98% efficient 4 watt 12V DC-DC converter EE13/7/3 N87 ferrite

Every time I get a 98% efficient measurement (or over 100% for that matter), I go back and take extremely careful measurements, and discover the true value was much less.

Seems like the diodes would eat roughly 6%.

You never did get back to us about how well that common choke worked as a transformer for that little 2W switcher...

Yup.

Ed

I'll do that next. I just received them late last week.

Paul

Yes, after I sent this, I realized that the correct value for 4.2/4.42 is

The diodes probably contribute about 0.3V * 0.4A = 0.12W each at 50% or

Thanks,

Paul

So here are the results for the Taiyo Yuden TLF9UA202WR54K1 common mode choke:

R=0.87 @100 Hz 4.29 @10 kHz L=2.97 mH, L(leak)=54.8 uH

Ugly test setup:

Open circuit output at 12V input:

Waveform with 33 ohm DC load:

Test results:

Input 12 VDC @ 0.11 A (1.32W), Output 6.39 V 194 mA (1.24 W) Eff: 94%

Actually I measured only 0.10 amps input but that would have produced an over unity result. The coil resistance seems to be the killer, but the efficiency still seems good, so perhaps it is the leakage inductance? Also this is still at 90 kHz, and a lower frequency might work better. As it is, it seems to be OK for a current-limited DC-DC converter such as a MOSFET or IGBT driver where average current is low. And not bad for an 85 cent part.

I also have some Kemet SU9H-R10008 chokes that may be capable of higher power. Test results will follow.

Paul

My calculation is showing something more like 53% efficiency.

Here are the results for the KEMET common mode chokes:

L(10 kHz) = 1.28 mH L(leak) = 15.8 uH R(10 kHz) = 2.41 ohms R(100 Hz) = 0.225 ohms

At 90 kHz:

At 71 kHz:

I added 0.01A to the input current to avoid going over unity ;)

Seems like this common mode choke is actually quite capable of being used as a 2W DC-DC converter for a current souce like the SCR gate drive, or the low average draw application of MOSFET or IGBT drive. And it is also a cheap $0.89 part.

Paul

That is just the voltage ratio, or regulation, perhaps, if you expect a 12 VDC output. But 12V at 0.11A input is 1.32 watts, and the output is 6.39V into 33 ohms, which is 1.24 watts. So losses appear to be only 80 mW.

Paul

Like I said earlier -- you only get (on the order of) a couple of watts, out of a CMC.

That such a figure should be roughly independent of size and type, may be non-obvious. But it is indeed supported by both measurement and theory.

Tim

So, what is the theory (or measurement) that supports your claim? I may have some larger CMCs from old PSUs that I could measure and test. :)

I had thought that the problem was with the intentionally lossy ferrite material and construction. I know that the dual isolated windings with wide separation makes for high leakage inductance, and the CMCs I got also use C-cores rather than E cores, which may add more losses. But the one with lower resistance and inductance seemed to provide much higher power, and I don't understand why even larger CMCs would not work for 5-10 watts or more, with similar good efficiency.

Here are the results for the KEMET common mode chokes:

L(10 kHz) = 1.28 mH L(leak) = 15.8 uH R(10 kHz) = 2.41 ohms R(100 Hz) = 0.225 ohms

At 90 kHz:

At 71 kHz:

I added 0.01A to the input current to avoid going over unity ;)

Thanks,

Paul

I found a CMC from an old computer PSU, about 1" cube, with about 4.6 mH inductance. It uses E-cores rather than the C-cores of the smaller CMCs. Here are the results using the same setup as before, at 71 kHz:

So, again, it seems that a CMC can handle a lot more than a couple of watts, with reasonably good efficiency.

It has quite a lot more AC resistance at 10 kHz than at 100 Hz, probably because of the thicker wire (about #22), so the skin effect would be more pronounced. I found a good article on eddy currents, skin effect, and transformer design for high frequency, showing the advantages of Litz wire and the disadvantages of multiple flat copper strips:

Thanks,

Paul

Ummm....care to explain how you get almost 2x secondary current as you have in primary with a 1:1 transformer??? My calc was based on voltage measurements alone and assumes all the loss is dissipative.

Good point! I believe the answer is that the leakage inductance acts as a buck converter, so that the square wave appearing on the primary, and the secondary, of the transformer, is effectively reduced to +/-7 volts or so. When I did a simulation with intentionally added primary inductance, heavy loads would reduce the voltage transfer, but did not seriously affect the efficiency. I would expect that the short circuit current would also be limited, and yet not cause excessive current or power from the DC supply. At that point, with no output voltage, efficiency would be zero, but losses might be only a few watts.

It's a feature, not a design problem ;)

Paul

Keep measuring small-signal parameters (Lp, LL, DCR), and saturation, of various parts.

Then apply your gained knowledge: calculate the maximum power transfer possible for a transformer of given Lp, LL, DCR and Phi_sat, into the best load resistance (and, reflect on how that impedance compares with common applications, like a nice and cozy 12V 300mA source).

If you need to look up the definitions of these terms, to be able to apply them -- don't be shy, and, you'll soon discover how powerful it is to use even a simple transformer model!

Nope! Core loss is only a second order effect, dependent on frequency and flux (applied voltage). Not current. Core loss doesn't prohibit you from getting power out the secondary. It does prevent you from reducing transformer losses arbitrarily.

Which as you already know, isn't a big deal (you've got quite reasonable efficiency figures). :-)

The wave-of-the-hand justification is that Lp and LL are essentially independent of size.

That is, it's a valuable property of a CMC, that it be a bad transformer. (A CMC /is/ a transformer, by the way.[1]) And, for the easy ways of constructing a bad transformer (namely, bank wound bobbins, or sector wound toroids), you get about so-and-so amount of leakage (giving a k ~= 0.98 or so).

There's also saturation.

If cores did not saturate, you could push the frequency as low as you care, and transmit a nearly arbitrary amount of power. You'd be limited by DCR, but not LL.

But the incremental value of Lp drops suddenly, when you deliver enough flux (V*sec) to the core. Which kind of screws that up.

Indeed, the value that Lp drops to (in hard saturation), is very similar to LL, and for very similar reasons!

In summary, the actual conclusion is that, because saturation flux goes up with size, you /can/ push more power. But not a crazy amount, and only in a relatively narrow frequency range (within a factor of 2, say). And that frequency corresponds with size, more or less.

So, using a CMC for big power (like, 100W+) just ain't gonna happen, even though that's an extremely accessible power level for small inverter transformers.

Tim

Accurate measurements at low power levels aren't easy, but it would help if the load was actually being measured, rather than calculated based on a 20% tolerance carbon composition part that was pulled from the junk box.

RL

The way the efficiency tanks at only slightly higher voltage is a bit suspicious too.

Cheers

Phil Hobbs

Nothing mysterious about that. Saturation of ferrite is dependent upon volt-seconds, and at that point losses increase quickly. It will be observed at higher voltage or lower frequency, and that is what I saw.

Subsequent tests were performed with a 33 ohm resistor that measured 33.3 ohms, so there might be 1% error there. I'm only looking for approximate values of efficiency and regulation.

Paul