High infrasonic goniometry applied to the detection …...High infrasonic goniometry applied to the detection of a helicopter in a high activity environment Dr. Vincent Chritin (1),

SPIE International Society for Optics and Photonics

SECURITY + DEFENCE Conference – Edinburgh, United Kingdom, 26-29 Sept. 2016

Target and Background Signatures | UAV Detection | www.spie.org | Manuscript 9997-3

High infrasonic goniometry applied to the detection of a helicopter in a

high activity environment

Dr. Vincent Chritin (1), Dr. Eric Van Lancker (1), Dr. Peter Wellig (2), Dr. Beat Ott (2)

(1) IAV Engineering, Chemin des Couleuvres 4A, CH-1295 Tannay, Suisse (www.iav.ch).

(2) Science and Technology, armasuisse, Feuerwerkerstrasse 39, CH-3602 Thun, Switzerland

1. PREAMBLE

A current concern of armasuisse is the feasibility of a fixed or mobile acoustic surveillance and recognition network of

sensors allowing to permanently monitor the noise immissions of a wide range of aerial activities such as civil or military

aviation, and other possible acoustic events such as transient events, subsonic or sonic booms or other. This objective

requires an ability to detect, localize and recognize a wide range of potential acoustic events of interest, among others

possibly parasitic acoustic events (natural and industrial events on the ground for example), and possibly high

background noise (for example close to urban or high activity areas). This article presents a general discussion and

conclusion about this problem, based on 20 years of experience totalizing a dozen of research programs or internal

researches by IAV, with an illustration through one central specific experimental case-study carried out within the

framework of an armasuisse research program.

2. GROUND ACOUSTIC SURVEILLANCE: A QUESTION OF RANGE, THUS OF SIGNAL-TO-

NOISE RATIO

Very generally speaking, the question of acoustic surveillance is to “detect (something happens), localize (where) and

recognize (what)”. But yet, whatever the sensing technology used, an acoustic signal of interest is only measurable

accompanied by a certain level background noise at the sensing location, (concurrent or parasitic signals reaching the

measuring sensor, background ambient noise, intrinsic noise of the sensor, or other). So, the detection, localization and

recognition are in the first place a matter of obtaining the highest possible signal-to-noise extraction capability.

Figure 1. (left) A variety of broadband and/or tonal aerial acoustic sources of a few Hz to a few hundred of Hz emission

[Free web illustrations]; (right) Pre-radar acoustic detection and tracking: a two-channels, directive acoustic goniometer

[Amplifiers used by the U.S. Naval Air Service, 1925. Source: Historical photographs]

For example, often, acoustic surveillance requires an ability to deal with very weak signal-to-noise ratios, such as 1/10 (-

20 dB) for example for sources of a few Hz of frequency and bandwidth (ex. Doppler-shifted helicopter blade-pass

frequency at a distance of about a dozen kilometres). This signal-to-noise extraction capability is always related to at

least four main considerations a) the acoustic emission characteristics of the sources (frequency content, temporal

http://2.bp.blogspot.com/-nJdamMVFbpQ/TiNiZZahybI/AAAAAAAAA4g/7DV8SEs7m8Q/s1600/acoustic-detector-1.jpg




enveloping and spatial directivity), b) the characteristics of the propagation path over the distance (geometrical

spreading, energetic losses, and, in all outdoor and long range applications: meteorological and atmospheric conditions

(obviously a calm and quiet environment (in the wavelengths of interest) and a straight propagation path will be more

favourable than opposite conditions)), c) the local conditions at the reception (background and ambient acoustic noise,

and again in all outdoor applications wind-induced turbulent noise on the acoustic sensors inlets themselves, and finally

d) the dynamic range and signal processing capability of the detection system (self-noise of the instrumentation). Instead

of detailing all these aspects, one can focus on the main central surveillance function of the localization, for the

following reason: whichever acoustic sensing technique used, localizing a distant acoustic event pre-requires the

detection of a coherent sound-wave, so that, definitively, localization and detection are almost always dual. There are

several techniques to measure sound wave direction, which are briefly reminded here. Historically, an “acoustic

goniometer” (gonio: angle) is a system that measures the direction of arrival (DOA) of a sound wave, and thus estimates

the source direction, on the principle of the determination of delays between waveforms picked-up between spatially

separated acoustic inlets. An acoustic goniometer is thus made up of an array of spatially separated microphones, called

antenna, arranged in a particular geometry, and a calculation algorithm. In the following historical example the

goniometer combines several two-channel systems, each possibly directive and orientable as a whole. Evidently without

computerized signal processing capability, the hearing capacity of an expert observer is the only way the acoustic

information delivered by the system can be analysed.

