RF Guideline Propagation Model Tuning
RF Guideline Propagation Model Tuning In ASSET
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 1 of 53
RF Guideline Propagation Model Tuning
Document History Version 1.0 Date 06/01 Author(s) Hans-Hubert
Rhrig Change Description First draft.
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 1 of 53
RF Guideline Propagation Model Tuning
CONTENTS1 2 3 INTRODUCTION ................................
................................ ................................
............................. 4 WHY TUNING A PROPAGATION MODEL?
................................ ................................
................... 5 IN COMMON USE PROPAGATION MODELS
................................ ................................
................ 6 3.1
OKUMURA-HATA-MODEL................................
................................ ................................
.............. 6 3.2 COST231-HATA MODEL
................................ ................................
................................ ............. 7 3.3 RACE-1043
CLUTTER MODEL ................................
................................ ................................
..... 8 3.4 EXTRA DETERMINISTIC METHODS
................................ ................................
................................ . 9 3.4.1 Common in use Knife-edge
Diffraction Methods ................................
................................ . 9 3.4.2 Effective Antenna Height
Calculation................................
................................ ................ 11 4 INDICATORS
OF PREDICTION M ODEL PERFORMANCE ................................
.......................... 12 4.1 BASIC STATISTICS
................................ ................................
................................ ..................... 12 4.2
PREDICTION ERROR STATISTICS OF AIRCOM INTERNATIONAL ASSET
................................ .............. 13 4.2.1 Displaying
Prediction Error in the 2D-View ................................
................................ ...... 13 4.2.2 Displaying
Received Level/Prediction Error vs.
Log(d)................................ ...................... 14
4.2.3 Asset Analyse Text File ................................
................................ ................................
... 15 4.3 PREDICTION ERROR STATISTICS OF
MEAANALYSE................................
................................ ........ 16 4.3.1 MeaAnalyse
output file _Summary.txt ................................
................................ ......... 16 4.3.2 MeaAnalyse
feature StandardDeviationVsMeanError
................................ ....................... 18 4.3.3
MeaAnalyse feature(s) ...versus Distance
................................ ................................
........ 19 5 INPUT DATA ................................
................................ ................................
................................ . 20 5.1 MAP
DATA................................
................................ ................................
................................ 20 5.1.1 Paper Maps
................................ ................................
................................ ..................... 20 5.1.2
Topographical Database ................................
................................ ................................
.. 21 5.2 CW SURVEY DATA ................................
................................ ................................
.................... 22 5.3 START PARAMETER VALUES OF PROPAGATION
MODEL ................................
................................ . 23 5.3.1 Aircom ASSET Standard
Macrocell Model ................................
................................ ....... 23 5.3.2 Classification of
Hata Adjustment Coefficients to ASSET
k-parameter.............................. 24 5.3.3 Enhancement of
the ASSET Standard Macrocell Model................................
................. 25 Clutter
Category................................
................................ ................................
............................. 27 6 WHICH COEFFICIENTS ARE TUNABLE?
................................ ................................
................... 28 6.1 ADJUSTMENT COEFFICIENTS OF HATA-MODELS
................................ ................................
............ 28 6.1.1 Intercept C1 and Frequency Coefficient C2
................................ ................................
....... 29 6.1.2 Base Station Heights Adjustment Coefficients C3
................................ ............................. 30
6.1.3 Path Loss Slope................................
................................ ................................
............... 31 6.1.4 Mobile Antenna Height Correction
................................ ................................
.................... 36 6.1.5 Clutter Adjustment L C
................................ ................................
................................ ....... 37 6.1.6 Diffraction Loss
LD and Adjustment Coefficient C6 ................................
............................ 38 7 THE CALIBRATION PROCESS
................................ ................................
................................ .... 39 7.1 SORT CW MEASUREMENT
DATA ................................
................................ ................................
40 7.2 F IRST CW MEASUREMENT ANALYSIS
................................ ................................
.......................... 40 7.3 F IND BEST SUITED EFFECTIVE
ANTENNA HEIGHT CALCULATION METHOD ................................
........ 40 7.4 T UNE BASE STATION HEIGHT ADJUSTMENT COEFFICIENTS
................................ .............................. 41
7.4.1 Tune base station height adjustment coefficient k5
................................ ........................... 41
7.4.2 Tune base station height with distance adjustment coefficient
k6 ................................ ...... 42 7.5 T UNE INTERCEPT
AND SLOPE COEFFICIENT ................................
................................ .................. 42 7.5.1
Intercept ................................
