paper eﬃcient resource utilization for heterogeneous ......appropriate for future connection....

IEICE TRANS. COMMUN., VOL.E96–B, NO.6 JUNE 20131577

PAPER

Efficient Resource Utilization for Heterogeneous Wireless PersonalArea Networks

Abolfazl MEHBODNIYA†∗a), Member, Sonia AISSA††b), Nonmember, and Fumiyuki ADACHI†††c), Fellow

SUMMARY Wireless personal area networks (WPANs) will play animportant role in next-generation communication networks. Currently, twotechnologies are being considered for the physical layer of WPANs, basedon the two ultra wideband (UWB) standards, namely, multiband orthogonalfrequency division multiplexing (MB-OFDM) UWB and direct-sequence(DS) UWB. The coexistence issue of these two types of WPANs in thesame coverage area, raises new issues and introduces new problems whichshould be dealt with to avoid performance degradation. In particular, effi-cient radio resource management (RRM) in such environments is challeng-ing. Indeed, the coexistence of heterogenous UWB based WPANs (UP-ANs) has an ad hoc nature, which requires RRM approaches that are dif-ferent from traditional infrastructure-based ones. In this paper, we proposenew algorithms for two RRM modules in heterogeneous UPANs, namely,radio access technology (RAT) selection and vertical handoff (VHO). Toimprove the overall performance of the system, our design considers pos-sible narrowband interference (NBI) in the environment as well as the linkoutage probability, in the decision process. We also provide an analyticalmodel based on a 4D Markov process to study the system in equilibriumand derive the performance metrics, namely, the new-call and handoff-callblocking probabilities, throughput and average carried traffic. Numericalresults and comparisons show that our design achieves enhanced perfor-mance in terms of throughput and grade of service (GoS).key words: radio resource management, heterogeneous networks, MB-OFDM, DS-UWB, vertical handoff, grade of service

1. Introduction

Wireless personal area networks (WPANs) are an impor-tant part of the future vision of seamless communications.WPANs provide short-range ad hoc wireless connectivity.Currently, there are two ultra wideband (UWB)-based pro-posals for the physical layer in WPANs: multiband orthog-onal frequency division multiplexing (MB-OFDM) [1] anddirect-sequence (DS) UWB [2]. In these networks, radioresource management (RRM) will play a major role in theproper assignment of resources and reducing interference.RRM consists of several modules, e.g., call admission con-trol [3], radio access technology (RAT) selection, vertical

Manuscript received June 12, 2012.Manuscript revised January 2, 2013.†The author was with INRS-EMT, University of Quebec, Mon-

treal, QC, Canada.††The author is with the INRS-EMT, University of Quebec,

Montreal, QC, Canada.†††The author is with the Department of Electrical and Com-

munication Engineering, Tohoku University, Sendai-shi, 980-8579Japan.

∗Presently, with the Department of Electrical and Communica-tion Engineering, Tohoku University, Sendai-shi, 980-8579 Japan.

a) E-mail: [email protected]) E-mail: [email protected]) E-mail: [email protected]

DOI: 10.1587/transcom.E96.B.1577

handoff (VHO) [4]–[6], power control, etc. RRM becomesespecially complicated in the case of heterogeneous UWBbased WPANs (UPANs), where two different technologiesneed to coexist in the same coverage area. Most stud-ies on RRM in heterogeneous networks, consider the prob-lem from a general perspective and most assume the con-ventional microcell-macrocell scenario where the networkshave different and partially overlapping coverage. To thebest of our knowledge, there have been no research resultson the RRM in heterogeneous WPANs, particularly whenthe PHY layer has different UWB technologies.

Surveying the previous studies that are most related toour topic, it is worth mentioning [7], [8] and [9]. In [7],an analytical model that considers different classes of callswas proposed for handoff management in cellular networks.In fact, the authors considered a priority handoff algorithmbased on channel reservation for real-time and non-real-timeservices in homogeneous cellular networks. The researchwork in [8] discussed VHO in fourth-generation heteroge-neous networks, by considering different parameters for theVHO policy design such as service type, costs and user con-ditions. In [9], Markov chain is used to model the RAT se-lection problem for a heterogeneous network consisting ofTDMA and wideband CDMA networks. Four different RATselection policies are studied in this model. This work doesnot consider VHO between the layers, and neither does ituse any of the system specifications from the PHY layer forthe RAT decision process.

In this paper, we propose algorithms for RAT selectionand VHO in heterogeneous UPANs. Our RAT selection al-gorithm considers the special characteristics of MB-OFDMand DS-UWB networks as well as the outage probability ofthe communication links to decide which network is mostappropriate for future connection. Technically, we call thecombination of these two networks, the system. Two differ-ent classes of calls are considered and decisions are madebased on the type of the call, the device (DEV)’s batterycondition and the DEV’s mobility. Our VHO procedure in-directly contributes to the overall system load balancing andachieving the grade of service (GoS) requirements. GoS isa measure of system performance based on the new-call andhandoff-call blocking probabilities. We have integrated anarrowband interference (NBI) verification policy into theRAT selection and VHO modules to adaptively compensatefor the effects of a reduction in GoS. Furthermore, a four-dimensional (4D) Markov chain is used to derive the perfor-mance metrics, namely, the new-call and handoff-call block-

Copyright c© 2013 The Institute of Electronics, Information and Communication Engineers

1578IEICE TRANS. COMMUN., VOL.E96–B, NO.6 JUNE 2013

ing probabilities, throughput and average traffic carried byeach network. Please note that our proposed RAT selectionand VHO algorithms are fundamentally different from con-ventional algorithms from two points of view. First, we haveproposed our algorithms for a new and original scenarioof heterogeneous UWB-based wireless personal networks,which is different from conventional heterogeneous networkscenarios. Conventional algorithms mostly aim at enhanc-ing the wireless coverage and maintaining the connectiv-ity which differentiate them from our work. Second, ourassumptions and triggering conditions are designed basedon the technical specifications of DS-UWB and MB-OFDMradios, as well as the type of connection and so are com-pletely different from conventional algorithms. As a resultour proposed algorithms are not directly comparable to con-ventional algorithms. The only algorithm in the literaturewith partial resemblance to our work is the service-base al-gorithm in [9], which we use as a benchmark for compar-isons in Sect. 4.

The remainder of the paper is organized as follows. InSect. 2, the system model and the proposed RAT selectionand VHO procedures are discussed. Section 3 presents thesystem analysis based on a 4D Markov model. Section 4addresses the numerical results and comparisons. Finally,concluding remarks are drawn in Sect. 5.

2. System Model and the Proposed RAT Selection andVHO Procedures

In this section we first describe our system model. Then, wepresent an overview of the UWB standards’ characteristicsand describe the proposed modules.

2.1 System Model

We consider a heterogeneous system composed of DS-UWBWPAN (DS-UPAN) and MB-OFDM UWB WPAN (MB-UPAN). Figure 1 illustrates the system model. In terms oftopology, each of these two networks can extend its cov-erage by forming several piconets and different DS-UPANand MB-UPAN piconets can have overlapping coverage. Apiconet consists of several DEVs, with one DEV assumingthe role of the piconet coordinator (PNC). In our system, weassume DEVs with low and high mobility, and consider thatthe DEVs are all multimode and support connection to eitherof the UPANs. Both networks can accept class-A (requiringhigh data rate) and class-B (requiring low data rate) calls.Here we use the general term, call, for a connection requestby a DEV. A call is also classified as being in a battery-critical condition (BC) or in a non-battery-critical condition(NBC). We assume that a session describer module is avail-able in the PNC of each network and is responsible for col-lecting the aforementioned information and for sensing thepresence of NBI in the environment. We also assume thatthe PNC in the DS-UPAN and the one in the MB-UPANhave the capability to exchange system-related information.Please note that this scenario is different from conventional

Fig. 1 System model.

infra-structured networks in the sense that only in the ini-tial phase does the PNC communicate with the DEVs, andthat for synchronization purpose. In the next phase, differ-ent transceiver pairs start their transmission directly. As apractical application for this scenario, we can mention thehigh-speed cordless connections to peripherals in office en-vironments, instant conferencing and automatic data syn-chronization on the move.

2.2 DS-UWB and MB-OFDM Characteristics

Before presenting the proposed RAT and VHO algorithms,it is important to recall the distinct characteristics of DS-UWB and MB-OFDM, which are taken into account inour design. Regarding the advantages and disadvantagesof these two technologies, in the literature, there are sev-eral comparative research works from different perspectives:mutual interference, robustness to multipath, performance,complexity, achievable range-data rate for WPAN applica-tions and effects of mobility. Below, we review some ofthese characteristics.

• Mobility: Not many studies dealt with the impact ofmobility on the performance of UWB systems. In [10],the authors investigated the effect of mobility on theperformance of DS-UWB, by varying the transmissiondata rate up to 500 Mbps. It was observed that the aver-age bit error rate of the system degrades for higher datarates in the presence of mobility. In our system, DS-UWB is hence not suitable for DEVs with high mobil-ity and requesting a class-A connection.

• Interference: In the presence of NBI, MB-OFDM hasan advantage over DS-UWB because of its subcarri-ers, which give it the ability to use the detect and avoidtechnique [11]. In this technique, the NBI frequenciesare first detected, and the subcarriers related to thosefrequencies are then nulled.

MEHBODNIYA et al.: EFFICIENT RESOURCE UTILIZATION FOR HETEROGENEOUS WIRELESS PERSONAL AREA NETWORKS1579

• Receiver Performance: Depending on the techniqueused at the receiver (matched filter, decision feedbackequalizer, etc...), the performance of the two technolo-gies may have a slight difference but, for practical pur-poses, we can consider the bit error rate performanceof both technologies to be similar [1], [2].

• Data Rate: Although MB-OFDM is an efficient tech-nology, it can not reach data rates as high as DS-UWB’s. This is due to its multiband structure, whichrestricts the transmission to some smaller subbands ofthe UWB frequency spectrum [1].

• Power Consumption: MB-OFDM technology leads tohigher power consumption in comparison to DS-UWB,especially at higher data rates, due to use of fast Fouriertransform [1]. Power consumption is a critical issue incertain applications such as wireless sensor networks.

2.3 RAT Selection and Handoff Algorithms

In this section, we first present our RAT selection process,then the proposed VHO procedure is discussed. The goalof these functions is to guarantee stable connection for theDEVs in the system and improved GoS. These two algo-rithms are designed based on the special characteristics ofDS-UPAN and MB-UPAN as reviewed in Sect. 2.2.

2.3.1 RAT Selection

The target network selection for each DEV is done by a co-operation between PNCs in the system. In the proposed al-gorithm, first the NBI condition is checked and if its valueis high, the decision is made regardless of the class of thenew call. At the next stage the link outage probability isverified and based on the call class type, mobility conditionand battery status, the proper assignment is performed. Thedetailed procedure is presented in Algorithm 1. In Algo-rithm 1, PNBI is the probability of existence of NBI in theenvironment with a decision threshold PNBI. In fact, this pa-rameter is measured by estimating the NBI power receivedby each DEV. There are a variety of techniques for detec-tion and estimation of NBI power in UWB communications.For DS-UPAN, we use an autocorrelation receiver similar to[12] for estimating the NBI power received by the DS-UWBreceiver. Later, this value is normalized and compared bythe threshold value for decisions in RAT selection and hand-off algorithms. For MB-UPAN, we use a receiver structuresimilar to [13], which estimates the interference power oneach subcarrier of the MB-UWB receiver. This estimationis later used to improve the accuracy of the channel esti-mation as well. In our model PHM =

ν2.5 is the probability

of the DEV having high mobility with a decision thresholdPHM and ν is the velocity of the DEV. The degree of mobil-ity can be determined by estimating the DEV’s velocity. Weassume that the velocity of each DEV can be estimated bythe help of an integrated global positioning system (GPS)sensor. We improve the precision by using a Grey predic-tion filter as in [14] to predict the current value of velocity

Algorithm 1 : RAT Selection1: loop2: if new call ∈ class-A then3: if (PNBI ≥ PNBI) then4: select MB-UPAN EXIT5: else if (λDS

A−out ≤ λDSA−out) ∧ (PHM < PHM) then

6: select DS-UPAN EXIT7: else if (λMB

A−out ≤ λMBA−out) ∧ (PBC < PBC) then

8: select MB-UPAN EXIT9: else

10: Reject Call11: end if12: end if13: if new call ∈ class-B then14: if (PNBI ≥ PNBI) then15: select MB-UPAN16: EXIT17: else if (λMB

A−out ≤ λMBA−out) ∧ (PBC < PBC) then

18: select MB-UPAN EXIT19: else if (λDS

B−out ≤ λDSB−out) then

20: select DS-UPAN EXIT21: else22: Reject Call23: end if24: end if25: end loop

based on the previous samples. To form PHM, this value isnormalized based on the numerical values considered for thevelocity range as explained in Sect. 4. PBC is the probabilityof the DEV being in the battery-critical condition (BC) witha decision threshold PBC and given by:

PBC =100 − Remaining Battery %

100(1)

We assume that the percentage of remaining batterypower is provided by the DEV’s power management mod-ule. Furthermore, λMB

A−out and λMBB−out are the link outage prob-

abilities in MB-UPAN for class-A and class-B with deci-sion threshold values λMB

A−out and λMBB−out, respectively. Sim-

ilarly, λDSA−out and λDS

B−out are the link outage probabilities inDS-UPAN for class-A and class-B with decision thresholdvalues λDS

A−out and λDSB−out. In practical systems, the aforemen-

tioned outage probabilities can be measured by calculatingthe percentage of time that the signal strength drops belowa certain threshold. For our analytical performance eval-uation, in Sect. 3 a procedure is proposed to derive theseprobabilities. Please note that we use the term probabil-ity to address the aforementioned parameters in this papermostly because we employ a probabilistic Markov modelto evaluate the performance of our proposed RAT selectionand handoff algorithms in Sects. 3 and 4. However, as weexplained the practical aspects of these parameters in thissection, they are external parameters and obtained directlyfrom the system power management and receiver modulesand may not necessarily be probabilistic in nature. Nom-inal values are chosen for decision thresholds PNBI, PHM,PBC, λMB

A−out, λMBB−out, λ

DSA−out and λDS

B−out in our numerical anal-ysis in Sect. 4. However, for practical systems, optimumvalues should be chosen with respect to technical specifica-


Algorithm 2 : VHO to MB-UPAN1: if (PDS

A−NB ≥ PDSA−NB) ∧ (PDS

B−NB ≥ PDSB−NB)

AND(PDS

A−VB ≥ PDSA−VB) ∧ (PDS

B−VB ≥ PDSB−VB) then

2: if PNBI ≥ PNBI then3: VHO to MB-UPAN4: else if call ∈ class-A then5: if (PHM ≥ PHM) ∧ (PNBC ≥ PNBC) then6: VHO to MB-UPAN7: end if8: else if call ∈ class-B then9: if (PNBC ≥ PNBC) then

10: VHO to MB-UPAN11: end if12: end if13: else14: no action15: end if

tions and based on a comprehensive simulation study whichis one of our future research objectives and out of the scopeof the present paper. Also note that the averaged values ofthe aforementioned parameters over time are used in the de-cision process.

2.3.2 VHO Algorithm

The goal in our VHO algorithm design between DS-UPANand MB-UPAN is two-fold. First, maximizing the end-usersatisfaction and the quality of service (QoS) they experi-ence. This is performed by proper reassignment of DEVsto another network by considering their present battery sta-tus, degree of mobility and the intensity of NBI which theyare exposed to. Second, optimizing the network resourcesand load balancing by taking into account the actual ratesfor handoff and new call blocking. The latter can also in-directly lead to increased end-user satisfaction as well. Toincrease the chance of accepting the handoff calls, we con-sider channels exclusively reserved for this purpose (guardchannels) in both networks. Algorithm 2 presents the VHOprocedure to MB-UPAN for calls originating from the DS-UPAN. In Algorithm 2, (PLM = 1 − PHM) is the probabilityof the DEV having low mobility with a decision threshold ofvalue (PLM = 1−PHM) and (PNBC = 1−PBC) is the probabil-ity of the DEV being in non-critical battery condition (NBC)with a decision threshold of value (PNBC = 1−PBC). Further,PDS

A−NB and PDSB−NB are the new-call blocking probabilities in

DS-UPAN for class-A and class-B with decision thresholdsPDS

A−NB and PDSB−NB, respectively. Corresponding parameters

for the handoff-call blocking probabilities are PDSA−VB, PDS

A−VB,PDS

B−VB and PDSB−VB. Algorithm 3 presents the VHO procedure

to DS-UPAN for calls originating from MB-UPAN. In Al-gorithm 3, PDS

A−NB and PDSB−NB are the new-call blocking prob-

abilities in DS-UPAN for class-A and class-B with decisionthresholds PDS

A−NB and PDSB−NB, respectively. Corresponding

parameters for the handoff-call blocking probabilities arePDS

A−VB, PDSA−VB, PDS

B−VB and PDSB−VB.

Algorithm 3 : VHO to DS-UPAN1: if (PMB

A−NB ≥ PMBA−NB) ∧ (PMB

B−NB ≥ PMBB−NB)

AND(PMB

A−VB ≥ PMBA−VB) ∧ (PMB

B−VB ≥ PMBB−VB) then

2: if PBC ≥ PBC then3: VHO to DS-UPAN4: end if5: else6: no action7: end if

3. 4D Markovian Performance Analysis

In this section, we use Markov based analysis to evaluate theperformance of the two proposed modules. We assume oneDS-UPAN and one MB-UPAN co-located in the same cov-erage area. We also assume that a DEV is connected onlyto one of these networks during its active session. In or-der to improve the GoS, we allocate guard channels to bothnetworks. There are a total number of CDS available chan-nels in DS-UPAN, from which Ch

DS channels are reservedfor handoff calls. For MB-UPAN, these values are CMB andCh

MB, respectively.The class-A and class-B calls are considered to have in-

dependent Poisson arrival rates. The corresponding call ar-rival rates are denoted λA and λB, respectively. We consideran exponential distribution for the channel holding time forboth types of calls with a mean of 1/μA and 1/μB for DS-UPAN and MB-UPAN, respectively.

Based on the above assumptions, the system can bemodelled as a birth-death process and a Markov-based anal-ysis can be performed. Here we consider a 4D Markovchain with states defined as P(a1, b1, a2, b2) where a1 is thenumber of class-A calls in MB-UPAN, b1 is the number ofclass-B calls in MB-UPAN, a2 is the number of class-A callsin DS-UPAN and b2 is the number of class-B calls in DS-UPAN. Both networks only accept new calls if (a1 + b1) ≤CT

MB and (a2 + b2) ≤ CTDS, where CT

MB = CMB − ChMB and

CTDS = CDS − Ch

DS.

3.1 State Transitions

Recalling the birth-death process properties, the only pos-sible transitions are to neighboring states. There are a totalof 24 possible transitions to or from state P(a1, b1, a2, b2)in our Markov chain model. Figure 2 shows the possiblebirth and death transitions from and to state P(a1, b1, a2, b2),and Fig. 3 shows the possible handoff transitions from andto this state. This assumption is valid because of the mem-oryless property of the arrival process and the exponen-tial distribution of the channel holding time. The systemis working under steady-state conditions. In the follow-ing, we present the possible transitions to or from stateP(a1, b1, a2, b2) and the corresponding transition rates, i.e.,C1, ...,C12 and D1, ...,D12.1- P(a1, b1, a2, b2) −→ P(a1 + 1, b1, a2, b2) In this case, the


Fig. 2 Possible birth and death transitions from and to state P(a1, b1, a2,b2).

Fig. 3 Possible handoff transitions from and to state P(a1, b1, a2, b2).

number of class-A calls being served by the MB-UPAN isincreased by one if 0 ≤ (a1+b1) < CT

MB. The transition rate,C1, is given by:

C1 =λAPNBI(PHM + PNBC)(1 −

(λMB

A−out

)CTMB−a1−1

).

(2)

As mentioned earlier, λMBA−out is the link outage probability

for class-A in the MB-UPAN, which imposes the channelquality constraint on our proposed RAT selection policy andit is defined as the event that the signal strength is below acertain level. To calculate the outage probability of class-A in MB-UPAN, we consider a reference DEV with class-A connection in MB-UPAN and similar to [15], λMB

A−out iswritten as:

λMBA−out=

CDS−1∑m=0

CMB−1∑n=1

λMBA−out|n,m

(CMB−1

n

) (CDS−1

m

)

× μnMBμ

mDS(1−μMB)CMB−1−n(1−μDS)CDS−1−m,

(3)

where μDS is the probability of finding active interfering

DEVs in the DS-UPAN, transmitting on the same time slotas the reference DEV (n=0), and μMB is the probability offinding active interfering DEVs in the MB-UPAN, transmit-ting on the same time slot as the reference DEV. Furtherdetails on calculation of λMB

A−out|n,m are provided in [16]. Itis important to mention that in (3), we sum up the effect ofall interferers in the DS-UPAN and the MB-UPAN for bothtypes of calls (class-A and class-B). In fact, λA is the sumarrival rate at the MB-UPAN and DS-UPAN networks, i.e.,we have λA = C1 +C3. The same argument stands for λB.2- P(a1, b1, a2, b2) ←− P(a1 + 1, b1, a2, b2) In this case thenumber of class-A calls being served by MB-UPAN is de-creased by one if a1 > 0. The transition rate, D1, is givenby:

D1 = (a1 + 1)μA. (4)

3- P(a1, b1, a2, b2) −→ P(a1, b1 + 1, a2, b2) In this case, thenumber of class-B calls being served by the MB-UPAN isincreased by one if 0 ≤ (a1+b1) < CT

MB. The transition rate,C2, is given by:

C2 =λBPNBI(PLM + PNBC)(1 −

(λMB

B−out

)CTMB−b1−1

),

(5)

where λMBB−out is the link outage probability for class-B calls

in MB-UPAN.4- P(a1, b1, a2, b2) ←− P(a1, b1 + 1, a2, b2) In this case, thenumber of class-B calls being served by the MB-UPAN isdecreased by one if b1 > 0. The transition rate, D2, is givenby:

D2 = (b1 + 1)μB. (6)

5- P(a1, b1, a2, b2) −→ P(a1, b1, a2 + 1, b2) In this case, thenumber of class-A calls being served by the DS-UPAN isincreased by one if 0 ≤ (a2 + b2) < CT

DS. The transition rate,C3, is given by:

C3 = λA −C1. (7)

6- P(a1, b1, a2, b2) ←− P(a1, b1, a2 + 1, b2) In this case thenumber of class-A calls being served by DS-UPAN is de-creased by one if a2 > 0. The transition rate, D3, is givenby:

D3 = (a2 + 1)μA. (8)

7- P(a1, b1, a2, b2) −→ P(a1, b1, a2, b2 + 1) In this case, thenumber of class-B calls being served by the DS-UPAN isincreased by one if 0 ≤ (a2 + b2) < CT

DS. The transition rate,C4, is given by:

C4 = λB −C2. (9)

8- P(a1, b1, a2, b2) ←− P(a1, b1, a2, b2 + 1) In this case, thenumber of class-B calls being served by the DS-UPAN isdecreased by one if b2 > 0. The transition rate, D3, is givenby:


D4 = (b2 + 1)μB. (10)

9- P(a1, b1, a2, b2) −→ P(a1+1, b1, a2−1, b2) In this case, thenumber of class-A VHO calls in the MB-UPAN is increasedby one if a2 > 0 and 0 ≤ (a1 + b1) < CMB. The VHO ratefor MB-UPAN, C5, is defined by:

C5 =

(λA

(1 − PDS

A−NB

)+C11

(1 − PDS

A−VB

))

× PNBI(PHM + PNBC)(1 −

(λMB

A−out

)CTMB−a1−1

),

(11)

10- P(a1, b1, a2, b2)←− P(a1 + 1, b1, a2 − 1, b2) In this case,the number of class-A VHO calls in the DS-UPAN is in-creased by one if a1 > 0 and 0 ≤ (a1 + b1) < CDS. TheVHO rate for DS-UPAN, D5, is defined by:

D5=

(C3

(1−PMB

A−NB

)+C5

(1−PMB

A−VB

)) (1−

(λDS

A−out

)CTDS−a2

),

(12)

where λDSA−out is the link outage probability for class-A in

DS-UPAN.11- P(a1, b1, a2, b2) −→ P(a1, b1 + 1, a2, b2 − 1) In this case,the number of class-B VHO calls in the MB-UPAN is in-creased by one if b2 > 0 and 0 ≤ (a1 + b1) < CMB. TheVHO rate for MB-UPAN, C6, is defined by:

C6 =

(λB

(1 − PDS

B−NB

)+C12

(1 − PDS

B−VB

))

× PNBI(PLM + PNBC)(1 −

(λMB

B−out

)CTMB−b1−1

),

(13)

12- P(a1, b1, a2, b2)←− P(a1, b1 + 1, a2, b2 − 1) In this case,the number of class-B VHO calls in the DS-UPAN is in-creased by one if b1 > 0 and 0 ≤ (a1 + b1) < CDS. TheVHO rate for DS-UPAN, D6, is defined by:

D6=

(C4

(1−PMB

B−NB

)+C6

(1−PMB

B−VB

)) (1−

(λDS

B−out

)CTDS−b2

),

(14)

where λDSB−out is the link outage probability for class-B in DS-

UPAN. As observed in (11) and (13), the VHO rate dependson the handoff and new-call blocking probabilities. Thereare a variety of other approaches to estimate the VHO rate.For further information on calculation of the handoff rate,the reader is referred to [6], [17]. Finally, the transition ratesC7, ...,C12 and D7, ...,D12 can be derived as explained forC1, ...,C6 and D1, ...,D6 in (2) to (14).

3.2 Steady State Balance Equations

It is rather impractical to write all the balance equations in

this paper and it is hard to summarize balance transition re-lations given that there are too many state boundaries forsuch a 4D Markov chain. Therefore, we divide the stateboundaries into three main groups and give an example foreach.

Case 1: In the states where 0 < (a1 + b1) < CTMB and 0 <

(a2 + b2) < CTDS, both the handoff and the new-calls are

admitted into the system. The steady state balance equationsfor this case are written as:

(C1 + · · · + C12)P(a1, b1, a2, b2) =

D1P(a1+1, b1, a2, b2)+D2P(a1, b1+1, a2, b2)

+D3P(a1, b1, a2+1, b2)+D4P(a1, b1, a2, b2+1)

+D5P(a1+1, b1, a2−1, b2)+D6P(a1, b1+1, a2, b2−1)

+D7P(a1−1, b1, a2, b2)+D8P(a1, b1−1, a2, b2)

+D9P(a1, b1, a2−1, b2)+D10P(a1, b1, a2, b2−1)

+D11P(a1−1, b1, a2+1, b2)+D12P(a1, b1−1, a2, b2+1).

(15)

Case 2: In the states where CTMB ≤ (a1 + b1) < CMB and

CTDS ≤ (a2 + b2) < CDS, only handoff calls are admitted and

the steady state balance equations are written as:

(C5 + · · · + C12)P(a1, b1, a2, b2) =

D5P(a1+1, b1, a2−1, b2)+D6P(a1, b1+1, a2, b2−1)

+D7P(a1−1, b1, a2, b2)+D8P(a1, b1−1, a2, b2)

+D9P(a1, b1, a2−1, b2)+D10P(a1, b1, a2, b2−1)

+D11P(a1−1, b1, a2+1, b2)+D12P(a1, b1−1, a2, b2+1).

(16)

Case 3: Border states, where any dimension of the state canterminate in one of the values: 0, CT

MB, CTDS, CMB or CDS,

e.g., the balance equations for state (0, 0, 0, 0) is given by:

(C1 +C2 + C3 + C4)P(a1, b1, a2, b2) =

D1P(a1 + 1, b1, a2, b2) + D2P(a1, b1 + 1, a2, b2)

+ D3P(a1, b1, a2 + 1, b2) + D4P(a1, b1, a2, b2 + 1)

(17)

and for state (0,CMB, 0,CDS), by:

(C7 +C8 + C9 + C10)P(a1, b1, a2, b2) =

+ D7P(a1 − 1, b1, a2, b2) + D8P(a1, b1 − 1, a2, b2)

+ D9P(a1, b1, a2 − 1, b2) + D10P(a1, b1, a2, b2 − 1).

(18)

The Markov chain normalization equation is given by:

CDS−a2∑b2=0

CDS∑a2=0

CMB−a1∑b1=0

CMB∑a1=0

P(a1, b1, a2, b2) = 1. (19)


3.3 Deriving the Performance Metrics

The steady state probabilities can be derived using the bal-ance equations, some of which are mentioned above, and thenormalization constraint given in (19). The total number ofpossible states is equal to:

NS =

CMB∑i=0

(CMB − i).CDS∑j=0

(CDS − j). (20)

As a result, the transition probability matrix will be of di-mension NS × NS .

After calculation of the system’s steady state probabil-ities, the performance metrics can easily be derived.

In the heterogeneous system, a handoff call is blockedwhen there are no more free channels available at the desti-nation network; or from an analytical point of view we cansay a handoff blocking occurs when the system is urged toenter an unfeasible state. As a result, we can write the sys-tem’s handoff blocking probability as:

PVB =PDSB−VB + PDS

A−VB + PMBB−VB + PMB

A−VB

=

CDS∑a2=0

CMB−a1∑b1=0

CMB∑a1=0

P(a1, b1, a2,CDS − a2)

+

CDS∑b2=0

CMB−a1∑b1=0

CMB∑a1=0

P(a1, b1,CDS − b2, b2)

+

CDS−a2∑b2=0

CDS∑a2=0

CMB∑a1=0

P(a1,CMB − a1, a2, b2)

+

CDS−a2∑b2=0

CDS∑a2=0

CMB∑b1=0

P(CMB − b1, b1, a2, b2). (21)

Similarly, the blocking probability of new-calls occurswhen the system is urged to enter unfeasible states or statesconsidered just for handoff. This probability is given by:

PNB =PDSB−NB + PDS

A−NB + PMBB−NB + PMB

A−NB

=

ChDS∑

x=0

⎛⎜⎜⎜⎜⎜⎜⎜⎜⎝CT

DS+x∑a2=0

CMB−a1∑b1=0

CMB∑a1=0

P(a1, b1, a2,CDS − a2)

⎞⎟⎟⎟⎟⎟⎟⎟⎟⎠

+

ChDS∑

x=0

⎛⎜⎜⎜⎜⎜⎜⎜⎜⎝CT

DS+x∑b2=0

CMB−a1∑b1=0

CMB∑a1=0

P(a1, b1,CDS − b2, b2)

⎞⎟⎟⎟⎟⎟⎟⎟⎟⎠

+

ChMB∑

x=0

⎛⎜⎜⎜⎜⎜⎜⎜⎜⎝CDS−a2∑

b2=0

CDS∑a2=0

CTMB+x∑a1=0

P(a1,CMB − a1, a2, b2)

⎞⎟⎟⎟⎟⎟⎟⎟⎟⎠

+

ChMB∑

x=0

⎛⎜⎜⎜⎜⎜⎜⎜⎜⎝CDS−a2∑

b2=0

CDS∑a2=0

CTMB+x∑b1=0

P(CMB − b1, b1, a2, b2)

⎞⎟⎟⎟⎟⎟⎟⎟⎟⎠ .(22)

The total throughput of the system can be written as:

RT =

CDS−a2∑b2=0

CDS∑a2=0

CMB−a1∑b1=0

CMB∑a1=0

P(a1, b1, a2, b2)

×(a1RMB

A + b1RMBB + a2RDS

A + b2RDSB

), (23)

where RMBA is the basic channel data rate of class-A in MB-

UPAN, RMBB is the one for class-B in MB-UPAN, RDS

A is thebasic channel data rate of class-A in DS-UPAN and RDS

B isthe one for class-B in DS-UPAN. Finally, the average num-ber of DEVs (class-A and class-B) in the system, which in-dicates the system’s carried traffic is given by:

TT =

CDS−a2∑b2=0

CDS∑a2=0

CMB−a1∑b1=0

CMB∑a1=0

P(a1, b1, a2, b2)

× (a1 + b1 + a2 + b2). (24)

4. Numerical Results

Throughout this section, we present numerical results re-garding the analysis of our Markov-based RRM approaches.To simplify the evaluation process, we consider one DS-UPAN and one MB-UPAN with CDS = 3 and CMB = 3 chan-nels, respectively. However, the system is easily scalable forhigher values of piconets and channels. In each network,one channel is reserved for handoff calls (Ch

DS = ChMB = 1).

A mobility model for 2D indoor environments similar to[10] is considered and typical values for the speed of lowand high mobility DEVs are chosen to be 0.6 m/s and 2 m/s,respectively. We assume that each class (A, B) adapts itstransmission rate according to the network it is connectedto, namely, 110 Mbps for MB-UPAN and 1000 Mbps forDS-UPAN. The numerical values assumed for the other pa-rameters are defined by: PNBI = 0.1, PBC = 0.2, PLM = 0.2,μA = μB = 0.1, unless otherwise stated.

An iterative algorithm is used to calculate the VHOrates in (11) and (13), starting from an initial value for C5

and C6. As previously mentioned, to compute the link out-age probability, we consider a general analytical model thataccounts for the fading effect based on a Nakagami-m fadingmodel [9], [18]. The provided results are compared with theservice-based RAT selection strategy presented in [9] wherethe selection is only based on the type of call. We chose thisalgorithm because of its partial resemblance, which makesit comparable to our proposed algorithm.

Figure 4 and Fig. 5 show the VHO blocking and new-call blocking probabilities versus class-A call arrival rate forthe DS-UPAN and the MB-UPAN and λB = 0.1. It is obvi-ous that our proposed algorithm has a better performancethan the service-based approach. Our algorithm uses moreinformation in the decision process for the RAT selection,such as the link outage probability which indirectly con-tributes to reducing the blocking probability for VHO aswell as for new calls. To investigate the effect of an in-crease in class-B call arrival rate, we set this value to 0.8and in Fig. 6 and Fig. 7 we derive the similar results as in


Fig. 4 VHO block. prob. vs. class-A call arrival rate (λA), λB = 0.1.

Fig. 5 New-call block. prob. vs. class-A call arrival rate (λA), λB = 0.1.

Fig. 6 VHO block. prob. vs. class-A call arrival rate (λA), λB = 0.8.

Fig. 7 New-call block. prob. vs. class-A call arrival rate (λA), λB = 0.8.

Fig. 4 and Fig. 5 for the VHO blocking and new-call block-ing probabilities, respectively. In Fig. 6, we observe that anincrease of 0.7 unit in class-B call arrival rate results in anincrease of about 3 percent in VHO blocking probability forservice based algorithm but this increase is only 2 percentfor our proposed algorithm. We also notice that is higher

Fig. 8 GoS vs. class-A call arrival rate.

Fig. 9 Total system throughput vs. class-A call arrival rate.

values of class-B call arrival rates, the difference betweenDS-UPAN and the MB-UPAN performance increases, how-ever it is less divergent. Similarly, as we observe in Fig. 7,the increase of 0.7 unit in class-B call arrival rate results inan increase of about 15 percent in new-call blocking proba-bility for service-based algorithm, while this increase is only10 percent for our proposed algorithm. For new-call block-ing probability, this increase reduces the difference betweenDS-UPAN and the MB-UPAN performance for our algo-rithm, while this difference remains the same for the service-based algorithm.

Figure 8 shows the GoS versus class-A call arrival rate.Technically, we define the GoS measure of the overall sys-tem performance by GoS=PNB + αPVB, where α > 1 isa weighting factor used to put more importance on hand-off blocking probability (e.g., 10). Indeed, blocking a con-nection already in progress is usually more annoying thanblocking a new call. From an overall system performancepoint of view, our algorithm still leads a better performance.It is also observed that at higher mobility the performancedegrades to some extent. This can be explained by the in-crease in the rate of VHO at high speeds. However, usingthe proposed algorithm the performance degradation is lesscompared to the service-based approach. Figure 9 shows thetotal system throughput versus class-A call arrival rate. Forlower probability of NBI in the environment (PNBI = 0.1),our algorithm gives better performance. This is indirectlyrelated to its better GoS performance. However, in the caseof higher probability of NBI (PNBI = 0.6), a slight degrada-tion is observed. Indeed, to improve the GoS, our algorithmforwards a larger number of connections to the MB-UPAN


Fig. 10 Average number of DEVs versus class-A call arrival rate for (a) the proposed scheme and lowprobability of NBI, (b) the proposed scheme and high probability of NBI, (c) the service-based schemeand low probability of NBI, and (d) the service-based scheme and higher probability of NBI.

in this case, which causes a reduction in system throughputbecause of the lower data rate per channel for MB-UPAN incomparison with DS-UPAN.

Finally, Fig. 10 shows the average number of DEVs (orthe carried traffic) versus class-A new-call arrival rate for(a) the proposed algorithm in the presence of lower proba-bility of NBI (PNBI = 0.1), (b) the proposed algorithm inthe presence of higher probability of NBI (PNBI = 0.6), (c)the service-based algorithm with PNBI = 0.1, and (d) theservice-based algorithm with PNBI = 0.6. As expected, theaverage traffic carried by the MB-UPAN increases when us-ing the proposed algorithm, under high probability for NBI.

5. Conclusion

Two RRM modules for RAT selection and VHO procedurein heterogeneous UWB-based WPANs (UPANs) were pro-posed in this paper. The RAT selection algorithm consid-ers the specifications of two types of UPANs (DS-UPANand MB-UPAN) and, based on the link outage probabil-ity and the existence of NBI in the environment, decidesto which network a DEV should be connected. The pro-posed VHO decision procedure receives real-time informa-tion of the new-call and handoff-call blocking probabilitiesand aims at balancing the load between the two networks.An analytical method based on 4D Markov modeling waspresented to assess the performance of our proposed RATselection and VHO algorithms. It was attested that our al-gorithm has a better performance in terms of reducing thenew-call and handoff-call blocking probabilities, and im-proving the GoS. For lower intensities of NBI in the environ-ment, the proposed algorithm improves the system through-

put significantly, whereas its performance is slightly lessthan service-based approaches at higher probabilities giventhat our algorithms tend to allocate more calls to the MB-UPAN to improve the GoS in this case. Considering thesystem’s average carried traffic for the MB-UPAN, our al-gorithms perform better under higher probability of NBI inthe environment.

References

[1] A. Batra, J. Balakrishnan, et al., “Multi-band OFDM physical layerproposal for IEEE 802.15 task group 3a,” IEEE P802.15-03/268r3,March 2004.

[2] R. Fisher, R. Kohno, et al., “DS-UWB physical layer submission to802.15 task group 3a,” IEEE P802.15- 04/0137r4, Jan. 2005.

[3] I. Stojmenovic, Handbook of wireless networks and mobile comput-ing, Wiley, 2002.

[4] E. Stevens-Navarro, Y. Lin, and V. Wong, “An MDP-based verticalhandoff decision algorithm for heterogeneous wireless networks,”IEEE Trans. Veh. Technol., vol.57, no.2, pp.1243–1254, March2008.

[5] A. Mehbodniya and J. Chitizadeh, “User location estimation inwireless networks with application to vertical handoff using patternrecognition,” Proc. 10th IEEE Symposium on Computers and Com-munications (ISCC 2005), pp.315–320, June 2005.

[6] A.H. Zahran, B. Liang, and A. Saleh, “Mobility modeling and per-formance evaluation of heterogeneous wireless networks,” IEEETrans. Mobile Comput., vol.7, no.8, pp.1041–1056, Aug. 2008.

[7] J. Wang, Q.-A. Zeng, and D. Agrawal, “Performance analysis ofa preemptive and priority reservation handoff scheme for integratedservice-based wireless mobile networks,” IEEE Trans. Mobile Com-put., vol.2, no.1, pp.65–75, Jan.-March 2003.

[8] J. McNair and F. Zhu, “Vertical handoffs in fourth-generation multi-network environments,” IEEE Trans. Wireless Commun., vol.11,no.3, pp.8–15, June 2004.

[9] X. Gelabert, J. Perez-Romero, O. Sallent, and R. Agusti, “A Marko-vian approach to radio access technology selection in heterogeneous


multiaccess/multiservice wireless networks,” IEEE Trans. MobileComput., vol.7, no.10, pp.1257–1270, Oct. 2008.

[10] F. De Rango, F. Veltri, and S. Marano, “The impact of mobil-ity users on DS-UWB system performance,” Proc. Canadian Con-ference on Electrical and Computer Engineering (CCECE 2008),pp.1667–1670, May 2008.

[11] J. Lansford, “Detect and avoid (DAA) for UWB: Implementationissues and challenges,” Proc. IEEE 18th International Symposiumon Personal, Indoor and Mobile Radio Communications (PIMRC2007), pp.1–5, Sept. 2007.

[12] Y. Alemseged, H. Harada, and K. Witrisal, “Detection and iden-tification of nbi for multichannel uwb autocorrelation receivers,”Proc. IEEE Wireless Communications and Networking Conference(WCNC 2009), pp.1–6, April 2009.

[13] N. Hadaschik, I. Zakia, G. Ascheid, and H. Meyr, “Joint narrowbandinterference detection and channel estimation for wideband ofdm,”Proc. 13th European Wireless Conference, pp.1–7, April 2007.

[14] X. Liu and L. Jiang, “A novel vertical handoff algorithm based onfuzzy logic in aid of grey prediction theory in wireless heteroge-neous networks,” J. Shanghai Jiaotong University (Science), vol.17,no.1, pp.25–30, 2012.

