an electroacoustic pipe organ using membrane driven pipes

Godfried-Willem RAES



When we first designed the Singing Bicycle Project  in 1976, calling it our 'Second Symphony' with some irony, we calculated a series of 24 pipes such that they would all be tuned to ideal harmonics of one single fundamental pitch.  Each pipe was excited by an impedance matched loudspeaker driven by the bicycle dynamo.  When mounted on the bicycle, the resonant frequency of the tube sounds at a specific speed, different for each bicycle. The piece was conceived as using 'just intonation' and hence the pipe lengths were cut to the 24-note series: 182 161 159,5 158 156 153,5 150 146 ,143 140 135,5 130,5 126,5 124 121,5 115,5 110 108 103,5 100,5 98,5 95,5 93,5 85,5 covering an interval of an octave and a fifth.  Although the symphony has been performed over a hundredth times since 1980, we never actually verified the acoustic and harmonic result against what we theoretically conceived...
In 2014 we performed a series of precise measurements and -not really amazing- found out that the initial concept belongs in the land of fairy tales. We were not really amazed because in the last twenty years we have been involved quite deeply into acoustic research on musical instruments and robotics. This research led us to the conviction that there is not such a thing as 'harmony' or 'just intonation' anywhere in the world of acoustics, be it strings, pipes ,let alone more 3-dimensional vibrating objects.

Loudspeakers driving cylindrical tubes in fact form a pretty complex acoustic system. The traditional acoustic theory with regard to open and closed pipes, does not seem to apply here. The pipes do not function merely as resonators, but are an integrated part of a complex oscillating system. The one end, closed by the loudspeaker does neither behave as a closed nor as an open end. The loudspeaker here behaves as a membrane driven in its center and shows off a highly nonlinear response curve. Thus we have to do with two mutually dependent vibrating systems. The table below gives the results of a long series of measurements, revealing this clearly.

Setup for the experiment:
Visaton loudspeaker, Type K50WP, 50 Ohms, 3 W. Frequency response 180-17000 Hz. Resonant frequency 300 Hz. For details and specifications, see spec. sheet. The response curve looks like this: Visaton_response
PVC Pipe 50 mm diameter (internal 46 mm), fault 0.1 mm. Measurement fault on pipe lengths: 0.5 mm. The speakers were glued firmly and airtight to the tubes using PVC cement glue. (Tangit). Here is an example:


The first column in the table gives the pipe length. The second columns gives the ratio length to internal diameter. The third column gives the resonant frequency for such a tube resonating at 1/4 wavelength and with end correction applied according to the textbook formula found in most books on acoustics : formula1 The fourth column gives the lowest resonant frequency as measured by sweeping the oscillator from near 0 Hz upwards. The next columns give the measured frequencies of the next clearly discernible overtones as well as their ratio to the fundamental. The excitation was always a pure sinewave.

Physical pipe length l/d fres
f0 f1, f1/f0 f2, f2/f0 f3 , f3/f0
48.5 mm
62 mm
78.5 mm
1008 [3.05]
2600 [7.88]
395 mm 8.59 160
170 362 [2.14] 664 [3.91] 1092 [6.42]
451 mm  (Mib 51)
155.6 310 [1.99]
340 [2.18]
583 [3.75]
933 [5.99]
465 mm 10.11
151 302 [2.00](*) 574 [3.80] 916 [6.06]
583 mm
326 [2.52] 462 [3.58]  
600 mm
323 [2.48]
489 [3.76] 989 [7.61]
627 mm 13.63
121 311 [2.57] 444 [3.66]  
683 mm
296 [2.64]
316 [2.82] 633 [5.65]
700 mm
287 [2.66]
428 [3.96] 619 [5.73]
774 mm
260 [2.65]
382 [3.90]
560 [5.71]
805 mm
287 [2.98] 385 [4.01] 950 [9.89]
847 mm 18.41
90 238 [2.64] 356 [3.95] 520 [5.77]
880 mm
262 [2.94] 352 [3.95]

700 [7.86]

865 [9.72]

924 mm
253 [2.99] 338 [4.00] 659 [7.80]
962.5 mm
242 [2.95] 330 [4.02] 628 [7.66]
1005 mm
235 [3.01] 359 [4.6] 772 [9.89]
1125 mm
189 [2.95] 250 [3.90] 296 [4.63]
384 [6.00]
1145 mm
205 [2.94] 322 [4.63] 410 [5.89]
1198 mm 26.04
67 201 [3.00] 311 [4.64]  
 1366 mm
59 175 [2.96]  277 [4.69]  356 [6.03]
 1425 mm
56 170 [3.03]  281 [5.02]  351 [6.27]
1562 mm
158 [2.98]
262 [4.94] 347 [6.55]

(*) The audible result we observed here is two separate tones, one octave different. The fundamental f0 sounds together with f1.

From the data, it will be clear that around the resonant frequency of the loudspeaker (300 Hz) either pulling or pushing of the total resonance at the corresponding pitch does occur. Below the loudspeaker resonance frequency, the pipes behave pretty well as pipes closed at one end, although even overtones are certainly present.  The larger the l/d ratio, the more the overtone series corresponds to a series off odd multiples of the fundamental, except for those overtones that fall in the region of the loudspeaker resonance.  Here are some measurements for 25 mm diameter tubes (internal diameter = 22 mm) , mounted on the speakers using a slightly conical adapter:


length L / d f0 f1, f1/f0 f2, f2/f0 f3, f3/f0
251.5 mm 11.4 169 511 [3.02] 955 [5.65] 1526 [9.03]
200 mm 9.09 190 572 [3.01] 1150 [6.05] 1876 [9.87]
186 mm 8.45 196 603 [3.08] 1199 [6.12] 2026 [10.34]
146 mm 6.6 220 695 [3.16] 1492 [6.78] 2605 [11.84]
125 mm 5.68 233 762 [3.27] 1715 [7.36] -

Measurements were performed using an analog sine wave generator with a digital frequency counter. Resolution 1 Hz. The output impedance is 50 Ohms and thus matches the nominal impedance of the drivers. The precision is better than 1%. Output voltage was kept constant at 3 V rms. for all measurements. For most tubes, some resonance was also observed around the octave and other even overtones, but at a level too low to be easily measured.  For measurement of the sound level and determination of the resonance peaks we used a calibrated sound level meter placed on a tripod at 30 cm distance from the open end of the tube.
The larger the l/d ratio becomes, the closer the first overtone comes to 3.0 times the fundamental. The second overtone seems to move from 3 times to 5 times the fundamental.  The factor for the third overtone is from our data not easily discerned. It also seems to increase with increasing tube length.

