Pervasive and Mobile Computing 7 (2011) 569–583

Contents lists available at ScienceDirect

Pervasive and Mobile Computing

journal homepage: www.elsevier.com/locate/pmc

A QoS-based dynamic pricing approach for services provisioning inheterogeneous wireless access networksAntonio Guerrero-Ibáñez a,∗, Juan Contreras-Castillo a,1, Antoni Barba b,2, Angélica Reyes c,3

a School of Telematics, University of Colima, Av. Universidad #333, Colonia Las Víboras, 28040 - Colima, Mexicob Telematics Engineering Department, Technical University of Catalonia, Jordi Girona 1-3, Edifici C3, 08028- Barcelona, Spainc Computer Architecture Department, Technical University of Catalonia, Avinguda del Canal Olímpic, 15 Edifici C4, 08860 Castelldefels, Spain

a r t i c l e i n f o

Article history:Received 17 March 2010Received in revised form 2 August 2010Accepted 1 October 2010Available online 8 October 2010

Keywords:Flat-rate modelHeterogeneous wireless access networksNetwork selection algorithmsPricing strategiesQuality of serviceUser satisfactionProfits

a b s t r a c t

In this paper we propose a simple QoS-based dynamic pricing approach for servicesprovisioning in a heterogeneous wireless access network environment which attemptsto increase user’s satisfaction level by firstly, maximizing the provided QoS level, andsecondly, by applying dynamic pricing strategies according to the QoS. These strategieswill allow service providers to maximize their profits. Simulation results demonstrate thatthe proposed dynamic pricing approach benefits both users and wireless service providers(WSPs). Results also suggest that users have better overall satisfaction due to a better QoSlevel and fairer prices. The analysis shows that our proposed pricing approach contributesto an increase inWSPs profits compared to the application of the flat-rate pricing model ina competitive market-model.

© 2010 Elsevier B.V. All rights reserved.

1. Introduction

In recent years, we have seen an increase in demand formobile computing and communication services. Development inwireless access technologies and sophisticated personal user devices are driving the way towards a heterogeneous wirelessaccess network (HWAN) environment which will enable anytime and anywhere communication. Various wireless servicessuch as 3rd generation cellular system, IEEE802.16 WiMAX r⃝, IEEE802.11 WiFi r⃝, and Bluetooth r⃝ are now getting popularall around the world. In order to support requests for seamless multimedia and high quality services, the interworkingintegrated network architecture over heterogeneous wireless networks have been reported [1–4].

Heterogeneous wireless access networks will create a market for the delivery of an extensive collection of novel andattractive services and contents. Obviously, this kind of environment will support services that have a variety of Quality ofService (QoS) requirements (such as: low latency, high bit rate, low error rate, among others); for example, multimedia overbroadband networks, voice over Internet Protocol (VoIP), and the Internet Protocol Television (IPTV) are more sensitive todelays, while file transfers are affected by loss [5].

This new market will promote the generation of a multitude of wireless service providers (WSPs), which can use a mixof wireless access technologies to provide these new services to end-users with competitive prices. It is therefore in the

∗ Corresponding author. Tel.: +52 312 3161075.E-mail addresses: antonio_guerrero@ucol.mx, jagigro@hotmail.com (A. Guerrero-Ibáñez), juancont@ucol.mx (J. Contreras-Castillo),

telabm@entel.upc.edu (A. Barba), mreyes@ac.upc.edu (A. Reyes).1 Tel.: +52 312 3161075.2 Tel.: +34 934016022.3 Tel.: +34 934137220.

1574-1192/$ – see front matter© 2010 Elsevier B.V. All rights reserved.doi:10.1016/j.pmcj.2010.10.003

570 A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583

best interest of WSPs to offer services that appeal to end-users and provide them with a best use to increase the technicalabilities of their terminals. From the service provider’s point of view, it will be necessary to define mechanisms to maximizeuser’s satisfaction level, allocating the most suitable QoS level to each requested service and the definition of competitivedynamic pricing strategies that adjust or adapt the price accordingly to the provided QoS.

Success in thismarketwill be based on the ability to offer an enhanced user experience. Pricing and resourcemanagementare important issues to end-users since the acceptance of the services is directly related to the perceived QoS and offeredprice. HWAN will change the traditional vision of service model from network-centric to user-centric. Unlike traditionalservice model where all management functions are controlled according to the service provider’s perspectives, the user-centric vision for the future considers end-users in a heterogeneous wireless network environment where they will be freeto ‘shop around’ not only for the service they need, but also for the available wireless access network which meets theircurrent service needs. Users will take advantage of this competition scenario and they will always connect to the networkthat can best satisfy their needs and preferences for the current application using novel algorithms to select the optimalaccess network in an intelligent way according to their needs.

In this situation, heterogeneous wireless networks look for the market evolution from traditional monopolies thatdictate usage conditions, to a user-centric service-oriented environment. However, within the current service pricing andprovisioning model, there is still a strong relationship between end-users with a single wireless service provider. End-usersget services from one provider for a period of time based on the contractual agreement. Prices for the various services arebased on the charging model for voice services (i.e. free, flat rate or any of their variations). In the flat-rate model price arefixed and do not fluctuate according to changes in network conditions. The price is paid monthly for the usage of a certainfacility. However, the current demands of quality of services into wireless service access have demonstrated the need formore sophisticated pricing strategies [6,7].

In this paper we present a QoS-based dynamic pricing approach for service provisioning in a heterogeneous wirelessnetwork environment where requirements of new services demand efficient and flexible pricing strategies and chargingmechanisms. We propose an adaptable scheme to changeable environments, which satisfies users’ demands whilemaximizing service providers’ revenue. Our approach, called QoSDPA (QoS-based Dynamic Pricing Approach), is based onthe IETF policy model [8]. In order to maximize user’s satisfaction, QoSDPA defines an access network selection mechanism,explained later on, that attempts to allocate the most appropriate network for the requested service, satisfying user’sdemands and optimizing the usage of resources. As a result, service providers may increase their profits too.

