Power supply multi-rail noise separation?

[paulca]

I am looking for a simple, robust noise suppression circuit.

My input power for this project is a 30-36V LFP (with BMS/balancer).  I will have, 0V 15V, 30V taps.  Allowing me to provide -15V,0V,+15V for amps and also low power linear 12V and higher power switched 12V + USB.

Loads are heavy weight MCUs, 4" color TFT LCD screens and a headphone amp.    I have no worries about the source supplying enough power, at a stretch (not planned, but considered), the pack could supply the desktop speakers, replacing it's internal SWPSU.

I am considering using a series of LPF circuits to supply bulk caps, creating high impedance in both directions, leaving the spikey demand power up to the cap.

I know these are limited as they significantly limit the average current capacity and efficiency.   "Capacitor" multiplier circuits are a limited option.

I also expect having some induction in there and having a pair of LRC circuits might be wise.  I just have no idea where to start calculating.

On the RC circuit....  If I decide the most I want to lose is 1V, then I need to select the resistor that will allow the capacitor to charge fast enough that the opamps can never drain it more than 1V.

Actually.  Induction.  Induction will dynamically limit current, so if a chunk of power gets sucked out of the cap by the opamp and the supply rushes to replace it the inductor will resist the current spike and smooth it out.

Is it a "PI" circuit when you put a LC and RC circuit is series?  Is this even what I'm looking for?

Basically a narrow path for current to trickle smoothly while either side can be spiking and noisey.  Hoping to prevent parasitic noise on rails.

[aliarifat794]

Pi can provide effective noise filtering across a broader frequency range. This option sounds good to me.

[Terry Bites]

Before considering filter requirements plan the wiring. Don't add too many filter components unitll you know how big your noise problem really is.

Keep analog supply and digital supply wiring separate. The only place where analog and digital power and ground wiring should meet is at the battery or battery switch.
Use CM chokes/ filters on all digital power lines to contain the nasties at source. ie on the display and micro pcbs or as close as you can physically get.
Keep wiring loop areas small, use twisted groups (pairs)wires.

Capacitance multipliers (gyrators) made from MOSFETs introduce unacceptably high voltage drops. If you want to use them, try Darlingtons or medium power types instead.
The headphone amp will only need to supply a few 100mW max so a ZTX651 will likely be good enough. Power supply regulation for audio amps can be very relaxed.
The LDOs in fig 1 could equally be capacitance multipliers. RFI filtering of the incomming audio is wise.

Voltage regulators will help keep the spikes out. If you can afford a bit of voltage drop on your small signal supply then consider using them.
An LDO will be a good choice. Adjustable regulators can reduce noise still further by bypassing the adj pin to ground.

The batteries are huge capacitors in their own right. Adding energy storage and bypass caps may be ineffective if not located at the point of load.
Low ESR types are a must. If not, they won't act as capacitors at high frequencies. Regular eclos may not work well.

https://www.omnicalculator.com/physics/lc-filter

[jonpaul]

EMI needs to be analyzed , see the many books and seminars on EMI filter design and DM, CM noise measurement.

Considet both input and load side filters.

Off the shelf   EMI filters  are reccomend.

See coilcraft,

j

[paulca]

The main noise that plagues me in the audio band is USB/PC/Digital chatter.  It may even be from the HDMI cables.  The kind of noise that changes as you move the mouse around on the computer.  It seems that anything which connects to either the USB, or connects to the same power supply which is powering the USB hub... suffers this noise.

It's the age old USB ground noise.  The noise IS in the audio band, so it's not like I can make a filter to remove it without removing the audio within that range as well (hi hats etc.).

Additionally there is buck converter whine in there somewhere.  I expect it's the USB hub stepping the 12V down to 5V.

In an ideal world I would remove all USB power connections and only use optical links to PCs.  The world is not ideal though and there will inevitably be a USB audio source needed.  In an ideal world I would pull all my cabling out of the lab/office and reroute noisy signal cables like HMDI far away from DC power lines and USB cables.  The world is not ideal and my lab/office is a noisy environment for DC power.  The whole point of this device is to try and decouple me from it.

I suppose there "is" the option of just making a new amp with an Optical-In (TOS), a DAC and my OPAs.  Run that and that alone on the battery.  The screen showing the battery status etc can be put on a push button, when it's pressed the MCU and screen are powered the MCU boots and shows the data.  Let go and it dies, disconnects, goes away alone with it's noise.

It's an expensive battery for a DAC amp, but it would be damn silent.  My only risk then becomes RF.

[paulca]

Oh and the amp itself does and will cause controversy and fear for my hearing.

2xOPA551s for L/R.  1xOPA551 for virtual ground <- no longer needed.   Gain is 11.

It's fed with an EQ with a static +10db on the 20-100Hz band (ish) and several mid range cut bands (one in the 'boxy speaker resonating mid bass range" to roll off the base boost more aggressively, wide one around 1kHz.  The rest of the EQ is unimportant, this explains the need for gain.  I want bass.  Bass is hard to drive.  If I accept the 'Sony standard 90db or 20mA per channel", it will not have the current to actually produce anything like 90db.  In order to get a robust bass bottom end I have found you need quite a lot of power to drive a pair of 34 Ohm headphones.

I do believe I 'chickened' out when I got to the final output stage and put a small ohm resistor in series to current limit the output to about 1.5W at maximum output voltage.  I even did the bode plots and used trial and error to get values which would match the output caps.

The amp itself is silent, as the amps are obviously fixed gain x11.  If you set the input pot to 0 then the headphones are perceptibly silent.  It's only when you open the volume the noise comes through from the analogue input.  So ... it's source is almost certainly coming from the audio MCU which is being supplied with USB audio at the moment.  I tried isolators, but they just didn't work.  Couldn't establish a connection with it set to "isolated" mode.  If I switched the power rails back in, it worked, but with noise.

