Network Selection in Heterogeneous WirelessEnvironment: A Ranking Algorithm

Joydip Dhar Ravi Kiran S Yashwanth Reddy Kj dhar (iiitm. ac. ravikiran. s (iiitmi acn kyashwanth_reddy (yahoo.com

#Information Technology, IIITM GwaliorABV-IIITM, Morena Link road, Gwalior-4740 10

Abstract-Vertical Handoff decision algorithms facilitate the Quality of Service (QOS) requirements factors play aterminal to be connected to the network which suffice the crucial role in selection process. Since the decision is based onrequirements of the application. It ushers the terminal to be multiple criteria and the importance of each ofthese may varyconnected to the most optimum network amongst those in the according to the requirement of the user, this has been reducedpurview of the mobile terminal. The selection criterion involves to the problem of multi attribute decision making (MADM)multiple attributes. Although many ranking algorithms havebeen proposed, there is a need to lay a stress on the user waoithmthe presence of a weight matrlx. Several MADMrequirements and the terminal power consumption. In this paper algorithms have been proposed namely TOPSIS, SAW, AHP,we propose a ranking algorithm keeping in view the user GRE but many of these suffer from ranking abnormalities.preferences and the application priorities. Having a close Hence a ranking algorithm which can remove theseproximity with TOPSIS algorithm, the idea proposed overcomes abnormalities and can score high on efficiency is required. Inthe ranking abnormalities and scores high on efficiency. the next section we proceed to look at the related work. The

subsequent section defines the variables. Later we explain inbrief the algorithm; and in the last section we discuss about

4G technologies are at the verge of making its own way to the results obtained.usher the research community to provide optimum access of II. RELATED WORKnetwork with the availability of very high data rates and Janise and Fang [1] describe in brief the evolution of 4Gefficient way of transmission. Also the increase in access of and vertical handoff process throwing a light on policytechnologies like WLAN, Bluetooth, 3GPP cellular networks decision and the context transfer. Farooq Ban and Victoretc would make heterogeneous IP infrastructure the future of Leung [2] proposed a new model based on network selectioncommunication. Evolution of value added services and easy approach. The additional nodes proposed in the paper i.e., thenetwork access at any point of time creates a responsibility on authentication node, service announcement and data collectionthe part of network provider to see to that the terminal has the node facilitate better sharing of information among the homeaccess to the most optimum network at any point of time. network and other networks thus providing better serviceService Level Agreements (SLA's) amongst the networkproviders have further created a demand for a selection dlvr vrtentokWenhui Zhang [3] gives a description of fuzzy approach formechanism amongst those available. network selection process and also compares it with the

Traditional handoff decision algorithm (Horizontal hand application of SAW and TOPSIS approaches.off decision algorithms) where the terminal used to switch to Balasubramaniam et al. [11] also use the AUP method in theirthat mobile base station having a high signal strength are decision making process. Tansir Ahmed, Kyandoghereunable to meet the requirements of the emerging technologies Kyamakya and Markus Ludwig in [8] propose a new anand the multimodal devices. These devices have created a newrequirement where the user has access to multiple interfaces innov war netor selectinbae onAUP.Farooq Banl and Victor Leung in [6] and [7] clearlyfor each application and also can select and switch to any of describe the TOPSIS process and its application in networkthe available based on the application requirements, the selection. In [6] the authors show how both compensatory andinterface capacities and the constraints of the mobile terminal. non-compensatory decision algorithms based on M\ADM easeThus the terminal has the facility of using multiple interfaces the process of ranking. In [7] iterative TOPSIS is definedsimultaneously which exists at that particular moment. This keeping in the view the ranking abnormalities of 1\ADMnew need has compelled the service provider to avert form thetraditional algorithms and move to Vertical Handoff Decision approaches and illustrated with an example.

