Physical Communication 7 (2013) 134–144

Contents lists available at SciVerse ScienceDirect

Physical Communication

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

Full length article

A fuzzy admission control scheme for high quality videodelivery over underlay cognitive radioPejman Goudarzi ∗Multimedia Systems Group, Information Technology Department, ICT Research Institute (ITRC), Teheran, Iran

a r t i c l e i n f o

Article history:Received 11 June 2012Received in revised form 9 December 2012Accepted 10 December 2012Available online 20 December 2012

Keywords:Quality of experienceCognitive radio networkAdmission controlCDMA

a b s t r a c t

In much of the traditional tight admission control algorithms for video cognitive usersin cognitive networks, cognitive users are admitted sequentially based on the strictquality of service and interference constraints imposed on the cognitive and primary usersrespectively. The sequential admittance of cognitive users may impose some form of thequeuing delay for time-sensitive cognitive users which may be unacceptable. On the otherhand, traditional admission control schemes do not consider the quality of experience(QoE) of video users for admitting newly incoming ones. For addressing these issues andobtaining a more flexible quality-centric admission control policy by which the admissionsystem can admit eligible cognitive users in parallel, and to cope with uncertainties inthe acceptable levels of the video quality for different cognitive users (which may usedifferent software/hardware with different capabilities) and interference levels imposedon the primary users, a soft admission control (SAC) technique (named FQAC) is proposedbywhich the admission probability level for the parallel cognitive users can intelligently becontrolled based on some linguistic input variables. Numerical analysis has been performedto validate the efficiency of the proposed quality-aware SAC mechanism.

© 2012 Elsevier B.V. All rights reserved.

1. Introduction

Cognitive radio networks (CRN) are designed to usethe potentially unused wireless spectrum holes in thelicensed band. The CRNs are categorized into two differentgroups which are underlay CRNs and overlay CRNs. Inunderlay CRNs which are the topic of interest in this work,there is no need for sophisticated spectrum/bandwidthestimation techniques (in contrastwith overlay CRNs). But,in underlay CRNs there always exists some interferencewhich is imposed on licensed primary users (the users ofthe primary network or PUs) from unlicensed secondary/cognitive ones (the users of secondary network or CUs).

Mitigating the imposed interference from CUs on PUs isalways an important challenging research area especiallyin the underlay CRNs. In the case that CUs are broadbandapplications such as video, suppressing the CU-induced

∗ Tel.: +98 21 84977351.E-mail addresses: pgudarz@yahoo.com, pgoudarzi@itrc.ac.ir.

1874-4907/$ – see front matter© 2012 Elsevier B.V. All rights reserved.doi:10.1016/j.phycom.2012.12.002

interference becomes even more challenging. In such sce-narios, the admission and power controlmechanismsmustbe designed such that while guaranteeing the stringentand hard quality requirement of secondary users, make theinterference power for the primary users below a targetvalue. Comparedwith overlay CRNs, the underlay cognitiveradio benefits from extremely simplified senders/receiversbecause it does not need sophisticated spectrum estima-tion techniques.

Broadband multimedia content delivery over wirelessnetworks has been converted to a reality by the emerg-ing 3G or 4G technologies and responded to the increasingmomentum in the network user demands. The emerginghigh-speed wireless access technologies and the require-ments of different wireless applications are expected tocreate a huge demand on spectral resources in the nextgeneration wireless networks. Achieving high spectrumutilization is, therefore, one of the most critical researchobjectives in designing wireless communication systemstoday.

P. Goudarzi / Physical Communication 7 (2013) 134–144 135

In fact, as discussed in a report by the Federal Commu-nication Commission (FCC) on spectrum usage, the spec-trum utilization varies from 15% to 85%, depending on thegeographical area [1]. Therefore, there is an increasing in-terest in developing efficient methods for spectrum man-agement and sharing. Cognitive radio is a new paradigmin wireless communications to enhance utilization of lim-ited spectrum resources. It is defined as a radio that isable to utilize available side information, in a decentralizedparadigm, in order to efficiently use the radio spectrum leftunused by licensed systems. CRN techniques exploit spec-trum opportunities in space, time, frequency while pro-tecting PUs from excessive interference due to spectrumaccess from the CUs. Meanwhile, Code Division MultipleAccess (CDMA) has been adopted as multiple access tech-nology for 3G and beyond due to its advantages such asuniversal frequency reuse, soft hand-off, inherent diver-sity, and high spectrum efficiency [2]. Multimedia servicesover IP-based CDMA wireless networks will be one of thekey applications in the cognitive radio field.

Quality of experience (QoE) is an essential feature as-sociated with successful video content delivery to theend-users. The main concern of most telecommunicationoperators about their video-based services is because ofvideo service assurance. Quality of experience ties togetheruser perception, experience and expectations to appli-cation and network performance, typically expressed byquality of service parameters. Quantitative relationshipsbetween QoE and Quality of Service (QoS) are required inorder to be able to build effective QoE control mechanismsontomeasurableQoS parameters. Against this background,the work proposed in [3] introduces a generic formula inwhich QoE and QoS parameters are connected through anexponential relationship, called IQX hypothesis. The for-mula relates changes of QoEwith respect to QoS to the cur-rent level of QoE, is simple tomatch, and its limit behaviorsare straightforward to interpret. It validates the IQX hy-pothesis for streaming services, where QoE in terms of thesubjective Mean Opinion Scores (MOS)1 [4] is expressed asfunctions of loss and reordering ratio, the latter of which iscaused by jitter. They conclude that the IQX hypothesis isa strong candidate to be taken into account when derivingrelationships between QoE and QoS parameters.

