Stefan Diepolder and Jan Kritzner
Chair for Communication and Distributed System, RWTH Aachen University, Templergraben 55, Aachen, Germany
Mobile TV, Switching Algorithms and Multimedia Delivery.
Today’s 3G networks enable the delivery of mobile television services to the user. However, the distribution
mechanism differs from conventional television broadcasting. For an effective mobile TV system both unicast
and multicast transmission are combined.
One weak point of digital television is channel switching. Both the underlying network and the data structure
impose partially unavoidable delays. In this paper we present a mobile television architecture which supports
fast stream switching and present different techniques to reduce at least the perceived switching duration. The
different switching scenarios between unicast and multicast are discussed in detail.
This paper presents work in progress, so both already implemented and future extensions are described.
Though the implementation has been done with a 3G background the algorithms can be applied to every
digital television scenario.
One important problem of digital television is chan-
nel switching. During the standardisation of today’s
digital television architectures the focus had been on
defining the channel characteristics similar to tradi-
tional analogue TV. However, this resulted in the
problem of slow switching between different chan-
nels. Especially the advent of conventional digi-
tal television via DVB-S and DVB-T (ETSI, 2004)
has confronted a large user base with this problem
(Knoche and McCarthy, 2005).
The next subsection will technically introduce the
problem of channel switching. Afterwards, the trade-
off between switching delay and video quality will be
discussed. In the next section first ways to overcome
the problem with advanced techniques are shown,
which are refined in section 3. First results are given
in the Conclusion.
Digital video compression technology is based on
reducing redundancy in the video stream. This redun-
dancy may either be spatial, i.e. within a frame, or
temporal, i.e. between different frames. This tempo-
ral correlation introduces dependencies between pic-
tures, i.e. some pictures can not be displayed without
others. I-frames are independent of other frames but
P-frames depend on previous frames. This backward
dependency chain ends with the previous I-frame. A
group of possibly dependent frames is called Group of
Frames (GOF). In case of a channel switch necessary
frames may not be available at the receiver, and de-
pending ones may be distorted during playback until
receiving the next I-frame of the new channel.
The next important factor is the time needed for
the access onto the data of the new channel. This
may either be a physical channel switch where a re-
ceiver has to be tuned to a new frequency, or some
logical switch where some transport connection has
to be opened.
Therefore, we see that there are two main causes
for the channel switch delay:
physical and logical transport channel switch,
video structure.
1.1 Mobile Television Switching
For mobile television several purely broadcast-based
standards like DVB-H (Faria et al., 2006) have
been defined. In this document an infrastructure
based on a 3GPP cellular network is proposed. We
combine a Packet Switched Streaming (PSS) server
(3GPP, 2005), which provides a unicast streaming ser-
vice, with a Multimedia Broadcast/ Multicast Service
Diepolder S. and Kritzner J. (2008).
In Proceedings of the International Conference on Wireless Information Networks and Systems, pages 213-218
DOI: 10.5220/0002021902130218
(MBMS) (3GPP, 2004) extension.
This way we are able to transmit frequently re-
quired channels via multicast and some less fre-
quently requested channels directly to the particular
receivers via unicast. Unicast mobile TV must have
a back-channel for signaling channel selection, and
the server can do the switch without the need of re-
establishing the connection to the client. For broad-
cast channels no interaction with the server is neces-
sary. Different multicast channels may or may not
share one bearer. The temporal relation between uni-
cast and multicast is unknown.
To speed-up the process of switching channels in
a mobile TV environment it is necessary to add some
extra functionality to mobile clients and to the packet
switching server architecture. An algorithm for the
switch from unicast to multicast may require access
of the PSS server to the multicast data streams. There-
fore, a unified server for multicast and unicast is nec-
essary (see figure 1).
Figure 1: Proposed End-to-End Mobile TV Client/Server
The client’s architecture has to be improved, also.
