Upload
gamze
View
218
Download
1
Embed Size (px)
Citation preview
ProgressiveDownloadfor MultimediaBroadcastMulticast Service
Zeki YetginDokuz Eylul University
Gamze SeckinVidiator Technology
The advent of new transport plat-
forms has revolutionized the way
users can access multimedia appli-
cations.1 Enabling technologies for
these platforms include Third Generation Part-
nership Project (3GPP) Multimedia Broadcast
Multicast Service (MBMS), Third Generation
Partnership Project 2 (3GPP2) Broadcast and
Multicast Service, and Digital Video Broadcast
for Handhelds, among others.2-4 Recently,
3GPP (see http://www.3gpp2.org/) introduced
support for IP multicasting services in the
MBMS architecture. This change lets providers
offer service to thousands of users simultane-
ously in a point-to-multipoint manner.
There are two delivery methods defined in
MBMS: download and streaming. Download
mode is for delivering files where transmission
reliability is important, while streaming is for
real-time media delivery. A third type, progres-
sive download, has drawn a great deal of atten-
tion recently as an alternative to MBMS
streaming. With progressive download, content
is streamed after an initial startup delay, and
the download continues in the background.
Progressive download substantially reduces the
user’s waiting time, and combines the advantages
of conventional streaming and downloading.
We believe MBMS should support progressive
download to increase user satisfaction and to
optimize MBMS radio resources. Moreover, adding
progressive-download capabilities to MBMS would
not require changing the infrastructure because
progressive download uses the MBMS standard’s
existing download-delivery mode. Currently,
however, progressive download remains an open
question in 3GPP. One of the main reasons for
the delay is that there is no research demonstrat-
ing the advantage of implementing progressive
download in MBMS. This article demonstrates
the possible advantages of using progressive
download for small multimedia files.
When doing the tests outlined here, we used
constant bit rate encoding because variable bit
rate makes time prediction difficult and requires
more capability on the receiving end. We also
used enhanced AAC+ and H.264 advanced
video coding (AVC) at 128 kbps.5 In addition,
we implemented downloading-time optimiza-
tion and application-layer interleaving. With
an effective interleaving mechanism, we
removed the negative effects of a reduction
in transmission cost and needed less error-
correcting overhead. Altogether, this article
compares four MBMS download systems:
legacy, progressive, interleaved-legacy, and
interleaved-progressive.
MBMS download delivery
MBMS download delivery is based on the File
Delivery over Unidirectional Transport (Flute)6
protocol used to deliver files over unidirectional
systems from one sender to many receivers.
Flute is an IETF protocol based on Asynchronous
Layered Coding (ALC),7 which makes Flute scal-
able and therefore preferable for unidirectional
systems. ALC is an instantiation of Layered Cod-
ing Transport (LCT)8 for Flute. LCT manages the
transport of an object identified by the Trans-
port Object Identifier (TOI) within a session
identified by the Transport Session Identifier
(TSI). ALC inherits LCT with asynchronous and
forward error correction (FEC) coding. Finally,
Flute inherits the ALC and provides capabilities
carried in-band or via File Delivery Table (FDT)
instances to signal the properties of the file,
including FEC coding descriptions, and to map
them to the ALC protocol.
With Flute, the only built-in reliability mech-
anism is provided by the FEC mechanism,
Feature Article
The project outlined
in this article focuses
on using progressive-
download
technology to
improve media
delivery in wireless
Multimedia
Broadcast Multicast
Services.
1070-986X/09/$25.00 �c 2009 IEEE Published by the IEEE Computer Society76
which provides reliable delivery of media con-
tent by appending repair symbols to the origi-
nal data prior to transmission across the
communication network. If some symbols are
lost during transmission, FEC allows the re-
ceiver to use repair symbols to recover the orig-
inal source block without retransmission. A
source block is a fragment of the original
media, with a block of k source symbols consti-
tuting a source block. Decoding algorithms
allow the recovery of the k source symbols
from any set of the n received symbols. Detailed
information for FEC can be found elsewhere.2,9
The most popular FEC codes are Raptor and
Reed Solomon.10,11 The MBMS system uses
Raptor codes due to their high performance.
A detailed overview of the complete proto-
col stack for MBMS downloads is described else-
where.12 A file in the application layer, called
an object in ALC terminology, is partitioned
into source blocks (SBs), each of which is div-
ided into k source symbols of size T each.
Each SB is forwarded to the FEC/Flute layer
where FEC encoding is individually applied to
each SB. The result is an encoding block (EB)
of N encoding symbols. Each encoding symbol
is identified by a source block number (SBN)
and an encoding symbol identifier (ESI). A
group of G consecutive encoding symbols that
share the same SBN is appended to an ALC/
Flute header (LHF ¼16 bytes). The result is a
Flute packet with payload P ¼ G � T.
