Near Field Acoustic Holography Based on an Array of

Particle Velocity Sensors

Finn Jacobsena and Yang Liub

Acoustic Technology, Ørsted•DTU, Technical University of Denmark, Building 352,

Ørsteds Plads, DK-2800 Kgs. Lyngby, Denmark
a
fja@oersted.dtu.dk; bgordan69@163.com

Abstract Planar near field acoustic holography is usually based on measurement of the
sound pressure in a plane. This paper describes a theoretical and experimental investiga-
tion of an alternative technique that involves measuring the normal component of the
acoustic particle velocity in a plane. A simulation study shows that there is no appreciable
difference between the quality of predictions of the pressure based on knowledge of the
sound pressure in the measurement plane and predictions of the particle velocity based on
knowledge of this quantity in the measurement plane. However, when the particle velocity
is predicted close to the source on the basis of the sound pressure measured in a plane fur-
ther away, high spatial frequency components corresponding to evanescent modes are not
only amplified by the distance but also by the wavenumber ratio (kz/k). By contrast, when
the sound pressure is predicted close to the source on the basis of the particle velocity
measured in a plane further away, high spatial frequency components are reduced by the
reciprocal factor (k/kz). For the same reason holography based on the particle velocity is
less sensitive to transducer mismatch than the conventional technique based on the pres-
sure. Thus all in all a particle velocity array will generally perform better than a pressure
array. These findings are confirmed by an experimental investigation made with a p-u
sound intensity probe produced by Microflown.

1. INTRODUCTION

Near field acoustic holography is a powerful experimental technique for analysing sound
fields near sources and for deducing important information about the nature of the source. In
planar near field acoustic holography the sound field is reconstructed from measurements at
discrete positions in a finite region in a plane [1, 2]. Usually it is the sound pressure that is
measured, but in principle there is the same information in the normal component of the
acoustic particle velocity. The reason for the fact that conventional near field holography is
based on measurements of the sound pressure rather than the particle velocity is of course
that it is easier to measure the sound pressure that the particle velocity. However, an acoustic
particle velocity transducer called ‘Microflown’ is now available [3, 4]. The purpose of this
paper is to examine the potential of particle velocity-based near field acoustic holography by
comparing its performance with the performance of conventional pressure-based holography.

2. THEORETICAL BACKGROUND

Planar near field acoustic holography is based on the fact that the sound pressure in one plane
can be expressed as the two-dimensional convolution of a ‘propagator’ and the sound pres-
sure in another plane. The theory was developed in the mid 1980s [1, 2]. Only a brief over-
view of the most fundamental relations will be presented here; see eg ref. [5] for more details.

The complex sound pressure is measured in a plane ( z = zm ) near the source under examina-
tion and a 2D spatial Fourier transform is calculated. The result is the wavenumber spectrum,
∞ ∞
P (k x , k y ) = ∫ ∫
j( k x x + k y y )
p ( x, y, zm )e dxdy . (1)
−∞ −∞

Since the pressure is given by the inverse 2D Fourier transform,

1 ∞ ∞
∫ ∫
− j( k x x + k y y )
p ( x, y , z m ) = P(k x , k y )e dk x dk y , (2)
(2π) 2 −∞ −∞

it can be seen that the wavenumber spectrum P (k x , k y ) for any given value of (k x , k y ) inside
the radiation circle (that is, when k x2 + k y2 ≤ k 2 ) may be interpreted as the amplitude of a
plane wave that propagates in the (k x , k y , k z ) direction,
− j( k x x + k y y + k z z )
P (k x , k y )e . (3)
Outside the radiation circle it is the amplitude of an evanescent wave. Thus the wavenumber
transform decomposes the sound field into plane and evanescent waves. The z-component of
the wavenumber is
⎧ k 2 − k 2 − k 2 for k 2 + k 2 ≤ k 2 ,
⎪ x y x y
kz = ⎨ (4)
⎪⎩− j k x2 + k y2 − k 2 for k x2 + k y2 > k 2 ,

where the sign in the last equation has been chosen so as to satisfy the Sommerfeld radiation
condition with the e jωt sign convention. It now follows that one can determine the wavenum-
ber spectrum of the sound pressure in another, parallel plane, z = zp , simply by multiplying
the wavenumber spectrum in the measurement plane with an exponential ‘propagator’,
− jk z ( zp − zm )
P(k x , k y , zp ) = P(k x , k y )e . (5)

