CHAPTER 2
LITERATURE SURVEY
21 REVIEW OF LITERATURES
The survey is divided into four sections such as BS parameters
selection using EMOA, BS location identification in real environment,
dynamic traffic demand model and evolutionary algorithms.
2.1.1 BS Parameters Selection Using EMOA
In general, maximum cell coverage with minimum number of base
stations is a NP-hard problem (Anderson and McGeehan 1994). Researchers
have applied various tools to find the optimum location of BS in the cellular
network planning. Integrated Cellular Network Planning Tool (ICEPT)
proposed by Tutschuka K et al (1997) considered four types of major design
aspects namely mobile subscriber, radio transmission, resource allocation and
system architecture
Every cellular organization desires to achieve the seamless
connectivity and maximum coverage through proper base station placement
(Theodore S.R, 2001). The design parameters like trans
ssion power, height,
azimuth, and tilt angle and path loss are normally taken into account during
commissioning of base station in the desired area. The determination of
coverage in Global System for Mobile Communication (GSM) network
(Hurley. S 2002) considers the non-adjustable constraints like traffic demand,
interferences, overlap and handover.BS placement is an important and challenging task in the cellular
mobile network planning and has been approached by various researches
u
ing different types of network planning model. Kapp Rawnsley and
Hurley S (2000) s
placement of base station with configuration and frequency assignment.
ggested network planning model with two phases namely
Allen S M et al (2001) compared the performance of the automated cellular
network planning with the manual planning in broadband wireless fixed
network.
Zimmermann et al (2003) recommended an advanced model of cell
planning using hexagonal structure, path loss propagation, antenna diagram
losses, handover and traffic demand. Augmented objectives are formulated by
combining three quality objectives with traffic and coverage constraints.
Hurley $ (2002) proposed the effective multiobjective BS placement model
u
ing four real-world datasets with coverage, site cost, traffic, interference
and handover as objectives using Simulated Annealing (SA) method. Antony
Roullier-Callaghan (2001) proposed cost and time efficient radio wave
coverage and planning tool (RAWCAPT) for its radio path planning using
Lidar datas
s. Coello (2005) introduced Oasys automatic cell planning tool to
determine optimum parameters like transmitted power, tilt and azimuth,
Algorithms such as heuri
ic, meta-heuristic:
uch as SA, Tabu
search, GA, and Greedy algorithms have been applied in previous methods
for base station placement. Deterministic heu
and Ulrich Behr 2000; Mathar R M and Niessen T 2000) constructively
produced feasible solutions which mainly concentrated on modeling of cell
stic algorithm (Xuemin Huang
plan, Meta-heuristic algorithm based on simulated annealing (Anderson H and
McGeehan JP 1994; Mathar R M and Niessen T 2000; Ralph-Michael Kling
and Prithviraj Banerjee 1987) and Tabu
search (Michel Vasquez and Jin-Kao
Hao 2001) produced local optimum results from initial neighborhood set ofsolutions. The application of GA in cell planning (Hurley $ 2002;
Caminada A et al 2002; Weicker et al 2003; Calegarie et al 1996; Molina et al
1999) has produced significantly better results from the initial identified
location coordinates.
The multiobjective base station placement models were proposed
by Whitaker R M and Hurley S (2005); Whitaker et al (2005); Raisanen et al
(2004); and Raisanen L and Whitaker R M (2005). Whitaker Roger M et al
(2005) demonstrated new graph theoretical model to derive improved lower
bounds on the minimum number of
s with coverage, traffic and
interference as objectives using back tracking routine, Whitaker Roger M
et al (2005) have also investigated the application of NSGA-II to optimize the
multiple objectives service coverage directly and total cost while imposing
constraints on pair-wise cell overlap.
The performance comparisons of various EMOAs such as PAES,
SPEA2, SEAMO and NSGA-II on three objectives such as traffic, cost and
coverage in 15x15 Km? cell network planning issue (Raisanen and Whitaker
2005; Whitaker Roger M et al 2005) are discussed. Recently, Raisanen L
(2008) has introduced a permutation-coded multiobjective evolutionary
approach with the measurement of five realistic network objectives such as
coverage, cost, traffic hold, interference and handover. All these research
work have not considered all the nec
wy design parameters, objectives and
constraints in the optimization proc:
awhole. Also, the epistatic relations
between design parameters and objectives are not cor
dered as only site
coordinates (x, ») in their optimization process.2.1.2 BS Location Identification in Real Environment
The challenge of choosing the optimum locations for base stations
in a microcell communications system is addressed using SA techniques
which offer a method to achieve near optimum solutions to complex
combinatorial optimization issues (Anderson H and McGeehan JP 1994). The
recommendations show that SA is a viable approach for handling practical
base station location identification in an urban microcell environment, Using
a simple propagation model, optimum transmitter sites can be successfully
selected automatically regardless of the initial position of the transmitters.
For the estimation and characterization of the expected teletraffic in
mobile communication networks, a new method is presented through the
geographic traffic model (Tutschku K and Tran-Gia 1998). This new model
obeys the geographical and demographical factors for the demand in mobile
communication services. The geographical information systems were used to
estimate the teletraffic demand in an early phase of the network design
process.
