LCR Meter Accuracy

[Wolfgang]

Hi David,

I've never seen the new AD parts in action in a measurement instrument. It looks that they are simply no competition for a "real" VNA from milliohms (PDNs) to 100Ks (crystals).

[JohnPi]

I just made a software Vector Network Analyzer (VNA) with my Rigol 1054Z and Siglent DS1025. It works from 1 Hz to 25 MHz and creates great plots and log files. I haven't added the software to de-embed the probes (or done extensive calibration), but that won't take long. Take a look at: https://github.com/jp3141/Vector-Network-Analyzer. You don't need anything extra other than a resistor.

Here's an example output:

p.s. I also have a DE-5000 and am happy with it, although I wish it also had a 10 kHz test frequency.

[Wolfgang]

Looks interesting, do you have an idea about the dynamic range your solution can provide ?

[JohnPi]

With the signal generator putting out 10 V, and the scope with 1 mV/division (closer to 100 uV/step) that calculates to 100 dB, but I don't expect much more than 70 dB. I just got it working and haven't had time to stress it yet (feel free to hack it). The scope only claims 80 dB or so isolation between channels.

It uses as much data as will fit in a 30 k sample window and extracts (correlates) the signal components at the test frequency. SNR is reasonable -- I think I pick up more from my environment and stray coupling (e.g. 10 V to 1 mV input range is actually hard to isolate). As you can see, though, the plots are quite smooth.

Here's a partial output from the log file showing 200 points per decade (stepping from 1 MHz to 10 MHz). Data is quite noise-free (Note the raw Sin/Cos phase varies because of the t=0 trigger point of the scope; it doesn't affect the results):

Code: [Select]

Sample 196, 9549925.860 Hz, 500 MS/s, 59948 points; 1145 cycles @       52.4/cycle
Ch1: Sin, Cos =    1.6443,   -1.5963; Mag =   4.58345, Phase =  -44.15 deg.
Ch2: Sin, Cos =    0.0063,    0.0021; Mag =   0.01333, Phase =   18.06 deg.
Ch2:Ch1 =  -50.73 dB @   62.21 deg.; Z =       1.3513 +j      2.5805

Sample 197, 9660508.790 Hz, 500 MS/s, 59986 points; 1159 cycles @       51.8/cycle
Ch1: Sin, Cos =    0.1239,    2.2847; Mag =   4.57618, Phase =   86.90 deg.
Ch2: Sin, Cos =   -0.0058,    0.0035; Mag =   0.01350, Phase =  148.84 deg.
Ch2:Ch1 =  -50.61 dB @   61.94 deg.; Z =       1.3825 +j      2.6099

Sample 198, 9772372.210 Hz, 500 MS/s, 59965 points; 1172 cycles @       51.2/cycle
Ch1: Sin, Cos =   -2.2754,   -0.2295; Mag =   4.57382, Phase = -174.24 deg.
Ch2: Sin, Cos =   -0.0026,   -0.0063; Mag =   0.01368, Phase = -112.60 deg.
Ch2:Ch1 =  -50.49 dB @   61.64 deg.; Z =       1.4156 +j      2.6388

Sample 199, 9885530.947 Hz, 500 MS/s, 59987 points; 1186 cycles @       50.6/cycle
Ch1: Sin, Cos =    0.0164,   -2.2851; Mag =   4.57037, Phase =  -89.59 deg.
Ch2: Sin, Cos =    0.0061,   -0.0033; Mag =   0.01385, Phase =  -28.25 deg.
Ch2:Ch1 =  -50.37 dB @   61.34 deg.; Z =       1.4482 +j      2.6658

Sample 200, 10000000.000 Hz, 500 MS/s, 60000 points; 1200 cycles @       50.0/cycle
Ch1: Sin, Cos =    1.5814,   -1.6418; Mag =   4.55909, Phase =  -46.07 deg.
Ch2: Sin, Cos =    0.0068,    0.0018; Mag =   0.01401, Phase =   14.97 deg.
Ch2:Ch1 =  -50.25 dB @   61.04 deg.; Z =       1.4830 +j      2.6969

Done
I intend to do a de-embedding step where I loop though the frequency steps without the unit-under-test connected and store that data to subtract from the measurements. However to expect 80 dB at 10 MHz needs a more sophisticated test fixture than a few crocodile clips.

