I bought a Roland System 100 101 keyboard years back, so I’ve been after a 102 expander since, with no luck. The last one I looked at on eBay went for over £1500 – it’s nice, but that seems a bit ridiculous. Although look at the panel, it’s bloody lovely, in the style of some austere-looking 70s science equipment.

So a while back I spent the evenings of a couple of weeks obsessively working on stripboarding the Roland System 100 filter.

This is a bit of the schematic, I’ve been staring at it a lot.

Most of the time has been spent wondering why it wasn’t working – but I happened across a post in the s-diy archives from Tony Allgood saying that all the System 100s he’d seen had the collector and base of the 2SC945Q transistors in the ladder connected together, much like the 303 – and this made it spring into life. (update: see comments at the bottom, I was being an idiot).

I had a look inside the keyboard just to check some values.

You can see the matched 945Q transistors at the bottom, marked with white paint.

Here’s a side view of the ladder, just to see how the transistors are connected.

If you squint a bit you can see the blob on the right hand legs where the middle leg is soldered. It’s even marked on the circuitboard, although the line goes the wrong way.

While I had the 101 open, I had a go at calibrating the square wave. The 50% trim pot is really touchy, trying to get that equally spaced waveform for the hollow square bonk was a trial, but it’s amazing how the ears can pick up on such tiny differences. It’s easier to get it just the wrong side of 50% and use the PWM slider to tune it in, although I think I’ve pretty much got it. Better still to replace the single turn 104 with a modern multi-turn preset, but I’d prefer to leave it alone while it works.

I used 2SC945GRs in for the 945Qs in the ladder. I didn’t match them but I did measure the betas and select the lower ones. The 945Q has a range of 135-270, and the 945GRs were mostly coming out between 260 and 290, so I selected a bunch between 210 and 250. I don’t know if it matters, but I was trying to cut down on any differences. The 945 were all pretty similar, it wasn’t hard to find matches (using Ian Fritz’s technique – PDF) for the matched pairs at the top of the ladder.

I put some old pulled 2SC2240 in place of the strictly unobtainable 2SC1000, although it’d be worth trying BC547 instead.

The dual ITS1276 transistor in the expo converter is also impossible to find, couldn’t even locate a datasheet for it, so I bunged in some matched 945GRs here and hoped for the best. Rob Keeble of Amsynth was planning on using a THAT transistor array (presumably THAT300) in his take on the SH-5 filter.

The input single op amp on the original is a TA7504M, so I stuffed a 741 in here. A TL071 works here too, it doesn’t seem to be critical. I used a CA1458 for the CV op amp.

Some of the things I got wrong were putting an electrolytic in the wrong way round, the TL071 powered from pin 7, not pin 8, as on the TL072, a track not cut on the output, and all of the 2SC transistors the wrong way round. I’d socketed most of them, so it wasn’t too bad to swap them, but… really.

Also I’d got the 2sk30a-y in the wrong way, which meant a drop in gain. Or at least, I thought I had – until I realised that it’s part of a sub-circuit to boost the gain going into the filter as the resonance increases, so the output level remains reasonably consistent.

Apart from the base-collector thing on the ladder transistors, the 101 schematic was wrong in a couple of places. The feedback capacitor C318 should be 47uF instead of 10uF – although I can’t think that would matter so much – and R371 in the VCF checkpoint output is marked as 10M, it should be 2.2M, or else you’ll get a distorted output. The 102 schematic looks right.

I’ve spent ages trying to get the resonance pot response right. It seems partly dependent on input level, needing 10vpp to sound authentic. The specified dual 100kA pot went nothing-nothing-nothing-slight-FEEDBACK, rather than the gentle increase in twang of the original.

Looking at the schematic for the LPF in the SH-5 which is really similar, the rotary for resonance is a 100kC reverse audio log pot. This makes more sense to me than a log pot, from memories of playing with guitar feedback and how twitchy it was.

Substituting a dual 100kC pot made the resonance much more controllable. I found it useful to unplug the feedback side of the pot (connections 90 and 91 on the right hand side) just to check that the gain increases when the resonance pot is turned up.

