The best sounding version (PCB "YB 3000A N4") has its LED label "TEMPO" misplaced at the power LED (above the power switch under the word "POWER"), while others place it correctly. It has only 4 AA batteries in one row (others have 6 or 2x 2 rows), 5 step volume switches and smooth plastic. This makes me conclude that it was the 1st generation and likely designed as direct successor of the similar looking Elite MC-2000, which SC-MC-2 CPU indeed flashes its power led to indicate tempo.

Another good sounding version has the PCB label "2030". The grey slide switches and their grooves are much wider and in different order (POWER, MICROPHONE VOLUME, MASTER VOLUME). It has 4 AA batteries in a changed compartment (2 rows of 2) and many case details differ from all others. But it contains many small PCBs interconnected with flimsy solder joints those may fail over time and need resoldering. Mine is detuned by 4 full notes.

I also found e.g. an awful sounding cheapened EK-001 version (PCB "3102 R"), which has 6 batteries and a 2.5mm phones-style AC-adapter jack. It is of rougher and brittle plastic (volume switch nib broken) with changed text size and icons (lowest volume setting labelled "OFF"), and its too big rear jack holes have a bezel of black plastic foil. The bottom is embossed "MADE IN CHINA" and has a plain black rectangular sticker covering a center hole. It distorted extremely when volume is not at maximum, because the amp IC was replaced with a lousy 4 transistor circuit with too low bias current (but still drains battery faster). It has a smaller thin sounding speaker and tuning trimmer was omitted, so mine played 1 full note detuned. Mine even had a way too long and stiff battery spring, so I could not get all 6 in (the plastic would surely break by using raw force, so I had to shorten it). Do not use a power supply on the 3102 R; mine had the jack polarity embossed wrongways, thus connecting like depicted for sure will destroy it. Also feeding >9V overheats its amplifier. 

In a similar variant "3102 R S1" the switch contacts fell off by too flimsy plastic.

Another quite similar looking version I found (PCB "3102R11", tempo led is farer left) has a proper amp IC, but the tempo led causes clicking and pressing buttons causes a thin buzz noise due to poor voltage stabilization.

circuit bending details

The mainboard hangs on too short and brittle (Angeltone style) grey ribbon cables those make the situation hard to examine. 

The hardware is mechanically a bit difficult to dismantle and I didn't want to ruin my very upgraded EK-001 (which is full of wire mess). So to make mainboard photos, I finally bought an unmodified 2nd specimen, took it apart and desoldered these cables to see the dark side of the moon. The CPU has indeed no label and also here adhesive film on it (doesn't look like falling apart); it is either bonded to the mainboard or a BGA package (which was uncommon in 1990th and thus unlikely). The behaviour hints that technically this is not a CPU (microcontroller) but a semi-analogue LSI made from gate logics. It has 3 pins unconnected (likely test pins) and apparently 4 individual percussion outs shorted for mixing (see HT3371 pinout). I haven't researched this further. After removing the panel PCB, the buttons are a bit difficult to put back into place, because they tend to tip over and the drumpad rubber contacts easily fall out. So before final reassembly screw only some top screws in and make a test run to see if all buttons work. Also the normal cables tear off easily, thus secure them with hotglue.

In my 2nd specimen I found 2 yellow paper strips (like sticker backing) under the panel buttons and at the lower panel PCB rim pieces of a foreign small rubber button with an "M" (may be "M+" from a calculator, which may hint that they were made in the same factory).

Although the front of the mainboard is hard to reach, fortunately all important traces are accessible from the back, and all components and many test pads are even labelled with tiny alphanumeric chars of trace material, which makes them easy to identify.

