Reducing rectifier noise in high voltage power supplies

[loudthud]

The following is an article written by John Camille that appeared in Sound Practices magazine. Thanks to Joe Roberts for allowing it to be reprinted. It discusses a technique to reduce rectifier noise in high voltage power supplies. Sound Practices was an audiophile publication focused around single ended tube amps. 16 issues were published back in the early to mid 90's. I posted it in the Debug Your Build forum but it's didn't get much traffic over there. The right column on the last page is the contents of the issue with the statement from the publisher.

[R.G.]

Yep. Ordinary speed silicon junctions use both majority and minority carriers through the depletion region, and when reverse biased, they conduct for a while until the slower minority carriers get swept out. Then they shut off with a "slam!" that makes any parasitic L-C attached to them ring, and make a squark of RF. This gets coupled to things in the amplifier.

That was where things stood in 1994, although as the article notes, there was progress being made with fast diodes for switching power supplies. Schottky diodes have no minority carriers, so they are both faster and don't slam off, but they are limited in voltage.

Today, nearly 20 years later, we have FREDs - Fast Recovery Epitaxial Diodes - and other fast, soft-recovery diodes that do not have the slam-off characteristics of ordinary power diodes. They are PN-junctions, but with special sauce to make them not do the door slamming that used to go on. Many of them have voltage ratings in the 800-1200V range, probably bigger now.

Ordinary diodes can be snubbed or filtered, much like the article says. One of the most common things is to use a series R-C network paralleled with each diode to (1) lower the RF ringing frequency with the cap, (2) introduce damping with the resistor and (3) not affect the diode's performance much at 120 Hz and lower.

[trobbins]

As a comparison for the authors obvious effort, it's a pity that the author didn't indicate what level he could get observable diode induced noise down to when using the far simpler technique of capacitor bypassing the PT secondary windings to localise the inductive energy in the recovery transient to the secondary winding.

[Mick Bailey]

How much diode switching noise do people hear? Looking at the majority of classic guitar amps with SS rectifiers they don't have snubbers. The PSU illustrated is putting out 1400v. An SE amp that's intending to reproduce full-bandwidth audio may be susceptible, but a PP guitar amp running 450v may not be.

[Steve Conner]

The snubber cap works fine and is often required to pass EMC.

However, when you're buying in small quantities for a hobbyist tube amp build, the difference in price between regular and ultrafast diodes is so small that it may work out cheaper than adding a snubber capacitor. You might like to try the UF4007, UF5408 or SF1600.

In solid-state hi-fi amps, the higher current ratings mean that 4 FREDs work out a lot more expensive than one cooking-grade bridge rectifier module.

I've seen diode switching noise on a distortion analyser, but I've never heard it.

[jimboyogi]

Copied from another thread http://music-electronics-forum.com/t34402/ , but here is what it looks like on the speaker output of my latest build. Using 1N4007s with no snubbers. I've got some UF4007s on order to try instead.

[trobbins]

One has to be very careful in presenting noise measurements, as the layout of parts, and the probing technique, can make different peoples results look worlds apart for exactly the same parts and loading.

I thought the author of the referenced paper indicated fairly well that his criteria for noise reduction was purely what he could see on the best measurement equipment he had. But then I thought he blew that somewhat analytical notion with "the noise reduction technique described above will remove several layers of grunge from the sound of an amplifier".

[Steve Conner]

Yeah! Guaranteed to turn Soundgarden into Barry White!

If only life were as simple as a tube amp power supply. I'm currently trying to trace the path of RF current at 320MHz in a badly laid-out 6 layer board.

[R.G.]

Here's an article I liked for encapsulating a lot of info on the topic:
http://www.hagtech.com/pdf/snubber.pdf

As I remember, the issue is involved with the layout dependent parasitic L-C circuits in the wiring to the diodes and the capacitances of the diode and wiring. These make for resonant circuits which ring at RF when excited by the transient step of the diode slamming off. This is then radiated or conducted to parts of the circuit that make the RF into a blip of signal at the rectifier frequency and induce a buzzy hum, something like the buzz from fluorescent lights.

Capacitors, being non-dissipative, lower the frequency of the ring, and may lower it enough to ruin some of the transmission modes. An R-C snubber actually damps the ring by eating the energy. Critical damping almost eliminates the ringing, if you can determine the value of R that introduces critical damping.

Fast, soft recovery diodes reduce the energy in the transient.

[trobbins]

RG - very good article by Hagerman - not many such articles focus on simple powering, as switchmode was and still is the bleeding edge for articles.

The thing for people to keep in mind is that the added bypass C or RC needs to be at the transformer winding, or half winding to CT, and any flying leads from the secondary need to be managed carefully (ie. not splayed in all directions to far flung parts). Also valve amp HT windings have a relatively large effective series resistance, which goes to dampening ringing better when just a C is used, and as seen in the two reference papers the resonant winding characteristics are about 20x different (560kHz to 26kHz). Note that Camille may have measured 26kHz as the full secondary winding response, which would be fine for a full bridge supply, whereas the rectifying current path for most valve amps would involve only half the winding (CT to one end of winding).

