You are on page 1of 16

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225000996

Noise Source Analysis of an Aeroengine with a new Inverse Method SODIX

Conference Paper · May 2008


DOI: 10.2514/6.2008-2860 · Source: DLR

CITATIONS READS

24 217

2 authors, including:

Ulf Michel
CFD Software Entwicklungs- und Forschungsgesellschaft mbH, Berlin
263 PUBLICATIONS   1,224 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Source location with microphone arrays View project

Wind-tunnel unsteadiness View project

All content following this page was uploaded by Ulf Michel on 23 June 2014.

The user has requested enhancement of the downloaded file.


14th AIAA/CEAS Aeroacoustics Conference (29th AIAA Aeroacoustics Conference) AIAA 2008-2860
5 - 7 May 2008, Vancouver, British Columbia Canada

Noise Source Analysis of an Aeroengine


with a New Inverse Method SODIX
Ulf Michel ∗ and Stefan Funke †
DLR, German Aerospace Center, Berlin, Germany

The directivities of all broadband sound sources of an aeroengine can be determined from measurements
with a line array of microphones. A new data analysis method SODIX (SOurce DIrectivity modelling in cross-
spectral matriX) was developed for this purpose and applied to the experimental data obtained with a high-
bypass ratio turbofan engine in an open air test bed. The method is based on modelling the matrix of the cross-
spectra of the microphone signals with a set of contributions from point sources with unknown directivities
assumed in positions along the engine axis. The source positions are defined with a separation of typically
0.4 wave lengths. The analysis has to be performed in narrow frequency bands. The influences of tones are
eliminated in the cross-spectral matrix of the microphone signals by interpolation in the frequency domain.
The unknown directivities of the point sources are determined with the side conditions that the values must
be non-negative and that the directivities and source distributions are smooth. The results of this non-linear
optimisation problem confirm that the various sources of an aeroengine have highly non-uniform directivities.
The far field can be predicted from the source distributions for positions over a large range of emission angles.
The predictions on a 45.72 m radius around the engine agree within about 1 dB with experimental data. These
far-field directivities can be separated into the contributions from inlet, casing, the two nozzles, and the jet.

I. Introduction
objective of this paper is the analysis of the sound sources of an aeroengine during noise tests in an open air test
T HE
bed. A proven method to accomplish this is the polar correlation technique by Fisher, Harper-Bourne, and Glegg
1 2
(1977) and by Tester and Fisher (1981). The signals of a microphone array installed on a polar arc (generally around
the exhaust nozzle of the engine) are evaluated with this method. The technique yields the strengths of point sources
on the engine and constants for a parametric model of jet mixing noise. A prediction of the far-field radiation is not
possible from these results because the directivities of the sources are not known for all emission angles.
Another common method for the localisation of sound sources is beamforming. The results are presented in
form of beamform maps, which are, mathematically, convolutions between the true source distributions and the point
spread functions of all sources.3 Beamforming with a line array of microphones was first applied to a turbofan engine
by Siller et al. (2001),4 primarily to study the core noise of an engine. A sliding subarray was used by them to study
the directivities of the sources. An estimation of the source strengths from the beamforming maps is possible only
in rare cases, e.g., when the source positions are spatially sufficiently separated. Sources along a line or distributed
over an area or over a source volume yield results that depend on the beam widths of the point-spread functions and
on sidelobes and aliases. Additional problems arise when the source directivities are non-uniform instead of uniform.
The consequence is that amplitudes of sound sources are very difficult to estimate from beamform maps. Siller et
al. (2001)4 tried to estimate the contribution of combustion noise and solved this problem by evaluating the signals of
a rumble probe in the combustor together with the time series of the beamform signal focussed on the nozzle exit.
The problem of estimating the source strengths can be solved with deconvolution techniques in which it is assumed
that sources are point monopoles with uniform directivities. The point spread functions of the array (or beamform maps
of the point sources) are calculated for every possible source position and for each narrow-band frequency of interest.
The source levels of the unknown sources have then to be determined with a least squares fit for the difference between
the measured beamform map and the estimate for the sum of point sources. This deconvolution procedure yields huge
and often poorly conditioned matrices. Special iterative procedures are required to solve them with the side condition
that only positive source levels are permitted.
A first procedure proposed to solve the complete inverse problem for beamform maps of microphone arrays
was published by Brühl and Röder (2000).5 The method was called source density modelling (SDM). Brooks and
Humphreys (2004-2006)6–9 developed a solution procedure called DAMAS and Dougherty (2005)10 proposed a sim-
plification DAMAS2, where a common point spread function is assumed for all sources as an approximation with
∗ Senior Scientist, DLR, German Aerospace Center, Institute of Propulsion Technology, Engine Acoustics Department, Müller-Breslau-Str. 8,
10623 Berlin, Germany, Honorary Professor Aeroacoustics, BTU Cottbus, Germany, ulf.michel@dlr.de, Member AIAA and DGLR
† Graduate student at TU Berlin

1 of 15

American
Copyright © 2008 by Ulf Michel. Published by the American Institute Instituteand
of Aeronautics of Astronautics,
Aeronautics and
Inc.,Astronautics
with permission.
the objective of reducing the computing time. The deconvolution of the beamform maps of moving sources is more
difficult, because the sidelobes of the point spread functions have frequencies that are different from the main lobe.
Guérin et al. (2006)11 proposed a method to compute an average point spread function for the broadband noise of
moving sources with the condition that the narrow-band levels of the sources are constant in neighbouring frequency
bands. An application of this procedure to flyover experiments is presented by Guérin & Siller (2008).12 Tones require
a different treatment as was shown by Guérin & Weckmüller (2008).13
Sijtsma14–16 developed a different approach (CLEAN-SC) to solve the inverse problem. He makes use of the fact
that the sidelobes in the beam pattern of a source are spatially correlated with the main lobe while the main lobes of
other sources are uncorrelated. CLEAN-SC iteratively removes the part of the source plot, which is spatially coherent
with the peak source and attributes it to a source of known strength. There is no need to compute point spread functions.
The resulting procedure is computationally rapid in contrast to the previously mentioned deconvolution methods.
Blacodon and Élias17, 18 and Blacodon19 used a different approach to determine the strengths of sources in assumed
positions. Their method (SEM) is based on modelling the cross-spectral matrix of the microphone signals. Instead
of determining the sources such that the beamform maps are best approximated, Blacodon and Élias determine the
sources for a best fit of the cross-spectral matrix. They generate a cross-spectral matrix for each possible point source
position assuming uniform directivities of the sources. The amplitudes of the sources are then determined with a least
squares fit between the modelled and the measured cross-spectral matrix. The beamforming map is not required for
their method.
It is the purpose of the present paper to develop a procedure for the determination of the source directivities based
on the cross-spectral matrix of a line array of microphones. The method of Blacodon and Élias18 is extended to include
the directivities of the sources. The method is described in section II. This section also explains, how the influence
of tones is removed from the cross-spectral matrix to restrict the investigation to the broadband part of engine noise.
The procedure is then applied to data that were acquired during a test of a modern high-bypass ratio turbofan engine
in an outdoor test facility. The measured sound pressure levels of the microphones in the line array are compared with
the model results in narrow bands for six frequencies from 100 Hz to 3150 Hz in octave steps in section III. Results
for the source distribution along the engine axis are shown in section IV for six one-third octave bands ranging from
100 Hz to 3150 Hz and for three emission angles 60, 90, and 120 deg relative to the centre of gravity of the engine
and the forward direction. Finally, the directivities are predicted in section V for the same six one-third octave bands
in the positions of the far-field microphones, which are located on a radius with 45.72 m around the centre of the test
rig. These far-field results are compared with measured data kindly made available by Rolls-Royce. The directivities
are separated into the contributions from the inlet, the casing, the two nozzles, and the jet.

