CN114152329B - A method for peak detection of underwater acoustic signal spectrum - Google Patents

Abstract

The invention discloses an underwater acoustic signal spectrum peak detection method. The signal waveform including the spectrum peak is depolarized to obtain B(n), the gain coefficient c(n) is calculated, c(n) is multiplied by B(n) to obtain an output signal C(n), the median filtering and smoothing of C(n) is obtained to obtain D(n), the Gaussian window is used to fit D(n) to obtain E(n), and the maximum point and the corresponding maximum value are extracted from E(n). The present invention can detect the spectral peaks existing in the signal waveform through the steps of absolute value depolarization, dual sliding window spectral peak enhancement, moving average filtering, Gaussian fitting, etc., effectively reducing the influence of the underwater acoustic channel on the original spectral peaks, and realizing the enhancement of spectral peak characteristics.

Description

A method for peak detection of underwater acoustic signal spectrum

technical field

The invention belongs to the technical field of feature extraction of underwater acoustic signals, and relates to a method for detecting spectrum peaks of underwater acoustic signals.

Background technique

When performing underwater acoustic signal identification or parameter estimation, it is often necessary to detect spectral peaks. For example, when identifying underwater acoustic frequency shift keying (FSK), the spectral peaks of the spectrum will be detected. When identifying underwater acoustic orthogonal frequency division multiplexing (OFDM) signals, the autocorrelation spectral peaks will be detected. For communication signals such as FSK, phase shift keying (PSK) and direct sequence spread spectrum (DSSS), it is necessary to detect the position of the spectral peak in the frequency domain.

Most of the spectral peak detection methods of signals in underwater acoustics come from radio. There are many classic spectral peak extraction methods in radio, such as direct spectral peak detection algorithm, three-point spectral peak detection algorithm, Gaussian fitting algorithm, polynomial fitting algorithm, etc. Although direct spectral peak detection has low computational complexity, it has poor anti-noise performance. Three-point spectral peak detection, Gaussian fitting, and polynomial fitting algorithms have high requirements on the shape of spectral peaks. Compared with common wireless channels, underwater acoustic channels are more complex and have complex time-space-frequency characteristics. In the frequency domain, they exhibit frequency-selective fading. When the signal-to-noise ratio is low or the channel time becomes obvious, the spectral peaks obtained by correlation will also become unstable. Therefore, it is very challenging to use the signal frequency domain or related waveforms to detect spectral peaks in underwater acoustic channels.

Contents of the invention

In view of the above-mentioned prior art, the technical problem to be solved by the present invention is to provide a method for detecting spectral peaks of underwater acoustic signals with inconspicuous spectral peak characteristics in underwater acoustic channels, which can effectively enhance spectral peak characteristics and suppress interference.

In order to solve the above-mentioned technical problems, a kind of underwater acoustic signal spectral peak detection method of the present invention comprises the following steps:

S1, represent the signal waveform that comprises spectrum peak with A (n), n is the position of the n sampling point of signal waveform, n=1,2,..., N, N is the number of sampling points, when spectrum peak waveform has negative polarity, make B(n)=|A(n)|, when spectrum peak waveform does not have negative polarity, make B(n)=A(n);

S2, taking the nth sampling point of B(n) as the center, establishing two sliding windows, the length of the short sliding window is 2s _win +1, the length of the long sliding window is 2l _win +1, and l _win =2s _win ; the gain coefficient c(n) is obtained by the energy difference of the two sliding windows, is the short sliding window energy of the nth sampling point, /> is the long sliding window energy of the nth sampling point, where /> The output signal is C(n)=B(n)·c(n), where n=1,2,...,N, when the value of c(n) is negative, ns _win ≤0, nl _win ≤0, n+s _win >N or n+l _win >N, let c(n)=0;

