Cooperation-based resource allocation and call admission for wireless network operators

Abstract

The increasing number of radio access technologies and the availability of multi-radio devices boost the need for novel resource
allocation schemes in cellular networks. This paper uses a cooperative game theoretic approach for resource allocation at
the network level, while utilizing simultaneous use of available radio interfaces at the device level. We model resource allocation
management using the well known bankruptcy model and apply Kalai-Smorodinsky bargaining solution method to find a distribution rule, based on which we propose resource
allocation and call admission control schemes. Performance analysis of our allocation and control schemes demonstrates significant
improvements over previous approaches in terms of utilization of the available bandwidth and the number of call drops. We
also study the performance of proposed approach for different operator policies.

KeywordsWireless networks–Resource allocation–Game-theory

Full text

Page 1

Telecommun Syst
DOI 10.1007/s11235-010-9412-1
Cooperation-based resource allocation and call admission
for wireless network operators
Manzoor A. Khan ·Ahmet C. Toker ·Fikret Sivrikaya ·
Sahin Albayrak
© Springer Science+Business Media, LLC 2010
Abstract The increasing number of radio access technolo-
gies and the availability of multi-radio devices boost the
need for novel resource allocation schemes in cellular net-
works. This paper uses a cooperative game theoretic ap-
proach for resource allocation at the network level, while
utilizing simultaneous use of available radio interfaces at
the device level. We model resource allocation management
using the well known bankruptcy model and apply Kalai-
Smorodinsky bargaining solution method to find a distrib-
ution rule, based on which we propose resource allocation
and call admission control schemes. Performance analysis
of our allocation and control schemes demonstrates signif-
icant improvements over previous approaches in terms of
utilization of the available bandwidth and the number of call
drops. We also study the performance of proposed approach
for different operator policies.
Keywords Wireless networks · Resource allocation ·
Game-theory
1 Introduction
We observe an increasingly heterogeneous landscape of
wireless access technologies, including UMTS, LTE, WiFi,
WiMAX, etc. These technologies are specialized for differ-
ent environments and user contexts. The development as
well as the business cycles of these technologies can as-
sure us that they will be available simultaneously for the
M.A. Khan (?) · A.C. Toker · F. Sivrikaya · S. Albayrak
DAI-Labor, Technische Universität Berlin, Ernst-Reuter-Platz 7,
10585 Berlin, Germany
e-mail: ManzoorAhmed.Khan@dai-labor.de
years to come. Consequently, there has been significant re-
search activity on the integration and inter-operability of
these fundamentally different access technologies, which
exhibit different service characteristics in terms of band-
width, coverage, pricing, and QoS support. The initial con-
cern for network operators was increased connectivity by
providing diversified methods of access for different types
of end devices. However, the emergence of multi-interface
terminals has shifted the simple connectivity issue to more
rewarding resource allocation problems, whose solutions
aimed at increasing the network efficiency and capacity as
well as improving users’ experience for ample amount of
services such as video on demand, video conferencing, and
a variety of other applications.
Common Radio Resource Management (CRRM) [17] is
the concept that multiple such radio access technologies
(RAT) can be combined in an operator network to diver-
sify the service offer, as well as make use of trunking gains.
The CRRM problem, which involves the allocation of call
requests of different service types to the different Radio Ac-
cess Networks (RANs), has been approached mostly from
a single operator perspective, where different Radio Access
Technologies (RATs) are deployed as radio access networks
belonging to the same operator.
Within this framework, one can discern two approaches.
On the one hand it was shown by Fruskär et al. [8] that
the optimal policy to maximize the combined utilized band-
width on the RANs is to associate individual RANs to a cer-
tain service type that they support better than the others, and
then to mix traffic only when one RAN is full. We call this
approach the service-based approach. On the other hand, the
work of Tolli [23] concentrates on the trunking gain that is
obtainable by balancing the load between different RANs.
This approach relies on a periodic measurement of the load