Good luck Steve.

[R.G.]

A very wise man said in another thread, "It is always best to suppress transients at their source."

For snubbing diode transients, the snubbing caps or RC networks need to be as close to the diode as you can get them. For bridges, I like to mount snubbing caps or RC pairs right at the leads of the bridge itself. For individual diodes, mount them right across the body of the diode. Snubbing transformers is more subtle. Transformers don't generate transients in themselves, they generate transients when something external interrupts current in their leakage inductances. Well, sometimes the primary inductance, but that's a one time thing, not the recurrent transients of diodes or other current breakers in circuit operations. Diode turn-off noise can be thought of as noisy transients that recur at regular intervals.

I prefer to twist the leads of low frequency transformers to get the current loop area down. Snubbing needs to be at the device creating the transient if you can get at it - for rectifiers, right at the diodes. For high frequency stuff, it's best to use surface mount snubbers as close to the pins as you can get them. If the transients you're protecting against come from outside the equipment and are unpredictable, things like MOVs and TVS devices at the transformer leads can help suppress the energy delivered to the transformer. Different situation, same idea - catch and dissipate the transient energy in resistances.

In low frequency transformers, the thing which drives the values of snubbers is usually not well controlled - the leakage inductances of the windings from one to the other. The main transformer inductances are hidden inside the transformers behind the leakage inductances where you can't get to them. Transformers of the same part number may have varying values of leakage from unit to unit. It's tough to get critical snubbing in situations like that.

Also, windings are separated by the leakages. The leakages act like they're in series with the leads inside the windings, and isolate one winding from another. So snubbing transients on one winding doesn't help on the other windings. Fortunately, if you snub well on one winding, the path through the winding capacitances doesn't carry much of the transients or snubbed remainders to other windings in low frequency transformers, courtesy of the leakages.

[trobbins]

The leakage inductance in the PT secondary winding that is passing the rectifying current through the diode that is turning off is the source of the energy that is causing the transient. The current that is trying to be turned off has to 'instantaneously' divert its path, and preferably via the smallest closed loop to the winding inductance. That's how I see it (?!)

If the diode didn't go in to reverse conduction, but stopped conducting at 0A, then the energy in the leakage inductance is zero, and hence a benign issue. Which has been the benefit of improved diode technologies, or of that olde technology called valves

Certainly providing a bypass snubber across the diode allows that current path to continue and damp down to zero, but as I see it that is a much larger and more uncontrolled loop as it often includes just the main filter capacitor and to a lower level through the OT primary.

[Steve Conner]

The leakage inductance might be the source of the energy, but the diode is what converts the energy from 50/60Hz to a troublesome RF frequency. Therefore the diode is the source of the RF and is what you should be applying the snubber to.

If my vacuum cleaner was interfering with my radio reception, I'd snub the vacuum cleaner motor, not the pole transformer out in the street.

Ideally you should minimise the loop area of the whole circuit and then it hardly matters where you put the snubber. It just needs to provide an alternate path so that when the diode current stops suddenly, the transformer winding current can divert somewhere else and tail off more gradually.

If one path has a loop area much different to the other, then more RF will be radiated as the current quickly snaps between paths. So ideally the two paths should share as much wiring as possible, and this is achieved by putting the snubber directly across the diode.

But when snubbing a bridge rectifier, I think it is acceptable to put a single RC snubber across the secondary of the transformer feeding it.

[trobbins]

Steve, as you indicate, if the diode and filter cap part of the loop has no inductance then there is effectively no difference between snubbing across the transformer winding, and snubbing across the diode except that the diode snubbing 'circuit' includes the filter cap in the current path. So I guess the subtlety we are discussing is related to (a) when the diode and filter cap part of the circuit has some inductance, and (b) the effect of having the transient energy take a path around the diode and filter cap loop, or another loop.

As you can see, I prefer to try and steer as little of the transient current through the filter cap, as that is one escape mechanism for 'noise', and similarly in oldish amps I come across, the ss diode and filter cap has a much larger loop area (and hence escape mechanism) than the placement of a snubber across the PT winding. Because I can't hear the noise anyway, I just go for the simpler cap across the secondary winding if I do anything at all, and don't go the extra distance in adding a diode bypass.

The vacuum cleaner motor is probably the dominant source of the interference in your example due to the motor winding inductance, but yes there is inductance on both ends of the circuit loop so breaking the loop at any point is best snubbed at both ends as the energy in the inductance at each end then mainly circulates locally with little transient current in the rest of the household wiring.

An example that I think has similarity is the typical single switched flyback smps primary - the snubbing can be across the (eg.) FET drain-source, or across the transformer primary - typically I have preferred to snub the primary winding as close as possible to the former terminals, and if anything is added to the FET then it is just a voltage clamp device.