The requirements of a specific set of QoS parameters(delay, jitter, packet loss, etc.) must be guaranteed for eachreal-time traffic transmitted over such wireless networks.However, for most real-time applications of wireless net-works, intrinsic and possibly large levels of interferenceor collisions in the physical or link layers caused by ra-dio transmission, media access protocols or time-varyingtopological changes provides challenging issues in guaran-teeing these stringent QoS requirements.

Admission control (AC) mechanisms are investigatedin networks to guarantee the desired levels of quality forexisting network users by blocking the unwanted newlyarriving ones. In other words, AC is a validation processin communication systems where a check is performedbefore a connection is established to see if current

1 The MOS is a subjective measure for the perceived video quality [4].

resources are sufficient for the proposed connection.Admission control should guarantee that all links receivetheir required QoS throughout their lifetime. Therefore,admission control should consider the long-term trafficconditions as well as transmission and interferenceconditions of both existing and the new links.

The AC problem in CRNs not only needs to provide qual-ity guarantees for the admitted CUs, but also it should guar-antee the interference constraints of CUs on the primarynetwork. Generally, in awireless network, there always ex-ists a tradeoff between two types of error that should beconsidered in the admission process. Type I errors occurwhen a new user is accepted incorrectly and cause exces-sive interference and hence outage of the ongoing users.Type II errors occur by tight admission rules which resultsin blocking. In a CRN, loose constraint in admitting newCUswhich is referred to as all-admission may cause excessiveinterference on PUs or other CUs. On the other hand tightadmission rules lead to lowutilization of opportunities andhigh blocking probability (the probability that the requiredlevel of quality cannot be offered). The former leads to vi-olation of the PUs interference constraints and the latterwill increase the outage probability (the probability thatthe mobile station is outside the service coverage area, oraffected by interference) of current CUs [5].

An admission control in an underlay CRN based onsingle ormultiple link removal has been investigated in [6].This method includes two phases: power control and linksremoval. In the first phase, the steady state powers ofall CUs are determined. Then, in the second phase, oneor multiple CUs are removed based on an interferencemeasure values until the remaining subset of CUs fulfill theQuality of Service (QoS) requirement and the interferencethreshold of primary network, the interference measuredefinition is based on the similar work in [7], where theusers have been removed gradually from network until theremaining set of CUs could be supported. In [8] differentrevenues have been considered for CUs. Then, the problemis how to find a subset of CUs such that the total revenueoutput of the network is maximized. For this purposethe problem is formulated in the optimization theoryframework to maximize the CUs revenue subject to theinterference constraint on the PU and QoS requirementof CUs.

These works and most of the existing, e.g. [6,8–10],however try to choose the best subset of CUs whichcan fulfill the two constraints of the underlay scheme,the interference temperature on PUs and QoS of CUs,regardless of their order of arrival. That is, the admissionprocedure has been simultaneously done for a set ofN CUs.Therefore, there is no guarantee on satisfying interferencelimits of the PUs or the QoS of current CUs during theadmission process. Furthermore, these schemes result inzero blocking and high outage probability for cognitivenetworks. However, in a real scenario ofwireless networks,simultaneous arrival of users is a rare event and typicallyfollows a statistical distribution in time.

In [5], the authors consider the arrival process ofCUs which better models the real scenarios. In theirmodel, some of the new arriving CUs may suffer blockingand consequently, with non zero blocking probability of

136 P. Goudarzi / Physical Communication 7 (2013) 134–144

CUs their outage probability will decrease. The authorsin [5] proposed a sequential interference aware admissioncontrol mechanism where a newly arrived CU is admittedaccording to the current status of the network, i.e., theaggregate interference on the PUs and the QoS of thecurrent active CUs.

In [11], Kim et al. develop a complete framework toperform joint admission control and rate/power allocationfor secondary users such that both QoS and interferenceconstraints are only violated within desired limits. Theyalso have investigated the throughput performance of pri-mary and secondary networks. In [12], based on the util-ity definition for cooperative users, Nie et al. have shownthat the channel allocation problem can be formulated asa potential game, and thus converges to a deterministicchannel allocation Nash equilibrium point. Alternatively,they have proposed a no-regret learning implementationfor some scenarios and shown that they have similar per-formance with the potential game when cooperation isenforced. In addition, their proposed learning algorithmaccommodates selfish users, and requires less knowledgeabout the game and less implementation overhead. Theyalso show that cooperation based spectrum sharing eti-quette improves the overall network performance at theexpense of an increased overhead required for informationexchange.

Soft computing techniques such as fuzzy logic is a formof many-valued logic or probabilistic logic; it deals withreasoning that is approximate rather than fixed and exact.In contrast with traditional logic they can have varyingvalues, where binary sets have two-valued logic, true orfalse, fuzzy logic variables may have a truth value thatranges in degree between 0 and 1. Fuzzy logic has beenextended to handle the concept of partial truth, wherethe truth value may range between completely true andcompletely false. Furthermore, when linguistic variablesare used, these degrees may be managed by specificfunctions, called membership functions.

In the current work, a fuzzy-based Quality-awareAdmission Control (QAC) algorithm has been developedwhich tries to sustain the QoEs associates with the real-time video CUs in an acceptable region while guaranteeingthe interference constraints of non real-time PUs. Theproposed admission control algorithm, allows a newlyarrived CU only if the QoE requirements of the existingCUs are satisfied and the interference level imposed onPUs be below a given threshold. The presented work inthis paper differs from all of the previous mentionedworks in considering the QoE requirements of the real-time CUswhen admitting newly arriving CUs. In summary,the paper’s main contributions are as follows. Firstly, afuzzy-based quality-aware admission control mechanismhas been proposed which with the help of an intelligentpower control mechanism; do not allow that the QoErequirements of the cognitive video users go below somepre-defined thresholds. On the other hand, the admissionand power control mechanisms are designed such that theinterference limits of the primary users are not violated.

