GSM Network and Services: Voice Coding

1
GSM Network and Services 2G1723 Johan Montelius

GSM Network and Services
Voice coding
2
From voice to radio waves
voice/source coding
channel coding
block coding
convolutional coding
interleaving
encryption
modulation
diff encoding
symbol coding
carrier modulation
burst building
3
Why digital?
link quality
speech
quality
analog
bit rate
compressed
digital
4
Analog vs Digital
If radio conditions are very good, an analog

connection could do the job and in many cases a
better job.
Digital coding allows:
compression
error detection and error correction
half duplex implementation
adaption to radio conditions
Digital drawback: processing, delay, voice only

codecs.
5
GSM voice codec
Sampled at 8 kHz, 13 bits per sample. Compare

this to regular A-law coding that samples at 8kHz
using 8 bits per sample.
Divides the sampled voice into 20 ms blocks for

the codec to work on. Each block is coded in 260
bits resulting in 13 kb/s.
The GSM codec is based on modeling of human

speech organs. Tuned for males!
Voice activation to allow discontinuous

transmission.
6
Remember
The 260 bits from the voice codec are encoded

using 456 bits that are interleaved into eight bursts
(half of each burst).
A 26-multiframe last for 120 ms and holds 24 TCH

burst, 1 SACCH burst and one idle frame.
24/4 = 6 voice samples every 26-multiframe.
26-multiframe takes 120 ms that is equal to 6 voice

samples of 20 ms.
One frame takes 120/26 = 4.6 ms.

7
GSM discontinuous transmission
sample
voice coder
voice activation det.
VAD
voice
DTX coder
frame
SP
comfort noise synth
SID
Silence Descriptor
8
DTX coder
The coder will detect if the sample is voice or

silence. When detected the Voice activity
detection will set a VAD bit to 1.
The voice coder will output a voice frame of 260

bits or a SID frame of 35 bits.
Depending on the VAD bit the DTX will output

voice frames or SID frames.
The DTX will pass the voice frames or in band

encoded SID frames to the channel coder.
9
DTX encoder
The channel coder will, based on the SP flag,

control when data is actually sent over the air:
any frame that contains speech information
the first frame after a speech frame since it

holds a SID frame
frames to keep the radio measurement

active
10
Comfort noise
If there is no noise in the background when

you speak you will think that the connection
is lost. This is extremely annoying
Since phone conversations are almost

always half duplex it is wasteful to keep a
full duplex connection open.
Why? Wasting what?

11
DTX decoder
voice
voice coder
bad frame
BFI
sample
DTX
voice
comfort noise synth
SID
12
DTX decoder
Correct voice frames are passed directly to the

voice decoder.
Correct SID frames are passed to the comfort

noise synthesizer.
If the Bad Frame Indicator (BFI) is set then
an incorrect voice frame is replaced by the

previous
an incorrect SID frame is replaced by the last

valid SID frame or last valid speech frame (that
probably contains noise)
13
Waveform vs Vocoder
Waveform codecs encode audio with no

knowledge of the signal.
Pulse Code Modulation (PCM) is one example that is

used in fixed line telephony.
Vocoder, or source coders, try to represent the

audio generator, in this case the vocal tract, and
represent a sample as a the parameters that define
the generator.
Vocoders can encode speech at a very low bit rate.

Even if the result does not sound like the original audio
signal.
14
Voice codec
0 1 2 4 8 16 32 64
0
1
2
3
4
5
6
7
8
10
!odecs
waveform
vocoder
hybrid
bit rate kb/s
quality
15
Hybrid coder
In a hybrid coder a vocoder is used to

encode the voice signal.
The decoded sample of the result is then

compared to the original sample.
The difference between the original and the

decoded result is encoded using a
waveform coder.
The result is a high quality low bit rate

