1 Dynamc SON-Enabed Locaton Management n LTE Networks Emad Aqee, Abdaah Moubayed, and Abdaah Sham Western Unversty, London, Ontaro, Canada e-mas: {eaqee, amoubaye, asham}@uwo.ca Abstract Wreess networks are facng varous chaenges that demand contnuous and rapd mprovement. Long-Term Evouton (LTE) s a preferred wreess technoogy because of ts satsfactory performance. Owng to an exponenta ncrease n demand and new potenta appcatons, the core network of LTE, whch s known as the Evoved Packet Core (EPC), s affected by a surge n sgnang caused by a varety of contro functons. The sgnang overhead decreases the users Quaty of Experence (QoE). The current study attempts to mprove the ntegence of ocaton management technques. As an extenson of our prevous study [1], a Sef-Organzng Network (SON) that enabes dynamc reconfguraton of ceto-tal/mme s ntroduced. Both centrazed and dstrbuted poong schemes are tested n terms of sgnang overhead and user power consumpton. A decomposton mode that reduces the orgna formuated probem to two sub-probems s proposed, each of whch s soved optmay. In addton, a smart ce-to- TAL seecton scheme s proposed to prortze potenta ces that mght be vsted by a user equpment (UE). Our method s shown to outperform severa state-of-the-art methods presented n the terature. Fnay, a heurstc agorthm s presented to obtan a ess compex souton than the optma one. Index Terms LTE, Locaton Management, Trackng Area (TA), Trackng Area Lst (TAL), TAU, Pagng, MME poong, SON I. INTRODUCTION Mobe networks have been attractng consderabe attenton from both ndustry and academa snce the eary 1980s, when the frst generaton (1G) of anaog ceuar networks was ntroduced as the Advanced Mobe Phone System (AMPS). Ten years ater, the second generaton (G) was aunched to support mobe phones and mted data connectons over the.5g extenson. Wth the support of Genera Packet Rado Servce (GPRS),.5G networks coud use crcut swtchng for voce appcatons and packet swtchng for data transmsson. The evouton contnued wth the ntroducton of the thrd generaton, whch provded faster speeds wth better effcency and quaty of servce. Subsequenty, the fourth generaton ntroduced two standards, namey WMAX and Long-Term Evouton (LTE), whch enhanced the capabty of packet swtchng to provde users wth better performance. LTE has ganed more attenton than WMAX, as t supports hgher speeds, better performance, and scaabe bandwdth. Despte these advancements, wreess technooges contnue to be chaenged by severe traffc that affects ther bandwdth. The proferaton of hand-hed devces and ther appcatons has caused sgnang traffc to exceed user data traffc. Sgnang traffc s ntated each tme there s a transmsson or recepton of packet streams between user equpments (UEs) and mobe networks, regardess of the actua sze of the data traffc. Noka Semens Networks have predcted that n the comng years, the ncrease n sgnang w be up to 50% faster than that n data traffc. Moreover, LTE suffers from greater sgnang overhead than 3G technooges because ts fat IP archtecture does not ncude a medum entty, such as a Rado Network Controer (RNC), between a base staton and the core network []. Thus, the average sgnang overhead of LTE s 4% greater than that of HSPA per subscrber. Pagng and Trackng Area Updates (TAU) have the most sgnfcant sgnang mpact on the Evoved Packet Core (EPC), specfcay on the Mobty Management Entty (MME). Pagng and TAU are essenta functons for UE ocaton management. They are used to track the user s ocaton and provde constant updates to the EPC. Pagng and TAU are defned as foows: 1) Pagng: messages sent by the MME to ocate a partcuar UE n a trackng area (TA). ) Trackng area update: messages sent by UEs to the MME when they move from one trackng area to another. To dentfy the UE ocaton, the LTE core network pages the atest trackng area that the UE was regstered to. The pagng sgna s receved by a of the ces that resde n the same trackng area. Addtonay, the UE w update the core network by sendng a trackng area update (TAU) sgna once t moves from one trackng area to another. Trackng Area (TA) s the area n whch the MME can ocate a specfc user wthn a defned set of ces. Ths technque s used n LTE and was orgnay nherted from prevous G and 3G technooges. However, TA has a number of mtatons that have ed to the ntroducton of the new concept of Trackng Area Lst (TAL), whereby severa TAs are grouped nto a snge TAL. TAL has the same functonaty as TA but wth the added fexbty of a set of TAs wthn the TAL. Thus, TAL can aevate the sgnang oad due to trggerng TAU each tme a UE moves from one ce to another. Varous mobty management technques have been nvestgated extensvey from dfferent perspectves, such as overappng, dynamc, and statc ce-to-tal assgnment. The current study s an extenson of our prevous study, whch ntroduced ce-to-tal assgnment wth two MME poong schemes. Centrazed and dstrbuted MME poong schemes were nvestgated n order to expore the dfference between both schemes statcay [1]. In ths context, we ntend to
enabe adaptve onne ce-to-tal assgnment n order to further nvestgate the proposed poong schemes. UEs are usuay n contnuous movement and ther coordnaton s not statc. Hence, the nta statc assgnment for ce-to-tal w graduay become neffectve over a specfc perod. Therefore, there s a need to revse the TA assgnment constanty n order to sut the current mobty state. Unke conventona TA, the TAL concept aows TA assgnment to be modfed wthout nterrupton of servce. Ths s an advantage of TAL over conventona TA, because TAL provdes greater fexbty to the system. Moreover, LTE aows for an auto-reconfguraton feature that adapts the network confguraton whenever there s a change n the UE statstcs, such as movement patterns and oads. In reease 8, 3GPP ntroduced the concept of a Sef- Organzng Network (SON) that provdes a methodoogy for pannng, managng, and optmzng mobe networks n order mprove performance effcency and system reabty. SON has been wdey accepted n ndustry and academa [3,4]. 3GPP has aso reeased dfferent use cases for LTE, offerng sef-optmzng and sef-heang paradgms. In ths context, adaptve TA st management can be used as an SON use case as n [5], whch can further reduce the sgnang oad. In addton, ntroducng dfferent technques that reate the TA st to the behavor of the mobe network woud further optmze the sgnang overhead. Thus, ce-to-tal assgnment can be engneered dynamcay whe the UE s n contnuous movement. The system keeps anayzng the mobty pattern and contnuousy updates the TA assgned to the st. Thus, the frequency of TAU w be reduced sgnfcanty. In ths study, the mobty pattern s obtaned usng a fud fow mode to estmate the handover correaton between ces. One way to measure the effcency of a dynamc system s to examne the UE battery fe. In our study, the UEs battery consumpton n statc and dynamc technques w be compared. The contrbutons of ths study can be summarzed as foows: 1) A dynamc ce-to-tal probem s proposed and formuated as a mxed nteger non-near programmng (MINLP) probem wth quadratc equaty constrants. Ths approach s dfferent from the prevous approach presented n [1], where the probem was soved statcay by fndng the optma assgnment once. The proposed dynamc technque s reazed through an SON scheme aong wth a new smart ce seecton approach nstead of the conventona rng-based ce seecton presented n the terature. ) The probem s soved usng a decomposton mode that dvdes the probem nto two sub-probems. The decomposton mode aows optma assgnment of ceto-tal dynamcay nstead of havng t known a pror n a statc fashon as n [1]. 