research article qos oriented multiobjective optimizer for...
TRANSCRIPT
Research ArticleQoS Oriented Multiobjective Optimizer for Radio ResourceManagement of LTE-A Femtocells
Ayesha Haider Ali and Muhammad Mohsin Nazir
Department of Computer Science Lahore College for Women University Jail Road Lahore 54000 Pakistan
Correspondence should be addressed to Ayesha Haider Ali ayeshaiqbalgmailcom
Received 8 March 2016 Accepted 16 June 2016
Academic Editor Pedro M Ruiz
Copyright copy 2016 A Haider Ali and M M Nazir This is an open access article distributed under the Creative CommonsAttribution License which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited
The future wireless networks support multimedia applications and require ensuring quality of the services they provide Withincreasing number of users the radio resource is becoming scarce Therefore how should the demands for higher data rates withlimited resources be met for Long Term Evolution-Advanced (LTE-A) is turning out to be a vital issue In this research paper wehave proposed an innovative approach for Radio Resource Management (RRM) that makes use of the evolutionary multiobjectiveoptimization (MOO) technique for Quality of Service (QoS) facilitation and embeds it with the modern techniques for RRMWe have proposed a novel Multiobjective Optimizer (MOZ) that selects an optimal solution out of a Pareto optimal (PO) set inaccordance with the users QoS requirements We then elaborate the scheduling process and prove through performance evaluationthat use of MOO can provide potential solutions for solving the problems for resource allocation in the advancement of LTE-A networks Simulations are carried out using LTE-Sim simulator and the results reveal that MOZ outperforms the referencealgorithm in terms of throughput guarantees delay bounds and reduced packet loss Additionally it is capable of achieving higherthroughput and lower delay by giving equal transmission opportunity to all users and achieves 100 accuracy in terms of selectingoptimal solution
1 Introduction
LTE-Advanced (LTE-A) is an emerging wireless accessnetwork technology that has been considered the 4th-Generation (4G) wireless system to offer higher bandwidthfor Internet access It offers higher data rates up to 1 Gbps [1]Radio Resource Management (RRM) in LTE-A system takesinto account existing spectrum both in time and in frequencydomains Fulfilling theQuality of Service (QoS) requirementsis more challenging in wireless networks due to
(i) limited radio resourcesspectrum(ii) channel conditions(iii) existence of multiple users with diverse QoS require-
ments
The foremost goal of LTE-A systems is to improve serviceprovisioning and reduce the cost of user and operators Thiscan be fulfilled by enhancing the quality of the system with
reference to data rates system capacity coverage throughputand reduced latency As certain traffic classesflows havestringent QoS requirements for example video and VOIPthus satisfying their requirements is essential to maintainnetwork operations smoothly As the radio resource is scarcethere is a need to have mechanisms to distribute this scarceand valuable resource efficiently to ensure QoS of individualusers for development of services for next generation telecomnetworks Thus our main objective in this research is todeal with QoS in terms of throughput and delay We aimto maximize throughput and minimize delay and these twoobjectives will form our objective function Figure 1 lists thedifferent applications and their QoS requirements in relationto delay and throughput
In LTE-A Physical Resource Block (PRB or simply RB)is the smallest user assignment resource unit for resourcescheduling In this allocation phase the bandwidth is dividedinto portions called Resource Chunks (RCs) [2] Variablenumbers of RCs are allocated to different User Equipment
Hindawi Publishing CorporationMobile Information SystemsVolume 2016 Article ID 7964359 13 pageshttpdxdoiorg10115520167964359
2 Mobile Information Systems
File download
Streaming video
Web browsing Email
Throughput sensitivity
Interactive video
Interactive voice applications
GamingStreaming
video
Delay sensitivity
Figure 1 Varying QoS requirement classification
middot middot middot N minus 2 N minus 1 N
N
1
1 2
PRBs
RCs
2 3 4 5 6
Figure 2 A total of M RCs dividing the total bandwidth into 119873PRBs
(UE) for various applications Figure 2 depicts this conceptTo guarantee QoS we need to select RB that will fulfill bothour objectives of throughput and delay guarantees
RRM is a set of techniques that are utilized to optimizethe usage of spectrum and equipmentThemotivation behindRRM functionality is to guarantee the provision of networkservices and QoS and optimize the system usage The real-ization of a wireless system is dependent on the bandwidthallocation and the QoS gains of the transmissionmedium [3]Now that data rate provisioning is heading towards the nextgeneration designing techniques for efficient radio resourceallocation is becoming more challenging These challengesresult due to the fact that the packets arrive randomly andthere is a lack of buffer space plus the QoS needs are erraticand thus resource management and allocation techniquesneed to be developed
Ourwork takes advantage of themultiobjective optimiza-tion (MOO) techniques that make use of a decision maker(DM) to take radio resource allocation decisions The solepurpose is to provide RRM technique that will ensure QoSprovisioning to all types of traffic specifically for increasedthroughput and reduced delay There exist evolutionaryalgorithms that have various schemes and techniques formanaging resources [4] In MOO the objective vectors areviewed as optimal if none of its elements can be furtherimproved without deteriorating at least one of the otherelements [5]
Most real world problems have more than one objectivefunction and when we have multiple objectives to attainwe need efficient solutions to solve the problem Also therelationship between objectives is generally rather perplexingand cannot be measured using the same standard thusmaking it hard to cumulate them into a single objectiveIn that case we do not have a solitary optimal solutionbut rather a collection of substitutes with varying trade-offswhich are called Pareto optimal (PO) sets Usually one ofthese PO solutions is selected at the end So in multiobjective
optimization we have two tasks first finding a PO set andsecond a decision-making processmodule that will select adistinct solution from this set also referred to as DM Thuswhen we have diverse goals with conflicting objectives theycannot be summed up into one objective function and thisencourages the use of various MOO techniques
Despite the fact that LTE deployment scenarios deliverhigh data rates they also showcase new challenges for inter-ference management and RRM To meet these challengespolicies are intended for the standard cellular networks Alsothe femtocells are ad hoc in nature and this fact tends tobind the scope of the algorithms under consideration Thusproficient RRM procedures are critical to restrain the impactof interference on LTE femtocells performance [6]
A framework for Time Domain (TD) scheduling withcongestion control and QoS provisioning is also introducedIt optimizes the QoS provisioning process for domains ofboth time and frequency by taking into account channel con-dition the status of the queues and the userrsquos QoS require-ments This framework enhances the process of resourcesutilization by considering the QoS requirements of variousservice classes The results indicate better QoS of Real Time(RT) traffic and fair resource sharing of available resources forthe Non-Real Time (NRT) traffic [7]
The idea of cross layer resource allocation for OrthogonalFrequency Domain Multiplexing (OFDM) scheme is consid-ered significant as it takes into account the channel conditionthe arbitrary nature of traffic QoS needs and fair distributionof resources between the users Time slots frequenciesand carriers are assigned dynamically to make this processeffective [8] A number of studies exist in literature that focuson the RRM of LTE-A considering the QoS requirements ofthe user Some make use of self-optimization some tend tofollow the resource scheduling approaches others focus onspectrum splitting using the time and frequency domainsand some have designed solution for RRM by using geneticalgorithms [9ndash15]
The algorithms that manipulate multiuser environmentare the MAX CI and Proportional Fairness (PF) algorithmsIn MAX CI algorithm RB is allocated to a user who tendsto achieve the highest data rates for the current time slotand tries to maximize overall system throughput The PFalgorithm also considers fairness and resources are allocatedon the basis of priority and allocation schemes [16]
However these algorithms target to improve utilizationof resources considering the channel conditions and interfer-ence mitigation only and the delay and throughput require-ments are not taken into much consideration A number oftechniques have also been proposed that aim to efficientlyuse the radio resources to provide guaranteed QoS for var-ious types of traffic [17ndash19] A Delay Prioritized Scheduling(DPS) algorithm [20] has been developed that manages thedistribution of RBs by choosing those RBs that best fulfill thedelay threshold condition With increasing users the delayalso tends to increase due to more video requests that arethroughput sensitive It is a packet scheduling algorithm thatselects RBs according to their SNR levels and then admitsand assigns the flow to the best possible RBs This is a two-way process which first selects a user according to a delay
Mobile Information Systems 3
threshold and then assigns RBs It focuses mainly on delayminimization
In the perspective of multiobjective optimization mostwork done is towards discovering a collection of near-Pareto optimal solutions algorithmically In multiobjectiveoptimization communication with the DM can be doneduring the phase of optimization or during the final decision-making part In many studies a human decision maker isinvolved after a variety of solutions have been sought out [21ndash23]
The process of multiobjective optimization is incompletewithout a decision-making activity that will take the finaldecision based on various alternatives available In thiscontext many interactive multiobjective optimization tech-niques are available under the title Multicriteria Decision-Making (MCDM) [24ndash26] All the techniques are differentfrom each other but all incorporate DM to provide infor-mation to help in taking the final decision The evolutionaryalgorithms (EA) suggest the use of natural evolution theoryfor optimization like the survival of the fittest theory ofDarwin They work with a set of solutions and can also beused to find a partial PO set
MOO has been researched for many years and is focusedon the theoretical aspects [27] A lot of approaches havebeen formulatedwithmathematical programming theory forexample nonlinear programming to solve the multiobjectiveoptimization problems [28] Varying interactive approacheshave also been used in which the information is given tothe DM and the DM specifies its preferences This proceduredepends on the type of problem and its mathematical prop-erties and scalar function [29]
A power-delay minimization scheme is also introducedthat makes use of linear programming [30] In this studymultiobjective problem is transformed into a solitary objec-tive by using weighted sum technique that aims to reduceboth the delay and transmit power Simulated annealing andgreedy heuristic algorithms are also part of this work Itstarget is not specifically LTE networks but mainly the IEEE80211 Wireless LANs (WLANs) and Green Wireless AccessNetworks (GWANs)
Table 1 depicts the classification of various techniquesused for RRM and for finding an optimal solution withmultiple objectives Based on the literature survey we cansee that there exist various techniques for QoS provisionin LTE-A using different methodologies Also the multiob-jective optimization offers a variety of solutions for solvingvarious linear and nonlinear problems Still there is a lackof these multiobjective optimization technique applicationsfor RRM of LTE-A targeted towards ensuring QoS As aresult we will propose a QoS awareMultiobjective Optimizer(MOZ) for the RRM of LTE-A which will take into accountthe provision of maximum throughput and minimumdelay
2 Materials and Methods
21 Problem Formulation To measure the performance ofLTE-A network we have a set of criteria such as through-put end-to-end transmission delay energy efficiency and
Table 1 Classification of literature on RRM and optimization
Approach [7] [16] [20] [21ndash23] [30] Proposed approachScheduling radic radic radic radic radic radic
User association radic radic radic mdash mdash radic
QoS aware radic radic radic mdash radic radic
Energy efficiency mdash radic mdash mdash radic mdashRB allocation mdash mdash radic mdash radic radic
LTE radic mdash mdash mdash radic mdashLTE-A mdash radic radic mdash mdash radic
MOO mdash mdash mdash radic radic radic
DM mdash mdash mdash radic mdash radic
transmission strength The purpose of the work presentedhere is to determine the trade-offs that arise while choosingperformance metrics for assigning resources to the UserEquipment (UE) The multiobjective Pareto optimizationconsists of three steps
(1) Multiobjective problem definition(2) Optimization (finding Pareto optimal solutions)(3) Decision-making (role of DM)
In the following we will elaborate the working of our QoSaware optimizer according to the above-mentioned threesteps and it will define the framework for our proposedoptimizer
Step 1 (problem definition) To solve our multiobjectiveoptimization problem we will involve DM to find the best(optimal) solution By optimal solution we are referring toPareto optimal solution that the DM considers as the bestoption In our case we have two objectives the first one ismaximum throughput and the second one is minimumdelayWe will manage the resources in such a way that we are ableto select RB that transmits the data according to the above-mentioned objectives
ParetoOptimization ProblemWewill solve themultiobjectiveoptimization problem that will take the form
MaximizeMinimize 1198911(119910) 119891
2(119910) 119891
119894(119910)
Subject to 119910 isin 119878
(1)
where 119894 is a set of objectives that will be minimized ormaximized according to its definition 119878 is a set of networkconstraints Here 119910 is the decision vector and 1199101015840 is thedecision variable We will call 1199101015840 isin 119878 Pareto optimal only ifthere is no other 1199101015840 isin 119878 such that
119891119896(119910) le 119891
119896(1199101015840) (where 119896 = 1 119894) (2)
This means that 1199101015840 is Pareto optimal only if there is noother possible vector (or solution)119910 thatwill deteriorate somecriterion without leading towards an increase in some othercriterion
4 Mobile Information Systems
Step 2 (Pareto front) In our case we have the followingobjective functions and the first one aims at maximizingnetwork utility through increased throughput (119879
119894)
Maximize119899
sum
119894=1
(119879119894)
Subject to119899
sum
119894=1
119879119894lt 119862119894119896(119905) where 119879
119894gt 0 forall119894 isin 119873
(119873 = total UE instances)
(3)
Considering channel 119896 and user 119894 we define the channelcapacity as
119862119894119896= RB log
2(1 +
120588
1205902sdot 119875) (4)
where RB is RB Bandwidth which is calculated as (totalbandwidthnumber of RCs) 119875 is Transmission Power 120588 isSNR space determined by Bit Error Rate (BER) and 1205902 isnoise power density
119879119894is the average throughput calculated as
119879119894= TUE
119894 [119905]1
120591+ 119879119894(119905 minus 1) (1 minus
1
120591) forall119894 isin 119873 (5)
where TUE119894[119905] is throughput of UE
119894in time instance 119905 120591 is
time constraint of the smoothing filter and119873 is total UETUE119894[119905] is calculated with the help of the following
equation
TUE119894 [119905] = RB log
21 + SNR (6)
where SNR = 119879119901times 119878UE(Noise + Interference) (119879
119901is
Transmission Power 119878UE is signal gain of UE)Our next objective is defined as follows
Minimize119873
sum
119894=1
(119863119894(119905))
Subject to 119863119894(119905) lt DB
119894
(7)
where DB119894is the delay budget or the upper bound which is
equivalent to 20ms in OFDMA networks [19 31] The delayexperienced by the UE
119894should be less than this upper bound
Here 119873 represents total active UE and 119863119894(119905) is the HOL
(Head of Line) delay for UE119894at time 119905 calculated as
119863119894(119905) =
119882119894
DT119894
(8)
where119882119894is the waiting time for UE
119894and DT
119894is the normal-
ized HOL delay obtained by dividing each userrsquos waiting timeby DB
119894
This process will generate a set of Pareto optimal solutionscalled the Pareto front As a result there is no single POsolution so the question is which solution to select from a setof PO solutionsThe answer lies in Step 3 that takes advantageof DM
Step 3 (the optimal solution) In this step the DM will selectthe best solution which is RB that will be allocated fortransmission Depending on the type of incoming traffic theDM will decide which strategy to follow according to thefollowing criteria
(i) If the traffic is delay sensitive preference will be givento minimizing the delay whereas throughput can becompromised to some extent
(ii) If the incoming packet is throughput sensitive it willemphasize maximum throughout whereas delay maybe compromised
This scheme will thus allow the delay sensitive applica-tions to sacrifice throughput for lower delays without anyeffect on the throughput sensitive applications
So we have 119896 and 119894 objective functions (represented by 119892)and a total of 119898 and 119891 constraints (represented by ℎ) Weassume that the constraints plus the objective function arethe functions of the decision vectors So the DMrsquos goal is asfollows
Maximize 119892 (119909) = 1198921(119910) 119892
2(119910) 119892
119894(119910)
Minimize 119892 (119910) = 1198921(119911) 119892
2(119911) 119892
119896(119911)
Subject to ℎ (119909) = ℎ1(119910) ℎ
2(119910) ℎ
119898(119910)
ℎ (119910) = ℎ1(119911) ℎ
2(119911) ℎ
119891(119911)
119909 = 1199091 1199092 119909
119894
119911 = 1199111 1199112 119911
119896
(9)
where 119909 is the decision vector for first objective function and119911 is the decision vector for second objective function 119874lowast isin119909 or 119911 is the feasible decision or solution to our optimizationproblem 119874lowast is Pareto optimal as we have no other bettersolution So here the DM will make use of a weighted MinndashMax Approach which is adopted from GameTheory [32 33]In this scheme we will compare the relative deviation froma separately attainable minima or maxima In our case it isthe deviation from maximum throughput and delay of thesystem We will denote this by 119863max119894 (delay variation) and119879min119894 (throughput variation) We will calculate the relativedeviations using
119879min119894 = TA119904minus 119879119894
(For throughput sensitive applications) (10)
TA119904is the maximum attainable throughput of the LTE
system calculated as follows(i) We will first calculate the total RBs assuming the
bandwidth of channel is 20MHz by using the following
carriers timesOFDM symbols times Slots times RBs (11)
where carriers are 12 OFDM symbols are 7 RBs are 100 Slotsare 2 The final value is 16800 RBs per frame
Mobile Information Systems 5
(ii) Second we consider modulation of 64 QAM and asingle modulation symbol carries 6 bits The total bits willthus be
16800
times 6 bits per symbol of modulation equal to 1008Mb(12)
(iii) Thirdly considering the 4-by-4 MIMO we get
4 times 1008Mb = 403Mb (13)
This is the peak data rate(iv) Finally we calculate the overhead which will be about
25 So we have
403Mb times 075 = 302Mb (14)
So we can calculate TA119904and RC and minimum 119879min119894 is
selected by the DM for transmission That is resource chunkRC119894as in the following
119874lowast= RC119894= min119879min119894 forall119894 isin UE (15)
Now let us talk about the delay sensitive applicationswhere we will find the feasible solution 119874lowast as follows
119863max119894 = DB119894minus 119863119894(119905)
(For delay sensitive applications)
119874lowast= RC119894= max119863max119894 forall119894 isin UE
(16)
So 119874lowast is chosen for transmission based on the followingdecision vectors
119874lowast=
argmin 119879min119894 forallUE119894
argmax 119863max119894 forallUE119894
(17)
The DM will take its decision based on the following setof constraints
119899
sum
119894=1
119879119894lt 119862119894119896(119905) where 119879
119894gt 0 forall119894 isin 119873 (18)
119863119894(119905) lt DB
119894forall119894 isin 119873 (19)
119873
sum
119894=1
RB119894le Total Badndwidth (20)
sum
119895isin119869
120579119894119895ge 1 forall119894 isin 119873 forall119869 isin Total BS (21)
Equation (18) refers to channel capacity constraint and(19) refers to delay bound constraint Equation (20) is RBbandwidth constraint Equation (21) ensures that at least 1 BScovers the active UE
22 The Decision-Making Process After the optimized solu-tion is sought out by theDM the trafficwill then be scheduledin output queues discussed in detail in the next subsection
Figure 3 depicts the detail of decision-making process insidethe DM Its details are explained as follows
(1) Traffic Type Determination The DM will take as input thePareto front and will first determine the type of traffic whichin our case is either throughput or delay sensitive
(2) Optimization Process After determining the traffic typethe following actions are taken
(i) For throughput sensitive applications it will calculatethe value of 119879min119894 based on the value of 119879
119894 and then
it will select the optimal solution 119874lowast = RC119894=
min119879min119894(ii) For delay sensitive applications it will calculate the
value of 119863max119894 based on the value of 119863119894(119905) and then
it will select the optimal solution 119874lowast = RC119894=
max119863max119894(3) RC Allocation Matrix Update After that the RC assign-ment matrix will be updated The matrix 119898
119896119899is represented
as
119898119896119899=
0 if RC119899is unassigned
1 if RC119899is assigned to service group 119896
(22)
where 119896 is type of traffic and 119899 is RC number [34]
(4) RC Availability Check If there are RCs availableleft forassignment then the process starts again andmoves to traffictype determination otherwise the process ends
In the following we explain in detail the construction ofPareto front
23 Constructing Pareto Front This step describes in detailthe process of constructing a representative Pareto front asfollows
Step 1 Determine the traffic type of the incoming packetAssume 119863
1 119863
119899are the alternative solutions in terms of
119894 and 119895 where 119894 represents throughput sensitive flows and 119895represents delay sensitive flows
Step 2 Find the largest 119894 such that 119863119894gt 119863119899(where 119863
119899
represents all other UE) and add119863119894to Pareto front
Step 3 Find the smallest 119895 such that 119863119895lt 119863119899(where 119863
119899
represents all other UE) and add119863119895to Pareto front
Step 4 Repeat Steps 2 and 3 until no such 119894 and 119895 exist and wehave the Pareto front of the form (119863
119894 119863
119899119863119895 119863
119899)
Note that 119863119894is sorted in order of decreasing throughput
and 119863119895is sorted in order of increasing delay The process
generates a set of candidate solutions that will undergo theoptimization phase for selecting a singled optimal solutionThe selection of the optimal solution is based on the efficiencyof the DM which not only considers the resource require-ments of the users but also sets priorities We now present analgorithm that will make use of the defined framework andconsiders the requirements of each step
6 Mobile Information Systems
Start
Throughput sensitive traffic Delay sensitive trafficDetermine from thePareto front the traffic
requirements
On the basis ofvarious values ofTi calculate Tmini
various values ofOn the basis of
Di(t) calculate Dmaxi
Choose optimalsolution based on
Olowast= RCi = minTmini
Choose optimalsolution based on
Olowast= RCi = maxDmaxi
Update RCallocation matrix
RCs left to be allocated Yes
No
End
Figure 3The decision-making process inside the DM for RC selection based on the objectives of maximum throughput andminimumdelay
24 The Multiobjective Optimizer Algorithm The detailedworking of proposed MOZ is outlined in Figure 4 TheMOZconsists of three main steps as follows
Step 1 The traffic classifier will take as input traffic fromvarious UE instances and will divide them into 2 queues oneconsists of packets that have stringent throughput require-ment and the other queue consists of packets from delaysensitive applications
Step 2 The information from Step 1 is then passed onto theoptimizer that will find out the Pareto front and will send itsdecision set to the decision maker
Step 3 The DM will make the final decision based on theinformation it has and according to its decision-makingcapabilities
Step 4 Finally119874lowast is sent out to the queue for scheduling andthe process then diverts to Step 2 to get the next 119874lowast later tobe scheduled
Step 5 The process is completed when either no RC is left orthere are no more incoming packets
Steps 2 and 3 are interrelated as the DM is way dependenton the input it gets from the Pareto front so constructing alegitimate Pareto front is itself a very important task forMOZThe decision-making process inside the DM is characterizedby a number of factors already explained and is one of themajor components of our proposed strategy
Compared to most algorithms and frameworks designedto ensure QoS and to manage RRM the distinct differencesof our proposed approach are as follows (1) it makes useof multiobjective optimization that deals with conflictingobjectives simultaneously instead of converting them intoa single objective (2) It makes use of a decision makerthat has the decision-making capability and works basedon the information it has and various algorithms used(3) Multiple UE instances are served based on the type ofincoming trafficflow instead of allocating RBs to just oneflow
Mobile Information Systems 7
Trafficclassification Optimization Decision-making Scheduling
Input trafficfrom variousUE instances
applicationsDelay sensitive
Throughput sensitiveapplications
Theoptimizationprocess todeterminethe Paretofront
DetermineO
lowast (theoptimalsolution)
Figure 4 The proposed MOZ
The goal of using MOZ is to always get an optimalsolution which is chosen from a set of solutions in the formof Pareto front
25 Queuing and Scheduling Now that we have the Paretooptimal solution in hand the packets are now ready tobe scheduled in their respective queues for transmission asfollows
Step 1 In this step we will define the priority metric for delaysensitive flows The priority of UE
119894at time 119905 119875
119889119894(119905) is
119875119889119894(119905) = 119863max119894 lowast 119866119894119895 (23)
where 119866119894119895is channel gain of UE
119894on RC
119895calculated as
119866119894119895= 10
Path loss10lowast 10
shadow fading10
lowast 10multipath fading10
(24)
Step 2 We will then form a delay queue for a given UE119894that
will ensure delay bounds as follows
119863119894(119905 + 1) = [119863
119894(119905) minus 120588119877
119894(119905)] +
119873
sum
119894=1
119876119894(119905) (25)
where 119876119894(119905) is the mean queue length at time interval 119905 and
119873 is total UE scheduled and 120588119877119894(119905) is the service rate In
each time interval the UEwith highest priority occupies frontposition in the queue
Step 3 For the priority metric of throughput sensitive flowswe will use the following
119875119905119894(119905) = 120590
119894lowast 119879min (119894) (26)
where 120590119894is the Head of Line (HOL) blocking and 119875
119905119894(119905) is the
priority metric for throughput sensitive applications In eachtime interval the UE instances are scheduled according to theascending order of their priority values
It follows the following constraints
119876119894119896
st forall119894 isin 119873 119896 isin delay sensitive traffic only (27)
119876119894119895
st forall119894 isin 119873 and 119895
isin throughput sensitive traffic only(28)
119894 = 119895 Queue Length
= Length (119876119894119896+ 119876119894119895)
(29)
Equation (27) ensures that the queue 119876119894takes as input
delay sensitive traffic only (28) ensures that the queue 119876119894119895
takes in throughput sensitive traffic only and (29) states thatqueue length is the total of both the queues
Thus the defined framework and the optimizer will worktogether to form our proposed model The whole process hasfour main goals
(1) Determining the target objectives(2) Based on the objectives forming a system model that
will take into account the objective functions to beminimized or maximized
(3) Constructing a representative solution pool to aid thedecision maker
(4) Introducing a decision-making entity that will notonly ease the task of scheduling but also find anoptimal solution to satisfy the defined objectives
In the following section we will define the simulationmodel and will discuss the results
3 Simulation and Results
31 Simulation Model Designing the model for LTE-A per-formance evaluation is generally more complex comparedto a 3G system An interface is used between variousevolvedNodeBs (eNBs ie the BS) in the LTE system toease the traffic load that occurs due to handovers In this
8 Mobile Information Systems
Table 2 Simulation parameters
Parameters ValuesCarrier frequency 2GHzBandwidth 10MHzNumber of subcarriers 300Number of RBs 45TTI 1msPath loss model Free space propagation modelTransmit power 30mWFading model RayleighPacket model Poisson modelMultiplexing MIMONumber of UE instances 20
section we will design the simulation model for testing andwill evaluate our system for examining delay throughputPacket Loss Ratio (PLR) the effect of SNR and cell spectralefficiency
We are considering a cell with the capacity of 10MHz andit consists of 45 RBs and the carrier frequency is 2GHz Allthe available RBs are controlled by the eNodeB The RBs aredivided among theUE instances in the formof RCsThe LTE-A parameters used for simulation are described in Table 2
We then calculate the number of bits in each RB (ie itsdata rate) It can be computed as follows
119887119894RC = NStti lowastNS119904 lowast 119878rb (30)
where 119887119894RC is achievable data rate NStti is number of slots in
each TTI NS119904is number of symbols in each slot and 119878rb is
number of subcarriers in each RBEach UE is assigned a queue according to the traffic type
and parameters as described in the system model previouslyAfter that we will calculate the channel gain of each UE
119894on
RC and we will now proceeed to calculate the SNRThe SNR values are reported to ENodeB by theUE at each
TTI All the users have a different value of SNR accordingto the experienced fading and multipath propagation thetransmit power (119875) and the total RBs (119873) as follows
SNR =119875 lowast 119866
119894119895
119873(Noise + Interference) (31)
The incoming packets will be treated by the eNodeBaccording to the parameters defined in Table 2 and thesystem model presented All packets have varying lengthand are streamed in queues according to their priorities andproperties Now for the delay sensitive packets the thresholdDB119894is set to be 20ms for each UEThus we can define the Average System Delay (AvD) with
a total simulation time of 119879 at time instance 119905 with totalnumber of UE instances equal to119873 as follows
AvD = 1119879
119879
sum
119905=1
1
119873
119873
sum
119894=1
119863119894(119905) (32)
We will then define AvTH which is the average through-put with trans
119894(119905) being the number of packetssec as follows
AvTH = 1119879
119873
sum
119894=1
119879
sum
119905=1
trans119894(119905) (33)
Now for evaluating the PLR by using the size of discardedpackets pdisc
119894(119905) and the sum of sizes of all packets psize
119894(119905)
we have
PLR =sum119873
119894=1sum119879
119905=1pdisc119894(119905)
sum119873
119894=1sum119879
119905=1psize119894(119905)
(34)
Theoverall spectrumefficiency (119878eff ) ismeasured in termsof total traffic usage factor 119879
119891by applying the formula
119878eff =119862tot119879119891
(35)
where 119879119891= 119903 times 119901 where 119903 is the resource usage factor and
119901 is total traffic 119862tot is number of available traffic channels inthe system
The queuing model is set up by considering 119883119905that
represents the number ofUE instances at a given time interval119905 isin 119879 and 119880
119905denotes the number of new UE instances
that arrive at time 119905 isin 119879 with common probability densityfunction 119891 on 119879 and we define
119883119905+1=
119880119905+1
119883119905= 0
(119883119905minus 1) + 119880
119905+1119883119905gt 0
119905 isin 119879 (36)
where 119883 = 1198830 1198831 is a Markov Chain that has the
probability matrix of the following form
119875 (0 119910) = 119891 (119910) 119910 isin 119879
119875 (119910 119911) = 119891 (119910 minus 119911 + 1)
119911 isin 119879 119910 isin (119911 minus 1 119911 119911 + 1 )
(37)
The chain119883 defined here is the queuing chain of distribu-tion 119891 and starting from 119910 isin 119879 a UE instance is served andnew UE instances arrive by the next time unit 119905 controlled by119891 So the probability of going from state 119910 to 119911 isin (119910minus1 119910 119910+1 ) is lfloor(119911 minus 119910 + 1)rfloor
32 Results Wewill now discuss the simulation results basedon the simulation model and proposed strategy In thispaper we have used the LTE-SIM [35] simulator availableonline at httptelematicspolibaitLTE-Sim It is crucial tocarry out performance evaluation for LTE-A systems and thisrequires the use of standard simulators LTE-Sim is designedaccording to the LTE-A standard with main features suchas multicell environment multiuser support eNodeB CQIUE resource allocation QoS management and schedulingstrategies
The network topology under consideration comprisesnetwork nodes (including eNodeBs and UE) distributedamong various cells We have tested our algorithms for both
Mobile Information Systems 9
MMEGW
MMEGW
Single-cell
eNBeNB
eNB
eNB
UE
UE
Multicell
EGW
eNBNBNBBBNBNBBBBBBBeNB
eNB
UE
UE
Figure 5 Single-cell and multicell simulation scenarios
5 10 15 20Users
DPSMOZ
0
200000
400000
600000
800000
1000000
1200000
Thro
ughp
ut (b
ps)
Figure 6 Video throughput
the single-cell and multicell scenario as shown in Figure 5The first scenario consists of many eNodeBs due to the factthat multiple cells are covered while in the second scenarioone eNodeB handles all UE
We will evaluate the performance of our proposed MOZwhile comparing it with that of DPS [20] For each UEdownlink VOIP and video flows are considered All flows areactive during the whole simulation Time for each simulationis around 100 sThe simulation is performedusing aWindowsmachine with Cygwin (Linux-like environment) a 26GHzCPU and 4GB RAM
321 Delay and Throughput The goal of achieving desiredlevel of QoS is very challenging due to Real Timemultimediaapplications which have strict delay and throughput con-straints Figures 6 and 7 show the throughput for video andVOIP flows with increasing number of UE instances Fromboth the figures it can be seen that the throughput achievedby MOZ is better than that of DPS algorithm althoughthe difference for VOIP flow is much less Throughput alsoincreases with increasing number of users For video up to15 users increasing pattern is observed and after 25 usersthroughput increases but gradually For VOIP flows thethroughput is still increasing although it is delay sensitiveThereason behind this increased throughput is that the systemresources are utilized according to the requirement of users
5 10 15 20Users
MOZDPS
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 7 VOIP throughput
5 10 15 20Users
DPSMOZ
00015
000155
00016
000165
00017
000175
00018
Del
ay (m
sec)
Figure 8 VOIP delay
and only the selected UE instances transmit on the best RBsAlso it is worth noticing that the throughput achieved byMOZ is much closer to the theoretical maximum In thisprocess decisions are made based on the channel quality ofeach UE which constantly measures the QoS parametersbased on the algorithms and techniques proposed After10 users the throughput for video using MOZ achievesperformance gains in order of 1 to 15 while for VOIP it isfrom 1 to 10
Figures 8 and 9 show the average delay of the MOZand DPS algorithm with increasing UE for both the videoand VOIP flows From Figure 8 it can be seen that thedelay performance of MOZ is slightly higher compared toDPS with less than 10 users and as number of UE instancesincreases MOZ shows better performance compared to theDPS algorithm because after 10 plus users there is morevideo streaming and it is not distinguishing VOIP and videoflows so the delay for DPS will increase in consecutive TTIsFigure 9 shows the performance of bothMOZandDPSwhichis very close because with increasing users video demand hasa possibility of rising
