4

I'm doing an LED project with a 36V Cree CXB3590 COB LED that I intend to underpower to experiment with efficiency and learn about boost converter design. I've made a 5uH inductor with 14 gauge magnet wire wound onto a T50 toroidal yellow/white(iron powder) core. I plan to use a 555 clock circuit and a 555 pulse generator with a horrible oscilloscope to determine the maximum pulse length for the inductor core and then experiment with frequency to compare total light output from the CXB3590 at different power levels to the output from single LEDs such as an XP-G or XR-E at the same power level. I'll be using PWM with a frequency limit ~1Mhz due to the iron powder core inductor. The large wire on the inductor and a decent MOSFET are intended to keep resistive losses as low as possible, hopefully negligible.

Because the CXB3590 is capable of using much more power than the boost converter circuit is likely to provide, I'm hoping to use it as both output diode and load, making the assumption that with each pulse the inductor will drive voltage up to the on voltage of the LED (~33-34V) and when that voltage is reached because the LED's effective resistance will become extremely low, it will not rise significantly higher as the inductor dumps its stored energy.

I'm hoping that this will result in the LED operating in a favorable efficiency band, operating in discontinuous mode and at close to the lowest voltage that will allow it to dissipate the power being pumped into it.

schematic

simulate this circuit – Schematic created using CircuitLab

So my question is this: For what reason(s) I would benefit from using the topology shown on the right, such as the inclusion of a capacitor helping clamp down the voltage the LED sees? the CXB3590 is rated at 86.4W, so at the ~3-10W I intend to operate it at, as I mention above it will likely operate discontinuously.

