audio engineering society conference papernando/publications/sphear.pdf · audio engineering...

Audio Engineering Society

Conference PaperPresented at the Conference on

Sound Field Control2016 July 18–20, Guildford, UK

This conference paper was selected based on a submitted abstract and 750-word precis that have been peer reviewed by atleast two qualified anonymous reviewers. The complete manuscript was not peer reviewed. This conference paper has beenreproduced from the author’s advance manuscript without editing, corrections, or consideration by the Review Board. TheAES takes no responsibility for the contents. This paper is available in the AES E-Library (http://www.aes.org/e-lib), all rightsreserved. Reproduction of this paper, or any portion thereof, is not permitted without direct permission from the Journal of theAudio Engineering Society.

The *SpHEAR project, a family of parametric 3D printedsoundfield microphone arraysFernando Lopez-Lezcano1

1CCRMA, Stanford University

Correspondence should be addressed to Fernando Lopez-Lezcano ([email protected])

ABSTRACT

This paper describes the *SpHEAR (Spherical Harmonics Ear) project, an evolving family of 3D printed, GNUPublic License/Community Commons licensed soundfield microphone designs. The microphone assembly is 3Dprinted as separate parts, one for each capsule holder and a microphone mount. The capsule holders interlocktogether like a 3D puzzle and create the microphone assembly. This strategy was chosen to have parts that can beprinted flat and without overhangs so they can be printed in low to medium price 3D printers that use fused-filamentfabrication technology. The 3D models currently include the TinySpHEAR, a four-capsule tetrahedral microphone,the Octathingy, an eight-capsule design, and the BigSpHEAR 12- and 20-capsule proof-of-concept platonic-solidmodels. The models are written in OpenSCAD and are completely parametric. The project also includes suggesteddesigns for the capsule interface electronics and preliminary calibration software written in Octave.

1 Introduction

The soundfield microphone was designed in the 1970’sby Michael Gerzon and Peter Craven to capture thespherical harmonics decomposition of a soundfield upto first order. It uses four capsules in a tetrahedralconfiguration, which are matrixed and equalized to de-rive the Ambisonic B-format signals that represent thesoundfield.

Current Ambisonic microphones include commercialsystems by SoundField/STL [4] and Core Sound [5], aswell as a number of experimental do-it-yourself (DIY)designs such as the one by Hemingson and Sariski [9](based on a design by Henry Walmsley), most of them

assembled from soldered wires that hold the capsulestogether.

The recent availability of affordable and accurate 3Dprinters makes it possible to “print” soundfield micro-phones. For example, a recent crowd-sourced prod-uct, the Brahma microphone [6], uses a 3D printedcore. Our goal is to print the whole microphone, froma simple first-order design to higher capsule count andhigher-order microphones, in part so that our studentscan build their own microphones as part of our spatialsound courses at Stanford University’s Center for Com-puter Research in Music and Acoustics (CCRMA).

Four- and eight-capsule designs are particularly attrac-

Lopez-Lezcano 3D printed microphone arrays

tive as there are reasonably priced USB audio inter-faces and SD-Card recorders with up to eight micro-phone inputs which make it easy to pair the micro-phone with a particular audio interface and calibratethem as a unit. Recent MOTU AVB/USB interfaceswith 24 inputs make designs with up to 24 capsulespractical.

2 TinySpHEAR, the tetrahedral design

We were interested in a design that could be printedin low- to medium-price 3D printers, like the one wehave in-house at CCRMA, an Ultimaker 2 Extended[7].This type of printer uses fused-filament fabricationtechnology (extrusion) and cannot produce intricateparts with overhangs such as a complete soundfieldmicrophone assembly. That lead to an approach where,like the original SoundField microphone, each capsuleholder is a separate part that is later assembled into acomplete microphone.

