Balanced Source Impedances ? compensation network quality?

[barefoot]

When designing op amp circuits I always balance the source impedances seen by the positive and negative inputs. It helps reduce distortion induced by the op amp's nonlinear input capacitances. This is just one of the many, many great bits of wisdom I've learned over the years from reading every word I can find from the op amp guru Walt Jung.

My question is, how important is the quality of the source impedance compensation network in a feedback loop? Let say, for example, we have a high pass Sallen-Key filter with a compensation network like I've illustrated here.

C1a and C2a are high quality polypropylene caps. Does it make any real difference if I use mylar caps for C1b and C2b? My thinking is it doesn't matter because the compensation network is only looking into the huge impedance of the negative input terminal. And the main goal is keep the source impedances close to one another, but the tiny variations caused by differences in capacitor types would be far smaller than even the component tolerances.

Am I right on this? I know it's splitting hairs, but I figure if I go to all the trouble of building high performance circuits, I should do it right. I want to save circuit board space and a little money by switching to mylar in the compensation networks. But I don't want to do it unless it represents a fairly insignificant change. I've tried it and I don't hear any difference. I guess I'm looking for some theoretical justification as well.

Thanks!
Thomas

 

[NewYorkDave]

That's a bizarre circuit.

You have a high-pass in your feedback network, which is going to result in rising gain at low frequencies, which is defeating the whole purpose of the circuit. And I'm not even getting into what a very bad idea it is to put phase-shifting elements in a feedback network...

 

[barefoot]

The circuit is perfectly fine and there is no gain or phase shift from the feedback loop since the input impedance is many orders of magnitude higher than the network impedance. If fact, this is a very high performance distortion reducing topography. My question pertains to the component quality in the feedback network.

 

[NewYorkDave]

[Edit: Thought about this while sitting on the crapper and realized what I was missing before. Nothing to see here...]

 

[PRR]

> You have a high-pass in your feedback network, which is going to result in rising gain at low frequencies

As long as the amp input is infinite impedance (it is certainly "high" compared to these external componenets): it is not high-pass. It is unity gain. Maybe 0.999 to 0.995 including amplifier input leakage and parasitic capacitance, not enough to see on your dB meter.

The amplifier input capacitance is small but often very non-linear (being mostly back-biased diodes, also called variable-capacitance diodes when sorted and sold for tuning uses). You can see nonlinear gate capacitance effects in Nelson Pass's Zen amplifiers: the MOSFET's 500-1500pFd gate capacitance against the feedback network dominates distortion above a few KHz. Chip input parasitics are smaller, but we use more chips than 400W MOSTFET stages, so even a little stage distortion "could" add-up in a many-stage box.

As for whether the impedance balance caps have to be "good": I just don't know. I suspect they are somewhat less critical than the filter caps in this plan, but maybe should not be crap. Time to build it three ways (very-fine, common, and crap) and ask the ear what it hears.

 

[barefoot]

[quote author="NewYorkDave"]
After reading this statement several times, I still have no idea what you're saying. [/quote]

Ok, I really don't want to get into a debate about it, but try this:

Ignore the filter components and imagine the circuit is a non-inverting amplifier with Gain = 1+ Z1/Z2. Z1 is the feedback network impedance and Z2 is the op amp input impedance (typical FET input).

[quote author="NewYorkDave"] snip..... --but this whole thing of trying to compensate the feedback network seems silly to me. Are any of Jung's papers about this online?[/quote]

http://www.elecdesign.com/Articles/ArticleID/7206/7206.html

The most relevant quote from that article is:

• "Extrapolating JFET-input op amps to even more sensitive topologies leads us to Sallen-Key active filters, which, by definition, use noninverting amplifiers (often unity-gain, JFET-based followers). For absolutely lowest distortion here, a mirror-image network "ZS(−)" can be used in the feedback path, in lieu of a direct connection. ZS(−) is simply a dummy RC component set, to mimic the real ZS(+) filter elements, as seen looking out from the op amp's (+) input.2 Other JFET-input op-amp circuits also can optimize RS, as described below."

Btw, I just recived a direct reply from Walt Jung on this subject. He writes:

• "Actually, a Sallen-Key filter is amazingly sensitive to this
  distortion. This was discussed in some more deatil on the OP176 data sheet
  (a pert that ADI, in their infinite wisdom, obsoleted), but you should
  still be able to find a data sheet and see what I'm talking about.
  
  My gut feeling is that it won't matter if you use mylar caps in
  the dummy network, unless things are really super fussy and you have
  extremely high resolution gear. If you are in doubt, make up a test stage
  and swap the caps in and out, listening carefully. If you measure it, I'm
  sure you'll see a difference by adding this network (vs. a wire). Of
  course, in the actual forward signal path, the very best caps are appropriate.
  
  Good luck with it, and let me know how it works out.
  
