RFC: Switching mode power supply schematics

[KW1252]

Here's my take on the occasionally emerging power source issue. Notional Semidestructor has quite a nifty design application on their website, "webench"; I got this design after playing with the dials a bit:



The values:

input range 9-14.2V
Imax=2.5A
efficiency= 91%
Peak to peak ripple voltage 0.01V
Peak to peak inductor ripple current 0.60A

Having a high-power 260kHz device on board is a source of worries though, the question is are they great enough not to take this route.

[jharvey]

What's the ripple? Theres a design with spice noted in the simulation directory of PUMA. The problem with using the switch mode supply directly was that the ripple of 1mV is 1 bit of the ADC, so it would induce noise on the ADC. So what Marcos is doing, is setting it to something like 6V using the linear reg to reg 5V with less ripple. This gets both the efficient switch mode / less heat, while maintaining the ripple of the linear reg.

If we can get the ripple low enough, I'd be happy to use it. Is the National Semi simulation file something you can post, do they let you download it and such?

[KW1252]

I attached a design printout from the site, to participate would require sign-up on the National Semiconductors board. The web software seems really clever though, with all the server-end services like thermal simulations so it's likely worth the hassle.

I do think the analog filtering circuits should have low pass filtering with a SMPS; however, I wouldn't worry too much about the LSB error on 10-bit ADC. The inherent error with all the sensors used here is between 2% and 5%. Even the MPX4250, one of the principal (and most accurate) sensors has the maximum error of +-3.45kPa (+-60mV).

What I find a greater problem is the minimum regulator input of 8 volts, which is a bit high for cranking conditions. A big cap on the power line might even things out quite a bit, but it's still a big issue.

[KW1252]

A mistake there - the first capacitor should be 22.0µF (and on the iteration 15.0µF). The decimal point was obscured on the image

[jbelanger]

What's the ripple? Theres a design with spice noted in the simulation directory of PUMA. The problem with using the switch mode supply directly was that the ripple of 1mV is 1 bit of the ADC, so it would induce noise on the ADC. So what Marcos is doing, is setting it to something like 6V using the linear reg to reg 5V with less ripple. This gets both the efficient switch mode / less heat, while maintaining the ripple of the linear reg.

If we can get the ripple low enough, I'd be happy to use it. Is the National Semi simulation file something you can post, do they let you download it and such?

Is the ripple the same no matter what the load is? I have yet to look at switching supplies but was told that you had to tune the components for a specific current output and that if you had a potentially big range in the output current, you would not get the same ripple amplitude at all loads. Is it the case?

Jean

[KW1252]

The design point load is 2.5A, it's an assumption on worst case scenario (everything installed and working at full load). The voltage ripple is probably much less with lighter load.

[jharvey]

On the one found at this link, https://github.com/jharvey/puma_JAH/blo ... motive.zip You can change the load to check the ripple. I seem to recall the worst ripple was fully loaded, however, you can change the load to check the tunning. Here's the PDF copy of the report. https://github.com/jharvey/puma_JAH/blo ... report.pdf

Oh also I just notice it was 4mV ripple, not 1mV.

[KW1252]

I was looking through the schematics, the analog inputs already have a low pass filtering with 1.6kHz cutoff frequency... shouldn't that pretty much cut all the SMPS-generated noise unless the entire board oscillates at some sub-harmonics below it?

[jharvey]

Good point, I really don't know if other filtering will remove the ripple or not. Testing is probably the only real way to know for sure.

[nitrousnrg]

Cool thread, let me put my 2 cents:

In Puma, there is a SMPS to step down vbat to 5.75v. This way, regulators doesn't get hot, and they (or at least the big one) attenuate the SMPS ripple by 30dB or more. Its capable of handling 30v inputs, but I can't see if it can work at 1 or 2 mA loads. Also, I have the *impression* that switching things shouldn't be supplying analog circuitry or sensors, but its mostly a prejudice.

Unfortunately, input voltage can't be under the specified output voltage (5.75v). That output voltage could be lower, though.

btw, design considerations demand to have 2 sources of 5v, at least on boards that follows the "always on" strategy.

This is an app that jharvey has found, and you can see the actual values I'm using


A quite important note:
Simulators like these aren't wise enough to suggest stocked items. I spent some hours with iterations like:
START
does it work?
is it the best combination so far?
are components available? -> yes -> DONE (ha! never happened)
are there similar components? -> no -> goto START
| yes
DONE

I checked the components suggested by National, and you have to get them from different distributors. Another restraint was the 85°C temperature limit. Not a big deal, I guess you can pick other components if they are similar enough.

In the original design calculated by ST I got 9mV of ripple. After playing around it went down to 7mv or so (jharvey got 4mV for a 5v output). It was a shame that the output capacitor and inductor weren't available. With similar and available components the best I had was a 14mV ripple at a 14V input (0.24%).

The moral is: keep in mind availability :-)

And, according to simulations, ripple doesn't depend on load, but in input voltage. Power losses are mostly load-dependent.