AU2005255537B2

AU2005255537B2 - Method for detecting targets

Info

Publication number: AU2005255537B2
Application number: AU2005255537A
Authority: AU
Inventors: Jorg Hurka; Dirk Neumeister
Original assignee: Atlas Elektronik GmbH
Current assignee: Atlas Elektronik GmbH
Priority date: 2004-06-17
Filing date: 2005-04-26
Publication date: 2010-03-04
Anticipated expiration: 2025-04-26
Also published as: EP1756983B1; ATE515843T1; IL179946A0; WO2005125064A1; KR20070031941A; EP1756983A1; DE102004029368A1; AU2005255537A1; NO20070285L; KR101099965B1

Abstract

The invention relates to a method for detecting targets emitting wavelength energy into a medium by means of entropy determination at time segments of electric reception signals of a receiver which is arranged in the medium remote from said target. In order to detect targets which emit particularly broad-band noisy low-energy signals, the time segments of the reception signals undergo linear transformation and the entropy of the reception of the transformed signals is determined and it is possible to infer upon said targets on the basis of modification of entropy in the transformed signals. Fourier analysis and/or wavelet analysis of the reception signals is carried out for linear transformation.

Description

METHOD FOR DETECTION OF TARGETS The invention relates generally to a method for detection of targets which emit wave energy into a medium. 5 In the case of one known method for detection of acoustic signals which have been transmitted or reflected by an underwater sound source and are masked by environmental noise (US 5668778), the received signal is sampled in or 10 at for example 30000 samples or points, and the amplitude values are associated with the sample order numbers. The amplitude of the received signal is quantized into, for example, 64 intervals of equal magnitude. The entropy H(X) is in each case calculated using 15 H(X) = p 1 (x) 102 Pi (X) in windows within the received signal with a window size of, for example, 1000 samples and an overlap of, for 20 example, 75%, with pi(x) being the probability of the samples containing an i-th amplitude interval. In order to determine pi(x), the total number of samples in the i-th amplitude interval is divided by the total number of samples in the window (1000 in the example). The entropy 25 value calculated in a window with a predetermined total number of digital samples (1000 in the example) is compared with a threshold. If the entropy value is below this threshold, then the presence of signals is deduced. If it is above this threshold, then exclusively 30 environmental noise is assumed. 868233_1.doc 2 This detection method is very highly appropriate when the signals emitted from the underwater sound source have a narrow bandwidth or are not very noisy, that is to say the received signals have a sufficiently good S/N ratio. In 5 the case of signals with a very broad bandwidth, such as those used by way of example for underwater communication, which are additionally also highly noisy, and thus have an extremely poor S/N ratio, for example of less than -9dB, this method leads to unsatisfactory detection results, and 10 to a high false alarm rate. Any discussion of documents, acts, materials, devices, articles or the like which has been included in the present specification is solely for the purpose of 15 providing a context for the present invention. It is not to be taken as an admission that any or all of these matters form part of the prior art base or were common general knowledge in the field relevant to the present invention as it existed before the priority date of each 20 claim of this application. Preferred embodiments of the present invention aim to provide a method by means of which targets which emit sound can be detected reliably in a very noisy 25 environment, to be precise irrespective of the nature of the signals emitted from the targets. In particular, one aim is also to be able to identify targets which emit only broadband and/or very noisy signals, such as machine noise, communication signals and the like, even in a 30 highly noisy-environment, with a low false alarm rate. Throughout this specification the word "comprise", or variations such as "comprises" or "comprising", will be understood to imply the inclusion of a stated element, 35 integer or step, or group of elements, integers or steps, 868233_1.doc 2A but not the exclusion of any other element, integer or step, or group of elements, integers or steps. In a broad aspect, the present invention provides a method 5 of detecting targets which emit acoustic energy into water, by means of entropy determination in time periods of electrical received signals in a receiver which is arranged remotely from the target in the water, wherein the time periods