Help design a power inductor

[Giaime]

Hello everybody, 1st post here. Long time lurker, big fan of the EEVBlog show.

I'm designing an "unusual" DC-DC LED driver for an hobby project of mine, and I would like to get some help designing the power inductor. I've dealt with PFC flyback transformers in the past.

Main characteristics of the DC-DC converter:
- boost topology, or, better, "boost-to-battery" (LEDs connected between boost output and supply input)
- fixed DC 48V input
- various output currents, from 400mA up to 1A
- 40-70V LED voltage, giving up to 120V boost output node voltage
- 1MHz fixed frequency operation, DCM mode (ILpeak from 3 to 6A depending on output voltage and current, typical Ton 350-450ns)
- inductance required: 4.7uH, 6.8uH, 10uH, saturation current more than 8-10A

I have already tried some "best offers" from off-the-shelf power inductor manufacturer, for example Wurth has a nice losses calculator: http://www.we-online.com/redexpert/
but the issue is: too high core losses, too high temperature rise. Without forced air, the best 4.7uH I could find is at 130°C, and only forced air can bring it down... So from my measurements it should dissipate a couple of watts, I would like to stay under 1W if possible with a nice 40-50K Trise.

Main design goals is not electrical efficiency (whole DC-DC converter, due to high frequency and DCM mode, is barely reaching 85%) but temperatures must be safe.

Since I will have to roll my own inductor, my questions are:
- does someone have experience with high ripple current, high frequency similar inductor designs?
- what could be the optimal core material to choose? And in general, where can I find a comparison of core material vs. applicable frequency range?
- what core size should be suitable?
- any tricks / tutorials to do the actual math for gap, number of turns, etc...

I was already given the suggestion of using Litz wire, with an E16 style core, air-gapped and try to keep the flux variation small by having a big gap and lots of turns.

Thanks in advance for the replies.

[DutchGert]

Why not just use this:

http://www.vishay.com/inductors/ihlp-power-inductors-hitemp/
http://www.vishay.com/inductors/ihlp-power-inductors/
http://www.vishay.com/inductors/ihth-power-inductors/

[Yansi]

DCM is feasible, sure. But not at 70W @1MHz.  Use CCM instead and the right type of core material. (iron powder toroid, possibly?)

You can jsut calculate the required inductance from this formula: U = L*di/dt. Only need to fill in the correct numbers. (Solving for operation duty cycle needed first. Just need to know that duty=0 -> 0V output, duty=50% -> output voltage = input voltage and duty approaching 100% means the output voltage will go towards infinity... in theory of course. Note: Applies only for CCM operation.)

The current ripple have to be kept small for iron powders, like 10-20% of nominal current, otherwise you will melt the core like in induction furnace.

To determine the power loss per volume unit, there are usually graphs provided in the datasheets.  (W/cm^3 derived from a known frequency and flux density peak value).

The maxB can be solved for using this equation: L*I = N*Bmax*S (beware that this formula will give you pk-pk flux density swing when current I will be equal to the full current ripple - those 10 to 20% of nominal current through the choke. To get Bmax you have to either fill the current in as Ipeak (half the ripple) or divide the B in half then).

Shouldn't be no problems calculate the required inductance and select the right powder core based on its power loss. Be prepared that such design is usually an iterative process, requiring trial and error, to fit the best core with best performance. (small size, low loss).

And I'd suggest lowering the operational frequency to 200-400k maximum.  1MHz 70W... meh. Could be done possibly, but I'd vote against.  Or are there any awful size/weight constraints on the design?

[Giaime]

Dear all,

thank you very much for your precious suggestions.

Let me clarify what is the scope of this: I'm building a tunable-white LED panel to light high-speed cameras (300-500 fps) experiments for a local hackerspace.
For color accuracy, I need quite precise PWM dimming, but we've seen that PWM dimming at lowish frequencies (500-1000Hz) is a problem with those cameras.
That's why the first prototype has 1MHz switching frequency and a PWM frequency of 20kHz, that's now generated by a signal generator and later replaced with an Arduino or something.

I've seen some appnotes that suggest DCM operation for the converter to obtain high accuracy while having a very high PWM frequency, and from my current accuracy measurements this seems to be confirmed.

I know DCM at 1MHz is quite a challenge and probably not the best way to do it, but this is an hobby project for fun 

The only problem is that the choke is burning hot  even with forced cooling.
I've tried some of them, also with the help of the Wurth calculation tool that I linked before: I simulated the design and tried out the best inductor that they could offer but still, too hot for me.

Usually off-the-shelf SMD inductors don't specify a suggested frequency range, or core losses estimation, so I decided to try and roll my own.

This is the reasoning I'm using to design it, it's a "backward" way of designing an inductor as far as tutorials on-line go:
- first I choose a couple of core materials (N49 from Epcos and 3F4 from Ferroxcube) that are suggested for such high frequencies, based on losses and electrical resistance (that works against eddy current losses);
- then I selected a couple of possible core sizes, starting from E cores: for exampe, E18/4/10 and E38/8/25 that are available with such materials (I will be using an air gapped core for the first trial);
- then from estimated thermal resistance and target temperature rise, I work out what are the kW/m^3 losses that I can accept;
- from this I calculate the peak B field from the material datasheet at my specified switching frequency;
- this as far as I understood, it is half of total B swing, so I double it and calculate N = (L*Ilmax)/(Bmax * As), which should be the minimum number of turns I can use. Going with an higher N, increasing air gap, can keep the same inductance while decreasing core losses.

This is as far as I am right now. Can someone confirm my understanding?
What I'm not sure also is if there is some other way to confirm the core size I need: from previous projects, it seems to me that an E18 is quite small to handle such power.

Thanks again for the replies 

[Zero999]

Do you have to do PWM dimming? Reducing the current is generally more efficient and will enable lower frequencies to be used.

[Giaime]

Hi Hero999,

unfortunately PWM dimming is mandatory here because LEDs will shift their color point (and emitted spectrum) as the instantaneous current changes, so to obtain precise colors I have to keep a constant peak current level and dim by PWM 

[megajocke]

That approach for calculating the minimum number of turns seems correct. Getting the winding AC resistance low may not be trivial however due to gap fringing field induced eddy current losses and winding proximity effect losses in a gapped inductor.

A low permeability powder core (Micrometals -2 material?) or even air core toroid with a single layer winding might make sense.

[poorchava]

Another thing: at that frequency and output voltage you'll need some seriously fast freewheeling  diode. Also,  you can definitely expect enormous EMI from the drain of the switching transistor. A snubber and ferrite beads might prove necessary. Good thing that boost ratio is not particularily high.

Sent from my HTC One M8s using Tapatalk.

[ajb]

Since you don't care about efficiency, have you considered a hybrid switching/linear approach?  You could do a constant-voltage boost converter at a more reasonable frequency (perhaps even an off-the-shelf module), then a PWMed linear current source.  Building a basic low-side 1A current source that can be turned on and off at 20kHz wouldn't be too difficult, and you can monitor the current source's head room during the on-time and trim the boost converter to minimize power dissipation.  Heck, if you only have one series string of LEDs, you might even be able to get away with just a resistor instead of the current source and just trim the boost converter voltage to keep the current within tolerance. 

I'm really curious what you're doing with high speed cameras that requires such tight control of the spectrum.  Also, the output spectrum is partly dependent on junction temperature, so while PWM dimming might produce more consistent results than constant current, how confident are you that it will be consistent enough for your application?

[Siwastaja]

+1 for ajb:

Do a constant-voltage supply with enough filtration that it has practically no AC component. Now you can use any type of converter and design it normally without your specific requirements.

Drive the LEDs with a series resistor and switch with NFET. Use the series resistor as a shunt to measure current (during on time) and adjust the supply voltage with a SLOW feedback loop to compensate for LED Vf changing with temperature. Now you can modulate the LEDs in any way and frequency you want with no problems whatsoever. The efficiency penalty can be very small, like 5%.