dynamic pci allocation on avoiding handover …home.cse.ust.hk/~tliay/paper/pci-2016.pdfresearch...

WIRELESS COMMUNICATIONS AND MOBILE COMPUTINGWirel. Commun. Mob. Comput. 2016; 16:1972–1986Published online 2 February 2016 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/wcm.2662

RESEARCH ARTICLE

Dynamic PCI allocation on avoiding handover confusionvia cell status prediction in LTE heterogeneous smallcell networksZhu Xiao1,2, Tong Li1*, Wei Ding3, Dong Wang1 and Jie Zhang3

1 College of Computer Science and Electronic Engineering, Hunan University, 410082 Changsha, China2 State Key Laboratory of Integrated Service Networks, Xidian University, Xi’an 710071, China3 Department of Electronic and Electrical Engineering, University of Sheffield, Sheffield S1 3JD, U.K.

ABSTRACT

In this paper, we investigate the inbound handover confusion in the two-tier macrocell-small cell networks with help of mo-bility prediction. Instead of studying the mobile user’s (MU) movement, we propose an analytical model for the activitystatus of small cells, which is to exploit the statistical property of inbound handover events that would happen in small cells.We design the cell status prediction algorithm to obtain the optimal prediction outcome of the next status for the small cells.On avoiding the handover confusion, we develop a dynamic allocation approach of physical cell identifier according to theprediction results. We design (i) the cell status prediction-based strategy, by which the dedicated PCIs will be assigned tothe small cells with busy activity status while the other small cells share the public PCIs, and (ii) an integrated strategy inorder to fully exploit the usage of PCI. We formulate the preference relation for small cells via reference signal receivedquality relation integrated with status prediction information using Bayesian average method. Simulation results reveal thatthe proposed algorithms yield higher accuracy than the conventional methods; in the meantime, handover confusions canbe reduced significantly during the inbound handover. Copyright © 2016 John Wiley & Sons, Ltd.

KEYWORDS

small cell; macrocell; inbound handover; mobility prediction; physical cell identifier; handover confusion

*Correspondence

Tong Li, College of Computer Science and Electronic Engineering, Hunan University, 410082 Changsha, China.E-mail: [email protected]

Received 11 February 2015; Revised 30 August 2015; Accepted 11 December 2015

1. INTRODUCTION

Mobile and wireless communications have undergonesustained and rapid growth in the first decade of the newcentury. Because of the developments of new standardssuch as Third Generation Partnership Project and the LongTerm Evolution (LTE), the cellular networks havesignificantly enhanced the wide-range network coverageand improved mobile broadband communications. In therecent, mobile data traffic has been approximately dou-bling every year, and this growth is continuing unabatedly[1]. The published forecast reports, by partners fromindustry such as Cisco and Qualcomn [2,3], predict anexponential increase of mobile and wireless data trafficthat would correspond to a 1000-fold increase in theforthcoming decade.

1972

In response to the sharp increase of mobile data demand,the current macrocell networks are required to be moredistributed, dense, and heterogeneous. The densification ofcellular networks is one of the recognized means of alleviat-ing this problem [4]. On the revolution to fifth generation(5G), heterogeneous networks (HetNets) and small cells arepromising techniques of increasing capacity and keepingpace with future mobile data demands [5].

In the HetNets, the macrocell is designed to provide a widerange of mobile coverage, which is overlaid by a diverse set oflow-power and low-cost small cells. By offloading traffic withfull reuse of frequency resources, small cells are able to pro-vide a cost-effective way to increase capacity and coveragebeyond the initial deployment of macrocell [6].

Because of the massive deployment of small cells,including picocells and femtocells, mobile users (MUs) move

Copyright © 2016 John Wiley & Sons, Ltd.

Dynamic PCI allocation on avoiding handover confusionZ. Xiao et al.

between close-by cells increases considerably; hence, cross-tier handover and mobility prediction play crucial roles to pro-vision seamless coverage and ensure service continuity [7,8].

1.1. Related work

There are two major remaining challenges whenimplementing inbound handover in two-tier HetNets [9].On one hand, mobility prediction together with the informa-tion of MU’s movement is better to be known in advance, sothat small cells can make proactive actions [10], that is, allo-cating resources or preventing unnecessary handovers at theright places and times. A variety of previous works havebeen proposed to address the mobility prediction issue, andmajority of them aim to predict the trajectory of the MUs[11,12]. In [11], the authors investigate spectral resource pro-visioning in the two-tier networks by designing a bandwidthcaching mechanism based on three types of mobility modelsfor MUs. The user movement probability distribution isstudied in [13] to describe the possibility that MU stays inthe femtocell/macrocell coverage or moves cross tiers, amovement prediction mechanism is designed to estimatethe likely path of an MU’s movement. It takes into accountof the variation of MU’s velocity and acceleration, wherethe assistance of Global Positioning System (GPS) or reason-ably accurate positioning capability is required to obtainMUs’ location. Consider that the probability of MU’s nextmove is related to the present location; mobility predictionapproaches are designed in [12,14] to predict the headingdirection of MU as well as estimate the target macro basestation (MBS) or small cell base station (SCBS) which theMUmay access, such as to avoid frequent handover. Never-theless, there are limitations for these methods to capture themobility characteristic of MUs because the relevance of theusers’ positions are based on predefined routes that cannotexactly represent the reality. Moreover, an MU can theoreti-cally turn up at anywhere in the coverage range and movealong in any direction; it is difficult to characterize the user’smoving behavior.

Statistical information and prediction methods abouthandovers have been studied in [15–18] to deal with the is-sues during handover. In [15], a prediction framework forthe channel quality after handover is proposed, in whicha Q-learning-based approach is designed to improve theprediction accuracy. To solve the problem of missingneighbor cells, a neighbor cell list management scheme isproposed in [16], which is based on the long-term userequipment (UE) measurement statistics and the increasingrate of recent measurements. For optimizing the neighborcell list, the authors in [17] derive the handover probabilityfrom the handover history, with purpose of reducing theredundant cell scanning and thus maximizing the energyefficiency. Via studying previous visited cell informationand potential handover, a distance-based neighborhoodscanning scheme is proposed in [18], in order to maxi-mize utilization of the small cells. Inspirited by thesemethods, we develop a cell-centric approach for cellstatus prediction (CSP) that is based on defining the

Wirel. Commun. Mob. Comput. 2016; 16:1972–1986 © 2016 John Wiley & SoDOI: 10.1002/wcm

activity status for small cells from the perspective ofhistory handover requests.

The other major challenge in inbound handover is theconfusion issue that stems from the large number of self-deployed SCBSs in the macrocell coverage. In LTEadvanced networks, Third Generation Partnership Projectsuggests to identify small cells to solve the confusion orcollision by allocating PCIs (physical cell identifiers) toeach individual small cell [19].