3) A new heurstc agorthm dffers from the one proposed n [1] and consttutes of two sub-probems n the same manner as the decomposton mode. The agorthm dynamcay dversfes the TALs among the ces whch heps n reducng the TAU sgnang oad. The remander of ths paper s organzed as foows. Secton II revews studes on mobty management technques. Secton III presents a detaed descrpton of the system mode and formuates the probem for both the centrazed and the dstrbuted schemes. Secton IV descrbes the heurstc agorthm deveoped to sove the optmzaton probem. Secton V dscusses the smuaton parameters and resuts. Fnay, Secton VI concudes the paper. II. RELATED WORK A consderabe amount of research effort has been devoted to the study of ocaton management n varous technooges, specfcay n terms of ocaton management, whch has not changed much n recent years. In ths context, we essentay hghght two major sources of sgnang that are drecty reated to ocaton management: TAU and pagng. Numerous studes have dscussed the ssue of sgnang burst caused by TAU and pagng. Most of these approaches have nvestgated the sgnang overhead when confgurng LA or TA havng the same propertes. Few researchers have addressed the sgnang overhead from the perspectve of TAL constructon, even though TAL has greater mportance and fexbty than LA or TA. In fact, most studes have faed to provde a rgd framework that can provde technca support and s appcabe to rea-word scenaros. As an exampe of prevous proposas for oder technooges, an overappng ocaton area mechansm was proposed n [6] for GSM technoogy. The purpose of the study was to mnmze the sgnang oad resutng from the png pong effect. The study ntroduced four seecton poces for determnng the ocaton area (LA) percentage. Smar reated studes have been conducted [7] [9]. Furthermore, TAU and pagng technques have been nvestgated for the purpose of reducng the sgnang oad. Severa studes [10] [1] have dscussed a number of methods, such as tmer-, veocty-, and movement-based ones. However, the aforementoned methods are not commony used n the current TA/TAL approach. Few studes have dscussed the sgnang oad n terms of steerng the oad through the contro-pane eements, such as MME or servce gateway (S-GW). In one study [13], a concept that enabes two modes of MME dstrbuton, namey centrazed and dstrbuted MME, has been ntroduced. The study proposed anayss of both archtectures n terms of the sgnang oad resutng from user mobty. The authors aso presented a comparson between mutcast and uncast pagng. Whe the study reported some fndngs usng dstrbuted and centrazed MME archtectures, t dd not dscuss the fundamentas of TA/TAL constructon that aow further exporaton of both archtectures. Furthermore, the authors dd not ustrate how the MMEs are aocated and mpemented n ther mode. Another study [14] has proposed a mode that supports hgh-mobty users by reocatng data pane gateways on the bass of the UEs mobty pattern, thereby mnmzng the reocaton frequency. In [15], the authors defned the concept of S-GW servce area, where a poo of S-GWs serves TAs or ces. The objectve
3 s to emnate frequent dsconnectons that occur when the UE moves to a dfferent ce or TA. The concept focuses specfcay on actve users whose Quaty of Servce (QoS) can be degraded sgnfcanty durng S-GW re-aocaton. A number of studes have proposed dynamc TA/TAL technques that can optmze the sgnang overhead perodcay [16] [18]. In [16], the authors expored the advantage of dynamc TAL confguraton by ntroducng a rue of thumb. The approach attempts to mnmze the sgnang overhead on the bass of the TAU and pagng correaton. It succeeded n reducng the overhead to a greater extent than conventona TA. In the proposed mode, dynamc confguraton s acheved after a fxed perod, whch does not yed accurate resuts n envronments havng UEs movng at dfferent speeds. In [17], the authors ntroduced an adaptve mode that constructs a sutabe TAL for each user n a set of ces represented as an ndvdua TA. The adaptve confguraton s trggered perodcay on the bass of a defned movement threshod. Ths mode does not ceary ndcate the feasbty of the souton when deang wth a arge-scae scenaro. Moreover, the authors dd not provde a detaed descrpton of the physca parameters of the system, such as the ce radus and the users veocty. In [18], the authors presented a dynamc mode that aows TAs to be confgured as user mobty patterns change. The agorthm tends to use the graph correaton coeffcent method, whch measures the smartes n user mobty behavor wthn a certan perod and trggers new TA constructon. The approach s expensve owng to the hgh ce-to-ta reconfguraton cost. Other studes have depoyed TA/TAL confguraton as an SON use case, e.g., [5]. The system mpements procedures and protocos based on the UE behavor pattern. Consequenty, the TA/TAL s confgured perodcay to mnmze the sgnang burst. The study proposed adaptng the TAL confguraton dependng on the user mobty pattern usng a set of femtoce mesh networks. Each TAL can be confgured dynamcay on the bass of the UE s mobty status. Despte the novety of the agorthm, the mode does not specfy some major desgn aspects, such as the advantage of overappng TALs n the ces, whch can contrbute sgnfcanty toward mnmzng the sgnang oad. TAL has been expored and modeed n [3,19] []. In [19], the authors proposed a mode for TAL that can be apped to a arge-scae scenaro, and they showed the advantages of TAL over the conventona TA concept. Thus, they tred to provde an abstracton of the TAL desgn and ts benefts. In [0], the same authors proposed an optmzaton mode based on overappng TALs, whch woud aow a ce to dstrbute dfferent TAL portons to a set of users resdng n that ce n order to aevate the TAU sgnang oad whst restrctng the pagng oad wthn a defned mt. In [1], the authors formuated a near programmng (LP) mode based on overappng TALs n arge-scae scenaros and compared t wth the conventona TA technque. Mobty management n LTE was aso nvestgated n []. Pagng and TAU were anayzed n terms of sgnang. Three sequenta pagng schemes were presented, namey ce-tal, TA-TAL, and ce-ta-tal, whereby the MME frst requests the ce to page the UE, and f the ce fas to aocate the UE, the MME