The rest of the paper is organized as follows. In Section 2the assumed CRN model has been defined. In Section 3the proposed quality-aware admission and power control

Fig. 1. Cognitive radio network settings.

mechanism have been introduced. In Sections 4 and 5,two proposed quality-aware admission and power con-trol mechanisms namely TQAC and FQAC have been in-troduced. Section 6 is devoted to the numerical analysisand finally in Section 7 some concluding remarks arepresented.

2. Cognitive radio model

Consider the hierarchical spectrum sharing problem ina spectrum underlay cognitive wireless network whereseveral secondary (cognitive) users and primary userstransmit data in a common frequency band (e.g., as in aCDMA-based wireless network) [6]. A communication ses-sion is established between a pair of users who wish tocommunicate with each other. The communication ses-sions between the primary users are referred to as pri-mary sessions, and the communication sessions betweenthe cognitive users are referred to as cognitive sessions.The transmission setting considered in this paper is illus-trated in Fig. 1. The set of primary sessions is denoted byN = {1, 2, . . . ,N}, and the set of the cognitive sessions isdenoted byM = {1, 2, . . . ,M}.

In this paper, it is assumed that all the cognitive sessionsin the set M are video streaming sessions and they areall single-hop sessions. Simultaneous communications ofprimary sessions and secondary sessions will interferewith each other. The quality-aware admission and powercontrol algorithms will be designed to guarantee that:(1) the interference received at the receiving nodes of theprimary sessions should be below an acceptable level, and(2) the secondary cognitive sessions (e.g., video streamingsessions) should meet the QoE requirements. In thetransmission setting as shown in Fig. 1, multiple primarysessions and secondary sessions share the commonchannel using CDMA technology.

In CDMA model, the spread spectrum bandwidth sdenoted by W , the power spectral density of the AdditiveWhite Gaussian Noise (AWGN) is denoted by N0. The

P. Goudarzi / Physical Communication 7 (2013) 134–144 137

channel gain from the sender j to the receiverm is denotedby gjm. The received Signal to Interference plus Noise Ratio(SINR) at the receiver of the secondary sessionm is denotedby γ m, which is given by Shu et al. [13] as follows:

γm =

WRm

g smP

sm

δ

j∈Mj=m

gjmPj +k∈N

gkmPk

+ N0W , ∀m ∈ M (1)

where δ is the orthogonality factor representing MultipleAccess Interference (MAI) from the imperfect orthogonalspreading codes. Typical values for δ are 2

3 and 1 for a chipof rectangular or sinusoid shape, respectively [13]. P s

m, g sm

and Rm are the transmission power of the transmitter inthe cognitive session m, channel gain for the cognitivesession m (from transmitter to receiver) and transmissionrate in the cognitive sessionm respectively. Assume BinaryPhase Shift Keying (BPSK) modulation is used in the CDMAsystem. The Bit Error Rate (BER) at the cognitive sessionmis given by:

bm = Q

2γm

, ∀m ∈ M (2)

where Q (·) is the Q -function [14].If a packet is received in error, it will be dropped at the

receiver. Assume that the bit errors occur independentlyin a packet. Therefore, the Packet Loss Rate (PLR) dueto transmission errors at the cognitive session m is thengiven by:

pm = 1 − (1 − bm)Lm , ∀m ∈ M (3)

where, Lm is the number of bits for the packet at thecognitive sessionm.

Based on the so-called IQX model presented in [3], theQoEs (which hereby are represented by Qm) are a functionof the PLRs associated with each video source as follows:

Qm1= ηe−ωqm + s, ∀m ∈ M (4)

where η, ω, s and qm is the quality of service parameterrelated to the cognitive sourcem and is defined as follows:

qm1= α0pm, ∀m ∈ M.

α0 is some positive constant which is calculated basedon standard experiments. In the current work and basedon [15], it is assumed that the mentioned constant is equalto 41.7. Also similar to [3] and by considering perfectquality for no packet loss, it is considered the η, ω and sparameters to be 3, 11 and 2 respectively.

An important point about the IQX model is the factthat this model is designed in accordance with the ITU-T standard G.1070 that can capture the effect of a widerange of video codecs such as MPEG-4 or H.264/AVCand is tested for different video sequences which containdifferent levels of motion and contextual complexity [16].

To guarantee the QoE requirement of each active CU,its Qm parameter must be above some pre-determinedthreshold λm, i.e. we must have:

Qm ≥ λm, ∀m ∈ M. (5)

Additionally, to fulfill the interference level constraints onthe primary users, the aggregate interference associatedwith CUs which is imposed on PU base stations should notexceed a pre-defined threshold T , i.e.:Mi=1

giPi ≤ T (6)

Pi and gi are the transmission power of the cognitive useri and the channel gain from this cognitive user to the basestation respectively.

Considering the above constraints, the admission prob-lem will be how to admit a new cognitive user withoutviolating the problem constraints in (5) and (6) during theadmission process. For this purpose, similar to [5], twoCRN-related margin parameters are considered which arePrimary InterferenceMargin (PIM) and the Cognitive Qual-ity Margin (CQM) as follows:

0 ≤ PIM 1= T −

Mi=1

giPi. (7)

It is clear that negative values of PIM violates the interfer-ence threshold.

Additionally, CQMm denotes themaximumqualitymar-gin that can be tolerated by the m’th active CU to protectits QoE. As QoE function Qm(·) is an increasing function ofthe SINR level at each cognitive user γm, it can be assumeda functional relationship between these two parameters asQm = f (γm) where f (·) is a monotone increasing functionand can be written based on (2) and (4) as follows:

f (γm)1= η exp

−ωα0

1 −

1 − Q

2γm

Lm

+ s.

