Sie sind auf Seite 1von 18

RESTRICTED

Asset Propagation Model Tuning

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 1 of 188
RESTRICTED

Table Of Contents
Page
1. Introduction...............................................................................................................................3
1.1 Purpose ................................................................................................................................3
1.2 Scope....................................................................................................................................3
1.3 References............................................................................................................................3
1.4 Definitions ...........................................................................................................................3

2. General Equation Background................................................................................................4

3. Database Requirements............................................................................................................5

4. Prediction Model Initial Conditions........................................................................................6


4.1 General.................................................................................................................................6
4.2 Creation of Model in Asset..................................................................................................7
4.3 Creation of Reference Site in Asset.....................................................................................7

5. Drive Test Data Import Into Asset ..........................................................................................7


5.1 File Import ...........................................................................................................................7
5.2 Data Verification .................................................................................................................8
5.3 Reference Site Model Tuning..............................................................................................8

6. Model Tuning ............................................................................................................................9


6.1 Numerical Analysis Process ................................................................................................9
6.2 Initial Tuning of K1 ...........................................................................................................10
6.3 Initial Tuning of K2 ...........................................................................................................11
6.4 Multi-site Drive Test Analysis. .........................................................................................11
6.5 Multi-site Drive Model Tuning. ........................................................................................12
6.6 Tuning of K7......................................................................................................................13
6.7 Tuning of Clutter Coefficients...........................................................................................13
6.7.1 Clutter Offsets........................................................................................................13
6.7.2 Clutter Heights and Separation ..............................................................................15

7. Document Control...................................................................................................................18
7.1 Authorisation .....................................................................................................................18
7.2 Amendment List ................................................................................................................18

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 2 of 188
RESTRICTED

1. Introduction

1.1 Purpose
The purpose of this document is to specify the recommended method of tuning the
Aircom – Asset, radio propagation prediction tool.

1.2 Scope
This document is intended for technical staff involved in the design of cellular mobile
phone networks.

1.3 References
Singtel Drive Test Methodology for Propagation Model Tuning
Optus
Aircom Asset User Reference Guide Version 4.1

1.4 Definitions
CBD Central Business District
CW Continuous Wave
ERP Effective Radiated Power
EIRP Effective Isotropic Radiated Power
RMS Root Mean Squared
DB Decibel

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 3 of 188
RESTRICTED

2. General Equation Background


The research work performed in Japan in the 1960’s by Okumura and Hata, culminated in the
formulation of the General Equation for radio propagation path loss in a mobile phone
environment. This equation has limitations, but contains terms which take into account the
path loss influence of the following effects,
1) Frequency.
2) Distance.
3) Antenna height.
4) Antenna pattern.
5) Ground reflection.
6) Clutter.
7) Diffraction into non-line of site regions.
The general equation is the recommended prediction method for modelling of mobile phone
base station coverage in Asset. This model provides a high accuracy for rural, urban and high-
density urban areas where the antenna height is greater than the surrounding structures.
Within an area like a CBD, the limitations become apparent.
For complex areas like a CBD, details of the individual buildings and structures need to be
present in the database. Also, the effect of reflection must be taken into account. A better
prediction model for this environment would use “ray tracing” techniques to predict the
“multi path” nature of the coverage patterns. These prediction techniques are not presented in
this document.
The general equation is essentially a linear prediction method and determines the radio signal
strength along a direct path away from the transmitting source. No account is included for
reflection from structures which, are alongside the direct path. The equation of path loss is
specified below.
Loss = K1+ K2Log(d) + K3(Hms) + K4Log(Hms) + K5Log(Heff) + K6Log(Heff)Log(d) + K7Diffn + C_Loss.

Where d = distance between base station and the mobile specified in Kilometres.
Hms = Height of the mobile above ground in metres. This is usually set to the value of 1.5m
Heff = The effective base station antenna height in metres.
Diffn = The diffraction loss. Alternative methods are available. Epstein–Peterson is
recommended.
K1 = Constant offset.
K2 = Fade slope constant.
K3 = Mobile antenna height factor. This parameter is normally set to zero.
K4 = Okumura – Hata multiplying factor for mobile height. Normally set to zero.
K5 = Effective antenna height gain. Options are available but “Slope Method” is preferred.
K6 = Log of the effective antenna height gain.
K7 = Diffraction loss

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 4 of 188
RESTRICTED

C_Loss = Loss due to clutter. Eg, urban, CBD, trees. Individual class coefficients are tuned.
In addition to the loss coefficient for clutter, provision is included to account for individual
clutter heights and separation distance between the clutter and the mobile. This provision
produces a more accurate result but dramatically increases the calculation time.
The general equation prediction method has less accuracy when the antenna height becomes
comparable with the height of the nearby clutter. In these situations, the model will over-
predict the potential coverage. It may be necessary to provide a separate model, specifically
tuned to compensate for the increased fading of this type of base station. Similarly, the
general equation will tend to under-predict the potential range of a base station located on a
high hilltop or mountain, which is typically 400-500m above the valley. A separate prediction
model for these situations may also be required.