of the received signals are subjected to 10 a linear transformation process, the entropy of the transforms is determined, and targets are deduced from changes in the entropy in the transforms. Embodiments of the invention have the advantage that, 15 since the entropy is not determined in the received signal itself, but is carried out in the linear transform of the received signal, it is, in particular, even possible to detect with a low false alarm rate those targets which emit broadband signals and/or signals which contain large 20 noise components and which furthermore contain only a small amount of energy, in a highly noisy environment. The method according to the invention finds application in water-borne sound technology or in underwater acoustic, 25 but can also be used for detection of 3 targets which emit acoustic or electromagnetic waves in air. According to one advantageous embodiment of the 5 invention, the linear transformation is carried out in preferably identical time periods in a received signal, which preferably follow one another at regular time intervals. In this case, a spatial area which is defined by the reception characteristic of the receiver can be 10 monitored for the entry of targets, and this can be used, for example, for monitoring harbor entrances and/or waterways. According to one advantageous embodiment of the 15 invention, the linear transformation is carried out in time periods of equal time from at least two received signals, which have been received from directionally selective reception sectors of the receiver. The direction selection of the receiver and the simultaneous 20 analysis of identical signal sections in a plurality of reception sectors which are preferably chosen to be adjacent to one another makes it possible not only to detect targets, but also to find their direction. 25 According to one preferred embodiment of the invention, to realize the linear transfomation of the time periods of the received signals, a Fourier analysis of the time periods of the received signals is carried out. The power spectra that are obtained are normalized, 30 preferably using the Euclidean vector norm, and the entropy of the normalized power spectra is calculated. The calculated entropy is compared with a threshold value, and the presence of at least one target is 4 deduced if the threshold value is significantly undershot. According to one preferred embodiment of the invention, 5 the amplitude of the normalized power spectra is in this case quantized into amplitude intervals of the same magnitude, and the frequency of the normalized power spectra is subdivided into preferably identical frequency intervals, with the frequency intervals resulting from 10 the Fourier analysis preferably being used as frequency intervals. The entropy is calculated from the distribution function of the normalized probability of the frequency intervals in the amplitude intervals. Alternatively, the entropy can also be calculated from 15 the distribution function of the normalized probability of the amplitude intervals which occur in the frequency intervals. In both cases, the probability is normalized using the total number of existing frequency intervals. 20 According to one advantageous embodiment of the invention, to realize the linear transformation of the time periods of the received signals, a wavelet breakdown of the time periods of the received signals is carried out, and the entropy is calculated from the wavelet 25 coefficients using i Ni S = K pi,k loga Pi,k where a=2 or a=e i=1 k=1 where k can assume the value -1 or +1, and Pi,k is the probability of the detail coefficients di,k, 1 is the 30 number of scales and Ni is the number of detail coefficients di,k on the i-th scale. The probability Pi,k is calculated using WO 2005/125064 PCT/EP2005/004440 -5 d 2 i,k 1 2 P =1+ e 2 where au is the noise standard deviation on each of the i scales. 5 The signals which are emitted from the targets can be detected to differently good extents by entropy calculation from the frequency spectra or by entropy calculation from the wavelength coefficients, depending 10 on the nature and form of the signals. However, since the signals which are emitted from the targets are unknown, it is advantageous to use both of the methods described here in parallel for target detection, so that the received signals in the receiver are subjected 15 not only to a Fourier analysis but also to a wavelet analysis, and the entropy is calculated in each case. A statement relating to a target can be made even more reliably by comparison of the detection results determined in this way by two different means. 