However, it is challenging to achieve this goal becauseof the limited number of PCIs, for example, maximum 504in LTE-A system [20]. This results in that several smallcells have to share the same PCI, and hence, confusion orcollision may occur when a handover request is triggeredby an MU. Figure 1 describes three typical handoverscenarios: (1) an MU at the top right tends to handoverfrom macro-tier to small cell tier, namely, the inboundhandover [9], the cross-tier PCI confusion occurs becausetwo alternative SCBSs have the same PCI; (2) two adjacentsmall cells at the bottom left share the same PCI, the single tiercollision, or the inter-small cell PCI collision probably comesout as an MU is moving out of one small cell to another; (3)single-tier PCI confusion, as shown at the bottom right, couldhappen when an MU handoff from its associated small celland handover toward other SCBSs that share one PCI. Thehandover confusion leads to negative consequences thathinder the seamless connection in the two-tier macrocell-small cell networks, such as radio link failure (RLF) and calldrop, which in turn deteriorate the quality of experience forthe MUs as well as degrade the system capacity [7,9].

To cope with this problem, cell global identifier (CGI)and E-UTRAN CGI were proposed in LTE and LTE-Abecause CGI is the unique ID for the individual smallcell [9,21]. This approach [9], by introducing CGI, couldcause large overhead and latency. The MUs should im-plement identity of the target cell in a short time interval,while CGI needs to be obtained by reading the systeminformation that requires a large measurement time gap(e.g., up to 160ms for LTE). In addition, the MU cannotreceive or transmit any data from or to the serving cellbecause of the large measurement gap. Therefore, it leads tounnecessary service interruption, such as call drop or hand-over failure. Therefore, effective PCI reuse approaches needto be investigated for avoiding confusion during handover.

1.2. Motivations and contributions

In this work, we study how to devise the CSP algorithm forsmall cells and then apply it to the design of dynamic allo-cation strategies of PCI (DA-PCI), which is used to avoidhandover confusion. Unlike the MU’s movement-basedprediction approaches, we investigate the activity statusof small cells, which is designed to capture the statisticalproperty of inbound handover events that happen in smallcells. Our approach is a cell-based scheme wherein theCSP algorithm is proposed such as to find out the optimalprediction output of the next status for the small cells.Within the CSP method, we look into whether the small

1973ns, Ltd.

Figure 1. PCI confusion and collision in handover.

Dynamic PCI allocation on avoiding handover confusion Z. Xiao et al.

cells have frequent hand-in requests from the MUs whileregardless of obtaining the movement of MUs.

We then develop a dynamic allocation scheme of PCI(DA-PCI) according to the prediction result, in order to en-hance the usage of PCI and reduce handover confusions.The PCIs available for the small cell tier are divided intotwo categories: dedicated PCIs and public PCIs. The moti-vation is to assign the individual dedicated PCIs to thesmall cells that are facing large handover requests, whilethe public PCIs can be shared by the other small cells. Inother words, a partition of PCIs from the entire availablePCIs are treated as exclusive for the “busy” small cells,in order to fully exploit the usage of PCI and reduce hand-over confusions. We design two novel strategies: (i) theCSP-based DA-PCI algorithm, by which the dedicatedPCIs will be assigned to the small cells that are with busystatus, while the other small cells share the public PCI, and(ii) an integrated strategy. Within the integrated strategy,we propose a criterion to distinguish the isolated and thenon-isolated (clustered) small cells. We then formulatethe location relation integrated by status prediction infor-mation of the small cells. Through adopting Bayesianaverage method, we derive the preference function in orderto allocate the dedicated PCI to the small cells in whichhandover confusion could most likely occur. Simulationresults demonstrate that the proposed method yields ahigher accuracy than the conventional methods; in themeantime, handover confusion and latency can be reducedsignificantly during the inbound handover.

The rest of the paper is organized as follows. InSection 2, we study the small cell status model and CSPalgorithm, and then in Section 3, we present two strategies

1974 Wirel. Comm

for dynamic allocation of PCI. The algorithm description ofDA-PCI via integrated strategy is presented in Section 4.The accuracy of prediction scheme and the effectivenessof DA-PCI are demonstrated by simulation results inSection 5. Finally, we conclude the paper in Section 6.

2. THE PROPOSED CELL STATUSPREDICTION ALGORITHM

In this section, we model the activity status for small cellsthrough investigating the statistical property of MUs’handover requests, instead of studying the MUs’ move-ment. We then derive the CSP algorithm, which is ableto achieve the optimal prediction output for the small cells.

We summarize the notation and description of keyparameters in Table I.

2.1. Analytical status model for small cells

Define the state space of small cells as two types of status,busy and normal, which are denoted by SI and SII, respec-tively. “busy” (SI) signifies a small cell that has larger hand-over request from MUs, and “normal” (SII) indicates theopposite situation in which fewmobility events would occur.

We first present how to define small cell status bymodeling the inbound handover requests from the MUsto the small cell tier. The handover requests that may beinitialized because of unnecessary handovers are also takeninto consideration when we derive the activity status ofsmall cells. Let Ns denote the number of small cells inthe two-tier networks and k = 1,2,…,Ns. Consider the nth

un. Mob. Comput. 2016; 16:1972–1986 © 2016 John Wiley & Sons, Ltd.DOI: 10.1002/wcm

Table I. Description for key parameters.

Notation Description

sI/sII Busy/normal status for small cellΔt Duration of status time periodλ The average arrival rate of MUs that enter the small cell tier within Δtλk The average arrival rate of MUs that enter small cell kSnk Present nth status of the small cell kNS Number of small cellsX i

k Number of handover requests to small cell k during the time period of the ith statusXk Average number of handover requests to small cell k within one time periodX k Estimate value of Xk

Snk Present nth status of the small cell kQ Average number of handover requests toward the small cell tier within one status

time period ΔtQ The practical value of QM Length of the historical observations for the status sequence that can be used to

calculate X k

Hn�1k Historical status sequence for small cell k before the nth status

L Length of the historical status sequence, which is used to calculate the probability of thesmall cell k’s status

HLk Previous status sequence with length L for small cell k before the nth status

Nk sI ;HLk

� �Occurrences of sI that has taken place in small cell k in historical status sequence HL

k

Snþ1k Prediction result of the status of the small cell k in the next status, that is, (n + 1)th status

NI Amount of small cells with isolatedNP Number of available PCIs for the small cell tier


status of the small cell k, Snk , we use Δt to describe theduration of status time period. Based on [22], the wirelesssmall cell network can record the handover request and thenconstruct the handover history. According to Jun-Bae andSeung-Que [23], the number of MUs enters the coverage ofsmall cell k follows the Poisson distribution with rate λk.Let λ denote the average arrival rate of MUs that enter theentire small cell tier within Δt. For the sake of simplicity,suppose all these MUs in the ith Δt could initialize inboundhandover requests to the small cells, which is denoted byX i

k. It is noteworthy that Xik is independently and identically

distributed for an arbitrary ith Δt.We then denote Xk as the average number of in-

bound handover requests to the small cell k from theMUs within a time period Δt, and theoretically, wehave X k ¼ E X i

k

� � ¼ λkΔt . Let Q denote the averagenumber of inbound handover toward the small cell tierthat has taken place in the time period Δt, which canbe given as

Q ¼ ∑NSk¼1X k

NS

$ %¼ ∑NS

k¼1λkΔtNS

$ %(1)

where b � c means the trunc operation. To define thestatus of the small cell k, if Xn

k≥Q then Sk = sI, other-wise, Sk= sII. Therefore, the probability of status ofthe small cell k is busy and can be expressed as

P Sk ¼ sIð Þ ¼ P Sik ¼ sI� � ¼ ∑∞

X ik¼Q

λkΔtð ÞX ik

X ik !

e�λkΔt

" #(2)


The probability of the status of the small cell k isnormal and can be expressed as

P Sk ¼ sIIð Þ ¼ P Sik ¼ sII� � ¼ ∑Q�1

X ik¼0

λkΔtð ÞX ik

X ik !

e�λkΔt

" #

(3)

Based on the aforementioned analysis, Q is constant for acertain λk. According to Equation (2) and Equation (3), theprobabilities P(Sk= sI) and P(Sk= sII) tend to be constantas well.

