Future Generation Computer Systems ( ) –

Contents lists available at ScienceDirect

Future Generation Computer Systems

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

A resource intensive traffic-aware scheme using energy-awarerouting in cognitive radio networksAthina Bourdena a,1, Constandinos X. Mavromoustakis b,∗, George Kormentzas a,1,Evangelos Pallis c,2, George Mastorakis c,2, Muneer Bani Yassein d,3

a University of the Aegean, Department of Information and Communication Systems Engineering, Samos, Greeceb University of Nicosia, Department of Computer Science, Nicosia, Cyprusc Technological Educational Institute of Crete, Department of Informatics Engineering, Estavromenos, Heraklion, Crete, Greeced Jordan University of Science and Technology, P.O. Box 3030, Irbid 22110, Jordan

h i g h l i g h t s

• New resource intensive traffic-aware scheme, incorporated into an energy-efficient routing protocol.• Cognitive radio routing methodology for energy-efficient manipulation of the secondary nodes.• The supported scheme by the framework enables nodes to have reliable transmission.• The proposed scheme associates the backward difference traffic moments with the Sleep-time duration.• The proposed framework is evaluated for the sharing efficiency using specified parameters.

a r t i c l e i n f o

Article history:Received 28 January 2013Received in revised form7 January 2014Accepted 17 February 2014Available online xxxx

Keywords:Cognitive radioRouting protocolsTVWSTraffic-aware energy conservationEnergy-efficient schemeUbiquitous computing andcommunications

Ad-hoc networks

a b s t r a c t

This paper proposes a resource intensive traffic-aware scheme, incorporated into an energy-efficientrouting protocol that enables energy conservation and efficient data flow coordination, among secondarycommunicating nodes with heterogeneous spectrum availability in distributed cognitive radio networks.The proposed scheme associates the backward difference traffic moments with the Sleep-time durationto tune the activity durations of a node for achieving optimal energy conservation and alleviating theuncontrolled energy consumption of wireless devices. Efficient routing protocol operation, as a matter ofmaximum energy conservation, maximum-possible routing paths establishments and minimum delaysis obtained, by utilizing a signallingmechanism, developed based on a simulation scenario that includes anumber of secondary communication nodes. The validity of the proposed resource intensive traffic-awarescheme and the energy-efficient routing protocol is estimated and verified, by conducting experimentalsimulation tests and obtaining performance evaluation results. The simulation results validated theefficiency of the proposed scheme and the effectiveness of the routing protocol, in terms of minimizingthe energy consumption and maximizing resources exchange between secondary communication nodesin a distributed cognitive radio network.

© 2014 Elsevier B.V. All rights reserved.

∗ Corresponding author. Tel.: +35 722841730; fax: +357 22357530.E-mail addresses: abourdena@icsd.aegean.gr (A. Bourdena),

mavromoustakis.c@unic.ac.cy, cmavrom@gmail.com (C.X. Mavromoustakis),gkorm@aegean.gr (G. Kormentzas), pallis@pasiphae.eu (E. Pallis),gmastorakis@staff.teicrete.gr (G. Mastorakis), masadeh@just.edu.jo (M.B. Yassein).1 Tel.: +30 2273082235; fax: +30 2273082009.2 Tel.: +30 2810379828; fax: +30 2810370311.3 Tel.: +962 2 7201000x43403; fax: +962 2 7201000.

1. Introduction

Cognitive Radio (CR) technology [1] is an emerging communica-tion paradigm that efficiently exploits radio spectrum resources toenable the deployment of future wireless networks. CR networkscomprised communication nodes, capable of adapting their tech-nical characteristics, based on interactions with the surroundingspectral environment. They can sense awide radio spectrum range,dynamically identify locally unused/unexploited frequencies andefficiently access them. This capability opens up the possibility ofdesigning new dynamic radio spectrum access policies with the

http://dx.doi.org/10.1016/j.future.2014.02.0130167-739X/© 2014 Elsevier B.V. All rights reserved.

2 A. Bourdena et al. / Future Generation Computer Systems ( ) –

purpose of opportunistically reusing under-utilized frequencies atlocal level, such as ‘‘television white spaces’’ (TVWS) [2]. TVWScomprise VHF/UHF radio spectrum portions that are either re-sulted by a switchover process from analogue to digital terrestrialtelevision, or are completely under-utilized due to frequency plan-ning principles (‘‘Interleaved Spectrum’’) [3]. Therefore, introduc-tion of CR networks in TVWS represents a disruption to the current‘‘command-and-control’’ paradigm of TV/UHF spectrum manage-ment. The exploitation of CR technology is highly intertwinedwiththe regulation models that would eventually be adopted [4,5] es-pecially in future computing systems. The flexibility in radio spec-trum access phase by CR networks caused new challenges alongwith increased complexity in the design of communication proto-cols at different layers. More specifically, the design and adoptionof efficient routing schemes, is a vital process for such an emerg-ing communication paradigm. CR networks are characterized bycompletely self-configuring architectures [6], where routing ischallenging and different from routing in a conventional wirelessnetwork. A key difference is that spectrum availability in a CR net-work highly depends on primary communication nodes presence.Therefore, a fixed Common Control Channel (CCC) is difficult to beexploited, towards establishing a stable routing path between sec-ondary communication nodes. The specific features of CR networkarchitectures pose new requirements in handling energy efficientresources alongwith an underlying reliable routing scheme. To thisend, this work considers the association of the routing mechanismutilized by CR systems with the traffic volume and the end-to-end mechanism for efficiently sharing the requested resources bynodes.

Energy conservation figures an important aspect for the highperformance deployment in ad-hoc CR networks. On one hand, theEnergy Conservation scheme has to be reactive so that the energylevels of wireless nodes will be tuned, according to the estimatedparameters (i.e. capacity, traffic [7] of the nodes). On the otherhand, an energy-efficient schemehas to take into consideration thebounded end-to-end delays of the transmissions. As the networklifetime is closely related to the transmission characteristics [8] ofa source node to a destination node and the underlying routingprotocol used [9], amechanism that combines the temporal traffic-aware behaviour of the node [10] and the efficient routing schemein an end-to-end path has to be investigated. In [8] the sleep-proxy nodes evaluate the duration of the activity periods of eachnode, according to the capacity and the estimated inter-clusteroverall energy consumed within a time frame. Towards furtherinvestigating the scheme proposed in [8], this work has applied thetraffic model and the characteristics of the volume of the trafficfor a specified time window frame to CR systems, supported bythe Backward Traffic Difference estimation. In order to minimizethe energy consumption the Backward Traffic Differencemeasuresthe volume of the incoming Traffic that is destined for each oneof the nodes within a time window frame. The Backward TrafficDifference [10,11] takes into consideration the repetition of theTraffic and estimates the Backward Difference for extracting thetime duration for which the node is allowed to Sleep.

In this context, this paper elaborates on the design, devel-opment and experimental evaluation of a resource intensivetraffic-aware scheme incorporated into an energy-efficient routingprotocol for distributed CR network architectures. Moreover, thejoint routing and traffic-aware methodologies were never com-bined in the past to offer energy conservation in CR systems. Morespecifically, a signalling mechanism combined with an energy ef-ficient scheme is proposed, based on the Backward Traffic Differ-ence estimationmethodology initially stated in [7]. The goal of thiswork is to achieve energy usage that scales with loading. This ispossible by using the incoming traffic aggregation for each nodeto adjust the volume of the traffic to the estimation of the activity

time period assigned for each node. In addition, this paper elabo-rates to describe the development and assessment through simu-lation, of a novel solution for linear scaling adjustment of energyusage with all loads on each secondary node without any packetloss. The key idea is based on traffic aggregation via a traffic-awaremechanism. This mechanism occurs on each secondary node toobtain an estimation and maximization for the time slot whenthe interfaces of each node are put to sleep. Based on the under-lying routing scheme and the volume of traffic that each nodereceives/transmits, the proposed scheme aims at minimizing theenergy consumption, by applying an asynchronous, non-periodicSleep-time assignment slot to the secondary wireless nodes. Fol-lowing this introductory section, Section 2 elaborates on the re-lated work and research motivation, while Section 3 presents thedesign and development of a novel green-aware routing protocol,offering energy efficient data transition, across secondary commu-nication nodeswith different TVWS availability. In order to achievean energy-efficient methodology, the proposed framework uses atraffic-aware Backward Traffic Difference scheme for estimatingthe duration of the sleep time according to the nodal traversed traf-fic. The proposed scheme can efficiently determine the ON and OFFdurations/periods of each node by adjusting the traffic onto the ac-tivity periods of each mobile node. The proposed scheme then ef-fectively provides a reflection of the activity of the traffic to theoverall energy consumed by nodes. Finally, Section 4 elaborates onthe performance evaluation analysis of the proposed research ap-proach, discussing experimental results and Section 5 concludesthis paper by highlighting directions for future research.