20 The invention will be described in more detail in the following text with reference to exemplary embodiments, which are illustrated in the drawing, in which: 25 Figure 1 shows a block diagram with functional units in order to illustrate the method for detection of targets, Figure 2 shows a time period in an example of a 30 received signal at the output of a beamformer in Figure 1, Figure 3 shows the power spectrum of two time periods in the received signal in Figure 2, 35 WO 2005/125064 PCT/EP2005/004440 -6 Figure 4 shows an illustration of the entropy calculated from the power spectra, as a function of time, 5 Figure 5 shows a block diagram with functional units in order to illustrate a modified method for detection of targets, Figure 6 shows a time period of an example of received 10 signal at the output of a beamformer as shown in Figure 5, Figure 7 shows a wavelet breakdown of the time period of the received signal in Figure 6 into three 15 levels or scales, and Figure 8 shows an illustration of the time profile of the entropy calculated from the wavelet coefficients shown in Figure 7. 20 The method described in the following text for detection of targets assumes that the targets which are located in a medium emit wave energy, that is to say they transmit wave energy themselves or reflect wave 25 energy striking them. In the exemplary embodiment of the method as described in the following text, the medium is in this case assumed to be water, in which an acoustic wave field propagates which is received by means of a receiver 10 which comprises a large number 30 of electroacoustic transducers 11, is remote from the target and is also referred to as a transducer array or sonar base. The electroacoustic transducers 11 are, for example, as is illustrated schematically in Figure 1, combined to form vertically aligned staves 12, which 35 are arranged preferably at equal intervals on a planar surface or on the casing surface of a cylinder. In the exemplary embodiment in Figure 1, three acoustic transducers 11 are illustrated per stave 12, although WO 2005/125064 PCT/EP2005/004440 -7 the number of electroacoustic transducers can be chosen as required. If the reception characteristic of the receiver 10 is not formed into a beam vertically, that is to say the vertical beam angle of the receiver 10 is 5 not small, then a single electroacoustic transducer 11 per stave 12 is adequate. The number of staves 12 which are arranged horizontally alongside one another may be chosen as required and is, for example, 96 in the case of a cylindrical base. One group of staves 12 is in 10 each case operated jointly. As a result of appropriate signal processing of the received signals from the transducers 11 in a group of jointly operated staves 12, the transducer group has a directional characteristic with one main direction of maximum 15 reception sensitivity, the so-called beam. This directional characteristic of the transducer group, also called the group characteristic, is scanned electronically in order to horizontally scan the medium in the area surrounding the receiver 10, in which case 20 the main direction of the directional characteristic is scanned horizontally in steps by addition and removal of in each case one or more staves 12 in the transducer array, with a cylindrical base of over 3600, by way of example. This electronic scanning results in a large 25 number of directional characteristics located alongside one another from the individual transducer groups with main directions of high reception sensitivity, which is referred to as a beam fan. The directional characteristics of the transducer group are formed in a 30 beamformer 13, at whose output an electrical group signal can be tapped off for each beam or for each transducer group. By way of example, the design and method of operation of the beamformer 13 are described in US 4 060 792. 35 The method for target detection will be described on the basis of the evaluation of one of the group signals from the beamformer 13, referred to in the following WO 2005/125064 PCT/EP2005/004440 -8 text as the received signal. The other group signals are evaluated in the same way, so that a target detection process is carried out in each beam, which corresponds to one reception sector with a defined 5 reception direction of the receiver 10. In principle, the evaluation of the received signals at the output of the beamformer 13 in order to detect targets is carried out in such a manner that signal sections of the received signals which are preferably of equal 10 magnitude are subjected to a linear transformation, the entropy of the transforms is determined, and targets are deduced from changes in the entropy in the transforms. If the aim is to detect targets in only a single beam, then the signal sections which are of 15 equal magnitude and in which a linear transform is carried out follow one another in time. If, on the other