When considering the practical handover scenario, itis difficult to obtain the exact values of λk and Xk.Instead, we try to find out the estimate of Xk in termsof the history observations of the inbound handoverevent, namely, the number of historical inbound hand-over requests of the small cell k. For the present nthstatus of small cell k, we can have the historic statussequence, which is given as Hn�1

k ¼ S1k ; S2k ;…; Sn�1

k

� �and n� 1 denotes the length of status sequence. In thefollowing, we present the definition for small cell status.

Definition 1 (small cell status). We denote X k ¼∑n�1

i¼n�MXik

� �=M , which is used to describe the average

number of inbound handover requests to the small cellk that has taken place in the last M time periods beforethe nth status time period. M is the length of the histor-ical status sequence that is used to calculate X k . Withthe increase of M, the difference between X k and Xk will

1975ns, Ltd.


approach zero. By replacing Xk with X k, Equation (1) canbe rewritten as

Q ¼ ∑NSk¼1X k

NS

$ %¼ ∑NS

k¼1∑n�1i¼n�MX

ik

M �NS

$ %(4)

For a specified M, if n ≤M, we have M= n� 1. The def-inition of the status of small cell k can be given by

Snk ¼sI ;X n

k≥QsII ;X n

k < Q

((5)

2.2. Cell status prediction algorithm

Let Nk sI ;HLk

� �denote the occurrences of SI that has taken

place in small cell k in the history status sequenceHLk . Here,

L denotes the length of HLk , which can be exploited to ob-

tain occurrences of sI (or sII) in the last L time periodsbefore the nth status. We calculate the probability of thesmall cell k’s status via

PL Sk ¼ sIð Þ ¼ Nk sI ;HLk

� �=L (6)

Through the same method, we can obtainPL Sk ¼ sIIð Þ ¼ Nk sII ;HL

k

� �=L . Similar with the relation

between Xk and X k, we would not be able to obtain the ex-act value of P(Sk); for this reason, we can replace P(Sk)with PL(Sk).

Lemma 1. Let Snþ1k denote the prediction result of the sta-

tus of the small cell k in the next period (i.e., (n+ 1)th Δt).The optimized status prediction judgment can be given as

Snþ1k ¼ argmax

s∈ sI ;;sIIf gPL Sk ¼ sð Þð Þ (7)

Proof. Let P Snþ1k ¼ sI

� �describe the probability of that

we predict the status for the small cell k in the (n+ 1)th

Figure 2. The error probability Pe versus the P Snþk

�

1976 Wirel. Comm

Δt as busy. Therefore, the error probability of the statusprediction can be expressed as

Pe ¼ P Snþ1k ¼ sII jS nþ1

k ¼ sI� �

þ P Snþ1k ¼ sI jS n�1

k ¼ sII� �

(8)

whereP Snþ1k ¼ sI

� �andP Snþ1

k ¼ sII� �

are the probabilitiesof the statuses for small cell k in the (n+ 1)th Δt are busyand normal, respectively. According to Equation (2) andEquation (3), for an arbitrary i, we have P Sik

� � ¼ P Skð Þ.Thus, Equation (8) can be rewritten as

Pe ¼P Sk ¼ sII jS nþ1k ¼ sI

� �þ P Sk ¼ sI jS n�1

k ¼ sII� �

(9)

By replacing P(Sk) with PL(Sk), we can obtain Pe from

Pe ¼ P Snþ1k ¼ sI

� ��PL Sk ¼ sIIð Þ

þ P Sn�1k ¼ sII

� ��PL Sk ¼ sIð Þ

(10)

Pe ¼ 1� 2�PL Sk ¼ sIð Þð ÞP Snþ1k ¼ si

� �þ PL Sk ¼ sIð Þ

(11)

Based on Equation (11), Pe can be illustrated inFigure 2, in which the case PL(Sk = sI) ≥ 0.5 and the casePL(Sk = sI)< 0.5 are presented in Figure 2(a) and (b),respectively.

In the case PL(Sk = sI) ≥ 0.5, the minimum of Pe appears

when P Snþ1k ¼ sI

� �= 1 (see “*” in Figure 2(a)). In the

case PL(Sk = sI)< 0.5, the minimum of Pe turns out when

P Snþ1k ¼ sI

� �= 0. Therefore, when PL(Sk = sI) ≥ 0.5, that

is, PL(Sk = sI) ≥PL(Sk = sII), we should predict the nextstatus for the small cell k is busy (sI) because we can surelyachieve the minimum Pe. In the same manner, the nextstatus for the small cell k should output normal (sII) whenPL(Sk = sI)< 0.5 (i.e., PL(Sk = sI)<PL(Sk = sII)).

1 ¼ sI�. (a) PL(Sk = sI) ≥ 0.5. (b) PL(Sk = sI)< 0.5.



In a word, the small cell status is related with the MUsthat may move close to the small cells and their attempts ofaccessing the small cell tier. Nevertheless, unlike theexisting works that focused on studying MU’s movement,we instead model the activity status for small cells fromexploiting the statistical property of MUs’ handover re-quests and hence regardless tracking the traveling ofMUs. Algorithm 1 summarizes the proposed algorithmfor small CSP.

Algorithm 1 Small Cell Status Prediction (CSP) Algorithm

Initialization: Initialize a set of small cells, USand Ns is its

cardinality, located randomly in a single macrocell.

Predict the (n+ 1)th status of a specified small cell and repeat.for k=1 to Ns do

a) Calculate Xknbased on the handover requests from MUs.

b)Compute Xk–

and Q–

according to Equation (4).

c) Determine the current status Sknfor the small cell k by

Equation (5).

d)Update the history sequence Hkn�1

and update PL(Sk) based

on Equation (6).

e) Obtain the (n+1)th prediction status Skn+1^ for small cell k by

applying lemma 1.

end for

Outcome: The (n+1) th statuses for the small cell set US.

For implementing the proposed CSP algorithm, twomemory units are required for each small cell. Onememory unit stores the handover requests of the last Mstatus periods and the other one stores the status of the lastL periods. In terms of the procedure, the complexity ofCSP algorithm is O(1), and the latency is relatively smalland somehow negligible. If we want to achieve a higheraccuracy and hence increase L and M, the memory unitssize should increase as well and the latency will raise afew but can be ignored.

