[sdiy] Three-ball OTA

[Rutger Vlek]

Hi all,

After a wonderful Christmas holiday, with lots of time for various forms of spice (both the food and the simulation program), I feel the need to bump up one of my old threads. As I mentioned on the list before, I have been working on several variations of discrete OTAs (mostly in Spice up to now). Partly, just for fun and to learn about OTAs, but also hoping to find something that would fit my needs for future synth building plans: (1) not relying on LM13700's, (2) avoiding use of trimmers (so low voltage offset on output), (3) decent performance for DC and audio (at least comparable to LM13700), (4) physically small (SMD) and affordable.

I think I now have found something that could work (haven't prototyped in real world yet, but looks good in Spice). I'm planning to use it in a cascaded OTA-type VCF with four stages, but I have a question regarding the expo convertor. After a previous discussion here under the same topic, I decided to go for current mirrors at the Iabc input of the discrete OTA, so my expo convertor is of a PNP type. It's very similar to the expo described here http://www.uni-bonn.de/~uzs159/rs20.png, except for different resistor/cap values. At the moment, I'm feeding the OTAs of all four filter stages from one transistor pair in the expo, but I have also seen designs using matched arrays of 5 transistors. What is the advantage in that? Is it simply a way to overcome a low-ish current gain of each individual transistor? Or is there more to it? As I haven't been able to find a decent PNP array of 5 matched transistors, I decided to try it with a matched PNP pair (relatively high beta though, around 290). With four stages requiring 2mA each for maximum filter frequency, the expo is now delivering about 8mA at maximum CV levels. Does this make sense so far? Or does anyone foresee problems when taking this design from Spice to the real world?

I'd also like the final version of the expo to be temperature compensated. Does it make sense change the 1.8k resistor to ground (see here http://www.uni-bonn.de/~uzs159/rs20.png) into a 1k tempco and adjust other resistors accordingly? Or is it not that simple (I've been confused a little with the possible effect of the bottom 47k pot that provides a fixed voltage offset to the CV input)?

I hope you can think along, because I feel the need for a bit of a sanity check on what I'm doing ;).

Thanks!

Rutger

On 23 sep 2011, at 14:52, René Schmitz wrote:

> Am 23.09.2011 13:35, schrieb Rutger Vlek:
>> Dear all,
>> 
>> Inspired by this (http://www.uni-bonn.de/~uzs159/ota3080.html) comprehensive explanation of an OTA, I have been playing around with an OTA modelled in Spice. Towards the end of the article a "three-ball" OTA is proposed, effectively leaving out one current mirror (the one that deals with the control current Iabc). Instead the article proposes to "suck out" current from the NPN differential pair directly with a NPN transistor.
>> 
>> This idea of a three-ball OTA seemed to make sense at first and performed well in simulation when I "sucked" current out of the differential pair with a simulated current source. However, as soon as I replaced the simulated current source with a more reallistic NPN expo convertor, distortion characteristics went up (in the wrong direction) tremendously. I'm having a hard time understanding this behaviour. Anyone willing to explain it to me?
> 
> The trouble here seems to be that the emitters of the differential pair are sitting near GND, if the voltage compliance of the current sink (i.e. standard expo current sink) is also GND. This will restrict your common mode voltage range on the diff pair. (all is well if you move the voltages up, e.g. centered around half of the positive supply.)
> Compare this to the two-ball OTAs as used in the ssm2040 clones by JH and myself.
> 
> Cheers,
> René
> 
> _______________________________________________
> Synth-diy mailing list
> Synth-diy at dropmix.xs4all.nl
> http://dropmix.xs4all.nl/mailman/listinfo/synth-diy