阅读说明：本技术 一种水声目标检测盲源分离方法 (Underwater sound target detection blind source separation method ) 是由 王耀宾 司纪锋 徐小亮 刘伟涛 于 2021-08-18 设计创作，主要内容包括：本发明涉及水声信号的处理方法,具体是一种水声目标检测盲源分离识别方法。一种水声目标检测盲源分离方法,包括：(1)从原始信号中提取多路独立分量信号；(2)对提取的独立分量信号进行模式识别,提取出表征共模干扰的独立分量信号；(3)从原始信号中抵消掉共模干扰信号,获得盲源分离后的目标水声信号。本发明通过目标检测盲源分离识别方法,可以有效分离和识别共模干扰信号,从而达到消除干扰信号的目的,具有很强的实用性。(The invention relates to a processing method of an underwater sound signal, in particular to a blind source separation and identification method for underwater sound target detection. An underwater acoustic target detection blind source separation method comprises the following steps: (1) extracting a plurality of independent component signals from an original signal; (2) performing mode identification on the extracted independent component signal, and extracting an independent component signal representing common-mode interference; (3) and canceling the common-mode interference signal from the original signal to obtain a target underwater sound signal after blind source separation. The invention can effectively separate and identify the common-mode interference signal through the target detection blind source separation and identification method, thereby achieving the purpose of eliminating the interference signal and having strong practicability.)

1. A blind source separation method for underwater sound target detection is characterized by comprising the following steps:

(1) extracting a plurality of independent component signals from an original signal;

(2) performing mode identification on the extracted independent component signal, and extracting an independent component signal representing common-mode interference;

(3) and canceling the common-mode interference signal from the original signal to obtain a target underwater sound signal after blind source separation.

2. The underwater acoustic target detection blind source separation method according to claim 1, characterized in that: in the step (1), an N-path component signal is separated from an original signal by adopting a FastlCA algorithm.

3. The underwater acoustic target detection blind source separation method according to claim 2, characterized in that: the step (2) specifically comprises the following steps:

respectively carrying out correlation operation on the separated N paths of component signals and the original N paths of signals to obtain a correlation value of each path of component signal and the corresponding original signal;

calculating the mean of the N correlation valuesSum variance S_jAs an identification feature;

selectingSatisfies the first reference intervalAnd S_jSatisfies a second reference interval S ═ S_l,s_h]And taking the 1-2 paths of component signals as component signals representing common-mode interference.

4. The underwater acoustic target detection blind source separation method according to claim 3, characterized in that: in the step (3), the common-mode interference component signal is cancelled out from the original signal by adopting normalization processing according to the energy.

5. The underwater acoustic target detection blind source separation method according to claim 3, characterized in that: the first reference intervalThe second reference interval S ═ 0,0.01]。

Technical Field

The invention relates to a processing method of an underwater sound signal, in particular to a blind source separation and identification method for underwater sound target detection.

Background

Interference signals which are often generated in the multichannel underwater sound signal acquisition process mainly come from a hydrophone array preamplifier circuit, an analog signal transmission medium, a signal conditioning circuit and the like. Common mode interference is one of the important interferences on electronic and electrical products, and causes thereof include impedance coupling and electrostatic and electromagnetic inductive coupling effects of a power supply system, and currents caused by potential differences between signal sources and grounds.

The common mode interference is eliminated mainly by solving the problem from hardware, and the method is a good choice for eliminating interference repair data by adopting a signal processing method when the sampling data containing the common mode interference cannot be completely eliminated from hardware or is processed.

The common mode interference signal forms interference on the A/D sampling signal, and the influence on sonar detection is shown as forming a false target in a positive horizontal direction on a direction course diagram, so the common mode interference signal can be called positive horizontal interference or in-phase interference. The common mode interference can be explained by a plane wave model and a beam forming theory in the array signal processing: the common mode interference is triggered simultaneously for A/D sampling of each array element of the hydrophone array, so that no time delay difference exists between interference signals acquired by each array element, namely the common mode interference can be regarded as a pseudo target positioned in a normal and horizontal direction, and no time delay difference exists when a radiation signal of the pseudo target reaches each array element.