hardware identification chart

PCB type	notes
"YB 3000A N4" "3-3000AN4" (back) (on daughterboard)	This is the classic high quality version I described here - a fairly complicated construction that likely came out first. It has 4 AA size batteries and smooth plastic. The power led is additionally mislabelled "TEMPO" while the tempo led has no text. The panel PCB is brown and single sided. The CPU is an unusual square black COB package (66 accessible pins, with individual percussions) on the back of a green daughterboard that also contains the analogue hardware with amp IC "JRC 386D" (LM386) and (as the only version) a tuning trimmer. Box is blue with yellow name "PORTABLE ELECTRONIC KEYBOARD".
"2030" PCBs are also marked "GS" and have suffix letters (buttons=B,  pads=C, cpu=D,  keys=E, switches=W, amp=H)	Also this simplified version sounds nice and works well. It has 4 AA batteries in a changed compartment (2 rows of 2) and many case details differ from all others. So the grey slide switches and their grooves are much wider and in different order (POWER, MICROPHONE VOLUME, MASTER VOLUME). Also sound and rhythm button order differ, drumpads are thicker, left panel groove is wider and the black more greyish. The bottom has an embossed circular logo (bird above water?) with text "MADE IN CHINA". Apparently it was not released in Germany (I ordered mine from Britain). Volume (each only 3 steps) can not be set lower than medium. Pitch is detuned by 4 full notes (F on C). The cheapish looking hardware consists of many small interconnected PCBs; all are single sided brown with green solder side. The CPU is a green COB module "GS-01 94HB" (normal blob, 60 accessible pins) that also contains some capacitors and resistors and is poorly soldered only at its edges (54 outer pins) between a narrow keys PCB (facing keys side) and a panel PCB. Additionally there is a narrow drumpad PCB and a switch PCB. Above the speaker is a small Yongmei-style amplifier PCB (2 transistors, 3 e-caps) mounted upside down (showing solder side). In my specimen the 'violin' button was blocked by a torn off silicone peg stuck under its grey contact mat.
"3102 R"	This cheapened version distorts very badly. Box is blue with colourful music notes and red name on yellow "Electronic Keyboard". It has 6 AA batteries and a phones-style 2.5mm AC-adapter jack. It is of rougher and brittle plastic; rear jack holes have a bezel of black plastic foil. The inner switch sliders are red. It has a thin sounding too small speaker (gap at the rim) and no tuning trimmer. (Mine also had a way too long and stiff battery spring.) The tempo led glows dim without rhythm. The double sided main PCB is green and the CPU on its back is a grey COB blob (no module, 62 accessible pins including hihat out). To its left and right are 2 single sided brown extension PCBs with additional key and button contacts and volume control resistors. This model distorts extremely when volume is not at maximum, because the amp IC was replaced with a lousy 4 transistor circuit. Danger: Do not use a power supply; mine had polarity embossed wrongways(!), so it will burn out when connected like that. Also feeding >9V overheats the amplifier.
"3102 RS, CW RS"	I saw this PCB photo on the internet. It is a close variant of my "3102 R", but the COB blob is black and some resistor values changed. Likely this was a minor revision.
"3102 R S1, CW RS1"	This looks exactly like "3102 R", only the rubber contact strip (visible between keys) is pink and the inner switch sliders are grey. Even the inside is almost identical, but the main PCB has an amplifier chip "LM386-3, GOLF B5" instead of the lousy transistor mess. The COB blob is black. The sound behaves much like "3102R11", i.e. the tempo led causes weak clicking and held buttons produce key matrix buzz.
"3102R10"	The rough plastic case looks almost like "3102 R", but the AC-adapter jack is round on a flat rectangular surface with a fine rectangular groove (like a frame) around; the phones jack has the lower half framed. The 2.5mm AC-adapter jack fortunately has correct polarity. The LEDs are a bit bigger (3mm) and the tempo led above the "up" button is slightly shifted to the left. The fine grooves above the keys have 2 different widths (top 3 rows narrower). The keys feel less responsive. Volume is higher and can not be set lower than medium, and as a flaw the tempo led crosstalks clicks into the sound. There is a small embossed bottom text "MADE IN CHINA" and a plain black rectangular center sticker. It has 6 AA batteries and contains only one brown single sided PCB with the analogue parts. Its back is green with all the key and button contacts and some black carbon trace bridges. The COB blob is directly bonded to this and has a strange diamond shape (not rhombic, but like a Superman logo). It has neither test nor individual percussion pins (i.e. only 60 pins accessible). Amp IC "LM386-1, GOLF I5". There is no tuning trimmer. Traktor e-mailed me photos of this (amp-IC "LM386-1, GOLF L5"), but later I bought my own specimen.
"3102R11"	This one is almost identical with "3102R10" and seems to be the newest and cheapest version. The rough plastic case looks almost the same, but the rear jacks are round on each a completely flat rectangular surface (no bezel) and the grooves above keys have all the same width. Mine came with a black sticker hiding the embossed "MADE IN CHINA", which may hint that the plastic injection mould was later reused in another country. Under the missing center sticker is no hole. Unlike "3102R10" the resistors are miniature size and amp IC "National Semiconductor M59BZ, LM 386N-1".
(unknown)	Traktor e-mailed me a link of this. It has one row of 6 AA batteries. Mic and main volume sliders (green) have only 4 instead of 5 steps. Buttons are yellow, long buttons blue. Bottom label "MADE IN CHINA" with likely a round embossed logo. Also other colour combinations seen on eBay (switches yellow, pads blue, (buttons red, long buttons green or vice versa)), (switches blue, pads yellow, (buttons green, long buttons red or vice versa)), (switches red, pads green, buttons yellow, long buttons blue). Box has yellow name "Electronic Keyboard".
"EK-900"	I saw this in the Elenco DIY kit service manual, which names it "AK-900", but its B/W photo says "Stereophonic EK-900". It has different case details than EK-001 with narrower slanted drumpads. So I bought one, which has a soldered small square COB module CPU "CIL-9038" (different demo) and amp IC "LM386". The schematics claims CPU "9037" (all 68 pins accessible).