MicroservicesOnDDD
  • 2,557
  • 1
  • 13
  • 48
K H
  • 4,023
  • 1
  • 11
  • 24
  • 1
    Don't have an answer, but this sort of topology is reminiscent of that used in the joule thief. – Hearth Oct 08 '19 at 01:05
  • @Hearth the COB won't arrive for about a month, so determining experimentally is an option, but I'd like to figure out as much as possible before then. I wonder if I should use a MOSFET or relay symbol for the switch to indicate the switch is pulse length fixed and frequency regulated? Unbelievably fast response. – K H Oct 08 '19 at 01:08
  • 2
    Looks OK. In 2nd cct as cap goes up in value you will convert to continuous LED action with ripple. CCt 1 will always be discontinuous. At low frequencies cct1 will always flicker and cct 2 may or may not deep-ending on cap. – Russell McMahon Oct 08 '19 at 07:22
  • 2
    LEDs have best luminous efficiency at low I. With pulsed I overall lumens will be lower - but not necessarily grossly so. If Pin is << PLEDmaxallowed all probably OK, but as P approaches PMax the pulses will cause power rating to be exceeded short term. Modern white LEDS have much lower Pabsmax/Pmax ratings than olde school LEDs or eg IR where high Ipulse is OK. "Fast diode" can be a FEt if desired to minimise voltage drop - but efficiency overall in this cct is liable to not be so marvellous than eg any Schottky of adequare rating will do. [I've used several million LEDs but far lower pwr]. – Russell McMahon Oct 08 '19 at 08:07
  • @RussellMcMahon Thank you! Good point. After reading your comment I realised that as long as I stay below the maximum pulse length, I would perhaps be better using PWM than PFM I think I got thinking about PFM because I may need to vary the frequency to find the acceptable pulse length part of my concern with using the bare inductor directly on the LED is that I don't know how fast the magnetic field will collapse and whether the LED's drop in effective resistance as voltage increases past the turn on point will respond as quickly. – K H Oct 08 '19 at 22:50
  • I was also thinking that my minimum frequency would be higher than 1khz, but there may be some merit to using PWM and keeping it arbitrarily high. – K H Oct 08 '19 at 22:53
  • 2
    Beware that your converter probably won't be terribly efficient. Or maybe even terribly inefficient. You're boosting by a factor of 20+ (assuming white LEDs), which is hard to do efficiently even with the best designs and components. Also, I suspect your target of 1MHz is pretty high, definitely experiment with lower frequencies. Finally, consider trying both topologies and measure which one performs better! – marcelm Oct 08 '19 at 23:33
  • 1
    @Marcel It will also be my first boost converter, so who knows? I hand wound the inductor with oversize wire, so unless it rings fiercely, resistor losses should be low. I'll probably start off with around 250khz. The COB contains many LEDs but I believe in S/P arrangement would explain the 36V operating voltage, so 2 or 3 to 1 boost depending on input voltage. 1Mhz is considered max sensible freq. For the core type I used so at that freq., core losses may be high. I'll probably try both but at my skill level, setup will be an effort so I may try the simpler one and sèe what the osc. says. – K H Oct 08 '19 at 23:46
  • 2
    Inductor current will be continuous across the switching transition. Voltage will rise to whatever is required to ensure this. On an open circuit that would be high enough to transfer the current into the stray capacitance plus I^@R losses in wiring and cap ESR = VERY high. With the LED there I_LED initial = I_inductor_at_switch_open. | With 12V in and ~= 36V out -> Pinstantaneous ~= 3 x Pin at switch open time. At say 1A switch turnoff, PLED = 1A and VLED is set by LED I/V Vf curve. Current and voltage will fall from there. ... – Russell McMahon Oct 09 '19 at 00:56
  • 2
    ... So LED POWER will peak at about 36V and fall across the inductor discharge so << 36W overall for 1A pulse. | At 1A in Iin mean is 0.5A or less (mean 0.5A during ton and 0 during toff with duty cycle of about 3:! at limit. So Pin About 6W at 1A so PLED less. So LED sees an about 36W pulse power for 6W DC in or 6:1+. So you can go to ABOUT 15W in without exceeding LED Pintypmax. Probably :-). YMMV. E&OE. – Russell McMahon Oct 09 '19 at 01:02
  • 1
    @RussellMcMahon Hmm it seems likely then that the second topology would perform significantly better unless I actively controlled both frequency and duty cycle to prevent unreasonable current. It had not even occurred to me that the voltage change could be that fast, but it makes sense if the current must be the same at time of switching. I'll start looking harder at the traditional topology. As far as the input power ripple, a big input cap will have to do until I figure out cuk converters. – K H Oct 09 '19 at 01:56
  • 1
    My inexperience is the reason I plan to rely on a fast diode rather than figure out how to high side switch a MOSFET above the source voltage while trying to learn the rest of this. As far as the output power spikes... Yeah that could really screw the efficiency experiment, especially given that I have no idea what to expect for current at switch turnoff without experimenting. I suppose I can actually calculate that based on time on, so NM. Looked up some terms... YMMV... Mileage may vary... Ok... Extremely disappointed E&OE doesn't mean "Experiment and Over Engineer"... – K H Oct 09 '19 at 02:02
  • 1
