Decoupling cap SRF

Isn't that a peak at a specific frequency?

You probaly just need to avoid exact multiples or octaves down type thing.

Decoupling caps have a _series_ resonance, so they work better at resonance than you'd expect from their capacitance alone.

If you make sure the reactance of the lead inductance is low enough, it doesn't matter how large the capacitor is.

But if you need a little extra in a one-off, picking a bypass that resonates at the frequency of your worst spur can sometimes help.

Cheers

Phil Hobbs

Find a capacitor that is series resonant at 22.5 MHz. From your data, it looks like a 0603 between 10nF and 100nF might get there.

The UHF/microwave guys call them J0 bypass or J0 DC block capacitors. The J0 DC block capacitors they use for input/output DC blocking.

Cheers, John

The resonant frequency for the 10nF X7R jibes with the AVX datasheet for a similar part.

What seems odd is that all the reactances are so high at resonance. Wouldn't the net reactive component be very close to zero at resonance, that is where the series X(l) = -X(c)?

Bob

Well, there is no reactance at resonance. And, none of his reactance tests are at resonance.

John

FYI, I've found the AVX SpiCap3 application great for finding such information.

I need to slow down when I read.

Bob

It probably wouldn't matter, but I'd use the 100nF, by habit. What puzzles me is the very different SRF for the two 10nF capacitors. The COG would have an ESL in the 1.8nH ballpark, much more than I'd expect of a 0603. How would that come about?

Tell us who makes it, so that I can avoid using it. ;-)

Jeroen Belleman

Oh! the 100nF unit's ESL is even worse, with over 3nH, if the SRF figure is to be believed!

Jeroen Belleman

Sorry to sound untechnical, but pick the cheapest :-). If you are using a moulded choke in series to the rail and have the layout right overall, there should be no problem. 100nF for decoupling at 22Mhz sounds a very high value. I would have thought that a 4.7 or 10nf or even less would be fine...

Regards,

Chris

I think the moral of the story is you have to be very careful with test fixtures and calibration to measure stuff like this.

A perfect 10n is 0.71 ohms at 22.5MHz.

My cheapy network analyzer measured an 0603 10n X7R 50v at 106MHz SRF and

An 0805 100n X7R 50v in the same fixture measured 35MHz SRF and 0.05 ohms at 22.5MHz.

SRF doesn't matter. A 10 uF 0603 would have an even lower SRF, but that doesn't make it any worse a bypass at 25 MHz. ESL is what matters, and the way the cap is connected matters most.

The best bypassing is when you have a solid ground plane, a thin dielectric, a nice power pour, and several bypass caps.

But usually it doesn't matter much. Most any bypassing scheme usually works, especially at low frequencies like 22 MHz.

How did you fixture the 0603 cap to the connector on the AIM?

I usually use TDR, to do time-domain measurements. I solder the parts along, or at the end of, a 50 ohm microstrip.

ftp://jjlarkin.lmi.net/TDR_0.1uF.JPG

ftp://jjlarkin.lmi.net/Mini_Melf_Fixture.JPG

John

I have a BNC to binding-post adapter fitted on the front of the AIM. I soldered wires to the ends of the 0603 caps and strung them between the binding posts. OK, so maybe I should've rounded Xs/Rs to one decimal place.

I'd guess that your fixturing has way more inductance, 10x at least, than the part does.

I cut fixtures out of copperclad, with an x-acto knife, and use 50 ohm microstrips (or CPW) and SMA edge-launch connectors (or the flange ones, which work better, if anything.) Like my mini-melf pic above. A loop area of a fraction of a square inch will have much more L than the cap does.

The advantage of TDR is that I can time resolve, and separate out, the impedances of cables and connectors. A network analyzer just sees it all in a couple of numbers.

John

John Larkin a écrit :

??? That's all the purpose of calibration. Just have your reference plane at the right place and your VNA only sees the cap.

Point taken about the superiority of TDR versus VNA. Thanks for the insight.

The AIM is calibrated by connecting standard resistances to the test fixture, which can be some distance away along a random length of transmission line. It's intended for measuring the impedance presented by an antenna at its feed point. I did the calibration with the binding post adapter fitted, so it should be at least partially compensated for. I know its only a rough measurement; but its good enough for most of the stuff I do; and I think it's superior to say the AADE LC meter because it sweeps 1 -

Did you first solder an 0603 resistor into the fixture, and normalize? That should work.

Of course, both a VNA and TDR can be fooled. For example, if there's a lot of loss in the cable (worst case, consider a 60 dB attenuator inline) there's no way either instrument can normalize that away.

John

Is there, anywhere, a "Dummy's Guide to How To Take a TDR Trace, and Calculate The Various Impedances On The Line" document or book available?

I think I can puzzle it out, when all of the Zs seen by the pulse or step-function are almost purely resistive... but I don't know how to handle reactances along the line (whether lumped, distributed, or from stubs). Would love to know! I recently lucked into a 7S12 plugin, with sampling and pulse-generating heads, for my Tek scope, and I can see it's capable of delivering a wealth of information... but I don't yet grok the language well enough to understand all of what it is telling me.

Try to find a 1S2 manual online. It has some nice theory.

I could put it up somewhere, but my PDF is 35 megabytes!

My 11801s just tell you line impedance on-screen. For big Cs or Ls, you can measure the time constant against 50 ohms. Small L and C require you to measure the area of a blip and do some math. All in the

John