................................ ................................
.......................... 43 7.5.2 Slope
................................ ................................
................................ ............................... 44
7.5.3 Near/Far Intercept and Slope Coefficients
................................ ................................
........ 44Author: Doc-ID: Date: H.H. Rhrig Lucent Technologies
Proprietary RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant
to Company Instructions Revision: Page: 1.1 2 of 53
RF Guideline Propagation Model Tuning
7.6 7.7 7.8 7.9 8
T UNE CLUTTER OFFSETS................................
................................ ................................
........... 45 F IND BEST SUITED KNIFE-EDGE DIFFRACTION METHOD
................................ ................................
... 46 T UNE THE DIFFRACTION ADJUSTMENT
COEFFICIENT................................
................................ ....... 46 REANALYZE, F INE TUNING
................................ ................................
................................ .......... 47
HOW TO USE MEAANALYSE ................................
................................ ................................
...... 47 8.1 8.2 8.3 8.4 GET THE BIN INFORMATION
................................ ................................
................................ ........ 47 GET SPREAD SHEETS WITH
MEAANALYSE ................................
................................ ................... 50 CREATE
CHARTS BY EXCEL ................................
................................ ................................
........ 50 CREATE CHART STANDARD DEVIATION VS. MEAN ERROR IN EXCEL
................................ ................ 52
9
MATH BASICS ................................
................................ ................................
.............................. 55 9.1 9.2 STATISTIC BASICS
................................ ................................
................................ ..................... 55
LOGARITHMIC BASICS ................................
................................ ................................
................ 56
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 3 of 53
RF Guideline Propagation Model Tuning
1
Introduction
To implement a mobile radio system, wave propagation models are
necessary to determine propagation characteristics for any
arbitrary installation. Predictions are required for a proper
coverage planning, for interference analysis as well as for cell
calculations, which are the basis for the RF network design and
optimization purposes. However, the radio propagation channel is a
very critical component for mobile radio communications systems.
The field strength level, at a given point, not only depends on its
distance from the transmitter, the frequency of transmission and
the antenna heights but also on the long-term and short-term
interferences caused by reflections of the natural environment
(terrain configuration, vegetation) and the man-made environment.
This influences the wave propagation in different ways. Well-known
empirical path loss prediction models like the model of Okumura
-Hata or the COST231Hata model estimates the median signal strength
in a small area and do not consider the path specific propagation
effects by detailed analytical expressions. The Hata models (or
other empirical methods) only use simple empirical expressions
extracted from curves get from the analysis of measurement data.
This has the advantage of implicitly taking all path specific
propagation effects of the environment (known or unknown) into
account mentioned above. However, each region or country and in the
end each city has the own specific character of topography,
vegetation and man-made structure have an effect on the wave
propagation. Therefore, empirical models must always be subjected
to stringent validation by testing it on measurement data sets
collected at locations and conditions (as well as at transmission
frequencies) which are in many cases other than used to produce the
model in the first place. The overall objective of the tuning
process is to adapt the propagation model to the local environments
characterized by CW measurement data, in conjunction with the
specific classification of the actual terrain database. But a tuned
propagation model is only good as the input data used to calibrate
it. Consequently, the results of the tuning process depends on
quality and quantity of the CW measurements, on the quality of used
terrain database as well as on the ability of the RF Planning tool
to support the user with suitable applications to the CW
measurement analysis process. Furthermore, the person who will
carry out the tuning process should have knowledge about the basic
mathematics and the basic wave propagation mechanism in different
mediums as well as knows the common in use propagation models,
effective antenna height calculation as well as knife-edge
diffraction methods. The purpose of this paper is not to describe
the perfect way of tuning empirical propagation models, because
there is no single correct way or ideal method. This paper tries to
give recommendations and methods useful for the tuning process with
help of the Aircom Asset CW Measurement Analyse Tool.
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 4 of 53
RF Guideline Propagation Model Tuning
2
Why tuning a propagation model?
The overall object of tuning a propagation model is to adapt the
path loss prediction model to the local environments and the
specific classification of the actual terrain database to improve
the coverage estimation (path loss prediction). Because: y Each
region or country has the own specific character of vegetation and
man -made structure that influence the wave propagation on
different ways. The Hata models based on the Okumura technique
adopts curves for urban areas based on the type and density of
buildings in Tokyo and it may not be transferable to cities in
Europe or North America. Indeed, experience with CW measurements in
the USA (e.g. South Carolina, Indianapolis and Boston) have shown
that the typical US urban environment lies is similar to Okumuras
definition of suburban. Empirical path loss prediction models like
the COST231-Hata model (see next chapter) are restricted to flat
terrain. In case of wavy (hilly) terrain or topographical obstacles
like mountains (obstruct the line of sight between BS and MS) the
Hata model has to combine by extra deterministic methods have to
use like knife-edge and/or effective antenna height calculation to
consider the influence of topography. Usually empirical models are
restricted to ranges of frequencies, antenna heights and distances.