    @RussellMcMahon I'll see if I can use this info to improve my question a bit before bounty time comes and if you'd be willing to answer at that point, it would probably be very valuable to my fellow experimental hobbyists. If noone else does I'll probably take some of the information you've given and try and whip up a few graphs. Thank you very much for the help! – K H Oct 09 '19 at 02:05
  • 2
    @KH _"I hand wound the inductor with oversize wire, so unless it rings fiercely, resistor losses should be low. I'll probably start off with around 250khz."_ - I'm more worried about core losses, skin effect, parasitic capacitances, parasitic inductances in the rest of the circuit, etc. Tight layout is very important in switch-mode converters. _"The COB contains many LEDs but I believe in S/P arrangement would explain the 36V operating voltage ..."_ - Oops sorry, I missed that! Yes, boost factor 3 is a lot more doable! – marcelm Oct 09 '19 at 08:31
  • 2
    I actually built a similar thing not too long ago as an experiment, a DIY 12V -> 40V LED boost converter with some cobbled together components (including a probably not terribly suited inductor). I managed to hit maybe 70% efficiency, but it did work. The optimal frequency I found experimentally was nowhere near 1MHz though, but <100kHz. For the classic design, try to minimize the switch-diode-cap loop area. For your alternative design, your LEDs become part of that loop so I'm not sure what that will do. – marcelm Oct 09 '19 at 08:37
  • 1
    @marcelm I think I'll add that to the list of good reasons that the traditional layout is likely better. Your mention of loop length caused me to also realise that the ~150mm lead wires may form a nasty antenna in addition to contributing to parasitics. I'll be doing my initial build on a solderless breadboard, so loop length initially will only be optimal for that, thus nonoptimal. If It's functional, I'll acid etch a board for it. Thanks for the advice! – K H Oct 09 '19 at 16:45
  • 2
    There are numerous bad assumptions on core losses and current spectrum. When driving with a pulse, if the rise time is 5% of the cycle, then you have 20x harmonic content. Considering eddy current losses are f^2. You may want to limit the harmonics to 300kHz and thus operate PWM at say 15kHz in continuous mode. This is kind of a Mickey Mouse bounty question. – Tony Stewart EE75 Oct 12 '19 at 15:54
  • 1
    @Sunnyskyguyee75 The simple nature of the question is partly why I decided to put a bounty on it. I'm hoping for an answer that will be useful at hobbyist skill level, and I'm willing to write it myself if necessary once I've learned enough about the relevant concerns. For instance I had no idea eddy current losses would vary with the square of frequency, or about most of what Russel has mentioned. I know this is simple by many of your standards, but for me it's a massive list of unknowns related to a small project I'll really enjoy. – K H Oct 12 '19 at 18:07
  • 1
    So if the harmonic content is multiplied by the inverse of the length of the rise time as a fraction of a cycle, that's another reason to try to make sure there is little dead time in the cycle. I'll have to think about whether anything else is implied by that. Thanks for your input! – K H Oct 12 '19 at 18:10
  • I suggest you buy some inexpensive LED light bulbs from the dollar store and remove the LED circuit boards and use those for your initial tests, because it would be a shame to fry your nice expensive COB as you're learning. At high frequency, the reverse voltage might fry the COB. A 60W bulb usually has 9 x 6v = 54v, so scratch the insulation off of the appropriate connection so you can use six LED's for 6 x 6v = 36V of forward voltage so as to mirror the functionality of your expensive COB. Frying the dollar store piece would be much better, if that's what's going to happen. – MicroservicesOnDDD Jan 28 '22 at 01:39
  • You may need to use a heat gun, or an electric frying pan, to heat up the aluminum PCB so that you can solder to the scratched connection so that you can get your string of 6 x 6V LEDs. – MicroservicesOnDDD Jan 28 '22 at 01:45
  • 1
    @MicroservicesOnDDD Well the challenge I'm most focused on right now is building voltage converters. If I don't succeed in building something I'm comfortable running when I'm away from home, I'll buy a potted driver for the COB. That said, I've had moderate success with a Boost converter and a Cuk converter. I don't think I'm likely to put the LED at risk until I get better at selecting/winding my inductors. I've pretty much ruled out attempting to use the LEDs as a blocking diode at my skill level, so given that, the COB actually feels very durable on the workbench. – K H Jan 28 '22 at 04:44
  • 1
    I had a house move in the middle and haven't learned enough to report back here yet, but I got as far as building a control circuit to limit output voltage, vary operating frequency via pot, vary output current via feedback loop and pot. I experimented with connecting this control circuit to boost converter and cuk style output stages and was able to run the LED at a blinding 1/3 of its total power for a few seconds with each. My oscilloscope is not useful up to the inductor's max 1mhz, and cannot confirm my suspicion that the inductor core is not good for anywhere near that frequency. – K H Jan 28 '22 at 04:50
  • @KH Thanks for letting me know. I'm highly interested in your journey and progress, as I am on a similar one. – MicroservicesOnDDD Feb 22 '22 at 12:06