In addition, when the QoS requirements cannot be fulfilled, a dynamic pricing strategy and a negotiation procedure aredefined. The main objective is to maintain a high QoS or to recover it as soon as possible when degradation occurs, thusobtaining the highest user’s satisfaction.

The rest of the paper is organized as follows: A brief description of related work is discussed in Section 2. Section 3describes in detail our QoS-based dynamic pricing approach. An evaluation of the performance of QoSPAMS is presented inthe Section 4. Finally, Section 5 shows the conclusions of our work.

2. Related work

The evolution of pricing strategies for telecommunication mobile services is in a continuous process. Several pricingstrategies have been proposed and some of themhave been implemented in the commercial environment. However, pricingin heterogeneouswireless access networks is still a challenge that requiresmore research. There are someworks that definea list of requirements that network designers must meet [9,10]. Basically, recent pricing works can be classified into twocategories: management architectures and control algorithms and strategies.

Regarding management architectures we can mention several research results. Ref. [11] discusses a charging systemcalled CAB. The main problem of this proposal is that the solution is based on a centralized architecture for accountingmanagement used in the current communication systems. However this architecture is not adequate for a heterogeneouswireless network environment where the accounting and pricing will be made in a distributed way.

A component-based charging in a next-generation multimedia network is proposed in [12]. The work analyzes the set ofcomponents that are relevant in the accounting process. However authors do not define a model to calculate the price.Ref. [13] proposes a comprehensive component-based accounting and charging architecture to support service sessionprovisioning across multiple domains. The architecture incorporates an interim accounting and charging mechanism toenable the processing and exchange of accounting information needed to update intermediate charges for separate servicecomponents and the user’s credit. However, these works do not consider situations in which different access technologiesand high users’ mobility are involved. In future mobile networks the complexity of accounting and charging increases indifferent forms e.g. in the seamless provisioning of service access across administrative domains (different network andservice providers), the use of different technology domains (cellular and/or wireless networks) and Quality of Services (QoS)constraints. Ref. [14] defines a role model for pricing and accounting that covers all participating entities of a distributedservice providing an environment for mobile networks. The model defines that accounting is configured to agree upon userprofiles and service classes. Themain disadvantage of all the proposedmodels is that they consider that the user has a strongassociationwith a singleWSP. However, the new vision of HWANs is focused on a user-centricmodelwhere users havemorefreedom of dynamically connect to anyWSP for any service and they can disconnect at anytime. In other words, there is notany contractual agreement for a fixed time period.

A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583 571

Regarding control strategies several solutions have been proposed in the literature. Ref. [6] presents a survey of pricingschemes. In this work the Pricing Schemes are classified, based on their ability to adapt to the needs of the WSPs and theirsubscribers during the entire service period, into Static-based Pricing and Dynamic-based Pricing Schemes. The PricingSchemes are also analyzed in detail and are further classified according to the factors involved in the price calculationof a service, i.e. the Service Level Agreement (SLA), the subscription type, the negotiation capabilities between WSPs andtheir subscribers, the network capacity, the available bandwidth and frequency spectrum, the network hops, and the BaseStations (BSs). The affected elements by the pricing network are also discussed, together with the performance evaluation ofthe presented pricing schemes. In [15] authors propose a new continuous timemodel with price and time sensitive demandto consider dynamic pricing, order cancellation ratio and different quality of service (QoS) levels in web networks. Theydefine a solution procedure to determine the optimal number of selling stages, the optimal price and service capacity. Theresults show that the proposed method is an effective method for solving dynamic pricing policy of service provider withdifferent QoS levels in web networks. However, sensitive-delay services are not considered. Authors in [16] studied thescalable connection management strategy for quality of enabled network service. The results demonstrated the incentivepricing control strategies are effective and practical.

In [17], a pricing policy for multiple competing ISPs (Internet Service Providers) using a threat strategy is presented.Ref. [18] shows a proposal in which authors use game theory to analyze the impact in the cost based on the economicinterests of a wireless access point owner and his/her paying client. In Ref. [19] several problems for resource allocation andbase-station assignment in CDMA (Code division multiple access) networks are studied. Other projects focus on evaluatingthe revenue maximization and pricing problems [7,20].

Unlike other pricing management proposals, which reflect the maximization of profits from the points of view eitherof service provider or users, our proposal considers both (service providers and users’ satisfaction) by using a flexible QoS-based pricing approach for heterogeneous wireless access networks.

Our approach includes a selection mechanism that allows users to choose the wireless access network that best adaptto their needs, and preserves the QoS level during the time the service is being used. Additionally, our QoSDPA algorithmdefines a dynamic pricing strategy that is applied when the QoS level is degraded. This strategy attempts to maintain a highlevel of user’s satisfaction, while maximizing the profits obtained by service providers.

3. Proposed pricing and accounting management approach

Pricing and accounting management are relevant topics for heterogeneous wireless access network developmentsbecause of the high and ambitious requirements in terms of heterogeneity, mobility, multiple domains, security, contextawareness, and composed services, among others. We focus on the pricing management by proposing the definition ofalgorithms and strategies that allow keeping an acceptable user’s satisfaction level and maximizing the obtained benefitsfor both users and service providers. We explain first the global scenario where our pricing approach is applied and in therest of the section we discuss in detail our dynamic pricing strategy approach for heterogeneous wireless access networks.

3.1. Global scenario

We assume a heterogeneous wireless access network environment as shown in Fig. 1. In this scenario, nwireless serviceproviders compose a competitive environment. All wireless service providers try tomaximize their profits. In order to reachtheir goals, they provide an array of heterogeneous services by using a wide variety of heterogeneous access networks. Weassume that different technologies (cellular networks, Wi-Fi, WiMAX, among others) are overlaid in the implementation.Each technology provides several features: better coverage, higher data rate, high security, etc. In this way,WSPs can exploitthe relative advantages of different types of technologies to expand their markets, to attract a major number of users, andto increase their revenues.

In this scenario we define services as wireless applications with different QoS requirements. Each service has its ownrequirements.We consider the four service classes defined by the 3rd Generation Partnership Project (3GPP): conversationalclass, streaming class, interactive class, and background class [5]. The main distinguishing factor between these QoS classesis how delay-sensitive to the traffic is: conversational class is meant for very delay-sensitive traffic while background classis themost delay insensitive traffic class. Conversational and streaming classes aremainly intended to carry real-time trafficflows. Table 1 summarizes the relevant parameters for each service class.

Users are willing to play a more central role by modeling and reshaping the service experiences upon their needs. Asmentioned, ourmodel is user-centricwhich improves the service offering profitable, value-added services faster and cheaperthan ever before. In this scenario we assume that end-users have more freedom connecting dynamically to any WSP forany service through their multi-mode devices and can disconnect at anytime, which means they do not have any strongrelationship with service providers. So WSPs and end-user only interact on a per-service or per-session basis.

From a user’s perspective, the objective is to select theWSP that best adapts to his/her needs. FromWSP perspective theobjective is to define dynamic pricing strategies for each service and the requested QoS level; this is important to considerbecause if prices are low,WSP would attract too many users leading to degradation in performance due to resource sharing.At the same time, if advertised prices are too high, it will lead to users turning down that WSP.

572 A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583

Fig. 1. Heterogeneous wireless network global scenario representation.

Table 1Summary of relevant parameters for service classes.

Service class Applications Relevant parameters

Conversational • Voice call, Delay, jitter• Videoconference

Streaming • Audio streaming, Jitter• Video streaming.

Interactive • Multimedia web browsing Delay, BER• Interactive remote games

Background • http downloading

BER, bitrate• FTP.• Electronic mail.• Html browsing

3.2. Management control policies

The costs of networkmanagement in heterogeneous wireless access networks can be highly reduced using policies; thusQoSDPA is based on the IETF policymodel. In thisway,we support the ability of a heterogeneouswireless network to adapt toever-changing situations and conditions.When new network resources become active, policies can change and the businessneeds and models vary accordingly. These changes are actuated through high-level policies and translated into low-levellocalized actions (Policy Enforcement Points) allowing the achievement of overall system goals.

Our approach specifies three policy classes: access, negotiation, and pricing and charging. The main objective of thepolicies is to satisfy user’s service demands by firstly, maximizing the provided QoS level; and secondly, when the QoS levelcannot be provided, by applying policies with fairer tariffs attempting to recover the user’s satisfaction level as soon aspossible. These measures should transcend in a set of benefits for the service providers.

Access policies allowmanaging the allocation procedure of the service requests.When a request arrives, the access policyexecutes a preliminary admission. When the network can provide the required QoS, the access policy triggers the selectionmechanism (which is explained later on) in order to select the best cell for the service to maximize the QoS provided to theuser. Otherwise, the access policy triggers the negotiation policy.

Negotiation policies can be used for new QoS levels allocation. These policies define a negotiation model used when theQoS level cannot be provided. Some considerations are analyzed during the negotiation process. Since dropping an ongoingrequest is more serious than blocking a new one, the negotiation policy treats handoff requests differently. This measure isrelated with the user’s perception about service continuity.

Each service provider has its own tariffs for the service. However, tariffs should consider a dynamic model according tocurrent network conditions. In our case, pricing policies help adjusting the price for the service in a dynamic way, withinthe changeable behavior of the environment.

A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583 573

3.3. Access network selection algorithm

In the HWAN environment, it is important to use the appropriate access technologies according to their location to enjoyservices with the optimal quality. Currently, there is not an automated way for the user terminal to intelligently select themost appropriate network from a heterogeneous mix of wireless and mobile access technologies in order to access thedesired service. Some related works have been made for Cooperative Wireless Networks [21]. There are works which havefocused on solving the problem from the service providers’ point of view, as shown in [22] where the main objective is tomaximize the overall network capacity. However, they do not consider user’s preferences in the decision. In [23] authorspropose a mechanism of access network selection which assumes a complete handoff decision process (handoff triggeringand network selection). The handoff trigger is decided based on the received signal strength (RSS). But they do not considera differentiation among the diverse services and their specific requirements.

Our proposed algorithm obtains the most efficient and suitable access network, by dynamically placing the (individual)service requests in any of the best available wireless access networks at any location within a complex heterogeneouswireless network environment, meeting the service’s QoS requirements and user’s preferences. The algorithm is called QoS-based Network Selection Algorithm (QNSA) and was designed to guide user’s behavior in cooperative way. The algorithmconsiders user’s preferences, network capabilities (e.g. bandwidth), current network conditions and service requirementsadding intelligence to the selection procedure through connection records.

The QNSA is based on the definition of a utility function and the use of a user connection record, named User ConnectionProfile (UCP). QNSA defines two essential mechanisms for selecting the best access network for the service: the utilityfunction, and the user connection profile (UCP). The following subsections describe the defined factors in detail.

