Discrete Schmitt trigger in Thomas Henry VCO-1

[tomfoy]

I'm trying to make sense of the oscillator in Thomas Henry's VCO-1. Why is the output taken from the emitters, and not the collector of the second transistor (like in the picture below). Is there even a square wave to be gotten from the emitter? I more or less get that the oscillator is an Integrator/Schmitt trigger, with the OTA used as a variable resistor. If anyone would care to explain what is going on with the Schmitt trigger, I'd be double plus appreciative. There is a little explanation on the site down below, but I'm still lost.



I was also thinking it might be possible to use an OTA as the integrator (like in the LM13700 data sheet Tri/Square oscillator application), instead of using an OTA and an op amp integrator. Any thoughts on this, anyone?

http://www.birthofasynth.com/Thomas_Henry/Pages/VCO-1.html
http://www.birthofasynth.com/Thomas_Henry/pdf/VCO-1/vco1_schem1.pdf

[R.G.]

I'm trying to make sense of the oscillator in Thomas Henry's VCO-1. Why is the output taken from the emitters, and not the collector of the second transistor (like in the picture below). Is there even a square wave to be gotten from the emitter? I more or less get that the oscillator is an Integrator/Schmitt trigger, with the OTA used as a variable resistor. If anyone would care to explain what is going on with the Schmitt trigger, I'd be double plus appreciative. There is a little explanation on the site down below, but I'm still lost.

Only the guy who did the design can say why, exactly, but I can make some guesses.

I've used this kind of Schmitt before. It's a bit of a pain to get right, but can do some interesting things, as well as being faster than an opamp version. As to why the emitter - I think it's because the emitter has a well defined voltage in all situations, and it probably fits the voltage he wanted for the input of the OTA. OTAs generally have a 25-100mV input range. The power supply is +/-15V, so the emitter goes above and below ground, I think.

When Q1 is fed a base voltage/current sufficient to turn it on, it saturates to some low voltage, perhaps only 50-500mV. That means that the 30V differntial through the 20K/10K sets the emitters to -5V, about. Q2 is turned off because its base is pulled to that same -5V, plus the Vce of Q1, but is also pulled down by the voltage divider of the 18K and 100K to -15, so it's off. When the voltage on the input 27K drops enough that Q1 can no longer keep a low enough voltage on its collector to keep Q2 off, Q2 starts conducting, and that pulls the emitters UP to a voltage determined by the 5.1K and 10K to -15. With Q1 off, Q2 keeps itself on with the current through the 20K and 18K in series. It stays that way until the input voltage rises enough to turn Q1 back on.

I'd have to do some more analysis to see what the hysteresis, symmetry, etc. is - this isn't what I was using it for, so my memory of the side effects is foggy.

I was also thinking it might be possible to use an OTA as the integrator (like in the LM13700 data sheet Tri/Square oscillator application), instead of using an OTA and an op amp integrator. Any thoughts on this, anyone?

It's possible. Thomas Henry has been tinkering with synthesizers for a long time, and I suspect that there's a lot that's NOT said about why various pieces were chosen. The OTA current source is a good idea for very wide range, and the JFET input integrator is a good, linear current-to-voltage converter/integrator. Presumably T.H. wanted this mix to get linearity, symmetry, etc.

I'd have to do some modelling and tinkering to make better guesses.

[tomfoy]

Thanks for illuminating that, R.G. I'm beginning to understand; it's very slick. Thomas Henry coyly hints at how it works, that tease; between that, what you've said, and playing around with it some, I should be able to get something going.

[PRR]

> Is there even a square wave to be gotten from the emitter?

One OR the other transistor is on (neglecting the very short transition).

R.G. shows the one condition gives -5V. The other is a 5K:10K ratio between the +/-15V rails.... +5V. So it's a fine symmetrical square-wave centered on zero. What's the destination? Switching one input of a '3080, other input referenced to.... zero.

If he'd took a collector, the output probably would not be symmetric around zero, or not without more thinking.

There's no advantage taking the greater swing from collector, because you can semi-switch the '3080 input with 50mV, don't take much. He pads-down the +/-5V swing with 100K:10K, +/-0.45V swing across the '3080 input pair, complete switching. (If the switching is sloppy you get frequency errors at the top of the audio band. This may be happening: note R32.)

[tomfoy]

R.G. shows the one condition gives -5V. The other is a 5K:10K ratio between the +/-15V rails.... +5V. So it's a fine symmetrical square-wave centered on zero.

Thank you, this makes it very clear.

[tomfoy]

So, I have another question about this circuit, if you guys are still interested in teaching:

Why is the triangle output 10V peak to peak? It's not obvious to me, since the square wave into the OTA is scaled down so much. What happens in the OTA and integrator stage that makes the triangle work out to be +/- 5V like the square wave? I'm guessing it's not coincidence that they're the same; if the square were, say, +/- 1.25V, would the triangle also be?

