LT1185-based lead acid battery charger

You need some capacitance on the input and output. Such LDOs are like the princess on the pea, for the output they need the ESR of this capacitor to be within a defined range. Else they can go berserk.

The datasheet says the usual on page 10, quote "The LT1185 is designed to allow output reverse polarity of several volts without damage or latch-up, so that a simple diode clamp can be used". So you have to provide a beefy diode. Best to also protect the FB pin via a diode but that one won't have to be big.

As for bypass-shorting you'd have to ask LTC but with many regulators that is ok as long as you make sure no BE junctions become reverse-biased past their pain threshold.

--
Regards, Joerg 

http://www.analogconsultants.com/

I will add a small tantalum on the output, but does it make any sense on input? Given the current involved (40A range) no cap is big enough and the rest of the device is powered with rectified AC instead of DC for exactly this reason: I don't want any electrolytic capacitor in the device. It must work flawlessly for decades.

OK, makes sense.

As far as I understand, by reverse polarity they mean Vout with respect to GND. In my case Vout is reverse polarized wrt Vin. From the GND point of view the polarity is always correct, especially when the 800uOhm NMOS (R12) enters the stage. So this is my primary concern: what would happen inside the IC then. Unfortunately, I have no LT1185, so I'll use LM337 as a test dummy.

Or another NMOS...

I've sent them a query. Thanks, Joerg.

Best regards, Piotr

I never use tantalums because I've seen them explode, out of the blue. Whenever possible I use a ceramic cap. But then again I never use LDOs either because I don't trust them unless I designed them myself on the transistor level. In this case you could use a ceramic cap with a resistor in series that puts it in the middle of the prescribed ESR range. And then pray that the capacitance that the battery presents in parallel to that will not spoil the whole scheme.

On the input I'd place at least a small ceramic cap, close to the regulator. Many reasons. Aside from stability it's also because you don't want strong external RF signals to leak into this regulator and mess with the regulator loop. AM stations nearby, shortwave, CB radios, et cetera.

IIUC it's not reversed but when there is no or low input voltage the output voltage is higher (meaning more negative) that the input voltage. When that goes too far it can kill a linear regulator IC unless you have a diode across it.

FETs can work as well, their body diodes are usually rated in a similar current class as the channel of the FET.

LTC is one of the best companies when it comes to applications support. Maybe even the best.

--
Regards, Joerg 

http://www.analogconsultants.com/

Agreed. My first attempt was to use a design based on discrete elements exactly because I understand better what is really going on in the circuit. It had an 8A NPN in DPAK soldered directly to an IMS based PCB. It had a nice current limiter set to 3.7A and I thought that IMS will be just the right thing to dissipate the power produced by the NPN under short circuit conditions. Unfortunately, while the board's thermal resistance is superior, its thermal impedance is mediocre and after 2 seconds the NPN exploded in a very nasty way. So what I really need is to have thermal/power limiter with a low inertia. As there are no power BJTs with a built-in thermistor, only two ways remain: use a overcurrent/thermally protected NMOS (for example an OMNIFET)

• a baroquesque driving circuit or a ready-made IC. The second option is not smarter -- just less crazy.

Good point.

Yes, I've checked several other specs and you are 100% right -- by "reversed" they DO mean exactly the Vin-Vout voltage alone, with no correspondence to GND. I have never used a regulator in this weird sector, so my questions may sound weird for experts -- I'm just trying to learn something new without too many bangs.

I meant that the FET will go in sync with the enable signal, but now I think that it is not the best idea.

So one more try: assume LM317 for less pins and the positive regulation side in order not to waste brainpower on sign transformations. Is it allowed connect a Schottky diode between Vout and the load, add a cap between Vout and GND and connect the feedback divider *after* the diode, i.e. to its cathode in order to compensate Uf(temp) changes ? Plus the usual clamping diodes between Vout/Vin and Adj/GND? The reverse current of a 5A Schottky can be as high as 15mA (at 100 degrees) according to the specs, so the clamping diodes are necessary. But the main question: will the feedback circuit happily adapt to this arrangement or is it a no-go? I'll check it, but the question is whether it is allowed, not whether it works with that particular part.

Few more square centimeters available and I would go for a decent SMPS, but unfortunately I cannot afford the real estate due to size limits. BTW, I am just a hobbyist.

Best regards, Piotr

Not sure about your application but I assume that a thermal shutdown will not be very desirable for the functionality of the whole gear.

Why not use a switcher instead? It can be much more efficient.

[...]

