DIY Roland System 100 VCO
I’ve read loads on the various forums about fat, vintage-y sounding voltage controlled oscillators. What is the secret? ask the people with too much time on their hands. “Let’s measure it” say some. “There’s some undefinable mojo” say others.
Does the System 100 VCO sound vintage and fat? What does that mean anyway?
Enough of all that, here’s a schematic.
My System 100 sounds generally lovely. The oscillator sounds slightly different to my Analogue Systems RS-95 oscillators, apparently itself a “vintage-y” sounding oscillator. I’d prefer to quantify such woolly terms with measurements, but when I started building this, I lacked test equipment beyond my laptop/audio interface inputs and a really cheap DSO Nano oscilloscope. Which actually turned out to be a problem.
(It must be vintage – it’s brown and the opamps are in cans).
‘Cos it’s bigger than the things I’ve stripboarded so far – and I couldn’t get it to work on a breadboard – I was super-careful about laying it all out and checking for shorts before plugging it in.
The expo converter at the heart of the System 100 oscillator uses this…
…the Fairchild uA726 heated matched transistor pair. It was apparently always quite costly, and has been officially out of production for a long time, which means that it’s even more expensive and harder to get hold of.
There are other ways to skin the expo converter cat, but I’d hoped to stick with the uA726. Mainly because it’s easy, it’s what the System 100 had, and I didn’t think I was clever enough to work out how to adapt the circuit for a tempco or a heated CA3046 (like on the Curetronic version of the 100m VCO). Also I fretted about the possible power draw of a heating up a 3046 to working temperature.
All this hand-wringing lead me to searching eBay, where a seller in Hong Kong was offering a pair for fifteen quid with a no-questions-asked refund policy.
With my usual timing, I read all the comments in the analogue-heaven archives about dodgy uA726s after I’d clicked buy-it-now, so it seemed quite likely I’d ordered a pair of fakes.
Here’s one of them – they’re outwardly identical.
Before I plugged the uA726 in, I bunged in a matched pair of BC547 with no temperature compensation just for testing the circuit.
Here’s the arrangement I used for testing, with a 10 pin DIP socket in the uA726 position.
I used some electrical tape just to stop the legs of the left hand transistor from touching.
I flipped the switch, the power supply lights glowed happily, and the VCO made some sounds like it was trying to oscillate, but not quite managing it. When waggling the pitch knob it would cut out, and then distort and glitch out in a mad FM sort-of way. This is as good as I got it.
After a lot of head scratching, on a whim I swapped the comparator CA3130 at IC205 for a CA3140, which made it oscillate without glitching. With the BC547 matched pair in place, the pitch drifted in a comical way.
On installing the 726, it warmed up nicely and the oscillator tracked reasonably. I’ve still got some fiddling to do to see if I can get it tracking as well as the oscillator in the 101 keyboard.
Other people have had success with the Korean 726s, Ramcur on Flickr tested some of the Hong Kong UA726HC on his Minimoog clone board and verified them to be working.
Although the CA3140 worked, it didn’t seem quite right. On taking tuning measurements, I found that the pulse width increased with the pitch, from 50.3% at C1 to 51.3% at C6. I measured the pulse width on my System 100, and that stayed steady up the octaves.
I checked my photographs of the 101 keyboard, and there really are two CA3130s there – here they are in the bottom left corner of the shot – so there must be something else going on.
Boringly I spent a couple of months (no really…) to get the CA3130 working with no luck. I looked at the SH-5 for inspiration and tried a bunch of different things, including swapping the CA3130s, checking the value of R227, using original 1S2473 diodes, and checked all the voltages, but nothing worked.
After some moping and a lot of swearing, and I came across a thread over at Muffwiggler on the Roland Jupiter 8 VCO, which mentions the length of the reset pulse being set by a capacitor from the comparator output to input. Here it is marked in blue:
Then I started having a look at the VCOs from the Rolands from around the same time and found an indication of the length of the reset pulse in the SH-1 service manual, again with using a small picofarad capacitor on the integator feedback:
Slinging in a 10pF capacitor across pin 3 and 6 of IC205 made it work. I was so glad, it was pathetic.
