A Hybrid Metaheuristic Approach to Optimize the Content
Transmission in Multimedia Systems
Arthur T. Gómez, Luan C. Nesi, Marcio G. Martins and Leonardo D. Chiwiacowsky
Master Program on Applied Computing, University of the Sinos Valley, Av. Unisinos, 950, São Leopoldo/RS, Brazil
Keywords: Content Transmission, Metaheuristics, Hybrid Algorithm, Genetic Algorithm, Tabu Search.
Abstract: The advent of the digital television in Brazil has allowed users to access interactive channels. Once
interactive channels are available, the users are able to find multimedia content such as movies and breaking
news programs, to send and/or receive emails, to access interactive applications and also other contents. In
this context, a high demand of requests from users is expected. Therefore, from the content provider's point
of view, the determination of transmission parameters is needed in order to ensure the best quality of
transmission to every user. The aforementioned identification problem is modelled as an optimization
problem and a solution procedure based on metaheuristic techniques is proposed. Genetic Algorithm and
Tabu Search metaheuristics are employed separately and coupled in a hybrid scheme to define the best
transmission policy, optimizing the transmission parameters, such as audio and video transmission rates.
Based on the experimental results, the hybrid algorithm has produced better solutions which meet the
quality requirements.
1 INTRODUCTION
Digital TV has promoted technological progress in
telecommunications area. Additionally to the quality
improvements on audio and video transmission,
Brazilian Digital Television System also includes an
interactive channel that allows users to access
interactive applications, on-demand systems and
other resources which are provided by digital
content suppliers. The television stations are
expected to provide an interactive channel through
which clients should access the multimedia content
delivery system. In this context, the interactive
channel makes possible the communication with the
provider multimedia server and data transmission
(Manhães et al., 2005).
In order to guarantee a better reception quality
for all clients, a proper configuration of transmission
parameters must be identified by the content
provider. However, it is not a simple task since a
variety of families of clients could be identified
according their reception features. Therefore, the
transmission parameters to be set should maximize
the use of the reception structure of every family of
clients. Given the set of possibilities related to the
reception features, a complex combinatorial problem
arises.
Thus, two different metaheuristic models are
proposed to optimize content delivery in a
multimedia system. They are employed separately
and also coupled in a hybrid scheme. The proposed
optimization algorithms are compared according to
the definition of the best transmission policy. So, the
use of a hybrid metaheuristic for optimizing the
transmission of multimedia content in Digital TV is
one of the contributions of this paper. Furthermore,
the development of a mathematical model in
accordance with both Brazilian Digital Television
System and International transmission rules stated
by ITU-T (ITU-T, 2000) should be emphasized.
This paper is organized as follows. Section 2
presents the proposed architecture. In Section 2.1,
the mathematical model is defined. In Section 2.2,
the proposed metaheuristic models are presented.
The experiments and results are discussed in Section
3. Finally, conclusions are presented in Section 4.
2 PROPOSED ARCHITECTURE
A prominent feature in the Brazilian Digital
Television System is the interactive channel. The
77
Gómez A., Nesi L., Martins M. and Chiwiacowsky L..
A Hybrid Metaheuristic Approach to Optimize the Content Transmission in Multimedia Systems.
DOI: 10.5220/0005063500770084
In Proceedings of the 11th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2014), pages 77-84
ISBN: 978-989-758-039-0
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
interactive channel is an additional resource that
enables users, individually and independent from the
others, to interact with broadcasters by sending or
receiving information and requests. To allow the
information exchange between de users and the TV
station a connection is needed, usually through an
internet network. The transmission dynamics is
depicted in Figure 1.
Figure 1: The transmission dynamics.
Based on this premise, the user authentication is
requested to connect to a multimedia content
delivery system. After the authentication, a
communication between the user and the content
provider is established. From this moment, users can
request specific programs, multimedia contents or
TV products to be broadcasted by the TV station.
All the communication between television station
and the user is performed through an additional
communication channel rather than the broadcasting
channel. In the Brazilian Digital Television System
this communication channel is named interactivity
channel, depicted in Figure 2.
Figure 2: Interactive channel.
After the connection is established, the server calls
from time to time an optimizer agent to determine
the best way to deliver the content to the end user. In
this context, the optimizer agent runs a metaheuristic
scheme to identify the set of transmission parameters
to guarantee the best use of the bandwidth with
maximum reception quality, according to the device
network capacity.
In Figure 3, the proposed system dynamics is
depicted.
Figure 3: The proposed system dynamics.
After the user connection is established, the control
is passed to the transmission content module. The
coordinator and optimizer agents are included in this
module. The coordinator agent controls the other
agents to run the metaheuristic with the objective to
determine the parameters for the best content
transmission policy. At the end of this process, each
optimizer agent transmits the results found by the
metaheuristics to the coordinator agent. Following,
an analysis is performed by the coordinator agent to
identify the best configuration that was chosen
according to the transmission content related do the
user. The content streaming could be transmitted
over any network model, for instance: IPTV
(Internet Protocol Television), broadcasting or
multicasting. After the transmission is finished, the
communication channel is closed.
2.1 Mathematical Formulation
This section presents the mathematical model for the
problem of optimal identification of transmission
parameters. The referred optimization problem could
be formulated as a multi-objective problem since a
compromise solution, based on the transmission
parameters, should be identified in order to ensure
video and audio quality for all the families of clients.
Nevertheless, in this work, the combinatorial
optimization was modelled as a single-objective
ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics
78
optimization problem by aggregating the objectives
in a weighted sum.
The objective function (OF) is defined based on
decision variables which represent five different
video qualities and two different audio qualities. The
constraints define the domain values of the decision
variables and assume a upper value for the
bandwidth. This mathematical formulation is used in
the optimization procedure to find the best possible
utilization of the server bandwidth in order to attend
all connected clients. This problem resembles a
classical combinatorial optimization problem, the
Selection Parts Problem. The problem of selecting
parts belongs to the class of NP-Complete
complexity problems (Kusiak et al., 1984).
(1)
1
2
51
;
subject to:
0.011.00;
(2)
1.0012.00;
(3)
2.005.00;
(4)
5.00210.00;
(5)
10.0018.00;
(6)
0.0960.256;
(7)
0.384511.00;
(8)
500;
(9)
,,,,,,0;
(10)
,,,1,2,,51∈
;
(11)
(12)
1
2
51
,
where:
• SB = total bandwidth of the television station
server (total flow system);
• vLD = LD (Low Definition) video quality;
• vSD = SD (Standard Definition) video quality;
• vHD = HD (High Definition) video quality;
• vP1 = intermediate configuration between LD and
SD;
• vP2 = intermediate configuration between SD and
HD;
• aST = stereo audio quality;
• a51 = multichannel 5.1 audio quality;
• nLD = number of clients connected as LD;
• nSD = number of clients connected as SD;
• nHD = number of clients connected as HD;
• nP1 = number of clients connected as P1;
• nP2 = number of clients connected as P2;
• nST = number of clients connected as stereo audio
quality;
• n51 = number of clients connected as multichannel
5.1 audio quality;
• α = importance level of the LD transmission;
• β = importance level of the SD transmission;
• γ = importance level of the HD transmission;
• δ = importance level of the P1 transmission;
• ω = importance level of the P2 transmission;
• θ = importance level of the Stereo Audio
transmission;
• ρ = importance level of the multichannel 5.1
transmission;
• All decision variables are assumed as real numbers.
The mathematical formulation shows that the
system bottleneck is the bandwidth available in the
AHybridMetaheuristicApproachtoOptimizetheContentTransmissioninMultimediaSystems
79
multimedia content delivery system. This is a
limited resource and has a high cost. Therefore, the
optimization procedure must provide a parameter
configuration which is able to utilize this resource in
the best possible way. The constraints are defined in
equations (2) to (12).
The equations (2) to (8) define the domain values
of the Video Quality (VQ) and Audio Quality (AQ)
in accordance with international standards defined
by ITU-T (ITU-T, 2000). These ranges are defined
according to the quality of video transmission (SD,
LD, HD, P1 and P2). The equation (9) limits the
bandwidth available in the multimedia server of the
TV station.
Equation (10) defines the parameters
,
,
,
,
,
and
which must be greater than zero. These
parameters are used to set a priority for a particular
type of transmission, when congestion on the server
is verified, with a large number of clients connected.
In this case, a specific video quality can be
prioritized, for example, the SD transmission, which
consumes a lower bandwidth than HD.
Equation (11) defines how many clients the
multimedia server is able to serve with a reasonable
quality.
Equation (12) defines that the sum of Video
Quality (LD, SD, HD, P1 and P2) and Audio Quality
(stereo and multichannel 5.1) should not exceed the
bandwidth available on the server.
2.2 Metaheuristics
The use of metaheuristics to solve the parameter
optimization problem in the context of IPTV content
transmission was addressed in (Weissheimer Jr.,
2011), where a model based on the application of
metaheuristics is presented to find the best
transmission parameters configuration on an IPTV
platform, assuming different types of receiving
devices and users.
A system to configure parameters of digital TV
video encoder using the H.264 standard has been
proposed in (Linck, 2011). The referred work is
based on Tabu Search and Genetic Algorithm
metaheuristics. A hybrid algorithm based on these
metaheuristics was developed. The Tabu Search
metaheuristic was used to intensify the search in
conjunction with the power of diversification of
Genetic Algorithms. This hybrid approach was
applied in the solution of combinatorial optimization
problems related to the encoding and decoding video
signals.
A new approach to perform the optimization of
bandwidth usage to ensure Quality of Service (QoS)
for IPTV broadcasts was developed in (Kandavanam
et al., 2009). A new algorithm combines Genetic
Algorithm and Variable Neighborhood Search
metaheuristics to solve the optimization problem.
Good results were achieved, since there were
significant improvements in the distribution of the
available internet link, the maximum use of
available internet link and also in the rejection rate
of service. In the present work, Genetic Algorithm
and Tabu Search were applied together resulting in a
hybrid algorithm. These techniques are the same
used in (Linck, 2011) and (Weissheimer Jr., 2011),
however in a different problem. Following, the
proposed optimization metaheuristics are presented
2.2.1 Genetic Algorithm
The Genetic Algorithm (GA) was proposed by John
Holland (Holland, 1975). It has been applied to
solve combinatorial optimization problems in
different areas such as mathematics, physics,
biology, engineering, industrial automation, among
others. According to Goldberg (1989), GAs are
search algorithms based on mechanisms of natural
selection and natural genetics. They employ the
survival-of-the-fittest principle and propose a
random exchange of information.
The initial population of the Genetic Algorithm
(GA) is randomly generated and should adequately
map the search space. The selection process used in
this work is based on the tournament method. The
population is sorted in descending order according
fitness, and a random number is drawn from a
uniform distribution in the interval [0, 1]. If it is less
or equal than 0.75, two individuals are chosen
randomly from the superior half of the population,
i.e. from the best individual portion. On the other
hand, if the random number is higher than 0.75 one
individual from the superior half is selected and the
other one is selected from the inferior half of
population (Filho and Tiberti, 2006). By using this
method, a worst fitted individual has the chance to
be crossed with better fitted individuals. This
strategy can lead the algorithm to unexplored areas
and thus improve the objective function value over
the generations.
The arithmetic crossover operator is responsible
for crossing the genetic information of parent
individuals to generate new individuals. It is used
with a predefined probability. The uniform mutation
is the genetic operator used to guarantee the search
space exploration and also maintain the diversity of
the population. It is also used with a predefined
probability.
ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics
80
The stopping criterion adopted in this work was
the maximum number of generations which was set
to 100.
2.2.2 Tabu Search
Tabu Search (TS) is a metaheuristic proposed by
Fred Glover (1986). It consists of a set of concepts
and practices used to solve combinatorial
optimization problems. This technique is used to
find approximate solutions to complex problems,
where the time to find the optimal solution is
exponential or factorial.
The search process is based on the generation of
neighborhood solutions, also called Local Search
(LS). The two basic elements of TS are the search
space and neighborhood structure. However, the
main difference between the TS and the LS is a list
that restricts the algorithm in order to prevent
reverse movements, called Tabu List, hence the
name Tabu Search. Through this mechanism, the
algorithm has the ability to escape from local
optima, spanning a wider search space and
producing better results (Gendreau et al., 2001). The
size of the tabu list is defined according to the
experiment. The aspiration criteria is used to ensure
the OF improvement if a movement is tabu.
Different neighborhood structures based on
different types of movements have been proposed to
generate new solutions in the Tabu Search (Glover
and Laguna, 1997), including swap, insertion and
removal. In this work a random removal/insertion
movement was used.
The stopping criterion adopted in this work was
the number of interactions without OF improvement.
This parameter is named Nmax and was set to 100
(Chung et al., 2011).
2.2.3 The Hybrid Algorithm
The proposed Hybrid Algorithm uses the concepts of
the GA and TS metaheuristics. It combines the
exploration capacity of GA together with the
exploitation power of TS.
In the Hybrid Algorithm the TS is initially
performed until the stopping criterion is satisfied,
i.e., until the number of iterations without
improvement in OF value reaches 100. The 20 best
solutions obtained in TS phase are used as the initial
population of the GA. Following, the GA is
performed until the maximum of 100 generations is
reached. After the end of a complete cycle
(TS+GA), the best solution obtained in GA is
returned to TS, from which a new hybrid
optimization cycle starts. In the hybrid algorithm,
together with the objective function defined in
equation (1) an additional criterion was used to
evaluate the solution quality. This criterion is
represented by the Harmonic Mean (HM) which is
also computed at the end of TS and GA phases. The
HM is a secondary criteria used to identify the best
solution when two different solutions present the
same OF value. The HM is computed as follow in
equation (13):
1
1
⋯
1
∑
1
,
(13)
where n is the total number of clients families,
0 is the transmission rate value assigned for
the i-th client. So, the hybrid algorithm has used as
stopping criterion the number of 20 complete cycles
(TS+GA) without improvement on both the
Objective Function (OF) and the Harmonic Mean
(HM).
3 EXPERIMENTS AND RESULTS
The experiments to validate the proposed algorithm
were made through the harmonic mean of the results
generated by the Genetic Algorithm, Tabu Search
and Hybrid Algorithm.
3.1 Tuning of Parameters
To avoid biased solutions, the OF weights were
tuned by normalization. Based on 100 runs of hybrid
algorithm with a random initial solution assumed,
the average value of each decision variable is
obtained. The corresponding weights (
,
,
,
,
,
and
) were computed by the quotient of the sum of
the average values and each average value. This
procedure allows to identify the contribution of each
decisions variable on the OF value. The respective
weight values are presented in Table 1.
Table 1: Definition of Objective Function Weights.
Variable Average Parameter Weight
vLD 0.392
58.079
vP1 1.219
18.689
vSD 2.694
8.452
vP2 6.033
3.775
vHD 11.349
2.007
aST 0.539
42.266
a51 0.548
41.594
Total 22.773
Beyond the weights used in the objective
AHybridMetaheuristicApproachtoOptimizetheContentTransmissioninMultimediaSystems
81
function, the parameters used in AG and TS
metaheuristics should be tuned as well. For AG
metaheuristic, different combinations of both
crossover and mutation probabilities were evaluated
for the solution of a small scale instance. From a set
of 100 runs performed for each possible
combination, the experiments showed the crossover
probability equal to 75% and the mutation
probability equal to 15% as the best tuning.
Parameters of TS metaheuristic should also be set by
performing tuning experiments. Different values of
both neighborhood size and the tabu list size were
evaluated. The best values for the parameters to be
tuned were identified by performing a set of 100
runs when solving a small scale instance. The tuning
experiments indicated a neighborhood size equal to
100 and a tabu list size equal to 25 for small instance
and 250 for medium and large instances.
3.2 Optimization Algorithm
for Quality of Transmission
The optimization algorithm aims to optimize the
quality of transmission of generic content that is
transmitted taking into account the number of
connected clients in a multimedia server.
Firstly, a validation experiment comparing the
performance between a linear relaxed optimization
algorithm and the Genetic Algorithm and Tabu
Search metaheuristics was carried out. Next, the
experiments were performed taking three different
test programs: (i) a small scale instance with 15
clients; (ii) a medium scale instance with 1,000
clients and (iii) a large scale instance with 15,000
clients.
Initially, a comparison experiment was carried
out assuming two different solution strategies. The
15 clients instance was solved using exact and
metaheuristic approaches. The Simplex Method,
through the use of LINGO software, had its
performance compared with GA and TS
metaheuristics.
In this experiment, the constraints were relaxed
and the decision variables (vLD, vP1, vSD, vP2 and
vHD) were limited in a range of 0.01 Mbit/s to 100
Mbit/s. Figures 4-6 show the results obtained by
using Lingo and the metaheuristics.
It should be noted that Lingo obtained the best
values, but it has prioritized only five clients. The
respective five clients can be identified in Figure 4
by the transmission values between 79 and 100
Mbit/s. On the other hand, Figures 5 and 6 show that
the metaheuristics, which employ the HM metric,
has not achieved best OF values, but it makes
Figure 4: Bandwidth distribution achieved by Lingo.
possible the inclusion of almost all clients in the
transmission.
Figure 5: Bandwidth distribution achieved by GA.
Figure 6: Bandwidth distribution achieved by TS.
In order to understand the behavior of the objective
function and the harmonic mean a comparative
graphic is shown in Figure 7.
Figure 7: Comparison between OF and HM.
Figure 7 presents the evolution of both the OF and
ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics
82
the HM values as the number of iterations increases.
It is possible to observe an improvement of both the
OF and HM values compared to the initial solution.
The initial value of the OF was 629.992 Mbit/s
while the final value was 733.116 Mbit/s. It
represents an improvement of approximately 16%
over the initial value. Considering the HM metric,
the initial value was 0.422 MBit/s and the final value
was 0.730 Mbit/s. It represents an improvement of
approximately 73% over the initial value
3.2.1 Small Scale Experiments
Figure 8 shows a comparison of the final results
obtained by GA, TS and HA based on the harmonic
mean metric for the small scale instance. These
results are average values computed from 100 runs
of each optimization scheme. In this experiment, it
was assumed that each video quality (LD, P1, SD,
HD and P2) is requested by three clients, totalling 15
clients. Concerning audio quality, it was assumed
that stereo audio is requested by five clients and 5.1
multichannel audio is requested by ten clients. The
Server Bandwidth (SB) was limited in 70 Mbit/s.
As it can be observed from Figure 8, GA and TS
have achieved similar results and the HA had the
best performance with a harmonic mean of
transmission rate equal to 0.73 Mbit/s with a
corresponding OF value of 733.116 Mbit/s.
Moreover, Fig. 6 also presents standard deviations
values (SDev) computed based on results of each
metaheuristic. Once more, HA has achieved best
results with smaller SDev values
Figure 8: Harmonic mean and standard deviation (small
scale problem - 15 clients).
3.2.2 Medium Scale Experiments
Figure 9 shows the respective results for an
experiment performed considering a medium scale
problem. Once more, these results are average
values computed from 100 runs of each optimization
scheme. In this experiment, it was assumed that each
video quality is requested by 200 clients, totalling
1,000 clients. Concerning audio quality, it was
assumed that stereo audio and 5.1 multichannel
audio were requested by 500 clients each. The server
bandwidth was limited in 700 Mbit/s. Once again,
the best result was achieved by HA metaheuristic,
with OF value equal to 7,917.530 Mbit/s and
harmonic mean of 0.374 Mbit/s. Also, the smaller
standard deviations value was achieved by HA
metaheuristic.
Figure 9: Harmonic mean and standard deviation (medium
scale problem - 1,000 clients).
3.2.3 Large Scale Experiments
Lastly, Figure 10 shows a comparison of the final
results obtained for a 15,000 clients instance.
Concerning video quality, the clients were grouped
in the same five groups as presented in last two
experiments, with 3,000 clients each group.
Regarding the audio quality, the total of 15,000
requests was equally divided, with 7,500 clients
requiring stereo audio quality and others 7,500
clients requiring 5.1 multichannel audio. The server
bandwidth was limited in 7,000 Mbit/s. Again, the
HA metaheuristic has achieved the best harmonic
mean results. Moreover, the standard deviation
values have shown once more the HA robustness
and effectiveness.
Figure 10: Harmonic mean and standard deviation (large
scale problem - 15,000 clients).
AHybridMetaheuristicApproachtoOptimizetheContentTransmissioninMultimediaSystems
83
4 CONCLUSIONS
This paper presented a model for the content
delivery optimization, based on the bandwidth
requirements on the multimedia content delivery
system. Based on the implementation and the
experiments, it was observed that the proposed
optimization model provided solutions which met
the specified requirements. Comparing the results
obtained by the metaheuristics, the hybrid algorithm
achieved the best results.
The main contribution of this work is the
proposal of a hybrid algorithm that generates good
quality solutions, optimizing content delivery in a
multimedia system. These solutions are applicable in
the Brazilian Digital Television System.
As future work, a more complex algorithm must
be developed, which contemplate specific data
traffic, features and behaviours. Also, a multi-
objective optimization model should be developed in
order to evaluate the proposed hybrid metaheuristic
robustness. Besides, additional experiments must be
performed with the proposed metaheuristics using
other mutation and crossover operators, for a better
intensification and with the aim of improving the
results in specific scenarios and environments.
REFERENCES
Chung, S. H., Wu, T. H., Chang, C. C., 2011. An efficient
tabu search algorithm to the cell formation problem
with alternative routings and machine reliability
considerations. Computers and Industrial
Engineering, 60(1)1, 7–15.
Gendreau, M., Laporte, G., Potvin, J.-Y., 2001. The
vehicle routing problem. In Society for Industrial and
Applied Mathematics ch. Metaheuristics for the
Capacitated VRP, Toth, P., Vigo, D. Editors.
Philadelphia, 129–154.
Glover, F., 1986. Future paths for integer programming
and links to artificial intelligence. Computers and
Operations Research, 13(5), 533– 549.
Glover F., Laguna, M., 1997. Tabu Search. Kluwer
Academic Publishers. Norwell/USA.
Goldberg, D. E., 1989. Genetic Algorithms in Search,
Optimization and Machine Learning, Addison-Wesley
Longman Publishing Co. Boston, 1
st
edition.
Gonçalves Filho, E. V., Tiberti, A.J., 2006. A group
genetic algorithm for the machine cell formation
problem. International Journal of Production
Economics, 102(1), 1 – 21.
Holland, J. H., 1975. Adaptation in natural and artificial
systems: an introductory analysis with applications to
biology, control, and artificial intelligence, University
of Michigan Press. Ann Arbor, USA.
ITU-T, 2000. Information technology – generic coding of
moving pictures and associated audio information:
system ITU-T recommendation H.222.0 (iso/iec
13818-1). Telecommunication Standardization Sector
of International Telecommunication Union, February.
Kandavanam, G., Botvich, D., Balasubramaniam, S.,
2009. An optimization based approach to maximizing
QoS assurance for IPTV triple play services on the
internet backbone. In IEEE 34th Conference on Local
Computer Networks, 406–413.
Kusiak, A., Dorf, A., Richard, C., 1984. Handbook of
Design, Manufacturing and Automation, John Wiley
and Sons. New York.
Linck, I. C. C., 2011. Um sistema de codificacão de vídeo
para TV Digital - SBTVD, Master’s thesis on Applied
Computing (in Portuguese), Universidade do Vale do
Rio dos Sinos, São Leopoldo, Brazil.
Manhães, M. A. R., Shieh, P. J., Lamas, A. C., Macedo, P.
E. O., 2005. Canal de interatividade em TV Digital. In
Caderno CPqD Tecnologia (in Portuguese), 29–36.
Weissheimer Jr., C. A., 2011. Desenvolvimento de um
algoritmo híbrido utilizando metaheurísticas aplicado
a uma plataforma de internet protocol television
IPTV, Master’s thesis on Applied Computing (in
Portuguese), Universidade do Vale do Rio dos Sinos,
São Leopoldo, Brazil.
ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics
84
