International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 792
ISSN 2229-5518
A Probabilistic Analysis of Network Selection Strategy for Heterogeneous Wireless Network A.Manokar, C.Amali
Abstract—Next generation wireless networks will integrate multiple radio access technologies (RAT) to provide seamless connectivity to mobile users. Unified network selection can be achieved through a vertical handover (VHO) in which connection can be handed over among different RAT. In the existing literature, different kinds of network selection algorithms have been proposed to select the optimal network in heterogeneous wireless environment. But, uncertainty associated with network selection process is not yet modeled. Usually dynamics of network selection occur due to diverse characteristics of networks and also dynamic behavior of mobile users. These dynamic factors cause the variation in the utility and price offered by the network. In this paper, the impact of variation in the utility and price on the network selection and transition probabilities is analyzed through Markov model. The network level quality of service (QoS) is evaluated using new call blocking and hand off call dropping probabilities in terms of steady state probabilities.
Index Terms— Heterogeneous wireless environment, quality of service (QoS), universal mobile telecommunications system (UMTS) , utility function, steady state probabilities , transition probabilities and wireless local area network (W LAN).
—————————— ——————————
The ultimate aim of the network selection algorithm is to select the best target network in order to ensure the service with better QoS at the lowest price. Thus the network selection behavior of user’s needs to be analyzed in terms of dynamic characteristics of networks and users. Dynamics of network selection in heterogeneous environment needs to be considered due to the following constraints.
Different applications demand diverse QoS requirements, which results in the dynamics of network selection. Hence, selecting the optimal network corresponding to the dynamics of user behavior may incur frequent handovers. The tradeoff is required between dynamics of user behavior and network selection process.
Dynamics of network selection may increase the handoff cost in terms of signaling overhead, energy and delay.
Due to diverse characteristics of channel conditions and network performance, network selection process is uncertain to the mobile users.
Thus the overall network selection dynamics can be described as a stochastic process as it changes over time in a probabilistic manner. Due to dynamics of the decision metrics, there is a need to take only the most relevant metrics into account to provide the best network dynamically to the user. In [1] multicriteria optimization technique was presented to provide complete solution for seamless connectivity in heterogeneous environment based on network
————————————————
• C.Amali is currently working towards Ph.D. degree with SRM
University, India, E-mail: amali.vec@gmail.com
conditions, application level QoS, mobile terminal (MT) battery level and user preference. But, mobility model was not considered. When the MT is moving with velocity, it is necessary to find out how much time the MT will stay in the target network. If the estimated residence time of the network is less than the threshold, it requires more number of handovers to complete the ongoing service. Residence time of the network is estimated using velocity and mobility pattern of the user. But, the probability of moving in/out of the network and uncertainty about the decision metrics was not addressed. VHO decision making based on time varying attributes always face the problem of uncertainty associated with the network selection. The impact of variation in decision metrics on the dynamics of network selection can be described through state transition probability using Markov model. The major contribution of the paper is
1) To study the impact of utility and price offered by the network on the probability of selecting the target network.
2) To derive the state transition and steady state network selection probabilities by considering the utility, mobility and service model.
3) To analyze the consistency level of network selection
and the network level QoS using the state transition model.
To attain network, service and application convergence, next generation networks must be designed in such a way that all the functions must work independent of characteristics of networks. Thus, the service providers can render their service to the users more efficiently with required throughput needed to support the fluctuating traffic demand generated by mobile users.
Amali et.al [1] proposed Multicriteria based network selection algorithm to provide solution for seamless application QoS and user preference. Racha Ben Ali et.al [2] presented the impact of VHO rate on the network level performance such as call blocking probability under different
IJSER © 2015 http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 793
ISSN 2229-5518
mobility scenarios. In [3] the performance of integration of cellular & WLAN and dynamics of network selection was analyzed based only on decision metrics. But, the concept of competition among users and network was not addressed for network selection process. Dusit Niyato et.al [4] presented the dynamic evolutionary game to model the dynamic of network selection and user mobility in heterogeneous environment. Reinforcement learning algorithm also proposed to learn the performance of different networks. Thus, it select best suitable network based in the QoS requirement of application. In practice, user discovers the networks with asymmetric prices and QoS. Uncertainty associated with decision making varies with time. In [5] Markov model is used to model the uncertainty present in the network selection and Bayesian price competition game is also proposed to investigate the impact of increased competition on the overall efficiency. In addition to the QoS
Fig.1. Heterogeneous wireless environment
related decision metrics used in conventional network
selection, users QoE is used in [6] to provide network
selection dynamic environment. In [7] provides the analysis
0
ζ
x − xl
x ≤ xl
of network selection problem associated with dynamic
m
− xl
xl ≤ x ≤ xm
(1)
demand, environment and dynamic network characteristics.