2 Answers2

4

ADD: The left cct cannot run in CCM (continuous conduction mode) because the switch also shorts out the LED to 0V. But besides that now you have a low ESR path with a discharged cap that draws as much current as the inductor can swing V+/DCR and causes a high Q current resonant cct and adding pulses adds fuel to the fire with burnout on the series resonant circuit.

That's a serious flaw on the left. 1st it starts a low frequency resonance, the as the current builds up in the inductor the cap voltage is discharged at the same time. So the series diode switch is essential by lowering the Q switching the Cap off and ON. and allowing the voltage to accumulate with current switched pulses **

Examine the impedance of every part and try to choose parts towards 0.1% of LED ON resistance. (incremental V/I) to minimize losses. ( i.e. 2 mOhm ballpark )

Assuming 36V LEDs: (although you would be wiser to use 72V LEDs)

Rled = 1.7Ω= 1V/0.6A = 35.5V/2A - 36.5V/2.6A (slope)
Fast Power Schottky Diode: ~ 2 mOhm @ 20A STPS5045SG-TR 50A rated
FET Switch: <5m Ohm = RdsOn @ 10Vgs

Operate gate drive frequency at 10k to 50kHz max ( lower is more efficient)
Operate duty cycle < 75% pref. <50% ON time.

enter image description here Choose large cores e.g. T50-26 = 0.5"D , AL= 33 nH/N² with 12 turns AWG16 magnet wire 3.2mΩ to get L = 33nH*240 = ~ 7.9 uH

Then with a very tight layout, attempt to achieve these current pulses in Falstad simulation.

LED Current = Cap Current (-ve ramp discharge) + Diode pulse current (+ve ramp charge)

Compute the efficiency from the floating power graphs with Average power.
+12V =-65.827 W
LED+ = 64.256 W
**loss = 1.571 W / 66W 100% = 2.5% @ 33kHz mostly in FET.
using low ESR , DCR, Rs (diode) components.

Tony Stewart EE75
  • 1
  • 3
  • 54
  • 182
  • 1
    I've been simulating a near identical circuit to the one you recommend and I've been learning a lot. I don't know how to simulate a not-purely resistive load in ltspice, but I managed to do basic voltage regulation by setting a clock pulse and having it skip pulses when output voltage is higher than desired and now I'm going to attempt current control while doing the same. You've been very helpful in terms of providing component values that will yield a functional hobby project design. There are 5 more days left on the bounty and I don't know if others will answer, – K H Oct 15 '19 at 02:19
  • 1
    but you haven't commented yet on why the circuit on the right would be your choice rather than that on the left. Is that because of the concerns mentioned in the comments just making it a really obvious choice to an engineer? – K H Oct 15 '19 at 02:20
  • 1
    Yes . The left cct cannot run in CCM (continuous conduction mode) because the switch also shorts out the LED to 0V. So the diode switches the current path and while raising the voltage slightly ---with a ramp -- with sufficiently large C, low ESR CAP – Tony Stewart EE75 Oct 15 '19 at 02:25
  • 1
    You can export a Falstad shortcut link in answers but the shortcut domain only works in comments. – Tony Stewart EE75 Oct 15 '19 at 03:13
  • 1
    I'll check out Falstad. I've never tried it before and just getting used to LTspice has been an effort for me. I'm not sure if you changed your screen name because you wanted to obfuscate your identity better, but that screenshot has your last name on it just so you know. I'll check out Falstad though. Very little chance it does logic gates in as stupid a way as LTspice does lol. – K H Oct 15 '19 at 03:16
  • 1
    Logic gates and Op Amps have slew rate control but 0 Ohms out, so add when Ron when desired. FET's have RdsOn controlled by Beta only 20m is like 300 Ohms, 10 is like 50mOhm – Tony Stewart EE75 Oct 15 '19 at 03:18
  • 1
    In LTSpice you start with a generic 5 input gate of whatever type and you have to manually specify pretty much everything, which was hard to figure out at my skill level. For most things I can use a chip model and look up the chip but they really could have made their default logic gates more intuitive. For op amps right now, I just use comparators and spend more money to get really fast rail to rail ones to make skill less of a factor so I can build more projects while in learning phase. I kind of understand what op amps do, like basically add and subtract voltages, but I'll have to – K H Oct 15 '19 at 03:24
  • 1
    learn some other stuff and then find and read some textbook chapters on them to really understand how to use them. I programmed a lot as a child so I'm much better with digital logic than with analog components. – K H Oct 15 '19 at 03:25
  • 2
    TO make Falstad more real, add 0.05 ~ 0.1 Ohm ESR to the 220uF Cap as per datasheet and < 10 mOhm for a good choke. – Tony Stewart EE75 Oct 15 '19 at 03:38
  • 1
    Congratulations on winning the bounty and thank you for your helpful answer. I may ask another thing or two in the comments once I understand this circuit better, but I'm learning a lot from your answer, @RusselMcMahon 's comments and experimenting. When my COB arrives in the mail I'll assemble it and perhaps add another answer with information on how things go. – K H Oct 17 '19 at 23:14
  • 1
    Design specs for DCR,ESR,Ron on each part and verify ΔV / ΔI result of each choice then Pd heatrise , as well as fast recovery Sch diode and remember radiation noise increases with current dI/dt and loop area EMI=V= LdI/dt * Area V/sqmm – Tony Stewart EE75 Oct 17 '19 at 23:41
  • Hi, Tony. You say "Assuming 36V LEDs: (although you would be wiser to use 72V LEDs)" and I would like to know why you say it's wiser to use 72V. Thanks. – MicroservicesOnDDD Jan 25 '22 at 20:37
  • 1
    @MicroservicesOnDDD just from the viewpoint of a cooler or cheaper PWM FET with 1/4 of the power dissipation and likewise for Inductor choices are better. – Tony Stewart EE75 Jan 25 '22 at 21:09
  • 1
    You could also consider an inexpensive NPN 3A switch for $0.15 with cheaper driver. – Tony Stewart EE75 Jan 25 '22 at 21:18
3

