An Enhanced Ant Colony Optimization for Routing Area Mobility

Prediction over Cellular Communications Network

Mohammad Sh. Daoud

, Aladdin Ayesh

, Mustafa Al-Fayoumi

and Adrian A. Hopgood

Faculty of Technology, De Montfort University, Leicester, U.K.

College of Computer Engineering and Sciences, Salman bin Abdulaziz University, Al-Kharj, Saudi Arabia

Shefﬁeld Business School, Shefﬁeld Hallam University, Shefﬁeld, U.K.

Keywords:

Ant Colony Optimization, LBSs, Mobility Prediction, Cellular Network, UMTS.

Abstract:

Cellular communication networks have become medium to provide various services. Most of the services

provided are based on the users’ locations, as in location-based services (LBSs); these services include both

common voice services as well as multimedia and integrated data services. Used techniques mostly suffered

from complex computation, accuracy rate regression and insufﬁcient accuracy. Nevertheless, in the cell side,

reducing the complexity cost and preventing the prediction algorithm to perform in two closer time slot. That’s

why using routing area should be able to avoid the cell side problems. This paper discusses An Enhanced Ant

Colony Optimization for Routing Area Mobility Prediction over Cellular Communications Network (EA-

CORA) which is based on developed ant colony Optimization.

1 INTRODUCTION

Mobile networks have become the platform that pro-

vides leading-edge Internet services, for instance a

person can solve problems in any place without any

need to go to his ofﬁce or to travel, just by us-

ing his mobile phone or laptop. These services in-

clude both common voice services as well as mul-

timedia and integrated data services. Integration of

the Internet Protocol (IP) with Third-Generation (3G)

wireless communication through the Universal Mo-

bile Telecommunications System (UMTS) All-IP net-

work was proposed by Third-Generation Partnership

Project (3GPP), as next-generation in the telecommu-

nications networks.

However, these networks are facing problems

such as fragile wireless link, consume resource, de-

nied of services and mobility of mobile users. The

mobility location is changing during the constantly

movement of mobile user’s. The cellular communi-

cations network is divided into cells, each cell cov-

ers a speciﬁc area within the network. The cell con-

tains Base Station (BS) that response to do commu-

nications with mobile users reside in the cell. Sev-

eral cells are grouped together belong to Routing Area

(RA). Consequently, the network consists a set of RA.

Mobile user being at the boundary of either cell or

RA and going to different one, the hand-off occurs

and the connection in some cases will be lost because

there are no resources to handle the mobile user at

new serving area. In contrast, if the resources are

enough at the new serving area often the connection

lose because the time when mobile user send a re-

quest message for re-located and be in the new area is

not enough to ﬁnish the hand-off procedure, specially

that happen at RA re-located. Finally, if the connec-

tion does not lose during the Hand-off, a service does

not deliver on the time to mobile users.

If the network has enough information about mo-

bile user and neighboured, appropriate artiﬁcial intel-

ligent systems are employed. These help the network

to predict the next displacement for mobile user with

high accuracy, then sensible resource will be saved,

delay time for delivering the services will decrease

and improve the network functionality such as pag-

ing, location update and Hand-off.

The EACORA is proposed in this paper to im-

prove the mobility prediction for Location-Based Ser-

vices, mobile user’s displacement is achieved by the

developed ant colony. EACORA works on the RA,

that means every RA classify as independent colony

and control their own. Variables pass through them

because each one of them needs to know the visibility

of his neighbours.

The main contribution of this paper targets the

LBSs cost by deploying a prediction technique that al-

419

Sh. Daoud M., Ayesh A., Al-Fayoumi M. and Hopgood A..

An Enhanced Ant Colony Optimization for Routing Area Mobility Prediction over Cellular Communications Network.

DOI: 10.5220/0004201004190424

In Proceedings of the 5th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2013), pages 419-424

ISBN: 978-989-8565-39-6

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

lows intelligent LBSs disclosing and hence minimizes

the computation cost, consumption of resources, re-

duce the message passing and the overall cost of the

location management process such as location update.

EACORA scheme utilizes geometrical and topologi-

cal techniques allowing users to receive desired ser-

vices timely fashion.