We all have to learn once in a while. I'd ask similar questions if it was a software topic :-)

I can't see why it would not be ok. The important thing is to use protection diodes so neither the output-input path nor that of the ADJ pin can ever become reverse-biased much.

However, in this arrangement you will be dissipating tons of power. The LM317 has major drop-out (I think around 2V at that current but haven't looked) and then the Schottky diode will be likely over 800mV unless you use a really fat one.

Look around, there are lots of very tiny switcher chips in the several-amps class available these days. Many can run at frequencies in excess of 500kHz so the inductor will be very small. If you find a synchronous one you won't even have the diode losses.

--
Regards, Joerg 

http://www.analogconsultants.com/

It will be a multichannel 12V lightbulb controller. The ~100Ah accu will allow them to work for several hours when the 230V line is not energized. It must be rock-stable (meaning decades), so I've removed all the quickly decaying elements like the electrolytic caps. Because the bulbs don't care about the exact shape of their powering current, the device is a mixed-mode design. If there power line works, which is the typical case, the bulbs are powered with abs(sin(50Hz)). Otherwise the accu is connected and they get 13.8V DC + some amount of PWM mean power correction.

I also want to dissipate as low power as possible because of cooling constraints. Hence the synchronous rectifier

• a bunch of very good MOSFETs. The + line is common instead of GND solely because the latter would require a PMOS and the best PMOS available is ~4x worse than an NMOS. However, it implies crazy driving in the charger part, but the current is so much lower -- I've decided it's a lesser evil.

The accu charger is just a supplemental functionality and is not expected to be operating very often -- mostly just to compensate the self-discharge current. It should be enabled when the lamps are (mostly) off, so my thermal budget can afford the additional 5 Watts.

The board is on IMS, so all the high power parts share a pretty decent radiator. The trick is to not enable all of them at once. :-)

Actually, there is a global thermal shutdown for safety purposes. It just didn't have a chance to intervene because of the high thermal impedance of IMS and my charger 1.0 exploded. :-) Lesson learned. A built-in thermal protection is a must-have.

A switcher would be great, but its size is considerably bigger than 3x TO220 which is the real estate I can assign to the task. A 5 Amp diode in SMC is quite big on its own... The LT1185 barely fits, and the main trick is to reuse the large GND trace, because Vin is connected to the tab. But I will rethink this possibility.

Then that's great, I have a viable candidate for the charger 2.0. I'll also try to cook something up on UC3843.

If self-limited, then that's acceptable. The poor NPN didn't have a limiter...

I assume 3V, but the LM317 was there just to show the key part of the new arrangement. LT1185 has 2 more pins and is negative. ;-)

The LT has 0.75V@3A and even if the Schottky adds another 0.75, it's just 1.5V -- 2x better than an unprotected LM317 at 2x higher current. It's pretty high, but I can live with that.

Best regards, Piotr

Ok, let me interject here. Contemporary synchronous buck regulators have a boost function which allows the use of NMOS throughout, for the reasons you have stated. If you have many of them on one board you could even contemplate a separate supply for the boost pin. That allows you to get to almost 100% duty cycle which means almost zero dropout.

What is IMS?

Whoops :-)

Nah, easy. I just designed a module with four switchers on there and I've got a 12V/30W switcher in the mix. It is oversized so could easily do twice that power. I just took a look at the layout and that whole 12V section isn't larger than 3x TO220. With switchers the external FETs do not have to be TO220, you can get LFPAK and even smaller. No diode in there, either.

That one is a bit long in the tooth, maybe use something more modern and with tons of gate drive.

Check out this stuff, just as an example how small things can be these days:

Yes, but LDO's aren't what I'd use in this case.

Just keep in mind that since you are designing for very high MTBF, every degree this runs cooler will add to the reliability. Less thermal cycling, less problems down the road.

--
Regards, Joerg 

http://www.analogconsultants.com/

1.5mm of aluminum + 100um of FR4 + 35um of copper:

I have no access to the Al2O3-insulated variant, but even the FR4-based one is so much better than generic FR4 in heat conduction. Soldering is a nightmare for exactly the same reason. :-)

Sure it is, but it has two advantages:

1. I already have 10 of them on the shelf. ;-)
2. It has a disconnected totem pole driver, while the remaining more modern switchers I own have their integrated MOSFET connected to the Vin line and in my case the Vin line is not filtered, which can cause a problem. IMHO I should provide a small low-power supply for the switcher and use an external NMOS to handle the power part.

Extremely nice pinout. I have LM2696 capable of 3 amps, but it is almost unroutable on a single-layer PCB and on IMS a via is definitely not what you want.

