orc volcano urban umts calibration

7/24/2019 Orc Volcano Urban Umts Calibration

1/93

SIRADEL Confidential - Copyright 2015

Workshop

UMTS Band I Volcano URBAN Model calibration

ORANGE CAMEROUN

30/09/2015


2/93

SIRADEL Confidential - Copyright 2015 2

Context

Volcano Urban Model review

Map Data overview

Volcano calibration guidelines

Volcano calibration workflow

Results, Conclusion and Recommendations

And now? Perspectives?

Agenda


3/93


Context


4/93


SIRADEL was founded in 1994 in Rennes,

Currently is also present in North America and China.

Partner with the world biggest telecom suppliers

SIRADEL[1/2]


5/93


SIRADEL has a large experience on RF models calibration, performing dozensof projects each year world wide.

Calibration using either CW or SCAN or both.

All wireless bands and technologies: 450MHz, 800MHz, 900MHz, 1800MHz,2100MHz, 2500MHz, 2600MHz, 5.2GHz, and even on milimetric bands (ex:60GHz).

SIRADEL[2/2]


6/93


Orange intend to improve 3G network quality.

To do so, Orange needs to dispose of accurate propagation models to improveradio planning.

Therefore Orange requested SIRADEL services to calibrate two Volcano URBANmodels on UMTS band 2100MHz, in Douala and Yaound.

SIRADEL perform CW and SCAN measurement in the UMTS band used byOrange in Douala and Yaound.

High resolution 3D map data was used to describe the environment.

Project


7/93SIRADEL Confidential - Copyright 2015

Volcano Urban Modelreview


8/93SIRADEL Confidential - Copyright 2015 8

Suitable to dominant and non-dominant antennas (rooftop, podium, buildingwalls, street-level)

Multi-path option for path-loss calculation (instead of DP - direct pathonly)

Model based on

Deygout method (to calculate multiple knife-edge diffraction)

UTD (Uniform Theory of Diffraction)

Multi-path path loss is computed as follows:

2D ray trajectories constructed in the horizontal plane (reflections and diffractions

on building facades + transmissions allowed)

Then a vertical analysis (3D) is performed along each single ray trajectory,unfolding the trajectory in vertical.

Urban ModelConcepts



Transmitter located well above surrounding environment

Most of the radio energy is propagated above top obstacles

Direct Path is the dominant path

Multiple paths effect is meaningless

Recommendation: Use Direct pathModelling

Urban ModelDominant Antenna

Site PDC @Yaound



Urban ModelNon Dominant Antenna

Antenna located on a Rooftop, Terrace or against Building facade withhigher buildings around

Radio energy propagated by diffraction and partly via wave-guidingeffects along streets

Direct path is seldom dominant

Multiple paths effect is significant even dominant

Recommendation: Use multi pathmodelling

Site La Libert @Douala Example of a major MP environment


11/93

To answer the demanding's of these two distinct environments, SIRADELpropose two different propagation mechanisms.


Urban Model

MP Contributions [1/4]


12/93


(5)

Multi-paths contributions in predictions include:

Direct path(1)with diffraction above rooftops

Reflections(2)/ Diffractions(3)on the vertical building facades

Transmission(4)permitted between Tx and the first interaction

Reflectionagainst building podiums(5)

Urban Model



13/93

Multi-paths contributions in predictions include:

Backward contributions

The first backward interaction should be in LOS

This kind effect its often seen on the live network when an obstacle is locatedin front of the cell and the cell became best server rearwards.


Urban Model


Site Stamatiades @Douala


14/93


Urban Model


MULTIPLE KNIFE-EDGE DIFFRACTION - The Deygout Method Extract vertical profile

Used by SIRADEL to describe the Tx-Rx direct path and multi path.

Uniform Theory of Diffraction (UTD)

Used by SIRADEL to describe multi path profiles.

In complex dense urban environments where the link path is described by

complex reflection and diffraction events, Deygout method start loosing his

accuracy.

Based on research and projects experience outcome, SIRADEL R&D hasmodified both UTD and Deygout original algorithms to enhancement

extraction accuracy and reduce computation time.


15/93


Urban Model

Multipath area definition

Beyond a given distance from transmitter, it has been proved that direct path

propagation is predominant and multiple-path insignificant.

Multiple path contributions is added value mainly nearby transmitters

To optimize efficiency, multi path prediction depends on the distance from thetransmitter:

*100 meters long

Near Region All multi-path contributions are calculated

Far Region Only the direct path is computed

TransitionRegion*

A linear transition is carried out from the

field strength calculated in near region and

the field strength calculated in far region


16/93


Urban Model

Signals Nature [1/7]

= + +Measurements

The signals are made of 3 components

Path Loss Determinist component, extracted with classical propagation laws

Shadowing due to Multi-path trajectories, extracted with Volcano assets using

accurate 3D map data + CW Calibration to extract specific environmental

characteristics.

