Toningenieurprojekt

DIY Mixed Order Ambisonics Microphone Array

Michael Romanov

Supervision: Winfried Ritsch

Graz, July 3, 2018

institut für elektronische musik und akustik

 Abstract
In this work a DIY-Mixed-Order-Ambisonics microphone is build. Several ap-
proaches to equalize spherical microphones are discussed and the easiest way used
by multichannel filters to achieve a good sounding result. The idea here is to build a
cheap microphone array that can compete with others available on the market. Also
it must be easy to build alone without assuming to much theory so everyone can do
this at home.
 Acknowledgements
I want to thank Dr. Franz Zotter for the fruitful discussions about the theory of
spherical microphone arrays and mathematical questions.
Also i would like to thank my colleagues Nils Meyer-Kahlen and Thomas Dep-
pisch for the support during the recording and listening session.
M.Romanov: Development of a MOA Microphone Array 4

Contents

1 Introduction 6

2 Ambisonics Microphone Theory 6

2.1 Coordinate System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Spherical Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4 Microphone encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.5 Spatial Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3 Generating Filters 9
3.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4 Microphone Construction 14
4.1 De-assembling Behringer Measurement MIc . . . . . . . . . . . . . . . . 14
4.2 Sphere Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.3 Building the microphone . . . . . . . . . . . . . . . . . . . . . . . . . . 17

5 Measurement 19
5.1 Measurement Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.2 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

6 Encoding and Filtering 23

6.1 Encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
6.2 Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

7 Microphone Comparison 27
7.1 Recording Session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
M.Romanov: Development of a MOA Microphone Array 5

7.2 Listening Session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

8 Conclusion and Outlook 29

M.Romanov: Development of a MOA Microphone Array 6

1 Introduction
In general spherical microphone arrays try to sample a sphere equidistantly [1]. Percep-
tually it has been shown, that the spatial resolution on the horizontal plane is higher then
for elevation [2]. This leads to the idea to build an microphone array that has a higher
resolved angular resolution for and less spherical spherical resolution. Combining a circu-
lar 2D array with a spherical 3D array is called Mixed Order Ambisonics and is discussed
in several publications [3] [4] [5].
Spherical microphone arrays are very expensive. In this work we try to build a micro-
phone array that is affordable. Equalisation of distortions that come up with the Ambison-
ics transformation can be cancelled by a spherical measurement techniques or analytic
filters, that will be discussed in this work. We try to show if it is possible to build an
microphone array that has not perfect microphone placement and will equalize manually
in a DAW using multichannel filters.

2 Ambisonics Microphone Theory

2.1 Coordinate System

To describe linear movements in space it is common to use cartesian coordinates. To de-
scribe circular movements it is common to use spherical coordinates. The transformation
from cartesian coordinates to spherical is performed like
p
r = x2 + y 2 + z 2 (1)
y
ϕ = arctan (2)
x
z z
ϑ = arccos p = arccos . (3)
2
x +y +z 2 2 r
The other way round the computation is performed as follows
x = r cos ϑ cos ϕ (4)
y = r cos ϑ sin ϕ (5)
z = r sin ϑ. (6)

2.2 Spherical Harmonics

To get an understanding of how Ambisonics works requires familiarity with the spherical
harmonic representation of directivity patterns. An standardized representation has been
published in the ambiX convention [6]. The spherical harmonics are defined as
(
sin(|m|ϕ), f or m < 0
Ynm (ϕ, ϑ) = Nn|m| Pn|m| (sin ϑ) (7)
cos(|m|ϕ), f or m ≥ 0.
M.Romanov: Development of a MOA Microphone Array 7
|m| |m|
whereby n denotes the order and m the degree. Pn are the Legendre functions and Nn
represents the normalization function and for the trigonometrical functions and for the
|m|
Legendre functions. For Nn in ambiX is defined with the SN3D normalization in
s
2 − δm (n − |m|)!
Nn|m| = . (8)
4π (n + |m|)!

2.3 Encoding
We can represent a sound source as a directivity pattern composed by an input signal
weighed with the sampled spherical harmonics at a specific direction. It is very similar to
the beam forming with microphone signals. Overlaying an omnidirectional microphone
with a figure-of-eight (dipole) microphone yields cardioid directivity pattern. Same hap-
pens in the spherical harmonics domain. Overlay the omnidirectional w-channel with the
spherical figure of eight patterns result in a spherical cardioid beam. Increasing the Am-
bisonics order leads the beam to be more focused like a super cardioid pattern. The focus
of the beam increases with the Ambisonics order n. To bring a mono signal s(t)in the
Ambisonics domain we just calculate the weights of every spherical harmonic Ynm (ϕ, ϑ)
up to the chosen Ambisonics order and a given direction (ϕ, ϑ), multiply the input signal
with the weight for every Ambisonics channel

χm m
n (t) = Yn (ϕ, ϑ)s(t), (9)

and we end up with the Ambisonics signal χm

n (t) with

k = (n + 1)2 (10)

Ambisonics channels for 3D Ambisonics. For the 2D case there are

k = (2n + 1) (11)

Ambisonics channels.

2.4 Microphone encoding

To transform any set spherical microphone signals into the Ambisonics domain we calcu-
late the weights of the spherical harmonics at microphone positions ϕi , ϑi what results in
a matrix y. The pseudo-inverse of this matrix maps the microphone signals s(t) into the
Ambisonics domain
γ(t) = (YT Y)−1 YT s(t). (12)

As we can see, the matrix needs to be inverted. To calculate a real inverse there is the
same amount of capsules as Ambisonics channels needed.
M.Romanov: Development of a MOA Microphone Array 8

This transformation of the microphone signals leads to distortions of the Ambisonics sig-
nals. To avoid this distortion a set of frequency dependant filters h(ω) needs to be applied
that equalizes the on-axis frequency response and yields undistorted Ambisonics signals

χ(t) = diag(h(ω)) γ(t). (13)

2.5 Spatial Filters

This kind of distortions are caused by the geometry of the microphone array, and are ba-
sically dependent on the radius of the used microphone array and the amount of capsules
used. In general this results in low frequency loss with growing Ambisonics order. On the
other hand by growing frequency spatial aliasing comes into play. Radial focusing filters
need to be applied to avoid the low frequency loss. These can be either calculated ana-
lytically or me measured. Analytical filter design is presented here, here, here and there.
In general this introduces a very high gain for low frequencies with growing Ambisonics
order. To avoid a very high gain at low frequencies a trade of between spatial resolution
in low frequencies and low frequency gain must be taken into account.
Several methods for calculating such filters are discussed in various papers [7] [8] [9]
[10].
Measurement of frequency responses can also lead to such filters. A method to do so is
discussed in [11].
A comparison of several model based and measurement based approaches to generate
encoding filters are discussed in [12]. It shows that measurement and matrix inversion
approaches lead to petter performance for low frequencies.
M.Romanov: Development of a MOA Microphone Array 9

3 Generating Filters

3.1 Theory
Angelo Farina et al. have published a measurement method for synthesizing virtual micro-
phones that can also be used calibrating spherical microphone arrays [11] for Ambisonics.
It is generally based on the idea that we try to synthesize virtual microphones that means
that a given microphone array is equalized and summed in a way that it represents a mi-
crophone with a target directivity pattern. In the presented method the virtual microphone
arrays can be any given target pattern. The idea is to measure the directivity pattern of
every single microphone of an spherical microphone array from multiple directions and
analyse the directivity pattern of every capsule. Given a target directivity pattern a set of
filters can be synthesized to create such a pattern by a matrix convolution.
So we take a given set of M real microphones and want to create V virtual microphones
with target directivity. For this conversion a set of filters is needed with the dimension
M xV . This can be implemented as FIR filters as they are always stable and can be
derived from measurements.
 signal-to-noise ratio and rejection of off-beam sounds decoding: t
are better than those obtainable employing traditional directly the
processing algorithms applied to the same input signals, signals, with
or dedicated top-grade ultra-directive microphones. microphone.
In principl
1 Development
M.Romanov: DESCRIPTION
of a MOAOF THE SYSTEM
Microphone Array 10microphones
practice we
1.1 Computation of the filters: the theory independent
The directiv
Given an array of transducers, a set of digital filters can
of order n is
be employed for creating the output signals. In our case
the expressio
the M signals coming from the capsules need to be
converted in V signals yielding the desired virtual
directive microphones: so we need a bank of M×V Qn ,
filters. As always, we prefer FIR filters.
Where Q1
x1(t) y1(t) cardioid mic
x2(t) y2(t)
SIGNAL
M inputs .. V outputs
. PROCESSOR ..
xM(t)
. Q1 ,
yV(t)

x1(t) y1(t)
h1,1(t)
x2(t)
.. h2,1(t)
.
xM(t)
hM,1(t)