I am not sure whether I can produce a reliable design of such an SMPS. I routinely use switchers, but mostly in arrangements very close to their application notes. This one is so much different -- the + pin of the accu is directly connected to the + side of the rectifier and the main NMOS is on the GND side. This is what I can't change. It means that the switcher must produce positive voltage which follows closely the shape of the rectified voltage in order to maintain the desired 13.8V difference. Not that hard, I can use a TL431 between the accu's terminal and a current mirror composed of 2xBC857 in order to bring the TL's state to the GND side. But how should the inductor and the diode be connected? Cathode to the common +, anode to the NMOs' drain and the - of the accu via an inductor to the drain and a capacitor parallel to the accu? And it must survive the reverse polarity in the Vin < Vbat sector.

I've invented something like that (assumed 5V instead of 13.8V for easier dividers):

but the simulator produces crazy output.

True. I would very much like to use a switcher, but I must be sure the design is safe. I don't want another explosion, so the linear regulator is just safer to me.

Best regards, Piotr

EMC will probably be an issue as well, I really don't like single-layer board. At least not for hi-rel stuff. A linear regulator would be tame in that respect but often they aren't when it come to susceptibility. Yet that's a test all of my designs have to pass. So the most I can use is a board with a fat copper core (expensive) or use 4oz copper.

The small low power supply is a good idea and it is almost the only way to really reach 100% duty cycle, so you can milk the input side to the limit. Provided there are no power factor requirements.

Thing is, with efficient switchers you may no longer need all this IMS stuff. My recent design delivers a total of up to 80W at various voltage levels and it all lives in 2" by 5" board space. I don't have it here but my client does and their engineer said that none of the semiconductors become too hot to touch, after running full bore for half an hour. We could run it well above 100W and it'll be ok.

Well, at some point one has to leave the trodden paths and head out into an adventure :-)

As the president of one client said, "the most fun projects are those where you get a serious knot in your stomach after making the commitment".

Can't quite follow here, I don't see any roadblocks. Why does the positive terminal of the battery have to be connected to the source? I thought that was only because you did not want to use P-channel devices. If you make a sync buck switcher with wo N-channel FETs, why not have this regulator in the positive path?

Of course there are also methods using a bridge converter and synchronous rectifiers that could go into the negative path but this gets more expensive.

Maybe because it has no compensation? It should be possible to make something like this work but there will be the losses in the diode.

You should go into the FB terminal with the feedback. Also, it is usually cheaper to use an optocoupler even though you don't need isolation.

In my younger days I had more linear regulators explode on me than switchers :-)

If you use a peak current controlled architecture there really isn't all that much that could go wrong in terms of a big bang. Of course, one must watch that the battery will never become overcharged. But that risk seems higher in a linear regulator, for example if the transistor shorts out (that happened to me).

--
Regards, Joerg 

http://www.analogconsultants.com/

Could it be inversion in the feedback path? Try feeding the current mirror into the FB pin instead of COMP pin?

No, I don't care about PFC. One specimen only, never going to be sold.

Not here, but there are 20 more heaters. :-)

4 big MOSFETs in the synchronous rectifier + 16 channels on OMNIFETs (basically NMOSes too). The IMS radiator is there for them, the ability to cool the charger is just a bonus.

Because a diffenent MOSFET is expected to be of N type. The big fat one which bypasses the charger completely, providing an 800uOhm path between the - terminal of the battery and GND. It must stand 40A and since P=I^2R, a PMOS is a no-go. It's the R12 part on my first diagram. Moving that NMOS to the high side would require more positive voltage than the most positive potential available, so at least one more charge pump or fancy voltage doubler. If this NMOS is on the low side, driving is easy, but the charger becomes a challenge.

LTSPICE gets really slow when optos are used. :-) In a real device I'll use an optocoupler.

Best regards, Piotr

a single part:

-Lasse

That's a lot. I'd consider a real heatsink, it's cheaper than this specialty stuff.

Why? Paralleling three of these would get you there:

It is not good to send 40A through a single device anyhow. If you absolutely want N-channel a small voltage generator would not be a big deal.

If you absolutely want N-channel a small voltage generator would not be a big deal. Much less of a challenge than a negative side switcher. Since this positive helper voltage does not even have to be regulated it would fall under the category "piece of cake" :-)

Then it wold be good to at least model it with a transfer function. But you need to go into the FB node and use the comp node for what it was intendedto do, compensation. Otherwise it is tough to get this stable.

--
Regards, Joerg 

http://www.analogconsultants.com/