Page 2

M.A. Khan et al.
situations on the RANs and allocating the service requests
to the RAN with the most capacity.
Although extensive research has been carried out on im-
proving vertical handovers and improving user Quality of
Experience (QoE) through service adaptation to suit the
characteristics of network interface, most of these research
confined to the use of a single network interface at any given
time to meet the requirements of applications (see e.g. [11,
18, 20, 22]). More recently, the possibility to use multiple
access technologies simultaneously and to split an applica-
tion’s resource requests among available RATs has been in-
vestigated. The “Ambient Networks” project [13] within the
EU Framework Program 6 (FP6) introduced a Generic Link
Layer (GLL), integrating different RATs at the link layer
for efficient interworking [5]. A byproduct of GLL is Multi-
Radio Transmission Diversity (MRTD), which allows split-
ting of data flow among multiple RATs. In [2] Bazzi et al.
also investigated the issue of using multiple RATs simulta-
neously at the terminal, concluding that parallel transmis-
sion over multiple RATs by a careful resource allocation
scheme allows one to achieve a throughput as high as the
sum of the throughput obtained by the use of each individ-
ual access technology.
As in many areas of the networking field, application of
game theory concepts to CRRM problem has been consid-
ered using both cooperative [15, 21] and non-cooperative/
competitive [4, 9, 14] game models to obtain efficient re-
source allocation schemes. Badia et al. provided a compari-
son between non-cooperative and cooperative models in re-
source allocation and demonstrated that collaborative strate-
gies are able to improve the overall system performance [1].
A bankruptcy game is used in [15] to model the problem,
but within a limiting scenario regarding the composition of
available access technologies. We provide a comparison in
Sect. 3 by applying our resource allocation to the scenario
considered [15].
In this paper, we address the issue of multi-radio resource
allocation in generic heterogeneous wireless networks us-
ing a cooperative game, where the network technologies
cooperate with each other to attain the ultimate goal of
user satisfaction. We use the Bankruptcy model and apply
Kalai-Smorodinsky Bargaining Solution (KSBS) to find the
distribution rule that best fits our objective of simultaneous
resourceallocationforchannelrequests.Wealsoprovideex-
tensions of our approach to handle the mobility of users be-
tween coverage areas with different composition of access
technologies.
2 Relevant background
We start by reviewing several basic definitions and concepts
that will be utilized in this work related to the bankruptcy
problem and bargaining solution of cooperative games. We
also provide a brief overview of different access technolo-
gies and their characteristics affecting our resource alloca-
tion problem.
2.1 Bankruptcy problem
Bankruptcy is a distribution problem, which involves the
allocation of a given amount of goods among a group of
agents, when this amount is insufficient to satisfy the de-
mands of all agents. The available quantity of the goods
to be divided is usually called estate and the agents are
called creditors. The question here is: How to distribute es-
tate amongst creditors? A number of distribution rules have
been proposed to deal with such problems. The solution to a
bankruptcy problem can be interpreted as the application of
an allocation rule that gives sensible distribution of estates
as a function of agents’ claims. Bankruptcy is a pair (E,C),
whereE representstheestatetobedistributedamongaset C
of the claims of n creditors, such that C = (c1,...,cn) ≥ 0
and 0 ≤ E ≤?n
In our formulation creditors represent the access net-
works and estate represents the required bandwidth by ap-
plications.
i=1ci. An allocation xiof the estate among
creditors should satisfy?n
i=1xi= E given that 0 ≤ xi≤ ci.
2.2 Bargaining solutions of cooperative games
Bargaining [10, 16] refers to the negotiation process (which
is modeled using game theory tools) to resolve the conflict
that occurs when there are more than one course of actions
for all the players in a situation, where players involved in
the games may try to resolve the conflict by committing
themselves voluntarily to a course of action that is benefi-
cial to all of them.
2.3 Kalai-Smorodinsky Bargaining Solution (KSBS)
Given a pair (S,d) that defines the general bargaining
problem, with S denoting the set of feasible utilities and
d representing the disagreement point, the unique Kalai-
Smorodinsky bargaining solution X∗= F(S,d) fulfills the
following axioms:
1. Individual rationality: Xi≥ difor all i.
2. Feasibility: X∗∈ S.
3. Pareto optimality: X∗should be Pareto optimal. A solu-
tion is Pareto optimal if it is not possible to find another
solution that leads to a strictly superior advantage for all
players simultaneously [19].
4. Translation invariance: ∀(S,d),∀h ∈ ?n: F(S + h,
d +h) = F(S,d)+h.

Page 3

Cooperation-based resource allocation and call admission for wireless network operators
Table 1 Notations and their
descriptions
Notation DescriptionNotation Description
N
Set of network technologies
Load balancing factor
Predefined offered bandwidth
Application class
User type
Uncongested network technology set in area a
Proportionate factor
Requested bandwidth of user k for service class c
Abstract coverage area
Disagreement point
Total capacity of network w
Current load on network w
Maximum utility
Operator per unit charged price
Sum of available bandwidths of all networks in area a νo
Oth
Operation threshold region
Set of operator policies
Network w available capacity
Potential-based policy
Hard-coded policy
Congested network tech. set
Network tech. utility function
Allocated bandwidth
Feasible allocation set
Congestion based policy
Bargaining solution
Decision boolean
Aggregated operator bandwidth
Operator per unit cost
Operator utility
ψM
˜ cw
πp
bk,c(π)
c
k
Wa
πh
Wa
βuw,k,c
rk,c
xk,c
aS
dπco
X∗
λ
ow
lw
˜ μ
pr
˜Ca
Bk,c(π)
ch
5. Individual monotonicity: Consider two bargaining prob-
lems (S1,d) and (S2,d) such that S1⊂ S2, and the range
of attainable utility by any player j is same in both
(S1,d) and (S2,d). Then individual monotonicity im-
plies that utility of player i ?= j is higher in (S2,d). In
other words, an expansion of the bargaining set in a di-
rection favorable to player i always benefits i.
2.4 Network technologies and characteristics
2.4.1 LTE
LTE is built on an all-new radio access network based
on OFDM (Orthogonal Frequency-Division Multiplexing)
technology with higher order modulation schemes such as
64QAM and sophisticated FEC (Forward Error Correction)
schemes, alongside complementary radio techniques like
MIMO and Beam Forming with up to four antennas per
station. In the downlink, it has the theoretical maximum of
300 Mbps and minimum of 100 Mbps per 20 MHz of spec-
trum, whereas the theoretical uplink rates can reach up to
75 Mbps per 20 MHz of spectrum. LTE performs optimally
in a cell radius of up to 5 km, however it still can deliver ef-
fective performance in cell sizes of up to 30 km radius. LTE
is very flexible and can be deployed in various frequency
bands using a mixture of channel bandwidths (1.25, 1.6, 2.5,
5, 10, 15, 20 MHz). Achievable data rate depends on pro-
vided bandwidth and MIMO technology. Considering mini-
mum requirements of 5 bps/Hz in downlink and 2.5 bps/Hz
is uplink, one can calculate supported data rates for follow-
ing allowed BWs. The size of MAC frames, the Transport
Blocks is decided by the MAC scheduler.
2.4.2 WLAN 802.11g
We consider IEEE 802.11g with Distributed Coordination
Function (DCF). 802.11g operates at a maximum physical
layer bit rate of 54 Mbps exclusive of forward error correc-
tion codes, or about 22 Mbps average throughput. Although
802.11g does not provide QoS guarantee, we assume that
bottleneck is the transport network and we apply Diffserv
there, i.e. given ∼22 Mbps WLAN throughput, the back-
bone network is based on a DSL link of 8 Mbps.
2.4.3 UMTS
We consider the UMTS cellular network with data rates up
to 2 Mbps in small-cell outdoor environments (or in indoor
environment), that uses Wide Band Code Division Multi-
ple Access (WCDMA) technology for UTRAN air interface.
WCDMA operates in Frequency Division Duplex (FDD)
mode, where the given data rate is achieved with spreading
factor 4, parallel codes (3 in downlink and 6 in uplink, 1/2
rate coding), frame length of 10 ms (38400 chips).
3 Model and assumptions
We summarize in Table 1 the notations used in the paper.
We consider an overall area A = {a1,...,aL} in a telecom-
munication coverage landscape L, consisting of an arbitrary
collection of coverage areas of various access technologies,
as demonstrated in Fig. 1. These access technologies ex-
hibit different service capabilities, i.e. they differ in band-
width capacity, coverage and other characteristics (as given
in Sect. 2.4). We assume that these network technologies