Two practical topologies surfaced. Both can be printedflat with no overhangs and have interconnection legswith “fingers” that lock the capsule holders together. Inthe first (figure 1) the legs extend horizontally from theside of the capsule holder; in the second they extendvertically (figure 2). The first one is easier to print andleads to the smallest possible microphone array radius,but is not a compact design. Because it is completelyflat, it is also suitable for fabrication using a laser cutterand appropriate materials. The second one is moredifficult to print and has a more compact design but isnot practical for larger radius microphones.

Fig. 1: TinySpHEAR with horizontal legs, 10mm cap-sule, 10mm radius

The 3D models are written in OpenSCAD, a Free Soft-ware constructive solid geometry (CSG) 3D modeling

Fig. 2: TinySpHEAR with vertical legs, 10mm cap-sule, 10.6mm radius

environment based on a built-in functional program-ming language [3]. The models are programs that are(almost) completely parametric, can be previewed on-screen, and then rendered into a 3D mesh which can beexported in a variety of computer-aided design (CAD)file formats.

Parameters include the dimensions of the microphonecapsule, the radius of the completed microphone as-sembly, dimensions of the interconnecting legs, type ofinterconnection leg, height of the mount, and so forth.Once these parameters are set, the rest of the geome-try of the microphone assembly is calculated automati-cally (for example the length of the legs and the depthof the connection slots) so that parts can be printed andassembled for that particular microphone.

A parameter in the Openscad program switches be-tween two views, one which shows the assembled mi-crophone and another that lays out all the parts on avirtual flat surface, ready to be printed (figure 3).

Fig. 3: Parts ready to be 3D printed

AES Conference on Sound Field Control, Guildford, UK, 2016 July 18–20 of 10


All code is available in a git repository1. The programsare licensed as GPL [1] (GNU Public License) andthe 3D models as CC BY-NC-SA [2] (Creative Com-mons). The repository will include all the necessary in-formation for building the microphone, including sug-gestions for the associated interface electronics as wellas calibration procedures so that the end result is a fullyfunctional microphone.

Fig. 4: The full 3D model and the first prototype,10mm capsules, 9.6mm microphone radius

3 *SpHEAR, expanding the family

To the four-capsule first-order design we have addedmicrophones of higher order that can use cardioid cap-sules (as opposed to rigid sphere designs that use omnidirectional capsules)[11][12]. We have designed andprinted parts that assemble into the Octathingy, byEric Benjamin and Aaron Heller[13] and experimen-tal proof-of-concept designs using platonic solids with12 and 20 microphone capsules (figure 5). We will usespherical t-designs[10] to create arrays with more than20 capsules.

All designs will share the same code and use the sameassembly procedure. Each capsule holder has betweenthree and five interconnecting legs at appropriate an-gles so that all parts click together to create the com-plete assembly. Depending on the model, there areone, two, or three different parts that need to be printedin sufficient quantity. All the parts are held together

1https://cm-gitlab.stanford.edu/ambisonics/SpHEAR/

through friction but are then glued together for the fi-nal assembly.

Fig. 5: Four-capsule TinySpHEAR, eight-capsule Oc-tathingy, 12- and 20-capsule platonic solid de-signs

The same approach will be used to design “rigid solid”arrays for omnidirectional capsules that can be assem-bled from individual parts that can be printed flat. Fu-ture work is also considered for multi-layer designs thattry to optimize low-frequency noise performance andhigh-frequency spatial aliasing for a better widebandhigh order microphone[15].

4 Results and discussion

All four existing mechanical designs have been 3Dprinted and assembled and we have built, calibrated,and tested a first prototype of the TinySpHEAR four-capsule tetrahedral microphone.

4.1 Capsules and electronics

Our first prototype uses 10mm Primo EM182 cardioidcapsules. We did not have the time needed to matchthe capsules, so this prototype uses random picks. Toconnect the capsule with our Zoom F8 test preamplifierand eight-channel recorder, we used a simple circuitconsisting of just a resistor and a capacitor to interfacethe signal to the preamplifier and power the capsulefrom the recorder’s P48 supply. This circuit and othersare described in a document hosted in the MicBuildersYahoo group[25]. It is so simple that the componentsfit into the shell of an XLR connector.

