whats the quietest switchmode power supply you have seen?

[SArepairman]

So what is the quietest switchmode power supply that you have seen in terms of Vpp and Vrms? Typically I see that linear bench power supplies have much better figures, but I have never worked with a switchmode -> linear hybrid supply.

 I am curious about just what the limits are.

[nickm]

You can get them down into the noise of the scope, <1mVpp with the right component values using the basic buck topology.  If you go crazy with the filtering you can get them even quieter.  The biggest obstacles are layout and measurement techniques.

[JoeN]

How would you go about measuring it with more precision than a normal oscilloscope?  My understanding is that most DSO scopes have about 8-10 bits of precision.  With 24bit ADCs over 1MSPS it would be easy enough to make an instrument to really look at the noise finely down to microvolts, wouldn't it?  Just not high-frequency noise like a scope allows you to look at.  Does any equipment manufacturer make something like this or would it be a one-off project?

[SArepairman]

How would you go about measuring it with more precision than a normal oscilloscope?  My understanding is that most DSO scopes have about 8-10 bits of precision.  With 24bit ADCs over 1MSPS it would be easy enough to make an instrument to really look at the noise finely down to microvolts, wouldn't it?  Just not high-frequency noise like a scope allows you to look at.  Does any equipment manufacturer make something like this or would it be a one-off project?

you use a pre amplifier specially built for testing noise.

[T3sl4co1l]

Arbitrary.  Not good enough?  More shielding, filtering, ferrite beads, LDOs.  I know of a electro-optics guy who's achieved nanovolt noise levels from noisy power rails.  The challenge is not if it can be done, but how many stages you can spare for the required attenuation.

Tim

[JoeN]

How would you go about measuring it with more precision than a normal oscilloscope?  My understanding is that most DSO scopes have about 8-10 bits of precision.  With 24bit ADCs over 1MSPS it would be easy enough to make an instrument to really look at the noise finely down to microvolts, wouldn't it?  Just not high-frequency noise like a scope allows you to look at.  Does any equipment manufacturer make something like this or would it be a one-off project?

you use a pre amplifier specially built for testing noise.

Can you suggest one or a link?  I am not familiar with this technique.  It would be appreciated if you can think of one without much trouble.

[David Hess]

How would you go about measuring it with more precision than a normal oscilloscope?  My understanding is that most DSO scopes have about 8-10 bits of precision.  With 24bit ADCs over 1MSPS it would be easy enough to make an instrument to really look at the noise finely down to microvolts, wouldn't it?  Just not high-frequency noise like a scope allows you to look at.  Does any equipment manufacturer make something like this or would it be a one-off project?

you use a pre amplifier specially built for testing noise.

Can you suggest one or a link?  I am not familiar with this technique.  It would be appreciated if you can think of one without much trouble.

Jim Williams as usually wrote on this specific subject:

http://www.linear.com/docs/4159

And he even made a video:



When I have done this in the past during design, I have just built a low noise amplifier on the spot or included one in the power supply.

[chickenHeadKnob]

How would you go about measuring it with more precision than a normal oscilloscope?  My understanding is that most DSO scopes have about 8-10 bits of precision.  With 24bit ADCs over 1MSPS it would be easy enough to make an instrument to really look at the noise finely down to microvolts, wouldn't it?  Just not high-frequency noise like a scope allows you to look at.  Does any equipment manufacturer make something like this or would it be a one-off project?

which 24 bit ADC can do 1 Msps?

[AndyC_772]

The problem with trying to measure noise on a switched mode supply is that it's often common mode, so what you measure with a scope (or any other instrument) depends greatly on exactly how you measure it.

It's quite easy to reduce differential mode noise - the variation in voltage between +V and GND at some particular physical location - down to tiny levels, and often that's all that's needed. A component placed at that location and connected to that supply won't 'see' any significant noise.

But: connect a scope probe's ground lead to some part of the ground plane and start probing, and you'll see switching spikes. They may not affect the operation of your circuit, because what you see on the scope is just an indication of the fact that 'ground' at your PCB is not the same as 'ground' at the scope.