(For very differently shaped case variants see below on this page.)

All hardware versions more or less suffer of tuning instabilities, i.e. pitch changes or howls when voltage drops by battery drain or poor contact of the simple power switch. Thus I recommend to install at least an easily accessible tuning trimmer. Also adding a mechanically better or electronically supported power switch (passing the operating current through a mosfet transistor (like a standby circuit) and only a control voltage through the switch contact) may be a good idea to get rid of the howl.

- switch repair

- version "3102 R" flaws fix

The hardware version "3102 R" has many flaws. Particularly it distorts horribly because the LM386 amplifier was improperly replaced with 4 transistors. This amp section has cut traces and stray wire bridges on the back, and looks like if the piecework factory ran out of correct parts and soldered in what they had to finish the job in time. The distortion is so high that decay envelopes end too soon and polyphonic play crunches badly; only at maximum volume distortion is reduced.

But even after improvement, model "3102 R" sounds cheap and thinner than the original "YB 3000A N4", so I recommend to get the real thing if you can find it, unless you like this one as a timbre variation. (The correctly working transistor version is not bad and has its own sound.) The transistors are not marked on PCB, so naming is by me. The lower right one (T3= "S9014, C140", NPN) mixes the separate hihat out (CPU pin HH) into the rest of the percussion (those are mixed inside the CPU; you may install here a separate hihat filter if you want). T3 has its collector grounded, hihat input through 22k resistor at base pin and rhythm (others) through 2k at emitter.

The amplifier is a very basic 3 transistor class AB push-pull circuit with neither drift compensation nor overheat protection. The upper right transistor (T4= "9014, B321", NPN) is the preamp, the lower left transistor (T2= "S8050, C334", NPN) forms the positive and the upper left (T1= "S8550, C326", PNP) the negative halve of the power amp. Their interconnected output goes through a 220uF capacitor to the speaker and through a 220k resistor as feedback loop to the base pin of preamp T4. To that base pin goes the volume control (resistor chain) output of the sound (that is mixed through resistors in various locations) and through an 1k resistor the microphone volume control.

To reduce distortion, an additional resistor Rx can be installed from the base of T1 (left side of the 220k resistor) to either a +Vs or GND trace nearby to increase bias current (not recommended).
 

Rx [Ohm] idle current [mA] wired against distortion without 19.5   very high 330 33 GND somewhat 200 40 GND low 150 47 GND low 100 61 GND lowest 150 40 +Vs low

The value can be between about 330 and 100 Ohm; lower value makes less distortion but increases battery drain. For test I had installed 150 Ohm against +Vs (idle current 40mA), which was still a bit distorted but didn't waste too much battery. But only 100 Ohm (high idle current 61mA) sounded clean. (For comparison: the idle current of the nicely smooth sounding classic EK-001 (version "YB 3000A N4") is only 11.5mA.)

Another working solution (less idle current) would be to decrease the 220k feedback resistor instead. Wiring e.g. 100k parallel noticeably reduces distortion, but also amplification and hence volume. Thus making it even lower is no help. So I recommend to replace the 220k resistor with 2x 22k in series, and connect between them a sufficiently big (I used 470 nF) capacitor against +Vs (at the 1k resistor of the collector of T2) as a lowpass. This way only the DC operating point is shifted to a reasonable level, while the audio feedback gets damped by the capacitor and so still has full volume. My idle current increased from 18.2 to only 23.7mA, which does not drain batteries too badly.

But regard that by lack of temperature drift correction, this amplifier tends to overheat when fed with >9V, and may burn out because hot transistors increase current further. Already at 10V I had >70mA, thus if you don't want batteries, I recommend to use a regulated power supply or install a 9V voltage regulator. The CPU itself runs on only 4.34V (through a zener diode + resistor) and the power supply jack is labelled "DC4.5-9V", thus using 1.2V rechargeables should be no problem. That is to say, the main flaw of the amplifier design is that bias voltage is set by a 220 Ohm resistor between the base pins of both output transistors. According to generic class AB template schematics, there instead always should be 2 diodes in series, because their nonlinear curves stabilize the bias. With my transistor exemplars 220 Ohm pulled the voltage way too low which caused the distortion. Likely the 220 Ohm resistor did its job well enough in a laboratory test specimen, but not in real life with out of specs transistors. (I expect that such cheap tablehooters are even made from rejected low-grade components to cut cost.) Some template schematics therefore recommend an additional trimmer between both bases to tweak bias, but never a fixed resistor without diodes.

So the proper solution is to remove the 220 Ohm bias resistor (at the base of T1 (located to its left), with trace connected to base of T2) and replace it with [2 silicon diodes (e.g. 1N48) in series]; their rings must face to the base of T1 (upper left transistor). Do not install them wrongways, else both output transistors conduct and will rapidly overheat when powered on. (Feel them with fingers and immediately remove batteries if they get hot.) For better drift compensation, these diodes should theoretically touch T1 and T2 as heat sensors, but it is mechanically easier to make them touch only T1. With this upgrade, the amp sounds quite clean and idle current 30mA (35mA when warm) is still bearable. (I conclude this may have been the originally intended behaviour if the transistors would exactly match their specs.) The only flaw is that the amplifier plays a little quiet. To make it louder, replace the original 220K feedback resistor with 330K (do not install the earlier suggested 2x 22k, which runs too hot), which also reduces idle current to 25mA. With this upgrade, the amp will run even with unregulated power supply without hum. At 8V ("4.5V" setting) idle current is 25mA, and even at 10.4V ("6V" setting) the current rises only to about 45mA idle, which seems safe to me. If your current is too high, solder an additional trimmer (about 1k?) between both bases (parallel to your diode chain) to adjust bias lower. The sound is quite ok now, although my classic EK001 version "YB 3000A N4" still plays louder with more bass.

(The necessary resistor values and behaviour may vary depending on the transistor exemplars. An eBay vendor of an identical looking used specimen (same case details and box) told me that his one did not noticeably distort (but he had to fixed some contact problems and installed a 9V battery clip), so it may be that not all "3102 R" suffer of distortion.)

The "3102 R" lacks a tuning trimmer and my pitch sounded over 1 full note too high. To fix this, replace the 27k clock resistor with a trimmer (I used 50k in series with 1k for safety) to make it adjustable.

My exemplar had a crooked and way too long battery spring, which made it impossible to insert all 6 AA batteries. (So I initially used one AAA instead.) Insertion with raw force for shure would have shattered the brittle plastic compartment bottom - possibly by design as planned obsolescence (if not immediately, then certainly after some weeks by material fatigue), so I shortened the spring to 1/3 its length and bent it into shape with 2 pliers.

The rubber contact strip under the keys was misaligned and is partially held in place only by double sided adhesive film, that may make some keys fail when the glue dries. To realign it without taking all PCBs out, move it with a piece of cable tie between the keys.

