SEPIC or buck+boost for cleaning up 12 volts?

[gxti]

How would you transform 10-15VDC (from a lead acid battery system) to 12V at 1A? SEPIC looks interesting but is more complicated than a buck and boost converter stitched together so I don't feel like I would be as successful designing one. I also need several other voltages at lower power levels, so I was considering buck to 5V @ 3A, then boost to 12. Any pitfalls to watch out for? Tips on SEPIC (or other buck-boost topologies)?

[Bored@Work]

If we say your buck converter has an efficiency of 80% on a good day, and your boost converter has an efficiency of 80$, too. Then the total efficiency is 64%. Can you afford to piss 36% of your battery power away?

[SeanB]

Follow the Application Notes for them, most will give a good range of component choices and design criteria. Look at the TI and AD converters, they have some good ones in your range.

If you want 2 converters better to go to 15-20V then do the second down to 12V, lower losses as currents are all lower.

[gxti]

If we say your buck converter has an efficiency of 80% on a good day, and your boost converter has an efficiency of 80$, too. Then the total efficiency is 64%. Can you afford to piss 36% of your battery power away?

Yes, thankfully. It's a battery backup not an offline system, and only a few watts load on batteries from a computer UPS that could last 10 minutes on a 500W load means I'll get hours of runtime. The batteries are near the end of their lifetime though so we'll see. The other nice thing is that the oscillator only draws max current (0.5A or 6W) on cold startup. Steady state is more like 0.15A.

If you want 2 converters better to go to 15-20V then do the second down to 12V, lower losses as currents are all lower.

The only reason I was considering it is if I could use the intermediate voltage, sounds like I should just focus on SEPIC. The LT3580 datasheet looks nice.

[HackedFridgeMagnet]

I guess the main question is "Do you really need 12 volt regulated supply?" and I suppose it is otherwise you wouldn't be asking.
If you're load is just another switched mode supply anyway, you can probably get away with just using the battery. It's always worth checking.

If you are going the DC converter route then the most efficient way is to do either a buck stepdown or a boost step up. Not both. You might be able to acheive this by using a 24v battery.

If this is impractical then I think Sepic is the way to go, but you could also consider Cuk.

I have heard that the Sepic is noisy.

The Cuk (pronounced chook, I think) suffers from needing switches rated at double the nominal voltage and it reverses the polarity of the output. Both of these issues might be ok for a low voltage battery powered system.

[Jason Long]

There are several manufactures that make buck/boost converter ics, some with stated efficiencies as high as 96%. Linear Technology and TI are the two that pop to mind...
Just do a parametric search volt/amp ranges you require. Most of the data sheets will provide sample circuits to make the circuit design easier.

[IanB]

The question was already asked, but it's worth repeating: is an accurately regulated 12 V supply truly needed?

If it's just a matter of cleaning up noise then bit of filtering should do the job.

Very few 12 V appliances need exactly 12 V. Most will be quite happy over a range of 11-14 V or wider. Anything designed for automobile applications will certainly be tolerant of voltage swings.

[gxti]

The question was already asked, but it's worth repeating: is an accurately regulated 12 V supply truly needed?

I should have explained this in the first post, but yes, unfortunately it is needed. The oscillator's heater has to re-settle after an input voltage change, which will cause a step in the internal temperature of the oven and thus a step in the oscillator frequency. The last thing I want is for the battery charger to have an influence on the oscillation frequency.

At this point I've got a schematic laid out for a LT3580 SEPIC. It was relatively easy, but that's because I skipped all the cheaper converter ICs whose datasheets didn't hold my hand enough. Something cheaper could definitely get the job done, but as a fraction of the total parts cost it's not bad at all because I've got 3 other rails as well, all MC34063s. Considering using a linear post-regulator for those as well.

[hlavac]

If we say your buck converter has an efficiency of 80% on a good day, and your boost converter has an efficiency of 80$, too. Then the total efficiency is 64%. Can you afford to piss 36% of your battery power away?

What? Buck-boost is not two converters in series, it's a switch mode converter topology capable of operating in both boost and buck mode - there are two switches... it operates as buck converter or boost converter depending on whether the input voltage is higher or lower than output voltage. Efficiency will be fine.

[gxti]

Design's finished, off it goes:

Schematic: http://partiallystapled.com/~gxti/circuits/2012/07/31-sfpower_sch.png
Routed: http://partiallystapled.com/~gxti/circuits/2012/07/30-sfsupply_tracks.png
3D top: http://partiallystapled.com/~gxti/circuits/2012/07/31-sfpower_front.jpg
3D bottom: http://partiallystapled.com/~gxti/circuits/2012/07/31-sfpower_back.jpg

[gxti]

Spot the backwards electro in the schematic! I didn't... luckily it bulged and oozed but didn't explode. You get what you pay for, and I paid for the good stuff.