The rest of the paper is organized as follows: Sec-

tion 2 discusses the previous work on mobility predic-

tion for LBSs and their limitations are described. The

proposed technique is introduced in section 3 and its

simulation model and result analysis is presented in

section 4. Finally, the conclusion and future work is

presented in sections 5.

2 PREVIOUS WORKS

Locating users as they move from one place to another

in a mobile computing environment is the key to pro-

viding continuous services with unrestricted mobility.

Therefore, the data management in this environment

is especially challenging for the need to process infor-

mation on the move, to cope with resource limitations,

and to deal with heterogeneity. One of the applica-

tions of mobile data management is LBSs which have

been identiﬁed as one of the most promising areas of

research and development (Barbar, 1999).

In the cell technique (Das and Sen, 1999; J. Bi-

esterfeld and Jobmann, 1997; Kubach, 2000; Kumar

and Venkataram, 2002; Shah and Nahrstedt, 2002; U

and Rothermel, 2000) a service area is partitioned into

several cells. The cell covering the mobile user will

page his or her device to establish a radio link in order

to track changes in the location of mobile users.

The cells broadcast their identities and the mobile

user periodically listens to cell identity and compares

it with the cell identity stored in its buffer. If the com-

parison indicates that the location has been changed

then the mobile user sends a location update message

to the network (Holma and Toskala, 2001).

Prediction techniques based on a cell technique

can be enhanced by heuristic methods and neural net-

works (Lu, 2003; Capka and Boutaba, 2004). Liou

and Lu (Lu, 2003) divided the cell into two areas,

edge and non-edge. The edge areas have neighbour-

ing cells, while the remaining areas are considered

as non-edge areas. When the mobile user is in a

cell’s edge area, the information is passed to a neu-

ral network which predicts from the neighbouring

cells the next cell to be visited. Another technique

captures some of the mobile user activity and paths.

These paths are progressively recorded, giving a his-

tory record which is used as an input to a neural net-

work to predict the next cell to be visited (Capka and

Boutaba, 2004).

The ﬁrst Ant Colony Optimization (ACO) al-

gorithm, called Ant System (AS)(Dorigo, 1996;

Dorigoa and Gambardella, 1997; Dorigo et al., 2000;

Dorigo and Di Caro, 1999), Dorigo et al., in (Dorigo,

1996) proposed that to solve the Traveling Sales-

man Problem (TSP). They proposed a new model

to combinative stochastic optimization, based on the

ants’ behavior, it is inspired from (S.Goss, 1990;

Deneubourg and Goss, 1989; Deneubourg et al.,

1983). This model is useful when use in greedy

heuristic to ﬁnd acceptable results at the early pro-

cessing, complex system which need to use the dis-

tributed computational to deal with the random space

variable, that mean avoiding premature convergence.

The mobility prediction technique in (Daoui et al.,

2008) uses the ﬁrst version of ACO, which the cell di-

viding has not used, stagnation of search is addressed,

and consuming computation cost because there was

no limitation for pheromone and the authors did not

use the modiﬁcations of ACO avoided.

Recent research ACO focuses on premature con-

vergence of the pheromone that the search concen-

trates at early state of search, which negatively affects

on the performance of ACO. It will lead to premature

stagnation of the search. Search stagnation is pro-

posed in (Dorigo, 1996) as the situation where all ants

follow the same path which is generated by other ants

and construct the same path over and over again, in a

sense, there are no new paths will be found anymore.

Map matching algorithm has been used for mobil-

ity prediction. Ren and Karimi developed the map

matching algorithm through using other techniques

such as Markov chain, hidden Markov fuzzy logic

to improve the mobility prediction for wheelchair.

In (Ren and Karimi, 2009a), the map matching al-

gorithm has been developed through its dependence

on the Markov chain and GPS sensor. The distance

and the direction between points which are recorded

by GPS are used. Prediction of the direction of

wheelchair users in sidewalk is considered as the out-

come from (Ren and Karimi, 2009a; Ren and Karimi,

2009b; Ren and Karimi, 2011).

