Research ArticleData-Driven Handover Optimization in Next Generation MobileCommunication Networks

Po-Chiang Lin Lionel F Gonzalez Casanova and Bakary K S Fatty

Department of Communications Engineering Yuan Ze University Taoyuan 320 Taiwan

Correspondence should be addressed to Po-Chiang Lin pclinsaturnyzuedutw

Received 1 April 2016 Revised 3 July 2016 Accepted 5 July 2016

Academic Editor Young-June Choi

Copyright copy 2016 Po-Chiang Lin et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Network densification is regarded as one of the important ingredients to increase capacity for next generation mobilecommunication networks However it also leads to mobility problems since users are more likely to hand over to another cell indense or even ultradense mobile communication networks Therefore supporting seamless and robust connectivity through suchnetworks becomes a very important issue In this paper we investigate handover (HO) optimization in next generation mobilecommunication networks We propose a data-driven handover optimization (DHO) approach which aims to mitigate mobilityproblems including too-late HO too-early HO HO to wrong cell ping-pong HO and unnecessary HO The key performanceindicator (KPI) is defined as the weighted average of the ratios of these mobility problems The DHO approach collects data fromthe mobile communication measurement results and provides a model to estimate the relationship between the KPI and featuresfrom the collected dataset Based on the model the handover parameters including the handover margin and time-to-triggerare optimized to minimize the KPI Simulation results show that the proposed DHO approach could effectively mitigate mobilityproblems

1 Introduction

The first generation (1G) mobile communication systemsenabled the release from traditional wireline place-to-placecommunications to wireless person-to-person communica-tions As technologies and demands evolve now the fifthgeneration (5G) mobile communication systems aim toconnect anything to anything all over the world

One of the engineering requirements for 5G is to achieve1000x data rate [1] There are three categories of technologiesthat would be combined together to achieve the 1000xcapacity gain

(1) Network densification and traffic offloading to im-prove the area spectral efficiency

(2) Millimeter wave (mmWave) technologies and accessto unlicensed spectrum to increase bandwidth

(3) Massive Multiple-Input and Multiple-Output (mas-sive MIMO) Multiuser MIMO (MU-MIMO) andCoordinated Multipoint (CoMP) to increase spectralefficiency

Among these categories of technologies network den-sification and traffic offloading are expected to provide themajority (40x to 50x) of the required capacity gain Networkdensification is a straightforward but effective method toincrease the network capacity by making cell size smallerThe advantages of cell shrinking include frequency reuse andreduction of resource competition among users within eachcell

However network densification leads to mobility prob-lems since users are more likely to hand over to another cellin dense or even ultradense mobile communication net-works Therefore supporting seamless and robust connec-tivity through such networks becomes a very importantissue Moreover as the number of base stations increasesthe installation configuration and maintenance efforts alsoincrease Mitigating these efforts to decrease the capitalexpenditure (CAPEX) and operational expenditure (OPEX)is also a critical issue [2]

In next generation mobile communication networksoptimizing the handover (HO) parameters to improve thesystem performance is critical The objective of handover

Hindawi Publishing CorporationMobile Information SystemsVolume 2016 Article ID 2368427 11 pageshttpdxdoiorg10115520162368427

2 Mobile Information Systems

optimization an important part of the Self-Organizing Net-work (SON) is to provide fast and seamless handover fromone cell to another while simultaneously keeping networkmanagement simple [3 4] The main goals of handoveroptimization include minimizing call drops minimizingradio link failures (RLF)minimizing unnecessary handoversand minimizing idle mode problems

In this paper we investigate the handover optimizationproblem in next generation mobile communication net-works We propose a data-driven handover optimization(DHO) approach which aims to mitigate mobility problemsThe DHO approach collects data from the mobile communi-cation measurement results and provides a model to estimatethe relationship between the key performance indicator(KPI) and features from the collected dataset Based on themodel the handover parameters which include the handovermargin (HOM) and time-to-trigger (TTT) are optimized tominimize the KPI which is the weighted average of variousmobility problem ratios Simulation results show that theproposed DHO approach could effectively mitigate mobilityproblems

The major contributions of this paper are threefold

(1) We consider several features including the locationmoving speed and moving direction of mobile sta-tions HOM and TTTThe proposed approach is thuscontext-aware and adaptive to mobile communica-tion environments

(2) We classify the mobility problems into five differenttypes and design the mobility problem identificationmechanism based on the combination of three condi-tional tests

(3) We adopt the neural network model to estimate theKPI function and to optimize the handover parame-ters HOM and TTT

The rest of this paper is organized as follows In Section 2we present a review of related work The HO problemformulation is described in Section 3 In Section 4 we providethe detailed description of the proposed DHO approach Per-formance evaluation and discussions are given in Section 5Finally Section 6 concludes this paper

2 Related Work

There exist several works for the handover optimization in thelong-term evolution (LTE)mobile communication networksJansen et al proposed a self-optimizing algorithm that tunesthe handover parameters of an LTE base station in order todiminish negative effects such as call dropping and handoverfailures [5] Their algorithm picks the optimal hysteresis andtime-to-trigger combination for the current network statusThe authors also proposed a weighted performance basedhandover optimization algorithm (WPHPO) that tunes thehysteresis and time-to-trigger in iterative steps [6] Li etal proposed a dynamic hysteresis-adjusting (DHA) methodwhich uses the network-allowedmaximumRLF ratio as a keyindicator [7] Lobinger et al investigated the coordination ofhandover parameter optimization and load balance in LTE

SON [8] Capdevielle et al investigated the joint interferencemanagement and handover optimization in LTE small cellsnetworks [9]They focus on two challenges for LTE small celldeployment intercell interferencemanagement andmobilitymanagement

Several works focus on the heterogeneousinter-RAThandover Ali and Saquib investigated cellularwireless LANhandover analysis for different mobility models in orderto exploit the heterogeneity of the wireless environment[10] They provided a medium selection method for verticalhandover Awada et al investigated the cell-pair specificoptimization of the inter-RAT handover parameters in SON[11]They provided the configuration paradigms for the inter-RAT handover thresholds Giacomini and Agarwal use QoSreputation and GM(1 1) prediction to make decision onvertical handover [12]They built on a novel reputation basedvertical handover decision rating system in the handoverdecision making progress Lopez-Perez et al characterizedthe relation between handover failure and ping-pong ratesin a heterogeneous network scenario [13] Rath and Panwarproposed a new prefetch-based fast handover procedure thatis designed to overcome the higher latency caused by the useof the public internet to connect the femtocell base stationwith the mobile core network [14] Xenakis et al proposed anovel handover decision policy for the two-tier LTE networkwith the goal to reduce the power consumption at mobilestations [15]The proposedmethod can adapt to the handoverhysteresis margin with respect to the prescribed SINR targetand measurement results

Some works introduced new concepts and architecturesto enhance user mobility in mobile communication net-works Taleb et al investigated the support of highly mobileusers [16] They introduced a data anchor gateway relocationmethod based on the information including user mobilityhistory information and user activity patterns and proposeda handover management policy to select a target base stationImran et al investigated how to empower SON with bigdata for enabling 5G in [17] They first characterized bigdata in next generationmobile communication networks andthen provided challenges in SON for enabling 5G includingunderutilized intelligence need for self-coordination needfor more transparent SON scalability energy efficiencyand need for a paradigm shift from reactive to proactiveSON The authors also proposed a framework for big dataempowered SON (BSON) which takes on the challengesmentioned above In this paper we leverage from their basicidea and investigate theHOproblemwhich is one of themostimportant issues in SON

3 Problem Formulation

We formulate the HO problem as follows Suppose thatX = [X

119863X119872] is a feature vector which includes two types

of variables X119863

and X119872 To be specific X

119872denotes the

measurement result which includes the location of MSmoving speed of MS moving direction of MS and otheruseful counters and flags described in Section 41X

119863denotes

the decision variables of the HO problem which is the han-dover parameter vector with two elements HOM and TTT

Mobile Information Systems 3

The details of these two handover parameters are describedas follows

(i) Handover Margin (HOM) (Also Known as Hysteresis)HOM is defined as the threshold of the difference in Signal toInterference plusNoise Ratio (SINR) between the serving andthe target cells When the SINR from the target cell is HOMbetter than that from the serving cell the HO condition issatisfied

(ii) Time-to-Trigger (TTT) An HO request would not besent immediately when the corresponding HO conditionmentioned above is satisfied Instead theHOcondition has tobe fulfilled for a certain period denoted as TTTWhen theHOcondition keeps satisfied for TTT an HO request is triggered

The proper values of HOM and TTT are critical toseamless connectivity Large HOM and TTT values lead tomore stable behavior but they may delay the HO decisionsunnecessarily whichmay cause problems On the other handsmall HOM and TTT values avoid long delay to trigger HOrequest but they may cause ping-pong HO and unnecessaryHO

The HO problem is formulated as

X119863opt = argmin

X119863119884 (1)

where 119884 is the corresponding key performance indicator(KPI) which depends on the feature vector X = [X

119863X119872]

The KPI should be designed to measure the mitigation ofmobility problems

There are four main challenges in the HO problem

(1) There are various kinds of mobility problems in nextgeneration mobile communication networks [18]Identifying the mobility problems would be the firstpriority in the HO problem

(2) The design of the KPI needs comprehensive andcareful considerations for these mobility problems

(3) The KPI 119884 can be formulated as a function of X thatis 119884 = 119891(X) However the exact form of 119891(X) isunknownTherefore one of the challenges we take onin this work is to provide a good estimation (X) tothe function 119891(X) With a good estimate (X)we canmake predictions of the KPI 119884 at some points X = xwhere x is a vector of some specific values of X

(4) The decision variables X119863consist of two elements

HOM and TTT Mitigating the mobility problems byjoint-optimizing HOM and TTT would be anotherchallenge to the HO problem

In Section 4 we describe the proposed DHO approach todeal with these main challenges

4 Proposed DHO Approach

In this section we provide the detailed description of the pro-posed data-driven handover optimization (DHO) approachFigure 1 shows the system architecture of the proposed DHO

Data collection

Right data

Data analysis

Big data

Applying the HO

HO parameter

HO parameteroptimization

engine

KPI function

KPI functionoptimization estimation

Key performanceindicator

estimationengine

parameters

Mobilityproblem

identification

Mobilecommunication

network

Figure 1 System architecture of the proposed DHO approach

approach First measurement results are collected from themobile communication network The dataset collected in thedatabase includes the feature vector described in Section 3and various counters and flags that are used to identify themobility problemsThe details of these counters and flags areprovided in Section 41 The DHO approach then analyzesand processes the data to calculate themobility problem ratio(MPR) of each mobility problem and obtain the KPI whichis saved in the database denoted as ldquoright datardquo in Figure 1The KPI function estimation engine uses the processed ldquorightdatardquo to provide the KPI function estimate After that theHO parameter optimization engine uses the KPI functionestimate to optimize the HO parameters including HOM andTTT Finally the optimal values ofHOMandTTT are appliedto corresponding base stations

In the following paragraphs we provide the detaileddescription of each building block in Figure 1

41 Mobility Problem Identification There would be fivedifferent mobility problems (MPs) in next generation mobilecommunication networks

(1) Too-Late HO Too-late HO trigger timing leads to lowSINR level of the serving cell A radio link failure (RLF)between the mobile station (MS) and the serving bast station(BS) occurs and then the MS try to reestablish a connectionto a nearby BS

(2) Too-Early HO Too-early HO trigger timing leads to a lowSINR level of the target cell An RLF between the MS andthe target BS occurs and then the MS try to reestablish aconnection to the original serving BS

(3) HO toWrong Cell HO from a source cell A to a target cellB which provides unstable SINR leads to an RLF At the sametime if there is another suitable cell C for theMS to reconnectto after the RLF the MS reestablishes connection with cell Cthat is neither the serving cell A nor the target cell B In thissituation instead of the failed HO from cell A to cell B HOfrom cell A directly to cell C would be a better decision

4 Mobile Information Systems

Table 1 Mobility problem identification

Mobility problem Close to lastHO RLF status

New target =previousserving

Too-late HO N Y Donrsquot careToo-early HO Y Y YHO to wrong cell Y Y NPing-pong HO Y N YUnnecessary HO Y N N

(4) Ping-Pong HO Cell A hands over aMS to cell B and cell Bhands over the sameMS back to cell A shortly after Since datatransmission is temporarily blocked during the connectiontransferring to the target cell these two HOs despite bothbeing successful should be avoided

(5) Unnecessary HO Cell A hands over a MS to cell B andcell B hands over the same MS to another cell C shortlyafter Those two HOs despite both being successful can becombined into one HO from cell A directly to cell C in orderto avoid unnecessary blocking of data transmission duringthe connection transferring time

In order to identify the mobility problems mentionedabove we design the mobility problem identification criteriaas Table 1 shows The mobility problem identification criteriadepend on three conditional tests

(1) Close to last HOWhether the occurrence time of the currentHO eventis close to that of the previous HO event

(2) RLF statusIs there an RLF for the current connection

(3) New target = previous servingWhether the new target cell is the same as the previousserving one

Table 1 shows that each of the five mobility problemswould be identified based on the test results of the threeconditional tests mentioned above

In the proposed DHO approach we design the followingcounters and flags to identify the mobility problems

(i) lastHoCounter to indicate the time interval since lastHO

(ii) lastHoThreshold to define a threshold value com-bined with lastHoCounter to identify whether theoccurrence time of the current HO event is close tothat of the previous HO event

(iii) rlfStatus to indicate whether an RLF occurs for thecurrent connection

(iv) newTagertCellId to indicate the new target cell ID(v) previousServingCellId to indicate the previous serv-

ing cell ID combined with the newTagertCellId toindicate whether the new target cell is the same as theprevious serving one

42 Design of KPI In the HO problem the KPI 119884 is definedas theweighted average of the variousmobility problem ratios(MPRs) There are five different kinds of mobility problemsThe KPI 119884 is calculated as

119884 =sum5

119896=1119908119896119877119896 (X)

sum5

119896=1119908119896

(2)

where 119908119896is the weight of the mobility problem 119896 which is

determined by mobile network operators and 119877119896(X) is the

MPR of the mobility problem 119896 which is a function of XMPR 119877

119896is defined as the probability of the event where the

mobility problem (MP) 119896 happens Consider

119877119896 (X) = Pr (MP

119896) =

119873119896

119873total (3)

where119873119896denotes the number of MP

119896 and119873total denotes the

total number of HO Each BS uses a sliding time window tocollect the statistics of 119873

119896and 119873total in order to obtain each

MPR and the KPI 119884

43 KPI Function Estimation In order to solve the HOproblem as (1) shows some previous solutions focus onanalyzing and obtaining the KPI function 119884 = 119891(X) Whenthe analytical formula 119891(X) is known solving the equation120597119891120597X

119863= 0 would get the optimal handover parameters

HOM and TTT However note that the function 119891 consistsof complex joint effects of the feature vector X = [X

119863X119872]

Under these complex effects it is difficult to analyze andobtain an accurate KPI function

In this paper we use machine learning techniques to esti-mate the KPI function To be specific we use the multilayerperception (MLP) which is one of the most popular neuralnetworks to perform the KPI function estimation [19]

The MLP consists of an input layer one or more hiddenlayers and an output layer In this paper we adopt the MLPwith one hidden layer consisting of four hidden neuronsThefeature vector X is applied to the input layer and propagatesthrough the network layer by layer from left to right Eacharrow in this figure represents an adjustable synaptic weightFrom our comprehensive experiments this architecture isempirically sufficient to model the relationship between KPIand the feature vector whereas the other experiments withmore hidden neurons provide similar results A simulationresult about the mean-square-error (MSE) of the KPI func-tion estimation with various numbers of hidden neuronsis shown in Figure 2 The epoch limitation is set as 1000This figure shows that four hidden neurons are sufficient toimprove the performance ofmodeling to the degree of 1119864minus10whereas the performance of two neurons only reaches thedegree of 1119864 minus 6 Applying more hidden neurons does notsignificantly improve the MSE

In MLP the backpropagation (BP) algorithm is used toadjust the neural network parameters in order to minimizethe error between the target (measured) KPI and the responsefrom the function estimate The detailed operation of theBP algorithm includes the forward pass and the backwardpassWe provide the detailed descriptions of the forward passand the backward pass in the following paragraphs

Mobile Information Systems 5

4 6 8 102Number of hidden neurons

10minus11

10minus10

10minus9

10minus8

10minus7

10minus6

10minus5

MSE

Figure 2 MSE of the KPI function estimation with various numberof hidden neurons

431 Forward Pass The purpose of the forward pass is toget the actual response of the MLP in accordance with thespecified input signal In the forward pass the feature vectorX(119899) at iteration 119899 is applied to the input layer and propagatesthrough the network layer by layer from left to right Thetop part of Figure 3 shows the signal flow of the forwardpass 119883

1(119899) denotes the HOM and 119883

2(119899) denotes the TTT

1198833(119899) to 119883

119898(119899) denote the other features When the signal

propagates through a synapse its value is multiplied bythe synaptic weight The input of a neuron is the synapticweighted sum of the output of its previous layer For examplethe input of the 119895th neuron in the 119897th layer is

V(119897)119895(119899) =

119898119897minus1

sum119894=0

119908(119897)

119895119894(119899) 119910(119897minus1)

119894(119899) (4)

where 119894 = 0 represents the bias termThe input of this neuronis applied to the activation function 120593(sdot) The activationfunction should be differentiable everywhere We adopt thecommonly used sigmoid function as

119910(119897)

119895(119899) = 120593 (V(119897)

119895(119899)) =

1

1 + exp (minusV(119897)119895(119899))

(5)

The output of this neuron propagates through the network tothe next layer and the same operation is performed againFinally the signal aggregates in the output layer and theestimated KPI function (119899) is obtained During the forwardpass the synaptic weights are all fixed

432 Backward Pass The purpose of the backward pass isto adjust the synaptic weights in order to minimize the errorbetween the actual response (119899) and the desired response119884(119899) The error signal 119890(119899) is defined by

119890 (119899) = 119884 (119899) minus (119899) (6)

The instantaneous square error 119864(119899) is defined as

119864 (119899) =1

21198902(119899) =

1

2(119884 (119899) minus (119899))

2

(7)

The goal of the back propagation algorithm is to minimize119864(119899) To achieve this goal the back propagation algorithmapplies a respective correction Δ120596

(119897)

119895119894(119899) to every synaptic

weight 120596(119897)

119895119894(119899) The correction Δ120596

(119897)

119895119894(119899) is designed to be

proportional to the gradient 120597119864(119899)120597120596(119897)119895119894(119899) that is

Δ120596(119897)

119895119894(119899) = minus120578

119886

120597119864 (119899)

120597120596(119897)

119895119894(119899)

(8)

The minus sign comes from the fact that lower 119864(119899) ispreferred 120578

119886is the adaptive learning rate proposed by Behera

et al in [20] which is expressed as

120578119886= 120583

100381710038171003817100381710038171003817100381710038172

1003817100381710038171003817J11987910038171003817100381710038172 (9)

where = 119884(119899) minus (119899) and J = 120597120597W119897 While consid-

ering the weight correction for 119897 = 2 that is the outputlayer according to the chain rule of calculus the gradient120597119864(119899)120597120596

(2)

1119894(119899) can be written as

120597119864 (119899)

120597120596(2)

1119894(119899)

=120597119864 (119899)

120597119890 (119899)

120597119890 (119899)

120597 (119899)

120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597120596(2)

1119894(119899)

(10)

Applying (4)ndash(7) to the four partial derivatives at the right ofthe equal sign in (10) it can be rewritten as

120597119864 (119899)

120597120596(2)

1119894(119899)

= minus119890 (119899) 1205931015840(V(2)1

(119899)) 119910(1)

119894(119899)

= minus120575(2)

1(119899) 119910(1)

119894(119899)

(11)

where 120575(119897)119895(119899) is the local gradient which is defined as

120575(119897)

119895(119899) = minus

120597119864 (119899)

120597V(119897)119895(119899)

(12)

While considering the weight correction for 119897 = 1 that isthe hidden layer by iterative backpropagation and chain rulemanners the local gradient 120575(1)

119895(119899) is

120575(1)

119895(119899) = 120593

1015840(V(1)119895

(119899)) 120575(2)

1(119899) 120596(2)

1119895(119899) (13)

The derivation of the local gradients at each layer is shown inthe bottom part of Figure 3 Finally every synaptic weight iscorrected following this formula

120596(119897)

119895119894(119899 + 1) = 120596

(119897)

119895119894(119899) + Δ120596

(119897)

119895119894(119899) (14)

The forward pass and the backward pass are executed iter-atively until the square error 119864(119899) reaches some satisfactorylevel or when the number of epoch exceeds some thresholdAfter that the handover parameter optimization is executed

6 Mobile Information Systems

X0(n) = +1

X1(n)

X2(n)

X3(n)

120596(1)10 (n)

120596(1)20 (n)

120596(1)30 (n)

120596(1)40 (n)

120596(1)1m(n)

120596(1)2m(n)

120596(1)3m(n)

120596(1)4m(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120575(1)1 (n) 120575(1)

2 (n) 120575(1)3 (n) 120575(1)

4 (n)

+1

minus1

120575(2)1 (n)

e1(n)

120596(2)10 (n)

120596(2)11 (n)

120596(2)12 (n)

120596(2)13 (n)

120596(2)14 (n)

120596(2)11 (n)

120596(2)12 (n)

120596(2)13 (n)

120596(2)14 (n)

(2)1 (n)

Y(n)

Y(n) = y(2)1 (n)

Xm(n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

(n)119831D

(n)119831M

Figure 3 The forward pass and the backward pass of the BP algorithm

44 Handover Parameter Optimization In this paper thehandover optimization problem is solved by the gradientdescent algorithm [21] Here the handover parameters X

119863

including HOM and TTT are adjusted adaptively To find thelocal minimum of the KPI function 119884 asymp (X) using gradientdescent one takes steps proportional to the gradient Supposethat at the 119899th time of adjustment the handover parametersareX119863(119899) and the KPI is119884(119899) At the next time of adjustment

the handover parameters X119863are set as

X119863 (119899 + 1) = X

119863 (119899) + ΔX119863 (119899) (15)

ΔX119863(119899) depends on the gradient of estimated KPI 119884(119899) with

respect to X119863(119899) that is

ΔX119863 (119899) = 120583

120597119884 (119899)

120597X119863 (119899)

(16)

where 120583 is the constant adjustment rateHow to obtain the gradient 120597119884120597X

119863is a problem leftThe

original BP algorithmdescribed in Section 43 only consists oftwo passes the forward pass and the backward pass In orderto obtain the gradient 120597119884120597X

119863 the proposed DHO approach

modifies the original BP algorithm to add the third pass theadjustment pass We provide the detailed description of theadjustment pass in the following paragraphs

The purpose of the adjustment pass is to obtain theestimated gradient 120597(119899)120597X

119863(119899) which is the key to adjust

the handover parameter in order to minimize the KPI Thesynaptic weights that have been adjusted well in the backwardpass in accordance with (14) are set as fixed in the adjustmentpass In the following we show how to obtain the estimatedgradient 120597(119899)120597X

119863(119899) The signal flow of the adjustment

pass is depicted in Figure 4First of all we define the local gradient 120582(119897)

119895(119899) of the

adjustment pass as

120582(119897)

119895(119899) =

120597 (119899)

120597V(119897)119895(119899)

(17)

This local gradient 120582(119897)119895(119899) is similar to the local gradient

120575(119897)

119895(119899) of the backward pass which is defined in (12) except

that the target of the partial derivative is changed to (119899)While considering the output layer the local gradient 120582(2)

1(119899)

is

120582(2)

1(119899) =

120597 (119899)

120597V(2)1

(119899)= 1205931015840(V(2)1

(119899)) (18)

Mobile Information Systems 7

Y(n) = y(2)1 (n)

120582(1)1 (n) 120582(1)

2 (n) 120582(1)3 (n) 120582(1)

4 (n) 120582(2)1 (n)

120597Y(n)

120597X1(n)

120597Y(n)

120597X2(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

+1

(2)1 (n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

X0(n) = +1

X1(n)

X2(n)

X3(n)

Xm(n)

120596(1)10 (n + 1)

120596(1)20 (n + 1)

120596(1)30 (n + 1)

120596(1)40 (n + 1)

120596(1)1m(n + 1)

120596(1)2m(n + 1)

120596(1)3m(n + 1)

120596(1)4m(n + 1)

120596(1)11 (n + 1)

120596(1)12 (n + 1)

120596(1)21 (n + 1)

120596(1)22 (n + 1)

120596(1)31 (n + 1)

120596(1)32 (n + 1)

120596(1)41 (n + 1)

120596(1)42 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

120596(2)10 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

(n)119831D

(n)119831M

Figure 4 Adjustment pass of the proposed DHO approach

While considering the hidden layer the local gradient 120582(1)119895(119899)

is

120582(1)

119895(119899) =

120597 (119899)

120597V(1)119895

(119899)=

120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119910(1)

119895(119899)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

= 120582(2)

1(119899) 120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899))

(19)

Using the results from (18) and (19) the gradient can bewritten as follows

120597 (119899)

120597119883119894 (119899)

=120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

= 120582(2)

1(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

(20)

where the last equality comes from (18) Since

V(2)1

(119899) =

4

sum119895=0

120596(2)

1119895(119899 + 1) 119910

(1)

119895(119899) (21)

the second term at the most right of (20) can be written as

120597V(2)1

(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(n)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899)) 120596(1)

1198951(119899 + 1)

(22)

where the third equality comes from the fact that 119910(1)0(119899) is

always equal to 1 and its partial derivative can be ignoredfrom the summation Combining (19) and (22) into (20) thegradient 120597(119899)120597119883

119894(119899) is obtained as

120597 (119899)

120597119883119894 (119899)

=

4

sum119895=1

120582(1)

119895(119899) 120596(1)

1198951(119899 + 1) (23)

The derivation of the local gradients at each layer and thegradient 120597(119899)120597119883

119894(119899) is depicted in Figure 4 Based on the

8 Mobile Information Systems

05 1 15 2 25 3 35 4 45 50X-coordinate (km)

0

05

1

15

2

25

3

35

4

45

5

Y-c

oord

inat

e (km

)

Figure 5 Topology of the simulated mobile communication net-work

result from (23) the handover parameter119883119894is adjusted as (15)

and (16) showIn order to cope with the time-varying channel condi-

tions we use online training and operate the three passesiteratively to model the instantaneous relationship 119884 = 119891(X)and optimize the handover parameters HOM and TTT

5 Performance Evaluation and Discussion

In this section we describe the simulation environmentsimulation results and discussions

In our simulation a mobile communication networkwhich consists of 100 small cells is built in a 5 km times 5 kmarea Figure 5 shows the coordinates of these small cell basestations Each cell has 5MHz bandwidth with 25 resourceblocks and 2GHz carrier frequency Each resource blockconsists of 12 subcarriers of size 15 kHz each A time slot is05ms in duration and the transmission time interval (TTI)is 1ms

In the simulation 100 MSs are uniformly distributedover the area with random initialized positions The mobilitytrace of these 100 MSs is generated by using the Manhattanmobility model which is widely used for urban areas andsuitable to implement realistic simulations [22ndash24] In theManhattanmobility model MSs move in either horizontal orvertical directions on an urban map At each intersection anMS employs a probabilistic approach to choose the directionof its movements The probabilities of going straight turningleft and turning right are 05 025 and 025 respectivelyThe blue line in Figure 5 shows an illustrative moving pathof an MS in the mobile communication network Whileconsidering the KPI the weights of mobility problems are setto be equal

An RLF will occur if the SINR level of the servingcell drops below minus65 dB before handover execution is

completed [18] We set this as our SINR threshold valuesuch that if the SINR from the source BS is lower than thethreshold an RLF will occur The simulation time is set as5000 seconds The operating point with 0 dB HOM and 0 sTTT is the starting operating point for all cells in the network

In this paper we use LTE-Sim an open source frameworkto simulate LTE networks as our basic simulation tool[25] The design and development of LTE-Sim support thefollowing aspects

(i) The Evolved Universal Terrestrial Radio Access (E-UTRAN) and the Evolved Packet System (EPS)

(ii) Single and heterogeneous multicell environments(iii) Multiusers environment(iv) User mobility(v) Handover procedures(vi) Frequency reuse techniques(vii) Quality-of-service (QoS) management(viii) Scheduling strategies

However LTE-Sim still lacks some important functional-ities and features that are required in this workTherefore wemodified LTE-Sim from several aspects Our modification toLTE-Sim includes the following

(1) Add

(i) radio link failure(ii) handover failure(iii) HOM and TTT(iv) proposed DHO approach

(2) Modify

(i) mobility pattern using the seed to generate thesame sequence of random numbers each timewe run the simulation

The simulation parameter values are shown in Table 2unless otherwise specified

Figure 6 shows the simulation result when the TX poweris set as 20 dBm The horizontal axis denotes the mobilityproblem index Indexes 1 to 5 mean the mobility problem oftoo-late HO too-early HO HO to wrong cell ping-pongHOand unnecessary HO respectively Index 6 means the KPIwhich is the weighted average of the five MPRs The verticalaxismeans the corresponding values ofMPR orKPIThe bluebars show the performance for a static HO parameter settingwith 0 dB HOM and 0 s TTT and the green ones for anotherstatic setting with 10 dBHOMand 512 s TTTThe yellow barsshow the performance of the proposed DHO approach Theresults shown in Figure 6 validate that the proposed DHOapproach can effectively mitigate the mobility problems andachieve better MPRs and KPI

Figure 7 shows the simulation result when the TX poweris set as 43 dBm Since the TX power is higher than the onein the previous scenario in Figure 6 it is less likely that an

Mobile Information Systems 9

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

Figure 6 Simulation result when the TX power is set as 20 dBm

Table 2 Simulation parameters

Parameter ValuesNumber of small cell base stations 100Number of mobile stations 100eNB Tx power 15 dBm20 dBm43 dBmCarrier frequency 2GHzMS speed 30 kmhMobility model ManhattanTTI 1msSubcarrier spacing 15 KHzResource block 180KHzSuper frame time 10msNoise figure 25Noise spectral density minus174 dBmSINR threshold minus65 dBSimulation time 5000 sHandover delay 30msSystem bandwidth 5MHz 25 RBsTTI

RLF occurs Therefore the MPRs and KPI in Figure 7 arelower than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

Figure 8 shows the simulation result when the TX poweris set as 15 dBm Since the TX power is lower than the onein the previous scenario in Figure 6 it is more likely that anRLF occurs Therefore the MPRs and KPI in Figure 8 arehigher than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

We compare the KPI improvement under the variousvalues of TX power The KPI improvement is calculatedas the difference between the initial-case KPI (HOM 0dB

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 7 Simulation result when the TX power is set as 43 dBm

DHO approach

2 3 4 5 61Mobility problem index

0

001

002

003

004

005

006

007

008

MPR

KPI

HOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 8 Simulation result when the TX power is set as 15 dBm

TTT 0 s) and the KPI of the proposed DHO methoddivided by the initial-case KPI Figure 9 shows that the KPIimprovement ranges from 15 to 20 It also shows that theKPI improvement is more significant when the TX power islower that is the transmission range of the base station issmaller

6 Conclusion

In this work we investigate the mobility problems innext generation mobile communication networks There aretotally five different mobility problems each with different

10 Mobile Information Systems

20 4315TX power (dBm)

0

5

10

15

20

25

KPI i

mpr

ovem

ent (

)

Figure 9 KPI improvement with various TX power

characteristics We propose a data-driven handover opti-mization (DHO) approach which aims to mitigate mobilityproblems The DHO approach consists of four parts

(1) Mobility problem identification(2) KPI design(3) KPI function estimation(4) Handover parameter optimization

The DHO approach collects data from the mobile com-munication measurement results and provides a model toestimate the relationship between the KPI and features fromthe collected dataset Based on the model the handoverparameters including the HOM and TTT are optimized tominimize the KPI which is a weighted average of differentmobility problem ratios Simulation results show that theproposed DHO approach could effectively mitigate mobilityproblems and provide better connectivity

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work was supported in part by Ministry of Scienceand Technology (MOST) Taipei Taiwan under Grants nosMOST 103-2221-E-155-005 MOST 104-2221-E-155-017 andMOST 105-2218-E-155-001

References

[1] J G Andrews S Buzzi W Choi et al ldquoWhat will 5G berdquo IEEEJournal on Selected Areas in Communications vol 32 no 6 pp1065ndash1082 2014

[2] K Kitagawa T Komine T Yamamoto and S Konishi ldquoAhandover optimization algorithm with mobility robustness forLTE systemsrdquo in Proceedings of the IEEE 22nd International

Symposium on Personal Indoor and Mobile Radio Communica-tions (PIMRC rsquo11) pp 1647ndash1651 Toronto Canada September2011

[3] 3GPP ldquoSelf-configuring and self-optimizing network (SON)usecases and solutionsrdquo TR 36902 V931 2011

[4] A Awada BWegmann I Viering and A Klein ldquoA SON-basedalgorithm for the optimization of inter-RAT handover param-etersrdquo IEEE Transactions on Vehicular Technology vol 62 no 5pp 1906ndash1923 2013

[5] T Jansen I Balan J Turk I Moerman and T Kurner ldquoHando-ver parameter optimization in LTE self-organizing networksrdquoin Proceedings of the IEEE 72nd Vehicular Technology Confer-ence Fall (VTC 2010-Fall rsquo10) pp 1ndash5 IEEE Ottawa CanadaSeptember 2010

[6] T Jansen I Balan S Stefanski I Moerman and T KurnerldquoWeighted performance based handover parameter optimiza-tion in LTErdquo in Proceedings of the IEEE 73rd Vehicular Technol-ogy Conference (VTC rsquo11) pp 1ndash5 May 2011

[7] W Li X Duan S Jia L Zhang Y Liu and J Lin ldquoA dynamichysteresis-adjusting algorithm in LTE self-organization net-worksrdquo in Proceedings of the IEEE 75th Vehicular TechnologyConference (VTC Spring rsquo12) pp 1ndash5 Yokohama Japan May2012

[8] A Lobinger S Stefanski T Jansen and I Balan ldquoCoordinatinghandover parameter optimization and load balancing in LTEself-optimizing networksrdquo in Proceedings of the IEEE 73rdVehicular Technology Conference (VTC rsquo11) pp 1ndash5 May 2011

[9] V Capdevielle A Feki and E Sorsy ldquoJoint interferencemanagement and handover optimization in LTE small cellsnetworkrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6769ndash6773 IEEE OttawaCanada June 2012

[10] T Ali and M Saquib ldquoWLANcellular handover analysis fordifferent mobility modelsrdquo in Proceedings of the IEEE Interna-tional Conference on Communications (ICC rsquo12) pp 2758ndash2762Ottawa Canada June 2012

[11] A Awada B Wegmann I Viering and A Klein ldquoCell-pairspecific optimization of the inter-RAT handover parameters inSONrdquo in Proceedings of the IEEE 23rd International Symposiumon Personal Indoor andMobile Radio Communications (PIMRCrsquo12) pp 1168ndash1173 Sydney Australia September 2012

[12] D Giacomini and A Agarwal ldquoVertical handover decisionmaking using QoS reputation and GM(11) predictionrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo12) pp 5655ndash5659 IEEE Ottawa Canada June2012

[13] D Lopez-Perez I Guvenc and X Chu ldquoTheoretical analysisof handover failure and ping-pong rates for heterogeneousnetworksrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6774ndash6779 IEEE OttawaCanada June 2012

[14] A Rath and S Panwar ldquoFast handover in cellular networkswith femtocellsrdquo in Proceedings of the International Conferenceon Communications (ICC rsquo12) pp 2752ndash2757 IEEE OttawaCanada June 2012

[15] D Xenakis N Passas and C Verikoukis ldquoA novel handoverdecision policy for reducing power transmissions in the two-tier LTE networkrdquo in Proceedings of the IEEE International Con-ference on Communications (ICC rsquo12) pp 1352ndash1356 OttawaCanada June 2012

[16] T Taleb K Samdanis and A Ksentini ldquoSupporting highlymobile users in cost-effective decentralized mobile operator

Mobile Information Systems 11

networksrdquo IEEE Transactions on Vehicular Technology vol 63no 7 pp 3381ndash3396 2014

[17] A Imran A Zoha and A Abu-Dayya ldquoChallenges in 5G howto empower SONwith big data for enabling 5Grdquo IEEE Networkvol 28 no 6 pp 27ndash33 2014

[18] H-D Bae B Ryu and N-H Park ldquoAnalysis of handoverfailures in LTE femtocell systemsrdquo in Proceedings of the Aus-tralasian Telecommunication Networks and Applications Con-ference (ATNAC rsquo11) pp 1ndash5 IEEE Melbourne AustraliaNovember 2011

[19] S Haykin Neural Networks A Comprehensive FoundationPrentice-Hall 1999

[20] L Behera S Kumar and A Patnaik ldquoOn adaptive learning ratethat guarantees convergence in feedforward networksrdquo IEEETransactions on Neural Networks vol 17 no 5 pp 1116ndash11252006

[21] J A Snyman Practical Mathematical Optimization An Intro-duction to Basic Optimization Theory and Classical and NewGradient-Based Algorithms vol 97 of Applied OptimizationSpringer New York NY USA 2005

[22] M Gudmundson ldquoCell planning in Manhattan environmentsrdquoin Proceedings of the 42nd Vehicular Technology Conference pp435ndash438 IEEE Denver Colo USA May 1992

[23] P-E Ostling ldquoImplications of cell planning on handoff per-formance in manhattan environmentsrdquo in Proceedings of the5th IEEE International Symposium on Personal Indoor andMobile Radio Communications Wireless NetworksmdashCatchingthe Mobile Future vol 2 pp 625ndash629 The Hague The Nether-lands September 1994

[24] M Karabacak D Wang H Ishii and H Arslan ldquoMobilityperformance of macrocell-assisted small cells in Manhattanmodelrdquo in Proceedings of the 79th IEEE Vehicular TechnologyConference (VTC-Spring rsquo14) pp 1ndash5 Seoul South Korea May2014

[25] 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

optimization an important part of the Self-Organizing Net-work (SON) is to provide fast and seamless handover fromone cell to another while simultaneously keeping networkmanagement simple [3 4] The main goals of handoveroptimization include minimizing call drops minimizingradio link failures (RLF)minimizing unnecessary handoversand minimizing idle mode problems

In this paper we investigate the handover optimizationproblem in next generation mobile communication net-works We propose a data-driven handover optimization(DHO) approach which aims to mitigate mobility problemsThe DHO approach collects data from the mobile communi-cation measurement results and provides a model to estimatethe relationship between the key performance indicator(KPI) and features from the collected dataset Based on themodel the handover parameters which include the handovermargin (HOM) and time-to-trigger (TTT) are optimized tominimize the KPI which is the weighted average of variousmobility problem ratios Simulation results show that theproposed DHO approach could effectively mitigate mobilityproblems

The major contributions of this paper are threefold

(1) We consider several features including the locationmoving speed and moving direction of mobile sta-tions HOM and TTTThe proposed approach is thuscontext-aware and adaptive to mobile communica-tion environments

(2) We classify the mobility problems into five differenttypes and design the mobility problem identificationmechanism based on the combination of three condi-tional tests

(3) We adopt the neural network model to estimate theKPI function and to optimize the handover parame-ters HOM and TTT

The rest of this paper is organized as follows In Section 2we present a review of related work The HO problemformulation is described in Section 3 In Section 4 we providethe detailed description of the proposed DHO approach Per-formance evaluation and discussions are given in Section 5Finally Section 6 concludes this paper

2 Related Work

There exist several works for the handover optimization in thelong-term evolution (LTE)mobile communication networksJansen et al proposed a self-optimizing algorithm that tunesthe handover parameters of an LTE base station in order todiminish negative effects such as call dropping and handoverfailures [5] Their algorithm picks the optimal hysteresis andtime-to-trigger combination for the current network statusThe authors also proposed a weighted performance basedhandover optimization algorithm (WPHPO) that tunes thehysteresis and time-to-trigger in iterative steps [6] Li etal proposed a dynamic hysteresis-adjusting (DHA) methodwhich uses the network-allowedmaximumRLF ratio as a keyindicator [7] Lobinger et al investigated the coordination ofhandover parameter optimization and load balance in LTE

SON [8] Capdevielle et al investigated the joint interferencemanagement and handover optimization in LTE small cellsnetworks [9]They focus on two challenges for LTE small celldeployment intercell interferencemanagement andmobilitymanagement

Several works focus on the heterogeneousinter-RAThandover Ali and Saquib investigated cellularwireless LANhandover analysis for different mobility models in orderto exploit the heterogeneity of the wireless environment[10] They provided a medium selection method for verticalhandover Awada et al investigated the cell-pair specificoptimization of the inter-RAT handover parameters in SON[11]They provided the configuration paradigms for the inter-RAT handover thresholds Giacomini and Agarwal use QoSreputation and GM(1 1) prediction to make decision onvertical handover [12]They built on a novel reputation basedvertical handover decision rating system in the handoverdecision making progress Lopez-Perez et al characterizedthe relation between handover failure and ping-pong ratesin a heterogeneous network scenario [13] Rath and Panwarproposed a new prefetch-based fast handover procedure thatis designed to overcome the higher latency caused by the useof the public internet to connect the femtocell base stationwith the mobile core network [14] Xenakis et al proposed anovel handover decision policy for the two-tier LTE networkwith the goal to reduce the power consumption at mobilestations [15]The proposedmethod can adapt to the handoverhysteresis margin with respect to the prescribed SINR targetand measurement results

Some works introduced new concepts and architecturesto enhance user mobility in mobile communication net-works Taleb et al investigated the support of highly mobileusers [16] They introduced a data anchor gateway relocationmethod based on the information including user mobilityhistory information and user activity patterns and proposeda handover management policy to select a target base stationImran et al investigated how to empower SON with bigdata for enabling 5G in [17] They first characterized bigdata in next generationmobile communication networks andthen provided challenges in SON for enabling 5G includingunderutilized intelligence need for self-coordination needfor more transparent SON scalability energy efficiencyand need for a paradigm shift from reactive to proactiveSON The authors also proposed a framework for big dataempowered SON (BSON) which takes on the challengesmentioned above In this paper we leverage from their basicidea and investigate theHOproblemwhich is one of themostimportant issues in SON

3 Problem Formulation

We formulate the HO problem as follows Suppose thatX = [X

119863X119872] is a feature vector which includes two types

of variables X119863

and X119872 To be specific X

119872denotes the

measurement result which includes the location of MSmoving speed of MS moving direction of MS and otheruseful counters and flags described in Section 41X

119863denotes

the decision variables of the HO problem which is the han-dover parameter vector with two elements HOM and TTT

Mobile Information Systems 3

The details of these two handover parameters are describedas follows

(i) Handover Margin (HOM) (Also Known as Hysteresis)HOM is defined as the threshold of the difference in Signal toInterference plusNoise Ratio (SINR) between the serving andthe target cells When the SINR from the target cell is HOMbetter than that from the serving cell the HO condition issatisfied

(ii) Time-to-Trigger (TTT) An HO request would not besent immediately when the corresponding HO conditionmentioned above is satisfied Instead theHOcondition has tobe fulfilled for a certain period denoted as TTTWhen theHOcondition keeps satisfied for TTT an HO request is triggered

The proper values of HOM and TTT are critical toseamless connectivity Large HOM and TTT values lead tomore stable behavior but they may delay the HO decisionsunnecessarily whichmay cause problems On the other handsmall HOM and TTT values avoid long delay to trigger HOrequest but they may cause ping-pong HO and unnecessaryHO

The HO problem is formulated as

X119863opt = argmin

X119863119884 (1)

where 119884 is the corresponding key performance indicator(KPI) which depends on the feature vector X = [X

119863X119872]

The KPI should be designed to measure the mitigation ofmobility problems

There are four main challenges in the HO problem

(1) There are various kinds of mobility problems in nextgeneration mobile communication networks [18]Identifying the mobility problems would be the firstpriority in the HO problem

(2) The design of the KPI needs comprehensive andcareful considerations for these mobility problems

(3) The KPI 119884 can be formulated as a function of X thatis 119884 = 119891(X) However the exact form of 119891(X) isunknownTherefore one of the challenges we take onin this work is to provide a good estimation (X) tothe function 119891(X) With a good estimate (X)we canmake predictions of the KPI 119884 at some points X = xwhere x is a vector of some specific values of X

(4) The decision variables X119863consist of two elements

HOM and TTT Mitigating the mobility problems byjoint-optimizing HOM and TTT would be anotherchallenge to the HO problem

In Section 4 we describe the proposed DHO approach todeal with these main challenges

4 Proposed DHO Approach

In this section we provide the detailed description of the pro-posed data-driven handover optimization (DHO) approachFigure 1 shows the system architecture of the proposed DHO

Data collection

Right data

Data analysis

Big data

Applying the HO

HO parameter

HO parameteroptimization

engine

KPI function

KPI functionoptimization estimation

Key performanceindicator

estimationengine

parameters

Mobilityproblem

identification

Mobilecommunication

network

Figure 1 System architecture of the proposed DHO approach

approach First measurement results are collected from themobile communication network The dataset collected in thedatabase includes the feature vector described in Section 3and various counters and flags that are used to identify themobility problemsThe details of these counters and flags areprovided in Section 41 The DHO approach then analyzesand processes the data to calculate themobility problem ratio(MPR) of each mobility problem and obtain the KPI whichis saved in the database denoted as ldquoright datardquo in Figure 1The KPI function estimation engine uses the processed ldquorightdatardquo to provide the KPI function estimate After that theHO parameter optimization engine uses the KPI functionestimate to optimize the HO parameters including HOM andTTT Finally the optimal values ofHOMandTTT are appliedto corresponding base stations

In the following paragraphs we provide the detaileddescription of each building block in Figure 1

41 Mobility Problem Identification There would be fivedifferent mobility problems (MPs) in next generation mobilecommunication networks

(1) Too-Late HO Too-late HO trigger timing leads to lowSINR level of the serving cell A radio link failure (RLF)between the mobile station (MS) and the serving bast station(BS) occurs and then the MS try to reestablish a connectionto a nearby BS

(2) Too-Early HO Too-early HO trigger timing leads to a lowSINR level of the target cell An RLF between the MS andthe target BS occurs and then the MS try to reestablish aconnection to the original serving BS

(3) HO toWrong Cell HO from a source cell A to a target cellB which provides unstable SINR leads to an RLF At the sametime if there is another suitable cell C for theMS to reconnectto after the RLF the MS reestablishes connection with cell Cthat is neither the serving cell A nor the target cell B In thissituation instead of the failed HO from cell A to cell B HOfrom cell A directly to cell C would be a better decision

4 Mobile Information Systems

Table 1 Mobility problem identification

Mobility problem Close to lastHO RLF status

New target =previousserving

Too-late HO N Y Donrsquot careToo-early HO Y Y YHO to wrong cell Y Y NPing-pong HO Y N YUnnecessary HO Y N N

(4) Ping-Pong HO Cell A hands over aMS to cell B and cell Bhands over the sameMS back to cell A shortly after Since datatransmission is temporarily blocked during the connectiontransferring to the target cell these two HOs despite bothbeing successful should be avoided

(5) Unnecessary HO Cell A hands over a MS to cell B andcell B hands over the same MS to another cell C shortlyafter Those two HOs despite both being successful can becombined into one HO from cell A directly to cell C in orderto avoid unnecessary blocking of data transmission duringthe connection transferring time

In order to identify the mobility problems mentionedabove we design the mobility problem identification criteriaas Table 1 shows The mobility problem identification criteriadepend on three conditional tests

(1) Close to last HOWhether the occurrence time of the currentHO eventis close to that of the previous HO event

(2) RLF statusIs there an RLF for the current connection

(3) New target = previous servingWhether the new target cell is the same as the previousserving one

Table 1 shows that each of the five mobility problemswould be identified based on the test results of the threeconditional tests mentioned above

In the proposed DHO approach we design the followingcounters and flags to identify the mobility problems

(i) lastHoCounter to indicate the time interval since lastHO

(ii) lastHoThreshold to define a threshold value com-bined with lastHoCounter to identify whether theoccurrence time of the current HO event is close tothat of the previous HO event

(iii) rlfStatus to indicate whether an RLF occurs for thecurrent connection

(iv) newTagertCellId to indicate the new target cell ID(v) previousServingCellId to indicate the previous serv-

ing cell ID combined with the newTagertCellId toindicate whether the new target cell is the same as theprevious serving one

42 Design of KPI In the HO problem the KPI 119884 is definedas theweighted average of the variousmobility problem ratios(MPRs) There are five different kinds of mobility problemsThe KPI 119884 is calculated as

119884 =sum5

119896=1119908119896119877119896 (X)

sum5

119896=1119908119896

(2)

where 119908119896is the weight of the mobility problem 119896 which is

determined by mobile network operators and 119877119896(X) is the

MPR of the mobility problem 119896 which is a function of XMPR 119877

119896is defined as the probability of the event where the

mobility problem (MP) 119896 happens Consider

119877119896 (X) = Pr (MP

119896) =

119873119896

119873total (3)

where119873119896denotes the number of MP

119896 and119873total denotes the

total number of HO Each BS uses a sliding time window tocollect the statistics of 119873

119896and 119873total in order to obtain each

MPR and the KPI 119884

43 KPI Function Estimation In order to solve the HOproblem as (1) shows some previous solutions focus onanalyzing and obtaining the KPI function 119884 = 119891(X) Whenthe analytical formula 119891(X) is known solving the equation120597119891120597X

119863= 0 would get the optimal handover parameters

HOM and TTT However note that the function 119891 consistsof complex joint effects of the feature vector X = [X

119863X119872]

Under these complex effects it is difficult to analyze andobtain an accurate KPI function

In this paper we use machine learning techniques to esti-mate the KPI function To be specific we use the multilayerperception (MLP) which is one of the most popular neuralnetworks to perform the KPI function estimation [19]

The MLP consists of an input layer one or more hiddenlayers and an output layer In this paper we adopt the MLPwith one hidden layer consisting of four hidden neuronsThefeature vector X is applied to the input layer and propagatesthrough the network layer by layer from left to right Eacharrow in this figure represents an adjustable synaptic weightFrom our comprehensive experiments this architecture isempirically sufficient to model the relationship between KPIand the feature vector whereas the other experiments withmore hidden neurons provide similar results A simulationresult about the mean-square-error (MSE) of the KPI func-tion estimation with various numbers of hidden neuronsis shown in Figure 2 The epoch limitation is set as 1000This figure shows that four hidden neurons are sufficient toimprove the performance ofmodeling to the degree of 1119864minus10whereas the performance of two neurons only reaches thedegree of 1119864 minus 6 Applying more hidden neurons does notsignificantly improve the MSE

In MLP the backpropagation (BP) algorithm is used toadjust the neural network parameters in order to minimizethe error between the target (measured) KPI and the responsefrom the function estimate The detailed operation of theBP algorithm includes the forward pass and the backwardpassWe provide the detailed descriptions of the forward passand the backward pass in the following paragraphs

Mobile Information Systems 5

4 6 8 102Number of hidden neurons

10minus11

10minus10

10minus9

10minus8

10minus7

10minus6

10minus5

MSE

Figure 2 MSE of the KPI function estimation with various numberof hidden neurons

431 Forward Pass The purpose of the forward pass is toget the actual response of the MLP in accordance with thespecified input signal In the forward pass the feature vectorX(119899) at iteration 119899 is applied to the input layer and propagatesthrough the network layer by layer from left to right Thetop part of Figure 3 shows the signal flow of the forwardpass 119883

1(119899) denotes the HOM and 119883

2(119899) denotes the TTT

1198833(119899) to 119883

119898(119899) denote the other features When the signal

propagates through a synapse its value is multiplied bythe synaptic weight The input of a neuron is the synapticweighted sum of the output of its previous layer For examplethe input of the 119895th neuron in the 119897th layer is

V(119897)119895(119899) =

119898119897minus1

sum119894=0

119908(119897)

119895119894(119899) 119910(119897minus1)

119894(119899) (4)

where 119894 = 0 represents the bias termThe input of this neuronis applied to the activation function 120593(sdot) The activationfunction should be differentiable everywhere We adopt thecommonly used sigmoid function as

119910(119897)

119895(119899) = 120593 (V(119897)

119895(119899)) =

1

1 + exp (minusV(119897)119895(119899))

(5)

The output of this neuron propagates through the network tothe next layer and the same operation is performed againFinally the signal aggregates in the output layer and theestimated KPI function (119899) is obtained During the forwardpass the synaptic weights are all fixed

432 Backward Pass The purpose of the backward pass isto adjust the synaptic weights in order to minimize the errorbetween the actual response (119899) and the desired response119884(119899) The error signal 119890(119899) is defined by

119890 (119899) = 119884 (119899) minus (119899) (6)

The instantaneous square error 119864(119899) is defined as

119864 (119899) =1

21198902(119899) =

1

2(119884 (119899) minus (119899))

2

(7)

The goal of the back propagation algorithm is to minimize119864(119899) To achieve this goal the back propagation algorithmapplies a respective correction Δ120596

(119897)

119895119894(119899) to every synaptic

weight 120596(119897)

119895119894(119899) The correction Δ120596

(119897)

119895119894(119899) is designed to be

proportional to the gradient 120597119864(119899)120597120596(119897)119895119894(119899) that is

Δ120596(119897)

