Discrete OTA filter prototyped

[Artis]

Hi to all!

Finally had free time at the end of that shitty past 2020 but at least I managed to finalise Discrete OTA filter on protoboard. Built using matched 2n3904 and 2n3906 transistors. Filter can be pushed in warm saturation with input volume, trough two pole low pass filter with band pass resonance and VCA at the end. It self oscillates, has an V/OCT input and with CV offset can be pushed down to few herts into LFO. When resonance is is pushed - it doesn't suffer from low frequency gain loss. Tracks well across 6 octaves. Has a Log FM and Linear VCA inputs with positive and negative volume pots. Audio output when pushed into saturation goes to +/-9Vpp but when running clean same 5Vpp and oscillates at +/-7Vpp. Levels can be brought back with output VCA. Works on +/-12V power supply (sorry MOTM users). I will soon try to record some demonstration video. Want to say thanks to fellow Wigglers who helped me before on this in this thread: Discrete OTA issues

Feel free to comment if you spot something weird or have an ideas how it could be improved with other bells and whistles.
Few comments and questions for experienced DIYers than me:

OTA cores are made with TL072 - not OPA 1662 as it is marked in drawing. Could be that there is a practical and audible reason to replace those TL072 op amps with lower bias / offset ones in the audio path? Does better op amps in audio chain makes such a difference?

OTA cores could be replaced with THAT 340. Haven't tested it yet but matched 2n3904/6 works fine.

Right now I wish that it could have better pitch stability. I made pitch controls with +/-9V voltage regulators and that improved a lot but some deviation is still going on. I suspect that the reason for pitch drifting are those Q12 Q13 Q14 transistors who needs to be closely matched and thermally coupled? Likely needs to have THAT 300 or similar? Maybe something else could improve pitch stability?

I tried but didn't found the way how to make High Frequency trim circuit for exponential converter. Then tracking could be improved over 6 octaves.

Should I need to make linear FM CV input and Exponential VCA CV inputs as well? Doesn't have an idea right now - will I miss them in the future.

[elektrouwe]

R4 has no function, shouldn't it be in series with Q1 base to balance R5 ?
Why the complicated differential driver Q1,Q2 ? next stages don't have diff. drive anyway...
temp. drift: your expo is temperature compensated, but the OTAs not (Gm)

[Synthiq]

It looks like the clipping of the signal by diodes D6 and D7 relies on the amplifier output current limit for correct function. I would consider adding a resistor between the output and diodes to reduce the maximum current.

I'm not sure of the function of the 50kohm RP2 driving the base of Q17. Since the base current is maybe 100 times smaller than the emitter current, the voltage drop across RP2 will be relatively low even at maximum resistance so the collector current will be almost constant when RP2 is changed from 0kohm to 50kohm.

[guest]

great work! i agree with whats been said above. is Q15,16 there for an eventual CV control of resonance? if you want lin/exp VCA control, you can replicate your expo converter from the VCF curcuit. lin input would be through R23. as for better exp tracking there are a few things you can do:

1. run your exp transistors at a lower current. replace R23 with 3.3M.
2. be sure that your exp transistors are run equally high as low (+/-100mV at the base).
3. use your 9V reference for R23.
4. increase R24,25 to increase the voltage swing at their juncture. this reduces opamp errors
5. bias the OTAs up or the exp down. right now they are at the same voltage, which means the transistors are saturated.

[MikeDB]

Shouldn't all those 100nF capacitors be on the +/- 9V side of the regulators, and across the power pins of each TL072 ?

I'd also add a small capacitor from output to -ve in on U5.3. It can easily go unstable in that configuration.

[Don T]

Better opamps in an audio path can indeed make a difference. I've never tried them in the middle of a filter or OSC, but I have when replacing opamps in between elements. Perhaps I'll breadboard this or a similar circuit with the stash of high-quality audio opamps I have lying around.

[Artis]

Hi to all!
Thanks for the great comments! Very much appreciated! Updated part here:

R4 has no function, shouldn't it be in series with Q1 base to balance R5 ?
Why the complicated differential driver Q1,Q2 ? next stages don't have diff. drive anyway...

