GB2181239A - A method of detecting sound impulses - Google Patents

Abstract

A method of detecting a sound impulse, e.g. from a gun or mortar, comprises the steps of feeding signals to a signal processing means, the signals DATA 0, DATA 1, DATA 2 being representative of the sound impulse detected from a plurality of sound detectors defining a detector cluster, processing the signals in the signal processing means to provide the angle of arrival of the sound impulse at the detector cluster and the time of arrival of the sound impulse at a reference point with respect to the detector cluster. A signal processor for carrying out the method is also described, involving correlation in the frequency domain, adaptive filtering and thresholding, (see also Figures 2 and 3, not shown). <IMAGE>

Description

SPECIFICATION Improvements in or relating to a method of detecting sound The present invention relates to a method of detecting sound, more particularly to a method of determining the time and angle of arrival of a sound impulse, and to a signal processorfor use in a sound location system for detecting sound emissions from, and locating the positions of, sources of quasiimpulsive sounds.

The signal processor may be employed in systems for locating guns and mortars fi ring at high or low rates, and may also be employed in sonar, seismic and the like sound location systems.

The basis of gun sound location is the measurement of characteristics (such as the time of arrival) of sound 'breaks' which are received at microphones distributed in some known configuration, and the calculation from the resulting data ofthe position of each gun, probably in Cartesian map co-ordinates.

In a future battle the rate offiring ofthe guns may be as high as 10 per second. To achieve sufficient location accuracy, an array length of approximately 15 to 20 kilometres or more is typically required. The speed of sound is such that it takes almost 60 sec- onds to travel 20 cm. Thus, in orderto collect data on sound emissions from a single gun from all microphones in a 20 km array, one needs to store that data foruptotwo minutes. As a result, the memory of a processorwill accumulate about 1200 x Breaks, where B is the number of microphone positions in the array.There are a number of methods of handling this sort of data, which fall into one oftwo groups: array processing or single source estimation methods. The former group accepts data in unprocessed form and estimates source positions by various means of cross-correlation. This type of processing isapplicabletocontinuousorquasi- continuous signals and so is not particularly app ropriate to gun location systems. The latter group of processing methods will only accept data that has been previously sorted into associated sets, but is appropriate to impulsive waveforms. The basic pro- cedure is to take these breaks from each microphone and to associate them with breaks from all the others, so that the computation of individual gun positions may be carried out.

The conventional sound location methods rely on a large measure of human intervention during the process of sound detection and source location, for both pattern recognition and time measurements.

The method relies on a man examining waveforms on a 'film reader' and correctly associating gun sounds together. This is a very slow process and it is notfeasible to consider manual systems for pro- cessing at the high rate offering given above. Furthermore, in addition to the large number of breaks in the data file, there will also be acoustic clutter, multi-path and noise, against which to discriminate. Accordingly, the present invention has as one objective the provision of a substantially automatic sound location method of determining the time and angle of arrival ofa sound impulse.

According to the present invention there is prov ided a method of detecting a sound impulse, the method comprising the steps offeeding signals to a signal processing means, the signals being repre sentative ofthe sound impulse detected from a plura lity ofsound detectors defining a detector cluster, processing the signals in the signal processing means to provide the angle ofarrival ofthesound impulse at the detector cluster and the time of arrival ofthe sound impulse at a reference point with re specttothe microphone cluster.

In a preferred embodiment ofthe present invention the signals representative ofthe sound impulse detected at the plurality of sound detectors defining the detector cluster are fed to the signal processing means viva sensor channels each ofwhich is associated with a respective sound detector in the detector cluster, the method comprising the steps of correlat ingsignalsfrom pairs of the sensor channels to pro- vide a plurality of correlated signals, processing each correlated signal to produce a respective correlogram, determining the positions of one or more peaks in each correlogram to provide data indicative of the time differential between the arrival ofthe signals in the pair of channels and data indicative of an estimated possible error in the time differential, processing the datato calculatethe angles of arrival ofthe sound impulse at each sound detector in the detectorclusterand estimating possible errors in said calculated angles of arrival.

The present invention also provides a signal processorfor processing signals representative ofthe sound impulse detected at a plurality of sound detectors defining a detector cluster, the signal pro cessorcomprising meansfor determining the angle of arrival of the sound impulse at the detector cluster and meansfordefining thetime of arrival ofthe sound impulse ata reference point with respect to the detector cluster.

In one embodiment of the present invention the sound detectors are microphones, the method being used, and the signal processorforming a part of, an automatic sound location system for locating guns or mortars.

The signal processor wil I for the pu rpose of the specific description hereinafter be referred to as the Cluster Processor in the sense that it processes signals receivedfromthedetectorcluster.

