I think the above gives a decent view of DA. My own experiments, limited as they were, follow results predicted for DA (Cole & Cole) and largely conform to experiments Bateman and others performed to test DA across differing dielectric types. Generally speaking, DA is worse the larger the capacitor and the higher dielectric constant.
How might this understanding apply to the Ncore? Here's my hunch. The Ncore draws a series of pulses from the power supply. Those pulses induce, among other disturbances, pulse-like voltage variations on the supply's complex dynamic impedance. The current pulses are for their part square wave composites of the amplified analogue waveform + switching harmonics. The psu voltage variations they induce will be a distorted version of those pulses. It is these variations, among others (like mains noise etc), that the electrolytic capacitors are placed to filter. Anything not filtered passes directly into the speaker. Because psu variations follow the music signal to some degree, and in other ways bounce around in some complex fashion (= a fashion that, imo, is sufficiently high to escape characterization, therefore not random), their presence in the audio band will muddy and distort the reproduction, quite audibly so.
I know the Ncore's feedback, in its limited way, addresses part of the distortion generated by these supply variations; so, too, does differential cancellation (PSRR). Because these are limited remedies, and because feedback introduces problems like higher-order distortion, the more one can eliminate noise from the rails before feedback and differential cancellation the better. I follow the rule that any node in a circuit is in its most general sense a voltage divider. The less noise appearing at any given node, the less is passed through to whatever next stage (or speaker) that node feeds.
All other things being equal, it seems to me that audio-band supply noise is perhaps the worst for class D amps. In addition to the normal range of supply disturbances affecting audible frequencies, class D amps also have the following:
• Switching noise modulating audio and audio+ frequencies. I maintain that frequencies beyond 20KHz are audible in some form of sensory perception. I can personally hear these frequencies as something of a pressure affecting my hearing, particularly where those frequencies are incoherent with the musical signal. I hear this effect with SACD and PWM amplifiers, and I suspect the use of feedback complicates the type of noise present (reference Pass' article on feedback-related IM effects).
• DA noise. My tentative theory concerning DA noise in class D amps is this: because DA recovery voltages are fractions of the original inciting frequency---the UHF pulses---this noise looks undoubtedly to reach downward into the audible (including audio+) frequencies.
It seems to me that part of what I hear wrong with class D HFs---their opacity, lack of sparkle and fineness, etc---traces to the above sources of noise, which are largely unique to class D design. The level of this noise looks to me to trouble just those HF subtleties I find missing in class D, including Ncore, reproduction.
And fwiw, and apart from DA, Ncore output capacitance looks to me insufficient to filter any such audio-frequency voltage variations. At 100Hz, 250uF has a capacitive reactance of about 6.4 ohms. Depending on the speaker's impedance, noise appearing at or near that frequency will virtually halve between the capacitors and the speaker. I'm uncertain what capacitance the Hypex smps has after its output, but from memory it seems that most of that capacitance resides one stage back in that supply.
______________________________
There's my long answer---the many posts above, including this one---to why I prefer to bypass output capacitors in this amp.