Also our results ensure that the delay always remainsbelow the threshold value The proposed MOZ achieves
10 Mobile Information Systems
DPSMOZ
5 10 15 20Users
0
002
004
006
008
01
012
014
Del
ay (m
sec)
Figure 9 Video delay
DPSMOZ
5 10 15 20Users
00280029
0030031003200330034003500360037
PLR
Figure 10 Packet loss ratio
performance gains by reducing delay from 1 to 5 for videoand 1 to 8 for VOIP after 10 users Also it is worth noticinghere that the delay achieved is way below the threshold whichwas set to 20ms
322 PLR Now let us compare the PLR performance ofMOZ and DPS algorithm shown in Figure 10 As expectedthe PLR is increasing as the number of users increasesbecause there are more packets that will be discarded as therewill not be sufficient RBs for transmitting all the packetsconsidering that the packets are nearing their delay thresholdHowever in comparison with the DPS algorithm MOZ iscapable of achieving a better PLR as it supports more usersby distinguishing packets according to their scheduling andresource needs It reduces the loss probability by exploitingthe gains at physical layer Note that the PLR counts only forphysical layer losses MOZ achieves a better PLR and packetloss is reduced from 2 to 14 as compared to the referencealgorithm
323 SNR Figure 11 shows the SNR performance of MOZwith reference to average system throughputWith increasing
5 10 15 20SNR (db)
MOZ
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 11 SNR versus throughput
10 20 30 40SNR (db)
MOZ
021
022
023
024
025
026
027
028
Spec
tral
effici
ency
(bps
)
Figure 12 Cell spectral efficiency
throughput it is observed that the SNR is also increasingwhich depicts better channel quality and less distortion andis one of the reasons we are able to achieve higher systemthroughputThe higher values of SNR are indicator of higherpercentage of data as compared to noise and our resultsshow that noise is around 004 of the actual signal whichis an indicator that the video and voice quality achieved areexcellent
324 Spectral Efficiency Then the spectral efficiency (inbitss) is computed and is depicted in Figure 12 At eachscheduling interval the RB is allocated to UE instancesaccording to their needs As the number of UE instancesincreasesMOZ still guarantees QoS constraints tomore flowswith a positive effect on overall system efficiency The systemsupports delay and throughput guarantees simultaneouslytherefore the spectral efficiency increases with increasingusers which is a good indicator of overall system gain andefficiency
325 Energy and Memory Consumption Figure 13 demon-strates the energy consumption in joules with respect to thenumber of users in the network The energy usage of MOZis compared to PF scheduling and DPS algorithm As the
Mobile Information Systems 11
5 10 15 20Users
DPSMOZPF
0
05
1
15
2
25
Ener
gy (j
oule
s)
Figure 13 Energy consumption versus no of nodes
5 10 15 20Users
MOZ
0
50
100
150
200
250
Mem
ory
Figure 14 Memory consumption versus no of nodes
number of users increases the level of energy consumptionalso rises for all algorithms As the eNodeB remains active allthe time this rise is expected But the energy consumptionlevel achieved by MOZ is the lowest due to the fact thatefficient RB selection takes place which reduces the delaysthus lowering levels of energy consumption Also only theoptimal solution is selected which in turn results in energygains and network lifetime
In Figure 14 the memory consumption in bits versus thenumber of users is depictedThe DM takes decision based onthe information it has and the solution depends on algorithmsthat run dynamically and this process is continuous untilthere are no more RBs to allocate Moreover it has beenrealized that the continuous assignment of RBs accordingto UE instancesrsquo needs utilizes less memory which in turnincreases network life time and performance and reducesdelays
326 PRB Utilization In this subsection we will show effectof the proposed algorithm on PRB utilization The scenariodepicts results according to two groups In the first groupsimulations are carried out without using the proposedMOZthe second group implements MOZ The performance is
5 10 15 20 25 30Video users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 15 Mean PRBs used for video
5 10 15 20 25 30VOIP users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 16 Mean PRBs used for VOIP
measured in terms of mean PRBs being used for both videoandVOIP users In high load the delay increases and a highernumber of PRBs are used when we do not employ the MOZalgorithm As a result after a specific limit no more requestscan be fulfilled because all PRBs are utilized Figure 15 depictsthis scenario as we can see that the average number of PRBsused withMOZ is less due to improved bandwidth utilizationstrategy better coverage and enhanced SNR But with higherload of 30 UE instances all the bandwidth is used up
In Figure 16 for VOIP users also the average number ofPRBs used is much less with increasing number of users
33 Summary of the Simulation Analysis Table 3 representsthe results and statistics obtained by using MOZ where 1 to 4represent the PO set The optimal solution 119874lowast for each set ofusers is bold
It can be clearly seen that the DM always selects the bestfeasible solution thus proving an accuracy of 100 in termsof optimal solution selection The optimization proceduregenerates a PO set for varying number of users and out ofthis set the DM selects the best solution
12 Mobile Information Systems
Table 3 Statistics of the PO set and the optimal solution Olowast selected using MOZ
Number of users PO set (average delay in ms) PO set (average throughput in bps)1 2 3 4 1 2 3 4
5 000252 000328 000173 000175 40722208000 66418738666 50126800000 3949574800010 000179 000197 000252 000328 1509846667 2871940000 4357810667 585808666715 000161 000167 000169 000175 40821598667 76296266667 93016094667 9334639733320 000161 000167 000159 000155 40821541334 76643476000 97387246667 101596778667
The evaluation and comparison of the simulation resultsshow that MOZ performs very well for both voice andVOIP scenarios and performs better than the referencealgorithm in terms of maximum achievable throughput andminimum delay The reason behind these results is the useof optimization process employed by the DM which alwaysselects the optimal solution for the arriving UE instancesbased on their QoS needs and priorities and channel qualityIt further optimizes the procedure by maintaining queues toimprove performance and achieves higher SNR and lowerPLRThe cell spectral efficiency increaseswith each addedUEthat indicates improved system performance and is a proof ofrobust system design
4 Conclusions
In this paper we have proposed a multiobjective optimizerthat helps in finding an optimal solution for selecting RBsfor transmission based on the information provided to thedecision maker It provides an optimal solution for RRM toensure QoS by incorporating themultiobjective optimizationtechniques that target to achieve higher throughput and lowerdelay A set of Pareto optimal solutions is originated and isforwarded to the DM which chooses the best solution outof this set We have adopted an interactive technique thatembeds the multiobjective optimization with the moderntechniques for RRM Furthermore the results are also pro-vided that prove the efficiency of our proposed MOZ Thistechnique is innovative as it makes use of evolutionary MOOtechniques (which are very powerful for finding optimalsolution) and provides its implementation for LTE-A and willprove helpful in solving other issues for 4G and emerging 5Gnetworks
In future we aspire to investigate the opportunities forimplementingMOO techniques for the evolving 5Gnetworksthat not only aim to deal with QoS issues but also target toachieve optimal energy consumption
Competing Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] S-B Lee I Pefkianakis A Meyerson S Xu and S LuldquoProportional fair frequency-domain packet scheduling for3GPP LTE uplinkrdquo in Proceedings of the 28th Conference on
Computer Communications (IEEE INFOCOM rsquo09) pp 2611ndash2615 Rio de Janeiro Brazil April 2009
[2] ldquoPhysical layer aspects for evolved UTRArdquo 3GPP TechnicalReport 25814 version 710 2006
[3] M Nazir and F Saleemi ldquoCooperative cognitive ecology in self-organizing networks a review articlerdquo International Journal ofComputer Science and Information Security vol 10 no 2 pp159ndash167 2012
[4] K Deb Multi-Objective Optimization with Evolutionary Algo-rithms John Wiley amp Sons New York NY USA 1st edition2001
[5] J B K Deb and K M R SłowinskiMulti-Objective Optimiza-tion Interactive and Evolutionary Approaches Springer BerlinGermany 2008
[6] A De Domenico and E C Strinati ldquoA radio resource man-agement scheduling algorithm for self-organizing femtocellsrdquoin Proceedings of the IEEE 21st International Symposium onPersonal Indoor and Mobile Radio Communications Workshops(PIMRC rsquo10) pp 191ndash196 IEEE Instanbul Turkey September2010
[7] R Kausar Y Chen K K Chai L Cuthbert and J SchormansldquoQoS aware mixed traffic packet scheduling in OFDMA-basedLTE-advanced networksrdquo in Proceedings of the 4th InternationalConference on Mobile Ubiquitous Computing Systems Servicesand Technologies (UBICOMM rsquo10) pp 53ndash58 Florence FranceOctober 2010
[8] Y J Zhang and K B Letaief ldquoAdaptive resource allocation andscheduling for multiuser packet-based OFDM networksrdquo inProceedings of the IEEE International Conference on Communi-cation vol 5 pp 2949ndash2953 June 2004
[9] C Stocchi N Marchetti and N R Prasad ldquoSelf-optimizedradio resource management techniques for LTE-a local areadeploymentsrdquo inProceedings of the 2nd International Conferenceon Wireless Communication Vehicular Technology InformationTheory and Aerospace amp Electronic Systems Technology (WirelessVITAE rsquo11) pp 1ndash5 Chennai India March 2011
[10] M Cenk Erturk H Aki I Guvenc and H Arslan ldquoFairand QoS-oriented spectrum splitting in macrocell-femtocellnetworksrdquo in Proceedings of the 53rd IEEE Global Communica-tions Conference (GLOBECOM rsquo10) pp 1ndash6 Miami Fla USADecember 2010
[11] V Chandrasekhar and J G Andrews ldquoSpectrum allocation intiered cellular networksrdquo IEEE Transactions on Communica-tions vol 57 no 10 pp 3059ndash3068 2009
[12] A Roy and S K Das Optimizing QoS-Based Multicast Routingin Wireless Networks A Multi-Objective Genetic AlgorithmicApproach Springer New Yok NY USA 2002
[13] D H Lorenz and A Orda ldquoQoS routing in networks withuncertain parametersrdquo IEEEACM Transactions on Networkingvol 6 no 6 pp 768ndash778 1998
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
2 Mobile Information Systems
File download
Streaming video
Web browsing Email
Throughput sensitivity
Interactive video
Interactive voice applications
GamingStreaming
video
Delay sensitivity
Figure 1 Varying QoS requirement classification
middot middot middot N minus 2 N minus 1 N
N
1
1 2
PRBs
RCs
2 3 4 5 6
Figure 2 A total of M RCs dividing the total bandwidth into 119873PRBs
(UE) for various applications Figure 2 depicts this conceptTo guarantee QoS we need to select RB that will fulfill bothour objectives of throughput and delay guarantees
RRM is a set of techniques that are utilized to optimizethe usage of spectrum and equipmentThemotivation behindRRM functionality is to guarantee the provision of networkservices and QoS and optimize the system usage The real-ization of a wireless system is dependent on the bandwidthallocation and the QoS gains of the transmissionmedium [3]Now that data rate provisioning is heading towards the nextgeneration designing techniques for efficient radio resourceallocation is becoming more challenging These challengesresult due to the fact that the packets arrive randomly andthere is a lack of buffer space plus the QoS needs are erraticand thus resource management and allocation techniquesneed to be developed
Ourwork takes advantage of themultiobjective optimiza-tion (MOO) techniques that make use of a decision maker(DM) to take radio resource allocation decisions The solepurpose is to provide RRM technique that will ensure QoSprovisioning to all types of traffic specifically for increasedthroughput and reduced delay There exist evolutionaryalgorithms that have various schemes and techniques formanaging resources [4] In MOO the objective vectors areviewed as optimal if none of its elements can be furtherimproved without deteriorating at least one of the otherelements [5]
Most real world problems have more than one objectivefunction and when we have multiple objectives to attainwe need efficient solutions to solve the problem Also therelationship between objectives is generally rather perplexingand cannot be measured using the same standard thusmaking it hard to cumulate them into a single objectiveIn that case we do not have a solitary optimal solutionbut rather a collection of substitutes with varying trade-offswhich are called Pareto optimal (PO) sets Usually one ofthese PO solutions is selected at the end So in multiobjective
optimization we have two tasks first finding a PO set andsecond a decision-making processmodule that will select adistinct solution from this set also referred to as DM Thuswhen we have diverse goals with conflicting objectives theycannot be summed up into one objective function and thisencourages the use of various MOO techniques
Despite the fact that LTE deployment scenarios deliverhigh data rates they also showcase new challenges for inter-ference management and RRM To meet these challengespolicies are intended for the standard cellular networks Alsothe femtocells are ad hoc in nature and this fact tends tobind the scope of the algorithms under consideration Thusproficient RRM procedures are critical to restrain the impactof interference on LTE femtocells performance [6]
A framework for Time Domain (TD) scheduling withcongestion control and QoS provisioning is also introducedIt optimizes the QoS provisioning process for domains ofboth time and frequency by taking into account channel con-dition the status of the queues and the userrsquos QoS require-ments This framework enhances the process of resourcesutilization by considering the QoS requirements of variousservice classes The results indicate better QoS of Real Time(RT) traffic and fair resource sharing of available resources forthe Non-Real Time (NRT) traffic [7]
The idea of cross layer resource allocation for OrthogonalFrequency Domain Multiplexing (OFDM) scheme is consid-ered significant as it takes into account the channel conditionthe arbitrary nature of traffic QoS needs and fair distributionof resources between the users Time slots frequenciesand carriers are assigned dynamically to make this processeffective [8] A number of studies exist in literature that focuson the RRM of LTE-A considering the QoS requirements ofthe user Some make use of self-optimization some tend tofollow the resource scheduling approaches others focus onspectrum splitting using the time and frequency domainsand some have designed solution for RRM by using geneticalgorithms [9ndash15]
The algorithms that manipulate multiuser environmentare the MAX CI and Proportional Fairness (PF) algorithmsIn MAX CI algorithm RB is allocated to a user who tendsto achieve the highest data rates for the current time slotand tries to maximize overall system throughput The PFalgorithm also considers fairness and resources are allocatedon the basis of priority and allocation schemes [16]
However these algorithms target to improve utilizationof resources considering the channel conditions and interfer-ence mitigation only and the delay and throughput require-ments are not taken into much consideration A number oftechniques have also been proposed that aim to efficientlyuse the radio resources to provide guaranteed QoS for var-ious types of traffic [17ndash19] A Delay Prioritized Scheduling(DPS) algorithm [20] has been developed that manages thedistribution of RBs by choosing those RBs that best fulfill thedelay threshold condition With increasing users the delayalso tends to increase due to more video requests that arethroughput sensitive It is a packet scheduling algorithm thatselects RBs according to their SNR levels and then admitsand assigns the flow to the best possible RBs This is a two-way process which first selects a user according to a delay
Mobile Information Systems 3
threshold and then assigns RBs It focuses mainly on delayminimization
In the perspective of multiobjective optimization mostwork done is towards discovering a collection of near-Pareto optimal solutions algorithmically In multiobjectiveoptimization communication with the DM can be doneduring the phase of optimization or during the final decision-making part In many studies a human decision maker isinvolved after a variety of solutions have been sought out [21ndash23]
The process of multiobjective optimization is incompletewithout a decision-making activity that will take the finaldecision based on various alternatives available In thiscontext many interactive multiobjective optimization tech-niques are available under the title Multicriteria Decision-Making (MCDM) [24ndash26] All the techniques are differentfrom each other but all incorporate DM to provide infor-mation to help in taking the final decision The evolutionaryalgorithms (EA) suggest the use of natural evolution theoryfor optimization like the survival of the fittest theory ofDarwin They work with a set of solutions and can also beused to find a partial PO set
MOO has been researched for many years and is focusedon the theoretical aspects [27] A lot of approaches havebeen formulatedwithmathematical programming theory forexample nonlinear programming to solve the multiobjectiveoptimization problems [28] Varying interactive approacheshave also been used in which the information is given tothe DM and the DM specifies its preferences This proceduredepends on the type of problem and its mathematical prop-erties and scalar function [29]
A power-delay minimization scheme is also introducedthat makes use of linear programming [30] In this studymultiobjective problem is transformed into a solitary objec-tive by using weighted sum technique that aims to reduceboth the delay and transmit power Simulated annealing andgreedy heuristic algorithms are also part of this work Itstarget is not specifically LTE networks but mainly the IEEE80211 Wireless LANs (WLANs) and Green Wireless AccessNetworks (GWANs)
Table 1 depicts the classification of various techniquesused for RRM and for finding an optimal solution withmultiple objectives Based on the literature survey we cansee that there exist various techniques for QoS provisionin LTE-A using different methodologies Also the multiob-jective optimization offers a variety of solutions for solvingvarious linear and nonlinear problems Still there is a lackof these multiobjective optimization technique applicationsfor RRM of LTE-A targeted towards ensuring QoS As aresult we will propose a QoS awareMultiobjective Optimizer(MOZ) for the RRM of LTE-A which will take into accountthe provision of maximum throughput and minimumdelay
2 Materials and Methods
21 Problem Formulation To measure the performance ofLTE-A network we have a set of criteria such as through-put end-to-end transmission delay energy efficiency and
Table 1 Classification of literature on RRM and optimization
Approach [7] [16] [20] [21ndash23] [30] Proposed approachScheduling radic radic radic radic radic radic
User association radic radic radic mdash mdash radic
QoS aware radic radic radic mdash radic radic
Energy efficiency mdash radic mdash mdash radic mdashRB allocation mdash mdash radic mdash radic radic
LTE radic mdash mdash mdash radic mdashLTE-A mdash radic radic mdash mdash radic
MOO mdash mdash mdash radic radic radic
DM mdash mdash mdash radic mdash radic
transmission strength The purpose of the work presentedhere is to determine the trade-offs that arise while choosingperformance metrics for assigning resources to the UserEquipment (UE) The multiobjective Pareto optimizationconsists of three steps
(1) Multiobjective problem definition(2) Optimization (finding Pareto optimal solutions)(3) Decision-making (role of DM)
In the following we will elaborate the working of our QoSaware optimizer according to the above-mentioned threesteps and it will define the framework for our proposedoptimizer
Step 1 (problem definition) To solve our multiobjectiveoptimization problem we will involve DM to find the best(optimal) solution By optimal solution we are referring toPareto optimal solution that the DM considers as the bestoption In our case we have two objectives the first one ismaximum throughput and the second one is minimumdelayWe will manage the resources in such a way that we are ableto select RB that transmits the data according to the above-mentioned objectives
ParetoOptimization ProblemWewill solve themultiobjectiveoptimization problem that will take the form
MaximizeMinimize 1198911(119910) 119891
2(119910) 119891
119894(119910)
Subject to 119910 isin 119878
(1)
where 119894 is a set of objectives that will be minimized ormaximized according to its definition 119878 is a set of networkconstraints Here 119910 is the decision vector and 1199101015840 is thedecision variable We will call 1199101015840 isin 119878 Pareto optimal only ifthere is no other 1199101015840 isin 119878 such that
119891119896(119910) le 119891
119896(1199101015840) (where 119896 = 1 119894) (2)
This means that 1199101015840 is Pareto optimal only if there is noother possible vector (or solution)119910 thatwill deteriorate somecriterion without leading towards an increase in some othercriterion
4 Mobile Information Systems
Step 2 (Pareto front) In our case we have the followingobjective functions and the first one aims at maximizingnetwork utility through increased throughput (119879
119894)
Maximize119899
sum
119894=1
(119879119894)
Subject to119899
sum
119894=1
119879119894lt 119862119894119896(119905) where 119879
119894gt 0 forall119894 isin 119873
(119873 = total UE instances)
(3)
Considering channel 119896 and user 119894 we define the channelcapacity as
119862119894119896= RB log
2(1 +
120588
1205902sdot 119875) (4)
where RB is RB Bandwidth which is calculated as (totalbandwidthnumber of RCs) 119875 is Transmission Power 120588 isSNR space determined by Bit Error Rate (BER) and 1205902 isnoise power density
119879119894is the average throughput calculated as
119879119894= TUE
119894 [119905]1
120591+ 119879119894(119905 minus 1) (1 minus
1
120591) forall119894 isin 119873 (5)
where TUE119894[119905] is throughput of UE
119894in time instance 119905 120591 is
time constraint of the smoothing filter and119873 is total UETUE119894[119905] is calculated with the help of the following
equation
TUE119894 [119905] = RB log
21 + SNR (6)
where SNR = 119879119901times 119878UE(Noise + Interference) (119879
119901is
Transmission Power 119878UE is signal gain of UE)Our next objective is defined as follows
Minimize119873
sum
119894=1
(119863119894(119905))
Subject to 119863119894(119905) lt DB
119894
(7)
where DB119894is the delay budget or the upper bound which is
equivalent to 20ms in OFDMA networks [19 31] The delayexperienced by the UE
119894should be less than this upper bound
Here 119873 represents total active UE and 119863119894(119905) is the HOL
(Head of Line) delay for UE119894at time 119905 calculated as
119863119894(119905) =
119882119894
DT119894
(8)
where119882119894is the waiting time for UE
119894and DT
119894is the normal-
ized HOL delay obtained by dividing each userrsquos waiting timeby DB
119894
This process will generate a set of Pareto optimal solutionscalled the Pareto front As a result there is no single POsolution so the question is which solution to select from a setof PO solutionsThe answer lies in Step 3 that takes advantageof DM
Step 3 (the optimal solution) In this step the DM will selectthe best solution which is RB that will be allocated fortransmission Depending on the type of incoming traffic theDM will decide which strategy to follow according to thefollowing criteria
(i) If the traffic is delay sensitive preference will be givento minimizing the delay whereas throughput can becompromised to some extent
(ii) If the incoming packet is throughput sensitive it willemphasize maximum throughout whereas delay maybe compromised
This scheme will thus allow the delay sensitive applica-tions to sacrifice throughput for lower delays without anyeffect on the throughput sensitive applications
So we have 119896 and 119894 objective functions (represented by 119892)and a total of 119898 and 119891 constraints (represented by ℎ) Weassume that the constraints plus the objective function arethe functions of the decision vectors So the DMrsquos goal is asfollows
Maximize 119892 (119909) = 1198921(119910) 119892
2(119910) 119892
119894(119910)
Minimize 119892 (119910) = 1198921(119911) 119892
2(119911) 119892
119896(119911)
Subject to ℎ (119909) = ℎ1(119910) ℎ
2(119910) ℎ
119898(119910)
ℎ (119910) = ℎ1(119911) ℎ
2(119911) ℎ
119891(119911)
119909 = 1199091 1199092 119909
119894
119911 = 1199111 1199112 119911
119896
(9)
where 119909 is the decision vector for first objective function and119911 is the decision vector for second objective function 119874lowast isin119909 or 119911 is the feasible decision or solution to our optimizationproblem 119874lowast is Pareto optimal as we have no other bettersolution So here the DM will make use of a weighted MinndashMax Approach which is adopted from GameTheory [32 33]In this scheme we will compare the relative deviation froma separately attainable minima or maxima In our case it isthe deviation from maximum throughput and delay of thesystem We will denote this by 119863max119894 (delay variation) and119879min119894 (throughput variation) We will calculate the relativedeviations using
119879min119894 = TA119904minus 119879119894
(For throughput sensitive applications) (10)
TA119904is the maximum attainable throughput of the LTE
system calculated as follows(i) We will first calculate the total RBs assuming the
bandwidth of channel is 20MHz by using the following
carriers timesOFDM symbols times Slots times RBs (11)
where carriers are 12 OFDM symbols are 7 RBs are 100 Slotsare 2 The final value is 16800 RBs per frame
Mobile Information Systems 5
(ii) Second we consider modulation of 64 QAM and asingle modulation symbol carries 6 bits The total bits willthus be
16800
times 6 bits per symbol of modulation equal to 1008Mb(12)
(iii) Thirdly considering the 4-by-4 MIMO we get
4 times 1008Mb = 403Mb (13)
This is the peak data rate(iv) Finally we calculate the overhead which will be about
25 So we have
403Mb times 075 = 302Mb (14)
So we can calculate TA119904and RC and minimum 119879min119894 is
selected by the DM for transmission That is resource chunkRC119894as in the following
119874lowast= RC119894= min119879min119894 forall119894 isin UE (15)
Now let us talk about the delay sensitive applicationswhere we will find the feasible solution 119874lowast as follows
119863max119894 = DB119894minus 119863119894(119905)
(For delay sensitive applications)
119874lowast= RC119894= max119863max119894 forall119894 isin UE
(16)
So 119874lowast is chosen for transmission based on the followingdecision vectors
119874lowast=
argmin 119879min119894 forallUE119894
argmax 119863max119894 forallUE119894
(17)
The DM will take its decision based on the following setof constraints
119899
sum
119894=1
119879119894lt 119862119894119896(119905) where 119879
119894gt 0 forall119894 isin 119873 (18)
119863119894(119905) lt DB
119894forall119894 isin 119873 (19)
119873
sum
119894=1
RB119894le Total Badndwidth (20)
sum
119895isin119869
120579119894119895ge 1 forall119894 isin 119873 forall119869 isin Total BS (21)
Equation (18) refers to channel capacity constraint and(19) refers to delay bound constraint Equation (20) is RBbandwidth constraint Equation (21) ensures that at least 1 BScovers the active UE
22 The Decision-Making Process After the optimized solu-tion is sought out by theDM the trafficwill then be scheduledin output queues discussed in detail in the next subsection
Figure 3 depicts the detail of decision-making process insidethe DM Its details are explained as follows
(1) Traffic Type Determination The DM will take as input thePareto front and will first determine the type of traffic whichin our case is either throughput or delay sensitive
(2) Optimization Process After determining the traffic typethe following actions are taken
(i) For throughput sensitive applications it will calculatethe value of 119879min119894 based on the value of 119879
119894 and then
it will select the optimal solution 119874lowast = RC119894=
min119879min119894(ii) For delay sensitive applications it will calculate the
value of 119863max119894 based on the value of 119863119894(119905) and then
it will select the optimal solution 119874lowast = RC119894=
max119863max119894(3) RC Allocation Matrix Update After that the RC assign-ment matrix will be updated The matrix 119898
119896119899is represented
as
119898119896119899=
0 if RC119899is unassigned
1 if RC119899is assigned to service group 119896
(22)
where 119896 is type of traffic and 119899 is RC number [34]
(4) RC Availability Check If there are RCs availableleft forassignment then the process starts again andmoves to traffictype determination otherwise the process ends
In the following we explain in detail the construction ofPareto front
23 Constructing Pareto Front This step describes in detailthe process of constructing a representative Pareto front asfollows
Step 1 Determine the traffic type of the incoming packetAssume 119863
1 119863
119899are the alternative solutions in terms of
119894 and 119895 where 119894 represents throughput sensitive flows and 119895represents delay sensitive flows
Step 2 Find the largest 119894 such that 119863119894gt 119863119899(where 119863
119899
represents all other UE) and add119863119894to Pareto front
Step 3 Find the smallest 119895 such that 119863119895lt 119863119899(where 119863
119899
represents all other UE) and add119863119895to Pareto front
Step 4 Repeat Steps 2 and 3 until no such 119894 and 119895 exist and wehave the Pareto front of the form (119863
119894 119863
119899119863119895 119863
119899)
Note that 119863119894is sorted in order of decreasing throughput
and 119863119895is sorted in order of increasing delay The process
generates a set of candidate solutions that will undergo theoptimization phase for selecting a singled optimal solutionThe selection of the optimal solution is based on the efficiencyof the DM which not only considers the resource require-ments of the users but also sets priorities We now present analgorithm that will make use of the defined framework andconsiders the requirements of each step
6 Mobile Information Systems
Start
Throughput sensitive traffic Delay sensitive trafficDetermine from thePareto front the traffic
requirements
On the basis ofvarious values ofTi calculate Tmini
various values ofOn the basis of
Di(t) calculate Dmaxi
Choose optimalsolution based on
Olowast= RCi = minTmini
Choose optimalsolution based on
Olowast= RCi = maxDmaxi
Update RCallocation matrix
RCs left to be allocated Yes
No
End
Figure 3The decision-making process inside the DM for RC selection based on the objectives of maximum throughput andminimumdelay
24 The Multiobjective Optimizer Algorithm The detailedworking of proposed MOZ is outlined in Figure 4 TheMOZconsists of three main steps as follows
Step 1 The traffic classifier will take as input traffic fromvarious UE instances and will divide them into 2 queues oneconsists of packets that have stringent throughput require-ment and the other queue consists of packets from delaysensitive applications
Step 2 The information from Step 1 is then passed onto theoptimizer that will find out the Pareto front and will send itsdecision set to the decision maker
Step 3 The DM will make the final decision based on theinformation it has and according to its decision-makingcapabilities
Step 4 Finally119874lowast is sent out to the queue for scheduling andthe process then diverts to Step 2 to get the next 119874lowast later tobe scheduled
Step 5 The process is completed when either no RC is left orthere are no more incoming packets
Steps 2 and 3 are interrelated as the DM is way dependenton the input it gets from the Pareto front so constructing alegitimate Pareto front is itself a very important task forMOZThe decision-making process inside the DM is characterizedby a number of factors already explained and is one of themajor components of our proposed strategy
Compared to most algorithms and frameworks designedto ensure QoS and to manage RRM the distinct differencesof our proposed approach are as follows (1) it makes useof multiobjective optimization that deals with conflictingobjectives simultaneously instead of converting them intoa single objective (2) It makes use of a decision makerthat has the decision-making capability and works basedon the information it has and various algorithms used(3) Multiple UE instances are served based on the type ofincoming trafficflow instead of allocating RBs to just oneflow
Mobile Information Systems 7
Trafficclassification Optimization Decision-making Scheduling
Input trafficfrom variousUE instances
applicationsDelay sensitive
Throughput sensitiveapplications
Theoptimizationprocess todeterminethe Paretofront
DetermineO
lowast (theoptimalsolution)
Figure 4 The proposed MOZ
The goal of using MOZ is to always get an optimalsolution which is chosen from a set of solutions in the formof Pareto front
25 Queuing and Scheduling Now that we have the Paretooptimal solution in hand the packets are now ready tobe scheduled in their respective queues for transmission asfollows
Step 1 In this step we will define the priority metric for delaysensitive flows The priority of UE
119894at time 119905 119875
119889119894(119905) is
119875119889119894(119905) = 119863max119894 lowast 119866119894119895 (23)
where 119866119894119895is channel gain of UE
119894on RC
119895calculated as
119866119894119895= 10
Path loss10lowast 10
shadow fading10
lowast 10multipath fading10
(24)
Step 2 We will then form a delay queue for a given UE119894that
will ensure delay bounds as follows
119863119894(119905 + 1) = [119863
119894(119905) minus 120588119877
119894(119905)] +
119873
sum
119894=1
119876119894(119905) (25)
where 119876119894(119905) is the mean queue length at time interval 119905 and
119873 is total UE scheduled and 120588119877119894(119905) is the service rate In
each time interval the UEwith highest priority occupies frontposition in the queue
Step 3 For the priority metric of throughput sensitive flowswe will use the following
119875119905119894(119905) = 120590
119894lowast 119879min (119894) (26)
where 120590119894is the Head of Line (HOL) blocking and 119875
119905119894(119905) is the
priority metric for throughput sensitive applications In eachtime interval the UE instances are scheduled according to theascending order of their priority values
It follows the following constraints
119876119894119896
st forall119894 isin 119873 119896 isin delay sensitive traffic only (27)
119876119894119895
st forall119894 isin 119873 and 119895
isin throughput sensitive traffic only(28)
119894 = 119895 Queue Length
= Length (119876119894119896+ 119876119894119895)
(29)
Equation (27) ensures that the queue 119876119894takes as input
delay sensitive traffic only (28) ensures that the queue 119876119894119895
takes in throughput sensitive traffic only and (29) states thatqueue length is the total of both the queues
Thus the defined framework and the optimizer will worktogether to form our proposed model The whole process hasfour main goals
(1) Determining the target objectives(2) Based on the objectives forming a system model that
will take into account the objective functions to beminimized or maximized
(3) Constructing a representative solution pool to aid thedecision maker
(4) Introducing a decision-making entity that will notonly ease the task of scheduling but also find anoptimal solution to satisfy the defined objectives
In the following section we will define the simulationmodel and will discuss the results
3 Simulation and Results
31 Simulation Model Designing the model for LTE-A per-formance evaluation is generally more complex comparedto a 3G system An interface is used between variousevolvedNodeBs (eNBs ie the BS) in the LTE system toease the traffic load that occurs due to handovers In this
8 Mobile Information Systems
Table 2 Simulation parameters
Parameters ValuesCarrier frequency 2GHzBandwidth 10MHzNumber of subcarriers 300Number of RBs 45TTI 1msPath loss model Free space propagation modelTransmit power 30mWFading model RayleighPacket model Poisson modelMultiplexing MIMONumber of UE instances 20
section we will design the simulation model for testing andwill evaluate our system for examining delay throughputPacket Loss Ratio (PLR) the effect of SNR and cell spectralefficiency