In addition to an existing conventional unicast PSS
client an advanced multicast receiver is necessary. It
consists of a multicast receiver, an audio/video de-
multiplexer, and a buffer. The buffer consists of sep-
arated queues for video and audio. The usage is eas-
ier given that the queues of the different channels are
logically separated. However, they may dynamically
share the same memory space because they are not
filled simultaneously. The concepts shall be ready
to be deployed at the application level because the
firmware of today’s mobiles is closed. However, in
the long run it is possible to integrate new functions
into the firmware.
The buffer needs an advanced interface to the ap-
plication enabling the direct access to different parts
of the stream. The normal functionality like fetching
the next frame from the buffer or maybe flushing is
not sufficient. Special access to different Presentation
TimeStamps (PTSs) is needed for improved switching
by commands like ”Get PTS of next Frame”, ”Skip to
PTS”, ”Get first PTS”, and ”Get last PTS”.
For our work we are assuming that the client is
capable of receiving more than one bearer at a time.
At the time of writing no such mobile user equipment
exists but it is planned to be available soon. Never-
theless, it can receive multicast and unicast simulta-
neously which is the basis of the enhanced channel
switching algorithms. In our scenario the MBMS ar-
chitecture may impose a delay of up to several sec-
onds onto the TV data. However, this is not important
for channel switching. There is no requirement for
temporal alignment between different channels.
As shown in figure 1 the proposed architecture is
combining PSS and MBMS into one. The dotted Con-
trol Channel from client application back to the PSS-
Server is necessary for some switching algorithms,
especially when switching from unicast to unicast and
from multicast to unicast.
1.2 Switching Delays and Problem
Several aspects affect the duration of a channel
switch. First there is the physical time needed to
switch a physical bearer. This delay occurs even when
the lower layers are technically able to receive more
channels simultaneously but the new channel the user
wants to switch to is unknown in advance. As an ex-
ample consider a switch between two multicast chan-
nels when the receiver is able to receive two multicast
bearers simultaneously. Given that both channels are
transmitted over one bearer the delay is zero. Given
that they are transmitted over different bearers, the
bearer carrying the new channel must be activated. If
the old bearer is received further on, the user can zap
back to the old channel with much less delay. When
unicast streaming is taken into account it is possible to
keep transport connections or sessions open to reduce
this kind of delay.
The first part of the video decoding delay is buffer-
ing time. Due to the fact that usually the same time of
the old channel has been buffered as will be buffered
for the new channel, the buffering delay can be con-
cealed by playing out the last channel. Yet, the user
usually wants a swift reaction onto his channel switch
command the old content is not what the user ex-
pects. Therefore, the buffering delay for mobile tele-
vision should be quite low. In case of multicast
streaming, the buffering delay only needs to accom-
modate the different transmission delays for frames
of different size. Unicast streaming protocols may
benefit from higher buffering delays given that lost or
corrupted data frames can be corrected by retransmis-
WINSYS 2008 - International Conference on Wireless Information Networks and Systems
sions. This results in a trade-off between the quick-
ness of frame switching and the overall quality. The
last delay which has been mentioned consists of the
time between the scheduled playout of the first re-
ceived video frame until the first undistorted frame
can be shown. Given that after a channel switch the
receiver has to begin the decoding at a random posi-
tion within a GOF, the expected duration of the dis-
tortion is half the duration of a GOF. Again, there
is a trade-off between video quality and the speed of
switching: Long GOFs result in optimal encoding ef-
ficiency but slow switches.
