Power Filtering Stage Review

[benwis]

Hi,
A few of you might remember this conversation:
https://www.eevblog.com/forum/projects/reducing-noise-for-24-bit-adc-weight-sensor/

As it turns out the existing stock is too valuable not to rework, so I get to design a new power stage to hopefully fix the issues.
I'm trying to get a clean(100mVpp ripple or less, preferably much less) 5V or 3.3V source to use for a 24 bit ADC. I've got 36V input coming out of a MeanWell PWM-36-90 power supply, fed into a V36SE05010NRFA 36 to 5V converter. I've added a 2 stage LR filter and ferrite beads to the input of the 36V to 5V converter, and the secondary LT1763 linear regulator stage, in addition to a 2 stage CR filter if necessary. And I added the recommended caps to the input and output of all the regulators. I found this http://cds.linear.com/docs/en/application-note/an101f.pdf, which recommends the beads.
Schematic Below:


What do you guys think?

[DBecker]

The PSRR of that regulator is listed as 50db minimum, 65db typical.

It's quite easy to surpass the figure with other regulators.  You should be able to get up to 90db with the right part and careful design.

Consider running two LDO regulators in series.  The first one can be cheap and low-spec, with the second having a low noise figure.  Doing this will allow dropping the size of the switching regulator output capacitor and eliminating the inductor, both of which will reduce radiated noise.

[benwis]

The PSRR of that regulator is listed as 50db minimum, 65db typical.

It's quite easy to surpass the figure with other regulators.  You should be able to get up to 90db with the right part and careful design.

Consider running two LDO regulators in series.  The first one can be cheap and low-spec, with the second having a low noise figure.  Doing this will allow dropping the size of the switching regulator output capacitor and eliminating the inductor, both of which will reduce radiated noise.

I think you're right about the LDO's PSRR. I could insert the LT3045 in place of the LT1763, and get 20-30dBish higher PSRR over the range. I don't think I can get rid of the output caps on the 5V side of the switcher, as I still have to power ~10A at 5V for other devices. I could parallel another LDO before the LT3045, just to get a lower PSRR, although there isn't a ton of voltage headroom...

EDIT: The MIC94310 is cheap and offers >50db PSRR from 10Hz to 10MHz

[whollender]

If you're looking for both 5V and 3.3V to be clean, you'll want some filtering after the 36-5 supply.  Better to get ~6V, then use 1 LDO for 5V, and 1 for 3.3V.

Some things to watch out for when looking at the filtering:

• Design the filter to reject starting around where the PSRR for the following LDOs starts to drop, or at least so that the main switching frequency and a few decades above have high rejection.
• Try to keep the filter as close to the source of the noise (switching supply) as possible, and keep the distances between the filter components small to minimize circulating currents to prevent the noise from radiating
• Make sure that the filter is well damped, or that the supply feeding it has a slow start option, so that the voltage overshoot on startup doesn't kill whatever's after the filter.

[benwis]

If you're looking for both 5V and 3.3V to be clean, you'll want some filtering after the 36-5 supply.  Better to get ~6V, then use 1 LDO for 5V, and 1 for 3.3V.

Some things to watch out for when looking at the filtering:

• Design the filter to reject starting around where the PSRR for the following LDOs starts to drop, or at least so that the main switching frequency and a few decades above have high rejection.
• Try to keep the filter as close to the source of the noise (switching supply) as possible, and keep the distances between the filter components small to minimize circulating currents to prevent the noise from radiating
• Make sure that the filter is well damped, or that the supply feeding it has a slow start option, so that the voltage overshoot on startup doesn't kill whatever's after the filter.

Thanks for the advice. I'm going to have to settle for "clean enough" on the 5V, as I don't have space for a linear LDO that can handle 10A. The input filter's corner frequency is 1kHz, and the switching frequency of the 36 to 5V switcher is 580kHz. I might have to reduce that somewhat, as the inductors and capacitors are pretty big. Maybe I'll bring that up to 5kHz and see if I can't reduce those sizes.

I don't think the PSU has a slow start option, can you provide any resources on how to properly damp an LC filter?

[DBecker]

Oh, I didn't realize that the 5V rail was providing 10A.  I though you had just stuck a bigger capacitor there in a shotgun attempt to reduce ripple.

With that power level you may have a few other sources radiating noise.  It's worth looking at the board and thinking in 3D about the cylindrical antennas (capacitors) and EM projectors (inductors) on the board.

The best way to avoid LC overshoot on the power supply filters is to avoid the L.  An inductor kills the transient response and can even destabilize a regulator.

[benwis]

Oh, I didn't realize that the 5V rail was providing 10A.  I though you had just stuck a bigger capacitor there in a shotgun attempt to reduce ripple.

With that power level you may have a few other sources radiating noise.  It's worth looking at the board and thinking in 3D about the cylindrical antennas (capacitors) and EM projectors (inductors) on the board.

The best way to avoid LC overshoot on the power supply filters is to avoid the L.  An inductor kills the transient response and can even destabilize a regulator.

That does seem like the best way I was trying to avoid the drop of an RC filter, but I guess I can live with losing a volt or two to get rid of these
pesky inductors. I simulated the filter in the schematic above(C is supposed to be 330u), and It goes all the way up to 77V!

I wouldn't be surprised if there are other noise sources. Most of the 5V power is going to charge a phone and to power about 2 meters of LED Strip lighting, which don't seem like they'd produce too much noise. Still, the board layout can use some work in that area.

I guess I'll just put in another 2 stage RC circuit, but to keep the R down I have to have a big cap of ~100uF. I thought that electrolytic caps performed poorly for higher frequencies, is there an alternative type that can handle 36V and performs well across the frequency range. I found some Ceramics, but they're 20 bucks each!

[whollender]

Oh, I didn't realize that the 5V rail was providing 10A.  I though you had just stuck a bigger capacitor there in a shotgun attempt to reduce ripple.

With that power level you may have a few other sources radiating noise.  It's worth looking at the board and thinking in 3D about the cylindrical antennas (capacitors) and EM projectors (inductors) on the board.

The best way to avoid LC overshoot on the power supply filters is to avoid the L.  An inductor kills the transient response and can even destabilize a regulator.

That does seem like the best way I was trying to avoid the drop of an RC filter, but I guess I can live with losing a volt or two to get rid of these
pesky inductors. I simulated the filter in the schematic above(C is supposed to be 330u), and It goes all the way up to 77V!

I wouldn't be surprised if there are other noise sources. Most of the 5V power is going to charge a phone and to power about 2 meters of LED Strip lighting, which don't seem like they'd produce too much noise. Still, the board layout can use some work in that area.

I guess I'll just put in another 2 stage RC circuit, but to keep the R down I have to have a big cap of ~100uF. I thought that electrolytic caps performed poorly for higher frequencies, is there an alternative type that can handle 36V and performs well across the frequency range. I found some Ceramics, but they're 20 bucks each!

If you aren't terribly worried about the noise on the +5V line, then just add the filter to the input of the 3.3V LDO.  You will need to be careful to make sure that there is plenty of capacitance at the input pin of the LDO, however, to make sure that it stays happy.  This kind of low current LDO generally only needs 10-20uF at the input to keep it stable and provide good dynamic performance.

[benwis]

There's a bit less board rework if I can get a clean enough 5V line, which is why I stuck it before the 36 to 5V switcher. Here's an updated schematic. I've replaced the LC filter with a 2 stage RC, and the LT1763 with a cheap MIC LDO and a LT3043. I'm hoping that'll do it.