Eventually one can calculate the sound pressure in the prediction plane z = zp by an inverse
Fourier transform,
1 ∞ ∞
∫ ∫
− j( k x x + k y y )
p ( x, y , z p ) = P(k x , k y , zp )e dk x dk y . (6)
(2π) 2 −∞ −∞

One can also calculate the particle velocity vector, and thus the sound intensity. The normal
component of the particle velocity is
1 ∞ ∞ k z − j( kx x + k y y )
u z ( x, y , z p ) =
(2π) 2 ∫ ∫
−∞ −∞
P ( k x , k y , zp )
ρ ck
e dk x dk y . (7)
Alternatively, one can measure the normal component of the particle velocity in the plane
near the source ( z = zm ), transform to the wavenumber domain, multiply with the propagator,
and transform back to the spatial domain. The result is the normal component of the particle
velocity in the prediction plane. If the wavenumber spectrum is multiplied by ρ ck k z before
the inverse Fourier transform is carried out one will get the sound pressure in the prediction
plane.

The above description is, of course, grossly oversimplified. In practice the measurement area
must obviously be finite, the sound field is sampled only at a finite number of discrete posi-
tions, and the 2D spatial Fourier transform is approximated by a 2D FFT. To reduce the in-
fluence of the finite aperture and the spatial sampling, zero padding and a tapered spatial
window should applied before the spatial Fourier transform is calculated. The spatial sam-
pling must satisfy the sampling theorem, one cannot sample very close to the source (where
there may be high spatial frequency components), and one should not sample too far away,
because then the evanescent modes will be buried in noise. Moreover, whereas forward pre-
diction (in which zp > zm ) is numerically stable, backward prediction, which involves pre-
dicting the sound field closer to the source than the measurement plane, is an unstable ‘ill-
posed’ inverse problem that requires regularisation because of the fact that the evanescent
waves are amplified exponentially with the distance. The standard regularisation technique
involves spatial low-pass filtering (multiplying with a window in the wavenumber domain).
See eg refs. [1, 2, 5, 6].

3. A SIMULATION STUDY

Initially a simulation study was carried out. To obtain realistic results, a model of a relatively
complicated source, a simply supported vibrating steel plate in an infinite, rigid baffle, was
developed. The dimensions of the plate were 0.45×0.75 m, and it was 5 mm thick. The criti-
cal frequency was about 2.4 kHz. The plate was driven by a harmonic point source placed
close to one of the corners, and this was modelled by a conventional modal summation. The
vibrating plate was divided into 64×64 rectangles, each of which was regarded as a point
source on the baffle, and the sound pressure radiated by the plate was calculated from the
corresponding approximation to Rayleigh’s first integral [5]. A virtual microphone array of
1.5×1.5 m with 32×32 microphones (corresponding to a sampling distance of 4.7 cm) was
placed 15 cm above the plate. The spatial window was an 8-point Tukey window [5], and
zeros were added outside the measurement area so as to reduce wrap-around errors [1]. The
regularisation was made using an exponential low-pass window [5]. On the radiation circle
the factor ρ ck k z goes to infinity, and therefore the wavenumber spectra were smoothed as
described in ref. [7]. The estimation plane was 5 cm above the plate. The calculations were
made at discrete frequencies with a resolution of 1 Hz, and the results were summed in one-
third octave bands.

Figure 1(a) shows the ‘true’ pressure in the 200-Hz band along a line across the prediction
plane, calculated directly from the approximated Rayleigh integral, a prediction based on the
pressure in the measurement plane, and a prediction based on the normal component of the
particle velocity in the measurement plane. There is good agreement between the ‘true’ pres-
sure and both predictions in a significant part of the prediction plane, but near the edges the
agreement deteriorates.
 85 130

80
120
Sound pressure level (dB re 20 uPa)

Particle velocity level (dB re 1 nm/s)

75
110
70
100
65

60 90

55
80
50
True pressure 70
Based on pressure
45 True velocity
Based on velocity
60 Based on velocity
40 Based on pressure

35 50
−0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8
Position Position

Figure 1: ‘True’ and predicted sound pressure (a) and particle velocity (b) in the 200-Hz one-third octave band.