About this time I finally got a decent oscilloscope (which would’ve really helped with tracking down the problem), so here’s a video of the reset pulse with extra cap, dancing about like a four year old at a wedding after too many sweets:
And here it is with a 10pF capacitor on the integrator feedback, just a picture because it stays still. Although the trace glows brighter as the pitch goes higher.
3µs is a bit long, so trying again with a 5pF capacitor we get:
which seems pretty close to the SH-1 spec. Going smaller with a 2.2pF capacitor it still resets happily with a pulse length of 1.3µs, and it’ll probably go lower than that.
I’m still interested to know why it didn’t work the first time round, or even more how my 101 keyboard works at all.
I note that the SH-5 and the System 100 have a similar arrangement of diode + resistor round the integrator, but most of the Roland VCO cores that come after use a low picofarad capacitor. Reading the 1S2473 datasheet, it seems like there would be some inherent capacitance there, maybe it’s not quite enough in this case to keep it cleanly resetting.
I made a more basic mistake when comparing the DIY VCO frequency with my System 100, wondering why it was wobbling around like that. Looking at it on the oscilloscope I could see the square wave flexing in and out in a suspiciously 50Hz-ish kind of way. I realised that stringing a long wire across the room to my (switched off) MC-4 for pitch CV was a bad idea – turning it on or removing the wire mostly fixed it, with the rest of the wobbliness coming from the unshielded pitch offset wire coming from a pot on a breadboard.
While researching pitch wobbliness, I stumbled across a post in the AH archives: Ritchie Burnett did some testing on analogue synths to check for oscillator pitch drifting, and found that the VCO in his SH-09 was modulated by interference from the nearby mains power cables. There are some power lines routed directly underneath my table which probably don’t help, so I’ve placed a grounded metal place underneath the circuit board in the hope that’ll cut down on the interference.
The voltage levels are close enough to the original, and the waveforms all look pretty similar, apart from the triangle which suffers from a much larger reset glitch, presumably from the capacitor I’ve added for lengthening the reset pulse.
Here are some octaves of C, all taken through the filter fully open, then through my version of the VCA, which unfortunately inverts at the moment. Not ideal but you get the idea. Warning – the tuning isn’t perfect… Saw:
And a triangle – the glitch makes it fuzzier than it should be:
And some random PWM fiddling:
The scaling measures as ok until it gets to the sixth octave where it goes sharp. Adjusting the scale has sort-of minimised it, but I’m wondering if increasing R224 (resistor in the integrator feedback loop) to the SH-5 standard of 3.3K or adding a trimmer here might help – I note there’s one on the SH-2 at this point which controls “linearity”. Starting off from C0 isn’t terribly realistic so it’s not as bad as it might seem.
The high notes don’t quite sound as pure as the original System 100, which I’m putting down to a 50Hz modulation being picked up from the mess of cables on my desk.
I had a quick go at just comparing a couple of loops from the original and from my clone. Here’s the original, sequenced from the MC-4 with resonance set about half-way:
and here’s the clone
And the original with a bit more resonance
and the clone
The pot positions are different on both, partly because some of the pot values are different on the clone (lower resistance pots = more control over the snappy area, especially on decay) but also some of the rotary pots on the clone don’t have any knobs on yet.
Despite having left both on for half-an-hour before I started this, the clone had drifted upwards slightly inbetween takes, which wasn’t too great.
This thing has been sitting on my desk for over a year now while I fret about sliders vs. rotary pots for a front panel for my 102 clone, and I’ve changed my mind about the uA726. Initially I thought it was a real one but it seems unlikely that they are exact clones given how expensive it would be to start making such a component again, so I wonder how it came to be. At the very least it’s some similar arrangement of transistors shoved into a can with a possibly dodgy heater circuit.
Now I’ve got a bit more confident with changing circuits, I’m leaning more towards remaking it with a heated CA3046 (looking at the Doepfer A110 for inspiration, see below for the relevant part of the schematic), or an LS318 matched NPN pair with a tempco.
But anyway, here’s the stripboard layout, and rather more helpfully here’s the DIYLC file in the rare case that you are masochistic enough to want to build this and you’ve somehow happened upon a bunch of cheap uA726.
If you find anything wrong with it or if you know why my oscillator needs the extra capacitor to stretch the reset pulse out I’d love to know.