119895119894(119899) = minus120578

119886

120597119864 (119899)

120597120596(119897)

119895119894(119899)

(8)

The minus sign comes from the fact that lower 119864(119899) ispreferred 120578

119886is the adaptive learning rate proposed by Behera

et al in [20] which is expressed as

120578119886= 120583

100381710038171003817100381710038171003817100381710038172

1003817100381710038171003817J11987910038171003817100381710038172 (9)

where = 119884(119899) minus (119899) and J = 120597120597W119897 While consid-

ering the weight correction for 119897 = 2 that is the outputlayer according to the chain rule of calculus the gradient120597119864(119899)120597120596

(2)

1119894(119899) can be written as

120597119864 (119899)

120597120596(2)

1119894(119899)

=120597119864 (119899)

120597119890 (119899)

120597119890 (119899)

120597 (119899)

120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597120596(2)

1119894(119899)

(10)

Applying (4)ndash(7) to the four partial derivatives at the right ofthe equal sign in (10) it can be rewritten as

120597119864 (119899)

120597120596(2)

1119894(119899)

= minus119890 (119899) 1205931015840(V(2)1

(119899)) 119910(1)

119894(119899)

= minus120575(2)

1(119899) 119910(1)

119894(119899)

(11)

where 120575(119897)119895(119899) is the local gradient which is defined as

120575(119897)

119895(119899) = minus

120597119864 (119899)

120597V(119897)119895(119899)

(12)

While considering the weight correction for 119897 = 1 that isthe hidden layer by iterative backpropagation and chain rulemanners the local gradient 120575(1)

119895(119899) is

120575(1)

119895(119899) = 120593

1015840(V(1)119895

(119899)) 120575(2)

1(119899) 120596(2)

1119895(119899) (13)

The derivation of the local gradients at each layer is shown inthe bottom part of Figure 3 Finally every synaptic weight iscorrected following this formula

120596(119897)

119895119894(119899 + 1) = 120596

(119897)

119895119894(119899) + Δ120596

(119897)

119895119894(119899) (14)

The forward pass and the backward pass are executed iter-atively until the square error 119864(119899) reaches some satisfactorylevel or when the number of epoch exceeds some thresholdAfter that the handover parameter optimization is executed

6 Mobile Information Systems

X0(n) = +1

X1(n)

X2(n)

X3(n)

120596(1)10 (n)

120596(1)20 (n)

120596(1)30 (n)

120596(1)40 (n)

120596(1)1m(n)

120596(1)2m(n)

120596(1)3m(n)

120596(1)4m(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120575(1)1 (n) 120575(1)

2 (n) 120575(1)3 (n) 120575(1)

4 (n)

+1

minus1

120575(2)1 (n)

e1(n)

120596(2)10 (n)

120596(2)11 (n)

120596(2)12 (n)

120596(2)13 (n)

120596(2)14 (n)

120596(2)11 (n)

120596(2)12 (n)

120596(2)13 (n)

120596(2)14 (n)

(2)1 (n)

Y(n)

Y(n) = y(2)1 (n)

Xm(n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

(n)119831D

(n)119831M

Figure 3 The forward pass and the backward pass of the BP algorithm

44 Handover Parameter Optimization In this paper thehandover optimization problem is solved by the gradientdescent algorithm [21] Here the handover parameters X

119863

including HOM and TTT are adjusted adaptively To find thelocal minimum of the KPI function 119884 asymp (X) using gradientdescent one takes steps proportional to the gradient Supposethat at the 119899th time of adjustment the handover parametersareX119863(119899) and the KPI is119884(119899) At the next time of adjustment

the handover parameters X119863are set as

X119863 (119899 + 1) = X

119863 (119899) + ΔX119863 (119899) (15)

ΔX119863(119899) depends on the gradient of estimated KPI 119884(119899) with

respect to X119863(119899) that is

ΔX119863 (119899) = 120583

120597119884 (119899)

120597X119863 (119899)

(16)

where 120583 is the constant adjustment rateHow to obtain the gradient 120597119884120597X

119863is a problem leftThe

original BP algorithmdescribed in Section 43 only consists oftwo passes the forward pass and the backward pass In orderto obtain the gradient 120597119884120597X

119863 the proposed DHO approach

modifies the original BP algorithm to add the third pass theadjustment pass We provide the detailed description of theadjustment pass in the following paragraphs

The purpose of the adjustment pass is to obtain theestimated gradient 120597(119899)120597X

119863(119899) which is the key to adjust

the handover parameter in order to minimize the KPI Thesynaptic weights that have been adjusted well in the backwardpass in accordance with (14) are set as fixed in the adjustmentpass In the following we show how to obtain the estimatedgradient 120597(119899)120597X

119863(119899) The signal flow of the adjustment

pass is depicted in Figure 4First of all we define the local gradient 120582(119897)

119895(119899) of the

adjustment pass as

120582(119897)

119895(119899) =

120597 (119899)

120597V(119897)119895(119899)

(17)

This local gradient 120582(119897)119895(119899) is similar to the local gradient

120575(119897)

119895(119899) of the backward pass which is defined in (12) except

that the target of the partial derivative is changed to (119899)While considering the output layer the local gradient 120582(2)

1(119899)

is

120582(2)

1(119899) =

120597 (119899)

120597V(2)1

(119899)= 1205931015840(V(2)1

(119899)) (18)

Mobile Information Systems 7

Y(n) = y(2)1 (n)

120582(1)1 (n) 120582(1)

2 (n) 120582(1)3 (n) 120582(1)

4 (n) 120582(2)1 (n)

120597Y(n)

120597X1(n)

120597Y(n)

120597X2(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

+1

(2)1 (n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

X0(n) = +1

X1(n)

X2(n)

X3(n)

Xm(n)

120596(1)10 (n + 1)

120596(1)20 (n + 1)

120596(1)30 (n + 1)

120596(1)40 (n + 1)

120596(1)1m(n + 1)

120596(1)2m(n + 1)

120596(1)3m(n + 1)

120596(1)4m(n + 1)

120596(1)11 (n + 1)

120596(1)12 (n + 1)

120596(1)21 (n + 1)

120596(1)22 (n + 1)

120596(1)31 (n + 1)

120596(1)32 (n + 1)

120596(1)41 (n + 1)

120596(1)42 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

120596(2)10 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

(n)119831D

(n)119831M

Figure 4 Adjustment pass of the proposed DHO approach

While considering the hidden layer the local gradient 120582(1)119895(119899)

is

120582(1)

119895(119899) =

120597 (119899)

120597V(1)119895

(119899)=

120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119910(1)

119895(119899)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

= 120582(2)

1(119899) 120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899))

(19)

Using the results from (18) and (19) the gradient can bewritten as follows

120597 (119899)

120597119883119894 (119899)

=120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

= 120582(2)

1(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

(20)

where the last equality comes from (18) Since

V(2)1

(119899) =

4

sum119895=0

120596(2)

1119895(119899 + 1) 119910

(1)

119895(119899) (21)

the second term at the most right of (20) can be written as

120597V(2)1

(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(n)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899)) 120596(1)

1198951(119899 + 1)

(22)

where the third equality comes from the fact that 119910(1)0(119899) is

always equal to 1 and its partial derivative can be ignoredfrom the summation Combining (19) and (22) into (20) thegradient 120597(119899)120597119883

119894(119899) is obtained as

120597 (119899)

120597119883119894 (119899)

=

4

sum119895=1

120582(1)

119895(119899) 120596(1)

1198951(119899 + 1) (23)

The derivation of the local gradients at each layer and thegradient 120597(119899)120597119883

119894(119899) is depicted in Figure 4 Based on the

8 Mobile Information Systems

05 1 15 2 25 3 35 4 45 50X-coordinate (km)

0

05

1

15

2

25

3

35

4

45

5

Y-c

oord

inat

e (km

)

Figure 5 Topology of the simulated mobile communication net-work

result from (23) the handover parameter119883119894is adjusted as (15)

and (16) showIn order to cope with the time-varying channel condi-

tions we use online training and operate the three passesiteratively to model the instantaneous relationship 119884 = 119891(X)and optimize the handover parameters HOM and TTT

5 Performance Evaluation and Discussion

In this section we describe the simulation environmentsimulation results and discussions

In our simulation a mobile communication networkwhich consists of 100 small cells is built in a 5 km times 5 kmarea Figure 5 shows the coordinates of these small cell basestations Each cell has 5MHz bandwidth with 25 resourceblocks and 2GHz carrier frequency Each resource blockconsists of 12 subcarriers of size 15 kHz each A time slot is05ms in duration and the transmission time interval (TTI)is 1ms

In the simulation 100 MSs are uniformly distributedover the area with random initialized positions The mobilitytrace of these 100 MSs is generated by using the Manhattanmobility model which is widely used for urban areas andsuitable to implement realistic simulations [22ndash24] In theManhattanmobility model MSs move in either horizontal orvertical directions on an urban map At each intersection anMS employs a probabilistic approach to choose the directionof its movements The probabilities of going straight turningleft and turning right are 05 025 and 025 respectivelyThe blue line in Figure 5 shows an illustrative moving pathof an MS in the mobile communication network Whileconsidering the KPI the weights of mobility problems are setto be equal

An RLF will occur if the SINR level of the servingcell drops below minus65 dB before handover execution is

completed [18] We set this as our SINR threshold valuesuch that if the SINR from the source BS is lower than thethreshold an RLF will occur The simulation time is set as5000 seconds The operating point with 0 dB HOM and 0 sTTT is the starting operating point for all cells in the network

In this paper we use LTE-Sim an open source frameworkto simulate LTE networks as our basic simulation tool[25] The design and development of LTE-Sim support thefollowing aspects

(i) The Evolved Universal Terrestrial Radio Access (E-UTRAN) and the Evolved Packet System (EPS)

(ii) Single and heterogeneous multicell environments(iii) Multiusers environment(iv) User mobility(v) Handover procedures(vi) Frequency reuse techniques(vii) Quality-of-service (QoS) management(viii) Scheduling strategies

However LTE-Sim still lacks some important functional-ities and features that are required in this workTherefore wemodified LTE-Sim from several aspects Our modification toLTE-Sim includes the following

(1) Add

(i) radio link failure(ii) handover failure(iii) HOM and TTT(iv) proposed DHO approach

(2) Modify

(i) mobility pattern using the seed to generate thesame sequence of random numbers each timewe run the simulation

The simulation parameter values are shown in Table 2unless otherwise specified

Figure 6 shows the simulation result when the TX poweris set as 20 dBm The horizontal axis denotes the mobilityproblem index Indexes 1 to 5 mean the mobility problem oftoo-late HO too-early HO HO to wrong cell ping-pongHOand unnecessary HO respectively Index 6 means the KPIwhich is the weighted average of the five MPRs The verticalaxismeans the corresponding values ofMPR orKPIThe bluebars show the performance for a static HO parameter settingwith 0 dB HOM and 0 s TTT and the green ones for anotherstatic setting with 10 dBHOMand 512 s TTTThe yellow barsshow the performance of the proposed DHO approach Theresults shown in Figure 6 validate that the proposed DHOapproach can effectively mitigate the mobility problems andachieve better MPRs and KPI

Figure 7 shows the simulation result when the TX poweris set as 43 dBm Since the TX power is higher than the onein the previous scenario in Figure 6 it is less likely that an

Mobile Information Systems 9

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

Figure 6 Simulation result when the TX power is set as 20 dBm

Table 2 Simulation parameters

Parameter ValuesNumber of small cell base stations 100Number of mobile stations 100eNB Tx power 15 dBm20 dBm43 dBmCarrier frequency 2GHzMS speed 30 kmhMobility model ManhattanTTI 1msSubcarrier spacing 15 KHzResource block 180KHzSuper frame time 10msNoise figure 25Noise spectral density minus174 dBmSINR threshold minus65 dBSimulation time 5000 sHandover delay 30msSystem bandwidth 5MHz 25 RBsTTI

RLF occurs Therefore the MPRs and KPI in Figure 7 arelower than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

Figure 8 shows the simulation result when the TX poweris set as 15 dBm Since the TX power is lower than the onein the previous scenario in Figure 6 it is more likely that anRLF occurs Therefore the MPRs and KPI in Figure 8 arehigher than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

We compare the KPI improvement under the variousvalues of TX power The KPI improvement is calculatedas the difference between the initial-case KPI (HOM 0dB

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 7 Simulation result when the TX power is set as 43 dBm

DHO approach

2 3 4 5 61Mobility problem index

0

001

002

003

004

005

006

007

008

MPR

KPI

HOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 8 Simulation result when the TX power is set as 15 dBm

TTT 0 s) and the KPI of the proposed DHO methoddivided by the initial-case KPI Figure 9 shows that the KPIimprovement ranges from 15 to 20 It also shows that theKPI improvement is more significant when the TX power islower that is the transmission range of the base station issmaller

6 Conclusion

In this work we investigate the mobility problems innext generation mobile communication networks There aretotally five different mobility problems each with different

10 Mobile Information Systems

20 4315TX power (dBm)

0

5

10

15

20

25

KPI i

mpr

ovem

ent (

)

Figure 9 KPI improvement with various TX power

characteristics We propose a data-driven handover opti-mization (DHO) approach which aims to mitigate mobilityproblems The DHO approach consists of four parts

(1) Mobility problem identification(2) KPI design(3) KPI function estimation(4) Handover parameter optimization

The DHO approach collects data from the mobile com-munication measurement results and provides a model toestimate the relationship between the KPI and features fromthe collected dataset Based on the model the handoverparameters including the HOM and TTT are optimized tominimize the KPI which is a weighted average of differentmobility problem ratios Simulation results show that theproposed DHO approach could effectively mitigate mobilityproblems and provide better connectivity

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work was supported in part by Ministry of Scienceand Technology (MOST) Taipei Taiwan under Grants nosMOST 103-2221-E-155-005 MOST 104-2221-E-155-017 andMOST 105-2218-E-155-001

References

[1] J G Andrews S Buzzi W Choi et al ldquoWhat will 5G berdquo IEEEJournal on Selected Areas in Communications vol 32 no 6 pp1065ndash1082 2014

[2] K Kitagawa T Komine T Yamamoto and S Konishi ldquoAhandover optimization algorithm with mobility robustness forLTE systemsrdquo in Proceedings of the IEEE 22nd International

Symposium on Personal Indoor and Mobile Radio Communica-tions (PIMRC rsquo11) pp 1647ndash1651 Toronto Canada September2011

[3] 3GPP ldquoSelf-configuring and self-optimizing network (SON)usecases and solutionsrdquo TR 36902 V931 2011

[4] A Awada BWegmann I Viering and A Klein ldquoA SON-basedalgorithm for the optimization of inter-RAT handover param-etersrdquo IEEE Transactions on Vehicular Technology vol 62 no 5pp 1906ndash1923 2013

[5] T Jansen I Balan J Turk I Moerman and T Kurner ldquoHando-ver parameter optimization in LTE self-organizing networksrdquoin Proceedings of the IEEE 72nd Vehicular Technology Confer-ence Fall (VTC 2010-Fall rsquo10) pp 1ndash5 IEEE Ottawa CanadaSeptember 2010

[6] T Jansen I Balan S Stefanski I Moerman and T KurnerldquoWeighted performance based handover parameter optimiza-tion in LTErdquo in Proceedings of the IEEE 73rd Vehicular Technol-ogy Conference (VTC rsquo11) pp 1ndash5 May 2011

[7] W Li X Duan S Jia L Zhang Y Liu and J Lin ldquoA dynamichysteresis-adjusting algorithm in LTE self-organization net-worksrdquo in Proceedings of the IEEE 75th Vehicular TechnologyConference (VTC Spring rsquo12) pp 1ndash5 Yokohama Japan May2012

[8] A Lobinger S Stefanski T Jansen and I Balan ldquoCoordinatinghandover parameter optimization and load balancing in LTEself-optimizing networksrdquo in Proceedings of the IEEE 73rdVehicular Technology Conference (VTC rsquo11) pp 1ndash5 May 2011

[9] V Capdevielle A Feki and E Sorsy ldquoJoint interferencemanagement and handover optimization in LTE small cellsnetworkrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6769ndash6773 IEEE OttawaCanada June 2012

[10] T Ali and M Saquib ldquoWLANcellular handover analysis fordifferent mobility modelsrdquo in Proceedings of the IEEE Interna-tional Conference on Communications (ICC rsquo12) pp 2758ndash2762Ottawa Canada June 2012

[11] A Awada B Wegmann I Viering and A Klein ldquoCell-pairspecific optimization of the inter-RAT handover parameters inSONrdquo in Proceedings of the IEEE 23rd International Symposiumon Personal Indoor andMobile Radio Communications (PIMRCrsquo12) pp 1168ndash1173 Sydney Australia September 2012

[12] D Giacomini and A Agarwal ldquoVertical handover decisionmaking using QoS reputation and GM(11) predictionrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo12) pp 5655ndash5659 IEEE Ottawa Canada June2012

[13] D Lopez-Perez I Guvenc and X Chu ldquoTheoretical analysisof handover failure and ping-pong rates for heterogeneousnetworksrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6774ndash6779 IEEE OttawaCanada June 2012

[14] A Rath and S Panwar ldquoFast handover in cellular networkswith femtocellsrdquo in Proceedings of the International Conferenceon Communications (ICC rsquo12) pp 2752ndash2757 IEEE OttawaCanada June 2012

[15] D Xenakis N Passas and C Verikoukis ldquoA novel handoverdecision policy for reducing power transmissions in the two-tier LTE networkrdquo in Proceedings of the IEEE International Con-ference on Communications (ICC rsquo12) pp 1352ndash1356 OttawaCanada June 2012

[16] T Taleb K Samdanis and A Ksentini ldquoSupporting highlymobile users in cost-effective decentralized mobile operator

Mobile Information Systems 11

networksrdquo IEEE Transactions on Vehicular Technology vol 63no 7 pp 3381ndash3396 2014

[17] A Imran A Zoha and A Abu-Dayya ldquoChallenges in 5G howto empower SONwith big data for enabling 5Grdquo IEEE Networkvol 28 no 6 pp 27ndash33 2014

[18] H-D Bae B Ryu and N-H Park ldquoAnalysis of handoverfailures in LTE femtocell systemsrdquo in Proceedings of the Aus-tralasian Telecommunication Networks and Applications Con-ference (ATNAC rsquo11) pp 1ndash5 IEEE Melbourne AustraliaNovember 2011

[19] S Haykin Neural Networks A Comprehensive FoundationPrentice-Hall 1999

[20] L Behera S Kumar and A Patnaik ldquoOn adaptive learning ratethat guarantees convergence in feedforward networksrdquo IEEETransactions on Neural Networks vol 17 no 5 pp 1116ndash11252006

[21] J A Snyman Practical Mathematical Optimization An Intro-duction to Basic Optimization Theory and Classical and NewGradient-Based Algorithms vol 97 of Applied OptimizationSpringer New York NY USA 2005

[22] M Gudmundson ldquoCell planning in Manhattan environmentsrdquoin Proceedings of the 42nd Vehicular Technology Conference pp435ndash438 IEEE Denver Colo USA May 1992

[23] P-E Ostling ldquoImplications of cell planning on handoff per-formance in manhattan environmentsrdquo in Proceedings of the5th IEEE International Symposium on Personal Indoor andMobile Radio Communications Wireless NetworksmdashCatchingthe Mobile Future vol 2 pp 625ndash629 The Hague The Nether-lands September 1994

[24] M Karabacak D Wang H Ishii and H Arslan ldquoMobilityperformance of macrocell-assisted small cells in Manhattanmodelrdquo in Proceedings of the 79th IEEE Vehicular TechnologyConference (VTC-Spring rsquo14) pp 1ndash5 Seoul South Korea May2014

[25] 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

The details of these two handover parameters are describedas follows

(i) Handover Margin (HOM) (Also Known as Hysteresis)HOM is defined as the threshold of the difference in Signal toInterference plusNoise Ratio (SINR) between the serving andthe target cells When the SINR from the target cell is HOMbetter than that from the serving cell the HO condition issatisfied

(ii) Time-to-Trigger (TTT) An HO request would not besent immediately when the corresponding HO conditionmentioned above is satisfied Instead theHOcondition has tobe fulfilled for a certain period denoted as TTTWhen theHOcondition keeps satisfied for TTT an HO request is triggered

The proper values of HOM and TTT are critical toseamless connectivity Large HOM and TTT values lead tomore stable behavior but they may delay the HO decisionsunnecessarily whichmay cause problems On the other handsmall HOM and TTT values avoid long delay to trigger HOrequest but they may cause ping-pong HO and unnecessaryHO

The HO problem is formulated as

X119863opt = argmin

X119863119884 (1)

where 119884 is the corresponding key performance indicator(KPI) which depends on the feature vector X = [X

119863X119872]

The KPI should be designed to measure the mitigation ofmobility problems

There are four main challenges in the HO problem

(1) There are various kinds of mobility problems in nextgeneration mobile communication networks [18]Identifying the mobility problems would be the firstpriority in the HO problem

(2) The design of the KPI needs comprehensive andcareful considerations for these mobility problems

(3) The KPI 119884 can be formulated as a function of X thatis 119884 = 119891(X) However the exact form of 119891(X) isunknownTherefore one of the challenges we take onin this work is to provide a good estimation (X) tothe function 119891(X) With a good estimate (X)we canmake predictions of the KPI 119884 at some points X = xwhere x is a vector of some specific values of X

(4) The decision variables X119863consist of two elements

HOM and TTT Mitigating the mobility problems byjoint-optimizing HOM and TTT would be anotherchallenge to the HO problem

In Section 4 we describe the proposed DHO approach todeal with these main challenges

4 Proposed DHO Approach

In this section we provide the detailed description of the pro-posed data-driven handover optimization (DHO) approachFigure 1 shows the system architecture of the proposed DHO

Data collection

Right data

Data analysis

Big data

Applying the HO

HO parameter

HO parameteroptimization

engine

KPI function

KPI functionoptimization estimation

Key performanceindicator

estimationengine

parameters

Mobilityproblem

identification

Mobilecommunication

network

Figure 1 System architecture of the proposed DHO approach

approach First measurement results are collected from themobile communication network The dataset collected in thedatabase includes the feature vector described in Section 3and various counters and flags that are used to identify themobility problemsThe details of these counters and flags areprovided in Section 41 The DHO approach then analyzesand processes the data to calculate themobility problem ratio(MPR) of each mobility problem and obtain the KPI whichis saved in the database denoted as ldquoright datardquo in Figure 1The KPI function estimation engine uses the processed ldquorightdatardquo to provide the KPI function estimate After that theHO parameter optimization engine uses the KPI functionestimate to optimize the HO parameters including HOM andTTT Finally the optimal values ofHOMandTTT are appliedto corresponding base stations

In the following paragraphs we provide the detaileddescription of each building block in Figure 1

41 Mobility Problem Identification There would be fivedifferent mobility problems (MPs) in next generation mobilecommunication networks

(1) Too-Late HO Too-late HO trigger timing leads to lowSINR level of the serving cell A radio link failure (RLF)between the mobile station (MS) and the serving bast station(BS) occurs and then the MS try to reestablish a connectionto a nearby BS

(2) Too-Early HO Too-early HO trigger timing leads to a lowSINR level of the target cell An RLF between the MS andthe target BS occurs and then the MS try to reestablish aconnection to the original serving BS

(3) HO toWrong Cell HO from a source cell A to a target cellB which provides unstable SINR leads to an RLF At the sametime if there is another suitable cell C for theMS to reconnectto after the RLF the MS reestablishes connection with cell Cthat is neither the serving cell A nor the target cell B In thissituation instead of the failed HO from cell A to cell B HOfrom cell A directly to cell C would be a better decision

4 Mobile Information Systems

Table 1 Mobility problem identification

Mobility problem Close to lastHO RLF status

New target =previousserving

Too-late HO N Y Donrsquot careToo-early HO Y Y YHO to wrong cell Y Y NPing-pong HO Y N YUnnecessary HO Y N N

(4) Ping-Pong HO Cell A hands over aMS to cell B and cell Bhands over the sameMS back to cell A shortly after Since datatransmission is temporarily blocked during the connectiontransferring to the target cell these two HOs despite bothbeing successful should be avoided

(5) Unnecessary HO Cell A hands over a MS to cell B andcell B hands over the same MS to another cell C shortlyafter Those two HOs despite both being successful can becombined into one HO from cell A directly to cell C in orderto avoid unnecessary blocking of data transmission duringthe connection transferring time

In order to identify the mobility problems mentionedabove we design the mobility problem identification criteriaas Table 1 shows The mobility problem identification criteriadepend on three conditional tests

(1) Close to last HOWhether the occurrence time of the currentHO eventis close to that of the previous HO event

(2) RLF statusIs there an RLF for the current connection

(3) New target = previous servingWhether the new target cell is the same as the previousserving one

Table 1 shows that each of the five mobility problemswould be identified based on the test results of the threeconditional tests mentioned above

In the proposed DHO approach we design the followingcounters and flags to identify the mobility problems

(i) lastHoCounter to indicate the time interval since lastHO

(ii) lastHoThreshold to define a threshold value com-bined with lastHoCounter to identify whether theoccurrence time of the current HO event is close tothat of the previous HO event

(iii) rlfStatus to indicate whether an RLF occurs for thecurrent connection

(iv) newTagertCellId to indicate the new target cell ID(v) previousServingCellId to indicate the previous serv-

ing cell ID combined with the newTagertCellId toindicate whether the new target cell is the same as theprevious serving one

42 Design of KPI In the HO problem the KPI 119884 is definedas theweighted average of the variousmobility problem ratios(MPRs) There are five different kinds of mobility problemsThe KPI 119884 is calculated as

119884 =sum5

119896=1119908119896119877119896 (X)

sum5

119896=1119908119896

(2)

where 119908119896is the weight of the mobility problem 119896 which is

determined by mobile network operators and 119877119896(X) is the

MPR of the mobility problem 119896 which is a function of XMPR 119877

119896is defined as the probability of the event where the

mobility problem (MP) 119896 happens Consider

119877119896 (X) = Pr (MP

119896) =

119873119896

119873total (3)

where119873119896denotes the number of MP

119896 and119873total denotes the

total number of HO Each BS uses a sliding time window tocollect the statistics of 119873

119896and 119873total in order to obtain each

MPR and the KPI 119884

43 KPI Function Estimation In order to solve the HOproblem as (1) shows some previous solutions focus onanalyzing and obtaining the KPI function 119884 = 119891(X) Whenthe analytical formula 119891(X) is known solving the equation120597119891120597X

119863= 0 would get the optimal handover parameters

HOM and TTT However note that the function 119891 consistsof complex joint effects of the feature vector X = [X

119863X119872]

Under these complex effects it is difficult to analyze andobtain an accurate KPI function

In this paper we use machine learning techniques to esti-mate the KPI function To be specific we use the multilayerperception (MLP) which is one of the most popular neuralnetworks to perform the KPI function estimation [19]

The MLP consists of an input layer one or more hiddenlayers and an output layer In this paper we adopt the MLPwith one hidden layer consisting of four hidden neuronsThefeature vector X is applied to the input layer and propagatesthrough the network layer by layer from left to right Eacharrow in this figure represents an adjustable synaptic weightFrom our comprehensive experiments this architecture isempirically sufficient to model the relationship between KPIand the feature vector whereas the other experiments withmore hidden neurons provide similar results A simulationresult about the mean-square-error (MSE) of the KPI func-tion estimation with various numbers of hidden neuronsis shown in Figure 2 The epoch limitation is set as 1000This figure shows that four hidden neurons are sufficient toimprove the performance ofmodeling to the degree of 1119864minus10whereas the performance of two neurons only reaches thedegree of 1119864 minus 6 Applying more hidden neurons does notsignificantly improve the MSE

In MLP the backpropagation (BP) algorithm is used toadjust the neural network parameters in order to minimizethe error between the target (measured) KPI and the responsefrom the function estimate The detailed operation of theBP algorithm includes the forward pass and the backwardpassWe provide the detailed descriptions of the forward passand the backward pass in the following paragraphs

Mobile Information Systems 5

4 6 8 102Number of hidden neurons

10minus11

10minus10

10minus9

10minus8

10minus7

10minus6

10minus5

MSE

Figure 2 MSE of the KPI function estimation with various numberof hidden neurons

431 Forward Pass The purpose of the forward pass is toget the actual response of the MLP in accordance with thespecified input signal In the forward pass the feature vectorX(119899) at iteration 119899 is applied to the input layer and propagatesthrough the network layer by layer from left to right Thetop part of Figure 3 shows the signal flow of the forwardpass 119883

1(119899) denotes the HOM and 119883

2(119899) denotes the TTT

1198833(119899) to 119883

119898(119899) denote the other features When the signal

propagates through a synapse its value is multiplied bythe synaptic weight The input of a neuron is the synapticweighted sum of the output of its previous layer For examplethe input of the 119895th neuron in the 119897th layer is

V(119897)119895(119899) =

119898119897minus1

sum119894=0

119908(119897)

119895119894(119899) 119910(119897minus1)

119894(119899) (4)

where 119894 = 0 represents the bias termThe input of this neuronis applied to the activation function 120593(sdot) The activationfunction should be differentiable everywhere We adopt thecommonly used sigmoid function as

119910(119897)

119895(119899) = 120593 (V(119897)

119895(119899)) =

1

1 + exp (minusV(119897)119895(119899))

(5)

The output of this neuron propagates through the network tothe next layer and the same operation is performed againFinally the signal aggregates in the output layer and theestimated KPI function (119899) is obtained During the forwardpass the synaptic weights are all fixed

432 Backward Pass The purpose of the backward pass isto adjust the synaptic weights in order to minimize the errorbetween the actual response (119899) and the desired response119884(119899) The error signal 119890(119899) is defined by

119890 (119899) = 119884 (119899) minus (119899) (6)

The instantaneous square error 119864(119899) is defined as

119864 (119899) =1

21198902(119899) =

1

2(119884 (119899) minus (119899))

2

(7)

The goal of the back propagation algorithm is to minimize119864(119899) To achieve this goal the back propagation algorithmapplies a respective correction Δ120596

(119897)

119895119894(119899) to every synaptic

weight 120596(119897)

119895119894(119899) The correction Δ120596

(119897)

119895119894(119899) is designed to be

proportional to the gradient 120597119864(119899)120597120596(119897)119895119894(119899) that is

Δ120596(119897)

119895119894(119899) = minus120578

119886

120597119864 (119899)

120597120596(119897)

119895119894(119899)

(8)

The minus sign comes from the fact that lower 119864(119899) ispreferred 120578

119886is the adaptive learning rate proposed by Behera

et al in [20] which is expressed as

120578119886= 120583

100381710038171003817100381710038171003817100381710038172

1003817100381710038171003817J11987910038171003817100381710038172 (9)

where = 119884(119899) minus (119899) and J = 120597120597W119897 While consid-

ering the weight correction for 119897 = 2 that is the outputlayer according to the chain rule of calculus the gradient120597119864(119899)120597120596

(2)

1119894(119899) can be written as

120597119864 (119899)

120597120596(2)

1119894(119899)

=120597119864 (119899)

120597119890 (119899)

120597119890 (119899)

120597 (119899)

120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597120596(2)

1119894(119899)

(10)

Applying (4)ndash(7) to the four partial derivatives at the right ofthe equal sign in (10) it can be rewritten as

120597119864 (119899)

120597120596(2)

1119894(119899)

= minus119890 (119899) 1205931015840(V(2)1

(119899)) 119910(1)

119894(119899)

= minus120575(2)

1(119899) 119910(1)

119894(119899)

(11)

where 120575(119897)119895(119899) is the local gradient which is defined as

120575(119897)

119895(119899) = minus

120597119864 (119899)

120597V(119897)119895(119899)

(12)

While considering the weight correction for 119897 = 1 that isthe hidden layer by iterative backpropagation and chain rulemanners the local gradient 120575(1)

119895(119899) is

120575(1)

119895(119899) = 120593

1015840(V(1)119895

(119899)) 120575(2)

1(119899) 120596(2)

1119895(119899) (13)

The derivation of the local gradients at each layer is shown inthe bottom part of Figure 3 Finally every synaptic weight iscorrected following this formula

120596(119897)

119895119894(119899 + 1) = 120596

(119897)

119895119894(119899) + Δ120596

(119897)

119895119894(119899) (14)

The forward pass and the backward pass are executed iter-atively until the square error 119864(119899) reaches some satisfactorylevel or when the number of epoch exceeds some thresholdAfter that the handover parameter optimization is executed

6 Mobile Information Systems

X0(n) = +1

X1(n)

X2(n)

X3(n)

120596(1)10 (n)

120596(1)20 (n)

120596(1)30 (n)

120596(1)40 (n)

120596(1)1m(n)

120596(1)2m(n)

120596(1)3m(n)

120596(1)4m(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120575(1)1 (n) 120575(1)

2 (n) 120575(1)3 (n) 120575(1)

4 (n)

+1

minus1

120575(2)1 (n)

e1(n)

120596(2)10 (n)

120596(2)11 (n)

120596(2)12 (n)

120596(2)13 (n)

120596(2)14 (n)

120596(2)11 (n)

120596(2)12 (n)

120596(2)13 (n)

120596(2)14 (n)

(2)1 (n)

Y(n)

Y(n) = y(2)1 (n)

Xm(n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

(n)119831D

(n)119831M

Figure 3 The forward pass and the backward pass of the BP algorithm

44 Handover Parameter Optimization In this paper thehandover optimization problem is solved by the gradientdescent algorithm [21] Here the handover parameters X

119863

including HOM and TTT are adjusted adaptively To find thelocal minimum of the KPI function 119884 asymp (X) using gradientdescent one takes steps proportional to the gradient Supposethat at the 119899th time of adjustment the handover parametersareX119863(119899) and the KPI is119884(119899) At the next time of adjustment

the handover parameters X119863are set as

X119863 (119899 + 1) = X

119863 (119899) + ΔX119863 (119899) (15)

ΔX119863(119899) depends on the gradient of estimated KPI 119884(119899) with

respect to X119863(119899) that is

ΔX119863 (119899) = 120583

120597119884 (119899)

120597X119863 (119899)

(16)

where 120583 is the constant adjustment rateHow to obtain the gradient 120597119884120597X

119863is a problem leftThe

original BP algorithmdescribed in Section 43 only consists oftwo passes the forward pass and the backward pass In orderto obtain the gradient 120597119884120597X

119863 the proposed DHO approach

modifies the original BP algorithm to add the third pass theadjustment pass We provide the detailed description of theadjustment pass in the following paragraphs

The purpose of the adjustment pass is to obtain theestimated gradient 120597(119899)120597X

119863(119899) which is the key to adjust

the handover parameter in order to minimize the KPI Thesynaptic weights that have been adjusted well in the backwardpass in accordance with (14) are set as fixed in the adjustmentpass In the following we show how to obtain the estimatedgradient 120597(119899)120597X

119863(119899) The signal flow of the adjustment

pass is depicted in Figure 4First of all we define the local gradient 120582(119897)

119895(119899) of the

adjustment pass as

120582(119897)

119895(119899) =

120597 (119899)

120597V(119897)119895(119899)

(17)

This local gradient 120582(119897)119895(119899) is similar to the local gradient

120575(119897)

119895(119899) of the backward pass which is defined in (12) except

that the target of the partial derivative is changed to (119899)While considering the output layer the local gradient 120582(2)

1(119899)

is

120582(2)

1(119899) =

120597 (119899)

120597V(2)1

(119899)= 1205931015840(V(2)1

(119899)) (18)

Mobile Information Systems 7

Y(n) = y(2)1 (n)

120582(1)1 (n) 120582(1)

2 (n) 120582(1)3 (n) 120582(1)

4 (n) 120582(2)1 (n)

120597Y(n)

120597X1(n)

120597Y(n)

120597X2(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

+1

(2)1 (n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

X0(n) = +1

X1(n)

X2(n)

X3(n)

Xm(n)

120596(1)10 (n + 1)

120596(1)20 (n + 1)

120596(1)30 (n + 1)

120596(1)40 (n + 1)

120596(1)1m(n + 1)

120596(1)2m(n + 1)

120596(1)3m(n + 1)

120596(1)4m(n + 1)

120596(1)11 (n + 1)

120596(1)12 (n + 1)

120596(1)21 (n + 1)

120596(1)22 (n + 1)

120596(1)31 (n + 1)

120596(1)32 (n + 1)

120596(1)41 (n + 1)

120596(1)42 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

120596(2)10 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

(n)119831D

(n)119831M

Figure 4 Adjustment pass of the proposed DHO approach

While considering the hidden layer the local gradient 120582(1)119895(119899)

is

120582(1)

119895(119899) =

120597 (119899)

120597V(1)119895

(119899)=

120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119910(1)

119895(119899)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

= 120582(2)

1(119899) 120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899))

(19)

Using the results from (18) and (19) the gradient can bewritten as follows

120597 (119899)

120597119883119894 (119899)

=120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

= 120582(2)

1(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

(20)

where the last equality comes from (18) Since

V(2)1

(119899) =

4

sum119895=0

120596(2)

1119895(119899 + 1) 119910

(1)

119895(119899) (21)

the second term at the most right of (20) can be written as

120597V(2)1

(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(n)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899)) 120596(1)

1198951(119899 + 1)

(22)

where the third equality comes from the fact that 119910(1)0(119899) is

always equal to 1 and its partial derivative can be ignoredfrom the summation Combining (19) and (22) into (20) thegradient 120597(119899)120597119883

119894(119899) is obtained as

120597 (119899)

120597119883119894 (119899)

=

4

sum119895=1

120582(1)

119895(119899) 120596(1)

1198951(119899 + 1) (23)

The derivation of the local gradients at each layer and thegradient 120597(119899)120597119883

119894(119899) is depicted in Figure 4 Based on the

8 Mobile Information Systems

05 1 15 2 25 3 35 4 45 50X-coordinate (km)

0

05

1

15

2

25

3

35

4

45

5

Y-c

oord

inat

e (km

)

Figure 5 Topology of the simulated mobile communication net-work

result from (23) the handover parameter119883119894is adjusted as (15)

and (16) showIn order to cope with the time-varying channel condi-

tions we use online training and operate the three passesiteratively to model the instantaneous relationship 119884 = 119891(X)and optimize the handover parameters HOM and TTT

5 Performance Evaluation and Discussion

In this section we describe the simulation environmentsimulation results and discussions

In our simulation a mobile communication networkwhich consists of 100 small cells is built in a 5 km times 5 kmarea Figure 5 shows the coordinates of these small cell basestations Each cell has 5MHz bandwidth with 25 resourceblocks and 2GHz carrier frequency Each resource blockconsists of 12 subcarriers of size 15 kHz each A time slot is05ms in duration and the transmission time interval (TTI)is 1ms

In the simulation 100 MSs are uniformly distributedover the area with random initialized positions The mobilitytrace of these 100 MSs is generated by using the Manhattanmobility model which is widely used for urban areas andsuitable to implement realistic simulations [22ndash24] In theManhattanmobility model MSs move in either horizontal orvertical directions on an urban map At each intersection anMS employs a probabilistic approach to choose the directionof its movements The probabilities of going straight turningleft and turning right are 05 025 and 025 respectivelyThe blue line in Figure 5 shows an illustrative moving pathof an MS in the mobile communication network Whileconsidering the KPI the weights of mobility problems are setto be equal

An RLF will occur if the SINR level of the servingcell drops below minus65 dB before handover execution is

completed [18] We set this as our SINR threshold valuesuch that if the SINR from the source BS is lower than thethreshold an RLF will occur The simulation time is set as5000 seconds The operating point with 0 dB HOM and 0 sTTT is the starting operating point for all cells in the network

In this paper we use LTE-Sim an open source frameworkto simulate LTE networks as our basic simulation tool[25] The design and development of LTE-Sim support thefollowing aspects

(i) The Evolved Universal Terrestrial Radio Access (E-UTRAN) and the Evolved Packet System (EPS)

(ii) Single and heterogeneous multicell environments(iii) Multiusers environment(iv) User mobility(v) Handover procedures(vi) Frequency reuse techniques(vii) Quality-of-service (QoS) management(viii) Scheduling strategies

However LTE-Sim still lacks some important functional-ities and features that are required in this workTherefore wemodified LTE-Sim from several aspects Our modification toLTE-Sim includes the following

(1) Add

(i) radio link failure(ii) handover failure(iii) HOM and TTT(iv) proposed DHO approach

(2) Modify

(i) mobility pattern using the seed to generate thesame sequence of random numbers each timewe run the simulation

The simulation parameter values are shown in Table 2unless otherwise specified

Figure 6 shows the simulation result when the TX poweris set as 20 dBm The horizontal axis denotes the mobilityproblem index Indexes 1 to 5 mean the mobility problem oftoo-late HO too-early HO HO to wrong cell ping-pongHOand unnecessary HO respectively Index 6 means the KPIwhich is the weighted average of the five MPRs The verticalaxismeans the corresponding values ofMPR orKPIThe bluebars show the performance for a static HO parameter settingwith 0 dB HOM and 0 s TTT and the green ones for anotherstatic setting with 10 dBHOMand 512 s TTTThe yellow barsshow the performance of the proposed DHO approach Theresults shown in Figure 6 validate that the proposed DHOapproach can effectively mitigate the mobility problems andachieve better MPRs and KPI

Figure 7 shows the simulation result when the TX poweris set as 43 dBm Since the TX power is higher than the onein the previous scenario in Figure 6 it is less likely that an

Mobile Information Systems 9

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

Figure 6 Simulation result when the TX power is set as 20 dBm

Table 2 Simulation parameters

Parameter ValuesNumber of small cell base stations 100Number of mobile stations 100eNB Tx power 15 dBm20 dBm43 dBmCarrier frequency 2GHzMS speed 30 kmhMobility model ManhattanTTI 1msSubcarrier spacing 15 KHzResource block 180KHzSuper frame time 10msNoise figure 25Noise spectral density minus174 dBmSINR threshold minus65 dBSimulation time 5000 sHandover delay 30msSystem bandwidth 5MHz 25 RBsTTI

RLF occurs Therefore the MPRs and KPI in Figure 7 arelower than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

Figure 8 shows the simulation result when the TX poweris set as 15 dBm Since the TX power is lower than the onein the previous scenario in Figure 6 it is more likely that anRLF occurs Therefore the MPRs and KPI in Figure 8 arehigher than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

We compare the KPI improvement under the variousvalues of TX power The KPI improvement is calculatedas the difference between the initial-case KPI (HOM 0dB

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 7 Simulation result when the TX power is set as 43 dBm

DHO approach

2 3 4 5 61Mobility problem index

0

001

002

003

004

005

006

007

008

MPR

KPI

HOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 8 Simulation result when the TX power is set as 15 dBm

TTT 0 s) and the KPI of the proposed DHO methoddivided by the initial-case KPI Figure 9 shows that the KPIimprovement ranges from 15 to 20 It also shows that theKPI improvement is more significant when the TX power islower that is the transmission range of the base station issmaller

6 Conclusion

In this work we investigate the mobility problems innext generation mobile communication networks There aretotally five different mobility problems each with different

10 Mobile Information Systems

20 4315TX power (dBm)

0

5

10

15

20

25

KPI i

mpr

ovem

ent (

)

Figure 9 KPI improvement with various TX power

characteristics We propose a data-driven handover opti-mization (DHO) approach which aims to mitigate mobilityproblems The DHO approach consists of four parts

(1) Mobility problem identification(2) KPI design(3) KPI function estimation(4) Handover parameter optimization

The DHO approach collects data from the mobile com-munication measurement results and provides a model toestimate the relationship between the KPI and features fromthe collected dataset Based on the model the handoverparameters including the HOM and TTT are optimized tominimize the KPI which is a weighted average of differentmobility problem ratios Simulation results show that theproposed DHO approach could effectively mitigate mobilityproblems and provide better connectivity

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work was supported in part by Ministry of Scienceand Technology (MOST) Taipei Taiwan under Grants nosMOST 103-2221-E-155-005 MOST 104-2221-E-155-017 andMOST 105-2218-E-155-001

References

[1] J G Andrews S Buzzi W Choi et al ldquoWhat will 5G berdquo IEEEJournal on Selected Areas in Communications vol 32 no 6 pp1065ndash1082 2014

[2] K Kitagawa T Komine T Yamamoto and S Konishi ldquoAhandover optimization algorithm with mobility robustness forLTE systemsrdquo in Proceedings of the IEEE 22nd International

Symposium on Personal Indoor and Mobile Radio Communica-tions (PIMRC rsquo11) pp 1647ndash1651 Toronto Canada September2011

[3] 3GPP ldquoSelf-configuring and self-optimizing network (SON)usecases and solutionsrdquo TR 36902 V931 2011

[4] A Awada BWegmann I Viering and A Klein ldquoA SON-basedalgorithm for the optimization of inter-RAT handover param-etersrdquo IEEE Transactions on Vehicular Technology vol 62 no 5pp 1906ndash1923 2013

[5] T Jansen I Balan J Turk I Moerman and T Kurner ldquoHando-ver parameter optimization in LTE self-organizing networksrdquoin Proceedings of the IEEE 72nd Vehicular Technology Confer-ence Fall (VTC 2010-Fall rsquo10) pp 1ndash5 IEEE Ottawa CanadaSeptember 2010

[6] T Jansen I Balan S Stefanski I Moerman and T KurnerldquoWeighted performance based handover parameter optimiza-tion in LTErdquo in Proceedings of the IEEE 73rd Vehicular Technol-ogy Conference (VTC rsquo11) pp 1ndash5 May 2011

[7] W Li X Duan S Jia L Zhang Y Liu and J Lin ldquoA dynamichysteresis-adjusting algorithm in LTE self-organization net-worksrdquo in Proceedings of the IEEE 75th Vehicular TechnologyConference (VTC Spring rsquo12) pp 1ndash5 Yokohama Japan May2012

[8] A Lobinger S Stefanski T Jansen and I Balan ldquoCoordinatinghandover parameter optimization and load balancing in LTEself-optimizing networksrdquo in Proceedings of the IEEE 73rdVehicular Technology Conference (VTC rsquo11) pp 1ndash5 May 2011

[9] V Capdevielle A Feki and E Sorsy ldquoJoint interferencemanagement and handover optimization in LTE small cellsnetworkrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6769ndash6773 IEEE OttawaCanada June 2012

[10] T Ali and M Saquib ldquoWLANcellular handover analysis fordifferent mobility modelsrdquo in Proceedings of the IEEE Interna-tional Conference on Communications (ICC rsquo12) pp 2758ndash2762Ottawa Canada June 2012

[11] A Awada B Wegmann I Viering and A Klein ldquoCell-pairspecific optimization of the inter-RAT handover parameters inSONrdquo in Proceedings of the IEEE 23rd International Symposiumon Personal Indoor andMobile Radio Communications (PIMRCrsquo12) pp 1168ndash1173 Sydney Australia September 2012

[12] D Giacomini and A Agarwal ldquoVertical handover decisionmaking using QoS reputation and GM(11) predictionrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo12) pp 5655ndash5659 IEEE Ottawa Canada June2012

[13] D Lopez-Perez I Guvenc and X Chu ldquoTheoretical analysisof handover failure and ping-pong rates for heterogeneousnetworksrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6774ndash6779 IEEE OttawaCanada June 2012

[14] A Rath and S Panwar ldquoFast handover in cellular networkswith femtocellsrdquo in Proceedings of the International Conferenceon Communications (ICC rsquo12) pp 2752ndash2757 IEEE OttawaCanada June 2012

[15] D Xenakis N Passas and C Verikoukis ldquoA novel handoverdecision policy for reducing power transmissions in the two-tier LTE networkrdquo in Proceedings of the IEEE International Con-ference on Communications (ICC rsquo12) pp 1352ndash1356 OttawaCanada June 2012

[16] T Taleb K Samdanis and A Ksentini ldquoSupporting highlymobile users in cost-effective decentralized mobile operator

Mobile Information Systems 11

networksrdquo IEEE Transactions on Vehicular Technology vol 63no 7 pp 3381ndash3396 2014

[17] A Imran A Zoha and A Abu-Dayya ldquoChallenges in 5G howto empower SONwith big data for enabling 5Grdquo IEEE Networkvol 28 no 6 pp 27ndash33 2014

[18] H-D Bae B Ryu and N-H Park ldquoAnalysis of handoverfailures in LTE femtocell systemsrdquo in Proceedings of the Aus-tralasian Telecommunication Networks and Applications Con-ference (ATNAC rsquo11) pp 1ndash5 IEEE Melbourne AustraliaNovember 2011

[19] S Haykin Neural Networks A Comprehensive FoundationPrentice-Hall 1999

[20] L Behera S Kumar and A Patnaik ldquoOn adaptive learning ratethat guarantees convergence in feedforward networksrdquo IEEETransactions on Neural Networks vol 17 no 5 pp 1116ndash11252006

[21] J A Snyman Practical Mathematical Optimization An Intro-duction to Basic Optimization Theory and Classical and NewGradient-Based Algorithms vol 97 of Applied OptimizationSpringer New York NY USA 2005

[22] M Gudmundson ldquoCell planning in Manhattan environmentsrdquoin Proceedings of the 42nd Vehicular Technology Conference pp435ndash438 IEEE Denver Colo USA May 1992

[23] P-E Ostling ldquoImplications of cell planning on handoff per-formance in manhattan environmentsrdquo in Proceedings of the5th IEEE International Symposium on Personal Indoor andMobile Radio Communications Wireless NetworksmdashCatchingthe Mobile Future vol 2 pp 625ndash629 The Hague The Nether-lands September 1994

[24] M Karabacak D Wang H Ishii and H Arslan ldquoMobilityperformance of macrocell-assisted small cells in Manhattanmodelrdquo in Proceedings of the 79th IEEE Vehicular TechnologyConference (VTC-Spring rsquo14) pp 1ndash5 Seoul South Korea May2014

[25] 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

Table 1 Mobility problem identification

Mobility problem Close to lastHO RLF status

New target =previousserving

Too-late HO N Y Donrsquot careToo-early HO Y Y YHO to wrong cell Y Y NPing-pong HO Y N YUnnecessary HO Y N N

(4) Ping-Pong HO Cell A hands over aMS to cell B and cell Bhands over the sameMS back to cell A shortly after Since datatransmission is temporarily blocked during the connectiontransferring to the target cell these two HOs despite bothbeing successful should be avoided

(5) Unnecessary HO Cell A hands over a MS to cell B andcell B hands over the same MS to another cell C shortlyafter Those two HOs despite both being successful can becombined into one HO from cell A directly to cell C in orderto avoid unnecessary blocking of data transmission duringthe connection transferring time

In order to identify the mobility problems mentionedabove we design the mobility problem identification criteriaas Table 1 shows The mobility problem identification criteriadepend on three conditional tests

(1) Close to last HOWhether the occurrence time of the currentHO eventis close to that of the previous HO event

(2) RLF statusIs there an RLF for the current connection

(3) New target = previous servingWhether the new target cell is the same as the previousserving one

Table 1 shows that each of the five mobility problemswould be identified based on the test results of the threeconditional tests mentioned above

In the proposed DHO approach we design the followingcounters and flags to identify the mobility problems

(i) lastHoCounter to indicate the time interval since lastHO

(ii) lastHoThreshold to define a threshold value com-bined with lastHoCounter to identify whether theoccurrence time of the current HO event is close tothat of the previous HO event

(iii) rlfStatus to indicate whether an RLF occurs for thecurrent connection

(iv) newTagertCellId to indicate the new target cell ID(v) previousServingCellId to indicate the previous serv-

ing cell ID combined with the newTagertCellId toindicate whether the new target cell is the same as theprevious serving one

42 Design of KPI In the HO problem the KPI 119884 is definedas theweighted average of the variousmobility problem ratios(MPRs) There are five different kinds of mobility problemsThe KPI 119884 is calculated as

119884 =sum5

119896=1119908119896119877119896 (X)

sum5

119896=1119908119896

(2)

where 119908119896is the weight of the mobility problem 119896 which is

determined by mobile network operators and 119877119896(X) is the

MPR of the mobility problem 119896 which is a function of XMPR 119877

119896is defined as the probability of the event where the

mobility problem (MP) 119896 happens Consider

119877119896 (X) = Pr (MP

119896) =

119873119896

119873total (3)

where119873119896denotes the number of MP

119896 and119873total denotes the

total number of HO Each BS uses a sliding time window tocollect the statistics of 119873

119896and 119873total in order to obtain each

MPR and the KPI 119884

43 KPI Function Estimation In order to solve the HOproblem as (1) shows some previous solutions focus onanalyzing and obtaining the KPI function 119884 = 119891(X) Whenthe analytical formula 119891(X) is known solving the equation120597119891120597X

119863= 0 would get the optimal handover parameters

HOM and TTT However note that the function 119891 consistsof complex joint effects of the feature vector X = [X

119863X119872]

Under these complex effects it is difficult to analyze andobtain an accurate KPI function

In this paper we use machine learning techniques to esti-mate the KPI function To be specific we use the multilayerperception (MLP) which is one of the most popular neuralnetworks to perform the KPI function estimation [19]