Fig. 1 is a schematic diagram of acquisition of hydrophone array signals, S1 and S2 are assumed 2 underwater target signals, and the 2 underwater target signals have a certain angle with a transducer array and can be identified and positioned through a beam forming algorithm. The common mode interference signal of the array appears as a false target S0 signal in the forward direction. For a/D sampling, each channel acquires a mixed signal of a target signal and a common-mode interference signal, and because there is little available prior information, and the frequency band of the common-mode interference signal has uncertainty, and the frequency spectrum coincides with the target signal, it cannot be removed by filtering.

Disclosure of Invention

The invention provides a blind source separation and identification method for underwater sound target detection, aiming at solving the technical problem that common-mode interference signals cannot be removed in the prior art.

The technical scheme adopted by the invention for solving the technical problems is as follows: an underwater acoustic target detection blind source separation method comprises the following steps:

(1) extracting a plurality of independent component signals from an original signal;

(2) performing mode identification on the extracted independent component signal, and extracting an independent component signal which can represent common-mode interference most;

(3) and canceling the common-mode interference component from the original signal to obtain a target underwater sound signal after blind source separation.

In a preferred embodiment of the present invention, in step (1), the FastlCA algorithm is used to separate N component signals from the original signal.

Further preferably, in the step (2), the method specifically includes:

respectively carrying out correlation operation on the separated N paths of component signals and the original N paths of signals to obtain a correlation value of each path of component signal and the corresponding original signal;

calculating the mean of the N correlation valuesSum variance S_jAs an identification feature;

selectingSatisfies the first reference intervalAnd S_jSatisfies a second reference interval S ═ S_l,s_h]And taking the 1-2 paths of component signals as component signals representing common-mode interference.

Further preferably, the first reference intervalThe second reference interval S ═ 0,0.01]。

Further preferably, in the step (3), the common-mode interference component is cancelled out from the original signal by using a normalization process according to the energy.

The invention is based on Independent Component (ICA) theory, when the common mode interference can not be eliminated from hardware or the data containing the interference signal is processed, the common mode interference signal can be effectively separated and identified by the target detection blind source separation and identification method, thereby achieving the purpose of eliminating the interference signal and having strong practicability.

Drawings

FIG. 1 is a schematic diagram of array signal acquisition;

FIG. 2 is a flowchart of a blind source separation method for underwater acoustic target detection according to an embodiment of the present invention;

FIG. 3 is a graph of array original frequency signals;

FIG. 4 is a graph of the independent components separated using the FastICA algorithm;

FIG. 5 is a graph of frequency domain characteristics of the front and back signals processed using the method of the present invention;

FIG. 6 is a chart of the time azimuth history before and after the processing by the method of the present invention.

Detailed Description

In order to facilitate an understanding of the invention, the invention is described in more detail below with reference to the accompanying drawings and specific examples. Preferred embodiments of the present invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

The invention provides a blind source separation and identification method for underwater sound target detection, which is specifically implemented by the flow shown in figure 2, in the embodiment, a hardware device is a 16-path array element, an underwater sound original frequency signal acquired by a hydrophone array is taken as a data processing object for separation and identification, as shown in figure 3, only the first 8 paths of signals and a local amplification signal are listed and are counted as A_i＝Orginal(Ch_i)。

A FastICA blind source separation algorithm is adopted, and the specific processing steps are as follows:

1. blind source separation

(1) The input mixed signal X is subjected to whitening processing, including averaging and decorrelation processing. Whitened mixed signalThenThe relationship to X can be described as: is a mixed signal sample covariance matrix.

(2) To pairPerforming linear transformation to enable the statistics of each row of the linear transformation result Y to be independent and the non-Gaussian property to be maximum, and taking Y as the estimation of a source signal S; based on the negative entropy of Y, an objective function is constructed and an optimization algorithm is used for solving the linear transformation W.

(3) From kuhn. tucker conditions, objective function f (w) is:

and solving the optimal solution of the objective function by using a Newton method, and iteratively solving w. Suppose that the t-th independent component is found to beThen w_tResult w of the nth iteration_t,n+1Can be expressed as:

w_t,n+1＝z/||z||₂ (3)。

in this embodiment, FastICA is used to separate 16 independent components, and the 16 component signals jointly represent the target signal and the co-directional interference signal. FIG. 4 shows the partially independent components isolated from FastICA.

(2) Pattern recognition

In addition to efficient separation, it is another key to identify the individual signals that are most representative of the common mode interference signal. Due to the uncertainty of FastICA, new requirements are put on pattern recognition, requiring better adaptability of the classifier to signal diversity. Specifically, the main factors affecting the independent component pattern recognition are scale uncertainty, rank order uncertainty, and the like.