3.3.1. Utility functionA utility function is a mathematical characterization that represents the costs and benefits obtained of a product. To

define our utility function we assume that there are M WSPs that are trying to cater the users and N services available. Wealso assume that a user is requesting a service Si which has a specific demand of QoS requirements in terms of delay, jitter,and bandwidth to satisfy user’s request. So, in our model as users pay for the service to the WSP from whom the servicesrequest is granted, each user has the freedom of selecting the best WSP based on his/her preferences. In this context, theobjective of any wireless service provider is to maximize its profits for the service provided, and users try to maximize theirbenefits based on the price they pay for and the quality of service received.

Based on those considerations we define the utility function as shown in (ec-1). The utility function defines a setof parameters which are evaluated in order to select the most suitable wireless access network for the desired service.Parameters are divided into categories: service parameters, network parameters, and user’s preferences. The category ofservice parameters refers to information of service requirements; the category of network parameters is related to thecurrent network conditions; and finally, the last category of user’s preferences represents relevant values for some factorsas price and QoS level. In the case of user’s preferences, these parameters are defined as weight. Users may specify theimportance or weights of each parameter which sum to 1, i.e., if the price is more important than the level of quality ofservice, then the user might define the values for price and QoS level as 60% and 40% respectively.

Service parameters used in the utility function are bandwidth (b), delay (d), jitter (j), error rate (e), and price (p). Theseparameters are based on the relevant parameters for each service class defined by 3GPP. However, the concept can extendto include other QoS parameters. Thus, we define the utility function for the access network k for the service s as follow:

Uk(s) = wpricefk,si(p)

+ wQoS

lk(sQoS)

(ec-1)

where lk(sQoS) represents the QoS level provided by service provider k and can be calculated as shown in (ec-2):

lk(si) =fk,si(b) × fk,si(d) × fk,si(j) × fk,si(e)

(ec-2)

fk,si(x), for x = b, d, j, e, p represents the evaluation functions for bandwidth, delay, jitter, error rate, and price. Thefunction for each parameter defines the difference between the estimated value for the specific service (Xdefined) and thecurrent value provided by the network (Xcurrent ). wprice and wQoS represent the user’s preferences for price and quality ofservice respectively.

Some considerations are taken into account when evaluation function is calculated for each parameter. For bandwidth,our main interest is the network to provide the best bandwidth for the service. The evaluation function f (xcurrent , xdefined)defines twopossible conditions for the evaluation of each parameter.We specify a value of zero if the network cannot providethe QoS level; otherwise, the evaluation function gets a value from the current and defined values. Defined values representthe minimum (bmin) andmaximum (bmax) values for bandwidth required to maintain the service with an acceptable quality.Consequently, we consider fk,si(x) for bandwidth as shown in (ec-3):

fk,s(b) =

0, bcurrent < bminfbcurrent , bdefined

, bmin < bcurrent < bmax

fbcurrent , bdefined

=

1

1 + e

(bmax+bmin)

2 −bcurrent . (ec-3)

574 A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583

The other parameters like delay, jitter, error rate and price need the network to provide the minor possible value. Ergo,we consider fk,s(x) as follow (ec-4):

fk,s(x) =

0, xcurrent > xmaxfxcurrent , xdefined

, xmin < xcurrent < xmax

fxcurrent , xdefined

=

1

1 + e

(xcurrent−xmin)xmin

. (ec-4)

The final value of the utility function will be in an interval from 0 to 1. This first evaluation reduces the set of candidatenetworks. If the QoS level that the network can provide is almost the same as the required level, the utility function will bealmost 1. So, the selected access network will be determined by:

WSPsel(s) = max(Uk(s)). (ec-5)We consider that the price is divided in small units (such as units of bandwidth or unit time period). As the user can

connect to any WSP for any service through their multi-mode devices and he/she can leave anytime, we can calculate theutility for user u for the service received by provider k during a time t and a price c as (ec-6):

Uuk(si) =

T−t=1

lWSPsel(si)

t −

T−t=1

lWSPsel(si) × c

t . (ec-6)

If the user was connected to n WSPs, the user’s utility for the complete service can be given by,

UuTotal(si) =

n−x=1

Uux(si). (ec-7)

The degree of user satisfaction therefore can be calculated based on user cost-performance expectation.We can calculatethe degree of satisfaction of user u for the service (Si) received by providerWSPsel during a time t and a price c as

Gusatisfaction (si,WSPsel) =

UWSPselT∑

t=1

lWSPsel(si) × c

t

. (ec-8)

This function relates the QoS level and price provided to end-users, and rates satisfaction on a 0 to 1 scale. Finally, thedegree of user’s satisfaction for the complete service can be given by:

GuTotal(si) =

n∑x=1

Gusatisfaction(si, x)

n. (ec-9)

3.3.2. User connection profileThis component is based on the idea that, if the user has a regular pattern, the locations and services that he/she accesses

are recorded. These recordswill give a certain intelligence levelwhich can be used in the selection process, sowe could avoidor reduce the processing time that the equipment has to spend in calculations during the selection activity. The learningmodel we propose is based on a track record, named User Connection Profile (UCP), which stores all information of theselection decisions made when the user accessed to services. UCP is formed by several fields as shown in Table 2.

UCP is applied on the user side as the number of service records can grow considerably according to each user’s activitythus overloading the network with a considerable delay in the network selection procedure in handover situations, whichcould affect the quality of delay-sensitive services. However, as previouslymentioned, we assume that end-users havemorefreedom connecting dynamically to anyWSP for any service through theirmulti-mode devices and can disconnect or changehis/her user’s preferences at anytime which in turn, could motivate end-users to try new services.

In order to create UCP, early decisions of the selection algorithm are made based on the utility function. When we haveseveral records about connection for that service, the decision ismade based on this profile, and if the selected network doesnot have the required resources for the service then the algorithm applies the utility function.

