Advanced Network Planning Telefónicaand Accenture Case Study

Advanced Network Planning

Telefónica and Accenture

Case Study

Todays Presenters

2

Markus Beckmann

Network Analytics Offering Lead

Accenture

Anni-Albers Str.11

D-80807 München

Tel: +49 175 57 68467

[email protected]

Alemu Abate

Engineering Access & Transport

Mobile Access

Telefónica Germany GmbH & Co. OHG

Georg-Brauchle-Ring 23-25

80992 München

Tel: +49 176 2442 4816

[email protected]

Telefónica Project Motivation and Objective

Accenture Approach and Solution

Discussion with Q&A

Agenda

3



Discussion with Q&A

Agenda

4

Network Operator Challenges

Why advanced analytics for network planning?

Today's challenge• High percentage of Smartphones

• Introduction of multiple data services

• Dynamic behaviour of customer, device & application

• Rapid growth of data, while voice & SMS decreases

> Simple trending of existing traffic is no longer enough

5

Network Operator Objectives

What to do going forward?• Paradigm change required to manage network capacity

• Profiling of customers, devices & applications behaviour

• Identify drivers/ variables of capacity demand

• Optimise investment between technologies (3G vs. LTE)

• Define “What-if” scenarios to verify new business strategy

> Apply analytics (tools) to monetize data traffic

6

Telefónica Advanced Network Planning

Feedback Loop to

Fine Tune Model

Forecasting Engine

f(App, Device, user, Take

rate time,...)

Statistical Analysis

Data Crunching

/ Correction

Data Processing

& Forecasting

Network PerformanceD

ata

Service

Level

Perform.

Traffic

Trend

System

ResourceUsage

CQI

Distribution

Network Configuration

Site/Sector

Capacity

Iub-

Connectivity

2G/LTE

Collocation

Clutter/

Cluster Information

Network Strategy

LTE Off-

Loading Potential

2G/3G/LTE

Interworking

Small

Cells

...

Input from Business Unit

Subscriber

Forecast

Regional

Market Share

Customer

Segment

Tariff

Groups

DPI Analysis

RNC Trace

Device

Profiling

Service

Profiling

External Trends

Market

Trend Indicators

...

External Data

Site/ Sector Level Network Data

Analysis Result per Site/ Sector

Forecasted

Traffic Load (Offered Traffic)

Required

Upgrade Type (Radio,

Baseband,...)

Time to Trigger

Upgrade (Lead Time)

Daily/ Seasonal

Trend Variation

Clustering/

Profiling of Sites/Sectors

...

Aggregated Analysis

Volume of

Upgrade per Cluster/ City

Volume of

Upgrade per

Vendor

Type of

Upgrades

Interactive

“What if”

Analysis

...

Required Output

7



Discussion with Q&A

Agenda

8

• Node (Base Station) Clustering

• Real Capacity Diagnosis

• Traffic Demand Forecasting

Accenture Project OverviewRadio Access

Network

Data ETL

• Traffic and capacity analysis

• What-if scenario simulation

• Investment optimization

Scope

• 3G data services

• Region: two major cities

Data Extraction, Transformation and Loading

• Extraction, cleaning, profiling and data loading in staging area

• Data include CDRs, performance counters, user information

etc.

Back-End Analytics

• Demand forecasting model per each node in RAN

• Real capacity diagnosis per each node in RAN

• Multidimensional clustering of RAN nodes

Front-End Functionality

• Detailed analysis of historic & future traffic per node

and compare against available capacity

• Traffic decomposition per node

(device, application, user breakdown etc.)

• Interactive ad-hoc reports that implement what-if scenarios

• Investment optimization proposals

Internal &

External Data

Sources

9

Hardware utilization

information (hourly

counters)

Traffic Busy Hour Capacity

Diagnosis (per Cell)

Radio link quality

information (hourly

counters)

User Busy Hour Capacity

Diagnosis (per Node)

Daily or Busy Hour Traffic

Forecast (per Node)

Daily or Busy Hour User

Forecast (per Node)

Daily/Node to Busy

Hour/Cell Traffic Mapping

Daily/Node to Busy

Hour/Node User Mapping

Target threshold & what-if

modifiers

Probability to reach

traffic capacity threshold

Probability to reach user

capacity threshold

Target threshold & what-if

modifiers

Weighting and

Prioritization Rules

and Dimensioning

Goals

Prioritized

Node

Upgrade List

Node Clustering Insights

Pricing Information

Capacity Diagnosis Demand Forecasting Gap Analysis Investment Optimization

Upgrade List

End-to-End Approach

10

• Several attributes per node

used for profiling (daily

traffic, signal to traffic ratio,

2G to 3G ratio, traffic per

user, device decomposition,

busy hour distribution etc.)

• Bayesian hierarchical

clustering algorithm applied

• The Hierarchical tree and

associated statistical

measures are used to

determine the optimal

number of clusters

Multidimensional Node Clustering

• Cluster are homogenous

group of nodes

• Detailed profile per cluster to

reveal the predominant

attributes

• Business logic is applied to

leverage the clustering

insights in network upgrade

11

Capacity Diagnosis Methodology

Example:

Detecting saturation

on hardware resources

utilization

Example:

Analyzing the effect

of radio link

quality to cell’s

maximum throughput

The capacity for each individual base station is derived using a data

driven approach rather than relying on generic vendor’s specifications

Traffic Capacity Factors• Limitation on downlink traffic throughput per cell

• Capacity mainly depends on radio link quality but can also be

affected by other factors

• Radio link quality is reported by the user equipment through

the CQIs

• Node’s capacity increased by adding more cells

User Capacity Factors• Limitation on simultaneous users per cell either on downlink or

uplink directions

• Additional limitation on simultaneous users per node (across

cells) either on downlink or uplink direction

• Capacity depends on Node B’s available hardware resources

• Node’s capacity increased by adding

more hardware resources

12

Demand Forecasting Methodology

Model Output (per node)

Goodness of fit metrics (R2 etc.)

Component Decomposition

Coefficient & Significance per Component

Forecasts with Confidence Limits per Component

Forecast validation metrics (sMAPE etc.)

Model Input (per node)

Historic information over 2 years (traffic & user demand, capacity

upgrades, special events etc.) are provided as input to the model

Unobserved Component Model (UCM) has been selected for solving the

forecasting problem

• It estimates automatically deterministic or stochastic unobserved

components

• It provides forecasts through the application of Kalman filtering

• It incorporates both autoregressive and explanatory regression terms

Dimensioning

Model Training

Model Testing

Model Refresh

Model Forecast

13

Generated Insights & Business Value

Lack of granular data-driven predictive insights

prohibits legacy approaches to utilize network

capacity upgrade budget at maximum efficiency

• Over-investment: 66% of total upgrades were planned on

nodes with low saturation risk and low value

• Under-investment: 80% of nodes with high saturation risk

were not the planned for upgrades

In the case study example roughly 41% of the

planned CAPEX would have to be reallocated to

tackle the under-investment problem and an

additional 25% could be to deferred to limit

over-investment

Increased Quality of Experience for end

customers and maximized Return on Investment

for the network operator can be achieved at the

same time

Under-investment

Over-investment

14

Interactive

What-If

Simulation

Interactive

Investment

Prioritization

Advanced Network Planning ToolA powerful front-end offers ad hoc data exploration in a consolidated data-mart

and evaluation of custom “What-If” scenarios

Web-based frontend

for Network Planners

Strategic Planning

with Analytical

Insights

15



Discussion with Q&A

Agenda

16

Discussion

17

18

Annex

Accenture Network Analytics

Capacity Planning

Provide accurate forecasts of

traffic volume and congestion to aid

network planners make the best

CAPEX investments for network

capacity

Service Assurance

Use Network Analytics to

understand how equipment,

devices and links enhance the

customer experience through

network event correlation

Network Control

Optimize resource allocation,

reduce OPEX to enable network

control factoring in congestion,

routing, and scheduling.

Solution Areas

Marketing

Engineering &

Planning

Portal (Business Views)

Batch and Online Processing

Batch ETLOnline

CPE

Engineering &

Planning


Real Time

Engine

Service

O&M

Analysis &

Reporting

Batch and Online Processing

Batch ETLOnline

CPE

Quality &

Operations

Analysis &

Reporting

Service

O&M

Real Time

Engine

Marketing

Engineering &

Planning


Analysis &

Reporting

Service

O&M

Real Time

EngineBatch and Online Processing

Batch ETLOnline

CPE

Quality &

Operations

Telefónica Germany Scope

19

Demand Forecasting: How it works?

20

Demand Forecasting: How accurate is it?

In these

exemplary

nodes,

forecasts (red)

are very close

to actual

demand (blue)

despite the

noisy nature of

the data.

21

“We Speak Analytics”

Accenture provides Business Analytics Services around the globe with more than 13,000

dedicated resources and a strong set of industry-leading insights

Depending on the underlying business requirements,

Accenture leverages its experience from more than

13,000 resources worldwide, including areas like:

• 6,500 Business Intelligence

• 3,500 Business Function Insights

• 3,000 Data & Technology Management

Accenture

Analytics

Business

ValueSupply

Chain

Commercial Business

Intellig.

Organisation

& Talent

Big

Data

22To find out more, click here

Advanced Network Planning Telefónicaand Accenture Case Study

Documents