By placing a ring shaped cardboard or sheet plastic bezel around the too small speaker to seal the gap, its efficiency and bass response may get slightly improved. But this speaker sounds crappy anyway, so you may want to replace it with a correctly sized one.

upper right ribbon cable pinout: (top to bottom)

1 = mixed sound out (to volume switch 3x 18k resistor chain)
2 = mixed sound in (from volume switch wiper to amp)
3 = mic sound in (from mic volume switch wiper to amp)
4 = mic sound out  (to mic volume switch 4x 24k resistor chain)
5 = tempo led + (750 Ohm to +Vs)
6 = /KEY1

The lower right ribbon cable has at the lowest pin GND; the rest are keys 2..6. The lower right resistor is hihat (CPU pin HH), the resistor above that is mixed percussion.

- version "3102 R S1" flaws fix

- versions "3102R10" & "R11" flaws fix

Traktor told me that version "3102R10" (I didn't own this one) has problems with beeps and pop noises those crosstalk from the tempo led. The problem seems to be that the other pole of this led pulls down the supply voltage line of the amp or analogue section of the CPU. Because the led is PWM modulated to show reduced brightness during others than the 1st beat of a bar, it can also cause beeps. Traktor told me, that disconnecting the positive pole of his tempo led and connecting it instead to the 750 Ohm resistor of the power led eliminates the noise. (This makes the led quite dim - for a better fix see below.) Also the keyboard matrix produces a thin buzz so long any buttons or drumpads (but not keys) are held down (which is particularly annoying if you want to install locking switches for the waveform arpeggiator synth mod).

I later found on eBay the almost identical version "3102R11", which has the same flaws. This is a very cheapened hardware design with only one large PCB. E.g. the speaker lacks screws and was simply welded by soldering iron into its plastic frame, so you can not take it out without breaking the plastic (use hotglue if necessary). Instead of a 4.7V zener diode (printed on PCB) only a resistor was installed (likely to cheapen production), but the tempo led click (quickly tested with parallel 3.9V zener diode) does not seem to be cause by this. Mine arrived defective; the slide switches failed, so the instrument only turned on by firmly holding power and volume sliders down with fingers. It turned out that the leaf switch contacts were crushed flat; after carefully bending them upward (do not break the flimsy welded plastic bolt) they work fairly ok. Possibly the contacts were crushed during poorly packaged shipping by weight of other parcels, thus do not lay heavy objects on these instruments. Like version "3102 R" pitch strongly fluctuates and howls down when the power switch makes poor contact or batteries fail. And also here the battery spring was a bit (less extreme) too long and thus to tight, so I took it out and squeezed it flat with pliers to make it shorter. The general timbre quality is cleaner (better?) than "3102 R", but below the classic "YB 3000A N4".

In my specimen the pitch was at least one note too low. The clock rate is controlled by the 3rd resistor from the left. Its value is 27k (printed on PCB). Replace it with a trimmer; I used 100k in series with 1k protective resistor, but 30k should be sufficient if you don't want to tune lower. The lower pin of the 27k is connected with the upper pin of the 220k (vibrato) to the right of it, thus the trimmer can be easily wired between both. You need to remove the 27k resistor (or cut the upper pin) only if you want to tune lower or have no high resistance trimmer. Mine sounds correct when total resistance is set to 23k (Holtek suggests 22k in schematics, thus the choice of 27k appears a bit strange).

A better fix for the clicking tempo led is to connect one of its pins through a capacitor to GND. The tempo led is wired to the CPU TEMLED output and its other end through a resistor from +Vs (positive supply voltage). The resistor is the middle one (labelled "750") at the right end of the PCB; solder at the left end of that resistor a capacitor against GND (e.g. the negative pin of the lower "220uF 16V" electrolytic cap next to it). A capacitor of at least 10nF (100 to 500nF recommended, I used 100nF) will sufficiently suppress the click. But you may also install smaller capacitors (e.g. switchable or through pots) to only modify the click as an additional percussion sound effect; the timbre here also depends on which end of the tempo led is wired to the capacitor.

I found no fix yet for the keyboard matrix buzz bug. Possibly serial resistors need to be installed into each keyboard matrix out line to reduce the current flow. Also installing a real 4.7V zener diode instead of the resistor may help to stabilize the CPU voltage better.
 

