inductor sizing

Tom,

Do you have a copy of Randy Rhea's (of Eagleware Genesys fame) paper, "A Multimode High-Frequency Inductor Model?" He does a pretty good job demonstrating how most real inductors are a lot closer to a transmission line model as you're discussing than the ECE 101 "fields and waves" model that's often proffered.

---Joel

Where can I get a copy of the paper? Google gives lots of referrals to the paper, but not the paper itself.

...Jim Thompson

--
|  James E.Thompson, P.E.                           |    mens     |
|  Analog Innovations, Inc.                         |     et      |
|  Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
|  Phoenix, Arizona            Voice:(480)460-2350  |             |
|  E-mail Address at Website     Fax:(480)460-2142  |  Brass Rat  |
|       http://www.analog-innovations.com           |    1962     |
             
         America: Land of the Free, Because of the Brave

Hi Tom, sorry for the confusion. The ferrite inductor is 25uH in a 2.75" cube. It is designed for 35Apk,

20.0Aac and 2.0Adc in a larger core than necessary because the OP will probably have to drop down to 100KHz in switching frequency and he needs a lot of wiggle room at his skills level. That said, a 2.25" cube would get the job done. My Rdc = 8.0mR is real but the Rac is a optimistic guestimate. Though I used Litz wire, the core gaps will cause havoc with the Rac. With the large Aac component, Rac is a dominate player and reducing it is an ongoing task. In winding ferrite cores with high Aac levels, we normally can not use many layers nor fill the available window because of proximity losses, air cores should exhibit the same or worse effects due to the flux being outside the core. Do you have a way of measuring your Rac at 200KHz? Your #24 Litz wire sounds too fat, maybe >#28 would serve you better. Tell us about your dimensions and layering.

Regards, Harry

See

You don't have to wind your own. ;-)

Cheers, Tom

Hi Joel,

Well, I'll be the first to admit that I'd get overloaded very quickly if I looked up every reference I come across -- and that particular paper is not one I have. I did read John Mezak's paper that was published in "RF Design" sometime in the early to mid 90's, describing his work on a computer model for the helical transmission line coil model, and if you go to Serge Stroobandt's wonderful website I referenced yesterday, you'll find a reasonable amount of discussion there about where the model came from, with links to some other good sites. I really like that Serge has provided all that background info and the links; so many sites with things like that just present a calculator and you have little or no way to build confidence that it was done right or to learn about what went into it. That's one of the reasons I trust Mezak's program--plus by this time I've had lots of opportunities to wind coils based on it and to measure them and see for myself that they agree with the predicted results plenty close enough for my engineering work. Mezak's program and the calculator at Serge's site give results that agree close enough for my purposes; as with ALL modeling, especially if you're pushing the limits, plan to test an actual circuit and make adjustments based on actual performance.

Cheers, Tom

We got some of those as samples... they are quite good, albeit quite spendy (but of course cheaper than making your own if someone's paying for your time...). Someone mentioned that initially low yields were one of the reasons the prices were so steep, so perhaps they've come down in the year or so since they introduced them as their yield has improved?

Hi Harry,

Thanks for the further explanations! I'm not shy about saying it's a breath of fresh air to find someone willing to actually discuss alternate ways of doing things--so many of the discussions get mired in "is so/is not" rhetoric. Thanks for helping keep things sane.

As I've mentioned before, my work normally is with RF filters. Almost all the Litz wire I have is based on 44 and 46 AWG stranding, and is dear enough that I'm not about to use a whole bunch of it (e.g. about

200 feet of 175/46) in a test coil I'm not going to be actually using for anything other than a test. So that left me last night digging out a coil of heavier Litz wire that a friend had given me, and discovering why he had given it to me: it's about 100 feet of 20/32, so equivalent to 19AWG, and that would be almost reasonable for 200kHz work--though really optimum for down around 10-20kHz. I fully agree that 24AWG would be way too big for the stranding! But what looked like a reasonably well-ordered coil turned out to be impossible to just unwind. So I spent an hour or so teasing it out (trying to keep it from kinking), and have it probably 3/4 done now. No coil yet. I also decided that I'm probably NOT going to cut it to do the proposed test--I'd want to cut it into something like 7 lengths and put them in parallel, and it's just worth too much to do that. In fact, for 200kHz, even 32AWG stranding is way too big. The Litz manufacturers suggest 40AWG stranding, typically. Anyway, that all leaves me in a quandary about how to make a test coil. For sure I'll wind all the 20/32 I have into one multi-layer coil, ID about 1.33" and length about 1.5" (I made a bobbin for it...), and I can measure that, but I'm not sure it will be all that representative of the rather different coil we'd really want: similar size, but fewer turns of heavier wire. I do expect the Q to be similar, though, if the form factor is similar, and the SRF probably scales in some reasonably nice way with inductance.

