The problem is that you are not appreciating the scale of the problem. So a little while ago, I was listening to the sample rate conversion for the ADC project. This used my 67 bit Lagrange interpolator running at 104 MHz, and I initially started simply with first order, and pS interpolation (this time interval is the resolution on the 104MHz time domain against the incoming 98MHz word clock). Ran simulations, and then listened, and I kept listening until I could hear no degradation. Eventually I ended up at 8th order Lagrange; but the really interesting thing was the time resolution. This had the effect of creating noise floor modulation that was dependent on the rate of change of input signal as well as amplitude; so a DC signal creates zero noise floor modulation, and as you double the frequency you double the noise floor modulation - so it's noise floor modulation that is dependent upon frequency and amplitude. The crazy thing was going to femto second resolution gave a 60dB improvement on simulation - and a big improvement on sound quality. I then went to atto second resolution; and here it was difficult to measure (from simulation) the noise floor modulation, so I started using 0dB 88kHz test signals (something you would never see in real life). Then I could see it at -300dB. Femto to atto second change was easily audible though. Then I went to zepto seconds, and now the noise floor modulation was not visible, even with 88kHz - so it would have been at -360dB levels. This is ridiculously small - but going from atto to zepto seconds resolution on the interpolator was still audible - and it sounded exactly like noise floor modulation, that is smoother and warmer.
What I concluded about this is that the target for noise floor modulation needs to be around -400dB. That is absolutely impossible to measure, and frankly it's an impossible engineering standard, which is why you can still hear subtle changes in cabling with the M scaler.
On the development of the M scaler, the ferrites were put into the OP drivers. And this indeed solved the problem - adding clip on 2GHz ferrites actually made the SQ worse (brighter indicative of more noise floor modulation) so I decided my treatments worked, so I submitted the design to Chord for production. Nick from Wave cables got a production M scaler, and sent me some prototypes of his cables - and indeed these sounded better, due to the solid ferrites he used.
So this situation is not a design fault, but merely illustrates how incredibly sensitive the human hearing system is. And in absolute terms we are hearing very,very,very tiny changes; the problem with listening tests is that something very tiny can become quite significant on an AB listening test. Remember too that this is the very top micrometer layer of icing on the cake! Moreover, if you want to perfectly isolate the RF noise, simply feed M scaler with an inexpensive battery lap-top power pack, and use optical inputs (although USB most of the time is fine).