Now, we propose new algorithms to speed up chan-
nel switching which are more intelligent compared
to the conventional ”play out what you have when
you should” approach. Generally we are not trying
to speed up the time until data can be received from
the network but to reduce the perceived time until the
presentation of the new channel starts. We will take
any possible combination of switches between unicast
and multicast into account, and propose algorithms
tailored to the problem as much as possible. Further-
more, we will distinguish between audio and video
2.1 Unicast to Unicast
Given that a user switches from channel 1 to channel
2, the unicast streaming system begins the transmis-
sion with the temporal nearest I-frame which is avail-
able in the transmission buffer. Therefore the client
sees some kind of virtual channel 2 which has some
temporal displacement. Therefore depending on the
time of their switching, two users viewing the same
TV channel which stand next to each other may have
a temporal displacement of up to one GOF-length in
the content they view. Usually this is acceptable and
people know this effect from the different delays for
analogue and digital conventional television. How-
ever, users may experience it as bad service if other
users see important information like goals in a soccer
game some seconds earlier than others. A more im-
portant problem is the strongly increased bandwidth
requirement given that a user zaps through multiple
channels. Then many large I-frames are transmitted,
and the bandwidth requirement exceeds normal val-
ues. Ideas to mitigate this problem are discussed in
section 3.
2.2 Multicast to Multicast
In the unicast-unicast scenario it is possible to reduce
the switching delay on the server side. In contrast to
the unicast-unicast scenario, in a pure multicast envi-
ronment the different multicast channels are simply
transmitted, possibly even without knowledge how
many subscribers use a certain service or how they
change channels. Therefore, a purely client-based so-
lution is necessary. A typical switching scenario de-
picting both the reception and playout over the time
is given in figure 2. A simpler graphic depicting the
playout process only can be found in the upper half of
figure 4. It is compared to the conventional approach
starting the video with the first I-frame and skipping
undecodable frames from the first GOF. Here d
and d
Bu f
denote the GOF and buffering duration re-
The new algorithm works as follows: After the
switching command the receiver tunes to channel 2
). When the first I-frame of the new channel
is received (after an average time of 0.5 · d
) it is
presented immediately. Until then either the playout
of the previous channel may continue or the old chan-
nel freezes to indicate a pending channel switch. Af-
ter the presentation of the first I-frame it remains on
the screen until its normal presentation time elapses
(dashed area). Then consecutive P-frames are de-
coded normally.
The early presentation of that I-frame may result
in dropping the last P-frames of the previous channel,
and the associated audio-data will be dropped. Fur-
thermore, the first undecodable P-frames of the new
channel will be dropped - but the audio data with
the according timestamps will be presented. There-
fore, the new channel begins with the presentation
of a still image, then audio is added, and afterwards
movement. This is the desired working scenario and
referred to as scenario 1.
However, it is not always possible to present the
first I-frame early enough. The scenario as described
above only works with sufficient buffer delay, i.e. the
I-frame has to be received before the playout of the
previous channel ends, as depicted in figure 2. Given
that the buffering delay is too low (figure 3) the old
channel ends and the picture has to freeze until the
first I-frame of the new stream has been received. This
is referred to as scenario 2 which can be found in the
lower half of figure 4. For this scenario two cases are
possible: The I-frame is received sufficiently early,
or audio data of the new channel could be presented
before the I-frame is received. Whether to drop this
audio data or not depends on psychological factors.
Playing back audio without the video may irritate the
viewer. However, a switch-command taking too long
Figure 2: Switch without buffer underruns (optimal case).
Figure 3: Switch resulting in an empty buffer due to late
Figure 4: Multicast Switching - Simplified Playout Only.
to show any effect is perceived as bad, also.
Whether the buffering delay is sufficient for avoid-
ing a freeze or not depends on the time of the switch.
However, the probability of the desired and the sub-
optimal case depends on both the buffering delay and
the GOF length: If the buffering delay is larger than
the GOF-length plus the switching delay, the I-frame
will definitely be available (first case). Otherwise the
probability for this optimal case is:
+ d
After the switch, it will take on average half a
GOF until an I-frame is received, hence the old chan-
nel will be shown for
E [d
] =
+ d
Then the first I-frame can be shown. Beginning with
the presentation of this first I-frame there is a period
of silence until the presentation time for the first audio
packet has been reached:
E [d
] = d
+ d
after which the normal audio playout begins, and the
channel switch is completed.