Measuring Rac is easy enough for me: I measure the Q and inductance, resonated with good polypropylene caps to the desired frequency. Assuming the caps have infinite Q puts all the loss in the coil and therefore puts an upper bound on Rac. Oh, and for a 25uH coil, Rac at

200kHz of around 15 milliohms would be a Q in excess of 2000! That one I'm _really_ not going to believe. ;-) I've measured the loss (through Q measurements) of a variety of power inductors I've salvaged from commercial supplies. Some of them have very disappointingly low Q (presumably were used at a place where the DC current was much greater than the AC through them), and I'm always happy when I find one with a Q at 100 or a bit more. I don't think I've ever found one with a Q very much above 100 in the 100kHz-200kHz region. Based on some other measurements made the same way on higher Q coils, I know it's not a limit of the test method. But this may not be a fair test, since the excitation is at a very low level compared with the level at which the part is used in the supply. On the other hand, I've seen a rule of thumb that core losses in high frequency transformers (including gapped flybacks??) are almost always several times the copper losses. If that's right, Rac in those cases must necessarily likewise be in an even larger ratio to Rdc.

'Nuff for now...

Cheers, Tom

I mentioned Serge's website, and how it has good links. One of the links that I had not previously looked into is

(and parts 2 and 3 of the same article). I think David does a great job of giving an overview about the need for different inductor models at different frequencies, and then diving into a discussion of the various calculation methods that have been developed, how accurate they are, where they're useful and so forth. But I should also point out that this is getting to be pretty bad basenote drift, since we've gotten into what normally would be considered "RF" coils, single-layer solenoid coils, rather far from the discussion about high-current inductors for 200kHz switching supplies!

Cheers, Tom

Thanks for the links, Tom, I'll take a look!

Got it untangled and wound onto the bobbin. At 200kHz, it's about 1.5 millihenries at a Q of about 60, not terribly good. I was hoping for something a bit higher. I suppose something closer to the recommended Litz stranding would help, and possibly also a different layup. I wouldn't expect Q to change much with a change of wire gauge/ inductance if the coil occupies the same volume. This was just layers back and forth. 1.33" ID, 2.03" OD, about 1.7" long. I didn't count the turns as I was untangling things. ;-) SRF is only about 500kHz, but assuming the effective shunt capacitance stays constant, the SRF of a 15uH coil would be about 5MHz. The effective capacitance may go down a bit with larger wire, and also with a different layup.

Cheers, Tom

Cool! I guess a variable-pitch spiral will behave similarly! I want slightly bigger inductors though* :)

Cheers Terry

• spiral of 20mm x 5mm Al strip

These are copies of the Piconics design; I assume the patents have expired. The Piconics parts are better and cheaper. These are commonly used in super-wideband bias tees, like the PSPL boxes.

John

Ok, so its my turn to do some measurements on a ferrite EC core (2.10"x1.10"x0.75") that yields about a 2.20" cube. I started with one layer of 7 turns Litz, (330x36 = 12AWG) and setting the gap to yield 25.0uH. The measured "Q" at 50KHz, 100KHz and 200KHz resulted in 190, 153 and 125 respectfully (?). I added another 7 turn layer, set the gap to yield 25uH and got readings of Qs at 114, 107 and 95. A final layer (3) was added of 7 turns, reset the gap and measured Qs at 89, 74 and 63. IMHO the only acceptable readings above are a single layer at 100KHz or less. The Q of 153 means that Rac = 0.103R which is 20X greater than Rdc. Now if I am using Litz in only one layer, what happens if I just use plain stranded, for flexibility, wire? I wound one layer of #14 AWG and got a Q of

138 at 100KHz. compared to 153 for Litz but that was #12AWG. Not much gain for Litz. So the OP should be using only one layer of non Litz winding at 100KHz or less. The air core's Q should be lower because of flux fields flying all over the place and inducing localized current loops (proximity effect) and shrinking the actual wires cross sectional area. Cheers, Harry

On Wed, 12 Mar 2008 13:10:53 -0700, "Harry Dellamano" wrote: . . .