The CQMm can be simply calculated for eachm from (1) and(5) as follows:

WRm

g smP

sm

δ

j∈Mj=m

gjmPj +k∈N

gkmPk + CQMm

+ N0W

≥ f −1 (λm) . (8)Hence, we must have:

0 ≤ CQMm1=

Wg smP

sm

δRmf −1 (λm)

−

N0Wδ

+

j∈Mj=m

gjmPj +k∈N

gkmPk

. (9)

It is tried to choose the proper values for PIM and CQM insuch a way that the aggregate interference on primary re-ceiver are kept below the certain level during the admis-sion phase and also the QoE requirements of the currentCUs are satisfied.

3. Proposed QAC algorithm

3.1. Power control

For satisfying theQoE requirements, each cognitive userhas to adapt its transmission power to meet the desirable

138 P. Goudarzi / Physical Communication 7 (2013) 134–144

Signal to Interference plus Noise Ratio (SINR). Therefore,an efficient power control algorithm is needed. TheDistributed Constraint Power Control (DCPC) algorithmin [17] has been adopted.

Pm(ℓ + 1) = minPmaxm , Pm(ℓ)

λm

γm

, ∀m ∈ M (10)

where, ℓ is the iteration number, Pmaxm is the maximum

admissible power for them’th CU.Starting with any initial power vector, DCPC will

converge to a unique stationary power vector→∗

P =P∗

1 , P∗

2 , . . . , P∗

M

if any solution exists [17].

3.2. QAC problem

Consider the case where the CUs arrive sequentially intime. The QAC problem is how to admit a new cognitiveuser when M active CUs exist. Both constraints (5) and (6)should be satisfied during the admission phase.

In the casewhen anewcognitive user arrives, the power

vector of all current CUs is set to→∗

P and a positive valuefor Pmax is assumed. Let the index of newly arrived CUbe denoted by ‘‘0’’ letter. When the new cognitive userbegins to transmit at power level P0, the interference levelimposed on the base station’s primary receiver will changeto

Mj=1 gjmPj + g0mP0 and the SINR of the current CUs will

change according to:

γm =

WRm

g smP

sm

δ

j∈Mj=m

gjmPj +k∈N

gkmPk

+ N0W + gomP0

. (11)

The gain g0m is the channel gain between the newlyarrived cognitive user 0 and cognitive user m. The extrainterference may violate the QoE requirements of thecurrent CUs and also the interference constraint on theprimary receivers.

To mitigate the QoE degradation of the current CUs,they can increase their transmission power to reachthe targeted QoE. But, increasing the powers in anuncontrolled way may increase the aggregate interferenceimposed on PUs beyond the interference threshold.Moreover, both (5) and (6) must be satisfied during theadmission phase. Hence, the power of current CUsmust beincreased by a proper boosting factor, which is determinedaccording to the current network condition. To do this, aproper set of PIM and CQMs have to be used to determinethe value of this boosting factor. For applying theseconcepts in a cognitive network, the following lemmasfrom [18] must be adopted.

Lemma 1. Let→∗

P be a stationary power vector which sup-ports allmobiles under noise condition κ to reach their desiredSINR. Assume that ς is an additional receiver noise vector sat-

isfying ς ≤ ε · κ , for some ε ≥ 0. If (1 + ε)→∗

P ≤→maxP ,

and the DCPC algorithm starts with (1+ε)→∗

P for the receivernoise ς + κ , then all mobiles are supported at any moment of

time. Moreover, we have→∗

P (ς + κ) ≤→∗

P (κ)(1 + ε) [18].

Note. As can be verified in the Lemma 1, the objectiveis that cognitive users meet their target SINR values byproper power control techniques. On the other hand, asmentioned earlier, there is a one-to-one relationship be-tween the target SINR and target QoE (through monotonef (·) function), so, we can easily extend the proposed powercontrol technique in the previous lemma to the case inwhich the objective is that the secondary users meet theirtarget QoE values.

Lemma 2. Starting with the boosted power vector, (1 +

ε)→∗

P ≤→maxP and executing the DCPC algorithm in the pre-

vious lemma will converge to a smaller stationary power vec-tor, i.e., the vector where all its elements are smaller than thecorresponding elements in (1 + ε)

→∗

P ≤→maxP [18].

Therefore, the objective is to find out the properboosting factor according to current CUs situation and in-terference threshold of PUs. To this aim, the appropriateboosting factor is calculated for the proposed settings interms of minimum and maximum values of the PIM andthe maximum power of CUs by deploying the above lem-mas. The maximum value of the PIM , i.e. PIMmax is definedin (7) and its minimum value is chosen to be equal to:

PIMmin = ∆(ε) · PIMmax, 0 < ∆(ε) < 1 (12)

where∆(ε) is a constant which should be selected accord-ing to ε. In fact, PIMmin indicates theminimumvalue of PIMafter the powers are boosted by the boosting factor (1+ε).

In addition, this parameter should be selected largeenough to reflect the minimum tolerable interference ofprimary system for admitting a new cognitive user, i.e. wehave:

PIMmin ≤ T −

Mi=1

gi(1 + ε) · Pi

= T −

Mi=1

giPi −Mi=1

gi · ε · Pi

= PIMmax − ε ·

Mi=1

giPi. (13)

Therefore, ε should be selected according to:

ε ≤PIMmax − PIMmin

Mi=1

giPi

. (14)

On the other hand, according to the maximum allowabletransmission power of current CUs, 1 + ε should satisfy:

1 + ε ≤ minPmax1

P1,Pmax2

P2, . . . ,

PmaxM

PM

. (15)

That is, the proper value of ε should be selected by (14)and (15) after choosing PIMmin in (12). On the other hand,the transmission power of the new CU is normally belowits target value during the admission process. It should bedecided how to adjust its power when the current CUsboost their power level. According to Lemma 2 by applyingthe DCPC algorithm on the boosted power vectors, the

P. Goudarzi / Physical Communication 7 (2013) 134–144 139

power of the current CUs will decrease in the consecutiveiterations.