3.4. Dynamic pricing strategy

Newapplication requirements anddevelopments inwireless access networks are leading theway towards an ‘‘enhanced’’next-generation wireless communication environment. This new environment will surpass the current ‘‘best effort only’’capability and evolve into a multi-service network able to accommodate differentiated classes of service to supportvarious types of applications and business requirements. In this new environment, pricing of wireless services is becomingincreasingly important mainly because it: (a) allows providers to recover their operating costs and finance future capacityexpansions; (b) can lead to a more efficient use of the network resources by providing sufficient incentives to users, and (c)enables the creation of a healthy market environment, where new network services can be introduced and sustained.

A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583 575

Table 2Components of the user connection profile.

Identifier Description

S_ID Service Identifier represents the type of service based on the service classification.N_ID Network identifier depicts the selected network to provide the service.UP User’s preferences represent the weight for each parameter at the moment of selection.

Fig. 2. Representation of the QoSDPA dynamic pricing strategy.

Several authors have applied economical concepts in the development of algorithms for mobile and wirelessnetworks [24]. Service providers consider factors like simplicity, scalability and the feasibility of implementation, in theselection of a pricing schema [25]. On the other hand, users demand pricing and charging schemes more adaptable andaccurate [26].

A few pricing schemes are widely used in the Internet and mobile environments today [27]: access-rate-dependentcharge, volume-dependent charge, or a combination of both. However, these works are based on a static pricing model [28],which is not enough for heterogeneous wireless network scenarios. In static pricing strategies, the price of the differentservices is either fixed or only changed at a specific period of the day. However, heterogeneous wireless networks needmore flexible, adaptable pricing schemes to the ever-changing environment conditions.

The dynamic pricing strategy considers the price as an additional network parameter that can be changed duringrelatively short periods of time to allow the network to operate with the optimum price according to the available resourcesand the existing demand. Consequently, this scheme provides benefits for both users and service providers.

Our approach considers a hybrid strategy where both static and dynamic pricing strategies are used to keep a high user’ssatisfaction level. So, we define a QoS-enabled dynamic pricing strategy that considers a simple method for the definition ofthe adequate tariff for the service. Using both strategies, our model allows WSPs to define a static price for specific periodsof time, and at the same time, applying dynamic pricing strategies that adjust the static price according to the assigned QoSlevel, see Fig. 2. The approach is based on the definition of subjective categories that represent QoS tolerance levels. Thesecategories are included within a profile that we called QoS Satisfaction profile (QSP).

3.4.1. Quality of service satisfaction profile (QSP)QoS satisfaction profile (QSP) is a subjective profile defined by service providers. The profile represents QoS tolerance

levels.Within this profile, the service providers define a nominal price for the servicewhich is the initial price for the service.Additionally, the service providers specify a minimal price in order to protect their profits.

One QSP is defined by each service that a WSP offers to end-users. Each category defined on QSP represents a QoSsatisfaction grade. For example, for a service k,WSP could create three categories: nominal, basic, andminimal, see Fig. 3. Thenumber of categories can vary depending of the service. If the number of categories is high, the complexity of the accountingprocess increases. However, if the number of categories is small, the approach would be similar to flat rate.

3.4.2. Tariff adjustment procedureThe tariff adjustment procedure defines the price that should be charged for the service according to the provider’s

strategy. Three situations are analyzed within the tariff adjustment procedure. The first case is a current networkenvironment where service provider can provide the QoS level defined by the nominal category. In this case, the price forthe service is the nominal tariff.

Tservice = Tnominal.

576 A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583

Fig. 3. Representation of quality satisfaction profile.

The second case represents a current network environment where wireless service provider cannot provide theappropriate QoS level. Thus the tariff is affected and dynamically changed based on the provided QoS level. This tariff couldbe calculated by (ec-10). The nominal tariff is affected by a decrement of price according to the QoS level.

T = Tnominal × Fadjust . (ec-10)

The factor that updates the nominal price is named adjustment factor (Fadjust ). This factor represents the discount appliedto the price for the degradation of the quality of service.

Finally, the last situation defines a network environment where the service provider is able to provide only a level lessthan the minimum specified. In this case, the provider could define any model with the minimal price or in the worst casescenario a free access while the service provider recovers the optimum QoS level. Consequently, the price could be definedas follow:

T = Tminimal.

4. Evaluation and results

In order to evaluate the performance of QoSDPA, we performed several simulations in a discrete-event simulatordeveloped in C++ [29]. The simulationmodel is based on the average of a set of 100 replications, and the adopted confidencelevel for themeanwas 95%. Each replication represents a simulation of 2 hrs of operation of thenetwork. QoSDPA is evaluatedwithin a detailed scenario with two wireless service providers competing. Each wireless service provider controlled twodifferent access networks (Wi-Fi and 3G).

We evaluated the access network selection mechanism to reflect its performance in the maximization of QoS level.Additionally, we evaluated the performance of the proposed dynamic pricing strategy in order to analyze the revenuesobtained by service providers when they apply the proposed strategy. Finally, we analyzed how pricing strategy affects theuser satisfaction level based on the relation price-QoS obtained during the use of the service.

Users generated two types of requests: handover and new connection requests. In the simulation model, we definedthe percentage of handover and new connection considering the distribution of the type of requests based on the studyperformed about user mobility in urban zones of Barcelona [30]. The study classifies the users in groups according to theirspeed. From the mobility’s point of view, the study considers the following user categories: pedestrian, automobile, andpublic transportation. Pedestrianmobility refers to peoplewalking,with an average speed of 4.82 km/h. Automobilemobilityrepresents car passengers with an average speed value of 50 km/h. Public transportation mobility refers to users moving byunderground or bus with an average speed value of 30 km/h and 40 km/h respectively. Results showed that the most usedtransportationmedia in big cities is the car, with values approaching 40%, whereas the public transportation and pedestrianusers had a similar weight of 30%. Nevertheless, a different situation occurs in Barcelona. Results in [30] demonstrated thatvehicles are used only by 20% of the population as a mean of transportation. The other means of transport have a similardistribution about 40%.