The invention mainly distinguishes the independent component which can most represent the common mode interference signal by theoretical analysis, induction and summarization and using the correlation operation of the signal. Because of the characterization of the common-mode signal, the correlation of the common-mode signal with all the original sampled signal channels is most correlated and stationary compared to the other independent components.

The specific steps of pattern recognition are as follows:

(1) respectively carrying out correlation operation on 16 paths of components separated from FastICA and original 16 paths of signals, and setting A_i＝Orginal(Ch_i) Collecting the signal for the ith path of the original signal, D_jSignal _ ICA (j) is a j-th Signal separated from the ICA, and is subjected to a correlation operation C_ij＝corrcoef(A_i,D_j)，C_ijThe correlation between the ith original signal and the jth component signal is shown in table 1.

TABLE 1 correlation operation List

Corrcoef	A1	A2	……	An-1	An
						D1	C11	C12	……	C1(n-1)	C1n
D2	C21	C22	……	C2(n-1)	C2n
						……	……	……	……	……	……
Dn-1	C(n-1)1	C(n-1)2	……	C(n-1)(n-1)	C(n-1)n
						Dn	Cn1	Cn2	……	Cn(n-1)	Cnn

The partial correlation values calculated in this example are shown in table 2.

TABLE 2 signal correlation coefficient

Corrcoef	A1	A2	A3	A4	A5
						D6	0.029	0.0166	-0.0253	-0.0320	-0.0187
D7	-0.0058	-0.0381	-0.0223	-0.0106	0.0032
						D8	-0.0116	0.0229	0.0224	-0.0160	-0.0136
D13	-0.0878	-0.0886	-0.0713	-0.1050	-0.1053
						D15	0.9923	0.7012	-0.9708	-0.8233	-0.2306

Because of the uncertainty of the FastICA component sign, the correlation coefficient with the original signal is negative by calculating ICA _ Sig 13. And comparing absolute quantities in the identification link. The common-mode signal component is judged only according to the correlation coefficient, and the method has great uncertainty and inadequacy, so that the average correlation coefficient is calculatedSum variance S_jAnd (4) parameters are comprehensively considered.

(2) Calculating the absolute correlation operation result corresponding to the j paths of component signalsMean value of quantitySum variance S_jTaking the mean and variance as identification features:

in this embodiment, the mean value of the absolute quantities of the correlation operation results corresponding to the partial component signals is calculatedSum variance S_jAs shown in table 3.

TABLE 3 characteristic quantity calculation

(3) Setting effective reference interval of mean and varianceS＝[s_l,s_h]。

Because the hardware environment is different, the interference degree is different, and the values are not completely the same. In general, non-co-directional interfering signalsTarget signal S_j>0.1, therefore, in the hardware test environment of the embodiment of the present invention, settings are setS＝[0,0.01]. The common-mode interference signal is characterized by 1 or multiple paths of component signals, and the 1 to 2 paths of component signals are identified as a table by combining the actual condition of the common-mode interference of the underwater acoustic array signalsCharacterizing a component signal of the common mode interference.

In this embodiment, the identification features in table 3 are judged and analyzed, and a path of component signals satisfying the condition is identified: d₁₃The component Signal is the component Signal most characteristic of common mode interference (Signal _ ICA (13)).

(4) The common mode interference component D which is identified is counteracted one by one according to the energy by adopting normalization processing from the original signal₁₃＝Signal_ICA(13)。

As shown in fig. 5, this embodiment lists 4 channels of signals for explanation, and it can be seen from the time domain that before cancellation, periodic interference signals exist in each channel, and the interference signals of each channel appear at the same time. The spectrum distribution of the interference signals before cancellation is wide as seen in the frequency domain. After the cancellation, the periodic interference signal is obviously eliminated in the time domain, and some frequency components are eliminated in the frequency domain.

Fig. 6 is a more intuitive MVDR azimuth process diagram, and it can be seen that there is a strong pseudo target co-directional interference signal in the forward direction of 0 degree before cancellation, the direction of 10 degrees is a real target signal, and the target signal is a periodic signal emitted by a signal source at intervals of 2.5 seconds and 1.5 seconds. The target signal is retained after the cancellation, and the forward pseudo target signal is eliminated. The method of the invention has very good effect on removing common-mode interference signals.