Mobility Load Balancing Enhancementfor Self-Organizing Network over LTE System

Sangchul Oh1(B), Hongsoog Kim1, Jeehyeon Na1, Yeongjin Kim1,and Sungoh Kwon2

1 5G Giga Communication Research Laboratory,Electronics and Telecommunications Research Institute (ETRI),

Daejeon 305–700, Korea{scoh,kimkk,jhna,yjkim}@etri.re.kr

2 School of Electrical Engineering, University of Ulsan, Ulsan 680–749, Koreasungoh@ulsan.ac.kr

Abstract. The objective of mobility load balancing (MLB) is to intelli-gently spread user traffic load out on a network in order to avoid degrada-tion of end-user experience and performance due to overloaded or con-gested cells. The load standard deviation (LSD) as a new key perfor-mance indicator (KPI) for MLB performance evaluation has been pro-posed in the paper. The paper aims to minimize the LSD over networklevel for equally spreading cell load out on a network with low radioresource control (RRC) signaling load. To support the MLB functionenhancement for self-organizing network (SON), the novel MLB algo-rithm has been proposed in this paper. The performance of the algorithmhas been analyzed and compared through computer simulations as well.

According to the results, we found that the proposed MLB algorithmcan reduce the LSD from 7.48 % to 60.74 %. On the other hand, weobserved that the proposed MLB algorithm required 10.79 % more han-dovers than the non-MLB operation.

Moreover, the overall number of RLF was produced as many as 140from the proposed MLB algorithm operation. This information indicatesthat MLB and mobility robustness optimization (MRO) coordination isneeded for reducing the number of RLF. Furthermore, looking at theimpacts of RLF, we can conclude that MRO should have higher prioritythan MLB.

Keywords: LTE · Self-organizing networks · Mobility load balancing

1 Introduction

A small cell access point (AP) is a nomadic or mobile access point with smallsize supporting cheap wired and wireless convergence services by connecting amobile phone with the Internet in indoor environments such as home and office.Although it is similar to a Wi-Fi access point in a functional aspect point ofview, it is different that a primary role of a small cell access point is to relay amobile phone call unlike a Wi-Fi AP.c© Springer International Publishing AG 2016O. Galinina et al. (Eds.): NEW2AN/ruSMART 2016, LNCS 9870, pp. 205–216, 2016.DOI: 10.1007/978-3-319-46301-8 17

206 S. Oh et al.

Small cell deployments are expected to occur in different scenarios, such asvenues, enterprise, dense urban residential and business areas. These scenariosinclude both uncoordinated deployments with customer and coordinated deploy-ments with the operator, or small cell vendors.

The self-organizing network (SON) function includes both network self-configuration and self-optimization functions. In a SON architecture aspect, SONfunctionalities are divided into two types: a distributed SON (dSON) and a cen-tralized SON (cSON).

The dSON algorithm is executed locally at the individual small cell to geta fast interaction with the direct neighborhood. UE specific information andrich input data sets are easier implemented locally by dSON. The dSON entitiesautonomously adjust local parameter settings within each Home eNB (HeNB),or interacting with neighbor HeNBs over X2 interface. On the other hand, thecSON algorithm requiring wide area visibility and parameter settings, withoutthe need for fast response times is performed at a central server (a.k.a. SONserver).

Therefore, a hybrid architecture for SON combines with interacting cSONand dSON components. A cSON entity collecting information like performancecounters retrieved from the small cell could be seen as a part of the networkmanager (NM) and/or of the element manager (EM) and provides guidelinesand parameter ranges to dSON functions.

The mobility load balancing (MLB) by cell reselection in radio resource con-trol (RRC) idle mode has been studied in [1], and the MLB for RRC connectedmode has been proposed in [2–4]. Although a few problems have been addressedin these papers, the MLB problems when both serving cell and neighbor cell werein overload status have not been addressed, yet. Moreover, the additional RRCsignaling load regarding early HO should be considered in MLB algorithm inorder to apply for a live LTE network. The objective of our MLB approach is tointelligently spread a load of cells out on a network with low RRC signaling loadnevertheless in this severe overload status. The load standard deviation (LSD)as a new key performance indicator (KPI) for MLB performance evaluation hasbeen proposed as well.

The rest of this paper is organized as follows. Section 2 describes the systemmodel and assumptions considered in this paper. Details of the proposed MLBalgorithm are presented in Sect. 3. We then provide the simulation environmentsand results in Sect. 4, and offer some concluding remarks in Sect. 5.

2 System Model

There are two load balancing (LB) strategies to perform load balancing by SONfunction over LTE system level. One is to modify downlink (DL) power suchas reference signal power or antenna tilt. However, degradation of coverage inreduced power cells can be archived by this strategy, and over-provisioning ofpower amplifiers in increased power cells can be required. This can be one ofthe reasons that the cost of small cell AP is increased. The other is to modify

MLB Enhancement for Self-Organizing Network over LTE System 207

handover (HO) parameter such as cell individual offset (CIO). The CIO is atarget parameter for mobility load balancing (MLB). Although this strategy canovercome the cons of DL power modification, free resource of neighbor cell isneeded to archive the pros of load balancing.

With regard to MLB, early HO is needed in case of overloaded cell, anddelayed HO is needed in case of normal loaded cell for MLB optimization.

The handover procedure to provide mobility management is one of the mainfeatures in the 3rd generation partnership project long term evolution (3GPPLTE). The HO procedure in 3GPP LTE is started with measurement report(MR) message transmission from a UE to its serving cell in a first stage. The UEmonitors the reference signal received power (RSRP) of reference symbols or ref-erence signal received quality (RSRQ) on downlink channel based on cell-specificreference signals (C-RS) periodically. If the conditions for MR transmission issatisfied with the signal strength, the UE sends the corresponding MR indicatingthe triggered event. In this paper, the event A3 and MR messages in [6] are usedfor our MLB optimization. Equation (1) shows the entering condition of eventA3 in case that a neighbouring cell becomes offset better than PCell.

Mn + Ofn + Ocn − Hys > Mp + Ofp + Ocp + Off, (1)

where Mn is the RSRP(or RSRQ) of the neighbouring cell. Ofn is the frequencyspecific offset of the frequency of the neighbour cell. Ocn is the cell specific offsetof the neighbour cell. Mp is the RSRP(or RSRQ) of the primary cell (PCell). Ofp

is the frequency specific offset of the primary frequency. Ocp is the cell specificoffset of the PCell. Hys is the hysteresis parameter for this event. Off is theoffset parameter for this event (i.e. a3-Offset).

In this paper Ofn and Ofp are ignored, since inter-frequency scenario isexcluded. Off is also removed, since target parameters for our MLB algorithmare Ocn and Ocp only. If there are multiple neighbors along with a UE, a UEhas several Mns, Ofns, and Ocns. They are considered as candidate target cellsin this paper.

Equations (2) and (3) show the resource block utilization ratio (RBUR) overa HeNB level and a network level respectively.

ρhk(t) =

∑mj=1 NUEj

(t)C × B

, (2)

where ρhk(t) indicates the RBUR on tth time slot over kth HeNB. NUEj

(t) indi-cates the number of the physical resource block allocated to jth UE connectedto kth HeNB on tth time slot. m is the number of UE connected to kth HeNB.C is the total number of carrier per a HeNB. B is the total number of physicalresource block (PRB) per carrier.

ρ(t) =K∑

k=1

ρhk(t) =

∑Mi=1 NUEi

(t)C × B × K

, (3)

208 S. Oh et al.

where ρ(t) indicates the RBUR on tth time slot over overall network level.NUEi

(t) indicates the number of the physical resource block allocated to ithUE on tth time slot over overall network level. M is the total number of UE overoverall network level. K is the total number of HeNB over overall network level.