Page 4

M.A. Khan et al.
Fig. 1 Network scenario
have two capacity regions, namely congested and uncon-
gested regions. A network technology is said to be in con-
gested region if its current available bandwidth drops below
some threshold value Oth. The congested network behaves
different from an uncongested one when offering resources
to any application request and this behavior is determined
by the load balancing factor, as we elaborate more on in the
next section. We assume that there are a number of appli-
cation users present in the coverage areas mentioned above.
These users are generating bandwidth requests for applica-
tions of different service classes using poisson distribution
and depending on these service classes network technolo-
gies offer different predefined amounts of bandwidth (fol-
lowing different operator policies as detailed in Sect. 3.1) to
the application requests. Users’ applications occupy differ-
ent amounts of bandwidth capacities of network technolo-
gies for some amount of time (holding time) and users leave
afterstayingconnectedforsomerandomintervalandrelease
the resources. To have a realistic model of cellular networks,
we also assume that users are mobile and move from one
area to the other, resulting in variable number of serving net-
work technologiesatdifferent times for the sameapplication
request.
3.1 Operator policies for offered bandwidth
We assume that future telecom operators will employ vari-
ous policies for offering bandwidth to the users. Hence the
predefined offered bandwidth of an operator is assumed to
be the function of operator policy denoted by π. The opera-
tor policies are strictly operator specific and are the elements
of a finite operator policy set ? = {π1,...,πz}. Here we de-
scribe a few different operator policies that we expect will
commonly be adopted by the future operators.
1. πp—supply the available capacity of a network technol-
ogy ˜ cwas the offered bandwidth. We term such a policy
as potential-based policy hereafter.
2. πh—supply the strict values (operator defined) as offered
bandwidth. We term such a policy as hard-coded policy
hereafter. In such settings operators are enabled to define
preferences over the association of services to the spe-
cific network technologies, which dictates that flow split-
ting is feasible over the whole operator capacity (both
Table 2 Offered bandwidths based on operator policies for video ap-
plication (network capacities in kbps)
π
LTE WLAN UMTSFlow splitting
πh
500
˜ cLTE
1000
200
˜ cWLAN
00
600
˜ cUMTS
00
Yes
Yes
No, unless congested
πp
πco
congested and uncongested regions). An example of of-
fered bandwidth using πhis shown in Table 2. As can
be observed, operator controls the flow split by varying
the hard-coded bandwidth offers; however, as soon as the
network technology gets into congestion region, the of-
fered bandwidth is derived by (1).
3. πco—supply the service specific offered bandwidth (op-
erator defined) unless the network congestion region is
reached. We term such a policy as congestion-based pol-
icy hereafter. In this policy the operator avoids the flow
splitting unless any of the network technology in a cov-
erage area gets into congestion region, in which case the
flow is split and load is shared among the available tech-
nologies. As long as the operator is in uncongested state,
the operator defines priorities over the application asso-
ciation to the available technologies, e.g. (i) audio →
UMTS,WLAN,LTE,(ii)video → LTE,WLAN,UMTS,
and (iii) data → WLAN, LTE, UMTS.
The offered bandwidth for πp is straightforward, i.e.
πp= ˜ cw, however when implementing the other policies πh
and πco, it is expected that at some stage (when the oper-
ator is in congestion region) the network technology might
not offer the operator defined bandwidth, thus in such case a
tuned bandwidth is offered, which we term as offered band-
width and is given as follows:
bk,c,w(πco∨πh)
⎧
⎩
where ψ = e
The load balancing factor ψ [21] is used to tune down-
ward the operator offered bandwidth in case any network
technology gets into congestion region. The offered band-
width of an uncongested network is increased (tuned up-
ward) by an amount that is equal to the sum of propor-
tional bandwidth allocated to all those portions of request
that could not have been supported by congested networks.
=
⎨
bk,c,w(πco∨πh)ψ
bk,c,w(πco∨πh)+˜ cw
˜ ca
−lw
if w ∈ Wa
if w ∈ Wa
if ?Wa? = 0
˜Caω
w
(1)
˜Cw+lw and ω =?
k∈Wabk,c,w(1−ψ).
4 Cooperative game theoretic resource allocation
In cooperative games [19] players cooperate, form coali-
tions, and bargain with each other to reach an agreement of