2. Related work and research motivation

Conventional routing algorithms exploited in wireless ad-hocnetworks, enable the optimization of network performance met-rics, such as end to end delay, switching delay and backoff de-lay. A rich literature on conventional routing protocols is availablebased on network-wide broadcast messages, without using any lo-cal hops information. Such approaches are not suited for wirelessCR networks, since there is no support for concurrently consider-ing radio spectrumavailability of secondary communicationnodes,as well as the effect on other primary nodes that share spectrumresources. In a general context, several research approaches havebeen recently proposed in [12–15], towards addressing routing is-sues in CR networking environments. In addition, a routing proto-col is proposed in [16], which is exploited to combine geographicalrouting and radio spectrum assignment, towards avoiding regionswith high presence of primary communication nodes. It also de-termines optimum routing path channel combinations that reducedelays in the network. A spectrum aware data adaptive routing al-gorithm is proposed in [17], where the end to end route selectiondepends on the amount of data to be transferred. Furthermore, theproposed routing protocol in [18] builds a forwarding mesh, basedon a set of available routes to the destination and opportunisticallyadapts during the forwarding process, according to the dynamic ra-dio spectrum conditions.Moreover, a joint approach of on-demandrouting and spectrumband selection is proposed in [19] for CR net-working environments and a delay based metric is used to evalu-ate the quality of alternative routes. Most of the previous schemesare based on on-demand routing protocols and discover paths be-tween source and destination communication nodes.

On the other hand, the routing mechanism has to be strictlyassociated with the Energy-Efficiency when the CR networkingarchitecture hosts wireless nodes requesting spectrum, via whichthe traffic will be transferred. Therefore the routing mechanism incollaboration with an energy-efficient scheme should guaranteethe end-to-end availability of requested resources, whereas itshould be able to significantly reduce the Energy Consumption. In

A. Bourdena et al. / Future Generation Computer Systems ( ) – 3

addition, the mechanism should be able to maintain the requestedscheduled transfers and the entire end-to-end connectivity. Manyrecent measurement studies [11] have convincingly demonstratedthe impact of Traffic on the end-to-end connectivity [20], and thusshowed the impact on the Sleep-time duration and the EnergyConsumption. Measures extracted in real-time using realistictraffic [11,20] have shown that the impact of the responsivenessof the routing scheme in regard to the end-to-end transmissionreliability is significant. Real-time communication networks andmultimedia systems, exhibit noticeable burstiness over a numberof time scales [21,22]. Based on the stochastic traffic modelling,the traffic in most of the cases can be expressed in timeexhibiting fractal-like characteristics [23]. The problem of hostinga scheme which, in collaboration with the routing mechanismused, takes into account the traffic characteristics in order toconserve energy has not yet been explored. The scheme will beable to tune the wireless interfaces of the nodes to the Sleepor Active state according to the incoming Traffic and a modelwhich considers the next Sleep-time duration. Notwithstanding,many Sleep-time scheduling strategies were introduced thatmodel the node transition between ON and OFF states. Existingscheduling strategies for wireless nodes could be classified intothree categories: the coordinated sleeping [24], where nodesadjust their sleeping schedules, the random sleeping [23], wherethere is no certain adjustment mechanism between the nodes inthe sleeping schedule with all the pros and cons [10], and on-demand adaptive mechanisms [25], where nodes enter into theSleep-state depending on the environment requirements whereasan out-band signalling is used to notify a specific node to go to sleepin an on-demand manner.

Although there are many schemes developed addressingdifferent Energy conservation methodologies, the combinationof a traffic-aware scheduling scheme with the routing protocolsupported by the CR networking architecture, has not yet beenexplored. The latter poses a fertile ground for the developmentof new approaches with the association of different parametersof the communication mechanisms, in order to reduce EnergyConsumption. Such schemes are classified into active or passivemechanisms. Active techniques conserve energy by performingenergy conscious operations, such as transmission scheduling andenergy-aware routing. Mavromoustakis et al. in [11] considerthe association of Energy conservation problem with differentparameterized aspects of the traffic (like traffic prioritization) andenable a mechanism that tunes the interfaces’ scheduler to sprawlinto the sleep state according to the activity of the traffic of acertain node in the end to end path in real-time.

The main target of the proposed scheme and the research ap-proach of this paper, is to exploit the incoming Traffic pattern inorder to minimize Energy consumption of secondary communica-tion nodes. The proposed scheme aims to minimize the consump-tion of the Energy of each secondary node in the CR system, bytaking into consideration the repetition pattern of the Traffic aswell as the delay limitation (bounded delay) of each transmis-sion. This reflective scheme considers the time-oriented continu-ity of the incoming traffic and the communication traffic volume(data and control packets) among peers in order to provide theenergy conservation schedules of the communicating secondarynodes. To this end, the proposed scheme estimates the BackwardTraffic Difference for extracting the time duration for which thenodes are allowed to sleep, overcoming at the same time the net-work partitioning problems and consolidating the delay limita-tions of the transmission in the scheme. The latter mechanism isperformed through themodelled framework, taking into consider-ation the overall volumeof traffic that traverses a secondary node—within a specified amount of time (duration window), and reflectsthis mechanism to the energy conservation modelled scheme.

Fig. 1. Secondary communication nodes operating over heterogeneous TVWS.

The proposed scheme, in order to enable further recoverabilityand availability of the requested resources, uses the promiscuouscaching [10] methodology in an opportunistic manner, in order tocache the packets destined for the node with turned-off interfaces(sleep state) onto intermediate nodes. The proposed frameworkand the utilized routing methodology enable, through the Back-ward Traffic Difference estimation, the next Sleep-time duration ofthe recipient node to be adjusted according to the activity durationand the volume of the traffic in collaborationwith the consolidatedrouting mechanism.

3. Energy efficient routing scheme based on backward trafficdifference estimation

The transmission of secondary communication nodes in anad-hoc CR network is based on radio spectrum opportunity,where routing has to take into account the availability of spec-trum in specific geographical locations at local level. Spectrumawareness, route quality and route maintenance issues have tobe investigated for different routing schemes, in order to en-able efficient data transfer across regions with heterogeneousradio spectrum availability, even when the network connectiv-ity is intermittent or when an end to end path is temporarilyunavailable. Fig. 1 illustrates a simulation scenario, where pri-mary nodes operate over specific channels in three geographicalareas (i.e. Areas A, B and C in Fig. 1). Secondary nodes (i.e. 43 nodeswere defined in simulation scenario) opportunistically operate, byutilizing remaining available channels in each geographical area(i.e. TVWS in Fig. 1). It has to be noted here that a CCC does notexist between secondary nodes, which are located in neighbouringgeographical areas (i.e. Areas A, B and C in Fig. 1). In this case, sec-ondary communication nodes, which are positioned in locationswith higher TVWS availability (e.g. locations outside areas A, B andC) operate as intermediate relay nodes, switching between alterna-tive channels. Therefore, such relay nodes enable ad-hoc connec-tions among secondary nodes, located inside areas A, B and C.

Taking into account this simulation scenario, spectrum aware-ness has to be investigated, regarding routing in such an ad-hocCR network, where secondary nodes are prohibited to operate onspectrum bands occupied by primary nodes. The main target ofrouting in this CR networking environment is to provide opti-mal, high throughput data transfer by efficiently selecting the bestrouting paths among secondary nodes. In this framework, a novelrouting protocol has to be adopted, in order to enable routing pathdiscovery capabilities, considering TVWS heterogeneity of differ-ent geographical areas. Route quality issues have also to be inves-tigated, since the actual topology of suchmulti-hop CR networks ishighly influenced by primary nodes behaviour and classical waysof measuring/assessing the quality of end-to-end routes (nomi-nal bandwidth, throughput, delay, energy efficiency and fairness)

4 A. Bourdena et al. / Future Generation Computer Systems ( ) –

Fig. 2. Message exchange process of the proposed routing protocol.

should be coupled with novel measures on path stability. In a gen-eral context, routing in an ad-hoc CR network over TVWS consti-tutes a rather important but yet unexplored problem, especiallywhen a multi-hop network architecture is considered. Therefore,a novel routing protocol is vital to be designed and developed, inorder to overcome the above mentioned challenges, towards es-tablishing and maintaining optimum routing paths, among com-munication nodes with different radio spectrum availabilities.

