power supply woes

[KTP]

I just added up the power supply requirements of the RGB led boards I am making and calculated worst case with each led element on full at 50mA, 1/6 duty cycle I will be drawing 3.6 amps at 5V (if I use 5V as the drive voltage for the allegro driver chip).  Ok, the kicker here is I need 80 boards...er, worst case then for each element of each RGB led fully on would be 288 amps at 5V! 

Since I want to power these off of at most 4 strings of cables, that would be almost 60 amps per cable...very beefy.

Sooo....I am thinking about incorporating a 24V to 5V dc-dc converter chip on each board, figuring that at 80% efficiency, I reduce the current to a more manageable 1 amp per board and 20 amps per each of the four strands.

The problem is I am having difficulty finding a dc-dc converter chip that can handle 3.6 amps and only requires a few external components.  Plenty that handle 3 amps, and some really expensive ones that handle 5 or 10 amps.  I also would like it to be adjustable such that maybe I supply the led driver with a bit less than 5V to limit waste.  I am looking at the LM2576 but it is only 3A.  http://www.onsemi.com/pub_link/Collateral/LM2576-D.PDF

Suggestions?

[Chasm]

Can you split the power supply on the board even more?
RGB, three converters. - but that is also getting expensive, fast.


Did you seee the TPS5450?

[Psi]

what kind of price range are you looking for?

http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=MC34167TVGOS-ND

Internal FET but you're still going to need diode and of course inductor.

[KTP]

what kind of price range are you looking for?

http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=MC34167TVGOS-ND

Internal FET but you're still going to need diode and of course inductor.

Thanks.  I missed that one because I filtered the 17,000 converters at digikey with "in stock" and "smt" and the smt version of that part is non stock.  I'd like to keep the total cost of the converter under $5 including catch diode and inductor, so that one fits the bill I think.  The whole board is getting really expensive (144 RGB leds, xmega micro, 6 allegro serial shift register drivers, RS485 communications).  My wife wants 80 of these boards....sheez.

[Psi]

Another option is to buy prebuilt dc/dc modules, you can get quite cheap ones considering they include the ic, inductor, diode and all other passives.

This one is 4.5v-13.8v input and 5V at 10A output (almost enough to run 3 of your boards with one module)
Effieiency is good too over 90% when load is 5A and above. Worst case 75% at 1A load
and it only costs  $9.50  or $7.6 in qty above 50

http://search.digikey.com/scripts/DkSearch/dksus.dll?vendor=0&keywords=507-1384-ND



That exact model is probably not an ideal choice as you want 24V input but it looks to be cheaper and easer to run 2 or 3 boards off something like this than build a psu on each board yourself.

[Zad]

Dare I ask what sort of project she has in mind that uses 1.44 kilowatts of LEDs? Might it be possible to serialise chains of LEDs so they may be driven by a higher voltage?

One tip that was kinda hidden in Dave's latest video is that if you are looking for high power devices, then they are unlikely to be surface mount.

[CafeLogic]

When you get over three amps internal switch DC/DC converters become impractical. What you need is typically termed a DC/DC controller. The difference is that the transistor is external. It's not a big deal. It is basically the same exact process with one extra component. The LM25088 would be one easy to use example. If you are really on a budget, and have room for large components, and don’t need high efficiency, you can use the MC34063 which does internal or external switching. If really don't have any room or you want to power multiple boards with one supply you can look for a synchronous buck converter IC. That is a buck converter with more than one inductor feeding into one bank of capacitors. The inductors are fired in a timed sequence and the result is the same as if you had one inductor working at twice the frequency which, of course, lets you use capacitors half the size. Synchronous bucks can pack huge amperage into tiny spaces. To give you an example, your typical $100 enthusiast “overclocker” motherboard or over-powered video card can deliver 100 low ripple amps to a processor core.

[Time]

Put these already-built-to-220-compatible ones in parallel?

http://www.dimensionengineering.com/DE-SW050.htm

Or maybe this guy?

http://www.dimensionengineering.com/DE-SWADJ3.htm

[CafeLogic]

I also forgot to point out that supplying those drivers with 5V will make a whole bunch of heat for no reason, try 3.6V. The other thing is, PC power supply's make a great way to get lots of DC amps form the wall socket, in case you didn't have a solution for that.

http://www.txmicro.com/680W-20-4-pin-Dual-Fan-ATX-PSU-w-SATA-PCIe-p-5013.html

[Eliminateur]

the problem is that "modern" ATX PSUs are made for 12V power, 5V is "deprecated".
(current PSUs have maybe 10A tops on 5V, maybe a tad more)
You'd need to find one of the P3-eras PSU that used 5V bulk(i have server PSUs from that era that push 35A @5V, 18A @3.3V and 9A @12V )

anyway, my two cents

[CafeLogic]

Specs from the one posted above:

# +3.3V, 30A
# +5V, 35A
# +12V, 38A
# -12V, 0.5A
# +5Vsb, 2.5A

I was referring to the 12V rail anyway since he wanted to limit the wire gauge.