Fast Fading unpredictable random variable, not considered by the model


17/93


Urban Model

Fast Fading Filtering[2/7]

Space-time filtering

Fast fadingmay introduce 30-40 dB signal variations along \2 distances

To remove fast variations in the measured signal (due to channel fluctuations,

etc)

To preserve only large variations (shadowing)


18/93


19/93


Clutter loss is applied to the predicted path loss according to followingrules and clutter/vector type where receiver is located:

Urban Model

Clutter Loss [4/7]


20/93


Urban Model

Parameters[5/7]

Propagation parameters for Volcano Urban Model (MP and DP)

Parameters Description Typical value Range

AfarCorrection of free space loss predicted at 1

meter for the receivers for direct path0 (dB) [-5 ; 5]

Bfar Correction of the distance dependence forthe receivers for direct path 24 (dB.dec-1) [18 ; 32 ]

Weight of the deterministic attenuation 0.7 [0.5 ; 1 ]

Antenna

Correction

Correction of the antenna diagram due to

the impact of the surrounding environmentHigh

Standard

Medium

High


21/93


Urban Model

Parameters[6/7]

Propagation parameters for Volcano Urban Model (MP only).

* Useful to predict coverage on ground level micro cells.

Parameters DescriptionTypical

valueRange

Anear/Bnear Free space corrections in the near area 0/23 (dB)[-5 ; 5]

[18 ; 32 ]

Radius

Multi path radius

800m

[0 ; [

Street Level

Mode2D ray tracing option* N/A N/A, Flat, Hilly


22/93


Urban Model

Parameters[7/7]

Propagation parameters for Volcano Urban Model (MP only).

Note:There is no limitation on the number of diffractions for vertical profileextraction (DP).

* For each single path

Parameters DescriptionTypical

valueRange

Wr Weighting of reflections 1 [0.5 ; 1 ]

Wd Weighting of diffractions 1 [0.5 ; 1 ]

Wb Weighting of backward 1 [0.5 ; 1 ]

Diffraction Maximum number of diffractions* 1 [0 ; 1 ]

Reflection Maximum number of reflections* 2 [0 ; 4 ]


23/93

Volcano Urban general settings

Onsettings tab the user is able to configure inputs, outputs and requirements of thesimulation.


Urban Model

Settings

Vertical profile extraction mode (Raster/Vector)

Absolute Rx height reference (Ground/Sea)

Required type of prediction (ex: Along linear)

Resolution accuracy for multi-resolution raster data

Rx height reference (Ground/Clutter)

Resolving mechanism for Tx inside building(Indoor/Outdoor)

Accuracy of the output coverage plot(Yes/No)


24/93


Volcano Types

Volcano type shall be linked to each clutter type imported from map data*, either

for vector or raster.

6 Volcano types available (+ Linear**):

* filled out automatically using menu.vol

**Only available with linear vectors, ex: GSM-R for Railways

3D high resolution

clutters

2D Low resolution

clutters (raster only)


25/93

This feature allows the user to filter out on Volcano Typeor Clutter Type. Path loss on excluded clutters is not computed*.

*Neverthless, the excluded clutter is taken into account for path loss calculation elsewhere.


Clutter exclusion (1/2)


26/93


Volcano returns the results for all clutters (raster or vector) except the

excluded ones.

Clutter exclusion (2/2)

Building Exclusion Building not excluded


27/93


Map DataOverview

G hi l D t L


28/93


To describe the environment different kinds of map data are available

Map data formats supported by Volcano in Atoll

Raster: Planet, Asset, BIL and MapInfo

Vector: MapInfo (TAB), Planet and Shape

Raster data:

Heights/Altitude (DTM Digital Terrain Model) Optional but highly recommended

Clutter/Land use (DLU Digital Land Use) If not available, polygon vector required to describe clutter Clutter heights/Building heights (DHM Digital Height Model) if not available, clutters height need to

be manually defined (or through vector data)

Elevation (DEM Digital Elevation Model, DEM = DTM + DHM) seldom used

High Resolution +-5mts Low Resolution +-30mtsDTM(pixel value = ground level

above mean sea level) DLU(pixel value = clutter type)

DHM(pixel value = clutter height

above ground level)

Geographical Data Layers

[1/4]

G hi l D t L


29/93


Raster data

Low resolution Versus High resolution

Low resolution (DTM + DLU)

Statistical description of the environment

Unable to distinguish objects (ex: houses, routes)

No information about clutter heights average user heights per clutter

Resolution usually between 20m & 100m

High resolution (DTM + DLU + DHM)

Deterministic description of the environment

Accurate description of the clutter types, able to distinguish buildings

objects (ex: houses, routes)

Accurate clutter height information

Resolution usually between 1m & 10m


[2/4]