3.1. Proposed underlying routing mechanism

Secondary nodes located outside geographical areas A, B andC in the above mentioned scenario, are able to operate over allavailable channels (i.e. c.40–c.60) and act as intermediate nodes,connected with a Geo-location database that includes informationregarding TVWS availability. They are also enhanced with routingmechanisms’ capabilities, enabling to determine routing paths be-tween secondary nodes with different radio spectrum availabil-ities in such areas. Towards enabling an optimum data transfer,among secondary communication nodes, a novel routing proto-col was designed, developed and evaluated, by conducting experi-mental simulations. This routing protocol is based on the exchangeof AODV-style messages [26] between secondary nodes, includ-ing two major steps in the route discovery process (i.e. route dis-covery and route reply step). This selection was made due to theunpredictable availability of the TVWS that requires hop-by-hoprouting, by broadcasting discovery packets only when necessary.During the route discovery step, a RREQ (route request) message,including TVWS availability of nodes is sent by the source node toacquire a possible route up to the destination node. Once the des-tination node receives the RREQ message, it is fully aware aboutthe spectrum availability along the route from the source node.The destination node then chooses the optimum routing path, ac-cording to a number of performance metrics (e.g. backoff delay,switching delay, queuing delay, number of hops, throughput) andassigns a channel to each secondary user along the route. It has to

Table 1Pseudocode of the basic steps of the proposed message exchange process.

Initiate New Flow ‘‘f ’’ with evaluation EnUpdate Intermediate Node ‘‘n’’ with neighbour statusk = number of intermediate nodes//Decision of node ‘‘n’’for (i=1; i++; i=k) {

if n = sending node ∥ next-hop node ∥ destination nodethen discard message

elseflow evaluation Eni

if Eni > Enthen flow accommodation

//Flow redirectionelse do

generate and broadcast redirection information messageflow evaluation Eniflow accommodation

until (receive route acceptance)generate and send RREP to source node

}

be noted here, that the evaluation of performance metrics is con-ducted, by each intermediate node during the routing path of theRREQmessage. More specifically, the evaluation of delaymetrics isrepresented as Eni (see Table 1), where E is the end-to-end delay inmilliseconds, while ni represents the ith intermediate node thatserves the flow. Also, En is defined as the delay that occurred dur-ing the RREQ message. In the next step of the proposed process,the destination node sends back a RREP (route reply) message tothe source node that includes information regarding channel as-signment.

Fig. 2 presents the detailed signalling mechanism of theproposed routing protocol for handling both RREQ and RREPmessages. A source node initiates a flow (i.e. New Flow in Fig. 2),transmitting a RREQ message to an intermediate node located ina neighbouring location. This intermediate node determines if it ispossible to accommodate the incoming flow based on informationstemming from theGeo-location database. In case that it is possible

A. Bourdena et al. / Future Generation Computer Systems ( ) – 5

to accommodate it, performance metrics is evaluated and RREQmessage is forwarded to the next hop. When the destination nodereceives this message, it is informed regarding TVWS availabilityalong the routing path from the source node. Destination nodereplies by sending a RREP message to the source node thatincludes relevant information, concerning channel allocation. Suchdata/information is mainly exploited to enable secondary nodes toset their channel of operation along the routing path. When sourcenode receives RREP message, routing path has been establishedand useful data transmission is initiated.

In the case when the intermediate node is not capable to ac-commodate the incoming flow (i.e. new flow in Fig. 2), a redirec-tion mechanism (redirection in Fig. 2) is in charge of informing thesource node, about the neighbouring node, which could possiblyact as an alternative intermediate node. The proposed routing pro-tocol determines a route, only when a source node wishes to initi-ate data flows to a destination node. Routes aremaintained as longas they are needed by the source node and the exploitation of se-quence numbers in the exchange messages guarantee a loop-freerouting process. Furthermore, the proposed routing protocol is areactive one, creating andmaintaining routes only if it is necessary,on a demand basis. Routes are maintained in routing tables, whereeach entry contains information, regarding destination node, nexthop, number of hops, destination sequence number, active neigh-bouring nodes for this route and expiration time of the flow. Thenumber of RREQ messages that a source node can send per secondis limited, while each RREQ message carries a time to live (TTL)value that specifies the number of times this message should bere-broadcasted. This value is set to a predefined value at the firsttransmission and increased during retransmissions, which occur ifno replies are received.

Towards further optimizing the proposed routing protocol, anassigning mechanism was designed and adopted to alleviate ser-vice load of intermediate nodes. This process is adapted to each in-termediate node, which is further able to determine if a neighbournode performs better during the process of routing paths establish-ment. More specifically, when a source node initializes a new flow,by sending a RREQ, the intermediate node is informed regardingthe status of neighbouring nodes from the geo-location databasethrough the CCC. Then, the intermediate node evaluates the newflow (i.e. evaluation of performance metrics ‘‘Eni’’) and encapsu-lates the evaluation results in a message that it is forwarded to allneighbouring nodes. Once neighbouring nodes receive a redirect-ing request, they check its validity with the corresponding flow,ensuring that they are not the source/destination nodes or next-hop nodes of that flow. Then the neighbouring nodes initiate aprocess, in order to evaluate the flow and they send to the inter-mediate node the result of the evaluation through a redirecting re-ply message. Once the intermediate node receives the redirectingreply from several of its neighbouring nodes it then selects the op-timumone, in order to serve/accommodate the incoming flow. Thebasic steps of the proposedmessage exchange process can be sum-marized in the pseudocode of Table 1.

For enabling Energy-Efficiency in the proposed frameworka Backward Traffic Difference (BTD) estimation methodology isused. The main additional contribution is that, in the proposedframework, the BTD estimation is bounded by the delay limitationsof the transmission, whereas it takes into consideration the hop-by-hop link delay as well as the total end-to-end delay of thetransmission. The latter should satisfy the delay requirements ofthe transmission. The designed model guarantees the end-to-endavailability of requested resources while it reduces significantlythe Energy Consumption and maintains the requested scheduledtransfers, in a mobility-enabled communication. The innovationadopted in this scheme is that each secondary mobile node usesdifferent assignment(s) of sleep–wake schedules based on the

incoming traffic difference that each node receives through time.The Sleep-time duration is assigned according to the BTD schemein a dissimilarmanner in order to enhance node’s lifetime,whereasit avoids mutation which, will result in network partitioning andresource sharing losses.

Assume that a mobile secondary node has already used thedepicted routing scheme of the previous section and establishedan end-to-end connection in order to transmit requested con-tent/packets. Routing occurs on the end-to-end basis and eachnode separately runs the traffic-aware mechanism using the BTDas is described in the following section. The mechanism measuresthe traffic that traverses each one of the nodes where, the BTD es-timation through the assigned time-window frame will affect theSleep-time duration and enable Energy conservation onto nodes asconducted simulation experiments show.