S3, the C (n) waveform obtained in S2 is carried out moving average filter, the output signal after moving average filter is: Where 2P+1 is the moving average filter length, n=1,2,...,N, when nP≤0 or n+P>N, let D(n)=C(n);

S4, do Gaussian fitting to D(n) obtained in S3, Gaussian window function is: Where L is the length of the Gaussian window function, -(L-1)/2≤n≤(L-1)/2, α is the width coefficient of the Gaussian window function, and α is inversely proportional to the standard deviation of the Gaussian function σ=(L-1)/2α; the output signal after Gaussian fitting is: /> Where L=2Q+1 is the length of the Gaussian window function, n=1,2,...,N, and then normalize E(n) to obtain the optimized spectrum peak waveform, when nQ≤0 or n+Q>N, let E(n)=D(n);

S5, extract all maximum values and corresponding maximum value points in E(n), and sort according to maximum value from large to small, if it is a single spectral peak, then extract the largest maximum value and corresponding maximum value points in E(n) as spectral peaks, if there are multiple spectral peaks, then extract the first m maximum values and corresponding maximum value points in E(n) as spectral peaks, m equals the number of spectral peaks.

Further, the moving average filter length in S3 is equal to the short sliding window length in S2.

Further, the length of the Gaussian window function in S4 is equal to the length of the short sliding window in S2.

Further, the length of the short sliding window in S2, the length of the moving average filter in S3 and the length of the Gaussian window function in S4 are equal.

Beneficial effects of the present invention: the present invention can detect the spectral peaks existing in the signal waveform through the steps of absolute value depolarization, dual sliding window spectral peak enhancement, moving average filtering, Gaussian fitting, etc., effectively reducing the influence of the underwater acoustic channel on the original spectral peaks, and realizing the enhancement of spectral peak characteristics. The present invention mainly has the application of two aspects:

In the first aspect, we can use the present invention to extract spectral peak features of non-cooperative underwater acoustic communication signals, and realize identification of non-cooperative underwater acoustic communication signals through these features. For example, underwater acoustic OFDM generally has a cyclic prefix structure. We perform spectral peak detection on the autocorrelation waveform of the signal to determine whether there are two correlation peaks. Another example is that the underwater acoustic FSK signal has multiple spectral peaks in the frequency domain. We perform spectral peak detection on the frequency domain waveform of the signal to determine how many spectral peaks exist, and determine the modulation order of FSK according to the number of detected spectral peaks.

In the second aspect, we can use the present invention to estimate each carrier frequency of the underwater acoustic FSK signal. Using the present invention to perform spectrum peak detection will finally obtain the maximum point and the corresponding maximum value, and the maximum point corresponds to the frequency of the signal in the frequency domain.

Description of drawings

Fig. 1 is the flowchart of the underwater acoustic signal spectrum peak extraction method provided by the present invention;

Fig. 2 is the channel impulse response diagram of testing provided by the present invention;

Fig. 3 is the original frequency-domain waveform obtained by Fourier transform of simulated 8FSK;

Fig. 4 obtains the gain coefficient on the basis of the original frequency domain waveform in Fig. 3;

Fig. 5 is the result of multiplying the gain coefficient of Fig. 4 and the original frequency domain waveform of Fig. 3;

Figure 6 is the result of moving average filtering to the waveform in Figure 5;

Figure 7 is the result of Gaussian fitting to the waveform in Figure 6;

FIG. 8 is the result of limiting the abscissa in FIG. 7 within 10 kHz to 14 kHz.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

In conjunction with Fig. 1, the present invention comprises the following steps:

S1. Use A(n) to represent the signal waveform including the spectral peak, and n is the position of the nth sampling point of the signal waveform. In order to eliminate the uncertainty of the bipolar signal, we first take the absolute value of A(n) to obtain B(n), that is, B(n)=|A(n)|, when there is no negative polarity in the spectral peak waveform, let B(n)=A(n);