Related WorkRecently, 3GPP working groups have been studying pro-
gressive-download technology for MBMS. According to the
packet-switched streaming service (PSS) specification, PSS
clients already support progressive download of media files
with HTTP connection over TCP/IP.1 But progressive down-
load in MBMS is still open to debate, with few projects actu-
ally analyzing it using formal methods.2
In an earlier project, we studied MBMS download optimi-
zation to reduce FEC overhead for both Reed Solomon and
Raptor FEC.3 In other projects, researchers have studied FEC
mechanisms for MBMS on two layers, namely the physical
layer and the application layer, with tradeoffs in applying
one or the other or suitable combinations of the two.4,5
In terms of interleaving, researchers have tested Reed-
Solomon and Raptor codes for MBMSs.6,7 Generally, re-
search approaches interleaving mechanisms above the FEC
layer as random symbol transmissions. However, random
transmission disables progressive download, so our recom-
mendation is to use an interleaving strategy that also enables
progressive downloading.
As an alternative to streaming, one project developed an
intermediate delivery mode that bridges the gap between
streaming and downloading and is based on MBMS down-
load delivery.8 However, the project’s new network elements
and requirements make the approach difficult to standardize
and implement.
In our current study, rather than focus on transmission-cost
optimization, we focus on download-time optimization be-
cause it provides a decrease in waiting time for MBMS down-
loads. We apply application-layer interleaving to progressive
downloading to decrease the wait time even more. Essen-
tially, this study combines our earlier work to focus on the
advantages of interleaved-progressive downloads.
References1. 3GPP TS 26.234, Transparent End-to-End Packet-Switched
Streaming Service (PSS); Protocols and Codecs, v7.2.0, 3GPP,
Mar. 2007.
2. Z. Yetgin and G. Seckin, ‘‘Progressive Download for 3G
Wireless Multicasting,’’ Proc. Future Generation Communica-
tion and Networking (FGCN), vol. 1, IEEE CS Press, 2007,
pp. 289-295.
3. Z. Yetgin and G. Seckin, ‘‘Reliable Download Analyses for
Multimedia Broadcast Multicast Services,’’ World Congress
on Engineering and Computer Science, IAENG, 2007,
pp. 389-394.
4. M. Watson and T. Stockhammer, ‘‘The MBMS Mobile Multi-
media Standard and Application-Layer Forward Error Correc-
tion,’’ white paper, Wireless Design and Development;
http://www.wirelessdesignmag.com/ShowPR.aspx?PUBCODE=
055&ACCT=0031786&ISSUE=0610&RELTYPE=PR&Cat=
13&SubCat=1&PRODCODE=M0210&PRODLETT=
A&CALLFROM=PR&CommonCount=0.
5. M. Luby et al., ‘‘Mobile Data Broadcasting over MBMS Trade-
offs in Forward Error Correction,’’ Proc. 5th Int’l Conf. Mobile
and Ubiquitous Multimedia, vol. 193, ACM Press, 2006, p. 10.
6. Efficient FEC Code for Reliable MBMS User Services, 3GPP TSG
System Aspects tech. doc. SP-050164, Siemens, Mar. 2005;
http://www.3gpp.org/ftp/tsg_sa/TSG_SA/TSGS_27/Docs/
PDF/SP-050164.pdf.
7. M. Luby et al., ‘‘Raptor Codes for Reliable Download Delivery
in Wireless Broadcast Systems,’’ Proc. Consumer Communications
and Networking Conf., vol. 1, IEEE Press, 2006, pp. 192-197.
8. Content Preloading for MBMS User Services, 3GPP TSG Sys-
tem Aspects tech. doc. S4-060385, Siemens, 2006; http://
www.3gpp.org/ftp/tsg_sa/WG4_CODEC/TSGS4_40/Docs/
S4-060385.zip.
Ap
ril�Ju
ne
2009
77
The Flute header contains FEC payload iden-
tifier that is the ESI and SBN of the first symbol,
along with TSI, TOI, and other information.
User datagram protocol (UDP) over IP distributes
the Flute packets, and additional headers are
appended to the original packet in UDP, IP, Log-
ical Link Control, Subnetwork Data Conver-
gence Protocol. At the Radio Link layer,
Protocol Data Units(PDU) are obtained from Ser-
vice Data Unit (SDU) and forwarded to the phys-
ical layer, as usual, where the data is encoded,
interleaved, and transmitted in small bursts.
System model
The legacy-download system is based on the
Vidiator (www.vidiator.com) MBMS download
prototype, which uses Reed Solomon FEC coding.