Online network selection algorithm is also proposed to offer
ζ
+ l
optimal network to the mobile user by providing
optimization between QoS and network handoff cost. In [8]
1
u( x ) =
x − x
m l
xm < x ≤ xu
mathematical model describing the optimization, complexity and performance in heterogeneous environment are discussed. In [9] Enhanced Media Independent Handover Function (MIHF) is proposed to provide complete
information about link layer and application layer in order to
ζ
u
x − x
1 − u m
x − x ζ
1 + u
x − x
maximize the end user satisfaction.
1
u m
x > xu
Consider the integration of complementary RAT such as Universal Mobile Telecommunications System (UMTS) based cellular network and IEEE 802.11 based wireless local area network (WLAN) as shown in Fig.1 to enjoy the diverse range
Where, x represents the decision metrics considered for
network evaluation. ζ determines the user sensitivity to the
variation of network characteristics. The overall utility function is calculated for each network using multiplicative
weighted utility function given by,
n wk
of services with better QoS, lowest price and seamless
roaming. Let’s consider WLAN as network 1 and UMTS as
ui = ∏k=1�uk (xk)�
i ∈ {1, 2} (2)
network 2 i.e. i ∈ {1, 2} . The method of modeling users
through utility, mobility and service model are considered in
the next subsections.
Considering the time varying characteristics of networks, different user requirements and mobility characteristics may select different networks. Thus, the networks in the coverage area have to be evaluated dynamically using appropriate utility function. Sigmoid based utility function is used to model the elasticity of applications and also to provide tradeoff between user demand and VHO rate. The degree of user sensitivity to QoS parameters for a particular service
Where, wk is the weight values assigned according to the
user preference and dynamic user demands. In practice,
network compete each other based on price and utility
offered by the network. Users pay minimum amount to connect with WLAN with no guaranteed QoS. But, cellular
network offers better QoS with high cost. Thus, the asymmetry between utility and price pose remarkable change on the probability of connecting to a network.
A probabilistic based mobility model is considered for the user moving across the integration of UMTS and WLAN. The probability of users present in the WLAN area is denoted
µ . Hence, 1 − Pµ corresponds to the probability of user
should be taken in to account in the utility calculation for
by Pr r
µ is the probability of
network selection mechanism. The QoS parameters that can be considered for utility calculation are data rate, delay, energy consumption and cost.
The elementary utilities of QoS parameters are calculated using sigmoidal function [1] given by (1) with threshold
value(xm ), lower (xl ) and upper(xU ) limits.
residing outside the WLAN coverage. Pout
user exiting the WLAN coverage. Under equilibrium
conditions, the average number of users entering in to the
WLAN area will be equal to the average number of users
exiting the WLAN area. Thus, the probability of a user
moving in to the WLAN area denoted by
µ µ
P
µ = Pr
out
Pin
µ
1−Pr
(3)
International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 794
ISSN 2229-5518
New calls arrive to the cellular network in a given time interval follow a Poisson distribution with mean arrival
rate λn. The call service time is exponentially distributed with
1
2) State 1: The user is outside the WLAN and connected with UMTS.
3) State 2: The user is inside the WLAN and connected
a mean of
by
. Hence the call completion probability is given
µ
with UMTS.
4) State 3: The user is inside the WLAN and connected
Pend = P(Tcall ≤ Tth ) (4)
Where, Tth is the time interval for the user state transition.
Considering exponentially distributed inter arrival time with
1
with WLAN.
a mean of
, the new call arrival probability is given byPnew.
λn
In the following analysis, it is assumed that utility and service
cost offered by the network and the probability associated
with user mobility and service model are known.
The acceptance probability or user satisfaction degree has been widely used in the network selection problem to analyze whether a user is convinced with the network selected using multicriteria based network selection algorithm. User satisfaction degree measures how much the user is satisfied with the perceived utility (u) and cost (c) offered by the network ‘i’. It can be modeled as [10]
Ai(u, c) = 1 − exp(−K uµ (ci)−ε ) (5)
Fig. 2. State transition diagram.
