I have done several experiments concerning IS of ceramic proton/ion conductors with the SI 1260 (but in combination with the SI 1296). Unfortunately I do not have access to ZView at the moment, so I am unable to have a look to your data. 

What is high frequency range in your case? I have in mind, that frequencies higher than 10^6 started to make problems in most of my measurements in unshielded setups.

Hi Thomas. I was hoping to get clean data up to 10^6 Hz and maybe some reasonable points up to 5x10^6 Hz. It is the frequency needed to see part of the bulk semicircle. I took a screenshot of the zview data. The measured data starts to deviate from the ideal situation somewhere between 10^4 and 10^5 Hz. 

Dear Yuxing,

It has been some years back that we did some critical assessment of Solartron Impedance Analyzer  and as I recall, the main problem has been a very poor compensation/calibration procedure of the instrument. We developed the software (along the lines of compensation/calibration procedure that is a standard in HP->Agilent->Keysight instruments), but Solartron was not interested.

All the stray impedances between the measuring plane of the instrument and your sample have to be eliminated. This is usually done by measuring the electrical impedance across the input 2 terminals of three elements,

THE SHORT ~ the stray series inductance,

THE OPEN ~ the stray parallel capacitance and

THE KNOWN LOAD (usually 50 Ohm but can be higher) ~ the stray series resistance. 

The three impedances form a "T" , the two "arms" are series inductance and series resistance, while the "foot" is the parallel capacitance. This three element impedance is in parallel with your sample , which is connected between the one arm of the T and the bottom of the foot.

Once the measured total impedance (T plus the sample) is corrected (sample impedance= total impedance with T impedance subtracted), you get the true sample impedance that agrees very well with Keysight instruments.

I hope that this will help

Best regards

Petr

Hi Petr,

Thank you very much for the information. Just want to check if I understand you correctly. The equivalent circuit with a test sample Z should look like that in the attachment?

Yuxing

Dear Yuxing,

yes, that is the idea. Once you get rid of the unwanted elements (Rs, Lstray and Cstray), you end up with true Zsample. If you draw the Law Terminal as a parallel line to High Terminal, you can see the "T", formed by the stray impedances, that I was trying to describe. In fact , this is an element,  describing the infinite coaxial line impedance (number of differential Ts going to infinity). 

With best regards

Petr

Hello Yuxing,

excuse my late answer, I had a few days off and I do not like to type complex answers on my cellphone.

I think Petr answer points to the right direction. I suppose, that there are instrumental artefacts, that contribute to your measurements. Measuring a known reference resistor or better RC would help to get rid of this. But I think a resistor of 50 Ohms like Petr suggested will not work for your sample, try to find a reference part with resistance in the range of kiloohms. (why are ALT Codes are not working for greek characters on this page?)

When looking on your Argand/Nyquist Plot you say, that you are "missing" your bulk semicircle in the high frequency range. What I can see there is "something" with a high resistance of nearly 2 kiloohm and with some inductive parts. I suspect, that you are not searching for a bulk resistance with such high values, so I suspect that this comes from your setup. I do not know your setup and especially your connections, but as I know from my own samples the connection between the cables coming from your instrument to your measurement setup could be the crucial point. We have worked with some sort of clamps, which caused problems at higher frequencies. Either a poor connection between your electrodes and your samples could be problematic, but I would not expect this at such high frequencies.

Dear Yuxing,

just a couple of further points :

1. I agree with Thomas. The LOAD for the compensation need not be 50 Ohms and a known impedance closer to your sample is better. Our own experience has been that a parallel R,C "calibration sample" (1 kOhm frequency independent resistor in parallel with  ~10 nF frequency independent capacitor led to even more clean data (sample impedance) than a purely dissipative 50 Ohm resistor.

2. Possible cause of the problem. Also here I agree with Thomas that it is probably un-compensated cables' impedance that might be the main problem. If that is so, then the cables'  self-inductance seems to be quite "huge".  I assume that positive theta in your plot is the capacitive response, while the negative theta is the inductive part of the response ?!. If that is not so, then we have a real problem ! The way to check the cables' influence is to measure your cables alone .

Best regards

Petr

I have one additional point (sounds crazy and I have no idea why, but for me it has worked):

To get better spectra always twist your cables (so that it look like plaited).

It will not help to get rid of your problem, bur my spectra always looked smoother. 

I usually have that problem when my range is wrong or when I use "auto" for the current range. Try to set it up manually.

Regarding the cable twisting, I was told by Solartron to do this, and it seemed to improve our measurements of a reference circuit. Another piece of advice was to provide a shunt between the grounds, which required wrapping a wire between the adjacent bnc shells.