We considered the pedestrian mobility as the percentage of new connection requests due to the almost null handoverprobability during the average duration of the service. On the other hand, we consider the percentage of users moving invehicle (20%) and in public transportation (40%) as the percentage of handover requests in the scenario of simulation (60%).

A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583 577

Table 3Services and their requirements.

Type of service Application Requirements

Delay-sensitive Videotelephony Bitrate → 4–25 kbps,Delay →< 200 kbpsAverage duration → 5 minError rate → 10−4

VoIP Bitrate → 4–25 kbpsDelay →< 150 msAverage duration → 3 minError rate → 10−4

Non delay-sensitive File transfer Bitrate → 32–150 kbpsDelay → up to 10 sAverage duration → 1–6 sError rate → 10−6

Web browsing Bitrate → 150–384 kbpsDelay → up to 15 sAverage duration → 1 minError rate → 10−6

Users were uniformly distributed over the service area. We assume that each user has a multi-mode terminal, whichallows them to connect to any of the networks if they are inside the coverage area. Four services were defined based onthe service classes specified by 3GPP. Each service used in the simulation scenario has a set of QoS requirements (Table 3).These requirements are based on the technical specifications of the group services and system aspects of the 3rd. GenerationPartnership Project (3GPP) [5].

WSPs apply different pricing strategies. First,wireless service provider (WSP1)defines a flat-rate strategywhere user paysa fixed price independently of the QoS level. Second, wireless service provider (WSP2) applies the QoSDPA dynamic pricingstrategy. We define two variations of the QoSDPA pricing strategy in order to analyze the performance of our approach. Thefirst strategy (QoSDPA-1) uses aQSP with three categories and the second strategy (QoSDPA-2) usesQSP with four categories.Different discounts are applied in each category. Nominal tariff is the same for both WSPs.

We evaluated the performance of single QNSA and QNSA with UCP (QNSA-UCP). We analyzed the rejected rate, theallocated QoS level, and the latency in handover situations which can affect the continuity of the service as factors thatcan have great impact on the user’s satisfaction level. Moreover, we examined the performance of the pricing strategy incomparison with other models like pricing based on a flat rate. Finally, we compared the impact of each pricing strategy inthe user’s satisfaction level and the number of categories defined by aWSP in the utilities for both end-users andWSPs. Theuser satisfaction is based on the user cost-performance expectation.

4.1. Simulation results

Fig. 4 shows the blocking probability obtained by the QNSA algorithm, considering a multi-service scenario. When thearrival rate is small, all requests were served by QNSA. However, as displayed, when the number of requests increases, QNSAobtained rejected rates less than 0.2. When we used UCP, the average rejected rate was 0.3. Due to the probabilistic valuesin which UCP is based, it obtained higher average rejected rates than QNSA.

Fig. 5 shows the variation of the QoS level with different arrival rate. In this case, we show the evolution of the throughputwhen the arrival rate increases for delay-sensitive (DS-QNSA) and non delay-sensitive (NDS-QNSA) services. When networkoperation initiates, both types of service are admitted in the network and receive the maximum required QoS, since thereare resources available.

As the arrival rates increases, different services are admitted and the network becomes saturated and the resources arereduced in order to accommodate the new sessions and handover requests. In this situation, QNSA allocates an acceptableQoS level for the service requests. Since the network resources are better allocated by QNSA, the degradation of the QoSlevel is minimal.

As initial conclusion,we can observe that the efficiency of the QSNA is due to a better evaluation of the available resourcesand the allocation of the required resources for each type of service. If the user receives a good QoS level his/her satisfactionlevel could be increased and WSPs would maximize their benefits.

Fig. 6 shows the average latency obtained by QNSA and UCP within the scenario. In the beginning of the networkoperation, both single QNSA and QNSA-UCP have a similar behavior. However, when the traffic increases, QNSA-UCP obtainsan average latency higher than 200 ms. Obtained values can affect the continuity of services like VoIP o IPTV that demandslatency times lower than 150 ms. However, when UCP is applied, the latency is reduced down to 116 ms when the trafficis small and almost 150 ms when the traffic increases. Single QNSA reduces the average latency almost 25% in comparisonwith QNSA-UCP in situation of high demand.

578 A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583

Fig. 4. Rejected requests rate obtained by QNSA algorithm in a scenario multi-service.

Fig. 5. The throughput variation forced by each algorithm in the scenario multi-service.

The second part of our evaluation is based on the pricing strategies. We analyze the profits obtained by each serviceprovider based on the pricing strategy applied. Additionally, we evaluated the impact that each pricing strategy has on theuser satisfaction level.

Fig. 7 shows an improvement of almost 100% in per-user utility when a QoSDPA approach is applied compared with anHWAN flat-rate approach, thus confirming that QoSDPA is more effective.

A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583 579

Fig. 6. Latency variation forced by each algorithm in the scenario multi-service.

Fig. 7. Average user’s utility obtained for each pricing strategy.

Fig. 8 shows the average revenue obtained by each pricing strategy within the defined scenario. As noted, the proposeddynamic pricing strategy obtained the best results for a WSP. In the beginning of the network operation, both strategiesobtain similar revenues with a difference of almost 7% of our proposed approach in comparison to flat-rate approach. Dueto almost all network resource is available, networks are used in an indistinct way.

580 A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583

Fig. 8. Percentage of WSP’s revenue according to the pricing strategy used.

Table 4Percentage of WSP revenues according to the adjustment factor.