So one cap swap later it's up and running, all 4 channels! I don't have an easy way to stress test everything though, I really need some homebrew dummy loads. But while the voltages look good across all input conditions, and everything seems to perform well under what load I can muster as well as a few accidental short-circuits, the noise is pretty bad. 50-120mV P-P bursts on the -5/+3.3/+5 rails, and 100-200 mV P-P on the +12 SEPIC rail when I'd really like everything to be < 20mV P-P.

What's the best approach for squashing high-frequency SMPS noise? Add an LC stage? Ceramic cap to bypass the big electro?

Poor bastard:

[AndyC_772]

You'll probably find that a 10uF ceramic capacitor soldered directly across the pins of each electrolytic helps a lot - but potentially even more significant is the way the noise is measured.

The correct way to probe a high frequency circuit is to use a scope probe with a very short ground spike. Most ordinary passive probes have a ground lead several inches long with a clip on the end, and these are completely useless for probing high frequency switching spikes. They'll show noise that's actually far worse than what really exists between the voltage output of the supply and the ground pin at the supply.

Note: the key words here are "at the supply". The output voltage must always be measured with respect to the correct ground point, and where high frequencies are concerned, cable inductance is a big factor. Ground at the PSU is not at the same potential as ground at the scope.

What you need to do is dismantle the probe a bit. Remove the ground lead with the clip and find a grounded point somewhere nearer the tip of the probe; sometimes there's a plastic shroud over the body of the probe which can be unscrewed to reveal the metal body of the probe, or there may just be a ground ring near the probe tip. Wrap a piece of tinned copper wire round the probe, ensuring it makes contact with the body of the probe as close to the tip as possible. Touch the end of this wire onto a nearby GND point on your power supply board while you touch the probe tip on the voltage rail you're measuring. The shorter you can make this ground spike the better.

Try making the measurement this way and see what the result is. If your scope probe can't be used in this way then a new probe may be in order, as you simply won't get accurate, representative results any other way.

Tip: to see the effect of the ground lead and clip, try clipping it to the probe tip itself, then touch that onto the ground of your power supply. If you see noise and spikes on the scope, then your probe is lying.

[Psi]

It can be done.
I had exactly the same issue and needed a stable 12V from a car.
I made a few prototypes that worked but in the end i just bought a pre-made dc-dc board because i wanted it to be over 90% efficient and it was going to take lots of work to get my 75-85% design up to 90+

This is the one i bought, which is 95% efficient and can supply up to 12A
http://store.mp3car.com/DSX12VD_140W_DC_DC_Regulator_p/pwr-025.htm




In case it helps you, this was my prototype designed around a TL494 and toroid transformer which i wound.

[T4P]

Well
1) TL494 will be quite tough to achieve even 85% with ...
2) At the speeds TL494's run it's even tougher

Well for one i always noticed that higher efficiency SMPS's always clock beyond 500KHz ...

I can be wrong again i'm open to other ideas

[gxti]

You'll probably find that a 10uF ceramic capacitor soldered directly across the pins of each electrolytic helps a lot - but potentially even more significant is the way the noise is measured.

Does using ceramics directly on the output add stability problems? I seem to remember more than a few switcher datasheets warning about low ESR. I've got some 10uF ceramics but they won't like the 12V rail, but I've also got a bunch of 50V 1uF that should work just as well in this application. I'll try them out after I get a better measurement.

What you need to do is dismantle the probe a bit. Remove the ground lead with the clip and find a grounded point somewhere nearer the tip of the probe; sometimes there's a plastic shroud over the body of the probe which can be unscrewed to reveal the metal body of the probe, or there may just be a ground ring near the probe tip.

These are the garbage probes that come with every Chinese scope under $1000 but luckily they have a ground ring right next to the tip so I'll rig up a ground probe for it. Come to think of it, I may not need to since I can probably touch the ring itself to the ground lead of the cap. Thanks for the advice, I did not think to check the instrument first.

Tip: to see the effect of the ground lead and clip, try clipping it to the probe tip itself, then touch that onto the ground of your power supply. If you see noise and spikes on the scope, then your probe is lying.

It also means my board is an EMI nightmare, which considering I have 4 switchers on a 2 layer board and no experience with EMC is extremely likely. I plan to put this all in a metal box, but the problem is there's stuff other than the power supply in there with it.

[gxti]

Well for one i always noticed that higher efficiency SMPS's always clock beyond 500KHz ...