The map matching techniques area proposed in

(Ren and Karimi, 2011; Ren and Karimi, 2009b;

Ren and Karimi, 2009a) suffered from many major

drawbacks. All of them are tested and evaluated for

wheelchair only in university camps sidewalk and it

works in outdoor only. These techniques are based

on GPS navigators. Therefore, anyone who needs to

use them must have GPS sensors. However, the GPS

sensors lead to extra physical cost bearing in mind

that they may not be applicable for all mobile devices.

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

420

Moreover, GPS suffers from inaccurate data in narrow

roads, high building and is believed to use higher-end

GPS relievers to improve the signal, instead of low-

end.

A new Splitting-based Displacement Predic-

tion Approach for Location-Based Services (SDPA)

(Daoud et al., 2011) has been developed to improve

prediction rate, minimizes consumption of resources,

and the overall cost of the location management pro-

cess comparing with PLM. Also, the SDPA reduces

the service area and the number of predicted routes

during the mobile user trip, by dividing the cell

into eight equivalent regions. Thus, the SDPA ap-

proach improved the location prediction probability

over PLM. The average complexity requirements for

usage space are smaller than for the PLM approach.

In addition, these techniques still work on cell level

which the cost of messages passing and executions

time are highly, because the SDPA and PLM executed

in tightly time slot.

3 EACORA TECHNIQUE

This section presents a EACORA. This technique is

based on a third generation mobile network, such as

UMTS.

3.1 EACORA Principles

The EACORA based on the responsibility to the RA

component instead of using the mobile user or cell.

This avoids the computation power required at mo-

bile users, which could be prettier, since power and

resource limitations are obstacles for mobile manu-

factures.

Figure 1: Routing Area Coverage.

The SGSN is managing the RA, each RA contains

one or more cells based on the radio speciﬁcations and

geographical features, as shown in ﬁgure 1.

The SGSN is responsible for managing and up-

dating the history displacements for all mobile users

which are residing in. Moreover, it handles the pre-

diction model to predict the next displacement for the

mobile user according to the current location, history

displacements and visibility to surround neighbors.

When a mobile user enters network, the SGSN

uses his and neighbors’ histories to make a relation

between them. Thus, improving the prediction per-

centage and handling any unusual movement. In con-

trast, if SGSN does not contain the history displace-

ments for the mobile user, it should use the history of

his neighbors.

Each RA is modeled by an ant colony and each

mobile user is modelled by an ant. An ant goes from

current RA to neighboring RA looking for food. In

the food searching, the ant prefers to go through the

usual paths or according to the displacement of his

neighbors.

3.2 EACORA Prediction

When a registration to the network is made for a mo-

bile user, the SGSN creates an ant to represent the

mobile user. Whilst, moving the ant will deposit a

pheromone on RA, this would be considered as the

communication channel between all ants in the cellu-

lar communications network.

At the ﬁrst entrance of mobile users to the net-

work, no pheromone would be found from any neigh-

bour, that’s why the movement goes randomly.

Over the time, each RA has its pheromone which

guides mobile users to the most preferable RA for the

future displacement.

Finding the probability of each RA, the previous

mobile user’s visibilities and the intensities of the

pheromones for all adjacent neighbours are required.

Suppose Ph is a vector of pheromone from 1 to A,

where A stands for the number of adjacent RAs. The

probability for the mobile user from current RA C

th to j RA expressed in equation 1.

RAi, j

(t) =

[τ

RAi j

(t)]

∗ [V

allC

i j

(t)]

∑

u∈Ph

(i)

[τ

(t)]

∗ [V

all

(t)]

(1)

Where P

RAi, j

the probability that the mobile user

at RA i at time t to RA j , t is the time factor,

τ the pheromone level and V

all

is the visibility -

memorization- of the mobile user. The visibility here,

all

, is obtained from the combination between local

and global visibility, accodring to euqation 2.

all

= P ∗ V

+ (1 − P) ∗ V

(2)

Where P between 0 and 1, V

is the local visibility

and V

is the global visibility.

