Cleaning power rail from switch mode power supply

You are on the right track.

I'll offer an improvement on the LDO and on the Pre_LDO filtering, just to left of LDO.

  • LDO should be a high-speed LDO. There are LDOs with 1ua Iddq, and their high_frequency PSRR is very poor.

  • The Pre_LDO filtering (L + R + C) must attenuate the high frequency trash/spikes/ringing, so the ADC comparators have clean VDD.

  • if the ADC is to make 1 microvolt decisions, then you need to plan on attenuating the SwitchReg high frequency trash to 1uV.

  • remember the capacitors have ESR as a limit to low_frequency attenuation, and capacitors have ESL (1nanoHenry to 10nanoHenry, plus PCB Vias at 1nH) that cause a worsening of high frequency attenuation.

  • thus you may want LRC and LRC, before the LDOv ( that is two networks, each being 2_pole) to overcome the attenuation floor of capacitors (ESR and ESL)

  • you must design the GROUND; do not share vias between capacitors, since that lets the 2 capacitors become a resonant LC filter, and you get PEAKING

  • use a wide GROUND STRIP, or even a GROUND PLANE

WARNING

Once you have designed and built a very_clean VDD, you must PROTECT IT from becoming polluted: Efields, Hfields, Ground Currents.

And magnetic shielding, such as standard thickness of copper foil --- 1.4 mils, 35 microns of the 1 ounce of copper per square foot, is

  • a few dB at 1MHz

  • a NEPER or 8.6dB at 4MHz

  • 3 NEPER or 26 dB at 40MHz

  • 10 NEPER or 86dB at 400MHz

but what about at 60Hertz powerline frequencies? ZERO

What about at the edge_speeds (10 microseconds?) of rectifier diode turn on and turn off frequencies? ZERO. Because is much slower than 1MHz.

Thus that CLEAN VDD will very vulnerable to adjacent power supplies.

How vulnerable?

We use a combined Biot_Savart and Faraday Law of Induction maths, like this

  • Vinduce = [ MU0 * MUr * Area / (2 * PI * Distance) ] * dI/dT

where we assume the loop (our rectangular region on the PCB) has Area, and that loop is coplanar with a Troublesome dI/dT from a long straight wire at Distance.

Substituting MU0 = 4 * PI * 1e-7 Henry/meter, and MUr = 1 (air, FR-4, copper) we get

  • Vinduce = [ 2e-7 * Area/Distance ] * dI/dT

Again, we want to compute a magnetic_field upset to our CLEAN_VDD. Any upset larger than 1 microvolt should be shielded against.

Now the assumptions: Distance = 1centimeter, Area = 1 centimeter by 1millimeter (this would be length of the PCB trace -- 1cm -- and height above the RETURN path for our CLEAN_VDD), and dI/dT (this is calculus derivative) of 1 ampere in 10 microSeconds as the power_supply diode current turns on at 60Hz (or 120Hz) rate.

  • Vinduce (remember, if > 1uV, its a problem)

  • Vinduce = [ 2e-7 Henry/meter * (1cm * 1mm) / 1cm ] * 1 amp/10 uSec

Now lets do the maths; notice everything is "1", so the maths is just powers of 10, and conversion factors

  • Vinduce = 2e-7 Henry/meter * 1mm * 1meter/1,000mm * 100,000 amp/second

  • Vinduce = 2e-7 * 1e-3 * 1e+5 = 2 * 10^(-7 -3 +5) = 2e-5 = 20 microVolts

Thus we have just computed, using our assumptions

  • 1cm between Power Supply and our CLEAN_VDD)

  • area of CLEAN_VDD trace (and its RETURN) is 1cm by 1mm height

  • turn on time of current thru rectifier diodes is 10 microsecond for 1 amp

That the deterministic trash induced, magnetically and at a SLOW edge of 10uS thus difficult to shield with thin copper foil (you will need steel), is

20 MICROVOLTS.

What to do? Shield it, with steel.