M6, module M7 and module M8 until the processing of all the radiation noise signals is completed, obtaining the enhanced radiation noise signals.

Images