Figure 1: Scheme of the signal processing

Assuming xm as the input signals of M microphones, yv

as the output signals of V virtual microphones and hm,v
the matrix of filters, the processed signals can be
expressed as:
M
y v (t ) xm (t ) hm,v (t ) (1)
m 1
Where * denotes convolution, and hence each virtual
microphone signal Figure is
1 –obtained summing
Rotation computing [?] the results of
the convolutions of the M inputs with a set of M proper
So a set ofFIR
realfilters.
microphones providing the signals xm (t) needs to be convolved with with
One of
a set of impulse the most
responses hm,v andused techniques
summed for deriving
to result a virtual microphonefilter
signal yv (t)
banks for generating virtual microphones with arbitrary
directivity is the Ambisonics
M
Figure 2: P
xmmethod:
(t) ∗ hm,v .first the M signals
X
yv (t) = (14)
are processed deriving an m=0 equal or smaller number of
v
spherical harmonics. Later these spherical harmonics
In Farinas approach the virtual microphones are final directivity patterns avoiding Am-
signals and
bisonics decoding areencoding.
added Thistogether
principle with proper gains,
allows synthesizing for
virtual microphones
The process
synthesizing the desired virtual microphones. This has one of sever
the advantage of allowing for the derivation of a large the solution
number of virtual microphones with a small additional simplificatio
effort, as most of the effort is represented by the identical. In
computation of the spherical harmonic signals. microphone
M.Romanov: Development of a MOA Microphone Array 11

having an arbitrary directivity patterns.

For this work we will use the more basic approach what is also discussed in Farinas
paper. Instead of synthesizing a complex given microphone patterns our target directivity
functions are the spherical harmonics. This allows us to create a Ambisonics microphone
that outputs normalized and equalized Ambisonics signals.
So the filters we want to derive are the spacial equalizing filters that equalize the on axis
impulse responses of the spherical harmonics. This way we end up with virtual micro-
phones that represent the spherical harmonics and we can apply all kind of Ambisonics
effects and manipulations to it.
A target spherical microphone directivity pattern Qn (ϕ, ϑ) is assumed, in our case this
represents the real valued spherical harmonic functions.
The characterization of the spherical microphone array can be described as a matrix of
measured anechoic impulse responses C generated with a on this distance equalized sound
source placed at a large number D of positions all around the probe.
So matrix C has the dimensions M ×D. To derive such set of filters h we should transform
the measured impulse responses c into the theoretical impulse responses p:
These theoretical responses p are derived from the target directivity pattern Qn (ϕ, ϑ) ap-
plied to a delayed unit-amplitude Dirac’s delta function δ. So for one specific direction d
out of D directions in total this results in

M
X
pd = cm,d ∗ hm . (15)
m=0

An easier way to compute such filters is in the frequency domain by computing the com-
plex spectra using FFT algorithm to the N -points-long impulse responses c, h and p this
results in

M
X
Pd = Cm,d,k · Hm,k (16)
m=0

for each frequency bin k whereby P , C and H denote the resulting complex spectra.
Now we can calculate a matrix of filters H for the V virtual microphones (here spherical
harmonics) by
This yields an over-determined system and doesn’t admit an exact solution. However, it
is possible to approximate a solution using the Least Squares method for matrix inversion
employing a regularization technique to avoid instabilities and excessive signal boost.
To calculate the Least Squares solution following system is applied.
 M.Romanov: Development of a MOA Microphone Array 12

atrix of measured least-squares method is shown in Figure 4:

med and the matrix
y employing some
Q k
east Squares plus
outputs of the
ose to the ideal
also inherently Ck Hk
and acoustical
ion, etc.). Figure 4: scheme of the Least Squared method with a
sed on a matrix of delay in the upper branch
obtained with the
D of positions all
In this scheme we observe the delay block , required
3.
for producing causal filters, and the resulting total
modelling error e, which is being minimized by the
least-squares approach
... c1,d ... c1, D In general, the frequency-domain representation of a
... c2,d ... c2, D Dirac’s delta delayed by n0 samples is given by:
... ... ... ... n0
j2 k
... cm ,d ... cm, D k e N (7)
... ... ... ... Figure 2 – Rotation computing [?]
Albeit various theories have been proposed for defining
... c M ,d ... cM , D the optimal value of the causalisation delay n0, we did
A delay δ is introduced here. This one is needed to produce causal filters. We can also
take the easy approach, setting n =N/2. Choosing N/2
observe a modelling error e that is needed by the Least0 Squares algorithm to be minimized.
ements from D In generalsamples is a safe choice, which creates inverse filters
in frequency-domain the delay function is represented as
rophones with their “main peak” close to their centre, and going
smoothly to zero at both ends. n0
orm the measured Furthermore, a regularizationδk = e
−j2πk
parameter
N is required in (17)
the denominator of the matrix computation formula, to
cribed theoreticalThe easiest way to apply a suitable delay is setting n0 to N/2. Hereby the peak of the
avoid excessive emphasis at frequencies where the
impulse response moves to the middle and the decay moves to the beginning and the end
signal is very low.
(4) So the solution formula, which was first proposed in
Kirkeby et al. [6], becomes:
e target impulse C k *MxD Q DxV e j k (8)
lying a direction- H k MxV *
C C
k MxD k DxM k I MxM
M.Romanov: Development of a MOA Microphone Array 13

of the impulse response.

As we want to avoid excessive emphasis at frequencies where the signal is very low a
regularisation parameter β is needed in the dominator of the matrix calculation.
Using the psoydo inverse and target pattern yields the searched filter

H = CT (CCT + β(ω)I)−1 Qe−iωτ (18)

The regularization parameter β depends on frequency ω and should be set up properly

according to the frequency range where the microphone array is designed to work properly.
In this frequency range it needs to be specified as a very small value. For frequencies
where conditioning problems could cause numerical instability of the solution we use
larger values e.g. very low and very high frequencies with a smooth transition between
the central band and the outer bands.
A a good choice for the spectral shape of the regularization parameter is to specify it
as a small, constant value inside the frequency range where the probe is designed to
work optimally, and as much larger values at very low and very high frequencies, where
conditioning problems are prone to cause numerical instability of the solution.
In general this is all what is needed to calculate such filters and is implemented in python
as a script that inputs impulse responses and results in the needed filter impulse responses
to equalize the microphone array and is part of the AAMA project (https://git.iem.
at/cm/AAMA)[13].

3.2 Model
Another approach to generate such filters is having a mathematical model of the micro-
phone. In general just the radius and the order of the microphone is needed. So assuming
shadowing effects the theoretical distortion of the microphone array can be calculated.
Assuming equal frequency response of the microphones and using only one measurement
from one direction is sufficient enough to fit all the curves of the model into this measure-
ment. So the needed filters can be easily calculated.
M.Romanov: Development of a MOA Microphone Array 14

4 Microphone Construction
We use 16 Behringer ECM8000 measurement microphone as they appear to have a pretty
flat frequency response and are pretty small standard deviation in frequency response.

Figure 3 – Frequency Response Behringer ECM8000

4.1 De-assembling Behringer Measurement MIc

To build this microphone we use 16 Behringer ECM8000 measurement microphones.
These are discussed in several forums as a wonder microphone.
To de-assemble these microphones we use a heat gun to get rid of the glue that holds the
cap with the capsule inside as shown in Fig. 4.

Figure 4 – Heat gun

After heating we carefully use a pincer and some paper to not scratch the material to get
off the capsule as can be seen in Fig. 5.
M.Romanov: Development of a MOA Microphone Array 15

Figure 5 – Pincer

In a last step we take out the screws and can take the PCB out of the aluminium tube.
The inner life of the microphone is shown in Fig. 6. It is needed to transform the 48V
symmetrical phantom power to the 3V the electret capsules need.