Your suggestion could work. I have done this with a single white LED where space and cost and simplicity were paramount. The boost diode wastes voltage and hence power. Your 33V string has about ten times the voltage that I had on yesteryears' birdsnest. The prospective power savings will not be great in percentage terms. You will waste more power in the LEDs because the current is very lumpy. The Ohmic losses in the LED bulk resistance will raise the LED junction temp. DC runs the LEDs coolest. Carefully look at what you are doing. The downside may be more than the upside making the design worse off.

JYelton
  • 32,302
  • 33
  • 134
  • 249
Autistic
  • 14,235
  • 2
  • 27
  • 65
  • 2
    I'm interested to see how the experiment plays out. At 3W, an xp-g LED will be at 1.1A whereas the test COB will be at only ~90mA avg and the XP-G will be operating at a less favorable section of the luminous efficiency band as well as suffering higher \$I^2R\$ losses. You make a valid point though, the resistive losses from the tiny trace wires on the COB will likely add up. I know LEDs have much greater efficiency when underpowered, but I'm not sure how this plays out when barely kept at the turn on point. The cob and test LEDs will be mounted with arctic silver on oversized heat sinks. – K H Oct 08 '19 at 03:47
  • 2
    Part of the test will be to adjust the COB and the test LED to similar light output and measure input power on both. I don't expect efficiency gains to justify the 10x-20x cost of the COB, but at least I'll have a COB to play with =). +1 for autism and helpfulness, but not definitive so I'll wait to see what others have to say. – K H Oct 08 '19 at 03:50
  • 4
    WARNING: Yellow and White core colour is a Micrometals trademark (really). If your core IS micrometals then it will meet spec sheet well. IF it is Asian and non micrometals then it may meet some specs. One of the specs which doesn't bother you here is lifetime. With heating the core ages and the binder fails and the core particles separate and core losses get higher and it gets hotter still and .... . Who would have thought that an iron powder core can suffer thermal runaway (at sloth speed). | I had buck converters made in Taiwan for a client. Spec said **MUST** use micrometals cores. .. :-) – Russell McMahon Oct 09 '19 at 01:06
  • 2
    @RussellMcMahon Provenance of the cores is aliexpress, so it is not certain they are cores, let alone iron or powder =). They do look toroidal though. Hopefully they took a good look at the product they were knocking off if they're(likely) not real product. There is no comparable storebought core to compare to, so I'll have to experiment to see how well they perform. Where I'm using questionable parts, I'm attempting to use them below their expected spec. I tried to find a reference to pre-calculate core saturation, but couldn't find formulas, let alone that I could understand. – K H Oct 09 '19 at 01:44
  • @RussellMcMahon -- Thank you for telling about iron powder core thermal runaway. Who would have thought? – MicroservicesOnDDD Feb 22 '22 at 18:01