If the parameter of the planned base stations are outside these
limitations, then the empirical model have to extent by analyzing
CW measurements. High-resolution terrain databases (e.g. pixel size
is from 20 meter up to 30 meter) are created by satellite images
(typical 10 meter resolution). However, the clutter database is the
result of a person, who interprets groups or cluster of
gray-pattern in the image and assign the marked area to the most
likely suitable clutter category. Furthermore, a satellite image
provides geo information about the local density and extent of
buildings, but it cannot give information about the local building
heights that also impact on wave propagation. Consequently, the
path loss prediction has to adapt to the topographical database by
the help of CW measurements. Some RF planning tools support extra
clutter attributes like clutter heights and separation. Using these
features can improve the accuracy of the coverage prediction. It is
recommended to validate the specified clutter information by CW
measurements.
y
y
y
y
Note:
Keep in mind, that the fitted propagation model is only
applicable to the local terrain database that was used for the
model tuning.
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 5 of 53
RF Guideline Propagation Model Tuning
3
In Common use Propagation Models
This chapter describes well-known propagation models used for
the coverage analysis in Macrocell environments like the model of
Okumura-Hata, the COST231-Hata model and the RACE model. The
propagation models mentioned below estimate the path loss by
empirical information. The empirical information based on the
analysis of RF propagation measurements. Empirical models calculate
the median signal for each pixel and cannot determine the local
mean signal that result from local effects of various multi-path
phenomena. The local mean signal levels have to distribute around
the pixel median with a log -normal probability distribution.
3.1
Okumura-Hata-Model
The Okumura-Hata prediction model is based on empirical
information obtained from measurements in Japan (Okumura 1965).
From the results of these measurements propagation curves in the
frequency ranges from 200 [MHz] up to 2[GHz] depending from the
distance (1- 100 [km]) to the transmitter have been extracted.
Curves are given for effective base station antenna heights in the
range 30 [m] 1000 [m] and for a mobile station antenna height of
1.5 [m]. The Hata formula (1980) is a mathematical fit for the
Okumura graphical measurement results. Four parameters are used for
estimation of the propagation loss by Hata's well known model:
frequency f, distance d, base station antenna height hBS and the
height of the mobile antenna hMS. In Hatas model, which is based on
Okamuras various correction functions the basic transmission loss,
Lb, in urban areas is: h BS f ! 69.55 26.16 lg 13.82 lg [m] [MHz]
[dB ] Lb Where: f hBS hMS d a(hMS) The model is restricted to:
Frequency f : Base station antenna height hBS: Receiver antenna
height hMS: Distance d from the site : 150 - 1500 [MHz] 30 - 200
[m] 1 - 10 [m] 1 - 20 [km] frequency in [MHz] base station antenna
height in [m] mobile antenna height in [m] distance between base
station and mobile station in [km] mobile antenna height correction
in [dB] h 44.9 6.55 lg BS [m] lg d a h [km] MS
(3.1.1)
The model of Okumura-Hata is restricted to quasi-smooth terrain
where the average height of terrain does not change more than 20
[m] and the actual elevations of the path profile undulate in a
range of no more than 10 m due to the average height. Furthermore,
the model of Okumura-Hata is limited to large and small
macro-cells, i. e. base station antenna heights above rooftop
levels adjacent to the base station.Author: Doc-ID: Date: H.H.
Rhrig Lucent Technologies Proprietary RFET-QA-REP-00-010-V01.00 12
September 00 Use Pursuant to Company Instructions Revision: Page:
1.1 6 of 53
RF Guideline Propagation Model Tuning
The Okumura-Hata formula is quite good in urban and suburban
areas. However, in rural areas over irregular terrain there is a
tendency to be too optimistic.
