Based on your 50dB @ 50kHz filter. This would be a multi stage differential filter approach for 300R load. It has some advantages in that the lower inductances would result in smaller (maybe off the shelf) parts but the disadvantage of more of them. As the inductance is smaller you would also benifit with higher SRF making it more effective for high frequency EMI for which a large inductor would be very poor due to its low SRF.

This type of filter would still have resonances that would require damping (look up series LR and shunt DQ damping) but due to the small values of capacitor would be light loading on your supply and achieve the levels of attenuation your targeting.

The capacitors would ideally be of Film type.

 

The response of your 4-pole filter looks very nice at 300 ohm load, but how often will the load be 300 ohms?
It will look like a complete mess at any other load impedance.
Stick to a 2nd-order filter. The Q will vary as the load impedance changes, so keep the filter resonance well away from both the output frequency and the PWM frequency, by choosing the filter turnover to be the geometric mean of the sampling frequency and the output frequency.

f(filter) = SQR(fPWM * foutput)

That puts your filter frequency around 1kHz, 5 or 6 octaves below your PWM frequency, so the PWM will be attenuated by 60 to 70dB

Don't just stick in random inductors, make sure you know the inductance, and that it won't saturate at the operating current.
Micrometals has an on-line inductor designer. Use it. Don't guess.
Use toroids with single-layer windings to keep the intrawinding capacitance low.

 

The response of your 4-pole filter looks very nice at 300 ohm load, but how often will the load be 300 ohms?
It will look like a complete mess at any other load impedance.
Stick to a 2nd-order filter. The Q will vary as the load impedance changes, so keep the filter resonance well away from both the output frequency and the PWM frequency, by choosing the filter turnover to be the geometric mean of the sampling frequency and the output frequency.

f(filter) = SQR(fPWM * foutput)

That puts your filter frequency around 1kHz, 5 or 6 octaves below your PWM frequency, so the PWM will be attenuated by 60 to 70dB

Don't just stick in random inductors, make sure you know the inductance, and that it won't saturate at the operating current.
Micrometals has an on-line inductor designer. Use it. Don't guess.
Use toroids with single-layer windings to keep the intrawinding capacitance low.

Its the peak load so not very often, it would need some damping as i suggested but I'm happy to be corrected if you think it that wouldn't work once the resonant peaks have been addressed. I only suggested it as the physical size of single stage may be rather large and require bespoke parts whereas the 4 pole may fall into COTS, it may not but hey ho thats life. For sure the 2nd Order is the simpler option.

I'll have to take a look at the MicroMetals inductor designer myself as i'm not familiar with it. So thanks for this.

Marc

 

Looking back at his circuit, although I’m not sure what Schmitt trigger U3a does, the system appears to be open-loop, so you are indeed correct that a four pole filter could be used, as there is no feedback loop, provided that both pole-pairs are kept well away from PWM or output frequencies. As Q increases at low load, the filter response won’t look too clever, but will still work!

As for availability of parts, you can always get micrometals cores and you can always get wire, though winding a big toroid might make your thumbs ache.

Micrometals used to have a clunky DOS program which you could download, which gave you half a dozen suggestions to fit the design, now they have a fancy on-line version which gives several thousand. . . .I think I preferred the old one.