--
Respectively. :-)

So the Q for the air-core guy of 7 layers (I think that's what it turned out to be) at 200kHz is pretty similar to the Q you got for 3 layers. -- For air core solenoid coils at 1MHz, I'm commonly getting close to 3 times improvement in Q for 175/46 Litz versus similar diameter solid, but these are single-layer with the turns spaced slightly--around 1.2* wire diameter. Interesting that the Litz and non-Litz in the case of your coils makes so little difference. It illustrates to me, once again, the need to try these things before jumping to confusions.

I wonder what kind of core materials they're using in the supplies that run at a few MHz. They seem to be able to get good efficiency.

Thanks for sharing your measurements.

Cheers, Tom

Gorgeous! Jumping to confusions.... I'll have to steal that.

and yeah, thanks for sharing. This has been a great thread, and helped me solve a problem. I had taken a T50-26 toroid with 450T of AWG34, and made it AWG44 instead, as I needed the series R. And the choke all of a sudden had an anti-resonance (at 3.6MHz) which needless to say made it less than useful.

I had figured it was multi-resonant becuase of the distributed inductance & high capacitance - it went from 8 layers to 1 layer. So I had some universal-wound (kinda) and halved the end-to-end capacitance, which increased the first resonance by 1.4x, and likewise with the 2nd resonance.

The magnetics vendor didnt believe me when I showed them an impedance plot, until I showed them a photo of the network analyser setup. but they couldnt explain it. And my sidekick inthe US (an EMC guy) hadnt come across this before either. Hurrah, I learned something new.

I had already concluded that the low perm single-layer winding would have distributed coupling, and the fix was pretty simple - go back to AWG34 and add a series R - but Toms reference sorted out exactly what was going on.

one of the best threads!

now all I have to do is figure out how I can exploit shell helix resonance. I must be able to do something useful with it....

Cheers Terry

Hi Harry,

Sorry for the late reply I was out of province for the last week.

For the feedback of an LCLC filter like this, are only one of the inductor currents and capacitor voltages required, or are both currents and voltage required?

For the LC filter version I use this algorithm for the feedback loop:

if inductor current > current_limit, pulse off

Duty cycle = ((desired voltage - current voltage) * gain) - (Inductor current * gain2)

I am not sure if I can use this for the LCLC filter as well.

Also do you have part#'s for the high flux inductor for this LCLC filter and the output capacitor? The output is 120VAC 60Hz at 3.6kW by the way.

I plan on using planar inductors instead of E cores and litz wire, but I will still send you the tequila for the design :D

cheers, Jamie

*The second LC stage is not required in the feedback loop and may be detrimental due to added lag. It is only needed to meet your output high frequency attenuation requirements which you have not yet revealed. *As we have stated before, the LCLC filter is a separate issue from the feedback loop. To simplify the feedback you may chose to feedback the output of the first LC and let the second LC run open loop. The FB loop must operate with the first LC but you can close the loop in any way your heart's desire as long as it stable.

• So, asked but never answered, what is your output noise requirement? Are you trying to meet FCC, DO-160, MS-461, CE....? How much must the 100KHz and harmonics be attenuated? Notice I said 100KHz because at 200KHz the AC resistance of the first inductor will eat you up unless you know of a way to reduce it. The best I could do was using a single layer winding, at 100KHz, and got a Q of 475. Earlier in this thread I reported a Q of 150 to PH but cleaned up my measurement methods and attained a true 475. The AC resistance was still 5X the Rdc and the biggest obstacle. The second LC will see little

100KHz current and can be any high flux core. I like the MPP-HF or a gapped MET-GLAS. Size, cost, availability and familiarity come into play here so there is no simple answer. *Sounds interesting, are these planar inductors something you will purchase from the boys in Florida or your in house design? What kind of Rac at 100KHz and 200KHz is attainable? *Tequila is only necessary for the closing ceremony.

Cheers, Harry

The output has to remain below 2.5% THD + noise.

I tried this in the simulator but never got the LCLC to regulate as well as a simple LC does.

Payton

or Himag both make these planar inductors. I got a quote for a 25uH 82Amp inductor from Payton for the 200kHz filter with 15Amp peak to peak ripple rating, and it has a specified 30watts loss. Its dimensions are

116mm by 66mm by 26mm high (big rectangular block).

I was way off with the original 250uH inductor, that would be huge at these currents and frequency, 25uH seems to be more than enough inductance (I think 15uH would work actually)

cheers, Jamie