Composite amps

I've only done it a little, in very special cases, but if the intent is to apply a slow DC offset correction, and the main amp and the DC trim amp don't overlap in frequency response, it seems to work fine.

If you want to make a general -6 dB/octave amp as a composite, the risk is probably saturating one of the amps in large-signal/slewing cases, or at leasy doing goofy things. A composite that clips clean would be a challenge.

A sorta similar case is where a fast signal needs to be DC coupled across a big DC offset. A capacitor is the fast path and some slow opamp thing does the DC part before the AC path decays. The gains have to both be the same, about 1.00 usually, and the frequency responses need to be matched, to get clean step response and no ISI.

Tek called this "feed-beside", a brutally fast but ugly signal path, and slow stuff in parallel to make it clean.

Not so easy!

Plus a lot a lot of hand work to get them to match. I've benefited greatly from their labours, but I have no interest in doing that myself!

What I'm mostly talking about is using a nice quiet accurate amp such as an ADA4898 plus a faster but less accurate thing such as a THS3091 or LM6171. The output amp is run at some fixed gain like 10, and the input amp is run at high enough gain that the combination is stable at quiescent conditions.

Other composite amps such as the one you mention or the common case of using a chopamp to control the offset voltage of some fast-but-ugly amplifier have a different set of problems.

Cheers

Phil Hobbs

So that would take the frequency response curve of the precision amp and just shift it up 20 dB?

That should work, as long as the first amp keeps rolling off -6 dB/octave past its normal Ft. You could tweak a bit if not.

Make the first amp limit clean at +-2 volts or whatever, to not wind up too much, and let the second, fast amp clip however it likes.

Or, when the second amp hits some swing limit, stuff current into the input of the first one. Some common-base things maybe.

This would need some Spicing but shouldn't be too bad.

This Wireless World article from '74 might be of interest:

And this one from Burr-Brown:

How so?

Cheers

Phil Hobbs

Yeah, that's the idea. Problem is that it has horrible transient response.

Cheers

Phil Hobbs

Guessing this stool-design problem probably only has three legs, the labor will in be picking the appropriate length of the legs to meet the requirements without ending up with the other kind of stool.

Sorry hadn't read the follow-up yet

No worries, I'm mainly interested in actual experiences. Spice we can all do. ;)

Cheers

Phil Hobbs

Not to change the subject (I'd never do that) but I have made a compound amp just to shift the power dissipation away from the front-end diff pair, off to another chip, to avoid nanovolt thermal hooks. I had to keep the feedback network low impedance to minimize Johnson noise, which required a lot of feedback current.

Not to change the subject, but I have a question. There is another type of amplifier that splits the signal into two paths - a high frequency path for an RF amplifier with poor DC drift and small DC offset capability, and a low frequency path for an amplifier with good DC characteristics and wide offset capability. I thought this was a compound amplifier, and once read an article in the HP Journal that described it.

But I can't find the article, and google is no help. Do you know the name of this kind of amplifier?

I've used emitter followers for similar things, most recently in a super low noise laser driver. It uses MC33078 op amps and 330-ohm feedback resistors in a 12-V system. It's basically a PNP current source with a two-pole decoupling network between base and 10-V supply. (The +10 rail is made by a couple of diodes from +12, so that the op amp + emitter follower has enough output swing.)

Cheers

Phil Hobbs

This is usually called a compound amplifier. Tektronix called something similar to this "feed-beside."

There are two ways to do this:

1. Split the signal with RC or bias tee circuits, amplify the AC and DC parts with separate amps, and combine at the output.

1. Build a compound amp, with optimized AC and DC paths, but treat it as a black-box opamp, and close a feedback loop around it.

I don't know of they have specific names. As Phil noted at the start here, it's tricky to manage the overlap with precision.

There is an RF power amp configuration that has a high-power amp with some distortion, and a paralleled low-power amp with correcting distortion behavior. That probably has a name. I think cell towers use that.

Dunno. I cordially dislike architectures like that on account of their weird settling behaviour. It's super hard to get the nonlinearities to match, so it's generally much better to put bandaids on the RF amp to make it behave properly. I've only done that once or twice, so I don't have too many specifics to contribute.

Cheers

Phil Hobbs

Feedforward.

Cheers

Phil Hobbs

Opamp data sheets and appnotes are pretty good at avoiding any mention of transient thermal effects, or static effects of output stage dissipation on input offset. I know it is a serious consideration in opamp chip layout.

They can measure their DC specs and AC/noise specs and large-signal swing specs with different setups. Sneaky devils.

A compound amp can avoid some of these problems.

Not to be confused with predistortion, I guess.

I'd play with the compound amp thing, but I need to force myself to do less interesting grunt work. Like revising proposals and replacing faucets. Hard to decide which is less appealing.

For temperature controllers you normally want the front end amp on the cold plate, but you don't want its dissipation to vary. I often put a lead-lag integrator inside the loop but off the cold plate. That forces the input amp's bias to stay very very stable, so there's no change in dissipation.

Cheers

Phil Hobbs