FreeCalypso RF calibration revisited

[Mychaela Falconia]

Hi DS!

Thank you for your observation regarding Tx power level tolerances.  I
am of course aware of the GSM 05.05 spec and of the quite generous
tolerances it gives.  However, take a look at the table on page 26 of
this ancient TI document:

ftp://ftp.freecalypso.org/pub/GSM/Calypso/rf_calibration.pdf

And at Table 4.8 on page 44 of this more recent TI document:

https://www.freecalypso.org/LoCosto-docs/Production%20test%20and%20calibration/i_sample_rf_test_and_calibration_13_03_04_01991%20-%20v026.pdf

Notice how in both cases TI had set the calibration target for the
highest power level in each band to less than the spec value.  For the
low frequency bands the top power level per the spec is 33 dBm, but TI
had their target set to 31.8 dBm in the earlier days and to 32.2 dBm
on the I-Sample (LoCosto).  For the high frequency bands, the
respective numbers are 30 dBm per the spec, 28.8 dBm in TI's old stuff
and 29.2 dBm on LoCosto.

When I was developing the Tx calibration code in fc-rfcal-tools, I
quickly saw why this reduction in the top calibration target number is
necessary.  There is a limit to how much output power the DUT can
produce, and once this limit is reached, one cannot get any more dBm
out no matter how high the APC DAC is set.  If the automated Tx
calibration program is given a calibration target number that is
higher than what the DUT can produce (whether it be a true DUT hw
limitation or a result of errors in the measurement setup), the
process will go haywire, hence the automated calibration program needs
to be given attainable target numbers, even if they have to be shifted
down a little from the ideal numbers given in the spec.

But the question for FreeCalypso device production is: what target
numbers should we use on *our* production line?  Should we use the
1.2 dB reduction in the top power level with 0.5 dB reduction in the
next one down like the ancient Sara document says and like FIC
(Openmoko's factory) appear to have done, or can we get a little
closer to the ideal and only reduce the top calibration target by
0.8 dB like TI did for LoCosto?

Whatever numbers I eventually decide on, I will need to be able to
justify them, and I would really like to be able to give a better
justification than "my CMU200 is out of calibration, I don't know if
the numbers it gives me are the real truth or not, I don't have the
money to get it properly calibrated, so I just set my txlevels targets
so that fc-rfcal-txband gives the appearance of passing in my setup".

The good news is that getting my CMU200 properly calibrated may be
more affordable than I previously thought.  I was told earlier by some
suit that it could cost somewhere around 15 kUSD, but then a different
person who appears to be quite a bit more knowledgeable about CMU200
instruments in particular as opposed to "general test & measurement
industry" told me that:

a) the cost was about $1600 the last time he checked, and

b) the only trustworthy shop to do this calibration would be R&S
themselves; for people like me who are based in USA-occupied Incalia,
R&S have a service facility on the East Coast of this continent.
Because I am on the opposite coast, careful packing and shipping of
the instrument will be required.

I have not yet tried calling R&S USA and asking them if the ~$1600
price to fully recalibrate a CMU200 is accurate - instead I figured
that I should save up the $1600 first, and when I have that much
money, start making more serious inquiries.  Because of some upcoming
family expenses, I probably won't start making those serious inquiries
into getting my CMU200 calibrated until the spring of 2018, and until
then I will have to ship FCDEV3B boards with Tx calibration that
merely *appears* to be spot-on in my setup - but like DS pointed out,
the GSM 05.05 spec tolerances are so generous that my current lousy
calibration is almost certainly within those tolerances.

M~