II. Methods
A. Set-up of microphone array
The test rig is shown in figure 1. The sound pressure is measured with a line array of microphones laid out on the
ground roughly parallel to the engine axis and spanning from positions far upstream of the intake to positions far
downstream. The microphones are visible in figure 1. The line array spans from the background on the left side of the
figure, passes the turntable and ends on the right side of the bottom edge of the figure. Some microphones near the
supporting structure of the inlet control device will likely be affected by sound waves scattered from these structures.
The microphone positions are shown in figure 2 in relation to the location of the engine. The distance between
the line array and the engine centre-line is about 9.5 m. The microphones of the array are in the acoustic far field of
the sources but in the geometric near field of the whole source distribution. The axial spacing of the microphones is
217 mm in the centre. In front of the intake and behind the primary nozzle the spacing increases such that the emission
angle changes by 1.25 deg between two microphones. The emission angle of the most forward microphone is 25 deg
relative to the inlet, the angle of the most rearward one is 146.25 deg relative to the primary nozzle.
The microphones consisted of electret microphone capsules (type Sennheiser KE4211-2), which were combined
by DLR with the necessary preamplifiers. The microphone capsules were installed in stainless steel tubes with an outer
diameter of 7 mm, which equals the diameter of the protecting grid of 1/4” measuring microphones and allows the use
of standard calibration equipment. 128 microphones were installed, of which M = 127 usable signals were available
for the analysis. The microphone signals were sampled with a frequency of 32786 Hz. The available signal length was
approximately 30 seconds and the FFT time segment length was chosen 8192 samples, yielding a segment length of
T = 0.2499 s and a frequency interval of ∆f = 4.002 Hz. The fast Fourier transforms (FFT) were performed with
a Hanning window and a 50% overlap, resulting in an averaging over about 240 time segments and a subsequently
statistically very stable cross-spectral matrix. Each frequency band consists of a matrix Cmn of M by M complex
values. The Matrix is Hermitian, Cmn = Cnm ∗
, where the ∗ indicates the conjugate complex element.
The engine was investigated in a range of power settings from almost idle to maximum power. The results shown
in this paper are for a test point in which the tones at multiples of the fan shaft frequency start to appear.

2 of 15

American Institute of Aeronautics and Astronautics


Figure 1: Open air noise test facility of Rolls-Royce plc in Hucknall. The large ”Golfball” is an Inflow Control Device
(ICD), a flow straightener to ensure a smooth inflow to the engine.

10
Microphones
y−Position / m

−5
−15 −10 −5 0 5 10 15 20 25
x−Position / m

Figure 2: Positions of the microphones in the horizontal plane of the coordinate system of the turbofan engine. The
line array is longer in the forward arc (x > 0) than in the rear arc. The jet extends substantially past the left end of
the array.

B. Definition of the source positions


It is assumed that acoustic sources are located in known axial positions along the centerline of the tested aeroengine. In
the jet region, the lateral source positions are assumed on the extended centerline of the engine. In the engine region,
the positions are assumed on the surface of the casing. The positions in front of the inlet are located on a cylindrical

3 of 15

American Institute of Aeronautics and Astronautics


extension of the inlet to reduce parallax effects when analysing the sources from different directions. The spacing
between the sources is chosen to be 0.4λ for frequencies 200 Hz ≤ f < 1600 Hz, where λ is the wavelength of the
narrowband or of the reference frequency of the one-third octave band to be analysed. The spacing for f = 200 Hz
is used for all smaller frequencies. A spacing of 0.8λ was chosen for frequencies f ≥ 1600 Hz to limit the number
of sources and consequently the computing time. The choice of 0.4 wavelengths was originally made to satisfy
an anticipated spatial sampling requirement for a continuous source distribution. However, it turned out that larger
spacings (and consequently fewer source locations) can be selected for higher frequencies without compromising the
quality of the results. The reason is that the phase difference of the resulting cross spectra of microphone signals
changes only little with source position. The source positions for the results shown in this paper start at 3.74 m in
front of the intake and end at 1 m+6300 m/(f /Hz) downstream of the primary nozzle. The downstream positions are
limited to −25 m. The resulting source counts are 50 for f ≤ 200 Hz, 100 for 400 Hz, 115 for 800 Hz, 92 for 1600 Hz,
and 157 for 3150 Hz.