Additional the model of Okumura-Hata considers the effects due
to land usage in the vicinity of the MS by empirical corrections
(recommended by ETSI). The clutter correction for suburban area is
defined by: L Suburban [dB ] d ! Lb 2 lg f 28[MHz] 2 5.4
(3.1.2)
The clutter correction for rural (quasi-open) area is defined
by:
[dB]
b
The clutter correction for rural (open) area is defined by:
[d ]
As L0 in equation (3.1.1) only applicable for a mobile antenna
height hMS=1.5 [m]. For other values of hMS the term a(hMS) is a
correction of the path loss L0. The corrections to the mobile
antenna height correction depends on the frequency range and the
land usage in the vicinity of the mobile station. If the mobile
station be in urban environment the adjustment to the mobile
antenna height is defined by:
In su ur an or rural environment the loss correction to the
mobile antenna height is defined by: ah MS Suburban ,Rural [dB
]
3.2
COST231-Hata Model
The COST-231 group has extended Hatas model to the frequency
band from 1500 [MHz] up to 2000 [MHz] by analyzing Okumuras
propagation curves in the upper frequency band. The repeated
analysis of the measured propagation curves of Okumura within this
fr quency range e resulted in a change of the term, which depends
on the frequency. Additionally a new correction factor was
introduced that increases the propagation path loss for
metropolitan centers. This combination is called
"COST231-Hata-Model". The basic path loss (Lb) in urban areas
is:Author: Doc-ID: Date: H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions Revision: Page: 1.1 7 of 53
a hMS Ur an [dB ]
3.2 l
h 11.75 MS 4.97 [m]
1.1 l
hMS f [MHz] 0.7 [m ] 1.56 l
L
ural(open )
f d ! Lb 4.78 lg MHz 18.33 lg MHz 40.94
2
2
f [MHz] 0.8
Rural(quasi - open)
d
f 4.78 lg MHz
2
18.33 lg
f 35.94 z
(3.1.3)
(3.1.4)
(3.1.5)
(3.1.6)
RF Guideline Propagation Model Tuning
Lb h BS f ! 46.30 33.90 lg 13.82 lg [m] [dB ] [MHz]
h 44.9 6.55 lg BS [m]
lg d a h [km] MS Cm
(3.2.1)
Where: f hBS hMS d Cm frequency in [MHz] base station antenna
height in [m] mobile antenna height in [m] distance between base
station and mobile station in [km] 0 [dB] for medium sized city and
suburban centers with medium tree density or 3 [dB] for
metropolitan centers
The mobile antenna height correction a(hMS) is defined by: hMS a
hMS f f ! 1.1 lg [MHz] 0.7 [m] 1.56 lg [MHz] 0.8 [dB ] The
COST231-Hata-Model is restricted to the following range of
parameters: Frequency f : Base station antenna height hBS: Receiver
antenna height hMS: Distance d from the site : 1500 - 2000 [MHz] 30
- 200 [m] 1 - 10 [m] 1 - 20 [km] (3.2.2)
The application of the COST231-Hata-Model is restricted to large
and small macro-cells, i. e. base station antenna heights above
rooftop levels adjacent to the base station. The clutter
corrections (equations 3.1.2, 3.1.3 and 3.1.4) mentioned in the
chapter before applicable to the COST231-Hata model as well.
3.3
RACE-1043 Clutter Model
The characterization (in view of different densities and/or
heights of buildings and vegetation) of the environment (clutter
category) in the vicinity of the mobile station is very important
for the path loss estimation. Because, the median signal and the
local mean signal distribution, at a given point, depends on the
land usage like vegetation and/or man-made structure (buildings).
The Okumura-Hata and COST231-Hata formulas treat only three
different types of land usage (urban, suburban and rural). However,
three land usage categories are not sufficient to characterize the
effects due to land usage in the vicinity of the MS. Within the
scope of the RACE-1043 working group, the three main clutter
classes (urban, suburban and open field) have been subdivided into
several clutter categories to distinguish between different
densities and heights of vegetations and buildings. Table 1 shows
the different clutter categories and the loss correction
recommended by RACE-1043.Clutter type W O1 O2 F1 F2Author: Doc-ID:
Date:
Description Water Open field, no obstructions Open field, few
obstructions Forest, low density with small trees or bushes Forest,
mostly higher and more densely packed treesRevision: Page:
Cm [dB] -29 -24 -19 -19 -91.1 8 of 53
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
RF Guideline Propagation Model Tuning
S1 S2 S3 U1 U2
Suburban, low density Suburban, leafy buildings Suburban,
density buildings Urban, low density, 2 - 5 floors Urban, high
density, more than 5 floors
-11 -8 -5 -3 0
Table 1: Clutter correction factors recommended by RACE-1043
3.4
Extra Deterministic Methods
The model of Okumura-Hata and the COST231-Hata model restricted
to flat terrain. Consequently, in case of wavy (hilly) terrain or
topographical obstacles like mountains (obstruct the line of sight
between BS and MS) the accuracy is generally decreased.
