DC DC Converter

Why you build such power supply while you can buy similar one for $20 ?

I build only thing not available for me in the market.

I think there is limitation for Buck converter regarding the maximum power you can get out of it, 13.8V x 20A =600 Watt is too much and out of its limit.

I believe the maximum limit of it is about 150 Watt.

Why ,i don't know,don't ask me.

First, 13.8V x 20A = 276W not 600W. Second, one can easily just purchase a 500W DC/DC converter. One can even buy a 5000W DC/DC converter if one really must have one.

sorry for the error, i mean about 300Watt which is too much for Buck converter,500Watt and 5000 Watt are not Buck ,could be Full Bridge or some other topology.

Oh I see now. You refuse to click on my links because you are certain that a 500 watt or even a 5000 watt buck converter cannot possibly be manufactured. Well I learned long a go to not argue with a fool that cannot face the truth. So go and run along in your anonymous fantasy world. Don't look at anything like a research article from Power Electronics Technology where a 100 kilowatt DC DC converter was made. Go along and live peacefully.

Thanks for updating my knowledge,the information i had from text books may be 20 years old.

Now, could you tell me how come the Buck Converter power capability has been increased from 150Watt to 5000Watt during the last 20 years, what are the new parts that made that progress possible ?

Thanks again for updating my knowledge.

The basic improvement has been from an increase in the current handling capability of transistors. I'm designing up a custom power supply design now that will control 500 amperes of current. 20 years ago the only semiconductors that could control this amount of current were silicon controlled rectifiers (SCR) that worked great for controlling when to turn ON but could not easily control OFF a flowing current. Today the current handling of a MOSFET has dramatically improved. One can also parallel a MOSFET for higher current handling capability. You will remember that if one tries to parallel a BJT without some complicated temperature measuring circuitry one doesn't get a real sharing because the transistor with the most heat dissipated has the lowest CE voltage drop. But a new switch topology has also been fabricated in the mean time called Isolated Gate Bipolar Transistor (IGBT) that combines several input impedance advantages of a MOSFET but with much higher amounts of current handling capability. There are several IGBT designs today that switch kiloamps of current.

To a lesser but not insignificant extent there has also been an improvement in core materials and fabrication over the years so that the higher switching frequencies used in today's switchers will have much lower parasitic core losses like eddy currents.

Well, what about the large spike generated when the MOSFET turns off in case of Buck converter ?

There should not be a large voltage spike if the fly-back diode turns on to maintain a current path for the inductor.

Can you refer me to a good FET to be used with Buck converter 20-30 amp.

Actually with only the single parameter of 20-30 amp drain current I can't give you a specific choice. But I went to Newark.com and this page of 30-40 amp MOSFET N channel transistors gave me a list of 440 different discrete transistors. Now you should also consider the threshold voltage, the V_DS hold off voltage and the ON value of R_ds to see which of all of these could be suitable. You should also indirectly consider the switching frequency to determine the suitability of the gate capacitance being charged and discharged by the controller.

As redfred said, it'd be helpful if you included a schematic. Yes, the '3842 is a widely-used classic jellybean converter IC. But it's best used for current-mode flyback, or boost step-up, rather than buck step-down conversion, especially when using only an inductor rather than a transformer. That's why there aren't any standard circuits in '3842 datasheets and app notes for high-current 24 to 14V buck conversion.

I see you like single-side PCBs. These create an interesting layout challenge, especially if you want to avoid jumpers, but a double-sided layout would give you room for fat traces, ground planes, etc., better suited for a high-power 280W design. I also see you like rows of paralleled capacitors, visually appealing? :-)

OK, to your question. Obviously you'll be needing higher currents in your inductor, so you'll need a bigger inductor, one with less inductance but with much larger wire and probably a larger core. If you can increase the switching frequency that can help keep its size in check, but it's still an issue. You'll need to run the MOSFET at higher currents, and you'll have 4x higher ripple currents, which means more low-esr capacitance. And your catch diodes will overheat. Fixing all this takes space your PCB doesn't have.