According to the classical definitions, active goniometry is used when the signal source is under control, while passive

goniometry refers to the localization of unknown sources of interest. A basic goniometer design is established according

to the application and adaptation of the Radar and Sonar techniques that have been extensively developed during the last

century (although Radar refers to electromagnetic waves and Sonar to the underwater acoustic equivalent, both deal with

the problem of echo detection and estimation of the direction of arrival of signals impinging on an array of sensors, while

in this case we only deal with passive techniques so that the source of interest is the acoustic emission of the target

itself). However, the diversity of the applications implies specific designs. The operational context summarizes the

relevant properties necessary to complete the goniometer design. For example, the medium is characterized by its

acoustic properties and background noise, and the sound source is defined by its signal properties: type, spectrum and

level. The distance separating the source from the antenna (source range) relative to the antenna extension determines the

propagation properties to consider: planar propagation for far sources, spherical propagation for close sources. If the

source is inside the antenna, another approach is considered (referred to as Distributed Array). When the acoustic source

is moving, such as in the present case of aircraft surveillance, many aspects have to be considered: the analysis rate and

the analysis window length can represent critical choices for reasonable source position updating. The configuration can

also involve one or several sources. The propagation can introduce a multi-path distortion such as specular reflections,

which is equivalent to having many delayed replicas of the same acoustic event, generated in different places in space.

Concerning other possible technical trends for acoustical localization, one can cite the particles velocity technologies

(heating technic as well as more ‘laboratory’ technics VIP or LDV). Compared to classical goniometry, the

complementary field of the velocity technologies might be intrinsically more sensitive to air flow and turbulence, which

can be a strongly limiting factor since in most applications the sound waves propagate in air flows and are locally

picked-up in turbulent air. Second, the intrinsic filtering behaviour must be low-pass oriented while high-pass is an

antinomic need for elimination of VLF noise, mostly parasitic. So, to come back, for the preceding reasons, to the main

goniometry field, there are also several possible concurrent approaches, that are discussed in the next section.

3. ACOUSTICS GONIOMETRY BACKGROUND AND THE FUNDAMENTAL QUESTION OF

GROUND ACOUSTICS SURVEILLANCE

From both ends of the ground acoustic surveillance problem (a certain acoustic source on one hand and a technical

detection capability on the other hand), and focusing to the general aircraft detection problematic, one can state that:

Concerning aircraft signals of interest, obviously the more the mass and horizontal or vertical-up speed of the aircraft,

and the bigger the dimensions of the blades or/and the jet engine (as well as possibly the vehicle dimensions

themselves), the higher the power and longer the wavelengths (thus, the propagation distance d). A noise-optimized

micro-drone will evidently produce much smaller wavelengths and total acoustic power than any conventional

aircraft. Thus, leaving aside the VLF ground acoustic surveillance (<10-2

to about 100 Hz, eg possible military

activities but almost no possible aircraft events) the problematic of acoustic detection and furtivity manifests itself in




about four magnitudes of metric range (from 102m to 10

6m) and three magnitudes of frequencies (from 10

0Hz to

103Hz), mainly depending on the category of the source of interest.

When dealing with d >> detection purposes in air and temporal or spatial detection methods, acoustic antennas can

be formed by a spatialized number of acoustic inlets of a centimetric (stand-alone integrated antenna), a metric

and up to a decametric (ex. local array of microphones, distributed array) spatial extension. Whatever the spatial

scale of the acoustic antenna providing DOA information, any single antenna can be a node or a substation of a more

general network of antennas, then providing, thanks to triangulation, a true localization capability (instead of only

DOA), which is evidently a main objective of acoustic surveillance.

To fix the ideas, a substation inter-distance of interest for triangulation of long distance, low frequency events typically

below a few Hz (ex. supersonic flying objects, subsonic booms, meteorological and atmospheric phenomena’s,) would

be about at least a dozen of km, while a substation inter-distance of interest for triangulation of events such as helicopters

or propelled UAV’s would typically be about a km, and detecting a micro-drone would require a meshing of about a

hundred meters. So in turn, the most general question of interest comes as follows:

Typologically speaking, given the physics of acoustics and the targeted field of application, what technological

orientation would be the most capable to cover five magnitudes of range and three magnitudes of frequencies?