An important observation was also that the resonant frequency shifts clearly upwards with increasing excitation amplitude. This may be attributed to a change in the velocity of sound in function of the pressure in a constricted volume.

The precision of the measurements presented here is not high enough to clearly prove the inharmonicity of the overtone series, let alone to derive laws on the underlying math. It is indicative though. Exact measurements would require 1 cent resolution for frequency measurement and hence, say for midi note 36 (f= 65.41) a time interval of at least 400 seconds. Of course the time interval required for 0.01% precision measurements, halves as the frequency to be measured doubles. To increase the reliability of our measurements using a function generator with frequency readout (time base limited to 1 second, hence frequency resolution = 1Hz), we used a Korg tuner with an analog readout and a precision of ca. 2 cents. This precision though is only valid for frequencies conforming to the scale of equal temperament at a given diapason (here set to 440 Hz). We had a Hameg function generator with a 0.1 Hz resolution on the read out available as well, but this piece of equipment makes use of digitally generated sine waves and hence the frequency runs up stepwise, preventing us to hit the exact resonance's.

We used the measurement data for 50 mm tubes to obtain a third degree equation, using a Gauss fit procedure. The equation becomes:

freq = 309.048 - 0.4617146 *  x  + 3.013809E-04 * x^2 - 7.109726E-08 * x^3

The tube length (x) is expressed in mm.

A 6th degree equation, obtained with a Gaussfit program calculated over 20 data pairs taken in the length traject between 395 mm and 1582 mm leads to errors less than 1%. It looks like this:

freq =  317.6497 -.4827873 x  + 2.720424E-04 x^2 + 2.693052E-08 x^3  -5.110973E-11 x^4 -1.656992E-14 x^5 + 1.248304E-17 x^6

Apparently, this approach does not seem to lead to a better understanding of the laws governing our observations, however it is quite useful for the calculation of pipe lengths extrapolated for parameters within the range of the data set as long as this data set is kept narrow.

Towards an experimental hybrid organ...

As the sound produced by these resonating tubes was pretty 'acoustic' in nature, we decided to design and build an automated instrument based entirely on this principle. The electric excitation signal is by default pure sinusoidal, but the waveform produced by the pipes at resonance is by far not a sinewave but contains quite a broad and varying spectrum clearly produced by the non linear behavior of the pipe resonators.  We baptized it 'Hybr' since the design is somewhere in between an electronic and an acoustic instrument. Each pipe is controlled by an individual oscillator of which excitation amplitude, envelope and microtuning can be fully controlled via MIDI commands. The experimentally determined overtones for each pipe can be added to the sine wave fundamental in programmable amplitude proportions. A form of inharmonic additive synthesis one could call this.  To this end we called in no less then 20 ARM 32 bit microprocessors (STM32F407) as each processor is capable of generating only two independent drive signals. The timbre of the instrument is characterized by a low  formant frequency of 300 Hz, caused by the resonant frequency of the loudspeakers. Of course this only holds for pitches below the resonant frequency.

The instrument belongs to the category of the organs. In traditional organ building, membrane driven pipes have been used occasionally in the 19th century: the diaphane register, mentioned and described by Audsley, is an example. This instrument however, is quite different and offers many more possibilities, as the resonance of the pipes can be fully controlled through electronic means. An important advantage over any other organ-type instrument is that here we can have a very broad control over dynamics, the tuning of the pipes being independent of wind pressure. The global sound volume can be controlled as well as the individual loudness of each individual pipe. Moreover, our implementation makes it possible to use the instrument in just about any imaginable tuning system, as each pipe can be detuned a quartertone up or down referred to equal temperament intonation.

Definitive pipes as made, tuned and measured::

