Triband status quo
==================

Our current Openmoko-based Calypso+RF modem core is very very good, but it has
one shortcoming compared to TI's Leonardo+ reference design: it is triband
rather than quadband.  This triband restriction stems from OM's use of discrete
antenna switch and SAW filter components as opposed to an integrated FEM (front
end module) like on Leonardo+.  In addition to the band restriction, our current
triband RF design suffers from one other very unpleasant problem: we have no
datasheet for the antenna switch component which we have to use.  We know from
Openmoko's BOM data that the manufacturer is Darfon and that the part number for
this antenna switch component is ASM4532T0P06-1, we are able to buy this part
from our Chinese grey market suppliers, we build our boards with these parts and
our boards do work perfectly fine when we get a good batch, but we have to do
this entire process blindly, without any datasheet or other documentation for
this mystery part.

While we lack any official documentation for our triband ASM, we know its basic
logical function: it has two Tx inputs (low band and high band Tx coming from
the PA), 3 Rx outputs going to 3 separate SAW filters for the 3 supported bands,
and 3 switch control inputs.  Two of these switch control inputs are Tx controls
(low band Tx enable and high band Tx enable) which appear to be the same across
all common RFFEs, whether they are 2-band (single region), triband or quadband.
The 3rd switch control input gets a logic high voltage applied to it during PCS
band Rx, thus it appears to be a switch that diverts the high band Rx path
between DCS and PCS SAW filters.  The same arrangement is found in most other
triband phones and modems from that era, i.e., they also have two Tx switches
and one Rx path switch selecting between DCS and PCS, plus 3 discrete SAW
filter components (outside the ASM) for the 3 bands.

This article outlines some ideas for how we may be able to move from this RFFE
to a different one, replacing our current mystery ASM with something less
mysterious and better documented, and improving our radio capability from
triband to quadband at the same time.

Epcos M034F
===========

TI's Leonardo+ and E-Sample boards used a magic component made by Epcos (the
canonical SAW filter manufacturer during that era) called M034 or M034F (the
exact proper designation is unclear).  It was an integrated quadband FEM,
integrating the antenna switch and SAW filters in one component package, with a
special twist.  The special twist is that even though there are 4 separate Rx
band SAW filters inside that M034 "chip" module, corresponding to its advertised
quadband capability, only 3 Rx signal path differential pairs come out of it,
neatly corresponding to the 3 LNA inputs on TI's Rita transceiver.  This twist
is important because even though the Rita transceiver itself is fully quadband
internally, it has only 3 LNA inputs, with GSM850 and EGSM bands sharing the
same LNA input while each of DCS and PCS get their own.

We do have an M034F.pdf datasheet for this magic component (came along with
Calypso and Leonardo docs), and the block diagram on page 6 shows the magic
quite clearly: there is a baseband-controlled switch selecting between EGSM Rx
and GSM850 Rx (in addition to the two usual Tx switches), this switch directs
the low band Rx path toward one of two different SAW filters, and the outputs
of those two filters are then joined.  The high band Rx path always goes to both
DCS and PCS band SAW filters, and each of those high band Rx SAW filters gets
its own output going to its own dedicated Rita LNA input.  Note the lack of a
baseband-controlled switch between DCS and PCS in the high band Rx path: this
switch is present in all triband RFFE designs I have seen, thus a big question
is raised as to how this magic M034 component functions without one.  I can
think of two possibilities:

Possibility 1: perhaps they do a 50/50 split of the total incoming energy
between DCS and PCS Rx paths, with each path suffering by 3 dB as a result.

Possibility 2: perhaps by virtue of integrating the ASM and the SAW filters
into a single monolithic FEM, Epcos found some way to have unswitched DCS and
PCS Rx without incurring that 3 dB penalty.  Perhaps they successfully
implemented some form of frequency diplexer such that out of the total incoming
energy picked up by the wideband antenna, DCS downlink frequencies go through
the DCS Rx SAW filter, PCS downlink frequencies go through the PCS Rx SAW
filter, and no needless losses are incurred.  This hypothesis is supported by
the observation that the available M034F.pdf document gives approximately the
same insertion loss numbers for all 4 Rx bands, i.e., the same between the
switched low bands and the unswitched high bands.  Note that they could not
have similarly eliminated the GSM850 Rx switch: both EGSM Rx and GSM850 Rx need
to go to the same LNA on the transceiver, thus a switch is needed somewhere.

I (Mother Mychaela) would absolutely love to play with an M034-based quadband
Calypso+Iota+Rita board in my lab with the trusty CMU200 instrument, and to see
how well it actually performs, especially in comparison with our current
OM-based triband version.  However, in all of my years of searching I have never
found a physical Leonardo board (any version), nor have we ever found any
Leonardo PCB layout files which would allow us to build a modern recreation -
thus the magic of M034 remains elusive.