The bottom line is your nice little PCB layout will be toast. Unless, that is, you try plan B. Plan B is to use three or four of your little 5A circuits in parallel. Build more copies of your board, and place them side-by-side. Just think of all those rows of capacitors to admire. In this scheme each converter will have a slightly different idea about the correct output voltage. The one that thinks it should be higher shoulders the burden up to its current-limit, and then the next lower one takes over adding more current.

This brings us back to your circuit. As a flyback or boost converter the '3842 operates in current-mode control. In this mode its output appears as a controlled current source, and with a few extra parts it can also be current limited, which is good for parallel operation with other converters.

But you've set aside current-mode-feedback regulation, and are using the '3842 rail-to-rail output driver as a buck converter, adding an n-channel MOSFET source follower and freewheeling catch diodes. The circuit seems not to have any current limit, is that right? If your circuit uses voltage-feedback regulation, and the output is too low it'll just keep the MOSFET turned on longer, attempting to raise the voltage. That would be very bad for experimenting with parallel operation. TI suggests two voltage-mode circuits (page 15 of their 2002 app note), shown at right, but your circuit board doesn't seem to use one.

An alternate approach would be to forget tightly controlling the output voltage and use a fixed duty cycle. E.g., 58% of 24V is 14V, at no load. The inductor and output caps simply act as an averaging filter. If you use that approach you can safely parallel the outputs. But once again, if there's an excessive load, you won't have current limiting to help save your ass. Either way, you'd better add input fuses for protection.

The duty-cycle approach isn't bad. You showed us your PCB. OK, here's my PCB for a 300-watt step-down or step-up converter that works with 20 to 50V on the high side and 5 to 20V on the low side. It's basically a high-current adjustable-ratio DC transformer. For example it can act as a charge equalizer between 14V and 42V batteries. I call it a bus converter.

Cool, huh?

My circuit does have current-limiting protection, but I also have an automotive plug-in fuse on the board (upper left corner). I use two-phase conversion with an LT3782 control IC. The power parts are on the top (red) and the control parts are surface-mount on the bottom (green). There's an 8-pin uP socket (upper right corner) serving other functions.

Like you, I also have a row of capacitors to admire. :-)

Oh, sorry, I put up the wrong PCB. That was a 300W 14V to 42V converter, but one way (step-up) only.

Here's my adjustable DC-transformer bus converter. It uses an IR2085 to create the voltage-ratio duty cycle, and is rated at about 120W. It does have current limiting, but it also has a fuse.

The pic shows the power parts on the top side; the 4-terminal current-sensing resistor and the control parts on are on the bottom (not shown).

Win,what is the name of that nice CAD software suitable for switching power supply design ?

I use Altium's PCAD, which they discontinued a few years ago, in favor of Altium Designer, which I'm learning. But I suspect most PCB CAD programs can do a pretty good job with switching-supply designs. The thing is, with all of them, you'll have to create your own custom parts. For example, I had to make inductors, fuses, 4-terminal resistors, etc. So one issue is, how hard is it to make new complex library items in your CAD program. There are a few other issues, like the easy of doing automatic copper pours, etc.

For 24V DC/120V DC Boost coveter , what is maximum power can i get out of it for single unit ( without parallel connections) using today's available technology ?

If the "single unit" is a half-bridge or H-bridge off-line forward converter operating in continuous mode you can easily get to 3 to 5kW with a single transformer and the associate mosfet switches. But with a 24V rather than 340Vdc input, 5kW would imply over 200A, which is a bit much, so we'll back down to say 2kW max. Now, if you're asking, how much power can I convert with one mosfet and a simple flyback inductor, well, that's another story. Let's call it no more than 250W.