The MLP consists of an input layer one or more hiddenlayers and an output layer In this paper we adopt the MLPwith one hidden layer consisting of four hidden neuronsThefeature vector X is applied to the input layer and propagatesthrough the network layer by layer from left to right Eacharrow in this figure represents an adjustable synaptic weightFrom our comprehensive experiments this architecture isempirically sufficient to model the relationship between KPIand the feature vector whereas the other experiments withmore hidden neurons provide similar results A simulationresult about the mean-square-error (MSE) of the KPI func-tion estimation with various numbers of hidden neuronsis shown in Figure 2 The epoch limitation is set as 1000This figure shows that four hidden neurons are sufficient toimprove the performance ofmodeling to the degree of 1119864minus10whereas the performance of two neurons only reaches thedegree of 1119864 minus 6 Applying more hidden neurons does notsignificantly improve the MSE

In MLP the backpropagation (BP) algorithm is used toadjust the neural network parameters in order to minimizethe error between the target (measured) KPI and the responsefrom the function estimate The detailed operation of theBP algorithm includes the forward pass and the backwardpassWe provide the detailed descriptions of the forward passand the backward pass in the following paragraphs

Mobile Information Systems 5

4 6 8 102Number of hidden neurons

10minus11

10minus10

10minus9

10minus8

10minus7

10minus6

10minus5

MSE

Figure 2 MSE of the KPI function estimation with various numberof hidden neurons

431 Forward Pass The purpose of the forward pass is toget the actual response of the MLP in accordance with thespecified input signal In the forward pass the feature vectorX(119899) at iteration 119899 is applied to the input layer and propagatesthrough the network layer by layer from left to right Thetop part of Figure 3 shows the signal flow of the forwardpass 119883

1(119899) denotes the HOM and 119883

2(119899) denotes the TTT

1198833(119899) to 119883

119898(119899) denote the other features When the signal

propagates through a synapse its value is multiplied bythe synaptic weight The input of a neuron is the synapticweighted sum of the output of its previous layer For examplethe input of the 119895th neuron in the 119897th layer is

V(119897)119895(119899) =

119898119897minus1

sum119894=0

119908(119897)

119895119894(119899) 119910(119897minus1)

119894(119899) (4)

where 119894 = 0 represents the bias termThe input of this neuronis applied to the activation function 120593(sdot) The activationfunction should be differentiable everywhere We adopt thecommonly used sigmoid function as

119910(119897)

119895(119899) = 120593 (V(119897)

119895(119899)) =

1

1 + exp (minusV(119897)119895(119899))

(5)

The output of this neuron propagates through the network tothe next layer and the same operation is performed againFinally the signal aggregates in the output layer and theestimated KPI function (119899) is obtained During the forwardpass the synaptic weights are all fixed

432 Backward Pass The purpose of the backward pass isto adjust the synaptic weights in order to minimize the errorbetween the actual response (119899) and the desired response119884(119899) The error signal 119890(119899) is defined by

119890 (119899) = 119884 (119899) minus (119899) (6)

The instantaneous square error 119864(119899) is defined as

119864 (119899) =1

21198902(119899) =

1

2(119884 (119899) minus (119899))

2

(7)

The goal of the back propagation algorithm is to minimize119864(119899) To achieve this goal the back propagation algorithmapplies a respective correction Δ120596

(119897)

119895119894(119899) to every synaptic

weight 120596(119897)

119895119894(119899) The correction Δ120596

(119897)

119895119894(119899) is designed to be

proportional to the gradient 120597119864(119899)120597120596(119897)119895119894(119899) that is

Δ120596(119897)

119895119894(119899) = minus120578

119886

120597119864 (119899)

120597120596(119897)

119895119894(119899)

(8)

The minus sign comes from the fact that lower 119864(119899) ispreferred 120578

119886is the adaptive learning rate proposed by Behera

et al in [20] which is expressed as

120578119886= 120583

100381710038171003817100381710038171003817100381710038172

1003817100381710038171003817J11987910038171003817100381710038172 (9)

where = 119884(119899) minus (119899) and J = 120597120597W119897 While consid-

ering the weight correction for 119897 = 2 that is the outputlayer according to the chain rule of calculus the gradient120597119864(119899)120597120596

(2)

1119894(119899) can be written as

120597119864 (119899)

120597120596(2)

1119894(119899)

=120597119864 (119899)

120597119890 (119899)

120597119890 (119899)

120597 (119899)

120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597120596(2)

1119894(119899)

(10)

Applying (4)ndash(7) to the four partial derivatives at the right ofthe equal sign in (10) it can be rewritten as

120597119864 (119899)

120597120596(2)

1119894(119899)

= minus119890 (119899) 1205931015840(V(2)1

(119899)) 119910(1)

119894(119899)

= minus120575(2)

1(119899) 119910(1)

119894(119899)

(11)

where 120575(119897)119895(119899) is the local gradient which is defined as

120575(119897)

119895(119899) = minus

120597119864 (119899)

120597V(119897)119895(119899)

(12)

While considering the weight correction for 119897 = 1 that isthe hidden layer by iterative backpropagation and chain rulemanners the local gradient 120575(1)

119895(119899) is

120575(1)

119895(119899) = 120593

1015840(V(1)119895

(119899)) 120575(2)

1(119899) 120596(2)

1119895(119899) (13)

The derivation of the local gradients at each layer is shown inthe bottom part of Figure 3 Finally every synaptic weight iscorrected following this formula

120596(119897)

119895119894(119899 + 1) = 120596

(119897)

119895119894(119899) + Δ120596

(119897)

119895119894(119899) (14)

The forward pass and the backward pass are executed iter-atively until the square error 119864(119899) reaches some satisfactorylevel or when the number of epoch exceeds some thresholdAfter that the handover parameter optimization is executed

6 Mobile Information Systems

X0(n) = +1

X1(n)

X2(n)

X3(n)

120596(1)10 (n)

120596(1)20 (n)

120596(1)30 (n)

120596(1)40 (n)

120596(1)1m(n)

120596(1)2m(n)

120596(1)3m(n)

120596(1)4m(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120575(1)1 (n) 120575(1)

2 (n) 120575(1)3 (n) 120575(1)

4 (n)

+1

minus1

120575(2)1 (n)

e1(n)

120596(2)10 (n)

120596(2)11 (n)

120596(2)12 (n)

120596(2)13 (n)

120596(2)14 (n)

120596(2)11 (n)

120596(2)12 (n)

120596(2)13 (n)

120596(2)14 (n)

(2)1 (n)

Y(n)

Y(n) = y(2)1 (n)

Xm(n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

(n)119831D

(n)119831M

Figure 3 The forward pass and the backward pass of the BP algorithm

44 Handover Parameter Optimization In this paper thehandover optimization problem is solved by the gradientdescent algorithm [21] Here the handover parameters X

119863

including HOM and TTT are adjusted adaptively To find thelocal minimum of the KPI function 119884 asymp (X) using gradientdescent one takes steps proportional to the gradient Supposethat at the 119899th time of adjustment the handover parametersareX119863(119899) and the KPI is119884(119899) At the next time of adjustment

the handover parameters X119863are set as

X119863 (119899 + 1) = X

119863 (119899) + ΔX119863 (119899) (15)

ΔX119863(119899) depends on the gradient of estimated KPI 119884(119899) with

respect to X119863(119899) that is

ΔX119863 (119899) = 120583

120597119884 (119899)

120597X119863 (119899)

(16)

where 120583 is the constant adjustment rateHow to obtain the gradient 120597119884120597X

119863is a problem leftThe

original BP algorithmdescribed in Section 43 only consists oftwo passes the forward pass and the backward pass In orderto obtain the gradient 120597119884120597X

119863 the proposed DHO approach

modifies the original BP algorithm to add the third pass theadjustment pass We provide the detailed description of theadjustment pass in the following paragraphs

The purpose of the adjustment pass is to obtain theestimated gradient 120597(119899)120597X

119863(119899) which is the key to adjust

the handover parameter in order to minimize the KPI Thesynaptic weights that have been adjusted well in the backwardpass in accordance with (14) are set as fixed in the adjustmentpass In the following we show how to obtain the estimatedgradient 120597(119899)120597X

119863(119899) The signal flow of the adjustment

pass is depicted in Figure 4First of all we define the local gradient 120582(119897)

119895(119899) of the

adjustment pass as

120582(119897)

119895(119899) =

120597 (119899)

120597V(119897)119895(119899)

(17)

This local gradient 120582(119897)119895(119899) is similar to the local gradient

120575(119897)

119895(119899) of the backward pass which is defined in (12) except

that the target of the partial derivative is changed to (119899)While considering the output layer the local gradient 120582(2)

1(119899)

is

120582(2)

1(119899) =

120597 (119899)

120597V(2)1

(119899)= 1205931015840(V(2)1

(119899)) (18)

Mobile Information Systems 7

Y(n) = y(2)1 (n)

120582(1)1 (n) 120582(1)

2 (n) 120582(1)3 (n) 120582(1)

4 (n) 120582(2)1 (n)

120597Y(n)

120597X1(n)

120597Y(n)

120597X2(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

+1

(2)1 (n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

X0(n) = +1

X1(n)

X2(n)

X3(n)

Xm(n)

120596(1)10 (n + 1)

120596(1)20 (n + 1)

120596(1)30 (n + 1)

120596(1)40 (n + 1)

120596(1)1m(n + 1)

120596(1)2m(n + 1)

120596(1)3m(n + 1)

120596(1)4m(n + 1)

120596(1)11 (n + 1)

120596(1)12 (n + 1)

120596(1)21 (n + 1)

120596(1)22 (n + 1)

120596(1)31 (n + 1)

120596(1)32 (n + 1)

120596(1)41 (n + 1)

120596(1)42 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

120596(2)10 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

(n)119831D

(n)119831M

Figure 4 Adjustment pass of the proposed DHO approach

While considering the hidden layer the local gradient 120582(1)119895(119899)

is

120582(1)

119895(119899) =

120597 (119899)

120597V(1)119895

(119899)=

120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119910(1)

119895(119899)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

= 120582(2)

1(119899) 120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899))

(19)

Using the results from (18) and (19) the gradient can bewritten as follows

120597 (119899)

120597119883119894 (119899)

=120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

= 120582(2)

1(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

(20)

where the last equality comes from (18) Since

V(2)1

(119899) =

4

sum119895=0

120596(2)

1119895(119899 + 1) 119910

(1)

119895(119899) (21)

the second term at the most right of (20) can be written as

120597V(2)1

(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(n)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899)) 120596(1)

1198951(119899 + 1)

(22)

where the third equality comes from the fact that 119910(1)0(119899) is

always equal to 1 and its partial derivative can be ignoredfrom the summation Combining (19) and (22) into (20) thegradient 120597(119899)120597119883

119894(119899) is obtained as

120597 (119899)

120597119883119894 (119899)

=

4

sum119895=1

120582(1)

119895(119899) 120596(1)

1198951(119899 + 1) (23)

The derivation of the local gradients at each layer and thegradient 120597(119899)120597119883

119894(119899) is depicted in Figure 4 Based on the

8 Mobile Information Systems

05 1 15 2 25 3 35 4 45 50X-coordinate (km)

0

05

1

15

2

25

3

35

4

45

5

Y-c

oord

inat

e (km

)

Figure 5 Topology of the simulated mobile communication net-work

result from (23) the handover parameter119883119894is adjusted as (15)

and (16) showIn order to cope with the time-varying channel condi-

tions we use online training and operate the three passesiteratively to model the instantaneous relationship 119884 = 119891(X)and optimize the handover parameters HOM and TTT

5 Performance Evaluation and Discussion

In this section we describe the simulation environmentsimulation results and discussions

In our simulation a mobile communication networkwhich consists of 100 small cells is built in a 5 km times 5 kmarea Figure 5 shows the coordinates of these small cell basestations Each cell has 5MHz bandwidth with 25 resourceblocks and 2GHz carrier frequency Each resource blockconsists of 12 subcarriers of size 15 kHz each A time slot is05ms in duration and the transmission time interval (TTI)is 1ms

In the simulation 100 MSs are uniformly distributedover the area with random initialized positions The mobilitytrace of these 100 MSs is generated by using the Manhattanmobility model which is widely used for urban areas andsuitable to implement realistic simulations [22ndash24] In theManhattanmobility model MSs move in either horizontal orvertical directions on an urban map At each intersection anMS employs a probabilistic approach to choose the directionof its movements The probabilities of going straight turningleft and turning right are 05 025 and 025 respectivelyThe blue line in Figure 5 shows an illustrative moving pathof an MS in the mobile communication network Whileconsidering the KPI the weights of mobility problems are setto be equal

An RLF will occur if the SINR level of the servingcell drops below minus65 dB before handover execution is

completed [18] We set this as our SINR threshold valuesuch that if the SINR from the source BS is lower than thethreshold an RLF will occur The simulation time is set as5000 seconds The operating point with 0 dB HOM and 0 sTTT is the starting operating point for all cells in the network

In this paper we use LTE-Sim an open source frameworkto simulate LTE networks as our basic simulation tool[25] The design and development of LTE-Sim support thefollowing aspects

(i) The Evolved Universal Terrestrial Radio Access (E-UTRAN) and the Evolved Packet System (EPS)

(ii) Single and heterogeneous multicell environments(iii) Multiusers environment(iv) User mobility(v) Handover procedures(vi) Frequency reuse techniques(vii) Quality-of-service (QoS) management(viii) Scheduling strategies

However LTE-Sim still lacks some important functional-ities and features that are required in this workTherefore wemodified LTE-Sim from several aspects Our modification toLTE-Sim includes the following

(1) Add

(i) radio link failure(ii) handover failure(iii) HOM and TTT(iv) proposed DHO approach

(2) Modify

(i) mobility pattern using the seed to generate thesame sequence of random numbers each timewe run the simulation

The simulation parameter values are shown in Table 2unless otherwise specified

Figure 6 shows the simulation result when the TX poweris set as 20 dBm The horizontal axis denotes the mobilityproblem index Indexes 1 to 5 mean the mobility problem oftoo-late HO too-early HO HO to wrong cell ping-pongHOand unnecessary HO respectively Index 6 means the KPIwhich is the weighted average of the five MPRs The verticalaxismeans the corresponding values ofMPR orKPIThe bluebars show the performance for a static HO parameter settingwith 0 dB HOM and 0 s TTT and the green ones for anotherstatic setting with 10 dBHOMand 512 s TTTThe yellow barsshow the performance of the proposed DHO approach Theresults shown in Figure 6 validate that the proposed DHOapproach can effectively mitigate the mobility problems andachieve better MPRs and KPI

Figure 7 shows the simulation result when the TX poweris set as 43 dBm Since the TX power is higher than the onein the previous scenario in Figure 6 it is less likely that an

Mobile Information Systems 9

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

Figure 6 Simulation result when the TX power is set as 20 dBm

Table 2 Simulation parameters

Parameter ValuesNumber of small cell base stations 100Number of mobile stations 100eNB Tx power 15 dBm20 dBm43 dBmCarrier frequency 2GHzMS speed 30 kmhMobility model ManhattanTTI 1msSubcarrier spacing 15 KHzResource block 180KHzSuper frame time 10msNoise figure 25Noise spectral density minus174 dBmSINR threshold minus65 dBSimulation time 5000 sHandover delay 30msSystem bandwidth 5MHz 25 RBsTTI

RLF occurs Therefore the MPRs and KPI in Figure 7 arelower than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

Figure 8 shows the simulation result when the TX poweris set as 15 dBm Since the TX power is lower than the onein the previous scenario in Figure 6 it is more likely that anRLF occurs Therefore the MPRs and KPI in Figure 8 arehigher than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

We compare the KPI improvement under the variousvalues of TX power The KPI improvement is calculatedas the difference between the initial-case KPI (HOM 0dB

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 7 Simulation result when the TX power is set as 43 dBm

DHO approach

2 3 4 5 61Mobility problem index

0

001

002

003

004

005

006

007

008

MPR

KPI

HOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 8 Simulation result when the TX power is set as 15 dBm

TTT 0 s) and the KPI of the proposed DHO methoddivided by the initial-case KPI Figure 9 shows that the KPIimprovement ranges from 15 to 20 It also shows that theKPI improvement is more significant when the TX power islower that is the transmission range of the base station issmaller

6 Conclusion

In this work we investigate the mobility problems innext generation mobile communication networks There aretotally five different mobility problems each with different

10 Mobile Information Systems

20 4315TX power (dBm)

0

5

10

15

20

25

KPI i

mpr

ovem

ent (

)

Figure 9 KPI improvement with various TX power

characteristics We propose a data-driven handover opti-mization (DHO) approach which aims to mitigate mobilityproblems The DHO approach consists of four parts

(1) Mobility problem identification(2) KPI design(3) KPI function estimation(4) Handover parameter optimization

The DHO approach collects data from the mobile com-munication measurement results and provides a model toestimate the relationship between the KPI and features fromthe collected dataset Based on the model the handoverparameters including the HOM and TTT are optimized tominimize the KPI which is a weighted average of differentmobility problem ratios Simulation results show that theproposed DHO approach could effectively mitigate mobilityproblems and provide better connectivity

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work was supported in part by Ministry of Scienceand Technology (MOST) Taipei Taiwan under Grants nosMOST 103-2221-E-155-005 MOST 104-2221-E-155-017 andMOST 105-2218-E-155-001

References

[1] J G Andrews S Buzzi W Choi et al ldquoWhat will 5G berdquo IEEEJournal on Selected Areas in Communications vol 32 no 6 pp1065ndash1082 2014

[2] K Kitagawa T Komine T Yamamoto and S Konishi ldquoAhandover optimization algorithm with mobility robustness forLTE systemsrdquo in Proceedings of the IEEE 22nd International

Symposium on Personal Indoor and Mobile Radio Communica-tions (PIMRC rsquo11) pp 1647ndash1651 Toronto Canada September2011

[3] 3GPP ldquoSelf-configuring and self-optimizing network (SON)usecases and solutionsrdquo TR 36902 V931 2011

[4] A Awada BWegmann I Viering and A Klein ldquoA SON-basedalgorithm for the optimization of inter-RAT handover param-etersrdquo IEEE Transactions on Vehicular Technology vol 62 no 5pp 1906ndash1923 2013

[5] T Jansen I Balan J Turk I Moerman and T Kurner ldquoHando-ver parameter optimization in LTE self-organizing networksrdquoin Proceedings of the IEEE 72nd Vehicular Technology Confer-ence Fall (VTC 2010-Fall rsquo10) pp 1ndash5 IEEE Ottawa CanadaSeptember 2010

[6] T Jansen I Balan S Stefanski I Moerman and T KurnerldquoWeighted performance based handover parameter optimiza-tion in LTErdquo in Proceedings of the IEEE 73rd Vehicular Technol-ogy Conference (VTC rsquo11) pp 1ndash5 May 2011

[7] W Li X Duan S Jia L Zhang Y Liu and J Lin ldquoA dynamichysteresis-adjusting algorithm in LTE self-organization net-worksrdquo in Proceedings of the IEEE 75th Vehicular TechnologyConference (VTC Spring rsquo12) pp 1ndash5 Yokohama Japan May2012

[8] A Lobinger S Stefanski T Jansen and I Balan ldquoCoordinatinghandover parameter optimization and load balancing in LTEself-optimizing networksrdquo in Proceedings of the IEEE 73rdVehicular Technology Conference (VTC rsquo11) pp 1ndash5 May 2011

[9] V Capdevielle A Feki and E Sorsy ldquoJoint interferencemanagement and handover optimization in LTE small cellsnetworkrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6769ndash6773 IEEE OttawaCanada June 2012

[10] T Ali and M Saquib ldquoWLANcellular handover analysis fordifferent mobility modelsrdquo in Proceedings of the IEEE Interna-tional Conference on Communications (ICC rsquo12) pp 2758ndash2762Ottawa Canada June 2012

[11] A Awada B Wegmann I Viering and A Klein ldquoCell-pairspecific optimization of the inter-RAT handover parameters inSONrdquo in Proceedings of the IEEE 23rd International Symposiumon Personal Indoor andMobile Radio Communications (PIMRCrsquo12) pp 1168ndash1173 Sydney Australia September 2012

[12] D Giacomini and A Agarwal ldquoVertical handover decisionmaking using QoS reputation and GM(11) predictionrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo12) pp 5655ndash5659 IEEE Ottawa Canada June2012

[13] D Lopez-Perez I Guvenc and X Chu ldquoTheoretical analysisof handover failure and ping-pong rates for heterogeneousnetworksrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6774ndash6779 IEEE OttawaCanada June 2012

[14] A Rath and S Panwar ldquoFast handover in cellular networkswith femtocellsrdquo in Proceedings of the International Conferenceon Communications (ICC rsquo12) pp 2752ndash2757 IEEE OttawaCanada June 2012

[15] D Xenakis N Passas and C Verikoukis ldquoA novel handoverdecision policy for reducing power transmissions in the two-tier LTE networkrdquo in Proceedings of the IEEE International Con-ference on Communications (ICC rsquo12) pp 1352ndash1356 OttawaCanada June 2012

[16] T Taleb K Samdanis and A Ksentini ldquoSupporting highlymobile users in cost-effective decentralized mobile operator

Mobile Information Systems 11

networksrdquo IEEE Transactions on Vehicular Technology vol 63no 7 pp 3381ndash3396 2014

[17] A Imran A Zoha and A Abu-Dayya ldquoChallenges in 5G howto empower SONwith big data for enabling 5Grdquo IEEE Networkvol 28 no 6 pp 27ndash33 2014

[18] H-D Bae B Ryu and N-H Park ldquoAnalysis of handoverfailures in LTE femtocell systemsrdquo in Proceedings of the Aus-tralasian Telecommunication Networks and Applications Con-ference (ATNAC rsquo11) pp 1ndash5 IEEE Melbourne AustraliaNovember 2011

[19] S Haykin Neural Networks A Comprehensive FoundationPrentice-Hall 1999

[20] L Behera S Kumar and A Patnaik ldquoOn adaptive learning ratethat guarantees convergence in feedforward networksrdquo IEEETransactions on Neural Networks vol 17 no 5 pp 1116ndash11252006

[21] J A Snyman Practical Mathematical Optimization An Intro-duction to Basic Optimization Theory and Classical and NewGradient-Based Algorithms vol 97 of Applied OptimizationSpringer New York NY USA 2005

[22] M Gudmundson ldquoCell planning in Manhattan environmentsrdquoin Proceedings of the 42nd Vehicular Technology Conference pp435ndash438 IEEE Denver Colo USA May 1992

[23] P-E Ostling ldquoImplications of cell planning on handoff per-formance in manhattan environmentsrdquo in Proceedings of the5th IEEE International Symposium on Personal Indoor andMobile Radio Communications Wireless NetworksmdashCatchingthe Mobile Future vol 2 pp 625ndash629 The Hague The Nether-lands September 1994

[24] M Karabacak D Wang H Ishii and H Arslan ldquoMobilityperformance of macrocell-assisted small cells in Manhattanmodelrdquo in Proceedings of the 79th IEEE Vehicular TechnologyConference (VTC-Spring rsquo14) pp 1ndash5 Seoul South Korea May2014

[25] 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

4 6 8 102Number of hidden neurons

10minus11

10minus10

10minus9

10minus8

10minus7

10minus6

10minus5

MSE

Figure 2 MSE of the KPI function estimation with various numberof hidden neurons

431 Forward Pass The purpose of the forward pass is toget the actual response of the MLP in accordance with thespecified input signal In the forward pass the feature vectorX(119899) at iteration 119899 is applied to the input layer and propagatesthrough the network layer by layer from left to right Thetop part of Figure 3 shows the signal flow of the forwardpass 119883

1(119899) denotes the HOM and 119883

2(119899) denotes the TTT

1198833(119899) to 119883

119898(119899) denote the other features When the signal

propagates through a synapse its value is multiplied bythe synaptic weight The input of a neuron is the synapticweighted sum of the output of its previous layer For examplethe input of the 119895th neuron in the 119897th layer is

V(119897)119895(119899) =

119898119897minus1

sum119894=0

119908(119897)

119895119894(119899) 119910(119897minus1)

119894(119899) (4)

where 119894 = 0 represents the bias termThe input of this neuronis applied to the activation function 120593(sdot) The activationfunction should be differentiable everywhere We adopt thecommonly used sigmoid function as

119910(119897)

119895(119899) = 120593 (V(119897)

119895(119899)) =

1

1 + exp (minusV(119897)119895(119899))

(5)

The output of this neuron propagates through the network tothe next layer and the same operation is performed againFinally the signal aggregates in the output layer and theestimated KPI function (119899) is obtained During the forwardpass the synaptic weights are all fixed

432 Backward Pass The purpose of the backward pass isto adjust the synaptic weights in order to minimize the errorbetween the actual response (119899) and the desired response119884(119899) The error signal 119890(119899) is defined by

119890 (119899) = 119884 (119899) minus (119899) (6)

The instantaneous square error 119864(119899) is defined as

119864 (119899) =1

21198902(119899) =

1

2(119884 (119899) minus (119899))

2

(7)

The goal of the back propagation algorithm is to minimize119864(119899) To achieve this goal the back propagation algorithmapplies a respective correction Δ120596

(119897)

119895119894(119899) to every synaptic

weight 120596(119897)

119895119894(119899) The correction Δ120596

(119897)

119895119894(119899) is designed to be

proportional to the gradient 120597119864(119899)120597120596(119897)119895119894(119899) that is

Δ120596(119897)

119895119894(119899) = minus120578

119886

120597119864 (119899)

120597120596(119897)

119895119894(119899)

(8)

The minus sign comes from the fact that lower 119864(119899) ispreferred 120578

119886is the adaptive learning rate proposed by Behera

et al in [20] which is expressed as

120578119886= 120583

100381710038171003817100381710038171003817100381710038172

1003817100381710038171003817J11987910038171003817100381710038172 (9)

where = 119884(119899) minus (119899) and J = 120597120597W119897 While consid-

ering the weight correction for 119897 = 2 that is the outputlayer according to the chain rule of calculus the gradient120597119864(119899)120597120596

(2)

1119894(119899) can be written as

120597119864 (119899)

120597120596(2)

1119894(119899)

=120597119864 (119899)

120597119890 (119899)

120597119890 (119899)

120597 (119899)

120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597120596(2)

1119894(119899)

(10)

Applying (4)ndash(7) to the four partial derivatives at the right ofthe equal sign in (10) it can be rewritten as

120597119864 (119899)

120597120596(2)

1119894(119899)

= minus119890 (119899) 1205931015840(V(2)1

(119899)) 119910(1)

119894(119899)

= minus120575(2)

1(119899) 119910(1)

119894(119899)

(11)

where 120575(119897)119895(119899) is the local gradient which is defined as

120575(119897)

119895(119899) = minus

120597119864 (119899)

120597V(119897)119895(119899)

(12)

While considering the weight correction for 119897 = 1 that isthe hidden layer by iterative backpropagation and chain rulemanners the local gradient 120575(1)

119895(119899) is

120575(1)

119895(119899) = 120593

1015840(V(1)119895

(119899)) 120575(2)

1(119899) 120596(2)