To conclude this section, the small cells status predic-tion is able to provide essential knowledge for the subse-quent PCI allocation for avoiding the handover confusion.The PCI is what the physical layer used to identify and sep-arate data coming from different base stations includingMBS and small cells. It is also used to achieve the channelsynchronization of MUs and cells, which is similar to theScrambling Codes from Universal Mobile Telecommuni-cations System (UMTS) [24]. In this regard, when we con-sider the inbound handover in two-tier HetNets, a portionof the available PCIs should be exclusive for the busy smallcells because they have large demand of data transmission.

3. DYNAMIC ALLOCATIONSTRATEGIES OF PCI

In this section, we propose a dynamic allocation algorithmof PCI (DA-PCI) to reduce confusion during inbound hand-over. Within the DA-PCI, PCI are divided into dedicatedand public, wherein the principle idea is assigning the


dedicated PCIs to the small cells with high hand-in re-quests, namely, the busy small cells. On the other hand,small cells with less handover possibility share public PCIs.

To be specific, we design two strategies for DA-PCI,the CSP-based strategy and the integrated strategy, whichare motivated in terms of two different concerns: (i) regard-less the geographical locations of the small cells, to ensurethe dedicated PCIs assigned to the small cells that are withstatus sI and (ii) to achieve the maximum usage of PCI, inother word, to find out the small cells that actually needdedicated PCIs by jointly considering the cell status andthe reference signal received quality (RSRQ) from theMUs during the handover.

We present the two strategies for DA-PCI in the follow-ing of this section.

3.1. DA-PCI: CSP-based strategy

Let NP denote the number of available PCIs for the smallcell tier. We design a criterion for the proportion whengrouping dedicated PCI and public PCI according to theprevious CSP. According to the current status and theprediction result of the (n+ 1)th status, the number of thededicated PCIs for the (n+ 1)th status period, Nnþ1

D , isproportional with Rnþ1

D;L :

Nnþ1D ¼ NP�Rnþ1

D;L

j k(12)

Rnþ1D;L describes the ratio of sI takes place in the previous L

status periods for all small cells:

Rnþ1D;L ¼ ∑NS

k¼1 Nk sI ;HLk

� �þ Δ� �Lþ 1ð Þ�NS

(13)

In Equation (13), Δ = 1 if Snþ1k ¼ sI , which is based on

the CSP result. Otherwise Δ= 0. The PCI allocation is car-ried dynamically to cope with the fast changing of smallcell activity status.

Because NS>NP, a certain number of public PCIshould be reserved. That is to say, ND should be smallerthan Np. Normally, Rn

D;L< 1, then ND<Np. In the extreme

situation, for instance, the vast majority of small cells havehigh handover possibility in the previous periods, RI mightapproach one, then ND could be equal to NP. In such case,we make at least one PCI as public; hence, ND=NP� 1.

In short, the “sI” small cells, with high probability fac-ing large handover requests, have the preference to assignthe dedicated PCI according to the CSP-based strategy,thus allowing to avoid the handover confusion that is likelyto happen in busy small cells.

In addition, because of the random manner of small celldeploying, some new small cells may turn on in the two-tier HetNets system. It is unable to obtain the activity his-tory for these new cells. To solve this, we set the new smallcells with sII, namely, the normal status, because we canassume there would be a small number of handover

1977ns, Ltd.


requests when a small cell is just deployed and switchedon. Therefore, the public PCI can be assigned for thenew small cells.

3.2. DA-PCI: an integrated strategy

With further consideration in order to avoid PCI confu-sion and obtain the high usage of insufficient PCIs, wepropose a novel integrated strategy for dynamic allocationof PCI. This approach has an advantage of deriving amore effective PCI allocation when comparing with theCSP-based strategy.

Proposition 1. For PCI allocation based only on the CSP,the DA-PCI cannot guarantee the full usage of the limitedPCI if the geographical location relation among smallcells is not properly taken into consideration.

Proof. Given the unplanned deploying of small cells, the“sI” small cells might be far apart from each other.Therefore, these “sI” small cells do not need dedicatedPCIs. In the contrary, the “sII” small cells sharing publicPCI should be assigned dedicated PCIs if they aregeographically clustered.

As shown in Figure 3(a), SCBS1 and SCBS2 are locatedfar away from each other. When the MUs move towardone small cell and are able to initialize inbound handoverrequest, for instance, MU1 approaches SCBS1 and receivesbetter signal from SCBS1 comparing with the receivedsignal from MBS; hence, this triggers inbound handoverto SCBS1. In the meantime, MU1 also receives downlinksignal from SCBS2 while very weak because of the largedistance between SCBS1 and SCBS2. Likewise, when

Figure 3. The distribution models of small cells. (a) Small cells can bare distributed cluste

1978 Wirel. Comm

MU2 moves to SCBS2 and intends to hand-in, SCBS2 isthe only candidate target cell. Under this geographical lo-cation assumption, we can conclude that SCBS1 andSCBS2 have non-overlapping handover triggering regions.Therefore, regardless the status of the small cells, they canshare public PCI.

On the other hand, we consider the case where theSCBSs are distributed closely as depicted in Figure 3(b).When MU3 has an attempt of inbound handover, thereare two competitive candidates, namely, SCBS3 andSCBS4, which are with sII and can share a public PCI ac-cording to the CSP-based strategy. Consequently, hand-over confusion could turn out. We can also observe thesimilar situation, where handover confusion could takeplace when MU4 moves to SCBS4 and SCBS5.

To summarize, Proposition 1 reveals the inadequacy ofthe CSP-based strategy for PCI allocation, because of themissing of location information of small cells. When welook into that MU5 moves to SCBS3 and SCBS6, handoverconfusion would not happen, although these two arelocated tightly close, because they are with sII and sI re-spectively and one of them (SCBS6) uses a dedicated PCI.

The aforementioned analysis proves the necessity of jointlyconsidering the location information and the activity status ofsmall cells when designing the PCI allocation method.

There are existing methods that utilize the small cell lo-cation information, for instance, PCI assignment usinggraph theory [25,26]. These methods rely on an assump-tion that geographic location of small cell should be knowna priori. Such an assumption is correct for the MBS as theposition of MBS is carefully selected and fixed in the net-work planning stage, while the location of SCBSs is uncer-tain because of the random manner of the small celldeployment. Moreover, placing indoor SCBSs is even

e treated as isolated, which can share public PCI. (b) Small cellsred and closely.



impossible to obtain the location information because ofvery weak GPS signal penetration into the buildings.

Because of the location dependency of these method,localization and location management for small cell tierare needed. In [18], the authors investigate the distanceestimation among small cells, in order to reduce theneighborhood scanning for handover purposes. In [27],the authors provide comprehensive understanding oflocation-aware self-organization network techniques thatutilize localization in LTE/LTE-A small cell networks.The localization information is the key part of enabling ad-vanced self-organization network methods. Rather thanattempting to obtain the specific geographic location forsmall cells, our goal is to devise a method to characterizethe location relation among small cells by means of theanalysis on RSRQ [7,19], by which we do not need todevelop accurate localization approaches. To achieve this,we design the preference relation for small cells, in whichthe location-relation and the CSP are deliberately integratedby using Bayesian average method [28].