Incorporating terrain information can make a substantial difference
to the prediction. Therefore, extra deterministic methods have to
use like knife-edge and/or effective antenna height calculation to
consider the influence of topography. Empirical models combined
with deterministic methods are semi-empirical propagation models.
3.4.1 Common in use Knife -edge Diffraction Methods
When topographical obstacles (hills, mountains) obstruct the
line of sight between the base station antenna and mobile station
additional diffraction losses occurs which depend on the height and
the location of obstacles (see Figure 1).
Figure 1: None line of sight (NLOS) condition
3.4.1.1
Bullington Method
The Bullington method calculates the diffraction loss over
multiple obstructions by considering a single equivalent knife-edge
positioned at the point of intersection of the transmitter and
receiver horizon paths. The total diffraction loss is taken as that
over the equivalent knife-edge obstruction. This method has the
advantage of being simple, but often significant obstacles can be
ignored leading to an optimistic estimate of field strength.
However, this knife-edge diffraction method achieve the lowest
standard deviation in many cases. 3.4.1.2 Epstein-Peterson
Method
The Epstein-Peterson technique is based on the assumption that
the total loss can be evaluated as the sum of attenuation due to
each respective significant obstruction.Author: Doc-ID: Date: H.H.
Rhrig Lucent Technologies Proprietary RFET-QA-REP-00-010-V01.00 12
September 00 Use Pursuant to Company Instructions Revision: Page:
1.1 9 of 53
RF Guideline Propagation Model Tuning
The diffraction loss from the obstacle is calculated by assuming
that the receiver is at the second obstruction. The loss from the
second obstacle is then calculated assuming the transmitter is at
the first obstruction and the receiver at the third. Furthermore,
the loss from a transmitter at the next obstacle to the receiver is
calculated. The total Epstein-Peterson diffraction loss is given by
the sum of all the losses calculated.
The Epstein-Peterson technique overcomes one of the problems of
the Bullington metho namely that d, important obstacles can be
ignored. However, it has been demonstrated that this method has
limitations when the obstructions are closely spaced.
3.4.1.3
Japanese Atlas Method
The Japanese Atlas technique is similar to the Epstein-Peterson
method and was proposed by the Japanese postal service. Again it is
based on the assumption that the total loss can be evaluated as the
sum of attenuation due to each obstruction. However, in contrast to
the Epstein-Peterson method the effective source is not the top of
the preceding obstruction but the projection of the horizon ray for
the obstruction to a point on the vertical plane through one of the
terminals. This method gives improved results when the obstructions
are closely spaced. 3.4.1.4 Deygout Method
The Deygout technique calculates a v-parameter for each edge,
the one with the largest is termed the main edge and its loss
calculated in the standard way. Additional losses for other
obstructions are calculated between the main edge and the
obstructed terminal. The total Deygout loss is given by the sum of
all losses calculated. In order to extend the technique to many
obstructions it is necessary to employ sub-main edges. These are
the next most significant edge(s) at either side of the main edge.
The loss form the sub-main-edge is calculated assuming a
hypothetical terminal located at the main -edge (ignoring any less
significant edges). This method provides accurate results where
there are two obstructions, with one being clearly dominant.
However, it tends to over-estimate losses where there is no
dominant obstruction. For 3 or 4 obstructions the Deygout method
gives the best results of any of the approximate methods. However,
for 4 or more obstructions Deygout will tend to overestimate the
loss.
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 10 of 53
RF Guideline Propagation Model Tuning
3.4.2
Effective Antenna Height Calculation
An important parameter with respect to the topography is the
determination of the effective antenna height. As the mobile
station moves, the effective base station antenna height changes.
Figure 2 illustrates some of the possibilities as well as the
different methods to determine the effective antenna height.
Figure 2: Possibilities to determine the effective antenna
height 3.4.2.1 Absolute Method
The absolute method uses the height of the base station antenna
above ground as the effective antenna height. The absolute method
is suitable in flat terrain. 3.4.2.2 Average Method
The average method is calculated as the base station antenna
height above the average terrain height across the area of the
prediction. The average method is suitable in flat or gently
rolling terrain. 3.4.2.3 Relative Method
The effective antenna height is determined as the relative
height of the base station antenna to the mobile station, if the
height above sea-level of the mobile station is lower than the
height above sealevel of the base station antenna. In the reverse
case, the height of the base station antenna above ground is the
effective antenna height. Otherwise this definition leads to
negative antenna heights. The relative method is reliable in
rolling hilly terrain where the mobile station is mainly below the
base station antenna.
