article3

Issue Date: September 2004Issue Date: September 2004

© 2004 Bechtel Corporation. All rights reserved. 17

INTRODUCTION

Since first deployed in 1992, European globalsystem for mobile communication (GSM)

networks have become a major commercialsuccess. Currently, penetration levels approach100 percent in some European countries. Therapid increase in subscriber numbers promptednetwork operators to increase investment innetwork infrastructure by embarking onaggressive network rollout projects aiming toexpand system coverage and capacity. Increasingdemand for mobile services and competition formarket share led many operators to dedicatemost of their resources to network deployment.Under these circumstances, the industryeventually developed the mindset of “roll outnow and optimize later.”

Until the late 1990s, most network operatorscompeted purely on coverage, considered themost important differentiator among the offeredservices. Having “signal bars” on phones was allthat mattered, even though calls were failing inmany cases. Users expected to see signal bars ontheir phones everywhere. Focused mainly on anagressive buildout strategy, several operatorscontinued to use default parameter valueswithout fully exploiting the available function-ality of the given system.

Network pre- and post-launch optimization is auseful mechanism to ensure good performanceafter commercial launch of the service. However,as the network expands and traffic increases, thebenefits of post-launch optimization may be lost.

Ongoing changes in functionality and parametersettings are necessary to provide optimum andconstant quality of service (QoS). All systemvendors continuously seek to improvefunctionality by adding improved features withevery base station subsystem (BSS) softwarerelease. If fully exploited, this continuousevolution of functionality can result insignificantly improved QoS and more efficientuse of network infrastructure.

This paper describes a review of the functionalityand parameter values of an Ericsson™ GSMnetwork containing approximately 150 basetransceiver stations (BTSs) with three cells perBTS and main sector configuration of 0-120-240degrees [1]. This review began as an optimizationproject 6 months after completion of the post-launch optimization phase. During this period,traffic increased substantially and the networkwas expanded to satisfy capacity demand as wellas to extend coverage.

A number of features were evaluated and fine-tuned. These features are listed below, followed bya short technical description of each and thephilosophy of the performed changes, the exactsettings selected, and a statistical evaluation of theresults. The seven functions discussed below applyto Ericsson BSS R8 and R9 (Releases 8 and 9).

• Frequency hopping

• Mobile station dynamic power control

• Cell load sharing

• Locating penalty timers

• Flow control timers

Michael Pipikakis [email protected]

GSM FUNCTIONALITY AND PARAMETER FINE-TUNING:A CASE STUDY

GSM FUNCTIONALITY AND PARAMETER FINE-TUNING:A CASE STUDY

Abstract—This paper documents the effect on the performance of an Ericsson™ global system for mobile communication (GSM) networkrealized by evaluating functionality and fine-tuning parameters after completing the pre- and post-launch optimization phases. Theseactions were carried out during a base station subsystem (BSS) performance optimization project attempting to further improve the qualityof service (QoS) because network traffic was increasing. Each feature changed is addressed separately, followed by a short technicaldescription of the philosophy of the changes performed, the exact settings selected, and a statistical evaluation of the results. This paperdemonstrates that network performance gains can be achieved from optimum use of the available functionality and parameter tuning. Thispaper can also serve as a reference for optimizing Ericsson CME201 systems.

________________________________

1 Ericsson’s GSM application

Bechtel Telecommunications Technical Journal 18

• Cell selection and access

• Signal strength measurement criteria in thelocating algorithm

FREQUENCY HOPPING

Frequency hopping (FH) means that multiplefrequencies are used to transmit speech or

data in a single connection. The basic principleinvolves transmitting consecutive bursts at

different frequencies. Each cell uses a predefinedset of frequencies, among which the connectionhops according to a specified pattern (i.e., cyclicor random) 217 times per second. The radioenvironment between a mobile station (MS) and aBTS is subject to variations due to multipathfading and cumulative interference. FH canimprove the radio environment, providingfrequency diversity against the multipath fadingand averaging the overall interference. See [2].

