Active sonar pulse signal detection method

Abstract

The invention discloses an active sonar pulse signal detection method. The method comprises the following steps: (1) preprocessing: completing operations such as time segmentation and power spectrum calculation on acquired data; (2) calculating fluctuation characteristics by using the power spectrum parameters, and arranging fluctuation characteristic parameters of each section according to a descending order; (3) estimating whether a pulse signal exists or not, the number of segments of the pulse signal and a segment sequence by using an MDL criterion; and (4) if the pulse signal exists, filtering the segmented signal, detecting the pulse width on the time domain based on a threshold method, and estimating the center frequency and the bandwidth on the frequency domain based on a 3-sigma threshold. The active sonar pulse signal can be effectively detected, and the method is still available under the condition that the signal-to-noise ratio is low. According to the method, the detection threshold setting can be avoided, so that the false alarm caused by the change of the noise level is avoided.

Description

Technical Field

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

[0002] Underwater signal detection is a primary method for human exploration of the marine environment. Existing signal detection methods mainly rely on energy detectors to measure signal energy and determine its presence. When background noise follows a Gaussian distribution, traditional energy detectors perform well. However, in the marine environment, some unknown sonars often emit pulse signals. Effective detection of these signals can be used to estimate the characteristics and location of unknown sonars, enabling early warning of potential threats. In the presence of marine noise interference, the detection performance of traditional energy detectors significantly degrades, or even completely fails, due to their sensitivity to pulse noise. The presence of pulse noise makes it impossible for traditional signal detection methods to distinguish between signal and noise, thus failing to accurately identify the presence of useful signals. Therefore, existing signal detection technologies cannot meet the requirements for pulse signal detection in the marine environment. Summary of the Invention

[0003] This invention proposes an active sonar pulse signal detection method to address the aforementioned problems. Its core principle is to utilize the MDL criterion to determine the presence of a pulse signal in a segmented time-series signal, and to estimate the time-domain and frequency-domain parameters of the pulse signal using smoothed filtering edge detection and a 3-σ threshold. This method can achieve good detection results even with low signal-to-noise ratios, significantly improving the capability of active pulse signal detection.

[0004] This invention discloses an active sonar pulse signal detection method, which includes the following steps:

[0005] Step 1: Preprocess the collected data;

[0006] Preprocessing includes segmenting each frame of data and calculating the power spectrum of each segment;

[0007] Step 2: Calculate the fluctuation characteristics of the signal segment using the standard deviation of the power spectrum, and sort the fluctuation characteristic parameters of each segment in descending order;

[0008] Step 3: Use the minimum description length criterion to estimate whether a pulse signal exists, the number of segments with pulse signals, and the segment sequence.

[0009] Step 4: If a pulse signal exists, filter the segmented signal and detect the pulse width in the time domain based on the threshold method, and estimate the center frequency and bandwidth in the power spectrum based on the 3-σ threshold.

[0010] Furthermore, in step 1, after obtaining a frame of data, this frame of data is divided into several segments at equal intervals, with M segments and N sampling points in each segment. Then, the total number of sampling points in this frame is L = MN.

[0011] Furthermore, in step 1, the power spectral density of each segment is calculated:

[0012]

[0013] In the formula, w(n) is the window function of the Fourier transform, x(n) is the acquired digital signal, and ω k Where is the digital frequency, N is the number of sampling points, k is the output sequence index, and n is the input sequence index.

[0014] Furthermore, in step 2, the standard deviation (std) of the power spectrum is calculated, and the standard deviation is used as the fluctuation characteristic of each segment:

[0015] P fluc =std

[0016] P fluc For segmented fluctuation characteristics;

[0017] The above fluctuation characteristic parameters are arranged in descending order.

[0018] Furthermore, in step 3, the MDL criterion is used to calculate:

[0019]

[0020] Where m is the number of segments of the signal to be estimated, M is the total number of segments, K is the number of samples on the power spectrum of each time segment, and Λ(m) is the likelihood function;

[0021] The likelihood function Λ(m) is:

[0022]

[0023] In the formula That is, the fluctuation characteristics of the i-th segment calculated above.

[0024] Furthermore, in step 3, the value of m that minimizes MDL(m) is selected as the estimate of the number of segments with pulse signals. When m = 0, it indicates that there are no pulse signals in the current frame; otherwise, it indicates that there are pulse signals in the current frame.

[0025] Furthermore, in step 3, when there are discontinuous frame numbers of pulse signals, the discontinuous frames are grouped separately, and the continuous frames are grouped into the same pulse group; and the signal parameters of each group are analyzed separately.