Regretfully, this interface circuit interacts badly withthe input stage of the Zoom F8 leading to low-frequency noise leakage into the signal which com-promised the low-frequency performance of our firstprototype. We built and tested with very good results avariant of the well known Schoeps circuit by Zapnspark[26]. A small PCB has been designed using KiCad forour next prototype and will be part of the git repository.



Our choice of the Zoom F8 was based on cost, but alsoon the fact that it has digitally-controlled preamplifiergain. That, coupled with the latest firmware in whichthe level controls can be ganged together, makes it anappropriate choice for recording a soundfield micro-phone with up to 8 capsules. Microphone and recorderare calibrated as a unit.

4.2 Soundfield Microphone Calibration

The soundfield microphone is the acoustic encoder forthe Ambisonic full-sphere sound system. It capturesthe soundfield at a point in space and then decomposesit into the zero-th and first-order spherical harmonics,which correspond to the acoustic pressure (a scalar, W)and acoustic particle velocity (a vector, X, Y, and Z) atthat point. One can do this directly with a monopole(omni) and three orthogonal dipole (figure-8) micro-phones, but it is difficult to position the four micro-phones in a compact arrangement in such a way thatthey do not interfere with each other acoustically.

Craven and Gerzon solved these problem by arrang-ing four identical cardioid-family microphones (a mix-ture of monopole and dipole responses) on the facesof a regular tetrahedron and then deriving the spheri-cal harmonic responses from the outputs of the fourcapsules. If the four capsules have perfectly flat fre-quency responses, matched polar patterns, and couldbe placed at a single point in space, the conversion fromcapsule outputs (“A Format”) to spherical harmonics(“B-format”) could be done by simple sums and differ-ences. The reality, however, is that the array has a finitesize, the capsules do not have flat frequency responseand identical polar patterns, and the geometry may beslightly off. Hence a series of correction filters andgain adjustments are needed to address these problems.A number of approaches have been taken in both com-mercial products and DIY projects. In the best of these,it is possible to achieve frequency and polar responsesthat are superior to the best available omni and figure-8microphones.

The radius of the array determines the frequency atwhich the B-format patterns begin to deviate from theirideal shapes. Therefore, a key design tradeoff is be-tween the radius of the array and the diameter of thecapsules. In principle, one can make the array verysmall by using very small capsules, however small cap-sules tend to have poor signal-to-noise ratio and poorlow-frequency response.

Gerzon’s original paper [8] proposes generating the B-format components using a scalar A-to-B conversionmatrix followed by filters that compensate for the non-coincident nature of the array at high frequencies. Dueto the fact that the array samples the sphere uniformlyand the B-format outputs are rotationally symmetric,the theoretical 4×4 matrix of filters reduces to a singlefilter for each of the components of the B-format signal(W, X, Y and Z), with the X, Y, and Z filters being ofidentical design.

The original SoundField microphones used analog cir-cuits for the A-to-B conversion and calibration. Inthe MkIV each capsule has low-frequency equaliza-tion and sensitivity adjustments, followed by a fixedA-to-B matrix and fixed coincidence correction filtersand finally independent gain adjustments for W, X, Yand Z[20]. Calibrating one of these microphones is avery labor intensive process.

Newer designs like the TetraMic or Brahma move theA-to-B conversion process and calibration to softwarerunning in a general purpose computer.

Another approach is Angelo Farina’s A-to-B conver-sion method[19], in which four on-axis impulse re-sponse measurements of the capsules and four on-axis impulse response measurements of the B-formatcomponents generate eight convolution filters throughNelson-Kirkeby inversion. The proposed software usesfour convolutions for the A-format equalization, fol-lowed by a fixed A-to-B matrix and finally four moreconvolutions that equalize the B-format components.