My 300W 14-to-42V converter pictured in the PCB layout in my first post, is a two-phase converter using an LT3782 controller IC, $9.63 at DigiKey, so each mosfet and inductor handles 150 watts. This is a current-mode controller, so paralleling outputs is fine, and if the sections are offset in time this reduces the output ripple current.

I suppose I could have pushed the design higher, but it seemed a sensible design. I de-rated it to 200W when used at elevated automotive engine-compartment temperatures. My biggest struggle was finding suitable low-esr high-temp output capacitors.

I see that LTC recommends their new $7 LTC3862 controller in place of the LT3782. See typical app dwg above. Introduced last year, it's suitable for use in multi-phase converter designs, nicely making a 4-phase design with two ICs.

They say it's suited for up to 12 phases. At 150W per phase, that'd be a 1.8kW converter, but I suppose you'd cry foul and assert the 6-chip 12-mosfet 12-inductor monster wasn't a "single unit."

Thank you for all, i just wonder regarding those replays, any way come to the point,,

now i added the schematic in that i am not used any fuse b'cause my equipment is highly sensitive that's why i use over voltage protector. my PCB and schematic not matched i know that .first i worked with common PCB, sure it is current mode design.

It's very hard to read your schematic (as I suppose you will see, trying yourself), because CR4 so badly downsamples submitted image files. But, peering at your PCB and dim drawing, I don't see a current-sensing resistor.

Wait a minute, is that a resistor in series with the catch diodes in your drawing (but not in your PCB layout)? Hmm, oops! Such a resistor would make a negative current-sense voltage, on the second, switch-off part of the cycle, both of which the UC3842 can't deal with.

But, wait! I can barely see the chip on CR4's copy of your schematic and it's not a UC3842, but an LTC part, an LT1242, is that right? Aha! Now this is getting interesting!! Even tho it's only an 8-pin chip, LTC has added some useful features over the original uc3842. For example, if you use an LT1242 instead of an UC3842, you can run it at up to 500kHz. Very nice, and well suited to increasing the energy you can get a small inductor to deliver.

But, wait again, I don't see that the enhanced LT1242 has the ability to work with negative CS on the OFF part of the switching cycle. Nope, sorry. So now I'm puzzled as to how your drawing could work, something is wrong. You really want a positive voltage on pin 3 during control-loop operation. LTC doesn't have an offset-voltage spec for pin 3, so I image some parts might seem to work with pin 3 = 0V, but whew, I dunno. Not good.

SFAICT, you still have a voltage-servo'ed output, not using current-mode, and no active current limit, so you can't simply parallel a bunch of your boards. BUT, depending on how your control is implemented (I can't make it out), you might be able to use one controller IC to control the rest of the boards, by routing a wire between them at the right place.

Perhaps you can send me a clean copy of your actual schematic by email, annotated if necessary, and I'll take a look -- winfieldhill 9at) yahoo (dot) com. Hah, that should confound the email harvesting programs! Also, an updated PCB image would be helpful.

Thanks.... I have tried to send you an email with a clear schematic diagram along with comments. But the mail got bounced.. please advise your correct email id. Or you may just send me a mail at below id

mail at hifiworld dot info

Hope to get your reply.

Please allow us to share you discussing the problem of mano's converter design and the way to solve it ?

Certainly. I studied the drawing and thought about ways to try using a UN3842 boost converter as a buck converter, and in the end realized it's a really bad idea. The lack of a proper positive inductor-current signal during the switch ON time is one problem, but another more serious one is that an n-channel mosfet can't be properly turned on, with a gate voltage 10V or so higher than its source, and hence its drain = 32V, so you need gate = 32+10 = 42V, etc. Once I realized that, I threw in the towel. The design might sort of work in spice, but in real life the mosfet will overheat.