G hi l D t L


30/93


Vector data

3D polygon vector data

Accurate description of shapesof clutters/buildings (2D) ~ DLU raster

describe both the shape and heights of clutters(3D) ~ DLU+DHM raster

Mandatory for RF predictions considering multi-paths

2D linear vector data (streets, railways, rivers)


[3/4]

3D Vector 2D Vector

G hi l D t L


31/93


Volcano is able to combine automatically:

Multi-resolution raster (High and Low resolution raster) Vector/Raster combination

Useful to shorten the processing time either on mixed environments (Urban\Rural)or on profile extraction (MP\DP),


[4/4]

LR Vs HR @Douala

U b M d l


32/93


Urban Model

Map Data settings

Since both 3D raster and vector are available, the user is able to choose

between 3 modes for vertical profile extraction.

Vector

Favourite

- Transmitter and receiver locations outdoor/indoor are estimated regarding vector data (Undefined portions are

filled afterwards by extracting complementary clutter information from the 3D raster layer).

- DTM raster and 3D vectors are used for the whole computation (Multipath and Direct path detection).

Raster

Favourite

- Vertical terrain profile between the transmitter and the receiver is preferably extracted from the raster layers

when available.

- Otherwise 3D vectors will be used for vertical profile extraction.

- 3D vectors still be used for multipath detection.

Raster only

- Only raster layers will be used to extract the vertical terrain profile between the transmitter and the receiver.

-3D vectors still be used for multipath detection.

- improve speed calculation => Recommended by SIRADEL


33/93


Volcano Calibration

Guidelines


34/93


Volcano tuning processCW calibration[1/2]

Measurement

Retrieve enough samples to characterize the clutter(it depends on the clutter size )

Samples spreading should be homogenous ( web spider)

Noise level marks the border

Accurate extraction of CW field configuration (Power, X/Y, HBA..)

Verification in Atoll

Copy CW configuration to atoll project (Power, X/Y, HBA..)

Analyze des samples(Filter Noise, exclude points inside buildings)

Analysis

Get Statistics for all the routes - Correlation, mean error

analyze high deviations and troubleshoot

Routes ValidationExtract Volcano Parameters for 33% of the routes

Apply extracted parameters on the remaining 66% of the routes

Compare Results and analyse

Final Values

Extract Volcano Parameters with all routes

Analysis

Verificationin Atoll

CWMeasurement

Validation

Final Values, A, B, W

V l i


35/93


Volcano tuning processCW model adjustment[2/2]

Validation& Antennacorrection

Verificationin Atoll

SCANMeasurement

ErrorAdjustment

Measurement

Measure the same routes as CW in SCAN mode

Extract Best Server

Verification in Atoll

Verify repeated scrambling codes(doublons)

Verify and correct 3G Atoll project as much as possible (X/Y, C-PICH, Azimut)

Analyze des samples (Exclude points inside buildings, or with wrong X/Y)

Validation & Antenna Correction

Apply CW calibrated model on SCAN Best Server => Get Statistics

Analyze Performance per SC (Correlation, Error, Std deviation)

Verify the best option for the Antenna Correction (Standard, High, Medium)

Error Adjustment

Compare SCAN measured Routes with the prediction based on CW model

Analyse and understand the gap

Adjust the model offset (error)


36/93


Volcano Calibration

Project management


37/93


Project

Scope of Work

Calibrate three urban models (Douala, Yaound and global) for UMTS Band I

(2100MHz).

Perform CW and SCAN measures on Douala and Yaound

14 routes (7 routes by town)

Contractual KPIs

*Calibration KssI with CW data

Global mean error = 0dB

Global standard deviation error ~6dB

Global correlation >85% Global Mean Error dispersion = 1.5dB

Mean error = 0dB

Global standard deviation error 70% Global Mean Error dispersion < 3dB

Targets* Commitments*

M t[1/5]


38/93


Measurement[1/5]CW Transmitter

As Radiating element we used the antenna Kathrein 741790, with 11dBi gain in the

frequency range 1920-2170MHz, mounted in a 6 meters mast.

As signal generated we used the ETSA radio transmitter, tuned on2152.5MHz with44dBm of output power.

M t[2/5]


39/93


Measurement[2/5]Mobile Station

For CW measurements we used a JDSU Receiver with E6474 Wireless Network

Optimization platform.

For SCAN measurements we used R&S TSMW, running ROMES platform.

Meas rement[3/5]


40/93


Measurement[3/5]Link Budget CW

Link Budget

*Approximated value after cable loss

Transmitter

Tx power Cable losses Tx antenna gain

Receiver

Rx antenna gain Cable losses

Frequency

Measured

Tx power

(dBm)

Tx antenna

gain (dBi)

Rx antenna

gain (dBi)

Rx cable

losses (dB)

Receiver

sensitivity

(dBm)

2152.5 MHz 42* 11 2,15 1,5 -124

Pathloss Free space losses

Measurement[4/5]


41/93


Measurement[4/5]Route

Spider web

Very dense near the site and wider further from the site

homogenous

Fair distribution of the bins in direction and distance around the transmitter

Contains a fair representation of the environment

Typical ranges for a Urban cell:

Dmax (Tx-Rx) up to 10Km

Length route > 50Km

bins > 10 000

Measurement[5/5]


42/93

Sometimes is not possible to shape the ideal spider web

Routes in bad conditions

Traffic jam

Dead ends


Measurement[5/5]Route


43/93


Site Survey

For this project we re-used the same sites of the previous project*.

*Except Ndokoti, replaced by Logbaba

Nom du site Longitude Latitude

Douala

Du Levant 9.672399516 4.086301652

La_Rose 9.686541508 4.041411402

Liberte 9.698922393 4.05590185

Madagascar 9.738478621 4.016136071

Stamatiades 9.698977532 4.046674689

Logbaba 9.76150 4.03975

Deido-Lycee 9.714712608 4.064273801

Nom du site Longitude Latitude

Yaound

Mendong 11.47091667 3.831222

H-Franco 11.51921928 3.882236

H-ville 11.51758611 3.872028

Mercure 11.51921206 3.864594

PDC 11.5013572 3.891482

Mvog-Betsi 11.48133333 3.866028

Polytechnique 11.50133056 3.863547


44/93


45/93


Frequencies [1/2]

Orange has 3 UMTS Band I carriers. Only 2 carriers are commercially on service, therefore we profited to use the spare

one to do the CW measurements.

SCAN measurements were done on the 2ndcarrier (expected to be less loaded).

1st Carrier 2nd Carrier

3rd Carrier


46/93


Frequencies [2/2]

With the JDSU Receiver we are able to measure simultaneously the CW signal and

the noise level.

The CW signal was defined on the center frequency of the 3rdOrange carrier.

The Noise window was defined on 2153MHz frequency.

Noise window used to detect external interferences.

NoiseBw12.5kHz

CW 2152.5 MHZ


47/93


Atoll Project [1/2]

In the first place we analysed the 3G Atoll project of Yaound and Douala

59 Cells inside buildings ( Building height > Cell Height) => Corrected (height building + 1

meter).

Around 100 cells with a wrong C-PICH power value => Corrected with OMC values


48/93


Atoll Project[2/2]

Yet, Impossible to quantify and correct

Sites wrongly placed

Wrong cell position on roof tops.

This play a major role on the error prediction mainly on the range [0-100] meters.


49/93

To characterize the environment we used high resolution map data developed by

SIRADELs cartography Team.

The map data is available for Douala and Yaound with 5 meters resolution.SIRADEL Confidential - Copyright 2015 49

Map Data [1/2]

DTM DLUDHM Vector


50/93

For Douala we also dispose of DLU and DTM low resolution layers.

These LR layers will be useful to predict CW coverage for the site Du_Levant, whichneeds to cross the river (LR clutter). Otherwise all the points on the other side of theriver will cross an invalid area, and therefore will result in invalid points.


DTM DLU

Map Data [2/2]

Volcano Urban Configuration


51/93

Define Map Data

* Vector Data still used for Multi path calculationSIRADEL Confidential - Copyright 2015 51

Volcano Urban Configuration[1/3]

Raster only used for verticalprofile extraction andindoor/outdoor detection*.

Reducing Time computation



52/93



Main settings

These settings are the

default ones, for theVolcano Urban model.

They will be automaticallyadjusted by thecalibration process.



53/93

Clutter Type is automatically linked to Volcano type through menu file


Buildings are excluded from

calibration and validation to filterout erroneous indoor meas. points.


Using Menu.vol each clutter typewill be automatically linked to a

Volcano type .


54/93


Volcano Calibration

CW Tuning

Volcano Urban


55/93


CW Yaound

Douala MP & DP*

Global

* Only step 3

SCAN Yaound

Douala

Global

Results, Conclusions and Recommendations

Volcano Urban

Calibration Workflow

1. Calibrate , A, B with 33% of the routes

2. Get statistics with 66% of the routes

3. Get , A, B for all routes

Analyze Performance per SC (Correlation, Error)

Verify which is the best option for the Antenna Correction (Standard, High, Medium)

Evaluate SCAN Vs Prediction

Measurements[1/2]


56/93


Measurements[1/2]Yaound

CW Route summary

*Averaged but not filtered

YAONDE

Nom du site Nb of points*Mendong 19942

H-Franco 18570

H-ville 21598

Mercure 14459

PDC 22460

Mvog-Betsi 17725

Polytechnique 17485

Mendong H-Franco H-Ville

MercurePDC Mvog-Betsi Polytechnique

Measurements [2/2]


57/93


Measurements [2/2]Yaound

Route measurements summary

To avoid CW samples reshape by noise, we filtered out all samples lower than -110dBm

Noise Level Average by Route CW Drive Test RxLev Mean Vs Distance

-118

-116

-114

-112

-110

Polytechniqu

e

Mercure

Mvog-Bet

si

H-Franc

o

PD

C

H-Ville

Mendon

g

RxLev(dB

m)

0

1000

2000

3000

4000

5000

6000

7000

-120

-110

-100

-90

-80

-70

-60

-50

-40

[0

;250]

[250

;500]

[500

;750]

[750;

1000]

[1000;

1250]

[1250;

1500]

[1500;

1750]

[1750;

2000]

[2000;

2250]

[2250;

2500]

[2500;

2750]

[2750;

3000]

[3000;

3250]

[3250;

3500]

[3500;

3750]

[3750;

4000]

[4000;

4250]

[4250;

4500]

[4500;

4750]

[4750;

5000]

dB

Distance (mts)

NbMeas Meas_Mean[dBm]

Model Tuning MP[1/4]


58/93


Model Tuning MP[1/4]Yaound

On the first stage we performed the model calibration using 1/3 of CW Routes.

Returning the following settings:

These good KPIs will be a constant throughout all project.

Route name NbPredPoints MeanError(dB) StdError(dB) RMSError(dB) CorrelationFactor

Polytechnique 18084 0.37 6.12 6.13 0.91

Mercure 14667 -0.09 6.39 6.39 0.91

Mvog-Betsi 18068 -0.3 6.61 6.62 0.87

Total 50819 0 6.38 6.38 0.9

Mean error dispersion(dB): 0.27

Parameters Value

a_near 2.75

a_far 0.08

b_near 23.06

b_far 24.05alpha 0.50

weighting_r 0.50

weighting_d 1.0

weighting_h 1.0



59/93


To validate the extracted model we apply it on the remaining 2/3 of the routes, and the

high performance should be upheld:

All routes presents good performances and can be used for calibration purposes.

These results shows the model accuracy when the model is spread over other urban

areas.



H-Franco 18788 -0.61 6.65 6.68 0.88

PDC 22603 -3 6.58 7.23 0.88

H-Ville 22156 -0.65 6.46 6.49 0.89

Mendong 20922 -1.75 5.87 6.13 0.92

Total 84469 -1.54 6.47 6.65 0.9




60/93


At last to obtain the final CW based model for Yaound, we use all the routes toextract the calibrated parameters.



H-Franco 18788 0.34 6.63 6.64 0.88

H-Ville 22156 0.29 6.44 6.45 0.89

Mendong 20922 -0.75 5.87 5.92 0.92

Mercure 14667 0.82 6.41 6.47 0.91

Mvog-Betsi 18068 0.68 6.61 6.64 0.87

PDC 22603 -2.01 6.57 6.87 0.88

Polytechnique 18084 1.32 6.1 6.24 0.91

Total 135288 0 6.47 6.47 0.9Mean error

dispersion(dB):0.9

Parameters Value

a_near 2.76

a_far 0.1

b_near 22.78

b_far 23.76

alpha 0.50

weighting_r 0.50

weighting_d 0.55

weighting_h 1.0

Final Parametersfor Yaound

Urban Model.

Model Tuning DP [4/4]


61/93


Model DP global

These results shows that the DP model also gives an accurate and trustableprediction of Yaound environment.

Model Tuning DP[4/4]Yaound

Final Parameters

for YaoundUrban Model DP.


H-Franco 18788 0. 35 6.64 6.65 0.88

H-Ville 22156 0.24 6.43 6.44 0.89

Mendong 20922 -0.75 6.89 5.94 0.92

Mercure 14667 0.69 6.55 6.58 0.9

Mvog-Betsi 18068 0.73 6.6 6.64 0.87

PDC 22603 -1.95 6.57 6.86 0.88Polytechnique 18084 1.35 6.11 6.26 0.91

Total 135288 0 6.49 6.49 0.9Mean error

dispersion(dB):0.89

Parameters Value

a_far 0.04b_far 23.76

alpha 0.5

Measurements [1/2]


62/93


Measurements [1/2]Douala


*Averaged but not filtered

Douala

Nom du site Nb of points (sampled)

Libert 13661

Du_Levant 14027

La_Rose 18117

Madagascar 14801

Stamatiades 13494

Logbaba 15079

Deido-Lycee 18429

Libert

Madagascar Du Levant StamatiadesLogbaba

Douala

Nom du site Nb of points*

Libert 14879

Du_Levant 14027

La_Rose 18117

Madagascar 14801

Stamatiades 13494

Logbaba 15079

Deido-Lycee 18429

Deido-lycee La_Rose

Measurements [2/2]


63/93


easu e e ts [ / ]Douala


To avoid CW samples reshape by noise, we filtered out all samples lower than -110dBm

-118

-116

-114

-112

-110

Liber

t

La_

Rose

Deido_

Lycee

Madagasc

ar

Du_

Leva

nt

Stamatiades

Nogbaba

RxLev(dBm)

0

1000

2000

3000

4000

5000

6000

7000

-120

-110

-100

-90

-80

-70

-60

-50

-40

[0;250]

[250;500]

[500;750]

[750;1000]

[1000;1250]

[1250;1500]

[1500;1750]

[1750;2000]

[2000;2250]

[2250;2500]

[2500;2750]

[2750;3000]

[3000;3250]

[3250;3500]

[3500;3750]

[3750;4000]

[4000;4250]

[4250;4500]

[4500;4750]

[4750;5000]

dB

Distance (mts)

NbMeas Meas_Mean[dBm]

Noise Level Average by Route CW Drive Test Mesure Vs Distance



64/93


g [ ]Douala

On the first stage we perform the model calibration using 1/3 of CW Routes

Weve got the following parameters


Libert 14448 0.76 5.67 5.72 0.9

La_Rose 17957 -0.99 5.6 5.68 0.89

Deido_Lycee 18345 0.37 5.18 5.19 0.91

Total 50750 0 5.52 5.52 0.9


Parameters Value

a_near 2.71

a_far 0.0

b_near 22.62b_far 23.62

alpha 0.5

weighting_r 0.5

weighting_d 0.71

weighting_h 0.92



65/93


Applying the calibrated parameters on the remaining 2/3 of the routes we obtain

the following KPIs:

Once again we obtain an good accuracy when the model is spread over theother areas.

All routes are inside the target KPIs and can be used for calibration.

g [ ]Douala


Madagascar 13845 0.41 5.77 5.78 0.92

Du_Levant 13916 1.25 4.84 5 0.9

Logbaba 13777 3.84 5.51 6.72 0.9

Stamatiades 13300 -0.1 5.27 5.28 0.9Total 54838 1.36 5.57 5.73 0.9




66/93


Finally to obtain the MP model, we calibrate the model with all 7 routes

g [ ]Douala

Final Parametersfor Douala Urban

Model.


Libert 14448 0.05 5.66 5.66 0.91

La_Rose 17957 -1.69 5.61 5.86 0.89

Deido_Lycee 18345 -0.31 5.17 5.18 0.91

Madagascar 13845 -0.3 5.76 5.77 0.92

Du_Levant 13916 0.51 4.85 4.88 0.93

Stamatiades 13300 -0.81 5.27 5.33 0.9Logbaba 13777 3.14 5.51 6.34 0.9

Total 105598 0 5.59 5.59 0.9


Parameters Value

a_near 2.7

a_far -0.01b_near 22.82

b_far 23.83

alpha 0.5

weighting_r 0.5

weighting_d 0.63

weighting_h 1

Model Tuning DP[4/4]


67/93


Model DP global

Also for Douala, the DP model achieve an accurate and trustable prediction ofthe environment.


Libert 14448 -0.01 5.77 5.77 0.9

La_Rose 17957 -1.76 5.61 5.88 0.89

Deido_Lycee 18345 -0.3 5.2 5.21 0.91

Madagascar 13845 -0.27 5.77 5.78 0.92

Du_Levant 13916 0.55 4.88 4.91 0.9

Stamatiades 13330 -0.78 5.29 5.34 0.9Logbaba 13777 3.17 5.5 6.35 0.9

Total 105598 0 5.62 5.62 0.9


g [ ]Douala

Final Parameters

for Douala UrbanModel DP.

Parameters Value

a_far -0.01

b_far 23.81alpha 0.5



68/93


g [ ]Global

Finally we calibrate the global model

On the first stage we perform the model calibration using 1/3 of CW Routes (3 Yaound+ 3 Douala).

Weve got the following parameters


Dou- Libert 14448 -0.02 5.65 5.65 0.91

Dou - La_Rose 17957 -1.76 5.62 5.89 0.89

Dou - Deido_Lycee 18345 -0.38 5.17 5.19 0.91Yde Polytechnique 18084 1.13 6.11 6.21 0.91

Yde Mercure 14667 0.65 6.4 6.43 0.91

Yde - Mvog-Betsi 18068 0.49 6.61 6.63 0.87

Total 101569 0 6.02 6.02 0.9


Parameters Value

a_near 2.74

a_far 0.06

b_near 22.84

b_far 23.83

alpha 0.5

weighting_r 0.5

weighting_d 0.88

weighting_h 0.98



69/93


Applying the calibrated parameters on the remaining 2/3 of the routes we obtain

the following KPIs:

This results proves the model accuracy when we spread the model in a globalway through Yaound and Douala.

g [ ]Global


H-Franco 18788 0.15 6.64 6.64 0.88

PDC 22603 -2.21 6.57 6.93 0.88

H-Ville 22156 0.11 6.45 6.45 0.89

Mendong 20922 -0.95 5.87 5.95 0.92

Madagascar 13845 -0.38 5.76 5.78 0.92

Du_Levant 13916 0.44 4.85 4.87 0.9Stamatiades 13300 -0.89 5.27 5.34 0.9

Logbaba 13777 3.06 5.51 6.31 0.9

Total 139307 -0.24 6.16 6.16 0.9




70/93


Finally to obtain the Global MP model, we calibrate the model with all 14 routes

g [ ]Global

Final Parametersfor Global MPUrban Model.

Route name NbPredPoints MeanError(dB) StdError(dB) RMSError(dB) CorrelationFactorDou- Libert 14448 0.12 5.66 5.66 0.9

Dou - La_Rose 17957 -1.63 5.61 5.84 0.89

Dou - Deido_Lycee 18345 -0.24 5.18 5.18 0.91

Yde Polytechnique 18084 1.27 6.11 6.24 0.91

Yde Mercure 14667 0.78 6.41 6.46 0.91

Yde - Mvog-Betsi 18068 0.63 6.61 6.64 0.87

Yde - H-Franco 18788 0.29 6.64 6.64 0.88

Yde PDC 22603 -2.06 6.57 6.89 0.88

Yde - H-Ville 22156 0.24 6.44 6.45 0.89Yde Mendong 20922 -0.81 5.87 5.93 0.92

Dou Madagascar 13845 -0.24 5.76 5.77 0.92

Dou - Du_Levant 13916 0.58 4.85 4.88 0.93

Dou Stamatiades 13300 -0.75 5.27 5.32 0.9

Dou - Logbaba 13777 3.2 5.51 6.37 0.9

Total 240886 0 6.1 6.1 0.9


Parameters Value

a_near 2.73

a_far 0.05

b_near 22.80

b_far 23.79

alpha 0.5

weighting_r 0.5

weighting_d 0.55

weighting_h 1

Model Tuning DP[4/4]


71/93


Model DP global

Global

Final Parametersfor Global Urban

Model DP.

Route name NbPoints MeanError(dB) StdError(dB) RMSError(dB) CorrelationFactorDou- Libert 14448 0.05 5.78 5.78 0.9

Dou - La_Rose 17957 -1.7 5.61 5.86 0.89

Dou - Deido_Lycee 18345 -0.24 5.2 5.21 0.91

Yde Polytechnique 18084 1.31 6.12 6.25 0.91

Yde Mercure 14667 0.65 6.55 6.58 0.9

Yde - Mvog-Betsi 18068 0.69 6.6 6.64 0.87

Yde - H-Franco 18788 0.31 6.64 6.65 0.88

Yde PDC 22603 -2 6.58 6.87 0.88

Yde - H-Ville 22156 0.19 6.43 6.44 0.89Yde Mendong 20922 -0.8 5.89 5.95 0.92

Dou Madagascar 13845 -0.2 5.77 5.78 0.92

Dou - Du_Levant 13916 0.61 4.87 4.91 0.9

Dou Stamatiades 13300 -0.72 5.29 5.33 0.9

Dou - Logbaba 15048 3.23 5.5 6.38 0.9

Total 240886 0 6.12 6.12 0.9


Parameters Value

a_far 0.02

b_far 23.78

alpha 0.5


72/93


Volcano Calibration

SCAN Adjustment

Model Adjustment [1/4]Y d


73/93


To verify and adjust the mean error of the calibrated model we use the SCAN data.

To avoid SC doublons we had to divide Yaound in three areas.

For KPIs analyses purposes, we merged each route in a single global route, thus theresults presented below consider only the global route.

Yaound



74/93

UMTS SCAN routes 115651* samples were used for SCAN evaluation

A mixed of suburban and urban areas was considered.

*After averaging and filtering


Yaound



75/93


Yaound

To adjust the Antenna Correction parameter we extracted the SCAN KPIs using the

calibrated model for the 3 available configurations (High, medium, standard). The correction method having the lower variance in the antenna range [-60:60] should

be employed.

The Antenna Correction method High has is the most stable .



76/93


Yaound

Finally using internal proprietary tools developed by SIRADEL R&D, which have proved

their accuracy throughout the years on many projects, we attain the error between CWmodel and SCAN measurements.

We got an error of -0.52dB.

These results shows how accurate is the Calibrated model.

Model Adjustment [1/4]D l


77/93


Once again to avoid SC doublons, we divided the Douala in two areas.

For KPIs analyses purposes, we merged each route in a single global route, thus theresults presented below consider only the global route.

Douala

Model Adjustment [2/4]D l


78/93


UMTS SCAN Best server routes 97034* samples were used for SCAN evaluation

Suburban areas are less represented because CW sites are mostly concentrated onthe city center and also because its harder to drive on the suburban side.

*After averaging and filtering

Douala

Model Adjustment [3/4]Douala


79/93


Douala

To adjust Antenna Correction parameter we extracted the SCAN KPIs over the 2 zones

using the calibrated model for the 3 available configurations (High, medium, standard). The correction method having the lower variance in the antenna range [-60:60] should

be employed.

The Antenna Correction method High has the lower error and is more stable .

Model Adjustment [4/4]Douala


80/93


Douala

Again using SIRADEL internal tools, we attain the error between CW based model and

SCAN measurements.

We got an error of -1.43dB.

Model Adjustment


81/93

Finally for the Global model, the error between CW based model and SCAN

measurements.

The error still around -0,83dB.


Global


82/93


Results, Conclusions andRecommendations

Results [1/2]


83/93

Reminding the targets and commitment KPIs:

All KPIs are inside the targeted objectives

Multi-Path

Direct-Path

*Calibration KPI with CW data SIRADEL Confidential - Copyright 2015 83

Results [1/2]

Route name MeanError(dB) StdError(dB) CorrelationFactor Mean error dispersion(dB)

Douala 0 5.59 0.90 0.97

Yaound 0 6.47 0.90 0.9

Global 0 6.1 0.90 0.92

Global mean error = 0dB

Global standard deviation error ~6dB

Global correlation >85%

Global Mean Error dispersion = 1.5dB

Mean error = 0dB

Global standard deviation error 70%

Global Mean Error dispersion < 3dB

Targets* Commitments*

Route name MeanError(dB) StdError(dB) CorrelationFactor Mean error dispersion(dB)

Douala 0 5.62 0.90 0.97

Yaound 0 6.49 0.90 0.89

Global 0 6.12 0.91 0.91

Results [2/2]


84/93

Extracted Parameters from CW calibrated model with the SCAN adjustment

included: Multi-Path

Direct Path


Results [2/2]

Yaound Douala Global

Parameters Value

a_near 2.24

a_far -0.42

b_near 22.78

b_far 23.76

alpha 0.50

weighting_r 0.50

weighting_d 0.55

weighting_h 1.0

Yaound Douala Global

Parameters Value

a_far -0.48

b_far 23.76

alpha 0.5

Parameters Value

a_far -1.44

b_far 23.81

alpha 0.5

Parameters Value

a_near 1.27

a_far -1.44

b_near 22.82

b_far 23.83

alpha 0.5

weighting_r 0.5

weighting_d 0.63

weighting_h 1

Parameters Value

a_near 1,9

a_far -0,78

b_near 22.80

b_far 23.79

alpha 0.5

weighting_r 0.5

weighting_d 0.55

weighting_h 1

Parameters Value

a_far -0.81

b_far 23.78

alpha 0.5

Conclusions


85/93


Co c us o s

CW routes presents outstanding performances both on Douala and Yaound confidence reinforced

Douala and Yaound Models presents very similar performances

Both DP and MP models reproduce the propagation environment with high

level of certitude Multi path trajectories doesn't play an important role on these two

environments

The Direct Path is the dominant one

Regarding the results, SIRADEL suggests DP Model utilisation. reduce computation time

prediction accuracy guaranteed

Recommendations


86/93


Use Atoll as an updated image of the live network.

After an optimisation Atoll project shall be updated (ex: C-PICH).

Review cells position.


87/93


And Now? Perspectives?

Perspectives [1/3]


88/93

On the Atoll project we found several sites placed either on wrong positions or

inside buildings.

A global site correction effort can bring important benefits to Orange networkplanning performance

being part of SIRADEL activity the site correction and detectionis often sought

by operators to improve their network planning accuracy.


p [ ]

Perspectives [2/3]


89/93

On Cameroun still exist other important villages where an HR map data can help

and ease Orange to improve his network quality, for example

Garoua 350 000 people

Bamenda 300 000 people

Maroua 270 000 people

Bafoussam 250 000 people

This amount of population represents an important source of revenue for Orange.

An accurate HR 3D map data will give the tools for an accurate optimisation and

site deployment in these towns. Allowing to reduce coverage holes, and

interference.

SIRADEL is available and has the resources to develop HR cartography for other

towns\areas in Cameroun.


p [ ]

Perspectives [3/3]


90/93

Since cities layout changes regularly, cartography should be updated to keep

accuracy.

SIRADEL suggests a cartography update on a 2/3 years basis.


p [ ]

Deliverables


91/93

Deliverables

6 Volcano Urban model calibrated in .vxf format

Douala, DP and MP

Yaound, DP and MP

Global, DP and MP

CW Atoll project with 14 CW Routes

SCAN Best Server Route for Douala and Yaound

Technical data sheets

Calibration Reports



92/93


Questions?


93/93

orc volcano urban umts calibration

Documents