OB-Xa auto tune demystified

Recently a half repaired OB-Xa hit the workshop. The main trouble was complete failure to auto tune. After fixing some obvious faults (damaged vias, poor quality trimpots) the auto tune issue remained. Although the serial number indicated “old” auto-tune the program LED advanced during auto tune, so the firmware has been upgraded to a revision C already.

Until now I have always ignored that the service manual does say that the Revision C software has a new auto-tune circuit – so I had to learn it the hard way. I measured the gate time of the period meter circuit (which is around 3.82ms for a well tuned VCO, 4 periodes of a high C at 1046,5Hz), measured the clock frequency which is half the crystal clock or 2.4576 MHz. This gives a period count of around 9400.

Now I disassembled the firmware and saw that the target count is not in the 9400 range, but twice of that – 18,788 ! I verified my assumptions by feeding a 523Hz clock from a DDS generator which instantly let auto-tune pass.

During this, I also used the great “XACA2” patch EPROM from Ricard Wanderlöf (read here http://butoba.net/homepage/synthhacks.html). We had an email chat for some days trying to evaluate what is happening and came to the conclusion that all out assumptions must be right and that there must be a hardware change as well.

But first my comparison between the XAAD firmware (the last of Revision A) and the XACA disassembly:

While the Revision C compares the period count with 18,788, the old revisions indeed had a target value half as high!

So the only solution to this can be that the period meter is clocked from the full crystal frequency of 4.9152 MHz for Revision C and higher. Half an hour(!) I made this conclusion I found the ECO #119 which explained how to change the upper board for use with the Revision C firmware.

Some times we learn it the hard way. Lots of techs I asked during the process weren’t aware of the hardware modification – or just forgot about it, 40 years later…

Interesting enough: the original schematics of the later mainboard type for the Revision F and G firmware still show the period meter clocked by 2.4576MHz, while the target value in the firmware remains at 18,788. So this must be an error in the diagram, probably not the only one.

Here’s the ECO #119:

Eventide H3000/H3500 OLED update

The H3000 is one of the examples that shows us that common OLED displays claiming to be HD44780 compatible are not, in several aspects.

We have three faults here:

  • missing first line in most of the Soft Functions pages, caused by the fact that a Clear command (0x01) does not “home” the cursor, although the WS-0010 datasheet (the most common OLED controller) says it does
  • stuck text “in ]” above Soft Key 2 that results from the firmware copying screens by reading the DDRAM and writing it back when needed, using the auto-increment function of the controller which behaves differently between HD44780 and WS0010
  • wrong characters (Japanese symbols) due to the OLED having four character sets, with the needed not being default

While changing the character set is as easy as using 0x39 instead of 0x38 for the standard LCD init sequence, selecting the western european charset, the first fault already required a small subroutine that homes the cursor after clearing the screen.

I did not even try to shift around code in the EPROM to obtain the space needed for the additional home command because the disassembly gives hints that certain routines are called from without the main firmware. Instead, I modified the chip selection by fitting a larger PLD instead of the PAL16L8 used for address decoding. This way I was able to use the 256 bytes of unused space at 0x8200 for my own routines.

This was necessary especially as the second problem, the stuck text, required to replace the routines that read and write, respectively, the screen from/to the DDRAM by new code that initially reads the current internal address of the display, keeps track of it while the display is read or written, and inserts a command to re-position the cursor when the pointer would normally wrap around into the second line on a standard HD44780 display.

Finally, a working OLED in a H3000 (with H3500 firmware)

ProPSU – a PSU Upgrade for the Prophet 5

The SCI Prohet 5 Synthesizer has thermal issues in the power supply. SCI’s attempt to make do with a single CT secondary winding results in a voltage of up to 25VDC before the 7805 series regulator. The lack of a decent heatsink makes this even worse. Several techs have been rebuilding the stock PSU with either an additional tranformer or by replacing the transformer with a new one with an additional winding, supplying around 9 volts to an additional rectifier.

This modification also requires some work on the reset circuitry, otherwise the RAM contents might be damaged on shutdown. Other typical work covers replacing the thermal interface material, the big filter capacitors and the tantalum capacitors on the PSU board.

I decided to somewhat standardize the PSU maintenance of Prophet 5’s by designing a drop-in replacement PSU board I called the ProPSU. In order not have to temper with the name plate I made a aluminium heat spreader that is tightened to the back (use of some thermal compound is highly advised!) by re-using three of the original screws. The ProPSU board itself will be, after the cable harness is soldered to the ProPSU, bolted to the readily attached heat spreader by means of 3 or 5 M3x16 machine screws.

