Julian Buhagiar - Meeting The Challenges Of Planning And Optimising Multi-System Rans - Barcelona Oct 2005.Pdf

Meeting the Challenges of Planning and Optimising Multi system RANs Meeting the Challenges of Planning and Optimising Multi System RANs Julian Buhagiar Product Manager Actix Ltd julian buhagiar@actix[.]

Meeting the Challenges of Planning and Optimising

Multi-System RANs

Julian Buhagiar Product Manager Actix Ltd.

julian.buhagiar@actix.com

Developing end-to-end 2/2.5/3G network dimensioning, capacity planning & traffic engineering

Optimising traffic distribution between network layers

Meeting the challenges of modelling multi-system networks

- data traffic services

Achieving maximum capacity and QoS for voice and data traffic

How do we develop end-to-end network dimensioning, capacity planning & traffic engineering?

If only…we could optimise

a cluster in the same time

as this afternoon’s session and be confident in the results…

The objectives of optimising networks for quality, capacity and coverage…

High level goals

ƒ Generating new revenue

ƒ Meeting rollout obligations

ƒ Increasing customer base

While:

ƒ Reducing required new sites

ƒ Reducing/delaying capex

ƒ Reducing opex

ƒ Reducing churn due to quality

ƒ Reducing time to launch

ƒ Maintaining (or reducing in

some cases) current headcount

Technical excellence

for the sake of technical excellence is

not a requirement

Coverage, capacity and quality in GSM,

EGPRS & UMTS

In GSM, EGPRS and UMTS all

three are fundamentally joined

together by the link budget:

ƒ Coverage – target

availability, propagation

characteristics

ƒ Capacity – design loading,

design user density

ƒ Quality – service(s) EbNo

and traffic

The combination of all three

leads to a single set of the

network design targets.

UMTS CPICH EcNo EGPRS BEP GSM RxQual

UMTS CPICH RSCP EGPRS CValue GSM RxLev

UMTS Pilot Pollution EGPRS Low C/I GSM Low C/I UMTS Soft Handover Overhead

EGPRS Cell Reselection GSM Hard Handover Overhead

Specific challenges for optimisation of UMTS

Capacity for a given equipment quantity is dependent on network spectral efficiency

ƒ The proportion of network loading available for traffic decreases when interference increases

It is possible to increase network capacity with no additional infrastructure deployment, or damage capacity by adding additional

infrastructure

UMTS operates on a single frequency

ƒ There is no frequency planning option to hide the interference problems caused by poor spectral efficiency of the network plan

The next cheapest options are significantly more expensive and time

consuming.

Optimising capacity and coverage

From Holma and Toskala:

The only improvement we can

make in downlink loading (for a

given quality) is by optimising

inter-cell interference (average i)

A 0.3 dB reduction in average

RSSI can lead to a 4% increase in

capacity on the downlink with the

same coverage

ƒ This is equivalent to reducing

the average i from 0.45 to

j

R W N E

1

0

1

νη

Impact of modifiying average intercell interference on

throughput vs range

600 700 800 900

Estimates from scanner data indicate

typical values of average i to be at least

0.38 to 0.50

Planning vs measured data

Automatic Cell and

Frequency Planning tools help

in decreasing interference

levels and improving

coverage This ensures better

coding scheme usage and

higher capacity

Measured data improves the

accuracy of the solution

Reduce analysis time

Reduce the number of

iterations we require prior to

Interference lev els [%]

Initial network

CellOpt ACP CellOpt

ACP+AFP

198 554

205

574

300 840

0 100 200 300 400 500 600 700 800 900

Initial network CellOpt ACP CellOpt ACP+AFP

Number of radios

Traffic [Erlang]

198 554

205

574

300 840

0 100 200 300 400 500 600 700 800 900

Initial network CellOpt ACP CellOpt ACP+AFP

Number of radios Traffic [Erlang]

Delta and model based solutions compared

Delta Based

Improvement Approach Shortcut

Accuracy Typically 4.5dB standard

Automatic vs Interactive Interactive

Data not uniform - unsuitable for full automation

Local knowledge and constraints part of process

Use Scenarios Fine Tuning

Course Optimisation Validation of coarse/break-out optimisation

Delta and model based solutions compared

Model Based

Improvement Approach Step Improvement

Accuracy Down to 7dB standard

deviation with improvements

Sampling Inherent difference in statistics

calculation

Information Coverage Uniform data but less certain

Automatic vs Interactive Data in uniform grid – suitable

for automation Possibility of breakout from local optima

Local knowledge and constraints require prior input and post-validation

Use Scenarios Initial Planning

Coarse Optimisation Break-out Optimisation

Typical multi-RAN Deployment

GSM and WCDMA are mostly co-sited for cost and convenience reasons

UMTS CPICH emitted power is ~10dB lower than GSM BCCH

In order to provide a good QoE to the users in all services, UMTS demands a CPICH coverage similar to the one GSM demands for the BCCH

WCDMA coverage typically is “less” than GSM

ƒ Exceptions exist for some indoor deployments and in close site proximity

Users handover from WCDMA to GSM when WCDMA coverage is inadequate

ƒ At the edge of “island” WCDMA coverage

ƒ When encountering coverage gaps or problems within the WCDMA coverage area

Typical multi-RAN Deployment

In order to be decoded:

ƒ GSM signal must be ~9dB above the noise in its 200kHz channel

ƒ GPRS signal must be ~12dB above the noise in its 200kHz channel

ƒ EDGE signal must be ~15dB above the noise in its 200kHz channel

ƒ If the signal from one cell approaches that of the serving cell in its 200kHz, we have bad quality

UMTS signal can be several dB below the noise level of its 5MHz channel,

because it is spread and dispread over the 5MHz width:

ƒ Every user signal is seen as noise by the other users, so every user contributes to the noise rise

ƒ As the noise level increases, an higher power is needed for the communications

UMTS has a bigger Spectrum Efficiency than GSM, taking into account:

ƒ the channel reuse factor

ƒ the average cell capacity

ƒ the channel bandwidth

Balancing Traffic & Load Distribution

Where is the traffic originated?

ƒ For calls it works the same way than in GSM

ƒ Building outdoor coverage following the lessons learned from GSM

ƒ Using the co-siting with GSM to save money and time

ƒ Most of the calls called indoors

Data increases indoor traffic

ƒ Who uses 384 kbps while driving or walking?

• Expensive and hard to build indoor coverage

ƒ Easy to predict the traffic with pure indoor solutions

ƒ Cell splitting may help to reduce expenses for indoor solutions but complicated antenna systems and power splitters increase managing and balancing problems

May become a source of hidden expenses!

Balancing Traffic & Load Distribution

Outdoor to indoor and vice versa

ƒ using gateway method: outdoor to indoor wanted only at the

entrances like parking place, ground floors

ƒ this can be achieved by tuning cell reselection triggering time, cell reselection hysteresis, cell reselection Quality Offset2

ƒ High hysteresis requires fast decision

ƒ Sintrasearch parameter will be used to define a trigger for intra frequency handover measurements

2G to 3G

ƒ if load in 2G cell increase, during voice call, the ISHO towards 3G can be achieved based on the load (i.e 75% load)

between different frequencies in the same cell

ƒ currently: based on Ec/No threshold during idle mode

ƒ near future: based on load during call setup

General UMTS capacity improvement

Reduce soft handover overhead

ƒ Antenna tilts

ƒ Antenna orientation

ƒ Leg addition threshold

ƒ Leg drop threshold

Tune holding time of dedicated channels

Fast RAB switching

Common channels (DSCH, HSDPA, HSUPA) for interactive & background traffic

Holding time tradeoff

ƒ Long – excellent QoE but excessive use of resources

ƒ Short – efficient resource usage but

• Delay due to frequent FACH-DCH switches & inter-RABs

• Poor QoE

• Higher Application Failures

General UMTS capacity improvement

Iub link

ƒ Tune VCC configurations

ƒ Overbooking

ƒ Scheduler for packet traffic

ƒ Extend link capacity

• Baseband processing power

ƒ Extend capacity

• Transmit Power

ƒ Transmit diversity

ƒ Amplifier with higher power

ƒ It is not possible to exceed the pole capacity!

Optimize Inter-system Handover parameters

ƒ Use RSCP and Ec/No as the main thresholds for triggering ISHO

Optimize Inter-system Handover parameters

ISHO Thresholds for buildings without 3G in-building solution

ƒ Typical RF conditions (RSCP & Ec/No correlation) within buildings

Observation (I)

ƒ High 3G in-building loss (15 ~ 25 dB depending on building layout)

ƒ Ec/No degrades rapidly when CPICH RSCP is closed to -110dBm

ƒ In low load network, Ec/No is still in good range at around CPICH RSCP of -100 dBm

EcNo

RSCP

Optimize Inter-system Handover parameters

ISHO Threshold Consideration (I) - For cell with good Ec/No

Early CPICH RSCP Threshold trigger

ƒ Can cause ping-pong effect between GSM and 3G network

ƒ LA/RA update is required when UE is in the new network

• increase signalling load

• degrade user experience for PS services (especially for streaming service because data disruption can be more than 25 sec)

ƒ Frequent compressed mode measurements

• higher power from the UE and 3G BTS

• can affect the cell capacity, coverage and quality

Late CPICH RSCP Threshold trigger

ƒ 3G to 2G ISHO may start too late

• handover process is not completed before link is lost

• higher probability of dropped call

ƒ Keeps the user in the 3G network longer

• Less ping pong effect

• Can lead to better user experience for PS service

Optimize Inter-system Handover parameters

Ave HO time for NRT service from Event 1F to receiving CCO command is 3.5 sec

Ave HO time for RT service is longer because BSIC verification is needed before HHO is executed

ƒ can have earlier threshold trigger for RT service (eg -100 dBm) as compared to NRT service (eg - 106 dBm)

Ec/No will degrade as loading and interference increase

ƒ Apart from CPICH RSCP trigger, CPICH Ec/No trigger must also be used (eg - 14dB)

Optimize Inter-system Handover parameters

CPICH RSCP of 3G macro cell drops rapidly after passing the 100dBm mark by:

-ƒ Between 5 ~ 10 dB per second for a vehicle moving along the ramp leading to basement car park

ƒ Between 10 ~ 20 dB per second upon entering the underground subway tunnel

Must ensure HO process is completed before link to macro 3G cell is lost within the tunnel or carpark

ƒ Early trigger for CPICH RSCP HO threshold is required (eg -95dBm)

ƒ MinRxlev of target 2G cell (in tunnel/carpark) should be lower than typical value

Early trigger may cause unnecessary ISHO to other 2G macro cells on street level and frequent compressed mode

measurement

ƒ 3G macro cell involved must have proper adjacency plan

Longer-term capacity improvement

Micro cells for hot spots

Shared channels (HSDPA, HSUPA) for I&B traffic

ƒ resource occupation only during transmission (minimum occupation time 2ms)

ƒ good user experience (fast switching, low latency)

ƒ efficient resource usage

Smart antennas

ƒ increases air interface capacity

ƒ availability?

ƒ costs?

Availability KPIs:

Attach set up time

Attach success rate

Detach success rate

Mobility KPIs:

LAU and RAU set up time

LAU and RAU success rate

Cell reselection time

Accessibility KPIs:

PdP Context set up time

PdP Context success rate

DNS and application server access

and many others

How does this optimization translate to Network KPIs?

Other common indicators:

DNS Lookup Time, TCP Packet Loss, Retransmission and Duplication, TCP RTT, TCP Connect and Session Completion Rate

How does this optimization translate to

Application KPIs?

FTP throughput low Non-optimised algorithms affect performance

Solutions to enhance EDGE services

Level and quality are good!

Too many cell

good for a long time

Stuck on MCS-3

And throughput doubles (30 kbps)!

Wrong neighbour definition!

Eventually it moves

to MCS-6…

Let’s filter on that Upload

4

8 8 13 10

App kbps UL LLC throughput RLC throughput

C/I BLER Cell reselections

C value per Coding Scheme Better parameters should move

the distribution to higher MCS and throughput values

FTP Downloads

6

2 1 2 0

dB and %

App kbps DL LLC DL throughput RLC DL throughput

# cell reselections C/I DL BLER

Problem

Unstable throughput

Radio analysis

Throughput BLER Cell ID

C/I < 10 dB

in this area

Long distance server

Ping Pong cell reselections

Resulting in Cell Reselections,

Interference and high BLER

Solution

Revise neighbour definition and

decrease interference and coverage

overlaps using automatic cell and

frequency planning tools

TCP analysis

Problem

Throughput is 62 kbps instead of

a potential application throughput

of over 100 kbps (2 timeslots are

used) and the average RTT is

well over 1 sec (measured on Gb)

Service summary

Good quality (BEP)

The mobile buffer limit

is reached and the throughput is affected

Throughput is unstable

in both Um and Gb

Diagnosis

TCP packet loss and out of

sequence increases the number

of packets waiting and the mobile

buffer saturates thus provoking

What would benefit the analysis?

What would benefit the analysis?

What would benefit the analysis?

What would benefit the analysis?

Optimise coverage and capacity in UMTS by reducing inter cell interference

Both automation and delta analyses are

complementary in practical processes.

Manage iRAT handovers via SHO reduction

Use different profiles for access parameters in indoor

& tunnel cells

Use process-driven software to facilitate planning and troubleshooting of multi-RAB management

Thank you for your attention

250+ customers, 200+ wireless operators

Over 5000 active users worldwide

Offices in UK, USA, Hong Kong, Sweden, China, Japan