The state transition probability matrix is constructed using the above 4 state conditions and given by
P00 P01 P02 P03
Where, µ>0 and ϵ >0 control the sensitivity to utility and
P10
P11 P12 P13
cost respectively and K is a positive constant representing the
P = �
20
P21
P22
P23
� (8)
satisfaction reference value. Utility (U) is calculated using
attributes such as data rate, delay etc. In the following section
P30
P31 P32 P33
user preference is assumed to be equally sensitive to utility and service cost. The acceptance probability can indeed be used as a network selection probability in the presence of single network or idle state. However network transition from network i to j in the heterogeneous environment can be modeled using transition and steady state probabilities.
Where, P01 represents the state transition from state 0 to
state1 and also include three events. i) The user is outside the
WLAN. ii) Connected to UMTS. iii) A new call is originated.
Whenever new call arrives, the probability of connecting with
WLAN and UMTS are given by P03 , P02 and P01 using
equations (5), (6) & (8).
µ )P
(9)
P01 = S2 (1 − Pr
new
Network transition from i to j occurs only if ui ≥ uj
Similarly P02 and P03is given by
and Pi ≤ Pj . It ensures that the user always be connected to
P = (1 − S )S Pµ P
(10)
the network with more utility at lowest cost thus the
02 1
µ
2 r new
probability of selecting the suitable network should be equal
to the user satisfaction probability of the network, when the
user is in idle state’0’. It can be expressed as
P03 = S1 Pr Pnew (11)
The probability of remaining connected to the network is
given by
S0i = Si (6)
P = β S (1 − Pµ )(1 − P
) (12)
Where, Si= Ai (U, c)
11 2 2
r end
µ
P33 = β1 (1 − S2 )S1Pr (1 − Pend ) (13)
Then, the network transition occurs when the user is not
µ (1 − P
) (14)
satisfied with the connected network and also user mobility
P22 = β2 (1 − S1 )S2Pr
end
corresponding transition probability must be linear function
of network selection probability and handover cost. It is
obtained as
Sij = ϒij S0j (7)
Where, β1 and β2 are scaling constants.
There is a probability that a user moving into the WLAN
and connected with either WLAN or UMTS. The
corresponding probability are given by
µ (1 − P
) (15)
Where, ϒij is the handover cost.
P12 = (1 − S1 )S2 Pin
end
µ (1 − P
) (16)
For the given system model of Fig.1, user can be in one of
the following states as shown in Fig.2:
P13 = ϒ21 (1 − S2 )S1 Pin
end
1) State 0: The user is idle condition and there is no call in
progress.
There is a probability that a user is moving out of the
WLAN and connected with UMTS. It can be expressed as
P = P = β S Pµ (1 − P ) (17)
International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 795
ISSN 2229-5518
When the user is in the WLAN coverage, the state
transition probability are given by P23andP32.
P = ϒ (1 − S )S Pµ (1 − P ) (18)
P = ϒ (1 − S )S Pµ (1 − P ) (19)
Where, S1 and S2 represents the network selection
probability of WLAN and UMTS respectively. After
calculating all the Pij , we can obtain Pi0 ∀ i ∈{0, 1, 2, 3} using
the Markovian property that sum of element of each row of
transition matrix is unity.
Assume that N1 and N2 are the number of channels in the WLAN and cellular network respectively. n1 & n2 are the number of users in the WLAN and UMTS respectively. Nh is the channels reserved to prioritize the hand off calls from WLAN to cellular network. nt is the total number of calls/users in the system. For the given system model, cellular network accepts both new and hand off calls only if n1 < Nth. where, Nth = N2 – Nh . A new call is blocked by cellular network if at least Nth channels are already occupied.
It can be denoted by P2 .
B = ∑N2 � nt � (π + π )n2
(1 − (π1 + π2))nt−n2
(28)
3
j=0
Pij = 1 => Pi0 = 1 − ∑3
Pij
(20)
n2 =Nth n2
Steady state probability is defined as the long term
probability of being in a particular state. Given that πR i
represent the steady state probability of being in states i ∈{0,
The hand off call dropping probability of cellular network is the probability that all the N2 channels are occupied. New calls cannot occupy Nh channels but hand off calls can occupy
Nh channels. It can be denoted by PD.
1, 2, 3}. It can be expressed as a row vector π.
