Dynamic SON-Enabled Location Management in LTE Networks · 2019-05-09 · Moreover, LTE suffers from greater signaling overhead than 3G technologies because its ﬂat IP architecture

1536-1233 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2017.2775221, IEEETransactions on Mobile Computing

1

Dynamic SON-Enabled Location Management inLTE Networks

Emad Aqeeli, Abdallah Moubayed, and Abdallah ShamiWestern University, London, Ontario, Canada

e-mails: {eaqeeli, amoubaye, ashami2}@uwo.ca

Abstract—Wireless networks are facing various challengesthat demand continuous and rapid improvement. Long-TermEvolution (LTE) is a preferred wireless technology because ofits satisfactory performance. Owing to an exponential increasein demand and new potential applications, the core networkof LTE, which is known as the Evolved Packet Core (EPC),is affected by a surge in signaling caused by a variety ofcontrol functions. The signaling overhead decreases the users’Quality of Experience (QoE). The current study attempts toimprove the intelligence of location management techniques.As an extension of our previous study [1], a Self-OrganizingNetwork (SON) that enables dynamic reconfiguration of cell-to-TAL/MME is introduced. Both centralized and distributedpooling schemes are tested in terms of signaling overhead anduser power consumption. A decomposition model that reducesthe original formulated problem to two sub-problems is proposed,each of which is solved optimally. In addition, a smart cell-to-TAL selection scheme is proposed to prioritize potential cells thatmight be visited by a user equipment (UE). Our method is shownto outperform several state-of-the-art methods presented in theliterature. Finally, a heuristic algorithm is presented to obtain aless complex solution than the optimal one.

Index Terms—LTE, Location Management, Tracking Area(TA), Tracking Area List (TAL), TAU, Paging, MME pooling,SON

I. INTRODUCTION

Mobile networks have been attracting considerable attentionfrom both industry and academia since the early 1980s, whenthe first generation (1G) of analog cellular networks wasintroduced as the Advanced Mobile Phone System (AMPS).Ten years later, the second generation (2G) was launched tosupport mobile phones and limited data connections over the2.5G extension. With the support of General Packet RadioService (GPRS), 2.5G networks could use circuit switching forvoice applications and packet switching for data transmission.The evolution continued with the introduction of the thirdgeneration, which provided faster speeds with better efficiencyand quality of service. Subsequently, the fourth generationintroduced two standards, namely WiMAX and Long-TermEvolution (LTE), which enhanced the capability of packetswitching to provide users with better performance. LTE hasgained more attention than WiMAX, as it supports higherspeeds, better performance, and scalable bandwidth.

Despite these advancements, wireless technologies continueto be challenged by severe traffic that affects their bandwidth.The proliferation of hand-held devices and their applicationshas caused signaling traffic to exceed user data traffic. Sig-naling traffic is initiated each time there is a transmission or

reception of packet streams between user equipments (UEs)and mobile networks, regardless of the actual size of the datatraffic. Nokia Siemens Networks have predicted that in thecoming years, the increase in signaling will be up to 50%faster than that in data traffic. Moreover, LTE suffers fromgreater signaling overhead than 3G technologies because itsflat IP architecture does not include a medium entity, suchas a Radio Network Controller (RNC), between a base stationand the core network [2]. Thus, the average signaling overheadof LTE is 42% greater than that of HSPA per subscriber.

Paging and Tracking Area Updates (TAU) have the most sig-nificant signaling impact on the Evolved Packet Core (EPC),specifically on the Mobility Management Entity (MME). Pag-ing and TAU are essential functions for UE location manage-ment. They are used to track the user’s location and provideconstant updates to the EPC. Paging and TAU are defined asfollows:

1) Paging: messages sent by the MME to locate a particularUE in a tracking area (TA).

2) Tracking area update: messages sent by UEs to the MMEwhen they move from one tracking area to another.

To identify the UE location, the LTE core network pagesthe latest tracking area that the UE was registered to. Thepaging signal is received by all of the cells that reside in thesame tracking area. Additionally, the UE will update the corenetwork by sending a tracking area update (TAU) signal onceit moves from one tracking area to another.

Tracking Area (TA) is the area in which the MME can locatea specific user within a defined set of cells. This technique isused in LTE and was originally inherited from previous 2G and3G technologies. However, TA has a number of limitations thathave led to the introduction of the new concept of TrackingArea List (TAL), whereby several TAs are grouped into asingle TAL. TAL has the same functionality as TA but withthe added flexibility of a set of TAs within the TAL. Thus,TAL can alleviate the signaling load due to triggering TAUeach time a UE moves from one cell to another.

Various mobility management techniques have been in-vestigated extensively from different perspectives, such asoverlapping, dynamic, and static cell-to-TAL assignment. Thecurrent study is an extension of our previous study, whichintroduced cell-to-TAL assignment with two MME poolingschemes. Centralized and distributed MME pooling schemeswere investigated in order to explore the difference betweenboth schemes statically [1]. In this context, we intend to



2

enable adaptive online cell-to-TAL assignment in order tofurther investigate the proposed pooling schemes. UEs areusually in continuous movement and their coordination is notstatic. Hence, the initial static assignment for cell-to-TAL willgradually become ineffective over a specific period. Therefore,there is a need to revise the TA assignment constantly in orderto suit the current mobility state. Unlike conventional TA, theTAL concept allows TA assignment to be modified withoutinterruption of service. This is an advantage of TAL overconventional TA, because TAL provides greater flexibility tothe system. Moreover, LTE allows for an auto-reconfigurationfeature that adapts the network configuration whenever thereis a change in the UE statistics, such as movement patternsand loads.

In release 8, 3GPP introduced the concept of a Self-Organizing Network (SON) that provides a methodology forplanning, managing, and optimizing mobile networks in orderimprove performance efficiency and system reliability. SONhas been widely accepted in industry and academia [3,4].3GPP has also released different use cases for LTE, offeringself-optimizing and self-healing paradigms. In this context,adaptive TA list management can be used as an SON usecase as in [5], which can further reduce the signaling load.In addition, introducing different techniques that relate theTA list to the behavior of the mobile network would furtheroptimize the signaling overhead. Thus, cell-to-TAL assignmentcan be engineered dynamically while the UE is in continuousmovement. The system keeps analyzing the mobility patternand continuously updates the TA assigned to the list. Thus, thefrequency of TAU will be reduced significantly. In this study,the mobility pattern is obtained using a fluid flow model toestimate the handover correlation between cells. One way tomeasure the efficiency of a dynamic system is to examine theUE battery life. In our study, the UEs battery consumption instatic and dynamic techniques will be compared.

The contributions of this study can be summarized asfollows:

1) A dynamic cell-to-TAL problem is proposed and for-mulated as a mixed integer non-linear programming(MINLP) problem with quadratic equality constraints.This approach is different from the previous approachpresented in [1], where the problem was solved staticallyby finding the optimal assignment once. The proposeddynamic technique is realized through an SON schemealong with a new smart cell selection approach insteadof the conventional ring-based cell selection presented inthe literature.

2) The problem is solved using a decomposition modelthat divides the problem into two sub-problems. Thedecomposition model allows optimal assignment of cell-to-TAL dynamically instead of having it known a prioriin a static fashion as in [1].

3) A new heuristic algorithm differs from the one proposedin [1] and constitutes of two sub-problems in the samemanner as the decomposition model. The algorithm dy-namically diversifies the TALs among the cells whichhelps in reducing the TAU signaling load.

