Application of Taboo Search and Genetic Algorithm in
Planning and Optimization of UMTS Radio Networks
Bahareh Jalili
Centre for Communication Systems Research (CCSR)
Department of Electronic and Electrical Eng.,
University of Surrey
Guildford, GU2 7XH, Surrey, United Kingdom
b.jalili@surrey.ac.uk
ABSTRACT
Planning and optimization of 3G networks is more than just
frequency allocation and coverage planning, due to the nature of
WCDMA coding. It usually involves solution of an NP-Hard
problem. In this paper we propose an effective method for
optimizing the Common Pilot Channel (CPICH) transmit power,
along with maximizing the number of served users and
minimizing the number of cell sites and compare use of two meta-
heuristic methods: Taboo Search (TS) and Genetic Algorithm
(GA) in planning and optimization of UMTS radio networks.
Categories and Subject Descriptors
I.6.3 [SIMULATION AND MODELING]: Applications
General Terms
Algorithms, Management, Performance, Design, Economics,
Verification.
Keywords
Planning, Optimization, UMTS, 3G Network, Taboo Search,
Genetic Algorithm.
1. INTRODUCTION
In cellular networks, significant cost is associated with Base
Station (BS) deployment and operation. By choosing optimum
number of sites, the costs of network setup and operation and
maintenance can be reduced significantly. Figure 1 shows the
UMTS planning and optimization process, which consists of two
main phases: planning and optimization. There is a preparation
phase before the planning, in which coverage and capacity
objectives need to be defined, the network planning strategy to be
selected, and the initial design parameters to be agreed on.
Planning phase itself comprises of dimensioning and detailed
planning [1]. While dimensioning provides rough estimation of
network layout and required resources, detailed planning deals
with computation of the detailed parameters for every single site.
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. To copy
otherwise, or republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee.
"IWCMC’10, June 28– July 2, 2010, Caen, France. Copyright © 2010
ACM 978-1-4503-0062-9/10/06/...$10.00"
143
Mehrdad Dianati
Centre for Communication Systems Research (CCSR)
Department of Electronic and Electrical Eng.,
University of Surrey
Guildford, GU2 7XH, Surrey, United Kingdom
m.dianti@surrey.ac.uk
After a network is successfully planned and deployed, there will
always be ongoing optimization during operation and
maintenance phase.
Figure 1. UMTS Planning and Optimization problem
Planning phase can be regarded as an optimization problem,
as the best possible set of sites and parameters should be chosen
among a predefined set of candidate site locations and network
parameters [2]. Compared to 2G systems, there are more
parameters involved in link budget analysis and coverage
planning of 3G networks [3]. For instance, parameters such as
CPICH power, level of interference, traffic type and distribution,
soft capacity, and cell breathing, all should be considered in
network planning and optimisation. Typically, this is considered
to be a special case of knapsack optimization problem, which
results in a computationally complex optimization problem that
can not be solved by classical techniques. Exhaustive search
techniques also cannot be applied due to the Non-Polynomial
growth of the search space. Thus, heuristic search techniques have
to be applied to provide sub-optimal solutions [4] in the literature.
In [5], a Taboo Search Scheme has been used in an iterative
method to find the power levels of the signals emitted by BSs,
considering both uplink and downlink. Global optimization of
UMTS Networks has been studied in [6], where in planning the
radio access network, the paper considers the uplink optimization
using Taboo Search and reaches acceptable results compared to
numerical results derived using CPLEX. In [7], a mathematical
model is proposed, which is suitable for small size problems, and
uses a local search heuristic to obtain suboptimal solutions. In [8],
randomized and reverse greedy procedures are implemented for
uplink optimization, and TS is used to further improve the results.
In [9], the cell planning problem is formulated as an integer linear
programming problem and a solution based on Taboo Search is
proposed.
This paper applies 2 methods of Genetic algorithm and
Taboo Search for optimization of UMTS Radio Access Networks
in order to be able to compare their efficacy in terms of the quality
of the solutions and computational complexity. The aim is to Here, Ut is the total number of users, Us is the number of served
achieve sub-optimal solutions with reasonable computational users, Pt represents total available CPICH power from all sites, Pu
complexity. Our results demonstrate that both proposed solutions is total used CPICH power by all sites, St and So show total
reduce the cost of network deployment and operation by reducing number of sites and total number of “on” sites- in contrast to an
number of sites and used CPICH channel power. At the same “off” site, or a site that does not need to be deployed, indicated
time, they help the network operators to increase their revenue by with a CPICH value of zero- and W1 to W3 represent the weight
increasing the number of served users. The comparison between associated with the parameter we would like to optimize. While
the two algorithms demonstrates that Taboo Search is more Ut, Pt and St are fixed numbers in the system, we intend to find the
efficient, in terms of optimality of the results. However, our sub-optimal values for Us, Pu and So in a way that the objective
results indicate that Taboo search results in slightly more function F ( f1 , f 2 , f 3 ) is maximized.
computational complexity, measured by required run-time.
The rest of this paper is organized as follows. In Section 2, the 2.1 Genetic Algorithm
proposed solutions are discussed after a brief introduction to Genetic Algorithm is a general-purpose search algorithm based on
system model. Section 3 represents and discusses the simulation the principles of evolution observed in nature. It is a guided,
scenario and the results. Finally, the concluding remarks are given random, heuristic search method which mimics the evolution and
in Sections 4. reproduction concepts as seen in nature and has its roots and even
terminology in Charles Darwin’s Evolution theory introduced in
2. PROPOSED SOLUTION 1858.
System model incorporated in this paper considers a UMTS
network with an area of m × n (km) which needs to be provided
with coverage. In the dimensioning phase a rough estimation of
the number of required sites (St) is made and the operators have a
set of candidate sites; S= {S1, S2,…,St}. Depending on the
requirements, the cell sites can be sectorized in order to improve
frequency reuse, and reduce interference. The cells need to
provide coverage for a total number of Ut users in the area,
according to the traffic model suggested in the dimensioning
phase. Each cell operates at a power of PCPICH, which determines
the coverage and cell boundaries by the amount of power
allocated to the CPICH channel, which can take different values
from a range of discrete power values; PCPICH = {P1, P2,…, Pmax}.
The model just includes case of single operator operating in the
area.
The UMTS planning and Optimization problems can not be
decomposed in to two separate sub-problems of capacity and
coverage planning, and these two should be taken into account
simultaneously. Besides the fact that the system is interference
limited, and there are too many parameters which affect the
network and need to be taken into account simultaneously, makes
this problem an NP-hard problem, which can not be solved in
reasonable complexity and computational times, yet sub-optimal
solutions can be obtained when the problem is tackled by heuristic Figure 2. Optimization model using Genetic Algorithm
algorithms. Here we introduce and use Taboo Search and Genetic
Algorithm in dealing with this problem. In solving a problem with Genetic Algorithm, we need to define
the objective function according to the parameters to be optimized
In this paper, we aim to minimize the cost associated with cell and the problem in hand. Then, we present the possible solutions
sites, by using the minimum number of cell-sites and PCPICH, and as chromosome strings, making sure that all the optimization
maximize the revenue by increasing the number of served users. parameters are placed inside chromosome structure. In the next
Hence, the objective function can be defined as: step, a random population of N chromosomes is generated and the
Link Budget Analysis (LBA) and fitness function evaluation of
max F ( f1 , f 2 , f 3 ) = W1 f1 + W2 f 2 + W3 f 3 (1) each chromosome in the current generation should be performed.
Finally, to generate the N offsprings for the next generation, we
Subject to: repeatedly evaluate the fitness of each chromosome in the
population, using the objective function, then select a pair of
W1 + W2 + W3 = 1 (2) individuals from the current population for mating using a random
selection method (Roulette Wheel, Tournament, Rank selection)
Where and finally, based on crossover probability (Pc) and mutation
probability (Pm), we apply cross over and mutation operators.
Us Pt − Pu St − So
f1 = ; f2 = ; f3 = (3) The flowchart in Figure 2 shows application of GA to our
Ut Pt St problem.

144
2.2 Taboo Search In this paper, binary encoding was used, where each site’s
First introduced by Fred Glover in 1986, Taboo Search is a PCPICH was represented using 2 bits to map different CPICH
meta-heuristic algorithm, which is used in solving combinatorial values, i.e. {0, 24, 30, 33} (dBm). At the beginning of the
optimization problems. Taboo search benefits the use of memory simulation a random CPICH power value is assigned to each
structure, in order to improve the performance of a local search. sector. In each iteration, the optimization algorithm changes the
When a potential solution is found, it is marked as taboo so that corresponding CPICH value and feeds it back to the system. Then
the algorithm does not re-visit that possibility repeatedly. the link budget analysis is performed, and the results are again fed
Like GA, Taboo Search, starts with defining the objective back into the optimization algorithm.
function and a presentation for the possible solution. A set of
possible solutions, called neighbourhood, is then created and
starting from the current solution. Next, LBA is performed and the
objective value is calculated, and the solution, which maximizes
the objective function, is found in that neighbourhood. Finally, a
new neighbourhood, which is a set of solutions that are found by
applying an appropriate transformation of the current solution, is
generated. In order for the algorithm to move away from a local
minimum, the search allows moves resulting in a degradation of
the objective function value, thus avoiding the trap of local
optimality. To prevent the search from cycling, solutions obtained
recently and moves that reverse the effect of recent moves are
considered taboo and they are kept in a “Taboo List” so that these
points are not re-visited in the next k iterations. Figure 3 depicts
the Taboo Search applied in solving our optimization problem.

Figure 4. System model and user distribution

In our implementation, NPSW (Network Planning Strategies
for Wideband CDMA) [10] is used for calculating the link budget.
Genetic Algorithm/ Taboo Search are used to define different
input parameters to NPSW. NPSW is a static simulator, developed
under MATLAB environment that estimates capacity and
coverage probabilities given a snapshot for user distribution, by
performing iterative uplink and downlink analysis until
converging with predefined quality of service criteria.
3.1 Genetic Algorithm Results
In applying the Genetic Algorithm, rank selection was used,
and the fittest individuals from each generation were kept in the
next generation. Numerous simulations were run in order to find
the optimum values for the governing parameters like Pm, Pc,
population size and number of iterations. Figure 5 shows the
objective function value of the fittest individual and each
generation’s average fitness for 50 iterations and the results of this
optimization can be summarised in Table 1.
Figure 3. Optimization model using Taboo Search Table 1: GA Results summary

Max First Final

3. SIMULATION RESULTS Possible Iteration Iteration
In the performed simulations, seven cell sites are assumed. No. of users 1000 764 797
Each site is divided into 3 sectors. One site is located at the centre Total CPICH 639 579 507
and the other six sites are located on a hexagonal grid, in an area On Sectors 21 19 17
of 13 × 13 Km. A total number of 1000 users are generated and Fitness Value N/A 0.5738 0.6267
are distributed randomly in the area to be covered. Hard blocking As it can be seen in Table 1, the number of served users has
is used if users can not be served. The CPICH power can take increased from 764 users in the initial setup to 797 users in the
different values of PCPICH = {0, 27, 30, 33} dBm with 0 final setup, while used CPICH and number of on air sites has
representing a cell site which is turned off or does not need to be decreased. In order to see the actual effect on the network Figure
deployed. The allocated Bandwidth is 5 MHz, the carrier 6 compares the interference in these two different stages. Red and
frequency of 2GHz, and the chip-rate is 3.84 MHz. The path loss brown shades indicate high levels of interference and we can see
model used is Okumura-Hata (flat earth without area correction that in the initial set up sectors 1, 3, and 9 have high levels of
based on a map) [10] model. The antenna height is assumed to be other to own cell interference (around 90 % to 100 %), and sectors
35 meters, with the user height of 1.5 meters. Figure 4 shows the 2, 10 and 14 have interference levels around 70%, where as, after
setup and initial site and user distributions. the optimization only two sector (3 and 14) with 70% interference
and the for the rest of the network this value is below 50%.

145
 iteration and the curve with lower values shows the average
population fitness in each iteration.

Figure 5. Objective function values for Genetic Algorithm

Figure 7. Objective function values for Taboo Search

The corresponding values of the optimized parameters are shown
in Table 2, where we see that while the number of served users is
highly improved, the used CPICH and the number of “on” sites
have been decreased.
Table 2: TS Results summary

Max First Final

Possible Iteration Iteration
No. of users 1000 735 811
Total CPICH 639 576 525
On Sectors 21 19 17
Fitness Value N/A 0.5541 0.6326
Figure 8 shows the other to own cell interference. As it can be
seen in the initial setup the interference levels are really high, with
two sectors with 100% interference, two sectors with 90%, two
sectors around 80% and one sector around 70% interference.
However, after the optimization, the interference levels are
meaningfully decreased, with only two sites with an interference
level of around 70% and all other sites below 50%.
3.3 Comparisons
The final values of objective function, after completion of the
optimisation process, are good criteria for comparing the results,
acting as a common parameter in both search methods. Figure 9
shows the objective function of different search methods. The
results show that while at the beginning of the search the two
methods improve at the same pace, there is point where Taboo
Search shows better improvement over Genetic Algorithm,
keeping its objective function higher through out the search,
except at iterations around 35, where decrease of the TS objective
Figure 6. GA- Interference in first and last iteration function can be described as the effect of diversification and for
the search moving to new neighbourhoods in order to discover the
3.2 Taboo Search Results search space more thoroughly. As we see, the search then picks up
In applying Taboo, various runs of simulation have been again and finally the search ends with higher value of the TS
performed, in order to fine tune the parameters affecting the objective function.
results. Figure 7 shows the final results. The curve with higher
values shows the objective function value of the solution in each

146

Figure 8. Interference in first and last iteration
Figure 9. Objective function: a comparison between TA & GA
In terms of computational times, for 50 Iterations, GA and TS
took 19566 and 20456 seconds respectively. The longer time
associated with Taboo Search can be considered as the time due to
the extra step in Taboo Search in updating the Taboo List and then
comparing each neighbour against the Taboo List.
147
4. CONCLUSIONS
In this paper, we applied Taboo Search and Genetic Algorithm in
planning and optimization of UMTS radio networks. The
provided results show that in general both methods were
successful in optimizing the network. However Taboo Search has
a much better performance over the GA in all areas and it
improves faster. Not only Taboo Search arrived at better objective
function values in the same number of iterations, but also it
arrived to configurations which could service more number of
users with less power and less “on” sites. The results were verified
in network simulations and improvements could be obviously
seen.
5. ACKNOWLEDGEMENTS
This work is supported by the Engineering and Physical Science
Research Council (UK) of the United Kingdom under India-UK
Advanced Technology Centre (IU-ATC) project.
6. REFERENCES
[1] Holma, H. and Toskala, A. “WCDMA for UMTS”. John
Wiley & Sons, Ltd., England, 3rd edition, 2004.
[2] Nawrocki, Dohler, Aghvami, “Understanding UMTS radio
network modelling, planning and automated optimization”
John Wiley & sons, Ltd, 2006.
[3] Laiho, J. Wacher, A. “Radio NW planning process and
methods for WCDMA”, Nokia networks,
http://lib.tkk.fi/Diss/2002/isbn9512259028/article3.pdf , Jan
2009.
[4] E. Amaldi, A. Capone, F. Malucelli, “Radio planning and
coverage optimization of 3G cellular networks” Sringer, pp
435-447, 2008.
[5] M.E. Aydin, J. Yang, and J. Zhang, “A Comparative
Investigation on Heuristic Optimization of WCDMA Radio
Networks”, Center for Wireless Network Design (CWIND),
Dept. of Computing and Information Systems, University of
Bedfordshire, UK,
http://www.springerlink.com/content/65044304857l6r5r/fullt
ext.pdf?page=1 , Dec 2008.
[6] M. St-Hilaire, S. Chamberland, S. Pierre,”A Taboo Search
Heuristic for the Global Planning of UMTS Networks,”
IEEE International Conference on Wireless and Mobile
Computing, Networking and Communications, pp 148 – 151,
2006.
[7] M. St-Hilaire, S. Chamberland, S. Pierre, “Uplink UMTS
Network Design , an integrated approach”,
http://www.sciencedirect.com.
[8] E. Amaldi, A. Capone, F. Malucelli, “Planning UMTS base
station location: optimization models with power control and
algorithms’, IEEE Trans. Wireless. Communication, pp.
939–952, 2003.
[9] C. Y. Lee and H. G. Kang, “Cell Planning with Capacity
Expansion in Mobile Communications: A Tabu Search
Approach”, IEEE Transaction on Vehicular Technology,
Vol.49, No.5, pp. 1678-1691, Sep. 2000.
[10] Wacher, A. Laiho, J. Sipila, K. Heiska, K. Heikkinen, K.
"Matlab Implementation of a Static Network Planning Tool
for WCDMA”, Documentation accompanying the RNP Tool.