FreeCalypso Minnie GSM MS development board
Design specification

0. Purpose

FC Minnie board is being produced for one primary purpose: to provide an
alternative to the horrible (in the opinion of the Mother of FreeCalypso)
GTM900-based application board design being developed by Sysmocom in the
project identified as OS#4030.  A situation in which the Horribilis board is
readily and cheaply available to the worldwide GSM hacker/enthusiast community
yet no better alternatives are available would be an unacceptable travesty,
hence I (Mother Mychaela) feel a moral imperative to develop, produce and make
available a better board, even though this venture will almost certainly come
at a loss in economic terms.

1. Overview of proposed new board

FC Minnie will be a single PCBA featuring the following key components:

* FC Tango module containing the Calypso GSM MS core;
* RF plumbing, terminating in a standard SMA female connector hanging off
  the edge of the board;
* A standard SIM 2FF socket;
* A USB subsystem built around FT2232H, providing two UART channels behind
  a single USB device.

The principal way of operating this board will be via USB-carried UARTs: the
operator will need to connect USB between her computer and FC Minnie board, and
two ttyUSB devices will appear, corresponding to Calypso Modem and IrDA UARTs.

FC Minnie is intended to be a GSM MS board of modem rather than handset type:
there are no hw provisions for connecting phone UI peripherals such as an LCD
or a keypad, and the board is NOT designed to operate untethered, without a
host computer connected via USB.  FreeCalypso family of projects includes a
plan for a different GSM MS board, named FC Venus, that will be designed to run
untethered with a 176x220 pixel LCD and a keypad matching the legendary
historical D-Sample board from TI - but FC Venus will be an expensive board,
reserved for those who truly share the Mother's vision for FreeCalypso, and
cannot compete with the OS#4030 application board coming out of Sysmocom.

1.1. Powering arrangement

The board will consist of two separate power domains: GSM and USB.  The GSM
domain will require a separate power supply, and cannot be powered from USB.
The battery-emulating power supply for the GSM MS core will be the same as used
for FCDEV3B and Caramel2 boards, with the same FC-standard power connector.
FreeCalypso HQ already has a large batch of these power supplies that were made
for us on a semi-custom basis, and we can provide them at a cost of about $7
per unit - hence from the cost perspective, there is no justification for
breaking FreeCalypso tradition and changing the design to a different powering
arrangement,

OTOH, USB will be the sole source of power for the FT2232H subsystem - if there
is no USB host connected, this subsystem will be unpowered.

1.1.1. Power domain boundary crossing

With USB and GSM sides of the board separately powered, there are two partial
power-down (PPD) scenarios to be concerned with:

PPD1: The USB host is plugged in, but there is no battery-emulating power supply
connected, or the GSM MS power supply is connected, but the chipset is in its
switched-off state per Iota VRPC.

PPD2: The GSM MS power supply is connected and the chipset is switched on in the
soft power switching sense, but there is no USB host connected and the FT2232H
subsystem is unpowered.

PPD2 is an error case: because this board is not designed to operate untethered,
the combination of booting the Calypso chipset and its flashed firmware without
a connected USB host is invalid.  However, this combination can happen very
easily by mistake, hence the hardware needs to handle it without entering a
state in which it would be subjected to serious adverse conditions such as
excessive current flow.  OTOH, PPD1 is a scenario that is expected to occur all
the time in standard FreeCalypso workflows: the USB host is connected but the
chipset has not been commanded to switch on yet, or the GSM MS has executed a
soft poweroff but the USB host hasn't been unplugged yet.

The set of LVC buffers inserted between FT2232H and Calypso I/O pins is expected
to provide correct graceful handling of both PPD scenarios, as detailed in
section 2.1.

1.2. FT2232H USB subsystem

Standard FreeCalypso workflows require that both Calypso UARTs be connected to
the development host side by side.  In the olden days of TI's own developers
working in TI offices with D-Sample and earlier development boards, the GSM MS
board would bring out two DE9F RS-232 ports, engineers had desktop PCs outfitted
with special multiport serial cards, and there was a pair of RS-232 cables for
each GSM MS development board.  In the present day USB is a much more practical
alternative, and the need to have both Calypso UARTs presented to the developer
side by side translates into a requirement for a two-channel USB-serial chip
that goes from one USB device to two UART channels.  Additional requirements
for this two-channel USB-serial chip are:

* The two channels should be symmetric from the chip's PoV, so that the choice
  of which channel goes to which Calypso UART can be made by FreeCalypso
  convention.  The latter convention says that in the absence of other ttyUSB
  devices, ttyUSB0 shall be Calypso Modem UART (AT command channel) and ttyUSB1
  shall be Calypso IrDA UART (rvinterf channel).

* The USB-serial chip needs to support non-standard baud rates, including an
  ability to handle 203125/406250/812500 bps, and this support should be
  available on both channels, so that no additional constraints are imposed on
  possible developer workflows.

Out of the repertoire of USB-serial chips which I (Mother Mychaela) am familiar
with, only FT2232x (either FT2232C/D or FT2232H) fit the bill.  My original
preference was for FT2232D, but that chip has just been discontinued, thus
FT2232H will have to be used instead.

1.3. Tango module signal bring-out trade-offs

The complete set of signals coming out of FC Tango module via its 80-pin system
interface connector is very rich, with most Calypso and Iota chipset signals
included.  Caramel-type boards (iWOW DSK and FC Caramel2) make all of these
signals accessible for play, with most of them going to the expansion interface
on a 56-pin header - but these boards cannot compete for the lowest possible
cost market niche.  Therefore, a board that is intended to compete with the
Horribilis board of OS#4030 on the metric of cost needs to be reduced in scope,
giving up advanced capabilities of FC Tango and providing only basic, modem-type
GSM MS functionality.

Out of the rich signal set coming out of FC Tango, only two non-essential
interfaces are brought out on FC Minnie: the main analog audio channel (see
section 2.7) and Calypso MCSI for digital audio (see section 2.8).  Those who
are interested in additional Calypso+Iota chipset signals available on FC Tango
will need to use a Caramel2 board, for as long as those remain in surplus, to be
succeeded later by another board tentatively named EvaTango, following TI's
original naming convention for component evaluation boards.

2. Detailed design

2.1. Dual UART interface across power domains

2.1.1. Signals from FT2232H to Calypso

There are 4 signals that need to connect in this direction (names from host DTE
perspective): TxD, RTS, DTR and TxD2.  If these signals were to be connected
directly between the two chips, there would be undesirable effects:

* FT2232H I/O operates at 3.3V; this voltage level is acceptable for Calypso I/O
  pins (CAL000/A spec says VDDS + 0.5 V, accounting for the forward drop voltage
  of clamping diodes inside the chip), but is not ideal.

* The main issue is partial power-down scenario PPD1, defined in section 1.1.1.
  If the outputs of a powered USB-serial chip are connected directly to I/O pins
  of an unpowered Calypso, current would feed from these outputs through Calypso
  I/O clamping diodes into the Calypso+Iota chipset's V-IO rail.  Series
  resistors can limit this current to a safe value, but we now know through
  experience that the only way to produce correct indicator LED behaviour is to
  eliminate this wayward current completely.

The solution is to insert an LVC buffer IC (either 74LVC125A or 74LVC541A) into
these 4 signal paths; this buffer IC needs to be powered from the Calypso+Iota
chipset's V-IO rail, which is brought out from FC Tango module.  With this LVC
buffer and the Calypso chip's I/O ring powered by the same supply rail, there is
no power domain boundary at Calypso inputs, and instead this boundary is moved
to the inputs of the LVC buffer.  Unlike Calypso and other common ASICs, LVC
logic ICs have special input and output structures that are specifically
designed for PPD applications, with a guaranteed very low Ioff spec.  LVC logic
family is also specifically designed to tolerate input voltages higher than the
IC's own supply, up to 5V - thus our 3.3V is well within safe design margins.

Buffer-translated DTR signal will go to Calypso GPIO3 per conventions of TI,
iWOW and FreeCalypso.  A 2.2 kOhm series resistor can be inserted in this path,
solely to protect the hardware from software misconfiguration, in case someone
erroneously configures this Calypso GPIO as an output.