The LSD is defined as

σ =

√∑n

t=1(ρ(t) − ¯ρ(t))2

n, (4)

where ¯ρ(t) =∑n

t=1 ρ(t)

n and n is the total number of time slot t.Our proposed MLB algorithm aims to minimize σ over network level in order

to spread their load equally out with low RRC signaling load.

3 Proposed MLB Algorithm

As stated before, the LSD in Eq. (4) is defined newly as a key performance indi-cator (KPI) in terms of MLB algorithm performance evaluation. An appropriateUE for early HO should be selected to satisfying this goal in the first stage, andthen CIO values of each UE should be adjusted within safty range to avoid aradio link failure (RLF) for early HO.

Table 1 shows Q-OffsetRange (a.k.a. cell individual offset CIO), defined in [6].The values are in dB. Value dB-24 corresponds to -24 dB, dB-22 corresponds to-22 dB and so on. The Q-OffsetRange was used to indicate a cell specific offsetsuch as Ocn or Ocp to be applied when evaluating triggering conditions for eventA3 measurement reporting.

Table 1. Q-OffsetRange

Q-OffsetRange ::= ENUMERATED {dB-24, dB-22, dB-20, dB-18,dB-16, dB-14, dB-12, dB-10, dB-8, dB-6, dB-5,dB-4, dB-3, dB-2, dB-1, dB0, dB1, dB2, dB3, dB4,dB5, dB6, dB8, dB10, dB12, dB14, dB16, dB18,dB20, dB22, dB24}

Figure 1 explains the basic concept and procedure of the MLB algorithm.Given that five UEs are connected to the HeNB1, three UEs are located at celledge area and two UEs are located at the center of cell area. When the RBURρHeNB1 of HeNB1 is more than overload threshold THoverload, the proposedMLB Algorithm 1 is performed over the HeNB1. If RRCConnectionReconfigura-tion in order to adjust Ocn or Ocp of event A3 should be sent to all UEs, heavyRRC signaling load is generated. To avoid this drawback, the appropriate userequipment (UE) selection procedure located in cell edge by periodic MR wasadded into our MLB algorithm. The UEs at cell edge are with RSRP values less

MLB Enhancement for Self-Organizing Network over LTE System 209

Fig. 1. The proposed MLB algorithm concept

than THserving and more than THtarget. Moreover, if both the serving cell andthe neighboring cell are in a similar overloaded status, the conventional MLBoperation make unnecessary HO or ping-pong HO as well as signaling overload.Therefore, we took THloadgap into account to resolve this defect. Ocn of eventA3 over the target cell is increased and Ocp of event A3 over the target cell isdecreased. Consequently early HO has been made forcibly.

Where H indicates the set of HeNBs. H = {h1, h2, ..., hK}, where K is thenumber of the HeNBs. U indicates the set of UEs located at cell edge with RSRPvalues less than THserving and more than THtarget. U = {u1, u2, ..., uM}, whereM is the number of the UEs. CT indicates the set of candidate target cells belongto U. CT = {ct1, ct2, ..., ctL}, where L is the number of the candidate target cellsbelong to U. ρs is the load of serving cell. ρctl is the load of lth candidate targetcell. THoverload is the threshold value with respect to overload. THloadgap isthe threshold value with respect to load differentiation between source cell andtarget cell.

210 S. Oh et al.

Algorithm 1. The proposed MLB algorithm1: Collecting the resource usage information of H;2: for h1 to hK do3: if ρhk > THoverload then4: Finding current active U located at cell edge and belong to hk;5: Finding CT belong to U;6: for u1 to uM do7: for ct1 to ctL do8: if ctl = max. RSRP then9: if (ρs − ρctl) > THloadgap then

10: Ocn += ΔQ-OffsetRange of A3 event over ctl;11: Ocp –= ΔQ-OffsetRange of A3 event over ctl;12: send RRCConnectionReconfiguration to um

13: end if14: end if15: end for16: end for17: end if18: end for

4 Simulation

4.1 Simulation Environments

The simulations assume the small cell parameter values modified from scenario#3 (dense) in [7] to consider practical smartphone test environment (LTE FDDBand 3). This paper simulates an urban environment where the users are walkingat 1 m/sec with circular footprint. Table 2 describes the simulation environmentsusing the detailed parameters and assumptions in this paper. In this work, weimplemented the MLB algorithm with an LTE model of the NS-3 simulationenvironment in [8]. Moreover, the RLF in [5] was implemented, since there wasno RLF implementation in current version of NS-3 simulation environment.