This reduction is an indicator of available capacity foradmitting a new CU and suggests updating PIMmin. Also toprotect the QoE of the current admitted CUs, CQMm, ∀m ∈

M should be updated according to this decrement based on(9). Hence, an appropriate value for the initial power of thenew user, P0, is given by:

P0 = minPIMmin

g0,CQMmin

g0

(16)

where, the minimum value of CQM is given by:

CQMm ≤ minm

WgmPmδRmf −1 (λm)

−

N0Wδ

+

j∈Mj=m

gjmPj

+

k∈N

gkmPk

(17)

where P =

P1, P2, . . . , PM

is the power vector of CUs

after the convergence of the DCPC algorithm. By applyingthese values for P0 and ε, we can retain the interference ofthe primary users and QoE of the secondary ones withintheir allowable limits.

In the sequel, it will be introduced two sub-types of QACalgorithm namely Tight Quality-aware Admission Con-trol (TQAC) and Fuzzy Quality-aware Admission Control(FQAC). In the first type, a crisp admission control pol-icy has been deployed which can give hard guarantee forthe cognitive quality and primary interference levels. Inthe second type, the evolutionary techniques have beenadopted to introduce a more flexible admission policy forthe cognitive quality control objective.

4. TQAC algorithm

The admission control which has been proposed inthis paper, employ PIM and CQMm as two importantparameters during the admission procedure of a newlycoming CU. The idea is that the power of current CUsare boosted by a factor of (1 + ε) and the power of thenew CU is adjusted according to (16). Note that TQAC isa modified version of the tight admission control S-SIACalgorithm presented in [5] which it is considered tight QoEguarantees in comparison with the tight QoS guaranteesadopted for S-SIAC in [5].

In this algorithm, a tight rule is applied on admittinga new cognitive user. Consequently, current CUs do notsuffer outage while the aggregate interference on primarynetwork does not violate the threshold in (6). However,CUs may meet high blocking probability due to a tightacceptance rule. The key feature of this algorithm is thatit can support both the QoE of the CUs and the interferencelimit of PUs.

Here, a new cognitive user that has a better link qualitywith respect to the current CUs may suffer blocking.The inherent trade-off between blocking and outage

probabilities can be used in this algorithm to mitigate thisproblem. For more flexibility in admission mechanism andto tackle the mentioned problem more effectively, Fuzzytechniques are added to the traditional tight QAC methodto introduce a soft version of QAC. This method applies asoft protection on the aggregate interference on PUs andthe perceived quality of the cognitive ones.

5. FQAC algorithm

In the TQAC algorithm, cognitive users are admittedsequentially based on the strict quality and interfer-ence constraints imposed on the cognitive and primaryusers respectively. The sequential admittance of cogni-tive users may impose some form of the queuing delayfor time-sensitive cognitive users which may be unaccept-able. To address this issue and to obtain a more flexibleadmission control policy by which the admission sys-tem can admit eligible cognitive users in parallel, and tocope with uncertainties in the acceptable levels of thevideo quality for different cognitive users (which may usesoftwares/hardwareswith different capabilities) and inter-ference levels imposed on the PUs, the soft fuzzy comput-ing techniques have been adopted bywhich the AdmissionProbability (AP) level for the parallel CUs can be intelli-gently controlled based on the PIM and CQMm parametersas fuzzy input variables. By definition, the AP is the per-centage of the eligible CUs which are simultaneously ad-mitted during each admission phase. A simple relationshipexists between the AP and the Blocking Probability (BP)which is AP = 1 − BP.

The proposed FQAC algorithm can be implemented by ajoint collaboration between the base station and cognitiveusers. The AP parameter might be intelligently controlledand calculated in a centralized manner using a simpleFuzzy Inference System (FIS) which can be implementedin the base station side. To do this, the CUs must send theirestimated CQMm values to the base station periodicallythrough exploiting some suitable control messages in thelink/network layer. The CUs can easily calculate the CQMmvalues because they have estimates of the γm value (SINR)and so can explicitly calculate the f (·) function and hencethe CQMm in their side based on (9). The base stationmakesdecisions on calculating the AP value by averaging thereceivedCQMm values in specific timeperiods. This average

CQMm value is denoted by CQM and is equal to:M

m=1 CQMmM .

An important point which must be mentioned hereis that by exploiting the fuzziness in the admissioncontrol system the cognitive network can support theimportant QoE-Scalability feature for giving service toheterogeneous mobile terminals with different servicequality requirements.

In the sequel, the FIS controller [19,20] is described indetail. This fuzzy model consists of four basic components.(a) Quality and interference metrics: These parameters arethe same CQM and PIM parameters discussed earlier in thetext. These parameters take three linguistic values namelyhigh (H), medium (M) and low (L) and behave as input tothe model.

140 P. Goudarzi / Physical Communication 7 (2013) 134–144

0

0.2

0.4

0.6

0.8

1

0

Low Medium HighD

egre

e of

mem

bers

hip

Fig. 2. Sample input membership functions.

0

0.2

0.4

0.6

0.8

1

0 0.5 1

AP

Deg

ree

of m

embe

rshi

p

Low Medium High

Fig. 3. Output membership function.

