Required bandwidth of the error amp in SMPS

Thread Starter

kender

Joined Jan 17, 2007
264
Folks,

I’m building a battery charger (500kHz boost), and designing my own analog control loop for it. Is there a way to estimate what bandwidth should the control loop OpAmp have? Is there a rule of thumb?

Any suggestion, insight or reference is really appreciated!

Cheers,
- Nick

EDIT: Parallel thread in the Yahoo group called "Switchmode": http://tech.groups.yahoo.com/group/switchmode/message/2443
 
Last edited:

SgtWookie

Joined Jul 17, 2007
22,210
500kHz is mighty zippy. Any particular reason you're going that high?

While it's true that you can decrease the size of the inductor as you go up in frequency, you're still limited by the amount of flux the core can handle - and as you increase in frequency, you incur greater switching losses; to the point where your switching losses negate the benefits of the higher frequency.

I suggest you look more in the range of 50kHz to 200kHz, and target somewhere in the middle of that. But without knowing more about your specifications, it's difficult to give any kind of decent answer.
 

Thread Starter

kender

Joined Jan 17, 2007
264
500kHz is mighty zippy. Any particular reason you're going that high?
My charger has to fit in a tight (narrow and thin) place, so space is at premium. Input power is abundant. Heat dissipation isn't a great concern (yet). As a result, I'm not concerned about losses too much.

More specs. Input: 5V. Output: 9V max. Current: 3A max.
 

crutschow

Joined Mar 14, 2008
23,494
I’m building a battery charger (500kHz boost), and designing my own analog control loop for it. Is there a way to estimate what bandwidth should the control loop OpAmp have? Is there a rule of thumb?
You need a bandwidth above the LC resonant frequency of the output inductor and filter. Generally this is in the neighborhood of a few kHz. So any general purpose op amp should have enough GBW for that.

You do know that you need compensation for this resonant frequency in the control loop to avoid loop oscillations(?). Here's a good reference on how to do that.
 

SgtWookie

Joined Jul 17, 2007
22,210
What are your available X/Y/Z dimensions? "Narrow" and "thin" are quite subjective.
What is your battery chemistry?
Or better yet, a link to the manufacturer's datasheet?
 

bountyhunter

Joined Sep 7, 2009
2,512
My charger has to fit in a tight (narrow and thin) place, so space is at premium. Input power is abundant. Heat dissipation isn't a great concern (yet). As a result, I'm not concerned about losses too much.

More specs. Input: 5V. Output: 9V max. Current: 3A max.
So, you want a small, narrow, thin, boost converter that delivers 27W.... and you don't think power dissipation and losses are going to be a problem?

I think you're in for a shock.:p

IMHO, you would be lucky to get 60% efficiency running at 500kHz which would give you about 10W dissipation in the supply at full power.
 

bountyhunter

Joined Sep 7, 2009
2,512
You need a bandwidth above the LC resonant frequency of the output inductor and filter. Generally this is in the neighborhood of a few kHz.
Probably not in this case. The whole purpose of going to 500kHz is to shrink the size of both the L and C on the output. 500kHz is above the resonant frequency of aluminum and Tantalum caps, so they can't be used. That forces the use of ceramics and size is very pricey for those.
 

SgtWookie

Joined Jul 17, 2007
22,210
Well, just for grins I ran some numbers through National/TI Webench; it came back suggesting a LM3481MM operating @ 569kHz, Pd @4.64W (I'd optimized it for footprint rather than efficiency). 640mm^2, or just under 1"^2. That's pretty tiny.

I'm attaching the .pdf it spit out.
 

Attachments

bountyhunter

Joined Sep 7, 2009
2,512
it came back suggesting a LM3481MM operating @ 569kHz, Pd @4.64W (I'd optimized it for footprint rather than efficiency). 640mm^2, or just under 1"^2. That's pretty tiny.
.
How are you going to get that 5W out in a board smaller than one sq inch? That thing is going to fry.
 

SgtWookie

Joined Jul 17, 2007
22,210
How are you going to get that 5W out in a board smaller than one sq inch? That thing is going to fry.
You tell me! You said that you wrote Webench! ;)

Slap that booger on a heat pipe, and let'er rip. ;)

[eta]
Seriously though, he said he needed it tiny. I just wanted to get some kind of a starting place; so I had Webench make it as small as possible, efficiency be hanged. The efficiency started off being 90% with a ~750mm^2 footprint. So now that we have an idea how small it can be, perhaps we can pop the efficiency back up, depending on what kind of space is available. Just a tad over 3W could be manageable, depending on what the enclosure is.
 
Last edited:

bountyhunter

Joined Sep 7, 2009
2,512
You tell me! You said that you wrote Webench! ;)
I was one who gave technical input..... which in most cases was ignored as you might have surmised.

Slap that booger on a heat pipe, and let'er rip. ;)

[eta]
Seriously though, he said he needed it tiny. I just wanted to get some kind of a starting place; so I had Webench make it as small as possible, efficiency be hanged.
I don't fault what you did, just pointing out that Webench (and most sims) are brain dead. That one aparrently has no clue about calculating power dissipation in components.

That design has an IC, a single surface mount FET, inductor and O/P cap that probably account for most of the losses. To get 5W safely out into the PCB and dissipated away will take a fairly large PCB. It doesn't do any good to get a design tiny if you can't manage the heat. It might be better to go up to a larger FET like a TO-263 package to help or use multiple FETs to spread the heat.
 

SgtWookie

Joined Jul 17, 2007
22,210
All I did was try to get some kind of starting place, but I said that already.

Yeah, I dislike it when I get ignored too; it makes me feel like I wasted my time.

It's up to Kender to tell us what kind of room there is to work with. Knowing if there was air circulation planned, a cold plate, etc. would also be really helpful. Then we'll at least have some idea of what we're dealing with.
 

Thread Starter

kender

Joined Jan 17, 2007
264
Guys, thanks for the insight.

- I have completely overlooked the resonant frequency of aluminum and Tantalum caps.
- I knew that a compensation network was needed, but I didn't have a good paper to follow.

Now I need to chew on this, make experiments, fry a few things.

I have 1" x 3" board to play on. 2 layers and 1oz copper, probably.
 

SgtWookie

Joined Jul 17, 2007
22,210
Psssstt.... Kender...

1) What are the dimensions of the space that your project has to go into?

2) Are there some provisions for cooling in this space, like perhaps forced air, a cold plate, even liquid cooling? Might get the power down to 3W or so, but that power has to go SOMEwhere.
 

bountyhunter

Joined Sep 7, 2009
2,512
Guys, thanks for the insight.

- I have completely overlooked the resonant frequency of aluminum and Tantalum caps.
You aren't the only one. National released a 1.6 MHz switcher IC that REQUIRED the use of ceramic caps (big letters on the front page of the data sheet). You would not believe how many whiner calls I got about how it didn't work from people using tantalum caps.:p
 

Thread Starter

kender

Joined Jan 17, 2007
264
1) What are the dimensions of the space that your project has to go into?
3" x 1" x 0.3"

2) Are there some provisions for cooling in this space, like perhaps forced air, a cold plate, even liquid cooling? Might get the power down to 3W or so, but that power has to go SOMEwhere.
Tough luck here. This power supply sits in a closed nylon pocket. Imagine tucking it into an organizer pocket of a backpack (where pens, glasses, business cards go). No air flow. If one side of the enclosure is made of aluminium, that would only help spread the heat from point sources. I don't have a definitive answer as to where the 3W can be dissipated. I need to experiment.
 
Last edited:
Top