You'd think this method would be OK for power amps because they
are pretty good in the cosmic scheme of things. But, when David
Hafler was using a signal subtraction-based methodology to promote
his amplifiers some years back, he ended up having to fix up the
phase response of his own products in ways that really didn't do
much for how they sounded under normal conditions.
Crown developed what they called a "null amplifier" that (supposedly)
performed this test correctly. Dr. Clayton Barclay, late of Crown, told me
that the Crown PL-1 amplifier could get close to a 90dB null on program
material. I've been trying to get my hands on a null amplifier, but they're
hard to come by.
Having to compensate for the amplifier's phase errors is an old problem. As
it's a static error *, and can in principle be compensated for with an
all-pass network, it's not significant.
* Well, not quite... As the average power level rises and the amplifier's
bandwidth narrows, the phase shift will increase.
You can find examples of applications of this methodology at
http://www.pcabx.com/product/amplifiers/index.htm. I by pushed the test
signals through these amps enough times so that their characteristic sonic
colorations are reasonably easy to hear. You can then back off on the
number of repetitions until you stop hearing the amp's colorations. In
every
case, that happens by the time the number of reps is one.
Well, well, well... Isn't _that_ convenient. That doesn't produce even the
faintest "Hmmm..." response?