ABBREVIATIONS, ACRONYMS, AND TERMS

BCCH broadcast control channel

BSC base station controller

BSIC base station identity code

BSS base station subsystem

BTS base transceiver station

C/I carrier-to-interference (ratio)

CLS cell load sharing

CP central processor

CTR cell traffic recording

DCR dropped call rate

DPC dynamic power control

FH frequency hopping

GSM global system for mobile communication

HSN hopping sequence number

MAHO mobile assisted handover

MS mobile station

QoS quality of service

SACCH slow associated control channel

SDCCH standalone dedicated control channel

TCH traffic channel

UL uplink

ERICSSON PARAMETERS

ACCMIN minimum signal strength to access the cell

BSRXSUFF received by the BTS sufficient signal strength level

CLSACC CLS traffic accept

CLSLEVEL CLS level

EVALTYPE evaluation type

HIHYST high-signal-strength hysteresis

HODWNQA handovers due to downlink signal quality

HOTOKCL handovers to K cells

HOTOLCL handovers to L cells

HOUPLQA handovers due to uplink signal quality

HYSTSEP signal strength level between high and low strength cells

KHYST K-criterion hysteresis

LCOMPUL uplink signal strength compensation factor

LHYST L-criterion hysteresis

LOHYST low-signal-strength hysteresis

MSRXSUFF received by the mobile sufficient signal strength level

PSSBQ penalty value for bad quality

PSSHF penalty value for failed handover

PTIMBQ penalty timer for bad signal quality

PTIMHF penalty timer for handover failure

QCOMPUL uplink signal quality compensation factor

QDESUL quality desired for uplink

RHYST region hysteresis

RXLEV measured signal strength level

RXQUAL measured signal quality

SSDES signal strength desired

TALLOC time between TCH allocations

TCALLS counter for TCH allocation attempts

TCONGAS congestion timer for immediateTCH assignments

TCONGHO congestion timer for handoverTCH assignments

TURGEN time for urgent handover

Parameter Adjustment and EvaluationCyclically sequenced baseband FH wasintroduced at launch in traffic channels (TCHs)and standalone dedicated control channels(SDCCHs). With this pattern, all availablefrequencies of a cell are used with a consecutiveorder in a call or signaling connection. Forinstance, a connection in a three-frequency (f1, f2,f3) cell will show the following burst-to-burstpattern:

…f3, f2, f1, f3, f2, f1, f3, f2, f1, f3, f2, f1,…

With reuse-pattern frequency planning, cyclichopping may result in connections in cells thatare reusing the same frequencies to get in phasewith one another, hopping “hand in hand” butlosing the benefit of interference averaging.

Random FH was proposed, which introduces apseudo random hopping sequence, according toparameter hopping sequence number (HSN). Upto 63 different FH patterns not correlated withone another can be defined. The burst-to-burstpattern would look as follows:

…f3, f1, f2, f2, f1, f3, f3, f2, f1, f1, f2, f1,…

Carefully choosing HSN values for cells using thesame frequency groups was expected to increasethe interference averaging gains of FH.

Random FH was introduced in all cells, with anHSN per cell based on the base station identitycode (BSIC) plan (HSN = 63 – BSIC), which wasalso planned to differentiate between co-channelcells.

Old values: HOP = ON, HSN = 0 => Results incyclic hopping (HOP is the Ericsson cell levelparameter to enable hopping)

New values: HOP = ON, HSN = (63 – BSIC) =>Results in random hopping

As can be seen from the results in Figure 1, aconsiderable improvement in QoS was achieved.The dropped call rate (DCR) decreased byapproximately 20 percent.

MOBILE STATION DYNAMIC POWER CONTROL

MS dynamic power control (DPC) is a featurethat controls the output power of an MS so

that the BTS receives a desired uplink signalstrength level. MS DPC helps reduce MS batteryconsumption, protects against possible BTSreceiver saturation, and reduces overall uplinkinterference.

