In reply to:
We need to separate two distinct issues here: (1) impedance linearity of maximum available power, and (2) power-level dependency of frequency responses under realistic loads.
The impedance of any speaker varies dramatically with frequency, regardless of the power output. It becomes a stronger factor as power levels increase.
In reply to:
the frequency response of your H/K 3470 under a realistic speaker load will be virtually flat (except for the tiny dips due to the finite output impedance as explained earlier) when driven at 1W, 10W or 100W into 4 ohms
That is a good point. However, I don't believe it's as clear cut as you're indicating. I have to fall back on my argument about transients. I think this is where the differences get revealed, because even if you're not sustaining 100wpc, music is very dynamic. Every single cymbal clash and kick-drum whack will instantaneosly demand a huge rate of current draw. You can calculate exactly how many farads of energy are stored up in capacitors, and how long it could supply energy at a steady voltage, but how does that current draw affect the nature of the sound, and is there a kind of hesitance in your wall current supplying the sudden jolt?
One very well-established, but counterintuitive, observation in high-end equipment design is the effect power supplies have on the sound of a component. I'm not just talking about amps; even preamps and digital sources are affected. The more "overhead" the power supply has, the less constrained the sound. This is one of the reason high end components weigh so much. It's oh-so-easy to measure continuous current draw situations, but with dynamics... well, remember the wonderful world of derivatives? If the current goes from 1 amp to 100 amps in .001 seconds, what does the current graph look like? A straight line?
This goes back to the square wave. Many amps do not actually produce great-looking square waves. It has to do with amplifier bandwidth: how high the frequencies go before the response tapers off. On a $9000 100wpc Linn Klimax Twin, Stereophile measured a 3dB rolloff at 58kHz. I will quote from the article: "This curtailed ultrasonic response also rounds off the leading edges of a 10kHz squarewave, but the waveform is refreshingly free from overshoot or ringing." The graph indicates this very obviously, and it also has an interesting curve when dropping down to the low part of the signal.
I'll stop here, as I can no longer make a strong assertion about exactly what causes the frequency-specific variation. But most measurements vary greatly with frequency, such as channel separation (by 10s of decibels), noise floor (often by 10s of decibels), and other things. Given the fact that plenty of people on Audioreview will comment on the "brightness" of a $400 Yamaha amp or the "warmth" of a Harman Kardon, I am inclined to believe there is some cause. But I'm pretty much out of steam on this argument, so maybe we can call it a draw?
-Cooper
