High gain composite amps with CFAs in-loop

There’s a lot of prior art about the TPA6120A2 at low gains, in-loop, for driving headphones. There are examples of various AD parts performing the same task, stretching back almost 40 years. I’m interested in dialing in to CFAs in-loop for bandwidth extension and line driving capability, which they seem to do quite well. I’m using a gain of 10 as a benchmark because it really separates the utility parts from the performers.

You guys know this, but just for reference: 24dBu to 28dBu differential into 600R is the standard, even though nominal level is 4dBu and typical loads are 10K. All of the 32-36V DSL and PLC drivers invented this century, and the composite video drivers before that, have the current. But not all have the headroom, and fewer have the ability to do a gain of 10 whilst putting out pro audio line levels. Only a handful have an Iq in the 5mA-per-amplifier neighborhood, and that is what I’m shooting for (the LM4562 and 5532 being the highest draw duals in a typical system). Ironically, ye olde AD812 seems to fit the bill.

My application requires a bandwidth helper for a differential microphone preamp, but there are certainly line stages where a CFA in loop at a gain of 5 or more could be useful. My circuit has the expected OPA1612 at the front end, with a CFA in-loop. A gain of 1000 overall, but a gain of 10 at the CFA will keep the 1612 at 100 maximum, and that’s one responsible way to do a 60dB main gain stage in a mic pre. There are other ways - simply using a pair of AD797s wrapped around a pair of BUF634As, or cascading AD8599s or OPA1612s for example, but I’m thinking that this is the best path to a single control element and a 60 degree phase margin at low gains too. I did look at decompensated VFAs, but I can’t find a high speed dual that can do 20MHz at a gain of 10 AND put out anything near line level. 20MHz is the bandwidth of an OPA1612 at a gain of 2, and the resultant mild peaking would disappear after the first coupla positions on the dial.


The THS6222 is the only dual CFA in the TI stable with a price point to match the TPA6120A2. It’s over a decade younger, and unlike the 6120, it is very clearly marketed for gains of 10 and up (in addition to unity gain and low gain stability). It has a bias adjustment that lets you take the Iq down, and while it’s not well documented, it hardly affects bandwidth - still over 150MHz. The data sheet does mention something about distortion, but there’s no graph about it.

The THS3112 is of the same era as the TPA6120A2 (or its progenitor the THS6012 rather), and comes in a hair below 5mA on a dual 15V supply. There are gain-of-8 specs, with an optimal Rf being 200R. I’m guessing a gain of 10 would necessitate an Rf in the 150R-180R range. I’m also guessing that the TPA6120A2 would like to see an Rf in the 220R-360R range for a similar gain. The THS6222 likes an Rf of 845R to 1K24 in its “high” gain range, which I think is an advantage — you can treat it just as you’d treat a good non-inverting audio BJT in a low impedance system.


Does anyone have experience with relatively high gain / high output applications for the TPA6120A2 or similar? The THS3112 is not commonly referenced, so any insight there is useful. But more to the point, what do people think about the THS6222?
Last edited:


I mean, I don’t particularly want a CFB, but if you look at the available high speed VFB duals, you see that bandwidth becomes compromised at higher levels. See THS4032 for example. Audio amps are way better at that. I suppose I could plot the bandwidth shrinkage for a VFB in loop with an audio amp and see whether the bandwidth reduction is commensurate at each step. Having the in-loop amp reduce faster than the wrapping amp is no bueno.
Last edited:
Actually…the in-loop op amp, if it’s a VFA with gain, is generally speaking more likely to reduce bandwidth faster than the wrapper amp. Because it’s outputting at the output level of the stage or the unit, whereas the wrapper has the benefit of having next to no load and a (significantly) lower output level. I had forgotten that in the last post. Hence CFA.
Joined 2003
Paid Member
The THAT 1580 says 40 dB gain at 2.7MHz:-

