FCDEV3B sleep mode investigation update

[Mychaela Falconia]

Hello FreeCalypso community,

Yesterday I got another batch of FCDEV3B boards from the assembly shop,
this time using RoHS solder and newly-bought Calypso, Iota and Rita
chips which are exactly the same versions as what Openmoko used.  This
board assembly run was another installment in the series of experiments
seeking to shed light on the very mysterious sleep mode bug, and the
results are:

* The sleep mode bug is still there, totally unchanged from the
  previous board builds which were done with SnPb solder and chips
  from 2013.  This result means that the chips from 2013 are not at
  fault, and that the slightly higher electrical resistance of SnPb
  solder is not at fault either.

* There are a total of 4 boards in this batch, but I was only able to
  test the RF tract (with the CMU200) on 3 out of the 4: one of the
  boards did not get the Q401 dual PNP transistor part populated (part
  shortage, more parts are on order), and without those transistors
  the Tx path cannot be turned on.  And on all 3 tested boards there
  is a problem in the RF Tx path:

* On two of the boards the high band Tx path (DCS and PCS bands) works
  fine, but when I turn on Tx for the low band (900 MHz), the output
  power levels are way low.  The Rx path is good for all 3 bands.  It
  is exactly the same behaviour as was seen on one of the boards in
  the very first batch, and points to a problem in the Tx path for the
  low band, the most likely suspects being the RF3166 PA (has separate
  internal paths for low and high bands) or the antenna switch (has to
  have separate paths for each Tx and Rx by the very nature of the
  switch business).

* On one of the 3 RF-tested boards the high band Tx path exhibits the
  same problem as the one for the low band: the power levels that come
  out are way low.  Furthermore, on that same board there is an oddity
  in the Rx path for the 1900 MHz PCS band: the measured "magic gain"
  (GMagic) is 181 half-dB units, whereas the expected value is around
  200.  That is a 10 dB difference, i.e., we've got 10 dB of extra
  unexpected loss somewhere in the PCS Rx path on this board, which
  does not happen on any other board or in the other bands (including
  the almost-identical DCS Rx path) on the same board.  The finger of
  suspicion points to the antenna switch.

It is certainly strange that all 3 of the RF-tested boards from this
batch of 4 exhibit almost the same breakage in the RF section, while
we haven't had too many problems in this area on our previous board
builds: the first batch had two defects in this area (one board with
similar low dBm output in the low band, and one with totally dead RF,
likely bad Rita chip), and the second batch had no defects at all.
The most likely explanation that comes to my mind is that either the
RF PA or the antenna switch (more likely the latter, given the PCS Rx
path strangeness on one of the boards) did not like the higher heat of
RoHS soldering, or perhaps some handling requirement was accidentally
violated.  I will talk to Technotronix folks about it later, but my
primary focus right now is on the sleep mode bug.

As for the sleep mode bug, the next relatively easy experiment I plan
on trying is to remove the Spansion S71PL129NC0HFW4B flash+RAM chip
from one of these boards and populate Openmoko-matching Samsung K5A3281
in its place.  I call this experiment relatively easy because my
friends at Technotronix have a BGA rework station and they seem to
have no difficulty with using it - otherwise it would be near
impossible.  I am hoping that we'll be able to perform this memory IC
swap experiment some time next week.

Hasta la Victoria, Siempre,
Mychaela aka The Mother