When not indicated otherwise, in the following I will refer to the classic version with PCB "YB 3000A N4" which sounds best. While the here discussed modifications likely all can be done with others too, you may need to figure out the wiring to find out where to connect parts. Their CPU pin order is likely the same, although unused pins may have been omitted on their COB.

power amp oscillation vs. blip percussion

The EK-001 blip percussion is output by the CPU through 4 connected pins (one pin in other versions) with coarse digital decay envelope that has strong zipper noise. Unlike early blip percussion, the decay curve has no visible linear sections. The 'cowbell' is plain squarewave between high B and C in octave 4. The 'base' is plain squarewave on C in octave 0 (1 octave below C1 with 'flute', hipass filtered). 'snare' and 'hihat' are made from shiftregister feedback noise. In rhythms the percussion generator can play accents. Powering the instrument on, fires all 4 drumpads simultaneously (at reduced volume), which may be a matrix- or software bug.

multipulse squarewave tone generator

According to my analogue scope, the multipulse patterns seem to be up to 16 bit long. Because the internal rotation is unknown, I sorted them by the longest sequence of  '1' as the start. They may be produced by a shift register feedback circuit or stored as 16 bit words in a small rom. To improve visibility, I omitted the redundant part of shorter periods.

The multipulse bit loops look like this:
 

preset sound	multipulse pattern	comment
music box	1100	plain squarewave (1 octave higher)
violin	1111011101101010	(same in Casio VL-1)
flute, banjo	11110000	plain squarewave
organ	1110101011001010	2 similar halves make this roar so bassy (same in Casio PT-50)
guitar	1111010010101110
horn	1111110000000000	squarewave 3:5 (1 octave lower)
piano	1111100000000000	squarewave 5:11 (1 octave lower) (same in Casio VL-1)

Although superficially influenced by early Casio mini keyboards (see VL-1), the analogue envelope control of EK-001 is very different and rather resembles analogue Yamaha  PortaSounds (e.g. PS-2).

The whole sound engine behaves very archaic, so decaying preset sounds ignore key press duration. The vibrato works by modulating the clock oscillator with a squarewave (about 6Hz, derived from clock speed) that temporary pulls clock rate lower. It seems to be operated by the same internal  LFO like the mandolin ring; with held notes it retriggers itself (changes phase to normal pitch) by pressing (trilling) any additional key. Pressing additional keys apparently restarts the LFO counter and so make vibrato and mandolin ring halt for a fraction of a second.

Attention: I have only incomplete draft schematics of my modification, thus this description may be partly inaccurate or even wrong where it is based on them.

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:
R.	= preset rhythm
O.	= preset sound ('orchestra')
P.	= drumpad

The volume slide switches crudely approximate potentiometers using a chain of each 4 resistors. You may replace them with a real pot. The upper right ribbon cable has this pinout (counted right to left):

1 = master volume common
2 = master volume max
3 = mic volume common
4 = mic volume max
5 = tempo LED (-)

important: Because pinouts differ among CPU variants, I also have marked the keyboard matrix pins by their "C" ("columns" = out pins) and "R" ("rows" = in pins) numbers found in Elenco AK-900 schematics. These are CPU pin names, not capacitors and resistors!

waveform arpeggiator synth

The synthesizer DIP switches basically do the same like holding multiple preset sound buttons. Important is to wire a diode in series with each switch to prevent unwanted matrix problems when pressing other buttons. Also the drum pads should be upgraded with each such a diode to prevent mess.

Theory of operation: It is unknown what exactly is going on inside the CPU, but I suspect it is something like this. Because the tiny hardware has extremely little RAM, all functions are set immediately during the keyboard matrix scan. So any held preset sound button immediately writes sound hardware registers for timbre and envelope duration at the moment the button is scanned. Thus when multiple buttons are held (which is simulated by the DIP switch) the sound registers get rewritten in the speed of the key matrix scanning loop, which makes the timbre rapidly toggle between the corresponding preset multipulse squarewave waveforms (or do other wicked stuff with bits of the shift register feedback generator) and so produce complex grainy digital timbres. The result sounds very similar like fast arpeggiator effects made with program loop synthesis on 8bit homecomputers. This is not to be confused with melodic note sequences played by a regular arpeggio; the tone switching happens here without restarting volume envelope and almost at sound frequency, and so rather makes a kind of very crude square frequency modulation. Also the envelope capacitor output is switched at that rate, which affects the note duration.