[R.G.]

The triangle peaks on a Schmitt-integrator are always the same as the trip points unless you do something else to scale them. The Schmitt tells the integrator "OK, go the other way now" when it trips. In this case, the trip points are at +5 and -5, so the triangle peaks are too.

[tomfoy]

Of course, !!! The Schmitt trigger doesn't switch until the triangle reaches that point. Thanks!

[tomfoy]

I played with this circuit some in simulation and on the breadboard, and I feel this thread needs a little closure, so here are my results.

Here is a simulation of the circuit to look at (the OTA is replaced with a fixed resistor here). Also notice that I'm using a +/- 12V supply instead of +/- 15V:

http://www.falstad.com/circuit/#%24+1+5.0E-6+16.817414165184545+62+5.0+50%0Ar+208+64+208+128+0+20000.0%0Ar+208+128+352+128+0+18000.0%0Aw+352+128+352+176+0%0At+352+176+432+176+0+1+-16.649341081059596+-1.2901468135319005+100.0%0At+144+176+208+176+0+1+0.5416196350818119+0.5745914259672329+100.0%0Aw+208+128+208+160+0%0Aw+432+160+432+128+0%0Aw+208+64+432+64+0%0Ar+432+64+432+128+0+5100.0%0Aw+208+192+208+224+0%0Aw+208+224+432+224+0%0Aw+432+192+432+224+0%0Aw+352+176+352+240+0%0Ar+352+240+352+304+0+100000.0%0Ar+208+224+208+304+0+10000.0%0Aw+208+304+352+304+0%0AR+208+64+144+64+0+0+40.0+24.0+0.0+0.0+0.5%0Ag+208+304+208+352+0%0Ap+144+176+144+304+0%0Aw+144+304+208+304+0%0AO+880+256+928+256+0%0Ax+217+183+238+187+0+16+Q1%0Ax+441+183+462+187+0+16+Q2%0Ar+144+176+96+176+0+27000.0%0Aa+720+256+816+256+0+24.0+0.0+1000000.0%0Ac+720+192+816+192+0+1.0E-7+-0.9330385410534632%0Aw+720+192+720+240+0%0Aw+816+192+816+256+0%0AR+720+272+672+272+0+0+40.0+12.0+0.0+0.0+0.5%0Aw+816+256+880+256+0%0Aw+96+400+96+176+0%0Aa+528+240+624+240+0+24.0+0.0+1000000.0%0Ar+528+176+624+176+0+100000.0%0Ar+432+224+528+224+0+100000.0%0Ar+720+240+624+240+0+100000.0%0Aw+528+176+528+224+0%0Aw+624+176+624+240+0%0Aw+528+256+528+288+0%0Aw+528+288+640+288+0%0Aw+640+288+640+256+0%0Aw+640+256+720+256+0%0Aw+720+256+720+272+0%0Aw+96+400+880+400+0%0Aw+880+400+880+256+0%0Ar+816+256+816+336+0+1000.0%0Aw+816+336+720+336+0%0Aw+720+336+720+272+0%0Ao+18+64+0+38+40.0+9.765625E-5+0+-1+in%0Ao+20+64+0+290+40.0+9.765625E-5+1+-1+out%0Ao+18+64+0+226+40.0+25.6+2+20+out+vs+in%0A

I love this Falstad circuit simulator.

If you move the out scope pin to the emitters you will see that the bottom of the square waveform follows the triangle input some.

The value of the 27k resistor on Q1's base affects how much the emitters follow the input, and the symmetry, offset, and amplitude of the triangle. Change the value of the 27k resistor to be higher or lower and you'll see what I mean. My goal was a small offset, that could then be divided down at the sine-shaper input to an insignificant offset (I'm using an OTA for shaping). I found that there's a certain resistor ratio which divides the Q2 collector output to a square symmetrical about ground, which can then be buffered by an op amp. Here is my circuit:

http://www.falstad.com/circuit/#%24+1+5.0E-6+32.755850052045055+62+5.0+50%0Ar+176+64+176+128+0+20000.0%0Ar+176+128+320+128+0+18000.0%0Aw+320+128+320+176+0%0At+320+176+400+176+0+1+0.490053923941403+0.5875204719539227+100.0%0At+112+176+176+176+0+1+-12.265631469006713+-8.127463190934481+100.0%0Aw+176+128+176+160+0%0Aw+400+160+400+128+0%0Aw+176+64+400+64+0%0Ar+400+64+400+128+0+5100.0%0Aw+176+192+176+224+0%0Aw+176+224+400+224+0%0Aw+400+192+400+224+0%0Aw+320+176+320+240+0%0Ar+320+240+320+304+0+100000.0%0Ar+176+224+176+304+0+10000.0%0Aw+176+304+320+304+0%0AR+176+64+112+64+0+0+40.0+24.0+0.0+0.0+0.5%0Ag+176+304+176+352+0%0Ap+112+176+112+304+0%0Aw+112+304+176+304+0%0AO+848+224+896+224+0%0Ax+185+183+206+187+0+16+Q1%0Ax+409+183+430+187+0+16+Q2%0Ar+496+128+496+224+0+1000000.0%0Ar+496+224+496+304+0+1500000.0%0Aw+496+304+320+304+0%0Aw+496+128+400+128+0%0Aa+560+208+656+208+0+24.0+0.0+1000000.0%0Aw+496+224+560+224+0%0Aw+560+192+560+160+0%0Aw+560+160+656+160+0%0Aw+656+160+656+208+0%0Aa+736+224+848+224+0+24.0+0.0+1000000.0%0Aw+736+240+736+304+0%0Ac+736+160+848+160+0+1.0000000000000001E-7+4.210717808078772%0Aw+736+160+736+208+0%0Aw+848+160+848+224+0%0Ar+656+208+736+208+0+100000.0%0Aw+848+224+848+384+0%0Aw+848+384+80+384+0%0Aw+80+384+80+176+0%0Aw+80+176+112+176+0%0AR+736+304+688+304+0+0+40.0+12.0+0.0+0.0+0.5%0Ao+18+64+0+38+20.0+9.765625E-5+0+-1+in%0Ao+20+64+0+290+40.0+9.765625E-5+1+-1+out%0Ao+18+64+0+226+20.0+25.6+2+20+out+vs+in%0A

At this point, you might ask, why not just use one of those nice op-amp and comparator IC's that Texas Instruments makes? I didn't have any success with this, but I still think it's possible. The problem is that the comparator has to drive the 100k/10k divider at the OTA input, and the output of the comparator is open-collector, so every resistor you hang off of it messes with the thresholds.

Well, hopefully anyone searching for information in the future will find this helpful.

[R.G.]

At this point, you might ask, why not just use one of those nice op-amp and comparator IC's that Texas Instruments makes? I didn't have any success with this, but I still think it's possible. The problem is that the comparator has to drive the 100k/10k divider at the OTA input, and the output of the comparator is open-collector, so every resistor you hang off of it messes with the thresholds.

It's most common to use an opamp or a comparator without open-collector output to do Schmitt/integrator LFOs. Most dual opamps will work find in similar applications if you don't demand super-fast edges on the Schmitt output. An opamp or comparator with active up and down drive on the output works fine; this is good because the two-opamp Schmitt-integrator is used for an LFO in many pedals.

[tomfoy]

I don't think the speed of the edges would be a real problem, but since the thresholds depend on the maximum output capability of the particular op amp you're using, I think the circuit would be less precise than the modified VCO-1.

A push-pull comparator could be great. I see Texas Instruments has the TLV2702, an op-amp and push-pull comparator together (3702 for two comparators). Max supply voltage is 16V, though, so you'd have to use something besides the +/- 12V rails. Oh, the unity gain bandwidth of the amplifier is 5.5 kHz; not so great for VCO use.


[Sorry about the time traveling edit, I didn't know you were still online.]

[R.G.]

Yeah, I just don't like how the thresholds depend on the maximum output capability of the particular op amp you're using.

They don't. They only depend on the saturation voltage of the opamp high and low. This is pretty solidly fixed within an opamp type in most cases, and in any case is not a hugely variable item with load resistors in the 2K+ range. In practice, I've never had a problem and I've done this several times.

I have no experience with push-pull comparators. I think I remember looking for them online and not finding much.

They're not as common as the open-collector type.

I understand what you mean about the edges, since the op-amp has to recover from saturation. I don't think this would really be a problem, since I'm running from LFO type frequencies up to around 4 or 5 kHz, but the thought of punishing an op-amp like that when I don't necessarily have to makes me cringe.

Most opamp-based function generators use opamps in exactly this kind of duty.

I'm pretty happy with what the buffered discrete thing is doing.

In that case, if it ain't broke, don't fix it.

[tomfoy]

I've looked at some nice designs that use op-amps as comparators, the Music from Outer Space VCO, for example, so I know it can work well.

[PRR]

If you are doing high musical pitches, the rise/fall time of generic (101-era) opamps makes pitch go flat in the high octaves.

ARP used an elaborate version of your first circuit, all discrete, to get good pitch scaling.

There's also "sticktion". Many opamps take time to recover from gross overload.

An LFO or a low-performance note oscillator could sure use some of today's speedy opamps as the Schmitt.