AIRCOM – Cingular Model Tuning Guidance
December 2004
Agenda •Model Calibration Experience • Model Calibration Process • Model Calibration Analysis
December 2004
December 2004
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 •Globe (Phillipines) GSM900 – 3 models
December 2004
December 2004
CW Measurements and Model Calibration Process Propagation Model Requirements Identification
Drive Route Definition
Site Selection
CW Survey Campaign Data Post Processing Data Data Validation Validation
Calibration
NO
YES Report
Pass Model?
December 2004
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. • Provide cost efficient Nominal Plan
December 2004
Site Selection
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 Distribution for Site Selection Spread of site heights representative of networkHeight sites heights within modelled region 6
Allow measurements in all clutter types
Rooftop sites are preferred in a case test transmitter has to be mounted
Ease of access
5
No blocking objects in close vicinity
Frequency
4
3 Frequency 2
1
Nothing unusual, we are characterising the majority of the network not the minority 0
10
20
30
40
50
60
70
80
90
100
More
-1 Height of Site
December 2004
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
Avoid large blocking objects as high building or long roof
Long enough to ensure sufficient data is captured
Check map data validity
December 2004
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
RF Signals 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
Omni Antenna with Transmitter attached through feeder.
In Vehicle, Receive equipment attached to roof mounted antenna
December 2004
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 every 0.8λ.
• 50 samples should be averaged and give the local mean.
December 2004
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
L
L
December 2004
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 in drive vehicle speed.
• Sampling in time triggering is not a problem since Lee states just minimum number of samples.
• Averaging over 40 λ is problem to implement in time triggering since there is not constant number of samples over 40 λ caused by speed variation.
• Whenever possible choose distance triggering.
December 2004
Total driving route per model • In order for model to be realistic, statistically sufficient number of data need 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 is recommended to have more than 10 sites per model.
• As stated before, in a case of modelling small geographical area with 3 sites, tuning can be performed with 10000 data or 22km per site.
• The more data the model is more realistic
December 2004
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.
December 2004
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.
December 2004
Data filtering • Filter clutter types that have less than 500 bins. Clutter offsets or them • • • • • •
will 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 reflection over big water surface, or driving along route which is higher than antenna. Unusually weak signal level can be caused by driving behind blocking object. Okumura –Hata can’t model above situations, therefore these data must be filtered out. With careful route planning filtering can be avoided. Having more than one file per site makes filtering during post processing much easier
December 2004
Filtering example-Driving above Tx antenna
December 2004
Filtering example-Blocking object
December 2004
Displaying CW measurements in Asset
Data Types-CW MeasurementsCW Signal To set up thresholds double click on CW Signal and specify thresholds under Categories tab The same goes for other options inside CW Measurements
December 2004
December 2004
Okumura-Hata • Okumura-Hata is a worldwide the most popular model in mobile telecommunication
• It is semi-empirical model. • Based on Okumura measurements in Tokyo in 1968 mathematical model was published in 1980 by Hata.
• Limitations:
Up to 2GHz
No less than 1km
Transmitter antenna always above mobile station antenna
December 2004
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 height Asset has 4 algorithms for calculating diffraction
December 2004
Asset improvements • K1 near and k2 near are designed to overcome Okumura-Hata limitation for close distances.
• Through Clutter Loss – takes into the account clutter profile along distance d from mobile station to base station.
• Advantages in improved accuracy/reduced standard deviation error and more realistic calculated predictions.
December 2004
Through Clutter Model Definition • Each clutter category is given Through Clutter Loss (dB/km) on the path between transmitter and receiver.
• Through clutter losses are linearly weighted. The clutter nearest the mobile station has the highest effect.
December 2004
Overview 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 and determine which one gives the lowest standard deviation Take note of second and third best
.
December 2004
CW Window • •
3g/Asset-Tools-Model Tuning
•
Highlight Site ID and click Remove button to remove particular file
Click Add to add measurements file from its destination, they mast have extension .hd
December 2004
Model setting • •
Tools-Model Tuning-Options
•
Select the model as a start tuning model. It is recommended to use default model
Select the resolution of mapping data
December 2004
Filter seting • • • •
Tools-Model Tuning-Options-Filter
• •
If you tune k7 click just NLOS
Set up distance filtering Set up signal level filtering Filter out clutter types with insufficient data by highlighting them Click antenna button if directional antennas were used
December 2004
Auto Tune • • •
Tools-Model Tuning-Auto Tune
• • •
Click Auto Tune under Tools tab
•
Through clutter offsets and clutter offsets are under Clutter tab
Set up deltas Click fix box next to the k factor you don’t want to tune Wait for results You can apply new parameters by clicking apply new parameters
December 2004
K parameters • K3 and K4 are not altered. This is because they relate to mobile height which in a typical cellular system is constant making these coefficients redundant.