[0026] Furthermore, step 4 also includes the following steps:

[0027] Step 41: Obtain the pulse width in the time domain;

[0028] If a pulse signal exists, filter it in the time domain and estimate its time-average power using the following formula:

[0029]

[0030] In the formula, x(n) is the received signal, W is the width of the smoothing window, and P... q This is a smoothed sequence of power changes over time. Let the number of points in the sequence be N, then...

[0031] Let the threshold for detecting pulse width on the time-domain power sequence be:

[0032] T = α·mean(P) q )+β·std(P q )+ξ·max(P q )+ζ·min(P q )

[0033] In the formula, α, β, ξ, ζ are weighting coefficients, mean() represents the mean, std() represents the standard deviation, max() represents the maximum value, and min() represents the minimum value;

[0034] The point where the power spectrum first reaches this threshold is considered the starting point of the pulse, and the second intersection of the power spectrum curve and the threshold is the ending point of the pulse. The pulse width is obtained by subtracting the starting point from the ending point.

[0035] Furthermore, step 42 also includes the following steps:

[0036] Step 42, estimate the center frequency and bandwidth based on the 3-σ threshold on the power spectrum, with the following sub-steps:

[0037] Step 421: Perform power spectrum estimation on the complete pulse signal;

[0038] Step 422: Search for all line spectra that cross the threshold and estimate the center frequency and bandwidth;

[0039] Step 423: Lower the threshold value, re-search for all line spectra that exceed the threshold, and estimate the center frequency and bandwidth again;

[0040] Step 424: If the estimated signal bandwidth and center frequency are within the allowable error range in the two estimations, then the estimation is considered to have obtained accurate estimates of the center frequency and bandwidth and the estimation stops; otherwise, continue to execute steps 422 to 424.

[0041] Furthermore, in step 421, the power spectrum estimation results are used to perform line spectrum detection using the 3-σ threshold detection criterion:

[0042] T PSD =mean(PSD(ω))+3×std(PSD(ω))

[0043] PSD(ω) is the power spectral density.

[0044] The beneficial effects achieved by this invention are:

[0045] This invention provides a method for detecting active sonar pulse signals, which can effectively detect active sonar pulse signals and remains usable even under low signal-to-noise ratio conditions. This method avoids setting a detection threshold, thereby preventing false alarms caused by changes in the noise level itself.

[0046] This invention provides an active sonar pulse signal detection method that can effectively detect pulse signals under low signal-to-noise ratio (SNR) conditions. For example, for typical pulse signals of several hundred milliseconds, the detection rate can reach over 90% even when the SNR is below -3dB.

[0047] This invention provides an active sonar pulse signal detection method that can dynamically adjust various thresholds in the algorithm, reducing the uncertainty caused by human intervention.

[0048] This invention provides an active sonar pulse signal detection method that can adapt to pulses of varying widths. When the pulse width is large, it can automatically connect multiple segments for detection. If each segment is processed independently, a wide pulse signal can easily be split into multiple narrow pulse signals, resulting in erroneous detection results. This application combines continuous segments of pulse signals for detection, resulting in detection results that are closer to the original, true signal, effectively improving the detection rate. Attached Figure Description

[0049] Figure 1 Flowchart of an active sonar pulse signal detection method;

[0050] Figure 2 Time-domain waveform of a single-frame simulated signal;

[0051] Figure 3 Piecewise power spectrum curves;

[0052] Figure 4 Standard deviation of power spectrum in each segment;

[0053] Figure 5 16 sets of MDL values;

[0054] Figure 6 Comparison of pulse signal before and after filtering;

[0055] Figure 7 Center frequency and bandwidth parameter extraction diagram. Detailed Implementation

[0056] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0057] like Figure 1 As shown, the present invention provides an active sonar pulse signal detection method, comprising the following steps:

[0058] Step 1, Preprocessing: Perform time segmentation and power spectrum calculation on the collected data;

[0059] Step 2, calculate fluctuation characteristics; use the standard deviation of the power spectrum to calculate the fluctuation characteristics of the signal segment, and arrange the fluctuation characteristic parameters of each segment in descending order;

[0060] Step 3: Use the MDL criterion to estimate whether a pulse signal exists, the number of segments with pulse signals, and the segment sequence.

[0061] Step 4: If a pulse signal exists, filter the segmented signal and detect the pulse width in the time domain based on the threshold method, and estimate the center frequency and bandwidth in the power spectrum based on the 3-σ threshold.

[0062] The detailed steps are as follows.

