analytical model for estimating the impact of changing the ...webpersonal.uma.es/~toril/files/2018...

Research ArticleAnalytical Model for Estimating the Impact of Changing theNominal Power Parameter in LTE

A B Vallejo-Mora 1 M Toril 1 S Luna-Ramırez 1 M Regueira2 and S Pedraza2

1Departamento de Ingenierıa de Comunicaciones ETS Ingenierıa de Telecomunicacion University of MalagaCampus Universitario de Teatinos sn 29071 Malaga Spain2Ericsson Severo Ochoa 51 29590 Malaga Spain

Correspondence should be addressed to A B Vallejo-Mora abvmicumaes

Received 4 January 2018 Accepted 13 May 2018 Published 14 August 2018

Academic Editor Piotr Zwierzykowski

Copyright copy 2018 A B Vallejo-Mora et al is is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

UpLink Power Control (ULPC) is a key radio resource management procedure in mobile networks In this paper an analyticalmodel for estimating the impact of increasing the nominal power parameter in the ULPC algorithm for the Physical Uplink SharedCHannel (PUSCH) in Long Term Evolution (LTE) is presented e aim of the model is to predict the effect of changing thenominal power parameter in a cell on the interference and Signal-to-Interference-plus-Noise Ratio (SINR) of that cell and itsneighbors from network statistics Model assessment is carried out by means of a field trial where the nominal power parameter isincreased in some cells of a live LTE network Results show that the proposed model achieves reasonable estimation accuracyprovided uplink traffic does not change significantly

1 Introduction

Coverage and Capacity Optimization (CCO) has beenidentified by mobile network operators as a relevant use caseof self-organizing networks [1ndash3] e aim of CCO is toensure an adequate trade-off between capacity and coverageduring network operation For this purpose CCO algo-rithms adjust key radio network parameters such as antennaconfiguration [4ndash8] or power settings in downlink [9ndash14]and uplink [15ndash25]

One of the most promising albeit challenging tech-niques for CCO in Long Term Evolution (LTE) is adjustingUpLink Power Control (ULPC) parameters ULPC de-termines user transmit power for different uplink physicalchannels It was conceived to avoid the near-far effect and toadjust received signal level to the limited dynamic range atthe receiver Without power control far users would receiveexcessive intracell interference from close users in the samecell as Single Carrier Frequency Division Multiple Access(SC-FDMA) is not totally orthogonal and it causes InterCarrier Interference (ICI) us far users with weak signal

could be drowned out by strong signals from close users dueto the limited quantization resolution of the Analog-to-Digital Coverter (ADC) at receiver By regulating usertransmit power ULPC also improves system capacity andcoverage ensuring Quality-of-Service (QoS) requirementsminimizing UL interference to other cells and reducing userbattery consumption

In the literature most studies have focused on the designand characterization of effective ULPC algorithms based onsimulations Due to its benefits the fractional ULPC (FPC)algorithm has been standardized for LTE [26] e behaviorof FPC is controlled by two parameters namely the nominalpower parameter and the path loss compensation factoreimpact of these parameters on the performance of FPC ininterference and noise limited scenarios is evaluated bymeans of simulation in [27] In [28] an analytical model isproposed for computing UL cell loads and blocking prob-abilities from the values of the nominal power parameter andthe path loss compensation factor in a simulator designed fornetwork planning purposes In [29] a theoretical frameworkfor modeling FPC in a computationally-efficient simulator is

HindawiMobile Information SystemsVolume 2018 Article ID 2458204 7 pageshttpsdoiorg10115520182458204

proposed for validating self-organizing algorithms All thesemodels are based on propagation traffic and radio linkquality predictions However to the authorsrsquo knowledge noanalytical model has been proposed to predict the impact ofchanging the referred parameters in an existing LTEscenario

In this work a novel analytical model for estimating theimpact of increasing the nominal received power parameterin the FPC algorithm for the Physical Uplink SharedCHannel (PUSCH) in LTE is presented Unlike previousmodels the proposed model is conceived for network op-timization purposes us the model is designed as anincremental model whose aim is to estimate the change inthe UL interference level and Signal-to-Interference-plus-Noise ratio (SINR) caused by modifying the current con-figuration of the nominal power parameter on a cell-by-cellbasis erefore it would be a predictive model that wouldallow the algorithm to react faster (ie in one shot) andmoreaccurately (ie no unnecessary changes and subsequentreversion) than working iteratively e predictive approachis very demanded in the current algorithms both for op-timization and for self-healing Although this model hasbeen conceived for optimization phase it could also be usedduring self-healing phase in order to compensate cells inoutage ensuring minimum UL SINR Moreover the modelcan be easily adjusted to any particular scenario based onavailable network performance statistics Model assessmentis carried out bymeans of a field trial whereP0 is increased insome cells of a live LTE network e rest of the paper isorganized as follows Section 2 outlines the FPC algorithmfor PUSCH in LTE Section 3 presents the proposed esti-mation model Section 4 shows the results of the model ina live network and Section 5 summarizes the mainconclusions

2 FPC Algorithm for PUSCH

FPC defines the transmit power of each User Equipment(UE) which is dynamically adapted to the changing radiopropagation conditions and fluctuations of interferencefrom users in neighbor cells e equation defining the valueof UE transmit power P (in dBm) in each PUSCH subframeis

P min Pmax P0 + αPL1113980radicradicradic11139791113978radicradicradic1113981open loop

+ΔTF + f ΔTPC( 11138571113980radicradicradicradicradicradic11139791113978radicradicradicradicradicradic1113981

closed loop

+ 10 middot log10 M1113980radicradicradicradic11139791113978radicradicradicradic1113981bandwidth factor

⎧⎪⎨

⎪⎩

⎫⎪⎬

⎪⎭

(1)

where Pmax is the maximum user transmit power P0 is thetarget received power per Physical Resource Block (PRB) atthe eNodeB α is the path loss compensation factor pa-rameter PL is the estimated user path loss M is the numberof PRBs assigned to the user by the scheduler at the eNodeBand ΔTF + f(ΔTPC) is a dynamic offset defined on a per-connection basis dependent of the modulationcodingscheme and closed-loop compensation [26] P0 consists ofa cell-specific power level common for all UEs in the cellreferred to as nominal power parameter and a UE-specific

offset used to correct for systematic offsets in a UErsquostransmission power setting (3GPP TS 36213) In this workonly the nominal power parameter is considered which canbe adjusted on a cell-by-cell basis Likewise full path losscompensation is assumed (ie α 1) as it is still a commonsetting and the default value in todayrsquos LTE networks beingalso used in the analyzed network during the field trial emaximum value of M assigned by the scheduler to a user islimited by the system bandwidth and the power headroom ofthe user (ie the ratio between the maximum user transmitpower and the UE transmit power per each PUSCH sub-frame given by (1)) [30] erefore in this paper it is as-sumed a specific UL scheduler implementation that stopsallocating PRBs to a user if it estimates based on UE powerheadroom reports [31] that UE has reached its maximumtransmit power

3 Analytical Model

e analytical model proposed here estimates the impact oninterference and connection quality statistics caused bychanging P0 in one or more cells in a live LTE system fromstatistics in the Network Management System (NMS) Forthis purpose the change in the average UL interference levelin a cell is first computed and the change in the averageSINR in the cell is derived later

To quantify such changes UL cell coupling is quan-tified that was utilized previously for congestion reliefalgorithm in Subway Areas by tuning Uplink PowerControl in LTE [32] A pair of cells (i j) is stronglycoupled when P0 changes in cell j affect significantly theinterference in cell i For this to happen several conditionsmust be satisfied (a) the interfering cell must be closedin electrical terms (ie small path loss between theinterfering users) (b) the load of the interfering cell mustbe high and (c) interference generated by users must beabove the noise floor In the absence of a precise propa-gation model for the live network in this work proximitybetween cells i and j is estimated by the ratio between thenumber of handovers from i to j compared to the totalnumber of handovers of source cell i (ie more handoversmeans more overlapping between cells) us the pro-portion of the UL interference in cell i coming fromusers in a specific neighbor cell n hereafter referred to asInterference Contribution Ratio (ICR) can be computed as

ICR(i n) I(i)

I(i) + N0(i)middot

HO(n i) middot L(n) middot P0(n)

1113936foralljisinN(i)HO(j i) middot L(j) middot P0(j)

(2)

where I(i) + N0(i) is the total received interference level inthe cell under study i I(i) is the sum of the interferencegenerated by users in neighbor cells N0(i) is the noise floorplus external interference from other systems (and hence thedependence of N0 of the specific cell i) N(i) is the set of cellsneighboring cell i HO(n i) is the number of outgoinghandovers from cell i to neighbor cell n L(n) is the load ofneighbor n and P0(n) is the nominal power of the neighbor

2 Mobile Information Systems

n (the contribution is greater if more transmit power) In theformula P0 and interference and noise levels are in linearunits (mW) and cell load is given by the PRB utilization ratio(ie L(n) isin [0 1])

ICR has been created in a heuristic way from theknowledge of what parameters have an effect on received ULinterference e first term in the product shows the ratio ofthe total undesired signal level due to users in surroundingcells while the second term shows the ratio of interferencedue to a particular neighbor

A cell i is considered a potential victim of the inter-ference generated by a source cell n if the interferencecontribution ratio is above 028 (ie ICRgt 028) Con-versely a cell n whose maximum interference contributionratio to its neighbor i ICR(i n) is below 028 should beable to increase its P0 without causing interference prob-lems in surrounding cells is criterion has been con-sidered in the algorithm for selecting which cells shouldincrease P0

In a real LTE network noise plus external interferencelevel on a cell basis can be inferred from the lowest bin in thehistogram of Received Strength Signal Indicator (RSSI) andP0(n) value is obtained from configuration file LikewiseHO(n i) and L(n) are NMS statistics For consistency theratio of UL interference in cell i not coming from itsneighbors ICRN0(i) is given by

ICRN0(i) 1minus 1113944foralljisinN(i)

ICR(i j)(3)

Once ICRs are obtained the change in average UL SINRin a cell i is computed from the change in the average re-ceived desired signal level (ΔS(i)) and received interferencelevel (ΔI(i)) in that cell as

Δ 1113956SINR(i) Δ1113955S(i) minusΔ1113955I(i)

ΔP0(i) middot (1minusPlim(i))minusΔ1113955I(i) (dB) (4)

whereΔP0(i) is the P0 increase (in dB) in cell i and Plim(i) isthe ratio of power-limited users in cell i which is available inthe NMS e equation reflects that the change in receiveddesired signal level is only due to non-power-limited users inthe cell under study [33]

e change in the average interference level in cell i dueto P0 changes in neighbor cells Δ1113955I(i) (in dB) is computed as

Δ1113955I(i) 10 log101113888ICRN0(i) + 1113944foralljisinN(i)

ICR(i j) middot1113956Lprime(j)

L(j)middot 10Δ

1113955S(j)101113889 (dB)

(5)

where L(j) and Lprime(j) are the load before and after increasingP0 respectively and Δ1113956S(j) ΔP0(j)(1minusPlim(j)) beingΔP0(j) the P0 change (increase or decrease in dB) in theneighbor cell j Briefly for the case of increasing P0 theincrease of interference in a cell i is larger when cell couplingICR(i j) is larger the increase of P0 ΔP0(j) is larger thepower-limited user ratio Plim(j) is smaller and theafterbefore load ratio is higher In contrast interferencedoes not change when ICR(i j) 0 forallj (note that in thiscase the term in the parenthesis is 1 from (3))

e only term in (5) that cannot be directly taken fromnetwork statistics is the cell load ratio 1113956Lprime(j)L(j) which iscomputed here as

1113956Lprime(n) L(n) middotRLE(n)

1113956RLEprime(n) (6)

where RLE(n) is the average radio link efficiency in theneighbor cell n before changing P0 and 1113956RLEprime(n) is the es-timate of the same indicator when P0 has changed Note that(6) considers that traffic does not change so that load changesare only due to changes in radio link efficiency RLE(n) and

1113956RLEprime(n) are computed from the average SINR values usingShannonndashHartley theorem [34] as

RLE(n) log2 1 + 10(SINR(n)10)1113872 1113873 (bpsHz) (7)

1113956RLEprime(n) log211138741 + 10(SINR(n)+Δ 1113956SINR(n)10)1113875 (bpsHz) (8)

where SINR(n) is the average SINR in cell n and Δ 1113956SINR(n)

is the estimated SINR change e former is included inNMS statistics but the latter must still be calculatedfrom (4) To break the loop in the use of equations (4)ndash(8)Δ1113955I(i) and Δ 1113956SINR(i) are computed in several steps Aninitial estimate Δ1113955I(i) is computed with (5) by assumingthat P0 changes do not affect cell loads (ie 1113956Lprime(j)L(j) 1)en a preliminary estimate Δ 1113956SINR(i) is obtainedwith (4) from which 1113956RLEprime(n) is obtained Later Δ1113955I(i)

is updated with changes in cell loads from which the finalvalue of Δ 1113956SINR(i) is obtained

4 Model Assessment

A field trial was conducted to check the impact of changingP0 in some cells of a live LTE network Model assessment iscarried out by comparing network performance measure-ments with estimates from the model

41 Assessment Setup e trial area comprises 124 outdoorcells in a densely populated urban scenario covering 10 km2e network includes a single carrier frequency at 2GHz with10MHz bandwidth UL antenna configuration is 1 transmitantenna in the UE and 2 receive antennas in the base stationInitially P0 is set to minus106 dBm in all cells of the scenario

e rules used to select cells increasing P0 are heuristiccriteria that take both cell capacity and connection qualityperformance into account Cells increasing P0 improvetheir spectral efficiency but they have to be carefully selectedto avoid deteriorating connection quality of neighborssignificantly due to excessive interference Note that thisimprovement is only achieved if the dynamic range at re-ceiver is not violated In another case a cell increasingP0 willdegrade greatly its own cell edge users In this study ULsignal level received at base station is always inside dynamicrange at receiver due to adequate power configurationerefore for this purpose of improvement in spectral ef-ficiency taking special care in the neighbors a cell is chosenfor increasing P0 if (a) it experiences low UL SINR valuesand at the same time neighbor cells do not suffer from UL

Mobile Information Systems 3

interference problems or (b) the cell is not strongly coupledto any neighbor cell so that users can increase their transmitpower without affecting any neighbor Cell coupling isobtained from initial P0 value handover load interferenceand RSSI statistics with (2) With these criteria 14 cells inthe scenario are selected for increasing P0 from minus106 tominus102 dBm is conservative value of 4 dB is selected asa tradeoff between observability and safety e larger themagnitude of changes the larger the impact on networkperformance indicators and the easier the analysis but atthe same time large parameter changes might cause un-necessary network degradation or instability In this sense itis worth noting that network usage is still very low in currentLTE networks especially in the UL is causes large trafficfluctuations between hours and days in a cell which makethe analysis complicated when introducing small networkparameter changes

e analyzed key performance indicators are the averageUL PRB utilization ratio () average UL interference level(dBmPRB) and average UL SINR (dB) All these indicatorsare collected on a per-cell basis in 4 office days before andafter changing P0 settings Busy hour (BH) measurementsare considered since network load is still small in the trialarea and hence interference is relatively low A single valueof these indicators is obtained for the before and after pe-riods which are then subtracted to obtain the difference

Network measurements ΔSINR(i) ΔI(i) and Lprime(j) arecompared with model estimates Δ 1113956SINR(i) Δ1113955I(i) and 1113956Lprime(j)

by means of linear regression analysis Goodness of fit isevaluated by checking the regression equation and the co-efficient of determination R2 To avoid border effects resultsare presented only for nonborder cells whose neighbors areall inside the cluster of cells where statistics are collected

During the trial unexpected changes of UL traffic de-mand took place in some cells between the before and afterperiodsese changes were due to the low traffic demand inthe trial area where the average PRB utilization ratio in theUL during the BH is less than 4 To include these un-controllable changes of traffic in the model a correctionfactor is introduced in (6) on a cell basis is factor is theratio between the traffic in the after and before periodMoreover these traffic variations cause that UEs connectfrom different positions in the cell which means that thenumber of power-limited users might change significantlyConsequently all nonborder cells will not be analyzed boththe interference and the SINR

To reduce the impact of these fluctuations in the trafficand power-limited users on the analysis and thus isolate theeffect of P0 changes the median value of both indicators hasbeen calculated for each nonborder cell in before and afterperiods In the case of the interference a nonborder cell isfiltered from the study when any relevant neighbor(HOgt 10) suffers from significant traffic or power-limitedusers variations (greater than 20 for traffic and 30 forpower-limited users) ese variations are calculated as therelative variation of the median value in after and beforeperiods on the median value in the before period In thecase of the SINR a nonborder cell is filtered by 2 reasons(a) traffic or power-limited users variations in relevant

neighbors that cause received interference level variations inthe cell under study as detailed above and (b) power-limitedusers variations (greater than 30) in the cell under studywhich can change the received signal level

42 Results e analysis is first focused on changes of cellload caused by P0 changes Only for this indicator bothborder and nonborder cells are considered since it isnecessary to have the load of all cells in the cluster which actas neighbors to others To calculate the load of border cellswhich have a lot of neighbors out of cluster it has beenassumed that the load in these neighbors is the average of theload for the rest of their neighbors in the cluster Further-more it is supposed that these cells out of cluster have notload and P0 variation Results show in Figure 1 that thecoefficient of determination is R2 092 with a slope of 116and an offset of minus0004 (ie 1113956Lprime(j) 116 middot Lprime(j)minus 0004) Allthese parameters (ie 092 116 and minus0004) have beenobtained from the linear correlation between real UL loadLprime(j) and estimated UL load 1113956Lprime(j) (using (6)) after P0changes using real data

e analysis is then focused on changes in interferencelevels Figure 2 plots measurements and estimates of ULinterference variations (ie ΔI(i) and Δ1113955I(i) resp) Each dotin the figure represents a nonborder cell that does not fulfillthe criterion for being filtered from the analysis It is ob-served that the slope of the regression equation is close to 1(ie 073) and the coefficient of determination is reasonable(ie R2 057) Note that in the figure some cells experi-ence a smaller UL interference (ie negative change) whichis not possible if P0 has increased is unexpected resultmight be due to changes in the spatial traffic distributioninside each cell mentioned previously which have beendetected from the comparison of power-limited ratios in thebefore and after periods in cells not changing P0 e changein the position of UEs between periods is again a conse-quence of the low traffic in the UL

e following analysis is focused on changes in con-nection quality Figure 3 shows measurements and estimatesof UL SINR variations (ie ΔSINR(i) and Δ 1113956SINR(i) resp)for nonborder cells that do not fulfill the criteria for beingeliminated from the study It is observed that the regres-sion equation is Δ 1113956SINR(i) 094 middot ΔSINR(i) + 025 withR2 038 Dots in the upper-right of the figure correspond tocells increasing P0 which experience larger increases in ULSINR However not all cells increasing P0 experiencea significant improvement in UL SINR but only those witha small ratio of power-limited users [33] Note that a P0increase leads to a higher transmit power only for users thatare not power limited (for which PltPmax typically close tothe base station) A closer analysis shows that some cellsincreasing P0 experience worse UL SINR is is explainedby the negative effect of increasing P0 on distant users in thesame cell (ie cell-edge users) which tend to be powerlimited (P Pmax) For these distant users an increase ofP0 is not translated into a higher transmit power whileinterference from other cells could be increased due to P0increase in those neighbor cells As a result cell-edge users


experience a lower UL SINR Whether UL SINR in a cell isincreased or decreased depends on the ratio of power-limited and non-power-limited users as included in themodel by the Plim parameter e rest of the dots in thelower-left of the figure correspond to neighbor cells of thoseincreasing P0 which experience a decrease in UL SINR

e impact on 95-ile UL SINR should be larger than onaverage UL SINR however for optimization purposes itis usually more convenient to work with average valuesNevertheless it will depend on the final goal us if theoptimization objective is an average improvement of the cellaverage SINR will be the parameter used Conversely if theaim is to improve for example the coverage 95-ile SINRshould be taken into account In this work the optimization isbased on the average cell level so the average SINR is analyzedin the model

Finally Table 1 compares measurements and estimatesaggregated in a cluster level considering all cells for the loadand only nonborder cells that are not filtered by any exposedcriterion for interference and SINR In the table it isshown that the average real change in interference leveland SINR is 65 less than the estimated values It is expected

that these differences between model estimates and realnetwork measurements reduce as UL traffic increases in LTEnetworks

During field trial cells changing P0 were decided in-telligently based on different rules explained in Section 41Unfortunately the experiment where all cells changed theirP0 could not be checked to compare results with the ana-lyzed experiment as it was not part of the optimizationprocess Nonetheless if all cells increased P0 without usingany criterion of cell selection the following behaviors couldbe observed

(i) Cells increasing their P0 could be improved in-significantly in terms of UL SINR due tomany power-limited users (only close users would be improved)

(ii) Degradation in neighbor cells in terms of ULinterference due to high UL coupling between cellsor initial interference problem that would beworsened when P0 is increased in other nearby cellsdespite not having high UL coupling

is basic approach would give way to not consider thetradeoff between benefit (ie more UL signal) and degra-dation (ie more UL interference) leading to possiblesituations where the degradation is greater than the benefitWith the proposed model the impact on UL SINR and ULinterference could be estimated in advance and to avoidunwanted effects

5 Conclusions

In this work an analytical model for estimating the changein interference level and SINR caused by modifying thenominal power parameter in the uplink power control

y = 09426x + 02458 R2 = 03825

ndash2

ndash1

0

1

2

3

4

ndash2 0ndash1 1 2 3 4

ΔSI

NR

(dB)

ΔSINR (dB)

Figure 3 Correlation of UL SINR variations

Table 1 Modeled and measured variations of network KPIs afterthe increase of P0

Key performance indicators Real EstimateAverage of UL radio utilization variation () 003 019Average of UL interference variation (dB) 006 017Average of UL SINR variation (dB) 011 035

y = 07291x + 0128 R2 = 05694

ndash1

ndash05

0

05

1

15

ndash1 ndash05 05 1 150

ΔI (

dB)

ΔI (dB)

Figure 2 Correlation of UL interference variations

60

50y = 1 1627x ndash 04315

R2 = 0924140

30

20

10

06050403020100

Lprime ()

Lprime (

)

Figure 1 Correlation of UL load after P0 changes


scheme of a live LTE network has been proposede modelcan be easily adjusted to any particular scenario based on P0parameter (configuration) handover RSSI SINR loadand power-limited user statistics available in the networkmanagement system Model assessment has been carried outin a field trial where the nominal power parameter is in-creased in some cells of a live LTE network Results showthat the performance of the proposed model is reasonableprovided that uplink traffic demand is stable in the networke flexibility and low computational load of the modelmakes it ideal for integration into self-tuning algorithms forthe nominal power parameter

More specifically this analytical model could be used inboth optimization and self-healing phases (ie compensa-tion of cells in outage) as it would allow to predict variationson certain KPIs when P0 is modified With optimizationalgorithms P0 is normally modified iteratively untilreaching a goal for these KPIs However this process couldbe carried out in one shot using the proposed modelNowadays an important objective is to get algorithms ca-pable of reacting very fast using for example a predictivemodel especially when they are working on self-healingphase Moreover iterative processes involve many timesunnecessary changes and subsequent reversion that wastestime and resources It is proposed as future work to extendthe model to high-level KPIs such as UL throughput and totest it on an LTE network with more UL load

Data Availability

e underlying customer data is not publicly available

Conflicts of Interest

e authors declare that they have no conflicts of interest

Acknowledgments

is work was funded by the Spanish Ministry of Economyand Competitiveness (TIN2012-36455 and TEC2015-69982-R) Ericsson Agencia IDEA (Consejerıa de Cien-cia Innovacion y Empresa Junta de Andalucıa Ref 59288)and FEDER

References

[1] NGMN Alliance ldquoNext generation mobile networks recom-mendation on SON and OampM requirementsrdquo TechnicalReport NGMN Alliance Frankfurt Germany December2008

[2] 3GPP TS 32500 Telecommunication Management Self-Organizing Networks (SON) Concepts and Requirements(Release 9) 3GPP Technical Specification Group Radio AccessNetwork Std 2009

[3] 4G Americas ldquoSelf-optimizing networks the benefits of SONin LTErdquo Technical Report 4G Americas Bellevue WA USA2014

[4] G Hampel K L Clarkson J D Hobby and P A Polakosldquoe tradeoff between coverage and capacity in dynamicoptimization of 3G cellular networksrdquo in Proceedings of the

IEEE 58th Vehicular Technology Conference (VTC-Fall) vol 2pp 927ndash932 Orlando FL USA October 2003

[5] V Bratu and C Beckman ldquoBase station antenna tilt for loadbalancingrdquo in Proceedings of the 7th European Conference onAntennas and Propagation (EuCAP) pp 2039ndash2043 GothenburgSweden April 2013

[6] Y Khan B Sayrac and E Moulines ldquoCentralized self-optimization in LTE-A using active antenna systemsrdquo inProceedings of the Wireless Days (WD) pp 1ndash3 ValenciaSpain November 2013

[7] A J Fehske H Klessig J Voigt and G P Fettweis ldquoCon-current load-aware adjustment of user association and antennatilts in self-organizing radio networksrdquo IEEE Transactions onVehicular Technology vol 62 no 5 pp 1974ndash1988 2013

[8] V Buenestado M Toril S Luna-Ramırez J M Ruiz-Avilesand A Mendo ldquoSelf-tuning of remote electrical tilts based oncall traces for coverage and capacity optimization in LTErdquoIEEE Transactions on Vehicular Technology vol 66 no 5pp 4315ndash4326 2017

[9] D Kim ldquoDownlink power allocation and adjustment forCDMA cellular systemsrdquo IEEE Communications Lettersvol 1 no 4 pp 96ndash98 1997

[10] D Kim ldquoA simple algorithm for adjusting cell-site transmitterpower in CDMA cellular systemsrdquo IEEE Transactions onVehicular Technology vol 48 no 4 pp 1092ndash1098 1999

[11] T Cai G Koudouridis C Qvarfordt J Johansson andP Legg ldquoCoverage and capacity optimization in E-UTRANbased on central coordination and distributed Gibbs sam-plingrdquo in Proceedings of the 2010 IEEE 71st Vehicular Tech-nology Conference (VTC 2010-Spring) Taipei Taiwan May2010

[12] A Engels M Reyer X Xu R Mathar J Zhang andH Zhuang ldquoAutonomous self-optimization of coverage andcapacity in LTE cellular networksrdquo IEEE Transactions onVehicular Technology vol 62 no 5 pp 1989ndash2004 2013

[13] S Fan H Tian and C Sengul ldquoSelf-optimization of coverageand capacity based on a fuzzy neural network with co-operative reinforcement learningrdquo EURASIP Journal onWireless Communications and Networking vol 2014 p 572014

[14] V Buenestado M Toril S Luna-Ramırez and J Ruiz-AvilesldquoSelf-planning of base station transmit power for coverageand capacity optimization in LTErdquo Mobile Information Sys-tems vol 2017 Article ID 4380676 12 pages 2017

[15] J Whitehead ldquoSignal-level-based dynamic power control forco-channel interference managementrdquo in Proceedings of theIEEE 43rd Vehicular Technology Conference pp 499ndash502Secaucus NJ USA May 1993

[16] C U Castellanos F D Calabrese K I Pedersen and C RosaldquoUplink interference control in UTRAN LTE based on theoverload indicatorrdquo in Proceedings of the IEEE 68th VehicularTechnology Conference (VTC Fall) pp 1ndash5 Calgary ABCanada September 2008

[17] B Muhammad and A Mohammed ldquoPerformance evaluationof uplink closed loop power control for LTE systemrdquo inProceedings of the IEEE 70th Vehicular Technology Conference(VTC-Fall) pp 1ndash5 Anchorage AK USA September 2009

[18] R Mullner C F Ball M Boussif et al ldquoEnhancing uplinkperformance in UTRAN LTE networks by load adaptivepower controlrdquo European Transactions on Telecommunica-tions vol 21 no 5 pp 458ndash468 2010

[19] S Yang Q Cui X Huang and X Tao ldquoAn effective uplinkpower control scheme in CoMP systemsrdquo in Proceedings of


the IEEE 72nd Vehicular Technology Conference (VTC-Fall)pp 1ndash5 Ottawa ON Canada September 2010

[20] A Awada BWegmann I Viering and A Klein ldquoOptimizingthe radio network parameters of the long term evolutionsystem using Taguchis methodrdquo IEEE Transactions on Ve-hicular Technology vol 60 no 8 pp 3825ndash3839 2011

[21] M Dirani and Z Altman ldquoSelf-organizing networks in nextgeneration radio access networks application to fractionalpower controlrdquo Computer Networks vol 55 no 2 pp 431ndash438 2011

[22] T Nihtila J Turkka and I Viering ldquoPerformance of LTE self-optimizing networks uplink load balancingrdquo in Proceedings ofthe IEEE 73rd Vehicular Technology Conference (VTC Spring)pp 1ndash5 Budapest Hungary May 2011

[23] S Xu M Hou K Niu Z He and W Wu ldquoCoverage andcapacity optimization in LTE network based on non-cooperative gamesrdquo Journal of China Universities of Postsand Telecommunications vol 19 no 4 pp 14ndash42 2012

[24] J A Fernandez-Segovia S Luna-Ramırez M TorilA B Vallejo-Mora and C Ubeda ldquoA computationally effi-cient method for self-planning uplink power control pa-rameters in LTErdquo EURASIP Journal on WirelessCommunications and Networking vol 2015 p 80 2015

[25] J A Fernandez-Segovia S Luna-Ramırez M Toril andC Ubeda ldquoA fast self-planning approach for fractional uplinkpower control parameters in LTE networksrdquo Mobile In-formation Systems vol 2016 Article ID 8267407 11 pages2016

[26] 3GPP TS 36213 Evolved Universal Terrestrial Radio Access(E-UTRA) Physical Layer Procedures (V1100) 3GPPTechnical Specification Group Radio Access Network Std2012

[27] C U Castellanos D L Villa C Rosa et al ldquoPerformance ofuplink fractional power control in UTRAN LTErdquo in Pro-ceedings of the IEEE 67th Vehicular Technology Conference(VTC Spring 2008) pp 2517ndash2521 Singapore May 2008

[28] K Majewski and M Koonert ldquoAnalytic uplink cell loadapproximation for planning fractional power control in LTEnetworksrdquo in Proceedings of the 14th International Telecom-munications Network Strategy and Planning Symposium(NETWORKS) pp 1ndash7 Warsaw Poland September 2010

[29] I Viering A Lobinger and S Stefanski ldquoEfficient uplinkmodeling for dynamic system-level simulations of cellular andmobile networksrdquo EURASIP Journal on Wireless Communi-cations and Networking vol 2010 article 282465 2010

[30] F D Calabrese C Rosa M Anas P H MichaelsenK I Pedersen and P E Mogensen ldquoAdaptive transmissionbandwidth based packet scheduling for LTE uplinkrdquo inProceedings of the IEEE 68th Vehicular Technology Conference(VTC-Fall) pp 1ndash5 Calgary AB Canada September 2008

[31] W Kim Z Kaleem and K Chang ldquoPower headroom report-based uplink power control in 3GPP LTE-A HetNetrdquoEURASIP Journal on Wireless Communications and Net-working vol 2015 p 233 2015

[32] A B Vallejo-Mora M Toril S Luna-Ramırez A Mendo andS Pedraza ldquoCongestion relief in subway areas by tuninguplink power control in LTErdquo IEEE Transactions on VehicularTechnology vol 66 no 7 pp 6489ndash6497 2017

[33] M Boussif C Rosa JWigard and RMullner ldquoLoad adaptivepower control in LTE uplinkrdquo in Proceedings of the EuropeanWireless Conference (EW) pp 288ndash293 Lucca Italy April2010

[34] C E Shannon and W Weaver e Mathematical eory ofCommunication e University of Illinois Press Urbana ILUSA 1949


Computer Games Technology

International Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom

Journal ofEngineeringVolume 2018

Advances in

FuzzySystems


Volume 2018


ReconfigurableComputing



Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

thinspArtificial Intelligence

Hindawiwwwhindawicom Volumethinsp2018


Civil EngineeringAdvances in


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia


Biomedical Imaging



Engineering Mathematics


RoboticsJournal of



Computational Intelligence and Neuroscience


Mathematical Problems in Engineering

Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018


Human-ComputerInteraction

Advances in


Scientic Programming

Submit your manuscripts atwwwhindawicom

proposed for validating self-organizing algorithms All thesemodels are based on propagation traffic and radio linkquality predictions However to the authorsrsquo knowledge noanalytical model has been proposed to predict the impact ofchanging the referred parameters in an existing LTEscenario

In this work a novel analytical model for estimating theimpact of increasing the nominal received power parameterin the FPC algorithm for the Physical Uplink SharedCHannel (PUSCH) in LTE is presented Unlike previousmodels the proposed model is conceived for network op-timization purposes us the model is designed as anincremental model whose aim is to estimate the change inthe UL interference level and Signal-to-Interference-plus-Noise ratio (SINR) caused by modifying the current con-figuration of the nominal power parameter on a cell-by-cellbasis erefore it would be a predictive model that wouldallow the algorithm to react faster (ie in one shot) andmoreaccurately (ie no unnecessary changes and subsequentreversion) than working iteratively e predictive approachis very demanded in the current algorithms both for op-timization and for self-healing Although this model hasbeen conceived for optimization phase it could also be usedduring self-healing phase in order to compensate cells inoutage ensuring minimum UL SINR Moreover the modelcan be easily adjusted to any particular scenario based onavailable network performance statistics Model assessmentis carried out bymeans of a field trial whereP0 is increased insome cells of a live LTE network e rest of the paper isorganized as follows Section 2 outlines the FPC algorithmfor PUSCH in LTE Section 3 presents the proposed esti-mation model Section 4 shows the results of the model ina live network and Section 5 summarizes the mainconclusions

2 FPC Algorithm for PUSCH

FPC defines the transmit power of each User Equipment(UE) which is dynamically adapted to the changing radiopropagation conditions and fluctuations of interferencefrom users in neighbor cells e equation defining the valueof UE transmit power P (in dBm) in each PUSCH subframeis

P min Pmax P0 + αPL1113980radicradicradic11139791113978radicradicradic1113981open loop

+ΔTF + f ΔTPC( 11138571113980radicradicradicradicradicradic11139791113978radicradicradicradicradicradic1113981

closed loop

+ 10 middot log10 M1113980radicradicradicradic11139791113978radicradicradicradic1113981bandwidth factor

⎧⎪⎨

⎪⎩

⎫⎪⎬

⎪⎭

(1)

where Pmax is the maximum user transmit power P0 is thetarget received power per Physical Resource Block (PRB) atthe eNodeB α is the path loss compensation factor pa-rameter PL is the estimated user path loss M is the numberof PRBs assigned to the user by the scheduler at the eNodeBand ΔTF + f(ΔTPC) is a dynamic offset defined on a per-connection basis dependent of the modulationcodingscheme and closed-loop compensation [26] P0 consists ofa cell-specific power level common for all UEs in the cellreferred to as nominal power parameter and a UE-specific

offset used to correct for systematic offsets in a UErsquostransmission power setting (3GPP TS 36213) In this workonly the nominal power parameter is considered which canbe adjusted on a cell-by-cell basis Likewise full path losscompensation is assumed (ie α 1) as it is still a commonsetting and the default value in todayrsquos LTE networks beingalso used in the analyzed network during the field trial emaximum value of M assigned by the scheduler to a user islimited by the system bandwidth and the power headroom ofthe user (ie the ratio between the maximum user transmitpower and the UE transmit power per each PUSCH sub-frame given by (1)) [30] erefore in this paper it is as-sumed a specific UL scheduler implementation that stopsallocating PRBs to a user if it estimates based on UE powerheadroom reports [31] that UE has reached its maximumtransmit power

3 Analytical Model

e analytical model proposed here estimates the impact oninterference and connection quality statistics caused bychanging P0 in one or more cells in a live LTE system fromstatistics in the Network Management System (NMS) Forthis purpose the change in the average UL interference levelin a cell is first computed and the change in the averageSINR in the cell is derived later

To quantify such changes UL cell coupling is quan-tified that was utilized previously for congestion reliefalgorithm in Subway Areas by tuning Uplink PowerControl in LTE [32] A pair of cells (i j) is stronglycoupled when P0 changes in cell j affect significantly theinterference in cell i For this to happen several conditionsmust be satisfied (a) the interfering cell must be closedin electrical terms (ie small path loss between theinterfering users) (b) the load of the interfering cell mustbe high and (c) interference generated by users must beabove the noise floor In the absence of a precise propa-gation model for the live network in this work proximitybetween cells i and j is estimated by the ratio between thenumber of handovers from i to j compared to the totalnumber of handovers of source cell i (ie more handoversmeans more overlapping between cells) us the pro-portion of the UL interference in cell i coming fromusers in a specific neighbor cell n hereafter referred to asInterference Contribution Ratio (ICR) can be computed as

ICR(i n) I(i)

I(i) + N0(i)middot

HO(n i) middot L(n) middot P0(n)

1113936foralljisinN(i)HO(j i) middot L(j) middot P0(j)

(2)

where I(i) + N0(i) is the total received interference level inthe cell under study i I(i) is the sum of the interferencegenerated by users in neighbor cells N0(i) is the noise floorplus external interference from other systems (and hence thedependence of N0 of the specific cell i) N(i) is the set of cellsneighboring cell i HO(n i) is the number of outgoinghandovers from cell i to neighbor cell n L(n) is the load ofneighbor n and P0(n) is the nominal power of the neighbor







ICR(i j)(3)








L(j)middot 10Δ

1113955S(j)101113889 (dB)

(5)












4 Model Assessment
























5 Conclusions


y = 09426x + 02458 R2 = 03825

ndash2

ndash1

0

1

2

3

4


ΔSI

NR

(dB)

ΔSINR (dB)




y = 07291x + 0128 R2 = 05694

ndash1

ndash05

0

05

1

15


ΔI (

dB)

ΔI (dB)


60

50y = 1 1627x ndash 04315

R2 = 0924140

30

20

10

06050403020100

Lprime ()

Lprime (

)





Data Availability




Acknowledgments


References












































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in









ICR(i j)(3)








L(j)middot 10Δ

1113955S(j)101113889 (dB)

(5)












4 Model Assessment
























5 Conclusions


y = 09426x + 02458 R2 = 03825

ndash2

ndash1

0

1

2

3

4


ΔSI

NR

(dB)

ΔSINR (dB)




y = 07291x + 0128 R2 = 05694

ndash1

ndash05

0

05

1

15


ΔI (

dB)

ΔI (dB)


60

50y = 1 1627x ndash 04315

R2 = 0924140

30

20

10

06050403020100

Lprime ()

Lprime (

)





Data Availability




Acknowledgments


References












































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in























5 Conclusions


y = 09426x + 02458 R2 = 03825

ndash2

ndash1

0

1

2

3

4


ΔSI

NR

(dB)

ΔSINR (dB)




y = 07291x + 0128 R2 = 05694

ndash1

ndash05

0

05

1

15


ΔI (

dB)

ΔI (dB)


60

50y = 1 1627x ndash 04315

R2 = 0924140

30

20

10

06050403020100

Lprime ()

Lprime (

)





Data Availability




Acknowledgments


References












































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in






Data Availability




Acknowledgments


References












































Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in


























Advances in

FuzzySystems


Volume 2018













Journal of

Journal of



Hindawi


Advances in

Multimedia


Biomedical Imaging





RoboticsJournal of









Volume 2018



Advances in




analytical model for estimating the impact of changing the ...webpersonal.uma.es/~toril/files/2018...

Documents