Power and 3.3V rails.

[paulca]

Most of my MCU circuits to date have been using in the order of a few 100mW.  For those circuits the MCU module onboard regulator has surficed for 90% of things.  Else ways, you just through any jelly bean 3.3V regulator on there and as the most jelly bean linear 3.3V regulators "claim" 800mA to 1Amp, you don't need to worry much about the power.

The trouble I am now facing is that my MCU dev board is pulling half an amp at 3.3V.  This has identified that the claimed 880mA at 12V the AS1117 claims, is assuming it has adequate heat sinking etc.  It IS providing the 3.3V at 500mA from a 5.8V supply, but it's doing it while at maximum package temperature of 80*C.... and it's my third in the this project.  The first one died due to inrush current.  The second one (on a module) died from simply overheating and releasing it's magic smoke.

Now, first approach.  "Bigger hammer".  I ordered a few LM317T 3.3V regulators in To-220 package.  I have some TO220 heatsinks.  The 317 claims 2.2A.  A small TO220 with heatsink can dissipate about what to air?  5W?  For 0.5Amps linear drop 5W is only 10V.  So, yes, it will probably be "ok" doing 0.5Amps.

The trouble is... I'm not finished.  I am about to add a second board to the project which itself will pull 0.5Amp or more.  So I need a different solution.  3.3V at 1Amp just seems wrong.  It's too small a voltage to be dealing with voltage drop on rails.  You can't even afford to drop 10% and (as I found out the hardware) boosting the 3.3 to 3.5 or 3.7 will fry ICs.

Am I heading in the right direction with a hierarchical power distribution network, rather than a central power supply?

Say, for example, my input is raw "automotive class 12V".  So that's going to be, in my case, anything from about 10V (rare) through (13V nominal) to 15V with 15 amps of current in the extreme.

If I diode/ferrite ring, inductor and bulk cap that input voltage.  Use an LM8712 to regulate that to 12V (max), distribute the 12V to each sub board.  Each sub board then linearly drops that to 5V which is distributed to each individual module to regulate down to 3.3V.

Damn.  That's not going to work.  Linear regulation is still going to just cost a lot of heat.  13V down to 3.3V at 1 Amp is 11.7 watts of heat, all I am doing above is dividing out who drops what parts of the heat.

I suppose I don't really have a choice but to go switch mode.

Twists and turns.  I want to avoid switch mode for "Audio noise" reasons.  I was hoping the DC power supply was to be provided by a battery, a bit like a UPS upside down.  It disconnects from the charger when in use and reconnects to the charger when idle.  By making that battery a high voltage and providing a center tap in the pack I can create a true floating -12V 0V +12V linear supply for the headphone amp.

So it seems as though, while that will work, the issue is in supplying those hungry 3.3V rails from +12V (or 24V if that way inclined to keep the pack balanced). 

So, I think I have answered my own question.  If I am to use a battery pack with a centre tap the power hungry 3.3V rail will absolutely need to be supplied by the full pack voltage to keep it balanced.  I am NOT linearly dropping an amp at 24V down to 3.3V.  No sir.  I'd like this to be useful in summer also.

Hmmm.  Now I need to find a convenient and quiet! DC-DC converter in module/daughter board form.  Such that I am not involved in the analouge calculations required there in

I still need to answer if 3.3V 0.5 amps is resonable input requirements for a PCB and if it really should be 12V+ at 0.1A and locally step it down with additional DC/DCs.

[paulca]

Ramblings in schematic form.

Question:  Should there be another stage, say ~24V down to 12V between PCBs and further dropping to 3.3 for the modules on that PCB?

If the power is plained on the PCBs.  The 3.3V at 0.5Amps for each PCB should be fine, right?

[Peabody]

I think you need to at least try a switching regulator.  Adafruit has small modules that use the MPM3610 buck converter.  The PWM is at 2MHz, so it's possible it would be quiet enough for your audio.  They make 3.3V and 5V versions, which can provide up to 1.2A from input voltages up to 21V.  That's just one example.  I don't know of another way to avoid the heat.  I have seen circuits where a buck converter is used to drop the voltage most of the way, followed by a linear regulator that removes most of the noise.  That might be an option.

[paulca]

Buck converters get a bad name for noise.

However what is absolutely crippling my system right now is USB noise.  The stuff is infectious and a pain in the ass as it's mostly common mode.  I was running the audio board of a switch mode bench supply and thought that was the source of noise, but switching my audio board off, leaving just the headphone amp running and the noise was still there.  Eventually I found that removing the USB signal cable from the PC and ... silence.  There IS noise a very, very faint whine, but that is only when the buck/boost converter is also powering abunch of other things at 2.5Amps.  When it's powering only the headphone amp and audio board it's silent....  "that" one is.  I have another one that whines like a child constantly.

That USB module is about to receive some surgery to at very least remove it's VBUS line it doesn't need but won't be helping.  It's VBUS power died anyway while supplying that 0.5Amp current, gave all it's fucks in one trail of blue smoke.  I can't isolate the ground and the module won't run through a "full speed" isolater.  I might just give up on it and go back to optical.

I have a 3A buck converter I'll try running the board on that and see how the noise pallete is.

[Peabody]

I found this on Youtube that might be useful:

[Infraviolet]

An MCU pulling half an amp, this sounds more like a problem with the dev board. I'd focus on whether that isn't due to a short or damage first before thinking about how to supply that much current, half an amp doesn't sound right at all for a microcontroller.

Depending what you are doing in the circuit overall, you could always use a switching regulator to provide the 3.3V rail, and then heavily filter the parts of that rail where noise would be a serious problem via use of Pi filters (two capacitors and an inductor, letter Pi configuration).

[paulca]

The board has an STM32H750 @ 480Mhz an Atmel/Xlinx-FGPA combo (USB audiophile board) and 2xPCM5102 DACs. 

The two big loads are the FPGA USB Bridge and the STM32.  Each pull close to 200mA.

I am, at least now, thinking it through and the case ventilation is going to need consideration and the STM32H7 might even like a heatsink!  Hopefully not a fan!

The audiophile USB board might be replaced with a smaller optical board.  All of the boards I have got from that manufacturer have been "hot" and "hungry".  Some with ICs sitting at 65*C constantly etc.  It might be time to move on from them.  I have another optical board to test, hopefully it pulls less power, makes less heat and isn't as fragile as the last two which died in suspicious circumstances.

On PI circuits.  I did a quick skirt around the power supply filtering space again and ... my determination... a battery is easier than working that stuff out.  Although, it would appear that's not an easy win either

[Jwillis]

Yes Buck converters do get a bad name for noise. But it very much depends on application and the frequency of the converter. For RF they probably wouldn't be the best choice but I could be wrong since I don't work with RF. Older SMPS were bad for audio because of the lower switching speeds. Newer converters work at much higher frequencies well beyond the audible range. Even cheap simple switchers actually have very clean outputs. Working with both LM2576 and LM2596 I find them to be not to bad. The LM2596 being the better of the two with the higher switching frequency.

[paulca]

A little MP1584 3 Amp buck module I had (10 of) lying in a box, powers it fine.  More efficient too obviously.   

Looks silent enough on the scope and sounds silent enough.  Also means I can just use the tiny little daughter boards and not have to deal with the buck layout.

[dobsonr741]

If MP1584 would not end up silent enough you could try a synchronous buck converter (no Schottky).