13 Jan
2019
13 Jan
'19
10:46 p.m.
Harald Welte wrote:
[...] the big difference is that the mobile has been
calibrated to
(if I remember) 2dB accuracy,
The GSM 05.05 spec indeed sets a tolerance of 2 dB under normal
conditions or 2.5 dB under extreme conditions on the mobile station's
maximum power output level for each band. These tolerances get looser
for the lower power control level, up to 5 dB under normal conditions
or 6 dB under extreme conditions for the lowest PCL.
However, as someone who is intimately familiar with the actual
calibration procedures performed by MS manufacturers (just did it
myself less than 48 hours ago for a new batch of boards), I can tell
you how it is actually done. There is a table of target power levels
in dBm, and the calibration station goes through the full set of
allowed PCLs (5-19 for EGSM and GSM850, or 0-15 for DCS and PCS) and
steers the Tx power output (as measured by the test instrument,
usually R&S CMU200) for each PCL to the dBm number in the table.
But there is a hack in this setup which most MS manufacturers fail to
disclose: with all GSM MS RF designs I have laid my hands on so far,
including the one I've inherited from Openmoko (OK, I admit that all
of my experience is with old stuff, not any of the recent stuff from
MTK or Spreadtrum, but don't blame me, blame MTK and Spreadtrum for
not publicly releasing their full chip docs and reference firmware
source code), the MS hardware is oftentimes not actually capable of
putting out the maximum power output prescribed by the spec (33 dBm
for low bands or 30 dBm for high bands) while staying in the linear
range of the PA (the range in which the relation between the APC
control voltage and RF Vout is mostly linear), and thus the calibration
procedure has to be fudged. In the case of Openmoko, they set the
calibration targets for the highest power levels to 31.8 dBm and
28.8 dBm instead of 33/30 (for low and high bands, respectively), the
next PCL down is set to 30.5/27.5 instead of 31/28, and then the
proper official numbers from the spec further down. I have to do the
same on my current GSM MS boards: when I tried putting out the full
33/30 dBm at the maximum PCL as the spec calls for, I can usually hit
it if I crank the PA really high, but then it goes out of its linear
range, and lacking better RF kung-fu, I have to assume that it's bad..
The maximum Tx power level fudging done by manufacturers in my or
Openmoko's position is 1.2 dB, which is less than the spec tolerance
of 2 dB, thus manufacturers like OM who did this fudging had no
problems passing type approval tests, but I really wish I could find
someone with more professional GSM MS design and manufacturing
experience who could explain what actually happens, why we are not
able hit the highest power levels while staying in the PA's linear
range (the PA datasheet says it can put out 34.2 dBm in the low bands
and 32.0 dBm in the high bands), and how the RF hw design might be
improved so we can put out the highest power levels without fudging.
Aside from the just-described fudging, the actual calibration tolerance
(the maximum error between what the calibration procedure aims for and
what actually comes out as measured by the CMU200) is less than 0.7 dB
- and those ~0.5-0.6 dB errors happen mostly toward the lower power
levels where the spec tolerances are the loosest; the maximum power
level is calibrated to the shifted target number to better than 0.1 dB.
M~