Noisy phantom rail...?

[Samuel Groner]

What was THAT all about?

With the addition of the capacitor you reduce the noise gain of the internal amplifier at high frequencies to unity. This is mostly notable for high output voltages where the voltage reference noise get's amplified much.

I reckon they will all end up getting some components tacked on to them...

While you're at it you might add a 36 V/5 W protection zener to make the regulators survive a potential voltage drop above their voltage rating.

Samuel

 

[Svart]

hmm I've noticed hash on my cards as well.. I'll give this a try. what value cap are you using?

 

[NewYorkDave]

If you're doing a retrofit to existing boards to get rid of the hash--and space and ease of installation are considerations--I would think that using a ceramic or film to bypass the ADJ pin should be adequate. If I'm understanding the problem correctly you only need a HF bypass to fix it.

As a (completely arbitrary) starting point, try a capacitor whose reactance at 1kHz is 1/10th or less the value of the lower voltage divider resistor.

 

[SSLtech]

[quote author="Svart"]hmm I've noticed hash on my cards as well.. [/quote]
...step away from the pipe...

:twisted:

I hit it with the big guns. 10µF "killed it to death". Maybe I'll try 1µF, or 0.1 or whatever, but for now I'll just make a mental note to provide for a capacitor across there. If the space is big enough, people can put any value they choose in there between 0.1 ceramic and 10µF electrolytic.

For now, I'm moving on from chasing this issue, as I came across it as a 'kink' in a design/build project that I'm deep into, which has a hard deadline in a few days. -The noise was killing me, since this project most assuredly requires a dynamic range well in excess of 100dB, and half a volt of white noise was starting to find wasy of creeping in in tiny amountsm where it DEFINATELY wasn't welcome!

-It's all good now though! The board may have space for both an electrolytic AND a ceramic, because the electroltic DEFINATELY does improve ripple rejection (during start-up) and gives the rail a pleasant 'soft-start' behaviour, whereas the ceramic will help to do the job of nailing any HF to the floor.[quote="JohnRoberts]When you removed power and unregulated rail decayed, the regulator output pegged full on and was quiet because it was no longer trying to regulate to it's noisy internal reference.[/quote]
Yes, I see that now. There was 8V margin, which decays VERY quickly since it's the first bit to drop off. It gave the illusion of being instantaneous in fact, when under load. -More reservoir would likely have reduced this illusion.

Keith

 

[Samuel Groner]

Perhaps we should clarify about what order of magnitude we're talking about--using a TL783 (without bypass capacitor) on my designs noise is just visible on the 5 mV scope setting, perhaps 3 mVpp? If its way more something else is wrong I'd say.

[Edit--posts crossed: Half a volt is really too much and should not be there even without bypass cap...]

Samuel

 

[Svart]

I have a lot of 1uf film caps, I'll try one of those..

 

[BYacey]

I've run into this before as well; the regulator is actually a high performance high speed amplifier. Putting too big a cap on the output slows the regulation down in a drastic manner but does indeed get rid of the hash from the gain stages in the regulator. For phantom power this probably isn't a concern because the load current is more or less steady, and the final phantom voltage isn't critical as long as it isn't wobbling around.

 

[Guest]

Here you go again:

[quote author="Wavebourn"]Is it's output shunted as close as possible to the chip by a tantalum cap?[/quote]

Tantalum there is what the Doctor prescribed.

 

[SSLtech]

[quote author="SSLtech"]And I tried clipping tantalums on as well. [/quote]

The surgeon had already performed that operation.

Keef

 

[Guest]

The Doctor prescribed as well 100 Ohm / 100 uF x 63V near each phantom powered pre (better before a switch, each cap grounded near the corresponding input. :grin:

 

[craigb]

This may have been posted before, but this site has some interesting info on reg. noise. Page 2 is specifically about the 317/337.

http://www.tnt-audio.com/clinica/regulators_noise1_e.html

 

[Samuel Groner]

Yep, seen that before. However, if the noise is really 0.5 V as SSLTech says it's not a noise problem but a stability problem. Proper grounding and layout might help more than the addition of capacitors--though this surely helps as well.