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 11 of 53
RF Guideline Propagation Model Tuning
3.4.2.4
Slope Method
The effective antenna height is defined as the line from the
base station antenna to the fictitious elongation of the mean
terrain in front of the mobile station towards to the base station.
The definition have to restrict by a limitation of the result
effective antenna height. Otherwise the slope method can lead to
negative antenna height or lead to very height antenna heights
(>200m). In conjunction with the Hata-models, the slope method
should be restricted to a minimum effective antenna height of 30[m]
and a maximum effective antenna height of 200[m]. The slope method
is suitable in very wavy areas where the terrain increases or slope
very strong in s front of the mobile station.
44.1
Indicators of Prediction Model PerformanceBasic Statistics
The statistics enable the user to assess the accuracy and
reliability of the specified propagation prediction model. The in
common use statistics are: y y y Mean error Root mean square (RMS)
error Standard deviation
The prediction error at a given point i is the result of the
measured signal subtracted by the predicted signal level:
?dBm A
The mean (median) prediction error is the sum of the prediction
error over all n points: Q 1 ! ?dB A n
i !1
n
Qerr r i ?dB A
The mean prediction error shows the tendency of the specified
propagation model. A positive mean prediction error signifies that
the prediction model is too pessimistic. Accordingly, a negative
value signifies that the prediction model is too optimistic. The
root mean square error (RMS) is defined by:
i !1
The RMS declares the overall variation range (mean prediction
error and standard deviation) of possible prediction error.
Consequently, the RMS error is greater than the standard deviation,
if the mean prediction error is unequal 0 [dB]. The RMS error is
equal the standard deviation, if the mean prediction is equal 0
[dB]. The standard deviation is a suitable indicator to assess the
accuracy of the prediction model and is defined by:
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
!
Q RMS ! ?dB A
1 n
n
Q err r i 2 ?dB A
Si nal _ measured
Si nal _ r dict d
?dBm A
Q err r i ! ?dB A
i
i
(4.1.1)
(4.1.2)
(4.1.3)
Revision: Page:
1.1 12 of 53
RF Guideline Propagation Model Tuning
W ! ?dB A
1 n 1
Q Q err r i ? A ? A dB dB i !1
A low standard deviation in conjunction with a low RMS error
indicates a well-tuned prediction model.
A typical statistical result by a well-tuned propagation
prediction model is a mean error of r3 [dB] and a standard
deviation of 8-9 [dB].
4.2
Prediction Error Statistics of Aircom International Asset
Aircom International Asset provides several features supports
the model calibration process. However, not all features suitable
for the model calibration. For additional guidance for using the
features of the Aircom Asset CW Measurement Analysis tool, please
have a look into the application note ASSET Standard Macrocell
Model Calibration provided by Aircom International. 4.2.1
Displaying Prediction Error in the 2D -View
Following items can be displayed on the Aircom Asset 2D-View: y
y y y Measurement Route Carrier Wave Route tags Carrie Wave Signal
Carrier Wave Signal Error
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
"
n
2
(4.1.4)
Revision: Page:
1.1 13 of 53
RF Guideline Propagation Model Tuning
Figure 3: 2D-View displaying Carrier Wave Signal Error
Especially for areas with wavy (hilly) terrain, displaying the
Carrier Wave Signal Error together with the Terrain Height can
point out graphical the coherence between prediction error and
influences of actual terrain along the direct propagation
prediction path or terrain features at the location of the MS. This
feature is useful to find a suitable effective antenna height
calculation method and/or knife-edge diffraction method.
4.2.2
Displaying Received Level/Prediction Error vs. Log(d)
Figure 4 shows the graph Received level vs. Log(d) (similar to
Prediction Error vs. Log(d)) produced by Asset. The different
colored data points represents different clutter types.
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 14 of 53
RF Guideline Propagation Model Tuning
Figure 4: Asset Graph of Received Level vs. Log(d) By this
feature it is impossible to find out specific adjustment
coefficients from a extend cloud of points. Furthermore, it is
difficult to find out tendencies. 4.2.3 Asset Analyse Text File
Aircom Asset will produce a text file similar to the shown in
Figure 5.
