I don't really understand this. Is a soft diode slower then a fast diode?
The "soft" recovery characteristic has to do with the di/dt of the "snap" when the diode turns off. Snap or step-recovery diodes are designed to make this as fast as possible by having the minority carriers which hold charge in the junction all run out at the same time. Soft recovery rectifiers are designed to make turn off after reverse recovery more gradual.
Standard recovery rectifiers, especially higher voltage ones, are sometimes produced using the same PIN structure that snap-recovery and varactor diodes use so it should not be surprising that they can behave like them. (1)
So we have moved to using more lossy diodes to prevent EMI problems?
That is sort of the case but the networks used to control EMI also lower efficiency.
Don't you want that stuff turned into high frequencies as much as possible so its easier filtered? Or is it more likely to cause problems if it gets into the circuit in the first place?
If the frequency components are too high, then they more easily couple through the parasitic elements of the filters and circuit. Placing small capacitors directly across the diodes removes the inductance produced by the loop area between the rectifier and bulk input capacitor. No amount of filtering will help once a large loop between the rectifiers and filter is spewing magnetic flux everywhere. (2)
Does this have anything to do with transformer shunting with capacitors like I have seen once? Perhaps to bypass its mixing nature?
It is likely as that is one way to control the EMI. The three steps I usually take in order are:
1. Use better diodes. (1)
2. Add capacitors directly across the diodes. I might do this in addition to using better diodes in a critical application.
3. Add ferrite beads in series with the diodes. I have seen this done but I am not sure why capacitors were not used instead. Maybe it is more effective or lower loss?
I do remember, in the 1970s and 1980s, seeing those caps in parallel to mains rectifier diodes in some (not all) products.
The reason given was to prevent RF issues, whatever that meant. No further details were provided.
Audio gear is particularly susceptible to this problem which results in a 120 Hz buzz but it shows up in test equipment also. (4) When ADCs are involved, it can manifest as excessive flicker. (3)
I also believe that an RC solution would dampen any spurious oscillations. But my personal opinion is that blindly applying a circuit without actually understanding what one is planning to achieve, or actually measuring its effect, is a recipe for problems.
Ideally a series RC snubber would be used but leaving the resistor out if it is not needed is less expensive. The capacitor values are usually 10s to 100s of picofarads for 1 amp diodes. For expediency, I typically just find the smallest capacitor value which was sufficient and then double it.
(1) Standard recovery rectifiers seem to be the worst but it varies between part number, manufacturer, and perhaps even lot. Some 1N4001s do it and some do not. So a better diode might simply be from a different manufacturer and you could always qualify your standard recovery rectifiers for having a soft recovery. But if you are not going to do that, then using fast recovery rectifiers which have less stored charge for line frequency rectification is usually the least expensive option. If you want to use diodes without worrying about it, then add the capacitors. A rough rule of thumb seems to be 4 to 10 times the diode's junction capacitance.
(2) Can of worms? The can is open. The worms are everywhere.
(3) This gets back to precision ADCs where once real flicker noise and other noise sources are accounted for, quantization noise should result in 1 count of flicker maximum and it is possible for 2 counts to indicate a real problem even at 16 bits. That may not seem like much but some ADCs and circuit designs even going back to the 70s and 80s are potentially that good. Some early 8-bit DSOs had noise levels this low.
(4) I see a business opportunity here selling special diodes, capacitors, and inductors for rectifiers used in audio.