codec.
16
GSM codec
short-term
LPC
short-term
filter 2080 bits
104kb/s
LTP
analysis
long-term
filter
36 bits (1.8 kb/s)
RPE
encoding
36 bits (1.8 kb/s)
188 bits
9.4 kb/s
Linear Predictive Coder
Long Term Prediction
Regular Pulse Excitation
RPE-LTP LPC
17
GSM codec
The resulting 260 bits are divided into:
Class 1A: 50 most significant bits
Class 1B: 132 significant bits
Class 2: 78 less significant bits
Classing is done by subjective evaluation of the

perceived speech quality (Mean Opinion Score,
MOS).
18
Channel coding
IA 50 b IB 132 b II 78 b
25 25 3
25 66 3 66 25 4
378 b after convolutional coding II 78 b
19
Other codecs
Half rate (HR): 6.5 kb/s used to increase capacity

since it only needs a half rate TCH. Speech
quality is considerably less.
Enhanced Full Rate (EFR): 12.2 kb/s but further

protected by CRC in a resulting 13 kb/s codec.
Better speech quality.
Adaptive Multi Rate (AMR): the codec adapts to

the radio conditions; high BER will use a low rate
codec that is better protected .
20
AMR
k"#s 12$2 10$2 6$7 4$75
%lock 244 204 134 5
!lass 1& 81 65 55 3
'ate (1#2) (1#3) (1#4) (1#5)
'aw 508 642 576 535
*uncturin+ 60 14 128 87
'esult 448 448 448 448
,n "and 8 8 8 8
-ra.e 456 456 456 456
21
BSS side coder
TRAU
BSC
MS
GSM voice
signaling
MSC
Alaw
!S"#
The TRAU needs not only the 13kb/s voice stream but also
BFI information etc. 3 kb/s signaling and error detection is
added resulting in a 16 kb/s stream.
4 streams will share a 64 kb/s ISDN channel

GSM Network and Services: Voice Coding

Uploaded by

Copyright:

Available Formats

GSM Network and Services: Voice Coding

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

GSM Network and Services: Voice Coding

Uploaded by

Copyright:

Available Formats

1

GSM Network and Services 2G1723 Johan Montelius

If radio conditions are very good, an analog

Digital coding allows:

error detection and error correction

half duplex implementation

adaption to radio conditions

Digital drawback: processing, delay, voice only

Sampled at 8 kHz, 13 bits per sample. Compare

Divides the sampled voice into 20 ms blocks for

The GSM codec is based on modeling of human

Voice activation to allow discontinuous

The 260 bits from the voice codec are encoded

A 26-multiframe last for 120 ms and holds 24 TCH

24/4 = 6 voice samples every 26-multiframe.

26-multiframe takes 120 ms that is equal to 6 voice

One frame takes 120/26 = 4.6 ms.

The coder will detect if the sample is voice or

The voice coder will output a voice frame of 260

Depending on the VAD bit the DTX will output

The DTX will pass the voice frames or in band

The channel coder will, based on the SP flag,

any frame that contains speech information

the first frame after a speech frame since it

frames to keep the radio measurement

If there is no noise in the background when

Since phone conversations are almost

Why? Wasting what?

Correct voice frames are passed directly to the

Correct SID frames are passed to the comfort

If the Bad Frame Indicator (BFI) is set then

an incorrect voice frame is replaced by the

an incorrect SID frame is replaced by the last

Waveform codecs encode audio with no

Pulse Code Modulation (PCM) is one example that is

Vocoder, or source coders, try to represent the

Vocoders can encode speech at a very low bit rate.

In a hybrid coder a vocoder is used to

The decoded sample of the result is then

The difference between the original and the

The result is a high quality low bit rate

The resulting 260 bits are divided into:

Class 1A: 50 most significant bits

Class 1B: 132 significant bits

Class 2: 78 less significant bits

Classing is done by subjective evaluation of the

Half rate (HR): 6.5 kb/s used to increase capacity

Enhanced Full Rate (EFR): 12.2 kb/s but further

Adaptive Multi Rate (AMR): the codec adapts to

You might also like