Memorization entity is used to calculate the visi-

bility variable (V), it is represented by a vector (n) and

its length based on the number of adjacent RAs A

An element of this vector either local or global repre-

sents the ant visibility of an adjacent RA. In a sense,

AnEnhancedAntColonyOptimizationforRoutingAreaMobilityPredictionoverCellularCommunicationsNetwork

421

the local memorization reﬂects the mobile user’s be-

haviour. Where, global memorization reveals all mo-

bile users’ behaviours at such RA. The local visibility

is managed by equation 3.



X +1 if the mobile user exists in Nr;

X = 1 if mobile user does not exist in Nr;

(3)

Where V

is a local visibility, L between 1 and

, X is a value starts from 1 and Nr is a local mem-

orization table which stores the adjacent RAs. Mean-

while, the global visibility is represented by equation



Y + 1 if the mobile user exists in Mr ;

Y = 1 if mobile user does not exist in Mr;

(4)

is a global visibility, G between 1 and A

for

each mobile user, Y start from 1, Mr is a global mem-

orization table which stores the adjacent RA, where

each mobile affects on this equation.

When a hard hand-off occurs for a mobile user,

the mobile user changes the RA to another one. At

this time, the mobile user deposits his pheromone

on the RA which has been just left. The amount of

pheromones is deposited on each RA represented by

equation 5.

∆τ

i, j



λ ∗ Q ∗ τ

staying in

if λ ∗ Q ∗ τ

staying in

< Q

Q if λ ∗ Q ∗ τ

staying in

>= Q

(5)

Where ∆ τ

i, j

is the pheromone quantity that would

be laid down on the RA by the mobile user when he

left RA i to RA j. Q is a constant which represents the

maximum quantity of pheromone that would be laid

on each RA. The value of Q is greater than zero > 0.

staying in

is the time that has been spent by the mobile

user in RA i, λ is a constant fraction which value is

0 < λ < 1. λ is used to prevent the pheromone amount

that has been laid from exceed the Q value since this

amount proportional increases over time.

When the value of ∆ τ

i, j

is less than Q, the mobile

user’s pheromone affects the pheromones that held by

RA in a proportional to the time spent in that RA. If

∆ τ

i, j

greater than or equal to Q, the mobile user is

spending very long time in the RA, that means the

mobile user is working or living there, this leads to

have pheromone quantity greater than Q, therefore the

quantity will lay down is all Q, to avoid the bias of the

quantity that may laid down and stagnation of search.

In EACORA after δT the evaporation process will

take a place to decrease the pheromone level at each

RA in the network, this is represented by equation 6.

acc RA

(t + 1) = τ

acc RA

(t) ∗ (1 − ρ) (6)

Where τ

acc RA

(t) represents the accumulative

pheromones in RA i, (1 − ρ) is the evaporation rate.

A small value of ρ carrying out to pheromone

evaporation slowly and the pheromone will accumu-

late more on a RA. A large value of ρ leads to forget

the behaviour of other mobile users and the prediction

turns into random way. In a case when ρ is equal 1 the

prediction becomes completely random. The value of

ρ affects on the prediction rate by permitting to forget

the behaviour of the elder mobile users and to remove

the bias mobile users’ behaviour.

Probabilities of all RAs that surround the RA

where the mobile user resides in are calculated in

equation 1, the highest probability would be taking in

consideration as next RA that mobile user will visit.

Hence, the next displacement is expressed in equation

= max(P

RAi, j

(t)) (7)

Where Next

is the next displacement.

4 DISCUSSION OF SIMULATION

AND RESULTS ANALYSIS

4.1 Parameter Setup and Environment

A simulator was created using Java programming lan-

guage for the EACORA, in which the algorithm based

on The developed ant colony model is implemented

and tested. Each experiment consisted of 10 different

iterations to improve accuracy. Each experiment took

ﬁve hours, as shown in table 1.

Table 1: Simulation Parameters.