  Walt Jung"

 

[NewYorkDave]

As you can probably guess, I failed to notice the first time around that there's no "Ri" on the inverting input and that the amp is unity-gain. The unusual topology threw me off.

No, I didn't smoke any crack today (yet), I'm just tired from getting up at 4 this morning... :sad:

 

[NewYorkDave]

See, this is why tubes are better... we only have to worry about Miller capacitance with those! :wink:

One thing that's wild is to play "make your own varactor" with any common diode or BJT, by varying the reverse bias. I used regular old 1N4007s rectifier diodes as varactors (aka varicap diodes) for tuning the tank circuits in my homemade theremin. The depletion region in the semiconductor acts like the dielectric in a capacitor. Varying the reverse bias is just like putting more or less space between the plates.

 

[PRR]

Barefoot-

Seems to me that nonlinear input capacitance effects should reduce as filter cap value increases. 470pFd against a 5pf-10pF gate capacitance will distort, but 47,000pF (0.05uF) will distort 100 times less. No?

Of course very large caps means small-value resistors and increased loading on the previous stage. And very large top-quality caps cost more, are big, and may have more parasitic inductance and mechanical effects.

As long as you have all those "excess" parts, why not look at inverting filters? Every follower has an inverting equivalent. They are unpopular because they are a pain to calculate, whereas the Sallen-Key topology is easier and Jung wrote a cook-book. But the inverter should not only eliminate gate capacitance variation, it should also eliminate Common-Mode distortion (which is non-negligible in many op-amps).

> Thought about this while sitting on the crapper

Done my best thinking there. Maybe we should move the PC to the potty-room.

 

[barefoot]

[quote author="PRR"]Seems to me that nonlinear input capacitance effects should reduce as filter cap value increases. 470pFd against a 5pf-10pF gate capacitance will distort, but 47,000pF (0.05uF) will distort 100 times less. No?

Of course very large caps means small-value resistors and increased loading on the previous stage. And very large top-quality caps cost more, are big, and may have more parasitic inductance and mechanical effects.[/quote]

Exactly, though I usually see the argument phrased the opposite way - reduce nonlinear input capacitance distortion (and noise) by using lower resistances, with the tradeoff of higher value, more expensive capacitors.

I use high quality PP and occasionally PS caps in my circuits, but I'm definitely not into the whole "boutique" thing.

[quote author="PRR"]As long as you have all those "excess" parts, why not look at inverting filters? Every follower has an inverting equivalent. They are unpopular because they are a pain to calculate, whereas the Sallen-Key topology is easier and Jung wrote a cook-book. But the inverter should not only eliminate gate capacitance variation, it should also eliminate Common-Mode distortion (which is non-negligible in many op-amps).
[/quote]

You mean Multiple-Feedback types? I guess I avoid them for two reasons:

1. My design software does not readily suit itself to these types. If I used this topography I could only optimize individual circuit elements, rather than treat the filter as a block with variable frequency, gain, and Q.

2. I may be wrong about this, but I think f, G, and Q are all interdependent with MFB filters, right? So, if you want to alter one parameter independently, you have to swap out all the components. This doesn't lend itself to the individual tuning (calibration) I do with my speakers.

Thomas

 

[Kev]

Thomas,
:?
what are you up to ?

Are you doing a driver compensation circuit in an active design ?

Back to the original question - is it components or circuit design that is the primary concern ?

Great to get a repy from Walt !! :thumb:

Last comment.
When using these left of centre topologies don't be so sure the optimisation software is going to give the right answers or even answers close enough to satisfy your high standards. You could end up waisting design time. Be sure to double check early results with some measured data.

yes,
I am making a huge assumtion as to what you are up to.

 

[barefoot]

Kev,

I'm designing crossovers... as usual. :grin:

I've already implemented this balanced impedance design. It performs beautifully. But now I'm adding some new features and my PCB real estate is getting tight. So, I'm mainly looking for ways to conserve space.

And LEAP gives excellent predictability for general circuit design. Of course, it doesn't model distortion or high frequency stability, but it never claimed to.

Thomas

 

[Kev]

cool
:thumb:

how old are you on LEAP ?
I didn't upgrade and am still DOS.

if you upgraded to WIN, I'd be interested to know your thoughts.
do you LMS ??

 

[barefoot]

I have the DOS version of LMS. From what I can see the windows version doesn?t offer a whole lot more than a simpler interface.

On the other hand LEAP 5 is WAY beyond the DOS version. Really there's no comparison. The list of new features and capabilities could go on for pages.

I use the circuit modeler to design preamps and all sorts of things, not just speakers. The only thing I find myself wishing it had is the ability to define relationships between components in the optimizer. For example, say you wanted to optimize the two capacitor values C1 and C2 in the circuit above. LEAP will change them to whatever values best fit the target response. But in some cases I might want to optimize with a constraint like C1=C2 or C1=2*C2. LEAP can't do that.

Otherwise I love it! :thumb:

 

[Kev]

Thanks Thomas.

something for me to think about.