The approach chosen within the framework of the armasuisse research program carried-out by IAV has been to

continuously develop competence into a generic acoustic goniometric framework that could allow a step-by-step

evaluation of most of the contemporary and future aircraft surveillance problems and bring advanced performance

demonstration in selected sub-applications case studies such as the one presented in the next section of this article

(helicopter in high activity environment). The next section presents the so called ‘spatio-temporal’, two-step localization

approach that has been preferred at the origin [1] and could be successfully consolidated in relation to the considered

applications, while the end of this section resumes the fundamentals of the different background techniques [2], [3], [4]

and [5], that have initially and already at an early stage led to the development of the more specific spatio-temporal

approach.

In brief, early systems were based on the conventional Delay & Sum beamforming algorithm, enhancements were

introduced by the Filter & Sum algorithm, and then by the Optimum beamforming. Then Sub-Spaced direction finding

methods were introduced as Super-Resolution techniques. The term beamforming techniques is used to describe the

ability of a system to perform a spatial filtering by “forming beams” in order to receive a signal radiating from a specific

direction while attenuating signals from other locations. The basic idea is to “steer” the array in one direction and to

calculate the output power. The steering locations, which result in maximum power, yield the DOA estimate. The pencil

beam is obtained by forming a linear combination of the sensor outputs. The beamformer output y, function of the

steering direction, is the weighted sum of the sensor output xm

M

m

mm kxwky1

][][ Eq. 1

where M is the sensor number, wm is the complex weighting and k refers to the time index of the sampled signal. The

overall block diagram of the beamformer is presented in Figure 2. Considering a source far from the array, the wave

fronts are perpendicular to the direction of propagation referenced by the angle . Under this condition, the time delay of

arrival between the sensors is a simple equation relating the angle , the inter-element distance d and the speed of sound

c. After being picked up and conditioned, the signals are digitized and introduced into the beamforming algorithm.




d

12 = d/c sin

1

2

G A D

G A D

G A D

G A D

+

DELAY Complex

WEIGHT Real

a (f) 1 1

a (f) 2 e

-j2f 12

a (f) 3 e

-2 j2f 12

FIR

a (f) M e

-(M-1) j2f 12

Source

Wave front

y[k]

x1[k]

x2[k]

xM[k]

x3[k]

x1(t)

x2(t)

x3(t)

xM(t)

3

M

Figure 2. Filter & Sum beamformer

In the basic case of a Uniform Linear Array (ULA) represented in Figure 3 (3-D space), the focalization in the direction

is obtained by introducing the beamformer weightings

( 1) sin

,

dj m

cm mw a e

1..m M Eq. 2

where the complex weight appears as a phase term multiplied by the amplitude weights am. The sensor outputs are thus

delayed in such a way that the signals impinging from the focalization direction given by are summed constructively

(signals in phase) although the components coming from other directions are summed destructively (statically random

phase). In the Delay & Sum algorithm, the amplitude weights are set to one, while in the Filter & Sum algorithm, the

weights are chosen to enhance the beam shape and reduce side lobe levels. For example, the Dolph-Tchebyshev

weighting forces the maximum level of the side lobe as presented in Figure 4.

Figure 3. Uniform Linear Array of five microphones

with inter-element spacing d; the steering direction is

referenced by the angles (,) or the angles (Azimuth,

Elevation)

Figure 4. Amplitude weight, Array response steered in

the direction =30°in case of Rectangular weighting

(Delay & Sum algorithm: gray line) and Dolph-

Tchebyshev (Filter & Sum algorithm: black line)

1 2 3 4 5 6 7 0

0.2

0.4

0.6

0.8

1

Weight Index

Am

plit

ude

-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90-40

-30

-20

-10

0

DO A (°)

Gain

(dB)




In order to increase the spatial filtering performances, the “Optimum beamformer” takes advantage of a priori

information. According to the minimization cost function, the weighting vector is based on the reference signal, when the

desired signal is known, or the SNR (Signal to Noise Ratio), when both signal and noise are known. If the locations of

competing sources are known, a Multiple Sidelobe Canceller (MSC) is implemented to eliminate the unwanted

contributions. The “Adaptive beamforming” arrays use weight vector adaptation algorithms such as the Least Mean

Square (LMS) or the Recursive Mean Square (RMS) to converge on the optimal weight vector. These techniques are

useful when there are a few discrete noise source directions that account for the acoustic interference. The “Matched