We first present the definition of a small cell k that iseither isolated or clustered from the perspective of RSRQ.The MUs can measure the RSRQ from the handover candi-date SCBSs when they move toward the vicinity of thesesmall cells [9,29].

Definition 2 (isolated small cells). Assume an MU movesclose to small cell k and the MU is able to reliably decodethe control signaling from SCBS k (it may also receivesignaling from other SCBSs), the small cell k can be proc-essed as isolated, if the following inequality is satisfied.

RSRQ MU ; jð ÞRSRQ MU ; kð Þ < ε;∀j∈ 1;NS½ �; j≠k (14)

With ε is used to describe the RSRQ difference that al-lows the small cell to choose the handover trigger parame-ters such as the hysteresis margin [9,30]. Inequality in (14)can be equivalent to

RSRQ MU ; kð Þ >> RSRQ MU ; jð Þ; ∀j≠k (15)

Recall Figure 3(a), which illustrates a typical isolatedcase, SCBS1 and SCBS2. It is thus reasonable to infer thatan isolated small cell is either far away from other smallcells, or strong signal attenuation between the small celland others. These both make a small cell as “isolated” ina sense of cell identification. As a result, when an MUtriggers handover to an isolated small cell, the other smallcells would not interfere this handover process.

Clustered small cellsWhen multiple small cells are geographically clustered,

as shown in Figure. 3(b), MUs can detect approximateRSRQs from those clustered small cells. On the otherhand, small cells are developed with purpose of offloading


the data and increasing capacity for the macrocell [4,5];hence, the area that small cells are clustered distributed islikely with high demand of mobile data. Therefore, aclustered small cell tends to be with “sI” and thus needs adedicated PCI as its neighbors can probably cause hand-over confusion when MUs move across the clustered area.

3.3. Preference of small cells via Bayesianaverage method

Let NI denote the amount of small cells with isolated.Clearly, no matter the activity status of the “isolated” smallcells are with, they share one public PCI. We note that thispublic PCI can also be shared by other non-isolated(clustered) small cells. Therefore, there are still NP PCIsremaining for the rest of (NS�NI) small cells.

Proposition 2. Consider the extreme case, all the non-isolated (clustered) small cells can have an individualdedicated PCI if NS�NI ≤ND holds; here, ND can beobtained from Equation (12).

Proposition 2 shows an extreme case, in which there area large number of small cells that can be treated as isolatedin the macrocell coverage, while the clustered small cellswith “sI” that probably need dedicated PCI are in aminority.

In general scenarios, the available PCIs are insufficientbecause of dense deploying of small cells, namely,(NS�NI)>>NP and in most cases NS�NI>ND. In otherwords, there is certain quota on the number of dedicatedPCIs that small cells should have. Hence, we form amatching problem to devise the PCI allocation strategy.We first define the preference function for small cellsas follows:

f k; nð Þ ¼ RD;LE yik� �þ∑NA

i¼1PL Snk� �

yikNA þ RD;L

(15)

where yik denotes the normalized RSRQ from small cell kby MU i. i= 1,…, NA, NA denotes the number of MUs(voters) that can detect the RSRQ from small cell k.PL Snk

� �can be obtain according to the analysis in

Section 2.2. RD,L can be derived from Equation (13), whichis nearly proportional to the occurrence frequencies ofstatus sI. Here, E yik

� �describes the average RSRQ for

(NS�NI) small cells collected by MUs, which can beexpressed as

E yik� � ¼ ∑Ns�NI

k¼1 ∑Nai¼1y

ik= Ns � NIð ÞNA (16)

As a result, f(k,n) in Equation (15) is used to evaluatethe average score for each small cells (except the isolatedsmall cells).

Through the preference relation, we introduce a rankingscheme via Bayesian average method [28], which charac-terizes the location relation for small cells by modelingRSRQ measurement integrated with the status prediction.

1979ns, Ltd.

for k=1 to (NS-NI) do

Compute E [ yki] using Equation (16).

Compute the preference function for small cell k by using

Equation (15) and the related analysis in Sec. II.

end for

Calculate the number of dedicated PCI ND according to

Equation (12).

if NS�NI ≤ND then

Assign individual dedicated PCI to each small cell inUCS.

else

Sort the preference sequence F(k,n) in descending order.

Find out the first ND small cells and assign individual


The preference sequence for the clustered small cells canbe written as

F k; nð Þ ¼ f k; nð Þ; k∈ 1;NS � NI½ � (17)

In a word, we rank the small cells by means of the givenpreference, which is able to guarantee the dedicated PCIassigned to the small cells with (i) the high hand-in re-quests (ii) clustered connection with others, in other words,the handover confusion is more likely to turn out in thesesmall cells.

dedicated PCI to each.

end if

for the rest of small cells, in total (NS-NI-ND), do

Classify the (NS-NI-ND) small cells into (NP-ND)

groups according to the preference Equation (13).

Each group shares one public PCI.

end for

4. ALGORITHM DESCRIPTION OFDA-PCI VIA INTEGRATEDSTRATEGY

Owing to the thorough study on the distribution propertiesof small cells, the total isolated small cells share one publicPCI regardless the cell status. In the meantime, this publicPCI can be also reused by the remaining small cells. Suchaction can aid in shrinking the number of small cells so thatwe can focus on the PCI allocation among the clusteredsmall cells.

The small cells, with busy status and clustered relationwith their neighbors, should have the priority to use dedi-cated PCIs. To this end, we deduce the preference functionby employing the Bayesian average method so that tocapture the “clustered connection” as well as cell statusinformation for the clustered small cells. Therefore, theproposed integrated strategy is able to make larger usageof the limited PCIs when comparing to the CSP strategy.The DA-PCI algorithm via integrated strategy consists oftwo phases, which is presented in Algorithm 2.

Algorithm 2 DA-PCI Algorithm: the Integrated Strategy

Initialization: Let UISdenote the isolated small cell set and NI

is its cardinality. Let UCSdenote the non-isolated small cell set

and (NS-NI) is its cardinality. InitializeUISand UC

Swith empty

sets.

Phase I – PCI sharing in isolated small cells

for k=1 to NS do

while An MU that enters the vicinity of small cell k do

Calculate RSRQ(MU,k).

Calculate RSRQ(MU, j), ∀ j∈ [1,NS], j ≠ k.

if Inequation is satisfied (13) then

Add the small cell into UIS.

else

Add the small cell into UCS.

end if

end while

end for

Assign one public PCI to UIS.