See the post I just placed, at the end I wrote "What about current-mode buck converters? After a bit of thought, my advice is to just go ahead with a proper full-fledged current-mode buck-controller IC, such as an LTC1624 with an n-channel mosfet and a catch diode, or an LT3741 with two mosfets. The latter is better for high currents, since you won't suffer from the catch-diode drop and dissipation. Also, you can parallel multiple circuits to get more power, if you need."

"Wait a minute, is that a resistor in series with the catch diodes in your drawing (but not in your PCB layout)? Hmm, oops! Such a resistor would make a negative current-sense voltage, on the second, switch-off part of the cycle, both of which the UC3842 can't deal with."

The UC3842 and SG3842 series of current-mode controller ICs were advanced when they were introduced nearly 40 years ago, and have become quite popular in the years since. First, that's because the current-mode operation with cycle-by-cycle current limiting is a good type of regulation, and second because the part quickly developed many second-source suppliers, and look-alike and improved versions (I must have nearly 100 different '3842 files in my computer). This also meant the part has become very cheap, about 15 to 25 cents or less in quantity.

But one thing has always frustrated me about the world of '3842 types, which all have a current-sensing circuit like the one above: the voltage range is rather high, up to 900mV, and the comparator doesn't have a tight offset spec suited for lower sense voltages. This means the part wastes excessive power in the current-sense resistor for low-voltage designs (this isn't so bad for high-voltage off-line designs). Also, the part doesn't have a differential input.

But it's very cheap and ubiquitous, and works well, so one is motivated to fix the problem. My solution (right) is to add a low-cost single-supply op-amp to the circuit, amplifying the sense voltage and ameliorating ground bounce issues at the same time.

Like the UC3842, the LM358 is widely sourced and very cheap, under 10 cents, making a suitable companion. It can handle up to 32V for its supply, or better yet, power it from the UC3842's 5.0-volt reference output. If you want a smaller part with better accuracy, and prefer not to waste the unused 2nd op-amp, an LMC7101 is a good choice, a TLV2221 would be another. (The mosfet should have a gate-source resistor (not shown) to insure it'll be OFF if the UC3842 is unpowered.)

What about current-mode buck converters? After a bit of thought, my advice is to just go ahead with a proper full-fledged current-mode buck-controller IC, such as an LTC1624 with an n-channel mosfet and a catch diode, or an LT3741 with two mosfets. The latter is better for high currents, since you won't suffer from the catch-diode drop and dissipation. Also, you can parallel multiple circuits to get more power, if you need.

Thanks for the good explanation,but practically which is better, should i run Buck converter in continuous mode or the discontinuous one ?

Ah, it's certainly a function of the exact voltages involved, but in general for power levels above 50W to 100W, continuous mode is a big advantage. Certainly at high power levels it's essential. As for the current/power mode threshold, in my case it's usually been a function of specific details like commercial inductor availability, etc. Let's hear from the experiences of others on this subject.

Thank Win,i have another question regarding Buck Converter,can the Buck converter to be designed as a constant current source rather than as constant voltage source ?

Yes, indeed. I suggested a "full-fledged current-mode buck-controller IC, such as an LTC1624 with an n-channel mosfet and a catch diode, or an LT3741 with two mosfets," pointing out that both are current-mode controllers, which means they both run their control loops based on cycle-by-cycle current limiting. So in the middle-frequency region (just below their switching frequency) they naturally look like current sources (at very high frequencies they have large capacitor outputs, which of course means they look like voltage sources up there).

It's true that under normal use their low-frequency control loop seeks to adjust the cycle-by-cycle current to enforce a fixed output voltage, but that's simply because we take a pair of voltage-divider resistors from the output and hook this up to the controller's FB or VFB point. If instead we take say the LTC1624's current-sense resistor located in the drain of its top mosfet, and we hook up a difference amplifier or an instrumentation amplifier to it, we'll have an honest current-feedback signal. And if we connect that to the FB terminal, now we will have implemented a power current source.

Thanks lot Win for your time and effort.