Figure 1 (b) shows a similar comparison of the ‘true’ normal component of the particle veloc-
ity and predictions based on the pressure and based on the particle velocity in the measure-
ment plane. The two predictions are in good agreement with the ‘true’ value’, but it is appar-
ent that the velocity-based prediction is far better than the pressure-based prediction near the
edges. Similar results have been seen in other frequency bands. Note also that the ‘true’ par-
ticle velocity, not unexpectedly, decays faster toward the edges and has a larger dynamic
range than the pressure. This is favourable for holography based on the particle velocity,
since it means that leakage cause by the spatial window is reduced. However, the main rea-
son for the consistent observation that velocity-based predictions of the pressure are consid-
erably better than pressure-based predictions of the velocity is probably the fact that the fac-
tor ρ ck k z used in the former case reduces high spatial frequencies whereas the reciprocal
factor of k z ρ ck used in the latter case amplifies them. Amplification of high spatial fre-
quencies increases the inherent numerical instability in backward propagation.

Postive intensity
Intensity level (dB re 1 pW/m2)

50

40

30

20
−0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8
Position
Negative intensity
Intensity level (dB re 1 pW/m2)

20

30

True intensity
40
Based on pressure
Based on velocity
Based on same quantities
50
−0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8
Position

Figure 2: ‘True’ and predicted sound intensity in the 200-Hz one-third octave band.

Figure 2 shows the ‘true’ normal component of the sound intensity in the 200-Hz band along
a line across the prediction plane, a prediction based the pressure in the measurement plane, a
prediction based on the corresponding normal component of the particle velocity, and a pre-
diction in which the pressure in the expression for the intensity has been calculated from the
pressure and the particle velocity in the same expression has been calculated from the particle
velocity. Note the low level compared with the sound pressure level, and note the regions of
negative intensity; the panel is a very inefficient radiator of sound at 200 Hz. It seems that the
prediction where the pressure has been based on the pressure and the velocity on the velocity
is slightly better than the prediction based solely on the velocity and much better than the
prediction based solely on the pressure. Similar observations have been made in other fre-
quency bands.
85 140

80 130
Sound pressure level (dB re 20 uPa)

Particle velocity level (dB re 1 nm/s)

75
120
70
110
65
100
60
90
55 True pressure
True velocity
Based on pressure 80
50 Based on velocity
Based on velocity
Based on pressure
Mag1
70 Mag1
45 Mag2
Mag2
Mag3
Mag3
40 Mag4 60
Mag4
Mag5
Mag5
35 50
−0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8
Position Position

Figure 3: ‘True’ and predicted sound pressure (a) and particle velocity (b) in the 200-Hz one-third octave band.
In (a) random amplitude mismatch has been introduced in the velocity ‘measurements’, and in (b) random
amplitude mismatch has been introduced in the pressure ‘measurements’.

85 140

80 130
Sound pressure level (dB re 20 uPa)

Particle velocity level (dB re 1 nm/s)

75 120

70
110
65
100
60
90
True velocity
55 True pressure Based on velocity
Based on pressure 80 Based on pressure
50 Based on velocity Phase1
Phase1 70 Phase2
45 Phase2 Phase3
Phase3 Phase4
Phase4 60 Phase5
40
Phase5
35 50
−0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8
Position Position

Figure 4: ‘True’ and predicted sound pressure (a) and particle velocity (b) in the 200-Hz one-third octave band.
In (a) random phase mismatch has been introduced in the velocity ‘measurements’, and in (b) random phase
mismatch has been introduced in the pressure ‘measurements’.

In real measurements with transducer arrays a certain amount of amplitude and phase mis-
match between the transducers can be expected. To examine the effect of such measurement
inaccuracies random amplitude mismatch, uniformly distributed between -0.5 dB and 0.5 dB,
has been introduced in the ‘cross predictions’, that is, in the predictions where the pressure is
calculated from the particle velocity and the particle velocity is calculated from the pressure.
Figure 3 shows the results of five outcomes of this stochastic experiment. It is apparent that
predictions from pressure to velocity are far more seriously affected than predictions from
velocity to pressure. The explanation is that the white noise introduced in the wavenumber
spectrum by these errors is amplified by the factor k z ρ ck in the pressure-based cross pre-
dictions and reduced by the reciprocal factor in the velocity-based cross predictions.