Phase II – Integrated strategy for PCI allocation

From Algorithm 1, we can obtain the status for small cells

setUSand the state transition probabilities. Consider

the clustered small cells set UCS:

1980 Wirel. Comm

5. SIMULATIONS AND DISCUSSION

In this section, we evaluate the performance for small CSPand the PCI allocation strategies. We consider a two-tierHetNets with a set of small cells located in a singlemacrocell, wherein parts of SCBSs are deployed indoor.Let rM and rS denote the radius of disc coverage formacrocell and small cell, respectively. We derive the activ-ity status of small cells by modeling the handover requestsfrom the MUs to the small cells. Notice that these handoverrequests also include the ones that are initialized because ofthe unnecessary handover. As the handover is proceeding,unnecessary handover may turn out, for instance, if an MUis moving with high speed and crossing the coverage areaof small cell. By considering MU’s mobility, that is, themoving speed, intelligently enhanced handover hysteresisdeciding algorithm can reduce the unnecessary handover[9]. Based on the finding in [31], we use reference symbolreceived power with hysteresis for MU to determinewhether it would trigger inbound handover, and in themeantime, to avoid unnecessary handover via consideringthe impact of MU’s velocity. The major simulation param-eters are summarized in Table II.

We first look into the accuracy of small CSP. Figure 4depicts the predictive results, which are calculated fromthe 20th status, and the length of status sequence is 200.The conventional MU mobility-based prediction, whichintends to capture MU’s movement, such as in [12,14],are compared with the proposed approach in terms of theaverage prediction accuracy. It is observed that theaccuracies of the proposed methods outperform the con-ventional methods that aim to capture the MU’s movement(the bottom two curves). We can see that the parameters(M, L)influence the prediction accuracy. The top twocurves represent the results of the proposed methods whenM= 20. The accuracy grows nearly linearly as the statuslength is larger than 50 when L= 40. After 120 statuses,the prediction accuracy keeps almost stable and up to


Table II. Key simulation parameters.

Scenario HetNets with co-channeldeployment small cells

MBS Tx Power 46 dBmSCBS Tx Power 20 dBmMBS height 30mSCBS height 10mMacrocell antenna gain 14 dBiSmall cell antenna gain 5 dBiPath loss ITU UMa/UMi [32]Log-normal shadowing Macrocell standard

deviation 8 dBSmall cell standarddeviation 4 dB

System bandwidth 10MHzCarrier frequency 2.6GHzMU velocity Uniform in

[5,10,15,20,30] km/hMU mobility wrap around [33]rM 500mrS 20mWall penetration loss 10 dB


0.93. The approximate accuracy can be obtained whenL= 20, but more random jitter can also be found from thegreen curve before 80 status.

To sum up, small cell status-based prediction is ableto capture the statistical property of MU’s hand-in re-quest, while the conventional methods fail to reflectthe moving property of each individual MU because ofthe randomness and irrelevance of MUs’ mobility be-havior. In Figure 4, the middle two curves based onthe proposed methods show the moderate predictionresults when M = 10, L = 20 and M = 10, L = 10, respec-tively. We can see that lower accuracies are finallyachieved (0.81 in black curve and 0.85 in red curve).

Figure 4. The average prediction ac


In addition, the curves tremble severely, and more his-toric statuses have to be experienced before enteringthe stable stage.

We then assess the performance of the proposedDA-PCI strategies. To verify the effectiveness by applyingthe predictive information, we observe the number of RLFcaused by the handover confusions that occur in 1 h withvarious numbers of small cells deployed under macrocellcoverage. Figures 5 and 6 depict the simulation results,where the available numbers of PCIs are set as 0.4*NS

and 0.8*NS, respectively. In this simulation, the proposedDA-PCI via CSP-based strategy, DA-PCI via the inte-grated strategy, the conventional PCI allocation based onMU movement prediction, and random PCI allocation assuggested in LTE [20,21] are compared. There are morehandover events and thus more confusions when thenumber of small cells increases, as shown by the upper-most curves with “triangle” in Figures 5 and 6, respec-tively. We can see that the handover confusions aredecreased in varying degrees when applying various PCIallocation methods, wherein the proposed methods canobtain better confusion avoiding results. By randomlyallocating PCIs, there are still large amount of confusionissues happen during the handover process as more smallcells are deployed. This implies that CGI has to be readso that large latency would be introduced and then thenumber of RLFs increases.

The confusion issue can be alleviated by exploitingpredictive information. Results in Figure 5 show thatthe proposed DA-PCI algorithm outperforms the conven-tional methods that are based on MU movement predic-tion. When comparing the two DA-PCI strategies, theRLF curves of two strategies grow in different way whenthe number of small cells exceeds 50. With the CSP-based strategy, the number of RLFs increases with arapid rise. When applying the integrated strategy, theRLF curve turns out to be a moderate rise.

curacy versus status sequence.

1981ns, Ltd.

Figure 5. The number of RLFs in 1 h when NP = 0.4*NS, M = 20, and L = 40.

Figure 6. The number of RLFs in 1 h when NP = 0.8*NS, M = 20, and L = 40.


When the available PCIs increase to NP=0.8*NS, asshown in Figure 6, the confusions and RLFs can be furtherreduced in contrary to that NP=0.4*NS. Two DA-PCI strat-egies can obtain better confusion avoiding performance thanthe other methods. The CSP-based strategy can obtain theapproximate performance with the integrated strategy whenthe small cells are sparsely deployed for instance, NS< 50.As NS grows, the RLFs increase with slow growth, whichis due to that DA-PCI via integrated strategy can furtherenhance the usage of the insufficient PCIs.

Figure 7 depicts how the number of RLF changes underthe various number of users. In this simulation, the numberof small cells, NS, is set to 200, and the available numberof PCIs is NP=0.8*Ns. As the average arrival rate of MUs,λ increases, which means more MUs enter the small cellsand hence more handover requests could be triggered; forthis reason, more RLFs take place because of the limited

1982 Wirel. Comm

PCIs. In terms of avoiding handover confusion and reducingRLFs, the proposed DA-PCI strategies can obtain better per-formance than the other methods with a certain MU arrivalrate. When comparing the two DA-PCI strategies, the inte-grated strategy outperforms the CSP-based strategy, whichis also validated from the results in Figures 5 and 6.

The throughput simulations are shown in Figures 8–10.In Figure 8, the throughput of small cells is given underfour cases with various ratio of the available PCIs to thesmall cell tier, namely, NP= 0.8*NS, NP= 0.6*NS,NP = 0.4*NS, and NP = 0.2*NS, respectively. The numberof small cells NS is 100. As we know, the handover confu-sion and the RLF can cause the degradation of the systemthroughput. All the cases of the throughput in Figure 8 aredropping down when the number of available PCIs getmuch insufficient. This is because that the confusions in-crease when the number of PCIs becomes less. Moreover,


2 4 6 8 10 12 14 16 18 200

200

400

600

800

1000

1200

1400

Average MU arrival rate in small cell tier

The

Num

ber

of R

LFs

DA-PCI, Integrated strategy

DA-PCI, CSP based strategy

MU movement based PCILTE Random PCI

Number of Handover

Figure 7. The number of RLFs with various MUs when NS = 200, NP = 0.8*NS, M = 20, and L= 40.