C. Evaluation of the source distribution


Blacodon and Élias18 compared the measured matrix Cmn with a modelled matrix consisting of the sum of the matrices
generated by each of the J unknown point sources, which yields the following equation for the modelled cross-spectral
matrix,
X J
mod
Cmn = gjm Sj gjn

, m, n = 1 . . . M (1)
j=1

where
gjm = eikrjm /rjm (2)
are the steering vectors between the source positions ξj (j = 1 . . . J) and the microphone positions xm (m = 1 . . . M ),
rjm = |ξj − xm | . (3)
Uniform directivities of the sources Sj are assumed in equation (1). The steering vectors determine the phase of the
cross spectra.
The goal is to determine the strengths Sj of the J sources such that the mean square error F (S) between the
measured and the modelled matrix becomes a minimum.
2
M
X XJ

F (S) = Cmn −
g S g ∗
jm j jn (4)
m,n=1 j=1

The condition for a minimum of F (S) is


∂F (S)
= 0, for 1≤j≤J (5)
∂Sj
which yields the set of J linear equations
J
X
Vij Sj = Ui , for 1 ≤ i ≤ J. (6)
j=1

Ui and Vij are given by18


M
X
Ui = gim Cmn gin

, (7)
m,n=1
M 2
X

Vij = gim gjm . (8)

m=1
The exact solution of equation (6) may yield a solution vector Sj with negative source levels for some of the source
positions ξj . In order to ensure that all Sj are positive, Blacodon and Élias18 replaced Sj in equation (4) by α2j and
searched for the minimum of equation
2
M J
X X
2 ∗
F (α) = Cmn −
g jm j jn .
α g (9)
m,n=1 j=1

4 of 15

American Institute of Aeronautics and Astronautics


This is a non-linear optimisation problem, which was solved with a procedure published by Shanno & Phua.20
Instead of solving the non-linear problem, one can alternatively solve the linear problem (4) with the constraint
that the Sj must be real and non-negative. This can be achieved with a slightly modified version of the Gauss-Seidel
procedure that was used by Brooks & Humphreys6 in DAMAS. The modifications consider that a linear set of complex
rather than real equations has to be solved and that the solution must be real.

D. Necessity to consider the directivities of the sources and source model for jet mixing noise
The radiations from the intake and the nozzles of an aeroengine are highly directive, which violates the assumption
of uniform directivities in equation (4). An additional problem is jet mixing noise, a volume source, in which the
sources are correlated over a considerable volume. It might be questioned if these sources can be described in terms
of independent point sources. However, this is in fact possible as was shown by Michel (2007),21 based on work of
Michalke (1977)22, 23 and Michel & Michalke (1981).24 The power-spectral density of the sound pressure of a jet in
a microphone position xm in the acoustic far field can be expressed as an integral over uncorrelated source volume
elements in the source positions ξj ,
Z
1 Wss (ξj , f )
Wpp (xm , f ) = Fs dV (ξj ). (10)
(4πa20 )2 r2
V

Wss is the power-spectral density of the source term in position ξj .21 r is the distance between ξj and xm . a0 is
the speed of sound in the ambiance of the jet. Fs is an interference integral, which considers the effects of source
interference within a coherence volume Vc in the source region in the vicinity of source position ξj .
Z s
r(ξj ) Wss (ξj + ηj )
Fs = γs cos(ψs + ψr ) dVc (ηj ), (11)
r(ξj + ηj ) Wss (ξj ) |{z} | {z }
Vc | {z }| {z } source coherence source interference
distance source strength

The extend of the coherence volume Vc and the necessary range of the separation vector ηj are defined by the region in
which the source coherence γs > 0. The two phases ψs and ψr describe the phase difference due to source convection
and due to retarded time differences between the two source positions ξj and ξj + ηj . Details can be found in Michel
(2007).21 The quantities r, Wss , γs , ψr , and possibly also ψs are functions of the microphone position, in addition
to the dependence on source position shown in equations (10) and (11). The influence of source interference on Fs
depends on how the coherence function γs decays with increasing source separation ηj and on the phase ψs , which
is primarily a function of the phase velocity of the disturbances in the jet. The interference integral Fs is highly
directional and the main cause for the directivity of jet mixing noise. The distance and strength correction terms may
be neglected if the coherence volume is small enough. Fs reaches a maximum value Fs,max when ψs + ψr = 0. This
describes the situation of Mach-wave radiation. Fs < Fs,max in all other directions, and Fs can reach small fractions of
Fs,max for an emission angle of 90 deg and in the forward arc, in qualitative agreement with experimental directivities
of jet mixing noise.
It can be concluded that jet noise sources can be modelled correctly by uncorrelated point sources if their directiv-
ities due to source interference are considered in the model.

E. Modelling of directivities with a sliding subarray


It was already proposed by Siller et al. (2001)4 to use a sliding subarray in beamforming to evaluate the directivities
of the sources. Blacodon & Élias18 proposed to determine the directivities of the sources by changing the position of
the microphone array and Blacodon19 demonstrated that this approach is successful. A uniform directivity is assumed
for all microphones of the array in each array or subarray position. The same procedure is applied here to yield first
estimates for the directivities of all sources. A subarray of 21 microphones is moved in increments of 5 microphone
positions over 126 microphones of the line array, yielding 22 subarrays and 22 directivity values for each of the J
sources. The directivity value for any required angle is then obtained by interpolation. Outside the available range of
angles the directivity values are assumed to be constant. The assumption of uniform directivities might be acceptable
for a subarray, it is not permitted for the whole array.
One problem with subarrays is that the linear set of equations (6) becomes less well-conditioned due to the smaller
number of microphones in the subarray yielding a smaller number of elements in the subset of the cross-spectral matrix.
The problem becomes even unsolvable if the number J of sources is larger than the number M of microphones. The
situation can be rectified by defining the smoothing function
J−1
X 2
Gs (D) = (Sj − 0.5(Sj−1 + Sj+1 )) (12)
j=2

5 of 15

American Institute of Aeronautics and Astronautics


and adding it to equation (4), which yields

G(D) = F (D) + σGs (D), (13)

where the slack variable σ has to be optimised experimentally. This equation has to be minimised instead of equation
(4). The effect of Gs (D) is a smoothing of the distribution of sources along the x-axis. Gs = 0 if the source levels of
all sources are equal to the arithmetic mean of the two neighbouring sources.
A minimum of G(D) requires that

∂G(D) ∂F (D) ∂Gs


= +σ = 0, (j = 1 . . . J) (14)
∂Sj ∂Sj ∂Sj

The partial derivative of equation (12) yields

∂Gs (D)
= −Sj−1 + 2Sj − Sj+1 , (j = 2 . . . J − 1) (15)
∂Sj

Equation (6) has to be modified accordingly.


Results obtained with this procedure were shown by Michel & Funke (2008).25

F. Evaluation of the source directivities (SODIX)


Instead of a source strength Sj with uniform directivity the directivities of the sources can be included in the modelling
of the cross-spectral matrix with the following definition.
J
X
mod
Cmn = gjm gjn

Djm Djn , (16)
j=1
| {z } | {z }
complex real

where Djm are the directivities of the sound pressures of the sources j (j = 1 . . . J) toward the microphones m
(m = 1 . . . M ) for a reference distance. Equation (2) for the steering vectors gjm is extended to include a uniform
mean flow Mach number Mf in the region between engine and line array as a first approximation for the entrainment
flow,
gjm = eikre,jm /re,jm , (17)
with the wave-normal distance re,jm between source ξj and microphone xm
rjm
re,jm = q , (18)
2 2
1 − Mf sin θjm − Mf cos θjm

where θjm and rjm are the geometric angle and geometric distance, respectively, between source position ξj and
microphone xm relative to the engine forward direction. This definition assumes that the phases of the sound field
of each source depend only on the wave-normal distance re,jm between source and microphone. This is exact for
monopoles but is still a good approximation in the acoustic far field within a single lobe of dipole and quadrupole
sources.
The goal of the procedure developed here is to determine the directivities Djm of the J sources toward the M mi-
crophones such that the mean square error F (D) between the measured and the modelled matrix becomes a minimum.
M
X
F (D) = Cmn − C mod 2 (19)
mn
m,n=1

A minimum of F (D) requires that


M
∂F (D) X
mod

=4 Djl ℜ gs,jl

gs,jm (Cmn − Cmn ) = 0, (j = 1 . . . J, l = 1 . . . M ) (20)
∂Djl m=1

Equation (20) yields a system of M J non-linear equations. ℜ means “real part of”. The solution procedure must
observe the side condition that all Djm are positive. In the case of M = 126 microphones and J = 100 sources we
have 12600 unknowns. The total number of independent real values in the cross-spectral matrix is M 2 , which would
in principle allow to determine the directivities of J ≦ M sources. However, the system is poorly conditioned. In

6 of 15

American Institute of Aeronautics and Astronautics


addition, in the case of high frequencies or for subarrays, we may have J > M , leading to an unsolvable problem. It
also may be questionable to include in the analysis the cross-spectra of microphones with a large separation, which
are poorly correlated.
Additional conditions are required to make solutions possible. Therefore, it is assumed that the directivities Djm
for each source j are relatively smooth for neighbouring microphones m and that the values for a given microphone
number m are similar for neighbouring sources j, especially in the jet source region. We obtain the following two
functions that have also to be minimised.
J M−1
X X 2
G1 (D) = (Dj,m − 0.5(Dj,m−1 + Dj,m+1 )) (21)
j=1 m=2

M J−1
X X 2
G2 (D) = (Dj,m − 0.5(Dj−1,m + Dj+1,m )) (22)
m=1 j=2

Instead of finding a minimum of F (D) a minimum has to be determined for


G(D) = F (D) + σ1 G1 (D) + σ2 G2 (D) (23)
where the slack variables σ1 and σ2 have to be optimised experimentally. A large value of σ1 would force more
uniform directivities, a large value of σ2 smooths the variation of the source strengths along the engine axis. Larger
values of σ1 and σ2 might be chosen for sources in the jet region than for those in the engine region. A minimum of
G(D) requires that
∂G(D) ∂F (D) ∂G1 ∂G2
= + σ1 + σ2 = 0, (m = 1 . . . M, j = 1 . . . J) (24)
∂Djm ∂Djm ∂Djm ∂Djm
for which we need in addition to equation (20) the partial derivatives of equations (21) and (22). For the microphones
3 . . . M − 2 we obtain
∂G1 (D)
= 0.5Dj,m−2 − 2Dj,m−1 + 3Dj,m − 2Dj,m+1 + 0.5Dj,m+2 , (m = 3 . . . M − 2, j = 1 . . . J), (25)
∂Djm
and for the sources 3 . . . J − 2
∂G2 (D)
= 0.5Dj−2,m − 2Dj−1,m + 3Dj,m − 2Dj+1,m + 0.5Dj+2,m , (m = 1 . . . M, j = 3 . . . J − 2). (26)
∂Djm
Equation (24) has to be solved with the side condition Djm ≧ 0. The addition of conditions (25) and (26) changes
the ill-conditioned or even unsolvable system (20) to a well-conditioned and robust one.
A procedure to find a minimum of a non-linear problem with constraint is required to solve the problem. The solver
for the unconstrained problem of Rasmussen26 (the Matlab routine minimize.m can be downloaded from the web) was
used to develop an own procedure. It is a conjugate gradient method requiring the function G(D) of equation (23)
and the derivatives of equation (24). Function G(D) with many thousands of unknowns Djm has naturally many local
minima and a good first estimate is required for a physically relevant solution and hopefully the global minimum of
the function G. The results with a sliding subarray of section E serve for this purpose.
The method described in this section shall be called SODIX (SOurce DIrectivity modelling in cross-spectral ma-
triX).

G. Tone removal in cross-spectral matrix


The analysis in this paper is restricted to the broadband noise radiation. Rather than removing the tones in the final
results it is proposed to remove the tones from the cross-spectral matrix. This is achieved by interpolation of the
cross-spectra in the frequency domain. Tones are identified by investigating the power spectra of all microphones.
Tone regions are defined where tones are found in any of these spectra. The tonal contribution in the power spectra is
removed by first removing the frequency bins of all tone regions and then replacing them by interpolated values. This
ensures that the likely broadband part is preserved in the frequency spectra. The cross spectra are treated in terms of
the coherence and phase spectra. The tone regions of the coherence spectra are treated just like the power spectra by
replacing the frequency bins of tone regions with interpolated values. The tone regions are also first removed in the
phase spectra. These cleared spectra are then unwrapped to eliminate the phase jumps of 2π. The gaps in the tone
regions are then filled with interpolated values. The amplitudes of the cross spectra between two microphones are then
recalculated using the cleaned power spectra of the two microphones and the cleaned coherence. The cleaned phases
are then used to calculate the real and imaginary parts of the cross-spectra.