Flat rate QoSDPA 2 cat. Flat rate QoSDPA 3 cat. Flat rate QoSDPA 4 cat. Flat rate QoSDPA 5 cat.

Low arrival rate 48 52 42.5 57.5 36 64 34 66High arrival rate 28 72 24.3 75.7 21.8 78.2 21.3 78.7

However, as time goes on, the different variants of QoSDPA pricing strategy obtain better revenues and in some casesthe increase goes up to 280% depending on the dynamic strategy applied. On the other hand, the graphic shows that theQoSDPA-2 obtains better results than QoSDPA-1, improving up to 25% the profits obtained by WSP.

Finally, the results obtained regarding the impact that pricing strategies introduce in the user satisfaction level arepresented. The user satisfaction is based on the user cost-performance expectation. Fig. 9 shows the effects introducedby the pricing strategies.

The results show that a flat-rate strategy affects the user satisfaction level considerably as the quality of service decreases.QoSDPA strategy reduces the impact in almost 39% in comparison with strategy based on fixed service plans. RegardingQoSDPA strategies, they have a similar behavior; however, QoSPAMS-2 obtained the best results, improving the usersatisfaction level up to 4% in comparison to QoSPAMS-1 when the QoS level is reduced to the minimal level.

Fig. 10 shows how the number of categories used in the adjustment factor affects the user’s utility function. Results showthat when adjustment factor is configured using 3 categories, the third one raises the user’s average profit more than doublewhen compared with flat rate. You can also see virtually no difference between the fourth and the fifth categories. In thisway, as the number of categories increases, the user’s utility function only shows a slight 7% increase between the thirdand the fourth category. Thus we conclude that an adjustment factor based on three or four categories is best suited for theuser’s utility function.

Finally, Table 4 presents the revenue’s percentage from a provider’s point of view, to implement a flat fee and ourproposed model with different levels of adjustment factors. With these results we conclude that the best number ofcategories to the adjustment factor is 4 in which both the end-user and WSP obtain the best gain.

5. Conclusions

Technological developments and the vision towards integration of emerging and existing technologies suggest theevolution towards a heterogeneous wireless access network environment attempting to give solution to the demands of

A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583 581

Fig. 9. Effects of pricing strategies and arrival rate in the user satisfaction level.

Fig. 10. Effects of adjustment factor and arrival rate in the user’s utility function.

the modern society. Among the demands, we can stand out satisfying a high mobility of the users and the seamless access,without restrictions neither in the coverage area, the access method, or the instant of time.

In this paper, we presented an approach based on dynamic pricing strategies that formulates the interaction betweenproviders and users in amarket-based environment. This solution, namedQoSDPA, is applied into amulti-WSP environment.QoSDPA considers the satisfaction of both service providers and users, valuing that, a user satisfied with a service is willingto use it in the future and to recommend the service to other users, producing an increase of profits for the service provider.

To achieve this,wehavedefined an access network selectionmechanism that attempts to choose the appropriate networkfor each requested service, maximizing the provided QoS level according to the user’s preferences. As a second measure,

582 A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583

when the suitable QoS cannot be fulfilled, we define a dynamic pricing strategy that allows wireless service providers toadapt the price according to the supplied QoS level. Pricing strategy is based on subjective categories that represent QoStolerance levels.

We proved by simulations that our proposed pricing approach helps both users and service providers to increase theirprofits. With the help of analytical computations and simulation, we demonstrated that the proposed dynamic pricingmechanism and QoS-enabled network selection strategy achieve better performance for both service providers and users.This efficiency is based on the provision of quality of services and the adjustment of tariffs applied to the services, improvingthe user’s satisfaction and resulting in an increase of revenues for service providers.

However, there are still some issues that need further study. More accurate tuning of parameters, like gains and losses,to ensure compatibility with the economic dimensions of a wireless service provider and the improvement of the currentcharging and billing scheme are essential. For example, the current charging and billing models are based on a monthly billor prepaid cards and a service contract. However the subscription model to wireless service on a session basis will demandnew charging and billing mechanism.

Other study may concern about the unstable dynamics of the user subscriptions. This kind of dynamics will have impacton the resource management as the users are free to select the service providers when he/she needs a connection. In thiscase,multiple service providerswill be included in an admission controlmanagement.We expect this to be a very interestingcase to analyze and will work on it.

Acknowledgement

This work was partially supported by PROMEP under grant No. promep/103.5/08/4113 provided to the first author.

References

[1] A.K. Salkintzis, G. Dimitriadis, D. Skyrianoglou, N. Passas, N. Pavlidou, Seamless continuity of real-time video across UMTS and WLAN networks:challenges and performance evaluation, IEEE Wireless Communications 12 (3) (2005) 8–18.

[2] A. Salkintzis, D. Skyrianoglou, N. Passas, Seamless multimedia QoS across UMTS and WLANs, in: Proc. IEEE VTC 2005, vol. 4, spring, May–June 2005,pp. 2284–2288.