Page 5

Cooperation-based resource allocation and call admission for wireless network operators
mutual benefits as opposed to non-cooperative games. In our
formulation of radio resource allocation problem as a coop-
erative game, different network technologies in any cover-
age area covered by a number of network technologies bar-
gain over the bandwidth requests generated by users for dif-
ferent applications, where each access network has its own
utility function as detailed in Sect. 4.1. The utility function
of network technologies is derived from the bandwidth that
these network technologies allocate to any application re-
quest above the disagreement point. The disagreement point
can be defined as the minimum desired utility that each net-
work expects by joining the game without cooperation. We
formulate the bandwidth resource allocation problem as the
bankruptcy problem and to find the utility distribution rule
(allocation of resources), we use the well-suited game theo-
retic approach Bargaining and a well-known bargaining so-
lution KSBS.
4.1 Network technology utility function
We assume that network utility function is strictly increas-
ing in the offered bandwidth, however offered bandwidth is
driven and bounded by the operator policies:
?˜ μλ
0 if λ = 0
where the decision boolean λ takes the value of 1, when
the call is admitted and 0 otherwise. The component ˜ μ · λ
provides the basis for load sharing by network technologies.
uw(bk,c(πz)) =
if bk,c≥ πz&& λ = 1
if w ∈ Wa
bk,c
(2)
4.2 Operator utility function
Operator aggregated utility can be defined as
?rk,c×(prw,k,c−chw,k,c)
νo(πz,Bk,c) =
if rk,c≤ Bk,c
otherwise0
(3)
where Bk,c=?
operator utility increases in offered bandwidth. The utility
component pw,k,c−chw,k,cof the utility function motivates
the operator to implement different policies.
w∈(Wa∪Wa)bw,k,c(πz) and p represents the
operator pricing scheme. Given any pricing scheme, the
Proposition 1 The network technologies are motivated to
share the load of congested network technologies, by in-
creasing their offered bandwidth within πhand πco.
Proof Combining the network utility function in (2) with
the operator utility in (3) dictates that the network technolo-
gies aggregated offered bandwidth when faced with a condi-
tion Bc,k< rc,kwould result in zero utility, which is further
translated into zero utility of network technology. Therefore
network technology strives to achieve Bc,k≥ rc,kas long as
the total aggregated capacity of network technology greater
or equal to requested bandwidth. Given the arguments, the
result follows.
?
4.3 Problem formulation
To define the bargaining problem for the resource allocation
bankruptcy problem, we start by defining the feasible utility
set. Let the user k ∈ K1generate bandwidth requests rk,cfor
application class c in an abstract coverage area a covered by
various network technologies. Upon receiving the request,
the operator is made available with the necessary informa-
tion2about the network technologies w ∈ {Wa,Wa}. Given
ra
k,cand Ba
gaining problem is a pair (S(ra
k,c,Ba
a compact and convex set, such that
k,cfor each possible value of k,c, and a, the bar-
k,c),d), where S ⊂ ?nis
S(ra
k,c,Ba
k,c) =
?
xa
k,c∈ ?n
?
+| xa
k,c≤ Ba
k,c,
w∈(Wa∪Wa)
xa
w,k,c≤ ra
k,c
?
(4)
Ba
operator then distributes the request amongst the technolo-
giesby implementingthe requestdistributionalgorithm(de-
tailedinSect.4.5).Theconsequenceofthedistributionalgo-
rithm is the bandwidth allocation by the network technology
to the request. The objective of the resource allocation prob-
lem is to find out the amount of optimal resource allocation
x∗
given disagreement point. S represents the set of all possi-
bilities of allocating bandwidth to the request, for which the
available networks are bargaining over. However the alloca-
tion should not exceed the requested bandwidth. Each net-
work technology within the coverage area allocates offered
bandwidth to the requests when the networks in coverage
area are all in the uncongested region, but the network tech-
nologies will offer tuned bandwidth when they are in the
congested region. However, in such a situation the load of
congested networks is shared by uncongested network tech-
nologies present within the same coverage area.3Therefore
our resource allocation management should satisfy the con-
k,cis driven by the operator policies (refer to Sect. 3.1), the
w,k,cto the generated request. d = (d1,...,dn) ∈ ?nis a
1We assume that there exist three different user types namely (i)
Excellent—more sensitive to application quality, (ii) Good, and (iii)
Fair—more sensitive to service costs.
2This information includes the current load, available bandwidth ca-
pacity, offered, and pre-defined offered bandwidths of the base station
(network technology).
3This is implemented by the operator policies detailed in Sect. 3.1.