A practical system using Gerzon’s calibration methodis the (unreleased) TetraCal[18] software package byFons Adriaensen, which complements his open-sourcefree software TetraProc[17] tetrahedral processor pro-gram, a very complete program for A-to-B conversionand monitoring of tetrahedral microphones.

TetraCal automatically generates a scalar A-to-B con-version matrix from a set of eight impulse responsesmeasured at regular angles in the horizontal plane. Thepost-matrix IIR parametric filters are then adjustedmanually by visual observation of the resulting fre-quency and phase response of each component of theB-format signal. In the hands of a skilled user thiscan produce a nearly perfectly calibrated Ambisonicmicrophone.

TetraProc itself can use calibration data generated byeither of the two methods outlined above. In one mode



of operation it applies low-frequency equalization to allfour capsules, followed by a scalar A-to-B matrix andpost-matrix IIR parametric filters for each B-formatcomponent. It also supports Farina’s method by usinga 4×4 convolution matrix that incorporates the capsuleequalization, A-to-B-format conversion and B-formatequalization in one step.

All the software based soundfield microphone calibra-tion systems that we are aware of are closed applica-tions. Furthermore, the systems we know about arenot fully automated and require manual operation bya skilled user. The human operator can distinguishbetween measurement artifacts and the actual micro-phone response and use some judgment as to whatspectral errors to correct and which to ignore. This isprobably unavoidable for best results, but our project istrying to see how far an automatic calibration processcan go in terms of the resulting sound quality of thefinished microphone.

The current results presented in this paper are a workin progress with software being written in the Octaveprogramming language.

4.3 Measuring the TinySpHEAR

The data used in the calibration process detailed be-low is a set of 16 impulse response measurements onthe horizontal plane of our first TinySpHEAR proto-type. The excitation source was an M&H M50 sin-gle driver speaker with a stated frequency response of100Hz to 20KHz +/- 5dB, and an EMM-61 microphonethat was used as a calibration reference in the substi-tution method. Fons Adriansen’s impulse responsesoftware, Aliki[16], running under Linux, was usedto measure the impulse responses using 10-second ex-ponential sweeps. The measurements took place in theCCRMA Stage, a small concert space (i.e., not in ananechoic chamber), and the first reflections from thefloor gave us approximately 5 mSecs of useful time do-main data (our software does not yet implement moresophisticated windowing methods as described by EricBenjamin[14]).

This first prototype of the TinySpHEAR microphonewas used to record a concert in the Bing Concert HallStudio at Stanford University on March 11th and 12th,2016 (“March Mattness, CCRMA in the Bing ConcertHall Studio”), which featured a 20.6 3D speaker de-ployment of the GRAIL (CCRMA’s speaker array for

concerts, “the Giant Radial Array for Immersive Listen-ing”). These recordings, which range from amplifiedacoustic instruments (ragas) to full 3D surround elec-troacoustic compositions, are being used to evaluatethe performance of the microphone calibration in infor-mal listening tests. The concert was also recorded us-ing a SoundField ST250 so comparisons can be madebetween the two (the two microphones were not colo-cated so the spatial image is slightly different).

4.4 Calibrating the TinySpHEAR

The current calibration process is a work in progresswith the initial goal of automatically creating a 4× 4matrix of filters that generates the four B-format sig-nals from the A-format capsule signals, compensatingfor the low frequency response of the capsules, differ-ences in sensitivity and directivity, and the effects ofthe non-coincident layout. The calibration requires 8to 16 equally-spaced impulse response measurementsin the horizontal plane.

The calibration software splits the frequency spectruminto two areas delimited by the transition frequency ofthe microphone. Below the transition frequency, scalarA-to-B matrices are calculated for each frequency bandand are used to design the frequency response of thefilters. Above the transition frequency, a common fil-ter for each B-format component is derived from anaverage of measurements of the RMS power of the im-pulse response for each frequency band. Both sets offilters (16 below the transition frequency, 4 above) aremerged into a single FIR filter for each of the 16 ele-ments of the A-to-B filter matrix.

