Martin,

'1. Lets assume the cable resistance is 0

Hi 301
As Cyclebrain pointed out, at 20KHz the amplitude error will be negligible. Measuring this deviation accurately, which is a fraction of a dB while of course not impossible, is tough. Therefore the measurement would probably have to be taken at some higher frequency where the measurement error of the instrument is small in comparison with the deviation.

As a note: I derived the equation for the L C network proposed by Cyclebrain. It reveals that its transfer function (Vout/Vin) = 1/(1-(2.pi.F)^2.L.C), for an ideal source (0ohms) and infinite load.

This illustrates
1. For an L C network. When the frequency (F)=0 the gain is 1. As F increases then the gain becomes > 1 i.e. positive. When (2.pi.F)^2 = 1/L.C the gain is infinite. Above this frequency the gain rolls off.

2. Considering the reactance of L and C and treating them as a potential divider is too simplistic

3. My original point was that the equation for the characteristic impedance of a cable (root L/C) is valid at all frequencies. The above Illustrates that should the L C derivation be applied to a cable, it is only valid when F=0, i.e. at DC.

Exactly, within the audio band, the effects are quite small. Even using a 500khz signal and scope, I really see no problem.

Although it appears to me in experimenting that changing the response as high as -3db at 500khz seems to make a sonic difference.

For whatever it is worth.

I guess it's time for me to bow out of this discussion.
It appears that we are both so sure of our "side", that nothing is going to change that. We are probably boring the crap out of everyone else.

I did mispeak about the 1/10 cable length to wavelength factor.

My simple voltage divider is not entirely correct as I left out phase factors for simplicity.

The complete formula for cable impedance does have frequency as a factor.
Zo = sqrt ( (R + 2 * pi * f * L ) / (G + j * 2 * pi * f * c) )

As the frequency becomes R and G become insignificant relative to the 2pifL and 2pifC terms. So it is simplified by dropping R and G. Then the 2pif values cancel leaving just L/C. At lower frequencies this simplification creates errors.