3. Database Requirements
In order to perform accurate predictions, an adequate database of the topology (terrain) and
morphology (clutter) is required. Higher resolution will produce a more accurate prediction.
The trade-off with the higher database resolution will be a slower calculation time. For rural
areas a 100m-pixel resolution is usually sufficient. Within a metropolitan area, 50m-pixel
resolution is preferred. Within a CBD a 5m resolution is preferred but a different prediction
model is required, and the Okumara-Hata general equation is not suitable.
The clutter database should classify the following regions,
1) Low Density Urban (Suburbs, detached single storey houses less than 6m high)
2) Medium Density Urban (Residential multistorey buildings less than 15m high)
3) High Density Urban (Residential multistorey buildings various heights up to 25m)
4) High Rise Industrial (Large factories up to 40m height, metal roof, brick structures).
5) CBD. (High rise buildings)
6) Trees. (A few categories here would be better. Such as “dense forest”, “open forest”)
7) Open areas.
8) Grasslands. (Crops, low scrub, pastures etc with less than 1m height vegetation).
9) Roads. (Wide major roads only).
A complete database of the antenna patterns is required. For the model tuning process only
the antenna pattern of the test antennas are required. However, once the prediction model is
tuned, patterns for each antenna used within the base station network must be included in the
prediction for each site.

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 5 of 188
RESTRICTED

4. Prediction Model Initial Conditions.

4.1 General
An extensive history of testing and experimentation has already taken place with optimum
tuning of the Okumura-Hata general equation. For this reason it is possible to specify several
initial conditions of the model. Fine-tuning may then be confined to the primary coefficients
and the specific differences of the localised clutter effects on the signal fading.

The following initial conditions can be specified and an explanation for these choices.

1) Asset Model Type = Standard Macrocell. Selects the general equation model.
2) Frequency = Selected as required. Example 900MHz, 1800MHz.
3) Mobile RX height = 1.5m
4) Earth Radius = 8491.2 Km
5) K1, K2, K7 = These parameters will be tuned.
6) K3 = 0
7) K4 = 0
8) K1 near, K2 near, d near = 0
9) K5 = -13.82 This value has been derived by Okumura-Hata. Standard value is suitable.
10) K6 = -6.55 This value has been derived by Okumura-Hata. Standard value is suitable.
11) Effective Height = Slope. This consistently produces higher accuracy than other methods.
It is particularly effective at predicting the effect of rising terrain when the signal strength
generally increases due to the path appearing more like free space. It provides up to 15dB
better accuracy than other methods. Other methods available are “Absolute”, “Average”,
and “Relative Height”. For more information on these options refer to the Aircom Asset
User Reference Guide.
12) Slope distance = 10 times the resolution. Ie. 50m resolution, distance = 500m
13) Heff min and Max. These parameters will be tuned.
14) Diffraction Method = Epstein-Peterson. This method has consistently provided more
accuracy. Other methods available are “Bullington”, “Deygout”, and “Japanese Atlas”.
For more information on these options refer to the Aircom Asset User Reference Guide.
15) Merge Knife-edges = 2 times the database resolution. I.e.50m resolution, Merge =100.
16) Clutter Offsets. These parameters will be tuned. Initially set all offsets to 0.
17) Clutter heights = Enabled. This is optional and depends on acceptance of longer
calculation time versus the desire to have the slightly higher accuracy. Initially set all
heights to 0.
18) Clutter separation = Enabled. This is optional and depends on acceptance of longer
calculation time versus the desire to have the slightly higher accuracy. The separation
distance is also tuned. Initially set all separations to 0.
19) Use Mobile Heights = Disabled
20) Mobile Height = 0

