A method for active sonar signal recognition

Abstract

The application discloses an active sonar signal recognition method, and belongs to the field of underwater acoustic signal processing. The method utilizes frequency down-conversion, down-sampling and low-pass filtering to reduce data quantity, filter out out-of-band noise, reduce the influence of noise on active sonar signal recognition, and reduce the calculation amount of subsequent signal processing. For CW signals, power spectrum analysis is used to determine whether the signals are CW signals. For LFM and HFM signals, the attenuation of delay autocorrelation spectrum peaks is compared in more detail, a signal spectrum peak distribution model is constructed by analyzing a large amount of data, the threshold value of signal recognition is corrected, and higher recognition accuracy is obtained under the condition of low signal-to-noise ratio.

Description

Technical Field

[0001] This invention belongs to the field of underwater acoustic signal processing, and specifically relates to an active sonar signal recognition method. Background Technology

[0002] Sound waves are the most effective transmission medium for long-distance underwater information transmission, and active sonar signals are an important means of underwater target detection. Various surface and underwater reconnaissance equipment continuously transmit active sonar signals into the water. Common active sonar signals are mainly divided into continuous (CW) signals, linear frequency modulation (LFM) signals, and hyperbolic frequency modulation (HFM) signals. Identifying the received active sonar signals and obtaining their signal patterns to determine the type of sonar platform would be of great significance.

[0003] The main challenge in identifying active sonar signals lies in distinguishing between LFM and HFM signals. CW signals and frequency-modulated signals (HFM, LFM) differ significantly in their time-frequency characteristics; CW signals exhibit a distinct line spectrum, while LFM / HFM signals lack discrete spectral lines. Power spectrum analysis can provide a preliminary distinction between CW and LFM / HFM signals, but differentiation between LFM and HFM signals remains difficult. Common algorithms for distinguishing LFM and HFM signals include instantaneous autocorrelation and time-frequency analysis. Instantaneous autocorrelation offers advantages such as low computational complexity and high timeliness, but its effectiveness is poor at low signal-to-noise ratios. Furthermore, existing time-frequency analysis and other methods involve high computational complexity.

[0004] Currently, several modulation recognition algorithms that combine time-frequency analysis and neural networks have emerged in existing technologies. These algorithms have better performance than traditional algorithms, but due to factors such as high computational complexity, poor real-time performance, and complex and variable communication environments, they have not been widely used. Summary of the Invention

[0005] In view of this, the present invention provides an active sonar signal identification method. The present invention can solve the problem of low identification accuracy of existing active sonar signal identification methods under low signal-to-noise ratio conditions.

[0006] The technical problem to be solved by this invention is achieved by the following technical solution:

[0007] An active sonar signal identification method includes the following steps:

[0008] (1) Receive signal data through a hydrophone, calculate the power spectrum S of the signal, and estimate the center frequency fc and bandwidth bw of the signal;

[0009] (2) The signal data is preprocessed by orthogonal downconversion, downsampling and low-pass filtering to obtain preprocessed signal data;

[0010] (3) Based on the bandwidth bw estimated in step (1), preliminarily determine whether the signal is a single-frequency signal or a frequency-modulated signal; if it is a single-frequency signal, proceed to step (4); if it is a frequency-modulated signal, proceed to step (5).

[0011] (4) Further determine whether the signal is a single-frequency signal or noise or interference through power spectrum analysis, and the identification process ends;

[0012] (5) Calculate the delay autocorrelation spectrum R of the signal. f ;

[0013] (6) Determine the delayed autocorrelation spectrum R f If the signal is a single peak, proceed to step (7); otherwise, determine that the signal is noise interference and the identification process ends.

[0014] (7) The delayed autocorrelation spectrum R f Sort by intensity, take the highest intensity spectral line I, and sort them in ascending order of intensity, denoted as F = {f1, f2, ..., f...} I The corresponding spectral line intensities are denoted as P = {p1, p2, ..., p}. I};

[0015] (8) Calculate the intensity decay rate of each spectral line relative to the first spectral line, and obtain the decay rate U = {u1, u2, ..., u I-1 Record the index of the spectral line with the largest decay rate; if the index value is greater than 2, the signal is determined to be a hyperbolic frequency modulated signal and the identification process ends; if the index value is 1 or 2, then execute step (9);

[0016] (9) Calculate the ratio L of the maximum value of the delayed autocorrelation spectrum intensity p1 and the average value of the spectral line intensity in P; if L is greater than the threshold α, the signal is determined to be a linear frequency modulated signal; otherwise, the signal is a hyperbolic frequency modulated signal; the identification process ends.

[0017] Furthermore, the specific method of step (2) is as follows:

[0018] (201) Perform quadrature downconversion on the signal data to shift the signal to the baseband;

[0019] (202) Calculate the cutoff frequency C = (bw / 2) / (fs / 2), where fs is the sampling rate after the last downsampling; if C < 0.1, then the signal data is downsampled by a factor of 2;

[0020] (203) Repeat step (202) until C≥0.1;

[0021] (204) For signal data that satisfies C≥0.1 or downsampled signal data, perform low-pass filtering with C as the cutoff frequency.

[0022] Furthermore, the specific method of step (3) is as follows:

[0023] If the bandwidth bw is less than the threshold f x If the signal is initially determined to be a single-frequency signal, step (4) is executed; otherwise, the signal is initially determined to be a frequency-modulated signal. x The value ranges from 340 to 360 Hz.

[0024] Furthermore, the specific method of step (4) is as follows:

[0025] (401) For the preprocessed signal data X(n)=x r (n)+jx i (n), calculate the signal around the estimated center frequency fc in the range of 2N. p +1 frequency point energy and E p :

[0026]

[0027] Where, N m x is the index of the center frequency point. r (n) represents the real part of the signal, x i (n) represents the imaginary part of the signal;

[0028] (402) Calculate the total frequency domain energy E all :

[0029]

[0030] Where nfft is the number of points used in the Fast Fourier Transform;

[0031] (403) Calculate the decision statistic:

[0032]

[0033] If stat > 0.85, the signal is determined to be a single-frequency signal; otherwise, the signal is considered noise or interference, and the identification process ends.

[0034] Furthermore, the specific method of step (5) is as follows:

[0035] (501) For the preprocessed signal data X(n), multiply X(n) by its function time delay complex conjugate to obtain the signal delay autocorrelation function R. t (n,τ):

[0036] Rt (n,τ)=X(n)×X * (n-τ)

[0037] Where τ is the delay amount;

[0038] (502) Calculate the spectrum R of the delayed autocorrelation function. f (n,τ):

[0039] R f (n,τ)=DFT[R t (n,τ)]

[0040] Wherein, DFT stands for Discrete Fourier Transform.

[0041] Furthermore, the specific method of step (8) is as follows:

[0042] (801) For P = {p1, p2, ..., p...} I} Calculate the intensity decay rate of each spectral line relative to the first spectral line:

[0043] u i =(p1-p i+1 ) / i

[0044] Among them, i=1,2,...,I-1;

[0045] All the obtained decay rates are denoted as U = {u1, u2, ..., u I-1 Record the index of the spectral line with the highest decay rate, UmaxLoc;

[0046] (802) If the value of index UmaxLoc is greater than 2, the signal is determined to be a hyperbolic frequency modulation signal and the identification process ends; if the index is 1 or 2, then step (9) is executed.

[0047] Furthermore, the specific method of step (9) is as follows:

[0048] (901) For P = {p1, p2, ..., p...} I}, calculate the average intensity

[0049]

[0050] (902) Calculate the maximum intensity p1 and the average intensity of the delayed autocorrelation spectrum. The ratio L:

[0051]

[0052] (903) For the ratio L, if L is greater than the threshold α, the signal is determined to be a linear frequency modulation signal; otherwise, the signal is determined to be a hyperbolic frequency modulation signal, and the threshold α is 1 to 1.05; the recognition process ends.

[0053] The beneficial effects of this invention are as follows:

[0054] 1. This invention utilizes down-conversion, downsampling, and low-pass filtering to reduce data volume and filter out out-of-band noise. This reduces the impact of noise on active sonar signal recognition and also lowers the computational load of subsequent signal processing.

