FreeCalypso > hg > gsm-codec-lib

comparison doc/EFR-rationale @ 122:b33f2168fdec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
doc/EFR-rationale article written
author Mychaela Falconia <falcon@freecalypso.org>
date Sat, 10 Dec 2022 08:51:01 +0000
parents
children 4af99bf8671a
comparison
equal deleted inserted replaced
121:b51295fcbbae 122:b33f2168fdec
1 Problem in need of solving
2 ==========================
3
4 At the time of the undertaking of Themyscira libgsmefr project (late 2022),
5 there did not exist any readily available library solution for GSM EFR codec.
6 The community of FOSS offers classic libgsm from 1990s for FR1 codec (it's an
7 implementation of GSM 06.10, on top of which we had to implement our own Rx DTX
8 handler) and opencore-amrnb for AMR (based on Android OpenCORE framework) - but
9 nothing for EFR. This situation creates a problem for anyone seeking to deploy
10 their own GSM network with a voice interface to PSTN or other networks: such
11 voice interface generally requires implementing a transcoder, and doing the
12 latter in turn requires a library that implements the codec to be supported.
13 In the present situation, anyone who wishes to implement a speech transcoder
14 for GSM networks can easily support FR1 and AMR codecs, but not EFR.
15
16 EFR is more than just 12k2 mode of AMR!
17 =======================================
18
19 There is a common misconception in the GSM hacker community that EFR is nothing
20 but the highest 12k2 mode of AMR, and that any library that implements AMR,
21 such as opencore-amrnb, is thus sufficient to support EFR as well. However,
22 the reality is more complex:
23
24 * If an AMR encoder operates with DTX disabled, such that the output contains
25 only speech frames and no SID, and the mode is forced to 12k2, then indeed a
26 simple reshuffling of bits will produce speech frames that can be fed to an
27 EFR decoder on the other end. Note that the two encoders (EFR and AMR 12k2)
28 will produce *different* encoded speech parameters from the same input, and
29 the decoded speech output on the other end will also be different, but the
30 two versions are expected to be equally good for human ears.
31
32 * In the other direction, if an EFR input stream contains only good speech
33 frames (no SID and no lost, FACCH-stolen or DTX-suppressed frames), one can
34 likewise do a simple bit reordering and feed these frames to an AMR decoder.
35 The output of this AMR decoder will once again be different from a proper
36 (bit-exact) EFR decoder for the same speech parameter inputs. but as long as
37 the EFR input stream is all good speech frames, the output will be good enough
38 for human ears.
39
40 * The real problem occurs when the EFR input stream contains SID frames and BFI
41 frame gaps, as will always happen in reality if this stream is an uplink from
42 a GSM call. AMR SID mechanism is different from that of EFR, and an AMR
43 decoder will NOT recognize EFR SID frames. A quick experiment confirms that
44 when a real GSM EFR uplink RTP capture is converted to AMR by non-SID-aware
45 bit reshuffling and then fed to amrnb-dec from opencore-amrnb, unpleasant
46 sounds appear in the output whenever GSM uplink goes into SID.
47
48 EFR reference code from ETSI
49 ============================
50
51 A published-source bit-exact implementation of GSM EFR encoder and decoder,
52 complete with all beyond-speech functions of DTX, VAD, comfort noise generation,
53 error concealment etc does exist in the form of reference code from ETSI.
54 However, this code has never been turned into a usable codec library by anyone
55 prior to us (at least not by anyone who freely published their work), and doing
56 such librification (producing an EFR analogue to what Android OpenCORE people
57 did with AMR) is no easy feat! The original EFR code from ETSI exhibits two
58 problems which need to be remedied in the librification project:
59
60 1) The original code maintains all codec state in global variables (lots of
61 them) that are scattered throughout. 3GPP reference code for AMR (naturally
62 later than EFR in chronological order) is better in this regard (in the AMR
63 version they gathered their global vars into structs and pass pointers to
64 these structs, although still many separately-malloc'ed structs instead of
65 single unified encoder state and decoder state), but we need the EFR version
66 for correct handling of all beyond-speech aspects, and this version is all
67 global vars.
68
69 2) These reference codes from ETSI/3GPP (both EFR and AMR versions, it seems)
70 were intended to serve as simulations, not as production code, and the code
71 is very inefficient.
72
73 Themyscira libgsmefr
74 ====================
75
76 Libgsmefr presented in this code repository is our current solution for EFR.
77 It is a library styled after classic libgsm for FR1, but its guts consist of a
78 librified derivative of ETSI EFR code. The problem of global vars has been
79 solved in this library version - they've been gathered into one unified struct
80 for encoder state and another unified struct for decoder state - but the problem
81 of poor performance (significantly worse than opencore-amrnb) still remains for
82 now.