Figure 6 – PCB
M.Romanov: Development of a MOA Microphone Array 16

4.2 Sphere Design

A publication on psychoacoustics show that the angular resolution of spatial hearing is
much higher for the horizontal plane as for the perception of height [2]. So the idea here
is to use a 4th -order circular 2D microphone array for azimuth, whereby a higher order
can be achieved by less amount of capsules. For the 3D part we use a 2nd -order 3D
microphone array. The coordinates used are shown in Tab. 1

Table 1 – Microphone Layout

lspk.idx elevation / ◦ azimuth / ◦

1 0 0
2 40 0
3 80 0
4 120 0
5 160 0
6 -160 30
7 -120 30
8 -80 30
9 -40 30
3 0 90
4 0 45
5 120 45
6 -120 45
7 60 -45
8 180 -45
9 -60 -45
M.Romanov: Development of a MOA Microphone Array 17

4.3 Building the microphone

This microphone array is build of "FIMO Light air" a modelling material that dries on
air or in the microwave. We use a styrofoam bowl to form the microphone and arrange 9
microphones on a circle and 6 in triangles.
The Microphone has an radius of 5cm.

Figure 7 – Sphere formed out of FIMO air light

The capsules are connected to the PCB using ribbon cable. The 16 PCBs output a ribbon
cable with 48 wires that will afterwards be connected via XLR to the audio interface.
M.Romanov: Development of a MOA Microphone Array 18

Figure 8 – Box holding 16 PCBs

All the 16 PCBs were packed into a box that can be placed nearby the microphone. The
output of the box.

Figure 9 – Box holding 16 PCBs

M.Romanov: Development of a MOA Microphone Array 19

5 Measurement

5.1 Measurement Setup

To measure the impulse responses c for D directions we install the spherical microphone
array on a turntable that can be turned in predefined steps that define our measurement
directions.
The measurements will be performed in an almost anechoic chamber at the IEM Peters-
gasse 116.

5.2 Calibration
We place the loudspeaker 1m away from the microphone array. We need to equalize
the “air-channel” between the loudspeaker and the microphone first to be sure that the
exponential sine sweep that we use as measurement signal has an equalized frequency
response. Therefore we place an calibrated measurement microphone on the position of
the sphere and do a measurement. Deconvolving the “air-channel” with the measuerement
signal in frequency domain

F F T (cair (t))
ccorrect (t) = IF F T ( ) (19)
F F T (csweep (t))

yields the filter to be applied to our impulse response to get an equal measurement signal
for our measurement on the turntable.
M.Romanov: Development of a MOA Microphone Array 20

The microphone was positioned in an anechoic chamber on an turntable that can be rotated
in 0.5◦ steps.

Figure 10 – Controllable turntable

It can be controlled over ethernet. So using a software e.g. Pure Data can be used to
automate the measurement. The software plays a sweep and records the responses to the
sweep on all 16 capsules.
For the measurements an Genelec 8030 Loudspeaker was positioned 1m away from the
microphone array. A reference measurement was recorded using one measurement mi-
crophone at the position of the array to equalize the channel between measurement micro-
phone and the loudspeaker as Fig. 11 shows.
M.Romanov: Development of a MOA Microphone Array 21

Figure 11 – Box holding 16 PCBs

4 measurement series with 2 repetitions have been performed. First one the microphone
in upward position (Fig. 13) and rotated in 10◦ steps the microphone number one facing
the loudspeaker in the beginning of the procedure.

Figure 12 – Azimuth Measurements

M.Romanov: Development of a MOA Microphone Array 22

Second measurement was performed with the first microphone facing the floor and the
top microphone facing the loudspeaker.

Figure 13 – Azimuth Measurements

For measurement three the microphone array was turned around for 90◦ and 225◦ for th
fourth one. When looking towards the loudspeaker the rotation was performed clockwise.
The full process results in 36 × 4 = 144 measurement positions that can be used for filter
generation and equalizing.
M.Romanov: Development of a MOA Microphone Array 23

6 Encoding and Filtering

6.1 Encoding
For the encoding of the microphone signals we use the MultiEncoder Plug-in from the
IEM Plugin Suite [14]. For the first 9 channels we position the sources according on
the horizontal plane using 4th -order, what results in only bringing signals to the circular
harmonics (see Fig. 14).

