Adaptive Link Assigment Applied in Case of Video
Streaming in a Multilink Environment
Péter Kántor 1, János Bitó 2
Budapest Univ. of Techn. and Economics, Dept. of Broadb. Infocomm. and Electrom. Theory
Goldmann Gy. tér 3., H-1111, Hungary
1 firstname.lastname@example.org, email@example.com
Abstract— Nowadays the importance of multilink networks
where end-to-end connections are provided by applying multiple
communication links simultaneously is rapidly growing due to
the expanding demand for high-speed Internet access and real-
time video transmission. Since channel conditions may vary in
time and space, the selection of the used communication links
depends on traffic load, latency, coverage and desired data rate.
Combining multilink architecture and adaptive video
transmission can be therefore an effective solution for
broadcasting video streams. In this paper the efficiency of the
adaptive transmission scheme will be presented for a terrestrial
heterogeneous multilink network which consists of WiMAX, LTE
and HSDPA communication links.
Keywords-component; multilink networks, adaptive video
transmission, quality enhancement, wireless communication,
Nowadays the demand for ubiquitous access to broadband
services is rapidly growing due to improved mobility. In
multilink systems devices are capable of communicating
through different links simultaneously, therefore
interoperability among existing communication links can be
exploited. These systems can use the same or different type of
access technology and may be owned by the same or different
network operators. One can choose the used communication
links according to one’s preferences, location and conditions of
the available channels. As the used channels may easily vary
according to the bandwidth, latency and area of coverage ,
selection and scheduling logic have to be present both in user
equipments as well as on the part of the network operator.
This paper investigates interoperability among WiMAX,
LTE and HSDPA links. In our previous works we studied the
case when the parallel links are part of a BFWA network; these
results can be found in . As a further investigation, the
present paper studies technologies employing different access
technologies. Moreover, differences among the applied carrier
frequencies, bandwidths, base station (or nodeB) arrangements
and antenna gains are considered as well.
Interoperability is defined as the ability of two or more
systems or components to exchange information and to use the
information that has been exchanged . Distinctions among
the employed technologies can be exploited in order to achieve
better Quality of Service (QoS) and more reliable connection
which minimizes the probability of the outage conditions.
However, the possibility of cooperation and interoperability
among the access networks is not enough in itself, it is
necessary to have applications which are interoperable as well.
In this paper we focused on the MPEG-2 video
transmission. For the sake of simplicity we restricted ourselves
to the so-called frame-based adaptivity; however, the proposed
concept of adaptive video transmission applies to the case of
other video coding methods as well. In the MPEG-2 standard
raw frames are compressed into three types of frames: intra-
coded frames (I frames), predictive-coded frames (P frames),
and bidirectionally-predictive-coded frames (B frames) .
These frames are coupled into Group of Pictures (GOPs). As
the importance and the size of the different frames are also
different, it is advantageous to transmit the most important
parts of the GOPs (i.e. I frames) through the link that has the
best quality and reliability, and less important parts (B frames)
through the link with low quality. Therefore applying adaptive
video transmission according to the channel conditions and
interoperability among the different links is a possible method
of improving the reliability of the video transmission. This
method can be considered as a special variety of Multiple
Description Coding (MDC) techniques .
Due to the applied user mobility models, the Bit Error Rate
(BER) on the investigated terrestrial links varies in time, and in
some cases significant degradation appears. Since the quality
of a communication link between the base station (or a nodeB)
and the terminal can drop, we proposed an adaptive video
transmission method which mitigates this degradation and
improves the QoS of the video stream. The solution could be to
redirect some parts of the terminals’ data flow (typically I or P
frames in the case of MPEG-2 video transmission) to another
nodeB or base station that is situated in another network.
We applied computer simulations to investigate adaptive
MPEG-2 video transmission in a multilink environment that
consists of WiMAX, LTE and HSDPA links. In this paper the
efficiency of the so-called frame-based adaptivity is presented.
The proposed concept can be adopted for the case of spatial
scalability as well.