[0063] Step 1, Preprocessing

[0064] The received signal enters the detector as a data stream, which is theoretically infinitely long. For ease of processing, it is artificially divided into "frames". Each frame of data includes several sampling points, and the number of sampling points is denoted as L.

[0065] After obtaining a frame of data, in order to facilitate the subsequent application of the Minimum Description Length (MDL) criterion, this frame of data is divided into several "segments" at equal intervals. Let the number of segments be M and the number of sampling points in each segment be N. Then the total number of sampling points in this frame is L = MN.

[0066] The power spectrum of each segment is calculated using the following formula:

[0067]

[0068] In the formula, w(n) is the window function of the Fourier transform, x(n) is the acquired digital signal, and ω kLet N be the digital frequency, n be the number of sampling points, n be the input sequence index, and k be the output sequence index; where k ∈ K, and K is the number of points in the power spectrum of each signal segment. Generally, the number of points in the power spectrum can be kept consistent with the number of points in the time-domain signal, i.e., K = N, or... in This indicates rounding up to the nearest integer.

[0069] Step 2, Calculate the fluctuation characteristics

[0070] The method for calculating the fluctuation characteristics of each signal segment is as follows: the standard deviation (std) of the statistical power spectrum is used as the fluctuation characteristic of each segment, i.e.:

[0071] P fluc =std

[0072] P fluc These are the segmented fluctuation characteristics.

[0073] The above fluctuation characteristic parameters are arranged in descending order.

[0074] Step 3: Use the Minimum Description Length (MDL) criterion to determine whether there is a pulse signal in the frame and the number of segments with pulse signals.

[0075] The MDL value is calculated using the MDL criterion as follows:

[0076]

[0077] Where m is the number of segments (degrees of freedom, m = 0, 1, ..., M-1) of the signal to be estimated, M is the total number of segments, K is the number of samples on the power spectrum of each time segment, and Λ(m) is the likelihood function;

[0078] The likelihood function Λ(m) is:

[0079]

[0080] In the formula That is, the fluctuation characteristics of the i-th segment calculated above.

[0081] The value of m that minimizes MDL(m) is selected as the estimate of the number of segments with pulse signals.

[0082]

[0083] when This indicates that there is no pulse signal in the current frame. It also allows us to determine the segment number where a pulse signal exists, and to identify the fluctuation feature P. flucSorted in descending order, the segment numbers containing the first m maximum values ​​are the segments containing pulse signals. Segment numbers that are consecutive belong to the same group. When the segment numbers are not consecutive, they can be divided into multiple pulse groups, and their signal parameters can be analyzed separately to refine the analysis of signals with short pulse widths and repetition periods.

[0084] Because the pulse width of sonar pulse signals cannot be predicted, if a segment containing a pulse signal is detected, processing each segment independently can easily split a wide pulse signal into multiple narrow pulse signals, resulting in erroneous detections. However, if consecutive segments containing pulse signals are detected together, the detection result will be closer to the original, true signal.

[0085] Step 4, estimate pulse parameters

[0086] Specifically, step 4 also includes the following steps:

[0087] Step 41: Obtain the pulse width in the time domain;

[0088] If a pulse signal exists, filter it in the time domain and estimate its time-average power using the following formula:

[0089]

[0090] In the formula, x(n) is the received signal, W is the width of the smoothing window, and P... q This is a smoothed sequence of power changes over time. Let the number of points in the sequence be N, then...

[0091] Let the threshold for detecting pulse width on the time-domain power sequence be:

[0092] T = α·mean(P) q )+β·std(P q )+ξ·max(P q )+ζ·min(P q )

[0093] In the formula, α, β, ξ, ζ are weighting coefficients, mean() represents the mean, std() represents the standard deviation, max() represents the maximum value, and min() represents the minimum value.

[0094] After setting the pulse detection threshold, the upper and lower edges of the pulse are determined according to the threshold criterion. The pulse width is then calculated based on the complete pulse signal segments detected from adjacent upper and lower edges.

[0095] Specifically, according to time-average power After obtaining the power spectrum of the signal, it can be understood as obtaining the envelope of the signal (e.g., Figure 6As shown, the power spectrum of the signal is close to its envelope. Using statistical measures of the power spectrum, including mean, standard deviation, maximum, and minimum values, a weighted sum is obtained to reach a threshold. The point where the power spectrum first reaches this threshold from left to right is considered the start of the pulse. Afterward, the power spectrum value remains above the threshold. The second intersection of the power spectrum curve and the threshold to the right is the end of the pulse. Subtracting the start time from the end time gives the pulse width.