The present invention will be described further, by way of example, with reference to the accompanying drawings in which Figure lisa block diagram illustrating two stages of a method carried out by a Cluster Processor according to an embodiment ofthe present invention; Figure2 is a block diagram illustrating a third stage ofthe method; and Figure3isa blockdiagram illustrating afourth stage of the method.

The basis ofthe method is to use the Cluster Pro cessorto correlate signals arriving at a detectorclus- ter in the form of a small array of spaced microphones (for example of the order of 30 metre sides) arranged in a predetermined pattern. From this the time and angle of arrival of an impulsive signal can be measured.

The purpose ofthe Cluster Processor is to determine the time and angle of arrival of a signal as accuratelyas possible. In orderthatthis objective may be met the method employed makes use of a number of stagers that enhance features ofthe signal.

These stages are: (a) Correlation ofthe signals from pairs of sensor channels. This is a coherent combination of two noisy signals and produces an improvementofupto 6 dB in signal to noise ratio.

(b) Determination ofthe positions of peaks in the correlograms. These indicate the differential time of arrival for signals atthetwo channels. An estimation ofthe measurement accuracy at this stage can be used to advantage in iaterstages of processing (principally in a further Processor).

(c) Determination of the possible angles of arrival relative to the cluster of sensors. At this stage, the estimate oftime measurement accuracy made in the previous section is used to produce an estimate of the angular measurement accuracy.

(d) Adaptive filtering and recombination of the individual channel signals, in orderto enhance the signal to noise ratio, SNR, before thresholding.

Thresholding is carried outto determine (relativeto a local clock) the time of arrival ofthe signal attheclus- ter.

The details of these stages are depicted in Figures 1 to 3 and described below.

Referring to Figure 1 the method stages (a) and (b) are depicted. The correlation process that is used is carried out inthefrequencydomain.This method, whilst not mandatory, provides a significant reduction in the processing requirement and makes the data available as a frequency domain representation (which is used in later stages ofthe design). Thus, time domain signalsfrom the sensors (three inthe illustrated case) are converted to their complex spectra (usually by means offastfouriertransform (FFT) techniques). These spectra may be filtered either by a fixed or adaptive filter function to maximisethe SNR. In the embodiment of Figure 1 adapt ivefiltering is used.The complex, element-wise product of any pair of these spectra is termed the crossspectrum for that sensor pair and the inverse Fox trier transform of such a cross-spectrum is the cross correlogram forthetwo channels. Ifthere is a signal present, the correlogram will contain a peak at a position corresponding to the difference in time of arrival of the signal atthetwo sensors. It is extracting thistime difference information that is generally difficult. In this case one requires not only the signal peak position but also an estimate ofthe accu racy with which it has been measured. The peak to mean ratio forthe correlogram is used to determine two thresholds separated by a variable amount that allows hysteresis in thethresholding process.The largest peak with the appropriate characteristics is found by searching from both ends of the correlogram. The centre ofthis peak is then located by approximating the sides of the peakto straight lines using a least squares method. The centre of the peak may be determined from the intersection of the two lines, and the accuracy ofthe peak centre from the average gradient and an estimate ofthe noise in the correlogram.

Referring to Figure 2 the method stage (c) is illustrated. In this stage the differential times of arrival are converted into angles of arrival for the sensor pairs.

This is accomplished by use of simpletrigonometry but, for a single pair of sensors and onetimedelay may result in an ambiguity. This ambiguity produces two angle estimates for each delay andthusforthe three sensors shown in the example system there are six angles estimated. These must be combined in such a way as to remove the ambiguities and to improve, if possible, the accuracy ofthe estimate.

The accuracy with which an angle can be calculated from the time delay and the sensor array geometry is dependent on the size of the errors in the time delay measurement and the angle itself. For a single pairthe error in angular measure will always be very large when the angle to be measured results from a source that is ata bearing close to the line joining the two sensors. Afunction has been gener- ated that allows the angular estimates from each pair of sensors to be combined in such a way as to weight each by the estimated error on it. At this time a search algorithm locates the set of three angles thatcor- respond most closely to the same direction resolving the ambiguity on the original sensor pair sets.The differential times have thus been converted into an estimate ofthe angle at which the source of the signal is located with respect to the cluster. In addition the weighting process produces an overall estimate of the error in this angle.