No pull resistors will be placed on the nets running from FT2232H outputs to
LVC buffer inputs.  In erroneous scenario PPD2 all LVC buffer inputs in this
direction and thus all Calypso UART inputs will sense logic low: LVC buffer
inputs appear at first glance to be floating, but they cannot reach anywhere
near the switching threshold, as any accumulated stray voltages will discharge
toward GND through the unpowered FT2232H chip's clamping diodes.

Having Calypso UART inputs sense low rather than high is not desirable, but
scenario PPD2 can only be an error case on this board, hence there is no
justification to expend more components and PCB real estate on a more
complicated circuit design (see FC Venus for example) that would produce logic
high levels at Calypso UART inputs in this scenario.

2.1.2. Signals from Calypso to FT2232H

There are 5 signals that need to connect in this direction (names from host DTE
perspective): RxD, CTS, DCD, RI and RxD2.  Just like in the case of signals
going the other way, connecting these signals directly between the two chips
would be problematic:

* Scenario PPD2: clamping diodes in the unpowered FT2232H chip will turn that
  chip's inputs (driven by Calypso outputs) into shorts to GND, thus Calypso
  outputs would be severely overloaded with high current flow.  This scenario
  is an error condition, but because it is very easily caused, we would rather
  not overstress the hw in this condition.

* Scenario PPD1: experience with Caramel2 boards shows that pull-up resistors
  on USB-UART inputs, including internal pull-ups inside FT2232x chips, can
  sometimes leak enough current into the clamping diodes of directly connected
  Calypso I/O pins to create visibly incorrect state on indicator LEDs.

The solution is to insert an LVC buffer IC (74LVC541A is needed here) into these
5 signal paths as well.  Which supply should this buffer IC be powered from:
Calypso V-IO or USB domain 3.3V?  Either option is expected to solve the LED
problem in scenario PPD1 (no path for current from the USB domain to flow into
the V-IO rail), but other considerations differ:

* If this LVC buffer is powered from Calypso V-IO, all FT2232H inputs will sense
  logic high (from the chip's internal pull-ups) in scenario PPD1.  However,
  the high current problem of scenario PPD2 is not only left unsolved, but made
  potentially worse, as LVC buffers have much higher drive strength than Calypso
  outputs.

* If this LVC buffer IC is powered from USB domain 3.3V rail, scenario PPD2 is
  solved: the interface between power domains occurs at a point where outputs
  from one domain hit Ioff-specified LVC buffer inputs in the other domain.
  However, in scenario PPD1 all FT2232H inputs will sense logic low rather than
  high: the LVC buffers will sense logic low inputs by the same seemingly-
  floating mechanism as in the previous section, and they will propagate this
  logic low to FT2232H inputs.

Logic low at LVCMOS-level (as opposed to RS-232 levels) UART inputs means a
break condition on the data line and asserted state for all control signals.
Having this state at host UART input when the target is unpowered and waiting
to be switched on is counter to usual conventions, but creating the opposite
(more conventional) state in this scenario without causing any other problems
with the interdomain interface would require significant extra circuit
complexity, translating into extra cost.  Thus breaking some customary
conventions about UARTs appears to be the appropriate compromise here.

2.2. Indicator LEDs

There will be 3 indicator LEDs on FC Minnie board: one green, one yellow and
one red, described in the following sections.

2.2.1. Green power-on status LED

FCDEV3B features a green LED that lights when the Calypso+Iota chipset is
switched on and goes out at other times.  The way in which this LED is
implemented on FCDEV3B is completely impervious to wayward current feeding into
the V-IO rail, hence the issue of this current feeding from USB power domain
never caused a problem on that board.  However, the method used on FCDEV3B
cannot be used on any board using a packaged modem module like iWOW-based Tango
or Huawei GTM900: the internal signal used for LED control on FCDEV3B is not
brought out by anyone.