Figure 14 – Encoding the 2D ring microphone signals

M.Romanov: Development of a MOA Microphone Array 24

The rest of the microphone signals are encoded 2th -order what can be observed in Fig. 15.

Figure 15 – Encoding the 3D spherical microphone signals

Afterwards the signals are summed up on the Ambisonics bus for further equalisation.
M.Romanov: Development of a MOA Microphone Array 25

6.2 Filtering
A set of multichannel filters are used to approximate the needed filters. For HOA micro-
phone arrays a tool was published in [15], but the current version does not support MOA
arrays.
The target filters to be applied to a microphone with 10cm radius can be seen in Fig. 16

Figure 16 – Sparta - array2sh [15]

Filters for the given microphone array. Thin lines represent the real needed filters with
very high gain in the low frequencies. The thick lines show the Thikonov filters that we
will approximate to boost low frequencies with maximum 30dB in the needed frequency
range. Hereby the blue lines represents the 0th -order filter, the red lines represent the 1th -
order filter, yellow line the 2th -order filter, purple line the 3th -order filter and the green
one represents the 4th -order filter.
As all orders have the same high shelf characteristic we put a mcfx-filter [16] with high
shelf characteristic on the sum bus of the encoded signals as can be seen in Fig. 17.

Figure 17 – mcfx-filter [16]

M.Romanov: Development of a MOA Microphone Array 26

For the other channels we use a low shelf and a peak filter for the lower frequencies as
shown exemplary for the 3th -order channels in Fig. 18.

Figure 18 – mcfx-filter [16]

For the other Ambisonics order channels we behave similarly at the needed frequencies.
With reaper Plug-in pin connectors we can decide what channels to apply for the filtering.
M.Romanov: Development of a MOA Microphone Array 27

7 Microphone Comparison

7.1 Recording Session

To compare the built microphone to some other available microphone arrays as set of test
recordings was created. Therefore we positioned 6 microphones (DIY-Mic, Eigenmike,
Oktava, Tetramic, CoreSound and Zoom H2N) next to each other in a seminar room what
can be seen in Fig. 19.

Figure 19 – Microphone comparison

A few scenes were recorded. One type was walking around in a circular trajectory with a
shaker. Second scene were positioning the shaker one meter away from the array in front,
left, right and back position for 0◦ and 45◦ elevation. Also for 90◦ elevation.
Another scene were three people talking, first speaker talk after each other and afterwards
all together. With this scene it can be tested how good virtual microphones work for any
array.
The positions we said before any recording.
Last set of recordings was done using a guitar making a figure of eight walktrough the
room. Using this scene it can be compared how natural the guitar sounds and also how
the impression of closeness changes during the movement.
M.Romanov: Development of a MOA Microphone Array 28

7.2 Listening Session

In a quick listening session the microphones were compared. The DIY-microphone array
sounds very similar to the Eigenmike in terms of sound colour, but introducing more noise
compared to the Eigenmike. Also more environmental noise such as the air conditioner
can be heard on the DIY microphone, this caused by not using a microphone suspension.
As expected we hear a better spatial resolution of the DIY-microphone that to the 1st -order
microphones but less than the Eigenmike. The DIY-microphone sounds more natural than
the Eigenmike.
In general it is a good sounding result of this simple approach. The microphone sounds
suitable and the easy encoding en filtering leads to a sufficient listening experience. How-
ever, the comparison between several filter approaches was not performed here due to a
lack of time.
However this was just a short listening session. A proper perceptive evaluation needs to
be done to scientifically prove this comparison using the recorded material. This could be
done similar to [17] [18] and [19].
M.Romanov: Development of a MOA Microphone Array 29