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 6 of 188
RESTRICTED

4.2 Creation of Model in Asset.


To create the new model, in Asset select under the “Tools” tab, the “Propagation Model
Editor” option. Select “Add” and enter a name for the new model. Note that write privileges
are required to perform this task. Enter the new model parameters using the recommended
defaults and assign the default numbers a described below for the un-tuned coefficients.
The primary coefficients of the model are K1 and K2. The K1 coefficient sets the overall
“size” of the coverage whereas K2 predominantly effects the level of coverage close to the
site. K6 effects how the prediction behaves in non-line-of-sight regions. These parameters are
tuned primarily by the influence of the terrain. The influence of the various clutter classes is
effectively fine-tuning of localised attenuation. Initial starting values are required for K1, K2
and K7 and the effective height. Default values of K1 = 140, K2 = 40 and K7 = 0.6 will be
suitable.
Also, use a minimum height of 10m and a maximum of 100m for the slope method of
effective height. Ensure all clutter offsets, heights and separation distances are set to 0.
Disable clutter heights.

4.3 Creation of Reference Site in Asset


From the range of test site locations, select a site, which would be considered “typical”
suburban. This site should be on relatively flat terrain and surrounded predominantly by low-
rise dwellings and buildings for several kilometres radius. The site antenna height should be
25-30m above ground or at least 20m higher than the surrounding clutter. This test case or
“reference site” will be used for the visual analysis of the prediction model performance when
compared to the drive test data.
Create the reference site in Asset and ensure the location, antenna type, antenna height,
transmit power level and prediction model are all correctly applied to the site configuration.
Other details normally entered in Asset to fully describe a base station are not required at this
stage. The details need only be sufficient for modelling of the reference test transmitter site.

5. Drive Test Data Import Into Asset


The drive test data should be stored on a file server or hard disc drive, under a directory,
which can be accessed by the Asset application. The different categories of drive test (ie
macrocell, microcell) need to be in separate directories and clearly labelled.

5.1 File Import


1) In the Asset application, select “Tools” and “CW Measurements”.
2) Select “Add” and choose the type of measurement file. The formatted files should contain
a header file and a data file. The header file contains the test transmitter set-up details and
is linked to the data file. The data file will contain the measurement point locations and a
signal strength reading.
3) Using the directory tree, search for the file locations. Select and “Add” the file. At this
stage select only one drive test file at a time and perform a data verification process. See
section 5.2 of this report.

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 7 of 188
RESTRICTED

5.2 Data Verification


It is assumed that Asset has available, a database of the local city streets or roads, as well as
the terrain and clutter information.
1) In Asset under the 2D-display view, load a roads or streets vector from the standard
vectors tab.
2) Select the “CW Measurements” tab and double click on “CW Signal”.
3) Select the “Categories” tab on the top of the window.
4) Add and create a range of signal strength thresholds using unique colours for each. A
suitable range of threshold and colours are indicated below.
-50dBm Bright Yellow
-60dBm Orange
-70dBm Blue
-80dBm Red
-90dBm Green
-110dBm Pale Yellow
These same colours and thresholds must also be applied to coverage predictions in
Asset, so a direct visual comparison can be made.
5) Choose a solid plot character and a circle shape is suitable with size ‘50’ for the plot
character.
Plot each drive test measurement one at a time and verify the accuracy of the navigation
compared to the vector data of the street database. Check the expected signal fading. If the
plot has numerous errors of navigation, missing data, unusual and unexpected signal levels
(too high, too low) the test may be unsuitable. Verify each drive test in this way before
performing the model tuning. Discard any drive test, which may be considered invalid.

5.3 Reference Site Model Tuning


1) Once each drive test has been checked using the procedure in section 5.2 and found to be
suitable, select the “Reference Site” and remove all other files from the CW Measurement
Analysis.
2) Confirm the file configuration using the “Info” tab under the CW Measurement Analysis.
Ensure that the essential details of Transmitter power, Transmitter test frequency,
Antenna height, Antenna type and Site location are all correctly entered. Note that
transmitter power must be specified in terms of EIRP and compensated for both the test
transmitter and test receiver antennas. If the recorded power was specified as ERP of the
test transmitter only, the power must be adjusted by +4.3dB. This is because Asset
predictions are based on isotropic transmitter and receiver antennas. For practical reasons,
the test transmitter power is usually specified in terms of ERP. Compensation is also
required however for the receive antenna which will provide signal strength
measurements in terms of normalised power relative to a dipole antenna. The
compensation between dipole and isotropic antennas is 2.15dB with a dipole having more
gain relative to the isotropic antenna. Twice this factor equates to 4.3dB.
3) Select the “Options” tab and under the “Model” tab, select the required resolution (this
should match the terrain/clutter database resolution)
Asset Propagation Model Tuning
Version: Draft 1
14/7/2003 Page 8 of 188
RESTRICTED

