circuit bending details

When I bough this instrument (used, from eBay), it hummed very badly at low volume setting and distorted at higher volume. I found out that it hummed due to bad GND wiring; I fixed this by soldering a thick wire from GND at the AC adapter jack to the GND pin of the power amplifier IC. Later it turned out that the 220uF electrolytic cap of the power amp was bad. After replacement it worked well, but the instrument sometimes does not reset properly (does nothing after power-on), so I also replaced the electrolytic 1uF reset capacitor with a more reliable foil cap.

When I had set the AC adapter volume to 9 or 12 volt (like indicated on the jack), the amp worked well, but the sound envelopes started to fluctuate and got stuck at full volume as continuous tones (see CIL-51 for exact behaviour). The instrument freaked out at the 7.5V setting of my unregulated AC adapter because the +5V voltage of the CPU was badly stabilized by a 5.6V zener diode "BZX85C 5V6" and a transistor, and thus ascended to >5.1V when input voltage was set high enough to make the amp work properly. Possibly someone damaged it by wrong polarity, because instead of a protection diode there was only a wire bridge soldered in. To fix this, I soldered a 47 kOhm trimmer in series to the (slightly burnt looking) 500 Ohm resistor that is wired parallel to the zener diode. By adjusting the trimmer, I found out that the CPU even works well with only 3.5V, but the envelope control of mine freaks out as soon the voltage rises only slightly above 5V. For normal use set it to about +4.7V as a safe value. (Lower is quieter and drowns in the hum of the amp.) 

The stuck envelopes at 5.1V may be suitable as a sound effect (they warble louder and quieter at a threshold voltage), but I don't know if this is safe, because overvoltage can cause chip cancer over time (known e.g. from overclocked Pentium PCs). At least the CPU in my keyboard did not run hot from it, and the margin between 5.0 and 5.1V does not look terrifying.

separate volume controls

keyboard matrix

The input lines are active- low, i.e. react on GND. Any functions can be triggered by a non- locking switch in series to a diode from one "in" to one "out" pin.
 

legend:
"o"	= keyboard key
R.	= rhythm
P.	= drumpad
C.	= chord
O.	= orchestra (main voice sound)
orange background	= easteregg (unconnected feature)

The matrix outs 26..32 can additionally drive each a panel LED through a transistor; for this the pauses between the short scan pulses (when the inputs are active) are pulled lo to light the LED (else hi).

•  4 additional rhythms

The most interesting eastereggs are the 4 additional OBS preset rhythms at {24,44,45,46}->32 due to the unusual alternative accompaniment patterns selectable by the variation button. The additional rhythms were named after those on Superb Sound EK-902 (see below); the "samba" sounds more like beguine, the "tango" more like heavy metal.

• fill-in

A button at 43->33 plays a rhythm fill-in pattern.

• manual bass

A button at 39->30 permits to play in the chord section a monophonic manual organ bass (sawtooth) instead of accompaniment. Unlike MC-3 here the main voice even stays 3 note polyphonic.

• weaker vibrato

A button at 46->33 adds a slightly weaker vibrato than the normal one.

• transposer

The "transpose b" button at 43->30 pitches the keyboard down in semitone steps (works only with accompaniment?). The "transpose reset" button at 24->30 transposes back to normal (not needed because the transpose button can do this too).

• one key play

A button at 46->34 steps through the sequencer contents note by note (works even during demo), which mutes the keyboard until the sequence is finished.

• 2 demo buttons

Buttons at 44->34 and 45->34 start both demos individually.

pinout CIL-51

By lack of schematics, all pin names were chosen by me (partly inspired by Casio naming conventions and the DSG YM2163). So do not depend on them - they may change if I find schematics or an official datasheet.
 

pin name purpose 1 GND ground 0V 2 RH1 drum audio out 3 RH2 snare+cymbal audio out 4 OR4 bass audio out 5 OR3 chord audio out 6 OR2 melody2 audio out 7 OR1 melody1 audio out 8 /NMI (hi) 9 /IRQ (hi) 10 +Vs supply voltage +4.7V 11 /RESET reset 12 GND ground 0V 13 L1 led mux? 14 L2 tempo led out 15   (lo out) 16   (lo out) 17 L3 led power out 18 GND ground 0V 19 GND ground 0V 20 GND ground 0V 21 KI1 key matrix in 22 KI2 key matrix in 23 KI3 key matrix in 24 KI4 key matrix in

pin name purpose 25 KO1 key matrix out 26 KO2 key matrix out + /variation led out 27 KO3 key matrix out + /program led out 28 KO4 key matrix out + /transpose led out 29 KO5 key matrix out + /manualBass led out 30 KO6 key matrix out + /fingered led out 31 KO7 key matrix out + /singleFinger led out 32 KO8 key matrix out + /rec led out 33 KO9 key matrix out 34 KO10 key matrix out 35 NC (high resistance) 36   (lo out) 37 +Vs supply voltage +4.7V 38 GND   39 KI5 key matrix in 40 KI6 key matrix in 41 KI7 key matrix in 42 KI8 key matrix in 43 KI9 key matrix in 44 KI10 key matrix in 45 KI11 key matrix in 46 KI12 key matrix in 47 X1 clock in (4 MHz) 48 X2 clock out