So does it sound like the original? I think so. The knob positions are a bit different because they’re tuned slightly differently, but just as a demo – here’s a square wave pattern recorded from the VCF check point of my System 100, no VCA

And here’s the same pattern, System 100 VCO check point to an input on my stripboard clone, recorded from the VCF check point, no VCA

Here’s a slow filter sweep at high res, the resonant peak picks out the harmonics in the square wave bass note.

Some more square wave acid-y silliness (now I’ve got that square wave tuned in I can’t help playing with it) – again no VCA, but with a TR-606 keeping time, the slow-ish envelope from the 101 keyboard struggling a bit to keep up.

And a bit more acid nonsense, some sawtooth wave cobblers, playing with ADSR and CV modulation of the filter, through a breadboarded version of the CA3080-based VCA.

Finally, the layout. I used DIY Layout Creator to lay the stripboard out, which is a bit buggy here and there, but works ok. Here’s the layout in PDF and DIY Layout Creator format.

I wouldn’t recommend it as a first build. In trying to save board space, I’ve made the layout rather tight. If you’re going to build it, I’d suggest zooming in really close and checking for track cuts underneath components. The numbering broadly follows the schematic, which is worth studying. Let me know if you find any problems with the layout.

I’ve been using 0.1″ headers for connecting wires from the stripboard to pots/sockets/power, which has worked out pretty well for testing and making alterations. I bought a bunch of rainbow coloured wires with the connectors already attached off eBay.

This seems obvious now, but next time, rather than jumping straight to laying out a stripboard, I would breadboard the circuit first. It’s what everyone says you should do, and I didn’t. Mostly because I couldn’t be arsed with fiddly breadboarding on my too-small slightly-melted board. It would have saved me time in the end.

There’s no panel design yet, mostly because I have all kinds of (possibly expensive) ideas about recreating the entire 102 expander on stripboards, and what kind of grim looking 70s style Roland front panel I could get built.

Remember actually making music, not just fiddling with a resistor and a voltmeter and swearing? No? Never mind. I liked the MS20 modulation generator, the skewing of the waveform is quite fun. Here’s the schematic.

Which turns into this (via Marjan Urekar’s additions)

which came out like this

Which works. After the obligatory tracking down of a few shorts, anyway. Here’s the ramp up on the square wave of an Analogue Systems RS-95 oscillator, it’s just as you’d expect. It’s really loud, turn it up so that it scares the shit out of the cat.

I need to do a bit more experimentation with the sizes of the capacitors. The smallest 1nF cap goes well into the audio range, but there’s something odd that happens to the waveform as it gets past a certain point. The 1uF cap allows the cycle to last for nearly two minutes. It’s just under 6v peak-to-peak, I’m wondering whether to boost it to 10v-pp. With the larger capacitors, the oscillation reset is revealed as being quite slow. On a slow speed setting with a ramp you get a fast slide down, like a zap, rather than a hard fall.

After scouring the s-diy archives for more on the MS20 LFO, hoping for a reset hack, I find that Tony Allgood had already done it on his Little LFO using a DG403 switch. Which renders my careful stripboarding a bit superfluous, but anyway. Taking the chance to do a bit of learning, but not that much, I’ve been experimenting with making it resettable. The accepted method for resetting this kind of circuit seems to be to short out the capacitor. Through some fiddling I learned that just shorting it wasn’t enough, it needed to be shorted to a negative voltage to make the waveform start from the bottom. Someone had previously recommended looking at the ARP Odyssey LFO for inspiration, this thing :

I’ve been using a 2N5460 for the JFET at Q2 and a BC550C at Q1, and pretty much just glued the stuff around the opamp on to the first opamp in the MS20 circuit. From experimenting a bit, I replaced the resistor at R9 with a pot so I could dial in the right voltage to reset to. So it works, or near enough. On a ramp down, it sounds ok – here I’m stabbing away at the (noisy) System 100 keyboard to reset the LFO.

But on a ramp up, it’s not quite so good.

Rather than getting falling instantly to -v, it does a sort of zap as though it is falling from +v max, rather than a smaller zap down from wherever it is at the time. Making C2 in the Odyssey circuit a bit smaller seemed to help a bit, I had it at 15nF for the experiments above. Making it too small just killed the LFO. I’m in two minds whether to keep fiddling and try and learn exactly what the circuit is doing to try and fix it, or just say fukkkkkkit, done, move on.