Figure 4 shows the results of similar stochastic experiments in which phase errors uniformly
distributed from -2° to 2° have been introduced in the cross predictions. The results are simi-
lar to the results shown in figure 3, but this amount of random phase mismatch appears to be
slightly less serious than random amplitude mismatch in the interval from -0.5 dB to 0.5 dB.
−5 −5
x 10 x 10
10 2
True intensity True intensity
Mag1 Mag1
8 Mag2 Mag2
Mag3 1.5 Mag3
Mag4 Mag4
6 Mag5 Mag5
Linear scale (W/m )

Linear scale (W/m2)

2

2 0.5

0
0
−2

−0.5
−4

−6 −1
−0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8
Position Position
−5
x 10
7
True intensity
6 Mag1
Mag2
Mag3
5
Mag4
Mag5
4
Linear scale (W/m2)

−1

−2

−3
−0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8
Position

Figure 5: ‘True’ sound intensity in the 200-Hz one-third octave band compared with (a) predictions based
solely on the pressure, (b) predictions based solely on the particle velocity, and (c) predictions based both on
the pressure and the particle velocity. Random amplitude mismatch has been introduced in all the ‘measure-
ments’.

Finally Figure 5 shows the effect of amplitude mismatch uniformly distributed from -0.5 dB
to 0.5 dB on predictions of the sound intensity. Whereas the predictions based exclusively on
pressure measurements are seriously affected the velocity-based predictions are reasonably
accurate, and somewhat better than predictions based both on both pressure and velocity.

4. EXPERIMENTAL RESULTS

To examine the validity of the results of the simulations some experiments have been carried
out. The sound source was a 3-mm steel plate with dimensions of 39 × 63 cm mounted in a
large baffle. The radiating plate was one of the side plates of a box; and one of the other side
plates of the box was driven via a Brüel & Kjær (B&K) 8200 force transducer by a B&K
4809 electrodynamic exciter fed with pink random noise. The box was suspended from a
support in such a way that it did not touch the rectangular hole in the baffle; the resulting
leaks around the plate were sealed by tape. The frequency responses between sound pressure
and the force signal (from a B&K 2635 charge amplifier) and between the normal component
of the particle velocity and the force signal were measured at 18 × 28 points in two planes
either using a single Microflown ½-inch p-u sound intensity probe or using an array of such
probes; see figure 6.

Since the measurements took place in an ordinary room the image source resulting from the
reflecting floor was taken into account; that is, the sound field was sampled down to one half
sampling distance above the floor, and the spatial transform was a 18 × 56 point transform.
The phase and amplitude calibration of the Microflown particle velocity transducers was car-
ried out in an impedance tube as described in ref. [8]. A robot moved the single transducer or
the array over the two measurement planes, one typically about 12 cm from the source plane,
and one typically about 4 cm from the source plane. A B&K 3560 ‘Pulse’ analyser in the
1/24 octave mode was used, and the measurements and the robot were controlled by B&K’s
STSF program. The data were, however, postprocessed using MATLAB with the same routines
as used in the simulations. Each set of 1/24 octave data was processed individually and then
summed so as to produce the sound pressure, the particle velocity and the sound intensity in
one-third octave bands.

Figure 6: The Microflown array in front of the steel panel in the baffle.
Figure 7 shows an example of the results determined using the single Microflown transducer.
Figure 7 (a) shows a comparison of the ‘true’, that is, directly measured sound pressure in the
plane close to the panel along a vertical line and predictions based on measurements of the
pressure and the normal component of the particle velocity in the plane further away. (The
entire prediction plane, doubled by the image source is shown, hence the symmetry about the
mid position.) It can be seen that velocity-based prediction is acceptable and somewhat more
accurate than the pressure-based prediction.

Figure 7 (b) shows the ‘true’ particle velocity in the plane close to the panel and predictions
based on measurements of the pressure and the particle velocity in the plane further away. It
is clear that the velocity-based prediction is by far the best.