3.2. Traffic-aware scheme for energy-efficient transmission

3.2.1. Traffic-driven middleware and supported mechanismsEfficient mobile sharing process is complex because its com-

ponents change in time and space in terms of connectivity, porta-bility, accessibility/availability andmobility. Towards reducing theimpact of these changes, the resource sharing application musthave a context-aware adaptive behaviour. Context-awarenessthrough traffic-aware adaptation is a fundamental concept forpervasive and ubiquitous environments. In collaboration with theproposed routing methodology used, this paper elaborates onthe traffic volume exploitation and manipulation, and its directimpact on the EC mechanism. Traffic-aware policy requires anactive scheme to be applied, through which, the traffic will re-flect a certain impact on the nodes taking into account the ECtrade-offs. Wireless devices should consider the incoming traffic,in order to adapt and reflect a certain feedback according to thetraffic volume to the energy conservation mechanism. A middle-ware, which hosts traffic changes and has a direct impact throughthe estimated scheme presented in the next section using a col-laborative traffic-aware scheme, is shown in Fig. 3. Fig. 3 depicts across layer interaction through a mechanism for traffic-awarenessin an end-to-end manner. In particular, real-time media traffic,such as voice and video typically have high data rate requirementsand stringent delay constraints, whereas wireless nodes gener-ally have limited or momentarily connectivity. The proposed mid-dleware enables data packets to be traversed and manipulatedthrough the utilized Wireless Data Link, Network, and Transportlayers, by considering the traffic awareness mechanism and themodel for volume estimation to be reflected on these layers. Theproposed traffic-aware scheme and the associated mechanismsevaluate (after the bootstrap process of the system) the estimated(quantified as Volume/Capacity) traffic that is destined for eachnode. In Fig. 3 theV i

k denotes the volume of traffic destined for nodek and stored onto node i using the promiscuous caching policy [10].In this way, it enables – through the proposed mechanism – an es-timation for the next sleep duration of the node—as presented inthe next section. This traffic-aware policy and the sleep durationevaluation are performed in an interactive way through the Back-ward Traffic Difference (BTD) using a certain window frame-size.These mechanisms are performed, in order to tune the wireless in-terface of each device to sleep/wake, according to the activity ofeach individual device in the resource exchanging path.

Packet classification methodology was utilized as in [16], inorder to mark the packets that are exchanged whether they aredelay sensitive or not. In turn, if packets are considered as delaysensitive, strict deadlines are applied by the sender, according tothe specifications set in the network. In the case where packetdeadlines cannot be satisfied, then cached packets of nearbynodes, enable recovery using the promiscuous caching [19]. This

6 A. Bourdena et al. / Future Generation Computer Systems ( ) –

Fig. 3. The traffic-aware interactive middleware mechanism with the associated influenced layers in the communication stack.

mechanism enables the resources’ replication and increases theresource sharing reliability [19]. The quantitative mechanismsshown in Fig. 3, are depicted in the following sections with thequantitative analysis.

The proposed scheme introduces high availability capabili-ties for resource sharing allowing for continuous operation andsmoother handling of system outages. The promiscuous cachingmechanism estimates the volume of traffic that is cached on anintermediate (active state) node in the path, in order to mea-sure the volume of traffic that is outage. The traffic-aware mid-dleware that hosts the resource intensive scheme allows a moreflexible system infrastructure that can adapt to dynamic changes inresource sharing application requirements and connectivity con-ditions. As the reflective middleware model is a principled andefficient way of dealing with highly dynamic environments, theproposed scheme yet supports a reflective and flexible adaptationof the traffic volumeV i

k. The traffic is considered in terms of the rep-etition pattern by estimating the Backward Difference for extract-ing the time duration for which the node is allowed to reduce theEnergy consumption by entering the Sleep state during the nexttime slot T . The middleware in collaboration with the proposedrouting scheme enables secondary nodes to exchange efficientlythe requested resources by evaluatingwithin a time framewindowthe incoming traffic volume as well as the incoming traffic that isdestined for these nodes. In order to enable recoverability of theincoming traffic, if a node is in the Sleep state of in no connectiv-ity range, then the traffic is cached using the promiscuous cachingconcept applied onto intermediate nodes in the path. The traffic-aware resource sharing scheme expands a cross-layer interaction(see Fig. 3) for Level 2 Medium Access Level (L2/wireless MACsleep/active time manipulation) and L3 (Layer 3/Network Layer)using the proposed routing methodology. In the proposed mid-dleware there are no strict associations among the tasks and thelayers. The traffic-aware middleware enables the data packets tobe traversed and manipulated through the utilized Data Link, Net-work, and Transport layers by considering the traffic awarenessmechanism and the model for volume estimation to be reflectedon these layers. The proposed traffic-aware mechanism evaluates(after the bootstrap process of the system) the estimated (quanti-fied as Volume/Capacity) traffic that is destined for each node. Inthis way it enables – through the proposed mechanism – an esti-mation for the next slot Sleep duration of the node as presented inthe next section.

The power control is provided by determining the transmitperiods and the associated power level such that the energyconsumed is steadily reduced. To this end, by using the Backwardtraffic-awaremechanismpresented in thenext section, the schemeaims to guarantee the resource sharing stability, whereas at thesame time to offer energy conservation. Since nodes in wirelessnetworks typically rely on their battery energy, the proposed

framework encompassed in a traffic-aware middleware, utilizesa reflective mechanism which hosts a traffic-aware scheme forconserving energy in CR wireless environments. The schemeevaluates the scheduled activity periods of each node, in orderto measure and estimate a ‘safe’ forecast time duration for thescheduled time that each node can safely sleep in order to conserveenergy.

3.2.2. Opportunistic resource sharing using backward differencetraffic estimation for energy conservation

When a source needs to send requested packets or stream ofpackets (file) to a destination where the destination node(s) mayhave moved or is/are set in the Sleep-state, then the requestedinformation will be missed and lost. This implies that, in a non-static multi-hop environment, there is a need tomodel the activityslots that a node experiences in contrast to the requested resourcesin the end-to-end path such that the resources can be efficientlyshared among users, whereas any redundant transmissions andlost packets/streams are avoided. The proposed scheme takes intoconsideration the incoming nodal traffic of the secondary nodesin the CR system, and estimates the Sleep-time duration of thenode according to the Backward Traffic Difference (BTD) using acertain window frame-size. Traffic that is being traversed in a pathis being forwarded on a hop-by-hop basis from one secondarynode to another node to another until the requested traffic reachesthe destination node. On one hand, if node is available and inActive-state, it receives the transfer (for example file) whereas,if the node receiving the file is not the destination, it forwardsthe packet to the destination node via other neighbouring hop-nodes in the path.4 On the other hand, if the next hop-node is notavailable to receive and process within a specified time-frame thetransmission to the next-hop node, then promiscuous caching [10]of the transmitted packet occurs in the path. This is performed inorder to buffer the packets that are intended for the destinationnode. Therefore the proposed traffic-based framework is focusingon the Traffic that is incoming for each node and for a specific time-window T . Data packets will be transferred from a source node toa destination node, according to the proposed routing proceduredescribed above. The spatial characteristics and the associatedmodelled traffic monofractality properties [10] were taken intoconsideration for modelling the energy schedules for a certaintime-frame in order to enable Energy Conservation. The trafficand the monofractal characteristics of it were considered in orderto enable greater associativity with the self-similar behaviourexpressed in [10,11], where the window of the traffic duration is

4 The path is constructed according to the routing scheme explored in theprevious section.

A. Bourdena et al. / Future Generation Computer Systems ( ) – 7

tw . In this work we consider the window to betw(s, d) = { lim

t→k·tFn(t) ∈ t(s, d) : RN(t) ≈ RN(t − kτ) ∀k < 2} (1)

where tw(s, d) is the time window measure for the multipathpair source–destination model and where the limit of it should bebounded into kτ time duration for the determined window size. kshould be less than 2 in order to satisfy themonofractality propertyof the repetition index RN(t) of the incoming traffic [10].

When nodes in the transmission path are expecting trafficthey keep their communication network interfaces in Active-state for time t . This means that if the transmission will delaywith dt > tactive where tactive is the active time duration of thewireless nodal interface, then nodes may set their interfaces toan Energy conservation state (Sleep-state). In this respect, thescheme enables the promiscuous caching [10] to be enabled. Thepackets that are destined for the certain node can be cached for aspecified amount of time (as long as the Node (i) is in the Sleep-state) in the 1-hop neighbour node (Node (i − 1) that is in Activestate) in order to be recoverable when the referred node entersthe Active state. When certain node has incoming traffic then thenode remains active for prolonged time. As a showcase this worktakes the specifications of the WiMax IEEE 802.16e (specificationsv.2005) [27] that are recommending the duration of the forwardingmechanism that takes place in a non-power saving mode lays inthe interval 0.1 ns < τ < 1 ps. This means that approx. 80 timesin ams the communication’s triggering action between nodes mayresult a problematic end-to-end transmission reliability/accuracy.Adaptive Dynamic Caching [8] takes place and enables the packetsto be ‘‘cached’’ in the 1-hop neighbouring nodes. Correspondingly,if node is no-longer available due to sleep-state in order toconserve energy (in the interval slot T = 0.08µs), then the packetsare cached into an intermediate node with adequate capacityequals to: Ctf ,k(s)(t) > Ctf ,i(t), where Ctf > α · Ci; where αi isthe capacity adaptation degree [10] based on the time duration ofthe capacity that is reserved on node N of Ck; where Ctf ,k(s)(t) isthe needed capacity where i is the destination node and k is thebuffering node (a hop before the destination via different paths).