The ProPSU can be used for both Rev.2 and Rev.3 Prophet 5’s, with either the +12/-5 volts rails for the 2708 EPROMs provided or missing. Both 15V supplies can be adjusted by means of multi-turn presets. An on-board reset circuit switches the “+20V” rail to the mainboard’s reset generator, simulating a quickly discharging filter cap when the +5V supply drops below 4.75 volts. On turn on, the delay of the reset circuit ensures that all voltages are stable.


Ready to use ProPSU for late Rev.3 or modified Prophet 5 (new EPROMs, no +12/-5V required)


Mounting detail: the heat spreader bar is attached to the bolts of the old PSU’s regulators, after that the ProPSU
PCB is mounted to the heat spreader (isolation washers not shown here)

Berlin Clock

berlinuhrThe “Berlin Clock”

also called “set theory clock” or “Mengenlehre-Uhr” in German.
The time is displayed by the count of illuminated fields, weighted by their value:
– lower row: 0 to 4 fields indicating 0 to 4 minutes PLUS
– 2nd row: 0 to 11 fields indicating (0..11) * 5 minutes = 0 to 55 minutes PLUS
– 3rd row: 0 to 4 fields indicating 0 to 4 hours PLUS
– upper row: 0 to 4 fields indicating (0..4) * 5 hours = 0 to 20 hours

The round field on top is illuminated for every other second. Additional, small LEDs in the top field show whether an alarm is set and activated.

The time shown on the photo?
4*5h + 3*1h + 1*5m + 2*1m = 23:07



The Berlin Clock was designed by Dieter Binninger in 1975. Some information can be found on Wikipedia: https://en.wikipedia.org/wiki/Mengenlehreuhr
Small table top models were sold by souvenir shops in the 1980s, as well as a wall clock. All of those suffer from the amount of heat generated by the transformer(s) – 1 for the table version, 3  for the wall clock – which makes the plastic brittle. To save or restore a Berlin Clock, it is highly advised to remove the transformer and use a wall-wart type power supply which is easily possible with the circuit introduced below. It can even recover a clock with a defective processor, which was a 4 bit controller with mask ROM made by Texas Instruments.
While the table version already used LEDs, the wall clock had miniature incandescent lamps, therefore the need of three transformers. I have modified a wall clock by installing high efficiency TOPLEDs in place of the lamps, effectively eliminating two of the three transformers and reducing the heat.

The Circuit

My circuit replaces the custom micro controller, the LED driver ICs and the whole power supply circuit board. I used an Atmel ATmega8 CPU which drives the LEDs by discrete PNP transistors for the high side and a ULN2003 array for the low side. The LED circuit board can remain in the original state; the 14 way DIP connector fits directly in the socket on the replacement board. I recommend to replace the old, not very efficient LEDs by modern high-efficiency types though. As for the original, the time base is derived from the mains frequency. Therefore the current firmware requires the clock to be run from a 50 Hz mains with a high long-time precision. The circuit board with the processor and driver ICs carries the contacts of the push buttons as well, so I suggest to remove all parts but the diodes shown in the photo below from the board and re-use it. 6 wires need to be soldered to the locations of the processor’s place as indicated in the circuit diagram.















Emptied original PCB with the 6 wires for the switch matrix attached. 3 diodes and 1 jumper wire remain in place, the other components need to be removed.


Replacement circuit fitted to the clock case. The 14 way connector from the LED board fits to the socket on the PCB. The screw terminal on the bottom connects to an AC wall-wart type power supply delivering 12v AC, current consumption is well below 200mA thanks to the used switching regulator which provides the 5 volts for the micro controller and LEDs. The speaker on top is a dynamic loudspeaker, no piezo, and without integrated electronics, with a DC resistance of about 50 ohms.


Solder side view – some SMD components are in use, capacitors and small resistors are size 0805, the resistors above the MC34063 on the right and the 4 56ohms resistors for the LEDs are size 1206. The unpopulated locations are reserves for a PCF8583 I²C real time clock circuit which would make the Berlin Clock independent from line frequency and preserve the time during mains outage. This function is not yet implemented in the current firmware. If anyone feels free to add it, please ask for the C code (WinAVR/AVR-GCC).