(b) Input membership functions: The ‘low’, ‘medium’ and‘high’ membership functions for input variables CQM andPIM are depicted as Gaussian membership functions inthe Fig. 2.(c) Output membership functions: The output of FIScontroller is AP parameter and its membership functionsare found based on a heuristic manner. For the topologyof the Fig. 4, its ‘low’, ‘medium’ and ‘high’ membershipfunctions are shown in the Fig. 3. A graphical sketch of theproposed FIS is depicted in Fig. 4.(d) Inference engine: The inference engine consists of thefollowing five simple rules:Rule1:If CQM is L and PIM is L then AP is M.Rule2:If CQM is L and PIM is H then AP is H.Rule3:If CQM is H and PIM is L then AP is L.Rule4:If CQM is H and PIM is H then AP is M.Rule5:If CQM is M or PIM isM then AP isM.

Fig. 4. Architecture of the proposed fuzzy AP controller.

The rationale behind selecting the above mentionedrules is as follows. In the network conditions wherecognitive and primary users are in good quality andinterference conditions i.e. we have small values of CQMand large PIM , we can highly admit newly arrived cognitiveusers by selecting large AP values. In the cases whereone of the CQM or PIM parameters are in mediumconditions, our network is assumed to be in close vicinityof high interference condition and so, we must select APparameter in a conservative manner to be in mediumlevels. In high network interference conditions which boththe cognitive quality and primary interferences are incritical levels (CQM is large and PIM is small) we mustselect the admittance parameter AP to its lowest valuewhich is selected to be low.

6. Numerical analysis

The numerical analysis section is comprised of twodifferent sparse and dense scenarios which have differentuser populations. The NS-2 network simulator has beenused for the analysis due to its extensive support forwireless networks [21].

6.1. Sparse scenario

ACDMAwireless networkwith a circular area of 1000mradius is considered. The CDMA wireless network consistsof one base station, two primary users (N = 2) and fourcognitive users (M = 4). The network setting is depictedin Fig. 1. The base station is located at the center, while theprimary and secondary users are randomly placed in thecircle area. It is assumed that there is one primary sessionand two secondary sessions.

The primary session is the uplink connectivity froma primary user to the base station. The two secondarysessions are all single-hop video streaming sessions. InCDMA model, the channel bandwidth is set to W =

10 MHz, the orthogonality factor δ is set to be 0.1 and thenoise power spectral density N0 = 10−13 W/Hz.

The maximum transmission power is set to 1 W for allsecondary sessions. The channel gain from the sender j tothe receiver m is given by gjm = 105/d4im, where djm is thedistance from the sender j to the receiverm.

The interference threshold at the primary receivingpoint (T ) is set to be 10−7 W.The transmission power at the

P. Goudarzi / Physical Communication 7 (2013) 134–144 141

Time

5 10 15 20 25 30 35 40 45 50

MO

S

1

2

4

5

3

Fig. 5. QoE performance of session 1 in FQAC, TQAC, S-SIAC and I-SMIRAmethods.

Table 1Normalized AWT and PWT comparison between differentmethods.

Method N AWT N PWT

TQAC 1 1FQAC 0.61 0.67S-SIAC 0.73 0.77I-SMIRA 0.8 0.91

primary user is set to be 0.5 W and packet length Lm is setto be 1024 for each cognitive user. At each video session,the Foreman CIF sequence is encoded using the JointScalable Video Model (JSVM) video codec at 30 frames persecond, using various quantization levels and for differentGroup of Pictures (GoP) lengths. The performance of thetwo secondary (cognitive) sessions 1 and 2 in the Fig. 1have been compared between the TQAC, FQAC, the S-SIACalgorithm presented in [5] and the I-SMIRA admissioncontrol algorithm presented in [6].

Packets are dropped only if they do not arrive at thereceiver by the play-out deadline. In this case, previous-frame concealment is used. Decoded video quality ismeasured in terms of the Eq. (4). For each experiment,the video sequence is looped almost for 200 times, andthe average values of all realizations are calculated, whichcan be interpreted as the expected performance of thealgorithm in a snapshot of time for the given network. It isassumed that newcognitive users arrive based on a Poissondistribution in the sequential TQAC algorithm.

In Figs. 5 and 6, the average QoE performances of twovideo cognitive sessions 1 and 2 have been compared forthe FQAC, TQAC, S-SIAC and I-SMIRAmethods. To calculatethe MOS values, Eq. (4) has been used for obtaining theQ values after calculating the packet loss associated witheachmethod. As it can be verified in Figs. 5 and 6, TQAC andFQAC have better quality performance with respect to thetraditional S-SIAC and I-SMIRA ones because the formertwo methods are designed in a QoE-centric manner whilethe latter two methods (S-SIAC and I-SMIRA) are designedbased on a QoS-centric manner which may not be a goodcandidate for the perceived video quality based on manyprevious researches such as [22] and references therein.

Another important feature for the FQAC method isthat in this method, the QoE level for the cognitive users

Time

MO

S

1

2

3

4

5

5 10 15 20 25 30 35 40 45 50

Fig. 6. QoE performance of session 2 in FQAC, TQAC, S-SIAC and I-SMIRAmethods.

has less fluctuation with respect to other methods. Thisproperty helps the existing cognitive users to experiencea more stable quality level during the media streaming(although this level may be lower with respect to TQAC).The fluctuation effect during the media playback is thedirect consequence of the interference effect imposed fromthe newly admitted users during the admission phases.

In another simulation, the normalized average waitingtime (NAWT) for the newly arrived cognitive users hasbeen compared in the FQAC, TQAC, S-SIAC and I-SMIRAmethods. The waiting times are normalized based on theTQAC algorithm. The simulation time is considered to be50 s. The average is taken based on 100 simulation runsduring the mentioned simulation time. The normalizedpeak waiting time (NPWT) is calculated in each methodduring the simulation runs. In Table 1, the results havebeen summarized. As it can be deduced from the Table 1,the lowest NAWT and NPWT values are associated withthe FQAC method. The reason is due to removing thetight quality guaranties in the admission process in FQACmethod in comparison with the other methods. In fact, theFQAC trades the waiting time for quality in comparisonwith TQAC method. This makes the FQAC method a goodcandidate for time sensitive and quality-scalable real-timestreaming applications such asmobile television ormobileconferencing.