To emulate MBMS link conditions,10,13,14 we
implemented a module for controlling transmis-
sion rate and packet loss, and played the media at
128 kbps consisting of 32 kbps enhanced AACþand 96 kbps H.264 AVC codec. The model
outlined in this section is valid for legacy-
download and progressive-download delivery
because progressive download is based on the
download mode of the MBMS.
Figure 1a shows the system model for non-
interleaved download deliveries. To support
progressive downloading, repair symbols are
sent just after source symbols. Each SBi is deliv-
ered to FEC layer where the result is EBi, which
includes N encoding symbols (ES). Each ES is
uniquely identified by its SBN and ES identifica-
tion (ESI). An encoding symbol group (ESG)
starting from ESI ¼ j for SBi is denoted as
ESGi,j and identified by the couple (SBN, ESI)
of the first encoding symbol, here (i, j). The
ESGs are packed into the Flute payload just
after the place reserved for Flute payload identi-
fication and transported until there are no
more encoding symbols to send.
Figure 1b illustrates the system model of
the interleaved download delivery. It shows the
sender side flow of the download delivery
with the SB interleaving of block-size b. That
is, b consecutive SBs constitute an interleaver-
block and are sent in parallel in the order of
ESIs. All encoding symbols in the interleaver-
block with ESI ¼ 1 are sent first. One require-
ment of our interleaving strategy is that each
Flute packet must include one encoding sym-
bol at a time. However, this process requires
Packetizer
Transport
IPheader
User-Datagram-Protocol header
Fluteheader
Flute payload ID = (i, j)Repeat until j ≥ K + N
(a) (b)
(i, j) ESi, j ESi, j + G
ESi, j
Encoding signal(ES) Gi, j
Encoding block (EB)i
SBi
Sourceblock (SB)1 SB2 SBi
File
Forward-error-correction encoding
SBz
K
K + N
j = j + G
ESi, j + G
Packetize ESk, s
Transport
IPheader
User-Datagram-Protocol header
Fluteheader
Flute payloadID = (k, s)
(k, s) ESk, s
EBi + bEBi + 1EBi
For k:i . . i + b
SB1 SBi SBi + bSBi + 1
File
Forward-error-correction encoding SBk
SBz
K
K + N
For s:1 . . K + N
For k:i . . i + b
Figure 1. System Models
for (a) noninterleaved
download delivery and
(b) interleaved
download delivery.
78
IEEE
Mu
ltiM
ed
ia
all the b EBs to be in memory before they are
sent. So the parameter b and SB size can be
used to adapt to different service conditions.
At minimum, the interleaver-block size should
be two, otherwise interleaving is not applicable.
However, increasing the interleaver-block size
consumes more memory and complicates the
cost of the download process. Memory-related
issues are beyond the scope of this study.
There are three parameters in turning the
problem into a formula: initial startup delay,
transmission cost, and gain. We define the ini-
tial startup delay as the minimum waiting time
required to start playing the media on the termi-
nal after the initialization of the MBMS down-
load service. The term ‘‘waiting time’’ implies
the initial startup delay in progressive down-
load, while it refers to the downloading time
in legacy downloads. The initial startup delay
is maintained on the sender side and sent to
the receiver. To predict the initial startup
delay, we consider two types of receivers: analyz-
ing receivers and actual receivers. Analyzing
receivers estimate the initial startup delays for
the target environment of the actual receivers.
We organize file transmission so that repair
packets are transmitted after each source block.
Figure 2 shows a worst-case analyzing receiver
where the sending rate is less than the media
play rate and both playing and downloading
end at the same time. By considering the
worst-case scenario, the analyzing receivers esti-
mate the worst-case waiting time for the actual
receivers. The timely manner property implies
that the sender transmits symbols at some con-
stant rate and the receiver can handle decoding
without causing any overflow in the buffer. We
can assume therefore that the receivers play the
media at some constant rate after waiting for a
certain time D. Each time a source block is
decoded on the receiver, i is incremented and
Ti is assigned to a time stamp, the time of writing
the source block to the file. At time Td, media
playback can be started by the receiver.
Congestion and buffering determine the ap-
proximation between time and downloaded
size in Figure 2. If the client suffers from buffer-
ing delays, the linearity approximation is good.
However, for the receiver doing a progressive
download, there is linearity between time and
media size as long as there is 100 percent reliabil-
ity and suitable initial startup delay.1 We define
partial receiving rate ri as the size of source block
decoded within a consecutive times Ti�1 and Ti.
ri ¼Fi � Fi�1
Ti � Ti�1¼ DFi
DTi
where F0 ¼ 0, T0 ¼ 0; Fi denotes downloaded
file size in Kbytes and ri denotes partial
receiving rate in kbps at time Ti. The total
number of source block is 1 . . . z, z.