hand, the aim is to detect and to track targets in all of the beams or reception sectors, then the signal sections which are of the same magnitude and are 20 subjected to a linear transformation are simultaneous signal sections in the individual group signals at the output of the beamformer 13. In the method as illustrated in Figure 1 for linear 25 transformation of the time periods of the received signals, each time period is subjected in a function block 14, which is in the form of a Fast Fourier Transform (FFT) means, to a Fourier analysis, which forms the power spectrum of the received signal in the 30 individual time periods. Figure 2 shows one example of a received signal which is tapped off in a time period 0 - approximately 13 s at the output of the beamformer 13. Although this 35 cannot be seen in the amplitude representation of the received signal plotted against time, this example of a signal is a synthetically generated signal, which in each case contains a noise component in the time 9 periods 0-5 s and 10-15 s, and a broadband signal component in the time period 5-10 s. The environmental noise, which is normally also superimposed on the signal in the receive signal, is suppressed in the example. 5 Figure 3 shows the power spectra obtained from the time periods 1-5 s and 5-10 s by the Fourier analysis. In this case, the power spectrum from the time period 1-5 s is represented by dashed lines, and that from the time period 10 5-10 s by solid lines. Because of the simplified representation, the signal structure can be seen in the power spectrum, to be precise the noise component in the time period 1-5 s (dashed lines) and the broadband signal component in the time period 5-10 s (solid lines). The 15 power spectra are normalized in the function block 15, preferably by using the Euclidean vector norm. For this purpose, all of the amplitude values of the Fourier or spectral components are divided by a normalization factor, which is obtained from the square root of the sum of the 20 squares of the amplitude values of all of the Fourier components in one time period. The entropy is calculated in the power spectra which had been normalized in this way, in the function block 16, which is in the form of an entropy detector. For this purpose, the amplitude of the 25 normalized power spectra is quantized into amplitude intervals of equal magnitude, and the frequency of the normalized power spectra is subdivided into preferably equal frequency intervals. The frequency intervals obtained from the Fourier analysis are preferably used as 30 frequency intervals. The probability of the frequency intervals which occur in each amplitude interval is determined and normalized for entropy calculation, and is associated with the quanta number of the amplitude intervals. The entropy S is then calculated from this 35 distribution function obtained in this way, using 10 S = - pi loga pi where a=2 or a=e (1), where pi is the relative probability or normalized probability of the frequency intervals which occur in the 5 i-th amplitude interval. The probability of the amplitude intervals is normalized by dividing the probability by the total number of the frequency intervals which occur in total in the respective power spectrum, such that: 10 Z pi=1 (2). Figure 4 shows the entropy S, which has been calculated in the "entropy detector" function block 16 and can be tapped off at the output of the entropy detector, as a 15 function of time. This clearly shows that the entropy value falls in the time period 5-10 s. This time period is identical to the time period 5-10 s in the received signal in which the broadband signal occurs. 20 The entropy calculated in this way is supplied to the "comparator" function block 17, which continuously compares the entropy value with a threshold Smin. If the threshold Smin is undershot, then the comparator 18 passes a signal to a display unit 18, which indicates target 25 detection. The magnitude of the threshold Smin is governed by the permissible false alarm rate. In a modification of the described method, the probability of the amplitude intervals which are 30 associated with the individual frequency intervals is determined, normalized and associated with the order number of the frequency intervals, in the entropy detector 16. The entropy S is calculated from the distribution function obtained in this way using WO 2005/125064 PCT/EP2005/004440 - 11 S = pj - log, pj where a=2 or a=e (3), Where pj is the relative probability or normalized 5 probability of the amplitude interval which occur in the j-th frequency interval. The probability is once again normalized by dividing the probability of the amplitude intervals by the total number of the frequency intervals which occur in total, such that: 10 pj = I (4).