All preset sound switching other than between 'violin' and 'organ' or 'flute' and 'organ' (but not 'violin' and 'flute') and vice versa switch the envelope of held notes to key release, which may hint to different internal synthesis parameters.

pitch control / tuning
 

The clock speed is set by a 50 kOhm trimmer between 2 CPU pins. Is has no external capacitor and is quite touch sensitive since it directly conducts the HF. My pitch control replaces the trimmer with a 220 kOhm pot. The CPU's clock "output" line (originally the trimmer edge pin?) I have additionally connected through another 22k log pot and a capacitor to GND to achieve a 2 level pitch/ speed control (coarse and fine). From the CPU's clock "input" line (originally the trimmer wiper?) there is a resistor connected through a capacitor to +5V. From the vibrato output pin of the CPU a 2nd resistor connects to that capacitor to modulate the clock speed. I wired a 4.7 MOhm pot in parallel to this resistor to make the vibrato depth controllable. Important is to use separately shielded cables for all lines to the additional pots, because they bear HF signals and make the instrument howl badly when left unshielded.

main voice envelope
 

The envelopes of the 2 main voice polyphony channels are controlled by each a 4.7µF electrolytic capacitor against +5V. In each channel I replaced it with 22nF (as the minimum value) and wired [a 1M log pot in series to the 4.7µF cap] parallel to it to make the envelope fade slower. Also I added [a 1M log pot in series to a 2.2k resistor] parallel to the 22nF to permit to make the envelope shorter and sound harder. Since it is much more comfortable to control the same parameter for both voices with a single pot, I used 2 stereo potentiometers instead of 4 normal pots in a way that the same function of both polyphony channels is controlled by one stereo pot.

rhythm volume
 

The rhythm from the CPU is going through a 5.6k and a 10k resistor to the amplifier. I replaced the 5.6k resistor with a 4.7 kLog pot (clockwise input at the CPU, wiper output at the 10k resistor) to control the rhythm/ organ volume balance.

voltage regulator

With a power supply the instrument hums badly and may easily burn out when PSU voltage is set to high, thus I replaced the 1A diode from the 9V input jack with a 7805 voltage stabilizer which I equipped with each a 1A diode in the input and output line, and [2 diodes in series] in the GND reference line to rise the output voltage to about 6V, which corresponds to the battery pack voltage.
 

pinout HT3371, CIL-9038

The Stereophonic EK-900 is an EK001 variant that according to schematics has its COB CPU "9037" (a renamed HT3371) soldered to the large main PCB instead of being directly bonded to a daughterboard full of analogue stuff. The EK-900 was released in 1997 by the company Elenco Electronics Inc. as the electronics kit "MODEL AK-900". In the Assembly and Instruction Manual (electronic_keyboard_kit.pdf) the depicted keyboard on the photo is labelled "Stereophonic MODEL: EK-900". I later bought a specimen with CPU "CIL-9038" (same shape).
 

The manual with schematics gives a coarse overview with block diagram about the "9037" CPU. It mentions that the sound frequency is divided from a 523252 kHz master clock (controlled by resistor), but does not explain details about the waveform and envelope generators. Also the HT3371 datasheet does not explain inner working and is very similar. The keyboard matrix and "Greensleeves" demo correspond to EK-001, so it appears to be identical. The CPU supports a 5x5 panel button matrix, but the 37 keys (notes C3 to C6) and demo button have individual inputs. I measured that the 2 main voice polyphony channels have each a 16 step multipulse squarewave tone out and multiplexed analogue envelope control line (to be connected with an external capacitor). The 4 percussions have individual audio outs (shorted externally for mixing). The blip percussion is made from plain squarewave (base, cowbell) and shift register feedback noises (snare, hihat) with coarse digital decay envelope (about 16 steps = 4 bit DAC?) of logarithmic shape and accent function. All 4 percussions can sound simultaneously. It may be that the envelope DAC output (see block diagram) is routed through an internal multiplexer running at clock frequency, because on oscilloscope the waveforms look dotted.