Figure 8. The average throughput in small cell tier under four cases with various number (NP) of the available PCIs. The number ofsmall cells, NS, is set to 100.


we can see that in Figure 8 the proposed DA-PCI canachieve better system throughput when compare with othermethods. In particular, when there are smaller number ofPCIs, for instance, NP= 0.4*NS and NP = 0.2*NS, the pro-posed integrated strategy is able to offer a considerableperformance advantage in terms of maintaining thethroughput at an acceptable level.

In Figures 9 and 10, the throughputs in small cell tierare presented with various number of small cells, namely,Ns = 100, Ns = 200, Ns = 300, Ns = 400, and Ns = 500, re-spectively. As Ns grows, then more small cells deployedwithin the macrocell coverage; hence, the throughputof small cells increases. When Ns> 300, the distributionof small cells turns to be relatively dense, large numberof handover events turn out, and in the meantime


handover confusion issues become severer because ofthe insufficient PCIs, for instance, NP = 0.6*NS andNP = 0.2*NS. The results from Figure 9 show that our pro-posed DA-PCI strategies can obtain better throughput im-provement than the other methods when Ns ≥ 100,especially for the dense small cell scenario, for example,when Ns ≥ 400. Similar results can be seen in Figure 10.In addition, the proposed integrated strategy still has ad-vantage of improving throughput as Ns increases. Theseresults are owing to the joint consideration of the cell ac-tivity status and the cell distribution relation, the integratedstrategy-based DA-PCI can maximize the usage of the PCIs,especially the limited dedicated PCIs. As a result, theselimited dedicated PCIs are ensured to assign to the small cellsthat are most likely facing handover confusions.

1983ns, Ltd.

Figure 9. The throughput in small cell tier under various number of small cells when NP = 0.6*NS.

Figure 10. The throughput in small cell tier under various number of small cells when NP = 0.2*NS.


6. CONCLUSIONS AND FUTUREWORK

In this work, we design an analytical status model for the smallcells, by focusing on investigating the statistical property ofMUs’ handover requests and thus regardless studying themove-ment of MUs. We develop a small cells status prediction algo-rithm, which is able to exploit the previous status and give outthe optimal prediction output for the small cells’ future status.

Based on the prediction, a pro-active action, dynamicallocation algorithm of PCI (DA-PCI) is proposed to avoidhandover confusion and reduce handover latency. Two

1984 Wirel. Comm

novel strategies are designed for DA-PCI: (i) the CSP-basedstrategy, by which the small cells with busy activity statuscan be assigned with dedicated PCIs, and (ii) an integratedstrategy, wherein we formulate the preference relation forsmall cells via RSRQ relation integrated with status predic-tion information using Bayesian average method. Simula-tion results demonstrate that our proposed approach, ratherthan tracking MU’s movement, is more feasible for theCSP to determine if a small cell faces large hand-in requests,as well as in resolving the handover confusion problem.

The proposed methods are applicable to deal with butnot restricted to cross-tier confusion or collision issues. In



terms of the future work, we will study resource allocationin wireless HetNets, such as pre-bandwidth reserving forthe incoming data requirement, by taking advantage ofthe prior knowledge of the small CSP.

ACKNOWLEDGEMENTS

This work was supported in part by the National NaturalScience Foundations of China (No. 61301148 and No.61272061), the Fundamental Research Funds for theCentral Universities of China, Hunan Natural ScienceFoundation of China.

REFERENCES1. Zander J, Mahonen P. Riding the data tsunami in the

cloud: myths and challenges in future wireless access.IEEE Communications Magazine 2013; 51(3):145–151. DOI:10.1109/MCOM.2013.6476879.

2. Mobile CVNI. Cisco visual networking index: globalmobile data traffic forecast update, 2011–2016. WhitePaper 2012.

3. Qualcomm. Bringing 1000× closer to reality: hyper-dense small cells deployments. May 2014.

4. Bhushan N, Li J, Malladi D, et al. Network densifica-tion: the dominant theme for wireless evolution into5G. IEEE Communications Magazine 2014; 52(2):82–89. DOI:10.1109/MCOM.2014.6736747.

5. Jungnickel V, Manolakis K, ZirwasW, et al. The role ofsmall cells, coordinated multipoint, and massive MIMOin 5G. IEEE Communications Magazine 2014; 52(5):44–51. DOI:10.1109/MCOM.2014.6815892.

6. Haider F, Gao X, You XH, et al. Cellular architectureand key technologies for 5G wireless communicationnetworks. IEEE Communications Magazine 2014;52(2): 122–130. DOI:10.1109/MCOM.2014.6736752.

7. Fehske A J, Viering I, Voigt J, et al. Small-cell self-organizing wireless networks. Proceedings of theIEEE 2014; 102(3): 334-350. DOI: 10.1109/JPROC.2014.2301595.

8. Hwang I, Song B, Soliman SS. A holistic view onhyper-dense heterogeneous and small cell networks.IEEE Communications Magazine 2013; 51(6):20–27. DOI:10.1109/MCOM.2013.6525591.

9. Xenakis D, Passas N, Merakos L, Verikoukis C. Mobilitymanagement for femtocells in LTE-advanced: key as-pects and survey of handover decision algorithms.IEEE Communications Surveys & Tutorials 2014;16(1): 64–91. DOI:10.1109/SURV.2013.060313.00152.

10. Taleb T, Ksentini A. QoS/QoE predictions-based ad-mission control for femto communications. IEEEInternational Conference on Communications 2012;5146-5150: . DOI:10.1109/ICC.2012.6364289.


11. Lee HY, Lin YB. A cache scheme for femtocellreselection. IEEE Communications Letters 2010;14(1): 27–29. DOI:10.1109/LCOMM.2010.01.091917.

12. Ulvan A, Bestak R, Ulvan M. Handover scenario andprocedure in LTE-based femtocell networks. TheFourth International Conference onMobile UbiquitousComputing, Systems, Services and Technologies 2010;213-218: . DOI:10.1007/s11235-011-9599-9.

13. Huang CJ, Chen PC, Guan CT, et al.A probabilistic mo-bility prediction based resource management scheme forWiMAX femtocells. IEEE International Conference onMeasuring Technology and Mechatronics Automation2010; 1: 295–300. DOI:10.1109/ICMTMA.2010.810.

14. Sung NW, Pham NT, Huynh T, Hwang WJ. Predic-tive association control for frequent handoveravoidance in femtocell networks. IEEE Communica-tions Letters 2013; 17(5): 924–927. DOI:10.1109/LCOMM.2013.031913.130127.

15. Becvar Z, Mach P, Strinati EC. Q-learning-based predic-tion of channel quality after handover in mobile networks.IEEE 25th Annual International Symposium on Personal,Indoor, and Mobile Radio Communication (PIMRC)2014; 1359–1364. DOI:10.1109/PIMRC.2014.7136380.

16. Watanabe Y, Matsunaga Y, Kobayashi K, Sugahara H,Hamabe K. Dynamic neighbor cell list management forhandover optimization in LTE. IEEE 73rd VehicularTechnology Conference (VTC Spring) 2011; 1–5.DOI:10.1109/VETECS.2011.5956456.