Compared to switching without an algorithm there
is no freeze of the old video, and audio can be pre-
sented on average half a GOF earlier than before.
The second scenario is more likely for short
buffering delays compared to the GOF-length:
= 1
+ d
The expected duration of the old channel freeze
before the first I-frame can be shown is
E [d
] = d
+ d
2.3 Multicast to Unicast
The switch from multicast to unicast is done similar to
the switch from unicast to unicast. The transport con-
nection is kept open if possible, and the client requests
the new channel from the server. Then the server
starts the unicast transmission of the new channel with
an I-frame which can be presented directly, and au-
dio is added when the buffer is filled sufficiently. It
is possible that the reserved bandwidth of the uni-
cast bearer is not available immediately but only af-
ter some seconds. This problem can only be solved
by over-provisioning of the unicast channel. Then the
amount of buffered playout time is larger than the du-
ration of the buffering process.
2.4 Unicast to Multicast
In the first step the switch from unicast to multicast
is identical to the switch between two multicast chan-
nels. The scenario with the freeze (scenario 2) ben-
efits when one I-frame of the new channel is trans-
mitted before stopping the unicast bearer. Then the
playout of the old channel ends with the presentation
of a frame from the new channel.
WINSYS 2008 - International Conference on Wireless Information Networks and Systems
In the latter sections simple algorithms and a basic ar-
chitecture for mobile TV have been introduced. They
have already been implemented. However, this paper
presents work in progress. More advanced ideas and
extensions are discussed now. One general technique
to reduce the experienced switching delay is adaptive
playout; it could be used for any switching scenario.
In the following it is shown which further improve-
ments could be made to the presented algorithms.
3.1 Unicast to Unicast
For unicast-unicast switching two improvements
could be made: One on the server side, and another
one by replacing the client architecture. One prob-
lem of a switch is the peak in the required bandwidth.
This happens because a new channel always begins
with an I-frame. Especially when a user simply zaps
through different channels quickly, the reserved rate
of the bearer would not be sufficient. This could
be solved by encoding specialised switching GOFs
with an I-frame of reduced quality, and consecutive
P-frames with better quantisers gradually improving
the video quality towards the end of the GOF. When
a new channel is requested the server transmits such a
switching GOF at the beginning, and afterwards con-
tinues with the presentation of the normal TV chan-
nel. The lower quality of the beginning of the first
GOF reduces the necessary bandwidth of a switching
event, and mitigates the effects of channel hopping.
Another approach to achieve the same goal could
be the usage of switching P-frames for the channel
switch instead of I-frames. However, it is not guaran-
teed that there is enough correlation between the old
and the new channel to enable a substantial reduction
in the required bandwidth.
On the client side advanced control over the play-
out buffer would be beneficial. Especially the de-
tection of the transition between one channel and
the other is important for advanced playout control,
e.g. a quick presentation of the first I-frame of the
new channel as done by a multicast-multicast switch.
Up to now the switching duration is identical to the
buffering time.
3.2 Multicast to Multicast
Given that the client is able to receive more than one
multicast channel at a given time it could be benefi-
cial to receive the data of both channels simultane-
ously. For example, conventional TV sets often have
a special button to go back to the previous channel.
Given that both of these channels are received such
a channel switch could be performed much faster. A
more complicated scenario would use a unicast con-
nection to support channel switches. When the user
changes from one multicast channel to the other the
server pushes one I-frame of the new channel to the
client. However, this technique has problems given
that the reserved bandwidth of the unicast bearer is
not available fast enough.
3.3 Multicast to Unicast
For multicast to unicast switches a good initial solu-
tion had already been found. The addition of adaptive
playout to this scenario would result in further im-
3.4 Unicast to Multicast
When the user switches from a unicast to a multi-
cast channel the server appends an I-frame of the new
channel to the old data stream which is eventually
(scenario 2) played out to give a first impression of
the program. For a quicker reaction it would be good
to present this I-frame as soon as it is available at the
client and not after the buffer has been played out.
