Peter Grenader on Waveshapers

Via Peter Grenader of Plan B via AH:

"Waveshapers don't really - or better put, shouldn't 'effect' a VCO per say, although they can play havoc if you're not careful. PWM circuits are notorious for this as many times they take the signal very close to the op amp's power rail, and depending which op amp you use it can cause real problems.

In any event, what a waveshaper does is take the core signal - meaning the waveform the oscillating engine creates on it's own, usually either Sawtooth or Triangle - and through a series of added circuits bends that into whichever waveform it needs at a given output. It's real trickery at times - reverse biasing of diodes, carefully timed disection to rearrange core signals into other shapes, etc. It's not an E.Q. Although one could filter a saw to it's first harmonic and have a sine, it wouldn't have the fidelity required. You have to do it other ways.

The waveshapers account for the bulk of a VCO's circuit. Along with the Expo converter, that's where the magic happens and it's really critical stuff. It's what makes a VCO sound as it does and designers guard their methods. For instance - while I've given some of my dealers - those who have in-house repair facilities - schematics of my VCO's core, I omit some of the waveshapers. If they've got a dead saw or sine for instance, I tell them what part needs to be replaced. If that doesn't do it, they send it back to me. I'm not the only one who's that paranoid.

There are three blocks to a VCO:

1) Expo Converter/Freq Control - This takes all the input controls - Freq Pot, Freq VC inputs, 1V/oct inputs, sums them to one voltage string, scales it as needed (somewhere around 18 mv/octave), and then converts that voltage to current through a circuit called an Expo Converter.

2) The Core. This is what does the oscillating. Basically it's a fixed-ratio envelope generator that gooses itself to start again, once per cycle. A secondary goose comes from the external SYNC input. The core can only manifest one waveform. There are a few things that determine what that waveform will be, namely the configuration of the circuit called an Integrator and where you place the capacitor in that integrator. Do it one way, you get a saw. Do it another way, you get a Triangle. The current from the expo converter gets injected into the core loop and that's what changes the oscillating frequency.

3) Waveshapers. One per waveform. They take the core signals and do the bending. One circuit for sine, one for triangle, one for saw, one for PWM, blah. Based on what waveform a given VCO's core produces, sometimes you've got to take the output from one waveshaper to make yet another waveform because it can't do it in one step. My core is triangle, I don't have to worry about this. They all take the triangle core signal to make their the output waveforms in single generation, thus all of the M15's waveforms remain in phase. All I do is fan the triangle to the string. But you do have to condition the core signals for the various applications. Some of the waveshapers need hearty signals to do their thing, some need comparatively low-amplitude signals, and you've got to scale your core signal for each.

So this doesn't sound all that bad, does it? and it isn't..not until you concern yourself with range, stability and tracking - this takes the bulk of the work in VCO design. It ain't easy and at times it's magic, but it's magic that must be repeatable under a plethora of operating environments. Back in the halcyon days, when musicians limited themselves to one system, designers could fine tune each module in their range to work with one another more efficiently. A wet dream compared to the Frakensynth 21st Century, all that goes out the window because people are using their Serge's with MOTM and Buchlas with Plan B's. Also know that many of these steps I've outlined are quite noisy. You have to deal with that. You have to assure your output waveforms retain their amplitude across the entire frequency range - that doesn't happen for free, either. There are mechanical considerations as well, specifically how the traces are thrown on the board. Some work as antennas which cause major problems if too long. Some can't be too close to others or interference will occur. Basically it's a huge undertaking and it's got to be right. You can get away with sonic defects in filters -people will propably prefer the results. With a VCO - no dice. Everything has to work correctly, in time and in spec and you have to find a way to do that affordably.

I'm skirting over much detail here, but I think this gives you an idea.

hope this helps," Pleas note this was sent to AH, so Peter will see questions sent there but not necessarily here although he has been known to drop by.

