From a post on audioasylum.com:
Although they seem kind of simple, rectifiers are not well understood from an RF-emission standpoint. If you think of them as really crude Class B (not Class AB!) high-power output stages, the mental picture will get clearer.
The rectifiers commutate the AC waveform just like a fast mechanical switch, and like a mechanical switch, the zero-crossing region is not dealt with cleanly. Conventional silicon diodes actually knock a hole out of the waveform thanks to charge-storage effects; Schottky diodes (which are not available for high voltage) don't have charge storage, so no divots in the waveform, but we're not home free yet.
The stray C's and L's of the power supply circuit "rings" when the rectifier shuts off; this can create pulse trains anywhere from 2 to 30 kHz, with a fairly high Q, commutated at a 120 Hz rate. This ringing is dynamically modulated by the current flowing through the rectifiers, so it isn't a steady-state phenomenon ... keep that in mind when you articles about snubbers.
Snubber circuits can quiet down the ringing quite a bit, but need to be precision-tuned to the exact power transformer and filter circuit components ... one size does not fit all. Snubber tuning requires pretty careful frequency measurements, and makes assumptions of linearity of the power supply reactive elements that may not be valid under dynamic conditions of varying current draw.
The measurements I saw used a Tek scope to zoom in at 10V/div on a 480V B+ supply; the scope probes looked at the differential signal coming off the full secondary of the power trans, and with different types of rectifier. The shape of the zero-crossing region is mostly controlled by the rectifier, and the way it behaves as it approaches shutoff.
The ugliest zero-crossing was conventional silicon, complete with a big hole dug out of it and a wavetrain of oscillation. This is at its worst in bridge-cap supplies, the type used in most solid-state equipment and DC heater supplies. Schottky and HEXFRED diodes were better, but the actual zero crossing still had a steep "Class B" appearance to it.
Vacuum tube rectifiers were quite a bit smoother through the zero crossing, thanks to losses in the plate circuit, but there were still significant differences between various types. If a high Rp was the sole reason for smooth crossover behaviour, you'd think a high-Rp rectifier like a 5R4-GY would be best, but that's not the case. The damper diodes had the smoothest transition region of any device we measured, making them suitable for preamp use as well as power amplifiers.
I've been messing with speaker-crossover filters since 1975, and filters have lots of drawbacks. They're made with components with weird nonlinearities that are hard to characterize, and getting rid of hash or assorted resonances takes a lot more dB of filtering than theory would indicate. The best filter is no filter, really, but as long as we use AC power we have to put up with these things. I'm not too worried about 60 or 120Hz, really, but midrange hash in the 1-5kHz region is another matter. Since rectifiers are prime sources of noise, I feel it makes sense to analyze rectifier types and PS supply topology with a view to minimizing EMI generation at the source, since the noise is readily propagated back through the power cord (which is why differences are audible) and into the chassis by electrostatic radiation.
I'm not really a fan of regulation unless it's shunt regulation, and I have reservations about the sonics of regulators that use internal feedback to create a low output resistance. Any time you have active circuits, questions of of slew rate and other dynamic nonlinearities arise ... and it seems a shame to build an unconditionally stable non-feedback amplifier and spoil it with a maginal regulator that is not dynamically stable. A common problem with ultrafast regulation is a tendency to briefly oscillate at a high frequency (as high as 100MHz!) as the regulator gets stressed by high-current-demand conditions. Normal 300MHz scopes cannot detect brief oscillations unless the oscillations are really huge, say 10V or more. It takes a spectrum analyzer to see the troublesome (and very audible) oscillations in the 10 millivolt region.
Brief oscillations are a lot more common in high-end equipment than you'd think, mostly because small manufacturers can't afford spectrum analyzers, and don't have engineers on staff with RF design experience. I'm pretty wary of ultrafast anything unless I've seen the spectrum analyzer output for myself ... and we're talking 100 MHz bandwidth and 80dB on-screen range to really do the job ... this isn't your run-of-the-mill PC digital sampling analyzer.
So there are a lot of not-very-pretty things that happen with rectifiers and ultrafast circuits (such as cascode-this or cascode-that). Can you hear them? You bet!!!
Matt Kamna helped me chase out a 3MHz oscillation in a fancy audiophile DAC, way down in the millivolt region (quite invisible on any scope), and the DAC sounded completely different (and much better) when the problem was removed. (It was caused by a + regulator for a front-panel LED that didn't have a decoupling cap! It was a design error, pure and simple. The easy and thorough fix was to disconnect the + rail power supply to the LEDs and the doggone regulator that fed them. All quiet on the western front after that.)