Enrique, Navarro and Vincent in [8] compare and contrastThi_ the available MIADM approaches. Enrique, Navarro, Wongaccess to the most optimum network in the region of overlay anYuiLnn[9prosamdebsdonM kv

networks. Thus the user need not bear the extra cost and so dcso rcs Temdlmksueo rbblt ndoes the network provider the extra burden on his/her part. nrdcst ikrwr ucin

978-1 -4244-1 878-7/07/$25.OO 2007 IEEE 41

III. DECISIONM\AKING CRITERIA each application varies, it is essential to consider only thoseThe attributes used for the selection algorithm use the networks for the second phase which have substantial data

same criteria that have been used in [5] for TOPSIS procedure. rates that can support the type of application. Also, as theHowever there was a need felt for considering another criteria battery power and the processor speed play an important rolefor the selection process. The attributes used are below. for the network to be selected they are compared with theCost Per Byte: This the amount charged by the network signal available networks at that moment to eliminate thoseoperator to which the user is subscribed or the operator whose from consideration which are far less compatible with theservices are in use. (Includes the roaming arrangements). terminal power consumptionTotal Bandwidth: It is the overall bandwidth available over

Ip;rcAt[on PrcfiIethe link accessed by the mobile terminal. It is measured interms ofbps (bits per second). proibssor

Available Bandwidth: This is the measure of per user Battery p=,_erbandwidth allotted by the network operator which is dynamic ;T aernob:ic terhlhal ha:s ;Wcess to r'

. .,., . .,^ , , . r E~~~~~~~~~oijle networks thlus there ne-ed1s tea be aand can change as per the utilization of the network. It is cutoff &1tHafdrorselectionmeasured in terms of bps (bits per second).Delay: The packet delay gives the measure in milliseconds of Ph3ethe average delay for the packet on the wireless link. Fig 2: Elimination Phase for the networks.Jitter: It is the delay variations (average) on the wireless linkof the network. It is measured in milliseconds. The application profile contains the type of application, thePacket Loss: It is the measure of average packet loss between requirements of the application and the battery power givesthe mobile terminal and the network operator. It is measured the current capacity of the terminal in terms of power usagefor every 1 million packets sent. and can be used to set the cut off limit for signal strengths (inSignal Strength: It is the measure of signal strength available fact we set the lower bound based on the battery power at theto the mobile terminal from the base station of the network. moment of the terminal and also under the assumption thatThis was the basis of traditional handoff procedures. It is maximum battery power will be used).measured in terms of dB-micro volts per meter (dBptV/m)Utilization: It is the percentage of utilization of the wireless B. Second Phase.link of the network used. It consists mainly oftwo steps

Terminal Power Consumption is one of the vital criteria to (i) Obtain the weight matrixbe considered for access of the network. For a mobile terminal (ii) Apply the ranking algorithm.having a low battery power, the optimum network must not 1) Obtaining the weight matrix. The total sevenonly satisfy the QOS factors but also should give due variables (AB TB U C D L J) are classified under threeimportance to the battery power available at that point of time. categories speed, cost and quality which are easily understoodFor a mobile with low batter power, the network with high by the user. The user then gives the input for the threesignal strength, moderate quality and cost may prove to be variables for each application in terms of linguistic variablesmore efficient than the network with best quality, low cost but (each variable having a value assigned to it) (Table givenlow signal strength. below)

TABLE IIV. TUE PROCESS LINGUISTIC VARIABLES WITH THEIR VALUES

Elicitthe user requirements. Linguistic xariabie Value assigned

V -ry high 9

H igh 7

Referto the rule base and obtain the 7x7 |ediurr}relative matrix.

VeryI1c) 1-

Obtain weight m-atrix. AH-P

The user is allowed to specify the priority as between two ofApply ranking algorithmi the adjacent linguistic variables given.

Then we have a set of heuristics predetermined byrequirement analysis of resources for each application which

Fig 1: Process flow ofthe algorithm give the relative priority between the two sets (AB, TB, U)and (D, L, J).These sets of heuristics are to be maintained for

A. First Phase. each application .Then the values of variables are obtained byIn this phase the available networks are short listed based multiplying user input and the ratio obtained from the

on the applications they support for the mobile terminal i.e., heuristics. With these values a 7x7 relative matrix is made andthe type of application (real time application, data services etc) then the eigen values and vectors are computed. Now theand the signal strength of the network. Since the data rate for eigen vector corresponding to the largest eigen value is taken