17. Becvar Z, Mach P, Vondra M. Self-optimizing neigh-bor cell list with dynamic threshold for handoverpurposes in networks with small cells. WirelessCommunications and Mobile Computing 2015;15(13): 1729–1743. DOI:10.1002/wcm.2456.

18. Vondra M, Becvar Z. Distance-based neighborhoodscanning for handover purposes in network with smallcells. IEEE Transactions on Vehicular Technology2015. DOI:10.1109/TVT.2015.2402753.

19. Golaup A, Mustapha M, Patanapongpibul LB.Femtocell access control strategy in UMTS andLTE. IEEE Communications Magazine 2009; 47(9):117–123. DOI:10.1109/MCOM.2009.5277464.

20. Zhang J, De la Roche G. Femtocells: technologies anddeployment. Wiley: New York, 2010; 211.

21. 3GPP. Evolved universal terrestrial radio access(E-UTRA) radio resource control (RRC) (release 10).TS 36.331 version 10.6.0, Jul 2012.

22. Katsaros D, Manolopoulos Y. Prediction in wireless net-works byMarkov chains. IEEEWireless Communications2009; 16(2): 56–64. DOI:10.1109/MWC.2009.4907561.

23. Seo JB, Lee SQ, Park NH, Lee HW, Cho CH. Queueingfor handover calls in a hierarchical cellular network. IEEE60th Vehicular Technology Conference (VTC Fall) 2004;5165–5169. DOI:10.1109/VETECF.2004.1405085.

1985ns, Ltd.

http://dx.doi.org/10.1109/MCOM.2013.6476879




http://dx.doi.org/10.1109/JPROC.2014.2301595

http://dx.doi.org/10.1109/JPROC.2014.2301595


http://dx.doi.org/10.1109/SURV.2013.060313.00152

http://dx.doi.org/10.1109/ICC.2012.6364289

http://dx.doi.org/10.1109/LCOMM.2010.01.091917

http://dx.doi.org/10.1007/s11235-011-9599-9

http://dx.doi.org/10.1109/ICMTMA.2010.810



http://dx.doi.org/10.1109/PIMRC.2014.7136380

http://dx.doi.org/10.1109/VETECS.2011.5956456

http://dx.doi.org/10.1002/wcm.2456

http://dx.doi.org/10.1109/TVT.2015.2402753


http://dx.doi.org/10.1109/MWC.2009.4907561

http://dx.doi.org/10.1109/VETECF.2004.1405085


24. Stefania S, Issam T, Matthew B. LTE, the UMTS LongTerm Evolution: From Theory to Practice SecondEdition. Wiley: New York, 2011. 136–158.

25. Oppolzer J, Bestak R. Cell identifier assignment intwo-tier femtocell networks. IEEE 77th VehicularTechnology Conference (VTC Spring) 2013; 1–5.DOI:10.1109/VTCSpring.2013.6692470.

26. Oppolzer J, Bestak R. Physical cell identifier assignmentin LTE-advanced networks. 2012 5th Joint IFIP Wire-less and Mobile Networking Conference (WMNC)2012; 70–74. DOI:10.1109/WMNC.2012.6416150.

27. Fortes S, Aguilar-García A, Barco R, et al. Manage-ment architecture for location-aware self-organizingLTE/LTE-a small cell networks. IEEE CommunicationsMagazine 2015; 53(1): 294–302. DOI:10.1109/MCOM.2015.7010548.

28. Savazzi S, Nicoli M, Carminati F, Riva M. A Bayesianapproach to device-free localization: modeling andexperimental assessment. IEEE Journal of SelectedTopics in Signal Processing 2014; 8(1): 16–29.DOI:10.1109/JSTSP.2013.2286772.

29. 3GPP.Requirements for support of radio resourcemanagement. TS 36.133 V10.6.0, Mar 2012.

30. Lopez-Perez D, Guvenc I, Chu X. Mobility man-agement challenges in 3GPP heterogeneous net-works. IEEE Communications Magazine 2012;50(12): 70–78. DOI:10.1109/MCOM.2012.6384454.

31. Xu Z, Zhu X, Mahato SB, et al. Dynamic userequipment-based hysteresis-adjusting algorithm inLTE femtocell networks. IET Communications 2014;8(17): 3050–3060. DOI:10.1049/iet-com.2014.0277.

32. 3GPP.Mobility enhancements in heterogeneousnetworks (release 11). TR36.839 v11.1.0, Jan 2013.

33. 3GPP. Small cell enhancements for E-UTRA andE-UTRAN physical layer aspects (release 12). TR36.872, May 2013.

AUTHORS’ BIOGRAPHIES

1986

Zhu Xiao received his PhD degree andMS degree in Communication andInformation System from XidianUniversity, China, in 2010 and 2007,respectively. From May 2010 toFebruary 2012, he was a researchfellow in the Department of ComputerScience and Technology, Universityof Bedfordshire, UK. Currently, he isa lecturer with the College of ComputerScience and Electronic Engineering,

Hunan University, China. His research interests includewireless communications, wireless localization, small cell,and heterogeneous networks.

Wirel. Commun. Mob. Comput. 2016

Tong Li received the BS degree inCommunication Engineering fromHunan University, China, in 2014. Heis currently pursuing the MS degreein Communication and InformationSystem at Hunan University. Hisinterests focus on power and resourcemanagement in future wireless andcellular networks.

Wei Ding has graduated from LutonUniversity in the UK for his postgrad-uate degree and became a researcherin the University shortly after. Hejoined Ranplan Wireless NetworkDesign Ltd as a research engineer whileparticipating in EU FP7 framework.Currently, he is also a part time PhDstudent in University of Sheffield, UK.

DongWang received the BS degree andPhD degree in Computer Science fromHunan University in 1986 and 2006.From December 2004 to December2005, he was a visiting scholar in Univer-sity of Technology Sydney, Australia.
Since 1986, he has been working withHunan University, China. Currently, heis professor of Hunan University. Hismain research interests include network
test and performance evaluation, wireless communicationsand mobile computing, and so on.

Jie Zhang received his MEng and
PhD degrees from the Department ofAutomatic Control and ElectronicEngineering, East China University of Science and Technology, Shanghai,China. Currently, he is a professor inthe Department of Electronic andElectrical Engineering, University ofSheffield, UK. Prof. Zhang foundedCWiND (Centre for Wireless Network
Design) in 2006. With his students and colleagues, hedeveloped CWiND into a well-known and world-leadingresearch group in radio propagation modeling, indoor,outdoor, indoor–outdoor radio network planning andoptimization, femtocell, SON, and so on in a short time.His research interests cover radio propagation, 5G, self-organizing network (SON), smart building, smart city andsmart grids, and so on.

; 16:1972–1986 © 2016 John Wiley & Sons, Ltd.DOI: 10.1002/wcm

http://dx.doi.org/10.1109/VTCSpring.2013.6692470

http://dx.doi.org/10.1109/WMNC.2012.6416150



http://dx.doi.org/10.1109/JSTSP.2013.2286772


http://dx.doi.org/10.1049/iet-com.2014.0277

dynamic pci allocation on avoiding handover …home.cse.ust.hk/~tliay/paper/pci-2016.pdfresearch...

Documents