π = [π0 π1 π2 π3 ] (21)
P = � nt
N2
� (π1 + π2 )N2 (1 − (π1 + π2 ))nt−N2 (29)
Markov chain under consideration is time homogeneous. Thus steady state probability should sum to unity.
π0 + π1 +π2 + π3 = 1 (22)
A new call is block by WLAN only if N1 channels are connected. Since there are no channels reserved for handoff
calls. New call blocking probability is equal to the handoff call dropping probability.
B = PD = � nt � (π )N1 (1 − π )nt−N1 (30)
The steady state probability vector is the left eigenvector for the state transition probability matrix corresponding to
P1 1
N1 3 3
eigenvalue 1. It can be expressed as
πP = π (23)
P00 P01 P02 P03
In numerical analysis, network level parameters are
analyzed under the assumption that the total number of
calls/users in the system is more than the maximum number of channels available in the networks.
P10
P11 P12 P13
In this section, the performance of the proposed scheme is
[π0 π1 π2 π3 ] �
20
P21
P22
P23
� = [π0 π1 π2 π3 ]
analyzed numerically using MATLAB. In network selection
P30
P31 P32 P33
algorithm, utility offered by the network is calculated using
By solving the above matrix we can obtain following equation
P01 π0 + (P11 − 1)π1 + P21 π2 + P31 π3 = 0 (24)
the decision metrics such as data rate, delay and energy
conception. The best suitable network is selected based on the values of utility and cost of service according to the
application QoS and user performance. Parameters of simulation are as follows: N1 =20, N2 = 60, Nh = 6, nt =100,
P02 π0 + P12 π1 + (P22 − 1)π2 + P32 π3 = 0 (25)
Nth =54, P
= 0.5, Pµ
=0.1, P
=0.4, ϒ
= ε = µ = K = β = 1.
r out
new
𝑖𝑗
P03 π0 + P13 π1 + P23π2 + (P33 − 1)π3 = 0 (26)
The equations 22, 24, 25 and 26 can be constructed in
matrix form to obtain the steady state probability πi.
The utility of the network is calculated using (2) for the network parameters considered for the given system model. Dynamics of the network selection process occurs due to uncertainty associated with decision metrics in heterogeneous
1 1 1 1
π0 1
P01
(P11 − 1) P21 P31
π1 0
wireless environment. This kind of dynamicity is described
�P03
P12 (P22 − 1) P32
� �π2 � = �
� (27)
using steady state probabilities. To analyze the dynamics of
P03
0
P13 P23 (P33 − 1) π3 0
the network selection process, normalized values of utility and service cost are directly considered.
Thus, steady state probabilities are obtained in terms of
network selection and state transition probability.
In the previous sections, dynamics of user network selection behavior is described using probability based mobility model and then modeled using state transition diagram. In this section network level QoS metrics such as new call blocking and hand off call dropping probability are
analyzed using steady state probability πi.
To show the impact of utility and cost offered by the network on the steady state probability of connecting to network, utility and service cost of UMTS network is kept constant at u2 =0.8 and c2 =0.6. The steady state probability of connecting to WLAN is plotted with variation in the utility of the network for various values of service cost. It is observed that when u1 < u2 and c1 < c2, network selection probability of network is slightly higher than UMTS. An increase in the u1 or decrease in c1 causes an increase in the steady state probability of WLAN. When u1 < u2 & c1 > c2 UMTS dominates the WLAN network. Thus the probability of
International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 796
ISSN 2229-5518
selecting the UMTS is very much greater than WLAN. When
value of Nth can be adjusted dynamically in order to
D
u1 > u2 & c1 < c2 , steady state probability π3become high due
minimize the P2
. It is observed that dynamics of network
to domination of WLAN network as shown in Fig. 3.
0.25
selection affects the steady state probability of user. The
variation of steady state probability degrades the network
level QoS in heterogeneous environment.
c =0.1
1
c =0.3
1
0.12
c =0.3&u =0.8
0.2
c =0.6 1 1
1
=0.9&u =0.6
0.1
c1 1
0.15
0.08
0.1
0.06
0.05
0.04
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Utility obtained at WLAN(u )
1
0.02
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
µ )
Fig.3. Variation of steady state probability of state 3(𝜋3)
with utility (u1) obtained for various cost at WLAN.