We define R as expected average receiving rate
(EARR) computed at time Tz by the analyzing re-
ceiver that predicts the average receiving rate of
the actual receiver. The analyzing receiver should
use all ri sequences up to Tz to find a single rate R
that best estimates the actual average receiving
rate of the receiver. To predict the initial startup
delay D as well as the download duration for
the actual receiver, an EARR function is defined.
Choosing a good EARR function is not critical
when it comes to its effect on waiting time for
small file sizes. So our choice is as follows:
Ri ¼ri þ Ri�1
2(1)
where Ri is our choice for the EARR at Ti, R0 ¼r1; i: 1 . . . z.
To understand the reasoning behind our
choice for such an EARR function, consider an-
other EARR function that averages all ri sequen-
ces. Let r1 ¼ 50, r2 ¼ 60, and r3 ¼ 70 kbps where
z¼ 3. Equation 1 results in R3¼ 62.5 kbps. How-
ever, the averaging EARR results in R3 ¼ 60 kbps.
Our choice reflects the recent changes in the ri
sequences. That is, the choice is more reactive to
the network conditions such as network conges-
tion and buffering delays. From Figure 2, we use
the following approximation for the expected
download duration of the actual receiver:
Dþ F
Rmedia¼ Tz ¼
F
Rz(2)
Ti–1
Fi–1
Fi
File
siz
e
Fz
D = TdTi Tz
Time
Figure 2. Download
size as a function of
time.
Ap
ril�Ju
ne
2009
79
Using Equation 2, the EARR and initial startup
delay approximated at the end of the file
download is as follows:
D ¼ Td ¼ F�1
Rz� 1
Rmedia
� �
where Td or D denote initial startup delay and
Rmedia denotes media play rate in kbps. F ¼ Fz
indicates media file size in Kbytes, while R ¼ Rz
is the EARR computed using Equation 1.
We define G as the gain in waiting time for
progressive download compared to legacy down-
load. We compute G using R and T as follows:
Tz ¼F
Rz
and
G ¼ 100� 1� Td
Tz
� �
where Tz denotes the duration of the legacy
download of the media in seconds and G is
the gain in waiting time with progressive
download.
Experimental results
We applied our experimental analyses to
two optimizations of the legacy download:
downloading-time optimization and transmission-
cost optimization. We extended the gains we
obtained in legacy download by using progres-
sive download and extensively analyzed pro-
gressive download to explore differences in
waiting time. As described here, we were able
to improve the gains obtained in progressive
download by implementing out interleaving
mechanism.
We conducted our experiments on a set of
computers, with each computer running a
copy of some type of MBMS for different con-
figuration parameters outlined in Table 1.
Each MBMS consists of a single server and a
single client. Considering server and client
on the same machine allows full control over
Table 1. Parameters studied across layers.
Parameter Unit Layer Experiment set System
File size Kilobyte Application {100,512} All
Transmission cost optimized FEC overheads Percent Application To be determined in this study Legacy
Transmission cost optimized waiting time Second Application To be determined in this study Legacy
Downloading time optimized FEC overheads Percent Application To be determined in this study Legacy
Downloading time optimized waiting time {Second,percent}
Application To be determined in this study Legacy
Gain in waiting time from optimizations {Second,percent}
Application To be determined in this study Legacy
Initial startup delay in progressive download Second Application To be determined in this study Progressive
Gain in initial startup delay from progressive
download
{Second,
percent}
Application To be determined in this study Progressive
Gain in downloading time from interleaving {Second,
percent}
Application To be determined in this study Interleaved
Gain in FEC overhead from interleaving Percent Application To be determined in this study Interleaved
Initial startup delay in interleaved-progressive
download
Second Application To be determined in this study Interleaved-
progressive
Gain in initial startup delay from Interleaved-
progressive download
{Second,
percent}
Application To be determined in this study Interleaved-
progressive
Symbol length Byte FEC {Service data unit � 48} All
Source block size Symbol FEC {10,50,80,100,120,150,180,200,230} All
IP packet size Byte IP {session data unit, or SDU} All
Service data unit block size Byte Core network {400,600,800,1000,1400} All
Protocol data unit block size (radio link control,
or RLC, block size)
Byte RLC radio l. {640,1280} All
PDU (RLC link layer) losses Percent RLC radio l. {1,5,10} All
Transmission rates Kbps All layer {64,128,256} All
Interleaver block size SourceBlock
FEC {2,3,4} Interleaved-progressive
Media play rate Kbps Application {128} All
80
IEEE
Mu
ltiM
ed
ia
our emulations. The MBMSs are actually our
emulation programs that use Reed Solomon
FEC coding. To simulate MBMS link condi-
tions, we implemented a transmission rate
and packet loss control module.