You still havent told us why you need the bandwidth or the gain. Audio is considered 20-20kHz. As long as the amplifier is flat in this region and well behaved outside this bandwidth you should be good to go, and a VFA is really what you need to be using (noise performance). If you are trying to look at frequencies much higher than this, then you should state what they are. My bet is a VFA will still be the correct choice.
@atavacron, I do fully support the idea of placing a large-bandwidth CFA at high gain (~10x at and above master's GBW point) in the loop because the bandwidth of the slave is maximized (to 500Mhz++ with a THS3491). And large bandwidth means little additional phase lag at the GBW point of the master amp. Compared to the established Groner/Polak composite with a VFA slave (schematic below), the compensation scheme for total gains of 10x should become simpler and more robust.


OTOH the net benefit over a VFA solution is questionable because implementation still needs clipping management etc so overall complexity is not that much reduced. But the current and bandwidth advantage alone of modern CFA's is temping.
  • Like
Reactions: 1 user
You still havent told us why you need the bandwidth or the gain.
Microphone preamplifier. Look at the bandwidth of a 40MHz VFA at Av = 1000. Now put a second 40MHz VFA at Av = 10 in the loop. See?

This is all fine and dandy if you are using a two pole control. But for one pole, at low gains, there is no way to open up the bandwidth of the in-loop amp, and you get peaking and eventually oscillation. If that VFA at Av = 10 is a CFA instead, with a flat bandwidth of at least 100MHz at that gain, you can leave it in the loop at that gain, even whilst the wrapping amp is at unity.

Mic preamps usually have at least 6dB gain, so a CFA in a 1612’s loop with the 1612 set to 6dB gain / ~20MHz only really needs a flat bandwidth of ~50MHz. The magic number seems to be 2.5x the maximum bandwidth / minimum gain of the wrapping amp in question.
@KSTR just on the Groner/Polak tip, amirite that there’s only sufficient phase margin at gains surrounding unity? Jan Didden made light of that in the main thread on this a few years back.

THS3491 sure is the flagship, with the Iq to match. THS3001 x 2 would do it. I’m unsure if TPA6120 will be happy at Av=10 but that’s the low hanging fruit. THS6022 looks like a good candidate, less Iq than the 6120. THS6222 probably has too much noise in the audio band to be fully cancelable.
Last edited:
@atavacron, When output voltage or current clipping is entered, the loop opens and the master opamp will try to slam its output into either rail. Besides the risk of killing the slave's input with overcurrent, recovery is excessively long and nasty because everything is fully saturated, with the risk of spurious oscillations. That must be avoided by taming the master opamp. For a Groner/Polak style composite in inverting config, a local feedback via diodes is closed when entering clipping and there is some limiting resistance in the slave's input.
I've been tinkering with composite amps using the (thru hole packaged) LM7171 as the downstream device. Maybe it might have enough grunt to succeed in this application?

Probably? The -3dB point at Av=2 is unusually high compared with UGBW. I have come across the LM6172 previously, but not this. I wonder if they made a dual.

It’s a good sign that some graphs were done at 20vpp. There’s one showing >5MHz flat at ~27Vpp at 1% THD, which beats some (or all?) of the more modern high speed VFAs. And there are some high gain indicators too. Have you done anything with it at Av=10 or so? Have you been able to mitigate the voltage noise in-loop? It’s high, like my THS6222 example.

This application, by the way, doesn’t necessarily require more than +16dBu, because a transformerless mic pre almost always has a +6dB output stage, and is generally speaking dual-driven. So even if you’re using a mythical +28dBu differential output standard, the BW-at-a-certain-level requirement is not as stringent as one might think.
Last edited:
Further, one could get rid of the two output balancers with their 8 pcs 0.1% of resistors by implementing a 0V common-mode servo loop around the master amps, only needing only two 0.1% precision resistors and one opamp, some compensation and two feed resistors to the (-)-inputs of the masters. Noise at low gains should be lower and the servo's noise and distortion contribution would be common-mode mostly.