LCR Meter Using SoundCard - Revisited and surprised at NO Discussion!

This is a good reference handbook from HP/Agilent if you want to dig deeper into the theory and practical measurements:

Cheers

Klaus

Not bad idea. Just spot checked going from 220 ohm down to 10 ohm and the C value, including the shape vs frequency held to within 0.5%, with the error tracking like the ohmeter may be slightly off for the two ranges, which makes sense.

But catching me waaaay off guard is that the Resr appeared to go up! way up! went from reasonable 0.1 up to 1.4 ohms. The cap was 60F, instead of an earlier 70F, but can't believe the Resr is THAT much of a thermometer!

again, measuring C at 20Hz, 40Hz, up to 1000Hz of this 1000uF 25Vdc axially leaded cap using either 220 or a 10 ohm resistor yielded the EXACT same values within 0.5% and in the low end where the capacitive reactance is higher the values track better than 0.1%

But the Resr! I'm starting to feel like the idea of measuring Resr directly is the pits. But should be able to see 0.1 with 10 ohm resistor, that's only 1% of the sense resistor and should be able to see 1% of 1%, so don't know where the problem is coming from. The 10k resistor measured around 0.1 ohms too [I'll look more cloesly at that data set]

Oddly, running the test to collect 1 second of data yields just about identical measurements to collecting 50mS of data. within error of less than 0.1% Running data collection for longer shows that there are some 'bad'frequencies like AC mains related and some SMPS running nearby, but mostly the two values obtained are identical.

I should put these numbers through octave while applying an error band. Maybe that will show where I went wrong because I was counting on something that is just NOT there.

Thanks for posting that URL. It has a lot of information in it, a good reference.

I haven't looked at the potential of distortion through the cap's dielectric. I had counted on the AC coupled signal simply biasing the dielectric into linear operation. If that doesn't happen I've got troubles. Hmmm, I do have a way to force a bias on the cap, like 5 Vdc through a 1MEG resistor and remove the impedance as part of the calibration; which means the bias network falls out and the cap can't ever be reverse biased by the drive.

I could try a simple test of running the interrogating signal at the same value, well approximately, whether it's 220 or 10 ohm. in series. But as you know, the source out of a SoundCard is around 560 ohm so going from

220 to 10 ohms changes the signal, but not that much.

Most network analyzers plot the Z vs f

If I plot Z vs f, you barely notice how the line is slightly curved due to the cap changing from around 880uF at 20 Hz down to 800 uF at 1000Hz. Maybe this is just nuance and doesn't matter, but it would be nice to have some explanation of whether it's real, or an artifact of the way I measured it.

What do other people see when they measure the capacitance of a 1000uF cap as a function of frequency? Any papers describing that?

the

of

thermometer!

EXACT

reactance

resistor,

measured

Thanks for trying it. But my mind is croggled now because i can't find any explanation for the data.

Have you posted some schematic of the test setup that i missed somewhere?

?-)

I think I've found the 'weakness' in my approach. ...I'm trying to use one resistor to do all. Big mistake! For example, the 10k, 220, and 10 ohms has NO difficulty finding the C within 0.5%, but the Resr is all over the place. Probably because the Resr is IN phase with the Rsense and there is a multiplier effect, meaning ratio of resistors to noise percentage! you can see how easy it is to swamp out a valid number very quickly. Whereas, phase shift always sticks out.

I found two sources of errors: the first is a theoretical value. For example trying to find the input Z of the voltage sense point, the resistor's accuracy is affected by the value of the sense resistor. As in, using a 10k to measure 10k Zin makes sense, but using 10 ohms to measure

10k does not. It's just I wanted to simplify the 'calibration' cycle by using a single resistor, but don't think that's going to be viable.

The second source of errors is an 'implementation' problem. Something is causing the measurements to 'shift' around. I just discovered the data set has some unexplained shifts in it. For example in calibrating to MATCH ch

1 to ch 2, instead of holding around 8 ppm accuracy; the calibration factor shifts over 100 ppm! Where does that come from? Could be a failure in the SoundCard, but more likely where I've been fighting physical stuff before - the connections. 1/4 inch guitar plugs are NOT reliable connections. Slightest corrosion, or pressure change, and you see the problem. People don't realize just how poor an unclamped connection can be. especially if you're tryiing to hold everything below 10 ppm. [Samtec is the ONLY connector manufacturing company I know of that makes connectors and willingly supplies third party measurement data of the 'quality' of those connections, data shown as Rseries in rms! and over time!] I didn't solder connections on the test board, to simplify/speed changing those connections, and not inject temp changes to the test. Instead of soldering these temporary connections, I've been using Pomona clip leads to press two wire leads together. Sadly, these clip leads 'appear' to hold well, but DON'T! simply clipping a resistor's lead to another wire using those thumb thingies. Even though you connect as securely as possible, let go, and let them sit for 10 minutes, the connection still continues to shift/walk around over 500 ppm!!!