Unless a miracle happens and we are able to obtain either a physical Leonardo+
board or a PADS PCB file for one, there is no quick or low-effort way to "just
try" this M034 RFFE.  Instead if we wish to build a FreeCalypso board with this
RFFE, it would have to be "the full 9 yards": a full-blown PCB design and layout
effort.  There is no way to just "drop" the M034 into our existing PCB design
in the place of our current triband RFFE, it would have to be either a very
disruptive RF section layout change or an entirely new PCB layout, making this
idea very open-ended - an open-ended venture with quite uncertain outcome, but
with a high dollar cost attached to it.  Given the massive effort required and
PCB layout labor costs, I currently have no active plans to pursue this idea
beyond hypothetical.

Commissioning a new custom RF FEM
=================================

Here is a wild thought: what if instead of twisting over backwards trying to
hammer an existing RF FEM like M034F into our not-quite-fitting PCB design, we
were to get an entirely new FEM made specially for us, made exactly the way we
need it?  If we were to venture that way, I would ask for a FEM very similar
conceptually to M034F, but with a few changes:

1) Instead of diplexing between DCS and PCS SAW filter inputs with a 50/50
   energy split, implement another switch (just like the GSM850 Rx switch) for
   PCS Rx, exactly the same way how it is done in classic triband designs like
   our current OM-based one.  This change should eliminate the extra 3 dB
   penalty which I assume (for lack of experimental data) must happen with the
   existing M034 FEM.  Or as an alternative to making this change, if someone
   who is more knowledgeable than me in this area can explain to me why it isn't
   necessary, I would accept that option as well.

2) I would ask for a rearranged pinout: the existing M034F pinout does not fit
   at all into our OM-based PCB layout, but it would fit much better with some
   rearrangement.

3) The hypothetical M034-like FEM would fit into our OM-based PCB layout a lot
   better if it were made a little smaller than the 8.2x5.5 mm size of M034F.
   Considering that the original M034F was created some 15-16 y ago, I assume
   that it should be possible to make a smaller version in 2020 or 2021 or
   whenever.

Timeline sequentiality
======================

All of the above ideas will be considered on a less hypothetical level after we
get our already-committed FCM40 product built.  FCM40 will be a modem module in
the same 56.5x36 mm form factor as Huawei GTM900 (with a mostly-compatible FPC
interface with only a few changes), featuring the same OM-based triband modem
core as FCDEV3B V2.  The reason for this sequencing is that our current FCDEV3B
suffers from a couple of issues which FCM40 is expected to solve:

1) FCDEV3B has a very tight PCB layout: not only do we have the tightly laid out
   core from GTA02, but also the whole board is quite small for the implemented
   peripheral complexity, imposing further constraints from all sides.  This
   tight and complex layout makes a poor choice of starting point for bold
   experiments like RFFE changes.

2) FCDEV3B is locked into Altium.  Layout data migration from Altium to FOSS
   appears to be much less feasible than migration from PADS to FOSS, thus
   freeing our PCB layout from the clutches of proprietary software will most
   likely require giving up (or rather setting aside) all of FCDEV3B new layout
   and going back to the GTA02 starting point, which is in PADS format rather
   than Altium.  Redoing all of FCDEV3B anew does not sound appealing at all,
   but the much simpler FCM40 board offers a perfect opportunity for a fresh
   start.

FCM40 will have exactly the same OM-based triband RFFE as our current FCDEV3B,
but it will be a much simpler board, and if we can get it done in FOSS instead
of continuing the Altium track, then we would have a very solid reference and a
good starting point for potential RFFE change experiments.

Firmware compatibility
======================

Our current FreeCalypso firmwares drive TSPACT RFFE control signals as follows
on FC hw family targets (CONFIG_TARGET_FCFAM):

TSPACT1 = Rx PCS band
TSPACT2 = Tx high bands
TSPACT4 = Tx low bands
TSPACT5 = Rx GSM850 band

The driving of TSPACT1, TSPACT2 and TSPACT4 matches the way these signals have
been assigned by Openmoko and thus the way they function on our current OM-based
triband RFFE, whereas TSPACT5 is a new signal which is not wired anywhere on
our current FCDEV3B.  This signal driving arrangement is expected to be
compatible with all 3 RFFE hw possibilities under consideration:

* On our current OM-based triband RFFE it works as is.

* If we use Epcos M034 or a semi-clone thereof that has the two Tx switches and
  a GSM850 Rx switch but no PCS Rx switch, then we will need to connect TSPACT2,
  TSPACT4 and TSPACT5 per the table above, and leave TSPACT1 unconnected.

* If we get a new M034-like FEM custom-made with a full set of all 4 switches,
  then all 4 TSPACT signals will need to be connected per the table above.