Signal current is usually scaled to about 1mA max. A rich class A VCA might run 3mA bias. 3mA bias at 0.7V is 2 milliwatts, temp rise about 1 deg C, essentially zero (either absolute or relative).
Remember that I looked at this 25 years ago, decided that VCAs were interesting but (unlike Gary) not my life's calling, and the little memory cells die off.
Yeah, well, sometimes shit happens. :grin:
Working class A, THD is low but noise is high. However, noise is constant. And matching is not as critical.
All correct. The only thing I would add is that matching is not so critcal for distortion performance, but even more important for offset vs. gain performance (control-port feedthrough) due to the high quiescent current. Hence, the persistance of a symmetry trim on the 2001. Of course, if you're not going to move the control port too fast, you may not care. Still, even a fast fader move might woof.
In class B noise is low, but rises with signal. For perfect devices, THD could be low; for real devices it can be nasty or very small. The increasing perfection of devices made discrete class B acceptable, and (I may be wrong!) I think it became the usual way to do it. Monolithic devices would match N to N and P to P, but getting the PNPs to be anything like the NPNs was tough when they were on the same chip with affordable pre-1990(?) processes. There were some VCAs built with a PNP array and an NPN array, so each side could use an optimized (and existing and mass-produced) process: at least each polarity matched, even if you had to trim the P/N differences away. I gather that good PNP/NPN wafers are now quite possible; though the up-front cost of getting time on those foundries may be hard to swing considering the very small market for top-end audio VCAs. Obviously Gary is in touch with those aspects, though he may not care to share everything he hears at work.
I would add that there are a couple of sources of distortion that become an issue with class B (or AB). One is the match of gain between the top and bottom of the gain cell (handling the negative and positive polarty signal currents, respectively). This is what the "symmetry" trim adjusts. The distortion resulting from a mismatch is almost purely 2nd harmonic -- a fact that some of our customers utilize deliberately. The "nasty" part would result from the wrong choice of input amplification stage for the VCA -- one whose open loop performance varies with the instantaneous transconductance of the gain-cell devices. Driving the gain cell from a heavily degenerated transconcuctance stage (class A, FWIW) makes the loop transmission essentially independent of the signal currents. (Still dependent on the gain setting, however.) This is one of the differences between our 218x devices and the older 215x devices.
Matching N's to N's and P's to P's is still the crucial part - but the P's still have to be "good" devices (i.e. adequate bandwidth, beta, log conformance.) The older 215X-series devices are built on an older foundry comlementary junction-isolated process. It has vertical PNPs that are pretty good, but still suffers from the relatively large collector-substrate capacitances inherent in a high-voltage JI process. And, since the gain-cell devices must be large for good log conformance (about 1/3 of the chip), those get pretty big. The 218x devices are built on a dielectrically isolated process in our own fab. The oxide isolation yields much reduced collector-substrate capacitance. The difficulty of maintaining the attention of foundries at our volumes was a big motivating factor in buying our own small fab. It's been a much longer and more expensive road to get the processes up and running than we ever expected, but it gives us a much greater level of control over what we're doing. This is starting to border on a sales pitch so I'll stop -- but we really do need to keep the volumes up in that fab :wink:
Gary