Field beamforming” takes into account the spatial inhomogeneity’s and medium’s variation inducing each sensor to pick

up different propagation directions and speed. The beamformer matches the measured field at the array, with replicas of

the expected field for all locations of sources. The replicas are determined by modelling the acoustic propagation for

each sensor. The general philosophy of beamforming techniques is to scan the space with an adequate spatial filtering in

order to build the beamformer output map and to determine the DOA corresponding to its maxima. This processing is

CPU costly and yields useless or not optimal information in many applications. Another inconvenience is that the

beamformers require array sizes that are comparable to the acoustic wavelength, as the method intrinsically implies a

narrow band approach. Thus these techniques are not adapted for large band signals (maximum band = 1 octave), such

as jet turbine noise and transient signals. Reasonable spatial resolution, related to the array aperture, is obtained to the

detriment of the number of sensors. Solutions with hundreds of sensors are common. Yet, the advantages of the

beamformers are that the theory is well known (extensive literature) and that dedicated DSPs can be found. A second

approach deals with the Subspace-based direction finding methods. These high-resolution methods are based on the

calculation of the spatial covariance matrix given by

1

/K H

kR x k x k K

Eq. 3

where is the spatial signal vector composed of the samples recorded at the M sensors, k is the index of the snapshot and

K is the number of snapshots considered. In the case of the minimum variance optimization algorithm, the principle is to

find the DOA () that maximizes the array output power

1

1,

, ,mv H

Pa R a

Eq. 4

where a is the steering vector. This second order Statistics-Based method is purely spatial. Results are updated for each

new sample. The signal to noise ratio is improved by performing a mean average on the K snapshots of the frame. An

initial estimation of the number of sources present is required. Distinction is made between signal Sub-Space methods

(Minimum Variance, Autoregressive, Subspace fitting, ESPRIT) and noise Sub-Space methods (Pisarenko, Music,

Eigenvector). The main advantage of the subspace methods is the resolution. The drawbacks are those of a spatial

approach. The spatial covariance matrix is calculated on a snapshot of the acoustic field at the array, thus the sensor

number is critical to extract the information and to cover the working frequency band. The minimization of the array

output power requires a 2D scanning in the case of a unique source. In a multi-sources configuration, the dimension of

the space to scan is multiplied by the a priori number of sources, thus drastically increasing the CPU load.

4. THE SPATIO-TEMPORAL APPROACH: LESS SENSORS, MORE CAPABILITY

To overcome the main drawbacks of the beamforming, spatial and other concurrent technics such as particles velocity-

based or other kinds of directive sensing, a new method called the spatio-temporal method was developed and is briefly

described in this section. The first advantage of this method is that a first temporal process allows handling broadband

signals as well as tonal signals, which is a very important applicability aspect for the diversity of sources of interest such

as propelled (tonal) vs jet engines and transient events (broadband). Secondly, whatever the expected situation, a

minimal number of four sensors forming three baselines is sufficient to manage the localization, which is technically a

very highly facilitating aspect when taking into account 1) the initial and long-term stability of acoustic transducers, in

particular phase and amplitude mismatches and 2), a very important point in the removing of ambiguities: with respect to

aliasing (low sensitivity) for (frequent) cases of sources emitting acoustic energy close above the Nyquist frequency (half

of the sampling rate). Furthermore, computations are restricted to the minimum, as the scan of the sources is achieved by

a one-dimension peak-picking procedure on the Nb baselines. So in combination with the precedent arguments, this

opens the possibility for scaling the general acoustics surveillance problematic into two very complementary modes of




technical implementation and architecture: a network of arrays of sensors operating either 1) each array as an

individual “event detector’ (reduction of information, i.e. full event signal processing and communication, which is

obtainedat the level of the array itself), or 2) each sensor of the array, thanks to the low total bandwidth required per

array (low sampling rate and reduced number of sensors), operates as a raw signal communication unit to a distant

aggregating advanced signal processing service.

To briefly mention the spatio-temporal method, it is carried out by a two-step procedure as presented in the next figure,

with the first step considering the estimation of time delay between the sensors taken two by two, and the second step

processing the localization itself.