sends another request to a ces resdng n the TA or TAL. Ths can aso be done n a dfferent order. The authors concuded that the resuts vary wth the number of users, pagng schemes, and pong cyces. Thus, the seecton of potenta pagng schemes w depend on severa factors, such as mnmzng the sgnang cost of pagng and TAU or reducng the number of pong cyces. Lasty, the performance of TAL was modeed and anayzed n [3]. The study nvestgated TAL modeng from severa aspects, e.g., mted number of TAs n the TAL, and ntroduced two ca-handng modes that can provde a more reastc vew of the system. The tota sgnang cost was cacuated and the optma TAL for UEs was determned. However, the authors dd not verfy the vadty of the mode n arge-scae scenaros. Moreover, the mode dd not emphasze the desgn of overappng TALs or the dstrbuton of TALs wthn the system. A. Probem Defnton III. SYSTEM MODEL The objectve of the probem s to mnmze the sgnang overhead due to TAU and pagng n the core network. Furthermore, our mode seeks to acheve oad baancng through dfferent MMEs. The same concept presented n our prevous study [1] w be apped aong wth an SON dynamc agorthm; the atter enabes perodc Ce-to-TAL/MME confguraton. The mode conssts of two poong schemes, namey centrazed and dstrbuted schemes. Tabe I summarzes the mportant notatons used n the system mode. B. Premnares In order to desgn a vad mode for mnmzng the sgnang oad, we frst show that pagng and TAU have a tght correaton that can be used for constructng a TAL. A TAL conssts of a number of TAs, each of whch can be represented as one ce. The sze of each TAL has a drect nfuence n terms of sgnang. In other words, as the number of ces accommodated nsde a TAL ncreases, the pagng sgnang oad ncreases and the TAU sgnang oad decreases, and vce versa. Ths s a resut of the TAU and pagng mechansms. A TAU sgna s trggered to update the users ocaton wthn the MME. Ths s done whenever a user moves to a ce that s n a dfferent TAL. A pagng sgna s trggered to ocate the user by messagng a the ces of the ast TAL to whch the user was regstered. As the number of ces wthn a st ncreases, fewer TAU are requred because the probabty that the user changes the st has decreased. However, pagng woud ncrease because the MME woud need to send messages to a arger group of ces n order to accuratey determne the ocaton of the user. Thus, the sgnang overheads due to TAU and pagng are nversey proportona. Ths can be carfed n the foowng equatons 1 and. Let us assume that X j s the ce-to-st assgnment, where ces and j resde n st. Further, the constant costs of pagng and TAU are denoted by C ρ and C u, respectvey. We can now express the nta cost functons of pagng and TAU caused by a UE as foows:
4 TABLE I TABLE OF NOTATIONS C T u () : Tota sgnang cost of TAU n ce. C u : TAU sgnang cost of UE movng from one ce to another ce j that s not wthn the same st. UE : Tota number of UEs served by ce. ρ : Pagng arrva rate. H j : Probabty that a user moves from ce to ce j. C ρ : Pagng cost of partcuar user equpment n ce. C T ρ () : Tota pagng cost n ce. Cuρ() : Tota pagng and TAU overhead n ce. L : Tota number of sts. N : Tota number of ces. : Indvdua st (ndex). κ : Maxmum number of TAs that are assgned to st. ω : Cost of MME reocaton durng handover. HX () : { Inter-st handover rate of users n ce. 1, f st beongs to MME M, O M = 0, otherwse Y T A = { 1, f ce beongs to T A, 0, otherwse Decson Varabes σ : { Usage rato of each st n ce. 1, f ces and j beong to, Xj = 0, otherwse gven by C u σ(1 Xj) () Equaton () s to cacuate the trackng area update cost generated by a specfc user movng from ce to ce j. Let us assume that a user moves from ce to ce 3 that s not wthn the same st (e.g, ce beongs to st 1 and ce 3 beongs to st ). The TAU cost woud be mutped by one for each two ces that are not wthn the same st. Note that the same reasonng gven for the pagng cost s used here for the vaue of σ and hence, σ 1 s set to 1 whereas σ s set to 0 because ce beongs to st 1 and not st. Therefore, the user woud ncur a sgnang overhead equa to the TAU cost C u. C. Desgn Hypothess As stated prevousy, the mode conssts of two MME poong schemes. The centrazed MME poong scheme aocates each TAL to an ndvdua MME eement. On the other hand, the dstrbuted MME poong scheme aocates each TA or ce to an ndvdua MME eement. Note that n our archtecture, a TA aways contans ony one ce; ths smpfes the probem and avods the ce-to-ta assgnment constrant. Further, n ths context, a ce or TA can be used nterchangeaby and w have the same constrant. Fgures 1 and show the centrazed and dstrbuted MME poong schemes, respectvey. The mode can have two eves of MME MME The cost functon of pagng a specfc UE wthn a certan ce s gven by ( ) C ρ σ + σjx j (1) j=1,j In order to carfy the equaton above (1), et us assume that ce 1 and are n the same st 1, the vaue of X 1 1 s set to 1, a other varabes are ether set to one or zero dependng on whether or not they beong to the same st. Equaton (1) cacuates the cost functon of pagng a specfc user beongng to partcuar ce whch s ce 1 n ths exampe. Thus, varabe X 1 1 woud be mutped by the usage rato of st 1 n ce 1 (denoted as σ 1 1) as we as a constant vaue of the pagng cost C ρ. The vaue of σ 1 1 s 1 snce ce 1 ony beongs to one st whch s the ony st provded to the ce. Smary, σ 1 s aso one snce ce beongs to one st ony. Therefore, the core network woud page every ce beongng to st 1 that the ntended user s wthn (ces 1 and n ths exampe). The cost functon of TAU generated by a specfc user crossng two ces that are not wthn the same TALs s TA/Ce TA/Ce TA/Ce TA/Ce TA/Ce TAL 1 TAL Fg. 1. Centrazed MME poong scheme TA/Ce confguraton: ce/ta-to-tal/mme assgnment, whch can be handed perodcay by the core network as ustrated n the frst sub-probem, and TAL/MME-to-UE assgnment, whch w be handed through each ce as expaned n the second sub-probem. Frst, n terms of ce-to-tal/mme assgnment, the system consders the mobty pattern of the UEs wthn the ces and dynamcay assgns or re-assgns the ces nsde the TALs n order to mnmze the TAU sgnang overhead. Second, n terms of TAL-to-UE assgnment, each node w assgn a number of TALs to a porton of UEs resdng wthn that ce wth a defned usage rato, denoted by σ. The decson varabe σ determnes the usage rato of each TAL wthn each ce. Ths resuts n overappng TAL assgnment to ces. Furthermore, ths technque can aso