Increasing the frequency adds to switching losses. I would think that all other things equal the lowest frequency switcher would have the best efficiency as long as the coils were sized appropriately, but there are lots of other reasons to use high frequencies and the efficiency problems can be dealt with.

The SEPIC converter I built uses ~1.5Mhz with a coupled inductor. Once I solve the noise problems I'll have to do some efficiency measurements. I don't have any efficiency goals so I am happy as long as it does well enough to not burst into flames under load. So far so good.

[SeanB]

Tin can in the shielded box with the switcher in it, and soldered seams with ferrite beads on the leads, and soldered in 10n feedthrough capacitors.

[AndyC_772]

Does using ceramics directly on the output add stability problems? I seem to remember more than a few switcher datasheets warning about low ESR. I've got some 10uF ceramics but they won't like the 12V rail, but I've also got a bunch of 50V 1uF that should work just as well in this application. I'll try them out after I get a better measurement.

Shouldn't be a problem. Try with and without. Very low ESR can (apparently!) be an issue for stability, but I'm not sure I've ever actually seen it make a power supply go unstable because of it. Do try putting a 1uF ceramic cap directly across the leads of one of your electrolytics, then probe across it with the scope.

These are the garbage probes that come with every Chinese scope under $1000 but luckily they have a ground ring right next to the tip so I'll rig up a ground probe for it. Come to think of it, I may not need to since I can probably touch the ring itself to the ground lead of the cap. Thanks for the advice, I did not think to check the instrument first.

This probing technique is so important I'm amazed it's not more widely taught. I think it was a Tektronix rep who first showed me.

It also means my board is an EMI nightmare, which considering I have 4 switchers on a 2 layer board and no experience with EMC is extremely likely. I plan to put this all in a metal box, but the problem is there's stuff other than the power supply in there with it.

That's quite possible! Though unless you actually need to get your product formally EMC tested, it may be completely unimportant.

The ceramic caps will help, but the amount of noise generated by a switching power supply is about 80% down to its physical layout. Keep the high current paths short, use wide copper traces to keep stray inductance to a minimum, and keep the overall layout as compact as possible. There are plenty of techniques for minimising noise and emissions, but get the prototype working functionally first.

[gxti]

Here's the top layer routing, if you care to criticize: http://partiallystapled.com/~gxti/circuits/2012/07/30-sfsupply_tracks.png

The bottom is all ground fill except for a single power trace near the edge.

[AndyC_772]

Well for one i always noticed that higher efficiency SMPS's always clock beyond 500KHz ...

I can be wrong again i'm open to other ideas

There are two major contributions to SMPS power loss - switching losses and conduction losses.

Conduction losses are due to the parasitic resistances in the circuit: primarily the resistance of the copper in the inductors, and the drain-source resistance of the transistors when they're switched on. These scale with the square of the current, ie. they follow P=I^2*R.

Switching losses come from the fact that when a power transistor switches between the off and on states, it has to go through a brief period where it conducts but still has significant resistance. When it's off there's no current, and so no power loss. When it's on, the current is high but the resistance is quite small, so power dissipation is once again relatively small. However, during the transition between the two, there's both a high current and a high resistance, and since P=I^2*R, power dissipation can be briefly very high indeed.

The total switching loss equals the amount of energy lost per switching event times the number of switching events per second, ie. it's proportional to switching frequency.

Low switching frequency therefore tends to mean low switching losses, but the trade-off is that peak currents have to be larger - so conduction losses increase. With higher switching frequencies there are greater switching losses (ie. more lossy switching events per second), but conduction losses reduce because the peak currents in the circuit reduce.

For any design there's always a trade-off between conduction and switching losses, so it follows that there will be an optimum frequency that gives the best overall efficiency.

[NiHaoMike]

I have designed a buck-boost converter for my PC, rated for up to 60A output at 12V. It's basically a buck converter integrated with a current doubler push-pull forward converter. It gets great efficiency since the MOSFETs I used (4x FDB7045L) have a very low RDSon for the buck converter and in boost mode, the push-pull converter is handling less than 17% of the power going through the unit.

[Psi]

Well
1) TL494 will be quite tough to achieve even 85% with ...
2) At the speeds TL494's run it's even tougher

Yeah, the efficiency was ~80% most of the time. Sometimes it would dip to 75% and i once measure 84%.
It was quite load dependent.

Slow frequency switch mode can be quite efficient, (20-50kHz).
With high frequency, like 100Khz-1Mhz you get losses from the rise/fall delay time where it's burning off energy.
This isn't much of a problem at low frequencies because the turn-on/off delay is such a small part of the entire pulse