Time Delay

Estimation

Localization 1 2, ,..., Sn n n

DOA sets Wave field

1 2, ,... Mp p p

Delay sets

1 2, ,...,SN

1 2, ,...,bN

Figure 5. Two–step procedure

The delays calculated for the appropriate Nb baselines (sensor pairs) feed the localization algorithm, which determines

the best DOA. The multi-sources configuration is handled by a recombination module, which reorganizes the delays

along each baseline into Ns delay vector [Nb x 1] corresponding to each source. The time delay estimation is a temporal

process as it is based on the generalized cross-correlation techniques. On the other hand, the localization module

introduces geometry and size of the antenna, the speed of sound and is a spatial process. The time delay estimation study

takes advantage of recent developments. G. C. Carter has edited “a collection of key contributions … in this fast-moving

field” [6] which regroups articles covering most aspects of the TDE. The methods considered can be classified as

classical methods, generalized methods (Roth, SCOT, PHAT, ML, CPSP), adaptive methods (LMS), and Higher-Order

Statistics-Based methods (Bi-spectrum, Cross-cumulants, Hologram). The classical and generalized methods were

chosen and are described in [1]. The LMS adaptive method consists in calculating a Finite Impulse Response filter that

automatically adapts its coefficients to minimize the mean square difference between two inputs. The FIR filter

introduces a delay opposite to the one existing between the two inputs. In the ideal situation, the filter weight

corresponding to the true delay would be unitary and all the other weights would be zero [7]. This adaptive method was

rejected because of updating constraints and CPU load. The Higher-Order Statistics-Based methods are based on the

hypothesis that the signals are non-Gaussian processes. The principle is that even cumulated Orders of Gaussian noise

vanish. Higher-Order Statistics-Based methods (the cross-correlation function is a second-order method) were

implemented in a MATLAB test framework. Yet, the frame length required is prohibitive for most of the applications.

The localization of close range sources is commonly performed by a hyperbolically fit algorithm. Successive iterations in

a least mean squares gradient descent search converges to the best DOA ([8], [9]). In [10], Brandstein solves the far-field

case by searching the DOA that minimizes the distance between the measured delays set and the delays set

corresponding to the investigated DOA. In conclusion, the finally implemented method can be defined as a spatio-

temporal closed-form goniometric solution, which consists in multiplying the delay vector by the inverse of the relative

position matrix and by the speed of sound, and under the appropriate taking into account of the inherent far-field

assumption in terms of localization errors. The estimation of the performances of the two-step localization procedure vs

the relevant parameters of the goniometry and by studying their influence on the performances, gives easy rules for the

antenna design, and finally ends up with a very flexible, very performant, easy to implement geometry in most ground

terrains or situations. Various applications, not described in this article, were tackled in parallel to this research and have

been validated in different armasuisse and/or IAV internal projects.

5. CASE STUDY DEMONSTRATION: THE AS2R EXPERIMENTAL SYSTEM

The as2r experimental system is an application example of the spatio-temporal design methodology answering to the

following case-study exercise: “Passively detect and locate , automatically, in real time, from the ground, in any weather

conditions, in urban mountainous regions, the intrusion of a medium size helicopter, or possibly a medium size propelled

UAV, that would be in altitude flight (altitude superior to the surrounding reliefs), or low-altitude non-hidden by a relief,

in a zone of typically 25 km2 around a central point”.




According to the preceding definition and the spatio-temporal design methodology, the technological solution was

defined as follows: a) a station made of two substations nodes, b) each substation consist of an antenna of eight

microphones within a ground footprint of R=15m, c) the eight microphones forming (see Figure 6) two different

imbricated patterns (pattern 1-5-6-7-8: provides the performance in precision, pattern 1-2-3-4-5: enables to remove

ambiguities); d) a time vector integration over 6.4 seconds (a duration in which the target has not moved too much), i.e. a

raw detection vector matrix of 8 microphones x 80 samples (80Hz sampling rate) x 6.4 sec., thus an intrinsic detection

information size of 4096 data. The nominal geometrical definition of the substation is defined as per explanation in the

next figure.

Microphones (8)

Mic. housing (8)

LAN+WLAN Central unit (1)

Figure 6. as2r substation geometrical array definition and example hardware

Note: Compared for example to a beamforming solution that would intrinsically need a considerable number of

microphones (for example, n=120), this solution has the main practical advantages that, with only 8 sensors, it has a

much more important raw information vector (ex. ratio of 4096/120 for the example); subsequently, it is much less

sensitive to amplitude and phase mismatch between microphones, and, finally, it is naturally much easier to protect from

wind (the size of the system and the very reduced number of sensors allow to establish very efficient protection to local

induced turbulence). These points are a critical advantage when needing to ultimately remove wind and background

noise from a signal-of-interest, so this is obviously a very important point in the targeted case-study application.




The demonstrator has successively been achieved in 2012 and has since been permanently operating on two successive

test sites.

