Multipurpose Audio
Transmission System
on the Internet
Takashi Kishida
Graduate School of Information Sciences,
Hiroshima City University, Japan
Overview
Introduction
Purpose
Audio communication scenes
Implementation
Evaluations
Conclusion
Reliability audio transmission
An ability to recover packet losses on networks
Short delay
End-to-end delays include processing delays of an
end node and transfer delays on a network
Ex.) the case of about 150 ms
delays
Trade-off between robustness and delays
Development of a multipurpose
audio transmission system to
adapt to various scenes
Distance Lecture
400ms
Audio conference
Conversation
100ms
Distance
Chorus
Broadcast mode
Broadcast
Distance Lecture
400msmode
High robustness
Audio conference
Conversation
mode Conversation
Original RAT
100ms
Distance
Chorus These two
Chorus
mode modes are
Shorter delays added.
0ms Low High
Robustness
2003.6.19 Human.Society@Internet Conference 9
Chorus mode
ー This mode is set up as to achieve shorter
delay
Correction
sender ) redundant packet each some
generating
packets
receiver
recovery from redundant packets in the
case ofAdvantage
packet losses
in a real-time application
We use Reed-Solomon block code
Reed-Solomon block code has advantage
in burst errors
Audio data
packet
・・・
12
packets
15
Tran packets
smit Redundant
packet
3 packets
Head data
er Reed-Solomon ( 15,12 ) block code
This code has an ability
2003.6.19
of recovery from less than 3 packets
Human.Society@Internet Conference 14
State of Implementation
All modes are completed
Almost all audio codec are
implemented in Broadcast mode
Confirmation of implementation
Ethernet
CPU PentiumⅢ CPU PentiumⅢ
600MHz
100Mbps 1GHz
2003.6.19 Human.Society@Internet Conference 17
Measurement of delays
Mode Delays Defined delay
Broadcast
143 [ms] any
(15,11)
Broadcast
138 [ms] any
(15,12)
Broadcast
138 [ms] any
(15,13)
Conversatio Less than 400
132 [ms]
n [ms]
Less than 100
Chorus 72 [ms] [ms]
These satisfy all conditions of the defined delay in each
mode
These values are almost same as processing delay.
Transfer delay in Human.Society@Internet
2003.6.19
practical networks is added.
Conference 18
FEC Performance measurement of
Broadcast mode
Loss
CPU PentiumⅡ
Packet loss generator
300MHz
generated
OS Vine Linux2.1
1,2,4,6,8,10
%
Ethernet
100Mbps
Measure after
decoding RS
codes
Host A Host Experimental
B
values
CPU PentiumⅢ
We compared
600MHz
the experimental values
CPU PentiumⅢ
1GHz
and
the
OS theoretical
Vine
2003.6.19
values
Linux2.5 Human.Society@Internet Conference
OS Vine Linux2.1 19
Result
5
The theoretical
Packet values
loss (15,13)Experiment
rate can aland
value thefrom
be decreased experimental values
11% to less than 1
Packet loss rate after applying FEC (%)
% by
are using (15,13)T
almost FEC.
the heoretical value
same. (15 ,
(15,12)Experiment al value
4
(15,12)T heoretical value 13)
(15,11)Experiment al value
(15,11)T heoretical value
3
2
(15 ,
12)
1
(15 ,
0
11)
0 2 4 6 8 10 12
2
(15 ,
12)
1
Requirement
Hiroshima-city Univ.
bandwidth
Audio :
-- Hiroshima Univ.
MRAT(160Kbps)
Jitter : 6ms
Movie :
Avg. packet loss :
Mpeg2ts(5Mbps)
0.120% Hiroshima Univ.
2003.6.19 Human.Society@Internet
RTT : Conference
8.5ms 22
Error recovery of packet
losses using Broadcast mode
7.0%
Before applying FEC
6.0% After applying FEC
Packet loss rate (%)
5.0%
4.0%
3.0%
2.0%
1.0%
0.0%
0 20 40 60 80 100
Time (sec)
The results
Packet lossesofare
error recovery
almost recoveredfor only by 100 seconds
using
as a typical
broadcast part during the seminar
mode
2003.6.19 Human.Society@Internet Conference 23
Distance
Chorus
Hakushima Elementary School
(Main melody)
Accompanime
nt
Accompanime Hiroshima City Univ
+Subntmelody
Main Experimental IP Accompanime
melody network nt
70 ~ 10Mbps, wide
75ms Sub area Ethernet Main+Sub
melod melody
y Accompanimen Requirement 512 kbps
Accompaniment bandwidth
t
+Main melody
Transfer delay 2.1 ms
Minami-Kanon Elementary Jitter 7 ms
(Sub melody)
2003.6.19 Human.Society@Internet Conference 24
Conclusion
Classification of audio communication
scenes
Development of a multipurpose audio
transmission system, MRAT, and its
evaluation
Some practical experiments such as a
distance chorus at multi-points and
distance seminars
We think practical
chorus is in
Metropolitan network
within an area of a
few hundreds
kilometers.It’s realizable as a regional
2003.6.19 network
Human.Society@Internet Conference 27
Bandwidth of MRAT
Encoding Bandwidth Bandwidth of
RS encode After RS encode
[kpbs]
[kbps]
Linear-16 512 ○ 640
μ-law 256 ○ 320
A-law 256 ○ 320
G726-40 160 ○ 200
G726-40 128 ○ 160
G726-40 96 ○ 120
G726-40 64 ○ 80
DVI 128 ○ 160
VDVI 128 × 160
GSM 52.8 ○ 66
2003.6.19 Human.Society@Internet Conference 28
The quality of sound
comparison of MRAT and RAT
accompanimen
t
Ideal tolerant delay
70ms
Main
melody
Sub
melody
2003.6.19 Human.Society@Internet Conference 30
Study of burst errors
We use Reed-Solomon (15,11) code
(15 packets in 1
block)
― During distance seminar
block numbers of recovery
blocks = 99.9584
%
All block numbers
It could be almost recovered at distance
seminar.
We think this is adequate value.
2003.6.19 Human.Society@Internet Conference 31
End-to-end delay bounds
Best medium
150m 400m
s s
http://www.mame.csi2.net/mrat/frame.ht