4) Select the model to be tuned. The new model should have been entered following section
4.2 of this report.
5) Under the “Filter” tab, exclude the “Null” and “Water” clutter by highlighting these items.
6) Adjust the calculation radius. Minimum of 0m and maximum of 100,000 will usually
allow all test data to be valid. The range of valid data may be adjusted if data at either
extreme of distance is found to be unsuitable.
7) Set the filter for upper and lower limits of the test receiver. These limits will depend on
the test receiver performance. Generally signal levels above –40dBm enter into the
saturation limit of a receiver and levels below –110dBm are affected by noise. These may
be suitable limits. Refer to the drive test receiver equipment specification or calibration
report for more details.
8) Select LOS (Line of Sight) and NLOS (Non Line of Sight) data visibility. This setting will
be changed later to tune different aspects of the prediction model.
9) Display the drive test on the 2D view in Asset.
10) Display the prediction of the Reference test site as an overlay on the drive test plot.
11) Examine the overall differences between the measurements and the prediction. This view
should be evaluated after each step change in the model tuning process to observe
progressively, the effects of the tuning.
12) Note how the measured signal fades and how the radii of the signal threshold step changes
relate to the corresponding step changes on the prediction. If the measurements at the
outer fringe of the plot are typically much higher than the prediction, then K1 may be too
high. If the measured threshold steps close to the site indicate a lower rate of change with
increasing distance than the prediction, then K2 of the model may be too high. In each
case the converse will be also be true. Many of the individual difference errors between
the measured and predicted at this stage will be related to the fact that the influence of
clutter has been disabled. A prediction with greater detail will be produced when this
feature is enabled and tuned.

6. Model Tuning
The model tuning will involve a combination of graphical analysis comparing the drive test
data with prediction plots as described in section 5.3, and numerical analysis using the
statistical study available in Asset. Relying only on the numerical analysis is not
recommended. The numerical analysis is a guide, which can offer insights into the potential
accuracy of a prediction model. It can also assist in finding an optimum value for a specific
coefficient by tuning for a RMS minimum around a particular coefficient value. However,
due to errors in the registration between the drive test measurement samples and the clutter
database, an accurate numerical analysis is not possible. The following steps 1 to 9 provide
the method of preparing the measurement data for the numerical analysis.

6.1 Numerical Analysis Process


1) With the drive test file for the reference site entered as per section 5.1 of this report, select
the “Analyse” tab. A window will provide a further set of options. Use the defaults as
selected but de-select the “Show Individual Bin” information. This option provides
excessive and non-essential information in the report.

2) The analysis process will produce an excel spreadsheet. A sample of this output is
presented in table 1.

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 9 of 188
RESTRICTED
File Site ID Site Name Num. Bins Mean Error RMS Error Std.Dev. Error Corr. Coeff.
H:\SJB\MSI Data\M90\30m90_9.hd 30m90_9 Survey 6528 1.5 7.8 7.6 0.8624

Model Num. Bin Mean Error RMS Error Std.Dev. Error Corr. Coeff.
Training G9 Model 6528 1.5 7.8 7.6 0.8624

Clutter Num. Bins Mean Error RMS Error Std.Dev. Error Corr. Coeff.
Open 129 3.3 6.8 6.0 0.9574 K1 = 130
Roads 2971 3.8 8.2 7.3 0.8546 K2 = 40
Vegetation 14 6.0 8.5 6.3 0.9451 K7=0.6
Low Density Urban 3124 -0.9 7.4 7.4 0.8381 Clutter Disabled
Medium Density Urban 23 2.8 4.0 3.0 0.3658
High Density Urban 124 2.5 6.5 6.0 0.6701
Light Industrial/Commercial 143 3.5 7.4 6.5 0.8210

Table 1. Model Tuning Analysis Output One Site.

3) This example has 1 drive test file, the reference site. In this example a 30m base station in
a suburban area. Copy this spreadsheet into a new model-tuning directory and give it a
label, which will provide an indication as to the progress of the tuning steps, ie.
K1RefTuneStep1. The settings of the parameters being tuned should also be entered on
the spreadsheet for reference. This table provides information about the mean and RMS
errors of the propagation model. The values are in decibel (dB). A target RMS error of
less than 7dB will provide a prediction model that has acceptable accuracy.