[3] D. Trossen, H. Chaskar, Seamless mobile applications across heterogeneous internet access, in: Proc. IEEE ICC’03, vol. 2, 11–15May 2003, pp. 908–912.[4] M. Kim, S.-U. Kim, S.-J. Cho, A study of seamless handover service and qos in heterogeneous wireless networks, in: Proc. 9th International Conference

on Advanced Communication Technologies, vol.3, Feb. 2007, pp. 1922–1925.[5] Group Services and System Aspects (3GPP) TS 23.107 v 8.0.0., Technical specification: quality of service (QoS) concept and architecture (Release 8),

December 2008.[6] A. Gizelis, D. Vergados, A survey of pricing schemes in wireless networks, in: IEEE Communication Surveys & Tutorials, vol. PP issue 99, July 25, 2010,

pp. 1–20.[7] M. Mandjes, Pricing strategies under heterogeneous service requirements, in: Proc. IEEE INFOCOM 2003, vol. 2, San Francisco, CA, March 30–April 3,

2003, pp. 1210–1220.[8] J. Strassner, Policy-Based Network Management, 1st ed., Elsevier, San Francisco, CA, ISBN: 1-55860-859-1, 2004.[9] Group Services and System Aspects (3GPP) TS 32.200 v 5.9.0., Technical specification: telecommunication management; charging management;

charging principles, October 2005.[10] Group Services and System Aspects (3GPP) TS 32.240 v 8.5.0., Technical Specification: telecommunication management; charging management;

charging architecture and principles (Release 8), December 2008.[11] M. Koutsopoulou, A. Kaloxylos, A. Alonistioti, L. Merakos, K. Kawamura, Charging, accounting and billing management schemes in mobile

telecommunication networks and the internet, in: Proc. IEEE Communications Surveys & Tutorials, vol. 6 issue 1, 2004, pp. 50–58.[12] F. Ghys, A. Vaaraniemi, Component-based charging in a next-generation multimedia network, IEEE communication magazine 41 (1) (2003) 99–102.[13] M. Van Le, B.J.F. van Beijnum, G.B. Huitema, A service component-based accounting and charging architecture to support interim mechanisms across

multiple domains, in: Proc. Network Operations and Management Symposium, vol. 1, Seoul, Korea, April 2004, pp. 555–568.[14] F. Eyermann, P. Racz, B. Stiller, C. Schaefer, T. Walter, Generic accounting configuration management for heterogeneous mobile networks, in: Proc.

3rd ACM International Workshop on Wireless Mobile Applications and Services on WLAN Hotspots, 2005, pp. 46–55.[15] W. Pan, W. Yue, S. Wang, A dynamic pricing model of service provider with different QoS level in web networks, in: Proc. International Symposium

on Information Engineering and Electronic Commerce (IEEC), IEEE Computer Society, Washington, DC, May16–17 2009, pp. 735–739.[16] X. Yue, Y. Jien, B. Xu, Research on incentive strategy for optimal provisioning and dynamic pricing control of monopoly networks, in: Proc. Chinese

Control and Decision Conference, CCDC 08, July 2–4, 2008, pp. 1181–1185.[17] S. Shakkottai, R. Srikant, Economics of network pricing with multiple ISPs, in: Proc. of IEEE INFOCOM, vol. 1, 2005, pp. 184–194.[18] J. Musacchio, J. Walrand, WiFi access point pricing as a dynamic game, IEEE/ACM Transactions on Networking 14 (2) (2006) 289–301.[19] S.C.M. Lee, J.W.J. Jiang, J.C.S. Lui, C. Dah-Ming, Interplay of ISPs: distributed resource allocation and revenue maximization, in: Proc. 26th IEEE

International Conference on Distributed Computing Systems, vol. 19 issue 2, 2006, pp. 204–218.[20] X. Wang, H. Schulzrinne, Pricing network resources for adaptive applications, in: Proc. IEEE/ACM Transactions on Networking, vol 14 issue 3, June

2006, pp. 506–519.[21] V.A. de Sousa Jr, R.A. de O Neto, F. de S Chavez, L.S. Cardoso, J.F. Pimentel, F.R.P. Cavalcanti, performance of access selection strategies in cooperative

wireless networks using genetic algorithms, in: Proc. 15th World Wireless Research ForumMeeting, WWRF’15, Paris, France, December 8–9, 2005.[22] J. Kim, E. Serpedin, D. Shin, K. Quaraqe, Handoff triggering and network selection algorithms for load-balancing handoff in CDMA-WLAN integrated

networks, in: Proc. EURASIP Journal on Wireless Communications and Networking, vol 2008.[23] G. Fodor, A. Furuskar, J. Lundsjo, On access selection techniques in always best connected networks, in: Proc. ITC Specialist Seminar on Performance

Evaluation of Wireless and Mobile Systems, August 2004.[24] T. Giles, J. Markendahl, J. Zander, P. Zettenberg, P. Karlsson, G. Malmgren, J. Nilsson, Cost drivers and deployment scenarios for future broadband

wireless networks – key research problems and directions for research, in: Proc. IEEE Vehicular Technology Conference Spring Conference, vol. 4, May17–19, 2004, pp. 2042–2046.

[25] T. Li, Y. Iraqi, R. Boutaba, Pricing and admission control for QoS-enabled internet, Computer Networks 46 (2004) 87–110.[26] S. Yaiparoj, F.C. Harmantzis, Dynamic Pricing with alternatives for mobile networks, in: Proc. IEEE Wireless Communications and Networking

Conference WCNC 2004, vol. 2, March 2004, pp. 671–676.

A. Guerrero-Ibáñez et al. / Pervasive and Mobile Computing 7 (2011) 569–583 583

[27] L. Badia, M. Lindström, J. Zander, M. Zorzi, Demand and Pricing effects on the radio resource allocation of multimedia communication services, in:Proc. IEEE Global Telecommunications Conference Globecomm ‘03, vol. 7, Dec. 2003, pp. 4116–4121.

[28] F. Caro, D. Simchi-Levi, Static Pricing for a network service provider, March 2006. Available at SSRN: http://ssrn.com/abstract=915806.[29] Omnet++ Community Site, Omnet ++ 4.1 Discrete event simulation system, http://www.omnetpp.org/.[30] M. Berjeno, F. Magrynià, Comparación de tiempos de trayectos metro-a pie-bici en la zona urbana de Barcelona, Bachelor Thesis, Technical University

of Catalonia, January 2007 (in spanish).