This work associates also the Backward Difference Trafficmoments with the Sleep-time duration in order to tune the Activedurations of a node according to the transmissions’ activities andthe expected traffic for the next time step. This is performed viathe BTD estimation which enables the capacity of the traffic C(t)that is destined for the Node i in the time slot (duration) t , and thetraffic capacity CNi(t) which is cached onto Node (i − 1) for time t ,to directly affect the Sleep-time of a node. The one-level BackwardDifference of the Traffic is evaluated by estimating the differenceof the traffic while the Node (i) is set in the Sleep-state for a period,as follows:∇CNi(1) = T2(τ ) − T1(τ − 1)∇CNi(2) = T3(τ − 1) − T2(τ − 2)...

∇CNi(n+1) = Tn(τ − (n − 1)) − T2(τ − (n − 2))

(2)

where ∇CNi(1) denotes the first moment traffic/capacity differencethat is destined for Node (i) and it is cached onto Node (i − 1) fortime τ , T2(τ ) − T1(τ − 1) is the estimated traffic difference whilepackets are being cached onto (i − 1) hop for recoverability asin [11]. The Traffic Difference is estimated so that the next Sleep-time duration can be directly affected according to the following:δ(C(T )) = Ctotal − C1, ∀Ctotal > C1, T ∈ {τ − 1, τ } (3)where the Traffic that is destined for Node (i), urges the Node toremain active for δ(C(T ))

Ctotal· Tprev > 0, Tprev is the previous Sleep-time

duration ({τ − 1, τ }) of the node. On the contrary with [1,4] thiswork measures the BTD within a certain transmission time-frame.This means that each transmission is bounded by a certain delay

limitation (time-duration tw(s, d)) which cannot be overtaken.When a node receives traffic, the traffic flow tf , can be modelledas a stochastic process [11,20] and denoted in a cumulative arrivalform as Atf = {Atf (T )}T∈N , where Atf (T ) represents the cumulativeamount of traffic arrivals in the time space [0 . . . T ]. Then, theAtf (s, T ) = Atf (T )−Atf (s), denotes the amount of traffic arriving intime interval (s, t]. Hence the next Sleep-time duration forNode (i)can be evaluated as a function of the Traffic that traverses the Node(i) provided that the amount of traffic arriving in time interval (s, t]is measured according to the total aggregated Traffic/Capacity thatthe channel can handle at time t . The next Sleep-time duration forNode (i) can be defined as:

Li(n + 1) =δ(C(T )|Atf (s, T ))

Ctotal· Tprev, ∀δ(C(T )) > 0,

tw(s, d) < 2δij

∆max

(4)

where δij is the delay that the transmission experiences to reachdestination j, ∆max is the max allowed delay-duration that thetransmission cannot overtake. The aggregated traffic destined forNode (i) should satisfy the sups≤T

Ntf =1 Atf (s, T ) − Ctf (T )

, for

traffic flow tf at time T , and Ctf (T ) represents the service capacityof the Node (i − 1) for this time duration. The delay that thetransmission experiences δij should satisfy the δij < dp, where dpis the maximum delay in the end-to-end path from a source to adestination and can be is evaluated as:

dp =

i−1i=0

δi + Ti (5)

where δi is the duration where the requested data was hosted ontoi-node, and T is the transmission delay. Then for obtaining theminimized energy consumed in the path ECt the following shouldbe satisfied:

arg minδij<dp

Ctf (T ) =δij < dp : Ctf (T ) = min f (ECt )

. (6)

Taking into consideration the above stochastic estimations, theEnergy Efficiency EEtf can be defined as a measure of the capacityof the Node (i) over the Total Power consumed by the Node, as:

EEtf (T ) =Ctf (T )

Total Power∀δij < dp. (7)

Eq. (7) above can be defined as the primary metric for the lifespanextensibility of the wireless node in the system.

The basic steps of the proposed scheme can be summarizedin the pseudocode of Table 2. In Table 2 the Algorithm starts byexamining the existing capacity of eachnode andwhether the nodeusing the caching capacity parameter can host a delay-intensivetraffic within a bounded delay requirement (δij < dp). Then, inline 2, the scheme examines whether each one of the nodes hascached capacity (this is the traffic that is destined for a node whichlays in the sleep-state). It then evaluates the traffic difference andthe difference in the traffic volume, and measures the activity ofthe node according to the incoming traffic that was buffered inthe 1-hop neighbouring node. The most importance evaluation inthe pseudocode comes through the estimation of the next Sleep-time duration for Node (i) which is then evaluated as a function ofthe Traffic that traverses the Node (i). This estimation is performed(line 7) and it is subject to the amount of traffic arriving in timeinterval (s, t], and is measured according to the total aggregatedTraffic/Capacity that the channel can handle at time t . The nextSleep-time duration for Node (i) can be defined as in Eq. (4) above.

8 A. Bourdena et al. / Future Generation Computer Systems ( ) –

Table 2Pseudocode of the basic steps of the proposed traffic-aware mechanism for energyconservation.

1: while ((Node (i) that there is C(t) > 0) && (argminδij<dp Ctf (T ) == True)) {2: while (CNi(t) > 0) { //cached Traffic measurement3: Evaluate (∇CNi(1));4: Calc (δ(C(T )) = Ctotal − C1, ∀Ctotal > C1, T ∈ {τ − 1, τ })5: if (Activity_Period =

δ(C(T ))

Ctotal· Tprev > 0),

6: //Measure Sleep-time duration

7: Li(n + 1) =δ(C(T )|Atf (s,T ))

Ctotal· Tprev, ∀δ(C(T )) > 0,

8: tw(s, d) < 2 δij∆max

9: Sleep (Li(n + 1)); //sleep duration for the upcoming slot10: } //while11: }// while

4. Performance evaluation analysis, experimental results anddiscussion

Several experimental tests were conducted, in order to validatethe efficiency of the proposed routing protocol and the resource in-tensive traffic-aware scheme. Performance evaluation resultswereextracted, by conducting exhaustive simulation runs and experi-mentation using the NS-2 [28] and the generated real traffic tracesfor implementing the proposed scenario. The energy consumptionmodel used in the simulation for the calculation of the amountof energy consumed is based theoretically on the specificationsof the WiMax IEEE 802.16e (ver. 2005) [27]. The extracted resultsare characterizing the trade-off issues between the performancein deploying the discussed scenario and the Energy consumptionof each secondary CR node by using the proposed traffic-orientedscheme. Results also encompass comparisons with other existingschemes for the throughput, the reliability and the accuracy of-fered by the proposed framework as well as EC efficiency convey-ing an estimated confidence interval (CI) of approximately 3% <CI < 5%. All confidence intervals were found to be less than 5% ofthe mean values of the certain examined parameters. The mobilitymodel adopted in this work is based on the probabilistic mobil-ity scenario derived by Fractional Random Walk. The probabilisticrandom walk mobility model was derived from the Brownian mo-tion [29], where nodes are moving according to certain probabili-ties with respect to the location and the time.

According to such simulation scenario, a number of data flowsare contending to pass through the same intermediate node, thusevaluation of delay metrics is crucial, for an efficient performanceof the proposed routing protocol. In this context, a number of delaymetrics [19,30], are evaluated, such as end to end delay, backoffdelay, switching delay and queuing delay. End to end delay fromthe source node up to the destination node is computed as theoverall sum of queuing delay and node delay:

DEnd-to-End = Dqueuing + Dnode. (8)

Node delay at an intermediate node i is based on switchingdelay and backoff delay and is computed as follows:

Dnode =

i1

(Dswitching + Dbackoff). (9)

Fig. 4 represents simulation results related with the perfor-mance comparison ofmean end-to-end delay, while the number ofactive flows is increasing for both version of the proposed routingprotocol. It is clear that when routing protocol incorporates the as-signing mechanism and the number of active flows in the networkis small, there is no important advantage, in terms of mean end-to-end delay. However, when the number of active flows is morethan three, intermediate nodes begin to suffer the accumulatingqueue, and flow redirection becomes necessary. Such results also

Fig. 4. Mean end-to-end delay for different number of simultaneous flows.