6.2 Initial Tuning of K1


1) In the CW Measurements “Options”, de-select the Non-LOS visibility data filter.
2) Run the analysis tool and keep a record of the spreadsheet output. Note the RMS error for
the particular K1 and K2 value used.
3) Change the value of K1 in the prediction model by a small increment (less than 5) with
either an increase or decrease in value depending on the initial observations from step 12
of section 5.3. For example, if the model appears to be over-predicting, then increase the
value of K1.
4) If K1 was increased in value but the RMS error also increased, then K1 must be adjusted
in the opposite direction, to a lower value.
5) This process is iterative until the lowest RMS error can be found. As the minimum RMS
error is approached, adjust K1 in steps of 1. If a range of values provide the same RMS
error, select a mid-point.
6) Evaluate the result on the Asset 2D-view presentation. Ensure the prediction is re-done
for the new values.

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 10 of 188
RESTRICTED

6.3 Initial Tuning of K2


1) Using the final tuned value of K1 established in step 6.2 above, apply the same process
for the adjustment of K2. If a range of values provide the same RMS error, select a mid-
point.
2) The analysis result will indicate the mean error. This value may be subtracted from K1 to
achieve the final optimised values for K1 and K2 at this stage of the process.
3) Evaluate the result on the Asset 2D-view presentation. Ensure the prediction is re-done for
the new values. At this stage the most accurate values for K1 and K2 have been found for
the “typical” macro-cell base station, using the Reference site. It is now necessary to
introduce the other drive test measurements to further fine tune K1 and K2 and begin
tuning the other parameters of the general equation.

6.4 Multi-site Drive Test Analysis.


1) As per section 5.2 “File Import”, in the CW Measurements Analysis application in Asset,
Add all the other drive test files which are suitable for model tuning. Do not include
microcells or drive test data that was found to be unsuitable.
2) For each drive test file, confirm the file configuration using the “Info” tab under the CW
Measurement Analysis. Ensure that the essential details of Transmitter power, Transmitter
test frequency, Antenna height, Antenna type and Site location are all correctly entered.
Note that transmitter power must be specified in terms of EIRP and compensated for both
the test transmitter and test receiver antennas. If the recorded power was specified as ERP
of the test transmitter only, the power must be adjusted by +4.3dB.
3) Perform steps 3-8 of section 5.3 if these settings have been changed.
4) Run the analysis again this time for the combined collection of drive test data. A table
similar to that shown in Table 2 will be produced.
5) Sort the output result for the different drive tests in terms of antenna height.
6) Indicate on the table the model parameter settings and save the table in a suitable
directory with a label indicating the tuning progress stage ie K2MultiTuneStep1, for later
reference.
7) Calculate the weighted mean and RMS error for each antenna height. This may be
performed by first, multiplying each mean error (or RMS error) by the number of bin
samples. Then the sum of all the weighted mean errors (or RMS errors) are divided by the
total number of bin samples. The results of this calculation for the example have been
inserted at the bottom of the Mean Error and RMS error column in table 2.
8) The overall-mean error will indicate the magnitude of the prediction error. A positive
mean error indicates under-prediction.
9) The RMS error indicates the spread of results and a RMS error of less than 7dB is an
acceptable target.

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 11 of 188
RESTRICTED

The weighted mean and RMS errors may appear to increase with lower antenna height. This
will be because as the antenna height gets closer to the height of the surrounding clutter, the
coverage potential will be reduced. The model will not have the ability to accurately predict
the effects of clutter at low antenna heights since individual buildings and trees do not exist in
the database. The model tuning process will attempt to produce an optimum model, which is
useful for a range of antenna heights. For base stations with very low antenna heights which
are within only a few metres of the height as the surrounding clutter, the prediction model
may have unacceptably high errors. For these special cases, a separate model may be
required.

File Site ID Site Name Num. Bins Mean Error RMS Error Std.Dev. Error Corr. Coeff.
H:\SJB\MSI Data\M90\30m90_9.hd 30m90_9 Survey 5005 -0.0 6.6 6.6 0.8250
H:\SJB\MSI Data\M100\32m100_9.hd 32m100_9 Survey 1323 0.9 8.7 8.6 0.8183
0.19 7.04
H:\SJB\MSI Data\m710\20m710_9.hd 20m710_9 Survey 2711 -2.2 6.8 6.4 0.7712
H:\SJB\MSI Data\M500\20m500_9.hd 20m500_9 Survey 2156 -6.5 8.7 5.8 0.8980
-4.1 7.64
H:\SJB\MSI Data\m710\15m710_9.hd 15m710_9 Survey 2766 -3.3 7.6 6.9 0.7372
H:\SJB\MSI Data\M500\15m500_9.hd 15m500_9 Survey 1479 -6.2 7.9 4.8 0.9015
-4.3 7.7