I changed the sequence last night using two resistors, instead of a single resistor. 10k was used to 'measure' the Zinput of ch 1, then 10 ohms was used to 'measure' C. Just before the test I 'reseated' the 1/4 inch guitar plugs, probably should NOT have, because after perturbing the system the resulting settling/shifting around absolutely ruined the data. Until I can resolve/solve the connection problems, not going to make a lot of progress. After typing this reply, I'm beginning to firmly believe the connection problem is preventing progress here, so I will go back and redress all. I know from experience using hardwired connections it is possible to get down below 10 ppm and with some averaging and games below

0.5 ppm relative, which would make measuring a pretty wide range of reactances a piece of cake.

Schematic?! What's to show! I'm talking drive out of channel 1 permanently connected to input channel 2 whose node then either gets connected directly to channel 1 input [to calibrate the two channels' ratio], or through a sense resistor to channel 1 [to measure channel 1's Zinput], or though a sense resistor to channel 1 in parallel with the DUT, the capacitor [to measure C and Resr], to GND.

Not much schematic there. just three set ups.

I should point out the change in C from 20Hz to 1000Hz is a drop of around

4%. Not much, may not be there, but the potential origin of such a shift, ...curiousity is killing me.

And someone pointed out that the reactance times the Resr is a constant! in the data. Xc*Resr is a constant. where does THAT come from?

More info, I cleaned the 1/4inch guitar plugs inside and plugs. Then CAREFULLY took a run of data to find that the simple calibration shifts around about 4,400 ppm !!! how much from clips on leads and how much from 'wet' 1/4 inch plugs, no idea!

So wiped each off, then soldered connections where possible and the shift immediately dropped to 150-200 ppm still not low enough but proves my point. that connections in an analog system are pure crap! If you want an accurate system you CANNOT have even a single 'temporary' connector. You must either use screw terminals, sacrificial crimp connections, or solder.!!!

I think the 200 ppm is still coming from the 'wet' 1/4 inch guitar plugs, but don't know yet. will investigate. Also, will see if data with that much noise produces anything worthwhile.

Again, the sequence is a three step cal and measurment:

1. both input channels connected together to the drive
2. from drive and channel 2 through 10k resistor to channel 1 to find Zin of channel 1
3. from drive/channel 2 through 10 ohms to channel 1 and C to measure C

However the soldered connections are: Drive and Channel 2 are soldered through 10k to channel 1 drive and hannel 2 are soldered through 10 ohm resistor to the C being measured.

so the drive is ALWAYS drug down by the 10 ohms to C and I used clips [spit, spit, curse begone!] to hold together leads to do step 1 and then step 3, but as you can see, step 2 everything is already connected.

I have a way to check the 'stabilitiy' of the data coming in for each step, I'm going to do that. if step 2 has excessive noise, then it's those !@#$#$@#$ 1/4 inch guitar connections! I feel like soldering cables straight to the PCB to avoid that problem! but that makes it difficult to disconnect later.

Hmmm. I did mention screw terminals, perhaps I can do that inside a shielded structure(s) a bit clumsy, but that may be what's necessary to get back down to that 8-10 ppm noise level I had at one time!

DUT,

Now i am worried. You seem to be treating the sound card inputs as if they were differential. They are not. That could be one contribution. Or i didn't understand the three setups.