[0096] Step 42, estimate the center frequency and bandwidth based on the 3-σ threshold on the power spectrum, with the following sub-steps:

[0097] Step 421: Perform power spectrum estimation on the complete pulse signal. The power spectrum estimation result is then used for line spectrum detection using a 3-σ threshold detection criterion. The 3-σ detection threshold is:

[0098] T PSD =mean(PSD(ω))+3×std(PSD(ω))

[0099] PSD(ω) is the power spectral density.

[0100] Step 422: Search for all line spectra that cross the threshold and estimate the center frequency and bandwidth.

[0101] Step 423: Lower the threshold value, re-search for all line spectra that exceed the threshold, and estimate the center frequency and bandwidth again.

[0102] In this embodiment, the threshold value is lowered by 0.9 times the initial threshold, and the center frequency and bandwidth are estimated again.

[0103] In step 424, if the estimated signal bandwidth and center frequency obtained in both steps are within the allowable error range (set appropriately according to the actual parameters), then the estimation is considered to have obtained accurate estimates of the center frequency and bandwidth, and the estimation stops. Otherwise, continue with steps 422 to 424.

[0104] Taking an exemplary pulse signal detection process as an example, the signal simulation parameters for this process are as follows: the signal type is a single-frequency continuous wave pulse signal (CW), the signal center frequency is 2kHz, the pulse period is 0.1s, the signal duty cycle is 40%, and the signal-to-noise ratio is 6dB. During the detection process, the number of signal processing points per frame is 16384, the sampling frequency is 32kHz, and the number of signal points per segment is 1024.

[0105] Using the signal parameters from the exemplary pulse signal detection process described above, a simulated signal frame is generated. The simulated signal is then processed using the method of this invention.

[0106] The present invention will now be described in detail with reference to a specific embodiment.

[0107] Simulation generates the signal to be detected, such as Figure 2 As shown, the signal length L = 16384 points.

[0108] In step (1), the above single-frame signal is divided into M = 16 segments of length N = 1024 points. Figure 2 The diagram is separated by vertical lines. The power spectrum is calculated for each segment, as shown below. Figure 3 As shown in the figure, the power spectrum curves of segment 1 and segment 3 are displayed respectively.

[0109] In step (2), for the power spectrum curves of the 16 segments, the standard deviation std of the power spectrum of each segment is calculated, and then the standard deviation is used as the fluctuation characteristic P of each segment. fluc ,like Figure 4 As shown, the fluctuation characteristic parameters of each segment are then arranged in descending order.

[0110] In step (3), the MDL value is calculated using the MDL criterion, and the m value that minimizes MDL(m) is selected as the estimate of the number of segments with pulse signals, such as... Figure 5 As shown, the minimum value of MDL is located at m=12, indicating that 11 signal pulses were detected, namely segments 1#, 2#, 4#, 5#, 7#, 8#, 10#, 11#, 13#, 14#, and 16#. These segments are not continuous and can be divided into 6 groups: Group 1: segments 1# and 2#; Group 2: segments 4# and 5#; Group 3: segments 7# and 8#; Group 4: segments 10# and 11#; Group 5: segments 13# and 14#; and Group 6: segment 16#.

[0111] In step (4), for each group of continuous pulse signal segments, filtering is performed in the time domain to estimate their time-average power. Figure 6 The graph shows the segmented time-average power curve for signal #2. It's evident from the graph that the filtered signal envelope is significantly larger. The threshold is calculated using the pulse width threshold formula, allowing for judgment... Figure 6 The 2# segment has a lower edge, and the lower edge is at point 355. Therefore, the pulse width of the 2# segment can be calculated as 355 / 32e3≈11ms. Adding the pulse width of the 1# segment in the same group as 1024 / 32e3≈32ms, the pulse width of the entire signal in the first group is 43ms.

[0112] Power spectrum analysis was performed on the continuous segments of each group of pulse signals, and the center frequency and bandwidth were estimated based on the 3-σ threshold, such as... Figure 7As shown, only the three adjacent frequencies centered at 2kHz are included when the threshold is exceeded. When 3-σ is reduced to 0.9 times the original threshold value, the three adjacent frequencies centered at 2kHz are also included. Therefore, the center frequency of the output signal is 2kHz and the bandwidth is 2×32e3 / 2048≈31Hz.

[0113] The above are merely specific steps of the present invention and do not constitute any limitation on the scope of protection of the present invention; all technical solutions formed by equivalent transformation or equivalent substitution fall within the scope of protection of the present invention; the parts of the present invention not described in detail are common knowledge to those skilled in the art.