H:\SJB\MSI Data\m710\13m710_9.hd 13m710_9 Survey 1912 -7.5 10.5 7.4 0.7109


H:\SJB\MSI Data\M500\13m500_9.hd 13m500_9 Survey 1458 -6.0 8.1 5.5 0.8860
-6.85 9.46

Model Num. Bin Mean Error RMS Error Std.Dev. Error Corr. Coeff. K1 = 135
Training G9 Model 18810 -3.2 7.9 7.2 0.7881 K2 = 43
K7 = 0.6
Clutter Num. Bins Mean Error RMS Error Std.Dev. Error Corr. Coeff. Clutter
Disabled
Open 1436 2.3 7.8 7.5 0.7630
Roads 5319 -2.6 8.0 7.6 0.7205
Vegetation 259 -0.9 7.5 7.5 0.5160
Low Density Urban 3162 -0.3 6.2 6.2 0.8125
Medium Density Urban 6303 -5.6 8.1 5.9 0.8391
High Density Urban 432 -2.7 6.8 6.2 0.7113
Light Industrial/Commercial 1802 -6.7 9.2 6.4 0.8645
CBD 97 -1.7 5.8 5.6 0.3981

Table 2. Model Tuning Analysis Output with Multiple Sites.

6.5 Multi-site Drive Model Tuning.


1) The tuning of K1 and K2 must be repeated for the multi-site case. The overall mean error
provides an indication of the K1 error so this value may be subtracted from K1 to
commence the first step of the tuning process. Follow the steps described in 6.2 and 6.3
for the process, tuning K1 first to reduce the overall mean error to 0, then tuning K2 to
produce the lowest RMS error. Ensure the Non-LOS data is excluded in the CW
Measurement Analysis filter.
Asset Propagation Model Tuning
Version: Draft 1
14/7/2003 Page 12 of 188
RESTRICTED

2) Evaluate the results on the Asset 2D view comparing the prediction with the drive test for
different example sites. Note that each new site must be entered into the Asset database
and the configuration entered to match the test transmitter. The revised values of K1 and
K2 will of course degrade the prediction of the “reference” site but will provide a more
useful model for other site locations. The degradation in the numerical analysis of the
reference site is expected and acceptable, provided the RMS error remains low.

6.6 Tuning of K7
1) In the Asset CW Measurement Options Filter, select the “NON-LOS” data and de-select
the “LOS” data.
2) Run the analysis process and evaluate the table produced. Consider small changes to K7
with steps less than 0.2. If an increase in K7 causes the RMS error to increase, try a
reduction in the value.
3) Once an optimum value for K7 has been found, try a few case studies by comparing the
prediction with drive tests. The case study examples must have situations in the drive test
where a non-line-of-sight situation occurs such as a drive through a cutting or river valley.
At the lower part of the valley, the terrain will block the path between the test transmitter
and the drive test vehicle. Proper tuning of K7 will provide more accuracy for the
coverage in these specific cases of the environment.
At this stage, the final values for K2 and K7 have been established. K1 will require further
fine-tuning following the influence of the clutter class tuning.

6.7 Tuning of Clutter Coefficients

6.7.1 Clutter Offsets


1) In the Asset CW Measurement Options Filter, select the “NON-LOS” data and also select
the “LOS” data.
2) In the Filter settings, exclude all clutter type except for “open” clutter type. If an open
clutter class is not provided, a similar clutter type may be suitable, such as “scrub”,
“fields”, “crops” and “rocky ground”. A suitable class which represents the major area of
morphology which has structures less than 1m height and has been used to represent open
areas of land such as parks, fields crops etc.
3) Run the analysis and evaluate the table produced.
4) Subtract the overall mean error of the model from the value of K1 and run the analysis
again. The overall mean error should now be zero and the model is now normalised to the
“open” clutter class.
5) In the Asset CW Measurement Options Filter include all the clutter types with the
exception of Water and Null.
6) Run the Analysis and evaluate the table produced.