The remainder of this paper is organized as follows. Section IIreviews studies on mobility management techniques. SectionIII presents a detailed description of the system model andformulates the problem for both the centralized and the dis-tributed schemes. Section IV describes the heuristic algorithmdeveloped to solve the optimization problem. Section V dis-cusses the simulation parameters and results. Finally, SectionVI concludes the paper.

II. RELATED WORK

A considerable amount of research effort has been devotedto the study of location management in various technologies,specifically in terms of location management, which has notchanged much in recent years. In this context, we essentiallyhighlight two major sources of signaling that are directlyrelated to location management: TAU and paging. Numerousstudies have discussed the issue of signaling burst caused byTAU and paging. Most of these approaches have investigatedthe signaling overhead when configuring LA or TA having thesame properties. Few researchers have addressed the signalingoverhead from the perspective of TAL construction, eventhough TAL has greater importance and flexibility than LAor TA. In fact, most studies have failed to provide a rigidframework that can provide technical support and is applicableto real-world scenarios.

As an example of previous proposals for older technologies,an overlapping location area mechanism was proposed in [6]for GSM technology. The purpose of the study was to min-imize the signaling load resulting from the ping pong effect.The study introduced four selection policies for determiningthe location area (LA) percentage. Similar related studies havebeen conducted [7]–[9].

Furthermore, TAU and paging techniques have been inves-tigated for the purpose of reducing the signaling load. Severalstudies [10]–[12] have discussed a number of methods, suchas timer-, velocity-, and movement-based ones. However, theaforementioned methods are not commonly used in the currentTA/TAL approach.

Few studies have discussed the signaling load in terms ofsteering the load through the control-plane elements, suchas MME or service gateway (S-GW). In one study [13], aconcept that enables two modes of MME distribution, namelycentralized and distributed MME, has been introduced. Thestudy proposed analysis of both architectures in terms ofthe signaling load resulting from user mobility. The authorsalso presented a comparison between multicast and unicastpaging. While the study reported some findings using dis-tributed and centralized MME architectures, it did not discussthe fundamentals of TA/TAL construction that allow furtherexploration of both architectures. Furthermore, the authors didnot illustrate how the MMEs are allocated and implementedin their model. Another study [14] has proposed a model thatsupports high-mobility users by relocating data plane gatewayson the basis of the UEs mobility pattern, thereby minimizingthe relocation frequency.

In [15], the authors defined the concept of S-GW servicearea, where a pool of S-GWs serves TAs or cells. The objective



3

is to eliminate frequent disconnections that occur when theUE moves to a different cell or TA. The concept focusesspecifically on active users whose Quality of Service (QoS)can be degraded significantly during S-GW re-allocation.

A number of studies have proposed dynamic TA/TAL tech-niques that can optimize the signaling overhead periodically[16]–[18]. In [16], the authors explored the advantage ofdynamic TAL configuration by introducing a ”rule of thumb”.The approach attempts to minimize the signaling overhead onthe basis of the TAU and paging correlation. It succeeded inreducing the overhead to a greater extent than conventionalTA. In the proposed model, dynamic configuration is achievedafter a fixed period, which does not yield accurate results inenvironments having UEs moving at different speeds. In [17],the authors introduced an adaptive model that constructs asuitable TAL for each user in a set of cells represented as anindividual TA. The adaptive configuration is triggered periodi-cally on the basis of a defined movement threshold. This modeldoes not clearly indicate the feasibility of the solution whendealing with a large-scale scenario. Moreover, the authors didnot provide a detailed description of the physical parametersof the system, such as the cell radius and the users velocity. In[18], the authors presented a dynamic model that allows TAs tobe configured as user mobility patterns change. The algorithmtends to use the graph correlation coefficient method, whichmeasures the similarities in user mobility behavior within acertain period and triggers new TA construction. The approachis expensive owing to the high cell-to-TA reconfiguration cost.

Other studies have deployed TA/TAL configuration as anSON use case, e.g., [5]. The system implements proceduresand protocols based on the UE behavior pattern. Consequently,the TA/TAL is configured periodically to minimize the signal-ing burst. The study proposed adapting the TAL configurationdepending on the user mobility pattern using a set of femto-cell mesh networks. Each TAL can be configured dynamicallyon the basis of the UE’s mobility status. Despite the novelty ofthe algorithm, the model does not specify some major designaspects, such as the advantage of overlapping TALs in thecells, which can contribute significantly toward minimizingthe signaling load.

TAL has been explored and modeled in [3,19]–[22]. In [19],the authors proposed a model for TAL that can be appliedto a large-scale scenario, and they showed the advantages ofTAL over the conventional TA concept. Thus, they tried toprovide an abstraction of the TAL design and its benefits.In [20], the same authors proposed an optimization modelbased on overlapping TALs, which would allow a cell todistribute different TAL portions to a set of users residingin that cell in order to alleviate the TAU signaling load whilstrestricting the paging load within a defined limit. In [21], theauthors formulated a linear programming (LP) model basedon overlapping TALs in large-scale scenarios and compared itwith the conventional TA technique. Mobility management inLTE was also investigated in [22]. Paging and TAU were ana-lyzed in terms of signaling. Three sequential paging schemeswere presented, namely cell-TAL, TA-TAL, and cell-TA-TAL,whereby the MME first requests the cell to page the UE, andif the cell fails to allocate the UE, the MME sends another

request to all cells residing in the TA or TAL. This canalso be done in a different order. The authors concluded thatthe results vary with the number of users, paging schemes,and polling cycles. Thus, the selection of potential pagingschemes will depend on several factors, such as minimizing thesignaling cost of paging and TAU or reducing the number ofpolling cycles. Lastly, the performance of TAL was modeledand analyzed in [3]. The study investigated TAL modelingfrom several aspects, e.g., limited number of TAs in the TAL,and introduced two call-handling models that can provide amore realistic view of the system. The total signaling costwas calculated and the optimal TAL for UEs was determined.However, the authors did not verify the validity of the model inlarge-scale scenarios. Moreover, the model did not emphasizethe design of overlapping TALs or the distribution of TALswithin the system.

III. SYSTEM MODEL

A. Problem Definition

The objective of the problem is to minimize the signalingoverhead due to TAU and paging in the core network. Fur-thermore, our model seeks to achieve load balancing throughdifferent MMEs. The same concept presented in our previousstudy [1] will be applied along with an SON dynamic algo-rithm; the latter enables periodic Cell-to-TAL/MME configu-ration. The model consists of two pooling schemes, namelycentralized and distributed schemes. Table I summarizes theimportant notations used in the system model.

B. Preliminaries

In order to design a valid model for minimizing the sig-naling load, we first show that paging and TAU have a tightcorrelation that can be used for constructing a TAL. A TALconsists of a number of TAs, each of which can be representedas one cell. The size of each TAL has a direct influence interms of signaling. In other words, as the number of cellsaccommodated inside a TAL increases, the paging signalingload increases and the TAU signaling load decreases, and viceversa. This is a result of the TAU and paging mechanisms.A TAU signal is triggered to update the users location withinthe MME. This is done whenever a user moves to a cell thatis in a different TAL. A paging signal is triggered to locatethe user by messaging all the cells of the last TAL to whichthe user was registered. As the number of cells within a listincreases, fewer TAU are required because the probability thatthe user changes the list has decreased. However, paging wouldincrease because the MME would need to send messages toa larger group of cells in order to accurately determine thelocation of the user. Thus, the signaling overheads due to TAUand paging are inversely proportional. This can be clarified inthe following equations 1 and 2. Let us assume that X l

ij isthe cell-to-list assignment, where cells i and j reside in list l.Further, the constant costs of paging and TAU are denoted byCρ and Cu, respectively. We can now express the initial costfunctions of paging and TAU caused by a UE as follows:



4

TABLE ITABLE OF NOTATIONS

CTu(i) : Total signaling cost of TAU in cell i.Cu : TAU signaling cost of UE moving from

one cell i to another cell j that is notwithin the same list.