R4 indeed it had no function. Series connection works better.
First diff amp works as input saturation and for limiting signal levels in to OTA cores. When resonance is brought in its still there when it reaches output VCA OTA core. Otherwise output VCA cuts all signals above +/- 9V and only resonance is left when signal crosses midpoint. At that point its almost only a square wave. Input amp brings much warmer saturation as well.

It looks like the clipping of the signal by diodes D6 and D7 relies on the amplifier output current limit for correct function. I would consider adding a resistor between the output and diodes to reduce the maximum current.

I'm not sure of the function of the 50kohm RP2 driving the base of Q17. Since the base current is maybe 100 times smaller than the emitter current, the voltage drop across RP2 will be relatively low even at maximum resistance so the collector current will be almost constant when RP2 is changed from 0kohm to 50kohm.

I tried to introduce resistance in to Zener Diodes and it brought up resonance levels and started to distort OTA cores again. Unfortunately there is no Zeners below 2.5V. So I tried this solution from AN701 application notes. It for my surprise made a possibility to balance resonance waveform symmetrical and to keep in reasonable levels. +/- 6V right now at the output. It eliminated any DC current from integrator output as well, but increased AC current twice as more. Now at +/-20 mA. If that current is over recommended TL072 value, maybe I should think about non inverting buffer?

RP2 50k - I wish that I could explain why but it just works very well.

great work! i agree with whats been said above. is Q15,16 there for an eventual CV control of resonance? if you want lin/exp VCA control, you can replicate your expo converter from the VCF curcuit. lin input would be through R23. as for better exp tracking there are a few things you can do:

1. run your exp transistors at a lower current. replace R23 with 3.3M.
2. be sure that your exp transistors are run equally high as low (+/-100mV at the base).
3. use your 9V reference for R23.
4. increase R24,25 to increase the voltage swing at their juncture. this reduces opamp errors
5. bias the OTAs up or the exp down. right now they are at the same voltage, which means the transistors are saturated.

Thanks! A lot of detailed suggestions.
Yes, I might do a CV control for resonance eventually as well.
Did a 3rd recommendation and that didn't change anything. Just needed to lower R23 value even more.
Tried 1st and 5nd recomendation about biasing exp transistors down but without luck. Tomorrow I might try again with OTAs biased up.
2nd thing - at the moment mV range at base of Q14 is -280/+190 mV. So I need to change R26 and R24 ratios to be within -/+ 100mV and maybe then I could increase R23 to Mohm range?

Shouldn't all those 100nF capacitors be on the +/- 9V side of the regulators, and across the power pins of each TL072 ?

Yes, it needs caps there! When investigating that I realised mistake with this design by choosing 79L09. Its no more available anywhere. Only negative regulators in stocks are 79L05. Looks like I need to redo all pitch CV control to +9V and -5V. Deeeeeeem! Or should I just forget it and stick to some configurable one like LM337LZ?

[MikeDB]

I tried to introduce resistance in to Zener Diodes and it brought up resonance levels and started to distort OTA cores again. Unfortunately there is no Zeners below 2.5V. So I tried this solution from AN701 application notes. It for my surprise made a possibility to balance resonance waveform symmetrical and to keep in reasonable levels. +/- 6V right now at the output. It eliminated any DC current from integrator output as well, but increased AC current twice as more. Now at +/-20 mA. If that current is over recommended TL072 value, maybe I should think about non inverting buffer?

Consider using two resistors from the supplies each feeding two series diodes to ground. This sets up good + and -1.4V references. Then a new resistor from the op-amp to R33, and Schottky diodes from the join of new resistor and R33 to the + and -1.4V references. That will save overloading (killing ?) the TL072 and give a more stable reference level.

Yes, it needs caps there! When investigating that I realised mistake with this design by choosing 79L09. Its no more available anywhere. Only negative regulators in stocks are 79L05. Looks like I need to redo all pitch CV control to +9V and -5V. Deeeeeeem! Or should I just forget it and stick to some configurable one like LM337LZ?