If i am reading this right is like these:

Cal:

ch1 --------------------------* | | ch2 --------------------------*

Common ---*

Measure 1:

ch1 --------------------------* | Rsense | ch2 --------------------------*

Common ---*

Measure 2:

ch1 --------------------------*----------* | | Rsense DUT | | ch2 --------------------------*----------*

Common ---*

I would like to try something else:

Cal:

ch1 --------------------------* | | ch2 --------------------------*

Common ---*

Measure 1:

ch1 --------------------------* | Rsense | ch2 --------------------------*

Common ---*

Measure 2:

ch1 --------------------------* | Rsense | ch2 --------------------------* | DUT | Common -----------------------*

Measure 3:

ch1 --------------------------* | DUT | ch2 --------------------------* | Rsense | Common -----------------------*

?-)

find

from

shift

an

plugs,

Zin

those

to

Lemme try to 'member better. Back in the day we had HP 4191, 4192 and

4194 LCR meters and lots of parts to measure. Also several DMMs and a couple of Keithley 619s and some digitizing OScopes of various makes and models. Quality instruments helped a lot. The whole thing was computer driven via GPIB.

The temperature chamber suitable fixtures held 20 parts each and had interesting designs. For leaded parts we did 4 terminal connections and for SMT we did 2 terminal connections that were "converted" to 4 terminal in the fixture. (The connection to the SMT contacting beefy parts was "4 terminal". The moving contact has a short length of ultra flexible wire.)

From the fixtures it went to DL-96 connectors and coax to a GPIB controlled relay DUT switch then on to the instruments on coax to the 4 BNC connectors on the front panel.

Didn't have any problem maintaining a few counts repeatability in the 4th digit (~ 400 ppm). We also compared with manufacturers fixtures right on the front panels. We could do something like DC to about 10 MHz with good comparability.

?-)

On Fri, 07 Feb 2014 22:34:39 -0700, josephkk wrote:

The first renditions were NOT correct, your second set is.

First set up your Soundcard for unbalanced operation [Once I get a handle on what's going on here, I'll go back and set up for differential operation where you get more dynamic range and more possibility of removing the effects of pickup.

Here is the schematic [I think]: LCR_Meter.asc

Version 4 SHEET 1 27404 680 WIRE -128 -272 -208 -272 WIRE 0 -272 -48 -272 WIRE 352 -272 0 -272 WIRE -208 -240 -208 -272 WIRE 0 -240 0 -272 WIRE -208 -128 -208 -160 WIRE 0 -128 0 -176 WIRE -80 -16 -192 -16 WIRE 352 -16 352 -272 WIRE 352 -16 -80 -16 WIRE 432 -16 352 -16 WIRE 512 -16 432 -16 WIRE 656 -16 592 -16 WIRE 704 -16 656 -16 WIRE 1008 -16 704 -16 WIRE -192 0 -192 -16 WIRE -80 0 -80 -16 WIRE 1008 32 1008 -16 WIRE -192 112 -192 80 WIRE -80 112 -80 64 WIRE 1008 128 1008 112 WIRE -80 224 -192 224 WIRE 704 224 704 -16 WIRE 704 224 -80 224 WIRE -192 240 -192 224 WIRE -80 240 -80 224 WIRE 1008 240 1008 192 WIRE -192 352 -192 320 WIRE -80 352 -80 304 FLAG -208 -128 0 FLAG 0 -128 0 FLAG -192 352 0 FLAG -80 352 0 FLAG -192 112 0 FLAG -80 112 0 FLAG 1008 240 0 FLAG 656 -16 1 FLAG 432 -16 2 SYMBOL voltage -208 -256 R0 WINDOW 123 24 124 Left 2 WINDOW 39 0 0 Left 2 SYMATTR Value2 AC 1 SYMATTR InstName Vs SYMATTR Value "" SYMBOL res -144 -256 R270 WINDOW 0 32 56 VTop 2 WINDOW 3 0 56 VBottom 2 SYMATTR InstName Rs SYMATTR Value 560 SYMBOL cap -16 -240 R0 SYMATTR InstName Ccable SYMATTR Value 100pF SYMBOL cap 992 128 R0 SYMATTR InstName C SYMATTR Value {Ctest} SYMBOL cap -96 240 R0 SYMATTR InstName C1 SYMATTR Value 2.5nF SYMBOL cap -96 0 R0 SYMATTR InstName C2 SYMATTR Value 2.5nF SYMBOL res -208 224 R0 SYMATTR InstName R1 SYMATTR Value 10k SYMBOL res -208 -16 R0 SYMATTR InstName R2 SYMATTR Value 10k SYMBOL res 496 0 R270 WINDOW 0 32 56 VTop 2 WINDOW 3 0 56 VBottom 2 SYMATTR InstName Rsense SYMATTR Value 10 SYMBOL res 992 16 R0 SYMATTR InstName R SYMATTR Value {Resr} TEXT 784 296 Left 2 !.param Ctest=1000uF Resr=0.10 TEXT -160 192 Left 2 ;Ch 1 TEXT -160 -56 Left 2 ;Ch 2 TEXT 1840 296 Left 2 ;right edge TEXT 256 280 Left 2 !.ac dec 1000 20 1kHz TEXT 440 -184 Left 2 ;NOTE:\nfor calibration of ch1 to ch 2 ratio, Rs=0\nfor measuring Zin of ch 1, Rs=10k\nfor measuring DUT after measuring Zin, Rs=10 TEXT 624 -288 Left 2 ;LCR Meter using SoundCard