  f0 l/d f1 f2 f3 L De Di SPL
165 [3.00]
219 [3.98]
270 [4.91]
175 [3.00]
233 [4.00]
288 [4.94]
183 [2.96]
245 [3.97]
286 [4.63]
36 65.4  26.4  195 [2.98]
 258 [3.94]  306 [4.68]  1216  50  46.4  8155.91
37 69.3  24.8  210 [3.03]
 274 [3.95]
 323 [4.66]
 1142  50  46.4  7656.19
38 73.4  23.3  218 [2.97]
 291 [3.96]
 344 [4.68]
 1073  50  46.4  8256.84
39 77.8 21.8 235 [3.25] 359[4.61] 772 [9.92] 1005 50 46.4  58.14
40 82.4 20.5 244 [2.96] 330 [4.00] 468 [5.68] 949.5 50 46.4  8258.79
41 87.3 19.9 262 [3.00] 352 [4.03] 700 [8.02] 880  50  46.4  60.02
42 92.5  17.9  274 [2.96]
377 [4.08]   523 [5.65]
 833  50  46.4  8260.80
43 98.0 16.8 193 [1.97] 260 [2.65] 384 [3.92] 780.5 50 46.4  8654.73
44 103.8  16.6  202 [1.95]
271 [2.61]  412 [3.97]   732  50  44  8855.52
45 110.0  15.6  212 [2.02]
291 [2.64]  428 [3.89]
 686  50  44  8857.16
46 116.5  14.6  231 [1.98]
 312 [2.68]
444 [3.82]    643  50  44  8457.84
47 123.5  13.7  246 [1.99]
328 [2.66]  460 [3.72]   601  50  44  8858.93
48 130.8  12.8  262 [2.00]
 342 [2.61]
484 [3.7]  563.5   50  44  9060.02
49 138.6  11.9  286 [2.06]
356 [2.57]  614 [4.43]   523.5  50  44  9161.54
50 146.8  11.1  302 [2.06]
 642 [4.37]
858 [5.84]   488.6  50  44  8962.48
51 155.6 9.8 310 [1.99] 340 [2.18] 583 [3.74] 451 50 46.4  62.94
52 164.8 10.9  341[2.07]  651[3.95]  1035 [6.28]
405 40 37  8864.58
53 174.6 9.8  387 [2.21]
 725 [4.15]
1160 [6.64]  362 40 37  9366.77
54 185 9.5  395 [2.13]
739 [3.99]   1199 [6.45]
352 40 37  8767.13
55 196 9.3  497 [2.53]
996 [5.08]  1596 [8.14]  263 32 28.4  9371.11
56 207.6 7.7  549 [2.64]
1146 [5.52]  1848 [8.90]  220 32 28.4  8972.83
57 220 7.3  545 [2.48]
1196 [5.44]  1964 [8.92]  208 32 28.4  72.70
58 233.1 6.18
763 [3.27] 1024 [4.39] 1728 [7.41]
132.2 25 21.4  9278.51
59 246.9  4.97  840 [3.4]
1900 [7.69]  -  109.5  25  22  8380.19
60 261.6  51.0  466 [1.78]
664 [2.54]  863 [3.30]   689  20  17  8369.99
61 277.2  38.3  470 [1.69]
 682 [2.46]
900 [3.25]  652  20  17  8670.14
62 293.7  35.7  513 [1.75]
726 [2.47]
962 [3.27]   607  20  17  71.66
63 311.1  33.2  538 [1.73]
 768 [2.47]
1004 [3.22]  565  20  17  8672.48
64 329.6  30.0 606 [1.84]  978 [2.97]  1129 [3.43]   511  20  17  74.54
65 349.2  28.1 628 [1.80]  909 [2.6]  1206 [3.46]  478   20  17  75.16
66 370  26.5 658 [1.78]  936 [2.53]  1243 [3.36]  451   20  17 83 75.97
67 392  31.4  711 [1.81]
973 [2.48]  1312 [3.35]   427  16  13.6  77.31
68 415.3  29.0  750 [1.81]
1005 [2.42]  1368 [3.3]   395  16  13.6  7878.23
69 440.0  27.2  764 [1.74]
1063 [2.42]  1450 [3.3]  370   16  13.6  78.55
70 466.2  25.2  776 [1.66]
1104 [2.37]  1544 [3.31]  342   16  13.6 85 78.82
71 493.9  23.4 820 [1.66]  1186 [2.4]  1665 [3.37]   318.1  16  13.6 86 79.78
72 523.2  21.3  893 [1.70]
 1325 [2.53]
1866 [3.56]   290.2 16   13  8781.25

Pipe 58, driven at resonance with our sine wave generator, shows up a 4.94 Vrms voltage across the speaker terminals. Power delivered to the speaker hence is 500 mW. The sound pressure level produced by the pipe is 92 dBA measured at 30cm distance from the open end of the pipe.

For all pipe diameters smaller than 50 mm, we used a conical adapter piece. This fundamentally changes the acoustic behavior of the system, as it introduces a third interacting resonator between the speaker and the pipe. The own frequency of this system  was found out to be situated between 80 Hz and 108 Hz, the exact value going down with increasing length of the coupled pipe. We could only avoid this by using speakers of decreasing diameters. Unfortunately, they do not seem to exist. Speaker systems for headphones and earplugs  do exist in small diameters down to 10 mm, but their power rating is limited to 100 mW. One of the reasons we selected the Visaton speakers of the type specified in this project is that they have mylar membranes and not cardboard, thus they are to a certain extend weather and moisture resistant. The only issue is that the pipes have to be kept free from objects, dust and granulates falling into them, as this will affect the sound.  Turning the instrument upside down and gently shaking it will be the cure.

Midi implementation and mapping:


Midi channel: 8 (counting 0-15)

Note Off: notes 33 to 72 , note release implemented.
Note On: notes 33 to 72 , velo implemented.


Keypressure: used to modulate the pitch of each individual sounding note. Value=64 gives standard pitch, values 63 to bend the pitch down, values 64 to 127 bend upwards. The range is 50 cents up or down. This implementation makes it possible to play the instrument in just about any imaginable microtuning. The value of the individual note microtuning settings is 'sticky', thus users do not have to send it with every note on command.  A note-off command will not reset the microtuning of the corresponding note. This implementation deviates from standard midi! Before using aftertouch, controller 124 should be set to value 64 to acquire the right scaling! Once this controller is set, each note should be initialized with it's right aftertouch.


#1: Wind noise or pitch instability. (Default value 0)
#3: Vibrato depth (FM) (default value 0)
#4: Vibrato speed (FM) (default value 0)
#5: Tremolo depth (AM) (default value 0)
#6: Tremolo speed (AM) (default value 0)
#7: global volume (default value 64)
#16: global attack control (attack time) (default value 1)
#17: global maximum velo amplitude (default value 127)
#18: global decay time (default value 1)
#19: global release time (default value 1)
#20: Global tuning (range, a quartertone up or down). Default value 64 for A= 440 Hz (*)
#23 to # 62: Amplitude of  the first inharmonic overtone on pipes for notes 33 to 72. The frequencies for these overtones correspond with the pipe table given above. (Default values 0)
#66: Power on/off switch. Power off disables the amplifiers by muting them but does not unpower them. Power off resets all controllers to their startup default values.
#67: Volume control for the common external audio input (default value 0)
#73 to #112: Amplitude of the second inharmonic overtone  for notes 33 to 72. The frequencies appear in the pipe table given above. (Default values 0)
#123: all notes off, preserving all controller settings including microtuning.
#124: initialization of aftertouch. In case aftertouch is used, this controller should be set to value 64. Once this controller is set, aftertouch for all notes should be set to 64 for all notes for normal pitch. In case no aftertouch is used, this controller should not be set.