Both client and server maintain a database
and create a record after each download. The
record contains evaluated target parameters
for the selected set of configuration parame-
ters shown in Table 1. We scheduled several
sets of configuration parameters to initiate
several downloads with different characteris-
tics. Each download experiment for one set
of configuration parameters goes through
100 download iterations. After performing
all the combinations of the configuration
parameters, we merged local databases in
each computer into a single database, and pro-
duced the experimental results using that
merged database.
Legacy download
Reed Solomon FEC performance is bad with
a small SB size. As shown in Figure 3 and Figure 4,
increasing SB size will decrease the FEC over-
head required for reliability, hence causing an
increase in FEC performance. However, the SB
size exceeded a threshold value, making the
download time increase despite reducing
the FEC overhead. This property provides
tradeoffs between downloading time and FEC
overhead. Together with SB size, effects of
other parameters, such as symbol length,
radio link control (RLC), block size, and ses-
sion data unit (SDU) size, should be consid-
ered. Downloading-time optimization aims
for faster downloading with reliability, which
also provides savings in initial startup time
for progressive-download delivery.
Increasing the SB sizes up to 80 symbols will
decrease the FEC overhead dramatically; there-
after the decrease slows. Additionally, small
downloads show more fluctuations in terms
of FEC overhead required for reliability. These
fluctuations can be seen in Figure 3a and 3b.
With transmission-cost optimizations, shown
in Figure 3a and 4a, FEC overhead is reduced
with the decreased SDU sizes and increased
number of symbols, increasing the download
time as shown in Table 2. With download-time
optimizations, shown in Figure 3b and 4b,
download time is reduced with the increase
in SDU sizes and decreased number of
symbols, causing the FEC overhead to in-
crease, as shown in Table 2.
Compared to both optimizations, transmission-
cost optimization is usually obtained at the
same or smaller SDU sizes and smaller symbol
length but higher SB size. While the SDU size
that optimizes FEC overhead is large but still
smaller than the RLC block size, for download-
time optimization the best SDU size is small
but still larger than the RLC block size. From
this observation, we can say in general that
SDU size should be equal to RLC block size.
Table 2 shows more detailed comparisons.
Generally speaking, as SB size increases, legacy-
download performance increases, but only up
to a point. Around that point there can be
tradeoffs between FEC overhead and download
time. In addition, download-time optimization
offers up to an 18 percent improvement and
Forw
ard-
erro
r-co
rrec
tion
over
head
(%
)
160150
130
110
90
6050
100
120100806040200
(a)
(b)
203040
7080
100
120
140
Source block size (number of symbols)
Forw
ard-
erro
r-co
rrec
tion
over
head
(%
)
180
150
130
110
90
6050
100
120100806040200
203040
7080
100
120
140
160170
Source block size (number of symbols)
RLC block 1,280 bytes, SDU size 1,000 bytes, 10% PDU loss
RLC block 1,280 bytes, SDU size 1,000 bytes, 5% PDU loss
RLC block 1,280 bytes, SDU size 1,000 bytes, 1% PDU loss
RLC block 640 bytes, SDU size 1,000 bytes, 1% PDU lossRLC block 640 bytes, SDU size 1,000 bytes, 10% PDU loss
RLC block 640 bytes, SDU size 1,000 bytes, 5% PDU loss
RLC block 1,280 bytes, SDU size 600 bytes, 10% PDU loss
RLC block 1,280 bytes, SDU size 1,000 bytes, 5% PDU loss
Radio link layer (RLC) block 1,280 bytes, service-data-unit(SDU) size 600 bytes, 1% protocol-data-unit (PDU) loss
RLC block 640 bytes, SDU size 600 bytes, 1% PDU lossRLC block 640 bytes, SDU size 400 bytes, 10% PDU loss
RLC block 640 bytes, SDU size 400 bytes, 5% PDU loss
Figure 3. Comparison
of (a) transmission-
cost optimization
and (b) downloading-
time optimization for
a 100 Kbyte file.
81
Ap
ril�Ju
ne
2009
saves roughly 8.3 seconds in waiting time over
FEC-optimized downloads.
Progressive download
Figure 5a and 5b show the initial startup
delays for 100 percent reliability for 100-Kbyte
and 512-Kbyte media files. Increasing SB size
decreases the initial startup delay for various
loss ratios and transmission rates. Table 3 com-
pares the experimental results for progressive
download to the results we obtained for our
legacy-download tests.
