Power supply LC filter degrades analog to digital part noise rejection.

[Georgy.Moshkin]

I am observing interesting effect: LC filter on power supply output worsens noise rejection performance between modules. Main power module (A) provides filtered output at 8v through 100uH + 1000uF LC filter. Two secondary modules (B) and (C) have 5V/3.3V LDOs and 100uH+68uF / 100uH+100uF input LC filters. Here is interesting observation:

case 1 - unmodified
(A) power module (LC on output)
(B) analog module (opamps, etc., LC on input)
(C) digital module (MCU, I2S ADC, etc., LC on input)
digital module (C) contaminates analog module (B) with noise (0-10kHz).
Noise starts right after MCU boot and initialization sequences are complete.

case 2 - removed 100uH inductor from module (A):
noise levels in analog part opamp preamplifier output are lowered by 20-30dB, very good performance

My hypothesis is that large inductor between 8v linear regulator and digital module is "slowing down" regulation, so digital module current spikes are larger at LC input of digital module. When 8V linear regulator output inductor is removed, 8V regulation becomes "fast" enough to deal with currents at LC input of digital module.
Maybe another explanation?

 

[Georgy.Moshkin]

Well, it seems that putting common inductor before two separate LC filters is a bad idea:
See how shorting inductor improves noise isolation between two outputs.

here is circuit (copy and paste https://falstad.com/circuit/) :

$ 1 0.00000125 11.086722712598126 50 5 43
g 112 272 112 304 0
w 112 208 112 144 0
w 112 144 224 144 0
l 224 144 320 144 0 0.0001 -0.19274470773201668
c 320 144 320 224 0 0.0001 5
g 320 224 320 272 0
l 432 144 528 144 0 0.0001 -0.28718527151876433
c 528 144 528 240 0 0.0001 4.830348764348988
g 528 240 528 288 0
l 464 336 528 336 0 0.0001 0.16184398959761928
c 528 336 528 400 0 0.0001 4.975707472825851
w 464 336 416 336 0
w 416 336 416 144 0
w 416 144 432 144 0
w 416 144 320 144 0
g 528 400 528 448 0
r 608 144 608 208 0 30
r 608 336 608 384 0 30
w 528 336 608 336 0
w 528 144 608 144 0
w 608 208 608 240 0
w 528 240 608 240 0
w 608 384 608 400 0
w 528 400 608 400 0
v 640 400 640 448 0 1 1000 5 0 0 0.5
g 640 448 640 480 0
w 640 400 640 368 0
c 640 368 640 288 0 0.00001 -1.7348790072879412
w 640 288 640 144 0
w 640 144 608 144 0
s 240 96 304 96 0 0 false
w 304 96 320 96 0
w 320 96 320 144 0
w 240 96 224 96 0
w 224 96 224 144 0
v 112 272 112 208 0 0 40 5 0 0 0.5
o 19 64 0 4099 10 1.6 0 2 19 3
o 18 64 0 4099 10 0.4 1 2 18 3
o 14 64 0 4099 20 6.4 2 2 14 3

 

[dick_freebird]

It's likely that the 1000uF cap of the power supply LC
filter has poor ESL and is no help for HF conducted noise.

Meanwhile the 100uH inductor prevents the power
supply's low output impedance from helping take
HF energy. It may also impose series resistance that
is significant enough to convert current noise to a
significant voltage noise "bonus".

Before declaring defeat you might try adding some
1uF, 0.1uF, 100pF ceramics chosen for good ESR/ESL, to
help out the bulk capacitor.

You might also, if you insist on a 2-stage filtering, make
the "digital" and the "analog" supplies use separate
paths all the way back to the PSU output -and ground-.
People think about decoupling as giving the current
spikes somewhere to go. But where do they go? Into
local ground. With some overspray. If the digital and
analog share a common ground (with only one PS
output filter, mustn't they?) the spikes dumped into
common ground will bump the analog supplies,
communicating through -their- decoupling caps.

Seen this in DC-DC chips, SW-node ground kick corrupts
the VIN and bothers current sense compare (falling edge
jitter).

 

[KlausST]

Hi,

For good assistance we need good informations first.

* wiring scheme (to detect GND loops and antennas...
* exact filter part names or links to datasheets
* information about HF sources (maybe it's a resonance)
* it seems you have an ADC: schematic, analog path. Maybe you are destroying reference signals with your filters.
* informations on how you obtained the values and what they mean. Sampling rate, resolution,
* and so on...

Klaus

 

[Georgy.Moshkin]

Inductance after linear regulator worsens performance. If inductance is shorted with wire, noise leakage is reduced:

 

[KlausST]

Hi,

None of the requested informations....

In your circuit, left to the red arrow, there is an "undamped node", just connected to L and C. It may cause a huge resonance.

But without good informations, it just guessing...


Klaus

 

[Georgy.Moshkin]

Narrowed it down to very small piece of board. In the end there is only two things left: powered opAmp first stage, MCU running at 180MHz, without ADC. Noise is measured using oscilloscope on opAmp output. I also removed power adapter and use 6x1.5v batteries. Shorting inductor clears noise out. Because power board is removed, for experiments use my own LC filter, some unshielded surface mount type 100uH, and capacitor is tantalum 100uF.

Let's focus on this inductor. Now I think we can't put any inductors after power supply if there are multiple power consumers (edit: if power consumers have input LC filters)

 

[Georgy.Moshkin]

None of the requested informations....

I will provide proof of concept board photos and schematic later.