[ogden]

I just made a software Vector Network Analyzer (VNA) with my Rigol 1054Z and Siglent DS1025. It works from 1 Hz to 25 MHz and creates great plots and log files.

Excellent work! Plots (luckily for you) are great indeed

I haven't added the software to de-embed the probes (or done extensive calibration), but that won't take long.

You mean open/short/load calibration? Indeed needed.

For owners of other than Siglent generator - could you possibly craft RLC software for manually configured generator frequency? I don't mind to manually set my generator to 100 KHz or 1MHz so I can measure capacitor or inductor at given frequency.

Some (not me) could maybe manufacture RLC add-on PCB with 1/10/100/1000 KHz generator on-board to have decent RLC meter that needs only Rigol 1054Z and your software

[Wolfgang]

Looks fine to me, congrats !!

 I had a similar idea of using a signal generator and a scope, but mine was using just one RIGOL MSO1104ZS (this one has built-in generators), so no extra generator would be needed. Instead of a swept input I planned to use a bandwidth-limited noise signal, and also a correlaction/coherence/FFT approach to fiddle out phase and gain.

More info can be found here:

electronicprojectsforfun.wordpress.com/a-loop-stability-measurement-solution-for-the-poor/

The intended purpose was Bode plots and the measurement of power supply output impedances.

The injector is ready, but what is still missing is a reference instrument to control my measurements. In late September, I will get my hands on a Keysight E5061B-3L5 (5Hz to 3GHz VNA) and on an Omicron Bode100, so I can make a lot of sample measurements that I can then compare with my approach. I will update my website as new results arrive.

Much success !

[JohnPi]

For owners of other than Siglent generator - could you possibly craft RLC software for manually configured generator frequency? I don't mind to manually set my generator to 100 KHz or 1MHz so I can measure capacitor or inductor at given frequency.

You could modify this line in the code:

Code: [Select]

  SDG1025.write("C1: BSWV FRQ, %11.3f" % TestF)to pause and ask the user to manually set the frequency. That's all it would take. However, don't expect any precision better than you set the signal source frequency -- the VNA correlates to this, and a frequency discrepancy will give unpredictable errors. With some work, you could add a frequency measurement loop to the software to remove this issue.

Some (not me) could maybe manufacture RLC add-on PCB with 1/10/100/1000 KHz generator on-board to have decent RLC meter that needs only Rigol 1054Z and your software

An Arduino (e.g. https://www.pjrc.com/store/teensy32.html would work well for this. It doesn't even need good sine waves -- because the VNA only extracts the fundamental sinusoid from the signal. You could also control the Arduino over serial/USB from Python also. Because the VNA measures both the forcing function and the response, it can use nearly any signal; you just need the frequency set correctly.

[JohnPi]

I had a similar idea of using a signal generator and a scope, but mine was using just one RIGOL MSO1104ZS (this one has built-in generators), so no extra generator would be needed. Instead of a swept input I planned to use a bandwidth-limited noise signal, and also a correlaction/coherence/FFT approach to fiddle out phase and gain.

That's a nice idea using a combined scope/generator.

Your SNR and dynamic range (30 dB only ?) will be horrible with a noise source and FFT. The scope doesn't have enough ADC resolution (8 bit) to remove the noise (in a reasonable time).

The Omicron Bode100 uses a 24-bit ADC and dedicated DSP to get great SNR and fast sweep speeds (mine takes about 100 points/minute).

[Wolfgang]

Its clear that the noise technique will need some averaging. On the other hand, I can take relatively long samples and I know what to look for - that could be used to cut noise down. On the other hand, noise instead of swept signal reduces the probability of dynamic range problems in case of resonances. Furthermore, the output noise spectrum can be screened for harmonics of the input.

Dont worry, its just a try and I am curious how far I can get.

[ogden]

You could modify this line in the code:

It was just "two cents" about how *you* can improve your "product".

Some (not me) could maybe manufacture RLC add-on PCB with 1/10/100/1000 KHz generator on-board to have decent RLC meter that needs only Rigol 1054Z and your software

An Arduino (e.g. https://www.pjrc.com/store/teensy32.html would work well for this. It doesn't even need good sine waves -- because the VNA only extracts the fundamental sinusoid from the signal. you could control an Arduino over serial/USB from Python also. Because the VNA measures both the forcing function and the response, it can use nearly any signal; you just need the frequency set correctly.