(*) If the tuning is set much different from 440Hz, the overall amplitudes will be affected, unless tuning follows the changes of resonance frequencies as a result of temperature changes.  The range is a quartertone up or down. Parameter value 64 equals A=440Hz)

Controller #13 can be used to change the waveform of the excitation.

Channel aftertouch and pitch bend are  not implemented on <Hybr>. Channel aftertouch would not be very meaningful for such a polyphonic instrument and pitch bend is contrary to the acoustic nature of the instrument. Moreover, global small pitch inflections can be performed using the tuning controller. Polyphonic pitch inflections can be performed by using he keypressure command.

In addition to the features described before, we implemented  an operational mode wherein the instrument can be used as an acoustic fourrier transformer with equal tempered frequency bins.  Therefore it is required to feed all 40 loudspeakers with one and the same audio signal, limited in bandwidth (fc= 1500 Hz)  to the range of the instrument. The sampling rate for this channel is pretty low (>= 3 kS/s) and the resolution is 12 bits. The common audio drive input requires a low impedance (< 600 Ohm)  line level signal. The level of this signal as presented to the pipes can be controlled with midi controller #67.

The design of this instrument may give rise to a lot of practical applications even in traditional organ building. In this project we used only cylindrical pipes, but obviously pipe shapes such as used in traditional pipe organs can be used as well, making it possible to have full sets of registers. The advantage of such membrane driven pipes in traditional organs being that they are by design to a much higher degree  free of maintenance as they do not require periodic tuning. Also, this technology is inherently free of wind noise caused by leaks as well as compressor motor noise.

A word of warning to our users: by design, electrically overloading the speakers such that they burn out is almost impossible. The same applies to the amplifiers. However, due to the resonance's occurring in the pipes, it is very well possible to ruin the speakers by excessive sound volumes. This occurs whenever the membrane displacement due to mechanical resonance exceeds the allowable maximum. As this limit is approached, a clear change in sound quality (distortion) will be observed. This should be understood as an imperative to reduce the volume, either by sending a lower setting for controller #7 (volume), or by using lower velocity values.

Technical specifications:

Design, research and construction: dr.Godfried-Willem Raes (2014)

Collaborators on the construction of this robot:

Music composed for <Hybr>:

Godfried-Willem Raes 'Namuda Study #49: Hybr', premiere January 13th 2015 by Dominica Eyckmans and Emilie De Vlam..

Godfried-Willem Raes 'Namuda Study #57: Tekstuur', premiere October 28th 2015 by Dominica Eyckmans.

Godfried-Willem Raes 'Bottom Boot', premiered september 2016 by Emilie De Vlam (This performance also includes <HybrHi> and <HybrLo>).

Kristof Lauwers 'Study for Hybr', 2015

Xavier Verhelst


Back to Logos-Projects page : projects.html Back to Main Logos page:index.html To Godfried-Willem Raes personal homepage... To Instrument catalogue Naar Godfried-Willem Raes' homepage

Construction diary:

03.10.2014: Construction of the first test-pipes based on the new designs for the Singing Bicycle Symphony.
04.10.2014: Construction of some prototype pipes with different diameters. Gray PVC pipe diameters we can easily get are 50 mm, 40 mm, 32 mm, 25 mm, 20 mm. The performed measurements lead us to design the pipes such that the ratio pipe-length to internal diameter stays larger than 9. A first version of a pipe calculation utility is programmed and debugged. pipes
05.10.2014: More prototype pipes build and measured. Interesting discovery: the resonant frequency depends on the excitation level.... It goes up with increasing sound pressure. Furthermore, the curve of resonance is not at all symmetrical. It has a slower slope rising than descending.
06.10.2014: Design of a  power amplifier to drive the speakers. Here is our first circuit: poweramp Note that all audio-in inputs are connected together such that all amps can be operated with a single audio signal. In this mode, the instrument operates as a chromatic spectrum transformer. The input impedance for this common input is 550 Ohms. The AXO input, connects to one of the two audio outputs of the digital synthesis boards using an ARM processor. The ANA input is used to connect to the outputs of individual analog modified sine wave generators. The diodes over the output terminals protect the amp against voltages higher than the supply voltage as these do occur at resonance over the speakers. However, we need at least 38 of these amps... A calculation of the costs leads to ca. 30 Euro per channel, so the total bill runs up to 1.200 Euro. As an alternative, we could also use ready made 4-channel car amplifier modules as produced by BaseTech, type AP-4012. The specs are 4 x 11.5 W rms output power. We ordered 10 such modules from Conrad at 25 Euro's a piece. So, this brings this part of the costs down to 250 Euro. There is no way to solve this problem cheaper by any design we could think of. The power supply for all these amplifiers should be rated for 10 A at 12 V DC.
07.10.2014: Purchase of extra lengths 50 mm PVC tube. Construction, measurement and fine tuning of pipes for A 33, Bb 34, B 35.  This is our testing set up: testingThick (25 mm) PVC plate material ordered from Eriks for the construction of the 'windchest' and pipe holder. Sketches made for the possible final design. Looks like it will have to become a six-wheeler, as we have to keep building height to the minimum required. Designing utility program for the calculation of the pipes much improved.
08.10.2014: Ten more pipes tuned and measured. As we run out of stock in 50 mm 1.8 mm thick pipe, we now use 50 mm 3 mm thick pipe material.
09.10.2014: Experiments performed with 32 Ohm speaker drivers from headsets. The power specifications for these are only 100 mW, but their diameter is more suitable for the smaller pipes. Sound pressure meter as used for our measurements: db-meter
10.10.2014: <Hybr> project temporarily postponed because of too many interfering projects.
11.10.2014: Some pictures of the pipes made so far:pipes
20.10.2014: Further work on the sketches for the robot.
25.10.2014: The ordered conical PVC adapter pieces as well as PVC pipe in smaller diameters came in from Eriks. We can continue building and tuning pipes.
26.10.2014: Twenty Axo boards commissioned from Johannes Taelman.
29.10.2014: Construction and measurement of pipes in 10 mm, 12 mm, 16 mm and 20 mm diameters, with cones to couple them to to 50 mm speakers. Data sets added to our calculation software for Hybr. For the highest pipe we will go for 16 mm outer diameter, although even 12 mm leads to a good sounding pipe, be it somewhat softer.  Discovery that with a conical piece and coupled to pipes much smaller than the loudspeaker diameter, the pipes behave acoustically as half wave resonators,  so as pipes open  at both ends.  The spectrum observed is compressed. This applies to 16 mm and 20 mm pipes.
30.10.2014: Extra PVC adapter pieces ordered from Eriks. Still waiting for the thick PVC plates to flow in. All 16 mm diameter pipes finished. Four 20 mm pipes ready. Calculation software adapted and measurement data entered. Here is a picture of the highest nine pipes:small hybr pipes Six 16 mm pipes and three 20 mm pipes.
31.10.2014: Design of the electronic circuitry. Decision to make use of 20 ARM STM32F4 discovery boards, each board steering two pipes. Design of a PC board to mount two of the boards on a single Eurocard with Weidmueller connector for 5 V and Midi, as well as a common analog input rail.ARM_Board The 1/8" stereo output jacks will be unsoldered and the connections redone using wires. A single Eurocard will then control four notes. The documentation for the ARM board is on the website: http://www.st.com/stm32f4-discovery. The software platform we will use to develop the firmware is Johannes Taelman's Axoloti project.
01.11.2014: Mounting PC board, Eurocard size, drawn to accommodate two ARM-discovery boards. Thus each board will have four note outputs and connects to its own four-channel power amp. Ten of these board will be required. Etching will be for tomorrow or later... The board (reduce to 25% for printing) can be downloaded here. As -by design- we have only 1000 mm width available for mounting, we decided to mount the board on swivels, such that they can be folded out for programming. If we mount them flat, the USB connectors required for uploading the firmware would become inaccessible
02.11.2014: Fifty cheap RCA phono plug connectors as well as two row 2.54 mm sockets for the ARM boards ordered from Farnell. Twenty small hinges bought on the flea market to be used for mounting our PC boards. Technical construction drawing worked out. It looks like it will be hard to keep the building height below 2 meters... Anyhow, that's still about 10 cm smaller than <Krum>...
03.11.2014: Start construction of the chassis: cutting and drilling stainless steel 40 x 40 x 4 T profile and  plate material 325 x 1000 x 3 mm. Central wheels will be 400 mm diameter. PVC adapter pieces delivered by Eriks.
04.11.2014: All 40 pipes finished now. Pipes 33 to 59 behave (more or less) as 1/4 wavelength resonators, pipes 60-72 as 1/2 wavelength resonators.
05.11.2014: Etching, drilling and soldering of the ten PC boards for the ARM-processors started.  Etching takes longer than we thought...Two boards completed so far.
06.11.2014: All boards exposed to UV and developed. Eight boards etched... more to come. A day of drilling in sight. Firmware discussion with Kristof Lauwers. Mattias Parent called in to help out with the drilling.

07.11.2014: Drilling, soldering and mounting works...eurocard board. This is the solder side of those boards: solder side

08.11.2014: Drilling and threading of the holes and M4 threads for the power amplifiers.
09.11.2014: Further soldering work on the boards.
10.11.2014: Running out of connectors. New order placed at Farnell. Five boards finished now.  Halfway that means. These can already be programmed.
11.11.2014: TIG-Welding works on the chassis parts.
12.11.2014: Already six ARM boards mounted on hinges. Still waiting for the thick PVC plates to make the pipe carrier plate. Nine boards soldered and ready.
13.11.2014: Two more PC boards for the ARM processors exposed, developed, etched and drilled. T-profiles welded to the large chassis plates. Start mounting of the power supply components. Fresh load of connectors came flowing in from Farnell.
14.11.2014: Start assembly of the power supply components. We stick to pure analog here, avoiding all spurious noises that could be caused by switch-mode power supplies. For both power supply voltages we make use of toroidal transformers.
15.11.2014: Wiring and testing of the power supplies.power

17.11.2014: All 10 quad power amps mounted on the back plate.amps All BaseTech amps mounted with M4 x 10 stainless steel bolts. Threads for the hinges  finished on the front plate. Here we have to use M4 x 6 countersunk bolts.
19.11.2014: Further tedious soldering work on the ARM carrier boards. Should be finished by the end of the day...
20.11.2014: Development task for the Axoloti patches for the ARM processors confined to Kristof Lauwers, with assistance from Johannes Taelman, who developed the firmware. For the wiring, we will call in the help of Mattias Parent. In the meanwhile, starting soldering of the midihub board.  Still waiting for the thick PVC material for the 'windchest'...
21.11.2014: Start writing test and debugging code in GMT.