1119895(119899) (13)

The derivation of the local gradients at each layer is shown inthe bottom part of Figure 3 Finally every synaptic weight iscorrected following this formula

120596(119897)

119895119894(119899 + 1) = 120596

(119897)

119895119894(119899) + Δ120596

(119897)

119895119894(119899) (14)

The forward pass and the backward pass are executed iter-atively until the square error 119864(119899) reaches some satisfactorylevel or when the number of epoch exceeds some thresholdAfter that the handover parameter optimization is executed

6 Mobile Information Systems

X0(n) = +1

X1(n)

X2(n)

X3(n)

120596(1)10 (n)

120596(1)20 (n)

120596(1)30 (n)

120596(1)40 (n)

120596(1)1m(n)

120596(1)2m(n)

120596(1)3m(n)

120596(1)4m(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120575(1)1 (n) 120575(1)

2 (n) 120575(1)3 (n) 120575(1)

4 (n)

+1

minus1

120575(2)1 (n)

e1(n)

120596(2)10 (n)

120596(2)11 (n)

120596(2)12 (n)

120596(2)13 (n)

120596(2)14 (n)

120596(2)11 (n)

120596(2)12 (n)

120596(2)13 (n)

120596(2)14 (n)

(2)1 (n)

Y(n)

Y(n) = y(2)1 (n)

Xm(n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

(n)119831D

(n)119831M

Figure 3 The forward pass and the backward pass of the BP algorithm

44 Handover Parameter Optimization In this paper thehandover optimization problem is solved by the gradientdescent algorithm [21] Here the handover parameters X

119863

including HOM and TTT are adjusted adaptively To find thelocal minimum of the KPI function 119884 asymp (X) using gradientdescent one takes steps proportional to the gradient Supposethat at the 119899th time of adjustment the handover parametersareX119863(119899) and the KPI is119884(119899) At the next time of adjustment

the handover parameters X119863are set as

X119863 (119899 + 1) = X

119863 (119899) + ΔX119863 (119899) (15)

ΔX119863(119899) depends on the gradient of estimated KPI 119884(119899) with

respect to X119863(119899) that is

ΔX119863 (119899) = 120583

120597119884 (119899)

120597X119863 (119899)

(16)

where 120583 is the constant adjustment rateHow to obtain the gradient 120597119884120597X

119863is a problem leftThe

original BP algorithmdescribed in Section 43 only consists oftwo passes the forward pass and the backward pass In orderto obtain the gradient 120597119884120597X

119863 the proposed DHO approach

modifies the original BP algorithm to add the third pass theadjustment pass We provide the detailed description of theadjustment pass in the following paragraphs

The purpose of the adjustment pass is to obtain theestimated gradient 120597(119899)120597X

119863(119899) which is the key to adjust

the handover parameter in order to minimize the KPI Thesynaptic weights that have been adjusted well in the backwardpass in accordance with (14) are set as fixed in the adjustmentpass In the following we show how to obtain the estimatedgradient 120597(119899)120597X

119863(119899) The signal flow of the adjustment

pass is depicted in Figure 4First of all we define the local gradient 120582(119897)

119895(119899) of the

adjustment pass as

120582(119897)

119895(119899) =

120597 (119899)

120597V(119897)119895(119899)

(17)

This local gradient 120582(119897)119895(119899) is similar to the local gradient

120575(119897)

119895(119899) of the backward pass which is defined in (12) except

that the target of the partial derivative is changed to (119899)While considering the output layer the local gradient 120582(2)

1(119899)

is

120582(2)

1(119899) =

120597 (119899)

120597V(2)1

(119899)= 1205931015840(V(2)1

(119899)) (18)

Mobile Information Systems 7

Y(n) = y(2)1 (n)

120582(1)1 (n) 120582(1)

2 (n) 120582(1)3 (n) 120582(1)

4 (n) 120582(2)1 (n)

120597Y(n)

120597X1(n)

120597Y(n)

120597X2(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

+1

(2)1 (n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

X0(n) = +1

X1(n)

X2(n)

X3(n)

Xm(n)

120596(1)10 (n + 1)

120596(1)20 (n + 1)

120596(1)30 (n + 1)

120596(1)40 (n + 1)

120596(1)1m(n + 1)

120596(1)2m(n + 1)

120596(1)3m(n + 1)

120596(1)4m(n + 1)

120596(1)11 (n + 1)

120596(1)12 (n + 1)

120596(1)21 (n + 1)

120596(1)22 (n + 1)

120596(1)31 (n + 1)

120596(1)32 (n + 1)

120596(1)41 (n + 1)

120596(1)42 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

120596(2)10 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

(n)119831D

(n)119831M

Figure 4 Adjustment pass of the proposed DHO approach

While considering the hidden layer the local gradient 120582(1)119895(119899)

is

120582(1)

119895(119899) =

120597 (119899)

120597V(1)119895

(119899)=

120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119910(1)

119895(119899)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

= 120582(2)

1(119899) 120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899))

(19)

Using the results from (18) and (19) the gradient can bewritten as follows

120597 (119899)

120597119883119894 (119899)

=120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

= 120582(2)

1(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

(20)

where the last equality comes from (18) Since

V(2)1

(119899) =

4

sum119895=0

120596(2)

1119895(119899 + 1) 119910

(1)

119895(119899) (21)

the second term at the most right of (20) can be written as

120597V(2)1

(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(n)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899)) 120596(1)

1198951(119899 + 1)

(22)

where the third equality comes from the fact that 119910(1)0(119899) is

always equal to 1 and its partial derivative can be ignoredfrom the summation Combining (19) and (22) into (20) thegradient 120597(119899)120597119883

119894(119899) is obtained as

120597 (119899)

120597119883119894 (119899)

=

4

sum119895=1

120582(1)

119895(119899) 120596(1)

1198951(119899 + 1) (23)

The derivation of the local gradients at each layer and thegradient 120597(119899)120597119883

119894(119899) is depicted in Figure 4 Based on the

8 Mobile Information Systems

05 1 15 2 25 3 35 4 45 50X-coordinate (km)

0

05

1

15

2

25

3

35

4

45

5

Y-c

oord

inat

e (km

)

Figure 5 Topology of the simulated mobile communication net-work

result from (23) the handover parameter119883119894is adjusted as (15)

and (16) showIn order to cope with the time-varying channel condi-

tions we use online training and operate the three passesiteratively to model the instantaneous relationship 119884 = 119891(X)and optimize the handover parameters HOM and TTT

5 Performance Evaluation and Discussion

In this section we describe the simulation environmentsimulation results and discussions

In our simulation a mobile communication networkwhich consists of 100 small cells is built in a 5 km times 5 kmarea Figure 5 shows the coordinates of these small cell basestations Each cell has 5MHz bandwidth with 25 resourceblocks and 2GHz carrier frequency Each resource blockconsists of 12 subcarriers of size 15 kHz each A time slot is05ms in duration and the transmission time interval (TTI)is 1ms

In the simulation 100 MSs are uniformly distributedover the area with random initialized positions The mobilitytrace of these 100 MSs is generated by using the Manhattanmobility model which is widely used for urban areas andsuitable to implement realistic simulations [22ndash24] In theManhattanmobility model MSs move in either horizontal orvertical directions on an urban map At each intersection anMS employs a probabilistic approach to choose the directionof its movements The probabilities of going straight turningleft and turning right are 05 025 and 025 respectivelyThe blue line in Figure 5 shows an illustrative moving pathof an MS in the mobile communication network Whileconsidering the KPI the weights of mobility problems are setto be equal

An RLF will occur if the SINR level of the servingcell drops below minus65 dB before handover execution is

completed [18] We set this as our SINR threshold valuesuch that if the SINR from the source BS is lower than thethreshold an RLF will occur The simulation time is set as5000 seconds The operating point with 0 dB HOM and 0 sTTT is the starting operating point for all cells in the network

In this paper we use LTE-Sim an open source frameworkto simulate LTE networks as our basic simulation tool[25] The design and development of LTE-Sim support thefollowing aspects

(i) The Evolved Universal Terrestrial Radio Access (E-UTRAN) and the Evolved Packet System (EPS)

(ii) Single and heterogeneous multicell environments(iii) Multiusers environment(iv) User mobility(v) Handover procedures(vi) Frequency reuse techniques(vii) Quality-of-service (QoS) management(viii) Scheduling strategies

However LTE-Sim still lacks some important functional-ities and features that are required in this workTherefore wemodified LTE-Sim from several aspects Our modification toLTE-Sim includes the following

(1) Add

(i) radio link failure(ii) handover failure(iii) HOM and TTT(iv) proposed DHO approach

(2) Modify

(i) mobility pattern using the seed to generate thesame sequence of random numbers each timewe run the simulation

The simulation parameter values are shown in Table 2unless otherwise specified

Figure 6 shows the simulation result when the TX poweris set as 20 dBm The horizontal axis denotes the mobilityproblem index Indexes 1 to 5 mean the mobility problem oftoo-late HO too-early HO HO to wrong cell ping-pongHOand unnecessary HO respectively Index 6 means the KPIwhich is the weighted average of the five MPRs The verticalaxismeans the corresponding values ofMPR orKPIThe bluebars show the performance for a static HO parameter settingwith 0 dB HOM and 0 s TTT and the green ones for anotherstatic setting with 10 dBHOMand 512 s TTTThe yellow barsshow the performance of the proposed DHO approach Theresults shown in Figure 6 validate that the proposed DHOapproach can effectively mitigate the mobility problems andachieve better MPRs and KPI

Figure 7 shows the simulation result when the TX poweris set as 43 dBm Since the TX power is higher than the onein the previous scenario in Figure 6 it is less likely that an

Mobile Information Systems 9

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

Figure 6 Simulation result when the TX power is set as 20 dBm

Table 2 Simulation parameters

Parameter ValuesNumber of small cell base stations 100Number of mobile stations 100eNB Tx power 15 dBm20 dBm43 dBmCarrier frequency 2GHzMS speed 30 kmhMobility model ManhattanTTI 1msSubcarrier spacing 15 KHzResource block 180KHzSuper frame time 10msNoise figure 25Noise spectral density minus174 dBmSINR threshold minus65 dBSimulation time 5000 sHandover delay 30msSystem bandwidth 5MHz 25 RBsTTI

RLF occurs Therefore the MPRs and KPI in Figure 7 arelower than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

Figure 8 shows the simulation result when the TX poweris set as 15 dBm Since the TX power is lower than the onein the previous scenario in Figure 6 it is more likely that anRLF occurs Therefore the MPRs and KPI in Figure 8 arehigher than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

We compare the KPI improvement under the variousvalues of TX power The KPI improvement is calculatedas the difference between the initial-case KPI (HOM 0dB

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 7 Simulation result when the TX power is set as 43 dBm

DHO approach

2 3 4 5 61Mobility problem index

0

001

002

003

004

005

006

007

008

MPR

KPI

HOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 8 Simulation result when the TX power is set as 15 dBm

TTT 0 s) and the KPI of the proposed DHO methoddivided by the initial-case KPI Figure 9 shows that the KPIimprovement ranges from 15 to 20 It also shows that theKPI improvement is more significant when the TX power islower that is the transmission range of the base station issmaller

6 Conclusion

In this work we investigate the mobility problems innext generation mobile communication networks There aretotally five different mobility problems each with different

10 Mobile Information Systems

20 4315TX power (dBm)

0

5

10

15

20

25

KPI i

mpr

ovem

ent (

)

Figure 9 KPI improvement with various TX power

characteristics We propose a data-driven handover opti-mization (DHO) approach which aims to mitigate mobilityproblems The DHO approach consists of four parts

(1) Mobility problem identification(2) KPI design(3) KPI function estimation(4) Handover parameter optimization

The DHO approach collects data from the mobile com-munication measurement results and provides a model toestimate the relationship between the KPI and features fromthe collected dataset Based on the model the handoverparameters including the HOM and TTT are optimized tominimize the KPI which is a weighted average of differentmobility problem ratios Simulation results show that theproposed DHO approach could effectively mitigate mobilityproblems and provide better connectivity

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work was supported in part by Ministry of Scienceand Technology (MOST) Taipei Taiwan under Grants nosMOST 103-2221-E-155-005 MOST 104-2221-E-155-017 andMOST 105-2218-E-155-001

References

[1] J G Andrews S Buzzi W Choi et al ldquoWhat will 5G berdquo IEEEJournal on Selected Areas in Communications vol 32 no 6 pp1065ndash1082 2014

[2] K Kitagawa T Komine T Yamamoto and S Konishi ldquoAhandover optimization algorithm with mobility robustness forLTE systemsrdquo in Proceedings of the IEEE 22nd International

Symposium on Personal Indoor and Mobile Radio Communica-tions (PIMRC rsquo11) pp 1647ndash1651 Toronto Canada September2011

[3] 3GPP ldquoSelf-configuring and self-optimizing network (SON)usecases and solutionsrdquo TR 36902 V931 2011

[4] A Awada BWegmann I Viering and A Klein ldquoA SON-basedalgorithm for the optimization of inter-RAT handover param-etersrdquo IEEE Transactions on Vehicular Technology vol 62 no 5pp 1906ndash1923 2013

[5] T Jansen I Balan J Turk I Moerman and T Kurner ldquoHando-ver parameter optimization in LTE self-organizing networksrdquoin Proceedings of the IEEE 72nd Vehicular Technology Confer-ence Fall (VTC 2010-Fall rsquo10) pp 1ndash5 IEEE Ottawa CanadaSeptember 2010

[6] T Jansen I Balan S Stefanski I Moerman and T KurnerldquoWeighted performance based handover parameter optimiza-tion in LTErdquo in Proceedings of the IEEE 73rd Vehicular Technol-ogy Conference (VTC rsquo11) pp 1ndash5 May 2011

[7] W Li X Duan S Jia L Zhang Y Liu and J Lin ldquoA dynamichysteresis-adjusting algorithm in LTE self-organization net-worksrdquo in Proceedings of the IEEE 75th Vehicular TechnologyConference (VTC Spring rsquo12) pp 1ndash5 Yokohama Japan May2012

[8] A Lobinger S Stefanski T Jansen and I Balan ldquoCoordinatinghandover parameter optimization and load balancing in LTEself-optimizing networksrdquo in Proceedings of the IEEE 73rdVehicular Technology Conference (VTC rsquo11) pp 1ndash5 May 2011

[9] V Capdevielle A Feki and E Sorsy ldquoJoint interferencemanagement and handover optimization in LTE small cellsnetworkrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6769ndash6773 IEEE OttawaCanada June 2012

[10] T Ali and M Saquib ldquoWLANcellular handover analysis fordifferent mobility modelsrdquo in Proceedings of the IEEE Interna-tional Conference on Communications (ICC rsquo12) pp 2758ndash2762Ottawa Canada June 2012

[11] A Awada B Wegmann I Viering and A Klein ldquoCell-pairspecific optimization of the inter-RAT handover parameters inSONrdquo in Proceedings of the IEEE 23rd International Symposiumon Personal Indoor andMobile Radio Communications (PIMRCrsquo12) pp 1168ndash1173 Sydney Australia September 2012

[12] D Giacomini and A Agarwal ldquoVertical handover decisionmaking using QoS reputation and GM(11) predictionrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo12) pp 5655ndash5659 IEEE Ottawa Canada June2012

[13] D Lopez-Perez I Guvenc and X Chu ldquoTheoretical analysisof handover failure and ping-pong rates for heterogeneousnetworksrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6774ndash6779 IEEE OttawaCanada June 2012

[14] A Rath and S Panwar ldquoFast handover in cellular networkswith femtocellsrdquo in Proceedings of the International Conferenceon Communications (ICC rsquo12) pp 2752ndash2757 IEEE OttawaCanada June 2012

[15] D Xenakis N Passas and C Verikoukis ldquoA novel handoverdecision policy for reducing power transmissions in the two-tier LTE networkrdquo in Proceedings of the IEEE International Con-ference on Communications (ICC rsquo12) pp 1352ndash1356 OttawaCanada June 2012

[16] T Taleb K Samdanis and A Ksentini ldquoSupporting highlymobile users in cost-effective decentralized mobile operator

Mobile Information Systems 11

networksrdquo IEEE Transactions on Vehicular Technology vol 63no 7 pp 3381ndash3396 2014

[17] A Imran A Zoha and A Abu-Dayya ldquoChallenges in 5G howto empower SONwith big data for enabling 5Grdquo IEEE Networkvol 28 no 6 pp 27ndash33 2014

[18] H-D Bae B Ryu and N-H Park ldquoAnalysis of handoverfailures in LTE femtocell systemsrdquo in Proceedings of the Aus-tralasian Telecommunication Networks and Applications Con-ference (ATNAC rsquo11) pp 1ndash5 IEEE Melbourne AustraliaNovember 2011

[19] S Haykin Neural Networks A Comprehensive FoundationPrentice-Hall 1999

[20] L Behera S Kumar and A Patnaik ldquoOn adaptive learning ratethat guarantees convergence in feedforward networksrdquo IEEETransactions on Neural Networks vol 17 no 5 pp 1116ndash11252006

[21] J A Snyman Practical Mathematical Optimization An Intro-duction to Basic Optimization Theory and Classical and NewGradient-Based Algorithms vol 97 of Applied OptimizationSpringer New York NY USA 2005

[22] M Gudmundson ldquoCell planning in Manhattan environmentsrdquoin Proceedings of the 42nd Vehicular Technology Conference pp435ndash438 IEEE Denver Colo USA May 1992

[23] P-E Ostling ldquoImplications of cell planning on handoff per-formance in manhattan environmentsrdquo in Proceedings of the5th IEEE International Symposium on Personal Indoor andMobile Radio Communications Wireless NetworksmdashCatchingthe Mobile Future vol 2 pp 625ndash629 The Hague The Nether-lands September 1994

[24] M Karabacak D Wang H Ishii and H Arslan ldquoMobilityperformance of macrocell-assisted small cells in Manhattanmodelrdquo in Proceedings of the 79th IEEE Vehicular TechnologyConference (VTC-Spring rsquo14) pp 1ndash5 Seoul South Korea May2014

[25] 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

X0(n) = +1

X1(n)

X2(n)

X3(n)

120596(1)10 (n)

120596(1)20 (n)

120596(1)30 (n)

120596(1)40 (n)

120596(1)1m(n)

120596(1)2m(n)

120596(1)3m(n)

120596(1)4m(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120575(1)1 (n) 120575(1)

2 (n) 120575(1)3 (n) 120575(1)

4 (n)

+1

minus1

120575(2)1 (n)

e1(n)

120596(2)10 (n)

120596(2)11 (n)

120596(2)12 (n)

120596(2)13 (n)

120596(2)14 (n)

120596(2)11 (n)

120596(2)12 (n)

120596(2)13 (n)

120596(2)14 (n)

(2)1 (n)

Y(n)

Y(n) = y(2)1 (n)

Xm(n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

(n)119831D

(n)119831M

Figure 3 The forward pass and the backward pass of the BP algorithm

44 Handover Parameter Optimization In this paper thehandover optimization problem is solved by the gradientdescent algorithm [21] Here the handover parameters X

119863

including HOM and TTT are adjusted adaptively To find thelocal minimum of the KPI function 119884 asymp (X) using gradientdescent one takes steps proportional to the gradient Supposethat at the 119899th time of adjustment the handover parametersareX119863(119899) and the KPI is119884(119899) At the next time of adjustment

the handover parameters X119863are set as

X119863 (119899 + 1) = X

119863 (119899) + ΔX119863 (119899) (15)

ΔX119863(119899) depends on the gradient of estimated KPI 119884(119899) with

respect to X119863(119899) that is

ΔX119863 (119899) = 120583

120597119884 (119899)

120597X119863 (119899)

(16)

where 120583 is the constant adjustment rateHow to obtain the gradient 120597119884120597X

119863is a problem leftThe

original BP algorithmdescribed in Section 43 only consists oftwo passes the forward pass and the backward pass In orderto obtain the gradient 120597119884120597X

119863 the proposed DHO approach

modifies the original BP algorithm to add the third pass theadjustment pass We provide the detailed description of theadjustment pass in the following paragraphs

The purpose of the adjustment pass is to obtain theestimated gradient 120597(119899)120597X

119863(119899) which is the key to adjust

the handover parameter in order to minimize the KPI Thesynaptic weights that have been adjusted well in the backwardpass in accordance with (14) are set as fixed in the adjustmentpass In the following we show how to obtain the estimatedgradient 120597(119899)120597X

119863(119899) The signal flow of the adjustment

pass is depicted in Figure 4First of all we define the local gradient 120582(119897)

119895(119899) of the

adjustment pass as

120582(119897)

119895(119899) =

120597 (119899)

120597V(119897)119895(119899)

(17)

This local gradient 120582(119897)119895(119899) is similar to the local gradient

120575(119897)

119895(119899) of the backward pass which is defined in (12) except

that the target of the partial derivative is changed to (119899)While considering the output layer the local gradient 120582(2)

1(119899)

is

120582(2)

1(119899) =

120597 (119899)

120597V(2)1

(119899)= 1205931015840(V(2)1

(119899)) (18)

Mobile Information Systems 7

Y(n) = y(2)1 (n)

120582(1)1 (n) 120582(1)

2 (n) 120582(1)3 (n) 120582(1)

4 (n) 120582(2)1 (n)

120597Y(n)

120597X1(n)

120597Y(n)

120597X2(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

+1

(2)1 (n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

X0(n) = +1

X1(n)

X2(n)

X3(n)

Xm(n)

120596(1)10 (n + 1)

120596(1)20 (n + 1)

120596(1)30 (n + 1)

120596(1)40 (n + 1)

120596(1)1m(n + 1)

120596(1)2m(n + 1)

120596(1)3m(n + 1)

120596(1)4m(n + 1)

120596(1)11 (n + 1)

120596(1)12 (n + 1)

120596(1)21 (n + 1)

120596(1)22 (n + 1)

120596(1)31 (n + 1)

120596(1)32 (n + 1)

120596(1)41 (n + 1)

120596(1)42 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

120596(2)10 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

(n)119831D

(n)119831M

Figure 4 Adjustment pass of the proposed DHO approach

While considering the hidden layer the local gradient 120582(1)119895(119899)

is

120582(1)

119895(119899) =

120597 (119899)

120597V(1)119895

(119899)=

120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119910(1)

119895(119899)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

= 120582(2)

1(119899) 120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899))

(19)

Using the results from (18) and (19) the gradient can bewritten as follows

120597 (119899)

120597119883119894 (119899)

=120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

= 120582(2)

1(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

(20)

where the last equality comes from (18) Since

V(2)1

(119899) =

4

sum119895=0

120596(2)

1119895(119899 + 1) 119910

(1)

119895(119899) (21)

the second term at the most right of (20) can be written as

120597V(2)1

(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(n)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899)) 120596(1)

1198951(119899 + 1)

(22)

where the third equality comes from the fact that 119910(1)0(119899) is

always equal to 1 and its partial derivative can be ignoredfrom the summation Combining (19) and (22) into (20) thegradient 120597(119899)120597119883

119894(119899) is obtained as

120597 (119899)

120597119883119894 (119899)

=

4

sum119895=1

120582(1)

119895(119899) 120596(1)

1198951(119899 + 1) (23)

The derivation of the local gradients at each layer and thegradient 120597(119899)120597119883

119894(119899) is depicted in Figure 4 Based on the

8 Mobile Information Systems

05 1 15 2 25 3 35 4 45 50X-coordinate (km)

0

05

1

15

2

25

3

35

4

45

5

Y-c

oord

inat

e (km

)

Figure 5 Topology of the simulated mobile communication net-work

result from (23) the handover parameter119883119894is adjusted as (15)

and (16) showIn order to cope with the time-varying channel condi-

tions we use online training and operate the three passesiteratively to model the instantaneous relationship 119884 = 119891(X)and optimize the handover parameters HOM and TTT

5 Performance Evaluation and Discussion

In this section we describe the simulation environmentsimulation results and discussions

In our simulation a mobile communication networkwhich consists of 100 small cells is built in a 5 km times 5 kmarea Figure 5 shows the coordinates of these small cell basestations Each cell has 5MHz bandwidth with 25 resourceblocks and 2GHz carrier frequency Each resource blockconsists of 12 subcarriers of size 15 kHz each A time slot is05ms in duration and the transmission time interval (TTI)is 1ms

In the simulation 100 MSs are uniformly distributedover the area with random initialized positions The mobilitytrace of these 100 MSs is generated by using the Manhattanmobility model which is widely used for urban areas andsuitable to implement realistic simulations [22ndash24] In theManhattanmobility model MSs move in either horizontal orvertical directions on an urban map At each intersection anMS employs a probabilistic approach to choose the directionof its movements The probabilities of going straight turningleft and turning right are 05 025 and 025 respectivelyThe blue line in Figure 5 shows an illustrative moving pathof an MS in the mobile communication network Whileconsidering the KPI the weights of mobility problems are setto be equal

An RLF will occur if the SINR level of the servingcell drops below minus65 dB before handover execution is

completed [18] We set this as our SINR threshold valuesuch that if the SINR from the source BS is lower than thethreshold an RLF will occur The simulation time is set as5000 seconds The operating point with 0 dB HOM and 0 sTTT is the starting operating point for all cells in the network

In this paper we use LTE-Sim an open source frameworkto simulate LTE networks as our basic simulation tool[25] The design and development of LTE-Sim support thefollowing aspects

(i) The Evolved Universal Terrestrial Radio Access (E-UTRAN) and the Evolved Packet System (EPS)

(ii) Single and heterogeneous multicell environments(iii) Multiusers environment(iv) User mobility(v) Handover procedures(vi) Frequency reuse techniques(vii) Quality-of-service (QoS) management(viii) Scheduling strategies

However LTE-Sim still lacks some important functional-ities and features that are required in this workTherefore wemodified LTE-Sim from several aspects Our modification toLTE-Sim includes the following

(1) Add

(i) radio link failure(ii) handover failure(iii) HOM and TTT(iv) proposed DHO approach

(2) Modify

(i) mobility pattern using the seed to generate thesame sequence of random numbers each timewe run the simulation

The simulation parameter values are shown in Table 2unless otherwise specified

Figure 6 shows the simulation result when the TX poweris set as 20 dBm The horizontal axis denotes the mobilityproblem index Indexes 1 to 5 mean the mobility problem oftoo-late HO too-early HO HO to wrong cell ping-pongHOand unnecessary HO respectively Index 6 means the KPIwhich is the weighted average of the five MPRs The verticalaxismeans the corresponding values ofMPR orKPIThe bluebars show the performance for a static HO parameter settingwith 0 dB HOM and 0 s TTT and the green ones for anotherstatic setting with 10 dBHOMand 512 s TTTThe yellow barsshow the performance of the proposed DHO approach Theresults shown in Figure 6 validate that the proposed DHOapproach can effectively mitigate the mobility problems andachieve better MPRs and KPI