• K7 is the diffraction parameter. It can be determined by tuning just NLOS data.
• All K parameters must keep the same polarity as in the original Okumura Hata model
K1, K2, K7 >0
K3, K5, K6 <0
• Above step can be easily fulfil by determining the delta range under Auto tune window
December 2004
Default K parameters
December 2004
k1,k2 near calibration • If model is not good close to the site, for example up to 700m, auto tune the model from 700m to 10k. Apply found k parameters.
• Tune model again with k5,k6 and k7 locked and filter out distances above 700m.
• Result will be k1near and k2 near. • If standard deviation is still bad try with other distances until you find the best fit.
December 2004
Clutter offset • Some through clutter offsets and clutter offsets need to be estimated due 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 the highest.
December 2004
Adjusting ME • Mean error is usually altered after estimation of clutter offsets. • ME can be easily bring back to 0 by changing k1 • If mean error is Δ change k1 to k1+ Δ
December 2004
December 2004
Model analyses • Make statistical analyses for ME and SD for different distance ranges. • In the range of interest, typically 1km to 4km, following requirements should be fulfilled
-1 < ME < 1
SD < 8
• If ME or SD is outside the above specified values, try with changing the dual slope distance or take the second best model from the initial tuning.
December 2004
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. • Distance used for tuning from 125m to 10km.
December 2004
Data Analysis for Coastal Urban 15m Distribution of bins per signal level 30000
Distribution of bins per clutter type
Number of bins
25000 20000
30000
15000
26856
10000
25000
5000
Number of bins
0
) ) ) ) ) ) 0) 0) 0) 0) 90 80 70 50 40 30 -5000 ( -12 -(6 -11 -10 -( -(-(-(-(-(00 00-( 0-( -7 0 -9 0 -8 0 -6 0 -5 0 -4 0 1 3 2 1 1 1 1 Signal level (dBm)
20000 14421
15000 10574
5062
Distribution of bins per distance
5000
3010
35000
0
1062
189
28
0
3
0 17791
15000 8700
10000 5000
373
1030
0-0.125
0.125-0.25
2899
891
0
8-16
>16
0 0.25-0.5
0.5-1
1-2
2-4
4-8
wa te re sid r en m tia ea l n ur de ba ns n e ur ba n bu ild in g vil la ge s in du op st ria en l in ur ba n fo re st de ns p a e rk ur s ba n hi gh sw am p
19351
20000
se a
25000
op en
Number of bins
4
29598
30000
9995
9056
10000
Clutter type
Distance (km)
December 2004
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
5.2
December 2004
Statistical Breakdown for ME and SD Mean error vs distance
Standard deviation distribution
1.5
9 8
1
Mean error
0.5
0 0.125 -0.5
0 -0.25
0.5 0.25-
0.5-1
1-2
2-4
4-8
8-16
Standard deviation
7 6 5 4 3 2
-1
1 -1.5
0
-0.2 0.125
-2 Distance (km)
50
0.5-1
2-4
8-16
Distance (km )
December 2004
Validation of Tuned Model-Site 1 Apoview site
Calibration whole range
No. of Bins
Mean Error
Standard Deviation Actual
10668
-1
6.1
125~250
53
4.3
5.6
250~500
368
0.4
7.5
500~1km
1153
-2.7
7.3
1km~2km
2324
-1.5
6.3
2km~4km
4383
0.4
5.9
4km~8km
2343
-2.4
5.1
8km~16km
44
-2.4
4.1
December 2004
Coverage plot – Site 1
December 2004
Validation of Tuned Model-Site 2 No. of Bins
Mean Error
6354
0.1
6.4
125~250
95
11.6
5.2
250~500
42
2.7
5.7
500~1km
252
-1.8
7.7
1km~2km
1620
-0.9
6.3
2km~4km
3228
1
6.4
4km~8km
1041
-1.6
4.8
8km~16km
76
-2.9
3.8
Banawa site Calibration whole range
Standard Deviation Actual
December 2004
Coverage plot – Site 2
December 2004