UEi : Total number of UEs served by cell i.ρ : Paging arrival rate.

Hij : Probability that a user moves from cell ito cell j.

Cρ : Paging cost of particular user equipmentin cell i.

CTρ(i) : Total paging cost in cell i.Cuρ(i) : Total paging and TAU overhead in cell i.

L : Total number of lists.N : Total number of cells.l : Individual list (index).κ : Maximum number of TAs that are as-

signed to list l.ω : Cost of MME relocation during handover.

HX l(i) : Inter-list handover rate of users in cell i.

OMl =

{1, if list l belongs to MME M ,0, otherwise

Y TAi =

{1, if cell i belongs to TA,0, otherwise

Decision Variables

σli : Usage ratio of each list l in cell i.

X lij =

{1, if cells i and j belong to l,0, otherwise

• The cost function of paging a specific UE within a certaincell i is given by

Cρ ∗L∑l

(σli +

N∑j=1,j 6=i

σljXlij

)(1)

In order to clarify the equation above (1), let us assumethat cell 1 and 2 are in the same list 1, the value of X1

12

is set to 1, all other variables are either set to one or zerodepending on whether or not they belong to the samelist. Equation (1) calculates the cost function of paging aspecific user belonging to particular cell which is cell 1in this example. Thus, variable X1

12 would be multipliedby the usage ratio of list 1 in cell 1 (denoted as σ1

1) aswell as a constant value of the paging cost Cρ. The valueof σ1

1 is 1 since cell 1 only belongs to one list which isthe only list provided to the cell. Similarly, σ1

2 is also onesince cell 2 belongs to one list only. Therefore, the corenetwork would page every cell belonging to list 1 that theintended user is within (cells 1 and 2 in this example).

• The cost function of TAU generated by a specific usercrossing two cells that are not within the same TALs l is

given by

Cu ∗L∑l

σli(1−X lij) (2)

Equation (2) is to calculate the tracking area update costgenerated by a specific user moving from cell i to cellj. Let us assume that a user moves from cell 2 to cell 3that is not within the same list (e.g, cell 2 belongs to list1 and cell 3 belongs to list 2). The TAU cost would bemultiplied by one for each two cells that are not withinthe same list. Note that the same reasoning given for thepaging cost is used here for the value of σ and hence, σ1

2

is set to 1 whereas σ22 is set to 0 because cell 2 belongs

to list 1 and not list 2. Therefore, the user would incur asignaling overhead equal to the TAU cost Cu.

C. Design Hypothesis

As stated previously, the model consists of two MMEpooling schemes. The centralized MME pooling scheme al-locates each TAL to an individual MME element. On theother hand, the distributed MME pooling scheme allocateseach TA or cell to an individual MME element. Note thatin our architecture, a TA always contains only one cell; thissimplifies the problem and avoids the cell-to-TA assignmentconstraint. Further, in this context, a cell or TA can be usedinterchangeably and will have the same constraint. Figures1 and 2 show the centralized and distributed MME poolingschemes, respectively. The model can have two levels of

MME

TA/Cell

TA/Cell

TA/Cell

MME

TA/Cell

TA/Cell

TA/Cell

TAL 1 TAL 2

Fig. 1. Centralized MME pooling scheme

configuration: cell/TA-to-TAL/MME assignment, which canbe handled periodically by the core network as illustratedin the first sub-problem, and TAL/MME-to-UE assignment,which will be handled through each cell as explained inthe second sub-problem. First, in terms of cell-to-TAL/MMEassignment, the system considers the mobility pattern of theUEs within the cells and dynamically assigns or re-assigns thecells inside the TALs in order to minimize the TAU signalingoverhead. Second, in terms of TAL-to-UE assignment, eachnode will assign a number of TALs to a portion of UEsresiding within that cell with a defined usage ratio, denotedby σli. The decision variable σli determines the usage ratioof each TAL within each cell. This results in overlappingTAL assignment to cells. Furthermore, this technique can also



5

TA/Cell

TA/Cell

TA/Cell TA/Cell

TA/Cell

TA/Cell

TAL 1 TAL 2

MME MME MME MME MME MME

Fig. 2. Distributed MME pooling scheme

provide load balancing through the MMEs in the centralizedpooling scheme, because each MME represents an individualTAL.

1) Cell/TA-to-TAL/MME Assignment: We attempt to de-termine the optimal cell/TA-to-TAL/MME assignment in thecentralized pooling scheme, while the distributed scheme con-siders cell-to-TAL assignment only. In the centralized scheme,each TAL can be allocated to each MME on a one-to-onebasis. Alternatively, in the distributed scheme, each cell isrepresented as a TA that is assigned to an individual MME.LTE has defined the maximum number of TAs that can beallocated to a TAL to be 16. As a basic definition for theassignment constraint, we can use the binary scalar decisionvariable X l

ij to determine whether cells i and j reside in thesame tracking area list. For example, if cell 1 and 2 are inthe same list 1, the value of X1

12 is set to 1, otherwise X112 is

set to 0. The same is applied to all the cells and lists in thesystem.The constant κ is the maximum number of TAs in the TALas described in the following equation:

N∑i

N∑j,j 6=i

X lij 6 κ, ∀l ∈ L (3)

The purpose of the above-mentioned constraint is to allocate amaximum of 16 cells/TAs to each TAL. However, to eliminatethe redundancy of cell/TA combinations in each TAL, anotherconstraint should be considered as follows:

X lij = X l

ji, ∀l ∈ L, ∀i, j, i 6= j ∈ N (4)

In this study, a dynamic algorithm is proposed to periodicallyupdate the cell-to-TAL configuration. A cell can have a numberof neighboring cells that are initially arranged as rings. Nu-merous studies, e.g., [3],[5],[23] have considered periodicallyexpanding or eliminating the cells arranged as rings. In fact,the ring-based assumption is not entirely efficient in terms ofminimizing the signaling load. This is because the conceptof rings might involve cells that have a low probability ofbeing visited owing to their mobility behavior. Conversely, ourtechnique involves analyzing the average mobility behavior ofthe UEs over a certain period and expanding or eliminatingspecific cells not strictly included in the entire ring. In other

words, the proposed technique selects a cell to be included oreliminated from a pool of cells that is not restricted directlyto the surrounding neighbors of the current cell. This cansignificantly minimize the signaling load caused by frequentTAL updates and lead to the inclusion of a greater numberof potential cells that are located outside the ring and havea higher probability of being visited by the UEs. Thus, it isan efficient scheme whereby cells can be chosen selectivelyrather than in terms of group of rings. Figure 3 shows thedifference between the proposed smart selection method andthe ring selection method. We note that the smart techniqueallows different selection shapes, such as the one depicted inorange. On the other hand, the ring selection technique, shownin green, can only include all the surrounding cells, regardlessof their probability of being visited.

Fig. 3. Ring neighbor selection versus smart cell selection

2) TAL-to-UE: Each TAL is assigned to UEs via theirserving cells. TALs are distributed among the users based onthe number of lists that the cell belongs to. This is becauseif a cell is assigned to multiple lists, the UEs need to haveaccess to these multiple lists.D. Problem Formulation

The original model that was proposed in [1] is used asa base model for both the centralized and the distributedpooling schemes. The objective of the model is to minimize thesignaling overhead resulting from both TAU and paging. Themodel is modified and divided into a bi-objective minimizationproblem as follows:• Objective function

min αN∑i

Cuρ(i) + βN∑i

HX l(i) (5a)

• Cost functions

Cuρ(i) = CTρ(i) + CTu(i), ∀ i ∈ N (5b)

HX l(i) =L∑l=1

UEi.Hi,j .(1−X lij) (5c)

• Centralized pooling-TAU cost

CTu(i) = UEi.Hij .Cu

[ L∑l=1

ω. OMl . σli(1−X l

ij)

],

∀ i, j, i 6= j ∈ N(5d)



6

• Distributed pooling-TAU cost

CTu(i) = UEi.Hij .Cu

[ L∑l

ω. Y TAi +

σLi (1−X lij)

],∀ i, j, i 6= j ∈ N

(5e)

• Paging cost

CTρ(i) = ρ .Cρ

[ L∑l=1

UEi.σli+

L∑l=1

N∑j,i 6=j

UEj . Xlij .σ

lj

], ∀ i ∈ N

(5f)

• ConstraintsL∑l=1

σli = 1, ∀ i ∈ N (5g)

N∑ij,i6=j

X lij 6 κ, ∀ l ∈ L (5h)

X lij = X l

ji, ∀l ∈ L, ∀ i, j, i 6= j ∈ N (5i)

0 ≤ σli ≤ 1 (5j)

OMl ∈ 0, 1 (5k)

X lij ∈ 0, 1 (5l)

The objective function (5a) is divided into two parts. The firstpart minimizes the signaling overhead caused by TAU andpaging and the second part minimizes the inter-list handover.Both parts are weighted with factors represented as α and β.These weights can be used by the service provider as controlparameters to prioritize an objective over the other. The firstpart of (5a) contains two cost functions, as shown in (5b),which combines the cost function of TAU in the centralizedscheme, as in (5d), and that in the distributed scheme, as in(5e), along with the second term, which has the paging costfunction, as in (5f). In the centralized scheme, the signalingload is determined by first calculating the number of UEs thatreside in a cell i and have a probability of moving to anothercell that is not within the same list, and then multiplyingthat number by the cost of inter-MME reallocation (ω) andthe usage ratio of each list (σli). In the distributed scheme,the cost of MME reallocation is multiplied by the number ofUEs that travel from one tracking area to another. The pagingcost function (5f) calculates the signaling load of pagingmessages that are triggered by the UEs inside the lists. Thecost considers the percentage of the overlapping lists used incell i multiplied by the decision variable X l

ij that determineswhether cell i and the neighboring cell j belong to the samelist. The second part of (5a) considers estimating the inter-listhandover rate of the average number of UEs in a cell i, asgiven in (5c). Constraint (5g) ensures fair usage by a set ofcells for every list/MME. Constraints (5j), (5k), and (5l) arethe boundary constraints.

E. Decomposition Model

The previous model is a mixed-integer nonlinear program-ming (MINLP) problem with quadratic equality constraints.This problem is a well-known NP-hard problem that is dif-ficult to solve to optimality [24]. In our previous study, thebinary decision variable was defined in the model and cell-to-TAL/MME assignment was done statically. In this context, wewant to have dynamic cell-to-TAL/MME assignment based onthe UE mobility patterns. Therefore, we propose a decompo-sition model that consists of two sub-problems, each of whichis defined as follows:Sub-problem 1. In a given set of cells that serve a number ofUEs, find the optimum cell-to-TAL/MME assignment allocatedperiodically to minimize the inter-list handover rate of UEstraveling from a cell to another cell that is not within thesame list TAL.Sub-problem 2. In a given set of TALs/MMEs that areoverlapping and used by a number of cells, find the optimumusage usage ratio of each TAL/MME to be given to a numberof UEs that are located inside a cell such that the signalingoverhead is minimized.

1) Sub-problem 1 Formulation:

min

N∑i

HX l(i) (6a)

HX l(i) =L∑l=1

UEi.Hi,j .(1−X lij), ∀ i, j, i 6= j ∈ N (6b)

N∑i

X lij ≥ 1, ∀ l ∈ L (6c)

(5h) and (5i) (6d)

(Hi,j − Limit).X lij ≥ 0, ∀ i, j, i 6= j ∈ N, ∀ l ∈ L (6e)

The objective function (6a) tries to create a suitable X lij that

tends to minimize the occurrence of UEs moving betweencells that are not within the same TAL. The cost function (6b)calculates the cost of the handover rate of UEs crossing cellsbelonging to different TALs. Constraint (6c) assigns at leasttwo cells to every list l. Constraint (6d) is used to satisfy theconditions of X l

ij formulation and limit the number of TAsinside each TAL. Finally, constraint (6e) prioritizes the cellshaving a higher probability of being visited for inclusion inthe same list, as indicated by the UE mobility patterns. Thisconstraint is calibrated and modified on the basis of the UEaverage speeds.

2) Sub-problem 2 Formulation:

minN∑i

Cuρ(i) (7a)

Cuρ(i) = CTρ(i) + CTu(i), ∀ i ∈ N (7b)

CTu(i) = (5d) or (5e) based on pooling scheme (7c)



7

CTρ(i) = ρ .Cρ

[ L∑l=1

UEi.σli+

L∑l=1

N∑j,i 6=j

UEj . Xlij .σ

lj

], ∀ i ∈ N

(7d)

L∑l=1

σli = 1, ∀ i ∈ N (7e)

0 ≤ σli ≤ 1 (7f)

X lij ∈ 0, 1 (7g)

The description of the objective function and significance ofthe constraints have been addressed in the previous subsection.

F. Mobility Pattern Model

We use the fluid flow model to simulate the mobilitybehavior of the users in the system. The fluid flow model isa well-known model that is commonly used in the literature.The model depicts the traffic flow rates of UEs moving outof a closed region represented as a cell or base station. For acertain cell i with perimeter L, UE density Ui, and averageUE velocity v, the average number of cell crossings per unittime is calculated as follows:

Ui.L.v

π(8)

In this context, cells are hexagonal in shape with side lengthR; hence, L is replaced with 6R.

G. SON Capability through MME

Self-Organizing Network is a paradigm that seeks to mini-mize the operation expenses related to network re-organizationin order to achieve higher efficiency and Quality of Experience(QoE) [25]. However, it is vital to deliver an updated versionof TAL due to the rapid changes in the mobility patternsof the UEs. The MME can be involved in triggering theself-optimizing capability through the network by sending theupdated TALs to the base stations. The purpose of dynamiccell-to-TAL/MME reassignment is to alleviate the signalingoverhead that results specifically from the movement of UEsfrom one cell to another cell that is not in the same TAL/MME.The dynamic algorithm will set a timer to trigger the desiredchanges in the TALs accurately and to distribute them amongthe cells. Two vital factors tend to have a major impact onthe algorithm: the average handover rate of the UEs betweenthe cells, and the given value of LIMIT that prioritizes thecombinations of cells to be allocated within the same list orMME. The parameter LIMIT can be chosen by the networkoperator as a control parameter to prioritize cells over othersin order to reduce the required TAU. Moreover, the value ofLIMIT is related to the average handover rate between cellsor the average UE velocity. Equation (6e) shows the relationbetween LIMIT and the average handover rate. We can seethat as the value of LIMIT increases, a greater number of cellshaving a higher handover rate between each other are includedin the same list. In our study, we assumed that LIMIT can have

different values related to the UE average velocity. A detaileddescription of the dynamic algorithm is provided in the nextsubsection.

H. Dynamic TAL Algorithm

The proposed dynamic algorithm mainly focuses on Sub-problem 1, which plays the role of allocating suitable combi-nations of cells in each TAL/MME. This tends to minimizethe handover rate between cells that are not in the sameTAL/MME. As mentioned earlier, a timer will be used totrigger the frequent update of cell-to-TAL/MME assignment.Furthermore, the TAL/MME is updated within a selective setof cells that are determined by the value of LIMIT, whichvaries periodically corresponding to the UE mobility patterns.

The timer is adjusted byR

v, which determines the average

time required for the UEs to cross a certain cell. Pseudo-code depicted in Algorithm 1 explains the mechanism of theproposed SON dynamic algorithm in the model. In line 1,the algorithm initializes the input values, which include thenumber of TALs/MMEs used in the system and the number ofTAs, which is equivalent to the number of cells. Further, boththe average velocity rate and the average number of UEs ineach cell are determined. Finally, the cell radius and the initialcell-to-list assignment are given. The output of this algorithmis the optimized cell-to-TAL/MME assignment. In lines 4-6,at a given time, the algorithm calculates the average crossingrate of UEs residing in each cell in order to determine thehandover rate Hij between the cells. Then, the LIMIT valueis calibrated to selectively choose the neighboring cells of anycell i that have the highest handover rate and allocate thoseneighboring cells to the same list or MME. In lines 7-8, theinitial assignment is modified by either eliminating or retainingthe old cells that had previously been allocated to the lists. Thisis achieved by solving Sub-problem 1. In lines 9-15, a timer isset to estimate the average time needed for a UE to leave thecells. The timer determines the optimal frequency with whichthe model should reassign the cells to the lists. Moreover, thesystem randomly changes the values of the UEs in every eventprior to the trigger time and keeps calculating the signalingcost by solving Sub-problem 2.

IV. HEURISTIC ALGORITHM

A. Algorithm Description

This section describes the development of a heuristic al-gorithm that depends on equal distribution of the TAL/MMEload among the relevant cells. This approach has the advantageof minimizing the TAUs among the cells in a less complexmanner. In this context, the current heuristic algorithm differsfrom the preceding one presented in [1]; it follows the sameconcept as the optimal formulation, which divides the orig-inal algorithm into two sub-problems. The first sub-problemprovides a new method for assigning cells to TALs/MMEs byselecting the cells having the highest values of UE crossingrates such that the TAU frequency is minimized. The cell-to-list assignment will also be performed dynamically by definingthe timer in the manner described above for the dynamic



8

Algorithm 1 SON Dynamic Algorithm1: Input: l = {1, 2, ..., L}: All lists = MMEs

TA = {1, 2, ..., N}: All tracking areasUE: Average number of UEs in cell ivRange = [v0 : vmax]v = Random such that v ∈ vRangeUi: Density of UEs in cell iR: Cell diameterUEi: Initial number of UEs in cell i ∈ TA.X lij : Initial random cell-to-TAL/MME assignment

2: Output: X lij : Optimized cell-to-TAL/MME assignment

3: for t ∈ TotalT ime do4: define UEijr = Ui.L.v

Π : Average crossing from i to j5: define Hij = UEijr

UEi6: update LIMIT7: Calculate Sub-problem 1:8: update X l

ij

9: for t′ ∈ (R/v) do10: Calculate Sub-problem 2:11: for i ∈ N do12: update UEi = Rand(UE)13: end for14: end for15: end for

algorithm (Algorithm 1). Pseudo-code shown in Algorithm2 describes Sub-problems 1 and 2 of the heuristic algorithmin detail. In line 2, the essential parameters required in thealgorithm are defined. The output of Sub-problem 1 is theheuristic cell-to-TAL/MME assignment. In lines 4-14, thealgorithm calculates the UE crossing rates based on theiraverage speeds and then sets the value of X l

ij that correspondsto the maximum crossing rates between the cells. Line 16defines the input of Sub-problem 2. The output of Sub-problem2 is the usage ratio of the lists in each cell. In lines 18-30,the usage percentage of each list to which a cell belongs isdetermined by performing a search of all the lists that servethe cells. The ratio is calculated by distributing the usage ofthe TALs/MMEs that serve the cells with equal percentages.This will accomplish a load balance on the MME side thatcannot be compromised by any given cell. Finally, in lines31-37, the signaling overhead of both schemes is calculatedusing the appropriate equations.

B. Algorithm Complexity

To justify the need for the heuristic algorithm, we mustfirst discuss the complexity of the decomposition algorithm.The original problem presented in (5) is intractable since themodel is a mixed-integer nonlinear programming (MINLP)problem with quadratic equality constraints, which is a well-known NP-hard problem that is difficult to solve to optimality[24]. Therefore, the decomposition model has been adoptedto solve it. Sub-problem 1 has O(L

Nκ ) complexity because

each list can have at most κ cells among N available cellswith possible overlap. Sub-problem 2 has O(LN2) complexitybecause the usage ratio of each list in all the cells is calculated

Algorithm 2 Heuristic Algorithm1: Sup-Problem 12: Input: l = {1, 2, ..., L}: All lists = MMEs

TA = {1, 2, ..., N}: All tracking areasUE: Average number of UEs in cell ivRange = [v0 : vmax]v = Random such that v ∈ vRangeUi: Density of UEs in cell iR: Cell diameterUEi: Initial number of UEs in cell i ∈ TA.X lij : Initial random cellto-TAL/MME assignment

3: Output: X lij : Optimized cell-to-TAL/MME assignment

4: define UEijr = Ui.L.vΠ : Average crossing rate from i to j

5: define Hij = UEijrUEi

6: for i ∈ N do7: for l ∈ L do8: if UEijr = MAX then9: SET X l

ij = 110: else11: continue12: end if13: end for14: end for15: Sup-Problem 216: Input: l = {1, 2, .., L}: All lists

TA = {1, 2, ..., N}: All tracking areasX lij : Cell-to-list binary indicator

17: Output: σli: Usage ratio of list l in cell i18: for i ∈ N do19: define NL = φ20: for l ∈ L do21: if i ∈ l then22: update NL = {NL ∪N}23: else24: continue25: end if26: for N ∈ NL do27: σli =

1||NL||

28: end for29: end for30: end for31: Calculate:

Cρ(i) as per (7d)32: if Centralized Scheme then33: Calculate:

CTu(i) as per (5d)34: else35: Calculate:

CTu(i) as per (5e)36: end if37: Calculate:

N∑i

Cuρ(i) =N∑i

(Cρ(i) + CTu(i))



9

for each cell. Hence, the total complexity of the decompositionalgorithm is O(L

Nκ )+O(LN2) = O(L

Nκ ). On the other hand,

the heuristic algorithm has much lower complexity. Usingthe proposed heuristic algorithm, Sub-problem 1 has O(LN)complexity because a decision as to whether each cell belongsto the list is taken in each iteration. Sub-problem 2 also hasO(LN) complexity because the usage ratio of each list withineach cell is calculated. Therefore, the total complexity of theheuristic algorithm is O(LN), which is much lower than thatof the decomposition algorithm.