Interesting.. You are not using DCT? (I did not read code and most likely will not)

Did you test using square wave? Any side by side results to confirm that "It doesn't even need good sine"?

[JohnPi]

Its clear that the noise technique will need some averaging.

You'll need to average over N^2 measurements to improve performance by a factor of N. Rather than look for a single frequency each time, you can extract all data for a single band simultaneously (i.e. with the same set of measurements). However you won't be able to cover (e.g.) 1 Hz to 10 MHz in one band -- perhaps a band could only cover 1 decade ?

My technique steps frequencies and stops & measures at each one; I'd like to understand how I could use a non-stop frequency sweep instead, but that would require good correlation between the expected frequency and correlation signal -- it's more math and coding than I want to do right now. Something like this is used for audio analysis, and is useful for finding distortions, but that's not what I am looking for -- I expect a linear system.

[JohnPi]

You could modify this line in the code:

It was just "two cents" about how *you* can improve your "product".

Some (not me) could maybe manufacture RLC add-on PCB with 1/10/100/1000 KHz generator on-board to have decent RLC meter that needs only Rigol 1054Z and your software

An Arduino (e.g. https://www.pjrc.com/store/teensy32.html would work well for this. It doesn't even need good sine waves -- because the VNA only extracts the fundamental sinusoid from the signal. you could control an Arduino over serial/USB from Python also. Because the VNA measures both the forcing function and the response, it can use nearly any signal; you just need the frequency set correctly.

Interesting.. You are not using DCT? (I did not read code and most likely will not)

Did you test using square wave? Any side by side results to confirm that "It doesn't even need good sine"?

No. Feel free to peruse the code if you are interested in the details.

[ogden]

No. Feel free to peruse the code if you are interested in the details.

I am interested as far as you are. [edit] it seems you are not because to three questions of mine you simply answered "no" 

Look, I like (idea of) what you made but don't need it. All my interest here is: talk you into improvements, obviously only in case you wish to.

[Wolfgang]

Hi, sorry for not saying earlier, my intended application for Bode plots and PDN impedance needs to go up to a few 100kHz only.

I am measuring linear power supplies only, because I need the for noise measurements (no switching residue allowed here).

Some sample PSUs:

https://electronicprojectsforfun.wordpress.com/power-supplies/battery-operated-power-supplies/

The issue here is not the RFI introduced by a switcher or line input, but the noise existing in the regulators itself. This, along the the
output impedances, is what I want to measure.

To the second part of your question:

In a swept approach as you have it, you measure one frequency at a time. You start your generator, let it settle, measure input and output by amplitude and phase and compute a transfer function point by point.

In a random appoach, you use white noise. White noise has a flat spectrum. If you time-sample the input signal (into 50Ohms) and FFT it, a flat line should be the result.
Now you do the same thing for the output, and you get an FFT that is not flat anymore and you also get a phase curve due to the delay created by your DUT. If you do this for a sufficiently long time, you obtain a gain/phase plot like by the classic way.

The classic swept approach is restricted to linear time independent systems. A lot of systems are *not* strictly linear, and Keysight has blown up their top line of VNAs to extract models for (moderate) nonlinear behaviour of DUTs using a harmonic balance approach. This approach seems to be most useful for mobile phone base station transmitters where linearity iss key due to simulataneous multichannel transmissions. The amps are *very* linear, and the PNA-X VNAs help to optimize them to perfection. For really nonlinear stuff like Paramps, Class C stuff, subharmonics, varactor mixers, ... I dont think the harmonic balance approach is useful.

When your input is a random signal, you could assume a parametrized nonlinear model of your DUT and fit the parameters so that the measured output signal is best explained.
My idea was that this could work better than a HB approach due to more flexibility in the models used.

[JohnPi]

I checked with a square wave (and added a -q command-line option). It's not as good as a sinusoid.

At low frequencies it's nearly identical, but at higher frequencies, the slight mismatch between scope sampling and whole (square wave) cycles leads to signals 'leaking' and therefore not being measured correctly.  For example, at 15.8 MHz, the scope samples 1901 full cycles at 31.5 samples/cycle -- the 0.5 non-integer cycle is what causes leakage and misalignment between the scope and input frequency.