According to that manual this is the "9037" pinout, which exactly matches the "Holtek HT3371" datasheet (die pads are numbered the same). The apparently identical pinout of the unlabelled EK-001 CPU is rotated, so pin 1 is in the middle of the CPU side at the 7th pin of the 10 pin ribbon cable (with the button matrix), counting anticlockwise from there. I.e. the 4 interconnected CPU pins are 62..65 (percussion outs) followed by 66..68. Then follows CPU pin 1. The CIL-9038 has the same pinout (thanks PHOBoS for documenting its schematics) and likely differs only in the demo "Evening of Moscow's Suburbs".

pin name purpose 1 /C3 key matrix out (buttons) 2 /C4 key matrix out (buttons) 3 /R4 key matrix in (buttons) 4 /R3 key matrix in (buttons) 5 /R2 key matrix in (buttons) 6 /R1 key matrix in (buttons) 7 /R0 key matrix in (buttons) 8 /KEY1 note key in (C3) 9 /KEY2 note key in 10 /KEY3 note key in 11 /KEY4 note key in 12 /KEY5 note key in 13 /KEY6 note key in 14 /KEY7 note key in 15 /KEY8 note key in 16 /KEY9 note key in 17 /KEY10 note key in 18 /KEY11 note key in 19 /KEY12 note key in 20 /KEY13 note key in 21 /KEY14 note key in 22 /KEY15 note key in 23 /KEY16 note key in 24 /KEY17 note key in 25 /KEY18 note key in 26 /KEY19 note key in 27 /KEY20 note key in 28 /KEY21 note key in 29 /KEY22 note key in 30 /KEY23 note key in 31 /KEY24 note key in 32 /KEY25 note key in 33 /KEY26 note key in 34 /KEY27 note key in

pin name purpose 35 /KEY28 note key in 36 /KEY29 note key in 37 /KEY30 note key in 38 /KEY31 note key in 39 /KEY32 note key in 40 /KEY33 note key in 41 /KEY34 note key in 42 /KEY35 note key in 43 /KEY36 note key in 44 /KEY37 note key in (C6) 45 VSS ground 0V 46 /DEMO demo button in 47 OUT1 channel 1 audio out 48 ENV1 channel 1 envelope analogue mux (capacitor 2.2uF to VDD) 49 OUT2 channel 2 audio out 50 ENV2 channel 2 envelope analogue mux (capacitor 2.2uF to VDD) 51 TEST1 test pin I/O (not used) 52 TEST2 test pin I/O (not used) 53 /TEMLED tempo led out 54 VDD supply voltage +4.7V 55 VIBRATO vibrato LFO out 56 RESET reset 57 TEST3 test pin I/O (not used) 58 OSC1 oscillator in (523252 kHz) 59 OSC2 oscillator out 60 TEST4 test pin in (not used) 61 TEST5 test pin in (not used) 62 CB percussion cowbell audio out 63 BD percussion base audio out 64 SD percussion snare audio out 65 HH percussion hihat audio out 66 /C0 key matrix out (buttons) 67 /C1 key matrix out (buttons) 68 /C2 key matrix out (buttons)

By block diagram and behaviour, these are not CPUs in the stricter sense, but LSI made from simple gate logics with very little software control. By the naming of its sister chip, the "9037" was possibly also released as "CIL-9037". But according to the HT3371 datasheet, the genuine creator appears to be Holtek. 

Apparently the test pins 51, 52 were omitted in EK-001 (empty gap between traces). CIL-9038 still has them.

The tempo led (pin 53) is PWM modulated to show reduced brightness during others than the 1st beat of a bar. (Seen on oscilloscope in version "3102 R".)

The percussion out pins 62..65 are shorted with each other, but certainly can be separated for individual analogue processing. In version "3102 R" the hihat pin 65 is separate, and externally connected through a transistor to the CPU pin that outputs the 3 other percussions.

note: If anybody owns an EK-001 with dead CPU (like those many corpses of failed circuit bending attempts), it would be interesting to donate it to a chip specialist for decapping (dissolving the COB blob in chemicals) to make microscope photos of the actual chip. I expect this to be a Pong-age hardware design made from only few thousand transistors. Decapping the dead chip may make it possible to figure out the exact algorithm of the blip percussion sound generator.
 

Successors of this hardware family are based on the Holtek CPU HT3421A, which was used in Yongmei MS-210B and a variant with changed rom in PlaySkool PS-635. Both are only monophonic, have a real key matrix, simple sequencer and 12 demos.