Time for another System 700 module recreation: this time I’m having a go at the later 703e 4-pole/24db OTA-based low pass filter.

It went together pretty easily with the schematic being near-enough correct, although as with the 704 VCA I had to add a 15nF capacitor on the 4001 +V pin to ground to banish some noise.

I made some mistakes this time as well. Initially I put the +VLED power protection diode in the wrong way because I got the silkscreen wrong, due to a mismatch between Kicad and the footprint in Osmond. Also it seems I can’t count, I put 13 pins on the connector rather than 15 – although it doesn’t really matter, they weren’t all used anyway.

Once I’d chopped it out we get this kind of thing, recorded using an MC-4 controlling two System 100 oscillators into the 703e in, no VCA, straight out into the mixer and Audiofire 4.

I’m pleased with how it sounds, although it does sound fairly similar to other Roland OTA filters. The closest thing I’ve got to it is an SH-09, which uses BA662s and mylar capacitors against the CA3080s and polystyrene caps in the 703e.

Another thing that’s different to the other Rolands is that each OTA has a trim pot attached. Here’s a redrawing of one of the poles showing one of the presets (RV9) at top left:

The System 700 user manual from Roland Japan includes calibration instructions for the rest of the modules, but not the 703e. Maybe there is some extra documentation around somewhere, but I’ve not seen any.

My guess was that they were for nulling out voltage offsets on each OTA, so once I’d worked out to pull the plug on the filter input and attach the oscilloscope to the output of each OTA, I could see a decent offset from ground. Twiddling the appropriate trimmer dragged the output back to 0V, and once I’d done each in turn, the waveform on R33 (before the output cap) was centred around ground.

Post-calibration I gave it a bit of a test, although it didn’t really sound any different. Here’s my second jam with the System 100 oscillators and envelope, the MC-4 providing the CV and gate and the 703e clone. Soz about the length, I was having fun:

Slide pots from Futurlec have arrived now so the original format 700 clone dream lives on. I’ve got more boards coming shortly so there should be plenty to do over the Christmas break.

Here’s my redrawing of the schematic. In the LED driver circuit, I suspect there should be a diode between the +14.4v power input and Vdd on the 4001 (as on the 702) but I’ve not tested that yet.

Stare lovingly, desperately at the service manual, then wake up and overlay the parts on the tracks.

Draw out the circuit in Kicad from the publicly available schematics. It might be right this time, who knows.

Export the net list in Orcad2 format then convert with a homebrew Java program to OsmondPCB format.

Grep out the list of parts from the Orcad2 net list and choose footprints for each part (“DIP8”, “RCR7”, “TO-92” etc) and then import all that into OsmondPCB to lay it out, using the service manual track diagram as a template:

Use the track diagram to trace from to help eliminate any errors

Check there’s no expensive mistakes being made in the Gerber files using a handy online web tool.

…looks passable, anyway (shoulda annotated all the pins) – send it off to China, and then two or three weeks later –

and then nine months later, when I’ve finished doing all the not-so important things, like moving house

And it worked, first time. Scarcely believable.

The published schematic actually is correct, although only one of the two envelopes is drawn out. It uses an N13T1 programmable unijunction transistor, like the System 100 and SH-5, although I reckon the 2N6027 could be used instead.

The timing capacitor is specified as a 0.33uF tantalum with a 3.3uF cap hanging off a switch for longer times, and it does get pretty clicky and tight – here’s everything set to a minimum, rise time of about 600µs.

Shorting out R101 gets us even faster, but I don’t think there’s much point:

and it all completes in just over 2ms:

Here’s a dodgy demo with some mistakenly applied Quadraverb, all System 100 with the 705 clone envelope:

And it sounds ok, yunno, although I don’t think it’s quite as good as the 100m, but that’s only going by memories ‘cos my stripboard version is buried in the loft somewhere. I’ll dig it out soon and compare – now I’ve got an oscilloscope I’m interested to try and work out why the 100m envelope is so nice.

Either way, another one off the list. I’ve put a big, possibly too optimistic order in with the ever-reliable Futurlec for a load of 30mm slide pots in the hope of recreating the 700 panels, and then hopefully I can stop dragging crocodile clips across my desktop to test a board or two at a time.

And it sounds alright. After I’d tried previously to get this going on a breadboard without success (and after my System 100 VCO woes) it was pretty satisfying to get it going so quickly.

As a bare minimum I found I needed something attached to the range input, or else I got nothing, or possibly just a subsonic rumble that I couldn’t hear.

Being a saw core, here’s the derived triangle with the reset glitch very apparent.

It doesn’t sound too fuzzy despite the glitch. Here’s the reset pulse when zoomed in. I had to take a photo to measure it ‘cos it was jumping about a lot, and my oscilloscope doesn’t do the storage thing, looks like about 2µs.

I’m blaming the pitch wandering on the thicket of wires and power cables taped under my desk. Fiddling with C5 might improve the reset speed, although possibly at the expense of stability. The SH-7 oscillator core looks similar to the 702 and uses a 10pF capacitor for reset pulses of less than 1µs.

Here’s the comparator/reset generator from the System 700 schematic:

Here’s the sine…

With the possibly suspect ua726, and a 10K resistor at R22 (as detailed in the parts layout rather than the 15K in the schematic), I found I had to effectively drop R9 to about 45K to give the width pot enough of a window to get it to track across the octaves.

My current test set-up doesn’t really lend itself to the greatest accuracy – which is a nice way of saying it’s a shit-tip – but after a lot of fiddling I got it to play acceptably in tune across between 0 and 5v, drifting significantly sharp at 6v.

…which apparently in the world of VCOs isn’t that great. I’d put a audio demo in here but it’d make your teeth go on edge.

It seemed odd that it drifted sharp higher up, if anything I would’ve thought the oscillator would go a bit flat.

Some later Roland oscillators swap the 3.3K resistor at R28 for a 4.7K trim pot (referred to as “linearity”), which I reckon compensates for the high frequency error. This is from the SH-09:

I think the 2N5484 I’ve subbed for the rare-as-hen’s-teeth NF510 is better (faster?), which means the 3.3K is compensating too much for the error.

Fixing this means either increasing R28 or perhaps switching the 2N5485 for a 2N4392, which the Jupiter 8 service manual recommends as the sub for the NF510. I need to fiddle further.

I’m having giddy dreams of building nine oscillators for the full ridiculous System 700 experience, but using ua726s isn’t realistic.

To that end I’ve long had a ‘726 replacement heated CA3046 planned out on stripboard, I just need to get round to building it. Annoyingly 3046s are getting harder to get hold of in DIP for now, so it’ll be cheaper to go surface-mount for the final PCB. I’ve never done much in the way of surface mount before so this’ll be another new thing to learn.

Going to need to hibernate for a month or so now while I do some other stuff, but I will return to System 700 fiddling.