Page 6

M.A. Khan et al.
ditions in (5) and (6),
?
?
a∈L
ra
k,c≤
?
w∈?
ba
k,c,w=
a∈L(Wa∪Wa)
ba
k,c,w≤
?
w∈?
a∈L(Wa∪Wa)
ow
(5)
w∈(Wa∪Wa)
Setting the value of disagreement point d in our bargain-
ing problem associated with bankruptcy problem (S(ra
Ba
k,c),d) influences cooperation among access technologies.
The existence of a disagreement point is natural, since it en-
dows one with a reference point from where utility compar-
ison can be made. The problem with disagreement point is
thatthereisnouniversallyacceptedcriteriontoselectit[24].
However selection of disagreement point in this paper is in-
fluenced by the objective of simultaneous allocation of re-
sources, which requires that all the available network tech-
nologies in the coverage area should participate in a game.
Therefore, in this paper, we keep the disagreement point as
zero which means that all network technologies will have
utility equal to zero if they do not collaborate. This also re-
alistically represents the situation for different RATs of a
single network operator.
?
w∈(Wa∪Wa)
x∗
w,k,c
(6)
k,c,
4.4 Application of KSBS to the bankruptcy problem
Let the solution of KSBS x∗be given by
x∗
k,c= fks(S,d)
(7)
where
fks(S,d) = d +μks(Xmax−d)
(8)
Here μksis the maximum value of μ such that d +
μ(Xmax−d) ∈ S and Xmaxis the ideal point. Since the dis-
agreement point in our formulation is zero, the problem re-
duces to:
fks(S,d) = μks(Xmax)
(9)
where Xmax in our formulation is analogous to the of-
fered bandwidth. For simplicity, consider the two credi-
tors bargaining problem, where the solution is applicable to
n-creditors. The ideal point for this problem is then defined
as (bw1
ideal point. Given the ideal point, (9) for this problem be-
comes
max,bw2
max), where bw1
maxand bw2
maxare the coordinates of
fks(S,d) = μks(bw1
max,bw2
max)
(10)
the unknown quantity in (10) dictates the solution, which
will be implemented by the request distribution algorithm.
Lemma 1 The recommendations made by KSBS, when ap-
plied to 0-associated bargaining problems, coincides with
the recommendation made by the proportional distribution
rule.
Proof Let x∗
vector x∗, the optimal allocation. These coordinates are in-
dexed by the operator technologies, which satisfies the re-
quirements of feasibility set, i.e. x∗
and convex feasibility set ‘S’. Let us assume that ideal point
for two player case is given by X(bw1
lem would be to minimize the distance between optimum
point (allocation) and ideal point, which is illustrated as fol-
lows:
??
+
as x∗
Plugging this value in (11), we get
??
+
Applying the first order condition to (12), we have
wi∀i ∈ {1,2} represent the coordinates of the
wireside in the compact
max,bw2
max), then the prob-
minx∗
w1,x∗
?
w2can be written in terms of x∗
w2
(x∗
w1)2+(x∗
w2)2
(bw1
max−x∗
w1)2+(bw1
max−x∗
w2)2?
w1, i.e. x∗
(11)
w2= rc,k− x∗
w1.
minx∗
w1
?
(x∗
w1)2+(rk,c−x∗
w1)2
(bw1
max−x∗
w1)2+(bw2
max−(rk,c−x∗
w1))2?
(12)
−2(rk,c−x∗
?
+
2
w1)+2x∗
w1)2+(x∗
max−x∗
(bw2
w1
2
(rk,c−x∗
−2(bw2
?
Minimizing the (13) for x∗
i.e.
w1)2
w1)+2(bw1
w1)2+(bw1
max−rk,c+x∗
max−rk,c+x∗
w1)
max−x∗
w1)2
(13)
w1, we get the optimum allocation,
x∗
w1=
bw1
max.rk,c
?N
Thus the optimal solution coincides with the principle of
proportional distribution, where the proportionality in our
case refers to the allocation proportional to offered band-
width. The solution corresponds to the proportional distrib-
ution [3].
i=1bwi
max
(14)
?
Equation (14) shows that the requested bandwidth is pro-
portionally distributed among the available network tech-
nologies within their coverage area based on their offered
bandwidth.
4.5 Request distribution algorithm
We develop our proportional bandwidth allocation algo-
rithm based on Lemma 1, as described next. The amount of

Page 7

Cooperation-based resource allocation and call admission for wireless network operators
allocated bandwidth by network w in area a to the request
of application of class c is given by
x∗
w[ra
where
k,c] = β ×bk,c,w(πz)
(15)
β =
⎧
⎪⎩
⎪⎨
ra
k,c
Ba
k,c
ra
k,c
˜Ca
if ?Wa? > 0
otherwise
(16)
and β is a proportionate factor.
4.6 Call admission control
The call admission control procedure in our work depends
on the amount of bandwidth available on all those networks
available at that time. An application request is admitted
only if the sum offered bandwidth (predefined and tuned)
by all the networks within the coverage area is more than
requested bandwidth. Hence the admission control scheme
is expressed as follows.
?
However for non-real-time applications, we assume that
user requests represent the user subscribed bandwidth. In
such a case, an operator not meeting the required level of
bandwidth causes an irritated user. Hence blocking such
a call is preferred operator strategy rather than adopting a
strategy where the user is admitted but service is not guar-
anteed. This in turn helps the reduction in user churn rate
in the long-term. This argument justifies the call admission
also for non-real-time applications.
x∗
w[ra
k,c] =
ra
k,c
0
if ra
otherwise
k,c≤?
w∈(Wa∪Wa)ba
k,c,w
(17)
4.7 Area handover
We also consider the resource allocation issue when the user
is mobile. User of an application may change the coverage
area as he moves, resulting in a varying set of available net-
works to the user at different instances of time, which is
analogous to the variable population bargaining problem.
A slightly modified version of our resource allocation al-
gorithm for mobility also achieves our objective of propor-
tional resource allocation in variable network scenario. Call
admission control is performed for each movement of a user
from his current coverage area to any destination cover-
age area, according to (17). The user then releases the por-
tion(s) of bandwidth allocated by all those networks that
were present in the previous coverage area but not present
in current coverage area. The following algorithm explains
this handover process:
?
xao→acu[ra
k,c] =
ra
k,c
0
if racu
otherwise
k,c≤?
w∈(Wa∪Wa)ba
k,c,w
(18)
and
cao
w(t +s) = cao
where aois the previous/old coverage area before the han-
dover, acuis the current coverage area after the handover, t
is the time before handover and t +s after the handover.
w(t)+xao[ra
k,c]
(19)
5 Numerical analysis
5.1 Scenario description
In this section we use a case study to assess the behavior
of the proposed scheme. We consider the scenario with dif-
ferent areas covered by a various number of technologies
as can be seen in the Fig. 1. The access technologies in-
clude UMTS, WLAN, and LTE with bandwidth capacities
2 Mbps, 6 Mbps, and 20 Mbps respectively. (The high data
rate of 20 Mbps for LTE is chosen specifically to present
the results more clearly, but it is still realistic considering
the 40 Mbps as its design goal.) Different coverage areas are
defined in terms of network technologies that cover these
areas:
1. a1= UMTS, LTE, WLAN
2. a2= LTE, WLAN
3. a3= LTE, UMTS
4. a4= WLAN, UMTS
For simulations, the arrival of requests is modeled by a
Poisson process, and the service class is chosen randomly
among voice, data, and video uniformly. The sizes of the re-
quests are assumed to be static, and are 60 kbps, 300 kbps,
and 450 kbps for voice, data, and video, respectively. Af-
ter the distribution algorithms, the allocated bandwidths are
subtracted from the bandwidth pools of the RANs, assuming
that users have an infinite channel holding time. This allows
us to simulate the overload conditions in the coverage areas.
We first apply our proportional resource allocation scheme
for three different operator policies, πh, πp, πco, in a station-
ary setting, where all three network technologies are avail-
able for the entire duration of study (area a1in Fig. 1). The
results are presented in Figs. 2 and 3, where the events on
the horizontal axis effectively represent time instances for
the arrival of application bandwidth requests. In Fig. 2, the
bandwidth amounts allocated by each available network to
the applications are given, while the number of admitted and
dropped calls are given in Fig. 3. Note that the values in
Fig. 3 are not instantaneous but cumulative; hence, e.g. the
adjacentrepeatingvaluesforcalldropmeanthatnocalldrop
has occurredbetweenthe correspondingtimeevents. We ob-
serve from Fig. 2 that a proportional amount of bandwidth is
allocated to the requests based on the offered bandwidth of
the network, which in turn depends on the operator policies.