The first step of the process splits the frequency rangecovered by the excitation speaker into fractional octavebands - 1/8-octave bands in the examples below.2 Foreach band the calibration program calculates an A-to-B matrix using Matlab/Octave functions included in anunpublished paper by Aaron Heller [23]. Each elementof a 4× 4 A-to-B matrix represents the contributionof a single capsule or A-format signal to one of the B-format components in that frequency band. The sameelement in all matrices traces the desired frequencyresponse for that filter.

The calculation of A-to-B matrices is valid as long asthe assumption that the capsules are coincident is true.

2A future version of the software will space the bands unevenly,broader bands in low frequencies, narrower in high frequencies.



Figure 6 shows the frequency response of the B-formatcomponents at 0 and 45 degrees azimuth incidence an-gle using these filters. It is reasonably flat in low andmid frequencies but starts to deviate from the desiredresponse at about 4 to 5 kHz. The observed transitionis an octave below the limit frequency of 11.25 kHzcalculated using Gerzon’s approximate formula (theprototype radius is 9.6mm).

Figure 7 shows the frequency responses of the result-ing filter set. Each one of the traces follows one of the16 FIR filters. The degree of vertical spread betweenfilters of the same order indicates how closely matchedthe capsules are. The difference in shape shows differ-ences in frequency response among the capsules andother effects.

300 500 1000 2000 3000 5000 7000 10000 15000 20000

6

3

0

-3

-6

-9

-12

-15

-20

-25

-30

-40

W

X

Y

Z

(a) 0 degrees azimuth

300 500 1000 2000 3000 5000 7000 10000 15000 20000

6

3

0

-3

-6

-9

-12

-15

-20

-25

-30

-40

W

X

Y

Z

(b) 45 degrees azimuth

Fig. 6: frequency response of B-format signals with fil-ters designed using only the A-to-B matrices

For frequencies above the transition frequency, weneed to use a different strategy to design our filters.In the following examples, we selected a transition fre-quency of 4 kHz by visual inspection of figure 6. The-ory or simulation can be used to arrive at a frequencyresponse for ideal capsules with exact orientation, flat

500 1000 2000 3000 5000 7000 10000 20000

-10

-5

0

5

10

15

20

W

XY

Z

Fig. 7: filter shapes designed using only the A-to-B ma-trices

frequency response and ideal polar patterns[21], butthe real frequency response will deviate from this idealshape. Other calibration systems tune the filters for thisfrequency range manually.

Our approach generates B-format signals using one ofthe scalar A-to-B matrices (about one octave below thetransition frequency), and extracts the RMS power ofeach B-format component for each measurement in theremaining frequency bands. These points trace the fil-ter frequency response above the transition frequency,one filter for each B-format component shared by all A-format signals. As our measurements are only made inthe horizontal plane, we average the X and Y frequencyresponses to create the Z filter.

It is impossible to design filters that will equalize the B-format components equally well in all directions. Thechoice of which of the 16 available measurements toaverage has therefore a big impact in the sound of thecalibrated microphone and is a crucial design decision.

We can average the measurements taken in the cardinaldirections (0, 90, 180, 270 degrees) which are the onesin which the microphone has the best performance,or average the diagonal directions, or average all ofthem (not quite a diffuse field response but somewhatsimilar). Choosing the cardinal directions means thatsounds coming from those directions will be correctlyequalized, and others will suffer a drop in response athigher frequencies. Averaging the diagonal directionswill have the opposite effect. Averaging all availablemeasurements sits in between those two extremes.

There is no correct choice and our software providesthe option of selecting which measurements to aver-age. It is easy to generate different filter matrices and



test them to decide which one sounds best for a partic-ular application. Experienced designers seem to pre-fer equalizing for the cardinal directions as the otheroptions result in microphones that tend to sound toobright. Other weighted averaging options will be ex-plored in the future.

