Network Analyser Measurements

[DF96]

Very good.

Interesting to see that the cap on its own, with short leads, only has a low impedance over about 3 decades in frequency. Of course, 'low' depends on context in the circuit so in a higher impedance situation the bandwidth would be wider.

[KSTR]

Thanks for the feedback, everyone.

Jan, tests on small signal supplies/regulators are on the list of course, but I really should build and test a clamping ac-coupled buffer/gain stage to avoid destroying the sensitive analyser input in case of any operator or other fault (+20dBm/2.8Vac max. and +-1V DC max.).

The active high-Z probe coming with the SNA-2 is ac-coupled but not overvoltage protected.

[KSTR]

Now getting closer to some "real" supply wiring....

A 33cm (1ft.) lenght of shorted twisted pair using thin hookup wire was attached to the jig :


Impedance came out much worse than expected :

Resistance is about 50mR but most important is the inductive rise, calculating to about 200nH!

[KSTR]

Replacing the short with a 1000uF/63V Panasonic FC :


Not much change in the impedance except for the added ESR of the cap (and the capacitive slope, of course). Inductance is dominated by the cabling :

[KSTR]

Adding a local 150nF film bypass :


Impedance plot... well, we've seen that kind of disaster before :


Clearly not a way to wire for example a chip amp supply, not even for temporary testing.
Tomorrow (hopefully) I'll try show a better way, striving for a reasonably low and flat impedance with a remote main cap bank and a mimimum of local bypassing...

[DF96]

Note that tight twisting reduces magnetic field pickup (as used for heater wiring and to reduce inductive coupling) but does not much affect the inductance of a shorted transmission line. Twisting is of course much better than not twisting, but the twist length is immaterial provided that it is short enough to keep the two wires next to each other. In fact tight twisting could make things slightly worse by lengthening the electrical length of the cable when compared with the physical length.

[KSTR]

Yes, that's right. In fact twisting I think would even increase inductance per unit length since the wires run not as parallel as they would if they ran straight with the same spacing so fields don't cancel as much as they could. I think a wide strip line made from copper foil glued together with minimum spacing should be best in terms of low resistance and inductance... gotta see if I find some 1" or 0.5" copper foil in the lab somewhere to check that.

[Elvee]

You seem to be new to network analyzers. Do you know that you can totally eliminate or calibrate your test jigs?
It makes life much easier, because you can eliminate the defects you introduce to make the practical use easier (lengths of wire, etc), and it makes the measurements much more accurate, because you measure exactly what you need.
In your case, you could eliminate the capacitance, inductance and residual transfer of your piece of PCB.

Calibration is a very powerful tool, it decuples the capacities of your instrument.

[metalsculptor]

Any wire will form a transmission line of some characteristic impedance, Network analysers are 50 ohm, twisted pair cables are around 100 ohms. A length of cable depending on what fraction of a wavelength it is will appear capacitive, resistive or inductive. Grab one of Agilent's manuals and some of their application notes for one of their netowrk analysers.
They are a powerful tool which will teach the importance of layout. For many components measuring the S11 parameters might be more useful than the S21 parameters

[marce]

The high quality of the caps is in fact a problem here, there are close to zero ohmic losses present that could absorb most of the resonant peak (at ~0.4R).

A good example of why X7R SMD capacitors are the best choice for decoupling purposes, especially high speed digital. Another interesting thing is these peaks are often not mentioned in data sheets and user guidelines for decoupling devices.
I look forward to more interesting measurements.