The target-detection method illustrated by means of the block diagram as shown in Figure 5 differs from the 15 method described with reference to Figure 1 in that the linear transformation of the time periods of the received signals is carried out at the output of the beamformer 13 by means of wavelet analysis or wavelet breakdown, and the entropy S is calculated from the 20 wavelet coefficients using Ni S ZZpi, 109a Pi,k where a=2 or a=e (5) I=1 k=I 25 where pi,k is the probability of the detail coefficients di,k, obtained during wavelet breakdown, is the number of scales or levels in the wavelet breakdown and Ni is the number of detail coefficients dk on the i-th scale. The probability Pik is calculated using 30 dij,k PA + e 2a. (6) where o is the noise standard deviation on each of the scales when i=1 to 1.

WO 2005/125064 PCT/EP2005/004440 - 12 Alternatively, the entropy can also be calculated using S=j N Pk In Pk (7) =1 k=1 i 5 where 1 is once again the number of scales and Ni is the number of wavelet coefficients on the i-th scale. log 2 pi,k may, of course, also be used instead of ln 10 By way of example, Figure 6 shows a time period of about 13 s of a synthetically produced signal emitted from the target, which is received in the receiver 18 and can be tapped off at the output of the beamformer 15 13. The received signal has a noise component in the time 1-8 s, has a broadband signal and noise component in the time period 8-10 s, and then has a noise component again. The wavelet breakdown which is applied to this received signal is illustrated by way of 20 example in three levels, which leads to three scales. Each wavelet breakdown results in a number of approximation coefficients and a number of detail coefficients. The detail coefficients dkk are numbered successively on the basis of their position (k=1 to Ni) 25 and are associated with a scale value i (i=1 to 1) . A modified time signal is generated from the approximation coefficients, and a wavelet breakdown is once again applied to it. As is illustrated in Figure 7, the first wavelet breakdown is applied to the 30 time period of the received signal as shown in Figure 6. The detail coefficients d~k which are in this case produced in level 1 are associated with the scale value i=1, and are occupied with the coefficient numbers k=1 to 517 corresponding to their position 35 (Figure 7a) . A modified time signal is generated from the approximation coefficients and has already been smoothed to a greater extent than the received signal WO 2005/125064 PCT/EP2005/004440 - 1.3 illustrated in Figure 6. The next wavelet breakdown is applied to this modified time signal, with the wavelet being expanded in a known manner. The detail coefficients d2,k which are in this case obtained in 5 level 2 are illustrated with their coefficient numbers k=1 to 264 associated with the scale value i=2 in Figure 7b. The approximation coefficients obtained from the second wavelet breakdown are once again joined together to form a new modified time signal, which has 10 once again been smoothed to a greater extent than the previously modified time signal. A wavelet breakdown is once again applied to this time signal, with the wavelet being expanded to an even greater extent in time. The detail coefficients d3,k obtained in level 3 15 are associated in Figure 7c, together with their coefficient numbers k=1 to 137, with the scale value i=3. The wavelet breakdown is terminated when the process of 20 joining the approximation coefficients together to form a new modified time signal results in a largely smoothed signal. The entropy S is calculated using equation (5) or equation (7) from the detail coefficients di,k, where i=1 to i=l and k=1 to k=Ni, 25 obtained in this way. The entropy are calculated from the detail coefficients dik using equation (5) or equation (7) in conjunction with equation (6) is illustrated as a function of time 30 in the diagram in Figure 8. This clearly shows that the entropy decreases to a major extent in the time period 8-10 s. This drop occurs in the time period in which the broadband signal component of the signal sent from the target occurs in the received signal. The entropy 35 calculated in the entropy detector 21 as shown in Figure 8 is supplied to the comparator 17, and a threshold Smin is once again applied to it. If the entropy values are below this threshold, then the WO 2005/125064 PCT/EP2005/004440 - 14 comparator 17 passes a signal to the display unit 18, which indicates target detection. In equation (1) and equation (3), the minus sign can be 5 replaced by a plus sign, and in equation (5) the plus sign can be replaced by a minus sign. In this case, target detection occurs when the calculated entropy is greater than a predetermined threshold.

Claims

1. A method of detecting targets, which emit acoustic energy into water, by means of entropy determination in time periods of electrical received signals in a receiver 5 which is arranged remotely from the target in the water, wherein the time periods of the received signals are subjected to a linear transformation process, the entropy of the transforms is determined, and targets are deduced from changes in the entropy in the transforms. 10

2. The method as claimed in claim 1, wherein the transformation is carried out in substantially identical time periods in a received signal. 15

3. The method as claimed in claim 2, wherein the substantially identical time periods follow one another at a regular time interval.