This CPU does not tolerate supply voltages above +5V. While my specimen works already at +3.5V, anything only slightly above +5V makes the volume envelopes get stuck. Despite the volume envelope has digital zipper noise steps, when supply voltage +Vs is risen slightly above 5V it immediately stops decaying and gets stuck where it is; at 5.1V it even turns a little louder because the output level proportionally(?) increases with voltage. Once +Vs is set below 5V again, it continues decaying from its actual level. This can be repeated many times, and even during stuck envelope a new key press will make the volume level immediately rise to attack level of that preset sound and stay there until +Vs is set low enough to decay again. The tone generator stops the channel once the envelope reaches zero (no matter how long this takes). At the threshold voltage of 5.05V there is a strange flutter on the stuck envelope and static noise on the output, which clearly indicates that it is an analogue phenomenon. Percussion envelopes are not affected.

My theory is that the volume envelope of each main voice channel inside the CPU is implemented by a VCA driven by an internal envelope capacitor, which is charged or discharged by short current pulses at a certain rate (PWM, similar like the external capacitors in Hing Hon EK-001). While +Vs has risen above 5V, the internal pulldown transistor stops working (by a zener diode effect?) and so the capacitor voltage at the VCA stays where it is. But the envelope stays in decay mode (generating discharge pulses) until a comparator detects that the output has reached zero, which will stop the tone generator to free the channel. Because the waveform DAC is time multiplexed among polyphony channels (seen on oscilloscope), likely the 4 envelope capacitors are switched like a BBD while the rest of the envelope circuit only exists once. Another evidence for the existence of such internal capacitors may be that pressing the OBS preset sound buttons during sustaining notes does not stop the notes but only switches their timbre, which is the typical behaviour of capacitor envelopes (although the same phenomenon could be also achieved in a fully digital system when it e.g. keeps its actual numerical volume values in down counting memory registers when sounds are switched). Possibly also the DSG sound IC of the original MC-3 hardware contains switched envelope capacitors, because both instruments sound and behave very similar.

The matrix out pins 26..32 can additionally drive each a panel LED through a transistor; for this the pauses between the short scan pulses (when the inputs are active) are pulled lo to light the LED (else hi). Another transistor gets its input voltage from pin 17 and base at pin 13 through a resistor; this may be some kind of auto-power-off for the panel leds (the amp stays on when pulled low - tested through 100 Ohm), or just blank out the contact poll spikes to keep unlit LEDs completely dark. But in opposite to the original MC-3DX CPU I see no row and column LED outs for the orchestra and rhythm buttons. Despite 4 MHz clock rate there is no button click or other software squarewave tone out, and this CPU is definitely no speed wonder (audible by somewhat choppy accompaniment), thus it may be more RISC or built around a completely different core. I also didn't find the typical clock divider out (MC-3DX pin 12). Some of the many GND pins may be fake and do something else (not tested).

The unused pin 8 is likely an NMI pin without supporting software. It is hi, and pulling it low immediately locks up the CPU (but unlike stopping clock, the DSG part still decays a sounding note to its natural end). Pulling it low during power-on completely messes up the pinout, so there is HF dirt (clock divider, sometimes serial data?) output through the matrix inputs and many others pins, which hints that the IC is based on a generic programmable microcontroller and not a simple LSI. Unused pin 9 only causes a crash when pulled lo during power-on, thus I suspect it to be a normal IRQ pin that makes only trouble when the interrupt vector table has not been initialized yet. A later manual reset works normal. The unused pins 15, 16 and 36 are lo outputs (tested with 1k pullup resistor).

By shitshot the main voice can turn into short blips ("Popcorn" sound, like release-only phase of the  'organ' envelope?), which is no selectable envelope of the original DSG Yamaha YM-2163 (according to its datasheet). Pressing the "sustain" button returns to a normal envelope (also after pressing twice to turn sustain off again) , which proves that the button press overwrites the envelope parameters from those of the preset sound. In other crash situations the percussion pins can output continuous noise, which is a valid DSG mode (intended to feed external analogue percussion envelope circuits).