Figure 2 shows the cell layout topology for the simulation. The load status ofHeNBs is expressed by green (normal status) and red (overload status) colors.The overload threshold was 0.7. This means that HeNB goes into the overloadstatus if it is spending PRB over 70 % comparing to total own PRB. The evolvedpacket core (EPC) includes core network interfaces, protocols and entities suchas the mobility management entity (MME), the serving gateway (S-GW), andthe packet data network gateway (PDN-GW). the application server provideUEs with LTE application service. Video service regarding QCI 2 (GBR videocall) is implemented in this simulation.

4.2 Simulation Results

Figure 3 presents RBUR ρ(t) per HeNB without an MLB algorithm. As we cansee in the figure, the fluctuation of RBUR per HeNB is kept on very low status,

MLB Enhancement for Self-Organizing Network over LTE System 211

Table 2. Simulation Parameters and Assumptions

Parameters Settings

System bandwidth per carrier 20MHz,Number of PRB (Physical Resource Block) = 100

Carrier frequency 1.8 GHz,[Band 3 frequency (fc)]DL = 1842.5 MHzUL = 1747.5 MHz

[Band 3 EARFCN]DL = 1575UL = 19575

Carrier number 1

Total BS TX power 24 dBm

Number of cells 10

Inter-site distance (ISD) 30m

Antenna pattern Omni-directional

Number of UEs 80

Path loss (PL)(non-line-of-sight (NLOS))

147.4 + 43.3 log10(R),distance R in km

Shadowing standard deviation (σ) = 4 dB

UE mobility model circular way (radius = 10m, speed = 1 m/sec)

deployment scenario urban

MAC scheduler channel and qos aware (CQA)

UE dropping * 50 % mobile UE: Randomly and uniformlydistributed over whole area.

* 50 % fixed UE: Randomly and uniformlydistributed over limited handover available area(cell edge) only.

THoverload 0.7

THloadgap 0.1

Hysteresis 2dB

time-to-trigger (TTT) 256msec

Simulation time 605 sec

since there is no MLB operation as well as identical UE circular mobility patternwhen time goes by.

However, we can see the big fluctuation of each RBUR in Fig. 4 due to theeffect of MLB operation. The overload compensation at network scope is per-formed by the forced early HO depending on MLB algorithm.

Figure 5 describes average RBURs ¯ρ(t) per HeNB with MLB and withoutMLB. The overall average values of RBUR with MLB operation and without

212 S. Oh et al.

Fig. 2. Simulation cell layout topology

Fig. 3. RBUR ρ(t) per HeNB without MLB

MLB Enhancement for Self-Organizing Network over LTE System 213

Fig. 4. RBUR ρ(t) per HeNB with MLB

Fig. 5. Average RBUR ¯ρ(t) per HeNB

214 S. Oh et al.

MLB operation have 8.1 and 7.7. While, their standard deviation values are 0.05and 0.12, respectively. This indicates we can get smaller deviation of averageRBUR by MLB operation. It means that MLB operation provides safety loadbalancing on network scope, although MLB operation requires bigger averagevalue of RBUR.

Fig. 6. Load standard deviation (LSD) KPI for MLB gain

As mentioned in previous section, we defined the LSD for new KPI in termsof MLB performance evaluation. Figure 6 shows we can reduce the LSD frommin. 7.48 % to max. 60.74 % by the MLB algorithm when time is increased.This information indicates that we obtained safety load balancing performanceincrease from min. 7.48 % to max. 60.74 % by the proposed MLB algorithm.