I learned on an HP Network Analyzer, very low f to 500MHz AND the instrument gave you best curve fit to match the model you select, like for a cap: C; or R and C; or Rs, C, L, and Rp! got really spoiled using that kind of stuff.

My goal here was to combine low cost, readily available insrumentation to create something very effective and otherwise possibly out of reach for a lot of people's budgets. Instrumentation that combines practically 'free' soundcard with some VERY simple lab equipment to then create a piece of instrumentation that would provide fairly accurate measurements, in situ, to quickly/easily identify those aging caps.

System only requires soundcard, resistor(s), ohmmeter ...and that's all! ooops and some cabling.

More info, I just soldered connections and sure enough the noise when measuring the ratio [where inputs are shorted together] dropped to around

200 into the 60 ppm range, which may, or may not, be coming from only the 1/4 inch guitar plugs now.

Interestingly, when measuring the Zin, the noise of ch 2 stayed constant as expected, but the noise of ch 1 jumped substantially to around 3 times the value of the noise in ch 2. interestingly this tracks the ratio of the square root of the increased impedance at the node. Went from approx 500 ohms with both attached to around 5k for ch 1 sqrt(10) is approx 3. I can't believe I'm down into Johnson noise of the resistors. more like a magnitude or more above it. but I'll check.

IMPORTANT CONCLUSION: Don't count on the connection made by taking a clip lead and clamping together two wires. IT DOES NOT MAKE A GOOD CONNECTION! mea culpa, I should KNOW better, but in defence, one does take short cuts to move faster, albeit as with most shortcuts, inevitably move slower.

FINALLY! got through the stuff.

Main problem was indeed the CONNECTIOONS! After soldering all the data became 'well-behaved' still had huge ppm error [in my mind] but was completely explainable by going back through and accounting for the input noise density function, both current and voltage types. The noise was actually less density function than I had measured earlier.

However, and this still has my mind boggled. Whether I use complete data, with all the !@#$#@!#$ noise, or whether I used 'fixed' approximations for things, like ratio of sensitivity for ch1 and ch 2 and then use fixed input impedances for the Zinput of ch 1, did NOT appreciably change the C and Resr measurements! just changed the characteristics of the noise. Still getting around 0.2% accuracy on the measurements.

So, I'm back to my questions:

1. Is the ratio of Resr/Xc supposed to be a related?
2. Does Resr skyrocket at lower frequencies? like up to 0.3 at 20Hz, then drop down to 'normal' 80 milliohms later?

Or, am I still fighting unknown errors here?

I could plot Q vs f it just seems strange that at 20Hz the Q is only 25.

Wet electros might well do something like that. At low speeds you have to push ions around, which is both slow and lossy.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

Speaking of... do you know offhand what kind of dynamics a supercap has? I expect the electric bilayer is that much more, well...dynamic than an electrolytic, which has electrolyte, but still is mostly about the aluminum oxide. The 'worst case' being actual batteries, which exhibit features of both, plus ionic potentials.

Anyway, ESR is all losses, not just the stuff that acts like it's in series (assuming the real component has many series and parallel elements). So, if you're calculating the loss component (which is the losses of all elements) and converting it to ESR, you'll see dielectric loss, absorption (at really low frequencies) and all other sorts of things.

Most things have some sort of intrinsic hysteresis or dielectric or absorption loss or whatever (and respectively for cored inductors), so that the Q tends to be fairly consistent with frequency, rather than inverse as a pure series resistance would be.

Published impedance curves often show real and imaginary components tracking (e.g., ceramic caps with a constant tan(delta) below resonance)... of course, whether that's real, or just tidied up data, could be debated.

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
Website: http://seventransistorlabs.com