We are considering a cell with the capacity of 10MHz andit consists of 45 RBs and the carrier frequency is 2GHz Allthe available RBs are controlled by the eNodeB The RBs aredivided among theUE instances in the formof RCsThe LTE-A parameters used for simulation are described in Table 2
We then calculate the number of bits in each RB (ie itsdata rate) It can be computed as follows
119887119894RC = NStti lowastNS119904 lowast 119878rb (30)
where 119887119894RC is achievable data rate NStti is number of slots in
each TTI NS119904is number of symbols in each slot and 119878rb is
number of subcarriers in each RBEach UE is assigned a queue according to the traffic type
and parameters as described in the system model previouslyAfter that we will calculate the channel gain of each UE
119894on
RC and we will now proceeed to calculate the SNRThe SNR values are reported to ENodeB by theUE at each
TTI All the users have a different value of SNR accordingto the experienced fading and multipath propagation thetransmit power (119875) and the total RBs (119873) as follows
SNR =119875 lowast 119866
119894119895
119873(Noise + Interference) (31)
The incoming packets will be treated by the eNodeBaccording to the parameters defined in Table 2 and thesystem model presented All packets have varying lengthand are streamed in queues according to their priorities andproperties Now for the delay sensitive packets the thresholdDB119894is set to be 20ms for each UEThus we can define the Average System Delay (AvD) with
a total simulation time of 119879 at time instance 119905 with totalnumber of UE instances equal to119873 as follows
AvD = 1119879
119879
sum
119905=1
1
119873
119873
sum
119894=1
119863119894(119905) (32)
We will then define AvTH which is the average through-put with trans
119894(119905) being the number of packetssec as follows
AvTH = 1119879
119873
sum
119894=1
119879
sum
119905=1
trans119894(119905) (33)
Now for evaluating the PLR by using the size of discardedpackets pdisc
119894(119905) and the sum of sizes of all packets psize
119894(119905)
we have
PLR =sum119873
119894=1sum119879
119905=1pdisc119894(119905)
sum119873
119894=1sum119879
119905=1psize119894(119905)
(34)
Theoverall spectrumefficiency (119878eff ) ismeasured in termsof total traffic usage factor 119879
119891by applying the formula
119878eff =119862tot119879119891
(35)
where 119879119891= 119903 times 119901 where 119903 is the resource usage factor and
119901 is total traffic 119862tot is number of available traffic channels inthe system
The queuing model is set up by considering 119883119905that
represents the number ofUE instances at a given time interval119905 isin 119879 and 119880
119905denotes the number of new UE instances
that arrive at time 119905 isin 119879 with common probability densityfunction 119891 on 119879 and we define
119883119905+1=
119880119905+1
119883119905= 0
(119883119905minus 1) + 119880
119905+1119883119905gt 0
119905 isin 119879 (36)
where 119883 = 1198830 1198831 is a Markov Chain that has the
probability matrix of the following form
119875 (0 119910) = 119891 (119910) 119910 isin 119879
119875 (119910 119911) = 119891 (119910 minus 119911 + 1)
119911 isin 119879 119910 isin (119911 minus 1 119911 119911 + 1 )
(37)
The chain119883 defined here is the queuing chain of distribu-tion 119891 and starting from 119910 isin 119879 a UE instance is served andnew UE instances arrive by the next time unit 119905 controlled by119891 So the probability of going from state 119910 to 119911 isin (119910minus1 119910 119910+1 ) is lfloor(119911 minus 119910 + 1)rfloor
32 Results Wewill now discuss the simulation results basedon the simulation model and proposed strategy In thispaper we have used the LTE-SIM [35] simulator availableonline at httptelematicspolibaitLTE-Sim It is crucial tocarry out performance evaluation for LTE-A systems and thisrequires the use of standard simulators LTE-Sim is designedaccording to the LTE-A standard with main features suchas multicell environment multiuser support eNodeB CQIUE resource allocation QoS management and schedulingstrategies
The network topology under consideration comprisesnetwork nodes (including eNodeBs and UE) distributedamong various cells We have tested our algorithms for both
Mobile Information Systems 9
MMEGW
MMEGW
Single-cell
eNBeNB
eNB
eNB
UE
UE
Multicell
EGW
eNBNBNBBBNBNBBBBBBBeNB
eNB
UE
UE
Figure 5 Single-cell and multicell simulation scenarios
5 10 15 20Users
DPSMOZ
0
200000
400000
600000
800000
1000000
1200000
Thro
ughp
ut (b
ps)
Figure 6 Video throughput
the single-cell and multicell scenario as shown in Figure 5The first scenario consists of many eNodeBs due to the factthat multiple cells are covered while in the second scenarioone eNodeB handles all UE
We will evaluate the performance of our proposed MOZwhile comparing it with that of DPS [20] For each UEdownlink VOIP and video flows are considered All flows areactive during the whole simulation Time for each simulationis around 100 sThe simulation is performedusing aWindowsmachine with Cygwin (Linux-like environment) a 26GHzCPU and 4GB RAM
321 Delay and Throughput The goal of achieving desiredlevel of QoS is very challenging due to Real Timemultimediaapplications which have strict delay and throughput con-straints Figures 6 and 7 show the throughput for video andVOIP flows with increasing number of UE instances Fromboth the figures it can be seen that the throughput achievedby MOZ is better than that of DPS algorithm althoughthe difference for VOIP flow is much less Throughput alsoincreases with increasing number of users For video up to15 users increasing pattern is observed and after 25 usersthroughput increases but gradually For VOIP flows thethroughput is still increasing although it is delay sensitiveThereason behind this increased throughput is that the systemresources are utilized according to the requirement of users
5 10 15 20Users
MOZDPS
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 7 VOIP throughput
5 10 15 20Users
DPSMOZ
00015
000155
00016
000165
00017
000175
00018
Del
ay (m
sec)
Figure 8 VOIP delay
and only the selected UE instances transmit on the best RBsAlso it is worth noticing that the throughput achieved byMOZ is much closer to the theoretical maximum In thisprocess decisions are made based on the channel quality ofeach UE which constantly measures the QoS parametersbased on the algorithms and techniques proposed After10 users the throughput for video using MOZ achievesperformance gains in order of 1 to 15 while for VOIP it isfrom 1 to 10
Figures 8 and 9 show the average delay of the MOZand DPS algorithm with increasing UE for both the videoand VOIP flows From Figure 8 it can be seen that thedelay performance of MOZ is slightly higher compared toDPS with less than 10 users and as number of UE instancesincreases MOZ shows better performance compared to theDPS algorithm because after 10 plus users there is morevideo streaming and it is not distinguishing VOIP and videoflows so the delay for DPS will increase in consecutive TTIsFigure 9 shows the performance of bothMOZandDPSwhichis very close because with increasing users video demand hasa possibility of rising
Also our results ensure that the delay always remainsbelow the threshold value The proposed MOZ achieves
10 Mobile Information Systems
DPSMOZ
5 10 15 20Users
0
002
004
006
008
01
012
014
Del
ay (m
sec)
Figure 9 Video delay
DPSMOZ
5 10 15 20Users
00280029
0030031003200330034003500360037
PLR
Figure 10 Packet loss ratio
performance gains by reducing delay from 1 to 5 for videoand 1 to 8 for VOIP after 10 users Also it is worth noticinghere that the delay achieved is way below the threshold whichwas set to 20ms
322 PLR Now let us compare the PLR performance ofMOZ and DPS algorithm shown in Figure 10 As expectedthe PLR is increasing as the number of users increasesbecause there are more packets that will be discarded as therewill not be sufficient RBs for transmitting all the packetsconsidering that the packets are nearing their delay thresholdHowever in comparison with the DPS algorithm MOZ iscapable of achieving a better PLR as it supports more usersby distinguishing packets according to their scheduling andresource needs It reduces the loss probability by exploitingthe gains at physical layer Note that the PLR counts only forphysical layer losses MOZ achieves a better PLR and packetloss is reduced from 2 to 14 as compared to the referencealgorithm
323 SNR Figure 11 shows the SNR performance of MOZwith reference to average system throughputWith increasing
5 10 15 20SNR (db)
MOZ
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 11 SNR versus throughput
10 20 30 40SNR (db)
MOZ
021
022
023
024
025
026
027
028
Spec
tral
effici
ency
(bps
)
Figure 12 Cell spectral efficiency
throughput it is observed that the SNR is also increasingwhich depicts better channel quality and less distortion andis one of the reasons we are able to achieve higher systemthroughputThe higher values of SNR are indicator of higherpercentage of data as compared to noise and our resultsshow that noise is around 004 of the actual signal whichis an indicator that the video and voice quality achieved areexcellent
324 Spectral Efficiency Then the spectral efficiency (inbitss) is computed and is depicted in Figure 12 At eachscheduling interval the RB is allocated to UE instancesaccording to their needs As the number of UE instancesincreasesMOZ still guarantees QoS constraints tomore flowswith a positive effect on overall system efficiency The systemsupports delay and throughput guarantees simultaneouslytherefore the spectral efficiency increases with increasingusers which is a good indicator of overall system gain andefficiency
325 Energy and Memory Consumption Figure 13 demon-strates the energy consumption in joules with respect to thenumber of users in the network The energy usage of MOZis compared to PF scheduling and DPS algorithm As the
Mobile Information Systems 11
5 10 15 20Users
DPSMOZPF
0
05
1
15
2
25
Ener
gy (j
oule
s)
Figure 13 Energy consumption versus no of nodes
5 10 15 20Users
MOZ
0
50
100
150
200
250
Mem
ory
Figure 14 Memory consumption versus no of nodes
number of users increases the level of energy consumptionalso rises for all algorithms As the eNodeB remains active allthe time this rise is expected But the energy consumptionlevel achieved by MOZ is the lowest due to the fact thatefficient RB selection takes place which reduces the delaysthus lowering levels of energy consumption Also only theoptimal solution is selected which in turn results in energygains and network lifetime
In Figure 14 the memory consumption in bits versus thenumber of users is depictedThe DM takes decision based onthe information it has and the solution depends on algorithmsthat run dynamically and this process is continuous untilthere are no more RBs to allocate Moreover it has beenrealized that the continuous assignment of RBs accordingto UE instancesrsquo needs utilizes less memory which in turnincreases network life time and performance and reducesdelays
326 PRB Utilization In this subsection we will show effectof the proposed algorithm on PRB utilization The scenariodepicts results according to two groups In the first groupsimulations are carried out without using the proposedMOZthe second group implements MOZ The performance is
5 10 15 20 25 30Video users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 15 Mean PRBs used for video
5 10 15 20 25 30VOIP users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 16 Mean PRBs used for VOIP
measured in terms of mean PRBs being used for both videoandVOIP users In high load the delay increases and a highernumber of PRBs are used when we do not employ the MOZalgorithm As a result after a specific limit no more requestscan be fulfilled because all PRBs are utilized Figure 15 depictsthis scenario as we can see that the average number of PRBsused withMOZ is less due to improved bandwidth utilizationstrategy better coverage and enhanced SNR But with higherload of 30 UE instances all the bandwidth is used up
In Figure 16 for VOIP users also the average number ofPRBs used is much less with increasing number of users
33 Summary of the Simulation Analysis Table 3 representsthe results and statistics obtained by using MOZ where 1 to 4represent the PO set The optimal solution 119874lowast for each set ofusers is bold
It can be clearly seen that the DM always selects the bestfeasible solution thus proving an accuracy of 100 in termsof optimal solution selection The optimization proceduregenerates a PO set for varying number of users and out ofthis set the DM selects the best solution
12 Mobile Information Systems
Table 3 Statistics of the PO set and the optimal solution Olowast selected using MOZ
Number of users PO set (average delay in ms) PO set (average throughput in bps)1 2 3 4 1 2 3 4
5 000252 000328 000173 000175 40722208000 66418738666 50126800000 3949574800010 000179 000197 000252 000328 1509846667 2871940000 4357810667 585808666715 000161 000167 000169 000175 40821598667 76296266667 93016094667 9334639733320 000161 000167 000159 000155 40821541334 76643476000 97387246667 101596778667
The evaluation and comparison of the simulation resultsshow that MOZ performs very well for both voice andVOIP scenarios and performs better than the referencealgorithm in terms of maximum achievable throughput andminimum delay The reason behind these results is the useof optimization process employed by the DM which alwaysselects the optimal solution for the arriving UE instancesbased on their QoS needs and priorities and channel qualityIt further optimizes the procedure by maintaining queues toimprove performance and achieves higher SNR and lowerPLRThe cell spectral efficiency increaseswith each addedUEthat indicates improved system performance and is a proof ofrobust system design
4 Conclusions
In this paper we have proposed a multiobjective optimizerthat helps in finding an optimal solution for selecting RBsfor transmission based on the information provided to thedecision maker It provides an optimal solution for RRM toensure QoS by incorporating themultiobjective optimizationtechniques that target to achieve higher throughput and lowerdelay A set of Pareto optimal solutions is originated and isforwarded to the DM which chooses the best solution outof this set We have adopted an interactive technique thatembeds the multiobjective optimization with the moderntechniques for RRM Furthermore the results are also pro-vided that prove the efficiency of our proposed MOZ Thistechnique is innovative as it makes use of evolutionary MOOtechniques (which are very powerful for finding optimalsolution) and provides its implementation for LTE-A and willprove helpful in solving other issues for 4G and emerging 5Gnetworks
In future we aspire to investigate the opportunities forimplementingMOO techniques for the evolving 5Gnetworksthat not only aim to deal with QoS issues but also target toachieve optimal energy consumption
Competing Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] S-B Lee I Pefkianakis A Meyerson S Xu and S LuldquoProportional fair frequency-domain packet scheduling for3GPP LTE uplinkrdquo in Proceedings of the 28th Conference on
Computer Communications (IEEE INFOCOM rsquo09) pp 2611ndash2615 Rio de Janeiro Brazil April 2009
[2] ldquoPhysical layer aspects for evolved UTRArdquo 3GPP TechnicalReport 25814 version 710 2006
[3] M Nazir and F Saleemi ldquoCooperative cognitive ecology in self-organizing networks a review articlerdquo International Journal ofComputer Science and Information Security vol 10 no 2 pp159ndash167 2012
[4] K Deb Multi-Objective Optimization with Evolutionary Algo-rithms John Wiley amp Sons New York NY USA 1st edition2001
[5] J B K Deb and K M R SłowinskiMulti-Objective Optimiza-tion Interactive and Evolutionary Approaches Springer BerlinGermany 2008
[6] A De Domenico and E C Strinati ldquoA radio resource man-agement scheduling algorithm for self-organizing femtocellsrdquoin Proceedings of the IEEE 21st International Symposium onPersonal Indoor and Mobile Radio Communications Workshops(PIMRC rsquo10) pp 191ndash196 IEEE Instanbul Turkey September2010
[7] R Kausar Y Chen K K Chai L Cuthbert and J SchormansldquoQoS aware mixed traffic packet scheduling in OFDMA-basedLTE-advanced networksrdquo in Proceedings of the 4th InternationalConference on Mobile Ubiquitous Computing Systems Servicesand Technologies (UBICOMM rsquo10) pp 53ndash58 Florence FranceOctober 2010
[8] Y J Zhang and K B Letaief ldquoAdaptive resource allocation andscheduling for multiuser packet-based OFDM networksrdquo inProceedings of the IEEE International Conference on Communi-cation vol 5 pp 2949ndash2953 June 2004
[9] C Stocchi N Marchetti and N R Prasad ldquoSelf-optimizedradio resource management techniques for LTE-a local areadeploymentsrdquo inProceedings of the 2nd International Conferenceon Wireless Communication Vehicular Technology InformationTheory and Aerospace amp Electronic Systems Technology (WirelessVITAE rsquo11) pp 1ndash5 Chennai India March 2011
[10] M Cenk Erturk H Aki I Guvenc and H Arslan ldquoFairand QoS-oriented spectrum splitting in macrocell-femtocellnetworksrdquo in Proceedings of the 53rd IEEE Global Communica-tions Conference (GLOBECOM rsquo10) pp 1ndash6 Miami Fla USADecember 2010
[11] V Chandrasekhar and J G Andrews ldquoSpectrum allocation intiered cellular networksrdquo IEEE Transactions on Communica-tions vol 57 no 10 pp 3059ndash3068 2009
[12] A Roy and S K Das Optimizing QoS-Based Multicast Routingin Wireless Networks A Multi-Objective Genetic AlgorithmicApproach Springer New Yok NY USA 2002
[13] D H Lorenz and A Orda ldquoQoS routing in networks withuncertain parametersrdquo IEEEACM Transactions on Networkingvol 6 no 6 pp 768ndash778 1998
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mobile Information Systems 3
threshold and then assigns RBs It focuses mainly on delayminimization
In the perspective of multiobjective optimization mostwork done is towards discovering a collection of near-Pareto optimal solutions algorithmically In multiobjectiveoptimization communication with the DM can be doneduring the phase of optimization or during the final decision-making part In many studies a human decision maker isinvolved after a variety of solutions have been sought out [21ndash23]
The process of multiobjective optimization is incompletewithout a decision-making activity that will take the finaldecision based on various alternatives available In thiscontext many interactive multiobjective optimization tech-niques are available under the title Multicriteria Decision-Making (MCDM) [24ndash26] All the techniques are differentfrom each other but all incorporate DM to provide infor-mation to help in taking the final decision The evolutionaryalgorithms (EA) suggest the use of natural evolution theoryfor optimization like the survival of the fittest theory ofDarwin They work with a set of solutions and can also beused to find a partial PO set
MOO has been researched for many years and is focusedon the theoretical aspects [27] A lot of approaches havebeen formulatedwithmathematical programming theory forexample nonlinear programming to solve the multiobjectiveoptimization problems [28] Varying interactive approacheshave also been used in which the information is given tothe DM and the DM specifies its preferences This proceduredepends on the type of problem and its mathematical prop-erties and scalar function [29]
A power-delay minimization scheme is also introducedthat makes use of linear programming [30] In this studymultiobjective problem is transformed into a solitary objec-tive by using weighted sum technique that aims to reduceboth the delay and transmit power Simulated annealing andgreedy heuristic algorithms are also part of this work Itstarget is not specifically LTE networks but mainly the IEEE80211 Wireless LANs (WLANs) and Green Wireless AccessNetworks (GWANs)
Table 1 depicts the classification of various techniquesused for RRM and for finding an optimal solution withmultiple objectives Based on the literature survey we cansee that there exist various techniques for QoS provisionin LTE-A using different methodologies Also the multiob-jective optimization offers a variety of solutions for solvingvarious linear and nonlinear problems Still there is a lackof these multiobjective optimization technique applicationsfor RRM of LTE-A targeted towards ensuring QoS As aresult we will propose a QoS awareMultiobjective Optimizer(MOZ) for the RRM of LTE-A which will take into accountthe provision of maximum throughput and minimumdelay
2 Materials and Methods
21 Problem Formulation To measure the performance ofLTE-A network we have a set of criteria such as through-put end-to-end transmission delay energy efficiency and
Table 1 Classification of literature on RRM and optimization
Approach [7] [16] [20] [21ndash23] [30] Proposed approachScheduling radic radic radic radic radic radic
User association radic radic radic mdash mdash radic
QoS aware radic radic radic mdash radic radic
Energy efficiency mdash radic mdash mdash radic mdashRB allocation mdash mdash radic mdash radic radic
LTE radic mdash mdash mdash radic mdashLTE-A mdash radic radic mdash mdash radic
MOO mdash mdash mdash radic radic radic
DM mdash mdash mdash radic mdash radic
transmission strength The purpose of the work presentedhere is to determine the trade-offs that arise while choosingperformance metrics for assigning resources to the UserEquipment (UE) The multiobjective Pareto optimizationconsists of three steps
(1) Multiobjective problem definition(2) Optimization (finding Pareto optimal solutions)(3) Decision-making (role of DM)
In the following we will elaborate the working of our QoSaware optimizer according to the above-mentioned threesteps and it will define the framework for our proposedoptimizer
Step 1 (problem definition) To solve our multiobjectiveoptimization problem we will involve DM to find the best(optimal) solution By optimal solution we are referring toPareto optimal solution that the DM considers as the bestoption In our case we have two objectives the first one ismaximum throughput and the second one is minimumdelayWe will manage the resources in such a way that we are ableto select RB that transmits the data according to the above-mentioned objectives
ParetoOptimization ProblemWewill solve themultiobjectiveoptimization problem that will take the form
MaximizeMinimize 1198911(119910) 119891
2(119910) 119891
119894(119910)
Subject to 119910 isin 119878
(1)
where 119894 is a set of objectives that will be minimized ormaximized according to its definition 119878 is a set of networkconstraints Here 119910 is the decision vector and 1199101015840 is thedecision variable We will call 1199101015840 isin 119878 Pareto optimal only ifthere is no other 1199101015840 isin 119878 such that
119891119896(119910) le 119891
119896(1199101015840) (where 119896 = 1 119894) (2)
This means that 1199101015840 is Pareto optimal only if there is noother possible vector (or solution)119910 thatwill deteriorate somecriterion without leading towards an increase in some othercriterion
4 Mobile Information Systems
Step 2 (Pareto front) In our case we have the followingobjective functions and the first one aims at maximizingnetwork utility through increased throughput (119879
119894)
Maximize119899
sum
119894=1
(119879119894)
Subject to119899
sum
119894=1
119879119894lt 119862119894119896(119905) where 119879
119894gt 0 forall119894 isin 119873
(119873 = total UE instances)
(3)
Considering channel 119896 and user 119894 we define the channelcapacity as
119862119894119896= RB log
2(1 +
120588
1205902sdot 119875) (4)
where RB is RB Bandwidth which is calculated as (totalbandwidthnumber of RCs) 119875 is Transmission Power 120588 isSNR space determined by Bit Error Rate (BER) and 1205902 isnoise power density
119879119894is the average throughput calculated as
119879119894= TUE
119894 [119905]1
120591+ 119879119894(119905 minus 1) (1 minus
1
120591) forall119894 isin 119873 (5)
where TUE119894[119905] is throughput of UE
119894in time instance 119905 120591 is
time constraint of the smoothing filter and119873 is total UETUE119894[119905] is calculated with the help of the following
equation
TUE119894 [119905] = RB log
21 + SNR (6)
where SNR = 119879119901times 119878UE(Noise + Interference) (119879
119901is
Transmission Power 119878UE is signal gain of UE)Our next objective is defined as follows
Minimize119873
sum
119894=1
(119863119894(119905))
Subject to 119863119894(119905) lt DB
119894
(7)
where DB119894is the delay budget or the upper bound which is
equivalent to 20ms in OFDMA networks [19 31] The delayexperienced by the UE
119894should be less than this upper bound
Here 119873 represents total active UE and 119863119894(119905) is the HOL
(Head of Line) delay for UE119894at time 119905 calculated as
119863119894(119905) =
119882119894
DT119894
(8)
where119882119894is the waiting time for UE
119894and DT
119894is the normal-
ized HOL delay obtained by dividing each userrsquos waiting timeby DB
119894
This process will generate a set of Pareto optimal solutionscalled the Pareto front As a result there is no single POsolution so the question is which solution to select from a setof PO solutionsThe answer lies in Step 3 that takes advantageof DM
Step 3 (the optimal solution) In this step the DM will selectthe best solution which is RB that will be allocated fortransmission Depending on the type of incoming traffic theDM will decide which strategy to follow according to thefollowing criteria
(i) If the traffic is delay sensitive preference will be givento minimizing the delay whereas throughput can becompromised to some extent
(ii) If the incoming packet is throughput sensitive it willemphasize maximum throughout whereas delay maybe compromised
This scheme will thus allow the delay sensitive applica-tions to sacrifice throughput for lower delays without anyeffect on the throughput sensitive applications
So we have 119896 and 119894 objective functions (represented by 119892)and a total of 119898 and 119891 constraints (represented by ℎ) Weassume that the constraints plus the objective function arethe functions of the decision vectors So the DMrsquos goal is asfollows
Maximize 119892 (119909) = 1198921(119910) 119892
2(119910) 119892
119894(119910)
Minimize 119892 (119910) = 1198921(119911) 119892
2(119911) 119892
119896(119911)
Subject to ℎ (119909) = ℎ1(119910) ℎ
2(119910) ℎ
119898(119910)
ℎ (119910) = ℎ1(119911) ℎ
2(119911) ℎ
119891(119911)
119909 = 1199091 1199092 119909
119894
119911 = 1199111 1199112 119911
119896
(9)
where 119909 is the decision vector for first objective function and119911 is the decision vector for second objective function 119874lowast isin119909 or 119911 is the feasible decision or solution to our optimizationproblem 119874lowast is Pareto optimal as we have no other bettersolution So here the DM will make use of a weighted MinndashMax Approach which is adopted from GameTheory [32 33]In this scheme we will compare the relative deviation froma separately attainable minima or maxima In our case it isthe deviation from maximum throughput and delay of thesystem We will denote this by 119863max119894 (delay variation) and119879min119894 (throughput variation) We will calculate the relativedeviations using
119879min119894 = TA119904minus 119879119894
(For throughput sensitive applications) (10)
TA119904is the maximum attainable throughput of the LTE
system calculated as follows(i) We will first calculate the total RBs assuming the
bandwidth of channel is 20MHz by using the following
carriers timesOFDM symbols times Slots times RBs (11)
where carriers are 12 OFDM symbols are 7 RBs are 100 Slotsare 2 The final value is 16800 RBs per frame
Mobile Information Systems 5
(ii) Second we consider modulation of 64 QAM and asingle modulation symbol carries 6 bits The total bits willthus be
16800
times 6 bits per symbol of modulation equal to 1008Mb(12)
(iii) Thirdly considering the 4-by-4 MIMO we get
4 times 1008Mb = 403Mb (13)
This is the peak data rate(iv) Finally we calculate the overhead which will be about
25 So we have
403Mb times 075 = 302Mb (14)
So we can calculate TA119904and RC and minimum 119879min119894 is
selected by the DM for transmission That is resource chunkRC119894as in the following
119874lowast= RC119894= min119879min119894 forall119894 isin UE (15)
Now let us talk about the delay sensitive applicationswhere we will find the feasible solution 119874lowast as follows
119863max119894 = DB119894minus 119863119894(119905)
(For delay sensitive applications)
119874lowast= RC119894= max119863max119894 forall119894 isin UE
(16)
So 119874lowast is chosen for transmission based on the followingdecision vectors
119874lowast=
argmin 119879min119894 forallUE119894
argmax 119863max119894 forallUE119894
(17)
The DM will take its decision based on the following setof constraints
119899
sum
119894=1
119879119894lt 119862119894119896(119905) where 119879
119894gt 0 forall119894 isin 119873 (18)
119863119894(119905) lt DB
119894forall119894 isin 119873 (19)
119873
sum
119894=1
RB119894le Total Badndwidth (20)
sum
119895isin119869
120579119894119895ge 1 forall119894 isin 119873 forall119869 isin Total BS (21)
Equation (18) refers to channel capacity constraint and(19) refers to delay bound constraint Equation (20) is RBbandwidth constraint Equation (21) ensures that at least 1 BScovers the active UE
22 The Decision-Making Process After the optimized solu-tion is sought out by theDM the trafficwill then be scheduledin output queues discussed in detail in the next subsection
Figure 3 depicts the detail of decision-making process insidethe DM Its details are explained as follows
(1) Traffic Type Determination The DM will take as input thePareto front and will first determine the type of traffic whichin our case is either throughput or delay sensitive
(2) Optimization Process After determining the traffic typethe following actions are taken
(i) For throughput sensitive applications it will calculatethe value of 119879min119894 based on the value of 119879
119894 and then
it will select the optimal solution 119874lowast = RC119894=
min119879min119894(ii) For delay sensitive applications it will calculate the
value of 119863max119894 based on the value of 119863119894(119905) and then
it will select the optimal solution 119874lowast = RC119894=
max119863max119894(3) RC Allocation Matrix Update After that the RC assign-ment matrix will be updated The matrix 119898
119896119899is represented
as
119898119896119899=
0 if RC119899is unassigned
1 if RC119899is assigned to service group 119896
(22)
where 119896 is type of traffic and 119899 is RC number [34]
(4) RC Availability Check If there are RCs availableleft forassignment then the process starts again andmoves to traffictype determination otherwise the process ends
In the following we explain in detail the construction ofPareto front
23 Constructing Pareto Front This step describes in detailthe process of constructing a representative Pareto front asfollows
Step 1 Determine the traffic type of the incoming packetAssume 119863
1 119863
119899are the alternative solutions in terms of
119894 and 119895 where 119894 represents throughput sensitive flows and 119895represents delay sensitive flows
Step 2 Find the largest 119894 such that 119863119894gt 119863119899(where 119863
119899
represents all other UE) and add119863119894to Pareto front
Step 3 Find the smallest 119895 such that 119863119895lt 119863119899(where 119863
119899
represents all other UE) and add119863119895to Pareto front
Step 4 Repeat Steps 2 and 3 until no such 119894 and 119895 exist and wehave the Pareto front of the form (119863
119894 119863
119899119863119895 119863
119899)
Note that 119863119894is sorted in order of decreasing throughput
and 119863119895is sorted in order of increasing delay The process
generates a set of candidate solutions that will undergo theoptimization phase for selecting a singled optimal solutionThe selection of the optimal solution is based on the efficiencyof the DM which not only considers the resource require-ments of the users but also sets priorities We now present analgorithm that will make use of the defined framework andconsiders the requirements of each step
6 Mobile Information Systems
Start
Throughput sensitive traffic Delay sensitive trafficDetermine from thePareto front the traffic
requirements
On the basis ofvarious values ofTi calculate Tmini
various values ofOn the basis of
Di(t) calculate Dmaxi
Choose optimalsolution based on
Olowast= RCi = minTmini
Choose optimalsolution based on
Olowast= RCi = maxDmaxi
Update RCallocation matrix
RCs left to be allocated Yes
No
End
Figure 3The decision-making process inside the DM for RC selection based on the objectives of maximum throughput andminimumdelay
24 The Multiobjective Optimizer Algorithm The detailedworking of proposed MOZ is outlined in Figure 4 TheMOZconsists of three main steps as follows
Step 1 The traffic classifier will take as input traffic fromvarious UE instances and will divide them into 2 queues oneconsists of packets that have stringent throughput require-ment and the other queue consists of packets from delaysensitive applications
Step 2 The information from Step 1 is then passed onto theoptimizer that will find out the Pareto front and will send itsdecision set to the decision maker
Step 3 The DM will make the final decision based on theinformation it has and according to its decision-makingcapabilities
Step 4 Finally119874lowast is sent out to the queue for scheduling andthe process then diverts to Step 2 to get the next 119874lowast later tobe scheduled
Step 5 The process is completed when either no RC is left orthere are no more incoming packets
Steps 2 and 3 are interrelated as the DM is way dependenton the input it gets from the Pareto front so constructing alegitimate Pareto front is itself a very important task forMOZThe decision-making process inside the DM is characterizedby a number of factors already explained and is one of themajor components of our proposed strategy
Compared to most algorithms and frameworks designedto ensure QoS and to manage RRM the distinct differencesof our proposed approach are as follows (1) it makes useof multiobjective optimization that deals with conflictingobjectives simultaneously instead of converting them intoa single objective (2) It makes use of a decision makerthat has the decision-making capability and works basedon the information it has and various algorithms used(3) Multiple UE instances are served based on the type ofincoming trafficflow instead of allocating RBs to just oneflow
Mobile Information Systems 7
Trafficclassification Optimization Decision-making Scheduling
Input trafficfrom variousUE instances
applicationsDelay sensitive
Throughput sensitiveapplications
Theoptimizationprocess todeterminethe Paretofront
DetermineO
lowast (theoptimalsolution)
Figure 4 The proposed MOZ
The goal of using MOZ is to always get an optimalsolution which is chosen from a set of solutions in the formof Pareto front
25 Queuing and Scheduling Now that we have the Paretooptimal solution in hand the packets are now ready tobe scheduled in their respective queues for transmission asfollows
Step 1 In this step we will define the priority metric for delaysensitive flows The priority of UE
119894at time 119905 119875
119889119894(119905) is
119875119889119894(119905) = 119863max119894 lowast 119866119894119895 (23)
where 119866119894119895is channel gain of UE
119894on RC
119895calculated as
119866119894119895= 10
Path loss10lowast 10
shadow fading10
lowast 10multipath fading10
(24)
Step 2 We will then form a delay queue for a given UE119894that
will ensure delay bounds as follows
119863119894(119905 + 1) = [119863
119894(119905) minus 120588119877
119894(119905)] +
119873
sum
119894=1
119876119894(119905) (25)
where 119876119894(119905) is the mean queue length at time interval 119905 and
119873 is total UE scheduled and 120588119877119894(119905) is the service rate In
each time interval the UEwith highest priority occupies frontposition in the queue
Step 3 For the priority metric of throughput sensitive flowswe will use the following
119875119905119894(119905) = 120590
119894lowast 119879min (119894) (26)
where 120590119894is the Head of Line (HOL) blocking and 119875
119905119894(119905) is the
priority metric for throughput sensitive applications In eachtime interval the UE instances are scheduled according to theascending order of their priority values
It follows the following constraints
119876119894119896
st forall119894 isin 119873 119896 isin delay sensitive traffic only (27)
119876119894119895
st forall119894 isin 119873 and 119895
isin throughput sensitive traffic only(28)
119894 = 119895 Queue Length
= Length (119876119894119896+ 119876119894119895)
(29)
Equation (27) ensures that the queue 119876119894takes as input
delay sensitive traffic only (28) ensures that the queue 119876119894119895
takes in throughput sensitive traffic only and (29) states thatqueue length is the total of both the queues
Thus the defined framework and the optimizer will worktogether to form our proposed model The whole process hasfour main goals
(1) Determining the target objectives(2) Based on the objectives forming a system model that
will take into account the objective functions to beminimized or maximized
(3) Constructing a representative solution pool to aid thedecision maker
(4) Introducing a decision-making entity that will notonly ease the task of scheduling but also find anoptimal solution to satisfy the defined objectives
In the following section we will define the simulationmodel and will discuss the results
3 Simulation and Results
31 Simulation Model Designing the model for LTE-A per-formance evaluation is generally more complex comparedto a 3G system An interface is used between variousevolvedNodeBs (eNBs ie the BS) in the LTE system toease the traffic load that occurs due to handovers In this
8 Mobile Information Systems
Table 2 Simulation parameters
Parameters ValuesCarrier frequency 2GHzBandwidth 10MHzNumber of subcarriers 300Number of RBs 45TTI 1msPath loss model Free space propagation modelTransmit power 30mWFading model RayleighPacket model Poisson modelMultiplexing MIMONumber of UE instances 20
section we will design the simulation model for testing andwill evaluate our system for examining delay throughputPacket Loss Ratio (PLR) the effect of SNR and cell spectralefficiency
We are considering a cell with the capacity of 10MHz andit consists of 45 RBs and the carrier frequency is 2GHz Allthe available RBs are controlled by the eNodeB The RBs aredivided among theUE instances in the formof RCsThe LTE-A parameters used for simulation are described in Table 2
We then calculate the number of bits in each RB (ie itsdata rate) It can be computed as follows
119887119894RC = NStti lowastNS119904 lowast 119878rb (30)
where 119887119894RC is achievable data rate NStti is number of slots in
each TTI NS119904is number of symbols in each slot and 119878rb is
number of subcarriers in each RBEach UE is assigned a queue according to the traffic type
and parameters as described in the system model previouslyAfter that we will calculate the channel gain of each UE
119894on
RC and we will now proceeed to calculate the SNRThe SNR values are reported to ENodeB by theUE at each
TTI All the users have a different value of SNR accordingto the experienced fading and multipath propagation thetransmit power (119875) and the total RBs (119873) as follows
SNR =119875 lowast 119866
119894119895
119873(Noise + Interference) (31)
The incoming packets will be treated by the eNodeBaccording to the parameters defined in Table 2 and thesystem model presented All packets have varying lengthand are streamed in queues according to their priorities andproperties Now for the delay sensitive packets the thresholdDB119894is set to be 20ms for each UEThus we can define the Average System Delay (AvD) with
a total simulation time of 119879 at time instance 119905 with totalnumber of UE instances equal to119873 as follows
AvD = 1119879
119879
sum
119905=1
1
119873
119873
sum
119894=1
119863119894(119905) (32)
We will then define AvTH which is the average through-put with trans
119894(119905) being the number of packetssec as follows
AvTH = 1119879
119873
sum
119894=1
119879
sum
119905=1
trans119894(119905) (33)
Now for evaluating the PLR by using the size of discardedpackets pdisc
119894(119905) and the sum of sizes of all packets psize
119894(119905)
we have
PLR =sum119873
119894=1sum119879
119905=1pdisc119894(119905)
sum119873
119894=1sum119879
119905=1psize119894(119905)
(34)
Theoverall spectrumefficiency (119878eff ) ismeasured in termsof total traffic usage factor 119879
119891by applying the formula
119878eff =119862tot119879119891
(35)
where 119879119891= 119903 times 119901 where 119903 is the resource usage factor and
119901 is total traffic 119862tot is number of available traffic channels inthe system
The queuing model is set up by considering 119883119905that
represents the number ofUE instances at a given time interval119905 isin 119879 and 119880
119905denotes the number of new UE instances
that arrive at time 119905 isin 119879 with common probability densityfunction 119891 on 119879 and we define
119883119905+1=
119880119905+1
119883119905= 0
(119883119905minus 1) + 119880
119905+1119883119905gt 0
119905 isin 119879 (36)
where 119883 = 1198830 1198831 is a Markov Chain that has the
probability matrix of the following form
119875 (0 119910) = 119891 (119910) 119910 isin 119879
119875 (119910 119911) = 119891 (119910 minus 119911 + 1)
119911 isin 119879 119910 isin (119911 minus 1 119911 119911 + 1 )
(37)
The chain119883 defined here is the queuing chain of distribu-tion 119891 and starting from 119910 isin 119879 a UE instance is served andnew UE instances arrive by the next time unit 119905 controlled by119891 So the probability of going from state 119910 to 119911 isin (119910minus1 119910 119910+1 ) is lfloor(119911 minus 119910 + 1)rfloor
32 Results Wewill now discuss the simulation results basedon the simulation model and proposed strategy In thispaper we have used the LTE-SIM [35] simulator availableonline at httptelematicspolibaitLTE-Sim It is crucial tocarry out performance evaluation for LTE-A systems and thisrequires the use of standard simulators LTE-Sim is designedaccording to the LTE-A standard with main features suchas multicell environment multiuser support eNodeB CQIUE resource allocation QoS management and schedulingstrategies
The network topology under consideration comprisesnetwork nodes (including eNodeBs and UE) distributedamong various cells We have tested our algorithms for both
Mobile Information Systems 9
MMEGW
MMEGW
Single-cell
eNBeNB
eNB
eNB
UE
UE
Multicell
EGW
eNBNBNBBBNBNBBBBBBBeNB
eNB
UE
UE
Figure 5 Single-cell and multicell simulation scenarios
5 10 15 20Users
DPSMOZ
0
200000
400000
600000
800000
1000000
1200000
Thro
ughp
ut (b
ps)
Figure 6 Video throughput
the single-cell and multicell scenario as shown in Figure 5The first scenario consists of many eNodeBs due to the factthat multiple cells are covered while in the second scenarioone eNodeB handles all UE
We will evaluate the performance of our proposed MOZwhile comparing it with that of DPS [20] For each UEdownlink VOIP and video flows are considered All flows areactive during the whole simulation Time for each simulationis around 100 sThe simulation is performedusing aWindowsmachine with Cygwin (Linux-like environment) a 26GHzCPU and 4GB RAM
321 Delay and Throughput The goal of achieving desiredlevel of QoS is very challenging due to Real Timemultimediaapplications which have strict delay and throughput con-straints Figures 6 and 7 show the throughput for video andVOIP flows with increasing number of UE instances Fromboth the figures it can be seen that the throughput achievedby MOZ is better than that of DPS algorithm althoughthe difference for VOIP flow is much less Throughput alsoincreases with increasing number of users For video up to15 users increasing pattern is observed and after 25 usersthroughput increases but gradually For VOIP flows thethroughput is still increasing although it is delay sensitiveThereason behind this increased throughput is that the systemresources are utilized according to the requirement of users
5 10 15 20Users
MOZDPS
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 7 VOIP throughput
5 10 15 20Users
DPSMOZ
00015
000155
00016
000165
00017
000175
00018
Del
ay (m
sec)
Figure 8 VOIP delay
and only the selected UE instances transmit on the best RBsAlso it is worth noticing that the throughput achieved byMOZ is much closer to the theoretical maximum In thisprocess decisions are made based on the channel quality ofeach UE which constantly measures the QoS parametersbased on the algorithms and techniques proposed After10 users the throughput for video using MOZ achievesperformance gains in order of 1 to 15 while for VOIP it isfrom 1 to 10
Figures 8 and 9 show the average delay of the MOZand DPS algorithm with increasing UE for both the videoand VOIP flows From Figure 8 it can be seen that thedelay performance of MOZ is slightly higher compared toDPS with less than 10 users and as number of UE instancesincreases MOZ shows better performance compared to theDPS algorithm because after 10 plus users there is morevideo streaming and it is not distinguishing VOIP and videoflows so the delay for DPS will increase in consecutive TTIsFigure 9 shows the performance of bothMOZandDPSwhichis very close because with increasing users video demand hasa possibility of rising
Also our results ensure that the delay always remainsbelow the threshold value The proposed MOZ achieves
10 Mobile Information Systems
DPSMOZ
5 10 15 20Users
0
002
004
006
008
01
012
014
Del
ay (m
sec)
Figure 9 Video delay
DPSMOZ
5 10 15 20Users
00280029
0030031003200330034003500360037
PLR
Figure 10 Packet loss ratio
performance gains by reducing delay from 1 to 5 for videoand 1 to 8 for VOIP after 10 users Also it is worth noticinghere that the delay achieved is way below the threshold whichwas set to 20ms
322 PLR Now let us compare the PLR performance ofMOZ and DPS algorithm shown in Figure 10 As expectedthe PLR is increasing as the number of users increasesbecause there are more packets that will be discarded as therewill not be sufficient RBs for transmitting all the packetsconsidering that the packets are nearing their delay thresholdHowever in comparison with the DPS algorithm MOZ iscapable of achieving a better PLR as it supports more usersby distinguishing packets according to their scheduling andresource needs It reduces the loss probability by exploitingthe gains at physical layer Note that the PLR counts only forphysical layer losses MOZ achieves a better PLR and packetloss is reduced from 2 to 14 as compared to the referencealgorithm
323 SNR Figure 11 shows the SNR performance of MOZwith reference to average system throughputWith increasing
5 10 15 20SNR (db)
MOZ
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 11 SNR versus throughput
10 20 30 40SNR (db)
MOZ
021
022
023
024
025
026
027
028
Spec
tral
effici
ency
(bps
)
Figure 12 Cell spectral efficiency
throughput it is observed that the SNR is also increasingwhich depicts better channel quality and less distortion andis one of the reasons we are able to achieve higher systemthroughputThe higher values of SNR are indicator of higherpercentage of data as compared to noise and our resultsshow that noise is around 004 of the actual signal whichis an indicator that the video and voice quality achieved areexcellent
324 Spectral Efficiency Then the spectral efficiency (inbitss) is computed and is depicted in Figure 12 At eachscheduling interval the RB is allocated to UE instancesaccording to their needs As the number of UE instancesincreasesMOZ still guarantees QoS constraints tomore flowswith a positive effect on overall system efficiency The systemsupports delay and throughput guarantees simultaneouslytherefore the spectral efficiency increases with increasingusers which is a good indicator of overall system gain andefficiency
325 Energy and Memory Consumption Figure 13 demon-strates the energy consumption in joules with respect to thenumber of users in the network The energy usage of MOZis compared to PF scheduling and DPS algorithm As the
Mobile Information Systems 11
5 10 15 20Users
DPSMOZPF
0
05
1
15
2
25
Ener
gy (j
oule
s)
Figure 13 Energy consumption versus no of nodes
5 10 15 20Users
MOZ
0
50
100
150
200
250
Mem
ory
Figure 14 Memory consumption versus no of nodes
number of users increases the level of energy consumptionalso rises for all algorithms As the eNodeB remains active allthe time this rise is expected But the energy consumptionlevel achieved by MOZ is the lowest due to the fact thatefficient RB selection takes place which reduces the delaysthus lowering levels of energy consumption Also only theoptimal solution is selected which in turn results in energygains and network lifetime
In Figure 14 the memory consumption in bits versus thenumber of users is depictedThe DM takes decision based onthe information it has and the solution depends on algorithmsthat run dynamically and this process is continuous untilthere are no more RBs to allocate Moreover it has beenrealized that the continuous assignment of RBs accordingto UE instancesrsquo needs utilizes less memory which in turnincreases network life time and performance and reducesdelays
326 PRB Utilization In this subsection we will show effectof the proposed algorithm on PRB utilization The scenariodepicts results according to two groups In the first groupsimulations are carried out without using the proposedMOZthe second group implements MOZ The performance is
5 10 15 20 25 30Video users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 15 Mean PRBs used for video
5 10 15 20 25 30VOIP users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 16 Mean PRBs used for VOIP
measured in terms of mean PRBs being used for both videoandVOIP users In high load the delay increases and a highernumber of PRBs are used when we do not employ the MOZalgorithm As a result after a specific limit no more requestscan be fulfilled because all PRBs are utilized Figure 15 depictsthis scenario as we can see that the average number of PRBsused withMOZ is less due to improved bandwidth utilizationstrategy better coverage and enhanced SNR But with higherload of 30 UE instances all the bandwidth is used up
In Figure 16 for VOIP users also the average number ofPRBs used is much less with increasing number of users
33 Summary of the Simulation Analysis Table 3 representsthe results and statistics obtained by using MOZ where 1 to 4represent the PO set The optimal solution 119874lowast for each set ofusers is bold
It can be clearly seen that the DM always selects the bestfeasible solution thus proving an accuracy of 100 in termsof optimal solution selection The optimization proceduregenerates a PO set for varying number of users and out ofthis set the DM selects the best solution
12 Mobile Information Systems
Table 3 Statistics of the PO set and the optimal solution Olowast selected using MOZ
Number of users PO set (average delay in ms) PO set (average throughput in bps)1 2 3 4 1 2 3 4
5 000252 000328 000173 000175 40722208000 66418738666 50126800000 3949574800010 000179 000197 000252 000328 1509846667 2871940000 4357810667 585808666715 000161 000167 000169 000175 40821598667 76296266667 93016094667 9334639733320 000161 000167 000159 000155 40821541334 76643476000 97387246667 101596778667
The evaluation and comparison of the simulation resultsshow that MOZ performs very well for both voice andVOIP scenarios and performs better than the referencealgorithm in terms of maximum achievable throughput andminimum delay The reason behind these results is the useof optimization process employed by the DM which alwaysselects the optimal solution for the arriving UE instancesbased on their QoS needs and priorities and channel qualityIt further optimizes the procedure by maintaining queues toimprove performance and achieves higher SNR and lowerPLRThe cell spectral efficiency increaseswith each addedUEthat indicates improved system performance and is a proof ofrobust system design
4 Conclusions
In this paper we have proposed a multiobjective optimizerthat helps in finding an optimal solution for selecting RBsfor transmission based on the information provided to thedecision maker It provides an optimal solution for RRM toensure QoS by incorporating themultiobjective optimizationtechniques that target to achieve higher throughput and lowerdelay A set of Pareto optimal solutions is originated and isforwarded to the DM which chooses the best solution outof this set We have adopted an interactive technique thatembeds the multiobjective optimization with the moderntechniques for RRM Furthermore the results are also pro-vided that prove the efficiency of our proposed MOZ Thistechnique is innovative as it makes use of evolutionary MOOtechniques (which are very powerful for finding optimalsolution) and provides its implementation for LTE-A and willprove helpful in solving other issues for 4G and emerging 5Gnetworks
In future we aspire to investigate the opportunities forimplementingMOO techniques for the evolving 5Gnetworksthat not only aim to deal with QoS issues but also target toachieve optimal energy consumption
Competing Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] S-B Lee I Pefkianakis A Meyerson S Xu and S LuldquoProportional fair frequency-domain packet scheduling for3GPP LTE uplinkrdquo in Proceedings of the 28th Conference on
Computer Communications (IEEE INFOCOM rsquo09) pp 2611ndash2615 Rio de Janeiro Brazil April 2009
[2] ldquoPhysical layer aspects for evolved UTRArdquo 3GPP TechnicalReport 25814 version 710 2006
[3] M Nazir and F Saleemi ldquoCooperative cognitive ecology in self-organizing networks a review articlerdquo International Journal ofComputer Science and Information Security vol 10 no 2 pp159ndash167 2012
[4] K Deb Multi-Objective Optimization with Evolutionary Algo-rithms John Wiley amp Sons New York NY USA 1st edition2001
[5] J B K Deb and K M R SłowinskiMulti-Objective Optimiza-tion Interactive and Evolutionary Approaches Springer BerlinGermany 2008
[6] A De Domenico and E C Strinati ldquoA radio resource man-agement scheduling algorithm for self-organizing femtocellsrdquoin Proceedings of the IEEE 21st International Symposium onPersonal Indoor and Mobile Radio Communications Workshops(PIMRC rsquo10) pp 191ndash196 IEEE Instanbul Turkey September2010
[7] R Kausar Y Chen K K Chai L Cuthbert and J SchormansldquoQoS aware mixed traffic packet scheduling in OFDMA-basedLTE-advanced networksrdquo in Proceedings of the 4th InternationalConference on Mobile Ubiquitous Computing Systems Servicesand Technologies (UBICOMM rsquo10) pp 53ndash58 Florence FranceOctober 2010
[8] Y J Zhang and K B Letaief ldquoAdaptive resource allocation andscheduling for multiuser packet-based OFDM networksrdquo inProceedings of the IEEE International Conference on Communi-cation vol 5 pp 2949ndash2953 June 2004
[9] C Stocchi N Marchetti and N R Prasad ldquoSelf-optimizedradio resource management techniques for LTE-a local areadeploymentsrdquo inProceedings of the 2nd International Conferenceon Wireless Communication Vehicular Technology InformationTheory and Aerospace amp Electronic Systems Technology (WirelessVITAE rsquo11) pp 1ndash5 Chennai India March 2011
[10] M Cenk Erturk H Aki I Guvenc and H Arslan ldquoFairand QoS-oriented spectrum splitting in macrocell-femtocellnetworksrdquo in Proceedings of the 53rd IEEE Global Communica-tions Conference (GLOBECOM rsquo10) pp 1ndash6 Miami Fla USADecember 2010
[11] V Chandrasekhar and J G Andrews ldquoSpectrum allocation intiered cellular networksrdquo IEEE Transactions on Communica-tions vol 57 no 10 pp 3059ndash3068 2009
[12] A Roy and S K Das Optimizing QoS-Based Multicast Routingin Wireless Networks A Multi-Objective Genetic AlgorithmicApproach Springer New Yok NY USA 2002
[13] D H Lorenz and A Orda ldquoQoS routing in networks withuncertain parametersrdquo IEEEACM Transactions on Networkingvol 6 no 6 pp 768ndash778 1998
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
4 Mobile Information Systems
Step 2 (Pareto front) In our case we have the followingobjective functions and the first one aims at maximizingnetwork utility through increased throughput (119879
119894)
Maximize119899
sum
119894=1
(119879119894)
Subject to119899
sum
119894=1
119879119894lt 119862119894119896(119905) where 119879
119894gt 0 forall119894 isin 119873
(119873 = total UE instances)
(3)
Considering channel 119896 and user 119894 we define the channelcapacity as
119862119894119896= RB log
2(1 +
120588
1205902sdot 119875) (4)
where RB is RB Bandwidth which is calculated as (totalbandwidthnumber of RCs) 119875 is Transmission Power 120588 isSNR space determined by Bit Error Rate (BER) and 1205902 isnoise power density
119879119894is the average throughput calculated as
119879119894= TUE
119894 [119905]1
120591+ 119879119894(119905 minus 1) (1 minus
1
120591) forall119894 isin 119873 (5)
where TUE119894[119905] is throughput of UE
119894in time instance 119905 120591 is
time constraint of the smoothing filter and119873 is total UETUE119894[119905] is calculated with the help of the following
equation
TUE119894 [119905] = RB log
21 + SNR (6)
where SNR = 119879119901times 119878UE(Noise + Interference) (119879
119901is
Transmission Power 119878UE is signal gain of UE)Our next objective is defined as follows
Minimize119873
sum
119894=1
(119863119894(119905))
Subject to 119863119894(119905) lt DB
119894
(7)
where DB119894is the delay budget or the upper bound which is
equivalent to 20ms in OFDMA networks [19 31] The delayexperienced by the UE
119894should be less than this upper bound
Here 119873 represents total active UE and 119863119894(119905) is the HOL
(Head of Line) delay for UE119894at time 119905 calculated as
119863119894(119905) =
119882119894
DT119894
(8)
where119882119894is the waiting time for UE
119894and DT
119894is the normal-
ized HOL delay obtained by dividing each userrsquos waiting timeby DB
119894
This process will generate a set of Pareto optimal solutionscalled the Pareto front As a result there is no single POsolution so the question is which solution to select from a setof PO solutionsThe answer lies in Step 3 that takes advantageof DM
Step 3 (the optimal solution) In this step the DM will selectthe best solution which is RB that will be allocated fortransmission Depending on the type of incoming traffic theDM will decide which strategy to follow according to thefollowing criteria
(i) If the traffic is delay sensitive preference will be givento minimizing the delay whereas throughput can becompromised to some extent
(ii) If the incoming packet is throughput sensitive it willemphasize maximum throughout whereas delay maybe compromised
This scheme will thus allow the delay sensitive applica-tions to sacrifice throughput for lower delays without anyeffect on the throughput sensitive applications
So we have 119896 and 119894 objective functions (represented by 119892)and a total of 119898 and 119891 constraints (represented by ℎ) Weassume that the constraints plus the objective function arethe functions of the decision vectors So the DMrsquos goal is asfollows
Maximize 119892 (119909) = 1198921(119910) 119892
2(119910) 119892
119894(119910)
Minimize 119892 (119910) = 1198921(119911) 119892
2(119911) 119892
119896(119911)
Subject to ℎ (119909) = ℎ1(119910) ℎ
2(119910) ℎ
119898(119910)
ℎ (119910) = ℎ1(119911) ℎ
2(119911) ℎ
119891(119911)
119909 = 1199091 1199092 119909
119894
119911 = 1199111 1199112 119911
119896
(9)
where 119909 is the decision vector for first objective function and119911 is the decision vector for second objective function 119874lowast isin119909 or 119911 is the feasible decision or solution to our optimizationproblem 119874lowast is Pareto optimal as we have no other bettersolution So here the DM will make use of a weighted MinndashMax Approach which is adopted from GameTheory [32 33]In this scheme we will compare the relative deviation froma separately attainable minima or maxima In our case it isthe deviation from maximum throughput and delay of thesystem We will denote this by 119863max119894 (delay variation) and119879min119894 (throughput variation) We will calculate the relativedeviations using
119879min119894 = TA119904minus 119879119894
(For throughput sensitive applications) (10)
TA119904is the maximum attainable throughput of the LTE
system calculated as follows(i) We will first calculate the total RBs assuming the
bandwidth of channel is 20MHz by using the following
carriers timesOFDM symbols times Slots times RBs (11)
where carriers are 12 OFDM symbols are 7 RBs are 100 Slotsare 2 The final value is 16800 RBs per frame
Mobile Information Systems 5
(ii) Second we consider modulation of 64 QAM and asingle modulation symbol carries 6 bits The total bits willthus be
16800
times 6 bits per symbol of modulation equal to 1008Mb(12)
(iii) Thirdly considering the 4-by-4 MIMO we get
4 times 1008Mb = 403Mb (13)
This is the peak data rate(iv) Finally we calculate the overhead which will be about
25 So we have
403Mb times 075 = 302Mb (14)
So we can calculate TA119904and RC and minimum 119879min119894 is
selected by the DM for transmission That is resource chunkRC119894as in the following
119874lowast= RC119894= min119879min119894 forall119894 isin UE (15)
Now let us talk about the delay sensitive applicationswhere we will find the feasible solution 119874lowast as follows
119863max119894 = DB119894minus 119863119894(119905)
(For delay sensitive applications)
119874lowast= RC119894= max119863max119894 forall119894 isin UE
(16)
So 119874lowast is chosen for transmission based on the followingdecision vectors
119874lowast=
argmin 119879min119894 forallUE119894
argmax 119863max119894 forallUE119894
(17)
The DM will take its decision based on the following setof constraints
119899
sum
119894=1
119879119894lt 119862119894119896(119905) where 119879
119894gt 0 forall119894 isin 119873 (18)
119863119894(119905) lt DB
119894forall119894 isin 119873 (19)
119873
sum
119894=1
RB119894le Total Badndwidth (20)
sum
119895isin119869
120579119894119895ge 1 forall119894 isin 119873 forall119869 isin Total BS (21)
Equation (18) refers to channel capacity constraint and(19) refers to delay bound constraint Equation (20) is RBbandwidth constraint Equation (21) ensures that at least 1 BScovers the active UE
22 The Decision-Making Process After the optimized solu-tion is sought out by theDM the trafficwill then be scheduledin output queues discussed in detail in the next subsection
Figure 3 depicts the detail of decision-making process insidethe DM Its details are explained as follows
(1) Traffic Type Determination The DM will take as input thePareto front and will first determine the type of traffic whichin our case is either throughput or delay sensitive
(2) Optimization Process After determining the traffic typethe following actions are taken
(i) For throughput sensitive applications it will calculatethe value of 119879min119894 based on the value of 119879
119894 and then
it will select the optimal solution 119874lowast = RC119894=
min119879min119894(ii) For delay sensitive applications it will calculate the
value of 119863max119894 based on the value of 119863119894(119905) and then
it will select the optimal solution 119874lowast = RC119894=
max119863max119894(3) RC Allocation Matrix Update After that the RC assign-ment matrix will be updated The matrix 119898
119896119899is represented
as
119898119896119899=
0 if RC119899is unassigned
1 if RC119899is assigned to service group 119896
(22)
where 119896 is type of traffic and 119899 is RC number [34]
(4) RC Availability Check If there are RCs availableleft forassignment then the process starts again andmoves to traffictype determination otherwise the process ends
In the following we explain in detail the construction ofPareto front
23 Constructing Pareto Front This step describes in detailthe process of constructing a representative Pareto front asfollows
Step 1 Determine the traffic type of the incoming packetAssume 119863
1 119863
119899are the alternative solutions in terms of
119894 and 119895 where 119894 represents throughput sensitive flows and 119895represents delay sensitive flows
Step 2 Find the largest 119894 such that 119863119894gt 119863119899(where 119863
119899
represents all other UE) and add119863119894to Pareto front
Step 3 Find the smallest 119895 such that 119863119895lt 119863119899(where 119863
119899
represents all other UE) and add119863119895to Pareto front
Step 4 Repeat Steps 2 and 3 until no such 119894 and 119895 exist and wehave the Pareto front of the form (119863
119894 119863
119899119863119895 119863
119899)
Note that 119863119894is sorted in order of decreasing throughput
and 119863119895is sorted in order of increasing delay The process
generates a set of candidate solutions that will undergo theoptimization phase for selecting a singled optimal solutionThe selection of the optimal solution is based on the efficiencyof the DM which not only considers the resource require-ments of the users but also sets priorities We now present analgorithm that will make use of the defined framework andconsiders the requirements of each step
6 Mobile Information Systems
Start
Throughput sensitive traffic Delay sensitive trafficDetermine from thePareto front the traffic
requirements
On the basis ofvarious values ofTi calculate Tmini
various values ofOn the basis of
Di(t) calculate Dmaxi
Choose optimalsolution based on
Olowast= RCi = minTmini
Choose optimalsolution based on
Olowast= RCi = maxDmaxi
Update RCallocation matrix
RCs left to be allocated Yes
No
End
Figure 3The decision-making process inside the DM for RC selection based on the objectives of maximum throughput andminimumdelay
24 The Multiobjective Optimizer Algorithm The detailedworking of proposed MOZ is outlined in Figure 4 TheMOZconsists of three main steps as follows
Step 1 The traffic classifier will take as input traffic fromvarious UE instances and will divide them into 2 queues oneconsists of packets that have stringent throughput require-ment and the other queue consists of packets from delaysensitive applications
Step 2 The information from Step 1 is then passed onto theoptimizer that will find out the Pareto front and will send itsdecision set to the decision maker
Step 3 The DM will make the final decision based on theinformation it has and according to its decision-makingcapabilities
Step 4 Finally119874lowast is sent out to the queue for scheduling andthe process then diverts to Step 2 to get the next 119874lowast later tobe scheduled
Step 5 The process is completed when either no RC is left orthere are no more incoming packets
Steps 2 and 3 are interrelated as the DM is way dependenton the input it gets from the Pareto front so constructing alegitimate Pareto front is itself a very important task forMOZThe decision-making process inside the DM is characterizedby a number of factors already explained and is one of themajor components of our proposed strategy
Compared to most algorithms and frameworks designedto ensure QoS and to manage RRM the distinct differencesof our proposed approach are as follows (1) it makes useof multiobjective optimization that deals with conflictingobjectives simultaneously instead of converting them intoa single objective (2) It makes use of a decision makerthat has the decision-making capability and works basedon the information it has and various algorithms used(3) Multiple UE instances are served based on the type ofincoming trafficflow instead of allocating RBs to just oneflow
Mobile Information Systems 7
Trafficclassification Optimization Decision-making Scheduling
Input trafficfrom variousUE instances
applicationsDelay sensitive
Throughput sensitiveapplications
Theoptimizationprocess todeterminethe Paretofront
DetermineO
lowast (theoptimalsolution)
Figure 4 The proposed MOZ
The goal of using MOZ is to always get an optimalsolution which is chosen from a set of solutions in the formof Pareto front
25 Queuing and Scheduling Now that we have the Paretooptimal solution in hand the packets are now ready tobe scheduled in their respective queues for transmission asfollows
Step 1 In this step we will define the priority metric for delaysensitive flows The priority of UE
119894at time 119905 119875
119889119894(119905) is
119875119889119894(119905) = 119863max119894 lowast 119866119894119895 (23)
where 119866119894119895is channel gain of UE
119894on RC
119895calculated as
119866119894119895= 10
Path loss10lowast 10
shadow fading10
lowast 10multipath fading10
(24)
Step 2 We will then form a delay queue for a given UE119894that
will ensure delay bounds as follows
119863119894(119905 + 1) = [119863
119894(119905) minus 120588119877
119894(119905)] +
119873
sum
119894=1
119876119894(119905) (25)
where 119876119894(119905) is the mean queue length at time interval 119905 and
119873 is total UE scheduled and 120588119877119894(119905) is the service rate In
each time interval the UEwith highest priority occupies frontposition in the queue
Step 3 For the priority metric of throughput sensitive flowswe will use the following
119875119905119894(119905) = 120590
119894lowast 119879min (119894) (26)
where 120590119894is the Head of Line (HOL) blocking and 119875
119905119894(119905) is the
priority metric for throughput sensitive applications In eachtime interval the UE instances are scheduled according to theascending order of their priority values
It follows the following constraints
119876119894119896
st forall119894 isin 119873 119896 isin delay sensitive traffic only (27)
119876119894119895
st forall119894 isin 119873 and 119895
isin throughput sensitive traffic only(28)
119894 = 119895 Queue Length
= Length (119876119894119896+ 119876119894119895)
(29)
Equation (27) ensures that the queue 119876119894takes as input
delay sensitive traffic only (28) ensures that the queue 119876119894119895
takes in throughput sensitive traffic only and (29) states thatqueue length is the total of both the queues
Thus the defined framework and the optimizer will worktogether to form our proposed model The whole process hasfour main goals
(1) Determining the target objectives(2) Based on the objectives forming a system model that
will take into account the objective functions to beminimized or maximized
(3) Constructing a representative solution pool to aid thedecision maker
(4) Introducing a decision-making entity that will notonly ease the task of scheduling but also find anoptimal solution to satisfy the defined objectives
In the following section we will define the simulationmodel and will discuss the results
3 Simulation and Results
31 Simulation Model Designing the model for LTE-A per-formance evaluation is generally more complex comparedto a 3G system An interface is used between variousevolvedNodeBs (eNBs ie the BS) in the LTE system toease the traffic load that occurs due to handovers In this
8 Mobile Information Systems
Table 2 Simulation parameters
Parameters ValuesCarrier frequency 2GHzBandwidth 10MHzNumber of subcarriers 300Number of RBs 45TTI 1msPath loss model Free space propagation modelTransmit power 30mWFading model RayleighPacket model Poisson modelMultiplexing MIMONumber of UE instances 20
section we will design the simulation model for testing andwill evaluate our system for examining delay throughputPacket Loss Ratio (PLR) the effect of SNR and cell spectralefficiency
We are considering a cell with the capacity of 10MHz andit consists of 45 RBs and the carrier frequency is 2GHz Allthe available RBs are controlled by the eNodeB The RBs aredivided among theUE instances in the formof RCsThe LTE-A parameters used for simulation are described in Table 2
We then calculate the number of bits in each RB (ie itsdata rate) It can be computed as follows
119887119894RC = NStti lowastNS119904 lowast 119878rb (30)
where 119887119894RC is achievable data rate NStti is number of slots in
each TTI NS119904is number of symbols in each slot and 119878rb is
number of subcarriers in each RBEach UE is assigned a queue according to the traffic type
and parameters as described in the system model previouslyAfter that we will calculate the channel gain of each UE
119894on
RC and we will now proceeed to calculate the SNRThe SNR values are reported to ENodeB by theUE at each
TTI All the users have a different value of SNR accordingto the experienced fading and multipath propagation thetransmit power (119875) and the total RBs (119873) as follows
SNR =119875 lowast 119866
119894119895
119873(Noise + Interference) (31)
The incoming packets will be treated by the eNodeBaccording to the parameters defined in Table 2 and thesystem model presented All packets have varying lengthand are streamed in queues according to their priorities andproperties Now for the delay sensitive packets the thresholdDB119894is set to be 20ms for each UEThus we can define the Average System Delay (AvD) with
a total simulation time of 119879 at time instance 119905 with totalnumber of UE instances equal to119873 as follows
AvD = 1119879
119879
sum
119905=1
1
119873
119873
sum
119894=1
119863119894(119905) (32)
We will then define AvTH which is the average through-put with trans
119894(119905) being the number of packetssec as follows
AvTH = 1119879
119873
sum
119894=1
119879
sum
119905=1
trans119894(119905) (33)
Now for evaluating the PLR by using the size of discardedpackets pdisc
119894(119905) and the sum of sizes of all packets psize
119894(119905)
we have
PLR =sum119873
119894=1sum119879
119905=1pdisc119894(119905)
sum119873
119894=1sum119879
119905=1psize119894(119905)
(34)
Theoverall spectrumefficiency (119878eff ) ismeasured in termsof total traffic usage factor 119879
119891by applying the formula
119878eff =119862tot119879119891
(35)
where 119879119891= 119903 times 119901 where 119903 is the resource usage factor and
119901 is total traffic 119862tot is number of available traffic channels inthe system
The queuing model is set up by considering 119883119905that
represents the number ofUE instances at a given time interval119905 isin 119879 and 119880
119905denotes the number of new UE instances
that arrive at time 119905 isin 119879 with common probability densityfunction 119891 on 119879 and we define
119883119905+1=
119880119905+1
119883119905= 0
(119883119905minus 1) + 119880
119905+1119883119905gt 0
119905 isin 119879 (36)
where 119883 = 1198830 1198831 is a Markov Chain that has the
