A discrete audio opamp I designed a while back and that's used in AVA products to this day will celebrate its tenth anniversary next year. So, I thought the time was right to say some things about it and discrete audio in general.
The discrete opamp in question came into being because I was unable to find a commercially available IC opamp that had the slew rate, GBW, and low frequency linearity I wanted. I could have the desired slew rate and GBW or the low frequency linearity, but not all three. So, I wanted to see if I could design something that did everything I wanted.
The plan was simple: Begin with a folded cascode input stage that's buffered through an emitter or source follower to a single-ended class A output stage. Then see where that takes you. A folded cascode because it has a lot of benefits for typical audio use, and a class A output stage because, initially at least, I didn't want to mess around with thermal tracking. I also decided to use passive loading as much as possible to minimize the introduction of potentially problematic parasitics or higher-order nonlinearities. The idea was that active loads could be introduced incrementally and carefully checked for pros and cons.
This resulted in a topology that used passive loads on all four legs of the folded cascode as well as on the follower between the input stage and the output. But to keep common mode rejection high, the input LTP got an active current source, and to keep output linearity high, the output stage also got an active load.
After the requisite design work, all the simulation models said that I was on track to have an opamp circuit that met all my criteria for open-loop gain, slew rate, GBW, and THD. But would reality concur?
We all tend to love our own creations. So, when I first grafted a prototype implementation of the circuit into a design, I didn't necessarily make that much of how much I enjoyed what I was hearing. But what I was hearing had none of the HF glare or stridency that I came to associate with audio opamps of the day, and it also had none of the LF and midrange bloat I had come to associate with high-speed opamps. (High-speed opamps are typically designed for video applications and have little need for LF linearity.) In short, it did exactly what I hoped it would. But personal bias is a strong force, so I didn't invest that heavily in my impressions.
It was only after playing it for Frank that I became confident that I had something noteworthy (pun intended?). Frank heard the same things I did and loved it so much that he asked me to design the circuit just as it was into new solid-state products, ASAP. This circuit is still in use today -- in the filter and output stages of the DVA Digital Preamplifier.
What about the planned revisions? Subsequent modeling showed that, contrary to expectations, active loading of either the cascode or the interstage buffer didn't yield a linearity benefit in this circuit. Sometimes you get lucky.
Some time after the above, I developed a version of the module that was further optimized for analog preamplifier line stages. Unlike the original design, which used matched BJTs in the front end, this one used expensive matched JFETs. Some of the supporting circuitry values were tweaked, but the topology is otherwise the same. This design too is still used today -- in the Vision SLR and Vision RB analog preamplifiers.
Downsides? The main one is something the user will never notice: Because the open loop gain in both modules is deliberately kept as flat as possible through the audio region, neither module has the incredibly high open-loop gain at DC that commercial IC opamps rely on to archive low, low output offsets.1 So, in both the DVA Digital Preamplifier and the Vision preamplifiers, additional circuitry is required to trim the offset to reasonable levels. Not a big deal. There are some additional issues that make designing with each of these modules kind of a PITA for the engineer, but again these don't impact the user. You'll never notice them.
Since these two modules were designed, many new and seriously improved IC opamps have been released, some as dedicated audio devices and some as general purpose ones. Without question, they have an important place in the audio designer's arsenal. But many has been the time that I've excitedly tried the "new best" opamp in my DAC or preamp, only to go back to my modules, in spite of what I wanted to hear. Newer devices are commendable for what they can do for what they cost. But for critical DAC and line stage applications, I'm still waiting for something that betters these discrete circuits.
Is there something magic about discrete circuits? No. There are only pros and cons. The biggest pro is arguably that isolating components thermally and electrostatically is trivial with discrete designs. Another pro is that their larger real estate makes it easier to dissipate power if you need to. Both of these are exploited in the two modules. The biggest con is probably that the increased real estate has the potential to introduce parasitics that impact VHF behavior. This means you might be able to squeeze a little more GBW out of an IC design than an equivalent discrete one. But both of these modules have GBW that's well in excess of what you'll find in a typical audio opamp. So, in this case, this isn't a significant concern.
Perhaps the most impactful difference between these modules and commercially available IC opamps is that they have been carefully tailored for their respective applications. This results in demands on the supporting circuitry that we have no issue putting up with but would be commercial suicide for something intended as a general-purpose device.
Sadly, the benefits that these modules confer come at a premium: They cost about an order of magnitude more than a decent audio opamp IC. Rest assured that we wouldn't ask our clients to pay this if we didn't feel it was worth it. I'm very grateful that they continue to make our clients so happy.
-
In many IC opamp designs, very high open loop gain at DC is also needed to stabilize the input stage at DC and low frequencies. The instability is the result of thermal modulation on the IC die from the output stage creating identical, simultaneous thermal offset voltages in both non-inverting and inverting transistors of the input LTP. Lots of VLF feedback "whips" the input stage into proper behavior, and this requires a lot of open loop gain. IC layout techniques can effectively mitigate against this effect, but I don't know how often this is done in practice. ↩︎