son in 4g mobile networks - vde-itg · pdf fileson in 4g mobile networks self-optimization...

SON in 4G Mobile Networks

Self-Optimization Techniques for Intelligent Base Stations

Bell Labs Stuttgart

Ulrich Barth

9. Fachtagung des ITG-FA 5.2, Oktober 2010

All Rights Reserved © Alcatel-Lucent 2008, XXXXX

Self- organizing Radio Access Networks

Motivation

Current situation for radio access network management� Deployment and maintenance become more and more complex and costextensive

� Trend to smaller cells, multi-band operation, heterogeneous mobile networks

� High manual intervention for configuration, capacity upgrade or in failure cases required

� High effort required for optimisation of system performance

� Deep system expertise required

� High effort necessary for measurement campaigns (drive tests)

� Different tools for planning, configuration, measurement/KPI acquisition and optimisation involved

increasing effort for network management and optimisation

�� new concepts for simplified network operation required


Self- organizing Radio Access Networks

Requirements

Management of radio access networks has to be self-organized in future � Automated configuration, optimization and fault management:

� towards real plug-and-play self-configuration

� continuous up to autonomous self-optimization

� fast self-healing mechanisms

� Paradigm change:

� to put network optimization know how into intelligent self-x algorithms

� to focus network management on high level monitoring and performance tuning

� High performance self-x algorithms required:

� fast convergence

� stable operation

� tuneable according to operator requirements

� managing mutual dependencies between self-x use cases


Self-X Architecture

� “NEM less” network management

� Fully autonomous, distributed

RAN optimisation

� Self-x functions in UE and eNB

� measurements, UE location info

� alarms, status reports, KPIs

� distributed self-x algorithms

� Network management in NM OSS

focussed on

� network planning

� alarm and performance monitoring

� high level performance tuning

Vision of fully distributed self-management

eNB

LTE RAN

Network Management

eNB

eNB

self-x

NM OSS

Itf-N

X2-Itf

self-x

self-x

RAN self-optimization

�� performance monitoring

�� KPIs

�� alarms

�� high level network

performance tuning

OSS: Operation Support SystemNEM: Network Element Manager

All Rights Reserved © Alcatel-Lucent 2010 5

Mobility Robustness (Handover Optimization)


Configuration Parameters for Handover in LTE

LTE handover performance has large impact on system

performance

� Configuration parameters

� Filtered RSRP values

� Handover Margin, i.e. hysteresis between source and target

� Time to trigger (TTT)

� Cell Individual Offset (CIO), add on handover margin

� Target

� High handover success rate

TTT(ms)

Handoverevent

from UEto eNB

FilteredRSRP

[dB]Source Cell

Target Cell

Time

X2 delay

Hyst(dB)

Handover

commandfrom eNBto UE

Radio LinkFailure (RLF)

threshold

Handover failuredue to RLF

TTT(ms)

Handoverevent

from UEto eNB

FilteredRSRP

[dB]Source Cell

Target Cell

Time

X2 delay

Hyst(dB)

Handover



threshold


A3 HO event

source cell UEtarget cell

HO command

Normalized HO Rate Vs Residual BLER for ; TTT=0 to 200 ms; 20ms step

0

5

10

15

20

25

30

35

0 1 2 3 4 5 6 7 8 9 10

BLER [%]

No

rmal

ized

HO

Rat

e

BLER: block error rate of HO command � RLF

Normalized HO rate:without slow/fast fading

SON algo: find best trade-off between minimum BLER and minimum HO rate (ping pong)


SON definition: too late Handover

RLF, as UE still associated to cell A

cell AUE

cell B

TTT(ms)

no uplink communication:no Handover event from UEto eNB A

FilteredRSRP

[dB]

Source Cell A

Target Cell B

Time

Hyst(dB)


threshold


TTT(ms)

no uplink communication:no Handover event from UEto eNB A

FilteredRSRP

[dB]

Source Cell A

Target Cell B

Time

Hyst(dB)


threshold


TTT(ms)

Handoverevent

from UEto eNB

FilteredRSRP

[dB]Source Cell A

Target Cell B

Time

X2 delay

Hyst(dB)

Handover



threshold


TTT(ms)

Handoverevent

from UEto eNB

FilteredRSRP

[dB]Source Cell A

Target Cell B

Time

X2 delay

Hyst(dB)

Handover



threshold


Alternative 2: a failure occurs in the

source cell during the HO procedure

Alternative 1: a failure occurs in the

source cell before the HO was initiated


Messages and algorithm for “too late” handover

SON Algorithm- sufficient measurements for decision of HO problem ?

- analysis of measurements and other data (own eNB and other eNB)

- modify HO parameters

- for all UE speed classes

- mobility state dependent

- start new measurement cycle

Update “Reestablishment”Statistics (PCI#A)

RLF INDICATION

UE eNB A eNB B

RLF


RRC Connection Reestablishment Request including PCI#A

UE connected to eNB A

Measurement Report (eNB B) HO Request

HO Request AcknowledgeRRC Connection Reconfiguration

RRC Connection Reestablishment

RRCConnectionReestablishmentComplete

UE Context Release


RLF INDICATION

UE eNB A eNB B

RLF


RRC Connection Reestablishment Request including PCI#A

UE connected to eNB A

Measurement Report (eNB B) HO Request

HO Request AcknowledgeRRC Connection Reconfiguration

RRC Connection Reestablishment

RRCConnectionReestablishmentComplete

UE Context Release


SON based HO optimization

Characteristics of simulation scenario “Frankfurt”

Available ray-tracing data

� Input data characteristics:

� 16 real world antenna locations in the city of

Frankfurt

� Tri-sectorized configuration (i.e. 48 cells)

� individual antenna type information, heights

and beam directions

� topographic map of region (4000m ×××× 4000m)

� Output data characteristics:

� Integer type pathloss data

per 10m ×××× 10m grid point per antenna

� Range: -88dB to -214dB



Algorithm test in simulation

� Event based simulation of HO

events, one algorithm instance per

simulated cell

� Interference level: 50% load

� mobile speed: 3, 30, 120 km/h

– 16 mobiles active per speed

– moving along given roads

– at crossroads, mobiles choose randomly next road

� same initial mobility parameters at

simulation start



HO success rate window controller

Simulation results (Simulated network operation: 3 days = 259200 s)

� A quick convergence of the cell global

parameters can be observed for frequently

visited cells (i.e. on many HO events)

� Choosing initial HO parameters above optimal

settings causes a HO success rate below given

limit

� The algorithm instance tunes the parameters

towards earlier decision for HO, achieving an

improvement of the cell total HO success rate

Gained Key Performance Indicators for TRX #31

0,0

10,0

20,0

30,0

40,0

50,0

60,0

70,0

80,0

90,0

100,0

0 28800 57600 86400 115200 144000 172800 201600 230400 259200

Simulated Time [s]

HO

Su

cces

s R

ate

[%]

0

3

6

9

12

15

18

21

24

27

30

Mea

n M

ob

ile R

estin

g T

ime

[s]

HO Success Rate

Mean Resting Time

Mobility Parameters for TRX #31

0,00

0,25

0,50

0,75

1,00

1,25

1,50

1,75

2,00

2,25

2,50

0 28800 57600 86400 115200 144000 172800 201600 230400 259200

Simulated Time [s]

Hys

tere

sis

[dB

]

0,0

0,2

0,4

0,6

0,8

1,0

1,2

1,4

1,6

1,8

2,0

Tim

e-T

o-T

rig

ger

[s]

Hysteresis

Time-To-Trigger


Coverage and Capacity Optimisation


Antenna Tilt Optimization

Optimization goals:

� sector coverage

� sector capacity

� based on downlink performance metric bits/sec/Hz

Approach:

� Distributed optimization:

� optimization of single cells together with closest neighbors

� targeting global optimum

� by adjusting the antenna tilt for each sector individually

� distributed approach, co-operating eNB and neighbours



Best Serving Sectors, Displaced Site Locations

� sites are displaced but playground borders

are kept fixed, with wrap around

� slow fading with area correlated

shadowing is invisible in best server plots

due to the equal attenuation of all sectors

at a certain point, independent of the

direction of the signalafter optimization



Metrics

Cell wide optimization approach

)1()()()1()(max

min5 dGGTGpBWGTBWM tf

G

Gpercentiletf ⋅⋅−+⋅⋅= ∫−

M: performance metricG: GeometryB: bandwidthp(G): probabilityTtf: throughput per MCSW: weighting factor

)( 51 percentiletf GTBM −⋅=coverage metric:

dGGTGpBM tf

G

G)()(

max

min2 ⋅= ∫

capacity metric:

� weighted coverage/capacity metric:

target: coverage target: capacity

next charts:

concrete simulation studies:operator tunable weighting

parameter W = 0.91



Performance Gains of Sector Average Performance and Sector Edge Perf.

+7%

+44%

gain of utility +21% of 160 → 193

Results:

� significant gains with respect

to equal tilts

� capacity performance +7%

� coverage performance +44%

Optimized Performance (red) [%]

Reference Performance 15°(blue) [100%]

Trade-off given by Utility Metric (green):- weight factor W between coverage and capacity

defines the slope of this line



Convergence Speed and Gains of Performance Metric

two reference curves for all sectors with 14 and 15 degrees equal downtilt

with optimized tilts

one cyle through all sectors equals 57 steps

Results:

� promising,

playground wide

improvement

� non-oscillating

� fast convergence


SON challenges


SON use case a

optimization algorithm a

Mutual interactions of SON optimization algorithms

SON use case b

KPI/measurement 1

KPI/measurement 2

KPI/ measurement 3

target functionmetric a

target functionmetric b

Parameter III

Parameter I

Parameter II

Radio System Radio System

Coupling by same control parameter: One parameter is modified bydifferent SON algorithms

Handover <-> Load balancing

Mutual impact on optimization target:One metric is influenced by parameters of different SON algorithms

Interference Coordination <-> Loadbalancing

optimization algorithm b

eNB

• different coupling mechanisms, different coupling strength

⇒ solution required to manage SON use case interworking !


www.alcatel-lucent.comwww.alcatel-lucent.com

son in 4g mobile networks - vde-itg · pdf fileson in 4g mobile networks self-optimization...

Documents