[0055] 2. Under low signal-to-noise ratio (SNR) conditions, there is no significant difference in the delayed autocorrelation (DAR) peaks of LFM and HFM signals. Compared to existing methods that simply compare the ratio of the maximum to the second-largest peak value to distinguish LFM and HFM signals, this invention provides a more detailed analysis and comparison of the DARCH peak attenuation of LFM and HFM signals. Based on a large amount of measured data, a model of the peak relationship is constructed, and the threshold value for signal recognition is corrected, achieving higher recognition accuracy under low SNR conditions. Attached Figure Description

[0056] Figure 1 This is a flowchart of the present invention;

[0057] Figure 2 This is a sequence diagram of the delayed autocorrelation spectrum peaks of an LFM signal under high signal-to-noise ratio conditions.

[0058] Figure 3 This is a sequence diagram of the delayed autocorrelation spectrum peaks of an HFM signal under high signal-to-noise ratio conditions.

[0059] Figure 4 This is a sequence diagram of the delayed autocorrelation spectrum peaks of an LFM signal under low signal-to-noise ratio conditions.

[0060] Figure 5 This is a sequence diagram of the delayed autocorrelation spectrum peaks of an HFM signal under low signal-to-noise ratio conditions. Detailed Implementation

[0061] The technical solution and effects of the present invention will be further described in detail below with reference to the accompanying drawings.

[0062] An active sonar signal identification method first calculates the power spectrum of pulse signal data received by a hydrophone to estimate its spectral center frequency and bandwidth, and preliminarily identifies it as a CW signal or a frequency-modulated signal (LFM, HFM) based on the bandwidth. For data initially identified as a CW signal, power spectrum analysis is used to further confirm whether the signal is indeed a CW signal. For data initially identified as a frequency-modulated signal, its instantaneous autocorrelation spectrum is calculated, and the signal is further identified based on the peak distribution of the autocorrelation spectrum.

[0063] The specific process of this method is as follows: Figure 1 As shown, the implementation steps are as follows:

[0064] Step 1: For the data received by the hydrophone, calculate its power spectrum S and estimate the center frequency fc and bandwidth bw of the signal.

[0065] Step 2: Preprocess the received data by orthogonal downconversion, downsampling, and low-pass filtering to obtain preprocessed signal data.

[0066] (201) The received data is orthogonally downconverted to shift the signal to the baseband.

[0067] (202) Cutoff frequency C = (bw / 2) / (fs / 2), where fs is the sampling rate after the previous downsampling. If C < 0.1, the received data is downsampled by a factor of 2.

[0068] (203) Iterate through step (202) until C≥0.1.

[0069] (204) The downsampled data is low-pass filtered at the C cutoff frequency.

[0070] Step 3: Based on the signal bandwidth bw estimated in step (1), preliminarily distinguish between CW signals and frequency modulation signals (LFM and HFM). If the signal is estimated to be a CW signal, proceed to step (4); if the signal is estimated to be a frequency modulation signal, proceed to step (5).

[0071] (301) If the estimated bandwidth of the signal bw is less than f x If the signal is initially estimated to be a CW signal, then step (4) is executed; if the estimated bandwidth bw is greater than or equal to f x Therefore, the initial estimate of the signal is a frequency-modulated signal (LFM and HFM). x 350Hz is usually selected.

[0072] Step 4: For data estimated to be CW signals, power spectrum analysis is used to determine whether the signal is indeed a CW signal, and the identification process ends.

[0073] (401) For the preprocessed signal X(n)=x r (n)+jx i (n), calculate the signal around the estimated center frequency fc in the range of 2N. p +1 frequency point energy and E p The index of the center frequency point is N. m ,

[0074]

[0075] (402) Calculate the total frequency domain energy E all ,

[0076]

[0077] (403) Calculate the decision statistic statistic.

[0078]

[0079] If stat > 0.85, the signal is considered a CW signal; otherwise, the signal is considered noise or interference, and the active sonar signal identification ends.

[0080] Step 5: For data estimated to be frequency-modulated signals, calculate the delay autocorrelation spectrum R. f .

[0081] (501) For the processed signal X(n), the signal is multiplied by its time-delay complex conjugate function to obtain the signal's delayed autocorrelation function:

[0082] R t (n,τ)=X(n)×X * (n-τ)

[0083] (502) Calculate the spectrum of the delayed autocorrelation function:

[0084] R f (n,τ)=DFT[R t (n,τ)]

[0085] Step 6: Determine whether the delayed autocorrelation spectrum is single-peaked. If the delayed autocorrelation spectrum has multiple peak groups or no obvious peak, the signal is determined to be noise interference, and the identification process ends; if the delayed autocorrelation is single-peaked, proceed to step (7).