For Reed Solomon FEC-protected downloads,
we obtained higher gains from progressive
downloads with higher transmission rates. We
observed improvements for 256, 128, and 64
kbps transmission rates at around 100, 80, and
40 percent on average. In general, we found
that progressive download provides savings in
initial startup delay of up to 32 seconds (a 32
to 100 percent gain) compared to legacy down-
load. Finally, progressive download provides
savings in initial startup delay of up to 40 sec-
onds (a 45 to 100 percent gain) compared to
FEC-optimized legacy downloads.
Interleaved-progressive download
We found improvement in initial startup
delay and FEC overhead when we applied inter-
leaving to progressive download. Table 4 com-
bines all results for our proposed MBMS.
We found that the improvement derived from
interleaving can save FEC overhead of up to 29
percent and reduce initial startup delay up to
40 percent in progressive download. We also
found improvement in initial startup delay
from interleaved-progressive download obtained
for 256, 128, and 64 kbps transmission rates are
around 100, 85, and 45 percent on average,
when compared to legacy download. Also, we
found that interleaved-progressive download pro-
vides savings in initial startup delay of up to 42
seconds (a 42 to 100 percent gain) compared to
legacy download. Finally, interleaved-progressive
Table 2. Comparison of optimizations for different transmission rates.
Transmission rate 256 Kbps Transmission rate
Transmission-cost
optimization
Download-time
optimization
Transmission-cost
optimization
File size
(Kbytes)
PDU loss
(%)
FEC o.
(%)
Downtime
(sec.)
FEC o.
(%)
Downtime
(sec.)
Download
time gain
(%)
FEC o.
(%)
Downtime
(sec.)
100 1 10 3.7 12 3.4 8 10 7.4
100 5 23 4.4 23 3.7 17 23 8.8
100 10 38 4.8 43 4.0 17 38 9.7
512 1 7 18.5 7 18.5 0 7 37.0
512 5 18 21.4 19 20.9 2 18 42.9
512 10 34 23.7 37 23.5 1 34 47.9
Forw
ard-
erro
r-co
rrec
tion
over
head
(%
) 170
150
130
110
90
6050
100
250200150100500
203040
7080
100
120
140
160
170
150
130
110
90
6050
100
203040
7080
100
120
140
160
Source block size (number of symbols)(a)
(b)
Forw
ard-
erro
r-co
rrec
tion
over
head
(%
)
250200150100500
Source block size (number of symbols)
RLC block 1,280 bytes, SDU size 800 bytes, 10% PDU loss
RLC block 1,280 bytes, SDU size 800 bytes, 5% PDU loss
RLC block 1,280 bytes, SDU size 1,000 bytes, 1% PDU loss
RLC block 640 bytes, SDU size 1,000 bytes, 1% PDU lossRLC block 640 bytes, SDU size 800 bytes, 10% PDU loss
RLC block 640 bytes, SDU size 800 bytes, 5% PDU loss
RLC block 1,280 bytes, SDU size 800 bytes, 10% PDU loss
RLC block 1,280 bytes, SDU size 600 bytes, 5% PDU loss
Radio link layer (RLC) block 1,280 bytes, service-data-unit(SDU) size 1,000 bytes, 1% protocol-data-unit (PDU) loss
RLC block 640 bytes, SDU size 400 bytes, 1% PDU lossRLC block 640 bytes, SDU size 400 bytes, 10% PDU loss
RLC block 640 bytes, SDU size 600 bytes, 5% PDU loss
Figure 4. Comparison of (a) transmission-cost optimization and
(b) downloading-time optimization for a 512 Kbyte file.
82
128 Kbps Transmission rate 64 Kbps
Download-time
optimization
Transmission-cost
optimization
Download-time
optimization
FEC o.
(%)
Downtime
(sec.)
Download
time gain
(%)
FEC o.
(%)
Downtime
(sec.)
FEC o.
(%)
Downtime
(sec.)
Download
time gain
(%)
12 6.8 7 10 15.3 12 13.8 10
23 7.4 17 24 18.5 32 15.3 17
43 8.1 17 42 21.3 53 17.5 18
7 36.9 0 8 82.7 8 74.5 10
19 41.4 4 20 88.3 21 85.3 3
37 47.1 2 37 104.2 41 99.9 4
Figure 5. Comparison
of initial startup
delays for a
(a) 100-Kbyte file and
(b) 512-Kbyte file.