File Summary
Overall Summary
Clutter Summary
Figure 5: Asset Analyse Text File
The following table shows options available in the Asset Analyse
text file. Options File SummaryAuthor: Doc-ID: Date:
1
#
in Information
Suitable to... point out failed measurements of one test
siteRevision: Page: 1.1 15 of 53
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
RF Guideline Propagation Model Tuning
2 3 4
Overall Summary Clutter Summary Bin Information
calibrate the intercept calibrate the clutter adjustment import
into a spreadsheet application, from where it is possible to
produce charts and graphs
Option 13 provide more detailed (numerical) information as
against the graph Received Level/Prediction Error vs. Log(d).
However, the option give only overall (summary) results and
provides no information about the progress of the prediction error
and the standard deviation with distance. The analysis of the
prediction error with distance is needed to calibrate the slope
adjustment coefficients or to determine near and far adjustment
coefficients. Option 4 provides the possibility to import the
results into a spreadsheet application like Excel. Unfortunately,
Excel can only process 32000 lines (equivalent to 32000 bins). In
many cases, the number of bins available for analysis will be more
than 32000.
4.3
Prediction Error Statistics of MeaAnalyse
The application MeaAnalyse creates several numerical statistics
in tabs separated ASCII text file for the import into Excel, from
where it is possible to produce charts and graphs. The purpose of
MeaAnalyse is to provide statistical outputs allow an easier and
faster analysis of the survey data (calibration of the propagation
model), for user not involved in the model calibration as well.
MeaAnalyse subdivides the survey data included for the CW
measurement analysis to the categories LOS&NLOS, LOS and NLOS
as well as to the used clutter. Furthermore, MeaAnalyse determine
not only the overall mean prediction error, RMS error and standard
deviation, but also for user defined segments (e.g. determine the
mean error, RMS error and standard deviation in segments of 500m)
of distances from the test site(s). MeaAnalyse processes the Aircom
Asset Analyse text file contains the bin information (option 4). A
guidance how to use MeaAnalyse together with Aircom Asset attached
in the Annex. MeaAnalyse creates 6 output ASCII text files:
_DistributedBins.txt _DistributedMeanError.txt
_DistributedRMSError.txt _DistributedStandardDeviation.txt
_DistributedStandardDeviationVsMeanError.txt _Summary.txt The first
5 outputs are spread sheets to import into Excel. The file
_Summary.txt can be displayed in a ASCII text editor. For the model
calibration process the files
_DistributedStandardDeviationVsMeanError.txt and _Summary.txt very
useful. For the graphical presentation of the propagation model
performance (e.g. mean error with distance) the first 4 outputs
very useful.
4.3.1
MeaAnalyse output file _Summary.txt
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 16 of 53
RF Guideline Propagation Model Tuning
The MeaAnalyse output file _Summary.txt provides numerical the
prediction error statistics like mean error (MeanErr), RMS error
(RMSErr) and the standard deviation (StdDev) as well as the number
of bins (Bins) considered for the CW measurement analysis. Figure 6
shows in detail the content of the MeaAnalyse output file
_Summary.txt.
Figure 6: MeaAnalyse output file _Summary.txt The survey data
(prediction error statistics) subdivided by Clutter Type (e.g.
Clutter Type : light_density_residential) included for the CW
measurement analysis in Aircom Asset. Each found clutter category
subdivided into LOS (receiver (pixel) has line of sight to base
station), NLOS (receiver (pixel) has none line of sight to base
station) and Total (summary of LOS and NLOS). Furthermore, the
prediction error statistics subdivided into distance Intervals
(user defined segments). In the example shown in Figure 6 the user
defined interval is 250m. In Aircom Asset the CW measurement
analysis performed for a radius of 5km. Consequently, the survey
data (prediction error statistics) subdivided in 20 segments. Below
the intervals, the line starting with Total shows the summary of
all bins assigned to this clutter category. At the end of ASCII
text file the table below the line Clutter Type : Total shows is
the overall statistics.
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 17 of 53
RF Guideline Propagation Model Tuning
4.3.2
MeaAnalyse feature StandardDeviationVsMeanError
The purpose of the MeaAnalyse feature
StandardDeviationVsMeanError (graph created by Excel) is to provide
a simple proceed to rate the propagation model performance after
altering adjustment coefficients or to compare the improvement by
using different knife-edge or effective antenna height calculation
methods. Figure 7 shows the typical graphical output created by
Excel. The chart shows the standard deviation (yaxis) versus the
mean prediction error (x-axis). Each ball represents one segment
(interval of distance from the test site(s), e.g. 250m). The size
of the balls shows graphical the number of the bins assigned to the
distance interval. The ball has the size 100% stands for the
interval contains (covered by the) highest number of bins.