[Zero999]

An efficient SMPS can be made using a couple of transistors. Obviously you'll need to use a larger transistor than the BC327 for 1A but the principle is the same.

http://www.romanblack.com/smps/smps.htm

[KTP]

The RGB matrix boards will be signal flags on our sailboat (so actually two boards per flag, 40 flags).  I need to be able to power them off of several deep cycle marine batteries (maybe several paralleled banks!   ).  I told my wife there is no market for this type of thing, but she really wants to build them so I am reluctantly agreeing.  The price per flag keeps creeping up the more things that get added...now another $5 to $10 for a smps on top of the $50 or so in parts for EACH side of the flag.  I do have to admit that it will probably look pretty cool on the water though..especially with every single RGB led on every flag controllable.  Here is a mockup I did using strip RGB leds from a roll bought off ebay.  I forget which marine flag letter this represents...'C' I think.  It was tough taking a picture because the thing is so bright, so this doesn't really do it justice.

[KTP]

I also forgot to point out that supplying those drivers with 5V will make a whole bunch of heat for no reason, try 3.6V. The other thing is, PC power supply's make a great way to get lots of DC amps form the wall socket, in case you didn't have a solution for that.

Yes, I am considering knocking down the led voltage supply to the drivers from 5V to something less.  The red diode in the RGB led only has a 2.4V max drop, but the green and blue have a 3.6V max drop, although typical drop is 3.3V.  Since I am pulsing these at 50mA, it probably will be closer to the max than the typical (I should actually measure this using my oscilloscope on a few of the leds I have).  The allegro A6282 serial shift register with current control which powers the column of my matrix will have some Vds drop.  The data sheet says this should be in the range of 1 to 3V, but I think it could get by on 0.6V or so looking at the curves.  There is also a small amount dropped across the Vds of the row mosfet (I am using IRF7104 P channel fets for these) but it should be small...say 0.1V or 0.2V max.  Add all this up worst case, you get 4.4V, so the 5V orginal plan wasn't too far off.  Still, some savings of waste power with a 4.5V supply or something might be good.

[CafeLogic]

I also forgot to point out that supplying those drivers with 5V will make a whole bunch of heat for no reason, try 3.6V. The other thing is, PC power supply's make a great way to get lots of DC amps form the wall socket, in case you didn't have a solution for that.

Yes, I am considering knocking down the led voltage supply to the drivers from 5V to something less.  The red diode in the RGB led only has a 2.4V max drop, but the green and blue have a 3.6V max drop, although typical drop is 3.3V.  Since I am pulsing these at 50mA, it probably will be closer to the max than the typical (I should actually measure this using my oscilloscope on a few of the leds I have).  The allegro A6282 serial shift register with current control which powers the column of my matrix will have some Vds drop.  The data sheet says this should be in the range of 1 to 3V, but I think it could get by on 0.6V or so looking at the curves.  There is also a small amount dropped across the Vds of the row mosfet (I am using IRF7104 P channel fets for these) but it should be small...say 0.1V or 0.2V max.  Add all this up worst case, you get 4.4V, so the 5V orginal plan wasn't too far off.  Still, some savings of waste power with a 4.5V supply or something might be good.

I took a look at the datasheet for that part and you are right. According to the graph on page 8, the current will start dropping out around 1V at 50mA. That really kind of sucks. Have you considered something like this:

http://us.element-14.com/allegro-microsystems/a6279eettr-t/ic-led-driver-constant-current/dp/19M5477

It is cheaper and it goes down to 0.15V. Why not just use a better PFET? I am guessing you have an extra reel or something. If you have all the leds in that column on (16 x 50ma) * 0.5Ohm RDS on = 0.4V drop. If you aren't married to it, I will share a Mouser secret. For some reason, all of the good buys on small smt power transistors are under the "small signal" category.

like this one:

http://my.mouser.com/ProductDetail/ON-Semiconductor/NTR4171PT1G/?qs=sGAEpiMZZMv4eh0jmGe024dsADQBMjpgoKQCXzVTxnQ%3d

14 Cents and 90mOhms at 2.5V gate drive

[KTP]

I took a look at the datasheet for that part and you are right. According to the graph on page 8, the current will start dropping out around 1V at 50mA. That really kind of sucks. Have you considered something like this:

http://us.element-14.com/allegro-microsystems/a6279eettr-t/ic-led-driver-constant-current/dp/19M5477

That is the first part I considered, but it is labeled as obsolete and a "last time buy"  I guess the A6282 with poorer specs is replacing it    They both have the same pinout though, so I *could* try to use the better part and fall back on the A6282.

As to the Pfet...busted!  I did have a reel of the IRF7104, but for the cost, you are right it might make sense to just buy a better one.  I really don't know if I am going to get reels of all of these parts or just cut-tape.  I will probably hand solder the first couple boards, but after that I may go crazy.  Still, getting the paste mask and setup for pick and place for only 80 boards will add another $1000 to $1500 to the costs.  Sigh...

[CafeLogic]

I think it is time the wife learned to solder, since she is the motivating force here. 

Try this: http://www.silvtronics.com/order-pcbassembly.php

they might be able to do a bit better than $1000-$1500

If you decide to go the toaster route, I use these dirt cheap mylar stencils: http://www.pololu.com/catalog/product/446
They work surprisingly well.

I don't know that the A6282 isn't as good, you are just driving it to the limit, that's why it has the high dropout. Everyone makes those drivers, ST, NXP, and TI so you have tons of options.

I hope those flags are really impressive when your done.