The performance of three different algorithms to solve the BS
location identification: the Greedy Algorithm (GR), GA and the Combination
Algorithm for Total optimization (CAT) are presented and results are given
for a typical test scenario (Molina et al 1999). The effectiveness and accuracy
of each algorithm is evaluated through comparison of its results with
measurements taken in a real environment.
The selection of suitable path loss model according to propagation
environment is very important for field strength calculations. Three path loss
models namely ECC-33, SUI and COST-231 Hata are compared to predict the
suitable one for Cambridge city (Abhyawardhana V S et al 2005). The10
significance of reflection coefficient and loss through the comparisons of
three path loss models Hata-Okumura, Walfish-Ikegami and ITU-R are
discussed (Minseok Jeong and Bomson Lee 2001, Wang et al 2009).
‘A new solution was given to the positioning BS of a mobile phone
network through heuristic approach based on the evolutionary paradigm
(Weicker et al 2003).Two standard multiobjective techniques are compared
and a new algorithm, the steady state Evolutionary Algorithm with Pareto
Tournaments (stEAPT) is recommended. The demonstration shows that
evolutionary algorithms are a strong enough tool to tackle the real-world
challenges of base station transmitter placement based on the real situation to
the region of Zurich.
Radio Network Design (RND) issues using a multi-objective
version of the algorithm CHC (Cross generational elitist selection,
Heterogeneous recombination and Cataclysmic mutation), which is
metaheuristic in nature is presented (Antonio Nebro et al 2007). The new
algorithm MOCHC (Multi-Objective CHC) is compared against a binary-
coded NSGA-II algorithm. The experimental results carried out in the real
environment indicate the better performance of MOCHC algorithm, In this
thesis a real urban area is considered with all environmental parameters and
suitable recommendations are given for best coverage.
2.1.3 Dynamic Traffic Demand Model
Earlier literatures, regarding traffic model, mainly focused towards
channel residing time and handover. A traffic model for mobile radio
telephone systems with cellular structure, frequency reuse, and handoff has
been proposed (Dachyoung Hong and Stephen S Rappaport 1986). ThreeWW
schemes for call traffic handling such as one nonprioritized and two priority
oriented are considered.
‘A mathematical formulation is developed for systematic tracking of
the random movement of a mobile station in a cellular environment.
Zonoozi, M M et al (1996) developed a mathematical model for the
systematic tracking of the random movement of a mobile station in a cellular
environment, This model characterizes different mobility related traffic
parameters of cellular mobile communication,
In most cellular environments, a user mobility model which
incorporates mobility parameters under the most generalized conditions is
proposed (Zonoozi M M and Prem Dassanayake 1997, Weihong Wang
et al 2007). This mobility model is used to characterize different mobility-
related traffic parameters of cellular systems which include the distribution of
the cell residence time of both new and handover calls, channel holding time,
and the average number of handovers.
‘A good mobility model is necessary to capture the importance of
Personal Communication Systems (PCS) network behavior (Orlik P V and
Rappaport S$ 1997).The recommended model must be good enough to fit
field data and make the resulting queuing model still tractable. The mobility
model, called hyper-Erlang distribution model is used to characterize the cell
residence time and obtain analytical results for the channel holding time
(Yuguang Fang et al 1998). This characterization is very important in
teletraffic analysis of PCS networks. The effective mobility model plays a
significant role for field data processing in PCS network design and
performance evaluation,12
In third generation systems, the influence of mobility on the
network performance will be strengthened due to the huge number of mobile
users in conjunction with the small cell size. In particular, accuracy of the
mobility model is very much important for the evaluation of system design
and network implementation cost i
ues (Tutschku K and Tran-Gia P 1998),
Markoulidakis J G et al (1997) proposed three basic types of mobil
namely City Area Model, Area Zone Model and Street Unit Model for the
ty models
analysis of the full range of mobile communications design issues. The author
highlighted the refined modeling approach through integrated mobility
modeling tool which improved the accuracy of input parameters of each basic
model,
Wang Ping et al (2009) recommended novel user mobility model
namely Markoy-based hierarchical user mobility model (MB-HUMM). This
model adopted two hierarchical structures namely Global User Mobility
(GUM) and Local User Mobility (LUM).The GUM depicted the macro
mobility of mobile user among cells in a Location Area (LA) and the LUM
depicted the micro mobility of mobile user in a cell. The definition of the
parameters as well as simulations to validate the model is discussed.
2.1.4 Evolutionary Algorithms
EAs are computer programs that attempt to solve complex tasks by
mimicking the processes of Darwinian evolution (Darwin C 1859). In EAs, a
number of artificial creatures search over the space of the problem. They
compete continually with each other to discover best areas of the search
space. From the optimization point of view, EAs methodologies have shown
their evident benefits in solving nonlinear, complex multimodal optimization
problems. Due to their robustness and easy applicability, the fact that they do
not need specialized information such as gradient information or smoothness13
of objective functions, These algorithms are capable of solving complex
optimization issues having discontinuous, non-convex and highly nonlinear
solution space. In addition, they can solve challenges that feature di
binary variables, which are difficult.