In Fig. 7, the blocking probability performance ofthe proposed FQAC, TQAC methods have been comparedwith the traditional S-SIAC method. As the philosophy ofdesigning the FQAC method is the reducing the blockingprobability and its associated waiting time, the blockingprobability performance of this method is better thanother ones and has a better average blocking probabilityperformance with respect to other methods. But this factcan result in an increased outage probability in thismethodwith respect to other ones as Fig. 8 shows.

This is because of the fact that there always exists atradeoff between the outage and blocking probabilities.The reason why the I-SMIRA method has a higher outageprobability is the fact that it clearly has a zero blockingprobability. As mentioned earlier, main application ofsuch soft admission control schemes are QoE scalableapplications such asmobile TV etc., in which the importantparameter is the perceived video quality and so this

142 P. Goudarzi / Physical Communication 7 (2013) 134–144

8 16SINR(dB)

Blo

ckin

g P

roba

bilit

y

0.15

0.3

0

Fig. 7. Blocking probability in FQAC, TQAC and S-SIAC methods.

8 16

0.15

0.3

SINR(dB)

Out

age

Pro

babi

lity

Fig. 8. Outage probability in FQAC, TQAC, S-SIAC and I-SMIRA methods.

increase in the outage probability has no significant effecton these types of applications.

As can be verified in these figures, there is no largeperformance gap between I-SMIRA and S-SIAC and theproposed algorithms in blocking and outage probabilities(∼3%), because the main concern here is enhancing theperceived quality of experience of the cognitive users byappropriate admission control techniques and so we payless attention on features like the blocking and outageprobabilities in the current work.

Another important concern is comparing the power/energy efficiency of the proposed methods with the tradi-tional ones. By definition, power/energy efficiency is theaverage number of successful transmissions (in bits) perpower/energy unit (in mW) and can be measured by theunit of (bits/mW) [23]. The mentioned average is taken intwo dimensions which are considered to be time and thenumber of the active cognitive users. We have computedand compared the normalized power efficiency (NPE) ofthe proposed algorithms versus the traditional ones in dif-ferent SINR values inwhich the normalization is performedwith respect to the bestmethod in each SINR value. The nu-merical results are summarized in the Table 2.

As can be verified in the Table 2, the proposed quality-centricmethods have showncomparableNPEperformance

Table 2NPE performance comparison between different methods.

Method SINR = −5 dB SINR = 0 dB SINR = 5 dB

TQAC 1 0.82 0.82FQAC 0.91 0.97 1S-SIAC 0.95 1 0.95I-SMIRA 0.73 0.85 0.85

in comparison with the traditional ones. As can be verifiedin this table, by increasing the SINR value, FQAC showssuperior NPE performance. This event can be due to thefact that in the low interference regimes (high SINR),FQAC can perform the admission process more efficientlydue to the lower probability of transmission errors (andhence larger probability of successful transmissions perpower unit) for cognitive users. In high and middleinterference regimes, TQAC and S-SIAC shows superiorNPEperformances respectively.

6.2. Dense scenario

In the current scenario, a more populated scenario isconsidered which contains one base station, 20 activeprimary users (N = 20) and 30 cognitive users (M =

30). Other simulation parameters are the same as thesparse scenario. The same as the sparse scenario, foreach experiment, the video sequence is looped almostfor 200 times, and the average values of all realizationsare calculated, which can be interpreted as the expectedperformance of the algorithm in a snapshot of time for thegiven network. We have depicted and compared the QoEperformances of two typical cognitive users 5 and 10 in thesimulation time (50 s) for different methods in the Figs. 9and 10. As can be verified in the Fig. 9, FQAC and TQAChave superior average QoE performance in comparisonwith the traditional I-SMIRA and S-SIAC methods even inthe mentioned dense scenario. The cause of the reducedquality level in the current scenario in comparison withthe sparse scenario is the increased interference due tothe larger number of concurrent cognitive users in densescenario. To calculate the required time for convergencein the FQAC and TQAC algorithms, we must first definea measure for calculating the convergence-time because,due to the facts that the time-varying background trafficassociated with the concurrent cognitive users alwaysexists during the simulation time and also the arrivaland departure events of different cognitive users duringsimulation time, the cognitive users’ quality of experiencealways fluctuates in time. So, we define the convergence-time for a cognitive user as the time t∗ after which theQoE of that cognitive user remains in a pre-determinedgap (g) around the optimal point. For more clearance, atypical plot of this concept for time-varying and dynamicscenarios is given in the Fig. 11. By considering the gapg = 1, the approximate required time for converging theFQAC and TQAC algorithms is about 7 and 23 s for cognitiveuser 5 and 4 and 40 s for cognitive user 10 respectively.The greater convergence time of the TQAC method is dueto more quality fluctuations in this method in comparisonwith the FQAC method. In other words, the reducedconvergence time in the FQAC algorithm is the result of

P. Goudarzi / Physical Communication 7 (2013) 134–144 143

5 10 15 20 25 30 35 40 45 50

Time

MO

S

1

2

3

4

5

Fig. 9. QoE performance of cognitive user 5 in different methods.

5 10 15 20 25 30 35 40 45 50

Time

MO

S

1

2

3

4

5

Fig. 10. QoE performance of cognitive user 10 in different methods.

Fig. 11. The convergence-time definition in dynamic scenarios.

more smooth admission phase in this algorithmwhich canbe translated into a higher convergence speed for activecognitive users. As an example, on average (by consideringall of the cognitive users), the proposed FQAC methodpresents about 27% quality improvement in comparisonwith the traditional I-SMIRA method.

7. Conclusions

In the current work, a quality-centric soft admissioncontrol scheme has been introduced. Traditional hard

admission control schemes do not consider the users’ per-ception for admitting newly arrived cognitive users. Theyconsider only some quality of service features such as SINRetc., for admission purposes. On the other hand, arrival ofnewly admitted cognitive users can have negative side ef-fects on the perceived QoE level of each cognitive user.In the current work we tried to address the mentionedissues by introducing the fuzzy logic concept in the ad-mission process. Furthermore, the users’ quality of expe-rience is modeled through the well-known IQX qualitymappingmodel. The simulation results show the improve-ments achieved in the proposed method with respect tosome traditional methods such as I-SMIRA and S-SIAC.

Furthermore, the blocking probability in the proposedFQAC schemewas lower than othermethods. Themain ap-plication of such algorithms is in QoE scalable applicationssuch as mobile television, etc., which can tolerate slightquality reductions during the session. Many open researchissues remain to consider after this work. For example in-corporating the Fuzzy 2 techniques in such applicationsmay result in improved performance because there alwaysexist some forms of the uncertainty in selecting the form ofthe input and output membership functions of the FIS dueto uncertainties in the quality of experience estimation insuch dynamic networks and different user requirements.On the other hand,more exactQoE estimationmethods canresult in improved system performance.

References

[1] Fcc, spectrum police task force report, 2002.[2] H. Jiang,W. Zhang, X. Shen, Q. Bi, Quality of-service provisioning and

efficient resource utilization in CDMA cellular communications, in:IEEE JSAC, vol. 24, pp. 4–15, 2006.

[3] M. Fiedler, T. Hossfeld, P. Tran-Gia, A generic quantitative relation-ship between quality of experience and quality of service, IEEE Net-work (2010).

[4] Methods for subjective determination of transmission, ITU-TRecommendation P.800, 1996.

[5] M. Farahani, A. Ghasemi, S.M. Razavizadeh, Sequential interference-aware admission control in underlay cognitive radio networks, In-ternational Journal of Information and Communication Technology2 (2010) 33–40.

[6] L. Le, E. Hossain, Resource allocation for spectrum underlay in cogni-tive radio networks, IEEE Transactions onWireless Communications7 (2008) 5306–5315.

[7] M. Andersin, Z. Rosberg, J. Zander, Gradual removals in cellular PCSwith constrained power control and noise, in: IEEE PIMRC, 1995,pp. 56–60.

[8] L. Zhang, Y. Liang, Y. Xin, Joint admission control and powerallocation for cognitive radio networks, in: IEEE ICASSP, 2007,pp. 673–676.

[9] H. Gu, C. Yang, Power and admission control for UWB cognitive radionetworks, in: IEEE ICC, 2008, pp. 4933–4937.

[10] J. Xiang, Y. Zhang, T. Skeie, Joint admission and power con-trol for cognitive radio cellular networks, in: IEEE ICCS, 2008,pp. 1519–1523.

[11] D.I. Kim, L.B. Le, E. Hossain, Joint rate and power allocation forcognitive radios in dynamic spectrum access environment, IEEETransactions on Wireless Communications 7 (2008) 5517–5527.

[12] N. Nie, C. Comaniciu, Adaptive channel allocation spectrumetiquettefor cognitive radio networks, in: IEEE DySPAN, 2005.

[13] T. Shu, M. Krunz, S. Vrudhula, Joint optimization of transmitpower-time and bit energy efficiency in CDMA wireless sensornetworks, IEEE Transactions on Wireless Communications 5 (2006)3109–3118.

[14] R.D. Yates, D.J. Goodman, Probability and Stochastic Processes,A Friendly Introduction for Electrical and Computer Engineering,second ed., John Wiley and Sons Inc., 2004.

144 P. Goudarzi / Physical Communication 7 (2013) 134–144

[15] H.l. Kim, S.G. Choi, A study on a QoS/QoE correlation model for QoEevaluation on IPTV service, in: The 12th International Conference onAdvanced Communication Technology, ICACT, 2010.

[16] Opinion model for video-telephony applications, ITU-T Recommen-dation G.1070, 2007.

[17] S.A. Grandhi, J. Zander, Constrained power control in cellular radiosystems, in: IEEE VTC, 1994, pp. 824–828.

[18] M. Andersin, Z. Rosberg, J. Zander, Soft and safe admission controlin cellular networks, IEEE Transactions on Networking 5 (1997)255–265.

[19] L. Zadeh, Fuzzy sets, Information and Control (1965).[20] T. Takagi, M. Sugeno, Fuzzy identification of systems and its

applications to modeling and control, IEEE Transactions on SystemsMan and Cybernetics 15 (1985) 116–132.

[21] The Network Simulator 2, Online available at:www.isi.edu/nsnam/ns/.

[22] Z. Wang, L. Lu, A.C. Bovik, Video quality assessment based onstructural distortion measurement, Signal Processing 19 (2004).

[23] C.E. Jones, K.M. Sivalingam, P. Agrawal, J.C. Chen, A survey ofenergy efficient network protocols for wireless networks, WirelessNetworks 7 (2001) 343–358.

Pejman Goudarzi was born in Shiraz-Iran in1972. He received his B.Sc. in Electronics fromSharif University of Technology Tehran-Iran in1995. He also received his M.Sc. and Ph.D.in Communications and Electrical Engineeringboth from Isfahan University of Technology,Isfahan-Iran in 1998 and 2004 respectively. Dr.Goudarzi is an assistant professor and facultymember in the Research Institute for ICT (exITRC) since 2004. His main research interestsare: wireless video communication, distributed

rate allocation algorithms and congestion control in data networks.