42

and normalized. Then the transpose of that eigen vector is iii. Normalize the decision matrix.taken and then converted into a diagonal matrix. This matrix (Ta r (A 1-1(C )i-)' ( )i (-)is our weight matrix. TB ) (AB--2) (J)2(C.2 D-< .2 (L )

ab/ab tb'ab u afib dab dfab liab jab (TB a (AB 3 (L 3 (Cn)3 (D- )3 (J >3 (L)ab/tb tb/tb u/tb ctb d/tb 1/tb j/tbo D=ab/u tb/u u/ut /u d/u 1/u j/u

RM = ab/c tb/c u/c c/c d/c I/c I/cab/d tb/d I/d c/d did l/'d j/d (TB- (AB- ) (C(-)J I- _

b"I tb/ 1 u/l s/1 d/1 11/ jIlabj tbj ij cj dj Ij jj iv. Obtain the weighted normalize decision matrix.

Where RM=relative matrix; ab, tb, u, c, d, 1, j are the WVN = W * Dvalues ofAB TB U D L J respectively.

(T )2 (ABt)2 (LT)2 (Cwn)22Dwn)2 (J- )2 (L-)2evl s 0 0 0 ti 0 ; ; :) s s .>ev2 0 V2 0 0 Qi 0 0 -=ev 0 0 eV3 0U 0 0

EigenValues of RIM = Weight matri=0 0 0 0 0 0. ev0tS) 0 0 (,> )E(AB-) (U (Cm> (P)-> (J1 (L-)-_

. 0~0 0 0 ev0d&7 0 & v. Obtain the median vector by finding the median of

each of the column in the normalized weightedWhen two attributes are compared they would be given values decision matrix.from the following table. - = : 1M M. mc m. M2 MI]

Score Rule vi. Now, obtain the delta matrix as defined in theI Equal preference to both following way.

3 One is moderately preferred over other (IT 11(- . ) -IMiu-(Ul ) Me-(C Mn-(D )i -(b)i IL-(L i)5 One is strongly preferred over the other (B.)i-(6 M6-(U )2 Mc- .)i NLIDu) g N -

7 Very strong preference over the other B (AB.)i46 ML(IJ)] MI-(C )h-(D )3 NX-(J| Xk-(L49 Etrere preference over the ot er

2) The ranking algorithm: -M, (. -. I, -(U ) M -(C 1 - , -(1X) -fL

i. Obtain the weight vector as described above and thusget the diagonal weight matrix. . ) . A

(AT11, 2r AA)::2A (A02): Ac> (A> ):2 ( L)Wm 0 0 0 0 0 0 A

00 0 0 0 0o o a a I . ...1

W= 0 0 0 NV 0 0 0 (Am>) (A9a)> (A7U)> (A2>E(A> ) (AL,

o a 0 a O o0 0 0 0 0 WI vii. Obtain the resultant matrix for consideration of ranksO 0 0 0 0 0 iN71L by summing up the values of the row elements.

ii. From the networks within the reach of the mobile | (N ( ( ( ( (ALterminal (after phase 1) obtain the decision matrix. ( (AY> + Oc)t + (rt + (4V+ (A|

TBi -AB IUl Ci Di Ji LIRT12 A2 U 2 C D2 J2 L2TB3 A] U3 C3ID3 f LI

TB~AB~L~ C~ D|J~~~ L~~~~ order give the resultant ranks of the networks.

43

V. SIMULATION and RESULTS order of each of the column does not change the value in theIn order to analyse the algorithm and the effectiveness we rank matrix will also not change since it is the summation of

have taken 5 networks [6] with the attributes values as given. the row elements. Thus the ranking will remain the same.The results and ranks are as depicted in the following figure. Hence, the algorithm is consistent.

The decision matrix (DM) is

VI. CONCLUSIONAB TB U C D L J(per1l )

(mhps) (mbs) |(%) (%) (MS} (mS)Net vorkl 0.2 10 100 400 50 100 M\ADM algorithms are very well known and have beenNeT 0 20 20 25 2 applied in many research areas. Coming to the application of802.l1b these algorithms for network selection, it has been found thatDM =-Ntworl3 2 54 2( 10 100 15 01 TOPSIS is one of the most preferred one compared to others.862.1 la

Netwotk4 5 100 40 5 150 30 20 It is proved that TOPSIS can change when a candidate at the8t02-rk5 5 100 20 30 100 2t 15 bottom of the ranking is removed from the comparison, and4G _______ _______ ______ _____ ______ ______ _______ the iterative approach gives a more consistent and accurate

final result. Terminal Power Consumption and User0.0274 0 0 0 0 0 0 preferences are not given due importance of the available