The plan for FC Minnie is to control the green LED with V-IO.  Putting a LED
plus a series resistor directly between V-IO and GND (powering the LED from
V-IO) would be a bad idea: when the Calypso+Iota chipset enters superdeep sleep,
the current budget of Iota regulator VRIO reduces to just 1 mA, which is too
little for a LED.  The plan for FC Minnie is to control this LED with the same
"digital transistor" (BJT plus bias resistors) circuit as used for the PWL LED
on Caramel2 (see section 2.2.3), but with V-IO in the place of PWL signal.
With 10 kOhm base resistor the load on V-IO is limited to 280 uA, yet the LED
(powered from raw VBAT) gets enough current to be visibly lit.

In order for this plan to produce the desired result of this LED lighting when
the Calypso+Iota chipset is switched on and going out upon soft poweroff, there
must not be _any_ wayward current feeding into the V-IO rail from the USB power
domain by any path.  The Mother's hope is that the buffer design of section 2.1
will work as intended and give us a reliable switch-on state indicator LED.

2.2.2. Yellow CLK13M status LED

FC Caramel2 board introduced an innovation: a yellow LED that lights when CLK13M
(brought out on FC Tango) is running and goes out when this clock is stopped
(chipset off or in deep sleep), causing this LED to flash when standard
FreeCalypso firmware with all sleep modes enabled is in idle mode listening on
PCH.  Unfortunately on some Caramel2 boards this LED would also light
erratically in scenario PPD1: when the V-IO rail rises above 0V as a result of
wayward current feeding while other Calypso power supplies are off, sometimes
Calypso outputs behave erratically, and some may unexpectedly drive high,
causing LEDs to turn on.

The plan is to replicate the CLK13M status LED circuit from Caramel2 verbatim
on FC Minnie; the hope is that the new design of section 2.1 will eliminate the
problem of wayward V-IO feeding and the LED will become fully reliable.

2.2.3. Red PWL LED

The remaining LED from FC Caramel2, a red LED controlled by Calypso PWL output,
will also be copied verbatim on FC Minnie, and the same hope as in section 2.2.2
applies here too.

Calypso PWL allows the intensity of light to be smoothly controlled by software,
and standard FC firmware exports this control to the user in the form of AT@PWL
command.  AT@PWL=0 turns this LED off, AT@PWL=255 turns it on at full
brightness, and all intermediate levels produce intermediate brightness.

2.3. Target boot control

FC Tango module brings out PWON and RESET boot control signals, although no
RPWON.  FC Minnie will provide both manual (tactile switches) and programmatic
boot control options.  There will be two manually operated buttons on the board,
PWON and RESET, and there will also be support for programmatic boot control
like on DUART28C.

DUART28C-style host-based boot control mechanism repurposes otherwise unused
RTS and DTR outputs of FT2232x Channel B, the UART channel that goes to the
data leads only Calypso debug UART.  The circuit consists of a pair of OD
drivers packaged in one 74LVC2G07 chip, with RTS driving PWON and DTR driving
RESET.

Correct operation with these circuit connections in place requires programming
a custom USB ID code into FT2232x EEPROM (see section 2.4) and patching the
Linux kernel ftdi_sio driver to apply a special quirk upon seeing this USB ID.
An additional complication is that Linux USB and tty subsystem maintainers are
refusing to mainline this quirk-adding patch, thus end users need to apply this
patch on their own systems in an act of principled defiance against obstinent,
assinine and tyrannical maintainers.

Because of these complications, this DUART28C-style boot control circuit needs
to be made optional.  The simplest solution is a pair of jumpers: the net from
the OD driver on Channel B RTS to Tango PWON will pass through one user-
removable jumper (2.54 mm header pin pair with a removable shorting block), and
the net from the OD driver on Channel B DTR to Tango RESET will pass through
another such jumper.

2.4. FT2232H USB subsystem details

The USB connector will be of mini-B type like on all other classic development
boards targeting this hacker community; the circuit from this USB connector to
FT2232H chip will be as prescribed by FTDI.  An external LDO regulator bringing
the bus-powering voltage from 5V down to 3.3V will need to be implemented, as
always required with FT2232H, and all components are then powered from this 3.3V
supply.

The FT2232H subsystem will include a 93C46 EEPROM.  FT2232H can work without an
EEPROM, but I as the Mother of FreeCalypso insist on always including this
configuration EEPROM.  FC Minnie will need to have a custom USB ID programmed
in this EEPROM when the two remote boot control jumpers are installed, and with
both PID options the EEPROM will always be programmed with custom textual ID
strings, allowing the board to be recognized among other USB devices sharing
the same VID:PID.