7 of 15

American Institute of Aeronautics and Astronautics


III. Comparison of line array data with SODIX results
The measured line array data are the basis of the computations. These data are now compared with the results of
the modelling in figures 3(a), 3(b), 4(a), 4(b), 5(a), and 5(b). The three vertical red lines indicate the locations of the
inlet and the two nozzles of the engine. The black lines are the measured results. The dash-dotted blue lines indicate
the simulations with the sliding subarray according to the method described in section E. These results serve as first
estimate for SODIX. The dashed red curves are the results of SODIX. These distributions agree almost perfectly with
the measured results. It can be concluded from figures 3(a) and 3(b) that the line array was too short in the jet direction
for low frequencies, because the downstream end of the array does not extend past the peak values of the sound
pressure levels. The smoothing effect of the functions G1 and G2 in equation (23) is visible in all figures, especially
in figures 5(a) and 5(b). The sigma values required for equation (23) of the shown results are σ1 = 0.0001J, where J
is the number of assumed sources, and σ2 = 0.0005. Within the jet region σ1 was increased by a factor of five and σ2
by a factor of two.
5 dB per Division

5 dB per Division
SPL / dB

SPL / dB

Measurement Measurement
Sliding subarray Sliding subarray
SODIX SODIX
−20 −15 −10 −5 0 5 10 15 −20 −15 −10 −5 0 5 10 15
Microphone Position −x / m Microphone Position −x / m

(a) 100 Hz (b) 200 Hz

Figure 3: Sound pressure levels of the microphones in the line array in narrow bands at f = 100 Hz and 200 Hz with
∆f = 4 Hz. Measured results are shown in solid black, the results of section E are shown in dash-dotted blue and the
results of SODIX (section F) are shown in dashed red. SODIX agrees almost perfectly with the measurements. Please
note that the x-axis is inverted in comparison to figure 2.
5 dB per Division

5 dB per Division
SPL / dB

SPL / dB

Measurement Measurement
Sliding subarray Sliding subarray
SODIX SODIX
−20 −15 −10 −5 0 5 10 15 −20 −15 −10 −5 0 5 10 15
Microphone Position −x / m Microphone Position −x / m

(a) 400 Hz (b) 800 Hz

Figure 4: Sound pressure levels of the microphones in the line array in narrow bands at f = 400 Hz and 800 Hz, with
∆f = 4 Hz. Measured results are shown in solid black, the results of section E are shown in dash-dotted blue and the
results of SODIX (section F) are shown in dashed red. SODIX agrees almost perfectly with the measurements. Please
note that the x-axis is inverted in comparison to figure 2.

8 of 15

American Institute of Aeronautics and Astronautics


5 dB per Division

5 dB per Division
SPL / dB

SPL / dB
Measurement Measurement
Sliding subarray Sliding subarray
SODIX SODIX
−20 −15 −10 −5 0 5 10 15 −20 −15 −10 −5 0 5 10 15
Microphone Position −x / m Microphone Position −x / m

(a) 1600 Hz (b) 3150 Hz

Figure 5: Sound pressure levels of the microphones in the line array in narrow bands at f = 1600 Hz and 3150 Hz
with ∆f = 4 Hz. Measured results are shown in solid black, the results of section E are shown in dash-dotted blue
and the results of SODIX (section F) are shown in dashed red. SODIX agrees almost perfectly with the measurements.
The effects of the cone of silence of jet noise radiation are visible at the downstream end of the microphone array.
Please note that the x-axis is inverted in comparison to figure 2.

IV. Source distributions in one-third octave bands


The results of SODIX permit to compute the sound radiation of the engine over a large range of emission angles
into the acoustic far field of the engine.
The influence of the emission angle is now studied in one-third octave bands for three far field positions on a circle
with a radius of 45.72 m around the centre of the engine with an angle of 60, 90, and 120 deg relative to the forward
direction and x = 0 (compare figure 2). Figure 6 shows the results for the frequency f = 100 Hz. The three red
vertical lines indicate the axial positions of the inlet and the secondary and primary nozzles of the engine. The vertical
lines indicate the strengths of all sources along the axis of the engine. Almost all sources for f = 100 Hz are located
in the jet region. A considerable source can be seen in the inlet only in the forward arc. It is interesting to see that the
source region extends to 25 m downstream of the nozzle, where the source strength is reduced by about 20 dB. Please
recall that the line array ended at x = −17 m (figure 2), which indicates that the source levels in the jet far downstream
are determined by extrapolation from smaller angles, which explains the spurious second peak seen in figures 6(b) and
6(c) for angles at 90 and 120 deg.
5 dB per Division

5 dB per Division

5 dB per Division
SPL / dB

SPL / dB

SPL / dB

−5 0 5 10 15 20 25 −5 0 5 10 15 20 25 −5 0 5 10 15 20 25
Source Position −ξ / m Source Position −ξ / m Source Position −ξ / m

(a) 60 deg (b) 90 deg (c) 120 deg

Figure 6: Source distribution in one-third-octave band 100 Hz as observed in a position with a distance of 45.72 m
from the engine for angles 60 deg (forward arc), 90 deg and 120 deg (rear arc). Source separation ∆ξ = 0.2λ. Please
note that the source resolution is substantially better than the wave length of λ = 3.34 m.

The corresponding results for a frequency f = 200 Hz are shown in figure 7. The situation has hardly changed in
comparison to 100 Hz. The relative contribution of the casing between inlet and secondary nozzle has increased.
The results for a frequency f = 400 Hz are shown in figure 8. The distance ∆ξ between adjacent sources is
smaller than for the the lower frequencies because ∆ξ is assumed to be 0.4 times the wave length λ for frequencies
above 200 Hz. We start to see a substantial noise emission from the inlet toward 60 deg. The source strength for
120 deg peaks at the location of the two nozzles.

9 of 15

American Institute of Aeronautics and Astronautics


5 dB per Division

5 dB per Division

5 dB per Division
SPL / dB

SPL / dB

SPL / dB
−5 0 5 10 15 20 25 −5 0 5 10 15 20 25 −5 0 5 10 15 20 25
Source Position −ξ / m Source Position −ξ / m Source Position −ξ / m

(a) 60 deg (b) 90 deg (c) 120 deg

Figure 7: Source distribution in one-third-octave band 200 Hz as observed in a position with a distance of 45.72 m
from the engine for angles 60 deg (forward arc), 90 deg and 120 deg (rear arc). Source separation ∆ξ = 0.4λ.
5 dB per Division

5 dB per Division

5 dB per Division
SPL / dB

SPL / dB

SPL / dB
−5 0 5 10 15 −5 0 5 10 15 −5 0 5 10 15
Source Position −ξ / m Source Position −ξ / m Source Position −ξ / m

(a) 60 deg (b) 90 deg (c) 120 deg

Figure 8: Source distribution in one-third-octave band 400 Hz as observed in a position with a distance of 45.72 m
from the engine for angles 60 deg (forward arc), 90 deg and 120 deg (rear arc). Source separation ∆ξ = 0.4λ.