V. PERFORMANCE EVALUATION

To evaluate the performance of the proposed models, weperformed MATLAB simulations at four different speeds,categorized as very slow (0–8 m/s), slow (8–16 m/s), normal(16–25 m/s), and fast (25–33 m/s). The simulation assumedan environment of 10 cells, each having a different numberof UEs. The number of UEs was distributed uniformly amongthe cells with an average of 100 users. Table II shows thesimulation parameters and their values.

TABLE IISIMULATION PARAMETERS & VALUES

Parameter ValueNumber of cells N 10Number of TAL L 3Average number of users 100 per cell iNumber of TAs 10Paging rate ρ 0.05UE speeds 0,8,16,25 and 33 m/sCell Radius 500 mTAU to Paging cost 10:1

The decomposition and heuristic models were implementedand compared with each other. We note that each sub-problemhas the same importance; thus, the weight factors α and βwere assigned values of 1. The simulation tested the modelsstatically and dynamically in the case of the centralizedand distributed schemes. The dynamic algorithm presentedin Algorithm (1) was implemented in the decomposition andheuristic algorithms, whereas the same analogy provided inour previous study [1] was used for the static algorithm.Moreover, a random algorithm was added in the evaluationfor the purpose of providing a simpler approach to replaceSub-problem 1, as it generates random cell-to-TAL/MMEallocation dynamically. Moreover, the random technique en-ables us to evaluate the effectiveness of frequent randomallocation without prior knowledge of the signaling overhead.This approach offers a less complex solution that does notaffect the core network. The total signaling overhead of theTAU and paging combination was employed as a performancemetric for the proposed algorithms. We used MATLAB’sbuilt-in functions intlinprog and linprog to optimally solveSub-problems 1 and 2, respectively.

A. Total Signaling CostFigure 4 shows the average total signaling overhead caused

by paging and TAU at different speeds for the decomposition-based algorithms, namely the dynamic, random, and static

5 10 15 20 25 30 35Average Speed (m/s)

1

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

2

Ave

rage

Tot

al C

ost -

Cen

tral

ized

×104

DynamicRandomStatic

Fig. 4. Total signaling overhead cost for the centralized scheme

algorithms, in the case of the centralized scheme. We note thatthe SON dynamic algorithm outperforms the static and randomalgorithms. On the other hand, the random optimal algorithmshows better performance in terms of the signaling overhead.This is because the random algorithm dynamically changesthe TAL in a random manner, whereas the static algorithmmaintains the same TAL at all times. We also note that theUE speed has a significant impact as it exceeds 25 m/s.

Figure 5 depicts the average total signaling overheadfor the centralized scheme when different paging rates areconsidered at different speeds. It is noted that the sametrend is observed as that shown in Figure 4 when changingthe average speeds of the users. The second observation isthat the total signaling overhead increases as the paging rateincreases. This is expected since the paging cost is dependenton the paging rate as illustrated in equation 5f. Figure 6

0.05 0.25 0.5Paging Rate

0

2

4

6

8

10

12

14

Ave

rage

Tot

al C

ost-

Cen

tral

ized

0-8 m/s8-16 m/s16-25 m/s25-33 m/s

Fig. 5. Total signaling overhead cost for different paging rates

shows the average total signaling overhead at different speeds



10

5 10 15 20 25 30 35Average Speed (m/s)

1.6

1.8

2

2.2

2.4

2.6

2.8

3A

vera

ge T

otal

Sig

nalin

g C

ost-

Dis

trib

uted

×105

DynamicRandomStatic

Fig. 6. Total signaling overhead cost for the distributed scheme

for the decomposition-based algorithms in the case of thedistributed scheme. The results show the same trends as thoseof the centralized scheme. Furthermore, we can conclude thatthe centralized scheme outperforms the distributed schemebecause of the frequent MME relocation in the latter case.

A comparison between the decomposition and heuristicalgorithms is shown in Figures 7 and 8. Figure 7 shows thecomparison in the case of centralized scheme, where it is ob-served that the heuristic approach shows acceptable behaviorat lower speeds compared to the decomposition model. This isbecause the decision variable sigma diversifies the TAL/MMEpool, thereby minimizing the TAU signaling and the overallsignaling cost. Figure 8 shows a comparison between thedecomposition and heuristic methods in the case of distributedscheme. The heuristic algorithm offers a near-optimal solutioncompared to the optimal decomposition model. This is aresult of the dominant value of the MME relocation weight,which has a major effect on the signaling in the case of thedistribution scheme.

Table III summarizes the average percentage improvementin the random and dynamic assignment algorithms comparedto the static algorithm. In the centralized case, the random al-gorithm shows an improvement of 8%–16.45%, while the dy-namic algorithm shows an improvement of 25.23%–34.98%.Similarly, the random algorithm results in an improvementof 6.53%–9.37% and the dynamic algorithm results in animprovement of 15.74%–18.88% in the distributed case.

TABLE IIIAVERAGE IMPROVEMENT PERCENTAGE

Parameter Random DynamicTotal Signaling Cost-Centralized [+8%,+16.45%] [+25.23%,+34.98%]

Total Signaling Cost-Distributed [+6.53%,+9.37%] [+15.74%,+18.88%]

Power Consumption [+16.6%,+17.7%] [+28.83%,+39.32%]

5 10 15 20 25 30 35Average Speed (m/s)

0.5

1

1.5

2

2.5

3

Ave

rage

Cen

tral

ized

Tot

al C

ost

×104

DecompositionHeuristic

Fig. 7. Total signaling overhead cost in the centralized scheme for decom-position algorithm vs. heuristic algorithm

5 10 15 20 25 30 35Average Speed (m/s)

1.6

1.8

2

2.2

2.4

2.6

2.8

3

Ave

rage

Tot

al S

igna

ling

Cos

t- D

istr

ibut

ed

105


Fig. 8. Total signaling overhead cost in the distributed scheme for decompo-sition algorithm vs. heuristic algorithm

B. Power Efficiency

In this subsection, we discuss the efficiency of the proposeddynamic approach in terms of power consumption. One wayto measure the efficiency of the SON dynamic model is toevaluate the UE battery life. It is estimated that each TAUprocedure consumes around 10 mW of a regular smart-phonebattery. The results in Figures 9 and 10 depict the averagetotal battery consumption for one UE in an hour. Figure 9compares the dynamic, random, and static approaches. Thefigure indicates significant power savings in the dynamicapproach compared to the static approach. In addition, therandom dynamic approach shows slightly lower power con-sumption than the static approach. Figure 10 compares thedynamic decomposition and heuristic solutions. It is clearthat the decomposition solution performs better. However, the



11

5 10 15 20 25 30 35Average Speed (m/s)

20

40

60

80

100

120

140

160

180

200

220A

vera

ge T

otal

UE

Pow

er C

onsu

mpt

ion

(mW

)

DynamicRandomStatic

Fig. 9. Average total UE power consumption (mW)

heuristic solution offers near-optimal results at slow speeds.Table III summarizes the power savings in the random anddynamic assignment algorithms compared to that in the staticalgorithm. The power savings range from 16.6% to 17.7% forthe random algorithm and from 28.83% to 39.32% for thedynamic algorithm.