The software is written in C (AVR-GCC). As written above, the current version uses the mains frequency of 50Hz as a timebase and makes no use of the RTC chip provided on the PCB layout. All functions of the clock have been programmed to resemble the original behaviour. The wrap-around at midnight works exactly as in the original: 23:59 -> 24:00 -> 00:01.


This project is free under the terms of the Creative Commons by-nc-sa licence (use and alter it for non-commercial purposes provided my name remains in the work).
The files needed to produce the PCB, build the circuit and program the micro controller can be downloaded here:

berlinuhr_schematic PDF file of the schematic diagram

berlinuhr_pcb PDF file of the PCB layout (size 100% when printed on A4)

berlinuhr_cam ZIP archive of CAM files (Gerber for solder side copper and solder mask plus drill file) for industrial PCB production

berlinuhr_firmware ZIP archive with hex file for programming the Atmel ATmega8 micro controller.
Make sure you set (write a “0”) the following fuse bits when programming: SUT0, CKSEL2, CKSEL3 and BODEN. All others should be unset (remain “1”).

APR Odyssey: CV/Gate for the White Face

Several approaches of adding CV/Gate to the Mk I Odyssey consist of not more than 3 or 4 1/8″ jacks with switch contact that interrupt and hook into the internal CV, gate and trigger lines. While the CV uses standard 1V/octave, the gate and trigger signals require a rather high amplitude of about 10 volts, with the additional impact that the gate level has a direct influence of the achievable output level of the AR envelope. Furthermore, both gate and trigger signals are needed for most modifications.

By adding a simple transistor circuit that resembles the levels generated by the Odyssey’s A board this ARP now works perfectly even with 5V gate and trigger levels and does not require the separate trigger signal anymore, although the trigger input jack is still provided for special purposes.

Odyssey Gate CV

The prototype for the improved CV / Gate modification in action


PPG 390 Drum Unit

One of the most rare PPG units found its way to my workshop recently: the 390 Drum Unit.
PPG 390 Drum Unit

Basically, this is a very rudimentary 8 bit sample player. It features 8 drum sounds in EPROMs on 4 voice cards. Two clock sources drive the address counters on the boards, which means that Sample Freq 1 defines the sampling rate of the odd channels, while Sample Freq 2 is in duty for the even numbered instruments.
[Block diagram to follow…]

Upon arrival one card was missing, instead I found a non working wire-wrapped vero board with a prototype for a single 4kByte instrument in two 2716s, and two 2732 EPROMs with instruments samples that won’t fit anywhere. I designed a PCB to replace the missing voice card, providing sockets for 2764 EPROMs so experimenting with other samples woule be easier as they can be programmes with cheap USB programmers.

PPG 390 Voice Card

Once completed the design I realized that I cloned the original cards so perfectly that I still don’t have a use for the 2732s – but my layout provides an easy solution, with two piggy-packed 4040 counters and some wires I modified my new voice card to work with the Noise and Hi-Hat EPROMs. Some more wires would allow to experiment with 2764 or even bigger EPROMs.

The clock frequencies are in the range of 13 to 30kHz, so the sound duration is between 70 and 160ms for the originally used 2716 EPROMs and 140 to 320ms for the 2732 sounds and would scale accordingly when using a larger EPROM.

PPG 390 Overview

Apart from this addition, the cards got guides (the blue things) to protect the connectors, new capacitors, additional capacitors in the output lines (originally it had a 3V DC offset), a new transformer, IEC mains jack, a mains fuse and a mains wiring according to current standards.

It is probably not an impressive drum unit sound-wise, but an interesting collectors item and I somewhat like the shaker … 😉

Posted in PPG

Advancing the Synthex

Four years ago I wrote “except the LFO section, because it’s strictly analogue”, but this did not really let me rest. So I started designing a new LFO board with some special features that can be controlled by the new processor board that is currently in development.

While the processor board design is near completion, a first prototype of the Advanced LFO has just entered the programming phase. Here’s a peek for those who are following the project for some time now. It’s alive!

Advanced LFO Board

RPL71 – Replacement for DAC71 series D/A converters

As stocks of working DAC71 series converters are depleted I designed a plug-in replacement for both the current (I suffix) and voltage (V suffix) types, including both CSB and COB coding. While the code is selected by a solder jumper, the voltage type replacement needs an additional SO8 OP Amp, like an OP07, fitted. The “TRIM” pin is not connected, instead of this a 20-turns potentiometer is added to the RPL71 board allowing fine adjustment of the on-board 10,00V reference. Where needed, a buffered output of the reference can be provided by adding another OP Amp and a few resistors.