Parameter Value

Number of cells 100

Cell radius 250 m

Transmission Rate 8 Mbps

Simulation time 18000 s

Iterations 10

Pause time 20 s

Velocity of UE

Slow Pedestrian 5.6 k/h

Fast Pedestrian 11.2 k/h

Slow Vehicle 44.8 k/h

Fast Vehicle 89.6 k/h

4.2 Experiments and Result Analysis

The experiments in this section made to obtain the op-

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

422

timal value for each parameter; the parameters were

tested to evaluate the RA mobility prediction tech-

nique, which include: Parameter-1:the factor which

affects on evaporation rate. Parameter-2: quantity

of the initial pheromone that would be laid on each

RA. Parameter-3: calculating the participation ra-

tio of local visibility compared with global visibility.

Parameters-4: determining the effectiveness ratio be-

tween pheromone and visibility, which represented by

α and β.

For parameter 1, the prediction rate is tested over

varied evaporation rates. Figure 2 shows the predic-

tion rate percentage over the change on factor ρ. The

experiments were carried out on a range of ρ between

[0, 1], where its increment was moved up by 0.1.

From the experiments depict the optimal values of ρ

were between 0.6 and 0.8 where the highest predic-

tion rates were achieved over the change on ρ. These

values encouraged the RA developed mobility predic-

tion technique to avoid search stagnation, ﬁnding new

solutions and prevent the deletion of any new solu-

tion.

Figure 2: Prediction rate according to the different values

of ρ.

Figure 3: Prediction rate over varied initial pheromone

quantities.

For parameter 2, the prediction rate over varied

initial pheromone quantities is examined. Figure 3

describes the initial pheromone quantity that would

be laid down on each RA before starting the run-

ning of the technique. The prediction rates between

88 and 53 over varied amounts of pheromone quan-

tity. The highest prediction rate was 88 when the

amounts of pheromone setup to 1, as a result the best

amount of pheromone to be laid down was 1 unit. The

use of small amount of initial pheromone would save

computation cost. For showing the effectiveness of

pheromone quantity that will lay down during the run-

ning of the algorithm the next parameter is addressed.

For parameter 3, the signiﬁcant combination of

local and global visibility for the mobile user is vali-

dated. Figure 4 shows the prediction rate according to

the changing between local and global visibility par-

ticipations. Based on equation 2 two values should be

avoided in order to utilise these two concepts, these

values are 0 and 1.

Figure 4: Prediction rate according to the changing values

of P.

When the value of P equals to zero, the local vis-

ibility would be eliminated that means the RA will

use the neighbours’ heuristic information to predict

the next displacement for a mobile user. Thus the mo-

bile user cannot visit any of its favourites RA. In other

words, the mobile user displacements would be pre-

dicted randomly.

To achieve balancing between local and global

visibility, P was set to 0.6. This value guarantee a

full participation of both local and global visibility, as

well as better prediction rate would be obtained, see

ﬁgure 4.

For parameter 4, the prediction rate is tested over

varied distance between alpha and beta. Figure 5

shows the effect that resulted from varying the val-

ues of alpha and beta in order to determine the best

values to gain the highest prediction rate. As shown

in ﬁgure 5, it was noticed that when both Alpha and

Beta were equalled the prediction rate was highest.

Therefore, both Alpha and Beta are going to be cho-

sen according to the results, meaning that the values

of Alpha should be chosen to be equal to values of

Beta to achieve better prediction rates.

5 CONCLUSIONS AND FUTURE

WORK

This paper introduced new routing area displacement

prediction technique for location based services. This

technique is based on developed ant colony Optimiza-

AnEnhancedAntColonyOptimizationforRoutingAreaMobilityPredictionoverCellularCommunicationsNetwork

423

Figure 5: Prediction rate according to the varied Alpha-

Beta.

tion which achieved a high prediction rate up to 88%.

In addition, this paper has obtained the optimal values

for all the parameters that improve the prediction rate

and reduce the complexity. More work needs to be

carried out, for example the different number of mo-

bile users effect on using the new technique have to

be tested. In addition, the effect of the mobile users’

history, complexity time and memory usage with the

current techniques.

REFERENCES

Barbar, D. (1999). Mobile computing and databases-a sur-

vey. IEEE Transactions on Knowledge and Data En-

gineering, 11:108–117.

Capka, J. and Boutaba, R. (2004). Mobility Prediction in

Wireless Networks Using Neural Networks, volume

3271/2004. Springer Berlin / Heidelberg.