22.11.2014: First tentative mount of the two large plates together.wheelmount Presenting the wheels, after cutting the spindle. Start construction of the holders for the small wheels. Seems we drilled the holes for the midihub board on the wrong side of the panel... Looks like we will have to leave it there, as changing it involves removing all the ARM boards again...hub side The space between the two plates is exactly 96 mm. News came from Eriks: the PVC plate material should be delivered on Monday...
23.11.2014: Continuing welding work on the small wheel holders. Looks like we will be running out of Argon gas soon...
24.11.2014: Finishing the small wheel holders. Order for extra wheels 125 x 35. Design of a polycarbonate protective plate for the frontal boards. Awaiting the delivery of the PVC plates... Firmware development by Kristof Lauwers. Continuing development of test code in GMT. Work session with Johannes Taelman. The PVC plates came in and so we could start the drilling of the holes in the pipe holder. The 50 mm holes pose some problem as the PVC melts in the hole saw and makes it very hard to remove the center from the saw. There must be a trick. Cooling with frozen CO2 didn't seem to help very much.drilling
25.11.2014: All 40 holes drilled and carefully honed to 50 mm. First two rows of pipes glued to the pipeholder. Rebuilding pipe for note 58 , using the same conical adapters as adjacent pipes. This pipe is now from 1.8 mm thick pipe and thus the length became a bit longer than that of the first made pipe.  Lookup table adapted accordingly.pipeholder
26.11.2014: All pipes glued to the pipeholder plate. Upper pipe holder for the longest pipes cut from 25 mm thick PVC. This measures 1000 x 100 x 25 mm and has 14 holes, diameter 50 mm. pipeholder low pipesDecided to use connectors between the speakers and the amplifier outputs as otherwise it seems pretty impossible to take the automate apart for repair and maintenance. First 30-pole connector mounted for the highest pipe rank. Two times 3 contacts remain uncommitted but may be used for 12 V  lights on the pipeholder. Extra 125 mm diameter spoke wheels ordered from Tente.glued pipes
27.11.2014: Start wiring of the loudspeakers and amplifiers. By the end of the day, we got almost halfway the wiring of the speakers and connectors on the pipeholder. The construction of the holders for the connectors took a lot of time. We used two 30 pole 3-row connectors taken from the telephone exchange company and one Bulgin 36 pole connector as used in our very first version of the player piano. We also constructed two lights on the top of the pipeholder. The lights have E14 (mignon)  sockets and should get 15 V/ 4 W bulbs. It will be possible to take of the entire pipeholder assembly from the stainless steel chariot just by loosening the three connectors and the bolts holding the robot together.
28.11.2014: DIN916 BIZK RVS-A1 rings bought at MEA. Mid row speaker wiring to connector finished. Wiring for the Bulgin connector on the last row drawn out.
29.11.2014: Wiring of all speaker connectors finished. 12 V and 24 V bulbs with E14 fittings ordered from Conrad. Common ground line constructed for all ten power amps. Common +12 V line made as well. These lines are made of 4 mm2 copper wire. We preserved the individual fuses in the positive power line of each individual amplifier. As the datasheet for the BaseTech amplifiers do not specify much, and we wanted to make sure not to blow all of them by connecting them to our non-stabilised power supply, we opened one to see what chips are used: two STA540 quad amplifier chips, each connected in bridge. The maximum supply voltage for these chips under operating conditions is 22 V, so we are safe with our 17 V peak power supply. Quiescent current is typical 80 mA, so with ten BaseTech amps, each using two chips, we should count on a quiescent current of no less than 1.6 A. The input filtering provided on the BaseTech amplifiers make use of a single TLO84 quad opamp chip.
30.11.2014: Finishing of midihub board. Firmware version 1.0 written and flashed into the microcontroller. Board mounted on the chassis.  Wiring can continue... Some 40 RCA connectors soldered. All 5 V power and midi lines wired from hub board to ARM boards. By the end of the day, 85% of all wiring finished. Power tests performed. No failures.
01.12.2014: Finishing wiring: signal input connectors and audio input lines.  Drilling of the two ground connection holes, one for mains ground connection, one for the negative 12 V supply line. Start construction of the audio-input board. Reichelt elektronik still has our connectors in their catalogue: FL830, Federleiste, 30-polig DIN41622. This is a picture of our male parts: These are rated for 8 A current per contact. Audio input board calculated and assembled. This is the circuit: audio-input The board measures 160 mm x 60 mm and is to be mounted on the outside of the backpanel. Nominal input signal level is 0 dB (775 mV rms). Input impedance is 600 Ohm. Maximum allowable signal: +6 dB (1.5 V rms or 4.3 Vpp), above this level clipping will occur. The input and output impedance of the Butterworth filter is 582 Ohms. At the output this is achieved by the parallel resistance of the 680 Ohm resistor and the load formed by the 20 ARM input impedance. At the input side it is achieved with the transformer. The outputs of the circuit have a DC offset of 1.5 V in order to accommodate the ARM inputs which have a 3.3 V range. We performed some measurements on our circuit and noticed it causes second harmonic distortion of the waveform on frequencies below 100 Hz if driven at full amplitude. By the end of the day, all wiring was done...
02.12.2014:  Start construction of the upper pipeholder and the handles. All performed using TIG welding. The vertical tubes are 50 mm x 2 mm x 480 mm and 50 mm x 1.5 mm x 480 mm. The thin pipe was very difficult to weld. Construction and placement of the holder for the back row pipes. This is secured with four M10 x 50 bolts.

