AIRCOM – Cingular
Model Tuning Guidance
Agenda
•
Model Calibration Experience
•
Model Calibration Process
CW and Model Tuning References
• 3GIS (Sweden) – UMTS, 6 models
• Belgacom (Belgium) – GSM 900, 6 models
• Swisscom (Switzerland) – GSM 900/1800 and UMTS, 9 models
• Inventis (Switzerland) – GSM R, 3 models
• Vodafone (Malta) – GSM 900, 2 models
• Globul (Bulgaria) – GSM 900, GSM1800
• Oniway (Portugal) – UMTS, 4 models
• Inquam (Portugal) – CDMA2000
• Blu (Italy) – GSM 1800
• Nortel (UK) – GSM 1800
• Ericsson (UK) – GSM 1800
• Dolphin (Belgium, Uk) – Tetra models
• KPN Base (Belgium) – GSM 900/1800, 8 Models, 4 for each
•TMN (Portugal) – GSM900, 1 model
•Mascom (Botswana) – GSM 900
• CHT Taiwan
• Brazil – GSM 900/1800, 5 models
• India BPL – GSM 900, 1 Model
•AWS (USA) – GSM 1900, 2 models
•TCI (Iran) – GSM 900, 5 models
•ESAT Digifone (Ireland) – UMTS 3 models
•Safaricom (Kenya) GSM 900 – 2 models
•Lucent (Riyadh) GSM 900 – 1 model
• Claro (Brazil) GSM1800 – 3 models
CW Measurements and Model Calibration
Process
Data Validation NO YESSite Selection Drive Route Definition CW Survey Campaign Data Post Processing Calibration
Report Model?Pass
Propagation Model Requirements Identification Data Validation
Aim of Model Calibration
•
Characterise the topology with network limits – identification of operating range for each model.• Minimise Standard Deviation Error.
• Provide zero mean error
• Determine model parameters in accordance to realistic propagation effects existing within proposed regions.
• Make sure calibrated model corresponds well with the collected data – data is essential.
Site Selection
Height Distribution for Site Selection
-1 0 1 2 3 4 5 6 10 20 30 40 50 60 70 80 90 100 More Height of Site F re q u e n c y Frequency
More or 10 sites per model. Less number of sites can be considered if
modelled geographical area is fairly small.
Within geographic region of model
Spread of site heights representative of network sites heights within
modelled region
Allow measurements in all clutter types
Rooftop sites are preferred in a case test transmitter has to be mounted
Ease of access
No blocking objects in close vicinity
Nothing unusual, we are characterising the majority of the network not the
CW Drive Route Definition
• Distance Must account for expected coverage propagation Must account for expected interference propagation
• Clutter
Sufficient measurement in all local clutter types ( >1000 )
• Roads
Avoid street canyons, tunnels, elevated roads, cuttings etc Mix of radial and tangential roads
• Miscellaneous
Do not plan a map along the roads with ground height above
the transmitter antenna. Okumura- Hata model can’t model this.
Good balance between measurements taken in LOS and NLOS
situations
Do not plan a route across a big water surface, if site is on the
one side of the lake, do not drive other lake side
Data in regions of terrain slope variation
Omni Antenna with Transmitter attached through feeder. In Vehicle, Receive equipment attached to roof mounted antenna RF Signals
CW Measurements
• Spectrum clearance During CW survey allocated test frequency shouldn’t be use for other purposes 10-15KHz bandwidth monitoring
Check restrictions on test frequency TX EIRP
• Equipment configuration
Accurate Radiated Power setting, EiRP should be greater than 40dBm Raw/Averaged data
Use Omni antenna with minimum vertical beamwidth of 12 degrees
Directional antenna can be used but in postproccessing everything beyond 3dBm should be dismissed
• Driving
Do not drive out of RX noise floor
Avoid street canyons, tunnels, elevated roads, cuttings etc Distance/Time triggering
Sampling - Lee Criteria
•
Lee Criteria – In order to eliminate fast fading from measurements,minimum 36 samples should be taken over 40λ. A local mean should
be found for the chosen number of samples.
•
Common practice is to take 50 samples which gives one sample every0.8λ.
Slow fading vs Fast fading
• Fast fading is fading due to multipath effect.
• Fast fading is characterized by Rayleigh probability distribution therefore can’t be
modelled by log normal distribution.
• Fast fading is superimposed onto signal envelope (slow fading) which we try to model.
• Slow fading is fading due to terrain and clutter.
• Slow fading follows log normal distribution.
• Okumura-Hata is log normal distribution
Distance triggering vs time triggering
•
Distance triggering allows us to easily apply Lee criterion.•
Time triggering is very difficult to follow Lee criterion due to change indrive vehicle speed.
•
Sampling in time triggering is not a problem since Lee states justminimum number of samples.
•
Averaging over 40 λ is problem to implement in time triggering sincethere is not constant number of samples over 40 λ caused by speed
variation.
Total driving route per model
•
In order for model to be realistic, statistically sufficient number of dataneed to be collected.
•
Aircom practise is to have at least 30000 data.•
30000 data gives total driven distance of 30000x40λ=198km or
20km per site for 1800MHz range.
•
If this distance is not achievable due to limitation in drivable roads it isrecommended to have more than 10 sites per model.
•
As stated before, in a case of modelling small geographical area with3 sites, tuning can be performed with 10000 data or 22km per site.
Data Post processing
• Depends on customer requirements:
Averaged Measurements – post processing involves simple conversion into Signia format supported by
Enterprise
Signia data file ( .dat ) contains longitude, latitude (decimal degrees) and received level (dBm) Every data file must have header file with identical name but with extension .hd.
Header file must have antenna type (identical name to one in Asset3g), Tx power, Tx antenna height,
coordinates.
It is common practice to include all gains and losses under Tx power value and leave other fields relevant to
gain/losses in the header blank. Therefore in a Tx field usually is put: Tx – Ct +Atg –Arg+Crl where
Tx-Tx power(dBm),
Ct-cable loss between transmitter and antenna (dB), Atg-transmitting antenna gain (dBi)
Arg-receiving antenna gain (dBi)
Crl-cable loss between receiver and receiving antenna (dB)
• It is important to get the projection system correctly so collected samples are lined up with the vectors in map data. If vectors are not aligned with measurements, during post process this should be adjusted.
CW Data Validation
• Compare the site data (photographs, surrounding clutter and terrain profile) to the Clutter and DTM layer of the map data provided.
• Check the driven routes against vectors within the map data.
• Filter out any invalid data that may cause anomalies in the calibration process
• Make sure that details relating to a site (EIRP, Location, Height, Antenna file) correspond to reports from CW Survey.
• Use Asset utilities to get visual representation of the received signal vs distance.
Data filtering
•
Filter clutter types that have less than 500 bins. Clutter offsets or themwill be estimated later in the model tuning process.
•
Filter out any file which shows extreme in signal level.•
Unusually high signal level at far distance can be caused by reflectionover big water surface, or driving along route which is higher than antenna.
•
Unusually weak signal level can be caused by driving behind blockingobject.
•
Okumura –Hata can’t model above situations, therefore these datamust be filtered out.
•
With careful route planning filtering can be avoided.•
Having more than one file per site makes filtering during postDisplaying CW measurements in Asset
Data Types-CW
Measurements-CW Signal
To set up thresholds double click
on CW Signal and specify
thresholds under Categories tab
The same goes for other options
Okumura-Hata
•
Okumura-Hata is a worldwide the most popular model in mobiletelecommunication
•
It is semi-empirical model.•
Based on Okumura measurements in Tokyo in 1968 mathematicalmodel was published in 1980 by Hata.
•
Limitations: Up to 2GHz
No less than 1km
Okumura-Hata in Asset
•
Asset uses slightly modified Okumura-Hata: Ploss =K1 + K2*log(d) + K3*Hms + K4*log(Hms) + K5*log(Heff) +
K6*log(Heff)*log(d) + K7*Ldiff + Lclutter
d is distance in km between Tx antenna and mobile station Hms is mobile station height
Heff is effective antenna height in metres Ldiff is a loss due to diffraction
Lclutter is a clutter loss
•
Asset has 4 algorithms for calculating effective antenna heightAsset improvements
•
K1 near and k2 near are designed to overcome Okumura-Hatalimitation for close distances.
•
Through Clutter Loss – takes into the account clutter profile alongdistance d from mobile station to base station.
•
Advantages in improved accuracy/reduced standard deviation errorThrough Clutter Model Definition
•
Each clutter category is given Through Clutter Loss (dB/km) on thepath between transmitter and receiver.
•
Through clutter losses are linearly weighted. The clutter nearest theOverview of Model Calibration
•
There must be project set up (map data, antennas, sites, propagation model)in order to start tuning
•
Load CW data•
Make appropriate filtering, usually: -110dBm to -40dBm 125m to 10000
•
Start with the default values for k parameters•
Do Auto Tune•
Try all combination of effective antenna height and diffraction algorithms anddetermine which one gives the lowest standard deviation
•
Take note of second and third bestCW Window
•
3g/Asset-Tools-Model Tuning•
Click Add to add measurements filefrom its destination, they mast have extension .hd
•
Highlight Site ID and click RemoveModel setting
•
Tools-Model Tuning-Options•
Select the resolution of mappingdata
•
Select the model as a start tuningmodel. It is recommended to use default model
Filter seting
•
Tools-Model Tuning-Options-Filter•
Set up distance filtering•
Set up signal level filtering•
Filter out clutter types withinsufficient data by highlighting them
•
If you tune k7 click just NLOS•
Click antenna button if directionalAuto Tune
•
Tools-Model Tuning-Auto Tune•
Set up deltas•
Click fix box next to the k factor youdon’t want to tune
•
Click Auto Tune under Tools tab•
Wait for results•
You can apply new parameters byclicking apply new parameters
•
Through clutter offsets and clutterK parameters
•
K3 and K4 are not altered. This is because they relate to mobileheight which in a typical cellular system is constant making these coefficients redundant.
•
K7 is the diffraction parameter. It can be determined by tuning justNLOS data.
•
All K parameters must keep the same polarity as in the originalOkumura Hata model
K1, K2, K7 >0
K3, K5, K6 <0
•
Above step can be easily fulfil by determining the delta range underk1,k2 near calibration
•
If model is not good close to the site, for example up to 700m, autotune the model from 700m to 10k. Apply found k parameters.
•
Tune model again with k5,k6 and k7 locked and filter out distancesabove 700m.
•
Result will be k1near and k2 near.•
If standard deviation is still bad try with other distances until you findClutter offset
•
Some through clutter offsets and clutter offsets need to be estimateddue to insufficient data.
•
Estimation is done relative to the clutter offsets with sufficient data.•
Clutter offsets must be realistic relative to each other.•
Water will have the smallest offset while building and forest theAdjusting ME
•
Mean error is usually altered after estimation of clutter offsets.•
ME can be easily bring back to 0 by changing k1Model analyses
•
Make statistical analyses for ME and SD for different distance ranges.•
In the range of interest, typically 1km to 4km, following requirementsshould be fulfilled
-1 < ME < 1
SD < 8
•
If ME or SD is outside the above specified values, try with changingthe dual slope distance or take the second best model from the initial tuning.
Example-Coastal Urban 900MHz
15m resolution map
•
Area considered: densely populated coastal cities.•
Used frequency: 935.2MHz•
Total of 10 sites were included in tuning process with 80260 points.•
Signal strenght threshold set to –40 to -110 dBm.Data Analysis for Coastal Urban 15m
Distribution of bins per signal level
-5000 0 5000 10000 15000 20000 25000 30000 -130-(-120) -120-(-110) -110-(-100) -100-( -90) -90 -(-80) -80 -(-70) -70-(6 0) -60 -(-50) -50 -(-40) -40 -(-30) Signal level (dBm) N u m b e r o f b in s
Distribution of bins per distance
373 1030 2899 19351 29598 17791 891 0 8700 0 5000 10000 15000 20000 25000 30000 35000 0-0.125 0.125-0.25 0.25-0.5 0.5-1 1-2 2-4 4-8 8-16 >16 N u m b e r o f b in s
Distribution of bins per clutter type
10574 4 0 14421 26856 9056 3010 1062 5062 9995 189 28 0 3 0 5000 10000 15000 20000 25000 30000 open sea wat er resi den tial mean urban dens e ur ban buildi ng villag es indu strial open in urban for est parks dens e ur ban high swa mp Clutter type N u m b e r o f b in s
Statistical Breakdown for Coastal Urban 15m
No. of Bins Mean Error Standard Deviation Actual Calibration whole range 80260 0 6.8 125~250 1030 -0.5 8.1 250~500 2899 -1.1 8 500~1km 8700 -1.4 7.7 1km~2km 19351 -0.1 7.4 2km~4km 29598 0.9 6.6 4km~8km 17791 -0.4 5.4 8km~16km 891 -1.6 5.2Statistical Breakdown for ME and SD
Standard deviation distribution0 1 2 3 4 5 6 7 8 9 0.125-0.250 0.5-1 2-4 8-16 Distance (km ) S tan d ar d d evi at io n
Mean error vs distance
-2 -1.5 -1 -0.5 0 0.5 1 1.5 0.125-0.25 0 0.25-0.5 0.5-1 1-2 2-4 4-8 8-16 Distance (km) M ean er ro r