5 MME MME MME MME MME MME TA/Ce TA/Ce TA/Ce TA/Ce TA/Ce TA/Ce TAL 1 TAL Fg.. Dstrbuted MME poong scheme words, the proposed technque seects a ce to be ncuded or emnated from a poo of ces that s not restrcted drecty to the surroundng neghbors of the current ce. Ths can sgnfcanty mnmze the sgnang oad caused by frequent TAL updates and ead to the ncuson of a greater number of potenta ces that are ocated outsde the rng and have a hgher probabty of beng vsted by the UEs. Thus, t s an effcent scheme whereby ces can be chosen seectvey rather than n terms of group of rngs. Fgure 3 shows the dfference between the proposed smart seecton method and the rng seecton method. We note that the smart technque aows dfferent seecton shapes, such as the one depcted n orange. On the other hand, the rng seecton technque, shown n green, can ony ncude a the surroundng ces, regardess of ther probabty of beng vsted. provde oad baancng through the MMEs n the centrazed poong scheme, because each MME represents an ndvdua TAL. 1) Ce/TA-to-TAL/MME Assgnment: We attempt to determne the optma ce/ta-to-tal/mme assgnment n the centrazed poong scheme, whe the dstrbuted scheme consders ce-to-tal assgnment ony. In the centrazed scheme, each TAL can be aocated to each MME on a one-to-one bass. Aternatvey, n the dstrbuted scheme, each ce s represented as a TA that s assgned to an ndvdua MME. LTE has defned the maxmum number of TAs that can be aocated to a TAL to be 16. As a basc defnton for the assgnment constrant, we can use the bnary scaar decson varabe Xj to determne whether ces and j resde n the same trackng area st. For exampe, f ce 1 and are n the same st 1, the vaue of X1 1 s set to 1, otherwse X1 1 s set to 0. The same s apped to a the ces and sts n the system. The constant κ s the maxmum number of TAs n the TAL as descrbed n the foowng equaton: Xj κ, L (3) j,j The purpose of the above-mentoned constrant s to aocate a maxmum of 16 ces/tas to each TAL. However, to emnate the redundancy of ce/ta combnatons n each TAL, another constrant shoud be consdered as foows: X j = X j, L,, j, j N (4) In ths study, a dynamc agorthm s proposed to perodcay update the ce-to-tal confguraton. A ce can have a number of neghborng ces that are ntay arranged as rngs. Numerous studes, e.g., [3],[5],[3] have consdered perodcay expandng or emnatng the ces arranged as rngs. In fact, the rng-based assumpton s not entrey effcent n terms of mnmzng the sgnang oad. Ths s because the concept of rngs mght nvove ces that have a ow probabty of beng vsted owng to ther mobty behavor. Conversey, our technque nvoves anayzng the average mobty behavor of the UEs over a certan perod and expandng or emnatng specfc ces not strcty ncuded n the entre rng. In other Fg. 3. Rng neghbor seecton versus smart ce seecton ) TAL-to-UE: Each TAL s assgned to UEs va ther servng ces. TALs are dstrbuted among the users based on the number of sts that the ce beongs to. Ths s because f a ce s assgned to mutpe sts, the UEs need to have access to these mutpe sts. D. Probem Formuaton The orgna mode that was proposed n [1] s used as a base mode for both the centrazed and the dstrbuted poong schemes. The objectve of the mode s to mnmze the sgnang overhead resutng from both TAU and pagng. The mode s modfed and dvded nto a b-objectve mnmzaton probem as foows: Objectve functon Cost functons mn α Cuρ() + β HX () Cuρ() = C T ρ () + C T u (), N HX () = UE.H,j.(1 Xj) =1 Centrazed poong-tau cost [ L ] C T u () = UE.H j.c u ω. O M. σ(1 Xj),, j, j N =1 (5a) (5b) (5c) (5d)
6 Dstrbuted poong-tau cost Pagng cost Constrants C T u () = UE.H j.c u [ L σ L (1 X j) ] ω. Y T A +,, j, j N [ L C T ρ () = ρ.c ρ UE.σ+ =1 j, j =1 ] UE j. Xj.σj, N (5e) (5f) σ = 1, N (5g) =1 j, j X j κ, L (5h) X j = X j, L,, j, j N (5) 0 σ 1 (5j) O M 0, 1 (5k) X j 0, 1 (5) The objectve functon (5a) s dvded nto two parts. The frst part mnmzes the sgnang overhead caused by TAU and pagng and the second part mnmzes the nter-st handover. Both parts are weghted wth factors represented as α and β. These weghts can be used by the servce provder as contro parameters to prortze an objectve over the other. The frst part of (5a) contans two cost functons, as shown n (5b), whch combnes the cost functon of TAU n the centrazed scheme, as n (5d), and that n the dstrbuted scheme, as n (5e), aong wth the second term, whch has the pagng cost functon, as n (5f). In the centrazed scheme, the sgnang oad s determned by frst cacuatng the number of UEs that resde n a ce and have a probabty of movng to another ce that s not wthn the same st, and then mutpyng that number by the cost of nter-mme reaocaton (ω) and the usage rato of each st (σ ). In the dstrbuted scheme, the cost of MME reaocaton s mutped by the number of UEs that trave from one trackng area to another. The pagng cost functon (5f) cacuates the sgnang oad of pagng messages that are trggered by the UEs nsde the sts. The cost consders the percentage of the overappng sts used n ce mutped by the decson varabe Xj that determnes whether ce and the neghborng ce j beong to the same st. The second part of (5a) consders estmatng the nter-st handover rate of the average number of UEs n a ce, as gven n (5c). Constrant (5g) ensures far usage by a set of ces for every st/mme. Constrants (5j), (5k), and (5) are the boundary constrants. E. Decomposton Mode The prevous mode s a mxed-nteger nonnear programmng (MINLP) probem wth quadratc equaty constrants. Ths probem s a we-known NP-hard probem that s dffcut to sove to optmaty [4]. In our prevous study, the bnary decson varabe was defned n the mode and ce-to- TAL/MME assgnment was done statcay. In ths context, we want to have dynamc ce-to-tal/mme assgnment based on the UE mobty patterns. Therefore, we propose a decomposton mode that conssts of two sub-probems, each of whch s defned as foows: Sub-probem 1. In a gven set of ces that serve a number of UEs, fnd the optmum ce-to-tal/mme assgnment aocated perodcay to mnmze the nter-st handover rate of UEs traveng from a ce to another ce that s not wthn the same st TAL. Sub-probem. In a gven set of TALs/MMEs that are overappng and used by a number of ces, fnd the optmum usage usage rato of each TAL/MME to be gven to a number of UEs that are ocated nsde a ce such that the sgnang overhead s mnmzed. 1) Sub-probem 1 Formuaton: HX () = mn HX () UE.H,j.(1 Xj),, j, j N =1 Xj 1, L (5h) and (5) (6a) (6b) (6c) (6d) (H,j Lmt).X j 0,, j, j N, L (6e) The objectve functon (6a) tres to create a sutabe Xj that tends to mnmze the occurrence of UEs movng between ces that are not wthn the same TAL. The cost functon (6b) cacuates the cost of the handover rate of UEs crossng ces beongng to dfferent TALs. Constrant (6c) assgns at east two ces to every st. Constrant (6d) s used to satsfy the condtons of Xj formuaton and mt the number of TAs nsde each TAL. Fnay, constrant (6e) prortzes the ces havng a hgher probabty of beng vsted for ncuson n the same st, as ndcated by the UE mobty patterns. Ths constrant s cabrated and modfed on the bass of the UE average speeds. ) Sub-probem Formuaton: mn Cuρ() Cuρ() = C T ρ () + C T u (), N C T u () = (5d) or (5e) based on poong scheme (7a) (7b) (7c)
