The error in that PDF file is that the "complex signals" are not normalized to a +/-1 Volt range, and they use a 3rd order polynomial for the distortion that gives increasing amount of distortion at higher input signal levels, so with nearly 5 Volts of peak voltage for the 7 tone signal (as opposed to just 1 Volt for the single sine wave), there is no surprise that the amount of distortion would be huge. Especially with the polynomial being applied four times in a row. For more realistic results the input signals should always have had a range of +/-1 Volt, and the distortion polynomial should have been scaled so that its output is also within the same range.
To illustrate the above point, here are a few graphs:
100 Hz sine wave with 1% 2nd order and 1% 3rd order distortion (y = (0.97*x + 0.02*x^2 + 0.04*x^3 - 0.01) / 1.02), the distortion is applied four times to simulate 4 amplifier stages (left channel is the input signal, right channel is the distortion residual):
101, 173, 307, 523, 919, 1607, and 2801 Hz sine waves, each at 1 / 6.0142 amplitude (peaks within +/- 1), the distortion is the same as above:
The same mix of sine waves, but with 1 / 2.5 amplitude for each tone (peak level = ~2.4):
With an out of range signal, like in the PDF, the distortion increases by orders of magnitude. However, once the peak level is normalized, the distortion peaks are not higher than with the simple sine wave input. Of course, that is to be expected, since the distortion is frequency independent, and can be described with a simple polynomial with no "memory" of any previous input. So, the article in the PDF file is somewhat misleading.
It is possible to test if an amplifier that uses negative feedback has unexpected high distortion when fed with a complex signal using null testing, and also double-blind listening tests. However, as far as I know, if the distortion+noise is low with any sine wave input (at any level and frequency within the range the amplifier is intended to be able to handle, and with a wide bandwidth measurement), then it is expected to be low also with music.
By the way, I recommend reading this article by Bruno Putzeys on negative feedback, if you have not already done so. It explains well the origin of some feedback myths, and also why they are wrong.
Edited by stv014 - 11/8/12 at 6:52am