[SOLVED] Passive filter design for power electronics

Hello everyone,

The output LC filter design is made by knowing the desired output voltage ripple and ripple current desired (e.g. in the buck converter), which means, we do not design the filter based on its frequency domain characteristic, which is how we are been though filters are designed in analog electronics.

That LC low pass filter in the output stage of a converter is simple to make because the output is assumed to be DC, and hence, the time domain characteristics are easy to determine.
My concern is, when trying to design a filter, but the time domain is difficult to determine (unlike the output stage of a DC/DC Converter e.g. Buck), we should employ the standard filter techniques from the frequency domain.

Consider e.g. the design of an 2nd order LC low pass filter for a full bridge rectifier. If one uses the frequency response of the filter and let us say that the cut off frequency is 1/10 of the fundamental frequency, how should one select the capacitor and the inductor ? If the cap is too big, the inductor is small and then the ripple current is big and hence lot of power dissipation.

The point I am trying to make is that one should relate the filter components to the time domain response as well, e.g. overshoot, no overshoot, slow response etc..

Should one go and select components by intuition, or go and follow the procedure of the normalized filters responses (Butterworth, Chebyschev, Bessel...) ?

Any comment is appreciated !

Thank you for your time !

CataM said:

Should one go and select components by intuition, or go and follow the procedure of the normalized filters responses (Butterworth, Chebyschev, Bessel...) ?

Click to expand...

An expert told me to see them as separate LC-filter stages, and that each stage should have a cutoff frequency about 10 times higher than the previous stage.
We didn't discuss the reasoning behind this, so I can't back it up with calculations.

Google " A closer look at filtering and time - frequency- mit press. It may not be directly related to your question but you might glean some useful information from it.

Standard design procedures for passive filters have prerequisites like well defined source and load impedance. Often load impedances are variable or even time variant for power supply filters which obsoletes a design according to nominal filter prototypes.

I think you are posing the question too general. Requirements can be quite different. I would specify the operation conditions and requirements for a particular filter and use appropriate time and frequency domain analysis and design methods.

Important parameters could be:
- load impedance range
- intended attenuation of ripple frequency
- requirements for load step behavior, e.g. maximal over-/undershoot
- maximum Q

Remember the post filter should have a corner frequency which is above the crossover frequency of the SMPS....usuallythey say three times more....otherwise the smps can go unstable.

- - - Updated - - -

Basso speaks a little about this in his book.

- - - Updated - - -

Also, post filter often not needed....software engineers often mess up their code, and then blame "noise from the SMPS" in order to buy themselves time.
Its often nonsense......your SMPS output capacitor bank will often get the ripple down well enough for your application.

FvM said:

Standard design procedures for passive filters have prerequisites like well defined source and load impedance. Often load impedances are variable or even time variant for power supply filters which obsoletes a design according to nominal filter prototypes.

Click to expand...

Yes, but I was thinking that designing the filter for the worst case scenario (i.e. full load), would make it able to follow standard filter prototypes.

I would specify the operation conditions and requirements for a particular filter and use appropriate time and frequency domain analysis and design methods.

Click to expand...

Ok, so the conclusion is that normalized filter responses (Butterworth, Chebyschev etc.) could not match the imposed requirements, which leads to use both time domain and frequency domain constraints in order to build your own custom filter response.

Standard filter design methods works if you can derive source and load impedances with a least a real (resistive) component at one side. I would go for low Q, Butterworth or Bessel.