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 13 of 188
RESTRICTED

7) For each clutter type in the model, subtract the overall mean error specified from the
tables for the specific clutter class. If a particular clutter class is not included in the table
or the number of samples in the Bins are low, this means the drive test did not pass very
often through the particular clutter zone, or did not pass at all through the zone. In this
situation, a large error may occur when tuning the coefficient of this particular clutter
class. The specific coefficient may therefore need to be set by careful study of the drive
tests and prediction plots to produce a realistic prediction based on experience and the
numerical analysis process should not be used.
8) Run the analysis and evaluate the table produced. The mean errors should now be zero
and the RMS error should be an acceptable level. If the examples with low antenna height
have high RMS errors (greater than 9dB), a separate model should be considered for
accurate prediction of these cases. The “low antenna height” model will typically only
require special values for K1 and K2 to obtain a more acceptable accuracy. The final
values of the clutter coefficients should be used on the special variants. Recalculate the
prediction of the “reference” site and plot the coverage on the Asset 2D view. This plot
should now reveal more detail and match the drive test data in many more locations than
before the clutter offsets were enabled.
9) If a “Roads” or “Streets” clutter class is available in the clutter database, there may be a
tendency to over predict these regions. The result will look on a coverage plot as though
the roads (or streets) are highlighted with more coverage than other surrounding regions.
To some extent this is true since the open area of a road allows for a more direct path to
the transmitter and coverage can be better on a wide road than within a narrow street for
example. This is also true when the road is aligned in a radial path with respect to the
transmitter. Unfortunately, the clutter class database does not usually accurately register
with the actual roads on which the drive was performed. For this reason the numerical
analysis will be misleading. It is recommended to “de-tune” the roads clutter loss to some
degree and it may also prove to be optimum to set the “roads” clutter offset to match that
of the most common clutter class in the database, such as Low Density Urban.
10) The clutter coefficient for water can be set to 0.
11) Although the clutter coefficients can be adjusted to get the numerical analysis optimised
for minimum errors, the eventual coefficients need to have sensible relativities. For
instance, the vegetation (trees) coefficient should be higher than the open class. Similarly,
the High Density Urban areas should have more attenuation than Low Density Urban
areas and the loss coefficient should be proportionally higher. Some exceptions to this
generality may occur however. A clutter class such as “Industrial” may appear to have a
low value of loss coefficient compared to High Density Urban for instance. This can occur
if the typical industrial zone has widely separated and low-rise buildings and structures.
Areas such as this will potentially have higher signal strengths because of the general
open nature of the area and coverage can also be enhanced by reflection off the buildings.
As a guide, the expected variation of the offset between the different clutter types should
be of the order of 1-4dB. Individual case studies can be done to look specifically at
various clutter zones to see if the model is performing as expected.
12) Finally, the coefficient offsets can be “rounded off” to whole numbers.

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 14 of 188
RESTRICTED

6.7.2 Clutter Heights and Separation

The influence of clutter heights and separation distance will further enhance the prediction
accuracy. Asset performs a diffraction loss calculation over the edge of the clutter and
therefore requires the clutter height and the distance between the edge of the clutter and the
mobile end point. Each clutter height needs to be set and the logical starting value depends
on the specific clutter zone. A residential area or Low-Density Urban class will typically
have a clutter height of 6m. This is because the typical building height of a single storey
house is about 6m. Local variations will of course change these values. The separation
distance defines the distance between a typical mobile phone and the nearest structure such
as a building within the clutter zone. For example, in a suburban street the width of the
street is typically 8m wide with a nature strip of 2m on each side. The house is set back
from the street by typically 15m. Therefore the separation distance between the mobile
phone on the street and the nearest structure will be about 20m. This could be used as a
starting value for tuning of this coefficient. In all cases the separation distance should be
less than or equal to the database resolution.
Some suggested starting values for clutter heights and separations are listed below.

Clutter type Clutter Height Clutter Separation


Low Density Urban 6m 20m
Medium Density Urban 10m 20m
High Density Urban 15m 20m
High Rise Industrial 20m 30m
CBD 20m 10m
Roads 0m 0m
Trees 10m 30m

The following procedure is a guide for tuning the clutter height and separation.

1) In Asset, enable clutter heights and enter the recommended default values for heights
and separation as listed above.
2) Ensure that the drive test data for multiple sites are entered in the CW Measurements
Analysis.
3) Run the CW Measurement analysis. Note that the numerical analysis now indicates a
high Mean and RMS error.
4) Subtract the overall mean error from the value of K1 and re-run the analysis. The
numerical analysis should now be restored with an overall mean error of zero and a low
value of RMS error. The RMS error may also be lower than the result achieved without
the clutter heights and separations enabled. Individual clutter class, mean and RMS
errors provide an indication of the accuracy of the clutter heights and separation values.
5) Select a specific clutter type one at a time and perform the tuning process described in
steps 5 to 9.
6) Keep the value of clutter separation fixed and adjust the value of clutter height in small
steps (less than 2).
7) After each iteration of the analysis, readjust the value of the clutter offset by subtracting
the clutter mean error. If the RMS error of the clutter in the analysis report increases
with an increase in the value of the clutter height, reduce the value of the clutter height.