probability matrix of the following form
119875 (0 119910) = 119891 (119910) 119910 isin 119879
119875 (119910 119911) = 119891 (119910 minus 119911 + 1)
119911 isin 119879 119910 isin (119911 minus 1 119911 119911 + 1 )
(37)
The chain119883 defined here is the queuing chain of distribu-tion 119891 and starting from 119910 isin 119879 a UE instance is served andnew UE instances arrive by the next time unit 119905 controlled by119891 So the probability of going from state 119910 to 119911 isin (119910minus1 119910 119910+1 ) is lfloor(119911 minus 119910 + 1)rfloor
32 Results Wewill now discuss the simulation results basedon the simulation model and proposed strategy In thispaper we have used the LTE-SIM [35] simulator availableonline at httptelematicspolibaitLTE-Sim It is crucial tocarry out performance evaluation for LTE-A systems and thisrequires the use of standard simulators LTE-Sim is designedaccording to the LTE-A standard with main features suchas multicell environment multiuser support eNodeB CQIUE resource allocation QoS management and schedulingstrategies
The network topology under consideration comprisesnetwork nodes (including eNodeBs and UE) distributedamong various cells We have tested our algorithms for both
Mobile Information Systems 9
MMEGW
MMEGW
Single-cell
eNBeNB
eNB
eNB
UE
UE
Multicell
EGW
eNBNBNBBBNBNBBBBBBBeNB
eNB
UE
UE
Figure 5 Single-cell and multicell simulation scenarios
5 10 15 20Users
DPSMOZ
0
200000
400000
600000
800000
1000000
1200000
Thro
ughp
ut (b
ps)
Figure 6 Video throughput
the single-cell and multicell scenario as shown in Figure 5The first scenario consists of many eNodeBs due to the factthat multiple cells are covered while in the second scenarioone eNodeB handles all UE
We will evaluate the performance of our proposed MOZwhile comparing it with that of DPS [20] For each UEdownlink VOIP and video flows are considered All flows areactive during the whole simulation Time for each simulationis around 100 sThe simulation is performedusing aWindowsmachine with Cygwin (Linux-like environment) a 26GHzCPU and 4GB RAM
321 Delay and Throughput The goal of achieving desiredlevel of QoS is very challenging due to Real Timemultimediaapplications which have strict delay and throughput con-straints Figures 6 and 7 show the throughput for video andVOIP flows with increasing number of UE instances Fromboth the figures it can be seen that the throughput achievedby MOZ is better than that of DPS algorithm althoughthe difference for VOIP flow is much less Throughput alsoincreases with increasing number of users For video up to15 users increasing pattern is observed and after 25 usersthroughput increases but gradually For VOIP flows thethroughput is still increasing although it is delay sensitiveThereason behind this increased throughput is that the systemresources are utilized according to the requirement of users
5 10 15 20Users
MOZDPS
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 7 VOIP throughput
5 10 15 20Users
DPSMOZ
00015
000155
00016
000165
00017
000175
00018
Del
ay (m
sec)
Figure 8 VOIP delay
and only the selected UE instances transmit on the best RBsAlso it is worth noticing that the throughput achieved byMOZ is much closer to the theoretical maximum In thisprocess decisions are made based on the channel quality ofeach UE which constantly measures the QoS parametersbased on the algorithms and techniques proposed After10 users the throughput for video using MOZ achievesperformance gains in order of 1 to 15 while for VOIP it isfrom 1 to 10
Figures 8 and 9 show the average delay of the MOZand DPS algorithm with increasing UE for both the videoand VOIP flows From Figure 8 it can be seen that thedelay performance of MOZ is slightly higher compared toDPS with less than 10 users and as number of UE instancesincreases MOZ shows better performance compared to theDPS algorithm because after 10 plus users there is morevideo streaming and it is not distinguishing VOIP and videoflows so the delay for DPS will increase in consecutive TTIsFigure 9 shows the performance of bothMOZandDPSwhichis very close because with increasing users video demand hasa possibility of rising
Also our results ensure that the delay always remainsbelow the threshold value The proposed MOZ achieves
10 Mobile Information Systems
DPSMOZ
5 10 15 20Users
0
002
004
006
008
01
012
014
Del
ay (m
sec)
Figure 9 Video delay
DPSMOZ
5 10 15 20Users
00280029
0030031003200330034003500360037
PLR
Figure 10 Packet loss ratio
performance gains by reducing delay from 1 to 5 for videoand 1 to 8 for VOIP after 10 users Also it is worth noticinghere that the delay achieved is way below the threshold whichwas set to 20ms
322 PLR Now let us compare the PLR performance ofMOZ and DPS algorithm shown in Figure 10 As expectedthe PLR is increasing as the number of users increasesbecause there are more packets that will be discarded as therewill not be sufficient RBs for transmitting all the packetsconsidering that the packets are nearing their delay thresholdHowever in comparison with the DPS algorithm MOZ iscapable of achieving a better PLR as it supports more usersby distinguishing packets according to their scheduling andresource needs It reduces the loss probability by exploitingthe gains at physical layer Note that the PLR counts only forphysical layer losses MOZ achieves a better PLR and packetloss is reduced from 2 to 14 as compared to the referencealgorithm
323 SNR Figure 11 shows the SNR performance of MOZwith reference to average system throughputWith increasing
5 10 15 20SNR (db)
MOZ
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 11 SNR versus throughput
10 20 30 40SNR (db)
MOZ
021
022
023
024
025
026
027
028
Spec
tral
effici
ency
(bps
)
Figure 12 Cell spectral efficiency
throughput it is observed that the SNR is also increasingwhich depicts better channel quality and less distortion andis one of the reasons we are able to achieve higher systemthroughputThe higher values of SNR are indicator of higherpercentage of data as compared to noise and our resultsshow that noise is around 004 of the actual signal whichis an indicator that the video and voice quality achieved areexcellent
324 Spectral Efficiency Then the spectral efficiency (inbitss) is computed and is depicted in Figure 12 At eachscheduling interval the RB is allocated to UE instancesaccording to their needs As the number of UE instancesincreasesMOZ still guarantees QoS constraints tomore flowswith a positive effect on overall system efficiency The systemsupports delay and throughput guarantees simultaneouslytherefore the spectral efficiency increases with increasingusers which is a good indicator of overall system gain andefficiency
325 Energy and Memory Consumption Figure 13 demon-strates the energy consumption in joules with respect to thenumber of users in the network The energy usage of MOZis compared to PF scheduling and DPS algorithm As the
Mobile Information Systems 11
5 10 15 20Users
DPSMOZPF
0
05
1
15
2
25
Ener
gy (j
oule
s)
Figure 13 Energy consumption versus no of nodes
5 10 15 20Users
MOZ
0
50
100
150
200
250
Mem
ory
Figure 14 Memory consumption versus no of nodes
number of users increases the level of energy consumptionalso rises for all algorithms As the eNodeB remains active allthe time this rise is expected But the energy consumptionlevel achieved by MOZ is the lowest due to the fact thatefficient RB selection takes place which reduces the delaysthus lowering levels of energy consumption Also only theoptimal solution is selected which in turn results in energygains and network lifetime
In Figure 14 the memory consumption in bits versus thenumber of users is depictedThe DM takes decision based onthe information it has and the solution depends on algorithmsthat run dynamically and this process is continuous untilthere are no more RBs to allocate Moreover it has beenrealized that the continuous assignment of RBs accordingto UE instancesrsquo needs utilizes less memory which in turnincreases network life time and performance and reducesdelays
326 PRB Utilization In this subsection we will show effectof the proposed algorithm on PRB utilization The scenariodepicts results according to two groups In the first groupsimulations are carried out without using the proposedMOZthe second group implements MOZ The performance is
5 10 15 20 25 30Video users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 15 Mean PRBs used for video
5 10 15 20 25 30VOIP users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 16 Mean PRBs used for VOIP
measured in terms of mean PRBs being used for both videoandVOIP users In high load the delay increases and a highernumber of PRBs are used when we do not employ the MOZalgorithm As a result after a specific limit no more requestscan be fulfilled because all PRBs are utilized Figure 15 depictsthis scenario as we can see that the average number of PRBsused withMOZ is less due to improved bandwidth utilizationstrategy better coverage and enhanced SNR But with higherload of 30 UE instances all the bandwidth is used up
In Figure 16 for VOIP users also the average number ofPRBs used is much less with increasing number of users
33 Summary of the Simulation Analysis Table 3 representsthe results and statistics obtained by using MOZ where 1 to 4represent the PO set The optimal solution 119874lowast for each set ofusers is bold
It can be clearly seen that the DM always selects the bestfeasible solution thus proving an accuracy of 100 in termsof optimal solution selection The optimization proceduregenerates a PO set for varying number of users and out ofthis set the DM selects the best solution
12 Mobile Information Systems
Table 3 Statistics of the PO set and the optimal solution Olowast selected using MOZ
Number of users PO set (average delay in ms) PO set (average throughput in bps)1 2 3 4 1 2 3 4
5 000252 000328 000173 000175 40722208000 66418738666 50126800000 3949574800010 000179 000197 000252 000328 1509846667 2871940000 4357810667 585808666715 000161 000167 000169 000175 40821598667 76296266667 93016094667 9334639733320 000161 000167 000159 000155 40821541334 76643476000 97387246667 101596778667
The evaluation and comparison of the simulation resultsshow that MOZ performs very well for both voice andVOIP scenarios and performs better than the referencealgorithm in terms of maximum achievable throughput andminimum delay The reason behind these results is the useof optimization process employed by the DM which alwaysselects the optimal solution for the arriving UE instancesbased on their QoS needs and priorities and channel qualityIt further optimizes the procedure by maintaining queues toimprove performance and achieves higher SNR and lowerPLRThe cell spectral efficiency increaseswith each addedUEthat indicates improved system performance and is a proof ofrobust system design
4 Conclusions
In this paper we have proposed a multiobjective optimizerthat helps in finding an optimal solution for selecting RBsfor transmission based on the information provided to thedecision maker It provides an optimal solution for RRM toensure QoS by incorporating themultiobjective optimizationtechniques that target to achieve higher throughput and lowerdelay A set of Pareto optimal solutions is originated and isforwarded to the DM which chooses the best solution outof this set We have adopted an interactive technique thatembeds the multiobjective optimization with the moderntechniques for RRM Furthermore the results are also pro-vided that prove the efficiency of our proposed MOZ Thistechnique is innovative as it makes use of evolutionary MOOtechniques (which are very powerful for finding optimalsolution) and provides its implementation for LTE-A and willprove helpful in solving other issues for 4G and emerging 5Gnetworks
In future we aspire to investigate the opportunities forimplementingMOO techniques for the evolving 5Gnetworksthat not only aim to deal with QoS issues but also target toachieve optimal energy consumption
Competing Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] S-B Lee I Pefkianakis A Meyerson S Xu and S LuldquoProportional fair frequency-domain packet scheduling for3GPP LTE uplinkrdquo in Proceedings of the 28th Conference on
Computer Communications (IEEE INFOCOM rsquo09) pp 2611ndash2615 Rio de Janeiro Brazil April 2009
[2] ldquoPhysical layer aspects for evolved UTRArdquo 3GPP TechnicalReport 25814 version 710 2006
[3] M Nazir and F Saleemi ldquoCooperative cognitive ecology in self-organizing networks a review articlerdquo International Journal ofComputer Science and Information Security vol 10 no 2 pp159ndash167 2012
[4] K Deb Multi-Objective Optimization with Evolutionary Algo-rithms John Wiley amp Sons New York NY USA 1st edition2001
[5] J B K Deb and K M R SłowinskiMulti-Objective Optimiza-tion Interactive and Evolutionary Approaches Springer BerlinGermany 2008
[6] A De Domenico and E C Strinati ldquoA radio resource man-agement scheduling algorithm for self-organizing femtocellsrdquoin Proceedings of the IEEE 21st International Symposium onPersonal Indoor and Mobile Radio Communications Workshops(PIMRC rsquo10) pp 191ndash196 IEEE Instanbul Turkey September2010
[7] R Kausar Y Chen K K Chai L Cuthbert and J SchormansldquoQoS aware mixed traffic packet scheduling in OFDMA-basedLTE-advanced networksrdquo in Proceedings of the 4th InternationalConference on Mobile Ubiquitous Computing Systems Servicesand Technologies (UBICOMM rsquo10) pp 53ndash58 Florence FranceOctober 2010
[8] Y J Zhang and K B Letaief ldquoAdaptive resource allocation andscheduling for multiuser packet-based OFDM networksrdquo inProceedings of the IEEE International Conference on Communi-cation vol 5 pp 2949ndash2953 June 2004
[9] C Stocchi N Marchetti and N R Prasad ldquoSelf-optimizedradio resource management techniques for LTE-a local areadeploymentsrdquo inProceedings of the 2nd International Conferenceon Wireless Communication Vehicular Technology InformationTheory and Aerospace amp Electronic Systems Technology (WirelessVITAE rsquo11) pp 1ndash5 Chennai India March 2011
[10] M Cenk Erturk H Aki I Guvenc and H Arslan ldquoFairand QoS-oriented spectrum splitting in macrocell-femtocellnetworksrdquo in Proceedings of the 53rd IEEE Global Communica-tions Conference (GLOBECOM rsquo10) pp 1ndash6 Miami Fla USADecember 2010
[11] V Chandrasekhar and J G Andrews ldquoSpectrum allocation intiered cellular networksrdquo IEEE Transactions on Communica-tions vol 57 no 10 pp 3059ndash3068 2009
[12] A Roy and S K Das Optimizing QoS-Based Multicast Routingin Wireless Networks A Multi-Objective Genetic AlgorithmicApproach Springer New Yok NY USA 2002
[13] D H Lorenz and A Orda ldquoQoS routing in networks withuncertain parametersrdquo IEEEACM Transactions on Networkingvol 6 no 6 pp 768ndash778 1998
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mobile Information Systems 5
(ii) Second we consider modulation of 64 QAM and asingle modulation symbol carries 6 bits The total bits willthus be
16800
times 6 bits per symbol of modulation equal to 1008Mb(12)
(iii) Thirdly considering the 4-by-4 MIMO we get
4 times 1008Mb = 403Mb (13)
This is the peak data rate(iv) Finally we calculate the overhead which will be about
25 So we have
403Mb times 075 = 302Mb (14)
So we can calculate TA119904and RC and minimum 119879min119894 is
selected by the DM for transmission That is resource chunkRC119894as in the following
119874lowast= RC119894= min119879min119894 forall119894 isin UE (15)
Now let us talk about the delay sensitive applicationswhere we will find the feasible solution 119874lowast as follows
119863max119894 = DB119894minus 119863119894(119905)
(For delay sensitive applications)
119874lowast= RC119894= max119863max119894 forall119894 isin UE
(16)
So 119874lowast is chosen for transmission based on the followingdecision vectors
119874lowast=
argmin 119879min119894 forallUE119894
argmax 119863max119894 forallUE119894
(17)
The DM will take its decision based on the following setof constraints
119899
sum
119894=1
119879119894lt 119862119894119896(119905) where 119879
119894gt 0 forall119894 isin 119873 (18)
119863119894(119905) lt DB
119894forall119894 isin 119873 (19)
119873
sum
119894=1
RB119894le Total Badndwidth (20)
sum
119895isin119869
120579119894119895ge 1 forall119894 isin 119873 forall119869 isin Total BS (21)
Equation (18) refers to channel capacity constraint and(19) refers to delay bound constraint Equation (20) is RBbandwidth constraint Equation (21) ensures that at least 1 BScovers the active UE
22 The Decision-Making Process After the optimized solu-tion is sought out by theDM the trafficwill then be scheduledin output queues discussed in detail in the next subsection
Figure 3 depicts the detail of decision-making process insidethe DM Its details are explained as follows
(1) Traffic Type Determination The DM will take as input thePareto front and will first determine the type of traffic whichin our case is either throughput or delay sensitive
(2) Optimization Process After determining the traffic typethe following actions are taken
(i) For throughput sensitive applications it will calculatethe value of 119879min119894 based on the value of 119879
119894 and then
it will select the optimal solution 119874lowast = RC119894=
min119879min119894(ii) For delay sensitive applications it will calculate the
value of 119863max119894 based on the value of 119863119894(119905) and then
it will select the optimal solution 119874lowast = RC119894=
max119863max119894(3) RC Allocation Matrix Update After that the RC assign-ment matrix will be updated The matrix 119898
119896119899is represented
as
119898119896119899=
0 if RC119899is unassigned
1 if RC119899is assigned to service group 119896
(22)
where 119896 is type of traffic and 119899 is RC number [34]
(4) RC Availability Check If there are RCs availableleft forassignment then the process starts again andmoves to traffictype determination otherwise the process ends
In the following we explain in detail the construction ofPareto front
23 Constructing Pareto Front This step describes in detailthe process of constructing a representative Pareto front asfollows
Step 1 Determine the traffic type of the incoming packetAssume 119863
1 119863
119899are the alternative solutions in terms of
119894 and 119895 where 119894 represents throughput sensitive flows and 119895represents delay sensitive flows
Step 2 Find the largest 119894 such that 119863119894gt 119863119899(where 119863
119899
represents all other UE) and add119863119894to Pareto front
Step 3 Find the smallest 119895 such that 119863119895lt 119863119899(where 119863
119899
represents all other UE) and add119863119895to Pareto front
Step 4 Repeat Steps 2 and 3 until no such 119894 and 119895 exist and wehave the Pareto front of the form (119863
119894 119863
119899119863119895 119863
119899)
Note that 119863119894is sorted in order of decreasing throughput
and 119863119895is sorted in order of increasing delay The process
generates a set of candidate solutions that will undergo theoptimization phase for selecting a singled optimal solutionThe selection of the optimal solution is based on the efficiencyof the DM which not only considers the resource require-ments of the users but also sets priorities We now present analgorithm that will make use of the defined framework andconsiders the requirements of each step
6 Mobile Information Systems
Start
Throughput sensitive traffic Delay sensitive trafficDetermine from thePareto front the traffic
requirements
On the basis ofvarious values ofTi calculate Tmini
various values ofOn the basis of
Di(t) calculate Dmaxi
Choose optimalsolution based on
Olowast= RCi = minTmini
Choose optimalsolution based on
Olowast= RCi = maxDmaxi
Update RCallocation matrix
RCs left to be allocated Yes
No
End
Figure 3The decision-making process inside the DM for RC selection based on the objectives of maximum throughput andminimumdelay
24 The Multiobjective Optimizer Algorithm The detailedworking of proposed MOZ is outlined in Figure 4 TheMOZconsists of three main steps as follows
Step 1 The traffic classifier will take as input traffic fromvarious UE instances and will divide them into 2 queues oneconsists of packets that have stringent throughput require-ment and the other queue consists of packets from delaysensitive applications
Step 2 The information from Step 1 is then passed onto theoptimizer that will find out the Pareto front and will send itsdecision set to the decision maker
Step 3 The DM will make the final decision based on theinformation it has and according to its decision-makingcapabilities
Step 4 Finally119874lowast is sent out to the queue for scheduling andthe process then diverts to Step 2 to get the next 119874lowast later tobe scheduled
Step 5 The process is completed when either no RC is left orthere are no more incoming packets
Steps 2 and 3 are interrelated as the DM is way dependenton the input it gets from the Pareto front so constructing alegitimate Pareto front is itself a very important task forMOZThe decision-making process inside the DM is characterizedby a number of factors already explained and is one of themajor components of our proposed strategy
Compared to most algorithms and frameworks designedto ensure QoS and to manage RRM the distinct differencesof our proposed approach are as follows (1) it makes useof multiobjective optimization that deals with conflictingobjectives simultaneously instead of converting them intoa single objective (2) It makes use of a decision makerthat has the decision-making capability and works basedon the information it has and various algorithms used(3) Multiple UE instances are served based on the type ofincoming trafficflow instead of allocating RBs to just oneflow
Mobile Information Systems 7
Trafficclassification Optimization Decision-making Scheduling
Input trafficfrom variousUE instances
applicationsDelay sensitive
Throughput sensitiveapplications
Theoptimizationprocess todeterminethe Paretofront
DetermineO
lowast (theoptimalsolution)
Figure 4 The proposed MOZ
The goal of using MOZ is to always get an optimalsolution which is chosen from a set of solutions in the formof Pareto front
25 Queuing and Scheduling Now that we have the Paretooptimal solution in hand the packets are now ready tobe scheduled in their respective queues for transmission asfollows
Step 1 In this step we will define the priority metric for delaysensitive flows The priority of UE
119894at time 119905 119875
119889119894(119905) is
119875119889119894(119905) = 119863max119894 lowast 119866119894119895 (23)
where 119866119894119895is channel gain of UE
119894on RC
119895calculated as
119866119894119895= 10
Path loss10lowast 10
shadow fading10
lowast 10multipath fading10
(24)
Step 2 We will then form a delay queue for a given UE119894that
will ensure delay bounds as follows
119863119894(119905 + 1) = [119863
119894(119905) minus 120588119877
119894(119905)] +
119873
sum
119894=1
119876119894(119905) (25)
where 119876119894(119905) is the mean queue length at time interval 119905 and
119873 is total UE scheduled and 120588119877119894(119905) is the service rate In
each time interval the UEwith highest priority occupies frontposition in the queue
Step 3 For the priority metric of throughput sensitive flowswe will use the following
119875119905119894(119905) = 120590
119894lowast 119879min (119894) (26)
where 120590119894is the Head of Line (HOL) blocking and 119875
119905119894(119905) is the
priority metric for throughput sensitive applications In eachtime interval the UE instances are scheduled according to theascending order of their priority values
It follows the following constraints
119876119894119896
st forall119894 isin 119873 119896 isin delay sensitive traffic only (27)
119876119894119895
st forall119894 isin 119873 and 119895
isin throughput sensitive traffic only(28)
119894 = 119895 Queue Length
= Length (119876119894119896+ 119876119894119895)
(29)
Equation (27) ensures that the queue 119876119894takes as input
delay sensitive traffic only (28) ensures that the queue 119876119894119895
takes in throughput sensitive traffic only and (29) states thatqueue length is the total of both the queues
Thus the defined framework and the optimizer will worktogether to form our proposed model The whole process hasfour main goals
(1) Determining the target objectives(2) Based on the objectives forming a system model that
will take into account the objective functions to beminimized or maximized
(3) Constructing a representative solution pool to aid thedecision maker
(4) Introducing a decision-making entity that will notonly ease the task of scheduling but also find anoptimal solution to satisfy the defined objectives
In the following section we will define the simulationmodel and will discuss the results
3 Simulation and Results
31 Simulation Model Designing the model for LTE-A per-formance evaluation is generally more complex comparedto a 3G system An interface is used between variousevolvedNodeBs (eNBs ie the BS) in the LTE system toease the traffic load that occurs due to handovers In this
8 Mobile Information Systems
Table 2 Simulation parameters
Parameters ValuesCarrier frequency 2GHzBandwidth 10MHzNumber of subcarriers 300Number of RBs 45TTI 1msPath loss model Free space propagation modelTransmit power 30mWFading model RayleighPacket model Poisson modelMultiplexing MIMONumber of UE instances 20
section we will design the simulation model for testing andwill evaluate our system for examining delay throughputPacket Loss Ratio (PLR) the effect of SNR and cell spectralefficiency
We are considering a cell with the capacity of 10MHz andit consists of 45 RBs and the carrier frequency is 2GHz Allthe available RBs are controlled by the eNodeB The RBs aredivided among theUE instances in the formof RCsThe LTE-A parameters used for simulation are described in Table 2
We then calculate the number of bits in each RB (ie itsdata rate) It can be computed as follows
119887119894RC = NStti lowastNS119904 lowast 119878rb (30)
where 119887119894RC is achievable data rate NStti is number of slots in
each TTI NS119904is number of symbols in each slot and 119878rb is
number of subcarriers in each RBEach UE is assigned a queue according to the traffic type
and parameters as described in the system model previouslyAfter that we will calculate the channel gain of each UE
119894on
RC and we will now proceeed to calculate the SNRThe SNR values are reported to ENodeB by theUE at each
TTI All the users have a different value of SNR accordingto the experienced fading and multipath propagation thetransmit power (119875) and the total RBs (119873) as follows
SNR =119875 lowast 119866
119894119895
119873(Noise + Interference) (31)
The incoming packets will be treated by the eNodeBaccording to the parameters defined in Table 2 and thesystem model presented All packets have varying lengthand are streamed in queues according to their priorities andproperties Now for the delay sensitive packets the thresholdDB119894is set to be 20ms for each UEThus we can define the Average System Delay (AvD) with
a total simulation time of 119879 at time instance 119905 with totalnumber of UE instances equal to119873 as follows
AvD = 1119879
119879
sum
119905=1
1
119873
119873
sum
119894=1
119863119894(119905) (32)
We will then define AvTH which is the average through-put with trans
119894(119905) being the number of packetssec as follows
AvTH = 1119879
119873
sum
119894=1
119879
sum
119905=1
trans119894(119905) (33)
Now for evaluating the PLR by using the size of discardedpackets pdisc
119894(119905) and the sum of sizes of all packets psize
119894(119905)
we have
PLR =sum119873
119894=1sum119879
119905=1pdisc119894(119905)
sum119873
119894=1sum119879
119905=1psize119894(119905)
(34)
Theoverall spectrumefficiency (119878eff ) ismeasured in termsof total traffic usage factor 119879
119891by applying the formula
119878eff =119862tot119879119891
(35)
where 119879119891= 119903 times 119901 where 119903 is the resource usage factor and
119901 is total traffic 119862tot is number of available traffic channels inthe system
The queuing model is set up by considering 119883119905that
represents the number ofUE instances at a given time interval119905 isin 119879 and 119880
119905denotes the number of new UE instances
that arrive at time 119905 isin 119879 with common probability densityfunction 119891 on 119879 and we define
119883119905+1=
119880119905+1
119883119905= 0
(119883119905minus 1) + 119880
119905+1119883119905gt 0
119905 isin 119879 (36)
where 119883 = 1198830 1198831 is a Markov Chain that has the
probability matrix of the following form
119875 (0 119910) = 119891 (119910) 119910 isin 119879
119875 (119910 119911) = 119891 (119910 minus 119911 + 1)
119911 isin 119879 119910 isin (119911 minus 1 119911 119911 + 1 )
(37)
The chain119883 defined here is the queuing chain of distribu-tion 119891 and starting from 119910 isin 119879 a UE instance is served andnew UE instances arrive by the next time unit 119905 controlled by119891 So the probability of going from state 119910 to 119911 isin (119910minus1 119910 119910+1 ) is lfloor(119911 minus 119910 + 1)rfloor
32 Results Wewill now discuss the simulation results basedon the simulation model and proposed strategy In thispaper we have used the LTE-SIM [35] simulator availableonline at httptelematicspolibaitLTE-Sim It is crucial tocarry out performance evaluation for LTE-A systems and thisrequires the use of standard simulators LTE-Sim is designedaccording to the LTE-A standard with main features suchas multicell environment multiuser support eNodeB CQIUE resource allocation QoS management and schedulingstrategies
The network topology under consideration comprisesnetwork nodes (including eNodeBs and UE) distributedamong various cells We have tested our algorithms for both
Mobile Information Systems 9
MMEGW
MMEGW
Single-cell
eNBeNB
eNB
eNB
UE
UE
Multicell
EGW
eNBNBNBBBNBNBBBBBBBeNB
eNB
UE
UE
Figure 5 Single-cell and multicell simulation scenarios
5 10 15 20Users
DPSMOZ
0
200000
400000
600000
800000
1000000
1200000
Thro
ughp
ut (b
ps)
Figure 6 Video throughput
the single-cell and multicell scenario as shown in Figure 5The first scenario consists of many eNodeBs due to the factthat multiple cells are covered while in the second scenarioone eNodeB handles all UE
We will evaluate the performance of our proposed MOZwhile comparing it with that of DPS [20] For each UEdownlink VOIP and video flows are considered All flows areactive during the whole simulation Time for each simulationis around 100 sThe simulation is performedusing aWindowsmachine with Cygwin (Linux-like environment) a 26GHzCPU and 4GB RAM
321 Delay and Throughput The goal of achieving desiredlevel of QoS is very challenging due to Real Timemultimediaapplications which have strict delay and throughput con-straints Figures 6 and 7 show the throughput for video andVOIP flows with increasing number of UE instances Fromboth the figures it can be seen that the throughput achievedby MOZ is better than that of DPS algorithm althoughthe difference for VOIP flow is much less Throughput alsoincreases with increasing number of users For video up to15 users increasing pattern is observed and after 25 usersthroughput increases but gradually For VOIP flows thethroughput is still increasing although it is delay sensitiveThereason behind this increased throughput is that the systemresources are utilized according to the requirement of users
5 10 15 20Users
MOZDPS
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 7 VOIP throughput
5 10 15 20Users
DPSMOZ
00015
000155
00016
000165
00017
000175
00018
Del
ay (m
sec)
Figure 8 VOIP delay
and only the selected UE instances transmit on the best RBsAlso it is worth noticing that the throughput achieved byMOZ is much closer to the theoretical maximum In thisprocess decisions are made based on the channel quality ofeach UE which constantly measures the QoS parametersbased on the algorithms and techniques proposed After10 users the throughput for video using MOZ achievesperformance gains in order of 1 to 15 while for VOIP it isfrom 1 to 10
Figures 8 and 9 show the average delay of the MOZand DPS algorithm with increasing UE for both the videoand VOIP flows From Figure 8 it can be seen that thedelay performance of MOZ is slightly higher compared toDPS with less than 10 users and as number of UE instancesincreases MOZ shows better performance compared to theDPS algorithm because after 10 plus users there is morevideo streaming and it is not distinguishing VOIP and videoflows so the delay for DPS will increase in consecutive TTIsFigure 9 shows the performance of bothMOZandDPSwhichis very close because with increasing users video demand hasa possibility of rising
Also our results ensure that the delay always remainsbelow the threshold value The proposed MOZ achieves
10 Mobile Information Systems
DPSMOZ
5 10 15 20Users
0
002
004
006
008
01
012
014
Del
ay (m
sec)
Figure 9 Video delay
DPSMOZ
5 10 15 20Users
00280029
0030031003200330034003500360037
PLR
Figure 10 Packet loss ratio
performance gains by reducing delay from 1 to 5 for videoand 1 to 8 for VOIP after 10 users Also it is worth noticinghere that the delay achieved is way below the threshold whichwas set to 20ms
322 PLR Now let us compare the PLR performance ofMOZ and DPS algorithm shown in Figure 10 As expectedthe PLR is increasing as the number of users increasesbecause there are more packets that will be discarded as therewill not be sufficient RBs for transmitting all the packetsconsidering that the packets are nearing their delay thresholdHowever in comparison with the DPS algorithm MOZ iscapable of achieving a better PLR as it supports more usersby distinguishing packets according to their scheduling andresource needs It reduces the loss probability by exploitingthe gains at physical layer Note that the PLR counts only forphysical layer losses MOZ achieves a better PLR and packetloss is reduced from 2 to 14 as compared to the referencealgorithm
323 SNR Figure 11 shows the SNR performance of MOZwith reference to average system throughputWith increasing
5 10 15 20SNR (db)
MOZ
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 11 SNR versus throughput
10 20 30 40SNR (db)
MOZ
021
022
023
024
025
026
027
028
Spec
tral
effici
ency
(bps
)
Figure 12 Cell spectral efficiency
throughput it is observed that the SNR is also increasingwhich depicts better channel quality and less distortion andis one of the reasons we are able to achieve higher systemthroughputThe higher values of SNR are indicator of higherpercentage of data as compared to noise and our resultsshow that noise is around 004 of the actual signal whichis an indicator that the video and voice quality achieved areexcellent
324 Spectral Efficiency Then the spectral efficiency (inbitss) is computed and is depicted in Figure 12 At eachscheduling interval the RB is allocated to UE instancesaccording to their needs As the number of UE instancesincreasesMOZ still guarantees QoS constraints tomore flowswith a positive effect on overall system efficiency The systemsupports delay and throughput guarantees simultaneouslytherefore the spectral efficiency increases with increasingusers which is a good indicator of overall system gain andefficiency
325 Energy and Memory Consumption Figure 13 demon-strates the energy consumption in joules with respect to thenumber of users in the network The energy usage of MOZis compared to PF scheduling and DPS algorithm As the
Mobile Information Systems 11
5 10 15 20Users
DPSMOZPF
0
05
1
15
2
25
Ener
gy (j
oule
s)
Figure 13 Energy consumption versus no of nodes
5 10 15 20Users
MOZ
0
50
100
150
200
250
Mem
ory
Figure 14 Memory consumption versus no of nodes
number of users increases the level of energy consumptionalso rises for all algorithms As the eNodeB remains active allthe time this rise is expected But the energy consumptionlevel achieved by MOZ is the lowest due to the fact thatefficient RB selection takes place which reduces the delaysthus lowering levels of energy consumption Also only theoptimal solution is selected which in turn results in energygains and network lifetime
In Figure 14 the memory consumption in bits versus thenumber of users is depictedThe DM takes decision based onthe information it has and the solution depends on algorithmsthat run dynamically and this process is continuous untilthere are no more RBs to allocate Moreover it has beenrealized that the continuous assignment of RBs accordingto UE instancesrsquo needs utilizes less memory which in turnincreases network life time and performance and reducesdelays
326 PRB Utilization In this subsection we will show effectof the proposed algorithm on PRB utilization The scenariodepicts results according to two groups In the first groupsimulations are carried out without using the proposedMOZthe second group implements MOZ The performance is
5 10 15 20 25 30Video users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 15 Mean PRBs used for video
5 10 15 20 25 30VOIP users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 16 Mean PRBs used for VOIP
