Sound Quality Vs. Measurements

[KSTR]

Please note that the photo shows the case with the probe not wired correctly, with the correct position right at the caps I see a |Z| of 0.3R at 10Mhz with an inductive slope (using the 220nF curve). This amounts to a equivalent inductance of 0.3/(2pi*10Mhz) = 5nH which looks not to far off for a 5mm lead spacing, connected right at the cap (probably too low from overcompensation, see below).
The non-perfectly inductive behaviour at higher freqs is from a) running only 75R as the current source and b) having used a 10R 0207 for normalization -- again right at the resistor body -- of the setup (including the probe of course), which will be slightly inductive at some point and thusly will overcompensate.

So it's not a fully confident measurement, agreed, but OTOH it is not that far off to render it useless, and the relative relationships are not affected much at all.

[KSTR]

Hi,

update on the parallel caps measurements.

I've build a small test jig using a piece of double sided blank PCB inserted into a coaxial cable to start with a few pF of PCB capacitance, attaching the caps along the edges for lowest possible inductance that can be had with their case style:

Reference value for the DB scale is 1R as before, but I widened the freq. range from 100Hz to 100MHz, and freq. linearity was checked with a 1206 1R and was found to be neglegible at the vertical scaling used, only a few dB down at the highest freqs. (hence no normalizing needed). Also the 50R mismatch of the cable proved irrelevant.

A set of capacitors was found im my parts bin that gave a smooth and low impedance. The 1000uF was not a PanaFC but a specimen with 10dB lower ESR (almost down to 10mR), then I added WIMA MKS2 with 3u3, 220nF and an unidentified ceramic with 22nF(IIRC). Further addition of a set of 3 220u PanaFC smoothed and lowered things again and this is the final result :

This cap bank results in an impedance magnitude lower than 30mR from 4kHz to 10Mhz, going down to 10mR and not exceeding 20mR for the most part. Using the GND pigtail (see photo above) to fully short the caps resulted in only 5dB lower impedance at high freqs (with the HF sighature of the caps still seen) so this is the inductive baseline that can be had with this type of construction (and baseline dropped 6dB with another pigtail short as expected).

Shown for comparison is a short with 2pcs of thin solder wire (as it has some resistance, here to be found at ~6mR and was confirmed with a 4-wire measurement with my Keithley 191). Both the "cap jig" curve and the "short" curve show perfect capacitve and inductive slopes, resp., confirming the bulk capacitance value (~1500uF) and the inductance of the solder wire short (4nH looks OK for this 2 sections of ~1cm wire length).
The rise a LF for the solder wire short is coming from carrier residual leakage it seems (as I'm on right on the edge of the analyser's lower freq range of 100Hz as spec'd.)

Not shown here, the bad influence of too small caps was noted, adding eg 4n7 showed a much peakier/ragged response in the 10Mhz++ region, giving more than 20dB impedance variations within an octave or so of frequency.


To check any influence of the High-Z probe I did a comparison, too. The setup :


And the result :

Nothing to worry about, the active Probe seems a bit noisier (it also has 10dB loss which is acounted for automatically by the analyser) and has a bit more inductance above 20Mhz if the GND lead is bent such as to minimize inductance. The W&G-guys did a good job there, I'd say. Again with the large scale displayed any small transfer function error of the probe, while present, is irrelevant.

A network analyser is a really powerful tool, any tiny variantion of resistance, capacitance and inductance is readily seen (eg I saw ESR rise when the 1000uF cooled down after soldering). I'll do many more experiments as time permits and will share any noteworthy findings (new thread in the power supply forum, then).

[bcarso]

Nice work!

[dvv]

It sure is.

[mr_push_pull]

can't see pics.

[KSTR]

Some forum quirk I think (I note there is no preview thumbs as well, but the pics are visible for me),

Maybe a moderator can fix it, in the meantime find them re-attached here :

[john curl]

I don't mean to be critical, but you should shorten the leads to more 'accurate' measurements. The extra inductance that you are adding by keeping the very long leads should DOMINATE the inductance of the measurement.

[KSTR]

John, you input is welcome of course. Are you talking about the GND lead of the probe or is it about another specific part of the jig (I can't figure it out)?

Eyeballing from the plots I see an inductive slope with a Z of about .3R @ 10Mhz which is quite smallish ~0.5nH total residual inductance for the whole shebang, even with a say +-10dB uncertanity I wouldn't speak of that dominating the measurement?

[fas42]

Yes, very nice work, good to see the LTspice exercises being confirmed so clearly. Personally, I would like to see 1mR at 100kHz, and then working carefully to keep below 10mR as you progress higher in frequency; I think 100mR at 100MHz, 1R at 1GhZ is about the best one can do unless you go crazy with RF techniques, etc.

Frank

[KSTR]

Thanks, Frank.
Now I'm finally able to back-match my sims to reality. I'll try building a few "super cap strips" with 10000uF bulk capacitance next where I will try to reach for your goal. It'll sure tackle my skills and that of the SNA-2 analyser, with its 180Mhz spec'd limit.