Thank the inductance of the scope probe's ground lead for that; Tektronix have a really good app note on probing technique which is well worth a read.

With that in mind, it's essential to specify exactly what you mean when you say a power supply is 'quiet'. Are you specifically looking for a very constant voltage between +V and GND at some prescribed location, or do you also require that both those rails are 'quiet' with respect to some third point (such as the ground at your scope, which is in turn connected to mains earth via a highly inductive string of power cables).

Common mode noise is more often than not a problem during EMC testing, because no amount of filtering between the output and (local) 'ground' is going to help. For what it's worth, the quietest commercial power supply I've come across in that respect is - believe it or not - the mains brick that came with my laptop. It's a 180W unit made by FSP, and we found it to be far quieter than there was any real need for it to be. A credit to its designers.

[David Hess]

How would you go about measuring it with more precision than a normal oscilloscope?  My understanding is that most DSO scopes have about 8-10 bits of precision.  With 24bit ADCs over 1MSPS it would be easy enough to make an instrument to really look at the noise finely down to microvolts, wouldn't it?  Just not high-frequency noise like a scope allows you to look at.  Does any equipment manufacturer make something like this or would it be a one-off project?

which 24 bit ADC can do 1 Msps?

Analog Devices makes one but noise and distortion limit it to about 16 useful bits:

http://www.analog.com/en/analog-to-digital-converters/ad-converters/ad7760/products/product.html

Texas Instruments has a similar converter which does a little better.

The way to go about this is just to use a preamplifier and normal oscillscope.

[chickenHeadKnob]

which 24 bit ADC can do 1 Msps?

Analog Devices makes one but noise and distortion limit it to about 16 useful bits:

http://www.analog.com/en/analog-to-digital-converters/ad-converters/ad7760/products/product.html

Texas Instruments has a similar converter which does a little better.

The way to go about this is just to use a preamplifier and normal oscillscope.

Had no idea delta sigma could be made to run that fast. TI device you talk about looks like ADS1675

$22 to $17 for 1 K price
575mW hot basterd!
93db signal to noise 103 db dynamic range, so yeah,  more like 17 to 18 bits
still those are unicorn specs, allmost.

Thanks

[BravoV]

Arbitrary.  Not good enough?  More shielding, filtering, ferrite beads, LDOs.  I know of a electro-optics guy who's achieved nanovolt noise levels from noisy power rails.  The challenge is not if it can be done, but how many stages you can spare for the required attenuation.

Tim

Just curious how is the regulation loop with those so many stages ? I mean like the transient response when driving a "dynamic" load ?

[AndyC_772]

Awful, and irrelevant, probably.

If you quieten down a power supply by adding filter components in series, then by definition you're putting some impedance in between the control loop and the load. Any dynamic current drawn by the load is going to result in a voltage change at the output of the PSU.

It's an important point to make to anyone who thinks they want an ultimately "quiet" power supply:

Quiet where exactly?

Even if you had an "ideal" power supply, as soon as it's connected to a load with a wire, there's an inductor and a resistor in series with it. If your load then draws a current pulse, then there *will* be a voltage dip across the load, and as soon as the current pulse stops being drawn, there *will* be a corresponding spike.

Non-steady loads always require local filtering, and if you have local filtering, it's highly debatable whether there is any benefit to including "heroic" filtering at the PSU itself.

[David Hess]

Filtering out the high frequency switching noise just takes a lossy ferrite bead or even the inductance from a couple inches of circuit board trace but switching frequency ripple needs a real LC filter.  Using a low noise design to start with is the most effective option though.

At low currents, RC filters work great.  Adding extra filtering at the point of load also serves to isolate noise between sections.

[T3sl4co1l]

Transient response need not be bad after heavy filtering, nor the voltage drop, even with multiple stages.  Keep the impedance low (relatively small Ls, big Cs) and well damped (use electrolytics or tantalums for the bulk, rather than polymer or ceramic).  Clean up high frequencies with ferrite beads and ceramics.

Speaking of impedances (from another thread ), power supply network design is a worthy, oft-forgotten subject.  Which, like so many things, is rife with poor understanding, crappy rules of thumb, and outright misinformation.

Tim