impedance analyzer AD2 - Accuracy of Impedance Analyzer

[ProbeT]

I am trying to make precise measurements of capacitance in the <100pF range. I am using the Impedance Analyzer board for the AD2. I may be asking too much of it but it seems to not represent the true capacitance variation with frequency. In measuring a 33pF COG disc ceramic, which I expect to have a constant capacitance, at least over 1lHz to 10kHz, I note that there is a change in the value with frequency of  0.1pF (~0.3%). Is this a realistic result?  I am using 1 MOhm loading resistor.

[attila]

Hi @ProbeT

Have you performed at least open compensation? 

The 100pF at 1Hz has about 1.6GOhms which with 1MOhm option is compared to 500kOhms (1M resistor | 1M probe). This is a very high 1/3k proportion which will have low precision with this measurement.
The 100pF at 1kHz has 1.6MOhms which is well comparable to 0.5M

The PCB and cabling can have several pF crosstalk which is eliminated/reduced by open compensation. The scope channel measurement inequalities are reduced by short-compensation.

See the related posts:

 

[ProbeT]

Hi Attila,

I have performed compensation for the range I am using, 1MOhm. I have redone everything as carefully as I can and I was able to get 1 pF capacitor repeatable to ~0.01pF with calibration in between. The same for ~150pF returned 5 values of 149.8 with one instance of 149.7 in between. There is though a persistent droop of about -1.5pF between 4kHz and 10kHz, where I used the difference cursors to obtain the results.  I can't at this stage reconcile the droop with what I understand to be the characteristics of a small value COG capacitors. I understand that you don't regard this device as being of precision instrument quality but I wonder if there is not a way to improve its performance.

Some questions:

Can you provide detailed information on how the calibration is performed?

Is there any way to accumulate the measured results that are available in the file -> export dialogue into a larger array with all results taken? E.g if you do 10 runs, so that the export contains the data for each of the runs.

Is there any advantage in having it auto choose the resistor during measurement to always give the best possible accuracy in any part of the spectrum?  I expect that it would require a calibration of all resistors before measurement.

 

 

 

 

 

 

 

 

 

 

[attila]

Hi @ProbeT

The ceramic capacitors are very sensitive to temperature, like touching, 22*C to 37*C, can make several % of difference.

You have Averaging expressed in time and Average option expressed in number of measurements. When both set the condition is AND logic.
The sweeps/scans while running are not averaged, the average/ing is performed for each step.

The constant voltage/current and custom modes are useful for many decade analysis or highly varying impedance.
These can change the resistor value to satisfy the requirement, but changing the resistor will change the reference impedance, so it might change the measurement result.
When using the above options it is good to have compensation on all the resistors.

Compensation:

 

Related post:

 

[ProbeT]

Hi Attila,

I was using NPO types and tweezers to handle etc, so temperature sensitivity is much less. I also have access to a gas filled reference capacitor, which is stable to better than 0.01% over a wide temperature range.

It may be a longer term project, but I will play around with the modes a little to see what the best arrangement is.

 

 

 

 

 

[ProbeT]

Further to above, I am now using a transconductance amplifier to measure the capacitor current at 1kHz so as to boost the signal to a good level and overcome the need to use high impedance.  The results are coming out very stable without using the impedance adapter, close to expected value without calibration. But I notice that the spectrum analyzer display does not seem to produce spectral lines that match the x-axis. When I set the spectral display to a narrow span to zoom in on the 1kHz line produced by the AD2 function generator the indicated frequency points appears to be out by at least one spectral line. Seems to be a plotting error?  I note also that I was expecting a spectral line to fall exactly at 1kHz as that was the frequency that I thought was coming out of the function generator, but that the table of spectral results had the result at 1000.02346096 Hz. Is this correct, and if so why does it not appear at 1000.023?

From the graph you can also observe that the magnitude of the points either side of the 1kHz peak are not symmetrical, which I expect is the consequence of issues with the the FFT processing.

[attila]

Hi @ProbeT

It would be useful if you could post some screenshots, Alt+PrtSc and Ctrl+V

With the default Auto Freq. Range option the sampling frequency is adjusted based on Stop frequency. You can select other options to have fixed sampling rate.
The cursor (quick measure) points to the peak BIN, which can be a bit off due to resolution.
The frequency measurements (FF, harmonics) use weighted average on the lobes to calculate peak more precisely.

 

At a different frequency: