noisepowerpower supply
What's the best solution for noise reduction in switching power supplies?
I'm going to use a switching dc to dc buck converter to supply an STM32 Cortex-M4 microcontroller.
But I'm a little concerned about reliability of the supply.
What considerations must be made when an mcu connected directly to a switching regulator that works at very high frequency?
Switching power is noisy, there's no doubt about it - typically anywhere from kilohertz up to the megahertz range, both CM and DM.
I also think you're getting your terms mixed up. I presume you meant to say transformer and bridge rectifier, not Wheatstone bridge.
You can use a switcher to feed a 317 regulator and get the benefit of cleaner output and less efficiency loss than a fully linear solution (mains frequency transformer et al)
A 317 is active and will reject ripple as a function of the voltage feedback network controlling the series-pass transistor in the device.
I'd consider this part with an extra stage of LC output filtering: -
I'd also consider adding 10nF (or whatever suits your bandwidth) capacitors on the ratiometric sensor outputs to ground to further reduce noise. Given that the outputs are ratiometric I'd also strongly consider measuring ratiometrically i.e. use Vcc as the reference for your ADCs - this cuts out a bunch of errors.
Best Answer
The switchreg will have parasitic ringing at 50X or 100X the switching frequency. Consider 100MHz ringing with 100mV amplitude. With close to ZERO ohms source impedance (the ringing is coming from a Power Circuit, after all).
You could just let that 100MHz ringing, at 100mV, simply enter your MCU. But that will trash any onboard ADC. And the clock-multiply PLL will be wonderfully jittery.
You need 3 regions of GND, to be successful
1) the SwitchReg
2) the filter: bead in series, low-esr cap to GND#2
3) the MCU
These 3 regions of GND should only connect through narrow pieces of foil.
simulate this circuit – Schematic created using CircuitLab
Here is the noise attenuation of the PI filter shown in the previous schematic, for 4 values of Rout of the SwitcherReg (5 / 0.5/ 0.05 / 0.005 ohms)
Notice the 12dB peaking for 0.005 ohms, the 3dB peaking for 0.05 ohms, then the well-dampened behavior for 0.5 and 5.0 ohms. Thus some discrete resistance between SwitchReg and the PI filter seems useful.
And here is the wide-spectrum filtering, assuming the 3 regions of GND are effective, for 0.005 Ohms Switcher Rout; above 3MHz, you get at least 47dB attenuation.
Quiet GND and VDD have to be designed; once quietness is achieved, guard it, protect it, do not expose GND and VDD to trash sources.
How quiet is quiet? 47dB below 100mV ringing of SwitchReg is 200:1 reduction, or 500 microVolts. Given 2.5 volt ADC fullscale, you have dynamic range of 5,000:1, only 12 bits. To reach 16 bits, VDD trash needs 24 dB more filtering. To reach 24 bits (and 24 bits ADCs will be trashed by all aggressors entering the VDD pin), you need 72 dB more filtering, or 2 or 3 more PI filters.
And once 24-bit clean GND and VDD are achieved, you better protect them.