Finally figure 7 (c) shows the ‘true’ sound intensity in the plane close to the panel and a pre-
diction based solely on pressure measurements in the plane further away, a prediction based
solely on particle velocity measurements in the same measurement plane, and a prediction
based on both pressure and particle velocity measurements. Again the velocity-based predic-
tion performs much better than the prediction based on measurements of the pressure, and
similar to a prediction based on measurements of both quantities.
80 130

75
120
Sound pressure level (dB re 20 uPa)

Particle velocity level (dB re 1 nm/s)

70
110

65
100
60

90
55
True pressure
Based on pressure True velocity
80
50 Based on velocity Based on velocity
Based on pressure

45 70
−1.5 −1 −0.5 0 0.5 1 1.5 −1.5 −1 −0.5 0 0.5 1 1.5
Position Position
1
True intensity
Based on pressure
0.8 Based on velocity
Based on Microflown

0.6
Linear scale (W/m2)

0.4

0.2

−0.2

−0.4
−1.5 −1 −0.5 0 0.5 1 1.5
Position

Figure 7: Measured sound pressure (a), particle velocity (b), and sound intensity (c) the 200-Hz one-third oc-
tave band compared with predictions based on the pressure and on the particle velocity in the measurement
plane.
 5. CONCLUSIONS

Planar near field acoustic holography has traditionally been based on measurements of the
sound pressure, but since an acoustic particle velocity transducer is now available this paper
has examined near field holography based on measurements of the normal component of the
particle velocity.

A simulation study has revealed that the particle velocity decays faster towards the edges of
the measurement region than the sound pressure and has a larger dynamic range; thus spatial
windowing has less serious consequences on velocity-based holography. Nevertheless, it has
generally been observed that the quality of pressure-to-pressure predictions is similar to the
quality of velocity-to-velocity predictions. However, velocity-to-pressure predictions are far
better than pressure-to-velocity predictions because of the fact that the wavenumber ratio that
enters into such cross predictions reduces high spatial frequencies in the former case but am-
plifies them in the latter case.

For the same reason amplitude and phase mismatch, which is likely to occur in measurements
with arrays of transducers, have a far more serious influence on pressure-to-velocity predic-
tions than on velocity-to pressure predictions, because such transducer mismatch introduces
high spatial frequencies, and the resulting errors are amplified exponentially with the distance
if the prediction plane is closer to the source than the measurement plane.

These findings have been confirmed by an experimental study in which the sound pressure
and the normal component of the particle velocity were measured at some distance from a
vibrating, baffled steel panel with an array of Microflown p-u sound intensity probes, and
used to predict the pressure, the normal component of the particle velocity, and the normal
component of the sound intensity in a plane closer to the panel. The velocity-based predic-
tions were consistently found to be better than the pressure-based predictions.

ACKNOWLEDGEMENTS

The authors would like to thank Microflown Technologies for lending us an array of p-u
sound intensity probes. We are also indebted to Brüel & Kjær for lending us a robot and a
multi-channel ‘Pulse’ front-end.

REFERENCES

[1] J.D. Maynard, E.G. Williams and Y. Lee, Nearfield acoustic holography: I. Theory of
generalized holography and the development of NAH, Journal of the Acoustical Soci-
ety of America 78, 1395-1413, 1985
[2] W.A. Veronesi and J.D. Maynard, Nearfield acoustic holography (NAH) II. Holo-
graphic reconstruction algorithms and computer implementation, Journal of the Acous-
tical Society of America 81, 1307-1322, 1987
[3] R. Raangs, W.F. Druyvesteyn and H.-E. de Bree, A low-cost intensity probe, Journal of
the Audio Engineering Society 51, 344-357, 2003
[4] H.-E. de Bree, The Microflown: An acoustic particle velocity sensor, Acoustics Austra-
lia 31, 91-94 2003
[5] E.G. Williams, Fourier Acoustics – Sound Radiation and Nearfield Acoustical Holo-
graphy, Academic Press, San Diego, 1999
[6] J. Hald, Beamforming and wavenumber processing, in Handbook of Signal Processing
in Acoustics, Springer Verlag, new York (in course of publication)
[7] E.G. Williams and J.D. Maynard, Numerical evaluation of the Rayleigh integral for
planar radiators using the FFT, Journal of the Acoustical Society of America 72, 2020-
2030, 1982
[8] F. Jacobsen and H.-E. de Bree, Measurement of sound intensity: p-u probes versus p-p
probes. Proceedings of NOVEM 2005, Saint Raphaël, France, 2005