Design analog power rails for stm32f373

[dotm]

Since the datasheet of the stm32f373 has no information about PSRR I'm quite stuck.
In my case the ADC should resolve at least 12bit @ 10ksps.
What I know (or will know after measuring tomorrow) is the ripple of my SEPIC converter for the VCC power rail at the estimated load. I also will know the lowest frequency of this ripple as well as the amplitude, thus I could estimate a resulting aliasing noise (in #LSB) and choose the correct ferrite bead to supress it.
But without PSRR ? How would you do it?

[Gallymimu]

I don't know the stm32f373, but doesn't it have a voltage reference input for the A/D reference and also an analog supply voltage input?  Normally we use a precision reference and/or a separate precision regulator to supply these sensitive analog voltages and then only worry about the PSSR of these secondary regulation devices.  I know that isn't what you want to here but you typically need to supply a voltage that is free from ripple and very low noise obviously and especially for the reference.

For anything reasonably precise we use a common mode choke on the input, as well as a differential mode LC filter set a good decade or two below the main supply ripple frequency and possibly an additional precision linear regulator or voltage reference depending.

[Ichan]

You make me ashamed, i never calculate any inductors i use  .

Agree with the last post, just use a another good ref source for the analog part.

-ichan

[dotm]

Well there is a difference in price, especially  in higher volume production. A choke should do the trick but as far as I can see it, I have to make the balancing act by not generating more noise that is caused by its serial resistance and choosing the desired attenuation. I will make some calculations this weekend and may keep this topic updated.

[solutions]

Use an external ADC if you need precision, including proper regulation, filtering and voltage references, and stop kidding yourself that you will get something for nothing.

12 bits is nothing to sneeze at, especially if it has to be accurate. If the part has a crappy analog design to it (I'm too busy to look at an unfamiliar processor design - sorry), which it sounds like it might, time to move on to one that does and stop falling in love with a small rectangular piece of plastic.

[dotm]

12 bits is nothing to sneeze at, especially if it has to be accurate. If the part has a crappy analog design to it (I'm too busy to look at an unfamiliar processor design - sorry), which it sounds like it might,

The discussed MCU has several 16 bit sigma delta ADCs on board with variable gain and offset calibration, so getting 12 bits ouf it should be possible. What did I write that gave you the impression that the analog part of it might be crappy?

[solutions]

The fact that the supplier allegedly doesn't have the balls to put the basics in a datasheet, unless it's taken care of internally and is guaranteed (likely the case). As I said, and as Gallymimu said, you're using a niche vs mainstream device, with which few have any experience/stories, which means...

"Did you ever notice it's the pioneers in the movie with the arrows in their backs?"

[bobcat1]

Hi

Simply place an LDO after your main supply (SEPIC) and you get a 70db(depend on LDO used) reduction.
The SEPIC should designed to be 300 to 400 MV above the LDO required input voltage.

All the best

Bobi