Fig. 5. End-to-end delay for the 1st flow versus probability of PU presence.

Fig. 6. Average end-to-end delay for ten simultaneous flows versus probability ofPU presence.

show that the mean end-to-end delay is less, in the case of the en-hanced routing protocol, in comparison to the basic version of itwithout incorporating the assigning mechanism.

Fig. 5 depicts simulation results of end-to-end delay for onesingle flow, when the probability of primary user presenceincreases, while Fig. 6 presents the same metric, but in this caseeach point represents the average of end-to-end delay for tensimultaneous flows for a certain value of primary user presenceprobability. From both figures it is clear that when the probabilityof primary user presence is getting higher, delay is increasing,while in the case of the basic routing protocol, delay increase ismore significant in comparison to the enhanced routing protocolincorporating the assigning mechanism. This result is reasonable,since the probability of the presence of an incumbent system isdetected as a route failure, introducing in thisway additional delay.

A. Bourdena et al. / Future Generation Computer Systems ( ) – 9

Fig. 7. Average end-to-end delay versus node distance.

Fig. 8. Number of Hops per flow.

Fig. 7 presents the average end-to-end delay that occurredamong the source and destination nodes as the distance betweenthem is increased. From this figure it is clear that the distanceaffects the delay among nodes. This result is reasonable since thelonger is the routing path, the more numerous are the primarynodes that affect the path, and the more significant are theeffects of the route range/diversity. It is further observed thatthe initial version of the proposed routing protocol adds higherdelays, as the distance is increasing rather than those occurredwhen an assigning mechanism is introduced, resulting in the mostoptimal routing paths between the source and destination nodes.Consequently, the longer is the path, the more significant are theeffects of the route diversity. Finally, Fig. 8 depicts the comparisonamong both versions of the proposed routing protocol, under thenumber of hops that are required, in order to make feasible allrouting paths between source and destination nodes, for each flowset according to the simulation scenario. This comparison resultsthat routing protocol, incorporating the assigning mechanismperforms better, since it makes the decision for routing pathestablishment at every hop.

Furthermore, Fig. 9 illustrates the lifespan of secondary nodes inthe transmission path in contrast to the number of hops. The pro-posed scheme is compared with existing similar schemes, show-ing significant increment in the lifespan extensibility, particularlywhen the number of hops increases. The comparative evaluationillustrates that the proposed routing protocol with the assignedmechanism behaves gradually better, and increases the lifespan ofeach secondary node.

The proposed scheme is also compared with other existingschemes in terms of the remaining energy dissipation of eachsecondary node. Fig. 10 shows the fraction of the remaining en-ergy compared with different EC schemes. In the case of peri-odic sleep and wake methodology, the fraction of the remain-ing energy is dramatically dropped whereas, using the traffic-

Fig. 9. Lifespan of secondary nodes with the number of hops in the transmissionpath.

Fig. 10. Fraction of remaining energy comparisons for different EC schemes.

Fig. 11. The Successful packet Delivery Ratio (SDR) with the end-to-end streamingdelay in the transmission path (ms).

aware sleep-scheduling in contrast to the limitations of the de-lay bounds, the scheme offers gradual consolidation of the reduc-tion of the remaining energy of the nodes. Fig. 11 shows the Suc-cessful packet Delivery Ratio (SDR) with respect to the end-to-endstreaming delay and the delay duration/total delay for k-hops inthe communicating path, for mobile secondary nodes. By compar-ing the proposed schemewith the scheme in [31], the BTD schemeoffers greater SDR in the end-to-end path. This outcome is evidentbecause of the adaptivity in the traffic volume and the recoverabil-ity mechanism hosted by the proposed scheme, which enables thepacket delivery ratio to be kept at high percentage values. Compar-ative results regarding the offered Throughput as well as the end-to-end latency for different fading and mobility models are shownin Fig. 12 respectively. The signal strength and the associated fad-ing characteristics are posing a major factor for the end-to-endreliable transmission, whereas as long as the path might be, the

10 A. Bourdena et al. / Future Generation Computer Systems ( ) –

Fig. 12. The Successful packet Delivery Ratio (SDR) with the total delay in the k-hops path (ms).

Fig. 13. The effects of channel’s fading and no-fadingwith the average throughput.

greater the SDR dissipation. Fig. 13 illustrates that the proposedmethodology shows robustness in the presence of the fadingRayleigh model used. The fading characteristics of the channels af-fect vertically the transmission rate of the channels, whereas theThroughput and the SDR both can be negatively affected. In ourconducted experiments the Rayleigh fading small scale fading of 2%with mean noise factor of 4 dB causes minor Throughput degrada-tion ranging from 7%–32%—in worst case affection. The Through-put remained at significantly high levels due to the recoverabilitypromiscuous caching methodology utilized. Additionally, the pro-posed scheme was evaluated for the end-to-end latency experi-enced by the secondary users since their initial request, for twomobility models, namely Uniform Random Mobility Pattern andRandom Waypoint mobility. The latter, the Random Waypointmodel, uses for the movement of mobile nodes a certain like-lihood based on their location, velocity and acceleration changeover time, whereas, Uniform Mobility uses only the uniform dis-tribution model to denote the next mobility/movement. Fig. 14 il-lustrates the measures of the end-to-end latency in ms with thenumber of mobile users during simulation. It can be easily spottedthat when the number of users increases, the end-to-end latencydecreases vertically, whereas the overall SDR remains at increas-ingly high levels.

Fig. 15, shows the Complementary Cumulative DistributionFunction (CCDF) for the resource sharing reliability with the num-ber of secondary users. The Complementary Cumulative Distribu-tion Function (CCDF or – as called – the tail distribution) showsthe distribution for reliable transmission with the number ofsecondary nodes/terminals and the variation by hosting fadingcharacteristics during transmission. Fig. 15 illustrates that theCCDF remains at high levels when the number of secondary usersis increasing. This occurs due to the cooperative promiscuous

Fig. 14. End-to-end latency in ms with the number of mobile users duringsimulation.

Fig. 15. CCDF Sharing Reliability with the Number of sharing secondary-users inthe CR system.

Fig. 16. The SDR with the total number of nodes.

caching, for which, when a missed packet cannot arrive at the des-tination node, it can then be recovered. It is evident that the CCDFcan be significantly increasedwhen the number of secondary usersincreases resulting in the awareness of an established best effortrouting path, which enables the secondary users to receive andtransmit with reliability the requested data. The SDR with the to-tal number of nodes and with the node speed (m/s) are shown inFigs. 16 and 17, respectively. Fig. 17 highlights the benefits of usingthe BTD with delay limitation in the transmission for two traffic-aware schemes. It is clearly illustrated that the proposed schemeoffers an increment in the SDR and overall higher packet deliveryrate. Fig. 18 illustrates the Packet Drop Ratio during simulation ex-perimentation in a log-scale magnitude for two different mobilityframeworks and for nodes without mobility. The Packet Drop Ra-tio is kept at low levels considering that there is mobility whichaggravates the success of packet delivery. It is worth mentioning

A. Bourdena et al. / Future Generation Computer Systems ( ) – 11

Fig. 17. Comparative SDR with the node speed (m/s).

that the Packet Drop Ratio in the presence of the generic proba-bilistic random walk mobility model derived from the Brownianmotion [29] as well as in the presence of mobility with distancebroadcasting of secondary nodes, is not increased showing sig-nificant robustness in the transmission reliability offered by theproposed scheme. Furthermore, Fig. 19 shows the system’sthroughput during simulation which is shown to be significantlyincreased when compared with the scheme in [8]. This is causedby the potential effects of the inter-cluster sleep-proxy schemeintroduced in [8] which is only based on incoming nodal trafficwhere the traffic volume is only considered in themeasured clusterrather than in the end-to-end exchanging path. In turn Fig. 20 illus-trates the average delay in dp bounded transmission with the aver-age number of secondary hop-nodes/users. The proposed schemeshows to slightly outperform the scheme in [8] taking into accountthe variability of delays that are minimized when more secondarynodes are participating in the transmission process, as depictedby Fig. 14, earlier. Moreover in Fig. 21 there is a comparative il-lustration for the SDRs with the transmission rates for three dif-ferent power ratios for CR secondary nodes in decibels (dB mW).Fig. 21 highlights the benefits of using the proposed intensivetraffic-aware scheme, incorporated into an energy-efficient rout-ing protocol, in contrast to the three different power ratios for CRsecondary nodes. Notwithstanding the proposed framework en-ables high transmission rates, when the 33 dB m is reached in thecommunication the SDR drops slightly decreasing the transmis-sion rates over 12 Mbps. Fig. 22 presents the comparative resultsfor the system’s overall EC (µW ) with the number of participat-ing secondary nodes, illustrating the supremacy of the proposedscheme minimizing the system’s overall EC by an average of 33%when compared with [31], and by an average of 17% when com-paredwith the scheme in [8]. The accuracy offered by the proposedframework for obtaining the system’s overall EC is measured witha confidence interval of approximately 3% > CI < 5%. Notwith-standing all confidence intervals were found to be less than 5% oftheir mean values, the system’s overall EC offered by the schemein [8] showed to be slightly above the proposed improved frame-workhosting the underlying efficient routing scheme, and it clearlyoutperforms the scheme proposed in [31].