Regards

That's OK, thank you for the interesting questions. Y'all can just vote me a Good Answer for one or two of my detailed posts that were helpful!

I was looking at the Mouser catalog, and as long as we're talking about the 100W to 500W territory, but still looking for jellybean classic easy-to-get low-cost parts, I think it's time to start thinking about real controllers, with push-pull mosfet forward-converting transformers, etc. For example based on the UC3825 family. Or some of its offspring, like the ON Semi MC34025 (links).

Hey guys, it's really not that hard to hand-wind a transformer when its running at say 200kHz and has a small number of turns.

I have an extra question please Win:

What are the advantages and disadvantages of using higher speed switcher ?

"What about current-mode buck converters?" I have heard further from manojtm, and it's clear that he'd like to use an UC3842 controller IC for his buck converter application. This is possible, but it presents us with several problems to solve.

The first is that we need high-side switching, whereas the UC3842 gives us a low-side pulse (we can't simply use an n-channel follower, because we need to provide 10V gate drive). The second is that we'd like to preserve current-mode operation, so we need a way to measure the inductor charging current and present it to the IC. Alternately, we could use voltage-mode control (and give up on paralleling PCBs), which can be invoked by adding parts as shown in page 15 of TI's 2002 app note, see post #9.

We'll take up the gate-pulse issue first. In the drawing below I've added a three transistor level-shifter to create a negative gate pulse, so we can use a P-channel high-side mosfet switch. In this circuit the bottom NPN transistor acts as a current sink to create a negative pulse across its collector resistor. The two resistors should have equal value. The NPN + PNP emitter followers conduct high currents for rapid charge + discharge of the mosfet's high gate capacitance, ensuring fast mosfet switching.

Next we'll deal with the current-mode issue. The UC3842 wants to see a rising inductor current during switch-ON, which it uses to control the ON time and regulate the loop. We can use a trick, and provide this in a buck converter by sensing the charging current in the output filter capacitors. But we don't want much voltage drop in the process, because now this is output ripple we're talking about, so once again adding a current-sense amplifier makes good sense. :-)

Note how the output filter capacitors are divided into two parallel groups. One capacitor is used for filtering, with a series resistor Rs for current sensing, and the rest (e.g., 3 caps in manojtm's PCB) are for energy storage and filtering. This group of output capacitors also needs a series resistor, e.g., Rs/3, to insure a frequency-independent current-ratio for the two paths, e.g., 1/4 to the single cap and its sense resistor. This second resistor also acts to filter out the Rs current signal from the output. To work well, Rs should be larger than the capacitor's esr.

With this type of scheme, the UC3842-based buck converter will be operating in current mode, with a maximum current limit at about Vs=50mV for G=20, and Icharge = 4 / G Rs, or 10A peak for Rs = 0.02 ohms. Multiple circuits can be paralleled to get more power.

Or, as mentioned earlier, we could use the UC3842 in voltage-mode.

I know this looks complicated, but that would be one motivation for using a more advanced PWM controller IC. OTOH, the UC3842 only costs 25 to 50 cents in volume (dramatically less than other controller ICs), and the other parts are also very cheap. All parts have multiple sources.

Here is the house of SMP's experts,i need your help:

I'm working on 1kw push-pull switching power supply design , I'm looking for suitable 2 heat sinks for the unit and to refer me to a supplier.

Thanks for help

"Or, as mentioned earlier, we could use the UC3842 in voltage-mode."

SFAICT, manojtm's application involves a 24 to 32V voltage-stable battery power source, powering a 13.8V transceiver with a wildly-varying current load. Current-mode controllers are good at isolating input-voltage steps (not a problem here), but not so good at dealing with transient load-current steps (a problem here), since they want to momentarily keep on delivering the same current. So it would appear voltage-mode control is preferred. The only problem with voltage control is difficulty in paralleling the outputs, which means we'll have to work until we have a single 300W version of the power supply.

