Commutation Noise in Rail-Switching Amps

[bob p]

I'm trying to find out if any of the major amp manufacturers have managed to fix the problem of commutation noise in rail-hopping amplifier designs. (aka: Class G)

It seems that I'm one of the unfortunate people who can hear commutation noise and finds it objectionable. Most of my experience with rail-switching amplifiers involves listening to Bob Carver's original rail-switching designs that were built back when the technology was patented and only Carver amps could use the technology. Now that the patents have long expired, amplifier companies like Crown, Crest, QSC and others have been making complete lines of rail-switching amps, employing either Class G or Class H operation.

I'd like to know if the commutation noise problem has ever been solved. It's certainly present in all of the old Carver designs, but I don't have any experience with any of the later Class G or Class H amps that have been recently manufactured by some of the companies with better reputations, so I thought I'd ask if they've been able to eliminate the problem.

If anyone happens to be familiar with the problem, can hear it, and knows whether or not there are current amplifier designs that eliminate the commutation noise problem, then I'd appreciate a point in the direction of the right circuits to look at. A link to a schematic of a rail-hopping amp that has solved the commutation noise problem would be greatly appreciated. thanks.

[Steve Conner]

Douglas Self found that the commutation noise was mostly reverse recovery transients in the rail-switching diodes, and using Schottky rectifiers improved it considerably. He wrote about it in his Audio Power Amplifier Design Handbook, which contains schematics of a Class-G design that claims pretty high performance.

Schottkies and FREDs probably weren't cost-effective back when Carver was designing Class-G amps, but they have got a lot better and cheaper over the years. They would probably be a nice upgrade for an old Class-G amp. You can even get 1200V silicon carbide ones that surely must be good for some audiophile mojo.

You probably won't ever eliminate the commutation noise, but you may be able to get it small enough that it'll be psychoacoustically masked. Of course masking depends on the spectral content of program material, it's unlikely to be of any help with reggae bass lines.

On a more subtle note, masking depends on not having familiarity with the program material beforehand. When I MP3 compress a track I wrote myself, I notice all the things that the codec threw away, because I spent time putting them there. Likewise, once you've trained yourself to hear Class-G commutation noise you may never be able to un-hear it.

[Jazz P Bass]

Here is a paper from QSC's Bob Lee.
Class G uses a separate transistor bank for each voltage rail while Class H uses an FET to switch the voltage rails.
Is one quieter than the other?
I do not know.
I know the switching circuitry in Class H is complicated.
Link: QSC's Bob LeeRedirect
Maybe drop him a line see what he has to say.

[bob p]

By any chance, Steve, would you know the replacement for the Motorola/Freescale MBR5025? It's the Schottky that Self recommends in his book, though the data sheets from ON have it marked as obsolete, without a replacement.

http://www.onsemi.com/pub/docs/pcn/10465.pdf

[Steve Conner]

Schottkies are all pretty much interchangeable as far as I know. I just shop for them by voltage and current. As Self points out, the required PIV is just the difference between the low and high voltage rails, so you could try something like the MBR2045 or MBR2060. http://www.diodes.com/datasheets/ds23016.pdf

These are dual diodes sharing a common cathode. I believe it is OK to just parallel the two sections.

[bob p]

Here is a paper from QSC's Bob Lee....
Link: Redirect

so i'm looking at that, and i'm wondering: who is this Bob Lee character, anyway?



why are the PS caps center point drawn as the amp output?
why is the amp output shorted to ground?
why are collectors shorted to ground?

that diagram makes me not want to take him seriously. it seems like whoever drew that didn't know what was doing.

edit: or maybe i just don't know what he's doing!

[bob p]

thanks for the recommendations, Steve. I was having trouble finding suitable parts using the parametric search at Newark, but it seems that Mouser and Digikey are giving somewhat better results.

The problem is that the ratings/specs for schottkys are kind of tricky. they list a 10us period for the surge spec. the problem is that the MR852 100A surge in my amp is for 8.3ms. A 60Hz hump. that's a bitch of a diode. It's designed to charge the PS cap on start up.

[loudthud]

There is a different architecture being used by QSC amps. Peavey also uses it, I think they call it Valve State. The power transistor totem pole goes to ground and the output is taken from the power supply center tap. The advantage is that the whole front end can run off of +/- 15V.

Bob Lee posts on the TalkBass forum. He seems like a good guy.

[Jazz P Bass]

Hear is QSC Theory of operation on the RMX line of amps.
Class H.

[Enzo]

Even Crown. It is often called "flying rail" topology.

The easy way to look at it is by imagining there was no ground. In a conventional amp, the speaker is connected between the "output" bus of the amp, and the power supply common. That common is connected to ground. In the flying rail amp, the speaker is connected between the "output" bus of the amp and the power supply common, but the output bus is grounded instead of the power supply.

[ReadyTeddy]

One advantage of the flying rail design is that the output transistors (depending on the topology) can often be screwed directly to the heatsink, without the use of an insulating gasket. This does improve the thermal coupling of the transistors.

[Steve Conner]

Here are some disadvantages...

The interwinding capacitance of the mains transformer couples noise from the AC line directly into the output, and inserts another pole into the feedback loop.

The output stage is not an emitter follower any more, it now operates as a common-emitter amplifier with gain dependent on the load. This is simply the price you pay for running the driver circuitry off +/-15V, the output stage must have gain if the driver is to drive it fully.

[bob p]

Thanks for the help, guys. This turns out to be a day when I'm learning something new that I hadn't planned on learning.

So this is all about where you want to put the ground reference in the design. When I looked at the topology diagram, I thought that the net gain was going to amount to saving a few pennies per amp on mica insulators by bolting the transistors directly to the chassis. This does give you a better thermal junction. I'm wondering if/where the performance gain comes along with the new ground reference paradigm. What's the point in doing all of this? Is this all about cooling?

I guess it could also be about saving on parts count if you can run the drivers off of the low voltage rail, but small parts are so cheap as to be barely significant.

But then if you do this, it seems that the entire amp has to be designed with the consideration for noise rejection from the power supply. For me, that turns out to be the most interesting part of the equation, but unfortunately we don't have that part of schematic to look at. One thing that is interesting is that this amp passes the output signal directly through the PSU caps. I'm thinking that the entire amp would have to be redesigned to focus on ripple and noise rejection in the power supply. as Steve noted, AC line noise is going to be a problem. In that respect, I'd be really interested in seeing the power supply section of the schematic to see how they tackle the problem.

Steve, thanks for pointing out the load/gain problem. I hadn't gotten far enough through thinking about this to realize that. Still on the first cup of coffee this morning.

Thanks again.

[Jazz P Bass]

Full Service manual: http://www.loudandclearproductions.com/library/QSC.pdf