Initi
al s
tart
up d
elay
(se
cond
s)
36
32
28
24
20
16
12
8
4
0140120100806040200
Initi
al s
tart
up d
elay
(se
cond
s)
160
0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
150
250200150100Source block size (number of symbols)
Source block size (number of symbols)(a)
(b)
5000
256 Kbps transmission rate, 1,000 bytes SDU size, 1% PDU loss256 Kbps transmission rate, 800 bytes SDU size, 10% PDU loss128 Kbps transmission rate, 800 bytes SDU size, 5% PDU loss64 Kbps transmission rate, 1,000 bytes SDU size, 1% PDU loss64 Kbps transmission rate, 800 bytes SDU size, 10% PDU loss256 Kbps transmission rate, 800 bytes SDU size, 5% PDU loss128 Kbps transmission rate, 1,000 bytes SDU size, 1% PDU loss128 Kbps transmission rate, 800 bytes SDU size, 10% PDU loss64 Kbps transmission rate, 800 bytes SDU size, 5% PDU loss
256 Kbps transmission rate, 1,000 bytes, service-data-unit (SDU)size, 1% protocol-data-unit (PDU) loss256 Kbps transmission rate, 1,000 bytes SDU size, 10% PDU loss128 Kbps transmission rate, 1,000 bytes SDU size, 5% PDU loss64 Kbps transmission rate, 1,000 bytes SDU size, 1% PDU loss64 Kbps transmission rate, 1,000 bytes SDU size, 10% PDU loss256 Kbps transmission rate, 1,000 bytes SDU size, 5% PDU loss128 Kbps transmission rate, 1,000 bytes SDU size, 1% PDU loss128 Kbps transmission rate, 1,000 bytes SDU size, 10% PDU loss64 Kbps transmission rate, 1,000 bytes SDU size, 5% PDU loss
83
Ap
ril�Ju
ne
2009
download provides savings in initial startup of up
to 46 seconds (a 49 to 100 percent gain) com-
pared to FEC-optimized legacy download.
Conclusion
We have explored MBMS from the progres-
sive-download viewpoint by analyzing many
factors to find efficient service parameters.
Our results indicate that interleaving provides
significant savings in FEC transmission cost
and improvement in initial startup delays for
progressive downloads. We hope this research
will help build support for implementing pro-
gressive-download protocols in MBMS. We
also hope our work will be beneficial to service
providers attempting in identifying and resolv-
ing various performance issues. MM
Acknowledgments
This work is part of EEEAG 104E163, a proj-
ect sponsored by Vidiator Technology Inc., The
Scientific and Technological Research Council of
Turkey (TUBITAK), and Dokuz Eylul University.
References
1. Support of Progressive Download in MBMS, 3GPP
TSG System Aspects tech. doc. SP-050164, BenQ
Mobile, 2006, http://www.3gpp.org/ftp/tsg_sa/
WG4_CODEC/TSGS4_39/Docs/S4-060267.zip.
2. 3GPP TS 26.346, Multimedia Broadcast/Multicast Service
(MBMS); Protocols and Codecs, v.7.3.0, 3GPP, 2007.
3. ETSI TS 102 472 v1.2.1, Digital Video Broadcast-
ing; IP Datacast over DVB-H: Content Delivery
Protocols, ETSI, 2006; http://www.dvb.org.
4. ‘‘Mobile TV: The Groundbreaking Dimension,’’
v2.21, white paper, UMTSF/GSMA Joint Mobile
TV Work Group, 2006; http://www.umts-forum.
org/component/option.com_docman/task,doc_
download/gid,1629/Itemid,12/.
5. ITU-T Rec. H.264 (03/05), Advanced Video Coding
for Generic Audiovisual Services, ISO/IEC 14496,
Oct. 2005.
6. T. Paila et al, Flute: File Delivery over Unidirectional
Transport, RFC 3926, IETF, Oct. 2004.
7. M. Luby et al., Asynchronous Layered Coding (ALC)
Protocol Instantiation, RFC 3450, IETF, Dec. 2002.
Table 3. Comparison of waiting time from progressive download.
Transmission rate 256 kbps Transmission rate
File size
(Kbytes)
PDU loss
(%)
FEC o.
(%)
Waiting time in
progress. D.
(sec.)
Waiting time in
legacy
D. (sec.)
Progressive
download gain
(%)
FEC o.
(%)
Waiting time in
progress. D.
(sec.)
100 1 12 0 3.4 100 12 0.6
100 5 23 0 3.7 100 23 1.1
100 10 43 0 4.0 100 43 1.8
512 1 7 0 18.5 100 7 4.9
512 5 19 0 20.9 100 19 9.4
512 10 37 0 23.5 100 37 15.1
Table 4. Summary for interleaved-progressive download.
Transmission rate 256 kbps Transmission rate
File size
(Kbyte)
PDU loss
(%)
FEC
overhead
(%)
Wait time in
interleaved-
progressive
(sec.)
