SMPS Output Filtering Limitations

[jmdejoanelli]

Hi all,

Just a quick question regarding switch mode power supply output LC filtering.

What stops us from putting an arbitrarily ordered filter (say like n=5) on the output of a SMPS to remove almost all output ripple and noise? I understand that cost and power dissipation in inductors may be two of the things, but what else?

I'm trying to wrap my head around the pros and cons of using linear vs SMPS power supplies in very low noise applications. I've had a look around online, but haven't really found anything that answers my question.

Thanks!
Josef

[Signal32]

Well you could just put a linear regulator after the SMPS to get rid of virtually all the noise / ripple. But, from what I understand it's the EMI that will get you if using SMPS in very low noise applications. 
For example the Keithley 2001 7.5 digit DMM has a special transformer coated in material that reduces EMI. https://doc.xdevs.com/doc/Keithley/2002/photo/xDevs.com/K2002_1-2254.jpg
And the transformer is running at 60hz, not hundred of kHz. So, basically you'dd have to isolate the entire SMPS in a metal can and I'm not even sure if that would drop the EMI to 0.

[peter.mitchell]

output transient response.
the more inductance in the output filter the longer the current flowing takes to change, so if a sudden load is applied, the output will sag, and if a large load is removed, the output will overshoot. then there is excessive output capacitance - if there is a lot of output capacitance then the output will take too long to change relative to the controllers output which will result in sub-harmonic oscillations.

[T3sl4co1l]

Nothing. It's often done in low noise applications.  Some people boggle how spectrum analyzers can possibly even be built with digital circuitry and switching supplies in the same enclosure; these people don't know about filtering and shielding.

What you want to avoid is having a lot of poles (and therefore a lot of delay) in the control bandwidth.  What's usually done is taking AC feedback from early in the filter, so that the loop is compensated for a stable output to that node.  The subsequent filter is designed to have minimal impact on the transient response (i.e., for a maximum overshoot of dV at a load step of dI, the filter is designed for a low Q and an impedance of dV/dI).

The DC feedback can be taken at the end of the filter, to provide more accurate DC regulation, or as long as the inductors have low enough DCR (so that Imax * DCR and overshoot dV doesn't blow the Vout(min) limit for the supply design).

Or, if the fundamental ripple is tolerable (as may be the case for analog circuits, where PSRR is good at lower frequencies), then an EMI filter can be added, with a much higher cutoff frequency.  Being above the loop cutoff frequency, it has little impact on stability, while filtering the crunchy MHz stuff.

Common mode filtering, or careful design of current loop paths (perhaps including cutting ground planes, but only very carefully), may also be necessary.

Tim

[ConKbot]

If your smps doesn't mind low esr ceramics on the output, (no min esr specification) load up the output with filtering caps. I like to use the TDK component characteristic viewer to get impedance vs freq data for different ceramic caps, and make sure that the impedance is low at first few harmonics of the switching frequency. If you can keep it under 20 mohm (on paper at least), the parasitic resistance in a small converter (50 mohms in the inductor, 50-100 in the switching fets maybe more) you can end up with a reasonable attenuation, with only the single pole. You definitely need to be mindful of your layout though, as stray inductance with all those high q caps will love to ring. 

[MosherIV]

Hi

"I'm trying to wrap my head around the pros and cons of using linear vs SMPS power supplies in very low noise applications."

It depends on the application. If it is audio, either will be ok. The high freq switching noise is well above audio freq.
If the application is some kind of low voltage instrumentation, linear is best due to lower noise. The frequency bandwidth plays a big part, usually switching freq is right in the band that you want to measure.

In general, linear psu has lower noise and is less efficient. Linear can leak mains hum into surrounding circuuts, best overcome by mechanical seperation or sheilding most sensitive circuits. Linear psu is easy to design with linear regulators
SMPS is more efficient (upto 97%) but is noisier. They are smaller and lighter to equivilant linear. There many more SMPS topologies to choose from, each with different characteristics. Due to the volumne of SMPS in use now, they have become cheaper than linear. SMPS circuits can be harder to get right in design and operation. Isolation transformers for some topologies are not available off the shelf, they must be designed and built to order.