view README @ 118:a4450ae8fd09

sip-manual-out UAC: use the new CSeq parsing function
author Mychaela Falconia <falcon@freecalypso.org>
date Thu, 29 Sep 2022 13:03:27 -0800
parents 97317ede320a
children b7cd66acb123
line wrap: on
line source

This Hg repository contains a work-in-progress named Themyscira Wireless system
software.  Themyscira Wireless (ThemWi) is an experimental GSM network operated
by Mother Mychaela of FreeCalypso at a semi-urban/semi-rural location in
Southern California, USA; this GSM network is operated with Osmocom CNI software
components, all running on a single Slackware Linux server.  ThemWi system sw
is a suite of daemon processes and command line tools that run on the same
machine as all those Osmocom sw components and provide some additional
functionality that is not provided "out of the box" by Osmocom, most important
of which is outside connectivity to USA PSTN.

We are currently experimenting with using bulkvs.com as our USA PSTN
connectivity provider.  Like most low-cost PSTN connectivity providers, they
provide the interface to PSTN in the form of a SIP trunk - while I would
absolutely love to get a traditional TDM trunk instead, with SS7 signaling,
such a toy would be far beyond my budget, hence I have to settle for SIP.
Our current status is:

* We have already obtained a block of USA phone numbers (NANP, chosen numbers
  from an exchange area local to us) from BulkVS.

* These BulkVS-sourced real NANP numbers have been entered as MSISDNs into
  OsmoHLR records for our test SIMs operating on ThemWi GSM.

* We can successfully dial calls from one ThemWi GSM phone to another, with our
  themwi-mncc switch understanding all dialing formats that are considered
  standard for cellular phone networks in USA (full international, or 11 digits
  starting with '1' but no '+', or 10 digits only), as well as our own non-
  standard shorthand dialing method with only 4 digits.

* Whenever someone dials one of our NANP numbers from the outside world, BulkVS
  servers send UDP SIP INVITE packets to our server.  Our inbound call gateway
  process is themwi-sip-in; this daemon process listens on UDP port 5060,
  accepts SIP calls from BulkVS (ultimately coming from global worldwide PSTN)
  and turns them into GSM MT calls in MNCC format, going through themwi-mncc
  and ultimately to OsmoMSC.

The following functionality remains to be implemented:

* As a counterpart to themwi-sip-in, there will be another process named
  themwi-sip-out that will serve as a gateway for outbound calls, going from
  GSM MO MNCC to outside PSTN via SIP.  The outbound SIP call functional part
  is already implemented in test prototype form in sip-manual-out.

* themwi-mgw will be our transcoding RTP bridge, speaking GSM codecs (FR and
  EFR are currently of most interest) on the side toward Osmocom components and
  G.711 (PCMU or PCMA) on the PSTN side.  Right now themwi-mgw is a working
  skeleton that allocates endpoints with RTP & RTCP UDP port pairs, but doesn't
  pass any traffic yet.

Differences from osmo-sip-connector
-----------------------------------

In the Osmocom community, the "standard" (or generally accepted) way to connect
a GSM voice network to the outside world is via osmo-sip-connector, an Osmocom
process that connects to OsmoMSC's MNCC socket on one end and talks SIP on the
other end.  Our combination of themwi-mncc, themwi-sip-in and themwi-sip-out
effectively takes the place of osmo-sip-connector.  Here are the principal ways
in which our solution differs from osmo-sip-connector:

* o-s-c is designed to connect to a local instance of a SIP PBX such as Asterisk
  or FreeSWITCH, as opposed to interfacing directly to an outside SIP trunk
  from/to a PSTN-via-SIP connectivity provider.  themwi-system-sw is different
  in this regard: we do NOT use Asterisk or FreeSWITCH or any other similar
  software of "spaceship" complexity, instead our themwi-sip-in and
  themwi-sip-out processes interface directly to our PSTN-via-SIP connectivity
  provider.

* o-s-c has no internal call switching function for calls from one local GSM
  phone to another, instead such switching is punted to the required Asterisk
  or FreeSWITCH etc.  With o-s-c, the calling phone's MO call is converted to
  SIP, then Asterisk or other PBX hairpins it back to o-s-c, and then o-s-c
  handles the destination call leg as a separate conversion from SIP back to
  GSM MNCC.  In our solution such local calls are switched internally inside
  themwi-mncc, staying native within GSM MNCC land and never turning into SIP.

* o-s-c is based on Sofia-SIP, which in turn uses glib - a very unpleasant
  dependency in this Mother's opinion.  In contrast, our implementation of SIP
  is 100% from scratch, written in plain C in the traditional Falconian coding
  style.

The need for RTP voice transcoding
----------------------------------

In the context of GSM voice codecs, the term "transcoding" is used in two
significantly different meanings:

* Transcoding from one lossy GSM codec to another effectively constitutes two
  lossy speech codecs running in tandem, and is a highly undesirable condition.

* Running a single GSM codec (not two in tandem), decoding from GSM to G.711 in
  one direction and encoding from G.711 in the other direction, is a standard
  required function for traditional voice gateways between GSM and PSTN.

In themwi-system-sw, we need to do transcoding in the second sense above.
BulkVS SIP call interface to PSTN does not support any of GSM codecs, they only
support G.711 and G.729, and the same situation is expected to hold with other
PSTN-via-SIP connectivity providers.  We certainly don't want to use G.729 - we
don't want to run two lossy speech codecs in tandem, first GSM and then G.729 -
hence the only codecs we speak on the PSTN-via-SIP side of our gateway are PCMU
and PCMA.  Therefore, we need to perform RTP transcoding in our themwi-mgw,
very similar to a traditional GSM TRAU.

On the GSM side, the two codecs of most interest to us at the present time are
the original FR and EFR - hence they will be the first to be supported.  Note
the big difference from other Osmocom-using GSM community networks which
typically prefer or even strictly require AMR instead!  Our reasons for focusing
on FR and EFR instead of AMR are:

* Our OsmoBSC time slot configuration is full rate channels only, no half rate
  channels.  HR channels are needed only for greater capacity of simultaneous
  calls, but with the total number of people *on the planet* who actively want
  GSM/2G as opposed to LTE or 5G being no more than maybe 10, the thought of
  exceeding the limit of 6 simultaneous call legs per cell (meaning 6 separate
  GSM phone handsets talking *at the same time*) is preposterous.

* Without any HR channels in OsmoBSC config, AMR means AMR-FR specifically, not
  AMR-HR.  The highest level of AMR-FR is identical with EFR - thus if we
  support EFR, do we really need AMR?

* The whole point of Themyscira Wireless is to provide service to *vintage*
  mobile phones.  Our current collection of vintage phones includes models that
  only support FR1 and EFR (Ericsson I888, Nokia 5190) and Calypso C05 which
  supports FR1, EFR and HR1, but not AMR.

* EFR is desirable because it gives better voice quality than FR1, but we must
  support FR1 too, so we can serve the very oldest of phones which support only
  FR1 and nothing else.

Voice codec restriction (forcing GSM phones to use EFR instead of AMR, or
forcing all the way down to FR1) is done in OsmoBSC config, with codec-list
setting under 'msc 0'.