Wait time in
interleaved-
legacy (sec.)
Gain in wait
time from
interleaved-
progressive
over legacy
(%)
Gain in FEC
overhead
from
interleaved-
progressive
(%)
FEC
over-
head
(%)
Wait time in
interleaved-
progressive
(sec.)
100 1 12 0 3.4 100 0 12 0.6
100 5 22 0 3.7 100 4 22 1.1
100 10 36 0 4.0 100 16 36 1.8
512 1 5 0 17.6 100 29 5 3.3
512 5 16 0 19.1 100 16 16 6.4
512 10 32 0 21.0 100 14 32 10.1
84
IEEE
Mu
ltiM
ed
ia
128 kbps Transmission rate 64 kbps
Waiting time
in legacy
D. (sec.)
Progressive
download gain
(%) FEC o. (%)
Waiting time
in progress.
D. (sec.)
Waiting time
in legacy
D. (sec.)
Progressive
download gain
(%)
6.8 92 12 7.5 13.8 45
7.4 85 32 9.0 15.3 41
8.1 77 53 11.3 17.5 36
36.9 87 8 42.5 74.5 43
41.4 77 21 53.3 85.3 38
47.1 68 41 67.9 99.9 32
256 kbps Transmission rate 256 kbps
Wait time in
interleaved-
legacy (sec.)
Gain in waiting
time from in-
terleaved pro-
gressive over
legacy down-
load (%)
Gain in FEC
overhead
from inter-
leaving (%)
FEC over-
head
(%)
Wait time in
interleaved-
progressive
(sec.)
Wait time in
interleaved-
legacy (sec.)
Gain in waiting
time from
interleaved-
progressive
over legacy
download (%)
Gain in FEC
overhead
from inter-
leaving (%)
6.8 92 0 10 7.6 13.9 45 17
7.3 85 4 20 9.2 15.5 40 38
8.0 78 16 38 10.9 17.1 38 28
35.3 91 29 7 39.7 71.7 47 13
38.4 85 16 19 47.8 79.8 44 10
42.1 79 14 37 58.2 90.2 42 10
8. M. Luby et al., Layered Coding Transport (LCT)
Building Block, RFC 3451, IETF, Dec. 2002.
9. Permanent Document On: Simulation Guidelines for
the Evaluation of FEC Methods for MBMS Download
and Streaming Services, 3GPP TSG System Aspects
tech. doc. S4-040348, Digital Fountain, 2004;
http://www.3gpp.org/ftp/tsg_sa/WG4_CODEC/
TSGS4_31/Docs/S4-040348.zip.
10. A. Shokrollahi, ‘‘Raptor Codes,’’ IEEE/ACM Trans. Net-
working, vol. 14, no. SI, June 2006, pp. 2551-2567.
11. I.S. Reed and G. Solomon, ‘‘Polynomial Codes
Over Certain Finite Fields,’’ SIAM J. Applied Mathe-
matics, vol. 8, no. 2, 1960, pp. 300-304.
12. T. Gasiba et al., ‘‘System Design and Advanced
Receiver Techniques for MBMS Broadcast Ser-
vices,’’ Int’l Conf. Communications, vol. 12, IEEE
Press, 2006, pp. 5444-5450.
13. Mapping of SDUs to Radio Blocks for FEC Simula-
tions, 3GPP TSG System Aspects tech. doc.
S4-AHP120, Nortel Networks, 2004; http://www.
3gpp.org/FTP/tsg_sa/WG4_CODEC/Ad-hoc_PSM/
Docs/S4-AHP120.zip.
14. 3GPP TS 26.946, Multimedia Broadcast/Multicast
Service (MBMS); User Service Guidelines, v6.1.0,
Sept. 2006.
Zeki Yetgin is working toward a PhD at the Depart-
ment of Computer Engineering, Dokuz Eylul Univer-
sity, Turkey. His research interests include reliable
multimedia transmission over wireless (broadcast)
channels, progressive download in MBMS, parallel
and distributed computing, learning and reasoning
in fuzzy environments, and image/speech processing.
Yetgin has an MS from the Computer Engineering
Department, Eastern Mediterranean University. Con-
tact him at [email protected].
Gamze Seckin is a principal engineer in the Network
Evolution and Strategy Team at T-Mobile USA. Her re-
search interests include asynchronous transfer mode,
IP, and mobile and wireless networks. Seckin has a
PhD from the Department of Computer Science and
Engineering, Arizona State University. She is a former
Fulbright Research Scholar, and a member of the IEEE
Communications Society and IEEE Women in Engineer-
ing Society. Contact her at [email protected].
85
Ap
ril�Ju
ne
2009