Figure 7 shows the simulation result when the TX poweris set as 43 dBm Since the TX power is higher than the onein the previous scenario in Figure 6 it is less likely that an

Mobile Information Systems 9

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

Figure 6 Simulation result when the TX power is set as 20 dBm

Table 2 Simulation parameters

Parameter ValuesNumber of small cell base stations 100Number of mobile stations 100eNB Tx power 15 dBm20 dBm43 dBmCarrier frequency 2GHzMS speed 30 kmhMobility model ManhattanTTI 1msSubcarrier spacing 15 KHzResource block 180KHzSuper frame time 10msNoise figure 25Noise spectral density minus174 dBmSINR threshold minus65 dBSimulation time 5000 sHandover delay 30msSystem bandwidth 5MHz 25 RBsTTI

RLF occurs Therefore the MPRs and KPI in Figure 7 arelower than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

Figure 8 shows the simulation result when the TX poweris set as 15 dBm Since the TX power is lower than the onein the previous scenario in Figure 6 it is more likely that anRLF occurs Therefore the MPRs and KPI in Figure 8 arehigher than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

We compare the KPI improvement under the variousvalues of TX power The KPI improvement is calculatedas the difference between the initial-case KPI (HOM 0dB

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 7 Simulation result when the TX power is set as 43 dBm

DHO approach

2 3 4 5 61Mobility problem index

0

001

002

003

004

005

006

007

008

MPR

KPI

HOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 8 Simulation result when the TX power is set as 15 dBm

TTT 0 s) and the KPI of the proposed DHO methoddivided by the initial-case KPI Figure 9 shows that the KPIimprovement ranges from 15 to 20 It also shows that theKPI improvement is more significant when the TX power islower that is the transmission range of the base station issmaller

6 Conclusion

In this work we investigate the mobility problems innext generation mobile communication networks There aretotally five different mobility problems each with different

10 Mobile Information Systems

20 4315TX power (dBm)

0

5

10

15

20

25

KPI i

mpr

ovem

ent (

)

Figure 9 KPI improvement with various TX power

characteristics We propose a data-driven handover opti-mization (DHO) approach which aims to mitigate mobilityproblems The DHO approach consists of four parts

(1) Mobility problem identification(2) KPI design(3) KPI function estimation(4) Handover parameter optimization

The DHO approach collects data from the mobile com-munication measurement results and provides a model toestimate the relationship between the KPI and features fromthe collected dataset Based on the model the handoverparameters including the HOM and TTT are optimized tominimize the KPI which is a weighted average of differentmobility problem ratios Simulation results show that theproposed DHO approach could effectively mitigate mobilityproblems and provide better connectivity

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work was supported in part by Ministry of Scienceand Technology (MOST) Taipei Taiwan under Grants nosMOST 103-2221-E-155-005 MOST 104-2221-E-155-017 andMOST 105-2218-E-155-001

References

[1] J G Andrews S Buzzi W Choi et al ldquoWhat will 5G berdquo IEEEJournal on Selected Areas in Communications vol 32 no 6 pp1065ndash1082 2014

[2] K Kitagawa T Komine T Yamamoto and S Konishi ldquoAhandover optimization algorithm with mobility robustness forLTE systemsrdquo in Proceedings of the IEEE 22nd International

Symposium on Personal Indoor and Mobile Radio Communica-tions (PIMRC rsquo11) pp 1647ndash1651 Toronto Canada September2011

[3] 3GPP ldquoSelf-configuring and self-optimizing network (SON)usecases and solutionsrdquo TR 36902 V931 2011

[4] A Awada BWegmann I Viering and A Klein ldquoA SON-basedalgorithm for the optimization of inter-RAT handover param-etersrdquo IEEE Transactions on Vehicular Technology vol 62 no 5pp 1906ndash1923 2013

[5] T Jansen I Balan J Turk I Moerman and T Kurner ldquoHando-ver parameter optimization in LTE self-organizing networksrdquoin Proceedings of the IEEE 72nd Vehicular Technology Confer-ence Fall (VTC 2010-Fall rsquo10) pp 1ndash5 IEEE Ottawa CanadaSeptember 2010

[6] T Jansen I Balan S Stefanski I Moerman and T KurnerldquoWeighted performance based handover parameter optimiza-tion in LTErdquo in Proceedings of the IEEE 73rd Vehicular Technol-ogy Conference (VTC rsquo11) pp 1ndash5 May 2011

[7] W Li X Duan S Jia L Zhang Y Liu and J Lin ldquoA dynamichysteresis-adjusting algorithm in LTE self-organization net-worksrdquo in Proceedings of the IEEE 75th Vehicular TechnologyConference (VTC Spring rsquo12) pp 1ndash5 Yokohama Japan May2012

[8] A Lobinger S Stefanski T Jansen and I Balan ldquoCoordinatinghandover parameter optimization and load balancing in LTEself-optimizing networksrdquo in Proceedings of the IEEE 73rdVehicular Technology Conference (VTC rsquo11) pp 1ndash5 May 2011

[9] V Capdevielle A Feki and E Sorsy ldquoJoint interferencemanagement and handover optimization in LTE small cellsnetworkrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6769ndash6773 IEEE OttawaCanada June 2012

[10] T Ali and M Saquib ldquoWLANcellular handover analysis fordifferent mobility modelsrdquo in Proceedings of the IEEE Interna-tional Conference on Communications (ICC rsquo12) pp 2758ndash2762Ottawa Canada June 2012

[11] A Awada B Wegmann I Viering and A Klein ldquoCell-pairspecific optimization of the inter-RAT handover parameters inSONrdquo in Proceedings of the IEEE 23rd International Symposiumon Personal Indoor andMobile Radio Communications (PIMRCrsquo12) pp 1168ndash1173 Sydney Australia September 2012

[12] D Giacomini and A Agarwal ldquoVertical handover decisionmaking using QoS reputation and GM(11) predictionrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo12) pp 5655ndash5659 IEEE Ottawa Canada June2012

[13] D Lopez-Perez I Guvenc and X Chu ldquoTheoretical analysisof handover failure and ping-pong rates for heterogeneousnetworksrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6774ndash6779 IEEE OttawaCanada June 2012

[14] A Rath and S Panwar ldquoFast handover in cellular networkswith femtocellsrdquo in Proceedings of the International Conferenceon Communications (ICC rsquo12) pp 2752ndash2757 IEEE OttawaCanada June 2012

[15] D Xenakis N Passas and C Verikoukis ldquoA novel handoverdecision policy for reducing power transmissions in the two-tier LTE networkrdquo in Proceedings of the IEEE International Con-ference on Communications (ICC rsquo12) pp 1352ndash1356 OttawaCanada June 2012

[16] T Taleb K Samdanis and A Ksentini ldquoSupporting highlymobile users in cost-effective decentralized mobile operator

Mobile Information Systems 11

networksrdquo IEEE Transactions on Vehicular Technology vol 63no 7 pp 3381ndash3396 2014

[17] A Imran A Zoha and A Abu-Dayya ldquoChallenges in 5G howto empower SONwith big data for enabling 5Grdquo IEEE Networkvol 28 no 6 pp 27ndash33 2014

[18] H-D Bae B Ryu and N-H Park ldquoAnalysis of handoverfailures in LTE femtocell systemsrdquo in Proceedings of the Aus-tralasian Telecommunication Networks and Applications Con-ference (ATNAC rsquo11) pp 1ndash5 IEEE Melbourne AustraliaNovember 2011

[19] S Haykin Neural Networks A Comprehensive FoundationPrentice-Hall 1999

[20] L Behera S Kumar and A Patnaik ldquoOn adaptive learning ratethat guarantees convergence in feedforward networksrdquo IEEETransactions on Neural Networks vol 17 no 5 pp 1116ndash11252006

[21] J A Snyman Practical Mathematical Optimization An Intro-duction to Basic Optimization Theory and Classical and NewGradient-Based Algorithms vol 97 of Applied OptimizationSpringer New York NY USA 2005

[22] M Gudmundson ldquoCell planning in Manhattan environmentsrdquoin Proceedings of the 42nd Vehicular Technology Conference pp435ndash438 IEEE Denver Colo USA May 1992

[23] P-E Ostling ldquoImplications of cell planning on handoff per-formance in manhattan environmentsrdquo in Proceedings of the5th IEEE International Symposium on Personal Indoor andMobile Radio Communications Wireless NetworksmdashCatchingthe Mobile Future vol 2 pp 625ndash629 The Hague The Nether-lands September 1994

[24] M Karabacak D Wang H Ishii and H Arslan ldquoMobilityperformance of macrocell-assisted small cells in Manhattanmodelrdquo in Proceedings of the 79th IEEE Vehicular TechnologyConference (VTC-Spring rsquo14) pp 1ndash5 Seoul South Korea May2014

[25] 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

Y(n) = y(2)1 (n)

120582(1)1 (n) 120582(1)

2 (n) 120582(1)3 (n) 120582(1)

4 (n) 120582(2)1 (n)

120597Y(n)

120597X1(n)

120597Y(n)

120597X2(n)

(1)1 (n)

(1)2 (n)

(1)3 (n)

(1)4 (n)

120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot) 120593998400(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

120593(middot)

+1

(2)1 (n)

y(1)1 (n)

y(1)2 (n)

y(1)3 (n)

y(1)4 (n)

X0(n) = +1

X1(n)

X2(n)

X3(n)

Xm(n)

120596(1)10 (n + 1)

120596(1)20 (n + 1)

120596(1)30 (n + 1)

120596(1)40 (n + 1)

120596(1)1m(n + 1)

120596(1)2m(n + 1)

120596(1)3m(n + 1)

120596(1)4m(n + 1)

120596(1)11 (n + 1)

120596(1)12 (n + 1)

120596(1)21 (n + 1)

120596(1)22 (n + 1)

120596(1)31 (n + 1)

120596(1)32 (n + 1)

120596(1)41 (n + 1)

120596(1)42 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

120596(2)10 (n + 1)

120596(2)11 (n + 1)

120596(2)12 (n + 1)

120596(2)13 (n + 1)

120596(2)14 (n + 1)

(n)119831D

(n)119831M

Figure 4 Adjustment pass of the proposed DHO approach

While considering the hidden layer the local gradient 120582(1)119895(119899)

is

120582(1)

119895(119899) =

120597 (119899)

120597V(1)119895

(119899)=

120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119910(1)

119895(119899)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

= 120582(2)

1(119899) 120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899))

(19)

Using the results from (18) and (19) the gradient can bewritten as follows

120597 (119899)

120597119883119894 (119899)

=120597 (119899)

120597V(2)1

(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

= 120582(2)

1(119899)

120597V(2)1

(119899)

120597119883119894 (119899)

(20)

where the last equality comes from (18) Since

V(2)1

(119899) =

4

sum119895=0

120596(2)

1119895(119899 + 1) 119910

(1)

119895(119899) (21)

the second term at the most right of (20) can be written as

120597V(2)1

(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597119883119894 (119899)

=

4

sum119895=0

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(119899)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1)

120597119910(1)

119895(119899)

120597V(1)119895

(n)

120597V(1)119895

(119899)

120597119883119894 (119899)

=

4

sum119895=1

120596(2)

1119895(119899 + 1) 120593

1015840(V(1)119895

(119899)) 120596(1)

1198951(119899 + 1)

(22)

where the third equality comes from the fact that 119910(1)0(119899) is

always equal to 1 and its partial derivative can be ignoredfrom the summation Combining (19) and (22) into (20) thegradient 120597(119899)120597119883

119894(119899) is obtained as

120597 (119899)

120597119883119894 (119899)

=

4

sum119895=1

120582(1)

119895(119899) 120596(1)

1198951(119899 + 1) (23)

The derivation of the local gradients at each layer and thegradient 120597(119899)120597119883

119894(119899) is depicted in Figure 4 Based on the

8 Mobile Information Systems

05 1 15 2 25 3 35 4 45 50X-coordinate (km)

0

05

1

15

2

25

3

35

4

45

5

Y-c

oord

inat

e (km

)

Figure 5 Topology of the simulated mobile communication net-work

result from (23) the handover parameter119883119894is adjusted as (15)

and (16) showIn order to cope with the time-varying channel condi-

tions we use online training and operate the three passesiteratively to model the instantaneous relationship 119884 = 119891(X)and optimize the handover parameters HOM and TTT

5 Performance Evaluation and Discussion

In this section we describe the simulation environmentsimulation results and discussions

In our simulation a mobile communication networkwhich consists of 100 small cells is built in a 5 km times 5 kmarea Figure 5 shows the coordinates of these small cell basestations Each cell has 5MHz bandwidth with 25 resourceblocks and 2GHz carrier frequency Each resource blockconsists of 12 subcarriers of size 15 kHz each A time slot is05ms in duration and the transmission time interval (TTI)is 1ms

In the simulation 100 MSs are uniformly distributedover the area with random initialized positions The mobilitytrace of these 100 MSs is generated by using the Manhattanmobility model which is widely used for urban areas andsuitable to implement realistic simulations [22ndash24] In theManhattanmobility model MSs move in either horizontal orvertical directions on an urban map At each intersection anMS employs a probabilistic approach to choose the directionof its movements The probabilities of going straight turningleft and turning right are 05 025 and 025 respectivelyThe blue line in Figure 5 shows an illustrative moving pathof an MS in the mobile communication network Whileconsidering the KPI the weights of mobility problems are setto be equal

An RLF will occur if the SINR level of the servingcell drops below minus65 dB before handover execution is

completed [18] We set this as our SINR threshold valuesuch that if the SINR from the source BS is lower than thethreshold an RLF will occur The simulation time is set as5000 seconds The operating point with 0 dB HOM and 0 sTTT is the starting operating point for all cells in the network

In this paper we use LTE-Sim an open source frameworkto simulate LTE networks as our basic simulation tool[25] The design and development of LTE-Sim support thefollowing aspects

(i) The Evolved Universal Terrestrial Radio Access (E-UTRAN) and the Evolved Packet System (EPS)

(ii) Single and heterogeneous multicell environments(iii) Multiusers environment(iv) User mobility(v) Handover procedures(vi) Frequency reuse techniques(vii) Quality-of-service (QoS) management(viii) Scheduling strategies

However LTE-Sim still lacks some important functional-ities and features that are required in this workTherefore wemodified LTE-Sim from several aspects Our modification toLTE-Sim includes the following

(1) Add

(i) radio link failure(ii) handover failure(iii) HOM and TTT(iv) proposed DHO approach

(2) Modify

(i) mobility pattern using the seed to generate thesame sequence of random numbers each timewe run the simulation

The simulation parameter values are shown in Table 2unless otherwise specified

Figure 6 shows the simulation result when the TX poweris set as 20 dBm The horizontal axis denotes the mobilityproblem index Indexes 1 to 5 mean the mobility problem oftoo-late HO too-early HO HO to wrong cell ping-pongHOand unnecessary HO respectively Index 6 means the KPIwhich is the weighted average of the five MPRs The verticalaxismeans the corresponding values ofMPR orKPIThe bluebars show the performance for a static HO parameter settingwith 0 dB HOM and 0 s TTT and the green ones for anotherstatic setting with 10 dBHOMand 512 s TTTThe yellow barsshow the performance of the proposed DHO approach Theresults shown in Figure 6 validate that the proposed DHOapproach can effectively mitigate the mobility problems andachieve better MPRs and KPI

Figure 7 shows the simulation result when the TX poweris set as 43 dBm Since the TX power is higher than the onein the previous scenario in Figure 6 it is less likely that an

Mobile Information Systems 9

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

Figure 6 Simulation result when the TX power is set as 20 dBm

Table 2 Simulation parameters

Parameter ValuesNumber of small cell base stations 100Number of mobile stations 100eNB Tx power 15 dBm20 dBm43 dBmCarrier frequency 2GHzMS speed 30 kmhMobility model ManhattanTTI 1msSubcarrier spacing 15 KHzResource block 180KHzSuper frame time 10msNoise figure 25Noise spectral density minus174 dBmSINR threshold minus65 dBSimulation time 5000 sHandover delay 30msSystem bandwidth 5MHz 25 RBsTTI

RLF occurs Therefore the MPRs and KPI in Figure 7 arelower than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

Figure 8 shows the simulation result when the TX poweris set as 15 dBm Since the TX power is lower than the onein the previous scenario in Figure 6 it is more likely that anRLF occurs Therefore the MPRs and KPI in Figure 8 arehigher than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

We compare the KPI improvement under the variousvalues of TX power The KPI improvement is calculatedas the difference between the initial-case KPI (HOM 0dB

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 7 Simulation result when the TX power is set as 43 dBm

DHO approach

2 3 4 5 61Mobility problem index

0

001

002

003

004

005

006

007

008

MPR

KPI

HOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 8 Simulation result when the TX power is set as 15 dBm

TTT 0 s) and the KPI of the proposed DHO methoddivided by the initial-case KPI Figure 9 shows that the KPIimprovement ranges from 15 to 20 It also shows that theKPI improvement is more significant when the TX power islower that is the transmission range of the base station issmaller

6 Conclusion

In this work we investigate the mobility problems innext generation mobile communication networks There aretotally five different mobility problems each with different

10 Mobile Information Systems

20 4315TX power (dBm)

0

5

10

15

20

25

KPI i

mpr

ovem

ent (

)

Figure 9 KPI improvement with various TX power

characteristics We propose a data-driven handover opti-mization (DHO) approach which aims to mitigate mobilityproblems The DHO approach consists of four parts

(1) Mobility problem identification(2) KPI design(3) KPI function estimation(4) Handover parameter optimization

The DHO approach collects data from the mobile com-munication measurement results and provides a model toestimate the relationship between the KPI and features fromthe collected dataset Based on the model the handoverparameters including the HOM and TTT are optimized tominimize the KPI which is a weighted average of differentmobility problem ratios Simulation results show that theproposed DHO approach could effectively mitigate mobilityproblems and provide better connectivity

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work was supported in part by Ministry of Scienceand Technology (MOST) Taipei Taiwan under Grants nosMOST 103-2221-E-155-005 MOST 104-2221-E-155-017 andMOST 105-2218-E-155-001

References

[1] J G Andrews S Buzzi W Choi et al ldquoWhat will 5G berdquo IEEEJournal on Selected Areas in Communications vol 32 no 6 pp1065ndash1082 2014

[2] K Kitagawa T Komine T Yamamoto and S Konishi ldquoAhandover optimization algorithm with mobility robustness forLTE systemsrdquo in Proceedings of the IEEE 22nd International

Symposium on Personal Indoor and Mobile Radio Communica-tions (PIMRC rsquo11) pp 1647ndash1651 Toronto Canada September2011

[3] 3GPP ldquoSelf-configuring and self-optimizing network (SON)usecases and solutionsrdquo TR 36902 V931 2011

[4] A Awada BWegmann I Viering and A Klein ldquoA SON-basedalgorithm for the optimization of inter-RAT handover param-etersrdquo IEEE Transactions on Vehicular Technology vol 62 no 5pp 1906ndash1923 2013

[5] T Jansen I Balan J Turk I Moerman and T Kurner ldquoHando-ver parameter optimization in LTE self-organizing networksrdquoin Proceedings of the IEEE 72nd Vehicular Technology Confer-ence Fall (VTC 2010-Fall rsquo10) pp 1ndash5 IEEE Ottawa CanadaSeptember 2010

[6] T Jansen I Balan S Stefanski I Moerman and T KurnerldquoWeighted performance based handover parameter optimiza-tion in LTErdquo in Proceedings of the IEEE 73rd Vehicular Technol-ogy Conference (VTC rsquo11) pp 1ndash5 May 2011

[7] W Li X Duan S Jia L Zhang Y Liu and J Lin ldquoA dynamichysteresis-adjusting algorithm in LTE self-organization net-worksrdquo in Proceedings of the IEEE 75th Vehicular TechnologyConference (VTC Spring rsquo12) pp 1ndash5 Yokohama Japan May2012

[8] A Lobinger S Stefanski T Jansen and I Balan ldquoCoordinatinghandover parameter optimization and load balancing in LTEself-optimizing networksrdquo in Proceedings of the IEEE 73rdVehicular Technology Conference (VTC rsquo11) pp 1ndash5 May 2011

[9] V Capdevielle A Feki and E Sorsy ldquoJoint interferencemanagement and handover optimization in LTE small cellsnetworkrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6769ndash6773 IEEE OttawaCanada June 2012

[10] T Ali and M Saquib ldquoWLANcellular handover analysis fordifferent mobility modelsrdquo in Proceedings of the IEEE Interna-tional Conference on Communications (ICC rsquo12) pp 2758ndash2762Ottawa Canada June 2012

[11] A Awada B Wegmann I Viering and A Klein ldquoCell-pairspecific optimization of the inter-RAT handover parameters inSONrdquo in Proceedings of the IEEE 23rd International Symposiumon Personal Indoor andMobile Radio Communications (PIMRCrsquo12) pp 1168ndash1173 Sydney Australia September 2012

[12] D Giacomini and A Agarwal ldquoVertical handover decisionmaking using QoS reputation and GM(11) predictionrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo12) pp 5655ndash5659 IEEE Ottawa Canada June2012

[13] D Lopez-Perez I Guvenc and X Chu ldquoTheoretical analysisof handover failure and ping-pong rates for heterogeneousnetworksrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6774ndash6779 IEEE OttawaCanada June 2012

[14] A Rath and S Panwar ldquoFast handover in cellular networkswith femtocellsrdquo in Proceedings of the International Conferenceon Communications (ICC rsquo12) pp 2752ndash2757 IEEE OttawaCanada June 2012

[15] D Xenakis N Passas and C Verikoukis ldquoA novel handoverdecision policy for reducing power transmissions in the two-tier LTE networkrdquo in Proceedings of the IEEE International Con-ference on Communications (ICC rsquo12) pp 1352ndash1356 OttawaCanada June 2012

[16] T Taleb K Samdanis and A Ksentini ldquoSupporting highlymobile users in cost-effective decentralized mobile operator

Mobile Information Systems 11

networksrdquo IEEE Transactions on Vehicular Technology vol 63no 7 pp 3381ndash3396 2014

[17] A Imran A Zoha and A Abu-Dayya ldquoChallenges in 5G howto empower SONwith big data for enabling 5Grdquo IEEE Networkvol 28 no 6 pp 27ndash33 2014

[18] H-D Bae B Ryu and N-H Park ldquoAnalysis of handoverfailures in LTE femtocell systemsrdquo in Proceedings of the Aus-tralasian Telecommunication Networks and Applications Con-ference (ATNAC rsquo11) pp 1ndash5 IEEE Melbourne AustraliaNovember 2011

[19] S Haykin Neural Networks A Comprehensive FoundationPrentice-Hall 1999

[20] L Behera S Kumar and A Patnaik ldquoOn adaptive learning ratethat guarantees convergence in feedforward networksrdquo IEEETransactions on Neural Networks vol 17 no 5 pp 1116ndash11252006

[21] J A Snyman Practical Mathematical Optimization An Intro-duction to Basic Optimization Theory and Classical and NewGradient-Based Algorithms vol 97 of Applied OptimizationSpringer New York NY USA 2005

[22] M Gudmundson ldquoCell planning in Manhattan environmentsrdquoin Proceedings of the 42nd Vehicular Technology Conference pp435ndash438 IEEE Denver Colo USA May 1992

[23] P-E Ostling ldquoImplications of cell planning on handoff per-formance in manhattan environmentsrdquo in Proceedings of the5th IEEE International Symposium on Personal Indoor andMobile Radio Communications Wireless NetworksmdashCatchingthe Mobile Future vol 2 pp 625ndash629 The Hague The Nether-lands September 1994

[24] M Karabacak D Wang H Ishii and H Arslan ldquoMobilityperformance of macrocell-assisted small cells in Manhattanmodelrdquo in Proceedings of the 79th IEEE Vehicular TechnologyConference (VTC-Spring rsquo14) pp 1ndash5 Seoul South Korea May2014

[25] 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

05 1 15 2 25 3 35 4 45 50X-coordinate (km)

0

05

1

15

2

25

3

35

4

45

5

Y-c

oord

inat

e (km

)

Figure 5 Topology of the simulated mobile communication net-work

result from (23) the handover parameter119883119894is adjusted as (15)

and (16) showIn order to cope with the time-varying channel condi-

tions we use online training and operate the three passesiteratively to model the instantaneous relationship 119884 = 119891(X)and optimize the handover parameters HOM and TTT

5 Performance Evaluation and Discussion

In this section we describe the simulation environmentsimulation results and discussions

In our simulation a mobile communication networkwhich consists of 100 small cells is built in a 5 km times 5 kmarea Figure 5 shows the coordinates of these small cell basestations Each cell has 5MHz bandwidth with 25 resourceblocks and 2GHz carrier frequency Each resource blockconsists of 12 subcarriers of size 15 kHz each A time slot is05ms in duration and the transmission time interval (TTI)is 1ms

In the simulation 100 MSs are uniformly distributedover the area with random initialized positions The mobilitytrace of these 100 MSs is generated by using the Manhattanmobility model which is widely used for urban areas andsuitable to implement realistic simulations [22ndash24] In theManhattanmobility model MSs move in either horizontal orvertical directions on an urban map At each intersection anMS employs a probabilistic approach to choose the directionof its movements The probabilities of going straight turningleft and turning right are 05 025 and 025 respectivelyThe blue line in Figure 5 shows an illustrative moving pathof an MS in the mobile communication network Whileconsidering the KPI the weights of mobility problems are setto be equal

An RLF will occur if the SINR level of the servingcell drops below minus65 dB before handover execution is

completed [18] We set this as our SINR threshold valuesuch that if the SINR from the source BS is lower than thethreshold an RLF will occur The simulation time is set as5000 seconds The operating point with 0 dB HOM and 0 sTTT is the starting operating point for all cells in the network

In this paper we use LTE-Sim an open source frameworkto simulate LTE networks as our basic simulation tool[25] The design and development of LTE-Sim support thefollowing aspects

(i) The Evolved Universal Terrestrial Radio Access (E-UTRAN) and the Evolved Packet System (EPS)

(ii) Single and heterogeneous multicell environments(iii) Multiusers environment(iv) User mobility(v) Handover procedures(vi) Frequency reuse techniques(vii) Quality-of-service (QoS) management(viii) Scheduling strategies

However LTE-Sim still lacks some important functional-ities and features that are required in this workTherefore wemodified LTE-Sim from several aspects Our modification toLTE-Sim includes the following

(1) Add

(i) radio link failure(ii) handover failure(iii) HOM and TTT(iv) proposed DHO approach

(2) Modify

(i) mobility pattern using the seed to generate thesame sequence of random numbers each timewe run the simulation