[paulca]

I asked chat GPT to help me design a filter.  It went straight to a basic LPF.  I ran with it anyway.  Gave it a 20Hz cut off and said I needed at most 1 watt of power.

1000uF _ 8.2KOhm.  Grand.

What about the power drop across that resistor is we assume a 0.5W load?

Only 0.48W!

I'll prompt it towards PI and LC filters next.

[MarkT]

I asked chat GPT to help me design a filter.  It went straight to a basic LPF.  I ran with it anyway.  Gave it a 20Hz cut off and said I needed at most 1 watt of power.

1000uF _ 8.2KOhm.  Grand.

Well that's obviously completely wrong at a glance....  Use a calculator for this.

[paulca]

I asked chat GPT to help me design a filter.  It went straight to a basic LPF.  I ran with it anyway.  Gave it a 20Hz cut off and said I needed at most 1 watt of power.

1000uF _ 8.2KOhm.  Grand.

Well that's obviously completely wrong at a glance....  Use a calculator for this.

In fairness that is after it aligned it to a standard resistor.

[Vovk_Z]

Recently I needed a post-SMPS filter for an output stage of a 50 W amplifier. 1 uH inductor (from an old motherboard) and 100-150 uF polymer capacitor work like a magic.

[David Hess]

A pi filter for each separate output is the way to go.  Since the filters are all starting at the same distribution point, the first capacitors are all in close proximity and in parallel so really an LC filter for each output.  A low value of resistance is placed across the inductor to quash its Q in place of a lossy ferrite core.  In the past the inductor might be wound around the low value resistor.

[paulca]

So... in the meantime (until I redesign a whole board) I have the battery and I have loads... and I have protoboard.

First test shows without filters of any kind, save an LM7815 regulator, the battery powers the current 16V single supply headphone amp... silently.  Of interest, it's also silent when connected to the rest of the apparatus.  Confirming the noise I am annoyed with is coming from/caused by/influenced by the power supply.

Next I need to see if I can power the digital audio box (STM32H7 based) from a separate LM7824.  It has it's own buck module capable of 24V and I don't want to drop too much linearly if I can avoid it.  The box pulls 200mA ish.

I might not need any filtering...  I doubt it though.  Something is going to pull all that USB ground garbage into the audio somewhere, I just know it.

Does this look sane? 


The right hand side I'm fine with, even without a darlington it will work.  It's only temporarily / testing.

On the CLC/PI filter.  I'm out of my depth.  If a capacitor multiplier is a "High school" level thing, inductors are definitely one level higher up from where I am.  I can't seem to get a straight answer it's always a cyclic this depends on that, which depends on that which depends on the original thing.  It seems to make sense of this, mathematically, we get into greek and I go blank.

It's not just impedence at a rising frequency, it's proper understanding on the impedence of the L and both Cs and how they interact.... with the variable load R throw in.  If you can get it to attenuate rising frequencies from as low as you dare, you are also likely to have to deal with the resoance of some random (calculable) higher frequency.

It's beyond me   I feel religated to stealing an example and YOLO or buying some cheap AliExpress module and hopeing it's worth bothering.

I'm presently looking for examples to pillage instead.

[T3sl4co1l]

Is that a common-ground (non isolated) module?  I guess being that it says "buck", it is.  This can introduce noise in the common-mode (between grounds) loop between equipment, and tends to be rather pernicious.  The C-mult filter can help, but you might still need a common mode choke and whatnot to keep the noise confined locally.

(Normally a buck module should be common ground, as in identical ground pins, but the fact that they often put connections on opposite ends of the board, and may have a DC offset between them besides (e.g. if low-side current sensing is used), means ground-loop EMI can be picked up, i.e. the ground path crosses the main switching loop.)

As for the filter, you typically want to design it for some cutoff frequency far enough below Fsw to be useful, and Zo a modest fraction of the DC operating conditions.  Notice that a filter only works when both ports are terminated into a resistance (or at least one for suitable prototypes), but we have no resistance in the circuit: the battery is as good as a short, or still reactive (consider stray inductance of the connecting cables), and the supply is negative-resistance if anything (current draw decreases as supply voltage increases).  We can't match to that at all, or, if we did we'd get an oscillator not a supply.

The solution is to use enough ESR (= Zo) in the capacitor(s) to dampen the filter; this necessarily costs HF response (adds zeroes to the transfer function), but stabilizes it by introducing resistance to dampen the system.  Losses can be added anywhere, really, but a deadass shunt resistor at the input or output is obviously out of the question (it would sink way more DC than the load proper!), but we can basically cap-couple it in place, blocking DC while still keeping it effective at the transition frequency (where the filter is most reactive, and therefore in need of damping).

The preferable way is to design the LC filter, and wire additional R+Cs in parallel with capacitor C, with Cs > 2.5 C and R = Zo.  This gives a zero in the transfer function, right in the middle where it's needed (for damping), then the hard-C-to-GND still gives good HF performance.

If electrolytics are used, the ESR likely comes along for free.

Tim

[paulca]

I have managed to get the noise on the scope and get some basic FFTs.

It's a right mess.  There are many audio band harmonics from the bottom up, almost equally spaced suggesting they are harmonics of the same signal?

So, while I have it on the breadboard (well when I finish work), I'm going to see if I can "fumble" a filter together.

Showing how little I have used the scope in anger in the past 6 months it took me 30 minutes of wondering why everything had a -14.50V offset and then realised you have to set the "channel" to AC coupling, not just the trigger!

Oh... and thanks for the reminder on putting a load on circuits.  I did chase the AC Coupled amplifier output up and down the DC range earlier before remembering that.  47Ohm on both output and it stopped migrating around.