Figure 7. as2r field test sites (top: two substations (aerial photograph: Reproduced by permission of swisstopo (BA16069)),

bottom: one substation)

For an acoustic system (which is intrinsically sensitive to background noise), the exceptional signal-to-noise ratio

performance of the system, and thus its global surveillance capability, is explained by obtaining a very high sensitivity to

coherent and tonal signals of interest vs a very low sensitivity to non-coherent, non-tonal (especially such as wind-

induced) background noise, combined with a very high robustness resisting the harshest meteorological and

environmental conditions. These three qualities are firstly obtained by the implementation of entirely self-designed high




infrasonic microphones (called “inframic” microphones), integrated into ground protections. Secondly by the use of an

(in this case) eight-channel dedicated conditioning electronical device, called the “as2r Box”, driven by a Linux-based

barebone also operating a wireless (3G), bit-synchronized permanent streaming of the full raw signals to a distant server.

Thirdly by the use on the distant server, of an advanced coherence-based signal processing technique (inspired by

Bedard/NOAA), and the spatio-temporal goniometry signal processing framework service that IAV has optimized during

approximately 20 years of intensive experimental monitoring in very diverse outdoor (and also indoor) applications. A

so-called ‘mobile reproduction’ of the as2r demonstrator has been developed as well, and can be deployed anywhere.

The fixed station demonstrator, formed of two substations to allow a triangulation layer service, is still permanently

operating, i.e. continuously streaming {2 substations*8 microphones*16 bits*SR=80Hz@1V/Pa} raw acoustic signals to

a central server; the server records the entirety of the data, and redirects it, so that any computer equipped with the as2r

GUI experimental software (and with access to the stream) immediately computes, and displays, live, all end-user

detection screens such as shown in screenshots examples Figure 8 to Figure 12.

Figure 8. Real-time as2r GUI examples during manoeuvres with helicopters hovering or flying over the test area

The demonstrator proved a high capacity of immediately showing the presence, recognizing the type (blade-pass

frequency signature) and providing the localization of any helicopter; in a typical range of 4 km2 under the worst

conditions (high wind, high background noise, small/low noise helicopter, masking reliefs), and up to over 100 km2

under the most favourable conditions (quiet atmosphere, low background noise, heavy/noisy machine). The following

figures show two examples of real-time GUI dashboards provided by the system during manoeuvres with helicopters

hovering or flying over the test area.




Figure 9. ACQ (Acquisition) real-time graphical window. In this example, substation 1 (left) detects a shutting off then

departing helicopter (vertical patterns), while substation 2 (right) detects 7 consecutive artillery transient events (horizontal

patterns).

Figure 10. TDE (Time-Delay Estimation) real-time graphical window. The sub-graphs show, the real-time inter-correlation

of the eighteen pairs of microphones forming the three sub-arrays, for each of the two substations.




Figure 11. LEQ (Equivalent Level) real-time graphical window. The left-hand graphs show the long-term (last 3 hours) time

history level and statistics in the frequency band of interest, and the right-hand graphs show the corresponding frequency

spectrum overlaid on helicopter type recognition (nominal blade-pass frequency and apparent Doppler shifted blade pass

frequency).

Figure 12. DOA (Direction Of Arrival) real-time graphical window. Localization of the acoustic source (here a EC635

helicopter) by triangulation between the almost simultaneous DOA from the two substations.




6. GLOBAL CONCLUSION ABOUT THE POTENTIAL OF GROUND ACOUSTIC

SURVEILLANCE

At this stage the fundamental question behind the problematic of ground acoustic surveillance of aircrafts activities arises

more precisely as follows: at the present state-of the art and near future, what technological orientation would

practically be the most capable to cover four magnitudes of kilometric range and three magnitudes of frequencies?

Acoustically speaking, it has been explained how the operational contexts of the territorial surveillance of aircrafts

present a large variety of emission properties (signal frequency, bandwidth, stationary properties, Doppler effect), of

configurations (far-field close range), of source types (stationary moving, spread localized), of source numbers

(competitive sources, image sources) and of environmental and instrumental conditions (altitude, temperature, wind,

noise, instrumental noise), and how, from this, finally in view of a permanent acoustic surveillance and recognition

system on a territory, the key factor to be stated far before any other consideration is the effective expected acoustic

bandwidth of the sources vs the distance range of interest, with the following first definitive answer: with a potential

frequency range of interest of about three orders of magnitude (from 100Hz to 10

3Hz, i.e. wavelengths from 0.34m to 340

m), and a potential range of interest from 10-1

km to 103km, governed by wavelengths and signal-to-noise ratio, such a

capability is practically impossible to cover entirely with a single sensing technique and single array dimensions. The

next subsequent question then is ‘which and how many technological sub-orientations would practically be the most

capable to cover four magnitudes of kilometric range and three magnitudes of frequencies?”