[0086] Step 7: Sort the delayed autocorrelation spectra by intensity, take the I-line spectrum with the highest intensity, sort it in ascending order of intensity, and denote the set of frequency points of the resulting I-line spectrum as F = {f1, f2, ..., f...} I The corresponding set of line spectral intensities is denoted as P = {p1, p2, ..., p}. I}

[0087] Step 8: Calculate the intensity decay rate of each spectral line relative to the first spectral line, and obtain the decay rate set U = {u1, u2, ..., u...} I-1 Record the index of the spectral line with the largest decay rate. If the index value is greater than 2, the signal is considered to be an HFM signal and the identification process ends; if the index value is 1 or 2, proceed to step (9).

[0088] (801) For the obtained set of line spectrum intensities P = {p1, p2, ..., p...} I} Calculate the intensity decay rate of each spectral line relative to the first spectral line:

[0089] u i =(p1-p i+1 ) / i

[0090] The decay rate set U = {u1, u2, ..., u} is obtained. I-1 The index of the spectral line with the highest decay rate is recorded as UmaxLoc.

[0091] (802) If the value of index UmaxLoc is greater than 2, the signal is considered to be an HFM signal and the recognition process ends; if the index is 1 or 2, then step (9) is executed.

[0092] Step 9: Calculate the ratio L of the maximum value p1 of the delayed autocorrelation spectrum intensity to the average value of the line spectrum intensity set P. If L is greater than the threshold α, the signal is an LFM signal; if L is less than the threshold α, the signal is an HFM signal. The identification process ends.

[0093] (901) For the obtained set of line spectrum intensities P = {p1, p2, ..., p...} I}, calculate the average intensity p:

[0094]

[0095] (902) Calculate the maximum value p1 of the delayed autocorrelation spectrum intensity and the average value of the line spectrum intensity set P. The ratio L:

[0096]

[0097] (903) For the ratio L, if L is greater than the threshold α, the signal is an LFM signal; if L is less than the threshold α, the signal is an HFM signal. The threshold value α for frequency modulation signal identification is usually between 1 and 1.05. The identification process ends.

[0098] The effectiveness of this method can be demonstrated by the following test results:

[0099] 1. Data testing conditions

[0100] The simulated active sonar signal data collected in a certain experiment were selected. The experiment collected CW, LFM and HFM signal data respectively, and the method of the present invention was used for signal identification.

[0101] 2. Data Test Results

[0102] The time-frequency characteristics of CW signals are significantly different from those of LFM and HFM signals. Therefore, the identification process for CW signals will not be elaborated upon here. The identification of LFM and HFM signals will be described in detail below. Figures 2-5The figure shows a comparison of the delayed autocorrelation spectrum peak sequences of the LFM and HFM signals obtained from data processing. As can be seen from the figure, under low signal-to-noise ratio (SNR) conditions, the LFM signal autocorrelation spectrum has no obvious sharp peaks and is not significantly different from the HFM signal autocorrelation spectrum. The method of this invention further analyzes the autocorrelation spectra of LFM and HFM signals under low SNR conditions, summarizes the characteristics of the signal peak models, and further distinguishes the two signals.

[0103] The collected data were processed using the method of this invention and conventional methods, and the recognition results are shown in Table 1:

[0104] Table 1 compares the accuracy of active sonar signal recognition between the present invention and traditional methods in the data test.

[0105]

[0106] The results show that the recognition accuracy of the method of the present invention is significantly improved compared with the traditional method under low signal-to-noise ratio conditions.

[0107] In summary, this invention achieves rapid and accurate identification of active sonar signals. By utilizing down-conversion, downsampling, and low-pass filtering, the amount of data is reduced and out-of-band noise is filtered out, minimizing the impact of noise on active sonar signal identification and reducing the computational load of subsequent signal processing. For CW signals, power spectrum analysis is used to determine whether the signal is a CW signal. For LFM and HFM signals, a more detailed comparison of their delay autocorrelation peak attenuation is conducted. By analyzing a large amount of data to construct a model of the signal spectral peak distribution, the signal identification threshold is corrected, and higher identification accuracy is achieved under low signal-to-noise ratio conditions.

