开云体育

Hi Reginald
Interesting approach but have you considered that the calibration kits used then should be from the same supplier else you will have a spread due to these are having different frequency characteristics and thus add an uncertainty not justifying the nanoVNA hardware investigation you are looking for. Just the short used can be with welded center pin and other with turned disk/center pin, which will introduce a couple of picosecond variation. Likewise for the open end cap there are many versions I have at least 4 different models where the one form Amphenol RF has a disk that close to the center conductor of the female adaptor on the nanoVNA that it lead to high delays and even often create a short. Other types have different mechanical structures, actually the best open is nothing used for frequencies up to 1GHz.
The load will definitely have frequency dependencies where both the resistance and inductive / capacitive component influence the frequency characteristic. There will as well be deviation from sample to sample and from brand to brand.
These comment not to kill your project but just to raise some concern
Kind regards
Kurt

-----Oprindelig meddelelse-----
Fra: [email protected] <[email protected]> P? vegne af Reginald Beardsley via Groups.Io
Sendt: 11. september 2019 00:21
Til: [email protected]
Emne: Re: [nanovna-users] Calibration data statistics

I'm suggesting something very basic. Accumulate enough data to make a meaningful statement of the mean and variance of S11 and S21 using the parts supplied with the unit by doing uncalibrated SOLT measurements with two open measurements. The bare SMA connector and the supplied open. Both ports for 5 cases, and if possible, a 6th case with both ports terminated with a 50 ohm load. No unknowns. Just poor quality "knowns".

Many of the people who buy these do not, and will not, have access to anything better. An Amphenol cal kit from SDR-Kits is roughly 1/2 the cost of the nanoVNA. The Rosenberger cal kit is 2x the price of a nanoVNA.

This is hobby science. The purpose is to provide a statistical estimate of measurement uncertainty. Fundamentally calibration consists of assuming that any deviation of the measured values from the theoretical values is device and fixture error. In this case we also have significant errors in the calibration standards.

Once we have sufficient, i.e. several thousand, measurements, one can make pretty good estimates of the quality of measurements a buyer can expect. Aside from the practical utility, I think it would be an interesting study in the quality of low budget Chinese electronics manufacturing.

To collect the data I'll set up an email account for people to send their results to and write a program which will collect the data. I already have software which will compute the mean, standard deviation and the probability density function at each frequency. Then on a regular basis I'll rerun the statistical codes, plot and post the results.

It will take me a few days to develop the data collection routines as I'll need programs for multiple platforms. If a python program proves portable to Windows, Linux, Solaris and MacOS I'll use that, The TDR program uses that. So users have strong incentive to install python. And I figure it's about time I learned it.

At the moment the question is: 101 samples from 70 kHz to 900 MHz? 901 samples from 1 MHz to 900 MHz? or something different?

Have Fun!
Reg