The results for a frequency f = 800 Hz are shown in figure 9. The radiation from the inlet becomes the most
prominent one for 60 deg. Surprising is the substantial noise radiation from the core nozzle into the forward arc.
The two other angles are dominated by the radiations from the two nozzles. The source position of the radiation
from the casing becomes more visible. Its contribution to the total noise emission is still small, but it can be clearly
identified, because of the large dynamic range of the resulting source distribution. The origin of the inlet radiation for
the radiation angle θ = 90 deg is located exactly on the inlet. This was achieved by selecting a uniform mean flow
Mach number Mf = 0.015.
5 dB per Division

5 dB per Division

5 dB per Division
SPL / dB

SPL / dB

SPL / dB

−6 −4 −2 0 2 4 6 8 10 −6 −4 −2 0 2 4 6 8 10 −6 −4 −2 0 2 4 6 8 10
Source Position −ξ / m Source Position −ξ / m Source Position −ξ / m

(a) 60 deg (b) 90 deg (c) 120 deg

Figure 9: Source distribution in one-third-octave band 800 Hz as observed in a position with a distance of 45.72 m
from the engine for angles 60 deg (forward arc), 90 deg and 120 deg (rear arc). Source separation ∆ξ = 0.4λ.

The results for a frequency f = 1600 Hz are shown in figure 10. Both nozzles can be distinguished in figure 10(b).
Their source positions are slightly moved downstream due to the convection effect of the jet flow. An additional source
can be seen on the casing of the engine. The jet and the two nozzles hardly contribute to the sound radiation into the
forward arc. Both nozzles dominate at 90 deg. The rear arc is dominated by the two nozzles and the region between
the two nozzles.
The results for a frequency f = 3150 Hz are shown in figure 11. It can be seen that the sources can still be localised
at this high frequency. This is very surprising because the microphone separations in the line array correspond to
between two and ten wavelengths for this frequency. A beamforming analysis would have been dominated by aliases
at this frequency and therefore useless. Surprising is the apparent sound radiation from positions in front of the inlet,
which might be caused by sound scattering at the inflow control device (ICD) at this high frequency, because the
wavelength has reduced to about 10 cm, a length comparable to the depth of the flow straightener and the supporting
frame around each of the pentagonal or hexagonal plane elements (see figure 1). The source levels of the sources
at the downstream end of the jet appear to be rising again for the angles 90 and 120 deg, which is a nonphysical

10 of 15

American Institute of Aeronautics and Astronautics


5 dB per Division

5 dB per Division

5 dB per Division
SPL / dB

SPL / dB

SPL / dB
−8 −6 −4 −2 0 2 4 6 −8 −6 −4 −2 0 2 4 6 −8 −6 −4 −2 0 2 4 6
Source Position −ξ / m Source Position −ξ / m Source Position −ξ / m

(a) 60 deg (b) 90 deg (c) 120 deg

Figure 10: Source distribution in one-third-octave band 1600 Hz as observed in a position with a distance of 45.72 m
from the engine for angles 60 deg (forward arc), 90 deg and 120 deg (rear arc). Source separation ∆ξ = 0.8λ.
5 dB per Division

5 dB per Division

5 dB per Division
SPL / dB

SPL / dB

SPL / dB
−8 −6 −4 −2 0 2 4 −8 −6 −4 −2 0 2 4 −8 −6 −4 −2 0 2 4
Source Position −ξ / m Source Position −ξ / m Source Position −ξ / m

(a) 60 deg (b) 90 deg (c) 120 deg

Figure 11: Source distribution in one-third-octave band 3150 Hz as observed in a position with a distance of 45.72 m
from the engine for angles 60 deg (forward arc), 90 deg and 120 deg (rear arc). Source separation ∆ξ = 0.8λ.

result. One possible explanation is that the microphone separation in the line array is too large for this high frequency.
Another explanation may be the simplified consideration of the influence of the entrainment flow between jet and
microphone array on the sound propagation. This influence is accounted for in a first approximation by assuming a
uniform flow speed in the computation of the steering vectors according to equations (17) and (18). The presented
results were computed assuming a Mach number of Mf = 0.015, which is likely a too low value for the downstream
jet region. The entrainment flow increases the wave normal distances between sources and microphones. The actual
flow field varies with axial and radial position relative to the engine resulting in curved ray paths for the sound waves.
A curvature also increases the wave-normal distance. An incorrect wave-normal distance has a larger influence on the
phases at higher frequencies.

V. Directivities in the far field and breakdown of main sources


Using the directivity of each individual source, one can compute the sound radiation to any position in the far
field over a large range of emission angles. The predictions for the positions of the far-field microphones located in
a radial distance of 45.72 m from the position x = 0 on the engine axis are compared in the following figures with
the experimental data provided by Rolls-Royce. The power spectra of these data were cleaned from the influences of
tones with the same procedure as the power spectra of the cross-spectral matrix as explained in section G.
The directivities of the one-third octave bands of 100 Hz and 200 Hz are shown in figures 12(a) and 12(b). The valid
range of the predictions is indicated by two vertical red lines. This is the range, where directivities are available from
measurements for all sources. Outside this range the directivities of at least part of the sources had to be extrapolated.
It can be seen that the predictions are slightly low by less than one dB. This difference might be explainable with the
coherence loss in the microphone signals of the line array due to turbulent decorrelation of the sound waves in their
propagation from the sources to the line array of microphones. This is a well-known problem in beamforming.27 The
differences between measurements and predictions are higher only in the downstream direction for angles θ > 130 deg.
The source directivities had to be extrapolated in this region.
The predictions with SODIX are separated into contributions from various source regions, inlet, casing, nozzles,
and jet. The source regions are defined as follows.

inlet ξ > 3.5 m


casing −0.5 m < ξ < 3.5 m
nozzles −3 m < ξ < −0.5 m
jet ξ < −3 m

11 of 15

American Institute of Aeronautics and Astronautics


5 dB per Division

5 dB per Division
SPL / dB

SPL / dB
20 40 60 80 100 120 140 20 40 60 80 100 120 140
Far Field Angle / Degree Far Field Angle / Degree

(a) 100 Hz (b) 200 Hz

Figure 12: Far-field directivity for one-third octave bands 100 Hz and 200 Hz in comparison with measured results
(green circles). The results of SODIX (solid black line) can be separated into the contributions from various source
regions, inlet (solid cyan), casing (magenta dash-dotted), nozzles (red dotted, includes aft fan, core, and some jet
noise), and jet (blue dashed). The results for all angles between the vertical red line at 74 deg and 150 deg contain
sources, whose directivities had to be extrapolated because the line array was too short in the downstream direction.

It can be concluded from figure 12(a) that the broadband sound radiation for 100 Hz is dominated by jet mixing
noise for all angles. The sound-pressure level for 200 Hz is dominated by the inlet radiation for angles θ < 30 deg.
The results for the higher frequencies 400 Hz and 800 Hz are shown in figures 13(a) and 13(b). The inlet is seen
to become more important for the broadband contribution in the forward arc. The cross-over angle being θ = 35 deg
for 400 Hz and about θ = 50 deg for 800 Hz.
5 dB per Division

5 dB per Division
SPL / dB

SPL / dB

20 40 60 80 100 120 140 20 40 60 80 100 120 140


Far Field Angle / Degree Far Field Angle / Degree

(a) 400 Hz (b) 800 Hz

Figure 13: Far-field directivity for one-third octave bands 400 Hz and 800 Hz in comparison with measured results
(green circles). The results of SODIX (solid black line) are separated into the contributions from various source
regions, inlet (solid cyan), casing (magenta dash-dotted), nozzles (red dotted, includes aft fan, core, and some jet
noise), and jet (blue dashed).

The results for the frequencies 1600 Hz and 3150 Hz are shown in figures 14(a) and 14(b). The inlet dominates in
the forward arc and the nozzle region in the rear arc. The cross-over angles are θ = 75 deg for 1600 Hz and θ = 80 deg
for 3150 Hz.

12 of 15

American Institute of Aeronautics and Astronautics


5 dB per Division

5 dB per Division
SPL / dB

SPL / dB
20 40 60 80 100 120 140 20 40 60 80 100 120 140
Far Field Angle / Degree Far Field Angle / Degree

(a) 1600 Hz (b) 3150 Hz

Figure 14: Far-field directivity for one-third octave bands 1600 Hz and 3150 Hz in comparison with measured results
(green circles). The results of SODIX (solid black line) are separated into the contributions from various source
regions, inlet (solid cyan), casing (magenta dash-dotted), nozzles (red dotted, includes aft fan, core, and some jet
noise), and jet (blue dashed).

VI. Conclusions
The directivities of the noise sources of an aeroengine in an open air test bed can be studied in detail with a new
data analysis program called SODIX (SOurce DIrectivity modelling in cross-spectral matriX). It is based on the cross-
spectra of the signals of a line array of microphones laid out approximately parallel to the engine axis in the geometric
near field of the engine. The method is applied in this paper to modelling the broadband part of the noise emission of
a high-bypass ratio turbofan. The cross-spectral matrix had to be cleaned from the influences of tones. The modelled
directivities for far-field positions agree extraordinary well with experimental data. The modelled far-field data are
within about one dB of the measured data. It may be concluded that the source model of Michel21 for jet mixing
noise is correct. It uses uncorrelated point sources with a directivity for the consideration of the effects of source
interference.
The new method provides the directivities of all sources of a jet engine. The maximum analysis frequency with
the current microphone layout is at least 1.6 kHz and possibly 3.2 kHz, which is surprisingly high, because the
microphone separations in the line array are between two and ten wavelengths for the latter frequency. Beamforming
requires separations of considerably less than one wavelength. The source resolution is a fraction of one wavelength,
which is also much better than beamforming results.
The method requires a choice made for two slack variables σ1 and σ2 , which are needed for the weighting of two
smoothing functions G1 and G2 (equations (21)) and (22)). These functions convert the poorly conditioned problem
to a well conditioned one. The shown results were computed with σ1 = 0.0001J, where J is the number of assumed
sources, and σ2 = 0.0005. The smoothing was increased within the jet region by increasing σ1 by a factor of five and
σ2 by a factor of two for sources inside the jet.
The results for the selected testpoint of the engine demonstrate that the radiation of broadband noise from the inlet
varies in the order of 15 dB as function of emission angle, peaking in the forward arc. Jet noise varies in the order of
10 to 15 dB peaking in the rear arc. The radiation from the nozzle region, which includes some jet noise, changes up
to 30 dB for higher frequencies, peaking in the rear arc. It is also surprising that the casing has a small but noticeable
noise emission, which is dominated by a contribution from ξ = 1.3 m.
Future tests should be performed with an array that extends further downstream to at least about x = −30 m for
a better modelling of the low-frequency portion of jet noise. This would increase the number of microphones in the
array from 128 to about 140. Raising the maximum frequency of the analysis to 4 kHz would require a further increase
of the number of microphones to about 180.
The method has a potential for improvement. Relatively easy to implement is an extrapolation of the source
directivities of the jet sources to higher angles outside the angle range available from the measurements. One example
is that the directivities of the sources at the downstream end of the jet region can be extended above the angles available
from experiment by using the directivities of the sources located further upstream.
A further refinement can be achieved by a better consideration of the influence of the entrainment flow between jet
and microphone array on the sound propagation. This influence is currently only accounted for in a first approximation
by assuming a uniform flow speed. The presented results were computed assuming a Mach number of Mf = 0.015,
which is likely too low for the downstream end of the jet. In addition, the actual entrainment flow field varies with
the radial position relative to the engine resulting in curved ray paths for the sound waves. At higher frequencies it
appears to be necessary to estimate the non-uniformity of the mean flow Mach number on the wave-normal distance
from the source to the microphone position. The mean-flow field of the jet itself is not to be included because this is
part of the sources in the description of equation (10).

13 of 15

American Institute of Aeronautics and Astronautics


Possible further applications of SODIX might be airframe noise measurements with microphone arrays in wind
tunnels with open or closed test sections. A big advantage in comparison to beamforming and deconvolution methods
is that the power spectra of the microphone signals can be excluded from the analysis without violating any mathe-
matical condition as is the case in beamforming. A further advantage of SODIX over beamforming is that a given
array geometry might be usable for higher frequencies than with beamforming. Moreover, the contribution of the
airframe noise sources to the power spectra in the positions of each array microphone can be modelled with SODIX.
The usual assumption in beamforming of a uniform directivity of each source is not necessary. The functions G1 and
G2 (equations (21) and (22)) would have to be reformulated for a two- or three-dimensional microphone array and a
two- or three-dimensional source distribution.

VII. Acknowledgements
The support of Rolls-Royce plc during the tests at the Hucknall outdoor test facility is acknowledged. Rolls-Royce
plc also provided the far-field noise data for validation of the method. The many discussions with Dr. Peer Böhning
of Rolls-Royce Deutschland were very helpful and are appreciated. Also acknowledged are the contributions of our
colleague Dr. Sébastien Guérin in the performance of the measurements, in the calibration of the data and in the
calculation of the cross-spectra. The discussions with our mathematician Sarah Schröder concerning the non-linear
optimisation are also much appreciated.