measured in terms of mean PRBs being used for both videoandVOIP users In high load the delay increases and a highernumber of PRBs are used when we do not employ the MOZalgorithm As a result after a specific limit no more requestscan be fulfilled because all PRBs are utilized Figure 15 depictsthis scenario as we can see that the average number of PRBsused withMOZ is less due to improved bandwidth utilizationstrategy better coverage and enhanced SNR But with higherload of 30 UE instances all the bandwidth is used up
In Figure 16 for VOIP users also the average number ofPRBs used is much less with increasing number of users
33 Summary of the Simulation Analysis Table 3 representsthe results and statistics obtained by using MOZ where 1 to 4represent the PO set The optimal solution 119874lowast for each set ofusers is bold
It can be clearly seen that the DM always selects the bestfeasible solution thus proving an accuracy of 100 in termsof optimal solution selection The optimization proceduregenerates a PO set for varying number of users and out ofthis set the DM selects the best solution
12 Mobile Information Systems
Table 3 Statistics of the PO set and the optimal solution Olowast selected using MOZ
Number of users PO set (average delay in ms) PO set (average throughput in bps)1 2 3 4 1 2 3 4
5 000252 000328 000173 000175 40722208000 66418738666 50126800000 3949574800010 000179 000197 000252 000328 1509846667 2871940000 4357810667 585808666715 000161 000167 000169 000175 40821598667 76296266667 93016094667 9334639733320 000161 000167 000159 000155 40821541334 76643476000 97387246667 101596778667
The evaluation and comparison of the simulation resultsshow that MOZ performs very well for both voice andVOIP scenarios and performs better than the referencealgorithm in terms of maximum achievable throughput andminimum delay The reason behind these results is the useof optimization process employed by the DM which alwaysselects the optimal solution for the arriving UE instancesbased on their QoS needs and priorities and channel qualityIt further optimizes the procedure by maintaining queues toimprove performance and achieves higher SNR and lowerPLRThe cell spectral efficiency increaseswith each addedUEthat indicates improved system performance and is a proof ofrobust system design
4 Conclusions
In this paper we have proposed a multiobjective optimizerthat helps in finding an optimal solution for selecting RBsfor transmission based on the information provided to thedecision maker It provides an optimal solution for RRM toensure QoS by incorporating themultiobjective optimizationtechniques that target to achieve higher throughput and lowerdelay A set of Pareto optimal solutions is originated and isforwarded to the DM which chooses the best solution outof this set We have adopted an interactive technique thatembeds the multiobjective optimization with the moderntechniques for RRM Furthermore the results are also pro-vided that prove the efficiency of our proposed MOZ Thistechnique is innovative as it makes use of evolutionary MOOtechniques (which are very powerful for finding optimalsolution) and provides its implementation for LTE-A and willprove helpful in solving other issues for 4G and emerging 5Gnetworks
In future we aspire to investigate the opportunities forimplementingMOO techniques for the evolving 5Gnetworksthat not only aim to deal with QoS issues but also target toachieve optimal energy consumption
Competing Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] S-B Lee I Pefkianakis A Meyerson S Xu and S LuldquoProportional fair frequency-domain packet scheduling for3GPP LTE uplinkrdquo in Proceedings of the 28th Conference on
Computer Communications (IEEE INFOCOM rsquo09) pp 2611ndash2615 Rio de Janeiro Brazil April 2009
[2] ldquoPhysical layer aspects for evolved UTRArdquo 3GPP TechnicalReport 25814 version 710 2006
[3] M Nazir and F Saleemi ldquoCooperative cognitive ecology in self-organizing networks a review articlerdquo International Journal ofComputer Science and Information Security vol 10 no 2 pp159ndash167 2012
[4] K Deb Multi-Objective Optimization with Evolutionary Algo-rithms John Wiley amp Sons New York NY USA 1st edition2001
[5] J B K Deb and K M R SłowinskiMulti-Objective Optimiza-tion Interactive and Evolutionary Approaches Springer BerlinGermany 2008
[6] A De Domenico and E C Strinati ldquoA radio resource man-agement scheduling algorithm for self-organizing femtocellsrdquoin Proceedings of the IEEE 21st International Symposium onPersonal Indoor and Mobile Radio Communications Workshops(PIMRC rsquo10) pp 191ndash196 IEEE Instanbul Turkey September2010
[7] R Kausar Y Chen K K Chai L Cuthbert and J SchormansldquoQoS aware mixed traffic packet scheduling in OFDMA-basedLTE-advanced networksrdquo in Proceedings of the 4th InternationalConference on Mobile Ubiquitous Computing Systems Servicesand Technologies (UBICOMM rsquo10) pp 53ndash58 Florence FranceOctober 2010
[8] Y J Zhang and K B Letaief ldquoAdaptive resource allocation andscheduling for multiuser packet-based OFDM networksrdquo inProceedings of the IEEE International Conference on Communi-cation vol 5 pp 2949ndash2953 June 2004
[9] C Stocchi N Marchetti and N R Prasad ldquoSelf-optimizedradio resource management techniques for LTE-a local areadeploymentsrdquo inProceedings of the 2nd International Conferenceon Wireless Communication Vehicular Technology InformationTheory and Aerospace amp Electronic Systems Technology (WirelessVITAE rsquo11) pp 1ndash5 Chennai India March 2011
[10] M Cenk Erturk H Aki I Guvenc and H Arslan ldquoFairand QoS-oriented spectrum splitting in macrocell-femtocellnetworksrdquo in Proceedings of the 53rd IEEE Global Communica-tions Conference (GLOBECOM rsquo10) pp 1ndash6 Miami Fla USADecember 2010
[11] V Chandrasekhar and J G Andrews ldquoSpectrum allocation intiered cellular networksrdquo IEEE Transactions on Communica-tions vol 57 no 10 pp 3059ndash3068 2009
[12] A Roy and S K Das Optimizing QoS-Based Multicast Routingin Wireless Networks A Multi-Objective Genetic AlgorithmicApproach Springer New Yok NY USA 2002
[13] D H Lorenz and A Orda ldquoQoS routing in networks withuncertain parametersrdquo IEEEACM Transactions on Networkingvol 6 no 6 pp 768ndash778 1998
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
6 Mobile Information Systems
Start
Throughput sensitive traffic Delay sensitive trafficDetermine from thePareto front the traffic
requirements
On the basis ofvarious values ofTi calculate Tmini
various values ofOn the basis of
Di(t) calculate Dmaxi
Choose optimalsolution based on
Olowast= RCi = minTmini
Choose optimalsolution based on
Olowast= RCi = maxDmaxi
Update RCallocation matrix
RCs left to be allocated Yes
No
End
Figure 3The decision-making process inside the DM for RC selection based on the objectives of maximum throughput andminimumdelay
24 The Multiobjective Optimizer Algorithm The detailedworking of proposed MOZ is outlined in Figure 4 TheMOZconsists of three main steps as follows
Step 1 The traffic classifier will take as input traffic fromvarious UE instances and will divide them into 2 queues oneconsists of packets that have stringent throughput require-ment and the other queue consists of packets from delaysensitive applications
Step 2 The information from Step 1 is then passed onto theoptimizer that will find out the Pareto front and will send itsdecision set to the decision maker
Step 3 The DM will make the final decision based on theinformation it has and according to its decision-makingcapabilities
Step 4 Finally119874lowast is sent out to the queue for scheduling andthe process then diverts to Step 2 to get the next 119874lowast later tobe scheduled
Step 5 The process is completed when either no RC is left orthere are no more incoming packets
Steps 2 and 3 are interrelated as the DM is way dependenton the input it gets from the Pareto front so constructing alegitimate Pareto front is itself a very important task forMOZThe decision-making process inside the DM is characterizedby a number of factors already explained and is one of themajor components of our proposed strategy
Compared to most algorithms and frameworks designedto ensure QoS and to manage RRM the distinct differencesof our proposed approach are as follows (1) it makes useof multiobjective optimization that deals with conflictingobjectives simultaneously instead of converting them intoa single objective (2) It makes use of a decision makerthat has the decision-making capability and works basedon the information it has and various algorithms used(3) Multiple UE instances are served based on the type ofincoming trafficflow instead of allocating RBs to just oneflow
Mobile Information Systems 7
Trafficclassification Optimization Decision-making Scheduling
Input trafficfrom variousUE instances
applicationsDelay sensitive
Throughput sensitiveapplications
Theoptimizationprocess todeterminethe Paretofront
DetermineO
lowast (theoptimalsolution)
Figure 4 The proposed MOZ
The goal of using MOZ is to always get an optimalsolution which is chosen from a set of solutions in the formof Pareto front
25 Queuing and Scheduling Now that we have the Paretooptimal solution in hand the packets are now ready tobe scheduled in their respective queues for transmission asfollows
Step 1 In this step we will define the priority metric for delaysensitive flows The priority of UE
119894at time 119905 119875
119889119894(119905) is
119875119889119894(119905) = 119863max119894 lowast 119866119894119895 (23)
where 119866119894119895is channel gain of UE
119894on RC
119895calculated as
119866119894119895= 10
Path loss10lowast 10
shadow fading10
lowast 10multipath fading10
(24)
Step 2 We will then form a delay queue for a given UE119894that
will ensure delay bounds as follows
119863119894(119905 + 1) = [119863
119894(119905) minus 120588119877
119894(119905)] +
119873
sum
119894=1
119876119894(119905) (25)
where 119876119894(119905) is the mean queue length at time interval 119905 and
119873 is total UE scheduled and 120588119877119894(119905) is the service rate In
each time interval the UEwith highest priority occupies frontposition in the queue
Step 3 For the priority metric of throughput sensitive flowswe will use the following
119875119905119894(119905) = 120590
119894lowast 119879min (119894) (26)
where 120590119894is the Head of Line (HOL) blocking and 119875
119905119894(119905) is the
priority metric for throughput sensitive applications In eachtime interval the UE instances are scheduled according to theascending order of their priority values
It follows the following constraints
119876119894119896
st forall119894 isin 119873 119896 isin delay sensitive traffic only (27)
119876119894119895
st forall119894 isin 119873 and 119895
isin throughput sensitive traffic only(28)
119894 = 119895 Queue Length
= Length (119876119894119896+ 119876119894119895)
(29)
Equation (27) ensures that the queue 119876119894takes as input
delay sensitive traffic only (28) ensures that the queue 119876119894119895
takes in throughput sensitive traffic only and (29) states thatqueue length is the total of both the queues
Thus the defined framework and the optimizer will worktogether to form our proposed model The whole process hasfour main goals
(1) Determining the target objectives(2) Based on the objectives forming a system model that
will take into account the objective functions to beminimized or maximized
(3) Constructing a representative solution pool to aid thedecision maker
(4) Introducing a decision-making entity that will notonly ease the task of scheduling but also find anoptimal solution to satisfy the defined objectives
In the following section we will define the simulationmodel and will discuss the results
3 Simulation and Results
31 Simulation Model Designing the model for LTE-A per-formance evaluation is generally more complex comparedto a 3G system An interface is used between variousevolvedNodeBs (eNBs ie the BS) in the LTE system toease the traffic load that occurs due to handovers In this
8 Mobile Information Systems
Table 2 Simulation parameters
Parameters ValuesCarrier frequency 2GHzBandwidth 10MHzNumber of subcarriers 300Number of RBs 45TTI 1msPath loss model Free space propagation modelTransmit power 30mWFading model RayleighPacket model Poisson modelMultiplexing MIMONumber of UE instances 20
section we will design the simulation model for testing andwill evaluate our system for examining delay throughputPacket Loss Ratio (PLR) the effect of SNR and cell spectralefficiency
We are considering a cell with the capacity of 10MHz andit consists of 45 RBs and the carrier frequency is 2GHz Allthe available RBs are controlled by the eNodeB The RBs aredivided among theUE instances in the formof RCsThe LTE-A parameters used for simulation are described in Table 2
We then calculate the number of bits in each RB (ie itsdata rate) It can be computed as follows
119887119894RC = NStti lowastNS119904 lowast 119878rb (30)
where 119887119894RC is achievable data rate NStti is number of slots in
each TTI NS119904is number of symbols in each slot and 119878rb is
number of subcarriers in each RBEach UE is assigned a queue according to the traffic type
and parameters as described in the system model previouslyAfter that we will calculate the channel gain of each UE
119894on
RC and we will now proceeed to calculate the SNRThe SNR values are reported to ENodeB by theUE at each
TTI All the users have a different value of SNR accordingto the experienced fading and multipath propagation thetransmit power (119875) and the total RBs (119873) as follows
SNR =119875 lowast 119866
119894119895
119873(Noise + Interference) (31)
The incoming packets will be treated by the eNodeBaccording to the parameters defined in Table 2 and thesystem model presented All packets have varying lengthand are streamed in queues according to their priorities andproperties Now for the delay sensitive packets the thresholdDB119894is set to be 20ms for each UEThus we can define the Average System Delay (AvD) with
a total simulation time of 119879 at time instance 119905 with totalnumber of UE instances equal to119873 as follows
AvD = 1119879
119879
sum
119905=1
1
119873
119873
sum
119894=1
119863119894(119905) (32)
We will then define AvTH which is the average through-put with trans
119894(119905) being the number of packetssec as follows
AvTH = 1119879
119873
sum
119894=1
119879
sum
119905=1
trans119894(119905) (33)
Now for evaluating the PLR by using the size of discardedpackets pdisc
119894(119905) and the sum of sizes of all packets psize
119894(119905)
we have
PLR =sum119873
119894=1sum119879
119905=1pdisc119894(119905)
sum119873
119894=1sum119879
119905=1psize119894(119905)
(34)
Theoverall spectrumefficiency (119878eff ) ismeasured in termsof total traffic usage factor 119879
119891by applying the formula
119878eff =119862tot119879119891
(35)
where 119879119891= 119903 times 119901 where 119903 is the resource usage factor and
119901 is total traffic 119862tot is number of available traffic channels inthe system
The queuing model is set up by considering 119883119905that
represents the number ofUE instances at a given time interval119905 isin 119879 and 119880
119905denotes the number of new UE instances
that arrive at time 119905 isin 119879 with common probability densityfunction 119891 on 119879 and we define
119883119905+1=
119880119905+1
119883119905= 0
(119883119905minus 1) + 119880
119905+1119883119905gt 0
119905 isin 119879 (36)
where 119883 = 1198830 1198831 is a Markov Chain that has the
probability matrix of the following form
119875 (0 119910) = 119891 (119910) 119910 isin 119879
119875 (119910 119911) = 119891 (119910 minus 119911 + 1)
119911 isin 119879 119910 isin (119911 minus 1 119911 119911 + 1 )
(37)
The chain119883 defined here is the queuing chain of distribu-tion 119891 and starting from 119910 isin 119879 a UE instance is served andnew UE instances arrive by the next time unit 119905 controlled by119891 So the probability of going from state 119910 to 119911 isin (119910minus1 119910 119910+1 ) is lfloor(119911 minus 119910 + 1)rfloor
32 Results Wewill now discuss the simulation results basedon the simulation model and proposed strategy In thispaper we have used the LTE-SIM [35] simulator availableonline at httptelematicspolibaitLTE-Sim It is crucial tocarry out performance evaluation for LTE-A systems and thisrequires the use of standard simulators LTE-Sim is designedaccording to the LTE-A standard with main features suchas multicell environment multiuser support eNodeB CQIUE resource allocation QoS management and schedulingstrategies
The network topology under consideration comprisesnetwork nodes (including eNodeBs and UE) distributedamong various cells We have tested our algorithms for both
Mobile Information Systems 9
MMEGW
MMEGW
Single-cell
eNBeNB
eNB
eNB
UE
UE
Multicell
EGW
eNBNBNBBBNBNBBBBBBBeNB
eNB
UE
UE
Figure 5 Single-cell and multicell simulation scenarios
5 10 15 20Users
DPSMOZ
0
200000
400000
600000
800000
1000000
1200000
Thro
ughp
ut (b
ps)
Figure 6 Video throughput
the single-cell and multicell scenario as shown in Figure 5The first scenario consists of many eNodeBs due to the factthat multiple cells are covered while in the second scenarioone eNodeB handles all UE
We will evaluate the performance of our proposed MOZwhile comparing it with that of DPS [20] For each UEdownlink VOIP and video flows are considered All flows areactive during the whole simulation Time for each simulationis around 100 sThe simulation is performedusing aWindowsmachine with Cygwin (Linux-like environment) a 26GHzCPU and 4GB RAM
321 Delay and Throughput The goal of achieving desiredlevel of QoS is very challenging due to Real Timemultimediaapplications which have strict delay and throughput con-straints Figures 6 and 7 show the throughput for video andVOIP flows with increasing number of UE instances Fromboth the figures it can be seen that the throughput achievedby MOZ is better than that of DPS algorithm althoughthe difference for VOIP flow is much less Throughput alsoincreases with increasing number of users For video up to15 users increasing pattern is observed and after 25 usersthroughput increases but gradually For VOIP flows thethroughput is still increasing although it is delay sensitiveThereason behind this increased throughput is that the systemresources are utilized according to the requirement of users
5 10 15 20Users
MOZDPS
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 7 VOIP throughput
5 10 15 20Users
DPSMOZ
00015
000155
00016
000165
00017
000175
00018
Del
ay (m
sec)
Figure 8 VOIP delay
and only the selected UE instances transmit on the best RBsAlso it is worth noticing that the throughput achieved byMOZ is much closer to the theoretical maximum In thisprocess decisions are made based on the channel quality ofeach UE which constantly measures the QoS parametersbased on the algorithms and techniques proposed After10 users the throughput for video using MOZ achievesperformance gains in order of 1 to 15 while for VOIP it isfrom 1 to 10
Figures 8 and 9 show the average delay of the MOZand DPS algorithm with increasing UE for both the videoand VOIP flows From Figure 8 it can be seen that thedelay performance of MOZ is slightly higher compared toDPS with less than 10 users and as number of UE instancesincreases MOZ shows better performance compared to theDPS algorithm because after 10 plus users there is morevideo streaming and it is not distinguishing VOIP and videoflows so the delay for DPS will increase in consecutive TTIsFigure 9 shows the performance of bothMOZandDPSwhichis very close because with increasing users video demand hasa possibility of rising
Also our results ensure that the delay always remainsbelow the threshold value The proposed MOZ achieves
10 Mobile Information Systems
DPSMOZ
5 10 15 20Users
0
002
004
006
008
01
012
014
Del
ay (m
sec)
Figure 9 Video delay
DPSMOZ
5 10 15 20Users
00280029
0030031003200330034003500360037
PLR
Figure 10 Packet loss ratio
performance gains by reducing delay from 1 to 5 for videoand 1 to 8 for VOIP after 10 users Also it is worth noticinghere that the delay achieved is way below the threshold whichwas set to 20ms
322 PLR Now let us compare the PLR performance ofMOZ and DPS algorithm shown in Figure 10 As expectedthe PLR is increasing as the number of users increasesbecause there are more packets that will be discarded as therewill not be sufficient RBs for transmitting all the packetsconsidering that the packets are nearing their delay thresholdHowever in comparison with the DPS algorithm MOZ iscapable of achieving a better PLR as it supports more usersby distinguishing packets according to their scheduling andresource needs It reduces the loss probability by exploitingthe gains at physical layer Note that the PLR counts only forphysical layer losses MOZ achieves a better PLR and packetloss is reduced from 2 to 14 as compared to the referencealgorithm
323 SNR Figure 11 shows the SNR performance of MOZwith reference to average system throughputWith increasing
5 10 15 20SNR (db)
MOZ
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 11 SNR versus throughput
10 20 30 40SNR (db)
MOZ
021
022
023
024
025
026
027
028
Spec
tral
effici
ency
(bps
)
Figure 12 Cell spectral efficiency
throughput it is observed that the SNR is also increasingwhich depicts better channel quality and less distortion andis one of the reasons we are able to achieve higher systemthroughputThe higher values of SNR are indicator of higherpercentage of data as compared to noise and our resultsshow that noise is around 004 of the actual signal whichis an indicator that the video and voice quality achieved areexcellent
324 Spectral Efficiency Then the spectral efficiency (inbitss) is computed and is depicted in Figure 12 At eachscheduling interval the RB is allocated to UE instancesaccording to their needs As the number of UE instancesincreasesMOZ still guarantees QoS constraints tomore flowswith a positive effect on overall system efficiency The systemsupports delay and throughput guarantees simultaneouslytherefore the spectral efficiency increases with increasingusers which is a good indicator of overall system gain andefficiency
325 Energy and Memory Consumption Figure 13 demon-strates the energy consumption in joules with respect to thenumber of users in the network The energy usage of MOZis compared to PF scheduling and DPS algorithm As the
Mobile Information Systems 11
5 10 15 20Users
DPSMOZPF
0
05
1
15
2
25
Ener
gy (j
oule
s)
Figure 13 Energy consumption versus no of nodes
5 10 15 20Users
MOZ
0
50
100
150
200
250
Mem
ory
Figure 14 Memory consumption versus no of nodes
number of users increases the level of energy consumptionalso rises for all algorithms As the eNodeB remains active allthe time this rise is expected But the energy consumptionlevel achieved by MOZ is the lowest due to the fact thatefficient RB selection takes place which reduces the delaysthus lowering levels of energy consumption Also only theoptimal solution is selected which in turn results in energygains and network lifetime
In Figure 14 the memory consumption in bits versus thenumber of users is depictedThe DM takes decision based onthe information it has and the solution depends on algorithmsthat run dynamically and this process is continuous untilthere are no more RBs to allocate Moreover it has beenrealized that the continuous assignment of RBs accordingto UE instancesrsquo needs utilizes less memory which in turnincreases network life time and performance and reducesdelays
326 PRB Utilization In this subsection we will show effectof the proposed algorithm on PRB utilization The scenariodepicts results according to two groups In the first groupsimulations are carried out without using the proposedMOZthe second group implements MOZ The performance is
5 10 15 20 25 30Video users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 15 Mean PRBs used for video
5 10 15 20 25 30VOIP users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 16 Mean PRBs used for VOIP
measured in terms of mean PRBs being used for both videoandVOIP users In high load the delay increases and a highernumber of PRBs are used when we do not employ the MOZalgorithm As a result after a specific limit no more requestscan be fulfilled because all PRBs are utilized Figure 15 depictsthis scenario as we can see that the average number of PRBsused withMOZ is less due to improved bandwidth utilizationstrategy better coverage and enhanced SNR But with higherload of 30 UE instances all the bandwidth is used up
In Figure 16 for VOIP users also the average number ofPRBs used is much less with increasing number of users
33 Summary of the Simulation Analysis Table 3 representsthe results and statistics obtained by using MOZ where 1 to 4represent the PO set The optimal solution 119874lowast for each set ofusers is bold
It can be clearly seen that the DM always selects the bestfeasible solution thus proving an accuracy of 100 in termsof optimal solution selection The optimization proceduregenerates a PO set for varying number of users and out ofthis set the DM selects the best solution
12 Mobile Information Systems
Table 3 Statistics of the PO set and the optimal solution Olowast selected using MOZ
Number of users PO set (average delay in ms) PO set (average throughput in bps)1 2 3 4 1 2 3 4
5 000252 000328 000173 000175 40722208000 66418738666 50126800000 3949574800010 000179 000197 000252 000328 1509846667 2871940000 4357810667 585808666715 000161 000167 000169 000175 40821598667 76296266667 93016094667 9334639733320 000161 000167 000159 000155 40821541334 76643476000 97387246667 101596778667
The evaluation and comparison of the simulation resultsshow that MOZ performs very well for both voice andVOIP scenarios and performs better than the referencealgorithm in terms of maximum achievable throughput andminimum delay The reason behind these results is the useof optimization process employed by the DM which alwaysselects the optimal solution for the arriving UE instancesbased on their QoS needs and priorities and channel qualityIt further optimizes the procedure by maintaining queues toimprove performance and achieves higher SNR and lowerPLRThe cell spectral efficiency increaseswith each addedUEthat indicates improved system performance and is a proof ofrobust system design
4 Conclusions
In this paper we have proposed a multiobjective optimizerthat helps in finding an optimal solution for selecting RBsfor transmission based on the information provided to thedecision maker It provides an optimal solution for RRM toensure QoS by incorporating themultiobjective optimizationtechniques that target to achieve higher throughput and lowerdelay A set of Pareto optimal solutions is originated and isforwarded to the DM which chooses the best solution outof this set We have adopted an interactive technique thatembeds the multiobjective optimization with the moderntechniques for RRM Furthermore the results are also pro-vided that prove the efficiency of our proposed MOZ Thistechnique is innovative as it makes use of evolutionary MOOtechniques (which are very powerful for finding optimalsolution) and provides its implementation for LTE-A and willprove helpful in solving other issues for 4G and emerging 5Gnetworks
In future we aspire to investigate the opportunities forimplementingMOO techniques for the evolving 5Gnetworksthat not only aim to deal with QoS issues but also target toachieve optimal energy consumption
Competing Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] S-B Lee I Pefkianakis A Meyerson S Xu and S LuldquoProportional fair frequency-domain packet scheduling for3GPP LTE uplinkrdquo in Proceedings of the 28th Conference on
Computer Communications (IEEE INFOCOM rsquo09) pp 2611ndash2615 Rio de Janeiro Brazil April 2009
[2] ldquoPhysical layer aspects for evolved UTRArdquo 3GPP TechnicalReport 25814 version 710 2006
[3] M Nazir and F Saleemi ldquoCooperative cognitive ecology in self-organizing networks a review articlerdquo International Journal ofComputer Science and Information Security vol 10 no 2 pp159ndash167 2012
[4] K Deb Multi-Objective Optimization with Evolutionary Algo-rithms John Wiley amp Sons New York NY USA 1st edition2001
[5] J B K Deb and K M R SłowinskiMulti-Objective Optimiza-tion Interactive and Evolutionary Approaches Springer BerlinGermany 2008
[6] A De Domenico and E C Strinati ldquoA radio resource man-agement scheduling algorithm for self-organizing femtocellsrdquoin Proceedings of the IEEE 21st International Symposium onPersonal Indoor and Mobile Radio Communications Workshops(PIMRC rsquo10) pp 191ndash196 IEEE Instanbul Turkey September2010
[7] R Kausar Y Chen K K Chai L Cuthbert and J SchormansldquoQoS aware mixed traffic packet scheduling in OFDMA-basedLTE-advanced networksrdquo in Proceedings of the 4th InternationalConference on Mobile Ubiquitous Computing Systems Servicesand Technologies (UBICOMM rsquo10) pp 53ndash58 Florence FranceOctober 2010
[8] Y J Zhang and K B Letaief ldquoAdaptive resource allocation andscheduling for multiuser packet-based OFDM networksrdquo inProceedings of the IEEE International Conference on Communi-cation vol 5 pp 2949ndash2953 June 2004
[9] C Stocchi N Marchetti and N R Prasad ldquoSelf-optimizedradio resource management techniques for LTE-a local areadeploymentsrdquo inProceedings of the 2nd International Conferenceon Wireless Communication Vehicular Technology InformationTheory and Aerospace amp Electronic Systems Technology (WirelessVITAE rsquo11) pp 1ndash5 Chennai India March 2011
[10] M Cenk Erturk H Aki I Guvenc and H Arslan ldquoFairand QoS-oriented spectrum splitting in macrocell-femtocellnetworksrdquo in Proceedings of the 53rd IEEE Global Communica-tions Conference (GLOBECOM rsquo10) pp 1ndash6 Miami Fla USADecember 2010
[11] V Chandrasekhar and J G Andrews ldquoSpectrum allocation intiered cellular networksrdquo IEEE Transactions on Communica-tions vol 57 no 10 pp 3059ndash3068 2009
[12] A Roy and S K Das Optimizing QoS-Based Multicast Routingin Wireless Networks A Multi-Objective Genetic AlgorithmicApproach Springer New Yok NY USA 2002
[13] D H Lorenz and A Orda ldquoQoS routing in networks withuncertain parametersrdquo IEEEACM Transactions on Networkingvol 6 no 6 pp 768ndash778 1998
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mobile Information Systems 7
Trafficclassification Optimization Decision-making Scheduling
Input trafficfrom variousUE instances
applicationsDelay sensitive
Throughput sensitiveapplications
Theoptimizationprocess todeterminethe Paretofront
DetermineO
lowast (theoptimalsolution)
Figure 4 The proposed MOZ
The goal of using MOZ is to always get an optimalsolution which is chosen from a set of solutions in the formof Pareto front
25 Queuing and Scheduling Now that we have the Paretooptimal solution in hand the packets are now ready tobe scheduled in their respective queues for transmission asfollows
Step 1 In this step we will define the priority metric for delaysensitive flows The priority of UE
119894at time 119905 119875
119889119894(119905) is
119875119889119894(119905) = 119863max119894 lowast 119866119894119895 (23)
where 119866119894119895is channel gain of UE
119894on RC
119895calculated as
119866119894119895= 10
Path loss10lowast 10
shadow fading10
lowast 10multipath fading10
(24)
Step 2 We will then form a delay queue for a given UE119894that
will ensure delay bounds as follows
119863119894(119905 + 1) = [119863
119894(119905) minus 120588119877
119894(119905)] +
119873
sum
119894=1
119876119894(119905) (25)
where 119876119894(119905) is the mean queue length at time interval 119905 and
119873 is total UE scheduled and 120588119877119894(119905) is the service rate In
each time interval the UEwith highest priority occupies frontposition in the queue
Step 3 For the priority metric of throughput sensitive flowswe will use the following
119875119905119894(119905) = 120590
119894lowast 119879min (119894) (26)
where 120590119894is the Head of Line (HOL) blocking and 119875
119905119894(119905) is the
priority metric for throughput sensitive applications In eachtime interval the UE instances are scheduled according to theascending order of their priority values
It follows the following constraints
119876119894119896
st forall119894 isin 119873 119896 isin delay sensitive traffic only (27)
119876119894119895
st forall119894 isin 119873 and 119895
isin throughput sensitive traffic only(28)
119894 = 119895 Queue Length
= Length (119876119894119896+ 119876119894119895)
(29)
Equation (27) ensures that the queue 119876119894takes as input
delay sensitive traffic only (28) ensures that the queue 119876119894119895
takes in throughput sensitive traffic only and (29) states thatqueue length is the total of both the queues
Thus the defined framework and the optimizer will worktogether to form our proposed model The whole process hasfour main goals
(1) Determining the target objectives(2) Based on the objectives forming a system model that
will take into account the objective functions to beminimized or maximized
(3) Constructing a representative solution pool to aid thedecision maker
(4) Introducing a decision-making entity that will notonly ease the task of scheduling but also find anoptimal solution to satisfy the defined objectives
In the following section we will define the simulationmodel and will discuss the results
3 Simulation and Results
31 Simulation Model Designing the model for LTE-A per-formance evaluation is generally more complex comparedto a 3G system An interface is used between variousevolvedNodeBs (eNBs ie the BS) in the LTE system toease the traffic load that occurs due to handovers In this
8 Mobile Information Systems
Table 2 Simulation parameters
Parameters ValuesCarrier frequency 2GHzBandwidth 10MHzNumber of subcarriers 300Number of RBs 45TTI 1msPath loss model Free space propagation modelTransmit power 30mWFading model RayleighPacket model Poisson modelMultiplexing MIMONumber of UE instances 20
section we will design the simulation model for testing andwill evaluate our system for examining delay throughputPacket Loss Ratio (PLR) the effect of SNR and cell spectralefficiency
We are considering a cell with the capacity of 10MHz andit consists of 45 RBs and the carrier frequency is 2GHz Allthe available RBs are controlled by the eNodeB The RBs aredivided among theUE instances in the formof RCsThe LTE-A parameters used for simulation are described in Table 2
We then calculate the number of bits in each RB (ie itsdata rate) It can be computed as follows
119887119894RC = NStti lowastNS119904 lowast 119878rb (30)
where 119887119894RC is achievable data rate NStti is number of slots in
each TTI NS119904is number of symbols in each slot and 119878rb is
number of subcarriers in each RBEach UE is assigned a queue according to the traffic type
and parameters as described in the system model previouslyAfter that we will calculate the channel gain of each UE
119894on
RC and we will now proceeed to calculate the SNRThe SNR values are reported to ENodeB by theUE at each
TTI All the users have a different value of SNR accordingto the experienced fading and multipath propagation thetransmit power (119875) and the total RBs (119873) as follows
SNR =119875 lowast 119866
119894119895
119873(Noise + Interference) (31)
The incoming packets will be treated by the eNodeBaccording to the parameters defined in Table 2 and thesystem model presented All packets have varying lengthand are streamed in queues according to their priorities andproperties Now for the delay sensitive packets the thresholdDB119894is set to be 20ms for each UEThus we can define the Average System Delay (AvD) with
a total simulation time of 119879 at time instance 119905 with totalnumber of UE instances equal to119873 as follows
AvD = 1119879
119879
sum
119905=1
1
119873
119873
sum
119894=1
119863119894(119905) (32)
We will then define AvTH which is the average through-put with trans
119894(119905) being the number of packetssec as follows
AvTH = 1119879
119873
sum
119894=1
119879
sum
119905=1
trans119894(119905) (33)
Now for evaluating the PLR by using the size of discardedpackets pdisc
119894(119905) and the sum of sizes of all packets psize
119894(119905)
we have
PLR =sum119873
119894=1sum119879
119905=1pdisc119894(119905)
sum119873
119894=1sum119879
119905=1psize119894(119905)
(34)
Theoverall spectrumefficiency (119878eff ) ismeasured in termsof total traffic usage factor 119879
119891by applying the formula
119878eff =119862tot119879119891
(35)
where 119879119891= 119903 times 119901 where 119903 is the resource usage factor and
119901 is total traffic 119862tot is number of available traffic channels inthe system
The queuing model is set up by considering 119883119905that
represents the number ofUE instances at a given time interval119905 isin 119879 and 119880
119905denotes the number of new UE instances
that arrive at time 119905 isin 119879 with common probability densityfunction 119891 on 119879 and we define
119883119905+1=
119880119905+1
119883119905= 0
(119883119905minus 1) + 119880
119905+1119883119905gt 0
119905 isin 119879 (36)
where 119883 = 1198830 1198831 is a Markov Chain that has the
probability matrix of the following form
119875 (0 119910) = 119891 (119910) 119910 isin 119879
119875 (119910 119911) = 119891 (119910 minus 119911 + 1)
119911 isin 119879 119910 isin (119911 minus 1 119911 119911 + 1 )
(37)