Daoud, M., Ayesh, A., Hopgood, A., and Al-Fayoumi, M.

(2011). A new splitting-based displacement predic-

tion approach for location-based services. In Systems,

Man, and Cybernetics (SMC), 2011 IEEE Interna-

tional Conference on, pages 392 –397.

Daoui, M., M’zoughi, A., Lalam, M., Belkadi, M., and

Aoudjit, R. (2008). Mobility prediction based on an

ant system. Comput. Commun., 31(14):3090–3097.

Das, S. and Sen, S. (1999). Adaptive location prediction

strategies based on a hierarchical network model in a

cellular mobile environment. In The Computer Jour-

nal, volume 42, pages 473–486.

Deneubourg, J.-L. and Goss, S. (1989). Collective patterns

and decision making. Ethology, Ecology and Evolu-

tion, 1(4):295–311.

Deneubourg, J.-L., Pasteels, J. M., and Verhaeghe, J. C.

(1983). Probabilistic behaviour in ants: A strategy of

errors? Journal of Theoretical Biology, 105:259–271.

Dorigo, M., Bonabeau, E., and Theraulaz, G. (2000). Ant

algorithms and stigmergy. Future Gener. Comput.

Syst., 16(9):851–871.

Dorigo, M. and Di Caro, G. (1999). The ant colony op-

timization meta-heuristic. McGraw-Hill Ltd., UK,

Maidenhead, UK, England.

Dorigo, M.; Maniezzo, V. C. A. (1996). Ant system: op-

timization by a colony of cooperating agents. IEEE

Transactions on Systems, Man, and Cybernetics Part

B: Cybernetics, 26:29 – 41.

Dorigoa, M. and Gambardella, L. M. (1997). Ant colonies

for the travelling salesman problem. Biosystems,

43(2):73–81.

Holma, H. and Toskala, A. (2001). Wcdma for umts: Radio

access for third generation mobile communications. In

Transactions on Networking, volume 9, pages 790–

800, New York, NY, USA. John Wiley and Sons.

J. Biesterfeld, E. E. and Jobmann, K. (1997). Location pre-

diction in mobile networks with neural networks. In

the International Workshop on Applications of Neural

Networks to Telecommunications, pages 207–214.

Kubach, U. (2000). A map-based, context-aware

hoarding mechanism. In Berichtskolloquium des

Graduiertenkollegs Parallele und Verteilte Sys-

teme,University of Stuttgart, Germany.

Kumar, B. P. V. and Venkataram, P. (2002). Prediction-

based location management using multilayer neural

networks. In Journal of Indian institute of science,

pages 7–21.

Lu, S.-C. L. H.-C. (2003). Applied neural network for lo-

cation prediction and resources reservation scheme in

wireless networks. In International Conference on

Communication Technology Proceedings, 2003, ICCT

2003, IEEE, volume 2, pages 958 – 961.

Ren, M. and Karimi, H. (2011). A fuzzy logic map match-

ing for wheelchair navigation. GPS Solutions, 15:1–

10. 10.1007/s10291-011-0229-5.

Ren, M. and Karimi, H. A. (2009a). A chain-code-based

map matching algorithm for wheelchair navigation.

Transactions in GIS, 13(2):197–214.

Ren, M. and Karimi, H. A. (2009b). A hidden markov

model-based map-matching algorithm for wheelchair

navigation. The Journal of Navigation, 62(03):383–

395.

S.Goss, R.Beckers, J. S.-J. (1990). How trail laying and

trail following can solve foraging problems for ant

colonies. in Behavioural Mechanisms of Food Selec-

tion, R.N.Hughes ed., NATO-ASI Series, G20:0.

Shah, S. H. and Nahrstedt, K. (2002). Predictive location-

based qos routing in mobile ad hoc networks. In

IEEE International Conference on Communications

(ICC ’02), volume 2, pages 1022–1027, New York,

NY, USA.

U, K. and Rothermel, K. (2000). An adaptive, location-

aware hoarding mechanism. In Proceedings. ISCC

2000. Fifth IEEE Symposium on Computers and Com-

munications, pages 615–620, Antibes-Juan les Pins,

France.

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

424