DC-DC converter in mixed signal application

[ZeroResistance]

What factors do we need to consider when selecting a DC-DC switching converter to power a mixed signal application.
The application would have very sensitive analog circuits along with the usual MCU and few other digital peripherals.
1. Is it a standard practice to have a linear regulator after the output of the DC-DC converter to reduce noise / ripple?
2. If the switching frequency of the DC-DC converter is outside the bandwidth of the signal of interest does that make it any less stringent on the design.

TIA

[floobydust]

DC-DC converters, like the 1W or 3W SIP modules are fairly noisy, they are around 300kHz, assuming it's an isolated module.

Most have poor regulation as they are simple oscillators, and the load+line regulation can be as bad as 7.5% so a linear Vreg is used downstream. Other modules are 1% if you shop around.

You get output noise which is easy get rid of with an LC filter, before any linear voltage regulator.
LDO's and such rarely have bandwidth above 10kHz, so 300kHz incoming is not good, it just passes through the voltage regulator IC.

I've had bipolar i.e. +/-15V units die or fail to startup without adding reverse diodes across the outputs. If your load current could be from +15V to -15V (not common GND), add diodes.

You get input-side noise current, which can make a mess of your rails on that side. Note the datasheets have "EMC recommended circuit" which is a CLC Pi filter.

The unknown enemy is capacitive coupling across the transformer windings (primary and secondary) with switching harmonics extending many MHz. You will find this out at the EMC lab, when the product fails conducted emissions tests. It can radiate out the primary or secondary side, depending on your design.

The radiated B-field from the module (transformer) induces noise in nearby pcb components and traces, so leave some room or put insensitive parts around it.

[perieanuo]

1.no
2.no


Envoyé de mon iPad en utilisant Tapatalk

[ZeroResistance]

DC-DC converters, like the 1W or 3W SIP modules are fairly noisy, they are around 300kHz, assuming it's an isolated module.

Most have poor regulation as they are simple oscillators, and the load+line regulation can be as bad as 7.5% so a linear Vreg is used downstream. Other modules are 1% if you shop around.

You get output noise which is easy get rid of with an LC filter, before any linear voltage regulator.
LDO's and such rarely have bandwidth above 10kHz, so 300kHz incoming is not good, it just passes through the voltage regulator IC.

I've had bipolar i.e. +/-15V units die or fail to startup without adding reverse diodes across the outputs. If your load current could be from +15V to -15V (not common GND), add diodes.

You get input-side noise current, which can make a mess of your rails on that side. Note the datasheets have "EMC recommended circuit" which is a CLC Pi filter.

The unknown enemy is capacitive coupling across the transformer windings (primary and secondary) with switching harmonics extending many MHz. You will find this out at the EMC lab, when the product fails conducted emissions tests. It can radiate out the primary or secondary side, depending on your design.

The radiated B-field from the module (transformer) induces noise in nearby pcb components and traces, so leave some room or put insensitive parts around it.

These are some pretty amazing points that you highlighted.
+/-  7.5% regulation is pretty bad, I have some datasheets that point out they could be as high as 20%.
So looks like a linear LDO is somewhat a must.
The noise that you highlighted where you said that it can't be removed with a linear LDO since their bandwidth is limited, so a filter at the input of the LDO is required correct?
Are there any capacitor switched versions of these power supplies available would they be lower noise than these? (but I guess they won't do isolation...

[ZeroResistance]

I see that most DC-DC converters operate around 100Khz, are there specifically LDO's with a wide band PSRR so that 100khz ripple / noise gets minimised?

[floobydust]

Minimizing Switching Regulator Residue in Linear Regulator Outputs - Banishing Those Accursed Spikes AN101F Jim Williams

Very good read on the topic.
Yes there are LDOs with high PSRR at 100kHz or higher. Since that changes with load and temperature, I play it safe and add a small inductor instead.

[Yansi]

Forget those old steaming piles SIP package DCDC converters. They are nasty shit.

As a non-isolated buck or boost. I've never head an issue in a sensitive application (mostly mixed signal audio designs). With careful layout design, the resulting noise is negligible with a bit of LC/ferrite bead cleaning action.