Of the several techniques developed within the evolutionary
computation field, the most popular evolutionary algorithms are: Genetic
Algorithm (Michalewiez Z 1996, Goldberg D E 1999), Evolutionary
Programming (Fogel D B 1992, Back T and Schwefel H P 1993, Yao X, et al
1999), Evolution Strategies (Fogel D B 1994, Kern S, et al 2004), Particle
‘Swarm Optimization (Shi Y and Eberhart R C 1998), Differential Evolution
(Storn R 1996, Storn R and Price K 1997), Tabu search (Glover F and
Laguna M 1997), Ant Colony Optimization (Dorigo M and Stutzle T 2000)
and Bacterial Foraging Algorithm (Passino K M_ 2002). Alll these techniques
have been applied successfully to numerous engineering applications.
One of the first approaches to utilize the concept of Pareto
optimality was Fonseca C M and Fleming P J (1993) MOGA, which tends
to be the favored approach for control engineers. Several excellent
surveys of EMOA activity can be found, namely, (Goldberg D E 1999),
Veldhuizen D A V and Lamont G B (2000), Coello C A C (2005), Deb K
(2001) and Fonseca C M and Fleming P J (1998a, 1998b). The essential
difference between a MOGA and a single objective GA is the method by
which fitness is assigned to potential solutions. Each solution will have a
vector describing its performance across the set of criteria. This vector must
be transformed into a scalar fitness value for the purposes of the GA. This
process is achieved by ranking the population of solutions related to each
other, and then assigning fitness based on rank. Individual solutions are
compared in terms of Pareto dominance.14
The formal definition of Pareto dominance and Pareto optimality
are given below (Deb K 2001). Recently,
carried out in crossover and mutation mechanisms of Real coded Genetic
several modifications have been
Algorithm (RGA) such as Simulated Binary Crossover (SBX), Parent Centric
Crossover (PCX) and non-uniform polynomial mutation to improve the
performance of RGA. Self-adaptive simulated binary crossover based RGA is
successfully applied to various engineering optimization challenges (Deb K
and Goel T 2001).
Nowadays, EMOAs are enjoying immense popularity, mainly
because of their ability to find out the widely spread Pareto optimal solutions
within a single simulation run, EMOAs developed so far can be categorized
into first generation and second generation techniques. First generation
EMOAs include Non-Pareto techniques and Pareto based techniques like
Vector Evaluated Genetic Algorithm (VEGA), Multiobjective Genetic
Algorithm (MOGA), Nondominated Sorting Genetic Algorithm (NSGA),
Niched Pareto Genetic Algorithm (NPGA) and others (Coello C A C 2005).
In general, first generation EMOAs have the following shortcomings:
1. A fitness sharing as a tool is used to keep diversity in the population
throughout the whole Pareto front. Hence, a fitness sharing factor is
required to be set.
2. Eli
ism is not incorporated which improves significantly the
performance of multiobjective algorithms.
The shortcomings of first generation algorithms are well addressed
in the second generation EMOAs by the use of elitism, Nondomination
ranking, and diversity preservation, In order to implement diversity, the
concepts of external archive and crowding distance (Deb K et al 2000)
methods have been employed. Archive based elitist second generation
EMOAs mainly include Pareto Archived Evolution Strategy (PAES), Strength15
Pareto Evolutionary Algorithm, (SPEA), Multiobjective Particle Swarm
Optimization (MOPSO) (Coello C A C 2005).
Deb K and Goel T (2002) proposed a NSGA-II algorithm, which is
computationally more efficient. It uses elitism and a crowded comparison
operator that keeps diversity without specifying any additional parameters.
During selection, the NSGA-II uses a crowded comparison operator which
takes into consideration both the nondomination rank of an individual in the
population and its crowding distance (i.e. nondominated solutions are
preferred over dominated solutions, but between two solutions with the same
nondomination rank, the one that resides in the less crowded region is
preferred),
NSGA-II does not use external memory as the previous algorithms.
Instead, its elitist mechanism consists of combining the best parents with the
best offspring obtained. Among EMOAs, NSGA-II is the most efficient and it
wes. The
is applied to solve su
sfully various engineering optimization
concept of CE was discussed to ensure better convergence by Deb K (2000).
Biao Luo et al (2008) have proposed the DCD for horizontal diversity in the
Pareto front. Recently, MNSGA-II algorithm has been proposed to
multiobjective optimization problem in power system after introducing
CE and DCD operator to improve both vertical and horizontal diversity
respectively (Jeyadevi S et al 2011).
The Real-coded Genetic Algorithm with Simulated Binary
Crossover (RGA-SBX) is chosen to identify the best obtained locations of
single objective BS placement in this work. MNSGA-II extends standard
evolutionary-based optimization techniques to allow individual treatment of
multiple objecti
concepts of CE (Deb K 2001) and DCD (Biao Luo et al 2008) to maintain
s simultaneously. In this work, MNSGA-II incorporates the
lateral diversity and horizontal diversity which are lacking in NSGA-IL