The frequency response of each B-format componentis concatenated to each low-frequency filter and the fin-ished FIR filters are transformed into close to minimumphase FIR filters by using the simple sinus ramp andflip over (SRFO) method outlined in [22]. This min-imizes unwanted phase shifts between the B-formatcomponents at high frequencies that otherwise degradeperformance of the A-to-B conversion.

Figure 8 shows the frequency response of the finishedfilters obtained by averaging the measurements in thecardinal directions. The filters include spline interpo-lation in the high frequency region to extrapolate thefrequency response above the speaker high frequencyresponse limit.

500 1000 2000 3000 5000 7000 10000 20000

-10

-5

0

5

10

15

20

W

XY

Z

Fig. 8: finished filter frequency responses, A-to-B ma-trices below 4 kHz, cardinal direction averag-ing above 4 kHz

Finally, figure 9 shows the 0 degree and 45 degree fre-quency response of the finished microphone using thisfilter matrix for the A-to-B conversion.

A different way of visualizing the same data is shownin figures 10 and 11. It shows the W and XY compo-nents frequency response in the cardinal directions (0,90, 180, 270 degrees) versus the diagonal directions(45, 135, 225, 315 degrees). In figure 10, we calibrateby averaging only the cardinal direction measurements,in figure 11 all directions. We can see that when we av-erage all measurements the on-axis response is brighter

300 500 1000 2000 3000 5000 7000 10000 15000 20000

0

-3

-6

-9

-12

-15

-20

-25

-30

-40

W

X

Y

Z

(a) 0 degrees azimuth

300 500 1000 2000 3000 5000 7000 10000 15000 20000

0

-3

-6

-9

-12

-15

-20

-25

-30

-40

W

X

Y

Z

(b) 45 degrees azimuth

Fig. 9: frequency response using A-to-B matrices be-low 4 kHz and average response for cardinaldirections above 4 kHz

above 7 kHz, while the off axis response is not atten-uated as much. Equalizing only for the cardinal direc-tions makes those signals flatter at the expense of moredrop in the off-axis directions.

Figures 12, 13, 14 and 15 show polar plots of the B-format signals for four different 1/8-octave bands infour different regions of the spectrum. Both omnidirec-tional and figure-of-eight plots show very good perfor-mance. Above the transition frequency we start to seethe expected contamination of the polar patterns withhigher-order spherical harmonics (the lobes of the fig-ure of eight patterns are narrower and there are obvioushigher order components in W and Z).

Figure 16 shows the polar patterns of a forward facingcardioid pattern created by summing the W and X sig-nals. As our FIR filters are not strictly minimum phasethere is still a small residual phase shift between W andX at high frequencies that slightly alters the shape ofthe polar patterns.

During the testing process a spurious resonance wasidentified at about 1.5 kHz. This is probably due to the



3000 5000 7000 10000 15000 20000

0

-3

-6

-9

-12

-15

-20

-25

-30

W (+12dB for clarity)

XY principal directions

XY diagonal directions

Fig. 10: principal directions (0, 90, 180, 270) and diag-onal directions frequency response for princi-pal directions averaging

3000 5000 7000 10000 15000 20000

0

-3

-6

-9

-12

-15

-20

-25

-30

W (+12dB for clarity)

XY principal directions

XY diagonal directions

Fig. 11: principal directions (0, 90, 180, 270) and diag-onal directions frequency response for averag-ing all directions

fact that the hole through which the cables are threadedinto the microphone mount was left open creating aresonant cavity. This highlights the attention to detailthat is necessary to create a good microphone.

Informal listening tests show that both the ST250 andan A to B encoder created by Fons Adriansen withTetraCal sound brighter than the calibrated prototypeat high frequencies as they are not calibrated for “flat”frequency response as is ours. We should include op-tions to create calibrations tailored for different uses(diffuse field versus free field, for example).

