Hi Warren,
There are a couple of real world takeaways from the measurements I think.
One point is that the trace is not anywhere near 60dB at all freqs. It is
only around 25dB RL on the white unit at the highest end in the first test.
To measure RL with a degree of confidence we need about 10dB of 'headroom'.
So that allows you to measure down to maybe 15dB with a few dB if error.
Errors in RL measurements are huge if you get close to the effective
directivity limit of the vna.
If you tried to measure a 20dB RL device then the actual result could be
anywhere in a 10 or 12dB range.
It does not behave a spectrum analyzer and this is not a noise floor. An
antenna measurement for example will often show a dip in the trace well
below this 'accuracy floor'.
That measurement will be rubbish and I will eat my hat if most people don't
fall for it.
The second point is that saved calibrations are pointless if the device
drifts all over the place after calibrating.
The white device drifts more and so for more confidence in that higher
range, it needs to be powered on 'for a while', then a cal done, then the
measurement taken.
That is the same procedure you use for a lab VNA.
The hugen device could be turned on, and a measurement taken right away
using a saved cal.
That procedure matches how you use any other antenna analyser.
Both of the above points are relevant to the simple 'antenna analyzer' use
case. Which is turn it on, attach it to the feedline, and take the
measurement.
The worse the accuracy is then the more expert you need to be to make use
of the device.
I have added the open port tests. After doing a calibration and leaving
overnight the white unit has drifted away from 0dB much more than the hugen
device.
This lends more weight to the above comment - the white unit should be
calibrated immediately before any use.
The bottom line is that I see clear differences between the different units
I have at least.
Roger
[image: image.png]
toggle quoted message
Show quoted text
On Tue, 3 Sep 2019, 5:35 AM Warren Allgyer, <allgyer@...> wrote:
Roger
I am reviewing your comparison results on my phone so I may have missed
something.
What I see is comparisons of the S11 noise floor in the 60-70 dB range
return loss. This is meaningless in the real world for all but the most
critical applications, like comparing the return loss of an N connector
versus an SMA for the example.
A return loss of 30dB equals a VSWR of 1.06:1. A return loss of 40 dB
equals a VSWR of 1.02:1 and is exponentially more difficult to achieve and
measure accurately. For all but the most critical lab measurements return
loss values greater than 30 dB are meaningless. They are also much more
sensitive to calibration errors.
Comparisons of noise floor are interesting and do indeed likely reflect
more care, shielding, and execution against a design. They do not, however,
represent an indication of comparative measurement accuracy at usable
return loss values.
The comparison you could do to measure real world comparative performance
would be to use a 100-500 ohm resistor and measure return loss and other
parameters at, say, 50, 450, and 850 MHz. I will eat my hat if you find
more than 1 dB of difference between the units and I expect it will be far
less than that. I do not have a Hugen unit but the variance between my
worse and worst units for these tests is less than 0.3 dB. NOTE: all units
being compared must have been carefully calibrated using the same set of
OSLIT loads!
In summary, a difference in noise floor between 60 and 70 dB is
meaningless in the real world and has no bearing on the accuracy of the
instrument for normal ranges of use.
WA8TOD
On Sep 2, 2019, at 8:57 AM, Roger Henderson <hendorog@...> wrote:
I have posted some comparisons here between my white 'Gecko/Salamander'
device and my hugen unit:
In all these traces, the lower the trace, the better. If the device is
theoretically perfect then the trace will be at some ridiculously low
value
as it will ultimately be limited by the mathematical precision in the
calculations.
This is what happens if I save measurements and transfer them to a PC,
do a
calibration and then apply the error correction to the load.
Since these are real physical devices, and they drift around so each
sweep
will be slightly different - even when measuring the same load and not
touching anything.
So perfection doesn't happen, and the trace will slowly float upwards as
the device shows its limitations.
First test file was saved after calibrating both devices right after
power
on.
Second test file was saved after leaving them both on for an hour and
then
calibrating.
Tests 3, 4 and 5 were saved after increasing lengths of time.
There is a clear advantage in the hugen trace after each of the
calibrations.
Over time the two traces eventually come together.
The white unit is quite poor at 900MHz unless it has a good amount of
warm
up time.
The white unit is much worse at the very lowest frequencies in the trace,
and note that it had a Start freq of 1MHz. This was my mistake as the
hugen
unit was starting at 50kHz.
The other test I should do is to see how quickly the trace drifts away
from 0dB - after calibration and with the test port open.
I know the white unit is worse there too, just from informally testing
it,
but it would be good to do it properly.
Also, it seems logical, well to me at least, that sealing up the sides to
limit air currents will improve these stability type results on both
devices. I haven't tried that though.
Roger
On Mon, 2 Sep 2019 at 19:18, Warren Allgyer <allgyer@...> wrote:
Hugen
With all due respect, the thread you cite says clearly there is no
operational difference among the fully shielded, "better" versions, the
"worse" clones, and the worst. No one has presented operational data
that
distinguishes clone performance at any level. The thread actually says
slightly better performance was had from unshielded clones.
The images presented on the data comparison are indeed cut-and-paste
images from advertisements but they are identical to the actual VNAs in
my
possession. I have posted an actual picture of my units as well so you
can
see the images used for identification are identical.
The data I have published shows no significant difference between the
two
clones.
So let's review:
1) No one, including you, have published any data or test results
demonstrating a significant performance advantage for the "real" unit
versus the clones.
2) No one, including you and me, has been able to publish data showing a
significant performance difference among the clones.
3) I have stated but not published data (I am happy to do so) where I
compared the results of my two clones to a spectrum analyzer/tracking
generator/RF bridge combination and found no significant variance in
results among all three devices.
The qualities of the original versus the clones...... so far..... have
been distinctions without differences. Those differences have been
cosmetic
only and no one has shown a resulting measurement discrepancy.
I challenge you or anyone on this board to publish actual data showing
that a clone performs significantly worse than an "original". I don't
think
that has been done. The thread you cite certainly does not support
measured
differences. If you have data please show it. If not then I think we all
need to stop confusing the readers here. Some people have actually
returned
what are likely perfectly good devices because of cosmetics.
WA8TOD