S2. Perform the following operations one by one for all sampling points: take the sampling point n of B(n) as the center, establish two sliding windows, one long and one short, the length of the short sliding window is 2s _win +1, the length of the long sliding window is 2l _win +1, and l _win =2s _win . The gain coefficient c(n) is obtained by the energy difference of the two sliding windows, and c(n) can be expressed as in When the value of c(n) is negative, ns _win ≤0, nl _win ≤0, n+s _win >N or n+l _win >N, set c(n)=0, and the output signal is C(n)=B(n)·c(n). By multiplying the gain coefficient c(n), when there is a peak in the signal waveform, the waveform at the peak is amplified, and the waveform at other places is suppressed;

S3. Carry out the following operations one by one for all the sampling points: perform moving average filtering on the C(n) waveform obtained in S2. This step can reduce random noise and roughly obtain the envelope shape of the spectral peak waveform. The output signal after moving average filtering can be expressed as Where 2P+1 is the length of the moving average filter, when nP≤0 or n+P>N, let D(n)=C(n);

S4. Carry out the following operations one by one for all sampling points: finally, Gaussian fitting is done to D(n) obtained in S3. This step can effectively smooth D(n), which is beneficial to the extraction of maximum value and spectral peak detection in D(n). The Gaussian window function can be expressed as Where L is the length of the Gaussian window function, -(L-1)/2≤n≤(L-1)/2, α is the width coefficient of the Gaussian window function, and α is inversely proportional to the standard deviation of the Gaussian function σ=(L-1)/2α. The output signal after Gaussian fitting can be expressed as /> Where L=2Q+1 is the length of the Gaussian window function, when nQ≤0 or n+Q>N, let E(n)=D(n), and then normalize E(n) to obtain the optimized spectrum peak waveform;

S5. Extract all maximum values and corresponding maximum value points in E(n), and sort them according to the size of the maximum values from large to small. If it is a single spectral peak, extract the largest maximum value and corresponding maximum value points in E(n) as spectral peaks, if there are multiple spectral peaks, extract the first m maximum values and corresponding maximum value points in E(n) as spectral peaks, and m is equal to the number of spectral peaks.

The spectrum peak refers to the frequency domain spectrum peak or the signal correlation spectrum peak. The bipolar signal in S1 refers to the situation where the signal waveform has negative polarity caused by correlation and other operations. In some cases, there will be no negative polarity. In addition, if the length of the sliding window is too short, the ability to suppress interference will be weakened. In practical application, it can be set according to experience value. If the length of the moving average filter in S3 is too short, it will easily form a large number of false envelopes, and if the length is too long, it will contain a lot of useless information. In order to ensure that the envelope of the spectral peak can be effectively extracted, the length of the moving average filter can be set to be consistent with the short sliding window length in S2. The length of the Gaussian window function in S4 can be consistent with the short sliding window in S2 and the moving average filter length in S3.

Embodiment is provided below in conjunction with specific parameters:

Combined with Figure 1, assume that the collected signal is an 8FSK signal, the sampling rate is 48kHz, the bandwidth of the 8FSK signal is 4kHz, and its carrier frequency is {10.25kHz, 10.75kHz, 11.25kHz, 11.75kHz, 12.25kHz, 12.75kHz, 13.25kHz, 13.75kHz}. The channel impulse response is shown in Figure 2. We first perform Fourier transform on the collected 8FSK signal to obtain the signal waveform A(n) including the spectral peak, and the signal waveform of A(n) is shown in Figure 3. The specific implementation of the present invention comprises the following steps:

S1. Since there is no negative polarity in A(n), the absolute value depolarization can be omitted, that is, B(n)=A(n);

S2. Set the length of the short sliding window as the length of the sampling point corresponding to the bandwidth of 200 Hz, respectively calculate the energy in the two sliding windows, and obtain the gain coefficient c(n) through the energy difference. The waveform of the gain coefficient c(n) is shown in dotted line in Fig. 4 . The final output signal is C(n)=B(n) c(n), and the waveform of C(n) is as shown in Figure 5;

S3. Perform moving average filtering on C(n) obtained in S2, obtain the filtered signal D(n), and the waveform of D(n) is as shown in Figure 6;

S4. Gaussian fitting is performed on D(n) obtained in S3. First, a Gaussian window G(n) is generated, and each sampling point in D(n) is fitted using G(n) to obtain a fitted signal E(n). The waveform obtained by normalizing E(n) is shown in FIG. 7 .

S5, extraction of the maximum point in Figure 7 and its corresponding maximum value as shown in Figure 8, from Figure 8, it can be found that there are 8 obvious spectral peaks, Figure 8 is a signal waveform that limits the abscissa of Figure 7 within the range of 10kHz to 14kHz, it can be found that the frequencies corresponding to these maximum points can correspond to 8 carrier frequencies of the 8FSK signal.

Claims (4)

1. The method for detecting the spectrum peak of the underwater acoustic signal is characterized by comprising the following steps of:

s1, using A (N) to represent a signal waveform containing a spectrum peak, wherein N is the position of the nth sampling point of the signal waveform, n=1, 2, & N, N is the number of the sampling points, let B (N) = |a (N) | when the spectral peak waveform has a negative polarity, and let B (N) = a (N) when the spectral peak waveform has no negative polarity;

s2, taking the nth sampling point of B (n) as the center, establishing two sliding windows, wherein the length of the short sliding window is2s _win +1, length of long sliding window 2l _win +1, and l _win ＝2s _win The method comprises the steps of carrying out a first treatment on the surface of the The gain factor c (n) is obtained by the energy difference of the two sliding windows, short sliding window energy, which is the nth sample point,/->Is the long sliding window energy of the nth sample point, wherein +.>The output signal is C (N) =b (N) ·c (N), where n=1, 2, N, when C (N) is negative, N-s _win ≤0、n-l _win ≤0、n+s _win > N or n+l _win Let c (N) =0 at > N;

s3, carrying out moving average filtering on the C (n) waveform obtained in the S2, wherein the output signal after the moving average filtering is as follows:where 2p+1 is the running average filter length, n=1, 2, the contents of the terms, N, let D (N) =c (N) when N-P is either less than or equal to 0 or n+p > N;

s4, performing Gaussian fitting on the D (n) obtained in the S3, wherein a Gaussian window function is as follows:where L is the length of the Gaussian window function, - (L-1)/2. Ltoreq.n.ltoreq.L-1/2, α is the width coefficient of the Gaussian window function, and α is inversely proportional to the standard deviation σ= (L-1)/2α of the Gaussian function; the output signal after gaussian fitting is: />Wherein l=2q+1 is gaussianThe length of the window function is chosen to be, n=1, 2, ·····, N, then normalizing E (N) to obtain an optimized spectrum peak waveform, let E (N) =d (N) when N-Q is either less than or equal to 0 or n+q > N;

s5, extracting all maximum values and corresponding maximum value points in the E (n), sequencing from large to small according to the maximum values, extracting the maximum value and the corresponding maximum value point in the E (n) as spectrum peaks if a single spectrum peak exists, and extracting the first m maximum values and corresponding maximum value points in the E (n) as spectrum peaks if a plurality of spectrum peaks exist, wherein m is equal to the number of the spectrum peaks.

2. The method for detecting the spectral peak of the underwater acoustic signal according to claim 1, wherein: the moving average filter length in S3 is equal to the short sliding window length in S2.

3. The method for detecting the spectral peak of the underwater acoustic signal according to claim 1, wherein: the gaussian window function length in S4 is equal to the short sliding window length in S2.

4. The method for detecting the spectral peak of the underwater acoustic signal according to claim 1, wherein: the short sliding window length in S2, the moving average filter length in S3, and the gaussian window function length in S4 are equal.