Code: [Select]

Sample 124, 15848931.925 Hz, 500 MS/s, 59972 points; 1901 cycles @       31.5/cycle
In principle there is a benefit in using a square wave since the amplitude of the fundamental sine wave in it is pi/2 higher -- thus improving SNR by 4 dB in theory. However the leakage at higher frequencies makes the overall result worse, and so I likely won't use it.

[ogden]

I checked with a square wave (and added a -q command-line option). It's not as good as a sinusoid.

Thank you for clarifying. Then for LCR adapter ad9833 + amp needed besides arduino. Still lot cheaper than Siglent DS1025. Well ... unless it is el-cheapo eBay "DDS generator.

I'd like to understand how I could use a non-stop frequency sweep instead, but that would require good correlation between the expected frequency and correlation signal -- it's more math and coding than I want to do right now.

Sample reasonable count of cycles (at least one) in the buffer so you can roughly calculate frequency from zero crossings and tune your numerically controlled local oscillator accordingly. After first correlation you will be able to re-calculate NCO frequency very precisely from phase rotation data (if any). This is more or less how communication modems lock onto transmitter frequency.

[bugi]

p.s. I also have a DE-5000 and am happy with it, although I wish it also had a 10 kHz test frequency.

Weird, my DE-5000 does have 10 kHz. After I power it on (at default settings), a single push on 'freq' switches from 1kHz to 10kHz.

[Gromitt]

Now there are other DE-5000's around, that will no doubt claim the same specifications as the IET Labs instruments, but I suspect are probably inferior.

How can they be inferior when they are the original.  IET DE-5000 was just a rebadge of DER EE DE-5000.

DE-6000 is a tweeked DE-5000, it's the same hardware.

[precaud]

Good project. I've put together several of these rigs over the years (hardware and software) and can offer some pointers.
: The key to accurate measurement is the de-embedding, fixturing, and repeatability.
: De-embedding (aka Open/Short/Load or OSL) is necessary at all freqs.
: Fixturing becomes more critical with increasing freq. Above 1MHz the fixturing becomes critical.
: Since you mentioned "with a single resistor" I'm guessing you're using the "Series-R" topology with the DUT to ground after it.
: Unless you're doing autoranging, and have high-sensitivity inputs, that topology is demanding on DSO dynamic range. An 8 bit DAC is going to be the limiting factor, especially at low Z's and high freqs. This is where swept FRA techniques hold the advantage.
: One can fairly easily get good results at middling freqs and Z's. The challenge is accuracy at the extremes.
: Ground loops are a killer at low levels (read up on "braid error"). With generator and DSO inputs all ground-referenced, you're guaranteed to have it.
: It really helps to have a set of components that have been measured at various freqs on known-accurate impedance meters to compare your results to.

Good luck!

[bugi]

Now there are other DE-5000's around, that will no doubt claim the same specifications as the IET Labs instruments, but I suspect are probably inferior.

How can they be inferior when they are the original.  IET DE-5000 was just a rebadge of DER EE DE-5000.

DE-6000 is a tweeked DE-5000, it's the same hardware.

IET may have had tighter manufacturing requirements (no missing components, no replaced components, discard any that do not meet some specifications, etc.). DER is now free to sell them out with whatever changes they like, and even the info on this forum already indicates there are indeed units with differences. Whether those differences will make them "out of specs" or "inferior" is something I have not seen any solid information about.

[Gromitt]

Now there are other DE-5000's around, that will no doubt claim the same specifications as the IET Labs instruments, but I suspect are probably inferior.

How can they be inferior when they are the original.  IET DE-5000 was just a rebadge of DER EE DE-5000.

DE-6000 is a tweeked DE-5000, it's the same hardware.

IET may have had tighter manufacturing requirements (no missing components, no replaced components, discard any that do not meet some specifications, etc.). DER is now free to sell them out with whatever changes they like, and even the info on this forum already indicates there are indeed units with differences. Whether those differences will make them "out of specs" or "inferior" is something I have not seen any solid information about.

The are still the same meter manufactured by DER EE (not DER). I don't think they would downgrade it under their own name.

IET did not manufacture the IET DE-5000 (or DE-6000), DER EE did.

[JohnPi]

Weird, my DE-5000 does have 10 kHz. After I power it on (at default settings), a single push on 'freq' switches from 1kHz to 10kHz.
[/quote]
You're correct -- I misremembered.

[JohnPi]

Thanks for the pointers.

: De-embedding (aka Open/Short/Load or OSL) is necessary at all freqs.