7 [ L C T ρ () = ρ.c ρ UE.σ+ =1 j, j =1 ] UE j. Xj.σj, N (7d) σ = 1, N (7e) =1 0 σ 1 (7f) X j 0, 1 (7g) The descrpton of the objectve functon and sgnfcance of the constrants have been addressed n the prevous subsecton. F. Mobty Pattern Mode We use the fud fow mode to smuate the mobty behavor of the users n the system. The fud fow mode s a we-known mode that s commony used n the terature. The mode depcts the traffc fow rates of UEs movng out of a cosed regon represented as a ce or base staton. For a certan ce wth permeter L, UE densty U, and average UE veocty v, the average number of ce crossngs per unt tme s cacuated as foows: U.L.v (8) π In ths context, ces are hexagona n shape wth sde ength R; hence, L s repaced wth 6R. G. SON Capabty through MME Sef-Organzng Network s a paradgm that seeks to mnmze the operaton expenses reated to network re-organzaton n order to acheve hgher effcency and Quaty of Experence (QoE) [5]. However, t s vta to dever an updated verson of TAL due to the rapd changes n the mobty patterns of the UEs. The MME can be nvoved n trggerng the sef-optmzng capabty through the network by sendng the updated TALs to the base statons. The purpose of dynamc ce-to-tal/mme reassgnment s to aevate the sgnang overhead that resuts specfcay from the movement of UEs from one ce to another ce that s not n the same TAL/MME. The dynamc agorthm w set a tmer to trgger the desred changes n the TALs accuratey and to dstrbute them among the ces. Two vta factors tend to have a major mpact on the agorthm: the average handover rate of the UEs between the ces, and the gven vaue of LIMIT that prortzes the combnatons of ces to be aocated wthn the same st or MME. The parameter LIMIT can be chosen by the network operator as a contro parameter to prortze ces over others n order to reduce the requred TAU. Moreover, the vaue of LIMIT s reated to the average handover rate between ces or the average UE veocty. Equaton (6e) shows the reaton between LIMIT and the average handover rate. We can see that as the vaue of LIMIT ncreases, a greater number of ces havng a hgher handover rate between each other are ncuded n the same st. In our study, we assumed that LIMIT can have dfferent vaues reated to the UE average veocty. A detaed descrpton of the dynamc agorthm s provded n the next subsecton. H. Dynamc TAL Agorthm The proposed dynamc agorthm many focuses on Subprobem 1, whch pays the roe of aocatng sutabe combnatons of ces n each TAL/MME. Ths tends to mnmze the handover rate between ces that are not n the same TAL/MME. As mentoned earer, a tmer w be used to trgger the frequent update of ce-to-tal/mme assgnment. Furthermore, the TAL/MME s updated wthn a seectve set of ces that are determned by the vaue of LIMIT, whch vares perodcay correspondng to the UE mobty patterns. The tmer s adjusted by R, whch determnes the average v tme requred for the UEs to cross a certan ce. Pseudocode depcted n Agorthm 1 expans the mechansm of the proposed SON dynamc agorthm n the mode. In ne 1, the agorthm ntazes the nput vaues, whch ncude the number of TALs/MMEs used n the system and the number of TAs, whch s equvaent to the number of ces. Further, both the average veocty rate and the average number of UEs n each ce are determned. Fnay, the ce radus and the nta ce-to-st assgnment are gven. The output of ths agorthm s the optmzed ce-to-tal/mme assgnment. In nes 4-6, at a gven tme, the agorthm cacuates the average crossng rate of UEs resdng n each ce n order to determne the handover rate H j between the ces. Then, the LIMIT vaue s cabrated to seectvey choose the neghborng ces of any ce that have the hghest handover rate and aocate those neghborng ces to the same st or MME. In nes 7-8, the nta assgnment s modfed by ether emnatng or retanng the od ces that had prevousy been aocated to the sts. Ths s acheved by sovng Sub-probem 1. In nes 9-15, a tmer s set to estmate the average tme needed for a UE to eave the ces. The tmer determnes the optma frequency wth whch the mode shoud reassgn the ces to the sts. Moreover, the system randomy changes the vaues of the UEs n every event pror to the trgger tme and keeps cacuatng the sgnang cost by sovng Sub-probem. A. Agorthm Descrpton IV. HEURISTIC ALGORITHM Ths secton descrbes the deveopment of a heurstc agorthm that depends on equa dstrbuton of the TAL/MME oad among the reevant ces. Ths approach has the advantage of mnmzng the TAUs among the ces n a ess compex manner. In ths context, the current heurstc agorthm dffers from the precedng one presented n [1]; t foows the same concept as the optma formuaton, whch dvdes the orgna agorthm nto two sub-probems. The frst sub-probem provdes a new method for assgnng ces to TALs/MMEs by seectng the ces havng the hghest vaues of UE crossng rates such that the TAU frequency s mnmzed. The ce-tost assgnment w aso be performed dynamcay by defnng the tmer n the manner descrbed above for the dynamc
8 Agorthm 1 SON Dynamc Agorthm 1: Input: = {1,,..., L}: A sts = MMEs T A = {1,,..., N}: A trackng areas UE: Average number of UEs n ce v Range = [v 0 : v max ] v = Random such that v v Range U: Densty of UEs n ce R: Ce dameter UE : Inta number of UEs n ce T A. Xj : Inta random ce-to-tal/mme assgnment : Output: Xj : Optmzed ce-to-tal/mme assgnment 3: for t T otat me do 4: defne UEr j = U.L.v Π 5: defne Hj = UEj r UE 6: update LIM IT 7: Cacuate Sub-probem 1: 8: update Xj 9: for t (R/v) do 10: Cacuate Sub-probem : 11: for N do 1: update UE = Rand(UE) 13: end for 14: end for 15: end for : Average crossng from to j agorthm (Agorthm 1). Pseudo-code shown n Agorthm descrbes Sub-probems 1 and of the heurstc agorthm n deta. In ne, the essenta parameters requred n the agorthm are defned. The output of Sub-probem 1 s the heurstc ce-to-tal/mme assgnment. In nes 4-14, the agorthm cacuates the UE crossng rates based on ther average speeds and then sets the vaue of Xj that corresponds to the maxmum crossng rates between the ces. Lne 16 defnes the nput of Sub-probem. The output of Sub-probem s the usage rato of the sts n each ce. In nes 18-30, the usage percentage of each st to whch a ce beongs s determned by performng a search of a the sts that serve the ces. The rato s cacuated by dstrbutng the usage of the TALs/MMEs that serve the ces wth equa percentages. Ths w accompsh a oad baance on the MME sde that cannot be compromsed by any gven ce. Fnay, n nes 31-37, the sgnang overhead of both schemes s cacuated usng the approprate equatons. B. Agorthm Compexty To justfy the need for the heurstc agorthm, we must frst dscuss the compexty of the decomposton agorthm. The orgna probem presented n (5) s ntractabe snce the mode s a mxed-nteger nonnear programmng (MINLP) probem wth quadratc equaty constrants, whch s a weknown NP-hard probem that s dffcut to sove to optmaty [4]. Therefore, the decomposton mode has been adopted to sove t. Sub-probem 1 has O(L N κ ) compexty because each st can have at most κ ces among N avaabe ces wth possbe overap. Sub-probem has O(LN ) compexty because the usage rato of each st n a the