03.12.2014: Mounting of the pipeholder on the base assembly. The frontal connector appears to have too short wires the reach the female. So, we extended the wires that were too short. The robot is ready to receive firmware and... to play. Kristof Lauwers is working on the patches in the Axoloti firmware. By accident we drilled a hole in the mid-row connector. We will have to replace it... Connectors ordered from Reichelt. Problems with uploading patches to the ARM boards..., so sounding <Hybr> is on hold for a while. Decided to change the two small gray wheels for pivoting wheels with green polyurethane tires as we encountered problems with the steering. The building height of these is 125 mm. Steering is perfect now. Connecting bolts and nuts finalized and tightened. Name plate 'Me fecit godfried-willem raes' mounted. Some ordered E14 fitting 12 V   lights came in from Conrad. We do not know as yet what colors to use here... green, tungsten or white?
04.12.2014: Running out of argon gas, so no welding possible... Stuck on Axoloti patch development...  White LED strip added on the front bottom side, below the ARM processor boards.
05.12.2014: Sawing and polishing of a large polycarbonate protective plate, thickness 8 mm,  for the front side of the robot. The standoff's on the underside have M10 x 140 bolts and a 90 mm long distance holder made in thick stainless steel. Still waiting for progress in the development of uploadable patches on the ARM boards.
06.12.2014: Construction of protection plate finalized. First test of the midi hub board. After correcting a minor bug in our GMT testing software the lights as well as the muting circuit (mapped on controller 66) do work to perfection. However as soon as we enable the amplifiers, a 100 Hz hum becomes audible through all the speakers.  The rest current of the amplifiers without signal totals some 2 A (calculated after the datasheet) and causes a voltage drop on the power supply voltage to 15.1 V. We measured the hum component on the power supply and indeed, it measures some 200 mV. To damp this we could either use a filter circuit with a power transistor, or an old passive solution: a third order butterworth low pass with a hefty inductance ( 6.8 mH, 0.12 Ohm resistance, copper wire diameter 2.5 mm). If we choose the -3dB cut off frequency at 1000 Hz, the damping at 100 Hz, a decade lower, should be -60dB.
07.12.2014: Mounting the low pass filter on the chassis. We used a piece of 20 x20 x 3 stainless steel  L-profile to mount the very large and heavy self inductance on the bottom side of the backpanel. It can be taken out by loosening the two M5 bolts.  For the electrolytic capacitor at the output we used a high quality 47000 uF/ 40 V Kemet type. The 12 V power connections at the filter output are accessible for measurements from the outside. Measurement of the hum component with this new filter gives a value of 3 mV, more than the calculated and expected value of 0.2 mV but still a substantial improvement. Any remaining hum must be due to ground loops, almost unavoidable in this design. Obviously the pipes for notes 43 and 44, resonating at 98 and 103.8 Hz respectively will amplify any residual 100 Hz hum, thus opposing the effectivity of our filter. Measurement of the rest-current (with mute disabled, so controller 66 on): 2.15 A.  Start cutting of a second polycarbonate protective plate for the backside. This plate will be less deep than the front plate. Sizes: 1000 mm x 220 mm x 8 mm.
08.12.2014: Further work on Axoloti patches by Kristof Lauwers. It still seems to be very shaky.  Waiting for a working version of an uploader...  Finishing of back panel polycarbonate plate in the meanwhile. To mount it, we had to replace the M10 x 70 cylinder bolts holding the pipeholder on the pipe mounting plate with countersunk type M10 x 60, with countersinking into the PVC plate. Otherwise it was impossible to mount the transparent backplate with two M8 x 25 bolts. backplateTestcode in GMT completed for all controller and features implemented so far by Kristof..
09.12.2014: We got <Hybr> as far as sounding two notes... 42 (F#) and 44 (G#), too soft for now but enough for evaluating the functionality of most of the controllers. Checked: Ctrl #20, tuning (quartertone up or down, default value 64), controller #1 (wind/randomness), #3,#4: tremolo, #5, #6: vibrato, #7 global volume. The audio signal level from the ARM board, being 30mV only, is too low to steer our amps. It needs to be at least a factor 10 (20 dB) higher. It should be possible to reach a 0 dB (775mV) level, even taking into account the 3.3 V ARM power voltage.
10.12.2014: Spare STA540 chips ordered from RS-components. Photoshoot with <Hybr>...
11.12.2014: Waiting for flash upload implementation under Axoloti from Johannes Taelman...
12.12.2014: In testing one of the ARM processors, we discovered a wiring bug in the Bulgin 33-pole connector for lowest rank of pipes. This is how the wiring ought to be:bulgin connector We resoldered the plug connector corresponding to this wiring diagram.
13.12.2014:  Output volume from the ARM board is now eight times higher. Tests with the notes <Hybr> can produce so far...
15.12.2014: Firmware upload seems to work now, although it's still a bit shaky and awkward on connecting as it requires two USB connectors for each board.  We got six notes playing so far... It is required to reboot the ARM boards after an upgrade of the patches.
16.12.2014: All firmware uploaded now together with Kristof's patches. Six notes seem to be shaky for some reason. Further debug required.
17.12.2014: A check of the voltages revealed a flaw in our design: the 5 V power supply is clearly underdimensioned now, as with all ARM boards working, the voltage sinks down to 3.5 V with shaky behavior for the midi signal levels as a consequence. We will have to replace the 15 VA transformer for the 5 V power supply with a 30 VA type. For test, we replaced the 5 V power supply by a Traco SMPS module capable of delivering 6 A. Indeed, now it works... All notes on Hybr do play now, but there will be still lots of fine tuning in the Axoloti patches required. First preliminary presentation and demonstration of <Hybr> under gesture control by the author at the occasion of the 'Snow' production of the robot orchestra.
18.12.2014: Further work on test code in GMT and preparation of some 'real' music sequences for evaluation. Although too late to be applied in <Hybr>, we traced some types of very small diameter speakers that would have been suitable -if not better- for sounding the notes 52-54 (Visaton type K40-50, 40 mm diameter), notes 55-57 (Visaton K36WP-50, 36 mm diameter), notes 58-66 (Visaton K28WP-50) and for notes 67 and up (Visaton K16-50, diameter 16 mm). They can be ordered from Reichelt and appear in their latest catalogue on page 1319. If ever, we would extend <Hybr>, we would certainly use these speaker types.
19.12.2014: The ordered female 30-pole connectors came in from Reichelt. Hybr demo for Laura Maes.
20.12.2014: Research and measurement of the required volume scaling for intonation.  Global sound production measured at 1 m distance from the instrument. With 4 pipes sounding, we easily get to 93 dBA.
21.12.2014: Extensive testing under GMT, with velo-scale lookups. Sound pressure level obtainable on a full cluster, with scaled velocities: 102 dBA. With proper settings for the envelope controllers, <Hybr>  is capable of producing very percussive sounds.
22.12.2014: Hybr demonstration for Lara Van Wynsberghe, as it should participate in the Drei Grosschen Oper production next year.
24.12 2014: Calculation of the measured overtones in fractional midi units, to comfort programming into the Axoloti patches. The appropriate columns were added to the pipe description and measurement table above.
28.12.2014: Hybr integration in the orchestration for the Dreigrosschenoper, by Lara Van Wynsberge.
03.01.2015: Hybr demonstrations for many friends. Very preliminary...
04.01.2015: Coding a Namuda study for Hybr...
05.01.2015: Further work on the Axoloti patches by Kristof Lauwers.
06.01.2015: The implementation of keypressure with a default value of 64 and a bipolar range for tuning seems very awkward. We are encountering some problems with the implementation of program change commands as well.
07-08.01.2015: New Axoloti patches uploaded in the ARM processors. Testing in PD and GMT. The controllers for the strength of the inharmonic overtones seem to lead to a crash condition on the ARM processors.
09.01.2015: Further fine tuning of the Axoloti patches by Kristof Lauwers. Conformation of the crash condition with the overtone controllers. Possible causes: audio buffer overflow, midi buffer overflow, too heavy load in the IRQ handlers, exceeding the number of controllers supported in Axoloti? We have to find this out and may have to contact Johannes Taelman for advice.
10.01.2015: Composition of the first Namuda Study for <Hybr> finished... It can be performed by Dominica Eyckmans and Emilie De Vlam.
13.01.2015: Hybr survived its premiere. Here is a link to a small picture album on the event.
19.01.2015: Further testing of crash conditions by Kristof Lauwers.
20.01.2015: Works on a treble extension for <Hybr> started: <HybrHi> may come...
23.01.2015: Continued work on the velocity scaling in the Axoloti patches by Kristof Lauwers.
07.02.2015: Stainless steel caps for the E14 light bulb sockets turned on the lathe. This element is similar as what we did for <HybrHi>.
08.02.2015: Namuda coding 'Seduction' for Hybr in combination with the other members of the robot orchestra. <Hybr> got a physical place in the orchestral setup.
09.02.2015: <Hybr> made its debut on television in the 'Iedereen Beroemd' series on VRT-TV1. The broadcast shows <Hybr> being programmed by Kristof Lauwers.
21.04.2015: Xavier Verhelst finished his composition for <Hybr> and <HybrHi>.
30.09.2015: Apparently controller 66 off does not reset the pitch-bends on the individual notes. This ought to be remedied in a next upgrade of the firmware for both <Hybr> and <HybrHi>.
19.10.2015: Check-up of <Hybr> and <HybrHi>. They survived the concert and the transportation to Dok19 very well.
25.10.2015: Fourier transform code tested with great success in Namuda Study #57, Tekstuur.
09.06.2016: Hybr demonstration and explanation for Berlin composers for the 'Wir sind die Roboter festival'.
18.08.2016: Considering to make a real bass version of <Hybr>, Could become <HybrLo>... if we only had the money.
19.09.2016: <HybrLo> is coming...
26.09.2016: <Hybr> and <HybrHi> transported to Berlin for participation in the 'Wir sind die Roboter' Festival.
03.10.2016: <Hybr> checked on return from Berlin. Note 66 (F#) is not working anymore. Repair required... Loudspeaker found to be fully burned out. Replacement was not an easy untertaking, as the speaker was glued on the PVC cone and in trying to remove it part of the conical foot broke. Thus we had to make an ring with internal 50 mm diameter to fit over the broken part, maintaining overall length. After all, quite an expensive repair as is took us 4 hours.
05.10.2016: Analysing the failure: If we recalculate the parameters again, we have under worst case conditions 5.65 V rms output from the power amps (with Uv=17V), leading to a power of only 638 mW to be dissipated in the 50 Ohm speaker. This is way below the rated 2 W power for these speakers. So, the only explanation for the burn out should be sought in exagerated cone excursion when sharp resonance occurs in the pipes.



