A few years ago, I developed an audio gain cell that was exceptionally fast for a fully discrete circuit and quite clean. That design ended up being adopted commercially, including by Audio by Van Alstine, who are using it in their DAC MK 5 and Vision preamplifier. I like to think that this gain cell is a key factor behind why the owner of a well-regarded manufacturer of luxury loudspeakers called the AVA DAC MK 5 one of the best sounding DACs he’d ever heard and a model for other manufacturer’s to live up to.
This gain cell project started as an experiment in engineering something that maintained what I liked about my favorite IC operational amplifiers while designing out what I didn’t. It was a back-burner project that took me a few years of on-again, off-again work to get to a point where I considered it done.
One of the guiding principles in the module’s design was to keep any mechanisms that might generate nonlinearity as simple and static as possible. The idea here was that simple and static sources of nonlinearity would be more likely to be low-order, which has a potential two-fold advantage: (a) Low-order nonlinearities tend to be subjectively less toxic than equivalent amounts of high-order nonlinearities, and (b) The harmonics produced by low-order nonlinearities are lower in frequency, which makes them better able to be suppressed by drooping HF loop gain, and this potentially leads to less out-of-band spuriae modulating into the audio band.
In the initial development, I used a single-ended class-A output stage because of the simple input impedance characteristics and static thermals. While I considered replacing this with a push-pull output stage for production, the single-ended output ended up working so well that I decided to use it as-is.
It took a lot of simulation and some workbench iteration to get things optimized, but overall the result performs subjectively and objectively better than I could have hoped for. It continues to impress me every day with its resolution and musicality.
The circuit has a fully differential input and plenty of open loop gain, which means it could be considered an operational amplifier. In spite of this, I’m not comfortable calling it an opamp. Its input bias current is too high to make it as general-purpose a device as most people expect from an opamp. Hence the expression “audio gain cell” or “audio module” whenever I talk about it. The high input bias current is a direct result of the high current I used in the BJT input stage to make the circuit as fast as it is. This and a couple other design decisions make working with the gain cell kind of a pain in the butt, but the sonic payoffs are worth it.
Naturally, I began to wonder whether I could design a circuit that maintains the virtues of the original but has lower input bias current. This would make it perfect for a range of applications that the current design isn’t as well-suited to. To that end, I’ve been working with a revision of the existing architecture as well as completely new architectures and have mode some good progress.
Even though it wasn’t a goal of the project, in the process of this work I’ve managed to increase my confidence about which measurable behaviors make different opamps, gain cells, and power amps have the subjective character they do. In particular, it’s interesting to hear what happens when different circuit architectures converge on the same selected engineering goals. Perhaps not surprisingly, they begin to converge in sound as well.
I’m very close to some production-ready solutions, so I’m excited.