Probability of user moving out of WLAN(Pout
D
In Fig.4 the variation of the new call blocking probability at UMTS with the probability of moving out of WLAN coverage is shown for various values of utility and service cost. Given c2 =0.6 and u2 =0.8 when u1 =0.6 and c1 =0.9 UMTS dominates WLAN. Thus, the probability of selecting the UMTS (S2 ) is higher than the network selection probability of
B
Fig.5. Variation of handoff call dropping probability (P2 )
with probability of user moving out of WLAN (Pµ ) obtained
for various cost and utility.
Network selection in heterogeneous wireless network always faces the problem of uncertainty associated with dynamic demands, environment and network selection. Thus,
WLAN (S1 ). This is the reason for an increase of P2
in the
the dynamics of network selection can be described as
UMTS. For the scenario with u1 =0.8 & c1 =0.3, WLAN becomes more suitable network than UMTS. This condition can reduce the number of users in the UMTS network and also new call blocking probability. It is inferred that dynamics of user behavior causes the dynamic network selections which in turn affects the network level QoS such as new call blocking probability.
0.12
=0.3&u =0.8
stochastic process using probability based Markov model.
The impact of variation in the utility and cost of service on
the network selection probabilities and network transition
probabilities is studied for the given system model. The overall network selection dynamics and the performance of
the networks are analyzed through steady state probability. Network level performance is also analyzed through blocking and dropping probabilities based on steady state probabilities and total number of user in the system. For future work, network level QoS such as call dropping probability will be
c1 1
c =0.9&u =0.6
analyzed in terms of user level QoS such as packet loss rate
0.1
1 1
(PLR) and delay.
0.08
0.06
0.04
0.02
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Probability of user moving out of WLAN(Pµ )
out
Fig. 4. Variation of new call blocking probability ( PB )
2
with probability of user moving out of WLAN ( Pµ )
obtained for various cost and utility.
In Fig.5 the variation pattern is same as that of PB except
2
[1] C.Amali, Dhanasree Jayaprakesh and B.Ramachandran, “Optimized network selection using aggregate utility function in heterogeneous wireless networks,” International Review on Computers and Software, vol. 9, no. 7 July 2014.
[2] R. B. Ali and S. Pierre, “On the impact of soft vertical handoff on
optimal voice admission control in PCF-based WLANs loosely coupled to 3G networks,” IEEE Trans. Wireless Commun., vol. 8, no.
3, pp. 1356–1365, Mar. 2009.
[3] W. Song, H. Jiang, and W. Zhuang, “Performance analysis of the
WLAN first scheme in cellular/WLAN interworking,” IEEE
Trans.Wireless Commun.,vol. 6, no. 5, pp. 1932–1952, May 2007.
[4] D. Niyato and E. Hossain, “Dynamics of networks selection in
heterogeneous wireless networks: An evolutionary game approach,” IEEE Transactions on Vehicular Technology, vol. 58, no. 4, pp. 2008-
2017, May 2009.
[5] A. Kumar, R. K. Mallik, and R. Schober, “WLAN service providers’
that the rate of increase of PB
is high compared to PD. This is
price competition with uncertainty in user demand,” IEEE
due to the fact that Nh channels are reserved for hand off calls in the UMTS. For the scenario with high hand off rate, the
Commun. Lett. vol. 17, no. 3, pp. 517–520, Mar. 2013.
International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 797
ISSN 2229-5518
[6] K. Piamrat, A. Ksentini, C. Viho, et al., “QoE-based network selection for multimedia users in IEEE 802.11 wireless networks,” in Proc. 2008 IEEE Local Computer Networks, pp. 388–394.
[7] Zhiyoung Du, Qihui Wu and Panlong Yang,”Dynamic user demand
driven online network selection,” IEEE Commun. Lett., vol. 18, no.
3, Mar. 2014.
[8] L. Wang and G. S. Kuo, “Mathematical modeling for network selection in heterogeneous wireless networks – A tutorial,” IEEE Commun. Surveys & Tutorials, vol. 15, no. 1, pp. 271–292, 2013.
[9] C.Amali, Dhanasree Jayaprakesh and B.Ramachandran, “Enhanced media independent network selection for heterogeneous wireless networks,” IETE Technical Review, vol. 31, no.5, pp.392-401, Oct
2014.
[10] L. Badia, M. Lindstrom, J. Zander, and M. Zorzi, “Demand and
pricing effects on the radio resource allocation of multimedia communication systems,” in Proc. 2003 IEEE GLOBECOM, San Francisco, CA, U.S.A.,pp. 4116–4121.