ces s cacuated Agorthm Heurstc Agorthm 1: Sup-Probem 1 : Input: = {1,,..., L}: A sts = MMEs T A = {1,,..., N}: A trackng areas UE: Average number of UEs n ce v Range = [v 0 : v max ] v = Random such that v v Range U: Densty of UEs n ce R: Ce dameter UE : Inta number of UEs n ce T A. Xj : Inta random ceto-tal/mme assgnment 3: Output: Xj : Optmzed ce-to-tal/mme assgnment 4: defne UEr = U.L.v Π : Average crossng rate from to j 5: defne Hj = UEj r UE 6: for N do 7: for L do 8: f UEr j = MAX then 9: SET Xj = 1 10: ese 11: contnue 1: end f 13: end for 14: end for 15: Sup-Probem 16: Input: = {1,,.., L}: A sts T A = {1,,..., N}: A trackng areas : Ce-to-st bnary ndcator X j 17: Output: σ : Usage rato of st n ce 18: for N do 19: defne N L = φ 0: for L do 1: f then : update N L = {N L N} 3: ese 4: contnue 5: end f 6: for N N L do 7: σ = 1 N L 8: end for 9: end for 30: end for 31: Cacuate: C ρ () as per (7d) 3: f Centrazed Scheme then 33: Cacuate: C T u () as per (5d) 34: ese 35: Cacuate: C T u () as per (5e) 36: end f 37: Cacuate: Cuρ() = N (C ρ () + C T u ())
9 for each ce. Hence, the tota compexty of the decomposton agorthm s O(L N κ )+O(LN ) = O(L N κ ). On the other hand, the heurstc agorthm has much ower compexty. Usng the proposed heurstc agorthm, Sub-probem 1 has O(LN) compexty because a decson as to whether each ce beongs to the st s taken n each teraton. Sub-probem aso has O(LN) compexty because the usage rato of each st wthn each ce s cacuated. Therefore, the tota compexty of the heurstc agorthm s O(LN), whch s much ower than that of the decomposton agorthm. V. PERFORMANCE EVALUATION To evauate the performance of the proposed modes, we performed MATLAB smuatons at four dfferent speeds, categorzed as very sow (0 8 m/s), sow (8 16 m/s), norma (16 5 m/s), and fast (5 33 m/s). The smuaton assumed an envronment of 10 ces, each havng a dfferent number of UEs. The number of UEs was dstrbuted unformy among the ces wth an average of 100 users. Tabe II shows the smuaton parameters and ther vaues. Parameter TABLE II SIMULATION PARAMETERS & VALUES Vaue Number of ces N 10 Number of TAL L 3 Average number of users 100 per ce Number of TAs 10 Pagng rate ρ 0.05 UE speeds 0,8,16,5 and 33 m/s Ce Radus 500 m TAU to Pagng cost 10:1 The decomposton and heurstc modes were mpemented and compared wth each other. We note that each sub-probem has the same mportance; thus, the weght factors α and β were assgned vaues of 1. The smuaton tested the modes statcay and dynamcay n the case of the centrazed and dstrbuted schemes. The dynamc agorthm presented n Agorthm (1) was mpemented n the decomposton and heurstc agorthms, whereas the same anaogy provded n our prevous study [1] was used for the statc agorthm. Moreover, a random agorthm was added n the evauaton for the purpose of provdng a smper approach to repace Sub-probem 1, as t generates random ce-to-tal/mme aocaton dynamcay. Moreover, the random technque enabes us to evauate the effectveness of frequent random aocaton wthout pror knowedge of the sgnang overhead. Ths approach offers a ess compex souton that does not affect the core network. The tota sgnang overhead of the TAU and pagng combnaton was empoyed as a performance metrc for the proposed agorthms. We used MATLAB s but-n functons ntnprog and nprog to optmay sove Sub-probems 1 and, respectvey. A. Tota Sgnang Cost Fgure 4 shows the average tota sgnang overhead caused by pagng and TAU at dfferent speeds for the decompostonbased agorthms, namey the dynamc, random, and statc Average Tota Cost - Centrazed 1.9 1.8 1.7 1.6 1.5 1.4 1.3 1. 1.1 10 4 Dynamc Random Statc 1 5 10 15 0 5 30 35 Fg. 4. Tota sgnang overhead cost for the centrazed scheme agorthms, n the case of the centrazed scheme. We note that the SON dynamc agorthm outperforms the statc and random agorthms. On the other hand, the random optma agorthm shows better performance n terms of the sgnang overhead. Ths s because the random agorthm dynamcay changes the TAL n a random manner, whereas the statc agorthm mantans the same TAL at a tmes. We aso note that the UE speed has a sgnfcant mpact as t exceeds 5 m/s. Fgure 5 depcts the average tota sgnang overhead for the centrazed scheme when dfferent pagng rates are consdered at dfferent speeds. It s noted that the same trend s observed as that shown n Fgure 4 when changng the average speeds of the users. The second observaton s that the tota sgnang overhead ncreases as the pagng rate ncreases. Ths s expected snce the pagng cost s dependent on the pagng rate as ustrated n equaton 5f. Fgure 6 Average Tota Cost-Centrazed 14 1 10 8 6 4 0 0-8 m/s 8-16 m/s 16-5 m/s 5-33 m/s 0.05 0.5 0.5 Pagng Rate Fg. 5. Tota sgnang overhead cost for dfferent pagng rates shows the average tota sgnang overhead at dfferent speeds
10 Average Tota Sgnang Cost- Dstrbuted 3.8.6.4. 1.8 10 5 Dynamc Random Statc Average Centrazed Tota Cost 10 4 3 Decomposton Heurstc.5 1.5 1 1.6 5 10 15 0 5 30 35 Fg. 6. Tota sgnang overhead cost for the dstrbuted scheme 0.5 5 10 15 0 5 30 35 Fg. 7. Tota sgnang overhead cost n the centrazed scheme for decomposton agorthm vs. heurstc agorthm for the decomposton-based agorthms n the case of the dstrbuted scheme. The resuts show the same trends as those of the centrazed scheme. Furthermore, we can concude that the centrazed scheme outperforms the dstrbuted scheme because of the frequent MME reocaton n the atter case. A comparson between the decomposton and heurstc agorthms s shown n Fgures 7 and 8. Fgure 7 shows the comparson n the case of centrazed scheme, where t s observed that the heurstc approach shows acceptabe behavor at ower speeds compared to the decomposton mode. Ths s because the decson varabe sgma dversfes the TAL/MME poo, thereby mnmzng the TAU sgnang and the overa sgnang cost. Fgure 8 shows a comparson between the decomposton and heurstc methods n the case of dstrbuted scheme. The heurstc agorthm offers a near-optma souton compared to the optma decomposton mode. Ths s a resut of the domnant vaue of the MME reocaton weght, whch has a major effect on the sgnang n the case of the dstrbuton scheme. Tabe III summarzes the average percentage mprovement n the random and dynamc assgnment agorthms compared to the statc agorthm. In the centrazed case, the random agorthm shows an mprovement of 8% 16.45%, whe the dynamc agorthm shows an mprovement of 5.3% 34.98%. Smary, the random agorthm resuts n an mprovement of 6.53% 9.37% and the dynamc agorthm resuts n an mprovement of 15.74% 18.88% n the dstrbuted case. TABLE III AVERAGE IMPROVEMENT PERCENTAGE Parameter Random Dynamc Tota Sgnang Cost- Centrazed [+8%, +16.45%] [+5.3%, +34.98%] Tota Sgnang Cost- Dstrbuted [+6.53%, +9.37%] [+15.74%, +18.88%] Power Consumpton [+16.6%, +17.7%] [+8.83%, +39.3%] Average Tota Sgnang Cost- Dstrbuted 3.8.6.4. 1.8 10 5 Decomposton Heurstc 1.6 5 10 15 0 5 30 35 Fg. 8. Tota sgnang overhead cost n the dstrbuted scheme for decomposton agorthm vs. heurstc agorthm B. Power Effcency In ths subsecton, we dscuss the effcency of the proposed dynamc approach n terms of power consumpton. One way to measure the effcency of the SON dynamc mode s to evauate the UE battery fe. It s estmated that each TAU procedure consumes around 10 mw of a reguar smart-phone battery. The resuts n Fgures 9 and 10 depct the average tota battery consumpton for one UE n an hour. Fgure 9 compares the dynamc, random, and statc approaches. The fgure ndcates sgnfcant power savngs n the dynamc approach compared to the statc approach. In addton, the random dynamc approach shows sghty ower power consumpton than the statc approach. Fgure 10 compares the dynamc decomposton and heurstc soutons. It s cear that the decomposton souton performs better. However, the