The chain119883 defined here is the queuing chain of distribu-tion 119891 and starting from 119910 isin 119879 a UE instance is served andnew UE instances arrive by the next time unit 119905 controlled by119891 So the probability of going from state 119910 to 119911 isin (119910minus1 119910 119910+1 ) is lfloor(119911 minus 119910 + 1)rfloor
32 Results Wewill now discuss the simulation results basedon the simulation model and proposed strategy In thispaper we have used the LTE-SIM [35] simulator availableonline at httptelematicspolibaitLTE-Sim It is crucial tocarry out performance evaluation for LTE-A systems and thisrequires the use of standard simulators LTE-Sim is designedaccording to the LTE-A standard with main features suchas multicell environment multiuser support eNodeB CQIUE resource allocation QoS management and schedulingstrategies
The network topology under consideration comprisesnetwork nodes (including eNodeBs and UE) distributedamong various cells We have tested our algorithms for both
Mobile Information Systems 9
MMEGW
MMEGW
Single-cell
eNBeNB
eNB
eNB
UE
UE
Multicell
EGW
eNBNBNBBBNBNBBBBBBBeNB
eNB
UE
UE
Figure 5 Single-cell and multicell simulation scenarios
5 10 15 20Users
DPSMOZ
0
200000
400000
600000
800000
1000000
1200000
Thro
ughp
ut (b
ps)
Figure 6 Video throughput
the single-cell and multicell scenario as shown in Figure 5The first scenario consists of many eNodeBs due to the factthat multiple cells are covered while in the second scenarioone eNodeB handles all UE
We will evaluate the performance of our proposed MOZwhile comparing it with that of DPS [20] For each UEdownlink VOIP and video flows are considered All flows areactive during the whole simulation Time for each simulationis around 100 sThe simulation is performedusing aWindowsmachine with Cygwin (Linux-like environment) a 26GHzCPU and 4GB RAM
321 Delay and Throughput The goal of achieving desiredlevel of QoS is very challenging due to Real Timemultimediaapplications which have strict delay and throughput con-straints Figures 6 and 7 show the throughput for video andVOIP flows with increasing number of UE instances Fromboth the figures it can be seen that the throughput achievedby MOZ is better than that of DPS algorithm althoughthe difference for VOIP flow is much less Throughput alsoincreases with increasing number of users For video up to15 users increasing pattern is observed and after 25 usersthroughput increases but gradually For VOIP flows thethroughput is still increasing although it is delay sensitiveThereason behind this increased throughput is that the systemresources are utilized according to the requirement of users
5 10 15 20Users
MOZDPS
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 7 VOIP throughput
5 10 15 20Users
DPSMOZ
00015
000155
00016
000165
00017
000175
00018
Del
ay (m
sec)
Figure 8 VOIP delay
and only the selected UE instances transmit on the best RBsAlso it is worth noticing that the throughput achieved byMOZ is much closer to the theoretical maximum In thisprocess decisions are made based on the channel quality ofeach UE which constantly measures the QoS parametersbased on the algorithms and techniques proposed After10 users the throughput for video using MOZ achievesperformance gains in order of 1 to 15 while for VOIP it isfrom 1 to 10
Figures 8 and 9 show the average delay of the MOZand DPS algorithm with increasing UE for both the videoand VOIP flows From Figure 8 it can be seen that thedelay performance of MOZ is slightly higher compared toDPS with less than 10 users and as number of UE instancesincreases MOZ shows better performance compared to theDPS algorithm because after 10 plus users there is morevideo streaming and it is not distinguishing VOIP and videoflows so the delay for DPS will increase in consecutive TTIsFigure 9 shows the performance of bothMOZandDPSwhichis very close because with increasing users video demand hasa possibility of rising
Also our results ensure that the delay always remainsbelow the threshold value The proposed MOZ achieves
10 Mobile Information Systems
DPSMOZ
5 10 15 20Users
0
002
004
006
008
01
012
014
Del
ay (m
sec)
Figure 9 Video delay
DPSMOZ
5 10 15 20Users
00280029
0030031003200330034003500360037
PLR
Figure 10 Packet loss ratio
performance gains by reducing delay from 1 to 5 for videoand 1 to 8 for VOIP after 10 users Also it is worth noticinghere that the delay achieved is way below the threshold whichwas set to 20ms
322 PLR Now let us compare the PLR performance ofMOZ and DPS algorithm shown in Figure 10 As expectedthe PLR is increasing as the number of users increasesbecause there are more packets that will be discarded as therewill not be sufficient RBs for transmitting all the packetsconsidering that the packets are nearing their delay thresholdHowever in comparison with the DPS algorithm MOZ iscapable of achieving a better PLR as it supports more usersby distinguishing packets according to their scheduling andresource needs It reduces the loss probability by exploitingthe gains at physical layer Note that the PLR counts only forphysical layer losses MOZ achieves a better PLR and packetloss is reduced from 2 to 14 as compared to the referencealgorithm
323 SNR Figure 11 shows the SNR performance of MOZwith reference to average system throughputWith increasing
5 10 15 20SNR (db)
MOZ
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 11 SNR versus throughput
10 20 30 40SNR (db)
MOZ
021
022
023
024
025
026
027
028
Spec
tral
effici
ency
(bps
)
Figure 12 Cell spectral efficiency
throughput it is observed that the SNR is also increasingwhich depicts better channel quality and less distortion andis one of the reasons we are able to achieve higher systemthroughputThe higher values of SNR are indicator of higherpercentage of data as compared to noise and our resultsshow that noise is around 004 of the actual signal whichis an indicator that the video and voice quality achieved areexcellent
324 Spectral Efficiency Then the spectral efficiency (inbitss) is computed and is depicted in Figure 12 At eachscheduling interval the RB is allocated to UE instancesaccording to their needs As the number of UE instancesincreasesMOZ still guarantees QoS constraints tomore flowswith a positive effect on overall system efficiency The systemsupports delay and throughput guarantees simultaneouslytherefore the spectral efficiency increases with increasingusers which is a good indicator of overall system gain andefficiency
325 Energy and Memory Consumption Figure 13 demon-strates the energy consumption in joules with respect to thenumber of users in the network The energy usage of MOZis compared to PF scheduling and DPS algorithm As the
Mobile Information Systems 11
5 10 15 20Users
DPSMOZPF
0
05
1
15
2
25
Ener
gy (j
oule
s)
Figure 13 Energy consumption versus no of nodes
5 10 15 20Users
MOZ
0
50
100
150
200
250
Mem
ory
Figure 14 Memory consumption versus no of nodes
number of users increases the level of energy consumptionalso rises for all algorithms As the eNodeB remains active allthe time this rise is expected But the energy consumptionlevel achieved by MOZ is the lowest due to the fact thatefficient RB selection takes place which reduces the delaysthus lowering levels of energy consumption Also only theoptimal solution is selected which in turn results in energygains and network lifetime
In Figure 14 the memory consumption in bits versus thenumber of users is depictedThe DM takes decision based onthe information it has and the solution depends on algorithmsthat run dynamically and this process is continuous untilthere are no more RBs to allocate Moreover it has beenrealized that the continuous assignment of RBs accordingto UE instancesrsquo needs utilizes less memory which in turnincreases network life time and performance and reducesdelays
326 PRB Utilization In this subsection we will show effectof the proposed algorithm on PRB utilization The scenariodepicts results according to two groups In the first groupsimulations are carried out without using the proposedMOZthe second group implements MOZ The performance is
5 10 15 20 25 30Video users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 15 Mean PRBs used for video
5 10 15 20 25 30VOIP users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 16 Mean PRBs used for VOIP
measured in terms of mean PRBs being used for both videoandVOIP users In high load the delay increases and a highernumber of PRBs are used when we do not employ the MOZalgorithm As a result after a specific limit no more requestscan be fulfilled because all PRBs are utilized Figure 15 depictsthis scenario as we can see that the average number of PRBsused withMOZ is less due to improved bandwidth utilizationstrategy better coverage and enhanced SNR But with higherload of 30 UE instances all the bandwidth is used up
In Figure 16 for VOIP users also the average number ofPRBs used is much less with increasing number of users
33 Summary of the Simulation Analysis Table 3 representsthe results and statistics obtained by using MOZ where 1 to 4represent the PO set The optimal solution 119874lowast for each set ofusers is bold
It can be clearly seen that the DM always selects the bestfeasible solution thus proving an accuracy of 100 in termsof optimal solution selection The optimization proceduregenerates a PO set for varying number of users and out ofthis set the DM selects the best solution
12 Mobile Information Systems
Table 3 Statistics of the PO set and the optimal solution Olowast selected using MOZ
Number of users PO set (average delay in ms) PO set (average throughput in bps)1 2 3 4 1 2 3 4
5 000252 000328 000173 000175 40722208000 66418738666 50126800000 3949574800010 000179 000197 000252 000328 1509846667 2871940000 4357810667 585808666715 000161 000167 000169 000175 40821598667 76296266667 93016094667 9334639733320 000161 000167 000159 000155 40821541334 76643476000 97387246667 101596778667
The evaluation and comparison of the simulation resultsshow that MOZ performs very well for both voice andVOIP scenarios and performs better than the referencealgorithm in terms of maximum achievable throughput andminimum delay The reason behind these results is the useof optimization process employed by the DM which alwaysselects the optimal solution for the arriving UE instancesbased on their QoS needs and priorities and channel qualityIt further optimizes the procedure by maintaining queues toimprove performance and achieves higher SNR and lowerPLRThe cell spectral efficiency increaseswith each addedUEthat indicates improved system performance and is a proof ofrobust system design
4 Conclusions
In this paper we have proposed a multiobjective optimizerthat helps in finding an optimal solution for selecting RBsfor transmission based on the information provided to thedecision maker It provides an optimal solution for RRM toensure QoS by incorporating themultiobjective optimizationtechniques that target to achieve higher throughput and lowerdelay A set of Pareto optimal solutions is originated and isforwarded to the DM which chooses the best solution outof this set We have adopted an interactive technique thatembeds the multiobjective optimization with the moderntechniques for RRM Furthermore the results are also pro-vided that prove the efficiency of our proposed MOZ Thistechnique is innovative as it makes use of evolutionary MOOtechniques (which are very powerful for finding optimalsolution) and provides its implementation for LTE-A and willprove helpful in solving other issues for 4G and emerging 5Gnetworks
In future we aspire to investigate the opportunities forimplementingMOO techniques for the evolving 5Gnetworksthat not only aim to deal with QoS issues but also target toachieve optimal energy consumption
Competing Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] S-B Lee I Pefkianakis A Meyerson S Xu and S LuldquoProportional fair frequency-domain packet scheduling for3GPP LTE uplinkrdquo in Proceedings of the 28th Conference on
Computer Communications (IEEE INFOCOM rsquo09) pp 2611ndash2615 Rio de Janeiro Brazil April 2009
[2] ldquoPhysical layer aspects for evolved UTRArdquo 3GPP TechnicalReport 25814 version 710 2006
[3] M Nazir and F Saleemi ldquoCooperative cognitive ecology in self-organizing networks a review articlerdquo International Journal ofComputer Science and Information Security vol 10 no 2 pp159ndash167 2012
[4] K Deb Multi-Objective Optimization with Evolutionary Algo-rithms John Wiley amp Sons New York NY USA 1st edition2001
[5] J B K Deb and K M R SłowinskiMulti-Objective Optimiza-tion Interactive and Evolutionary Approaches Springer BerlinGermany 2008
[6] A De Domenico and E C Strinati ldquoA radio resource man-agement scheduling algorithm for self-organizing femtocellsrdquoin Proceedings of the IEEE 21st International Symposium onPersonal Indoor and Mobile Radio Communications Workshops(PIMRC rsquo10) pp 191ndash196 IEEE Instanbul Turkey September2010
[7] R Kausar Y Chen K K Chai L Cuthbert and J SchormansldquoQoS aware mixed traffic packet scheduling in OFDMA-basedLTE-advanced networksrdquo in Proceedings of the 4th InternationalConference on Mobile Ubiquitous Computing Systems Servicesand Technologies (UBICOMM rsquo10) pp 53ndash58 Florence FranceOctober 2010
[8] Y J Zhang and K B Letaief ldquoAdaptive resource allocation andscheduling for multiuser packet-based OFDM networksrdquo inProceedings of the IEEE International Conference on Communi-cation vol 5 pp 2949ndash2953 June 2004
[9] C Stocchi N Marchetti and N R Prasad ldquoSelf-optimizedradio resource management techniques for LTE-a local areadeploymentsrdquo inProceedings of the 2nd International Conferenceon Wireless Communication Vehicular Technology InformationTheory and Aerospace amp Electronic Systems Technology (WirelessVITAE rsquo11) pp 1ndash5 Chennai India March 2011
[10] M Cenk Erturk H Aki I Guvenc and H Arslan ldquoFairand QoS-oriented spectrum splitting in macrocell-femtocellnetworksrdquo in Proceedings of the 53rd IEEE Global Communica-tions Conference (GLOBECOM rsquo10) pp 1ndash6 Miami Fla USADecember 2010
[11] V Chandrasekhar and J G Andrews ldquoSpectrum allocation intiered cellular networksrdquo IEEE Transactions on Communica-tions vol 57 no 10 pp 3059ndash3068 2009
[12] A Roy and S K Das Optimizing QoS-Based Multicast Routingin Wireless Networks A Multi-Objective Genetic AlgorithmicApproach Springer New Yok NY USA 2002
[13] D H Lorenz and A Orda ldquoQoS routing in networks withuncertain parametersrdquo IEEEACM Transactions on Networkingvol 6 no 6 pp 768ndash778 1998
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
8 Mobile Information Systems
Table 2 Simulation parameters
Parameters ValuesCarrier frequency 2GHzBandwidth 10MHzNumber of subcarriers 300Number of RBs 45TTI 1msPath loss model Free space propagation modelTransmit power 30mWFading model RayleighPacket model Poisson modelMultiplexing MIMONumber of UE instances 20
section we will design the simulation model for testing andwill evaluate our system for examining delay throughputPacket Loss Ratio (PLR) the effect of SNR and cell spectralefficiency
We are considering a cell with the capacity of 10MHz andit consists of 45 RBs and the carrier frequency is 2GHz Allthe available RBs are controlled by the eNodeB The RBs aredivided among theUE instances in the formof RCsThe LTE-A parameters used for simulation are described in Table 2
We then calculate the number of bits in each RB (ie itsdata rate) It can be computed as follows
119887119894RC = NStti lowastNS119904 lowast 119878rb (30)
where 119887119894RC is achievable data rate NStti is number of slots in
each TTI NS119904is number of symbols in each slot and 119878rb is
number of subcarriers in each RBEach UE is assigned a queue according to the traffic type
and parameters as described in the system model previouslyAfter that we will calculate the channel gain of each UE
119894on
RC and we will now proceeed to calculate the SNRThe SNR values are reported to ENodeB by theUE at each
TTI All the users have a different value of SNR accordingto the experienced fading and multipath propagation thetransmit power (119875) and the total RBs (119873) as follows
SNR =119875 lowast 119866
119894119895
119873(Noise + Interference) (31)
The incoming packets will be treated by the eNodeBaccording to the parameters defined in Table 2 and thesystem model presented All packets have varying lengthand are streamed in queues according to their priorities andproperties Now for the delay sensitive packets the thresholdDB119894is set to be 20ms for each UEThus we can define the Average System Delay (AvD) with
a total simulation time of 119879 at time instance 119905 with totalnumber of UE instances equal to119873 as follows
AvD = 1119879
119879
sum
119905=1
1
119873
119873
sum
119894=1
119863119894(119905) (32)
We will then define AvTH which is the average through-put with trans
119894(119905) being the number of packetssec as follows
AvTH = 1119879
119873
sum
119894=1
119879
sum
119905=1
trans119894(119905) (33)
Now for evaluating the PLR by using the size of discardedpackets pdisc
119894(119905) and the sum of sizes of all packets psize
119894(119905)
we have
PLR =sum119873
119894=1sum119879
119905=1pdisc119894(119905)
sum119873
119894=1sum119879
119905=1psize119894(119905)
(34)
Theoverall spectrumefficiency (119878eff ) ismeasured in termsof total traffic usage factor 119879
119891by applying the formula
119878eff =119862tot119879119891
(35)
where 119879119891= 119903 times 119901 where 119903 is the resource usage factor and
119901 is total traffic 119862tot is number of available traffic channels inthe system
The queuing model is set up by considering 119883119905that
represents the number ofUE instances at a given time interval119905 isin 119879 and 119880
119905denotes the number of new UE instances
that arrive at time 119905 isin 119879 with common probability densityfunction 119891 on 119879 and we define
119883119905+1=
119880119905+1
119883119905= 0
(119883119905minus 1) + 119880
119905+1119883119905gt 0
119905 isin 119879 (36)
where 119883 = 1198830 1198831 is a Markov Chain that has the
probability matrix of the following form
119875 (0 119910) = 119891 (119910) 119910 isin 119879
119875 (119910 119911) = 119891 (119910 minus 119911 + 1)
119911 isin 119879 119910 isin (119911 minus 1 119911 119911 + 1 )
(37)
The chain119883 defined here is the queuing chain of distribu-tion 119891 and starting from 119910 isin 119879 a UE instance is served andnew UE instances arrive by the next time unit 119905 controlled by119891 So the probability of going from state 119910 to 119911 isin (119910minus1 119910 119910+1 ) is lfloor(119911 minus 119910 + 1)rfloor
32 Results Wewill now discuss the simulation results basedon the simulation model and proposed strategy In thispaper we have used the LTE-SIM [35] simulator availableonline at httptelematicspolibaitLTE-Sim It is crucial tocarry out performance evaluation for LTE-A systems and thisrequires the use of standard simulators LTE-Sim is designedaccording to the LTE-A standard with main features suchas multicell environment multiuser support eNodeB CQIUE resource allocation QoS management and schedulingstrategies
The network topology under consideration comprisesnetwork nodes (including eNodeBs and UE) distributedamong various cells We have tested our algorithms for both
Mobile Information Systems 9
MMEGW
MMEGW
Single-cell
eNBeNB
eNB
eNB
UE
UE
Multicell
EGW
eNBNBNBBBNBNBBBBBBBeNB
eNB
UE
UE
Figure 5 Single-cell and multicell simulation scenarios
5 10 15 20Users
DPSMOZ
0
200000
400000
600000
800000
1000000
1200000
Thro
ughp
ut (b
ps)
Figure 6 Video throughput
the single-cell and multicell scenario as shown in Figure 5The first scenario consists of many eNodeBs due to the factthat multiple cells are covered while in the second scenarioone eNodeB handles all UE
We will evaluate the performance of our proposed MOZwhile comparing it with that of DPS [20] For each UEdownlink VOIP and video flows are considered All flows areactive during the whole simulation Time for each simulationis around 100 sThe simulation is performedusing aWindowsmachine with Cygwin (Linux-like environment) a 26GHzCPU and 4GB RAM
321 Delay and Throughput The goal of achieving desiredlevel of QoS is very challenging due to Real Timemultimediaapplications which have strict delay and throughput con-straints Figures 6 and 7 show the throughput for video andVOIP flows with increasing number of UE instances Fromboth the figures it can be seen that the throughput achievedby MOZ is better than that of DPS algorithm althoughthe difference for VOIP flow is much less Throughput alsoincreases with increasing number of users For video up to15 users increasing pattern is observed and after 25 usersthroughput increases but gradually For VOIP flows thethroughput is still increasing although it is delay sensitiveThereason behind this increased throughput is that the systemresources are utilized according to the requirement of users
5 10 15 20Users
MOZDPS
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 7 VOIP throughput
5 10 15 20Users
DPSMOZ
00015
000155
00016
000165
00017
000175
00018
Del
ay (m
sec)
Figure 8 VOIP delay
and only the selected UE instances transmit on the best RBsAlso it is worth noticing that the throughput achieved byMOZ is much closer to the theoretical maximum In thisprocess decisions are made based on the channel quality ofeach UE which constantly measures the QoS parametersbased on the algorithms and techniques proposed After10 users the throughput for video using MOZ achievesperformance gains in order of 1 to 15 while for VOIP it isfrom 1 to 10
Figures 8 and 9 show the average delay of the MOZand DPS algorithm with increasing UE for both the videoand VOIP flows From Figure 8 it can be seen that thedelay performance of MOZ is slightly higher compared toDPS with less than 10 users and as number of UE instancesincreases MOZ shows better performance compared to theDPS algorithm because after 10 plus users there is morevideo streaming and it is not distinguishing VOIP and videoflows so the delay for DPS will increase in consecutive TTIsFigure 9 shows the performance of bothMOZandDPSwhichis very close because with increasing users video demand hasa possibility of rising
Also our results ensure that the delay always remainsbelow the threshold value The proposed MOZ achieves
10 Mobile Information Systems
DPSMOZ
5 10 15 20Users
0
002
004
006
008
01
012
014
Del
ay (m
sec)
Figure 9 Video delay
DPSMOZ
5 10 15 20Users
00280029
0030031003200330034003500360037
PLR
Figure 10 Packet loss ratio
performance gains by reducing delay from 1 to 5 for videoand 1 to 8 for VOIP after 10 users Also it is worth noticinghere that the delay achieved is way below the threshold whichwas set to 20ms
322 PLR Now let us compare the PLR performance ofMOZ and DPS algorithm shown in Figure 10 As expectedthe PLR is increasing as the number of users increasesbecause there are more packets that will be discarded as therewill not be sufficient RBs for transmitting all the packetsconsidering that the packets are nearing their delay thresholdHowever in comparison with the DPS algorithm MOZ iscapable of achieving a better PLR as it supports more usersby distinguishing packets according to their scheduling andresource needs It reduces the loss probability by exploitingthe gains at physical layer Note that the PLR counts only forphysical layer losses MOZ achieves a better PLR and packetloss is reduced from 2 to 14 as compared to the referencealgorithm
323 SNR Figure 11 shows the SNR performance of MOZwith reference to average system throughputWith increasing
5 10 15 20SNR (db)
MOZ
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 11 SNR versus throughput
10 20 30 40SNR (db)
MOZ
021
022
023
024
025
026
027
028
Spec
tral
effici
ency
(bps
)
Figure 12 Cell spectral efficiency
throughput it is observed that the SNR is also increasingwhich depicts better channel quality and less distortion andis one of the reasons we are able to achieve higher systemthroughputThe higher values of SNR are indicator of higherpercentage of data as compared to noise and our resultsshow that noise is around 004 of the actual signal whichis an indicator that the video and voice quality achieved areexcellent
324 Spectral Efficiency Then the spectral efficiency (inbitss) is computed and is depicted in Figure 12 At eachscheduling interval the RB is allocated to UE instancesaccording to their needs As the number of UE instancesincreasesMOZ still guarantees QoS constraints tomore flowswith a positive effect on overall system efficiency The systemsupports delay and throughput guarantees simultaneouslytherefore the spectral efficiency increases with increasingusers which is a good indicator of overall system gain andefficiency
325 Energy and Memory Consumption Figure 13 demon-strates the energy consumption in joules with respect to thenumber of users in the network The energy usage of MOZis compared to PF scheduling and DPS algorithm As the
Mobile Information Systems 11
5 10 15 20Users
DPSMOZPF
0
05
1
15
2
25
Ener
gy (j
oule
s)
Figure 13 Energy consumption versus no of nodes
5 10 15 20Users
MOZ
0
50
100
150
200
250
Mem
ory
Figure 14 Memory consumption versus no of nodes
number of users increases the level of energy consumptionalso rises for all algorithms As the eNodeB remains active allthe time this rise is expected But the energy consumptionlevel achieved by MOZ is the lowest due to the fact thatefficient RB selection takes place which reduces the delaysthus lowering levels of energy consumption Also only theoptimal solution is selected which in turn results in energygains and network lifetime
In Figure 14 the memory consumption in bits versus thenumber of users is depictedThe DM takes decision based onthe information it has and the solution depends on algorithmsthat run dynamically and this process is continuous untilthere are no more RBs to allocate Moreover it has beenrealized that the continuous assignment of RBs accordingto UE instancesrsquo needs utilizes less memory which in turnincreases network life time and performance and reducesdelays
326 PRB Utilization In this subsection we will show effectof the proposed algorithm on PRB utilization The scenariodepicts results according to two groups In the first groupsimulations are carried out without using the proposedMOZthe second group implements MOZ The performance is
5 10 15 20 25 30Video users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 15 Mean PRBs used for video
5 10 15 20 25 30VOIP users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 16 Mean PRBs used for VOIP
measured in terms of mean PRBs being used for both videoandVOIP users In high load the delay increases and a highernumber of PRBs are used when we do not employ the MOZalgorithm As a result after a specific limit no more requestscan be fulfilled because all PRBs are utilized Figure 15 depictsthis scenario as we can see that the average number of PRBsused withMOZ is less due to improved bandwidth utilizationstrategy better coverage and enhanced SNR But with higherload of 30 UE instances all the bandwidth is used up
In Figure 16 for VOIP users also the average number ofPRBs used is much less with increasing number of users
33 Summary of the Simulation Analysis Table 3 representsthe results and statistics obtained by using MOZ where 1 to 4represent the PO set The optimal solution 119874lowast for each set ofusers is bold
It can be clearly seen that the DM always selects the bestfeasible solution thus proving an accuracy of 100 in termsof optimal solution selection The optimization proceduregenerates a PO set for varying number of users and out ofthis set the DM selects the best solution
12 Mobile Information Systems
Table 3 Statistics of the PO set and the optimal solution Olowast selected using MOZ
Number of users PO set (average delay in ms) PO set (average throughput in bps)1 2 3 4 1 2 3 4
5 000252 000328 000173 000175 40722208000 66418738666 50126800000 3949574800010 000179 000197 000252 000328 1509846667 2871940000 4357810667 585808666715 000161 000167 000169 000175 40821598667 76296266667 93016094667 9334639733320 000161 000167 000159 000155 40821541334 76643476000 97387246667 101596778667
The evaluation and comparison of the simulation resultsshow that MOZ performs very well for both voice andVOIP scenarios and performs better than the referencealgorithm in terms of maximum achievable throughput andminimum delay The reason behind these results is the useof optimization process employed by the DM which alwaysselects the optimal solution for the arriving UE instancesbased on their QoS needs and priorities and channel qualityIt further optimizes the procedure by maintaining queues toimprove performance and achieves higher SNR and lowerPLRThe cell spectral efficiency increaseswith each addedUEthat indicates improved system performance and is a proof ofrobust system design
4 Conclusions
In this paper we have proposed a multiobjective optimizerthat helps in finding an optimal solution for selecting RBsfor transmission based on the information provided to thedecision maker It provides an optimal solution for RRM toensure QoS by incorporating themultiobjective optimizationtechniques that target to achieve higher throughput and lowerdelay A set of Pareto optimal solutions is originated and isforwarded to the DM which chooses the best solution outof this set We have adopted an interactive technique thatembeds the multiobjective optimization with the moderntechniques for RRM Furthermore the results are also pro-vided that prove the efficiency of our proposed MOZ Thistechnique is innovative as it makes use of evolutionary MOOtechniques (which are very powerful for finding optimalsolution) and provides its implementation for LTE-A and willprove helpful in solving other issues for 4G and emerging 5Gnetworks
In future we aspire to investigate the opportunities forimplementingMOO techniques for the evolving 5Gnetworksthat not only aim to deal with QoS issues but also target toachieve optimal energy consumption
Competing Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] S-B Lee I Pefkianakis A Meyerson S Xu and S LuldquoProportional fair frequency-domain packet scheduling for3GPP LTE uplinkrdquo in Proceedings of the 28th Conference on
Computer Communications (IEEE INFOCOM rsquo09) pp 2611ndash2615 Rio de Janeiro Brazil April 2009
[2] ldquoPhysical layer aspects for evolved UTRArdquo 3GPP TechnicalReport 25814 version 710 2006
[3] M Nazir and F Saleemi ldquoCooperative cognitive ecology in self-organizing networks a review articlerdquo International Journal ofComputer Science and Information Security vol 10 no 2 pp159ndash167 2012
[4] K Deb Multi-Objective Optimization with Evolutionary Algo-rithms John Wiley amp Sons New York NY USA 1st edition2001
[5] J B K Deb and K M R SłowinskiMulti-Objective Optimiza-tion Interactive and Evolutionary Approaches Springer BerlinGermany 2008
[6] A De Domenico and E C Strinati ldquoA radio resource man-agement scheduling algorithm for self-organizing femtocellsrdquoin Proceedings of the IEEE 21st International Symposium onPersonal Indoor and Mobile Radio Communications Workshops(PIMRC rsquo10) pp 191ndash196 IEEE Instanbul Turkey September2010
[7] R Kausar Y Chen K K Chai L Cuthbert and J SchormansldquoQoS aware mixed traffic packet scheduling in OFDMA-basedLTE-advanced networksrdquo in Proceedings of the 4th InternationalConference on Mobile Ubiquitous Computing Systems Servicesand Technologies (UBICOMM rsquo10) pp 53ndash58 Florence FranceOctober 2010
[8] Y J Zhang and K B Letaief ldquoAdaptive resource allocation andscheduling for multiuser packet-based OFDM networksrdquo inProceedings of the IEEE International Conference on Communi-cation vol 5 pp 2949ndash2953 June 2004
[9] C Stocchi N Marchetti and N R Prasad ldquoSelf-optimizedradio resource management techniques for LTE-a local areadeploymentsrdquo inProceedings of the 2nd International Conferenceon Wireless Communication Vehicular Technology InformationTheory and Aerospace amp Electronic Systems Technology (WirelessVITAE rsquo11) pp 1ndash5 Chennai India March 2011
[10] M Cenk Erturk H Aki I Guvenc and H Arslan ldquoFairand QoS-oriented spectrum splitting in macrocell-femtocellnetworksrdquo in Proceedings of the 53rd IEEE Global Communica-tions Conference (GLOBECOM rsquo10) pp 1ndash6 Miami Fla USADecember 2010
[11] V Chandrasekhar and J G Andrews ldquoSpectrum allocation intiered cellular networksrdquo IEEE Transactions on Communica-tions vol 57 no 10 pp 3059ndash3068 2009
[12] A Roy and S K Das Optimizing QoS-Based Multicast Routingin Wireless Networks A Multi-Objective Genetic AlgorithmicApproach Springer New Yok NY USA 2002
[13] D H Lorenz and A Orda ldquoQoS routing in networks withuncertain parametersrdquo IEEEACM Transactions on Networkingvol 6 no 6 pp 768ndash778 1998
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mobile Information Systems 9
MMEGW
MMEGW
Single-cell
eNBeNB
eNB
eNB
UE
UE
Multicell
EGW
eNBNBNBBBNBNBBBBBBBeNB
eNB
UE
UE
Figure 5 Single-cell and multicell simulation scenarios
5 10 15 20Users
DPSMOZ
0
200000
400000
600000
800000
1000000
1200000
Thro
ughp
ut (b
ps)
Figure 6 Video throughput
the single-cell and multicell scenario as shown in Figure 5The first scenario consists of many eNodeBs due to the factthat multiple cells are covered while in the second scenarioone eNodeB handles all UE
We will evaluate the performance of our proposed MOZwhile comparing it with that of DPS [20] For each UEdownlink VOIP and video flows are considered All flows areactive during the whole simulation Time for each simulationis around 100 sThe simulation is performedusing aWindowsmachine with Cygwin (Linux-like environment) a 26GHzCPU and 4GB RAM
321 Delay and Throughput The goal of achieving desiredlevel of QoS is very challenging due to Real Timemultimediaapplications which have strict delay and throughput con-straints Figures 6 and 7 show the throughput for video andVOIP flows with increasing number of UE instances Fromboth the figures it can be seen that the throughput achievedby MOZ is better than that of DPS algorithm althoughthe difference for VOIP flow is much less Throughput alsoincreases with increasing number of users For video up to15 users increasing pattern is observed and after 25 usersthroughput increases but gradually For VOIP flows thethroughput is still increasing although it is delay sensitiveThereason behind this increased throughput is that the systemresources are utilized according to the requirement of users
5 10 15 20Users
MOZDPS
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 7 VOIP throughput
5 10 15 20Users
DPSMOZ
00015
000155
00016
000165
00017
000175
00018
Del
ay (m
sec)
Figure 8 VOIP delay
and only the selected UE instances transmit on the best RBsAlso it is worth noticing that the throughput achieved byMOZ is much closer to the theoretical maximum In thisprocess decisions are made based on the channel quality ofeach UE which constantly measures the QoS parametersbased on the algorithms and techniques proposed After10 users the throughput for video using MOZ achievesperformance gains in order of 1 to 15 while for VOIP it isfrom 1 to 10
Figures 8 and 9 show the average delay of the MOZand DPS algorithm with increasing UE for both the videoand VOIP flows From Figure 8 it can be seen that thedelay performance of MOZ is slightly higher compared toDPS with less than 10 users and as number of UE instancesincreases MOZ shows better performance compared to theDPS algorithm because after 10 plus users there is morevideo streaming and it is not distinguishing VOIP and videoflows so the delay for DPS will increase in consecutive TTIsFigure 9 shows the performance of bothMOZandDPSwhichis very close because with increasing users video demand hasa possibility of rising
Also our results ensure that the delay always remainsbelow the threshold value The proposed MOZ achieves
10 Mobile Information Systems
DPSMOZ
5 10 15 20Users
0
002
004
006
008
01
012
014
Del
ay (m
sec)
Figure 9 Video delay
DPSMOZ
5 10 15 20Users
00280029
0030031003200330034003500360037
PLR
Figure 10 Packet loss ratio
performance gains by reducing delay from 1 to 5 for videoand 1 to 8 for VOIP after 10 users Also it is worth noticinghere that the delay achieved is way below the threshold whichwas set to 20ms
322 PLR Now let us compare the PLR performance ofMOZ and DPS algorithm shown in Figure 10 As expectedthe PLR is increasing as the number of users increasesbecause there are more packets that will be discarded as therewill not be sufficient RBs for transmitting all the packetsconsidering that the packets are nearing their delay thresholdHowever in comparison with the DPS algorithm MOZ iscapable of achieving a better PLR as it supports more usersby distinguishing packets according to their scheduling andresource needs It reduces the loss probability by exploitingthe gains at physical layer Note that the PLR counts only forphysical layer losses MOZ achieves a better PLR and packetloss is reduced from 2 to 14 as compared to the referencealgorithm
323 SNR Figure 11 shows the SNR performance of MOZwith reference to average system throughputWith increasing
5 10 15 20SNR (db)
MOZ
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 11 SNR versus throughput
10 20 30 40SNR (db)
MOZ
021
022
023
024
025
026
027
028
Spec
tral
effici
ency
(bps
)
Figure 12 Cell spectral efficiency