I intend to do this -- I just got the software in a state where it works well enough to move to those steps. In fact, while I do have nice smooth plots, I have not confirmed that they are accurate.
Do you have any examples or photos of a fixture setup that might be suitable ?

: Since you mentioned "with a single resistor" I'm guessing you're using the "Series-R" topology with the DUT to ground after it.

Yes. I haven't determined what is the best value to use -- it will depend on the DUT. I may need de-embedding for different values; that'll depend on the accuracies I can achieve.

: Unless you're doing autoranging, and have high-sensitivity inputs, that topology is demanding on DSO dynamic range. An 8 bit DAC is going to be the limiting factor, especially at low Z's and high freqs. This is where swept FRA techniques hold the advantage.

Yes, it autoranges. One nice feature of the Rigol is it has arbitrary values for the y-scale. I basically set the y V/div to make the measurement span +/- 3 divs. If the Q isn't too high and levels don't change too much between frequency steps, this works well. In fact the Rigol only uses +/- 100 counts for full screen display, so there is some over range available. Later I'll add a check for over range and rescale if needed. For the 1st frequency measurement, I do it twice -- once to get a good autorange value.
On the Rigol I also calculate and send in an arbitrary x-axis value; it rounds to the next (higher) step, and I read back the actual timescale.

: Ground loops are a killer at low levels (read up on "braid error"). With generator and DSO inputs all ground-referenced, you're guaranteed to have it.

I haven't noticed a problem -- not even 60 Hz pickup.

: It really helps to have a set of components that have been measured at various freqs on known-accurate impedance meters to compare your results to.

Definitely. However I'm most interested in the general trend of the responses, I don't need expect better than a few % accuracy. My main use will be  for measuring power supply loop gain characteristics and output impedance.

[precaud]

Do you have any examples or photos of a fixture setup that might be suitable ?

I'll post a pic of one later. I was going to make several, but ended up finding a set of the Omicron fixtures for pretty cheap and so I use them.

Yes. I haven't determined what is the best value to use -- it will depend on the DUT. I may need de-embedding for different values; that'll depend on the accuracies I can achieve.

Yes, there are interesting tradeoffs that come with the choice of reference resistor value (and Load resistor value, too). The higher the Ref R, the greater the dynamic range between channels is needed for low-Z. For a "one-size-fits-all" fixture I settled on 33 Ohms.

Yes, it autoranges. One nice feature of the Rigol is it has arbitrary values for the y-scale. I basically set the y V/div to make the measurement span +/- 3 divs. If the Q isn't too high and levels don't change too much between frequency steps, this works well. In fact the Rigol only uses +/- 100 counts for full screen display, so there is some over range available. Later I'll add a check for over range and rescale if needed. For the 1st frequency measurement, I do it twice -- once to get a good autorange value.

Autoranging comes with its own set of problems. Now the inaccuracies from range switching is part of the equation. It puts you squarely in the territory of real LCR meters with this - various "range resistors" and receiver gain ranges, all with their own freq and level trims, to cover a wide Z range. OSL does not compensate well for this...

: Ground loops are a killer at low levels (read up on "braid error"). With generator and DSO inputs all ground-referenced, you're guaranteed to have it.

I haven't noticed a problem -- not even 60 Hz pickup.[/quote]

You'll see it when you start measuring things below say 100mOhm...

: It really helps to have a set of components that have been measured at various freqs on known-accurate impedance meters to compare your results to.

Definitely. However I'm most interested in the general trend of the responses, I don't need expect better than a few % accuracy.[/quote]

That's easy to do above say 1 Ohm. But below 100mOhm, a couple % is difficult...

My main use will be  for measuring power supply loop gain characteristics and output impedance.

For output impedance I would definitely concentrate on the sub-100mOhm aspect... most regulated power supplies have output Z well below 1 Ohm...

[ogden]

Autoranging comes with its own set of problems. Now the inaccuracies from range switching is part of the equation. It puts you squarely in the territory of real LCR meters with this - various "range resistors" and receiver gain ranges, all with their own freq and level trims, to cover a wide Z range. OSL does not compensate well for this...

Particular tool shall be considered as RF I-V impedance meter, with all the consequences (of inability to measure extreme Z ranges). Just proper I-V bridge is needed and that's it. Well, maybe two bridges - one optimized for low-Z and another for hi-Z.