4. The method as claimed in claim 1, wherein the 20 transformation is carried out in periods of substantially equal time from at least two received signals, which have been received from directionally selective reception sectors of the receiver. 25

5. The method as claimed in claim 4, wherein the received signals are taken from spatially adjacent reception sectors.

6. The method as claimed in any one of claims 1-5 30 wherein, to realize the linear transformation of the time periods of the received signals, a Fourier analysis of the time periods of the received signals is carried out, their normalized power spectra are formed, and the entropy of the normalized power spectra is calculated. 35 16

7. The method as claimed in claim 6, wherein the Euclidean vector norm is used for normalization of the power spectra. 5

8. The method as claim in claim 6 or 7, wherein the amplitude of the normalized power spectra is quantized into amplitude intervals of the same magnitude, and the frequency of the normalized power spectra is subdivided into substantially identical frequency intervals, wherein 10 the probability of the frequency intervals which occur in the amplitude intervals is determined and is normalized, wherein the normalized probability is in each case associated with the quanta numbers of the amplitude intervals, and wherein the entropy S in the distribution 15 function obtained in this way is calculated using S = K 1: pi loga pi where a=2 or a=e (1), where K can assume the value -l or +1, and pi is the 20 normalized probability of the frequency intervals which occur in the i-th amplitude interval, and wherein the probability is normalized using the total number of the frequency intervals which occur in total in the respective power spectrum, such that: 25 pii= (2).

9. The method as claimed in claim 6 or 7, wherein the amplitude of the normalized power spectra is quantized 30 into substantially identical amplitude intervals, and the frequency of the power spectra is subdivided into frequency intervals of equal magnitude, wherein the probability of the amplitude intervals which are associated with the individual frequency intervals is 17 determined, normalized and is in each case associated with the order numbers of the frequency intervals, and wherein the entropy S is calculated from the distribution function obtained in this way using: 5 S = K I pj where a=2 or a=e (3), where K can assume the value -l or +1 and Pj is the normalized probability of the amplitude intervals which 10 occur in the j-th frequency interval, and wherein the probability is normalized using the total number of the frequency intervals which occur in total in the respective power spectrum, such that: 15 Zpj=1 (4).

10. The method as claimed in claim 8 or 9, wherein the frequency intervals which are obtained from the Fourier analysis are used as frequency intervals. 20

11. The method as claimed in any one of claims 1-5, wherein, to realize the linear transformation of the time periods of the received signals, a wavelet breakdown of the time periods of the received signals is carried out 25 and the entropy S is calculated from the wavelet coefficients using S Ni S = K p 1P loga Pi,k where a=2 or a=e (5), i=1 k=1 30 where K can assume the value -1 or +1 and Pi,k is the probability of the detail coefficients di,k, 1 is the number of scales and Ni is the number of detail coefficients on 18 the i-th scale, and wherein the probability Pi,k is calculated using d'i.k PiA=1+ e 2' (6), where ai is the noise standard Oeviation on the i-th scale.

12. The method as claimed in claim 11, wherein the 10 wavelet breakdown is carried out in a plurality of levels or transformation steps, with approximation coefficients and detail coefficients being obtained in each wavelet breakdown, wherein the wavelet breakdown in the first level is applied to the time signal defined by the 15 individual time periods, and in each subsequent level is applied to a modified time signal generated from the approximation coefficients, and wherein the detail coefficients dj,k which are obtained from one level of a wavelet breakdown are in each case plotted on the basis of 20 their coefficient number k on one of i successive scales, which corresponds to the respective level of the wavelet breakdown.

13. The method as claimed in any one of claims 1-12, 25 wherein the calculated entropy S is compared with a threshold value, and the presence of at least one target is deduced if the threshold value is undershot or overshot, depending on whether the predetermined value K is -1 or +1. 30

14. A method of detecting targets, which emit acoustic energy into water, said method substantially as hereinbefore described with reference to the accompanying drawings. 35