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

W: 604 to 659 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

X: 604 to 659 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

Y: 604 to 659 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

Z: 604 to 659 Hz

Fig. 12: B-format polar plots around 600 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

W: 1216 to 1327 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

X: 1216 to 1327 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

Y: 1216 to 1327 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

Z: 1216 to 1327 Hz

Fig. 13: B-format polar plots around 1.2 kHz

5 Future Developments

This is the first paper in a series that will describe theevolution of the *SpHEAR Project.

A second tetrahedral prototype is being built using theZapnspark interface circuit for a much improved low-frequency noise floor. Another prototype will use thePrimo EM200 capsules, a larger 14mm design that hasbetter low frequency and noise performance at the ex-pense of a lower transition frequency.

We also plan to build and test the Octathingy[13] anddevelop a practical calibration method for it. We arevery interested in evaluating a first step at second order



-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

W: 5383 to 5875 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

X: 5383 to 5875 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

Y: 5383 to 5875 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

Z: 5383 to 5875 Hz

Fig. 14: B-format polar plots around 5 kHz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

W: 10840 to 11831 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

X: 10840 to 11831 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

Y: 10840 to 11831 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

Z: 10840 to 11831 Hz

Fig. 15: B-format polar plots around 10 kHz

capture, albeit only in the horizontal plane. We hopeto follow this with a 12 capsule design which shouldallow for full periphonic second order capture.

The entire calibration process needs a complete rethinkand rewrite. While the goal is not a “push one buttonand you are done” program, a much better and inte-grated software package needs to be written.

6 Summary

Given access to a 3D printer these designs are quiteeasy to build and are suitable for testing and experi-mentation by students of our courses and the commu-nity at large. Soldering skills are needed for wiring the

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

507 to 553 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

2056 to 2244 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

4932 to 5383 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

6999 to 7639 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

9932 to 10840 Hz

-1 -0.5 0 0.5 1

-1

-0.5

0

0.5

1

14094 to 15382 Hz

Fig. 16: virtual cardioid polar plots showing effects ofslight B-format component phase shift

capsules and associated electronics, and careful mea-surements are necessary for a successful calibration,but the 3D printer makes it possible to create micro-phone arrays with precise capsule orientations and verygood performance. The parametric models can adaptthe design to use other capsules and/or change the sizeor geometry of the array. The measurements of thefilter matrix created with our preliminary calibrationsoftware and informal listening tests verify we are ableto build and calibrate a microphone with very goodperformance characteristics.

7 Acknowledgments

Many thanks to Aaron Heller, without his advice, soft-ware and encouragement I would only have four cap-sules mounted in a precise tetrahedron and not muchmore. Aaron together with Eric Benjamin (and others)listened to many long email threads detailing my workand provided useful comments and thoughtful advice.Their experience in the microphone and Ambisonicsworlds has been invaluable. Thanks also to Joe An-derson for his help in putting together the first versionof this paper and its submission to this conference andFons Adriaensen for his software and advice. Also toElliot Kermit-Canfield for his help in measuring themicrophone and to Jay Kadis for help in finding andbuying the proper electronic components as well ashelping to wire the first prototype. John Granzow andthen Romain Michon helped deal with the world of 3Dprinting, it can be frustrating but in the end it is alsorewarding. And Sasha Leitman, of course, without thefabrication facilities she maintains in the MaxLab atCCRMA this project would not exist.



References

[1] Free Software Foundation, GNU General PublicLicense, http://www.gnu.org/licenses/gpl-3.0.en.html

[2] Creative Commons, CC Attribution NonCommercialShareAlike 4.0 International,http://creativecommons.org/licenses/by-nc-sa/4.0/

[3] OpenSCAD, The Programmers Solid 3D CADModeller, http://www.openscad.org/

[4] TLS Products, SoundField Microphones,http://www.tslproducts.com/soundfield-type/soundfield-microphones/

[5] Core Sound, TetraMic,http://www.core-sound.com/TetraMic/1.php

[6] Brahma: Affordable Ambisonics Microphone,https://www.kickstarter.com/projects/1569945514/brahma-affordable-ambisonics-microphone/description