8 Conclusion and Outlook

An Mixed-Order-Microphone array was build to prove if it is possible to achieve such an
microphone with less then 500 Euro costs. Several techniques for equalisation of spherical
harmonics channels were discussed. A set of measurements (144 directions) were made
for further evaluation. To equalize the microphone a set of multichannel filters were used
in a DAW that achieved a suitable result.
In general it is cheap and quite easy to build spherical microphone arrays. But at the end
an multi channel audio interface, pre-amps and a computer is needed to make recordings.
This makes field recordings pretty complicated. In a further work an piece of hardware can
be designed to make a 16-channel field recorder. An approach can be to use the dsPIC33
Digital Signal Controllers. They combine DSP capabilities with micro-controllers and
can process 16 channels of 16 bit audio in 44100 Hz sampling rate on one chip and can
be easily programmed in C. In combination with four Analog Devices 4 channel - pream-
s/adc an field recorders can be designed that includes equalization of the microphones due
convolving and directly writing to an SD-card. This way a very handy tool can be created
in a further work.
Also a set of recordings were made to compare 6 different spatial audio microphones.
As a next step it would be interesting to make an perceptual evaluation using the material
recorded during this work in regard to sound colouration, source extent, depth graduation.
It can also be evaluated how the several equalizing approaches effect the quality of the
microphone array.
M.Romanov: Development of a MOA Microphone Array 30

References
[1] F. Zotter, Analysis and Synthesis of Sound-Radiation with Spherical Arrays. PhD
thesis, December 2009.
[2] R. David, D.R. Perrott and K. Saberi, “Minimum audible angle thresholds for
sources varying in both elevation and azimuth,” 1990.
[3] S. Favrot,; M. Marschall, “Metrics for performance assessment of mixed-order Am-
bisonics spherical microphone arrays,” 2012.
[4] M. Marschall, S. Favrot, and J. Buchhol, “Robustness of a mixed-order Ambisonics
microphone array for soundfield reproduction,” April 2012.
[5] J. Trevino, S. Koyama, S Sakamoto and Yoiti Suzuki, “Mixed-order Ambisonics
encoding of cylindrical microphone array signals,” 2014.
[6] C. Nachbar, F. Zotter, E. Deleflie, A. Sontacchi, “AMBIX - A Suggested Ambisonics
Format,” June 2011.
[7] S. Lösler, F. Zotter, “Comprehensive Radial Filter Design for Practical higher-order
Ambisonic Recording,” 2015.
[8] F. Zotter, M. Frank, C. Haar, “Spherical microphone array equalization for Ambison-
ics,” 2015.
[9] R. Baumgartner, H. Pomberger and M. Frank, “Practical Implementation of Radial
Filters for Ambisonic Recordings,” 2015.
[10] F. Zotter, “A Linear-Phase Filter-Bank Approach to Process Rigid Spherical Micro-
phone Array Recordings,” June 2018.
[11] A. Farina, A. Capra, L. Chiesi, L. Scopece, “A Spherical Microphone Array For
Synthesizing Virtual Directive Microphones In Live Broadcasting And In Post Pro-
duction,” October 2010.
[12] A. Politis, H. Gamper, “Comparing Modeled and Measurement-Based Spherical
Harmonic Eencoding Filters For Spherical Microphone Arrays,” October 2017.
[13] W. Ritsch, “Affordable Ambisonics Microphones Array - DIY higher order ambison-
ics microphone kits - https://git.iem.at/cm/AAMA,” 2017.
[14] D. Rudrich, “IEM Plug-in Suite - https://plugins.iem.at,” 2018.
[15] L. McCormack, S. Delikaris-Manias, A. Farina, D. Pinardi, V.Pulkki, “Real-Time
Conversion of Sensor Array Signals into Spherical Harmonic Signals with Applica-
tions to Spatially Localized Sub-Band Sound-Field Analysis ,” May 2018.
[16] M. Kronlachner, “mcfx - multichannel audio plugin-suite
http://www.matthiaskronlachner.com/?p=1910,” 2018.
[17] E. Bates, S.Dooney, M. Gorzel, H. O’Dwyer, L. Ferguson and F. M. Boland, “Com-
paring Ambisonic Microphones Part 1,” July 2016.
[18] E. Bates, S.Dooney, M. Gorzel, H. O’Dwyer, L. Ferguson and F. M. Boland, “Com-
paring Ambisonic Microphones Part 2,” May 2017.
[19] E. Kurz, F. Pfahler, M. Frank, “Comparison of first-order Ambisonics microphone
arrays,” September 2015.