5 10 15 20 25 30 35Average Speed (m/s)

20

40

60

80

100

120

140

160

180

200

220

Ave

rage

Tot

al U

E P

ower

Con

sum

ptio

n (m

W)


Fig. 10. Average total UE power consumption (mW) for decompositionalgorithm vs. heuristic algorithm

C. Related Work vs. Our Approach

This subsection compares the latest related methods inthe literature with our approach. Three studies have beenconsidered in the comparison: [5],[3], and [23]. The authorsproposed three different approaches that dynamically solvethe problem of cell-to-TA/TAL assignment for the purposeof minimizing the total signaling overhead due to TAU and

paging signals. Specifically, two of the studies, i.e., [5] and[23], have proposed a SON mechanism to enhance the intel-ligence of their algorithms. However, we note that none ofthe aforementioned studies have included MME realization intheir solutions, which is an important factor in the signalingoverhead. Consequently, both MME realizations are adaptedin the compared studies in order to ensure accurate andfair comparison. All the compared studies use the ring-basedapproach for cell-to-TAL assignment.

Figures 11 and 12 show the total signaling overhead inthe centralized and distributed schemes. The key differencebetween our technique and other related methods is that ouralgorithm uses a smart technique for choosing the candidatecells to be assigned in TAL, whereas the other methods use theconventional cell-to-TAL assignment [3],[5], [23]. The smartselection technique outperforms the ring-based techniquesbecause the smart selection assignment of cells alleviates thefrequent TAU signaling by decreasing the probability of theUE moving from one cell to another cell that is not in the sametracking area list. Furthermore, the smart selection techniqueincludes a greater number of cells that are more likely to bevisited by the UE; thus, it is very efficient, especially for UEsmoving at high speeds. Another key metric for the variationin the results is the overlapping TAL, which increases theprobability of fewer TAU updates caused by the UE whenit travels from one cell to another. Finally, the method fortriggering the dynamic configuration or SON technique is alsoan important factor that affects minimization of the signalingoverhead. The SON technique is used in [5], and it facilitatesthe transition between stages through timers and an activationthreshold for triggering the dynamic configuration. The authorsstatically solved the problem of cell-to-TAL assignment untilthe activation threshold was reached. Although the activationthreshold is not defined clearly, the technique used is toocomplicated to be implemented in a real-world scenario,especially with the rapid variation in UE velocity. AnotherSON technique was proposed in [23], where the authorsintroduced an overlapping TAL scheme solved statically in amanner similar to [5] until a threshold was reached. The maindifference in this study is the introduction of the overlappingTAL technique, which offers the advantage of lower TAUsignaling overhead. However, no available SON technique forsolving TAL assignment has adapted the triggering schemebased on the velocity variation related to the UEs. Our schemecontinuously adapts the TAL assignment based on the averageUE velocity.

Figure 13 shows the power consumption of the relatedmethods and our method. The comparison has been made atdifferent speeds. Our method achieves greater power savings,which vary from +32.7% to +46.9%.

Table IV summarizes the given comparison in terms ofthe average improvement percentage between our method(SON Dynamic/Smart) and the related methods. Our methodoutperforms the related methods significantly in terms of thecentralized and distributed schemes as well as in terms ofpower consumption. For instance, the SON Dynamic/Smartalgorithm in the case of the centralized scheme outperformsthe method proposed in [23] by +23.87% to +28.82% which



12

TABLE IVAVERAGE IMPROVEMENT PERCENTAGE COMPARED WITH RELATED METHODS

Algorithm Used Overlapping SON-Enabled SON Dynamic/SmartAlgorithm Centralized

SON Dynamic/SmartAlgorithm Distributed Power Consumption

Ref [3]√

× [+30.8%,+35.69%] [+17.51%,+18.49%] [+36.48%,+39.85%]Ref [23]

√ √[+23.87%,+28.82%] [+14.62%,+16.11%] [+28.73%,+32.66%]

Ref [5] ×√

[+36.69%,+42.37%] [+20%,+21.27%] [+42.84%,+46.91%]

5 10 15 20 25 30 35Average Speed (m/s)

0.8

1

1.2

1.4

1.6

1.8

2

2.2

2.4

Ave

rage

Cen

tral

ized

Tot

al C

ost

104

Our algorithmRef [3]Ref [23]Ref [5]

36.6% 30.8% 23.8%

42.3%

28.8%

35.6%

Fig. 11. Total signaling overhead cost comparison between our algorithm(centralized) and related methods

5 10 15 20 25 30 35Average Speed (m/s)

1.6

1.8

2

2.2

2.4

2.6

2.8

3

3.2

Ave

rage

Dis

trib

uted

Tot

al C

ost

105


21.2%

14.6%

18.4%

16.1%

17.5%20%

Fig. 12. Total signaling overhead cost comparison between our algorithm(distributed) and related methods

5 10 15 20 25 30 35Average Speed (m/s)

20

40

60

80

100

120

140

160

180

200

220

Ave

rage

Tot

al U

E P

ower

Con

sum

ptio

n (m

W)


46.9%

39.9%32.7%

Fig. 13. Average total UE power consumption (mW) comparison betweenour algorithm and related methods

uses overlapping and SON techniques. The lower percentageis taken in the case of the lowest speed while the higherpercentage is taken in the case of the highest speed, as shownin Figure 11.

VI. CONCLUSION

SON, a new concept introduced in release 8 of 3GPPfor LTE networks, is a promising paradigm that enablesself-planning, self-managing, and self-optimizing of networks.Therefore, SON has been widely accepted across different ap-plications. A number of SON use cases have been proposed toovercome increasing operational expenses. In this context, weproposed a SON approach to alleviate the signaling overheadcaused by TAU and paging. We used SON as an enablerto perform dynamic cell-to-TAL/MME reconfiguration. Ourapproach can be considered as a SON use case that signif-icantly minimizes the signaling overhead. Furthermore, twoschemes used in the previous work, namely the centralizedand distributed MME pooling schemes, were implementedand investigated dynamically. We used the well-known fluidflow model to simulate the movement of UEs within thesystem. The model consists of two sub-problems derivedfrom our original NP-hard problem formulation. The first sub-problem is a binary integer programming problem, whereasthe second is a linear programming problem. In addition,



13

a new smart selection method was proposed to intelligentlyselect the potential cells in the TAL/MME. Our method wasshown to outperform conventional ring selection, which iscommonly used in the literature. Finally, a less complexheuristic solution was proposed, which is easy to implementand gives a sub-optimal result. The results showed that thedynamic decomposition solution also achieves greater powersaving than previous methods.

ACKNOWLEDGEMENTSThis work was funded in part by Royal Commission for

Jubail and Yanbu through the Saudi Cultural bureau in Canada.This support is greatly appreciated.

REFERENCES

[1] E. Aqeeli, A. Moubayed, and A. Shami, “Towards intelligent LTEmobility management through MME pooling,” in IEEE 58th GlobalCommunications Conference (GLOBECOM’15), Dec. 2015.

[2] “Signaling is growing 50% faster than data traffic,” Nokia SiemensNetwork, 2012.

[3] T. Deng, X. Wang, P. Fan, and K. Li, “Modeling and performanceanalysis of tracking area list-based location management scheme in ltenetworks,” IEEE Transactions on Vehicular Technology, vol. PP, no. 99,pp. 1–1, 2015.