A.Lay-out of the Simulated Multilink Network
The lay-out of the simulated multilink network with a size
of 4 km x 4 km is shown in Fig. 1. We applied the Manhattan
mobility model, which uses a grid road topology in an urban
area . Users can move along the streets, LTE and HSDPA
nodeBs are located at every junction, while WiMAX base
This work was supported by the Celtic MARCH project.
stations are located only at the junctions marked with the dark-
gray colored circles (at J1, J3, J5, J7 and J9 in Fig. 1). According
to the lay-out the distance between the closest HSDPA and
LTE nodeBs is 1 km, and the distance between two WiMAX
base stations is
Figure 1. Lay-out of the simulated multilink network
We assumed that radio waves are totally shadowed by the
buildings because of their attenuation, therefore only line of
sight (LOS) links are considered. Moreover, users always
connect to the closest base station/nodeBs simultaneously;
therefore this simplification does not affect the validity of the
B. Parameters of the Links of the Multilink Network
Assumed parameters of the investigated links for different
access technologies are given in Table 1. The applied
transmitter power, carrier frequency and antenna gains are
chosen to be realistic values. Bit rates of the different types of
links were presumed to be equal to each other.
TABLE I. PARAMETERS OF THE SINGLE LINKS
The path attenuation was calculated between the actual
serving base station/nodeBs and the moving user, according to
the instantaneous value of the distance between them. Effects
of the urban environment and user mobility were taken into
account at the specification of the minimal required signal-to-
noise ratio (SNR). The Gaussian noise power was set to -81.5
dB for all the access technology types.
C. Types of User Mobility
In the current paper two types of user mobility was
investigated: the Manhattan mobility model with pedestrian
and vehicle (i.e. car) speeds. In that model users can move only
along the streets, probabilities of turning to the left, right, up
and down were ¼ at every junction, respectively. If a user
reaches the edge of the simulated area, they turn back to the
direction from which he came. The pedestrian speed was 3.6
km/h; speed of the car was 30 km/h.
III.THE FRAME-BASED ADAPTIVE VIDEO TRANSMISSION
In case of a multi-linked environment it is advantageous to
split the video stream into separate sequences with different
priority regarding the overall QoS of the video stream. Our
goal was to provide reliable connection for the most important
frames of the video sequences (e.g. I frames), so we reroute the
important parts of the stream according to the channel
conditions. The quality of the transmitted video stream can be
enhanced with the proposed method highlighted in Fig. 2.
Figure 2. The concept of frame-based adaptive video transmission in the
case of the investigated multilink environment
Since the intra-coded frames are used as reference frames
for the other frames, I frames have to be transmitted at the
highest priority. On one hand, transmission of predictive-coded
frames depends on the I frames and on the other hand, P frames
are reference frames for the B frames. Therefore P frames
should be handled as the second important frames and have to
be directed to the best of the remaining possible links. Since
there is no predicted frame from B frames, they can be
transmitted over the lowest trustable links. The receiver at the
terminal reschedules the frames structure and restores the video
stream as it can be observed in Fig. 2.
As it was mentioned above, our method can be considered
as a special MDC technique. However, in contrast to
techniques elaborated for ad hoc networks (see  and ), in
a multilink network the independent access networks have their
own error protection and scheduling procedures. Therefore a
multilink gateway (MLG) has to be present at the transport
logic in order to apply these techniques, so the instantaneous
channel conditions, delays and throughputs would be known at
the server side. Moreover, using extra forward error correction
(FEC) would increase the required data flow, but by applying
the proposed adaptive transmission this is not necessary.
In this paper we compared the non-adaptive and our
proposed adaptive method. In both cases we exploit the
bandwidth aggregation capability of the multilink system and
assume that there is three times as mach bandwidth available
compared to the single link solution. We assumed the
following GOP structure of I, B and P frames:
By the scheduling of the frames the differences in size
among the three types of frames have to be considered. Since P
transmission and constant bit rate
Figure 5. The schedule of segments in split GOPs assuming adaptive
transmission and constant bit rate
In the second case the frame-based adaptive video
transmission is investigated. Schedule of segments in a split
GOP can be seen in Fig. 5 for the case when the first link has
the best quality. Therefore the I frame and the first P frame of
the GOP are transmitted here. If the second link has the second
best quality, segments of the rest of the P frames are scheduled
in this split GOP. In that case the third link has the lowest
quality, it serves the less important B frames. Since the
transmission is adaptive, it can be guaranteed that the important
frames are transmitted through the highest quality links.
Due to the different parameters of the investigated access
networks, the three links of the system have different SNR
values that are measured at the TS in the downlink. The useful
signal is sent by the base station or nodeB that transmits the
actually examined segment. SNR values are calculated in 1800
locations along the user movement, the length of the simulated
time was 3000 seconds.
A frame is considered successfully transmitted if the actual
SNR (depending on the used link and the position of the user)
is larger than the value that belongs to the 10−3 BER. This
threshold depends on the type of the actually used access
network, the applied modulation, coding and the MPEG-2
profile. Since results of many link-level simulations are
available in the relevant literature, we set these threshold
values from ,  and . The assumed modulation types,
code rates, channel models, user movement speeds and
threshold values are given in Table 2.