This is only possible if the advanced buffer control
with direct data access is available.
Even more interesting would be the transmission
of some kind of prefix data which would enable the
client to seamlessly switch to the new channel, e.g.
by making the first received P-frames of the multicast
transmission decodable. However, due to the possibly
high difference in the forward trip times of the unicast
and the multicast system some kind of synchronisa-
tion technique is necessary determining which data of
the new multicast channel will be missing.
One of the main problems of today’s mobile TV sys-
tems, slow channel switching, has been analysed, and
a mobile TV system is developed to achieve a faster
channel switch. The different alternatives for the
transmission of mobile TV have been compared, and
the causes for slow channel switching have been anal-
ysed. To accelerate the switching process, algorithms
have been proposed which are implemented in a mo-
bile TV system using enhanced buffer management.
Slow channel switching is caused by the structure
of digital video, and this cannot be solved because up
to now it is the only way to achieve good compression
ratios. The aim was to attenuate the effect of the video
structure, and to improve the perceived quality.
The proposed algorithms for unicast and multicast
switching gain an improvement regarding the switch
delay (multicast) or resource utilisation (unicast) re-
spectively. Moreover, the switching algorithms guar-
antee a switch without distortions in the presented
video by avoiding decoding useless P-frames. The
search for the first I-frame is done quickly when the
unicast switching algorithm is used, and it introduces
no relevant delay regarding the channel switch.
The proposed archicture and the algorithms have
been implemented on top of the Network-Integrated
Multimedia Middleware (NMM) (Lohse, 2005), and
will be evaluated. NMM is a framework for me-
dia processing building flow graphs through special
nodes. Due to the object oriented design the nodes
are extensible. To build our system we have added
some functionality to the framework and especially to
the Switching Node, which implements the different
switching algorithms, the Media Source Node which
feeds the video content into the graph, and the dis-
play node, that shows the result of the delivered video
at the end of the switching process.
The delay of multicast-multicast switches is domi-
nated by waiting for the first I-frame of the new chan-
nel. Therefore, a fast channel switch is possible if
the I-frame is received early. The proposed multicast
switching algorithm improves every case because I-
frames are presented as fast as possible. The gain due
to this immediate presentation roughly matches the
buffering delay, in initial simulations we measured an
average gain of 0.8 seconds, which is a noticeable im-
provement over the simple switching. Despite this im-
provement, a multicast switch takes at the average 2.8
seconds which is caused by the long GOF structure of
4 seconds for our simulations.
3GPP (2004). Multimedia Broadcast/Multicast Service;
Stage 1 Release 6, 3GPP TS 22.146 V6.6.0. 3GPP.
3GPP (2005). Transparent End-To-End Packet-Switched
Streaming Service (PSS): Protocols and Codecs (Re-
lease 6), 3GPP TS 26.234 V6.3.0. 3GPP.
ETSI (2004). Digital Video Broadcasting (DVB): Fram-
ing structure, channel coding and modulation for dig-
ital terrestrial television (DVB-T), EN 300 744 V1.5.1.
European Telecommunications Standards Institute.
Faria, G., Henriksson, J. A., Stare, E., and Talmola, P.
(2006). DVB-H: Digital Broadcast Services to Hand-
held Devices. Proceedings of the IEEE, 94(1):194–
Knoche, H. and McCarthy, J. D. (2005). Design require-
ments for mobile TV. In MobileHCI ’05: Proceed-
ings of the 7th international conference on Human
computer interaction with mobile devices & services,
pages 69–76, New York, NY, USA. ACM Press.
Lohse, M. (2005). Network-Integrated Multimedia Middle-
ware, Services, and Applications. PhD thesis, Depart-
ment of Computer Science, Saarland University, Ger-
WINSYS 2008 - International Conference on Wireless Information Networks and Systems