throughput it is observed that the SNR is also increasingwhich depicts better channel quality and less distortion andis one of the reasons we are able to achieve higher systemthroughputThe higher values of SNR are indicator of higherpercentage of data as compared to noise and our resultsshow that noise is around 004 of the actual signal whichis an indicator that the video and voice quality achieved areexcellent
324 Spectral Efficiency Then the spectral efficiency (inbitss) is computed and is depicted in Figure 12 At eachscheduling interval the RB is allocated to UE instancesaccording to their needs As the number of UE instancesincreasesMOZ still guarantees QoS constraints tomore flowswith a positive effect on overall system efficiency The systemsupports delay and throughput guarantees simultaneouslytherefore the spectral efficiency increases with increasingusers which is a good indicator of overall system gain andefficiency
325 Energy and Memory Consumption Figure 13 demon-strates the energy consumption in joules with respect to thenumber of users in the network The energy usage of MOZis compared to PF scheduling and DPS algorithm As the
Mobile Information Systems 11
5 10 15 20Users
DPSMOZPF
0
05
1
15
2
25
Ener
gy (j
oule
s)
Figure 13 Energy consumption versus no of nodes
5 10 15 20Users
MOZ
0
50
100
150
200
250
Mem
ory
Figure 14 Memory consumption versus no of nodes
number of users increases the level of energy consumptionalso rises for all algorithms As the eNodeB remains active allthe time this rise is expected But the energy consumptionlevel achieved by MOZ is the lowest due to the fact thatefficient RB selection takes place which reduces the delaysthus lowering levels of energy consumption Also only theoptimal solution is selected which in turn results in energygains and network lifetime
In Figure 14 the memory consumption in bits versus thenumber of users is depictedThe DM takes decision based onthe information it has and the solution depends on algorithmsthat run dynamically and this process is continuous untilthere are no more RBs to allocate Moreover it has beenrealized that the continuous assignment of RBs accordingto UE instancesrsquo needs utilizes less memory which in turnincreases network life time and performance and reducesdelays
326 PRB Utilization In this subsection we will show effectof the proposed algorithm on PRB utilization The scenariodepicts results according to two groups In the first groupsimulations are carried out without using the proposedMOZthe second group implements MOZ The performance is
5 10 15 20 25 30Video users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 15 Mean PRBs used for video
5 10 15 20 25 30VOIP users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 16 Mean PRBs used for VOIP
measured in terms of mean PRBs being used for both videoandVOIP users In high load the delay increases and a highernumber of PRBs are used when we do not employ the MOZalgorithm As a result after a specific limit no more requestscan be fulfilled because all PRBs are utilized Figure 15 depictsthis scenario as we can see that the average number of PRBsused withMOZ is less due to improved bandwidth utilizationstrategy better coverage and enhanced SNR But with higherload of 30 UE instances all the bandwidth is used up
In Figure 16 for VOIP users also the average number ofPRBs used is much less with increasing number of users
33 Summary of the Simulation Analysis Table 3 representsthe results and statistics obtained by using MOZ where 1 to 4represent the PO set The optimal solution 119874lowast for each set ofusers is bold
It can be clearly seen that the DM always selects the bestfeasible solution thus proving an accuracy of 100 in termsof optimal solution selection The optimization proceduregenerates a PO set for varying number of users and out ofthis set the DM selects the best solution
12 Mobile Information Systems
Table 3 Statistics of the PO set and the optimal solution Olowast selected using MOZ
Number of users PO set (average delay in ms) PO set (average throughput in bps)1 2 3 4 1 2 3 4
5 000252 000328 000173 000175 40722208000 66418738666 50126800000 3949574800010 000179 000197 000252 000328 1509846667 2871940000 4357810667 585808666715 000161 000167 000169 000175 40821598667 76296266667 93016094667 9334639733320 000161 000167 000159 000155 40821541334 76643476000 97387246667 101596778667
The evaluation and comparison of the simulation resultsshow that MOZ performs very well for both voice andVOIP scenarios and performs better than the referencealgorithm in terms of maximum achievable throughput andminimum delay The reason behind these results is the useof optimization process employed by the DM which alwaysselects the optimal solution for the arriving UE instancesbased on their QoS needs and priorities and channel qualityIt further optimizes the procedure by maintaining queues toimprove performance and achieves higher SNR and lowerPLRThe cell spectral efficiency increaseswith each addedUEthat indicates improved system performance and is a proof ofrobust system design
4 Conclusions
In this paper we have proposed a multiobjective optimizerthat helps in finding an optimal solution for selecting RBsfor transmission based on the information provided to thedecision maker It provides an optimal solution for RRM toensure QoS by incorporating themultiobjective optimizationtechniques that target to achieve higher throughput and lowerdelay A set of Pareto optimal solutions is originated and isforwarded to the DM which chooses the best solution outof this set We have adopted an interactive technique thatembeds the multiobjective optimization with the moderntechniques for RRM Furthermore the results are also pro-vided that prove the efficiency of our proposed MOZ Thistechnique is innovative as it makes use of evolutionary MOOtechniques (which are very powerful for finding optimalsolution) and provides its implementation for LTE-A and willprove helpful in solving other issues for 4G and emerging 5Gnetworks
In future we aspire to investigate the opportunities forimplementingMOO techniques for the evolving 5Gnetworksthat not only aim to deal with QoS issues but also target toachieve optimal energy consumption
Competing Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] S-B Lee I Pefkianakis A Meyerson S Xu and S LuldquoProportional fair frequency-domain packet scheduling for3GPP LTE uplinkrdquo in Proceedings of the 28th Conference on
Computer Communications (IEEE INFOCOM rsquo09) pp 2611ndash2615 Rio de Janeiro Brazil April 2009
[2] ldquoPhysical layer aspects for evolved UTRArdquo 3GPP TechnicalReport 25814 version 710 2006
[3] M Nazir and F Saleemi ldquoCooperative cognitive ecology in self-organizing networks a review articlerdquo International Journal ofComputer Science and Information Security vol 10 no 2 pp159ndash167 2012
[4] K Deb Multi-Objective Optimization with Evolutionary Algo-rithms John Wiley amp Sons New York NY USA 1st edition2001
[5] J B K Deb and K M R SłowinskiMulti-Objective Optimiza-tion Interactive and Evolutionary Approaches Springer BerlinGermany 2008
[6] A De Domenico and E C Strinati ldquoA radio resource man-agement scheduling algorithm for self-organizing femtocellsrdquoin Proceedings of the IEEE 21st International Symposium onPersonal Indoor and Mobile Radio Communications Workshops(PIMRC rsquo10) pp 191ndash196 IEEE Instanbul Turkey September2010
[7] R Kausar Y Chen K K Chai L Cuthbert and J SchormansldquoQoS aware mixed traffic packet scheduling in OFDMA-basedLTE-advanced networksrdquo in Proceedings of the 4th InternationalConference on Mobile Ubiquitous Computing Systems Servicesand Technologies (UBICOMM rsquo10) pp 53ndash58 Florence FranceOctober 2010
[8] Y J Zhang and K B Letaief ldquoAdaptive resource allocation andscheduling for multiuser packet-based OFDM networksrdquo inProceedings of the IEEE International Conference on Communi-cation vol 5 pp 2949ndash2953 June 2004
[9] C Stocchi N Marchetti and N R Prasad ldquoSelf-optimizedradio resource management techniques for LTE-a local areadeploymentsrdquo inProceedings of the 2nd International Conferenceon Wireless Communication Vehicular Technology InformationTheory and Aerospace amp Electronic Systems Technology (WirelessVITAE rsquo11) pp 1ndash5 Chennai India March 2011
[10] M Cenk Erturk H Aki I Guvenc and H Arslan ldquoFairand QoS-oriented spectrum splitting in macrocell-femtocellnetworksrdquo in Proceedings of the 53rd IEEE Global Communica-tions Conference (GLOBECOM rsquo10) pp 1ndash6 Miami Fla USADecember 2010
[11] V Chandrasekhar and J G Andrews ldquoSpectrum allocation intiered cellular networksrdquo IEEE Transactions on Communica-tions vol 57 no 10 pp 3059ndash3068 2009
[12] A Roy and S K Das Optimizing QoS-Based Multicast Routingin Wireless Networks A Multi-Objective Genetic AlgorithmicApproach Springer New Yok NY USA 2002
[13] D H Lorenz and A Orda ldquoQoS routing in networks withuncertain parametersrdquo IEEEACM Transactions on Networkingvol 6 no 6 pp 768ndash778 1998
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
10 Mobile Information Systems
DPSMOZ
5 10 15 20Users
0
002
004
006
008
01
012
014
Del
ay (m
sec)
Figure 9 Video delay
DPSMOZ
5 10 15 20Users
00280029
0030031003200330034003500360037
PLR
Figure 10 Packet loss ratio
performance gains by reducing delay from 1 to 5 for videoand 1 to 8 for VOIP after 10 users Also it is worth noticinghere that the delay achieved is way below the threshold whichwas set to 20ms
322 PLR Now let us compare the PLR performance ofMOZ and DPS algorithm shown in Figure 10 As expectedthe PLR is increasing as the number of users increasesbecause there are more packets that will be discarded as therewill not be sufficient RBs for transmitting all the packetsconsidering that the packets are nearing their delay thresholdHowever in comparison with the DPS algorithm MOZ iscapable of achieving a better PLR as it supports more usersby distinguishing packets according to their scheduling andresource needs It reduces the loss probability by exploitingthe gains at physical layer Note that the PLR counts only forphysical layer losses MOZ achieves a better PLR and packetloss is reduced from 2 to 14 as compared to the referencealgorithm
323 SNR Figure 11 shows the SNR performance of MOZwith reference to average system throughputWith increasing
5 10 15 20SNR (db)
MOZ
0
10000
20000
30000
40000
50000
60000
70000
Thro
ughp
ut (b
ps)
Figure 11 SNR versus throughput
10 20 30 40SNR (db)
MOZ
021
022
023
024
025
026
027
028
Spec
tral
effici
ency
(bps
)
Figure 12 Cell spectral efficiency
throughput it is observed that the SNR is also increasingwhich depicts better channel quality and less distortion andis one of the reasons we are able to achieve higher systemthroughputThe higher values of SNR are indicator of higherpercentage of data as compared to noise and our resultsshow that noise is around 004 of the actual signal whichis an indicator that the video and voice quality achieved areexcellent
324 Spectral Efficiency Then the spectral efficiency (inbitss) is computed and is depicted in Figure 12 At eachscheduling interval the RB is allocated to UE instancesaccording to their needs As the number of UE instancesincreasesMOZ still guarantees QoS constraints tomore flowswith a positive effect on overall system efficiency The systemsupports delay and throughput guarantees simultaneouslytherefore the spectral efficiency increases with increasingusers which is a good indicator of overall system gain andefficiency
325 Energy and Memory Consumption Figure 13 demon-strates the energy consumption in joules with respect to thenumber of users in the network The energy usage of MOZis compared to PF scheduling and DPS algorithm As the
Mobile Information Systems 11
5 10 15 20Users
DPSMOZPF
0
05
1
15
2
25
Ener
gy (j
oule
s)
Figure 13 Energy consumption versus no of nodes
5 10 15 20Users
MOZ
0
50
100
150
200
250
Mem
ory
Figure 14 Memory consumption versus no of nodes
number of users increases the level of energy consumptionalso rises for all algorithms As the eNodeB remains active allthe time this rise is expected But the energy consumptionlevel achieved by MOZ is the lowest due to the fact thatefficient RB selection takes place which reduces the delaysthus lowering levels of energy consumption Also only theoptimal solution is selected which in turn results in energygains and network lifetime
In Figure 14 the memory consumption in bits versus thenumber of users is depictedThe DM takes decision based onthe information it has and the solution depends on algorithmsthat run dynamically and this process is continuous untilthere are no more RBs to allocate Moreover it has beenrealized that the continuous assignment of RBs accordingto UE instancesrsquo needs utilizes less memory which in turnincreases network life time and performance and reducesdelays
326 PRB Utilization In this subsection we will show effectof the proposed algorithm on PRB utilization The scenariodepicts results according to two groups In the first groupsimulations are carried out without using the proposedMOZthe second group implements MOZ The performance is
5 10 15 20 25 30Video users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 15 Mean PRBs used for video
5 10 15 20 25 30VOIP users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 16 Mean PRBs used for VOIP
measured in terms of mean PRBs being used for both videoandVOIP users In high load the delay increases and a highernumber of PRBs are used when we do not employ the MOZalgorithm As a result after a specific limit no more requestscan be fulfilled because all PRBs are utilized Figure 15 depictsthis scenario as we can see that the average number of PRBsused withMOZ is less due to improved bandwidth utilizationstrategy better coverage and enhanced SNR But with higherload of 30 UE instances all the bandwidth is used up
In Figure 16 for VOIP users also the average number ofPRBs used is much less with increasing number of users
33 Summary of the Simulation Analysis Table 3 representsthe results and statistics obtained by using MOZ where 1 to 4represent the PO set The optimal solution 119874lowast for each set ofusers is bold
It can be clearly seen that the DM always selects the bestfeasible solution thus proving an accuracy of 100 in termsof optimal solution selection The optimization proceduregenerates a PO set for varying number of users and out ofthis set the DM selects the best solution
12 Mobile Information Systems
Table 3 Statistics of the PO set and the optimal solution Olowast selected using MOZ
Number of users PO set (average delay in ms) PO set (average throughput in bps)1 2 3 4 1 2 3 4
5 000252 000328 000173 000175 40722208000 66418738666 50126800000 3949574800010 000179 000197 000252 000328 1509846667 2871940000 4357810667 585808666715 000161 000167 000169 000175 40821598667 76296266667 93016094667 9334639733320 000161 000167 000159 000155 40821541334 76643476000 97387246667 101596778667
The evaluation and comparison of the simulation resultsshow that MOZ performs very well for both voice andVOIP scenarios and performs better than the referencealgorithm in terms of maximum achievable throughput andminimum delay The reason behind these results is the useof optimization process employed by the DM which alwaysselects the optimal solution for the arriving UE instancesbased on their QoS needs and priorities and channel qualityIt further optimizes the procedure by maintaining queues toimprove performance and achieves higher SNR and lowerPLRThe cell spectral efficiency increaseswith each addedUEthat indicates improved system performance and is a proof ofrobust system design
4 Conclusions
In this paper we have proposed a multiobjective optimizerthat helps in finding an optimal solution for selecting RBsfor transmission based on the information provided to thedecision maker It provides an optimal solution for RRM toensure QoS by incorporating themultiobjective optimizationtechniques that target to achieve higher throughput and lowerdelay A set of Pareto optimal solutions is originated and isforwarded to the DM which chooses the best solution outof this set We have adopted an interactive technique thatembeds the multiobjective optimization with the moderntechniques for RRM Furthermore the results are also pro-vided that prove the efficiency of our proposed MOZ Thistechnique is innovative as it makes use of evolutionary MOOtechniques (which are very powerful for finding optimalsolution) and provides its implementation for LTE-A and willprove helpful in solving other issues for 4G and emerging 5Gnetworks
In future we aspire to investigate the opportunities forimplementingMOO techniques for the evolving 5Gnetworksthat not only aim to deal with QoS issues but also target toachieve optimal energy consumption
Competing Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] S-B Lee I Pefkianakis A Meyerson S Xu and S LuldquoProportional fair frequency-domain packet scheduling for3GPP LTE uplinkrdquo in Proceedings of the 28th Conference on
Computer Communications (IEEE INFOCOM rsquo09) pp 2611ndash2615 Rio de Janeiro Brazil April 2009
[2] ldquoPhysical layer aspects for evolved UTRArdquo 3GPP TechnicalReport 25814 version 710 2006
[3] M Nazir and F Saleemi ldquoCooperative cognitive ecology in self-organizing networks a review articlerdquo International Journal ofComputer Science and Information Security vol 10 no 2 pp159ndash167 2012
[4] K Deb Multi-Objective Optimization with Evolutionary Algo-rithms John Wiley amp Sons New York NY USA 1st edition2001
[5] J B K Deb and K M R SłowinskiMulti-Objective Optimiza-tion Interactive and Evolutionary Approaches Springer BerlinGermany 2008
[6] A De Domenico and E C Strinati ldquoA radio resource man-agement scheduling algorithm for self-organizing femtocellsrdquoin Proceedings of the IEEE 21st International Symposium onPersonal Indoor and Mobile Radio Communications Workshops(PIMRC rsquo10) pp 191ndash196 IEEE Instanbul Turkey September2010
[7] R Kausar Y Chen K K Chai L Cuthbert and J SchormansldquoQoS aware mixed traffic packet scheduling in OFDMA-basedLTE-advanced networksrdquo in Proceedings of the 4th InternationalConference on Mobile Ubiquitous Computing Systems Servicesand Technologies (UBICOMM rsquo10) pp 53ndash58 Florence FranceOctober 2010
[8] Y J Zhang and K B Letaief ldquoAdaptive resource allocation andscheduling for multiuser packet-based OFDM networksrdquo inProceedings of the IEEE International Conference on Communi-cation vol 5 pp 2949ndash2953 June 2004
[9] C Stocchi N Marchetti and N R Prasad ldquoSelf-optimizedradio resource management techniques for LTE-a local areadeploymentsrdquo inProceedings of the 2nd International Conferenceon Wireless Communication Vehicular Technology InformationTheory and Aerospace amp Electronic Systems Technology (WirelessVITAE rsquo11) pp 1ndash5 Chennai India March 2011
[10] M Cenk Erturk H Aki I Guvenc and H Arslan ldquoFairand QoS-oriented spectrum splitting in macrocell-femtocellnetworksrdquo in Proceedings of the 53rd IEEE Global Communica-tions Conference (GLOBECOM rsquo10) pp 1ndash6 Miami Fla USADecember 2010
[11] V Chandrasekhar and J G Andrews ldquoSpectrum allocation intiered cellular networksrdquo IEEE Transactions on Communica-tions vol 57 no 10 pp 3059ndash3068 2009
[12] A Roy and S K Das Optimizing QoS-Based Multicast Routingin Wireless Networks A Multi-Objective Genetic AlgorithmicApproach Springer New Yok NY USA 2002
[13] D H Lorenz and A Orda ldquoQoS routing in networks withuncertain parametersrdquo IEEEACM Transactions on Networkingvol 6 no 6 pp 768ndash778 1998
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mobile Information Systems 11
5 10 15 20Users
DPSMOZPF
0
05
1
15
2
25
Ener
gy (j
oule
s)
Figure 13 Energy consumption versus no of nodes
5 10 15 20Users
MOZ
0
50
100
150
200
250
Mem
ory
Figure 14 Memory consumption versus no of nodes
number of users increases the level of energy consumptionalso rises for all algorithms As the eNodeB remains active allthe time this rise is expected But the energy consumptionlevel achieved by MOZ is the lowest due to the fact thatefficient RB selection takes place which reduces the delaysthus lowering levels of energy consumption Also only theoptimal solution is selected which in turn results in energygains and network lifetime
In Figure 14 the memory consumption in bits versus thenumber of users is depictedThe DM takes decision based onthe information it has and the solution depends on algorithmsthat run dynamically and this process is continuous untilthere are no more RBs to allocate Moreover it has beenrealized that the continuous assignment of RBs accordingto UE instancesrsquo needs utilizes less memory which in turnincreases network life time and performance and reducesdelays
326 PRB Utilization In this subsection we will show effectof the proposed algorithm on PRB utilization The scenariodepicts results according to two groups In the first groupsimulations are carried out without using the proposedMOZthe second group implements MOZ The performance is
5 10 15 20 25 30Video users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 15 Mean PRBs used for video
5 10 15 20 25 30VOIP users
Without MOZWith MOZ
0
5
10
15
20
25
30
PRBs
Figure 16 Mean PRBs used for VOIP
measured in terms of mean PRBs being used for both videoandVOIP users In high load the delay increases and a highernumber of PRBs are used when we do not employ the MOZalgorithm As a result after a specific limit no more requestscan be fulfilled because all PRBs are utilized Figure 15 depictsthis scenario as we can see that the average number of PRBsused withMOZ is less due to improved bandwidth utilizationstrategy better coverage and enhanced SNR But with higherload of 30 UE instances all the bandwidth is used up
In Figure 16 for VOIP users also the average number ofPRBs used is much less with increasing number of users
33 Summary of the Simulation Analysis Table 3 representsthe results and statistics obtained by using MOZ where 1 to 4represent the PO set The optimal solution 119874lowast for each set ofusers is bold
It can be clearly seen that the DM always selects the bestfeasible solution thus proving an accuracy of 100 in termsof optimal solution selection The optimization proceduregenerates a PO set for varying number of users and out ofthis set the DM selects the best solution
12 Mobile Information Systems
Table 3 Statistics of the PO set and the optimal solution Olowast selected using MOZ
Number of users PO set (average delay in ms) PO set (average throughput in bps)1 2 3 4 1 2 3 4
5 000252 000328 000173 000175 40722208000 66418738666 50126800000 3949574800010 000179 000197 000252 000328 1509846667 2871940000 4357810667 585808666715 000161 000167 000169 000175 40821598667 76296266667 93016094667 9334639733320 000161 000167 000159 000155 40821541334 76643476000 97387246667 101596778667
The evaluation and comparison of the simulation resultsshow that MOZ performs very well for both voice andVOIP scenarios and performs better than the referencealgorithm in terms of maximum achievable throughput andminimum delay The reason behind these results is the useof optimization process employed by the DM which alwaysselects the optimal solution for the arriving UE instancesbased on their QoS needs and priorities and channel qualityIt further optimizes the procedure by maintaining queues toimprove performance and achieves higher SNR and lowerPLRThe cell spectral efficiency increaseswith each addedUEthat indicates improved system performance and is a proof ofrobust system design
4 Conclusions
In this paper we have proposed a multiobjective optimizerthat helps in finding an optimal solution for selecting RBsfor transmission based on the information provided to thedecision maker It provides an optimal solution for RRM toensure QoS by incorporating themultiobjective optimizationtechniques that target to achieve higher throughput and lowerdelay A set of Pareto optimal solutions is originated and isforwarded to the DM which chooses the best solution outof this set We have adopted an interactive technique thatembeds the multiobjective optimization with the moderntechniques for RRM Furthermore the results are also pro-vided that prove the efficiency of our proposed MOZ Thistechnique is innovative as it makes use of evolutionary MOOtechniques (which are very powerful for finding optimalsolution) and provides its implementation for LTE-A and willprove helpful in solving other issues for 4G and emerging 5Gnetworks
In future we aspire to investigate the opportunities forimplementingMOO techniques for the evolving 5Gnetworksthat not only aim to deal with QoS issues but also target toachieve optimal energy consumption
Competing Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] S-B Lee I Pefkianakis A Meyerson S Xu and S LuldquoProportional fair frequency-domain packet scheduling for3GPP LTE uplinkrdquo in Proceedings of the 28th Conference on
Computer Communications (IEEE INFOCOM rsquo09) pp 2611ndash2615 Rio de Janeiro Brazil April 2009
[2] ldquoPhysical layer aspects for evolved UTRArdquo 3GPP TechnicalReport 25814 version 710 2006
[3] M Nazir and F Saleemi ldquoCooperative cognitive ecology in self-organizing networks a review articlerdquo International Journal ofComputer Science and Information Security vol 10 no 2 pp159ndash167 2012
[4] K Deb Multi-Objective Optimization with Evolutionary Algo-rithms John Wiley amp Sons New York NY USA 1st edition2001
[5] J B K Deb and K M R SłowinskiMulti-Objective Optimiza-tion Interactive and Evolutionary Approaches Springer BerlinGermany 2008
[6] A De Domenico and E C Strinati ldquoA radio resource man-agement scheduling algorithm for self-organizing femtocellsrdquoin Proceedings of the IEEE 21st International Symposium onPersonal Indoor and Mobile Radio Communications Workshops(PIMRC rsquo10) pp 191ndash196 IEEE Instanbul Turkey September2010
[7] R Kausar Y Chen K K Chai L Cuthbert and J SchormansldquoQoS aware mixed traffic packet scheduling in OFDMA-basedLTE-advanced networksrdquo in Proceedings of the 4th InternationalConference on Mobile Ubiquitous Computing Systems Servicesand Technologies (UBICOMM rsquo10) pp 53ndash58 Florence FranceOctober 2010
[8] Y J Zhang and K B Letaief ldquoAdaptive resource allocation andscheduling for multiuser packet-based OFDM networksrdquo inProceedings of the IEEE International Conference on Communi-cation vol 5 pp 2949ndash2953 June 2004
[9] C Stocchi N Marchetti and N R Prasad ldquoSelf-optimizedradio resource management techniques for LTE-a local areadeploymentsrdquo inProceedings of the 2nd International Conferenceon Wireless Communication Vehicular Technology InformationTheory and Aerospace amp Electronic Systems Technology (WirelessVITAE rsquo11) pp 1ndash5 Chennai India March 2011
[10] M Cenk Erturk H Aki I Guvenc and H Arslan ldquoFairand QoS-oriented spectrum splitting in macrocell-femtocellnetworksrdquo in Proceedings of the 53rd IEEE Global Communica-tions Conference (GLOBECOM rsquo10) pp 1ndash6 Miami Fla USADecember 2010
[11] V Chandrasekhar and J G Andrews ldquoSpectrum allocation intiered cellular networksrdquo IEEE Transactions on Communica-tions vol 57 no 10 pp 3059ndash3068 2009
[12] A Roy and S K Das Optimizing QoS-Based Multicast Routingin Wireless Networks A Multi-Objective Genetic AlgorithmicApproach Springer New Yok NY USA 2002
[13] D H Lorenz and A Orda ldquoQoS routing in networks withuncertain parametersrdquo IEEEACM Transactions on Networkingvol 6 no 6 pp 768ndash778 1998
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
12 Mobile Information Systems
Table 3 Statistics of the PO set and the optimal solution Olowast selected using MOZ
Number of users PO set (average delay in ms) PO set (average throughput in bps)1 2 3 4 1 2 3 4
5 000252 000328 000173 000175 40722208000 66418738666 50126800000 3949574800010 000179 000197 000252 000328 1509846667 2871940000 4357810667 585808666715 000161 000167 000169 000175 40821598667 76296266667 93016094667 9334639733320 000161 000167 000159 000155 40821541334 76643476000 97387246667 101596778667
The evaluation and comparison of the simulation resultsshow that MOZ performs very well for both voice andVOIP scenarios and performs better than the referencealgorithm in terms of maximum achievable throughput andminimum delay The reason behind these results is the useof optimization process employed by the DM which alwaysselects the optimal solution for the arriving UE instancesbased on their QoS needs and priorities and channel qualityIt further optimizes the procedure by maintaining queues toimprove performance and achieves higher SNR and lowerPLRThe cell spectral efficiency increaseswith each addedUEthat indicates improved system performance and is a proof ofrobust system design
4 Conclusions
In this paper we have proposed a multiobjective optimizerthat helps in finding an optimal solution for selecting RBsfor transmission based on the information provided to thedecision maker It provides an optimal solution for RRM toensure QoS by incorporating themultiobjective optimizationtechniques that target to achieve higher throughput and lowerdelay A set of Pareto optimal solutions is originated and isforwarded to the DM which chooses the best solution outof this set We have adopted an interactive technique thatembeds the multiobjective optimization with the moderntechniques for RRM Furthermore the results are also pro-vided that prove the efficiency of our proposed MOZ Thistechnique is innovative as it makes use of evolutionary MOOtechniques (which are very powerful for finding optimalsolution) and provides its implementation for LTE-A and willprove helpful in solving other issues for 4G and emerging 5Gnetworks
In future we aspire to investigate the opportunities forimplementingMOO techniques for the evolving 5Gnetworksthat not only aim to deal with QoS issues but also target toachieve optimal energy consumption
Competing Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] S-B Lee I Pefkianakis A Meyerson S Xu and S LuldquoProportional fair frequency-domain packet scheduling for3GPP LTE uplinkrdquo in Proceedings of the 28th Conference on
Computer Communications (IEEE INFOCOM rsquo09) pp 2611ndash2615 Rio de Janeiro Brazil April 2009
[2] ldquoPhysical layer aspects for evolved UTRArdquo 3GPP TechnicalReport 25814 version 710 2006
[3] M Nazir and F Saleemi ldquoCooperative cognitive ecology in self-organizing networks a review articlerdquo International Journal ofComputer Science and Information Security vol 10 no 2 pp159ndash167 2012
[4] K Deb Multi-Objective Optimization with Evolutionary Algo-rithms John Wiley amp Sons New York NY USA 1st edition2001
[5] J B K Deb and K M R SłowinskiMulti-Objective Optimiza-tion Interactive and Evolutionary Approaches Springer BerlinGermany 2008
[6] A De Domenico and E C Strinati ldquoA radio resource man-agement scheduling algorithm for self-organizing femtocellsrdquoin Proceedings of the IEEE 21st International Symposium onPersonal Indoor and Mobile Radio Communications Workshops(PIMRC rsquo10) pp 191ndash196 IEEE Instanbul Turkey September2010
[7] R Kausar Y Chen K K Chai L Cuthbert and J SchormansldquoQoS aware mixed traffic packet scheduling in OFDMA-basedLTE-advanced networksrdquo in Proceedings of the 4th InternationalConference on Mobile Ubiquitous Computing Systems Servicesand Technologies (UBICOMM rsquo10) pp 53ndash58 Florence FranceOctober 2010
[8] Y J Zhang and K B Letaief ldquoAdaptive resource allocation andscheduling for multiuser packet-based OFDM networksrdquo inProceedings of the IEEE International Conference on Communi-cation vol 5 pp 2949ndash2953 June 2004
[9] C Stocchi N Marchetti and N R Prasad ldquoSelf-optimizedradio resource management techniques for LTE-a local areadeploymentsrdquo inProceedings of the 2nd International Conferenceon Wireless Communication Vehicular Technology InformationTheory and Aerospace amp Electronic Systems Technology (WirelessVITAE rsquo11) pp 1ndash5 Chennai India March 2011
[10] M Cenk Erturk H Aki I Guvenc and H Arslan ldquoFairand QoS-oriented spectrum splitting in macrocell-femtocellnetworksrdquo in Proceedings of the 53rd IEEE Global Communica-tions Conference (GLOBECOM rsquo10) pp 1ndash6 Miami Fla USADecember 2010
[11] V Chandrasekhar and J G Andrews ldquoSpectrum allocation intiered cellular networksrdquo IEEE Transactions on Communica-tions vol 57 no 10 pp 3059ndash3068 2009
[12] A Roy and S K Das Optimizing QoS-Based Multicast Routingin Wireless Networks A Multi-Objective Genetic AlgorithmicApproach Springer New Yok NY USA 2002
[13] D H Lorenz and A Orda ldquoQoS routing in networks withuncertain parametersrdquo IEEEACM Transactions on Networkingvol 6 no 6 pp 768ndash778 1998
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mobile Information Systems 13
[14] A M Brintrup J Ramsden H Takagi and A TiwarildquoErgonomic chair design by fusing qualitative and quantitativecriteria using interactive genetic algorithmsrdquo IEEE Transactionson Evolutionary Computation vol 12 no 3 pp 343ndash354 2008
[15] R Kamalian H Takagi and A Agogino ldquoOptimized designof MEMS by evolutionary multi-objective optimization withinteractive evolutionary computationrdquo in Genetic and Evolu-tionary ComputationmdashGECCO 2004 Genetic and EvolutionaryComputation Conference Seattle WA USA June 26ndash30 2004Proceedings Part II vol 3103 of Lecture Notes in ComputerScience pp 1030ndash1041 Springer Berlin Germany 2004
[16] S Stefania T Issam and B Matthew The UMTS Long TermEvolution Forum Theory to Practice John Wiley amp Sons NewYork NY USA 2009
[17] J Puttonen N Kolehmainen T Henttonen M Moisio andM Rinne ldquoMixed traffic packet scheduling in UTRAN longterm evolution downlinkrdquo in Proceedings of the IEEE 19thInternational Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC rsquo08) pp 1ndash5 Cannes France Septem-ber 2008
[18] P Won-Hyoung C Sunghyun and B Saewoong ldquoSchedulingdesign for multiple traffic classes in OFDMA networksrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo06) pp 790ndash795 Istanbul Turky 2006
[19] T Janevski Traffic Analysis and Design of Wireless IP NetworksArtech House Norwood Mass USA 2003
[20] K Sandrasegaran H A M Ramli and R Basukala ldquoDelay-Prioritized Scheduling (DPS) for real time traffic in 3GPP LTEsystemrdquo in Proceedings of the IEEE Wireless Communicationsand Networking Conference (WCNC rsquo10) pp 1ndash6 IEEE SydneyAustralia April 2010
[21] A Machwe I C Parmee and J C Miles ldquoMulti-objectiveanalysis of a component based representation within an inter-active evolutionary design systemrdquo in Proceedings of the 7thInternational Conference in Adaptive Computing andDesign andManufacturing 2006
[22] K C Tan T H Lee D Khoo and E F Khor ldquoA multiobjectiveevolutionary algorithm toolbox for computer-aided multiob-jective optimizationrdquo IEEE Transactions on Systems Man andCybernetics Part B Cybernetics vol 31 no 4 pp 537ndash556 2001
[23] C M Fonseca and P J Fleming ldquoMulti-objective optimizationandmultiple constraint handlingwith evolutionary algorithmsrdquoIEEE Transactions on Systems Man and Cybernetics Part ASystems and Humans vol 28 no 1 pp 38ndash47 1998
[24] K Miettinen Nonlinear Multi-Objective Optimization KluwerBoston Mass USA 1999
[25] V Chankong andY YHaimesMulti-ObjectiveDecisionMakingTheory and Methodology North-Holland New York NY USA1983
[26] K Miettinen and M M Makela ldquoInteractive bundle-basedmethod for nondifferentiable multiobjective optimizationNIMBUSrdquo Optimization vol 34 no 3 pp 231ndash246 1995
[27] F Y EdgeworthMathematical Psychics An Essay on the Appli-cation of Mathematics to the Moral Sciences C Kegan Paul ampCo London UK 1881
[28] H Kuhn and A Tucker ldquoNonlinear programmingrdquo in Proceed-ings of the Second Berkeley Symposium on Mathematical Statis-tics and Probability J Neyman Ed pp 481ndash492 University ofCalifornia Press Berkeley 1951
[29] P Korhonen ldquoInteractive methodsrdquo in Multiple Criteria Deci-sion Analysis J Figueira S Greco and M Ehrgott Eds State
of the Art Surveys pp 641ndash665 Springer New York NY USA2005
[30] F Moety S Lahoud B Cousin and K Khawam ldquoA heuristicalgorithm for joint power-delay minimization in green wirelessaccess networksrdquo in Proceedings of the International Conferenceon Computing Networking and Communications (ICNC rsquo15) pp280ndash286 IEEE Garden Grove Calif USA February 2015
[31] S Jun N Yi A Liu and X Haige ldquoOpportunistic schedulingfor heterogeneous services in downlink OFDMA systemrdquoin Proceedings of the IEEE WRI International Conference onCommunications and Mobile Computing( CMC rsquo09) pp 260ndash264 Yunnan China January 2009
[32] H Jutler ldquoLinear model with several objective functionsrdquoEkonomika I Matematiceckije Metody vol 3 pp 397ndash406 1967(Polish)
[33] R Solich ldquoLinear programming problem with several objectivefunctionsrdquo Przeglad Statystyczny vol 16 pp 24ndash30 1969 (Pol-ish)
[34] A H Ali M Nazir A Afzaal and A Sabah ldquoA traffic schedulerfor radio resource management of Long Term EvolutionmdashAdvanced (LTE-A)rdquo Bahria University Journal of Information ampCommunication Technologies vol 8 no 1 2015
[35] G Piro L A Grieco G Boggia F Capozzi and P CamardaldquoSimulating LTE cellular systems an open-source frameworkrdquoIEEE Transactions on Vehicular Technology vol 60 no 2 pp498ndash513 2011
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014