TABLE II. PARAMETERS OF THE ASSUMED CHANNEL TYPES
LinkChannel Modulation Code rateThreshold
QPSK 1/2 11dB
QPSK 1/2 21 dB
QPSK 1/214 dB
QPSK 1/224 dB
QPSK 1/212 dB
If a frame lasts for more than one segment, each segment
has to be received by the terminal properly, therefore the SNR
value has to exceed a critical value during the whole frame.
A.Quality of the Simulated Channels
Pedestrian Case 1)
The random path of the examined pedestrian user was:
J3→J6→J5→J2, when Ji refers to the junction in Fig 1. This
means that the simulated user starts from the centre of the
simulated network (2 km, 2 km; J3) and heads to the west.
After 1 km it turns to the right (at the J6) and after another 1 km
it turns to the right again (at the J5). At the beginning of the
simulation the user is close to a WiMAX base station and LTE
and HSDPA nodeBs, so the channel quality is nearly perfect as
it can be observed at the 1st GOP time in Fig. 6, which shows
the downlink SNR values of the multilink network.
0 50 100150200 250300
Figure 6. Downlink SNR values on the links of the multlink network and the
applied tresholds in the case of pedestrian channel
Getting away from J3, the quality of each link is decreasing,
while after 0.5 km, at the transmission of the 50th GOP the
quality of the LTE and HSDPA networks begins to improve, as
the junction J6 gets closer to the terminal. At J6 (when the 100th
GOP is transmitted) the quality of the WiMAX network is not
sufficient, since at this time the threshold criterion is not
satisfied. When the user heads toward the J5, the quality of the
WiMAX channel gets better. At J5 (when the 200th GOP is
transmitted) the user is also near to a WiMAX base station and
LTE and HSDPA nodeBs, so the qualities of the channels are
perfect again. At the end of the path the quality of the WiMAX
channel is not sufficient again, as it can be observed in Fig. 6.
2) Case of Vehicle
The random path of the examined vehicle was:
In this case the path was longer compared to the pedestrian’s
case, because of the same time of the simulation but with
different user speed (30 km/h).
The SNR curves of the different links are shaped just as in
Fig. 6, however when the user is at the edge of the network (i.e.
around the [0, 2] km and [2, 0] km positions) the WiMAX base
station is located too far from the vehicle. Therefore the SNR
drops as it can be observed in Fig. 7 around the transmission of
the 120th and the 170th GOPs. Moreover, the threshold criteria
many times cannot be satisfied in this case, so the video
transmission will be highly unreliable. Here we emphasize, that
the value of the thermal noise was chosen to be rather
unfavourable, therefore we could demonstrate the efficiency of
the proposed frame-based adaptive video transmission method.
0 50 100 150200250300
Figure 7. Downlink SNR values on the links of the multlink network and the
applied tresholds in the case of vehicular channel
In case of non-adaptive link assignment split GOPs
containing the segments of the I frames are forwarded to the
LTE link. Split GOPs containing the segments of the two P
frames are transmitted through the WiMAX link, while the
third type of split GOPs are transmitted through the HSDPA
link. The schedule of segments in split GOPs for the case of
non-adaptive transmission can be seen in Fig. 4 assuming
Link1 corresponds to LTE, Link2 to WiMAX and Link3 to
HSDPA, respectively. In Fig. 8 the number of properly
transmitted frames can be seen in the case of the pedestrian
channel. As it was mentioned at Fig. 6, around the transmission
of the 100th and the 300th GOPs the condition of the WiMAX
channel is not sufficient, therefore only the I frame, one P
frame and two B frames are transmitted properly (see Fig. 12
as well). Neither the frames that are scheduled to the WiMAX
link (namely two P and two B frames) nor the B frames
predicted from these P frames can be decoded, in spite of the
fact that the quality of the HSDPA link is good.
Number of the transmitted
Figure 8. Number of different types of properly transmitted frames in the
case of pedestrian channel and non-adaptive transmission
In Fig. 9 the number of properly transmitted frames can be
seen in the case of the vehicular channel. As it was mentioned
at Fig. 7, links are almost inefficient in that case; therefore the
number of successfully transmitted frames drops frequently.
Number of the transmitted
Figure 9. Number of different types of properly transmitted frames in the
case of vehicular channel and non-adaptive transmission
C. Adaptive Transmission
For the case of the adaptive link assignment the schedule of
segments in the split GOPs can be seen in Fig. 5. Split GOPs
containing the segments of the I frames are forwarded to the
link with the best quality (according to the difference between
the threshold and the actual SNR), the second type of split
GOPs are forwarded to the second best link and the third
remaining link serves the split GOPs containing the B frames.
