Noise floor in class AB amplifiers

[filthyone]

Hi everyone,
as the title suggests, i would like for this thread to become a place where we exchange opinions and suggestions on the subject of noise floor in class AB amplifiers and ways to reduce it.

Personaly, (i hope others too) i find the noise performance of an amplifier extremely important, far more important than say THD or output power capability. All the amplifiers i have built have some sort of noise with no input signal (usually white noise, no hum), some being better than others, as i have to get my ear very close to the tweeter. That still bothers me! I know it can be made to perfom better, but i am all out of tricks!

I am an amateur diyer with not much to offer here, but i would like to hear some general ideas (or, even better, some specific approaches) used by more experienced diyers on taming the noise floor and getting a better S/N ratio.

To get the conversation started, some topics might include power supply decoupling (i think that's an important one), IS/VAS supply RC decoupling, maybe transistor families with lower noise figures, thermal/johnson noise of critical resistors, shielding and general layout rules etc...


Thanks for any input. I hope for me and others to become smarter and to better understand the mechanisms of noise and the contribution of each mechanism to the general noise performance.

 

[Mark Johnson]

Many people measure (and listen to!) power amplifier "noise" with the input signal shorted to ground. Is that what you do?

They believe you can't possibly get a quieter input than a dead short. Therefore all AC signal(s) present at the amplifier output, must have been created by the amplifier itself.

 

[filthyone]

Yep, by saying "with no input signal" i meant to say with input shorted. Actually, as i remember from my main diy amp, the noise floor doesn't change at all wheather the input is shorted or not.

 

[xrk971]

Many of my Class AB amps are dead quiet with inputs shorted. A good way check is to connect a pair of 96dB/mW headphones (assuming DC offset not an issue). If there is hiss or hum you will hear it. If you can use your speaker amp as headphone amps without noise, you did your job well. Not all power amps can pass this test. I find using simpler topologies (fewer actives ) helps. 7 or 8 at most.

 

[DF96]

Why "AB amplifiers"? Noise largely comes from the input stage, which is likely to be the same whatever the class of the output stage.

 

[filthyone]

Headphones, with their extreme sensitivity, seem a great idea for checking noise and also checking for improvements and mods. Thanks for that!

Why "AB amplifiers"?

Mainly because that's what i exclusively have/use.
Right, so noise comes mainly from input stages... that's some info i was missing, thanks!

 

[filthyone]

I should also mention that sometimes i experience random noise that comes and goes with no apparent period, which i'm certain comes from the mains or possibly the ground. I live in a neighbourhood with lots of industrial buildings that probably "dirty up" the mains...
Any ways of getting rid of those kinds of interferences?

 

[DPH]

I find using simpler topologies (fewer actives ) helps. 7 or 8 at most.

Let's think about the case of so many extremely low noise opamps, which most of our power amplifiers share topology. Hundreds of transistors in an opamp, depending on the specific design, as it's usually better to design in a transistor than a passive from a thermal/area budget perspective. So our statement doesn't necessarily make sense when:

Noise largely comes from the input stage, which is likely to be the same whatever the class of the output stage.

In other words, manage your input stage (both the diff pair and its current sources in most topologies) well, and enjoy a very quiet amplifier.

 

[xrk971]

I don’t think the low noise opamps I would consider low noise as having hundreds of transistors. Maybe 20. Really quiet low distortion buffers like LME49600 have about 10 if you give each CCS 2 actives (Fig. 2).

http://www.ti.com/lit/ds/symlink/lme49600.pdf

Having low noise PSU’s is also important if you are chasing low noise floors.

 

[DPH]

Simplified schematic. Notice no current mirrors/etc? Easy to add tons of transistors. But, yes, I exaggerated with 100's (although depending on the process, any diode is a diode-connected transistor, etc, etc)

Anyhow, quiet input pair, oversized rails in the gain portion of a discrete amplifier and big resistors in the current sources will go a long long ways towards low noise.

 

[nigelwright7557]

Yep, by saying "with no input signal" i meant to say with input shorted. Actually, as i remember from my main diy amp, the noise floor doesn't change at all wheather the input is shorted or not.

It depends on the input impedance.
A 10k solid state input impedance is always going to be quieter than 1Meg input impedance valve amp.

 

[filthyone]

...Anyhow, quiet input pair, oversized rails in the gain portion of a discrete amplifier and big resistors in the current sources will go a long long ways towards low noise.

Well, i suppose by "oversized rails in the gain portion" you mean voltage wise and by "big resistors" you mean the physical size (the wattage). Did i get this right?

Also, how would oversized rails help with noise figures? I don't get it. Can you elaborate?

 

[DPH]

http://www.ti.com/lit/an/sboa046/sboa046.pdf

Read "noise reduction of current sources"

Hopefully someone else links another source of better theory of operation, but it's the one I found quickly.

 

[filthyone]

Ok, i did not know that about current sources. Very interesting document, i will read the rest of it soon, thanks.

Now what about the oversized rails argument? Would you like to explain?

 

[DPH]

Well, running that large resistor in the current source eats voltage headroom a lot. There's other reasons, but that's the primary one in regards to low noise.

 

[filthyone]

Ok, fair enough. Now it all comes together. Thank you very much. I will read more about current sources (i tend to always think of them as ideal, and it never occurred to me that they can be improved)

 

[DPH]

Glad to help, what little I did. Good luck out there!

 

[peranders]

I would suggest that you do a google search for "noise input stage amplifier".

When you evaluate noise performance you must have the current source impedance. A shorted input will say only that the noise will be higher in a real situation. Normally noise in power amplifiers is not a problem but if you anyway want to have it as low as possible, then you should study the design of the input stage all noise comes from this single stage.

 

[MarcelvdG]

I should also mention that sometimes i experience random noise that comes and goes with no apparent period, which i'm certain comes from the mains or possibly the ground. I live in a neighbourhood with lots of industrial buildings that probably "dirty up" the mains...
Any ways of getting rid of those kinds of interferences?

That seems to be an EMC issue rather than a circuit noise issue. The usual countermeasure is experimenting to find where the interference comes from and adding filters.

Have you tried putting a big ferrite ring around your mains cable (and preferably all other cables), close to the amplifier? If you have a mains connection with safety earth, you could try a standard mains filter.

Does the amplifier have a passive low-pass filter at the input? Is the issue still there when you place the loudspeakers very close to the amplifier and use very short loudspeaker cables?

 

[BesPav]

I would suggest that you do a google search for "noise input stage amplifier".

Why google?

Thoroughly read this:
http://www.ti.com/lit/an/slva043b/slva043b.pdf
Then this:
http://www.ti.com/lit/an/sboa015/sboa015.pdf
And this:
http://www.ti.com/lit/an/slyt174/slyt174.pdf

Then design your amps inverting with 2k input resistance. Try to achieve at least 80 dB feedback depth at 20 kHz. Try to use common base/gate everywhere.