Following the suggestions in TI's 2002 app note, slua257 (link), see post #9, we've sketched a candidate voltage-control-mode buck-regulator circuit. (TI suggests using their UCC38C42 version of the '3842, which looks worth looking into.)

We've used the RC OSC pin, isolated with an emitter-follower, plus a resistive divider, to drive the CS current-sense control with a 900mV peak ramp, simulating a rising inductor current.

We also added an inverting level-shifting gate-driver for a proper P-channel MOSFET buck switch. The driver is fast and can source-sink about 0.5A into the FET's gate. Note how the level-shift circuit uses a current sink, so a fixed-amplitude gate pulse is delivered to the mosfet, independent of the Vin voltage, from 22V or so up to over 50V. The rest of the power circuitry is conventional.

The control-loop compensation looks complicated, according to TI, but we don't need a particularly-fast feedback loop, and as a result we can no doubt simplify the circuit.

If we want to go for broke and create a single 300W buck converter, the pressure will be on us to find a suitable 20A inductor (or paralleled inductors), MOSFET and Schottky diodes, and low-esr input and output capacitors.

One big bonus, no awkward 2-milliohm 1-watt current-sense resistor!

A reader, Raffy, pointed out that the regulator ICs have a 16-volt UVLO under-voltage lockout. So the 12V zener diode will prevent the circuit from working. The zener is only needed if the input voltage can go above 25V for the LT1242 (or 30V for the UC3842), and if used it should instead be say 20V (or 27V). Alternately an UC3843 with a 9V max startup lockout could be used.

"with one mosfet and a simple flyback inductor, well, that's another story. Let's call it no more than 250W."

That figure '250W' is very close to what i said about the maximum power capability of single Buck/Boost convert, 150 Watt 20 years ago and now 250 Watt.

Why not a several MOSFETs in parallel can increase Buck/Boost converter capability to much more than 250W for a single unit ?

Agreed, MOSFET current handling capability is but one aspect of all of the other parameters that must come together for any buck converter design to be feasible. Also 250W is a good rule of thumb for an upper limit for a single stage buck converter. Higher amounts of power will be much more easily designed using different circuit topologies. But I don't believe that 250W or any number near this is the theoretical power limit for a buck converter.

I consider Buck/Boost converter is the most ever reliable converter, but unfortunately its power handling capability is limited to about 250W.

Very neatly summed up. I bow to the master.

Up to this moment 178 viewers have seen this thread only "two" of them shared with information and the rest preferred to take information only and not to give any thing.

Amazing !!!

185 viewers now. It's not clear how many found the subject interesting once they landed here.

Well, it's a rather esoteric subject, how to make a buck converter from a jellybean boost converter IC, and push it up to 300 watts. It continued in part because I found the idea intriguing, and because the O.P., manojtm, has a special interested in using the UC3842, and continued with offline email questions and encouragement.

At any rate shortly I'll make at least one more post to advance the topic towards a satisfactory finishing point, in my mind at any rate. Or maybe manojtm, or someone else, will have more to say, sparking further posts. That'd be fine. It'd be nice to find out how it all works out.

Pictured, 8-pin and 14-pin versions of the UC3842.

hi, win i completed my project . i tested 16A load with my circuit wow !!!! it can handle more than 20A or more if you use IRFP4668 MOSFET anybody interested i will post my final circuit thanks all

Yes, indeed, we're still interested. Please post away! What frequency does it run at? Tell us about the inductor(s), etc. The IRFP4668, $8 at DigiKey, very good. But maybe you don't need a 200V part - a lower-voltage part with the same R_DS(on) will have less capacitance.

But Manoj, after you place the post, check to see if you can read the drawing. If not, divide the figure into two pieces and post both of the new pieces as well. Note, when you make a post, you have 15 minutes to edit it and repost, and only the final post shows.

Manojtm , I want your complete project with irfp4668.thk