Page 8

M.A. Khan et al.
Fig. 2 Resource allocation for
three policies in area a1
Fig. 3 Number of active calls
vs. call drops in area a1
It can be observed that operator policies πpand πhbehave
in a similar fashion in terms of network utilization, i.e. the
load is distributed among different access technologies, sat-
urating each technology around the same time. As soon as
UMTS gets into the congestion region, it starts offering the
tuned bandwidth and its load is shared by WLAN and LTE.
After event-15, LTE starts sharing the loads of both UMTS
and WLAN until its capacity is exhausted and call drops are
observed after event-19 (see Fig. 3).
On the other hand the operator policy πcoresults in dif-
ferent saturation points, i.e. at event-1, event-5, and event-18
for UMTS, WLAN, and LTE, respectively. The impact of
avoiding flow splitting during below congestion level can be
analyzed by studying the call blocking. Call blocking in this
case starts earlier, i.e. event-14, as compared to other ap-
proaches, where the call blocking start at event-19. Further-
more a 29.5% decrease in active, and an increase of 38.2%
in call blocking advocates the superiority of policies πpand
πhover the policy πh.
Observation
policy if it is interested in avoiding flow splitting complex-
ities or reducing the signalling cost involved in splitting
flows.
In order to analyze the behavior of our proposed algo-
rithms with different operator policies in case of mobility,
we consider different scenarios covering different situations
of changes in the set of available access networks. This mo-
bility scenario is depicted as a path labeled A to F in Fig. 1.
We start with calls admitted in area a1and assume that the
users of those applications move to area a2. Since the three
networks are already in congestion region, few call drops for
policies πhand πpare observed as users move to area a2,
whereas about 17% more calls are blocked when operator
implements πco. This can be seen near the right edge of
the region labeled as B in Fig. 4. No call drops are ob-
served when the users move from a2to a1, which can be
observed by a straight line in the region labeled as C. In
region D, users move from area a1to a3loosing the cov-
erage of WLAN, which results in releasing bandwidth of
Anoperatorimplementsthecongestionbased

Page 9

Cooperation-based resource allocation and call admission for wireless network operators
Fig. 4 Number of active calls
vs. call drops in mobile scenario
Fig. 5 Resource allocation for
three policies in mobile scenario
WLAN and then WLAN gets out of the congestion region
(see Fig. 5). LTE shares its load with UMTS as soon as
UMTS is available when users move from area a3to a1as
shown in region E, and no call drops are observed in this
region. Now when the users move to area a4, loosing the
coverage of LTE, call drops are observed, inevitably. The re-
source utilization pattern in the mobile scenario pattern can
be viewed in Fig. 5.
5.2 Comparison to previous approaches
Wenowtrytoassesstheeffectivenessofourresourcealloca-
tionschemeincomparisontosimilarexistingworkinthelit-
erature. First, we provide comparisons with [12], where the
authors analyze the performance of their proposed schemes
(LessVoice and Random) and other existing heuristics (First
Fit, BestFit, WorstFit) in multi-access, multi-service envi-
ronment in terms of call blocking probabilities. We use the
same simulation parameters as used by the referred papers
to obtain the call blockingprobabilityusing our approach.In
addition, we set the values of additional parameters required
inourscheme,namelypredefinedoffered bandwidth, bk,c,w,
and congestion region threshold, CRw, as given in the first
column of Table 3.
Figure 6 shows that the call blocking probability for
our resource allocation scheme varies around 5 percent,
and are both significantly less than the results presented in
the referred paper for different algorithms with voice and
elastic/non-elastic applications. Carefully tuning the para-
meters in our approach can result in even less call drop
probability. We observe that there is a trade-off between the
eventual call drop probability and the time call drops start
increasing, as depicted in Fig. 6.

Page 10

M.A. Khan et al.
Fig. 6 Comparison of our resource allocation scheme to those in [12]
Fig. 7 Comparison of our resource allocation scheme to that in [15]
Table 3 Algorithm parameters
Parameter GSM
Value (1)
WCDMA
Value (1)Value (2) Value (2)
b(voice1)
b(voice2)
b(data)
CR
12
27
48
30
120
110
300
30
32
70
123
40
150
200
50
35
Change in performance of call blocking probability with
different values of predefined offered bandwidth and con-
gestionregionthresholdvalueisnatural,astheamountofal-
located bandwidth to request of bandwidth depends on pre-
defined offered bandwidth. Network operators can control
the allocation of bandwidth to any class of application by
tuning predefined offered bandwidth and their control on the
congestion region enable operators to always allocate some
fixed amount of resources to application requests, until net-
work congestion threshold value is reached and allocate less
resources afterwards.
As noted at the beginning of this paper, [15] is very rel-
evant to our work, as it also considers simultaneous use of
multiple interfaces (as opposed to [12]) and also uses a col-