Yes I'd use a configurable one - keep supplies symmetric. Even Moog went that way eventually :-)

[guest]

Did a 3rd recommendation and that didn't change anything. Just needed to lower R23 value even more.
Tried 1st and 5nd recomendation about biasing exp transistors down but without luck. Tomorrow I might try again with OTAs biased up.
2nd thing - at the moment mV range at base of Q14 is -280/+190 mV. So I need to change R26 and R24 ratios to be within -/+ 100mV and maybe then I could increase R23 to Mohm range?

the main issue is the saturated transistors, and bumping up the OTAs is the easier solution for that. some matching pullups at the bases should do the trick. youll need to add the same offset to the bias generating opamp as well. another approach would be to use an opamp to create a bias voltage at 1V or so, that you then bias all your OTAs to. once that is set, youll probably find that you dont need as large of a sweep at the transistor bases anymore. if youre still having issues with HF, reducing the reference current will help a lot. but, increasing R23 will require that the integration capacitors get smaller to keep the frequency range the same.

ive built a discrete OTA very similar to this (but was very low cost/low performance - ran of single 5V rail), and it sounds amazing. absoutely love it. it saturates quite nicely on its own, so maybe you dont need that first diff amp?

[Artis]

An update.

Consider using two resistors from the supplies each feeding two series diodes to ground. This sets up good + and -1.4V references. Then a new resistor from the op-amp to R33, and Schottky diodes from the join of new resistor and R33 to the + and -1.4V references. That will save overloading (killing ?) the TL072 and give a more stable reference level.

I made you suggested circuit and result was the same as with my circuit. Op amp output still was at -/+ 20mA. I starting to think that my resonance works kind of backwards. When I remove feedback from U3.1 to R33 it goes in full power of selfosciliation and op amps current output drops almost completely. So the limiting itself increases current output. All that resonance circuit is kind of suppressing resonance when Q17 is at the ground and releasing it when base of Q17 is biased up. Tried buffering it with another op amp and it works well but limiting isn't working anymore after buffer.

the main issue is the saturated transistors, and bumping up the OTAs is the easier solution for that. some matching pullups at the bases should do the trick. youll need to add the same offset to the bias generating opamp as well. another approach would be to use an opamp to create a bias voltage at 1V or so, that you then bias all your OTAs to. once that is set, youll probably find that you dont need as large of a sweep at the transistor bases anymore. if youre still having issues with HF, reducing the reference current will help a lot. but, increasing R23 will require that the integration capacitors get smaller to keep the frequency range the same.

ive built a discrete OTA very similar to this (but was very low cost/low performance - ran of single 5V rail), and it sounds amazing. absoutely love it. it saturates quite nicely on its own, so maybe you dont need that first diff amp?

Tried biasing OTAs up with resistors and then with another voltage source and they started to react asymmetrically. I suspect i need Caps at the start and end of the filters to keep biasing up. But that unfortunately leaves filter without capacity to pass CV. I need to go back and to figure out how to do proper down biasing of exponential converter. Tried to bias expos down with negative voltage source but without luck. They works only at ground level. I feel that my understanding of proper transistor biasing is too short at the moment. :(
But hey - there is some good news! Managed to increase R23 to suggested 3.3M. tried to go even higher even to 8M but that didn't improve things much. Now I reduced necessary voltage swing at Q14 base approximately to -140 / +90 mV. I had to change resistors for tuning pots to fit new range.

I am exploring possibility of removing first diff pair but it still sounds much better with it at the front.

[guest]

if it sounds better with the first diff amp, then definitely keep it in!

here is my VCF schematic, in case it helps any:



again, its a zero-frills version, there is no compensation on the expo, etc, but i liked how it sounded. it was part of the minimodular project that i never ended up finishing (and now that the AE modular format exists, there really isnt a point to finishing). the OTAs are biased with an opamp, and the exp converter with a resistor divider. you are correct that the inputs/outputs of the VCF core need to get biased up and then back down to keep the external circuits at 0V. you can do this and keep DC, but you have to add offset resistors. so maybe it is easier to just offset the exp pair. the thing to watch out for there, is that your thermistor has to go between the 2 bases of the transistors; not between one of the bases and ground. using an opamp to create the -1V would be the best way of doing this, with the one base going straight to the opamp, and a resistor from the opamp to the CV mixer opamp to subtract the right amount of voltage.