Therefore links used to transmit the three types of split GOPs
vary according to the channel qualities.
050 100150 200250300
Number of the transmitted
Figure 10. Number of different types of properly transmitted frames in the
case of pedestrian channel and adaptive transmission
In Fig. 10 the number of properly transmitted frames can be Download full-text
seen according to the adaptive transmission for the pedestrian’s
case. It can be observed, that the number of the transmitted P
and B frames raises compared to the non-adaptive method (Fig.
9) when the WiMAX link is in outage (around the 100th and
300th GOPs), since split GOPs containing two P frames are
always transmitted through the second best link, therefore P
frames can always be decoded properly (see Fig. 12 as well).
In Fig. 11 the number of properly transmitted frames can be
observed for the case of vehicular channel and adaptive
transmission. Comparing with Fig. 9 it can be stated, that
adaptive transmission and scheduling enhances the quality
when at least one of the used links is operable. When every
link is under the threshold, no gain can be accomplished.
050 100150200250 300
Number of the transmitted
Figure 11. Number of different types of properly transmitted frames in the
case of vehicular channel and adaptive transmission
D. Visual Demonstration
For the sake of visual palpability, effects of the frame-based
adaptivity are illustrated in Fig. 12. Here can be observed, how
this method can overcome the phenomenon of lost frames
which emerges in multilink networks. As it was mentioned
before, for the case of pedestrian channel at the 100th GOP the
WiMAX quality is not sufficient. Applying non-adaptive
transmission is not beneficial since two P frames and six B
frames are lost, therefore the video quality in the cases of fast
motions drops observably as it can be seen as the freeze on the
video in Fig. 12. However, when adaptive transmission is
applied, all P frames can be transmitted properly, therefore
only B frames are lost. In that case the drop of video quality is
less observable, as it can be seen in Fig. 12.
In the present paper an adaptive frame-based transmission
in a multilink environment composed by LTE, WiMAX and
HSDPA links for MPEG-2 video transmission was proposed.
By applying this method, the video quality can be enhanced
and the most important frames can be transmitted with the
highest possible probability, since the possibilities of the
multilink network are exploited. The method can be adopted
for spatial scalability and layered video streams as well.
IEEE 802.16 Working Group on Broadband Wireless Access Standards.
Available at: http://grouper.ieee.org/groups/802/16
P. Kántor and J. Bitó, "Adaptive Video Transmission over Multilink
Networks", 52nd International Symposium ELMAR-2010, Zadar,
Croatia, pp. 269-272,15-17 September 2010, ISBN 978-953-7044-11-4
IEEE Standard Computer Dictionary: A Compilation of IEEE Standard
Computer Glossaries. Institute of Electrical and Electronics Engineers
J. Watkinson, “MPEG-2”, Focal Press, 2nd edition; 1994
D. G. Andersen, A. C. Snoeren, and H. Balakrishnan, “Best-path versus
multi-path overlay routing,” in Proc. 3rd ACM SIGCOMM Conf.
Internet Measurements, Oct. 2003, pp. 91–100.
W. C. Y. Lee, Mobile Communications Engineering. New York, NY,
USA: McGraw-Hill Inc., 1982.
D. Jurca, P. Frossard and A. Jovanovic, “Forward Error Correction for
Multipath Media Streaming,” IEEE Transactions on Circuits and
Systems for Video Technology, vol. 19, no. 9, pp.1315-1326, 2009.
G. Xie, M. N. S. Swarmy and M. O. Ahmad, "Joint optimal multipath
routing and rate control for multidescription coded video streaming in
Ad Hoc networks", IEEE Transactions on Multimedia, 2008, 10(8):
Andrews, J. G., A. Ghosh, and R. Muhamed, "Fundamentals of
WiMAX", Prentice-Hall, February 2007.
 A. Osman and A. Mohammed, “Performance evaluation of a low-
complexity OFDM UMTS-LTE system” in Proceedings of the IEEE
Vehicular Technology Conference (VTC '08), pp. 2142–2146,
Singapore, May 2008.
 M. Wrulich and M. Rupp, “Computationally efficient MIMO HSDPA
system-level modeling”, EURASIP J. Wireless Communications and
Networking, vol. 2009
Figure 12. Comparison of the efficiency of the non-adaptive and adadptive transmission for the case of the 100
th GOP and pedestrian channel environment