Samuel

 

[JohnRoberts]

I'm a little too lazy to dig up the schematic to make a 317 do 48V phantom but if it directly regulating it with a simple resistor divider the noise of the 317 Vref will be scaled up by the ratio of the two resistors which could be quite a bit. Since the regulators are optimizing other things than noise there could be noise voltage associated with reference voltage, and noise current associated with that reference port.

At least one of the data sheets shows a cap at the port to ground.

JR

 

[Svart]

I don't recall ever getting this kind of noise with lm317s before though. I think it's likely centered around "floating" the part via resistor to get it to take 50v+ on the input.

 

[Guest]

[quote author="Svart"]I don't recall ever getting this kind of noise with lm317s before though. I think it's likely centered around "floating" the part via resistor to get it to take 50v+ on the input.[/quote]

It may be too high voltage between input and reg (-, or ground) pins causing undesired Zener effect.

Anyway, a tantalum cap between output and reg (ground) pins always helps me.

Also, I used an emitter follower before LM317 with Zener to it's base from - pin to limit input voltage when I used them for +300V regulators. I used as well 78xx regulators that was in a bin appropriately selecting feedback resistors, no problems at all with noises / oscillations.

 

[JohnRoberts]

Output noise in 317 is specified as a % of output voltage (.003%). It is indeed scaled up with output voltage as you would expect from the topology. They specifiy the noise as RMS and only typical, not max. So the typical 150 mV rms isn't much of a stretch to .5V p-p. OTOH If it's .5V rms then it's some 10dB worse than typical and perhaps a concern.

JR

 

[SSLtech]

[quote author="Wavebourn"]The Doctor prescribed as well 100 Ohm / 100 uF x 63V near each phantom powered pre (better before a switch, each cap grounded near the corresponding input. :grin:[/quote]

Also already done. -In fact configured as a 2-stage R-C-R-C.

However, as I mentioned at the beginning, an ounce of prevention is worth a pound of cure.

Generating an amount of noise which is deemed to be too much, and then hoping that it can be filtered it sufficiently is different from NOT generating noise, and then adding a filter for belt-and-braces protection.

There are no phantom on-off switches, this is not a typical application... It's just a 48V low-current supply which has to be quiet.

[quote author="John Roberts"]the typical 150 mV rms isn't much of a stretch to .5V p-p. OTOH If it's .5V rms then it's some 10dB worse than typical and perhaps a concern.[/quote]
I quoted the number from the 'height' of the fuzz on my scope, AC coupled, probing the regulator output. At 500mV/division, the line was about 1 division tall, so yes I was really looking at peak-peak. It was a little dimmer at the upper and lower extremities of course, but definately there.

Keith

 

[bcarso]

A rough rule-of-thumb used when eyeballing noise on a scope is that rms is about 1/5 of peak-to-peak. So it sounds as if you are within spec.

 

[Dan Kennedy]

The 317 needs the adjust pin bypassed to be adequately quiet, as you've discovered. In a 48V circuit it also needs the protection diodes both from input to output and from adjust to output. It also needs a zener, as Samuel recommended from input to output to limit the voltage across the device.

The zener needs to withstand the short circuit current, if the output is directly connected to the load, in the case of a phantom regulator, a 1W is sufficient.

It makes a simple 317 circuit a bit partsy, but they're all cheap and cheerful parts.

 

[Samuel Groner]

So the typical 150 mVrms isn't much of a stretch to 0.5 Vpp. OTOH If it's 0.5 Vrms then it's some 10 dB worse than typical and perhaps a concern.

John, you might have forgotten to scale your figure by 100 (that's the trick with the %)--Matlab says it should be 1.44 mVrms typical for a 48 V output. This is consistent with the TL783 noise figure I posted upthread and which I have consistently measured on several different 48 V implementations. The LM317 has the same typical noise specification as the TL783 so I still insist on things beeing rather a stability problem than inherent noise. If it's solved with a capacitor--OK, why not (there are reasons why it could improve stability) but I'd recommend checking the layout and grounding as well, at least for future implementations.

Samuel