[7] Ultimaker 2 Plus Extended 3D printer,https://ultimaker.com/en/products/ultimaker-2-plus

[8] Michael Gerzon, “The Design of Precisely CoincidentMicrophone Arrays for Stereo and Surround Sound”,50th Audio Engineering Society Convention, PreprintL-20, London, 1975

[9] Dan T. Hemingson, Mark Sarisky, “A PracticalComparison of Three Tetrahedral AmbisonicMicrophones”, Audio Engineering Society ConventionPaper 7676, 126th Convention, 2009

[10] R. H. Hardin and N. J. A. Sloane, “McLaren’sImproved Snub Cube and Other New SphericalDesigns in Three Dimensions”, Discrete andComputational Geometry, 15 (1996), pp. 429-441

[11] Sébastien Moreau, “Étude et réalisation d’outilsavancés d’encodage spatial pour la technique despatialisation sonore Higher Order Ambisonics:microphone 3D et contrôle de distance”, Thèse deDoctorar, Spécialité: Acoustique, 2006, ÉcoleDoctorale de L’Université du Maine, Lemans, France

[12] Peter Plessas, “Rigid Sphere Microphone Arrays forSpatial Recording and Holography”, Diploma thesis inElectrical Engineering - Audio Engineering, 2009,IEM Institute of Electronic Music and AcousticsUniversity of Music and Performing Arts Graz, Austria

[13] Eric Benjamin, “A second-order soundfieldmicrophone with improved polar pattern shape”,Audio Engineering Society Convention Paper 8728,133rd Convention, San Francisco, 2012

[14] Eric Benjamin, “Extending Quasi-AnechoicElectroacoustic Measurements to Low Frequencies”,Audio Engineering Society Paper 6218, 117thConvention, San Francisco, 2004

[15] Abhaya Parthy, Craig Jin and André van Schaik,“Acoustic Holography with a Concentric Rigid andopen Spherical Microphone Array”, Acoustics, Speechand Signal Processing, 2009. ICASSP 2009. IEEEInternational Conference

[16] Fons Adriansen, “Aliki, an impulse responsemeasurement tool”,http://kokkinizita.linuxaudio.org/linuxaudio/

[17] Fons Adriansen, “A Tetrahedral Microphone Processorfor Ambisonic Recording”, Linux Audio Conference2007, TU Berlin, Germany

[18] Fons Adriansen, “TetraProc and TetraCallScreenshots”, http://kokkinizita.linuxaudio.org/linuxaudio/tetra-pict.html

[19] Angelo Farina, “A-format to B-format conversion”,http://pcfarina.eng.unipr.it/Public/B-format/A2B-conversion/A2B.htm

[20] Sound Field Microphone Mark 4 Control ModuleInput Circuit,http://www.ai.sri.com/ajh/ambisonics/schematic-5.pdf

[21] Christof Faller and Mihailo Kolundzija, “Design andLimitations of Non-Coincidence Correction Filters forSoundfield Microphones”, Audio Engineering Society,126th Convention, 2009, Munich, Germany

[22] Robert Mores and Ralf Hendrych, “Two alternativeminimum-phase filters tested perceptually”, AudioEngineering Society, 140th Convention, 2016, Paris,France

[23] Aaron Heller, “Derivation of the A-to-B matrix for acoincident array of first-order microphones”,unpublished paper, 2007.

[24] Aaron Heller, Eric Benjamin, “Calibration ofSoundfield Microphones using the Diffuse-FieldResponse”, Audio Engineering Society ConventionPaper 8711, 133rd Convention, San Francisco, USA

[25] Richard Lee, “Simplest possible P48V circuit withPanasonic WM61a”, SimpleP48wm61.doc & pdf,posted to the MicBuilders Yahoo group.

[26] Zapnspark, “Generic Back Electret CondenserMicrophone”, http://www.sdiy.org/oid/mics.html,2007


audio engineering society conference papernando/publications/sphear.pdf · audio engineering...

Documents