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 15 of 188
RESTRICTED

8) Tune the height value for a minimum RMS error for the clutter class (not the overall
model RMS error). If the clutter height does not appear to tune to a minimum RMS
error value or the tuning is “vague”, select a suitable default value for the clutter height.
Poor tuning can occur if insufficient drive test data is available for a particular clutter
class, or if the clutter data is inaccurate. Adjusting the clutter value to a height, which is
unrealistically lower than the actual real world height is not recommended. However, If
the particular clutter class only occurs in the actual environment in small areas or
pockets, the influence on the path loss may also be small. Experience and best
judgement may be required when setting the final clutter height in these cases.
9) Evaluation of the final clutter height tuning using a prediction plot and examination of
“case studies” will also assist in determining a suitable clutter height.
10) When all the clutter heights have been tuned, commence tuning the clutter separations
one at a time in a similar manner, using steps 5 to 9. Adjustment of the clutter
separation may be done using larger steps such as 5 to 10m.

The tuning of the clutter separation usually has a low sensitivity and a well-defined
minimum in the RMS error may be difficult to find. In situations where it is not possible to
obtain a clear minimum on the RMS error, experience and best judgement may be required
to set a suitable value.

11) When the Clutter heights and separations have all been set to an optimum value, ensure
that the CW Measurements Options filter has selected all clutter types except Null and
Water.
12) Run the analysis once more and note the clutter mean errors.
13) Subtract the mean errors from each clutter offset in the prediction model. Round off the
Clutter offsets to the nearest integer.

6.7.3 Effective Height Limits

At this stage the Effective Height will have the default settings of minimum 10m and
maximum of 100m. If the lower limit is reduced below 10m, as may be expected for a base
station, which has a low antenna height, a problem may appear. The method Asset uses to
apply the antenna pattern to the prediction, allows all characteristics of the antenna pattern
to modulate the prediction. This includes the pattern nulls on the underside of the main
lobe. The underside nulls may affect the actual coverage but this effect becomes less
significant when the distance is less than a few hundred metres distance from the base
station. At short distances, the effects of reflection become dominant and counteract the
influence of the antenna pattern underside nulls. To simulate the effect of these near field
reflections, it is necessary to use a suitable value for Effective Height minimum. If the
minimum is reduced to 0 for instance, the effect will be to reveal coverage holes quite close
to the base station, which are created by the antenna pattern. This is not found in practice
and will look unrealistic.

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 16 of 188
RESTRICTED

At the other extreme, the maximum value of Effective Height needs to be tuned based on a
specific characteristic of a base station. The specific characteristic is the relative height
difference between the base station antenna height and the predominant height of the area
receiving coverage.
As an example, a typical suburban base station with 25m height on relatively flat terrain
will only require a maximum Effective Height value of approximately 50m. This limit will
be valid for base station heights up to 50m in this type of environment.
Alternatively, the base station may be located on the top of a mountain and provide
coverage to a valley below. In this situation the relative height difference may be as much
as 300m to1000m. A much greater value of Effective Height is required for these
“mountain” classes of base stations, and a value of 500 to 1000m may be suitable.
If the Effective Height Maximum is not adjusted accordingly, an under prediction will
occur for the mountain site with a low value set. Conversely, an over-prediction of the
coverage will occur for a flat terrain suburban site if a high value is used for the Effective
Height maximum.
It is usually not necessary to have a large number of different Effective Height model
variants. A “standard” and a “Mountain” are usually sufficient.
To effectively tune the “Mountain” model Effective Height maximum value, it is necessary
to have drive test data from a test case which will create this condition. The drive test will
also need to collect measurements from a great distance since the range of the “mountain”
site can be very large. The tuning of the Effective Height maximum will be done to make
the more distant prediction more accurate.

The tuning method using the numerical analysis may not be effective and case studies of
drive tests compared graphically to the prediction may prove to be the best method for
deciding on a suitable value of the Effective Height maximum.

The model tuning process is complete.

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 17 of 188
RESTRICTED

7. Document Control

7.1 Authorisation
A. Author Stephen J. Bentley

B. Authorised

C. Approved

7.2 Amendment List


Version Date Section Nature of Amendment Amendment
Author
Draft 1 All Document compilation Stephen J. Bentley

Asset Propagation Model Tuning


Version: Draft 1
14/7/2003 Page 18 of 188

Das könnte Ihnen auch gefallen