Update: More via Peter on AH:
"One must remember there are bunches of ways to create an oscillator core. Charge pump, 555 timers (yikes!), 4046 PLL's (I would recommend this over the 555), Feedback loops with logic gates, etc. And as a result different methods must be used to shape these core signals into the final waveforms.

Sometimes you have to use the output of one waveshaper to create another final waveform - the core signal will not allow you to manifest all of them, so you do it in steps.

OK - if one were to draw on paper four waveform types of the exact same freq, - sine, saw, tri and square, it would be done (I assume) with all four starting their cycle at the same time, going through their pattern so to speak, and ending at the same time. By doing this, you would be showing these four waveforms in phase with one another. But sometimes you can't do that. For instance, common square to saw shapers need the trailing edge of the squarewave to do it's thing, not the rising edge. Given that that square is 50% duty cycle, then the saw pattern begins 180 degrees out of phase with the core signal, and other waveforms which can produce their outputs from the core signal directly.

Next question - so where does that squarewave come from?
Along with the final output, most cores produce a timing pulse as well - the signal used to 're-goose' the core to begin it's cycle again. It's produced by a carefully tuned comparator circuit which pops a high output at the exactly the right time to open and close what's known as the flyback, a switch that opens the closes the current loop that charges the integrator. Open the loop, the integrator discharges. close it, it charges. The comparator is fed the core signal - so it's output states are governs by the very core it's enabling. You bias the comparator to trigger at the right voltasge level. Do it right, your core singal's fidelity is maintained. Do it wrong, your triangle is leaning over on it's side or pitifully low in amplitude.

Some waveshapers use the core output, some use the comparator output.

Here's a really odd analogy: Let's say the VCO core is a bathtub. The liquid is the core waveform, the faucet is the current loop and the plug, and the guy working it is the comparator. He plugs up the drain, the tub fills at a certain point, he pulls the plug, the water begins to drain. This goes on over and over, the water level going up and down. They are fed from and effect one another. But they are doing their thing out of phase. If the plug gets pulled when the tub is empty, it'll remain that way. If it's pulled when it's full, then the cycle is intact. If it's not pulled at the righ ttime, the water won't make it's target level. There's your visual.

So tying my original statement to this model - Some waveshapers need the empty tub, some need the full tub. Waveforms which are produced from these two concurrently are out of phase with one another - while at the same frequency, they don't begin and end at the same time."

"I need to add that the circuit I described - the current loop/comparator pair, is one type of circuit used to create a core waveform -there are other methods which are easier (on paper) to design than this. But, they suck. Linearity is often poor, the range is pitiful, amplitude levels are hard to maintain throughout that range, they're sensitive to temperature variations, they are unstable as hell. I see core VCO circuits that use a 555, or worse yet a logic gate pair to create their core signal, in which in order to address these variables the guy was forced to hang scads of parts all over the place. If the purpose of the exercise was to prove that a stable VCO can be made in this way, cool but real world - it's not going to work predictably. So instead of hacking it, it's best just to use a core which will give you the fidelity, linearity, range and stability required to do the job. If electronic components were finite - a 1k being EXACTLY 1K, a 10uf cap being EXACTLY that - then these types of alternate circuits would stand a chance. But things don't work that way real world."

"More ramblings on alternate methods -

I know a guy named Jerry Steckling - brilliant guy, who used to make mobile recording studio installations out of the panels used to for walk-in restaurant freezers which were f'ing incredible, who later went to Skywalker for a number of years and became a very big cheese up there, who now makes his own speakers, multichannel speaker systems and amplifiers. He had a pet project to produce a speaker which used a flame as the oscillating body instead of the driver. Wack idea, but on paper it's possible, and if anyone could do it, Jerry could. We had many jokes about this - how big is YOUR flame - that sort of thing, but while he knew it was possible, he'd never go that route obviously for anything more than an experiment followed by a good larf.

There are many ways to produce a VCO core, but....

you get the idea."