From our experience one pertinent solution would be as follows:

Long range 101 to 10

3 km Short-range 10

-1 to 10

1 km Very short-range 10

-1 km

100Hz to 10

1Hz 10

1Hz to 10

2Hz 10

2Hz to 10

3Hz

Surveillance of powerful ground or

aerial acoustic events, including

transient events (heavy supersonic

vehicles, subsonic booms, heavy

artillery, other)

Surveillance of conventional ground

or aerial acoustic events, including

transient events (helicopters,

aircrafts, UAV’s, artillery, other)

Surveillance of ground or aerial

acoustic events, including transient

events (small aerial vehicles to mini-

drones, firings, other)

Territorial, permanent, with typical

station inter-distance: about 20 to

100km or more

Zonal with a mesh of permanent or

mobile substations, typical station

inter-distance of about 1 to a few km

Local mesh of permanent or mobile

nodes, nodes inter-distance about

100m

Each station a spatio-temporal

goniometer formed of an array of 4

infrasonic microphones

Each station a spatio-temporal

goniometer formed of an array of 4

(basic azimuthal precision) to 8

(maximal precision) high infrasonic

microphones

Each node a spatio-temporal

goniometer formed of an array of for

example 4 (basic azimuthal precision)

to 24 (maximal precision) low-audio

microphones

Indicative ground footprint R=50m Indicative ground footprint R=15m Indicative ground footprint R=0.5m

Similar existing system/background:

ARFANG project


AS2R project


FIRST NODE project

Figure 13. Ground acoustic surveillance example typology (three ranges)

For any of the above designs, a generic solution is available around a common signal processing and full antenna design

framework. In all implementations, the temporal step tackles the problem of the Time Delay Estimation along the

antenna baselines, whereas the second step introduces the antenna geometry, in order to estimate the Direction of Arrival

per se. Then a detection module uses the temporal and spatial properties of the potential sources and propagation




situations to produce the additional detection information. Over the past 20 years, this generic spatio-temporal approach

has been applied to numerous situations, among them is the present helicopter detection case study.

For example in a combined scenario oriented to cover both the long and short range (100Hz to 10

2Hz), plus punctually

the third range (102Hz to 10

3Hz) an estimation of a global coverage of the Swiss territory would typically lead to two

different approaches 1) for long and short range (first two columns in Figure 13) around 100 permanent ground stations

i.e. some hundreds to a thousand of sensors (channels) that would be disseminated in a certain mesh and a possible

architecture of such a system could be a centralized signal processing service that would receive permanent continuous

real-time data from all sensors, and 2) for very-short range (third column in Figure 13) a second fixed or mobile network

of small acoustic nodes for example with a network of nodal stand-alone easily deployable antennas, with integrated

computation and communicating only low bandwidth vectored event detections on a private network such as a dedicated

LPWAN (Low-Power Wide-Area Network). For the permanent ground stations network for long and short range, the

main practical difficulty would probably be the finding of eligible long-term installation sites for a sufficient number of

stations. Nevertheless once such a contingency is solved, this system could be classified in the category of ‘network of

sensors’ and could build up a permanent source with very low maintenance for ‘big data’ and very long term acquisition

and analysis.

Figure 14. Multi-range ground acoustic surveillance example

7. TECHNOLOGICAL POTENTIAL OPTIMIZATION

Future works can be undertaken to enhance this two-step localization method. Indeed, the resolution of the second order

time delay estimator is limited to 1/B. If the distance between the sources or the path difference is so small that the time

delays induced in a sensor pair are below 1/B, super-resolution techniques have to be considered. The price to be paid

will be additional CPU load and a longer observation time. Nevertheless, one important weak point to be mentioned

concerning the multiple source tracking is: intrinsically, since the acoustic goniometry is passive and dominated by the

more powerful and coherent signal, a multiple-source tracking is possible to some extent but localization will in some

cases become utopic for the discrimination of one or several identical or similar and simultaneous sources. Anyway,

even without localization, one important observation of the as2r case study is that the simple DAQ dashboard already

produces very significant surveillance information for observers.