11 Average Tota UE Power Consumpton (mw) 0 00 180 160 140 10 100 80 60 40 0 Dynamc Random Statc 5 10 15 0 5 30 35 Fg. 9. Average tota UE power consumpton (mw) heurstc souton offers near-optma resuts at sow speeds. Tabe III summarzes the power savngs n the random and dynamc assgnment agorthms compared to that n the statc agorthm. The power savngs range from 16.6% to 17.7% for the random agorthm and from 8.83% to 39.3% for the dynamc agorthm. Average Tota UE Power Consumpton (mw) 0 00 180 160 140 10 100 80 60 40 0 Decomposton Heurstc 5 10 15 0 5 30 35 Fg. 10. Average tota UE power consumpton (mw) for decomposton agorthm vs. heurstc agorthm C. Reated Work vs. Our Approach Ths subsecton compares the atest reated methods n the terature wth our approach. Three studes have been consdered n the comparson: [5],[3], and [3]. The authors proposed three dfferent approaches that dynamcay sove the probem of ce-to-ta/tal assgnment for the purpose of mnmzng the tota sgnang overhead due to TAU and pagng sgnas. Specfcay, two of the studes,.e., [5] and [3], have proposed a SON mechansm to enhance the ntegence of ther agorthms. However, we note that none of the aforementoned studes have ncuded MME reazaton n ther soutons, whch s an mportant factor n the sgnang overhead. Consequenty, both MME reazatons are adapted n the compared studes n order to ensure accurate and far comparson. A the compared studes use the rng-based approach for ce-to-tal assgnment. Fgures 11 and 1 show the tota sgnang overhead n the centrazed and dstrbuted schemes. The key dfference between our technque and other reated methods s that our agorthm uses a smart technque for choosng the canddate ces to be assgned n TAL, whereas the other methods use the conventona ce-to-tal assgnment [3],[5], [3]. The smart seecton technque outperforms the rng-based technques because the smart seecton assgnment of ces aevates the frequent TAU sgnang by decreasng the probabty of the UE movng from one ce to another ce that s not n the same trackng area st. Furthermore, the smart seecton technque ncudes a greater number of ces that are more key to be vsted by the UE; thus, t s very effcent, especay for UEs movng at hgh speeds. Another key metrc for the varaton n the resuts s the overappng TAL, whch ncreases the probabty of fewer TAU updates caused by the UE when t traves from one ce to another. Fnay, the method for trggerng the dynamc confguraton or SON technque s aso an mportant factor that affects mnmzaton of the sgnang overhead. The SON technque s used n [5], and t factates the transton between stages through tmers and an actvaton threshod for trggerng the dynamc confguraton. The authors statcay soved the probem of ce-to-tal assgnment unt the actvaton threshod was reached. Athough the actvaton threshod s not defned ceary, the technque used s too compcated to be mpemented n a rea-word scenaro, especay wth the rapd varaton n UE veocty. Another SON technque was proposed n [3], where the authors ntroduced an overappng TAL scheme soved statcay n a manner smar to [5] unt a threshod was reached. The man dfference n ths study s the ntroducton of the overappng TAL technque, whch offers the advantage of ower TAU sgnang overhead. However, no avaabe SON technque for sovng TAL assgnment has adapted the trggerng scheme based on the veocty varaton reated to the UEs. Our scheme contnuousy adapts the TAL assgnment based on the average UE veocty. Fgure 13 shows the power consumpton of the reated methods and our method. The comparson has been made at dfferent speeds. Our method acheves greater power savngs, whch vary from +3.7% to +46.9%. Tabe IV summarzes the gven comparson n terms of the average mprovement percentage between our method (SON Dynamc/Smart) and the reated methods. Our method outperforms the reated methods sgnfcanty n terms of the centrazed and dstrbuted schemes as we as n terms of power consumpton. For nstance, the SON Dynamc/Smart agorthm n the case of the centrazed scheme outperforms the method proposed n [3] by +3.87% to +8.8% whch
1 TABLE IV AVERAGE IMPROVEMENT PERCENTAGE COMPARED WITH RELATED METHODS SON Dynamc/Smart SON Dynamc/Smart Agorthm Used Overappng SON-Enabed Power Consumpton Agorthm Centrazed Agorthm Dstrbuted Ref [3] [+30.8%, +35.69%] [+17.51%, +18.49%] [+36.48%, +39.85%] Ref [3] [+3.87%, +8.8%] [+14.6%, +16.11%] [+8.73%, +3.66%] Ref [5] [+36.69%, +4.37%] [+0%, +1.7%] [+4.84%, +46.91%] Average Centrazed Tota Cost.4. 1.8 1.6 1.4 1. 1 10 4 Our agorthm Ref [3] Ref [3] Ref [5] 36.6% 30.8% 3.8% 4.3% 35.6% 8.8% 0.8 5 10 15 0 5 30 35 Fg. 11. Tota sgnang overhead cost comparson between our agorthm (centrazed) and reated methods Average Tota UE Power Consumpton (mw) 0 00 180 160 140 10 100 80 60 40 0 Our agorthm Ref [3] Ref [3] Ref [5] 39.9% 3.7% 46.9% 5 10 15 0 5 30 35 Fg. 13. Average tota UE power consumpton (mw) comparson between our agorthm and reated methods 3. 10 5 Our agorthm uses overappng and SON technques. The ower percentage s taken n the case of the owest speed whe the hgher percentage s taken n the case of the hghest speed, as shown n Fgure 11. Average Dstrbuted Tota Cost 3.8.6.4. 1.8 Ref [3] Ref [3] Ref [5] 0% 17.5% 14.6% 1.% 18.4% 16.1% 1.6 5 10 15 0 5 30 35 Fg. 1. Tota sgnang overhead cost comparson between our agorthm (dstrbuted) and reated methods VI. CONCLUSION SON, a new concept ntroduced n reease 8 of 3GPP for LTE networks, s a promsng paradgm that enabes sef-pannng, sef-managng, and sef-optmzng of networks. Therefore, SON has been wdey accepted across dfferent appcatons. A number of SON use cases have been proposed to overcome ncreasng operatona expenses. In ths context, we proposed a SON approach to aevate the sgnang overhead caused by TAU and pagng. We used SON as an enaber to perform dynamc ce-to-tal/mme reconfguraton. Our approach can be consdered as a SON use case that sgnfcanty mnmzes the sgnang overhead. Furthermore, two schemes used n the prevous work, namey the centrazed and dstrbuted MME poong schemes, were mpemented and nvestgated dynamcay. We used the we-known fud fow mode to smuate the movement of UEs wthn the system. The mode conssts of two sub-probems derved from our orgna NP-hard probem formuaton. The frst subprobem s a bnary nteger programmng probem, whereas the second s a near programmng probem. In addton,
