LC filter basics for SMPS

[casper.bang]

I'd like to add an LC filter after a switching power supply (based on TI's lm2596). The data sheet mentions stabilizing using a post-ripple LC filter with a 3uH inductor and a 180uF low-ESR cap. Since these parts are hard to source for me, and I noticed many various configurations of similar power supply filters (incl. Dave's) I wonder how critical these components are. The LC formula of 1/(6.28*SQRT(0.000003*0.00018)) results in a cut-off frequency of 6849Hz.

With the lm2596's switching frequency of 150KHz, why is this filter so low; there's after all quite a way from 7kHz to 150kHz.

If I can not source a 180uF low-ESR cap or a 3uF inductor, can I simply play around with the LF factor as long as it multiplies up to roughly 7kHz or are there other aspects I should be aware of?

I noticed aluminum Nichicon caps are mentioned in the datasheet for the main switching components; which manufacturers or types (tentalytes?) would the optimal very low-ESR I could use for the purpose?

Thanks,
/Casper

[codeboy2k]

The filter cutoff frequency is the -3db point, so remember that at the cutoff or Fc of the filter, approx. 7 kHz, the attenuation will be 3db (-3db from the input).   The attenuation continues at -6db  per octave or -20db per decade as the octaves/decades go up.

so at 150kHz, your switching frequency, the attenuation will be between 27 - 30 db down from where it started.  This is what you want.  This is why the filter is much lower then your switching frequency. I'd even be tempted to go lower.

You can play with the LC values to get somewhere near 7khz, it's not crucial. And I'd even go lower, down to 1Khz or so, but you should use an online LC filter tool, or an offline tool like ELSIE to figure out the values.  There are many pairs of LC values that the formula 1/(2*pi SQRT(L*C) ) = 7Khz (or 1Khz)

which values for L and C should you choose?  the answer lies with the impedance of the source and load.

You can figure the source impedance is very low, and the load impedance will also be low, since it's a power supply. So if you use an online filter tool like this one http://www.wa4dsy.net/filter/hp_lp_filter.html then you get the following, for a 1 kHz filter with a 1 Ohm source and load (the tool only goes down to 1 Ohm)



This is a 3rd order filter. It's a good one.  It will attenuate down 60db at 150Khz.  The L's are 160uH and the C can be 330uF.

If you only want to use a 2nd order filter, i.e. a single L followed by a C to ground, then choose 225uH and 220uF.

The datasheet's choice of 3uH and 180uF is probably more to do with trading off size and an acceptable  25-30db attenuation at 150kHz. More attenuation requires either moving the 3db point to a lower frequency, or choosing a higher order filter. But choosing a higher order filter just to use small parts doesnt help if you need more of them; it just takes up more board real-estate. Furthermore, depending on the real voltage level of the ripple, it may be the case that 25-30db attenuation is actually all that is needed, at which point a 7Khz 2nd order filter of 3uH and 180uF is good enough. How and where you trade off these issues is dependent on your design goals.

As for brands, well you only have to watch a few teardown videos here to know which ones are good and which are bad. Basically, everyone likes Panasonic,Rubycon, Sanyo, Nichicon, Nippon, TDK, and United Chemi-Con.  There are probably others, but that's a good list.

[casper.bang]

What an great and detailed reply. I think I'll be borrowing a scope at work to try and see these filters do their job. Much appreciated!

[smashedProton]

You may want to lower the cutoff frequency past 60 if you have problems with mains hum.  But that requires a huge ass inductor...

www.GarrettBaldwin.com

[Harvs]

Remember to include the cap esr in your calcs. That's why they stipulated that in the filter requirements, can make a significant difference to smps filters.

[codeboy2k]

What an great and detailed reply. I think I'll be borrowing a scope at work to try and see these filters do their job. Much appreciated!

You're welcome. Yes, definitively build 'em out and, analyze them and see how they perform, you get a better "feel" for it all after that.

[casper.bang]

Remember to include the cap esr in your calcs. That's why they stipulated that in the filter requirements, can make a significant difference to smps filters.

That ESR thing confuses me. Some spec sheets (digikey) mentions ESR's around 160 while some suggest using values of 10-20 for SMPS.

You're welcome. Yes, definitively build 'em out and, analyze them and see how they perform, you get a better "feel" for it all after that.

I borrowed a Rigol DS1102CD from the hardware guys at work, but I find it rather hard to see the small ripple (noisy and only 234 vertical pixels). I'll try to go through the oscilloscope tutorial here on the forum over the next couple of days; I may be missing something (filter, average or something).