The second way to enhance the goniometry entails better knowledge of the propagation properties. The propagation

phenomena affecting the goniometry such as the wave vector deviation, the diffraction effects or the reflections, would

deserve specific work. This would enable us, either to develop a calibration procedure, or to determine the optimal

configuration and design in order to free the goniometry from these phenomena. It is worth noting here that the systems

in competition with the goniometer are submitted to the same difficulties. Yet, the goniometry presents the advantages of




using antennas with few sensors and small extensions. The advice that could be given when handling goniometry

problems would concern the spatio-temporal scales. The visualization of both spatial and temporal properties of the

propagating waves and their comparison to those of the goniometer, are indeed fundamental when tackling antenna

design. The signal’s central wavelength, the spreading and range of the source should always be compared to the

baseline lengths and the antenna extension. The period of the central frequency, the inverse bandwidth and the stationary

properties of the emission have to be related to the observation time. This should help finding on one hand a good

compromise between the search for precision and resolution by designing large baselines and, on the other hand the

realization of an unambiguous localization, which requires a condensed spatial sampling. The influence of geometry is

perhaps the most difficult to handle. It should be kept in mind that the relevant information for the localization is

extracted from the projections of the incoming wave vector on each baseline. In relation with the a priori information on

the source location, every potential DOAs should be tested in terms of spatial distribution of the projection inside the

antenna. The antenna should be oriented in order to present the maximum extension to the sector to cover, as the

baselines present higher performances in the Broadside directions. The propagation phenomena should be visualized at

the scale of the antenna. A perturbing phenomenon should always be handled as the superposition of a steady

component, which is homogenous in respect to the antenna and to very local components affecting each sensor

differently. Whereas the local perturbations deteriorate the localization performance, the steady perturbations do not

affect it but deliver in fine an apparent DOA, which can be very far from the source direction. Although it is difficult to

list all the potential applications, it is possible to say that every DOA application can be solved by this approach. The

question is then “Considering the operational context of my application, does the effort required to achieve the

performances justify the use of this two-step goniometry more than another technique?” This work has demonstrated that

the best results were obtained for the goniometry of a unique stationary localized source, generating broadband low pass

signals in a homogenous medium, without wind and noise: it is important to keep in mind that a judicious use of

goniometry requires the user to balance the operational conditions with the performances to achieve.

8. BIBLIOGRAPHY

[1] E. Van Lancker, “Acoustic goniometry: A spatio-temporal approach”, EPFL PhD Thesis, Presses Polytechniques

Romandes, 2002.

[2] D.H. Johnson and D.E. Dudgeon, “Array Signal Processing: Concepts and Techniques,” Prentice Hall Signal

Processing Series. Prentice-Hall, Englewood Cliffs, NJ, 1993.

[3] H. Krim and M. Viberg, “Two decades of array signal processing research,” IEEE Signal Processing Magazine,

vol. 13, no. 4, pp. 67-94, July 1996.

[4] V. Madisetti and D.B. Williams, “The Digital Signal Processing Handbook,” IEEE Press, 1998.

[5] B.D. Van Veen and K.M. Buckley, “Beamforming: A versatile approach to spatial filtering,” IEEE ASSP-

Magazine, pp. 4-24, April 1988.

[6] G.C. Carter, “Coherence and Time Delay Estimation: An Applied Tutorial for Research, Development, Test, and

Evaluation Engineers,” IEEE Press, Trade Cloth, April 1993.

[7] F.A. Reed, P.L. Feintuch, and N.J. Bershad, “Time Delay Estimation using the LMS Adaptive Filter-Static

Behaviour,” IEEE Trans. on Acoustics, Speech and Signal Processing, vol. 29, no. 3, pp. 571-577, June 1981.

[8] M. Omologo and P. Svaizer, “Acoustic Source Location in Noisy and Reverberant Environment using CSP

Analysis,” Proc. of ICASSP, Atlanta, GA USA, vol. 2, pp. 921-924, 1996.

[9] H.F. Silverman and S.E. Kirtman, “A two-stage algorithm for determining talker location from linear microphone-

array data,” Computer, Speech, and Language, vol. 6, no. 2, pp. 129-152, April 1992.

[10] M.S. Brandstein, “A Framework for Speech Source Localization Using Senor Arrays,” PhD thesis, Brown

University, Providence RI, May 1995.

[11] D.V. Rabinkin, “Optimum sensor placement for microphone array,” PhD thesis, The State University of New

Jersey, New Brunswick, 1998.

Documents

High infrasonic goniometry applied to the detection …...High infrasonic goniometry applied to the detection of a helicopter in a high activity environment Dr. Vincent Chritin (1),