In Fig. 23 the average energy consumed with the power ratioin decibels (dB) of the measured power referenced to one mW incontrast to the simulation time, is presented. The experimentalevaluationswere extracted in the presence of fading and no-fadingcommunicating obstacles and characteristics as indicated in the in-troduction. The Energy Efficiency (bytes/mW), which is defined asthe service capacity/total energy consumed as in Eq. (8), in con-trast to the delay requests is presented in Fig. 24. Taking intoconsideration the estimations of the previous section, the EnergyEfficiency EEtf is defined as in Eq. (8) as ameasure of the capacity of

Fig. 18. Packet Drop Ratio during simulation experimentation (log-scale).

Fig. 19. System throughput during simulation.

Fig. 20. Average delay in dp bounded transmission.

Fig. 21. SDR with the transmission rates for three different power ratios for CRsecondary nodes in decibels (dBmW).

12 A. Bourdena et al. / Future Generation Computer Systems ( ) –

Fig. 22. Comparative results for the system’s overall EC (µW) with the number ofparticipating secondary nodes.

Fig. 23. Average energy consumed (J/bit) with the simulation time and the powerratio in decibels (dB) of the measured power referenced to one mW.

Fig. 24. Energy efficiency with respect to the delay per request (transmission) inan end-to-end manner.

the Node (i) over the Total Power consumed by the Node. Results ob-tained in Fig. 24 show that the network lifetime can be significantlyprolonged when the Traffic-aware scheme is applied. By compar-ing the results obtained through simulation experiments for thescheme developed in [10] which takes into consideration the re-gional capacity and the remaining capacity of each mobile nodeas well as the comparison with the periodic Sleep/Wake schedul-ing, the proposed scheme offers greater Energy-Efficiency, while itminimizes the delay per request.

5. Conclusions and further research

This work proposes an efficient routing mechanism where,in collaboration with the underlying BTD scheme, it enablesenergy conservation and reliable data flow among secondarycommunication nodes with heterogeneous spectrum availability

in CR systems. The proposed routing scheme establishes an end-to-end optimal path whereas, secondary nodes in CR systemscan efficiently and, in a collaborative manner, share requesteddata/resources. The hosted traffic-aware scheme enables the self-tunability of the sleep schedule of each node to be applied throughthe BTD, measured within a certain transmission time-frame.Within the proposed framework, the bounded end-to-end delayof the transmission is taken into consideration for each secondarynode, aiming to impact the EC through the modelled traffic-awaremechanism. The performance evaluation through simulationshows that the proposed routing scheme in collaboration withthe BTD mechanism, manipulates the energy consumption ofeach secondary node/device effectively, and outperforms incontrast to similar traffic-aware schemes. Moreover, the traffic-aware management scheme can significantly reduce the energyconsumed and can keep the throughput response of the systemat relatively high-levels. The comparative measurements withother similar schemes show that the proposed methodology canefficiently conserve the energy, by offering at the same timesignificantly high SDRs, and can significantly extend the lifetimeof each secondary node in the CR network. Furthermore, efficientrouting protocol operation, as a matter of maximum-possiblerouting paths establishments and minimum delays was validated,by adopting the proposed message exchange mechanism thatwas developed based on the simulation scenario defined above.Towards evaluating the performance of the routing protocol in thisrespect, a large number set of experimental tests was conductedunder controlled simulation conditions, where various secondarysystems were concurrently/simultaneously communicating in ad-hoc connections, accessing the available TVWS. The obtainedexperimental results verified the validity of the proposed routingmechanism, towards enabling an efficient communication amongsecondary nodes located in areas with different TVWS availability.

Further streams in our on-going research include the evaluationof our scheme using real-time measurements and real-time veri-fication using the existing infrastructure. Issues to be consideredare the topology formation, using social collaboration as well asgeographical profiles, in order to face potential partitioning prob-lems.Moreover, the usage of traffic engineeringmodels, in order toexplicitly express the behaviour of such dynamically changing sce-narios. Additionally we are working towards the expression of ourschemewith the combined infinitesimal perturbation analysis andapply a stochastic algorithm into the performance gradient of thesystem. Finally, several optimization methods will be adopted, to-wardsminimizing delays occurred during the routing paths of dataflows and maximizing the number of established paths. This com-prises an open-end research issuewithmany research concepts forfuture examination.

Acknowledgements

This section is dedicated to host our thanks to the reviewers fortheir valuable comments, which helped us to significantly improveour paper presentation and quality of our research work.

References

[1] I.F. Akyildiz, W.Y. Lee, M.C. Vuran, S. Mohanty, NeXt generation/dynamicspectrum access/cognitive radio wireless networks: a survey, Comput. Netw.50 (13) (2006) 2127–2159.

[2] A. Bourdena, E. Pallis, G. Kormentzas, G. Mastorakis, A prototype cog-nitive radio architecture for TVWS exploitation under the real timesecondary spectrum market policy, Phys. Commun. (2013). Elsevier.http://dx.doi.org/10.1016/j.phycom.2013.11.003.

[3] A. Bourdena, E. Pallis, G. Kormentzas, C. Skianis, G. Mastorakis, Real-timeTVWS trading based on a centralised CR network architecture, in: Proc. IEEEGlobecom2011, IEEE InternationalWorkshop on Recent Advances in CognitiveCommunications and Networking, Texas, Houston, USA, 05–09 December,2011, pp. 994–999.

A. Bourdena et al. / Future Generation Computer Systems ( ) – 13

[4] A. Bourdena, E. Pallis, G. Kormentzas, G. Mastorakis, A radio resourcemanagement framework for TVWS exploitation under an auction-basedapproach, in: Proc. IEEE 8th International Conference on Network andService Management, IEEE CNSM2012, Las Vegas, USA, 22–26 October, 2012,pp. 204–208.

[5] A. Bourdena, E. Pallis, G. Kormentzas, H. Skianis, G. Mastorakis, QoS provision-ing and policy management in a broker-based CR network architecture, in:Proc. IEEE Globecom 2012, Anaheim, California, USA, 03–07 December, 2012.

[6] E. Hossain, D. Niyato, Z. Han, Dynamic Spectrum Access and Management inCognitive Radio Networks, first ed., Cambridge University Press, 2009.

[7] C.X. Mavromoustakis, Using backward traffic difference estimation forefficient energy saving schedules in wireless device, IEEE CommSoft E-Lett.1 (1) (2012) 1–6.

[8] Marios C. Charalambous, Constandinos X. Mavromoustakis, Muneer BaniYassein, A resource intensive traffic-aware scheme for cluster-based energyconservation in wireless devices, in: Proceedings IEEE 14th InternationalConference on High Performance Computing and Communications (HPCC-2012) of the Third International Workshop on Wireless Networks andMultimedia (WNM-2012), 25–27 June 2012, Liverpool, UK, pp. 879–884.

[9] H. Shpungin, Energy efficient online routing in wireless ad hoc networks, in:The 8th Annual IEEE Communications Society Conference on Sensor,Mesh andAdHocCommunications andNetworks, IEEE SECON’11, Salt Lake City, UT, USA,2011.