The simulation parameter values are shown in Table 2unless otherwise specified

Figure 6 shows the simulation result when the TX poweris set as 20 dBm The horizontal axis denotes the mobilityproblem index Indexes 1 to 5 mean the mobility problem oftoo-late HO too-early HO HO to wrong cell ping-pongHOand unnecessary HO respectively Index 6 means the KPIwhich is the weighted average of the five MPRs The verticalaxismeans the corresponding values ofMPR orKPIThe bluebars show the performance for a static HO parameter settingwith 0 dB HOM and 0 s TTT and the green ones for anotherstatic setting with 10 dBHOMand 512 s TTTThe yellow barsshow the performance of the proposed DHO approach Theresults shown in Figure 6 validate that the proposed DHOapproach can effectively mitigate the mobility problems andachieve better MPRs and KPI

Figure 7 shows the simulation result when the TX poweris set as 43 dBm Since the TX power is higher than the onein the previous scenario in Figure 6 it is less likely that an

Mobile Information Systems 9

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

Figure 6 Simulation result when the TX power is set as 20 dBm

Table 2 Simulation parameters

Parameter ValuesNumber of small cell base stations 100Number of mobile stations 100eNB Tx power 15 dBm20 dBm43 dBmCarrier frequency 2GHzMS speed 30 kmhMobility model ManhattanTTI 1msSubcarrier spacing 15 KHzResource block 180KHzSuper frame time 10msNoise figure 25Noise spectral density minus174 dBmSINR threshold minus65 dBSimulation time 5000 sHandover delay 30msSystem bandwidth 5MHz 25 RBsTTI

RLF occurs Therefore the MPRs and KPI in Figure 7 arelower than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

Figure 8 shows the simulation result when the TX poweris set as 15 dBm Since the TX power is lower than the onein the previous scenario in Figure 6 it is more likely that anRLF occurs Therefore the MPRs and KPI in Figure 8 arehigher than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

We compare the KPI improvement under the variousvalues of TX power The KPI improvement is calculatedas the difference between the initial-case KPI (HOM 0dB

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 7 Simulation result when the TX power is set as 43 dBm

DHO approach

2 3 4 5 61Mobility problem index

0

001

002

003

004

005

006

007

008

MPR

KPI

HOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 8 Simulation result when the TX power is set as 15 dBm

TTT 0 s) and the KPI of the proposed DHO methoddivided by the initial-case KPI Figure 9 shows that the KPIimprovement ranges from 15 to 20 It also shows that theKPI improvement is more significant when the TX power islower that is the transmission range of the base station issmaller

6 Conclusion

In this work we investigate the mobility problems innext generation mobile communication networks There aretotally five different mobility problems each with different

10 Mobile Information Systems

20 4315TX power (dBm)

0

5

10

15

20

25

KPI i

mpr

ovem

ent (

)

Figure 9 KPI improvement with various TX power

characteristics We propose a data-driven handover opti-mization (DHO) approach which aims to mitigate mobilityproblems The DHO approach consists of four parts

(1) Mobility problem identification(2) KPI design(3) KPI function estimation(4) Handover parameter optimization

The DHO approach collects data from the mobile com-munication measurement results and provides a model toestimate the relationship between the KPI and features fromthe collected dataset Based on the model the handoverparameters including the HOM and TTT are optimized tominimize the KPI which is a weighted average of differentmobility problem ratios Simulation results show that theproposed DHO approach could effectively mitigate mobilityproblems and provide better connectivity

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work was supported in part by Ministry of Scienceand Technology (MOST) Taipei Taiwan under Grants nosMOST 103-2221-E-155-005 MOST 104-2221-E-155-017 andMOST 105-2218-E-155-001

References

[1] J G Andrews S Buzzi W Choi et al ldquoWhat will 5G berdquo IEEEJournal on Selected Areas in Communications vol 32 no 6 pp1065ndash1082 2014

[2] K Kitagawa T Komine T Yamamoto and S Konishi ldquoAhandover optimization algorithm with mobility robustness forLTE systemsrdquo in Proceedings of the IEEE 22nd International

Symposium on Personal Indoor and Mobile Radio Communica-tions (PIMRC rsquo11) pp 1647ndash1651 Toronto Canada September2011

[3] 3GPP ldquoSelf-configuring and self-optimizing network (SON)usecases and solutionsrdquo TR 36902 V931 2011

[4] A Awada BWegmann I Viering and A Klein ldquoA SON-basedalgorithm for the optimization of inter-RAT handover param-etersrdquo IEEE Transactions on Vehicular Technology vol 62 no 5pp 1906ndash1923 2013

[5] T Jansen I Balan J Turk I Moerman and T Kurner ldquoHando-ver parameter optimization in LTE self-organizing networksrdquoin Proceedings of the IEEE 72nd Vehicular Technology Confer-ence Fall (VTC 2010-Fall rsquo10) pp 1ndash5 IEEE Ottawa CanadaSeptember 2010

[6] T Jansen I Balan S Stefanski I Moerman and T KurnerldquoWeighted performance based handover parameter optimiza-tion in LTErdquo in Proceedings of the IEEE 73rd Vehicular Technol-ogy Conference (VTC rsquo11) pp 1ndash5 May 2011

[7] W Li X Duan S Jia L Zhang Y Liu and J Lin ldquoA dynamichysteresis-adjusting algorithm in LTE self-organization net-worksrdquo in Proceedings of the IEEE 75th Vehicular TechnologyConference (VTC Spring rsquo12) pp 1ndash5 Yokohama Japan May2012

[8] A Lobinger S Stefanski T Jansen and I Balan ldquoCoordinatinghandover parameter optimization and load balancing in LTEself-optimizing networksrdquo in Proceedings of the IEEE 73rdVehicular Technology Conference (VTC rsquo11) pp 1ndash5 May 2011

[9] V Capdevielle A Feki and E Sorsy ldquoJoint interferencemanagement and handover optimization in LTE small cellsnetworkrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6769ndash6773 IEEE OttawaCanada June 2012

[10] T Ali and M Saquib ldquoWLANcellular handover analysis fordifferent mobility modelsrdquo in Proceedings of the IEEE Interna-tional Conference on Communications (ICC rsquo12) pp 2758ndash2762Ottawa Canada June 2012

[11] A Awada B Wegmann I Viering and A Klein ldquoCell-pairspecific optimization of the inter-RAT handover parameters inSONrdquo in Proceedings of the IEEE 23rd International Symposiumon Personal Indoor andMobile Radio Communications (PIMRCrsquo12) pp 1168ndash1173 Sydney Australia September 2012

[12] D Giacomini and A Agarwal ldquoVertical handover decisionmaking using QoS reputation and GM(11) predictionrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo12) pp 5655ndash5659 IEEE Ottawa Canada June2012

[13] D Lopez-Perez I Guvenc and X Chu ldquoTheoretical analysisof handover failure and ping-pong rates for heterogeneousnetworksrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6774ndash6779 IEEE OttawaCanada June 2012

[14] A Rath and S Panwar ldquoFast handover in cellular networkswith femtocellsrdquo in Proceedings of the International Conferenceon Communications (ICC rsquo12) pp 2752ndash2757 IEEE OttawaCanada June 2012

[15] D Xenakis N Passas and C Verikoukis ldquoA novel handoverdecision policy for reducing power transmissions in the two-tier LTE networkrdquo in Proceedings of the IEEE International Con-ference on Communications (ICC rsquo12) pp 1352ndash1356 OttawaCanada June 2012

[16] T Taleb K Samdanis and A Ksentini ldquoSupporting highlymobile users in cost-effective decentralized mobile operator

Mobile Information Systems 11

networksrdquo IEEE Transactions on Vehicular Technology vol 63no 7 pp 3381ndash3396 2014

[17] A Imran A Zoha and A Abu-Dayya ldquoChallenges in 5G howto empower SONwith big data for enabling 5Grdquo IEEE Networkvol 28 no 6 pp 27ndash33 2014

[18] H-D Bae B Ryu and N-H Park ldquoAnalysis of handoverfailures in LTE femtocell systemsrdquo in Proceedings of the Aus-tralasian Telecommunication Networks and Applications Con-ference (ATNAC rsquo11) pp 1ndash5 IEEE Melbourne AustraliaNovember 2011

[19] S Haykin Neural Networks A Comprehensive FoundationPrentice-Hall 1999

[20] L Behera S Kumar and A Patnaik ldquoOn adaptive learning ratethat guarantees convergence in feedforward networksrdquo IEEETransactions on Neural Networks vol 17 no 5 pp 1116ndash11252006

[21] J A Snyman Practical Mathematical Optimization An Intro-duction to Basic Optimization Theory and Classical and NewGradient-Based Algorithms vol 97 of Applied OptimizationSpringer New York NY USA 2005

[22] M Gudmundson ldquoCell planning in Manhattan environmentsrdquoin Proceedings of the 42nd Vehicular Technology Conference pp435ndash438 IEEE Denver Colo USA May 1992

[23] P-E Ostling ldquoImplications of cell planning on handoff per-formance in manhattan environmentsrdquo in Proceedings of the5th IEEE International Symposium on Personal Indoor andMobile Radio Communications Wireless NetworksmdashCatchingthe Mobile Future vol 2 pp 625ndash629 The Hague The Nether-lands September 1994

[24] M Karabacak D Wang H Ishii and H Arslan ldquoMobilityperformance of macrocell-assisted small cells in Manhattanmodelrdquo in Proceedings of the 79th IEEE Vehicular TechnologyConference (VTC-Spring rsquo14) pp 1ndash5 Seoul South Korea May2014

[25] 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

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

Figure 6 Simulation result when the TX power is set as 20 dBm

Table 2 Simulation parameters

Parameter ValuesNumber of small cell base stations 100Number of mobile stations 100eNB Tx power 15 dBm20 dBm43 dBmCarrier frequency 2GHzMS speed 30 kmhMobility model ManhattanTTI 1msSubcarrier spacing 15 KHzResource block 180KHzSuper frame time 10msNoise figure 25Noise spectral density minus174 dBmSINR threshold minus65 dBSimulation time 5000 sHandover delay 30msSystem bandwidth 5MHz 25 RBsTTI

RLF occurs Therefore the MPRs and KPI in Figure 7 arelower than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

Figure 8 shows the simulation result when the TX poweris set as 15 dBm Since the TX power is lower than the onein the previous scenario in Figure 6 it is more likely that anRLF occurs Therefore the MPRs and KPI in Figure 8 arehigher than those in Figure 6 Moreover the proposed DHOapproach can still effectively mitigate the mobility problemsand provide better connectivity to MSs

We compare the KPI improvement under the variousvalues of TX power The KPI improvement is calculatedas the difference between the initial-case KPI (HOM 0dB

0

001

002

003

004

005

006

007

008

MPR

KPI

1 3 4 5 62Mobility problem index

DHO approachHOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 7 Simulation result when the TX power is set as 43 dBm

DHO approach

2 3 4 5 61Mobility problem index

0

001

002

003

004

005

006

007

008

MPR

KPI

HOM 10dB TTT 512 sHOM 0dB TTT 0 s

Figure 8 Simulation result when the TX power is set as 15 dBm

TTT 0 s) and the KPI of the proposed DHO methoddivided by the initial-case KPI Figure 9 shows that the KPIimprovement ranges from 15 to 20 It also shows that theKPI improvement is more significant when the TX power islower that is the transmission range of the base station issmaller

6 Conclusion

In this work we investigate the mobility problems innext generation mobile communication networks There aretotally five different mobility problems each with different

10 Mobile Information Systems

20 4315TX power (dBm)

0

5

10

15

20

25

KPI i

mpr

ovem

ent (

)

Figure 9 KPI improvement with various TX power

characteristics We propose a data-driven handover opti-mization (DHO) approach which aims to mitigate mobilityproblems The DHO approach consists of four parts

(1) Mobility problem identification(2) KPI design(3) KPI function estimation(4) Handover parameter optimization

The DHO approach collects data from the mobile com-munication measurement results and provides a model toestimate the relationship between the KPI and features fromthe collected dataset Based on the model the handoverparameters including the HOM and TTT are optimized tominimize the KPI which is a weighted average of differentmobility problem ratios Simulation results show that theproposed DHO approach could effectively mitigate mobilityproblems and provide better connectivity

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work was supported in part by Ministry of Scienceand Technology (MOST) Taipei Taiwan under Grants nosMOST 103-2221-E-155-005 MOST 104-2221-E-155-017 andMOST 105-2218-E-155-001

References

[1] J G Andrews S Buzzi W Choi et al ldquoWhat will 5G berdquo IEEEJournal on Selected Areas in Communications vol 32 no 6 pp1065ndash1082 2014

[2] K Kitagawa T Komine T Yamamoto and S Konishi ldquoAhandover optimization algorithm with mobility robustness forLTE systemsrdquo in Proceedings of the IEEE 22nd International

Symposium on Personal Indoor and Mobile Radio Communica-tions (PIMRC rsquo11) pp 1647ndash1651 Toronto Canada September2011

[3] 3GPP ldquoSelf-configuring and self-optimizing network (SON)usecases and solutionsrdquo TR 36902 V931 2011

[4] A Awada BWegmann I Viering and A Klein ldquoA SON-basedalgorithm for the optimization of inter-RAT handover param-etersrdquo IEEE Transactions on Vehicular Technology vol 62 no 5pp 1906ndash1923 2013

[5] T Jansen I Balan J Turk I Moerman and T Kurner ldquoHando-ver parameter optimization in LTE self-organizing networksrdquoin Proceedings of the IEEE 72nd Vehicular Technology Confer-ence Fall (VTC 2010-Fall rsquo10) pp 1ndash5 IEEE Ottawa CanadaSeptember 2010

[6] T Jansen I Balan S Stefanski I Moerman and T KurnerldquoWeighted performance based handover parameter optimiza-tion in LTErdquo in Proceedings of the IEEE 73rd Vehicular Technol-ogy Conference (VTC rsquo11) pp 1ndash5 May 2011

[7] W Li X Duan S Jia L Zhang Y Liu and J Lin ldquoA dynamichysteresis-adjusting algorithm in LTE self-organization net-worksrdquo in Proceedings of the IEEE 75th Vehicular TechnologyConference (VTC Spring rsquo12) pp 1ndash5 Yokohama Japan May2012

[8] A Lobinger S Stefanski T Jansen and I Balan ldquoCoordinatinghandover parameter optimization and load balancing in LTEself-optimizing networksrdquo in Proceedings of the IEEE 73rdVehicular Technology Conference (VTC rsquo11) pp 1ndash5 May 2011

[9] V Capdevielle A Feki and E Sorsy ldquoJoint interferencemanagement and handover optimization in LTE small cellsnetworkrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6769ndash6773 IEEE OttawaCanada June 2012

[10] T Ali and M Saquib ldquoWLANcellular handover analysis fordifferent mobility modelsrdquo in Proceedings of the IEEE Interna-tional Conference on Communications (ICC rsquo12) pp 2758ndash2762Ottawa Canada June 2012

[11] A Awada B Wegmann I Viering and A Klein ldquoCell-pairspecific optimization of the inter-RAT handover parameters inSONrdquo in Proceedings of the IEEE 23rd International Symposiumon Personal Indoor andMobile Radio Communications (PIMRCrsquo12) pp 1168ndash1173 Sydney Australia September 2012

[12] D Giacomini and A Agarwal ldquoVertical handover decisionmaking using QoS reputation and GM(11) predictionrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo12) pp 5655ndash5659 IEEE Ottawa Canada June2012

[13] D Lopez-Perez I Guvenc and X Chu ldquoTheoretical analysisof handover failure and ping-pong rates for heterogeneousnetworksrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6774ndash6779 IEEE OttawaCanada June 2012

[14] A Rath and S Panwar ldquoFast handover in cellular networkswith femtocellsrdquo in Proceedings of the International Conferenceon Communications (ICC rsquo12) pp 2752ndash2757 IEEE OttawaCanada June 2012

[15] D Xenakis N Passas and C Verikoukis ldquoA novel handoverdecision policy for reducing power transmissions in the two-tier LTE networkrdquo in Proceedings of the IEEE International Con-ference on Communications (ICC rsquo12) pp 1352ndash1356 OttawaCanada June 2012

[16] T Taleb K Samdanis and A Ksentini ldquoSupporting highlymobile users in cost-effective decentralized mobile operator

Mobile Information Systems 11

networksrdquo IEEE Transactions on Vehicular Technology vol 63no 7 pp 3381ndash3396 2014

[17] A Imran A Zoha and A Abu-Dayya ldquoChallenges in 5G howto empower SONwith big data for enabling 5Grdquo IEEE Networkvol 28 no 6 pp 27ndash33 2014

[18] H-D Bae B Ryu and N-H Park ldquoAnalysis of handoverfailures in LTE femtocell systemsrdquo in Proceedings of the Aus-tralasian Telecommunication Networks and Applications Con-ference (ATNAC rsquo11) pp 1ndash5 IEEE Melbourne AustraliaNovember 2011

[19] S Haykin Neural Networks A Comprehensive FoundationPrentice-Hall 1999

[20] L Behera S Kumar and A Patnaik ldquoOn adaptive learning ratethat guarantees convergence in feedforward networksrdquo IEEETransactions on Neural Networks vol 17 no 5 pp 1116ndash11252006

[21] J A Snyman Practical Mathematical Optimization An Intro-duction to Basic Optimization Theory and Classical and NewGradient-Based Algorithms vol 97 of Applied OptimizationSpringer New York NY USA 2005

[22] M Gudmundson ldquoCell planning in Manhattan environmentsrdquoin Proceedings of the 42nd Vehicular Technology Conference pp435ndash438 IEEE Denver Colo USA May 1992

[23] P-E Ostling ldquoImplications of cell planning on handoff per-formance in manhattan environmentsrdquo in Proceedings of the5th IEEE International Symposium on Personal Indoor andMobile Radio Communications Wireless NetworksmdashCatchingthe Mobile Future vol 2 pp 625ndash629 The Hague The Nether-lands September 1994

[24] M Karabacak D Wang H Ishii and H Arslan ldquoMobilityperformance of macrocell-assisted small cells in Manhattanmodelrdquo in Proceedings of the 79th IEEE Vehicular TechnologyConference (VTC-Spring rsquo14) pp 1ndash5 Seoul South Korea May2014

[25] 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

20 4315TX power (dBm)

0

5

10

15

20

25

KPI i

mpr

ovem

ent (

)

Figure 9 KPI improvement with various TX power

characteristics We propose a data-driven handover opti-mization (DHO) approach which aims to mitigate mobilityproblems The DHO approach consists of four parts

(1) Mobility problem identification(2) KPI design(3) KPI function estimation(4) Handover parameter optimization

The DHO approach collects data from the mobile com-munication measurement results and provides a model toestimate the relationship between the KPI and features fromthe collected dataset Based on the model the handoverparameters including the HOM and TTT are optimized tominimize the KPI which is a weighted average of differentmobility problem ratios Simulation results show that theproposed DHO approach could effectively mitigate mobilityproblems and provide better connectivity

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work was supported in part by Ministry of Scienceand Technology (MOST) Taipei Taiwan under Grants nosMOST 103-2221-E-155-005 MOST 104-2221-E-155-017 andMOST 105-2218-E-155-001

References

[1] J G Andrews S Buzzi W Choi et al ldquoWhat will 5G berdquo IEEEJournal on Selected Areas in Communications vol 32 no 6 pp1065ndash1082 2014

[2] K Kitagawa T Komine T Yamamoto and S Konishi ldquoAhandover optimization algorithm with mobility robustness forLTE systemsrdquo in Proceedings of the IEEE 22nd International

Symposium on Personal Indoor and Mobile Radio Communica-tions (PIMRC rsquo11) pp 1647ndash1651 Toronto Canada September2011

[3] 3GPP ldquoSelf-configuring and self-optimizing network (SON)usecases and solutionsrdquo TR 36902 V931 2011

[4] A Awada BWegmann I Viering and A Klein ldquoA SON-basedalgorithm for the optimization of inter-RAT handover param-etersrdquo IEEE Transactions on Vehicular Technology vol 62 no 5pp 1906ndash1923 2013

[5] T Jansen I Balan J Turk I Moerman and T Kurner ldquoHando-ver parameter optimization in LTE self-organizing networksrdquoin Proceedings of the IEEE 72nd Vehicular Technology Confer-ence Fall (VTC 2010-Fall rsquo10) pp 1ndash5 IEEE Ottawa CanadaSeptember 2010

[6] T Jansen I Balan S Stefanski I Moerman and T KurnerldquoWeighted performance based handover parameter optimiza-tion in LTErdquo in Proceedings of the IEEE 73rd Vehicular Technol-ogy Conference (VTC rsquo11) pp 1ndash5 May 2011

[7] W Li X Duan S Jia L Zhang Y Liu and J Lin ldquoA dynamichysteresis-adjusting algorithm in LTE self-organization net-worksrdquo in Proceedings of the IEEE 75th Vehicular TechnologyConference (VTC Spring rsquo12) pp 1ndash5 Yokohama Japan May2012

[8] A Lobinger S Stefanski T Jansen and I Balan ldquoCoordinatinghandover parameter optimization and load balancing in LTEself-optimizing networksrdquo in Proceedings of the IEEE 73rdVehicular Technology Conference (VTC rsquo11) pp 1ndash5 May 2011

[9] V Capdevielle A Feki and E Sorsy ldquoJoint interferencemanagement and handover optimization in LTE small cellsnetworkrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6769ndash6773 IEEE OttawaCanada June 2012

[10] T Ali and M Saquib ldquoWLANcellular handover analysis fordifferent mobility modelsrdquo in Proceedings of the IEEE Interna-tional Conference on Communications (ICC rsquo12) pp 2758ndash2762Ottawa Canada June 2012

[11] A Awada B Wegmann I Viering and A Klein ldquoCell-pairspecific optimization of the inter-RAT handover parameters inSONrdquo in Proceedings of the IEEE 23rd International Symposiumon Personal Indoor andMobile Radio Communications (PIMRCrsquo12) pp 1168ndash1173 Sydney Australia September 2012

[12] D Giacomini and A Agarwal ldquoVertical handover decisionmaking using QoS reputation and GM(11) predictionrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo12) pp 5655ndash5659 IEEE Ottawa Canada June2012

[13] D Lopez-Perez I Guvenc and X Chu ldquoTheoretical analysisof handover failure and ping-pong rates for heterogeneousnetworksrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo12) pp 6774ndash6779 IEEE OttawaCanada June 2012

[14] A Rath and S Panwar ldquoFast handover in cellular networkswith femtocellsrdquo in Proceedings of the International Conferenceon Communications (ICC rsquo12) pp 2752ndash2757 IEEE OttawaCanada June 2012

[15] D Xenakis N Passas and C Verikoukis ldquoA novel handoverdecision policy for reducing power transmissions in the two-tier LTE networkrdquo in Proceedings of the IEEE International Con-ference on Communications (ICC rsquo12) pp 1352ndash1356 OttawaCanada June 2012

[16] T Taleb K Samdanis and A Ksentini ldquoSupporting highlymobile users in cost-effective decentralized mobile operator

Mobile Information Systems 11

networksrdquo IEEE Transactions on Vehicular Technology vol 63no 7 pp 3381ndash3396 2014

[17] A Imran A Zoha and A Abu-Dayya ldquoChallenges in 5G howto empower SONwith big data for enabling 5Grdquo IEEE Networkvol 28 no 6 pp 27ndash33 2014

[18] H-D Bae B Ryu and N-H Park ldquoAnalysis of handoverfailures in LTE femtocell systemsrdquo in Proceedings of the Aus-tralasian Telecommunication Networks and Applications Con-ference (ATNAC rsquo11) pp 1ndash5 IEEE Melbourne AustraliaNovember 2011

[19] S Haykin Neural Networks A Comprehensive FoundationPrentice-Hall 1999

[20] L Behera S Kumar and A Patnaik ldquoOn adaptive learning ratethat guarantees convergence in feedforward networksrdquo IEEETransactions on Neural Networks vol 17 no 5 pp 1116ndash11252006

[21] J A Snyman Practical Mathematical Optimization An Intro-duction to Basic Optimization Theory and Classical and NewGradient-Based Algorithms vol 97 of Applied OptimizationSpringer New York NY USA 2005

[22] M Gudmundson ldquoCell planning in Manhattan environmentsrdquoin Proceedings of the 42nd Vehicular Technology Conference pp435ndash438 IEEE Denver Colo USA May 1992

[23] P-E Ostling ldquoImplications of cell planning on handoff per-formance in manhattan environmentsrdquo in Proceedings of the5th IEEE International Symposium on Personal Indoor andMobile Radio Communications Wireless NetworksmdashCatchingthe Mobile Future vol 2 pp 625ndash629 The Hague The Nether-lands September 1994

[24] M Karabacak D Wang H Ishii and H Arslan ldquoMobilityperformance of macrocell-assisted small cells in Manhattanmodelrdquo in Proceedings of the 79th IEEE Vehicular TechnologyConference (VTC-Spring rsquo14) pp 1ndash5 Seoul South Korea May2014

[25] 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

networksrdquo IEEE Transactions on Vehicular Technology vol 63no 7 pp 3381ndash3396 2014

[17] A Imran A Zoha and A Abu-Dayya ldquoChallenges in 5G howto empower SONwith big data for enabling 5Grdquo IEEE Networkvol 28 no 6 pp 27ndash33 2014

[18] H-D Bae B Ryu and N-H Park ldquoAnalysis of handoverfailures in LTE femtocell systemsrdquo in Proceedings of the Aus-tralasian Telecommunication Networks and Applications Con-ference (ATNAC rsquo11) pp 1ndash5 IEEE Melbourne AustraliaNovember 2011

[19] S Haykin Neural Networks A Comprehensive FoundationPrentice-Hall 1999

[20] L Behera S Kumar and A Patnaik ldquoOn adaptive learning ratethat guarantees convergence in feedforward networksrdquo IEEETransactions on Neural Networks vol 17 no 5 pp 1116ndash11252006

[21] J A Snyman Practical Mathematical Optimization An Intro-duction to Basic Optimization Theory and Classical and NewGradient-Based Algorithms vol 97 of Applied OptimizationSpringer New York NY USA 2005

[22] M Gudmundson ldquoCell planning in Manhattan environmentsrdquoin Proceedings of the 42nd Vehicular Technology Conference pp435ndash438 IEEE Denver Colo USA May 1992

[23] P-E Ostling ldquoImplications of cell planning on handoff per-formance in manhattan environmentsrdquo in Proceedings of the5th IEEE International Symposium on Personal Indoor andMobile Radio Communications Wireless NetworksmdashCatchingthe Mobile Future vol 2 pp 625ndash629 The Hague The Nether-lands September 1994

[24] M Karabacak D Wang H Ishii and H Arslan ldquoMobilityperformance of macrocell-assisted small cells in Manhattanmodelrdquo in Proceedings of the 79th IEEE Vehicular TechnologyConference (VTC-Spring rsquo14) pp 1ndash5 Seoul South Korea May2014

[25] 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