2.5. RF plumbing

The RF interface on FC Tango module is a microcoaxial connector of a type
compatible with Hirose U.FL and Sunridge MCB2 series, but a GSM MS development
board needs to bring out its RF interface on an SMA female connector.  The plan
for FC Minnie is to use the same approach as was successfully implemented on
FC Caramel2: a microcoaxial cable assembly from Sunridge that goes from an
MCB2G plug to an SMA connector that has PCB mounting legs only for mechanical
securement and grounding, while the RF signal never passes through the main PCB.

2.6. SIM socket

The SIM 2FF socket on FC Minnie will be of the same hinged type as featured on
FCDEV3B and Caramel2 boards.  The placement and orientation of this socket on
the board will need to be chosen so that it will work conveniently with SIMtrace
FPC cables; FCDEV3B is good in this regard, but Caramel2 is not.

2.7. Analog audio

Only the main analog audio channel will be brought out, not auxiliary.  The main
audio channel will be brought out on a 2.5 mm TRRS jack in the pinout that was
originally established by iWOW with their DSK and has since been adopted by
FreeCalypso, same as on FC Caramel2.  FreeCalypso HQ has a large batch of
FC-HDS4 headsets that have been custom-made for us in this pinout, hence it
makes no sense to implement any other arrangement.

2.8. Digital audio

Calypso MCSI is a PCM interface to the DSP part of Calypso that can be
configured (using the standard facilities of DSP ROM code plus TI's official
patches) to carry digital voice during calls, in 13-bit linear PCM sample
format, 8000 samples per second.  This interface was first brought out and
experimented with on FCDEV3B, but it is also brought out on FC Tango module.
This interface will be brought out on FC Minnie board on a 5-pin header, in the
same pinout as on FCDEV3B.

The Mother of FreeCalypso has a plan to produce a gateware design for the common
Lattice iCEstick FPGA board that will turn this Icestick into an interface
adapter for Calypso PCM voice, ferrying digital PCM samples to and from a Linux
host by way of FT2232H UART channel on the Icestick itself.  In this plan the
PCM interface will be connected with jumper wires directly from the Icestick
board to the MCSI header on FCDEV3B, FC Minnie or FC Caramel2.

3. Production notes

3.1. Manufacturing and test process overview

FC Minnie boards will be assembled at Technotronix in Anaheim, California, USA
and then production-tested at FreeCalypso HQ.  The assembly performed at
Technotronix will include permanent mounting of a Tango module onto each Minnie
board, whereas the subsequent production test process at FreeCalypso HQ will
include the following steps:

* FT2232H EEPROM programming;
* Loading FreeCalypso firmware into Tango module flash;
* Preening FFS for FC Tango firmware and for FC Minnie pinmux config;
* Testing the GSM RF tract in all 4 bands with a CMU200;
* Testing the SIM socket by inserting an FCSIM1 card and verifying correct
  SIM communication;
* Testing the analog audio interface by inserting an FC-HDS4 headset and
  commanding the firmware to generate a beep;
* Other minor hardware tests as feasible.

3.2. Permanent coupling of Minnie board and Tango module

FC Tango modules are not meant to be casually swappable once mounted on an
application board such as FC Minnie:

* The module features 4 grounding legs which must be soldered; these grounding
  legs are essential, as they carry power supply return current during GSM Tx
  bursts.  The need for soldering and desoldering should already be seen as a
  deterrent to casual (unnecessary, just for the heck of it) module swapping,
  but given that many hackers have absolutely no difficulty with soldering,
  other factors should be considered too:

* The fine-pitch board-to-board connector with 80 pins is extremely delicate,
  and can be easily damaged by unnecessary mating and unmating cycles;

* The microcoaxial connector for the RF interface is likewise very delicate and
  subject to the same considerations.

If a *legitimate* need to remove and replace the Tango module does arise (for
example, if some component inside the module goes bad and you need to either
repair or replace the Tango module), it can certainly be done - but it should
NOT be done casually, just for the heck of it.