0 0o0196 0 0 0 0 0 procedures. The algorithm proposed overcomes the above0 0 00.152 0 0 0 0 abnormalities and scores high on efficiency by consuming less

Wegtnatrix= 0 0 0 0.1918 0 0 0[Veightniatrix 00 0 0 031970 0 0 memory and at an optimum processor speed.0 0 0 O 0).3197 0 0

0 0 0 0 0 0.1066 0 REFERENCES0J 0 0 0 0 0 0J.3197

- [1] Janise Mc Nair and Fang Zhu. Vertical Handoff infourthThe ranks of the networks are generation multi-network environments IEEE Wireless

Communications June 2004.Ntwk#,1 Ntwk#,2 NtwkU3 Ntw&M4 Ntwkm'; [2] Farooq Bari and Victor Leung. Service Delivery over

Heterogeneous Wireless Systems. Networks Selection-0.5974 -0.0429 0.08al1 0.02 97 0.0536 Aspects. IWCMC'06, July 3-6, 2006, Vancouver BritishRan k-5 Fan k-4 Rank- I Ran k-3 Rank-2 Columbia, Canada.

[3] Wenhui Zhang. Handover Decision Using FuzzyMADMThis method has also been tested and found not to have the in Heterogeneous Networks. IEEE Communicationsdrawback of TOPSIS when the process is done iteratively. Society WCNC 2004.

[4] Dr. Rainer Haas and Dr.Oliver Meixner. An IllustratediMiD.s Ntk#1 Ntk#2 Ntk#3 Ntkt4 Ntk45 Guide to AHP, University OfNatural Resources and1 -0.5974 -0_0429 0I081i 0.0297 0_0536 Applied Life Sciences, Vienna.

Rank-s Rank-4 Rank-. Rank-3 Rank-2 [5] Felix Naumann. Data Fusion and Data Quality; Berlin-2_-0.1558 0 :50 -TO1145 _0=o11 Brandenburg Graduate School in Distributed Information

___---_ Rank-4 Rank-i Rank-3 Rank-2- Systems (DFG grant no.GRK 316), 1998.3 -0.0027 -0.1359 _0 1198 [6] Farooq Bari and Victor Leung. AutomatedNetwork Select-

Rank-I. Rank-3 Rak-2 -ion in a Heterogeneous Wireless Network Environment.4 0.0603 -0.0603 IEEE Network January/February. 2007.

___-- ---------R--nk-i ___-_ k-2 [7] Farooq Bari and Victor Leung. Multi-Attribute NetworkSelection by Iterative TOPSISfor HeterogeneousWireless Access.2007 IEEE.

MIADM algorithm is said to be ideal if the removal of the [8] Enrique Stevens-Navarro and Vincent W.S. Wong.bottom ranked candidate does not alter the ranking. However, Comparison between Vertical HandoffDecisionsTOPSIS algorithm did not meet this requirement and so Algorithmsfor Heterogeneous Wireless Networks.iterative TOPSIS was proposed in [5]. [9] Enrique Stevens-Navarro, Vincent W.S.Wong, and Yuxia

For each of the attributes namely Cost, Delay, Jitter, Loss Lin. A Vertical HandoffDecision Algorithm forand utilization the affect ofremoval of bottom ranked network Heterogeneous Wireless Networks.will either increase or decrease the median. In either of the [10] Tansir Ahmed, Kyandoghere Kyamakya and Markuscases the difference (Median - Attribute value of weighted Ludwig. Design and Implementation ofa Context-Awarenormalized matrix) would vary respectively. However as we Decision Algorithmfor Heterogeneous Networks.subtract from a common term the order for the each of column SAC'06, April, 23-27, 2006, Dijon, France.will remain the same (in the delta matrix). The same would be [11] 5. Balasubramaniam, J. Indulska, Handovers betweenthe case in case of the attributes Total bandwidth and available Heterogeneous Networks in Pervasive Systems, IEEEbandwidth (where the delta matrix attribute being Attribute Procon Communication Technology, April 2003.value of weighted normalized matrix- median). Now, as the

44