Page 11

Cooperation-based resource allocation and call admission for wireless network operators
Fig. 8 Number of active calls
vs. traffic intensity (calls/min)
Fig. 9 Call blocking
probability vs. simulation steps
laborative game model with a different solution method for
resource allocation. Thus we also compare the performance
of our approach with [15], this time for average number of
connections with unequal connection arrival rate. As the re-
ferred paper also discusses bandwidth splitting using coop-
erative games, we used the exact same parameters and sim-
ulated discretely for different areas using our resource allo-
cation approach. The results are demonstrated in Fig. 7.
We also compare the proposed approach to service-based
andcapacity-based approaches,forwhichweimplementthe
allocation schemes as described in the Introduction section.
The simulations are run in area a1, which has the cover-
age of all access technologies. We compute the call block-
ing probability in area a1as a function of the simulation
steps. We also plot the number of accepted requests as a
function of traffic intensity in calls per minute, assuming a
simulation time of 100 minutes. The results for these re-
gions are given in Figs. 8 and 9. These results are com-
pared with the results obtained by the capacity-based and
service-based approaches discussed earlier. In the capacity
based approach the players (network technologies) report
their available bandwidths to the operator. The service re-
quest is then allocated to the RAN with the largest amount
of free bandwidth. In the service-based approach, service
classes are associated with certain RANs, and are allocated
to other RANs only if the associated RANs are overloaded.
In this scheme voice is allocated to GSM (implemented in
simulation for this comparison). Then to UMTS, and fi-
nally to LTE. Data is associated to UMTS, and allocated to
LTE in overload. For video the sequence is UMTS and then
LTE. In this scheme the individual RANs submit the type
of traffic they can support to the CRRM. CRRM chooses

Page 12

M.A. Khan et al.
the RANs that are willing to support the service class of the
request.
Our approach outperforms both the service-based and
capacity-based solutions. In the area a1, where only the we
can support 12% more calls, with the same call blocking
probability as the service-based approach.
6 Realization of the proposed flow splitting approach
The proposed approach of proportional flow splitting can be
realized by implementing the Multiple Care of Addresses
(MCoA) of Mobile IPv6 (MIPv6) as mentioned in RFC
5648. MCoA is employed for using various care of ad-
dresses simultaneously. In MIPv6 the Mobile Equipment
(ME)registersitsactiveinterfaceswithitsHomeAgent (HA)
through MCoA registration mechanism and also instructs its
HA by sending filter rules. HA starts intercepting ME des-
tined traffic in its home network and sorts out, using filter
rules sent by ME. This enables ME to use its various inter-
faces for receivingtraffic simultaneouslyhencerealizingour
proposed splitting approach. In addition, flow management
identification option in MIPv6 header is used to establish
flow bindings between ME and HA/CN, just like a regular
location update procedure used to inform receiver about the
current location of ME. Moreover, the filter rules are also
communicated using this flow binding. These flow bindings
can be refreshed, removed, and can also get expired. A flow
binding message is usually piggybacked on its associated
CoA’s binding message. In order to manage flows across
available interfaces, another pair of extensions [6] can be
used. The intrinsic problem in multi-path flow is the out of
order packet delivery, which has adverse effect on TCP end-
to-end performance and even some strictly delay sensitive
applications. To avoid such TCP performance degradation,
the solution in [7] can be employed.
7 Conclusion
In this paper, we have presented a game theoretic ap-
proach for resource allocation using cooperative games,
where available network technologies cooperate to simul-
taneously allocate resources to the application requests. The
amount of allocation by each network technology is deter-
mined by a distribution rule that is found by applying Kalai-
Smorodinsky Bargaining Solution to our resource alloca-
tion problem. Based on the distribution rule for resource
distribution, we developed resource allocation, call admis-
sion and area handover algorithms and the proof of con-
cept is presented by simulating our approach in different
scenarios especially when users of applications are mobile.
We also compared our approach with similar existing ones,
demonstrating superior performance in terms of call drop-
ping probability.
References
1. Badia, L., Taddia, C., Mazzini, G., & Zorzi, M. (2008). Multira-
dio resource allocation strategies for heterogeneous wireless net-
works. EURASIP Journal of Advanced Signal Processing.
2. Bazzi, A., Pasolini, G., & Andrisano, O. (2008). Multiradio re-
source management: parallel transmission for higher throughput?
EURASIP Journal of Advanced Signal Processing.
3. Bosmans, K., & Lauwers, L. L. (2007). Comparisons of nine rules
for the adjudication of conflicting claims. Katholieke Universiteit
Leuven Centrum voor Economische, ces0705.
4. Das, S. K., Lin, H., & Chatterjee, M. (2004). An econometric
model for resource management in competitive wireless data net-
works. IEEE Network, 18, 20–26.
5. Dimou, K., Agero, R., Bortnik, M., Karimi, R., Koudouridis, G. P.,
Kaminski, S., & Lederer, H. (2005). Generic link layer: a solution
for multi-radio transmission diversity in communication networks
beyond3G.InProc.IEEE62ndvehiculartechnologyconf.(Vol.3,
pp. 1672–1676).
6. draft-ietf-mext-binary-ts.
7. draft-ietf-mptcp-architecture-01.
8. Gabor, F., Anders, F., & Johan, L. (2004). On access selection
techniques in always best connected networks. In ITC specialist
seminar on performance evaluation of wireless and mobile sys-
tems.
9. Halder, N., & Song, J. B. (2007). Game theoretical analysis of ra-
dio resource management in wireless networks: a non-cooperative
game approach of power control. IJCSNS International Journal of
Computer Science and Network Security, 7(6), 184–192.
10. Kalai, E. (1983). Solutions to the bargaining problem, vol. 556.
Center for mathematical studies in economics and management
science.
11. Luo, J., Mukerjee, R., Dillinger, M., Mohyeldin, E., & Schulz, E.
(2008). Investigation of radio resource scheduling in WLANs cou-
pled with 3G cellular network. IEEE Communications Magazine,
41, 108–115.
12. Mariz, D., Cananea, I., Sadok, D., & Fodor, G. (2006). Simulative
analysis of access selection algorithms for multi-access networks.
In Proc. international symposium on a world of wireless mobile
and multimedia networks WoWMoM.
13. Niebert, N., Schieder, A., Abramowicz, H., Malmgren, G., Sachs,
J., Horn, U., & Prehofer, C. (2004). Ambient networks: an archi-
tecture for communication networks beyond 3G. IEEE Wireless
Communications, 11, 14–22.
14. Niyato, D., & Hossain, E. (2006). Bandwidth allocation in 4G het-
erogeneous wireless access networks: a noncooperative game the-
oretical approach. In Proc. IEEE global telecom. conf. GLOBE-
COM (pp. 1–5).
15. Niyato, D., & Hossain, E. (2006). A cooperative game framework
for bandwidth allocation in 4G heterogeneous wireless networks.
In Proc. IEEE international conf. on communications ICC ’06
(Vol. 9, pp. 4357–4362).
16. Osborne, M. J., & Rubinstein, A. (2005). Bargaining and Markets.
UCLA Department of Economics.
17. Perez-Romero, J., Sallent, O., Agusti, R., Karlsson, P., Barbaresi,
A., Wang, L., Casadevall, F., Dohler, M., Gonzalez, H., & Cabral-
Pinto, F. (2005). Common radio resource management: functional
models and implementation requirements. In IEEE 16th personal
and indoor and mobile radio communications (Vol. 3).
18. Pries, R., Andreas, & Staehle, D. (2006). A network architecture
for a policy-based handover across heterogeneous networks. In
Proceedings of OPNETWORK.
19. Rasmusen, E. (2006). Games and information: an introduction to
game theory. New York/Oxford: Wiley/Blackwell.
20. Song, W., Zhuang, W., & Cheng, Y. (2007). Load balancing for
cellular/WLAN integrated networks. IEEE Network, 21, 27–33.