14 of 15

American Institute of Aeronautics and Astronautics


References
1 Fisher, M. J., Harper-Bourne, M., and Glegg, S. A. L., “Jet engine noise source location: The polar correlation technique,” J. Sound Vib.,
Vol. 51, 1977, pp. 23–54.
2 Tester, B. J. and Fisher, M. J., “Engine noise source breakdown: theory, simulation and results,” AIAA-Paper 81-2040, 1981, AIAA 7th
Aeroacoustics Conference, October 5–7, Palo Alto, California, USA.
3 Johnson, D. H. and Dudgeon, D. E., Array Signal Processing, Concepts and Techniques, P T R Prentice Hall, Englewood Cliffs, 1993.
4 Siller, H. A., Arnold, F., and Michel, U., “Investigation of aero-engine core-noise using a phased microphone array,” AIAA Paper 2001-2269,
2001, 7th AIAA/CEAS Aeroacoustics Conference and Exhibit, Maastricht, Netherlands, May 28-30, 2001, Collection of Technical Papers. Vol. 2
(A01-30800 07-71).
5 Brühl, S. and Röder, A., “Acoustic noise source modelling based on microphone array measurements,” J. Sound Vib., Vol. 231, 2000,
pp. 611–617.
6 Brooks, T. F. and Humphreys, W. M., “A Deconvolution Approach for the Mapping of Acoustic Sources (DAMAS) Determined from Phased
Microphone Arrays,” AIAA Paper 2004-2954, 2004, 10th AIAA/CEAS Aeroacoustics Conference, Manchester, Great Britain, May 10-12, 2004.
7 Brooks, T. F. and Humphreys, W. M., “Three-Dimensional Applications of DAMAS Methodology for Aeroacoustic Noise Source Defini-
tion,” AIAA Paper 2005-2960, 2005, 11th AIAA/CEAS Aeroacoustics Conference, Monterey, California, May 23-25, 2005.
8 Brooks, T. F. and Humphreys, W. M., “A deconvolution approach for the mapping of acoustic sources (DAMAS) determined from phased
microphone arrays,” J. Sound Vib., Vol. 294, 2006, pp. 856–879.
9 Brooks, T. F. and Humphreys, W. M., “Extension of DAMAS Phased Array Processing for Spatial Coherence Determination (DAMAS-C),”
AIAA Paper 2006-2654, 2006, 12th AIAA/CEAS Aeroacoustics Conference, Cambridge, Massachusetts, May 8-10, 2006.
10 Dougherty, R. P., “Extensions of DAMAS and Benefits and Limitations of Deconvolution in Beamforming,” AIAA Paper 2005-2961, 2005,
11th AIAA/CEAS Aeroacoustics Conference, Monterey, California, May 23-25, 2005.
11 Guérin, S., Weckmüller, C., and Michel, U., “Beamforming and deconvolution for aerodynamic sound sources in motion,” BeBeC, 1st Berlin
Beamforming Conference, 2006.
12 Guérin, S. and Siller, H., “A Hybrid Time-Frequency Approach for the Source Localization Analysis i of Acoustic Fly-over Tests,” AIAA
Paper 2008-2955, 2008, 14th CEAS/AIAA Aeroacoustics Conference, Vancouver, British Columbia, Canada, 5-7 May 2008.
13 Guérin, S. and Weckmüller, C., “Frequency-domain reconstruction of the point-spread function for moving sources,” BeBeC, 2nd Berlin
Beamforming Conference, 19-20 February 2008, http://www.bebec.eu, 2008.
14 Sijtsma, P., “CLEAN Based on Spatial Source Coherence,” AIAA Paper 2007-3436, 2007, 13th AIAA/CEAS Aeroacoustics Conference,
Rome, Italy, May 21-23, 2007.
15 Sijtsma, P., “CLEAN based on spatial source coherence,” International Journal of Aeroacoustics, Vol. 6, 2007, pp. 357–374.
16 Sijtsma, P., “Tutorial: Improving resolution with CLEAN-SC,” BeBeC, 2nd Berlin Beamforming Conference, 19-20 February 2008,
http://www.bebec.eu, 2008.
17 Blacodon, D. and Élias, G., “Level Estimation of Extended Acoustic Sources Using an Array of Microphones,” AIAA Paper 2003-3199,
2003, 9th AIAA/CEAS Aeroacoustics Conference, Hilton Head, South Carolina, May 12-14, 2003.
18 Blacodon, D. and Élias, G., “Level Estimation of Extended Acoustic Sources Using a Parametric Method,” Journal of Aircraft, Vol. 41,
2004, pp. 1360–1369.
19 Blacodon, D., “Analysis of the Airframe Noise of an A320/A321 with a Parametric Method,” Journal of Aircraft, Vol. 44, 2007, pp. 26–34.
20 Shanno, D. F. and Phua, K. H., “Remark on “Algorithm 500: Minimization of Unconstrained Multivariate Functions [E4]”,” ACM Trans.
Math. Softw., Vol. 6, No. 4, 1980, pp. 618–622.
21 Michel, U., “Influence of Source Interference on the Directivity of Jet Noise,” AIAA Paper 2007-3648, 2007, 13th AIAA/CEAS Aeroacous-
tics Conference, Rome, Italy, May 21-23, 2007.
22 Michalke, A., “On the Effect of Spatial Source Coherence on the Radiation of Jet Noise,” J. Sound Vib., Vol. 55, 1977, pp. 377–394.
23 Michalke, A., “Some Remarks on Source Coherence Affecting Jet Noise,” J. Sound Vib., Vol. 87, 1983, pp. 1–17.
24 Michel, U. and Michalke, A., “Prediction of Flyover Jet Noise Spectra,” AIAA Paper 81-2025, 1981, AIAA 7th Aeroacoustics Conference.
25 Michel, U. and Funke, S., “Inverse method for the acoustic source analysis of an aeroengine,” 2nd Berlin Beamforming Conference, Berlin,
Germany, 19-20 February 2008, http://www.bebec.eu, 2008.
26 Rasmussen, C. E., Evaluation of Gaussian Processes and other Methods for Non-Linear Regression, Ph.D. thesis, Graduate Department of
Computer Science in the University of Toronto, 1996.
27 Dougherty, R. P., “Turbulent Decorrelation of Aeroacoustic Phased Arrays: Lessons from Atmospheric Science and Astronomy,” AIAA
Paper 2003-3200, 2003, 9th AIAA/CEAS Aeroacoustics Conference, Hilton Head, South Carolina, May 12-14, 2003.

15 of 15

American Institute of Aeronautics and Astronautics

View publication stats

You might also like