Shown above is a DAC71-CSB-I replacement board (no I-to-V OP Amp on the upper right side) and without a reference outputs (missing parts on the lower corner) which is equipped with a 14- instead of a 16-bit DAC to save 10€ as this board is tailored for use in Sequential Circuits Prophet-5 synthesizers where the lower 2 bits are not used anyway. The actual use case is shown below.



Variophon: Electrical safety – naaah…

The Realton Variophon, a quite rare electronic wind instrument, is a frightening example how ignorant engineers can be when it comes to safety. While the whole circuit is a very interesting design and far from amateurish, the power supply is a completely no-go. To illustrate what I’m writing about, here’s a photo of the amplifier PCB that carries the mains fuse and switch connections:

  •  mains input is a standard IEC jack with the PE left unconnected, although exposed metal parts are present, but no proper isolation
  • the mains wiring are – 2 AWG28 litz wires from the IEC jack to the board above and 2 AWG 28 wires back to the mains transformer
  • those AWG28 wires are part of a 20 conductor flat cable, entering the PCB in the photo via pins 19 and 20 of the header
  • the track passing the mounting screw, that is actually in direct contact with the metal front plate, is isolated with a small nylon washer
  • the first and only fuse can be seen after the mains switch
  • next to the 4 header pins carrying 230VAC are secondary circuits which are in direct contact with the metal parts of the instrument

The only valid bet would be what happens first – electric shock or fire. This one obviously needs a serious rework.



PPG 1020: Digital keyboard?

The PPG 1020 is known as the stable successor of the 1002 synthesizer. This was achieved by use of a special oscillator architecture that combines digital octave division and a closed-loop approach including a VCO. The VCO covers the whole range(4 keyboard octaves + approx one more for the “digital modulation” feature), but being within a closed loop, it is stabilized to some degree. The details will be topic of a later blog post. For now we only need to know that we need a digital word that determines the octave and a control voltage to drive the VCO to one out of 12 semitones.

Even without the use of a processor (the 1020 does not have a computer of any kind) it would be straight forward: scan the 49 keys, somehow divide the number of the pressed key modulo 12, the result will be the (already digital) octave number, the remainder drives a DAC to generate the CV.

This “somehow” was actually done by using synchronously counting registers – a chain of 2 binary counters, one turning over at 12, the other incrementing whenever the first one turns over, and a chain of shift registers scanning the keys. The first “high” (pressed key) in the serial pulse train from the shift registers will latch the counter outputs in the next cycle, so we have a low-note priority here.

The counters mentioned above drive the octave control word and a digital-to-analog converter built from two 4051 multiplexers and a chain of accordingly weighted resistors. The reference voltage to that chain is generated on the main board and derived from the setting of the tune knob and the pitch bend slider.

For modulation purposes, an offset is digitally imposed. Since the modulation source from the main board is an analog control voltage, analog-to-digital conversion is necessary. A discrete A/D converter was built from a constant current source generating a linear voltage ramp on a capacitor. This slope is than compared with the modulation voltage in order to reset the above mentioned counters, thereby shifting the generated octave control word and the control voltage. With this approach it is possible to modulate the pitch across an octave border, whereas the analog part of the oscillator control would be limited to variation within an octave.

The comparator in this A/D converter is built from discrete transistors, yet integrated into a CA3086. The same circuit, but under processor control, can be found on the analog panel board (ANIN) of the Wave Computer 360!

Here’s the component side of the keyboard PCB:
PPG 1020 keyboard PCB components side

While the copper areas in the right top and middle right are of no electrical use, the long copper stripe on the lower edge is actively driven by a delayed clock signal. One might wonder what the reason for this could be – a view on the solder side and some circuit analysis reveals the secret:
PPG 1020 keyboard PCB solder side

Every key has a 2x1cm big copper area connected to the J-wire on the opposite side. Electrically, each key is capacitively coupled to a delayed clock signal – a signal that is logical “0” whenever the shift registers sample the key lines. This loose coupling is easily overriden by a J-wire touching the bus bar, which is connected to the positive supply voltage by a resistor of just 1k.

The last question about this is: was this only done to eliminate the need for 49 pull down resistors (the shift registers are CMOS devices and would not work properly with open inputs), or were other aspects like debouncing in mind?

A 3.3meg resistor added between one key line and ground suggests that the capacitive pull-down idea did not work 100%.