Replacement of the broken mid-row female connector

Further development and debugging of patches on the Axoloti platform. Tasklist:

Thorough testing of the analog input functions.

Implementation of lead battery operation.

Construction of a suitable flightcase.

Testing and evaluation with the definitive firmware.

Maintenance information:

Circuit diagrams:

ARM-boards and audio amplifiers:


Power supply circuits:


The 6.8 mH self inductance is made by InterTechnik Gmbh, type FE130, its intended application being loudspeaker cross over filters. These coils have an open E core and are wound with 2.5 mm diameter copper wire.

Midihub board:

hub board

Wired board:

Firmware for the PIC microprocessor on this board.
Hex-dump of the assembled code

Audio input circuit with low pass filter:

audio input

Prototype board:

Last update: 2016-10-05

by Godfried-Willem Raes

Further reading on this topic (some in Dutch):

Audsley, George Ashdown 'The Art of Organ-Building', ed. Dover Inc, NY,1965, (first edition: 1905)  ISBN 0-486-21314-5
D'Appolito, J. 'Luidspreker-meettechniek', ed. Segment BV, Beek , Nederland, 2000, ISBN: 90 5381 116 8
Raes, Godfried-Willem , Expression control in musical automates
Raes, Godfried-Willem, HybrHi
Raes, Godfried-Willem, HybrLo
Taelman, Johannes 'AXO development platform' (Hasselt, 2014)

Technical data sheet, design calculations and maintenance instructions:

 Maintenance and disassembly instructions:

Before undertaking any repair on this robot we strongly advise to first read through our construction diary, as this will clarify fully how the machine was originally designed and assembled.

To upgrade the firmware in the ARM-processors, or to repair any boards on the front side the transparent protective cover should be removed.

For repairs, exchanging fuses and adjustments on the BaseTech amplifiers on the backside, similarly, the backplate should be removed.

If for any reason, the 12 V DC voltage would fail, the most likely cause is either a blown fuse or the position of the green switch on the black box containing the toroidal transformer.  To access any circuitry sandwiched between both bottom plates, remove the upper pipeholder first. Make sure to disconnect the two 30-pole connectors as well as the Bulgin 36-pole connector on the back.

Removal of the midihub board, holding the 18F2525 processor is possible without removing any bolts other than the M3 bolts wherewith the board is mounted on the side panel. First remove all Weidmueller connectors from the board and make sure to note the place where they should go back before loosening any connectors.

Robody picture with <Hybr>: