I am designing an LED driver that uses 1 switching current regulator to drive 4 series connected LED's (RGBW)

The connector goes to a string of LED's all connected in series, RED, GREEN, BLUE and WHITE.

Each LED has a PNP transistor that shunts current around it, allowing me to PWM the LEDs independently - it all works, but the inductor gets really hot.
The LED Current is set at about 730 mA.

The circuit runs from 24 volts, consumes 1.75W with all the LED's shunted OFF.
With all the LED's on, it burns almost 10 watts.

The switching frequency with all the LED's off is 65 kHz, - the inductor is clearly not going into saturation.
I ran a test at DC, running 730 mA through the inductor and it barely gets warm, so clearly the heating is due to core losses.

Why would there be so much loss at the pokey frequency of 65 KHz?

 

Just curious, when you say "really hot" how hot is that?
Are the waveforms you posted representative of the waveforms present when you have all of the LEDs shunted?

 

Yes- the wave forms shown are indeed with all LED's shunted.

The temps reached (65 C) are not crazy, but this board needs to by tiny, so I need to squeeze the most efficiency out of it.
The thermal image shows that the inductor and schottky diode are dissipating the lions share of the power.

 

It is very rare for somebody to be so well prepared to answer that question. Excellent!

The inductor datasheet is sparse in the losses vs frequency department. 65 kHz might be pushing it in your application if a 40° temperature rise is a problem.

It could also be eddy current loss in the copper, especially where the copper is very close to the core because of the large magnetic field perpendicular to the core changing through the conductor. Whether it is copper losses (which looks to be about 250 to 300 mW for the I2R loss at DC), eddy losses in the copper, or core losses and the temperature is a problem the best solution is probably to find an inductor rated for higher current and/or higher frequency use.

 

The "Q" of this inductor is listed as '20' in the data sheet, measured at around 800 kHz, that's the only clue I see regarding it's performance Vs frequency.

Where does one go from here? looking at the parameters for inductors on the Digikey website, it's hard to know what to try next.
Seems like I should investigate different core materials, at least then I have a prayer of getting closer to what I need.

 

Short of making your own the best thing might be to contact the sales representatives or better yet, sales engineers at a few of your favorite inductor manufacturers and see if they can make recommendations. That or pour over the catalogs until you find some promising parts to try.

 

The core is listed as ferrite, and unless it is really really horrible ferrite the losses should be small. The inductance vs. DC bias curve is not what I would expect for gapped ferrite, which tends to saturate quite abruptly. The curve looks more like powdered iron to me, be even then loss at 65 kHz shouldn't be too bad.

The DC resistance is high, so there are lots of turns which means lots of proximity effect, but it also means very fine wire that is probably 100% skin at 65 kHz. Still, if the build height (id - od difference) is large, prox. may be a significant contributor to loss.

There is nothing about the current waveform that even hints at approaching saturation.

Ferrite normally exhibits lower core loss than any of the alternative materials. The next best (or superior, depending on circumstances) material for inductors with DC bias is molybdenum permalloy powder (MPP), but it is expensive. For many of the small inductors the manufacturers are secretive about the core material. The IHLP series from Vishay is very good but probably a lot more expensive.

Ferrite thermal conductivity is generally dismal, so it makes it hard to try to get some clues from rate of temperature change at the surface

 

I found a few guides from various inductor manufacturers, none of which really help me to get an intuitive feel for which core material or physical configuration is best for my application.