13 a new smart seecton method was proposed to ntegenty seect the potenta ces n the TAL/MME. Our method was shown to outperform conventona rng seecton, whch s commony used n the terature. Fnay, a ess compex heurstc souton was proposed, whch s easy to mpement and gves a sub-optma resut. The resuts showed that the dynamc decomposton souton aso acheves greater power savng than prevous methods. ACKNOWLEDGEMENTS Ths work was funded n part by Roya Commsson for Juba and Yanbu through the Saud Cutura bureau n Canada. Ths support s greaty apprecated. REFERENCES [1] E. Aqee, A. Moubayed, and A. Sham, Towards ntegent LTE mobty management through MME poong, n IEEE 58th Goba Communcatons Conference (GLOBECOM 15), Dec. 015. [] Sgnang s growng 50% faster than data traffc, Noka Semens Network, 01. [3] T. Deng, X. Wang, P. Fan, and K. L, Modeng and performance anayss of trackng area st-based ocaton management scheme n te networks, IEEE Transactons on Vehcuar Technoogy, vo. PP, no. 99, pp. 1 1, 015. [4] New son use-case: Trackng area optmzaton,, 3GPP TSG-RAN WG3 Std. R3-071 594, Aug 007. [5] J. Ferragut and J. Mangues-Bafauy, A sef-organzed trackng area st mechansm for arge-scae networks of femtoces, n 01 IEEE Internatona Conference on Communcatons (ICC 1), June 01, pp. 519 5134. [6] S.-R. Yang, Y.-C. Ln, and Y.-B. Ln, Performance of mobe teecommuncatons network wth overappng ocaton area confguraton, IEEE Transactons on Vehcuar Technoogy, vo. 57, no., pp. 185 19, March 008. [7] D. Gu and S. Rappaport Stephen, Mobe user regstraton n ceuar systems wth overappng ocaton areas, n 1999 IEEE 49th Vehcuar Technoogy Conference (VTC 99), vo. 1, Ju 1999, pp. 80 806 vo.1. [8] K.-H. Chang and N. Shenoy, A -d random-wak mobty mode for ocaton-management studes n wreess networks, IEEE Transactons on Vehcuar Technoogy, vo. 53, no., pp. 413 44, March 004. [9], Performance of an overapped macro-ce based ocaton area scheme, n IEEE Internatona Conference on Communcatons (ICC 03), vo. 1, May 003, pp. 475 481 vo.1. [10] A. Bar-Noy, I. Kesser, and M. Sd, Mobe users: to update or not to update? n 13th IEEE Networkng for Goba Communcatons (INFOCOM 94), Jun 1994, pp. 570 576 vo.. [11] G. Wan and E. Ln, A dynamc pagng scheme for wreess communcaton systems, n 3rd Annua ACM/IEEE Internatona Conference on Mobe Computng and Networkng, ser. MobCom 97. New York, NY, USA: ACM, 1997, pp. 195 03. [Onne]. Avaabe: http://do.acm.org/10.1145/6116.6147 [1] S. Tabbane, An aternatve strategy for ocaton trackng, IEEE Journa on Seected Areas n Communcatons, vo. 13, no. 5, pp. 880 89, Jun 1995. [13] I. Wdjaja, P. Bosch, and H. La Roche, Comparson of MME sgnang oads for ong-term-evouton archtectures, n 009 IEEE 70th Vehcuar Technoogy Conference Fa (VTC 09-Fa), Sept 009, pp. 1 5. [14] T. Taeb, K. Samdans, and A. Ksentn, Supportng hghy mobe users n cost-effectve decentrazed mobe operator networks, IEEE Transactons on Vehcuar Technoogy, vo. 63, no. 7, pp. 3381 3396, Sept 014. [15] A. Kunz, T. Taeb, and S. Schmd, On mnmzng servng GW/MME reocatons n LTE, n IEEE 6th Internatona Wreess Communcatons and Mobe Computng Conference, ser. IWCMC 10. New York, NY, USA: ACM, 010, pp. 960 965. [Onne]. Avaabe: http://do.acm.org/10.1145/1815396.1815615 [16] S. Razav, D. Yuan, F. Gunnarsson, and J. Moe, Dynamc trackng area st confguraton and performance evauaton n LTE, n 010 IEEE GLOBECOM Workshops (GC 10 Wkshps), Dec 010, pp. 49 53. [17] Y. W. Chung, Adaptve desgn of trackng area st n LTE, n 011 Eghth Internatona Conference on Wreess and Optca Communcatons Networks (WOCN 11), May 011, pp. 1 5. [18] M. Tor, S. Luna-Ramrez, and V. We, Automatc repannng of trackng areas n ceuar networks, IEEE Transactons on Vehcuar Technoogy, vo. 6, no. 5, pp. 005 013, Jun 013. [19] S. Razav, D. Yuan, F. Gunnarsson, and J. Moe, Expotng trackng area st for mprovng sgnang overhead n LTE, n 010 IEEE 71st Vehcuar Technoogy Conference (VTC 10-Sprng), May 010, pp. 1 5. [0] S. Modarres Razav and D. Yuan, Mtgatng mobty sgnang congeston n LTE by overappng trackng area sts, n Proceedngs of the 14th ACM Internatona Conference on Modeng, Anayss and Smuaton of Wreess and Mobe Systems, ser. MSWM 11. New York, NY, USA: ACM, 011, pp. 85 9. [Onne]. Avaabe: http://do.acm.org/10.1145/068897.068947 [1] S. Razav and D. Yuan, Reducng sgnang overhead by overappng trackng area st n LTE, n 014 7th IFIP Wreess and Mobe Networkng Conference (WMNC 14), May 014, pp. 1 7. [] R.-H. Lou, Y.-B. Ln, and S.-C. Tsa, An nvestgaton on LTE mobty management, IEEE, vo. 1, no. 1, pp. 166 176, Jan 013. [3] S. Modarres Razav, D. Yuan, F. Gunnarsson, and J. Moe, On dynamc sgnang congeston mtgaton by overappng trackng area sts, J. Netw. Comput. App., vo. 53, no. C, pp. 164 17, Ju. 015. [Onne]. Avaabe: http://dx.do.org/10.1016/j.jnca.015.03.009 [4] P. Bonam, L. T. Beger, A. R. Conn, G. CornuéJos, I. E. Grossmann, C. D. Lard, J. Lee, A. Lod, F. Margot, N. Sawaya, and A. WäChter, An agorthmc framework for convex mxed nteger nonnear programs, Dscret. Optm., vo. 5, no., pp. 186 04, May 008. [Onne]. Avaabe: http://dx.do.org/10.1016/j.dsopt.006.10.011 [5] Sef-optmzng network, the benefts of son n te,, http://www.4gamercas.org/ 011. Emad Aqee receved hs B.Sc. and M.Sc. degrees from KAU Unversty. He worked wth the Roya Saud Ar Defense Forces from 007 to 01. Snce August 01, he has been empoyed at Yanbu Unversty Coege, Roya Commsson for Juba and Yanbu. He s currenty pursung hs Ph.D. n Eectrca and Computer Engneerng at the Unversty of Western Ontaro. Abdaah Moubayed receved hs B.E. degree n Eectrca Engneerng from the Lebanese Amercan Unversty, Berut, Lebanon, n 01 and hs M.Sc. degree n Eectrca Engneerng from Kng Abduah Unversty of Scence and Technoogy, Thuwa, Saud Araba, n 014. He s currenty pursung hs Ph.D. (started September 014) n Eectrca and Computer Engneerng at the Unversty of Western Ontaro, London, Ontaro, Canada. Hs research nterests ncude wreess communcaton, resource aocaton, and wreess resource vrtuazaton. Abdaah Sham receved hs B.E. degree n Eectrca and Computer Engneerng from the Lebanese Unversty n 1997 and hs Ph.D. degree n Eectrca Engneerng from the Graduate Schoo and Unversty Center, Cty Unversty of New York, n September 00. Further, n September 00, he joned the Department of Eectrca Engneerng at Lakehead Unversty, Thunder Bay, Ontaro, Canada, as an Assstant Professor. Snce Juy 004, he has been at Western Unversty, where he s currenty a Professor n the Department of Eectrca and Computer Engneerng. Hs current research nterests are n the areas of network optmzaton, coud computng, and wreess networks.