The MS DPC algorithm is implemented on thebase station controller (BSC) and performed for

both TCHs and SDCCHs. The algorithmcalculates a power order according to BTSreceived signal strength and BTS measuredquality. The first term introduces MS powerreduction based on a desired value—signalstrength desired (SSDES)2. The second termintroduces compensation for bad quality,according to a desired value for signal quality—quality desired for uplink (QDESUL)2. The MSpower capabilities are a limiting factor. The MSpower cannot be reduced beyond the minimumoutput power of the MS (for phase 2 MSs, thedynamic power range is 8 dBm to 33 dBm).

Parameter Adjustment and EvaluationMS DPC was initially introduced with thefollowing settings for desired values andweighting factors:

• SSDES = –94 dBm

• QDESUL = 10

• Uplink signal strength compensation factor(LCOMPUL)2 = 50

• Uplink signal quality compensation factor(QCOMPUL)2 = 30

These initial values correspond to an aggressivepower-down regulation aiming to minimizeuplink interference. However, it was observedfrom analyzing drive test files and cell trafficrecording (CTR) files that the settings could leadto performance deterioration. For instance, aconnection with received signal strength(RXLEV)2 = –80 dBm and received signal quality(RXQUAL)2 = 5, given the previous settings,would be further down-regulated in steps of2 dB, despite the obvious quality problem.

After studying the case, a more reasonable valueof SSDES = –88 dBm was introduced, whileQDESUL was set to zero. Also, compensationfactor LCOMPUL, which introduces a slope in the

September 2004 • Volume 2, Number 2 19

1/12 1/22 2/1 2/11 2/21 3/3 3/13 3/23 4/2 4/12Date (MM/DD)

Figure 1. Network DCR After Implementation of Random Hopping Sequence

2.3

2.0

1.7

1.4

1.1

0.8

0.5

DCR

(%)

Cyclic FH Random FH

Considerablenetwork

performancegains can bemade by fullyutilizing the

availablefunctionality

and fine-tuningthe networkparameters

using statistics to evaluate the results.

_________________________________

2 An Ericsson DPC parameter

power reduction, was set to 100. This settingcorresponds to maximum uplink regulation (noslope) because the algorithm was expected towork rapidly on “good” signals. Qualitycompensation parameter QCOMPUL was set to60 to enhance up-regulation in case of inter-ference and to give the connection a chance toovercome the bad quality by increasing theoutput power. For a more detailed description ofthe algorithm, see [3].

Figure 2 shows the positive effect of the changeson the MS DPC settings in terms of droppedconnections due to uplink quality and uplinksignal strength. The indices “min-ERLANG/UL_QUAL-DROP” (minutes of traffic carriedbefore a call drop due to uplink signal qualityoccurs) and “min-ERLANG/UL_SS-DROP”(minutes of traffic carried before a call drop dueto uplink signal strength occurs) were used.

The indices presented in Figure 2 are TCH dropsdue to bad uplink quality and low uplink signalstrength related to the traffic carried by thesystem. The indices “minErl/UL_QA_DROP”and “minErl/UL_SS_DROP” express the minutesof traffic the system carries before a drop occursdue to bad uplink quality or low uplink signalstrength. The minute-Erlang method was usedbecause it is more sensitive to changes and thusmore accurately evaluates the effectiveness of

optimization activities. The minute-Erlang perdrop index is inversely proportional to the DCR index.

Due to the new settings for SSDES, the averagepower received on the uplink is greater thanbefore, so the risk of a connection dropping due toweak signal strength on the uplink shoulddecrease. Since the main reason for uplink qualityis also believed to be the strength of the MStransmitted signal power, bad quality drops on theuplink should also decrease with the new settings.

In Figure 3, the improvement trend can beverified by examining the handover reasons dueto uplink (UL) quality.

CELL LOAD SHARING

Cell load sharing (CLS) is a feature thatdistributes traffic among neighboring cells at

high traffic load to reduce congestion and betteruse the available resources.

The CLS algorithm works by monitoring trafficload for every cell in terms of idle TCHs. Whenthe number of idle TCHs in a given cell,expressed as a percentage of the total, falls belowthe CLS level (CLSLEVEL)3, traffic is shifted fromthis cell to prevent it from being congested.Connections close to the cell border, within anarea determined by region hysteresis (RHYST)3,are handed over to any neighboring cellconsidered suitable to accept traffic, i.e., whosepercentage of idle TCHs is greater than the valueCLS traffic accept (CLSACC)3.