Claims

1. A method for detecting active sonar pulse signals, characterized in that, The active sonar pulse signal detection method includes the following steps: Step 1: Preprocess the collected data; Preprocessing includes segmenting each frame of data and calculating the power spectrum of each segment; Step 2: Calculate the fluctuation characteristics of the signal segment using the standard deviation of the power spectrum, and sort the fluctuation characteristic parameters of each segment in descending order; Step 3: Use the minimum description length criterion to estimate whether a pulse signal exists, the number of segments with pulse signals, and the segment sequence. Step 4: If a pulse signal exists, filter the segmented signal and detect the pulse width in the time domain based on the threshold method, and estimate the center frequency and bandwidth in the power spectrum based on the 3-σ threshold.

2. The active sonar pulse signal detection method according to claim 1, characterized in that, In step 1, after obtaining a frame of data, the frame of data is divided into several segments at equal intervals, with M segments and N sampling points in each segment. Then the total number of sampling points in the frame is L = MN.

3. The active sonar pulse signal detection method according to claim 2, characterized in that, In step 1, the power spectral density of each segment is calculated: In the formula, w(n) is the window function of the Fourier transform, x(n) is the acquired digital signal, and ω k Where is the digital frequency, N is the number of sampling points, k is the output sequence index, and n is the input sequence index.

4. The active sonar pulse signal detection method according to claim 1, characterized in that, In step 2, the standard deviation (std) of the power spectrum is calculated, and the standard deviation is used as the fluctuation characteristic of each segment: P fluc =std P fluc For segmented fluctuation characteristics; The above fluctuation characteristic parameters are arranged in descending order.

5. The active sonar pulse signal detection method according to claim 1, characterized in that, In step 3, the MDL criterion is used to calculate: Where m is the number of segments of the signal to be estimated, M is the total number of segments, K is the number of samples on the power spectrum of each time segment, and Λ(m) is the likelihood function; The likelihood function Λ(m) is: In the formula That is, the fluctuation characteristics of the i-th segment calculated above.

6. The active sonar pulse signal detection method according to claim 5, characterized in that, In step 3, the value of m that minimizes MDL(m) is selected as the estimate of the number of segments with pulse signals. When m = 0, it indicates that there are no pulse signals in the current frame; otherwise, it indicates that there are pulse signals in the current frame.

7. The active sonar pulse signal detection method according to claim 5, characterized in that, In step 3, when there are discontinuous frame numbers of pulse signals, the discontinuous frames are grouped separately, and the continuous frames are grouped into the same pulse group. The signal parameters of each group were analyzed separately.

8. The active sonar pulse signal detection method according to claim 1, characterized in that, Step 4 also includes the following steps: Step 41: Obtain the pulse width in the time domain; If a pulse signal exists, filter it in the time domain and estimate its time-average power using the following formula: In the formula, x(n) is the received signal, W is the width of the smoothing window, and P... q Let N be the smoothed power sequence over time. Let the threshold for detecting pulse width on the time-domain power sequence be: T=α·mean(P q 0+β·std(P q )+ξ·max(P q )+ζ·min(P q 0 In the formula, α, β, ξ, ζ are weighting coefficients, mean() represents the mean, std() represents the standard deviation, max() represents the maximum value, and min() represents the minimum value; The point where the power spectrum first reaches this threshold is considered the starting point of the pulse, and the second intersection of the power spectrum curve and the threshold is the ending point of the pulse. The pulse width is obtained by subtracting the starting point from the ending point.

9. The active sonar pulse signal detection method according to claim 1, characterized in that, Step 42 also includes the following steps: Step 42, estimate the center frequency and bandwidth based on the 3-σ threshold on the power spectrum, with the following sub-steps: Step 421: Perform power spectrum estimation on the complete pulse signal; Step 422: Search for all line spectra that cross the threshold and estimate the center frequency and bandwidth; Step 423: Lower the threshold value, re-search for all line spectra that exceed the threshold, and estimate the center frequency and bandwidth again; Step 424: If the estimated signal bandwidth and center frequency are within the allowable error range in the two estimations, then the estimation is considered to have obtained accurate estimates of the center frequency and bandwidth and the estimation stops; otherwise, continue to execute steps 422 to 424.

10. The active sonar pulse signal detection method according to claim 8, characterized in that, In step 421, the power spectrum estimation results are used to perform line spectrum detection using the 3-σ threshold detection criterion: T PSD =mean(PSD(ω))+3×std(PSD(ω)) PSD(ω) is the power spectral density.

Images