[10] C.X. Mavromoustakis, K.G. Zerfiridis, On the diversity properties of wirelessmobility with the user-centered temporal capacity awareness for EC inwireless devices, in: Proceedings of the Sixth IEEE International Conference onWireless andMobile Communications, ICWMC 2010, September 20–25, 2010-Valencia, Spain, pp. 367–372.

[11] C.X. Mavromoustakis, C.D. Dimitriou, G. Mastorakis, Using real-time backwardtraffic difference estimation for energy conservation in wireless devices,in: Proceedings of the Fourth International Conference on Advances in P2PSystems, AP2PS 2012, September 23–28, 2012, Barcelona, Spain, pp. 18–23.

[12] H. Khalife, S. Ahuja, N. Malouch, M. Krunz, Probabilistic path selection inopportunistic cognitive radio networks, in: Proc. IEEE Globecom, Novem-ber–December 2008.

[13] H. Ma, L. Zheng, X. Ma, Y. Luo, Spectrum aware routing for multihop cognitiveradio networks with a single transceiver, in: Proc. Cognitive Radio OrientedWireless Networks and Comm., CrownCom, 15–17 May 2008, pp. 1–6.

[14] H.-P. Shiang, M. van der Schaar, Delay-sensitive resource management inmulti-hop cognitive radio networks, in: Proc. IEEE DySPAN, October 2008.

[15] C. Xin, L. Ma, C.-C. Shen, A path-centric channel assignment framework forcognitive radio wireless networks, Mob. Netw. Appl. 13 (5) (2008) 463–476.Kluwer.

[16] K.R. Chowdhury, M. Di Felice, Search: a routing protocol for mobile cognitiveradio ad-hoc networks, Comput. Commun. J. 32 (18) (2009) 1983–1997.

[17] R. Pal, Efficient routing algorithms formulti-channel dynamic spectrum accessnetworks, in: Proc. of IEEE DySPAN 2007, Dublin, Ireland, 2007, pp. 288–291.

[18] I. Pefkianakis, S.H.Y. Wong, S. Lu, SAMER: spectrum aware mesh routingin cognitive radio networks, in: Proc. of IEEE DySPAN 2008, October 2008,pp. 1–5.

[19] G. Chen, W. Liu, Y. Li, W. Cheng, Spectrum aware on-demand routing incognitive radio networks, in: Proc. of IEEE DySPAN 2007, Dublin, Ireland, 2007,pp. 571–574.

[20] C.X. Mavromoustakis, H.D. Karatza, Real time performance evaluation ofasynchronous timedivision traffic-aware and delay-tolerant scheme in ad-hocsensor networks, Int. J. Commun. Syst. (IJCS) 23 (2) (2010) 167–186. Wiley.

[21] J. Yu, A.P. Petropulu, Study of the effect of the wireless gateway on incomingself-similar traffic, IEEE Trans. Signal Process. 54 (10) (2006) 3741–3758.

[22] X. Zhuang, S. Pande, A scalable priority queue architecture for high speednetwork processing, in: Proc. 25th IEEE International Conf. ComputerCommun., INFOCOM’06, 2006, pp. 1–12.

[23] H. Cunqing, P.Y. Tak-Shink, Asynchronous random sleeping for sensornetworks, ACM Trans. Sensor Netw. (TOSN) (2007).

[24] Q. Cao, T. Abdelzaher, T. He, J. Stankonic, Towards optimal sleep schedulingin sensor networks for rare-event detection, in: The 4th InternationalSymposium on Information Processing in Sensor Networks, 2005.

[25] I.H. Jawhar, J. Wu, Resource scheduling in wireless networks using directionalantennas, EURASIP J. Wirel. Commun. Netw. 21 (9) (2010) 1240–1253.

[26] RFC: http://www.ietf.org/rfc/rfc3561.txt.[27] WiMAX Forum Specification (2011-09-20), IEEE to WiMAX Forum, last

accessed 29th October 2012.[28] NS-2 Simulator, at: http://www.isi.edu/nsnam/ns/ (last accessed: October

2012).[29] T. Camp, J. Boleng, V. Davies, A survey of mobility models for ad hoc

network research, Wirel. Commun. Mobile Comput. (WCMC) 2 (5) (2002)483–502. Special Issue on Mobile Ad Hoc Networking: Research Trends andApplications.

[30] G. Cheng, W. Liu, Y. Li, W. Cheng, Joint on-demand routing and spectrumassignment in cognitive radio networks, in: IEEE International Conference onCommunications, ICC’07, 2007, pp. 6499–6503.http://dx.doi.org/10.1109/ICC.2007.1075.

[31] I. Rhee, A. Warrier, J. Min, L. Xu, DRAND distributed randomized TDMAscheduling for wireless ad-hoc networks, in: Mobile Ad Hoc Networking andComputing, 2006, pp. 190–201.

Athina Bourdena received her B.Sc. from the Departmentof Applied Informatics and Multimedia of TechnologicalEducational Institute of Crete in 2007 and herM.Sc. in Dig-ital Communications and Networks from the Departmentof Digital Systems of the University of Piraeus in 2009. Sheis currently a Ph.D. candidate in the field of radio resourcemanagement in cognitive radio networks at the Universityof the Aegean in theDepartment of Information&Commu-nication Systems Engineering in Greece.

Constandinos X. Mavromoustakis is currently an Asso-ciate Professor at the Department of Computer Science atthe University of Nicosia, Cyprus. He received a five-yearDiploma in Electronic and Computer Engineering fromTechnical University of Crete, Greece,M.Sc. in Telecommu-nications from University College of London, UK, and hisPh.D. from the department of Informatics at Aristotle Uni-versity of Thessaloniki, Greece. He serves as the Chair ofC16 COMSOC chapter of the Cyprus IEEE section. His re-search interests are in the areas of spatial and temporalscheduling, energy-aware self-* scheduling and adaptive

behaviour in wireless and multimedia systems.

George Kormentzas received the Diploma in Electricaland Computer Engineering and the Ph.D. in ComputerScience both from the National Technical University ofAthens (NTUA), Greece, in 1995 and 2000, respectively. Heis currently serving as an Assistant Professor in ICSD De-partment at the University of the Aegean. His research in-terests are in the fields of traffic analysis, network control,resource management and quality of service in broadbandnetworks. He has published extensively in internationalscientific journals, edited books and conference proceed-ings. He is amember of pronounced professional societies,

an active reviewer and guest editor for several journals and conferences.

Evangelos Pallis received his B.Sc. in Electronic Engineer-ing from Technological Educational Institute of Crete in1994, his M.Sc. in Telecommunications from Universityof East London in 1997 and his Ph.D. in Telecommunica-tions from University of East London in 2002. He currentlyserves as an Associate Professor in Technological Educa-tional Institute of Crete in Applied Informatics and Mul-timedia Department, and acts as the director of Researchand Development of Telecommunication Systems Labora-tory. His research interests are in the fields of wirelessbroadband and mobile networks and network manage-

ment. He has more than 80 publications in international scientific journals, con-ference and workshop proceedings.

George Mastorakis received his B.E. in ElectronicEngineering from UMIST in 2000, his M.Sc. in Telecom-munications from UCL in 2001 and his Ph.D. in Telecom-munications from University of the Aegean in 2008. Heis serving as an Assistant Professor in the Department ofCommerce and Marketing and as a research associate inResearch & Development of Telecommunications SystemsLaboratory at Technological Educational Institute of Cretein Greece. His research interests include cognitive radionetworks, networking traffic analysis, radio resourceman-agement and energy efficient networks. He has more than

60 publications in various international conferences proceedings, workshops, sci-entific journals and book chapters.

Muneer Bani Yassein received his B.Sc. degree in Com-puting Science and Mathematics from Yarmouk Univer-sity, Jordan in 1985 andM.Sc. in Computer Science, fromAlAl-Bayt University, Jordan in 2001 and his Ph.D. degree inComputer Science from the University of Glasgow, UK, in2007. He is currently an associate professor in the Depart-ment of Computer Science at Jordan University of Scienceand Technology (JUST), His research includes the develop-ment/analysis of the performance of probabilistic flood-ing behaviours in MANETs and the refinement of servicediscovery and routing algorithms for mobile device com-

munications in heterogeneous network environments. Dr. Bani Yassein is mem-ber of IEEE and he is a member of the technical programs of several journals andconferences.