If you need isolated DCDC, I'd recommend Murata NXJ series (or similar) converters. They have very low pri-sec capacitance, which is great for reducing EMI.

[Doctorandus_P]

Read a bit about this silentswitcher:

https://linearaudio.nl/silentswitcher

It switches, but has residual noise in the uV range.
It's what modern electronics can do.

An then, there is of course also:
https://duckduckgo.com/html?q=switching+regulators+for+poets

[David Hess]

1. Is it a standard practice to have a linear regulator after the output of the DC-DC converter to reduce noise / ripple?

Line rejection of a linear regulator at high frequencies is poor so unless tighter regulation is required, it is usually better to use just an RLC filter to reduce high frequency noise.

2. If the switching frequency of the DC-DC converter is outside the bandwidth of the signal of interest does that make it any less stringent on the design.

Yes, limiting the analog bandwidth below the switching frequency is a big help.

[bson]

What kind of converter and what currents?

I've built a bunch of buck converters for going from USB Vbus to 3.3V and the LM3671 is brilliant - absolutely no ripple even at the full 600mA load.  And it's tiny, easy to copy onto any schematic and add to the layout.  It barely even registers on an IR camera at the full 600mA current, where a linear regulator gets scorching HOT at even 250mA.

So, depends on what you're converting.

[ZeroResistance]

What kind of converter and what currents?

Good question!
I need a converter primarily for isolation about 4200Vac rms / 6KV dc to comply with standards.

[radiolistener]

use very high frequency dc-dc, for example 1-2 MHz, it will increase gap between noise spurs and will minimize risk of influence.

[Yansi]

Also good to note, never ever use switchers with those stupid hysteretic controllers (34063, LT1074 atc... being the fine specimens of this). You won't get the in band noise away ever!

Modern MHz range converters can be made very silent with a tight good layout and a bit of postfiltering (LC, ferrite bead).

[ZeroResistance]

Minimizing Switching Regulator Residue in Linear Regulator Outputs - Banishing Those Accursed Spikes AN101F Jim Williams

Very good read on the topic.
Yes there are LDOs with high PSRR at 100kHz or higher. Since that changes with load and temperature, I play it safe and add a small inductor instead.

Great link! Thanks for sharing. And inductor would work well to reduce the 100Khz switch ripple however what about the high frequency > 10Mhz noise spikes that appear along with the ripple at the output of the DC-DC converter would you need a ferrite bead for that?

[floobydust]

For DC-DC output differential-mode noise going into the LDO, something like a 10uH inductor works, up to 10's MHz which is enough. I find ferrite beads alone don't attenuate enough in the low MHz.

Monitoring mains voltages using an optoisolator and DC-DC for power does have drama.

Most important and not known is the DC-DC converter's common-mode noise ends up getting out on the mains wiring that you are doing the voltage monitoring of. This is a big trap, also with the ADUM isolators too. Even Analog Devices misses the boat AN-0971 thinking the PCB layout is the fix 

The isolation voltage rating of the DC-DC is usually a 1 minute hi-pot test value, NOT an operating value. The insulation's working value, for years of use needs to be considered. It helps to have decent UL/CSA approvals and the proper overvoltage category i.e Cat. III for industrial products. Don't forget a fuse if the isolation (opto or DC-DC) fails, as raw mains will burn up a board badly. Even a 1A fuse is needed when you stare at the circuit for a long time.

[David Hess]

For DC-DC output differential-mode noise going into the LDO, something like a 10uH inductor works, up to 10's MHz which is enough. I find ferrite beads alone don't attenuate enough in the low MHz.

They make special "broadband" ferrite beads which have extended low frequency performance while keeping a high self resonate frequency.

One thing to watch out for when using ferrite beads is not to drive them into saturation with excessive DC current.

[ZeroResistance]

Monitoring mains voltages using an optoisolator and DC-DC for power does have drama.

I have low voltage 12VDC going into the DC-DC converter. What exactly does monitoring the input do?