Claims

1. A method for active sonar signal identification, characterized in that, Includes the following steps: (1) Receive signal data through a hydrophone, calculate the power spectrum S of the signal, and estimate the center frequency fc and bandwidth bw of the signal; (2) The signal data is preprocessed by orthogonal downconversion, downsampling and low-pass filtering to obtain preprocessed signal data; (3) Based on the bandwidth bw estimated in step (1), make a preliminary judgment on whether the signal is a single-frequency signal or a frequency-modulated signal; If it is a single-frequency signal, proceed to step (4); if it is a frequency-modulated signal, proceed to step (5). (4) Further determine whether the signal is a single-frequency signal or noise or interference through power spectrum analysis, and the identification process ends; (5) Calculate the delay autocorrelation spectrum R of the signal. f ; (6) Determine the delayed autocorrelation spectrum R f If the signal is a single peak, proceed to step (7); otherwise, determine that the signal is noise interference and the identification process ends. (7) The delayed autocorrelation spectrum R f Sort by intensity, take the highest intensity spectral line I, and sort by intensity from smallest to largest, denoted as... The corresponding spectral line intensity is denoted as ; (8) Calculate the intensity decay rate of each spectral line relative to the first spectral line to obtain the decay rate. Record the index of the spectral line with the largest decay rate; if the index value is greater than 2, the signal is determined to be a hyperbolic frequency modulated signal, and the identification process ends; if the index value is 1 or 2, then proceed to step (9); the specific method is as follows: (801) For Calculate the intensity decay rate of each spectral line relative to the first spectral line: Among them, i=1,2,...,I-1; All the decay rates obtained are denoted as Record the index of the spectral line with the highest decay rate, UmaxLoc; (802) If the value of index UmaxLoc is greater than 2, the signal is determined to be a hyperbolic frequency modulation signal and the identification process ends; if the index is 1 or 2, then step (9) is executed. (9) Calculate the ratio L of the maximum value of the delayed autocorrelation spectrum intensity p1 to the average value of the spectral line intensities in P; if L is greater than the threshold If the signal is positive, it is determined to be a linear frequency modulated (LFM) signal; otherwise, it is a hyperbolic frequency modulated (HFM) signal; the identification process ends. The specific method is as follows: (901) For Calculate the average strength : (902) Calculate the maximum intensity p1 and the average intensity of the delayed autocorrelation spectrum. The ratio L: (903) For the ratio L, if L is greater than the threshold If the threshold is met, the signal is determined to be a linear frequency modulated (LFM) signal; otherwise, the signal is determined to be a hyperbolic frequency modulated (HFM) signal. The value ranges from 1 to 1.05; the recognition process ends.

2. The active sonar signal identification method according to claim 1, characterized in that, The specific method for step (2) is as follows: (201) Perform quadrature downconversion on the signal data to shift the signal to the baseband; (202) Calculate the cutoff frequency Where fs is the sampling rate after the previous downsampling; if C < 0.1, the signal data is downsampled by a factor of 2; (203) Repeat step (202) until C≥0.1; (204) For signal data that satisfies C≥0.1 or downsampled signal data, perform low-pass filtering with C as the cutoff frequency.

3. The active sonar signal identification method according to claim 1, characterized in that, The specific method for step (3) is as follows: If the bandwidth bw is less than the threshold f x If the signal is initially determined to be a single-frequency signal, then step (4) is executed; otherwise, the signal is initially determined to be a frequency-modulated signal. x The value ranges from 340 to 360 Hz.

4. The active sonar signal identification method according to claim 1, characterized in that, The specific method for step (4) is as follows: (401) For preprocessed signal data Calculate the signal around the estimated center frequency fc within 2N. p +1 frequency point energy and E p : Where, N m The index of the center frequency point, Let be the real part of the signal. This represents the imaginary part of the signal; (402) Calculate the total frequency domain energy E all : Where nfft is the number of points used in the Fast Fourier Transform; (403) Calculate the decision statistic: If stat > 0.85, the signal is determined to be a single-frequency signal; otherwise, the signal is considered noise or interference, and the identification process ends.

5. The active sonar signal identification method according to claim 1, characterized in that, The specific method for step (5) is as follows: (501) For preprocessed signal data ,Will Multiplying it by its time-delay complex conjugate yields the signal's delayed autocorrelation function. : in, This is the delay amount; (502) Calculate the spectrum of the delayed autocorrelation function. : Wherein, DFT stands for Discrete Fourier Transform.

Images