[4] “New son use-case: Tracking area optimization,,” 3GPP TSG-RAN WG3Std. R3-071 594, Aug 2007.

[5] J. Ferragut and J. Mangues-Bafalluy, “A self-organized tracking arealist mechanism for large-scale networks of femtocells,” in 2012 IEEEInternational Conference on Communications (ICC’12), June 2012, pp.5129–5134.

[6] S.-R. Yang, Y.-C. Lin, and Y.-B. Lin, “Performance of mobile telecom-munications network with overlapping location area configuration,”IEEE Transactions on Vehicular Technology, vol. 57, no. 2, pp. 1285–1292, March 2008.

[7] D. Gu and S. Rappaport Stephen, “Mobile user registration in cellularsystems with overlapping location areas,” in 1999 IEEE 49th VehicularTechnology Conference (VTC’99), vol. 1, Jul 1999, pp. 802–806 vol.1.

[8] K.-H. Chiang and N. Shenoy, “A 2-d random-walk mobility model forlocation-management studies in wireless networks,” IEEE Transactionson Vehicular Technology, vol. 53, no. 2, pp. 413–424, March 2004.

[9] ——, “Performance of an overlapped macro-cell based location areascheme,” in IEEE International Conference on Communications (ICC’03), vol. 1, May 2003, pp. 475–481 vol.1.

[10] A. Bar-Noy, I. Kessler, and M. Sidi, “Mobile users: to update ornot to update?” in 13th IEEE Networking for Global Communications(INFOCOM ’94), Jun 1994, pp. 570–576 vol.2.

[11] G. Wan and E. Lin, “A dynamic paging scheme for wirelesscommunication systems,” in 3rd Annual ACM/IEEE InternationalConference on Mobile Computing and Networking, ser. MobiCom ’97.New York, NY, USA: ACM, 1997, pp. 195–203. [Online]. Available:http://doi.acm.org/10.1145/262116.262147

[12] S. Tabbane, “An alternative strategy for location tracking,” IEEE Journalon Selected Areas in Communications, vol. 13, no. 5, pp. 880–892, Jun1995.

[13] I. Widjaja, P. Bosch, and H. La Roche, “Comparison of MME signalingloads for long-term-evolution architectures,” in 2009 IEEE 70th Vehic-ular Technology Conference Fall (VTC’09-Fall), Sept 2009, pp. 1–5.

[14] T. Taleb, K. Samdanis, and A. Ksentini, “Supporting highly mobileusers in cost-effective decentralized mobile operator networks,” IEEETransactions on Vehicular Technology, vol. 63, no. 7, pp. 3381–3396,Sept 2014.

[15] A. Kunz, T. Taleb, and S. Schmid, “On minimizing servingGW/MME relocations in LTE,” in IEEE 6th International WirelessCommunications and Mobile Computing Conference, ser. IWCMC ’10.New York, NY, USA: ACM, 2010, pp. 960–965. [Online]. Available:http://doi.acm.org/10.1145/1815396.1815615

[16] S. Razavi, D. Yuan, F. Gunnarsson, and J. Moe, “Dynamic tracking arealist configuration and performance evaluation in LTE,” in 2010 IEEEGLOBECOM Workshops (GC’10 Wkshps), Dec 2010, pp. 49–53.

[17] Y. W. Chung, “Adaptive design of tracking area list in LTE,” in 2011Eighth International Conference on Wireless and Optical Communica-tions Networks (WOCN’11), May 2011, pp. 1–5.

[18] M. Toril, S. Luna-Ramirez, and V. Wille, “Automatic replanning oftracking areas in cellular networks,” IEEE Transactions on VehicularTechnology, vol. 62, no. 5, pp. 2005–2013, Jun 2013.

[19] S. Razavi, D. Yuan, F. Gunnarsson, and J. Moe, “Exploiting trackingarea list for improving signaling overhead in LTE,” in 2010 IEEE 71stVehicular Technology Conference (VTC’10-Spring), May 2010, pp. 1–5.

[20] S. Modarres Razavi and D. Yuan, “Mitigating mobility signalingcongestion in LTE by overlapping tracking area lists,” in Proceedingsof the 14th ACM International Conference on Modeling, Analysisand Simulation of Wireless and Mobile Systems, ser. MSWiM ’11.New York, NY, USA: ACM, 2011, pp. 285–292. [Online]. Available:http://doi.acm.org/10.1145/2068897.2068947

[21] S. Razavi and D. Yuan, “Reducing signaling overhead by overlappingtracking area list in LTE,” in 2014 7th IFIP Wireless and MobileNetworking Conference (WMNC’14), May 2014, pp. 1–7.

[22] R.-H. Liou, Y.-B. Lin, and S.-C. Tsai, “An investigation on LTE mobilitymanagement,” IEEE Transactions on Mobile Computing, vol. 12, no. 1,pp. 166–176, Jan 2013.

[23] S. Modarres Razavi, D. Yuan, F. Gunnarsson, and J. Moe, “On dynamicsignaling congestion mitigation by overlapping tracking area lists,” J.Netw. Comput. Appl., vol. 53, no. C, pp. 164–172, Jul. 2015. [Online].Available: http://dx.doi.org/10.1016/j.jnca.2015.03.009

[24] P. Bonami, L. T. Biegler, A. R. Conn, G. CornueJols, I. E. Grossmann,C. D. Laird, J. Lee, A. Lodi, F. Margot, N. Sawaya, and A. WaChter,“An algorithmic framework for convex mixed integer nonlinearprograms,” Discret. Optim., vol. 5, no. 2, pp. 186–204, May 2008.[Online]. Available: http://dx.doi.org/10.1016/j.disopt.2006.10.011

[25] “Self-optimizing network, the benefits of son in lte,,”http://www.4gamericas.org/ 2011.

Emad Aqeeli received his B.Sc. and M.Sc. degreesfrom KAU University. He worked with the RoyalSaudi Air Defense Forces from 2007 to 2012. SinceAugust 2012, he has been employed at Yanbu Uni-versity College, Royal Commission for Jubail andYanbu. He is currently pursuing his Ph.D. in Elec-trical and Computer Engineering at the Universityof Western Ontario.

Abdallah Moubayed received his B.E. degree inElectrical Engineering from the Lebanese AmericanUniversity, Beirut, Lebanon, in 2012 and his M.Sc.degree in Electrical Engineering from King Abdul-lah University of Science and Technology, Thuwal,Saudi Arabia, in 2014. He is currently pursuing hisPh.D. (started September 2014) in Electrical andComputer Engineering at the University of West-ern Ontario, London, Ontario, Canada. His researchinterests include wireless communication, resourceallocation, and wireless resource virtualization.

Abdallah Shami received his B.E. degree in Elec-trical and Computer Engineering from the LebaneseUniversity in 1997 and his Ph.D. degree in Elec-trical Engineering from the Graduate School andUniversity Center, City University of New York, inSeptember 2002. Further, in September 2002, hejoined the Department of Electrical Engineering atLakehead University, Thunder Bay, Ontario, Canada,as an Assistant Professor. Since July 2004, he hasbeen at Western University, where he is currentlya Professor in the Department of Electrical and

Computer Engineering. His current research interests are in the areas ofnetwork optimization, cloud computing, and wireless networks.

Documents

Dynamic SON-Enabled Location Management in LTE Networks · 2019-05-09 · Moreover, LTE suffers from greater signaling overhead than 3G technologies because its ﬂat IP architecture