A little bit of history
Yet Another Poweramp
Another way to skin the cat
After the PGP Amp was completed I was contemplating for quite some time the next steps into my power audio journey. After all, probably the most performant audio power amp ever was successfully built and tested, so anything that would follow could barely improve (if at all) the PGP performances.
Probably the loudest critic I've heard about the PGP amp was on the complexity. True, the PGP amp is very complex, and there's good reason for that. That's a consequence of something that could be called "feedback conservation law". Don't look up this concept, you won't find it in any Control Theory book; it's actually a byproduct of the Lurie's Bode integrals theory. From a very simplified perspective, this is just another way to illustrate the Gain x Bandwidth = constant principle.
The bottom line impact of the "feedback conservation law" is that, given a base amp, you can't provide more feedback without affecting the stability margins. If you look at the Gain vs. Frequency graph, the area ("the Bode integral") under the gain curve is constant, and the only way to get more loop gain for further performance improvements is (theoretically) to start with a very high bandwidth base amp and, to isolate the residual poles, use N poles to brickwall rolloff the frequency response and N-1 zeroes to bring the phase back and comply with the Bode stability criteria at ULGF. Of course, the larger N, the higher the sensitivity, so potentially the Cherry NDFL (probably the most known implementation of this concept, used in the PGP amp) could be the largest order (three) that's practically feasible.
For these reasons, implementing the Cherry NDFL requires an extra gain stage. Add the CMCL loops and you got a very complex front end. What other options do we have here?
Well, first immediate idea is to push the unity loop gain frequency (aka ULGF) as high as possible. This was the idea behind the first finalized YAP amp (v2.1) that was for so long promised to be published here. YAP 2.1 uses a local current feedback, MOSFET based, output stage, with an ULGF no less than 8MHz. As a consequence, the input stage/driver could successfuly be implemented as a two pole compensated amp with an ULGF of 2.8MHz and hence almost four times the loop gain of a standard Blameless amp, while keeping the amp rock solid stable. For an apple to apple comparison, YAP 2.1 has about 1.2ppm THD20 into 8ohm at 100W output power so, for all practical purposes, performance wise, is at par with the PGP amp. However, it was built with four pairs of vertical MOSFET power devices, so it could easily drive 4ohm loads (at about 4ppm distortions), therefore addressing one of the PGP Amp limitations.
Subsequent releases of the YAP amp, up to v3.2, were attempted improvements around the same concept. I still have the boards of all these amps, some of them installed on huge heatsinks, but none of these amps were actually ever built into a case, with all the frills that a finished amp should have. Reason is, I was somehow deeply unhappy with all these amps, and that's because of their "conditional stability". "Conditional stability" doesn't mean that the amps are anywhere near close to explode in a burst of oscillation. At the designed closed loop gain, these amps have the regular 70-80 degrees of phase margin, while the gain margins are well over 10-12dB. However, because of the multiple pole compensation inherent phase dip, there are closed loop gains at which the Bode stability condition is no longer satisfied. Now, when can an amp have such a (lower) closed loop gain? During the power on/off and during clipping. The clipping issue was addressed by frequency compensated active clamps, while the power sequencing takes care of the power on/off transients. The net result? Again, very complicated designs, by no significant margin simpler than the PGP amp. From a complexity perspective, all these YAP amps were ultimately a failure. The big lesson learned from the early YAP builds was the technology and PCB layout, allowing an ULGF of 8MHz. One day, time permitting, I will post these amps schematics and PCBs, they are though very good amps, certainly worth building by the audio enthusiast.
The other option towards better performance is to improve the base amp linearity so that, for a given closed loop output distortion level, less loop gain is required. This leads immediately to the idea of abandoning the MOSFET output stage. Whatever the MOSFET proponets will ever claim, the reality is that bipolar output stages have, for the same output current capability, less distortions. Reason is the MOSFET Crss which is strongly nonlinear, in particular when Vgd approaches zero (that is, when the output voltage approaches the rails). This, and the almost complete extinction of the power MOSFETs intended for linear applications (that is, N/P channel parts designed for matched transconductances rather than switching parameters) made me think about implementing a bipolar output stage.
On the other side, I was yearning for a powerful amp (300...400W continous power into 4ohm, able to drive 800W bursts into 2ohm loads), simple enough to be affordable, that would still qualify as ultra high performance and, perhaps the most important feature, that would not require special lab equipment to adjust for optimum performance (recall the distortion trimming in the PGP amp?). Somehow arbitrary, I have set the distortion performance goal (THD20 and IMD 19+20KHz) at 10ppm (0.001%) into 8ohm, at 200W output level, and 25ppm (0.0025%) into 4ohm, at 400W output level, all low order harmonics.
Meet YAP v3.2, my first adventure in the power bipolar audio amps.