[Artis]

Update!

the thing to watch out for there, is that your thermistor has to go between the 2 bases of the transistors; not between one of the bases and ground. using an opamp to create the -1V would be the best way of doing this, with the one base going straight to the opamp, and a resistor from the opamp to the CV mixer opamp to subtract the right amount of voltage.

Finally managed to bias expos down. Red your previous comments on another my thread and this one and found the missing bias point. Now it works like a charm at -1V. Filter feels and sounds bit angrier as well. Thanks! Need to work on CV controls again but that will be fast thing to fix.
Now I have one empty op amp left. Need to figure out what to do with it. Maybe buffer for resonance or separate power supply for VCA OTA core.

I'm not sure of the function of the 50kohm RP2 driving the base of Q17. Since the base current is maybe 100 times smaller than the emitter current, the voltage drop across RP2 will be relatively low even at maximum resistance so the collector current will be almost constant when RP2 is changed from 0kohm to 50kohm.

I reworked resonance control circuit as well with Q15 Q16 as NPN transistors and different bias on Q17. I think this now makes much more sense?

One mystery that remains for me - resonance limiter circuit. In updated schematic it needs to be there for stable resonance across all frequency spectrum. If removed resonance peaks at 15 kHz and reduces in gain quite fast as I go down. At the moment I don't understand what is going on there. Current goes +/- 25mA from op amp. Reading TL027 data sheet it says that Short-circuit current level is +/- 26 mA. That means that practically op amp output is short cir-cued? When I remove limiter and let filter oscillate fully - op amp current output drops to +/-0.5 mA. I am thinking about buffer from integrator to R33. If I understand op amps correctly - op am will stop current flowing because of its input impedance and I will be able to limit voltage at op amps feedback resistor with some parallel Zener diodes.

[guest]

connect Q12,13 base dirctly to -1V bias point, dont use resistors. this will make it more accurate. mixing some of the -1V into the CV mixer will allow the CV to move with -1V bias in case it also moves. also, you may want to consider getting rid of D4,5 as they may distort your CV signal.

[Synthiq]

I'm not sure of the function of the 50kohm RP2 driving the base of Q17. Since the base current is maybe 100 times smaller than the emitter current, the voltage drop across RP2 will be relatively low even at maximum resistance so the collector current will be almost constant when RP2 is changed from 0kohm to 50kohm.

I reworked resonance control circuit as well with Q15 Q16 as NPN transistors and different bias on Q17. I think this now makes much more sense?

In the original design, differential pair Q15/Q16 was driven by a current source (the collector of Q17). Now it is driven by th eemitter of Q17 so it is a low impedance source so Q15/Q16 will operate quite differently. If I wanted to adjust the current in Q15/Q16, I would use a PNP transistor for Q17 with R31 as the emitter resistor to VCC and drive the base from the RP2 wiper and connect the ends to VCC and ground. R31 is selected to give the maximum desired collector current when the wiper is at ground.

One mystery that remains for me - resonance limiter circuit. In updated schematic it needs to be there for stable resonance across all frequency spectrum. If removed resonance peaks at 15 kHz and reduces in gain quite fast as I go down. At the moment I don't understand what is going on there. Current goes +/- 25mA from op amp. Reading TL027 data sheet it says that Short-circuit current level is +/- 26 mA. That means that practically op amp output is short cir-cued? When I remove limiter and let filter oscillate fully - op amp current output drops to +/-0.5 mA. I am thinking about buffer from integrator to R33. If I understand op amps correctly - op am will stop current flowing because of its input impedance and I will be able to limit voltage at op amps feedback resistor with some parallel Zener diodes.

This looks like a standard state variable filter and what you call resonance control is actually the damping control that reduce the resonance when you apply feedback via Q15/Q16. There is a 360 degree phase shift from the base of Q4 to the output of U3.2 and this signal is then fed back to the base of Q4 via R18 so with the damping circuit disconnected it is no surprise if the filter oscillates. So if the objective is to limit the amplitude when oscillating, it would make more sense to try to increase the feedback in the damping circuit at large amplitudes instead of limit the feedback with zener diodes. I would try to connect the zener diodes across resistor R33 to see if it helps.