[15] J.-P. Linnartz, “Exact analysis of the outage probability in multiple-user mobile radio,” IEEE Trans. Commun., vol.40, no.1, pp.20–23,Jan. 1992.

[16] A. Mehbodniya and S. Aissa, “Outage and BER analysis forultrawideband-based WPAN in Nakagami-m fading channels,” IEEETrans. Veh. Technol., vol.60, no.7, pp.3515–3520, Sept. 2011.

[17] D. Sarkar, T. Jewell, and S. Ramakrishnan, “Convergence in the cal-culation of the handoff arrival rate: A log-time iterative algorithm,”EURASIP Journal on Wireless Communications and Networking,pp.1–11, 2006.

[18] E. Vanem, S. Svaet, and F. Paint, “Effects of multiple access al-ternatives in heterogeneous wireless networks,” Proc. IEEE Wire-less Communications and Networking Conference (WCNC 2003),pp.1696–1700, 2003.

Abolfazl Mehbodniya received his Bach-elor’s degree and his Master’s degree in elec-trical engineering from Ferdowsi University ofMashhad, Iran in 2002 and 2005 and his Ph.D.degree from the National Institute of ScientificResearch-Energy, Materials, and Telecommu-nications (INRS-EMT), University of Quebec,Montreal, QC, Canada in 2010. Dr. Mehbodniyais the the recipient of the Japan society for pro-motion of science (JSPS) postdoctoral fellow-ship and is currently a research fellow at grad-

uate school of engineering, Tohoku university. His research interests arein wireless communications, ultra wideband communications, interferencemanagement, radio resource management and cooperative relay networks.

Sonia Aıssa received her Ph.D. degreein Electrical and Computer Engineering fromMcGill University, Montreal, QC, Canada, in1998. Since then, she has been with the NationalInstitute of Scientific Research-Energy, Mate-rials, and Telecommunications (INRS-EMT),University of Quebec, Montreal, Canada, whereshe is currently a Full Professor. From 1996 to1997, she was a Researcher with the Departmentof Electronics and Communications of KyotoUniversity, Kyoto, Japan, and with the Wireless

Systems Laboratories of NTT, Kanagawa, Japan. From 1998 to 2000, shewas a Research Associate at INRS-EMT, Montreal. From 2000 to 2002,while she was an Assistant Professor, she was a Principal Investigator inthe major program of personal and mobile communications of the Cana-dian Institute for Telecommunications Research (CITR), leading researchin radio resource management for code division multiple access systems.From 2004 to 2007, she was an Adjunct Professor with Concordia Univer-sity, Montreal. In 2006, she was Visiting Invited Professor with the Gradu-ate School of Informatics, Kyoto University, Japan. Her research interestslie in the area of wireless and mobile communications, and include radioresource management, performance evaluation, design and analysis of mul-tiple antenna (MIMO) systems, and cross-layer design and optimization,with a focus on cellular, ad hoc, and cognitive radio networks. Dr. Aıssawas the Founding Chair of the Montreal Chapter IEEE Women in Engineer-ing Society in 2004–2007, a Technical Program Cochair for the WirelessCommunications Symposium (WCS) of the 2006 IEEE International Con-ference on Communications (ICC 2006), and PHY/MAC Program Chairfor the 2007 IEEE Wireless Communications and Networking Conference(WCNC 2007). She was also the Technical Program Leading Chair forthe WCS of the IEEE ICC 2009, and is currently serving as Cochair forthe WCS of the IEEE ICC 2011. She has served as a Guest Editor of theEURASIP journal on Wireless Communications and Networking in 2006,and as Associate Editor of the IEEE WirelessCommunicationsMagazine in2006–2010. She is currently an Editor of the IEEE Transactions onWire-less Communications, the IEEE Transactions on Communications and theIEEE Communications Magazine, and Associate Editor of the Wiley Secu-rity and Communication Networks Journal. Awards and distinctions to hercredit include the Quebec Government FQRNT Strategic Fellowship forProfessors-Researchers in 2001–2006; the INRS-EMT Performance Awardin 2004 for outstanding achievements in research, teaching and service;the IEEE Communications Society Certificate of Appreciation in 2006 and2009; and the Technical Community Service Award from the FQRNT Cen-ter for Advanced Systems and Technologies in Communications (SYTA-Com) in 2007. She is also co-recipient of Best Paper Awards from IEEEISCC 2009, WPMC 2010 and IEEE WCNC 2010; and recipient of NSERC(Natural Sciences and Engineering Research Council of Canada) DiscoveryAccelerator Supplement Award.


Fumiyuki Adachi received the B.S. andDr. Eng. degrees in electrical engineering fromTohoku University, Sendai, Japan, in 1973and 1984, respectively. In April 1973, hejoined the Electrical Communications Labora-tories of Nippon Telegraph&Telephone Corpo-ration (now NTT) and conducted various typesof research related to digital cellular mobilecommunications. From July 1992 to December1999, he was with NTT Mobile Communica-tions Network, Inc. (now NTT DoCoMo, Inc.),

where he led a research group on wideband/broadband CDMA wirelessaccess for IMT-2000 and beyond. Since January 2000, he has been withTohoku University, Sendai, Japan, where he is a Professor of Electricaland Communication Engineering at the Graduate School of Engineering.His research interests are in CDMA wireless access techniques, equaliza-tion, transmit/receive antenna diversity, MIMO, adaptive transmission, andchannel coding, with particular application to broadband wireless commu-nications systems. From October 1984 to September 1985, he was a UnitedKingdom SERC Visiting Research Fellow in the Department of ElectricalEngineering and Electronics at Liverpool University. He is an IEICE Fel-low and was a co-recipient of the IEICE Transactions best paper of the yearaward 1996, 1998, and 2009 and also a recipient of Achievement award2003. He is an IEEE Fellow and was a co-recipient of the IEEE Vehicu-lar Technology Transactions best paper of the year award 1980 and again1990 and also a recipient of Avant Garde award 2000. He was a recipientof Thomson Scientific Research Front Award 2004, Ericsson Telecommu-nications Award 2008, and Telecom System Technology Award 2010.

paper eﬃcient resource utilization for heterogeneous ......appropriate for future connection....

Documents