Page 13

Cooperation-based resource allocation and call admission for wireless network operators
21. Suliman, I. M., Pomalaza-Rez, C., Lehtomki, J., & Oppermann, I.
(2004). Radio resource allocation in heterogeneous wireless net-
works using cooperative games. In Proc. Nordic radio, symposium
2004/finnish wireless communications workshop.
22. Taha, A.-E. M., Hassanein, H. S., & Mouftah, H. T. (2008). Verti-
calhandoffsasaradioresourcemanagementtool.ComputerCom-
munications, 5, 950–961.
23. Tolli,A.,Hakalin,P.,&Holma,H.(2002).Performanceevaluation
of common radio resource management (CRRM). In IEEE inter-
national conference on communications (Vol. 5, pp. 3429–3433).
24. Vartiainen,H.,&Foundation,Y.J.(2003).Bargainingwithoutdis-
agreement. Discussion papers.
Manzoor A. Khan received the
Bachelors of Engineering degree
in Electronic Engineering from the
Mehran University of Engineering
and Technology (MUET), Pakistan,
in 2001, the M.S. in computer sci-
ence degree from Balochistan Uni-
versity of Engineering, Information
Technology and Management Sci-
ences, Pakistan, in 2005. He is pur-
suing his Ph.D. at DAI Labor, Tech-
nical University Berlin since 2007.
He is the author of several scholarly
articles and book chapters. His re-
search interest includes the resource allocation, network selection al-
gorithms, and representation of user Quality of Experience (QoE).
Ahmet C. Toker graduated with
high honors from the Middle East
Technical University, Ankara in
2003 with a B.S. in Electrical and
Electronics Engineering. He pur-
sued a M.Sc. in Electrical Engi-
neering from University of Texas at
Austin in 2005. Since 2005, he is
a research scientist and PHD can-
didate in DAI-Labor of Technical
University of Berlin. He is the au-
thorofvariousscholarlyarticlesand
patents. His research interest is the
application of AI formalizations to
the cooperative solution of resource allocation problems in All-IP net-
works.
Fikret Sivrikaya received his Ph.D.
degree in Computer Science from
RensselaerPolytechnic
NY, USA. Before joining DAI-
Labor at Technische Universität
Berlin, he was as a research assis-
tant at Rensselaer Polytechnic Insti-
tute and earlier worked for a few
telecommunications companies in
Istanbul, Turkey. He is currently the
director of competence center Net-
work and Mobility at DAI-Labor.
He is responsible for coordinating
project and research activities in the
Institute,
area of telecommunication technologies. His research interests include
wireless communication protocols, medium access control and rout-
ing issues in multi-hop ad-hoc networks, distributed algorithms and
optimization.
Sahin Albayrak holds a profes-
sorship chair for Agent Technolo-
gies in Business Applications and
Telecommunication (AOT) at the
Technische Universität Berlin. He
is a member of the Institute of
ElectricalandElectronicsEngineers
(IEEE), Association for Computing
Machinery (ACM), Gesellschaft für
Informatik, American Association
for Artificial Intelligence (AAAI).
In addition, he is a member of
the Deutsche Telekom Laboratories
(T-Labs) steering board. Research
areas are next generation telecommunication services, applications and
networkinfrastructures,service-centricarchitectures,serviceengineer-
ing, agent-oriented modelling, agent architectures, agent programming
languages, telecommunication services, e-/m-commerce, mobility sup-
porting services, 3G- and Beyond-3G services, supply chain manage-
ment, autonomous security and smart systems.