00596369

8/3/2019 00596369

1/5

DETERMINATION OF OPTIMAL HANDOVER BOUNDARIES

DISTRIBUTION ANALYSIS OF MOBILE MEASUREMENTREPORTSIN A CELLULAR NETWORK IBASED ON TRAFFIC

C. Chandra T. JeanesMotorola1501 W. Shure Dr. IL27-AR3205Arlington Heights, 11. 60004USA

Abstract: Traffic distributions in a cellular networktend to follow periodic patterns where local conges-tions occur. By optimizing handover boundaries be-tween neighboring cells, opportunities for load sharingcan be exploited to maximize the capacity of the cel-lular network as a whole. We propose a method fordetermining optimal handover boundaries by analyz-ing historical measurement reports data, as an indica-tion of traffic distribution patterns within the network;and modeling the problem as an optimization problemwith a set of nonlinear constraints.I INTRODUCTION

To achieve ubiquitous coverage and call continuitythe coverage areas among cells in a cellular networkenvironment tend to be highly overlapped. Undernormal conditions, mobiles are served by cells thatprovide the strongest signal strength and handoversare initiated when a neighbor cells received signal be-comes stronger than the servers received signal, af-ter some appropriate averaging window and hystere-sis margin to avoid ping-pong handovers. In the casethat more than one neighbor cell is stronger than theserving cell then handover is ideally attempted to thestrongest neighbor.In a live network, the traffic load experienced byneighboring cells tends to vary at different times ofthe day and commonly follows predictable patternsaccording to rush hours and centers of activities. Of-ten times, handing over to the strongest neighbor willnot achieve the most capacity out of the network be-cause some cells are more heavily loaded than others.By allowing mobiles to be served by cells with lowerreceived signals but are within an acceptable qualitylevel, localized congestion can be avoided and a highercall carrying capacity can be achieved by the com-bined network. This can be achieved by allowing lessheavily loaded cells to serve mobiles beyond its ideal

W.H. Leung

boundary and distributing the traffic load more evenlyamong the cells through the use of handover preferencemargins. These handover margins reflect the loadingof cells to offset the strongest received signal criteriafor selecting handover target cells. When the overallgenerated traffic capacity in the network stays withinthe offered load, but moving concentrations of trafficexist, then dynamic adjustment of handover bound-aries offers the least costly method for increasing lo-calized capacity, without adding additional resources(e.g. carriers, new cells).Various congestion relief techniques to handle tem-porary traffic concentrations have been proposed inthe literature. These techniques range from similarmethods of changing boundary parameters to morecostly methods that introduce additional hardware.More costly methods, such as using ada tive anten-nas to dynamically target traffic coveragefi ] have beenpropclsed, but these methods introduce new hardwareto the system. Representative works in the first cate-gory include directed retry 13, overlapping cells withchannel reuse partitioning [2 , and using an additionaloffset parameter to reflect cell loads to control han-dover decisions [4, 1. While these works propose sim-ilar schemes for moving handover boundaries t o relievelocal (congestion,a method of determining the optimalhandover margins (boundaries) to maximize the over-all network capacity has not been addressed.In this paper, we present an approach to deter-mine optimal configurations of handover boundariesbetween neighboring cells to maximize overall networkcapacity that is based on an analysis of historical mea-surement report data. Unlike the various methodsthat have been proposed, where the system performs areactive action to deal with congestion, our approachattempts to be more predictive of possible future con-gestion, by analyzing traffic patterns from previousmeasurement data.The rest of this paper is organized as follows.Section 2 provides a brief background of the basichandover algorithm in the GSM system. Section 3

0-7803-3659-3/971 0.00 1997 EEE 305

8/3/2019 00596369

2/5

presents our approach to determining optimal han-dover margins configuration to maximize network ca-pacity, by using nonlinear programming technique.Section 4 provides experimental results of this ap-proach. Finally, we conclude in section 5.I1 BACKGROUND

In GSM cellular system, the handover algorithm is0 signal level (RxLev)0 signal quality (RxQual)0 mobile distance (signal propagation delay)

pathloss difference between server and neighborwith corresponding thresholds for each handover crite-ria. While the signal level, signal quality, and mobiledistance thresholds are based on hard performance re-quirements power budget is based on a less stringentcriteria to minimize the pathloss between the servingcell and the mobile. The corresponding threshold forpower budget, the handover margin, is then used tooffset the minimum pathloss criteria to prevent ping-pong at cell boundaries and as a "preference level"among the feasible serving cells.In an environment where the traffic is unevenly dis-tributed among cells, the handover margin can be usedas an offset to prefer covering cells, which are less heav-ily loaded.

based on four types of indications:

cell (power budget)

I11 APPROACHOur approach relies on past measurement data topredict the traffic distribution patterns and fit the

handover margin threshold parameters to the mea-surement data in order to maximize the carried ca-pacity of the entire network. Our analysis is based onmeasurements of downlink power level of the servingand surrounding base stat ions, which are reported bythe mobile to its serving base station at every mea-surement period. At every measurement period, thefollowing value is calculated at the base station, basedon the measurement reports:PBGT(sln ) M P L ( s )- P L ( n )where

PBGT(s, n) = the power budget of a neigh-bor cell n with respect to server cell sPL(n) = pathloss from mobile to neighborcell n ,PL(s) = pathloss from mobile to serving cellS

WhenP B G T ( s ,n )2 , ,

where H,, is the handover margin from serving cell sto neighbor cell n , a handover caused by power budget

Yo mobiles

I

Figure 1: Distribution function of power budgetof neighbor n

is triggered. When more than one neighbor satisfiesthe above condition] then the neighbor with the high-est power budget value is selected as the target cellfor a handover. In order to set the H,, of variouscells to the appropriate values, we need to know howmuch traffic would be absorbed by neighbor n by asetting of Hsn.n approximation could be made bylooking at past measurements of mobiles in the bound-ary region of the serving cell having various values ofpower budgets for various neighbors n. Capturing thepercentage of mobiles having particular power budgetmeasurements will tell us how many mobiles are likelyto handover t o a particular neighbor cell when H,, isset to a certain value. We are only interested, however,in measurements of mobiles at the boundary regionsthat meet the following conditions:

The measurement of RxLev(n) (signal strength)for neighbor n is above RxLev-Threshold whereRxLev-Threshold is a measure of "good enough"signal level. This condition indicates whether theneighbor can adequately serve the mobile.PBGT(s,n) is among the three highest measuredfor the mobile. This condition ensures tha t themobile is handed over to cells that are close to thelocation of the mobile. Capturing only the threehighest measurements also simplifies the formula-tion of the problem, as will be seen later.

By capturing the power budget measurements ofmobiles and those measurements where the receivedsignal strength from corresponding neighbors cells arebelow RxLev-Threshold, we can determine the distri-bution function F ( P B G T ( s , ) )of mobiles that can'thandover to a given nei hbor cell when the handovermargin is set to PBGTTs,n) . On a per neighbor cellbasis, the percentage of mobiles that will be absorbedfrom the traffic currently covered by the serving cell,when H, , is set to a given power budget value is then:1- F(Hsn)

. Fig. 1 illustrates the relationship between the dis-tribution function F ( ) and the percentage of mobilesthat are absorbed by a neighbor cell.

306

8/3/2019 00596369

3/5

with the following nonlinear constraints:ranges over all neighbors of cell i:For each i , j where i ranges over all cells and j

\f Hi kT i

Figure 2: Illustration of variable definitions

We can now formulate the handover margin param-eter optimization problem as a nonlinear optimizationproblem.Let the following variables define the parameters inthe system:Ti = Traffic currently originated within cell is cover-age, including served and blocked calls (Erlangs).Si = Amount of traffic (ErXangs) from that will beretained by cell i with the optimized settings ofvarious handover margin neighbor parameters.Gjj = Amount of traffic (Erlangs) out of 2?: that celli gives up to cell j when the handover margin isset to H i j .H i j = Handover margin setting from cell i to cell j .Ri = Resource (offered load) of cell i (Erlangs).

Fig. 2 illustrates the definition of these variables.In Fig. 2 cell i is shown to have two neighbor cells: jand k. The traffic that is currently generated withincell i is bounded by the heavy circle. The handovermargins to the two neighbor cells are set to H i j andH i k respectively and with these settings, the distribu-tion of traffic that cell i is giving up to cell j and kare denoted by the cross hatched regions ( G i j , G i k ) .The traffic that cell i retains is the shaded area thatis left from the heavy circle after subtracting the crosshatched areas, and is denoted by Si . Note that sinceTi includes the blocked calls tha t are originated in celli , whereas Si, Gij, and Gik denote the amount of traf-fic that can actually be served by the cells according toeach cells offered load,Z 2 S i + ~ j , - n e i g h b o r s of Gij.

The optimization that we are interested in can beformulated using the following two objective functions:

iccel ls jeneighbors of i ieeel lsand

iccel ls jcneighbors of i

ckcneighbors of i ; k # j+ c1,mcneighbors of i ; ,m# jx - y < H i j - H i l , ~ -Z < H i j - H i m ) ) *T ,

where 2 E PBGT(i , j ) ,y E PBGT(i , ) in the firstsummation and 2 E PBGT(i, ) , y E PBGT(i , ), andz E PBGT(i,m) in the second summation.For each cell i:

Si 52?:- Gij (4 )jcneighbors of i

For each cell j :Gi j+S j 5 R j (5)

ieneighbors of jFor each neighbor cells pair i, j:

H i j +P < - H j i ( 6 )where P > 0.where Hmin and H,,, are not restricted in sign.In the above formulation, the first objective func-tion tries to maximize the overall traffic carried by allthe cells in the network. The second objective functionmakes a trade-off between maximizing capacity withmaintaining good signal quality. Equation (2) tries tokeep the setting of the handover margins as close tothe ideal as possible, where the ideal is the situationwhen a handover is generated at the point when thesignal strengths of the neighbor and serving cell areequal (handover margin equals 0). With this objectivefunction, when there are more then one lightly loadedneighbors the overflow traffic will be partit ioned to theneighbors such that the path losses are minimized.In (3 ) , F ( ) denotes the distribution function of thepower budget measurements and P ( )denotes the cor-responding joint probability functions. Equation (3 ) ,essentially states that the amount of traffic that cell igives up to cell j is bounded by the percentage of mo-biles having power budget measurements above thehandover margin setting H i j . The additional termsin the multiplier are needed to reflect the overlappingregicins of the traffic distribution among the neigh-bor cells. The first summation is the overlap be-tween any pair of neighbors, that may result in traf-fic attributed to neighbor j being actually absorbedby neighbor k. A mobile will actually be absorbedby neighbor k when its power budget measurementfor neighbor j is greater than H i j , but the differencebetween the power budget measurements for neigh-bors j and k is less than H j j - H i k . In such a case,

Also, Z , S i , G i j , R i 2 0 and Hmin 5 Hij 5 Hmaz

307

8/3/2019 00596369

4/5

For each cell i:

j cneighbors of iwhere UT , is the slack in (4), that indicates how muchmore resources are required to handle the additionaltraffic in cell i.Similarly, from (5), the following can be derived:For each cell j:

Figure 3: Relative placements of handover mar-gin boundaries and the guard margin requiredbetween mutual neighbors to avoid ping-pong

PBGT( i , )- i j < PBGT(i ,k) H i k . The joint den-sity function in the first summation term exactly re-flects the amount of traffic overlap that is absorbed bycell k out of the traffic tha t has already been at tribu tedto cell j in the first distribution function term. Thesecond summation term denotes the amount of overlapbetween any three neighbor cells. The first summationterm would have reduced any three-way distributionoverlap twice, therefore this amount has to be addedback into the percentage calculation. Since our dis-tribution functions only take into account the threehighest power budget measurements, a three variablejoint distribution is as far as we need to consider to ac-count for overlapping regions among neighboring cells.

Equation (4), is a constraint relationship amongthe traffic that each cell retains, the amount that itgives up to its neighbors and the total traffic gen-erated within its boundaries. Equation (5), reflectsthe resource constraint of each cell. This constraintstates that the amount of traffic that each cell ab-sorbs from its neighbors and the amount of traffic thatit retains (not including blocked calls) are boundedby its resources (offered capacity). Equation (6), is aconstraint on the settings of the handover margins ofneighboring cells to prevent ping-pong handovers. Fig.3 , illustrates the relative placements of the handovermargin boundaries between neighboring cells that arenecessary to avoid ping-pong. The parameter P de-notes the distance between the two handover marginboundaries.

Having formulated the handover boundaries opti-mization problem as a constrained nonlinear optimiza-tion, we can use an optimization tool to solve our ob-jective function with its associated constraints to ob-tain optimal values for the handover margin parame-ters H . . Our analysis optimizes over sets of measure-ment xata that are gathered over a period of time,exhibiting similar traffic distribution pattern. Eachdistribution pattern may then yield different sets ofoptimal handover margin settings. In the fu ture, clus-ter analysis may be used to analyze similar distribu-tion patterns.The results from the above analysis could also beused to determine the sites that need additional re-sources because of very high traffic concentration.From (4), the following can be derived:

ieneighbors of jwhere U R ~s the slack between the amount of resourcesavailable in a cell and how much traffic its serving.The slack value indicates if additional resources areavailable in the cell to possibly increase its physicalcoverage. Both of the above slack information wouldhelp in the planning and deployment of additional re-sources for various cells in the network.

IV EXPERIMENTAL RESULTSOur experimental results are based on measurementreports data on a subset of carriers within six nearbycells in a live network. The measurement reports areanalyzed for the power budget distributions of everysignificant pair of server-neighbor cells reported, wherethe significance level is set to 1%frequency of occur-rence within the entire sample. We assume a nor-mal distribution of the data, which were validated tomodel the actual measurements quite well. The num-ber of significant neighbors for each of the six cellsare 10, 16, 16, 14, 13, and 24 respectively, with a to-tal of 71 significant cells in the cluster being analyzed.RxLewThreshoZd was set to -110 dBm in determining

this distribution function. From the computed powerbudget distributions, the objective functionsand the set of constraint equations (3) , (4),are generated. The maximum and minimum allow-able handover margins were set to Hmin -10 andH,,, = 15 respectively.The set of generated constraint optimization prob-lem are then processed using MATLAB to computethe optimized handover margins to maximize the car-ried load of the network with 71 cells. The weight-ing factors for the two objective functions that werefound to work well are 1 and 85 respectively. We as-sume that the six cells have a congestion profile thatreflect blocking within 0-15%. For the remaining cells,channel utilization was set at 95%. Table 1 summa-rizes the original utilization (including blocked callsand optimized utilization (theoretical carried load) a -ter adjustments of handover margins within each cell.Table 1 also indicates the total percentage (sum) ofblocked calls and the theoretical percentage being car-ried after optimization of the handover margin. Due tospace limitation, the power budget distributions andresulting optimized handover margins between pairsof neighbor cells are not shown here. These resultsshow that quite significant capacity increase can beobtained from adjusting handover boundaries within

308

8/3/2019 00596369

5/5

Table 1: Original (including blocked calls) andoptimized utilization (carried load) within each

LOrig. 95% 95% 95% 95% 95% ,_Optim. 95% 96% 96% 100% 100%Cells 21 22 23 24 25Orig. 95% 95% 95% 95% 95%

. Optim. 100% 100% 1oo%loo%loo%~Cells 26 27 28 29 30Orig. 95% 95% 95% 95% 95%. Optim. 96% 96% 100% 100% 96% .Cells 31 32 33 34 35Orig. 95% 95%. Optim. 96% 95%Cells 36 37 I 38 I 39 I 40 fl

the acceptable quality and resource constraints of anetwork with typically high utilization.

U UOrig. 95% 95% -. Optim. 95% 95% 95 95

Cells 41 42 43 44 45Optim. 96% 95% 95% 95% 95%

1 Cells 46 47 48 49 50 IOrig. 95% 95% 95% 95% 95% .

Y Orig. 95%- Optim. 95%

Cells 51Orig. 95%

. Optim. 95%Cells 56Orig.

. Optim. 96%Cells 61Orig. 95%

. Optim. 95%Cells 66

U95%95%52 I 53 I 54 I 55 I]

95%95%57 58 59 60

95% 96% 100% 95% ~--95%95%95%95%95%62 63 64 65

95%mi5% 95 9667PE

V CONCLUSIONO ar approach assumes that traffic concentrations

withiin regions in a cellular network tend to followperiodic patterns, therefore by analyzing past mea-surement data, we should be able to plan for similartraffic distributions that tend to recur in the future.By anticipating for similar traffic distributions, local-ized congestions could be redistributed to less con-gested neighboring cells to maximize the capacity ofthe whole network.Th,e optimization model that we developed allowsthe determination of the maximum capacity that canbe achieved by moving handover boundaries; the set-ting of the handover margins to achieve that maximumcapacity; and pinpoints areas where redistribution oftraffic to overlapping cells can no longer increase thetotal carried capacity within the area, and therefore,addit ional resources may be required. Our experimen-tal results show that significant capacity increase canbe achieved by adjust ing handover boundaries withinthe prescribed quality and resource constraints of thenetwork. The tool that we developed can be used toguide the optimization of an already deployed networkto adjust handover boundaries according to changingtraffic concentrations.

ACKNOWLEDGEMENTSWe thank Kenneth Haas for writing the patent ap-plication for this work.References13. Eklundh, Channel Utilization and Blocking

]?robability in a Cellular Mobile Telephone Sys-tem with Directed Retry, IEEE Trans. of Com-munications, Vol. COM-34, Apr. 1986.Eai-Po Chu, Stephen Rappaport, OverlappingCoverage with Reuse Partitioning in Microcel-lular Communication Systems, IEEE VehicularTechnology Conference, 1995.IMitsuhiko Mizuno, Takeo Ohgane, Applicationof Adaptive Array Antennas to Radio Commu-nications, Electronics and Communications in,Japan, Part 1, Vol77, No . 2, 1994.!3S. uek, Wai-Choong Wong, Rajiv Vijayan,]David J. Goodman, A Predictive Load Sharing9cheme in a Microcellular Radio Environment,lEEE Trans. on Vehicular Technology, Vol. 42,No. 4,Nov. 1993.R. Bodin, A. Norefers, Load Sharing Controljor a Mobile Cellular Radio System, Patent No.15241685, Aug. 31, 1993.

309