Referring to Figure 3 the method stage (d) is illustrated. The objective ofthis stage is to use information gained in the other processes to produce a filter forthe sensor data. This filter is derived primarily from the cross-spectrum data (see Figure 1 ) and a knowledge of the basic characteristics of the particular signals concerned. Once the filter characteristic has been derived each ofthe original spectra is filtered using it and then the resulting frequency domain data are converted (by means ofthe inverse Fouriertransform) back into the time domain. This process is functionally equivalentto producing a time domain filter but because the spectra are already available it is more computationally efficient. The three signals that result from this process are then shifted by the amounts calculate in stages (a) and (b) so thatthey are in phase and then added.This results in a further improvement in signal to noise ratio of Wothethresholding stage.

Thethreshold process is similartothat used inder- iving the peak positions in the cross-correlogram.

The peak to mean ratio of the output signal is means urged and the amount of noise is estimated. From these data two thresholds are calculated. A search is then carried outto locate the first peak in the output waveform. The centre of this peak is calculated by fitting the sides of the peak (by a least squares approximation) to straight lines. The centre of the peak is given by the intersect ofthetwo lines and an estimate of the error on the peak position can be calculated from slopes ofthe peak sides and the estimated noise in the output channel.

In a practical system the information generated (time of arrival) would then be referenced to a local clock.

The Cluster Processor is utilised as an integral part of an Automatic Sound Location System design, which in one embodiment includes: an array of microphone clusters, distributed over tens of kilometers, Cluster Processors, which, from the outputs of the microphones generate time and angle of arrival data, a Course Processor, which takes the outputs from a number of Cluster Processors and performs associations between the data in order that sets originating from a particular source may be related, a Fine Processor, in which the associated sets of data are further refined to provide the best estimate ofsource position.

Clusters, each consisting of a set of three or more microphones and an attendant 'Cluster Processor' are distributed over a distance perhaps 15 to 25 km.

Claims (9)

1. A method of detecting a sound impulse, the methodcomprisingthesteps offeeding signalsto a signal processing means, the signals being representative of the sound impulse detected from a plurality of sound detectors defining a detector cluster, processing the signals in the signal processing meansto providetheangleofarrival ofthesound impulse at the detector cluster and the time of arrival ofthe sound impulse ata reference point witch re specttothe detector cluster.

2. A method of detecting a sound impulse as claimed in claim 1 wherein the signals representative ofthe sound impulse detected at the plurality of sound detectors defining the detector cluster are fed tothesignal processing meansviasensorchannels each of which is associated with a respective sound detector in the detector cluster, the method compris ingthestepsofcorrelatingsignalsfrom pairs ofthe sensor channels to provide a plurality of correlated signals, processing each correlated signal to produce a respective correlogram, determining the positions of one or more peaks in eachcorrelogramto provide data indicative ofthetime differential be tween the arrival ofthesignals in the pairofchannels and data indicative of an estimated possible error in the time differential, processing the data to calculate the angles of arrival ofthesound impulse at each sound detector in the detector cluster and estimating possible errors in said calculated angles of arrival.

3. A signal processorfor processing signals representative ofthe sound impulse detected at a plurality of sound detectors defining a detectorclus- ter, the signal processorcomprising meansfordetermining the angle of arrival ofthesound impulse at the detector cluster and meansfordefining the time ofarrival ofthe sound impulse art a reference point with respect to the detector cluster.

4. Asignal processor as claimed in claim 3 wherein the signal processor is arranged to receive the signals representative of the sound impulse via sensor channels each of which is associated with a respective sound detector in the detector cluster, means being provided for correlating signals from pairs ofthesensorchannels to provide a plurality of correlated signals and for processing each correlated signal to produce a respective correlogram, means being provided for determining the positions of one or more peaks in each correlogram to provide data indicative ofthe time differential between the arrival ofthe signals in the pair of channels and to provide data indicative of an estimated possible error in the time differential, and means being provided for processing the data to calculate the angles of arrival of the sound impulse at each sound detector in thedet- ector cluster and to estimate possible errors in said calculated angles of arrival.

5. Asignal processor as claimed in claim 3 or4 wherein the sound detectors are microphones.

6. A method of detecting a sound impulse sub stantially as hereinbefore described with reference to, and as illustrated in, Figures 1,2 and 3 ofthe ac- companying drawings.

7. Asignal processorfor processing signals representative ofthe sound impulse detected at a plurality of sound microphones defining a microphone cluster, the signal processor being substantially as hereinbefore described with reference to, and as illustrated in, Figures 1,2 and 3 ofthe accompany- ing drawings.

8. An automatic sound location method of determining the time and angle of arrival of sound impulses from guns or mortars, the method being as claimed in any one of claims 1,2 or 6.

9. A signal processor for processing signals representative ofthe sound impulses from guns or mortars, the signal processor being as claimed in any one of claims 2 to 5 or claim 7.