The MLB algorithm makes early HO by adjusting CIO parameters such asOcn and Ocp forcibly, so that more number of HO is need. We can call this MLBcost for MLB operation. As can be seen in Fig. 7, we found that extra HOs areneeded more as much as 10.79 % for the MLB operation.

Figure 8 presents the RLF count per HeNB generated by early HO that isthe result of MLB operation. The overall number of RLFs is 140 and HeNB1 hasthe greatest number of RLFs since there are most of UEs that are the targetsfor early HO around HeNB1. This RLF count took a consideration into MLBcost for normal MLB operation. This histogram shows us that MLB and MROcoordination is needed for avoiding RLF increase prior to MLB operation. MROshould have higher priority than MLB for the clean RLF.

MLB Enhancement for Self-Organizing Network over LTE System 215

Fig. 7. Overall handover count for MLB cost

Fig. 8. RLF count for MLB cost

5 Conclusions

In this paper, we discussed important properties such as MLB gains and coststhat take place in the proposed MLB algorithm operation over LTE system.The objective of this paper was to minimize LSD over network level in orderto spread their load equally out with low RRC signaling load. To analyze and

216 S. Oh et al.

evaluate the performance of the proposed MLB algorithm, LTE small cell urbanenvironments where users are walking at 1 m/sec with circular footprint weresimulated.

As we can see in the simulation results of Sect. 4.2, we found that the pro-posed MLB algorithm can reduce LSD from min. 7.48 % to max. 60.74 %. On theother hand, we observed that 10.79 % of handovers happened more for the pro-posed MLB operations compared to the number of handovers without an MLBalgorithm. Moreover, the overall RLFs are produced as many as 140 times fromthe proposed MLB algorithm operation. This information indicates that MLBand MRO coordination is needed for reducing the number of RLF. Furthermore,looking at the impacts of RLF, we can conclude that MRO should have higherpriority than MLB.

We leave the study of the performance evaluation with regard to RRC sig-naling load impact and threshold of load gap THloadgap effect in the proposedMLB algorithm for further research.

Acknowledgments. This work was supported by Institute for Information &communications Technology Promotion(IITP) grant funded by the Korea govern-ment(MSIP)(No. R0101-16-244, Development of 5G Mobile Communication Technolo-gies for Hyper-connected smart services.

References

1. Yamamoto, T., Komine, T., Konishi, S.: Mobility load balancing scheme based oncell reselection. In: The Eighth International Conference on Wireless and MobileCommunications Mobility, pp. 381–387. IARIA, Venice (2012)

2. Zia, N., Mitschele-Thiel, A.: Self-organized neighborhood mobility load balancingfor LTE networks. In: IFIP Wireless Days, pp. 1–6 (2013)

3. Lobinger, A., Stefanski, S., Jansen, T., Balan, I.: Coordinating handover parameteroptimization and load balancing in LTE self-optimizing networks. In: IEEE Vehic-ular Technology Conference Spring, pp. 1–5. IEEE Press, Yokohama (2011)

4. Marwat, S., Meyer, S., Weerawardane, T., Goerg, C.: Congestion-aware handover inLTE systems for load balancing in transport network. ETRI J. 36, 761–771 (2014)

5. 3GPP: Evolved Universal Terrestrial Radio Access (E-UTRA); Requirements forsupport of radio resource management. TS 36.133, 3GPP (2010). http://www.3gpp.org/DynaReport/36133.htm

6. 3GPP: Evolved Universal Terrestrial Radio Access (E-UTRA); Radio Resource Con-trol (RRC); Protocol specification. TS 36.331, 3GPP (2010). http://www.3gpp.org/DynaReport/36331.htm

7. 3GPP: Small cell enhancements for E-UTRA and E-UTRAN - Physical layeraspects. TS 36.872, 3GPP. (2013). http://www.3gpp.org/DynaReport/36872.htm

8. NS3: discrete-event network simulator 3. version 3.23, NS3 Consortium (2015).https://www.nsnam.org/