Drawbacks of the feature are the increasednumber of handovers and a considerable increasein BSC central processor (CP) load. For a detaileddescription of the functionality and algorithm,see [4].

Parameter Adjustment and EvaluationThe CLS feature was introduced networkwide tocope with unevenly distributed traffic amongcells, to use the available resources efficiently,and to increase the total capacity. The original parameter set was CLSLEVEL = 23,CLSACC = 55, and RHYST = 75, meaning thatCLS evaluations for a cell started when the idlenumber of TCHs fell below 23 percent, while acell accepted CLS traffic only if 55 percent ormore of its resources were idle.

Statistical analysis indicated that with thesesettings, the success rate of CLS handovers waspoor, mainly because of the high values of


5/5 5/7 5/9 5/11 5/13 5/15 5/17 5/19 5/21 5/23 5/25 5/27Date (MM/DD)

Figure 2. Effect of New MS Power Control Settings

1,400

1,300

1,200

1,100

1,000

900

800

700

600

500

300

280

260

240

220

200

180

160

Num

ber o

f Min

utes

Old MS DPC Settings New MS DPC Settings

minErl/UL_QA_DROP

minErl/UL_SS_DROP

4/25 5/2 5/9 5/16 5/23 5/30Date (MM/DD)

Figure 3. Effect on Handovers due to UL Quality of new MS Power Control Settings

24

23

22

21

20

19

18

(%)

New MS DPC Settings

min

Erl/

UL_S

S_D

RO

P

_________________________________

3 An Ericsson CLS parameter


CLSLEVEL and CLSACC. A more reasonablesetting was introduced, where a cell would moreeasily accept CLS handovers (CLSACC = 25) andwould not start CLS evaluations as soon(CLSLEVEL = 15). Also, RHYST was set to 100,maximizing the area around the nominal cellborder where CLS could take place.

In Figure 4, the impact of the change can be seen.Cell load sharing became more effective, sinceCLS calculations were limited, practicallymaintaining the same number of successful CLShandovers. This development had a positiveeffect on the BSC CP load.

LOCATING PENALTY TIMERS

Penalty timers for bad signal quality(PTIMBQ)4 and for handover failure

(PTIMHF)4 specify the time in seconds for whichthe respective penalty values in decibels, namelypenalty value for bad quality (PSSBQ)4 andpenalty value for failed handover (PSSHF)4, areapplied to a cell’s neighbors.

When an urgent handover is successfullyperformed that resulted from bad quality due todownlink, uplink, or both, the originating cell ispenalized with PSSBQ decibels to preventimmediate hand-back to this cell. The original cellis penalized because bad radio conditions mightstill be in effect there; also, the original badquality cell is most likely the best cell from a strictly signal strength point of view. Under asimilar philosophy, handover to a cell where a handover failure occurred is inhibited for a timedetermined by timer PTIMHF [5].

Parameter Adjustment and EvaluationPenalty values PSSBQ and PSSHF were both setto 50 dB to remove the penalized cells from thelocating algorithm evaluations. However, thelengths of the timers, PTIMBQ = 10 sec andPTIMHF = 5 sec (original settings), were thoughtto be insufficient to give radio conditions in thepenalized cell a chance to improve. The lengths ofthe two timers must be carefully chosen, on theother hand, to predict handover performance offast-moving subscribers. A very high value maylead to call drops due to handover being inhibitedfor a time not matching the user’s mobility. Thenew time settings selected were PTIMBQ = 15 secand PTIMHF = 12 sec.

Figure 5 shows the effect of this change inhandover performance. The term “ping-pong”indicates the percentage of handovers back to the

originating channel within 10 seconds. Hand-over success improved as a direct result of reducing the possibility of attempted connectionsto a cell suffering from poor quality.

The reduction of mobile connections lost duringhandover can be seen in Figure 6. In addition tothe improvements in ping-pong effect andhandover success rate, the timer change also had a positive effect on network dropoutperformance. In a typical GSM network, nearly

3/6 3/13 3/20 3/27 4/3 4/10 4/17 4/24 5/1 5/8 5/15 5/22 5/29Date (MM/DD)

Figure 4. Effect of CLS Parameter Changes

1,000,000

800,000

600,000

400,000

200,000

0

Num

ber o

f Att

empt

s

4/17 4/20 4/23 4/26 4/29 5/2 5/5 5/8 5/11 5/14 5/17 5/20 5/23 5/26 5/29 6/1Date (MM/DD)

Figure 5. Effect of Penalty Timer Changes on Network Handover Performance

96.5

96.0

95.5

95.0

94.5

94.0

93.5

93.0

25

20

15

10

5

0

(%)

4/15 4/25 5/5 5/15 5/25 6/4Date (MM/DD)

Figure 6. Effect of Penalty Timer Changes on Percent of Mobiles Lost During Handover

0.95

0.90

0.85

0.80

0.75

0.70

0.65

0.60

0.55

0.50

(%)

Penalty Timer Changes

Success

Ping-pong


CLS AttemptsCLS Success

New CLS Settings

_________________________________

4 An Ericsson locating algorithm parameter

Ping

-pon

g H

ando

vers

(%)


30 percent of the total dropped calls occur duringhandover, which is considered a sensitive task inthe radio environment.

A more reliable way to assess the overall dropoutperformance is to determine the MSs lost duringhandover in relation to the total traffic. This datais shown in Figure 7, where a clear and steadilyincreasing trend is apparent for the index“minErl/MSLOST” (minute-Erlangs per MS lostduring handover).

FLOW CONTROL TIMERS

Flow control timer time between TCHallocations (TALLOC)5 gives the time in slow

associated control channel (SACCH) periods(480 msec) between consecutive TCH allocationattempts, from the channel allocation algorithm,if the first TCH allocation attempt fails. The timeris used during assignment when the BSCattempts to find an idle TCH for data or speechand also during handover. No candidate list isprepared from the locating algorithm before thetimer expires unless an urgency is detected, inwhich case the new list for handover is sentwithin the time specified by the timer for urgenthandover (TURGEN)5.

Parameter Adjustment and EvaluationParameter TALLOC specifies the pace at whichallocation attempts counted by the Ericsson BSCcounter for TCH allocation attempts (TCALLS)5

are repeated when congestion is counted by theEricsson congestion timer for immediate TCHassignments (TCONGAS)5 or by the Ericssoncongestion timer for handover TCH assignments(TCONGHO)5. A decision was made to changethe original setting from two SAACH periods tofour to limit the number of allocation attemptsper event (assignment or handover). Multipleallocation attempts increase the overall measured

congestion because congestion timers TCONGASand TCONGHO count every allocation attempt.By increasing TALLOC, the measured figures forcongestion during handover and assignment will be closer to the true, customer-perceivedcongestion.

Figure 8 shows the measured congestion trendafter the change was performed, indicating thatthe overall measured congestion rate during thebusy hour is reduced. Reducing the number ofchannel allocation attempts can also have apositive effect on the BSC CP load.

CELL SELECTION AND ACCESS

Some of the parameters controlling MS idle modebehavior during cell selection and system access

are critical for the system’s performance. Minimumsignal strength to access the cell (ACCMIN)6 is acell-level parameter that determines the minimumreceived signal strength at the MS required to accessthe system. When an MS first tries to camp to a cell,the MS decodes ACCMIN, which is transmitted onthe system information messages of the broadcastcontrol channel (BCCH), and compares it to theactual signal strength the MS measures. If ACCMINis higher, the MS is not allowed to camp to the cellbecause the MS is considered to be at poor radioconditions.

Parameter Adjustment and EvaluationDepending on the setting of ACCMIN, the cellradius (in idle mode) can be modified. ACCMINwas originally set to –107 dBm to improve thecustomer perception of the available coverage.However, such perceived improvement wasachieved at the risk of an increased number of callset-up failures, since MSs at poor radio conditionswere allowed to access the system. Additionally,the mobile equipment static sensitivity is limitedto approximately –104 dBm for most of thehandsets available, so lower signals are notpractically measurable.

A lower ACCMIN value also meant that fewersubscribers were able to respond to pagingmessages and that poor paging performancecould result [6].

To improve call set-up performance andminimize the risk of SDCCH droppedconnections, ACCMIN was set to the quotedmobile static sensitivity of –104 dBm and theSDCCH drop rate was monitored. The expectedimprovements were verified by a 22 percentreduction in SDCCH drops, as shown in Figure 9.

4/17 4/24 5/1 5/8 5/15 5/22 5/29Date (MM/DD)

Figure 7. Effect of Penalty Timer Changes on Mobiles Lost During Handover in Relation to Traffic

400

350

300

250

200

150

100

min

Erl/

MSL

OST


minErl/MSLOST

_________________________________

5 An Ericsson flow control parameter

_________________________________

6 An Ericsson access parameter

SIGNAL STRENGTH MEASUREMENT CRITERIA INTHE LOCATING ALGORITHM

The locating algorithm implemented in theBSC controls cell selection in dedicated (i.e.,

call) mode and determines handover decisions.The main objectives of handover are to maintaincall continuity and quality and to control cell sizeand handover borders to minimize total networkinterference.

The inputs to the locating algorithm are signalstrength and quality measurements from the MS(the so-called mobile assisted handover [MAHO])and from the BTS. The output is a list of candidatecells for handover, ranked in descending orderaccording to preferences and constraints intro-duced by other features and by the settings of thealgorithm itself. The locating algorithm workscontinuously for all active MSs and completes acycle every SAACH period (480 msec).

The signal strength measurements reported bythe MS and the BTS are evaluated according tocomparison criteria that can be selected withdifferent settings in the locating algorithm. Thefirst is the signal strength or K criterion and thesecond is the path loss or L criterion. They are usedto compare reported values for serving andneighboring cells to determine the optimum cellranking and the handover borders.

In the K-criterion mode, the comparisons areperformed purely according to the receivedsignal strength (i.e., cells measured with highersignal strength are ranked higher). Hence, anincrease in the output power of a cell signifiesexpansion of its service area. This criterion seeksto maximize the carrier-to-interference (C/I) ratioby maximizing “C.”

In the L-criterion mode, path loss is taken intoaccount. Cells with lower path loss are rankedhigher, and the output power of each cell doesnot affect the calculations. The criterion actuallyfavors cells with low output power; thus,improvement in C/I ratio is attempted bydecreasing the total interference. However,L ranking can sometimes lead to a locally lowerC/I ratio than K ranking. Two cell rankingalgorithms are available, set by BSC parameterevaluation type (EVALTYPE)7 [5]:

• Ericsson-1-2, which uses both L andK ranking. The candidate cells are separatedinto high- and low-signal cells by comparingreceived signals to the following parametersfor downlink and uplink, respectively:received by the mobile sufficient signal

strength level (MSRXSUFF)7 and received bythe BTS sufficient signal strength level(BSRXSUFF)7. High-signal-strength cells areranked according to the L criterion and therest according to the K criterion.

• Ericsson-3, where ranking is performed onlyaccording to the K criterion, but two separatehysteresis values are used.

Parameter Adjustment and EvaluationBefore this exercise, only the K criterion was usedfor handover calculations. The hysteresis was setto K-criterion hysteresis (KHYST)7 = 4 dB.Hysteresis is a signal strength offset that is addedto the actual reported value for the serving cell toprevent unnecessary ping-pong handovers at theborder between two cells. The L criterion wasintroduced in an attempt to further improvenetwork handover performance. Sufficientcondition parameters MSRXSUFF = –86 dBm andBSRXSUFF = –92 dBm determine the breakingpoint between L and K ranking. Cells reportingwith signal strength values greater than bothlevels are considered suitable for L ranking,where an increased hysteresis value, L-criterionhysteresis (LHYST)7 = 7 dB, is used. Theremaining cells are K ranked with a hysteresisKHYST = 4 dB.


4/30 5/5 5/10 5/15 5/20 5/25 5/30 6/4 6/9Date (MM/DD)

Figure 9. SDCCH Drop Rate Before and After ACCMIN Change

3.0

2.5

2.0

1.5

1.0

0.5

SDCC

H D

rop

Rat

e (%

)

5/11 5/13 5/15 5/17 5/19 5/21 5/23 5/25 5/27 5/29 5/31 6/2 6/4 6/6 6/8Date (MM/DD)

Figure 8. Difference of Measured Congestion After Flow Control Timer Change

50

45

40

35

30

25

20

15

10

5

0

(%)

Bid Congestion Handover

Bid Congestion Assignment

TALLOC = 2 TALLOC = 4

_________________________________

7 An Ericsson locating parameter

The Ericsson-3 algorithm was also tested. Themain difference from the previous K-ranked-onlyalgorithm is that, depending on the receiveddownlink signal strength, one of two hysteresisvalues is used. The signal strength level between high and low strength cells(HYSTSEP)7 = –86 dBm parameter specifieswhether the serving cell is a high or low strengthcell, allowing a larger high-signal-strengthhysteresis (HIHYST)7 = 7 dB or a smaller low-signal-strength hysteresis (LOHYST)7 = 4 dB tobe applied. The purpose of the high hysteresisvalues for both tested algorithms is to preventunnecessary handovers in the cell borders whenradio conditions permit.

In Figure 10 the reduction in the total number ofhandovers in the system due to the increasedhysteresis in both testing cases can be verified. Itis noteworthy that the L-criterion algorithmseems to introduce the highest (25 percent)reduction in the handovers, as expressed by thehandovers per call index.

Figure 11 shows the following handover areas:handovers to K cells (HOTOKCL)7, handovers toL cells (HOTOLCL)7, low hysteresis (LOWHYST),(HIHYST), handovers due to uplink signal quality(HOUPLQA)7, and handovers due to downlinksignal quality (HODWNQA)7.

The portion of handovers performed with theL criterion in the first case and with the HIHYSTvalue in the second can well justify the previousdeviation. Up to 30 percent of total handovers inboth cases take place with the use of the increasedhysteresis values, which means that thehandovers are actually delayed. The result is atotal handover reduction, if averaged over thewhole network.

As already mentioned, handover is considered atask with a high risk of call drop. Figure 12 showsthe effect of the tested settings in call dropperformance of the handover algorithm.

Handover dropouts, expressed as a percentage oftotal handovers, may initially convey that thesituation worsened with the new settings.Nevertheless, what matters is the absolutenumber of failures actually experienced by thesubscriber; since the total number of handoversdecreased, this difference is not substantial. To emphasize this point, the index“minErl/MSLOST,” giving Erlang minutes oftraffic carried out per handover dropout, is alsodepicted. Inspecting this index, it is clear that theL-criterion algorithm appears much improved,while the performance of the Ericsson-3algorithm is rather ambiguous.


5/14 5/16 5/18 5/20 5/22 5/24 5/26 5/28 5/30 6/1 6/3 6/5 6/7 6/9Date (MM/DD)

Figure 10. Handovers per Call per Evaluation Criterion

3.0

2.5

2.0

1.5

1.0

0.5

(%)

5/1 5/5 5/9 5/13 5/17 5/21 5/25 5/29 6/2 6/6Date (MM/DD)

Figure 11. Handover Causes per Evaluation Criterion

3.0 x 106

2.5 x 106

2.0 x 106

1.5 x 106

1.0 x 106

0.5 x 106

0

Num

ber o

f Han

dove

r Cau

ses

K Only K+L Ericsson-3Criterion

HOTOKCL LOWHYST HIHYST

HOTOLCL HOUPLQA HODWNQA K Only K+L Eric-3

5/1 5/4 5/7 5/11 5/14 5/17 5/20 5/23 5/26 5/29 6/1 6/4 6/7Date (MM/DD)

Figure 12. Mobiles Lost During Handover per Evaluation Criterion

0.90

0.85

0.80

0.75

0.70

0.65

0.60

0.55

0.50

280

250

220

190

160

130

100

(%)

min

Erl/

MSL

OST

K Only K+L Eric-3

Over Total Handovers

minErl/MSLOST

5/1 5/4 5/7 5/11 5/14 5/17 5/20 5/23 5/26 5/29 6/1 6/4 6/7Date (MM/DD)

Figure 13. Handover Success Rate and Ping-pong Rate per Evaluation Criterion

96.5

96.0

95.5

95.0

94.5

94.0

93.5

93.0

25

20

15

10

5

0

(%)

K Only K+L Eric-3

HO Success

Ping-pong

Ping

-pon

g H

ando

vers

(%)


The superiority of the L-criterion algorithm overthe Ericsson-3 algorithm is also apparent inFigure 13; the ping-pong handovers (i.e.,handovers back to the originating cell within10 seconds) are reduced in both cases. Thisreduction is a direct consequence of the hysteresisvalues of 7 dB introduced in both algorithms.However, the L-criterion algorithm shows thebest performance in this field, meaning that moreaccurate and reliable handover decisionsaccompany this algorithm, exactly as predictedby theory.

The only disadvantage of the L-criterion algorithmappears to be the handover success percentage,half a decimal unit below the previous figures. Thesame applies for the Ericsson-3 algorithm, whichcan be attributed to the lower number of handovercommands. It can be assumed that, due to variousradio problems, a significant number of handoverfailures always exist in the network. This assumedvalue can be highlighted or hidden, in a statisticalsense, depending on the volume of the totalsample. It is believed that careful optimization andindividual neighbor cell inspection of thenetwork’s handover performance can furtherimprove this figure.

As a result, the K-L combination algorithm waseventually introduced. Further improvement canbe achieved by fine-tuning sufficient levelparameters BSRXSUFF and MSRXSUFF toidentify a balanced breakpoint for cell ranking.Also, different LHYST values can be tried.

CONCLUSIONS

All of the optimization-related changes weremade in a controlled manner so that their

effectiveness could be measured and evaluated.At the end of the project, the average daily DCRwas reduced by 30 percent, and the averageminute-Erlang per drop was increased by almost45 percent. At that point, a foundation wascreated for further fine-tuning as the networkexpands in response to increases in traffic andsubscriber base.

As has been shown, considerable networkperformance gains can be made by fully utilizingthe available functionality and fine-tuning thenetwork parameters using statistics to evaluatethe results.

TRADEMARK

Ericsson is a trademark or registered trademarkof Telefonaktiebolaget LM Ericsson.

REFERENCES

[1] “Radio Network Parameters and Cell DesignData” – Ericsson CME20 Documentation.

[2] “User description, Frequency Hopping” –Ericsson CME20 Documentation.

[3] “User description, MS Dynamic Power Control” –Ericsson CME20 Documentation.

[4] “User description, Cell Load Sharing” –Ericsson CME20 Documentation.

[5] “User description, Locating” – Ericsson CME20Documentation.

[6] “User description, Idle Mode Behaviour” –Ericsson CME20 Documentation.

BIOGRAPHYMichael Pipikakis is a networkplanning and wireless tech-nology manager for Bechtel'sEurope, Africa, Middle East,and Southwest Asia Region. Hesupports ongoing and newprojects and new businessdevelopment; writes guidelinesand procedures for mobilenetwork design, planning, and

optimization; and participates in technology forums.

Michael is a mobile networks specialist with 17 years of experience in the telecommunications industry,including more than 11 years in RF planning, design,optimization, and management of the end-to-endperformance of cellular networks.

Before joining Bechtel, Michael held variousmanagement positions in the Vodafone Group's radiosystem design and optimization department anddevelopment department over a 10-year period;worked for Cellnet UK and GEC Marconi UK; and wasa telecommunications operator in the Greek Navy.From 1999 to 2003, he was a member of the VodafoneGlobal Forum for UMTS design harmonization.

Michael has a BEng Honors in Electronics Engineeringwith Computing and Business from KingstonUniversity in Surrey, England, and an HND in RadioCommunications Systems Design from the PolytechnicSchool of Athens, Greece. He is a member of theInstitution of Electrical Engineers.

article3

Technology