Figure 7: MeaAnalyse feature StandardDeviationVsMeanError
The closer the balls together and the closer all balls to the
origin the more fitted the propagation model to the measurements.
The example shown in Figure 7 the effective antenna height
calculation method Slope (second chart) gives a better fit to the
measurement as against the Absolute method (first chart). Because,
in the second chart the balls closer together.
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 18 of 53
RF Guideline Propagation Model Tuning
4.3.3
MeaAnalyse feature(s) ...versus Distance
The overall mean prediction error and the overall standard
deviation (in the end two values only) no suitable indicators to
judge the quality/reliability of the calibrated path loss
prediction model. A mean prediction error of 0 [dB] expresses
nothing on that the calibrated prediction model is faultless within
the radius of 3km from the test site or that the prediction error
is 0 [dB] at distances greater than 10km . For the RF planning
(design) process it is important to know the expected mean
prediction error and standard deviation (apropos Fade Margin)
within the (designed) cell radius. Furthermore, it is important to
know the mean prediction error on far distances impacts on the size
of the service area and the results of the interference analysis.
The Aircom Asset CW Measurement Analysis tool supports no suitable
features provide that. Only the Bin information processed in Excel
(or other spread sheet applications) allows the user to obtain the
information. Unfortunately, Excel can process only 32000 lines
correspond to 32000 bins (measured points). In may cases more than
32000 bins available for (included in) the CW measurement analysis.
Therefore, MeaAnalyse creates several spread sheets for further
easy and uncomplicated process in Excel. The Figures 8 show charts
display the number of bins, the mean error, the RMS error as well
as the standard deviation versus the distance created from the
output MeaAnalyse ASCII text files _DistributedBins.txt,
_DistributedMeanError.txt, _DistributedRMSError.txt and
_DistributedStandardDeviation.txt. For the examples below the
curves show the summary of the clutter types included in the CW
measurement analysis. But it is possible to created the charts
(statistics) for a single clutter category (e.g. urban) as
well.
Figure 8: MeaAnalyse features after imported/processed in
Excel
It is recommended to include such charts into the model tuning
report. Because, it also provides information about tuned model to
people were not involved in the model calibration process.
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 19 of 53
RF Guideline Propagation Model Tuning
55.1
Input DataMap Data
5.1.1 Paper Maps Paper maps (with scales from 1:10.000 to
1:250.000) provide extra geo-information (e.g. distribution and
density of the morphology) of the area surrounding the test site
and mobile station. Approximate all RF planning supports the import
of scanned maps. For many regions or countries, scanned paper maps
are available in the World Wide Web. Figure 9 shows the detail a
1:25.000 paper map (source jpeg format) of the area around
Entroncamento (Portugal). Figure 10 shows the detail of a 1:50.000
paper map (source jpeg format) of the city Columbia in the USA.
Figure 9: Scanned paper map (scale 1:25.000) of Entroncamento
(Portugal)
Figure 10: Scanned paper map (scale 1:50.000) of Columbia City
(USA)Author: Doc-ID: Date: H.H. Rhrig Lucent Technologies
Proprietary RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant
to Company Instructions Revision: Page: 1.1 20 of 53
RF Guideline Propagation Model Tuning
5.1.2
Topographical Database
The topographical database is the integral component in the
coverage analysis. The DTM (digital terrain model) data base
provides topographic concerning location and shape of obstacle
information (terrain features like hills, mountains or valleys)
impacts on wave propagation. The clutter database revealing
geographic distribution of natural (e.g. open, water forest) and
built-up features (man made like urban, suburban, village) impact
on wave propagation. Typical DTM and the land usage information
(clutter) are structured by raster data. A raster element (pixel)
represents the average terrain height above sea level or a clutter
code for a square area of, for example, 100m x 100m. Terrain
databases are created from topographical paper maps and/or from
satellite images and are available in different types of resolution
and number of clutter categories. The typical used terrain
databases are: y Countrywide data sets are recommended for
Macrocell coverage analysis in rural and semi -rural environments.
The database almost created from 1:100.000 to 250.000 topogra
phical paper maps, depending on country. Typical is a 5-class land
usage map. The typical resolution is from 50mx50m up to 100mx100m.
The planimetric accuracy is (x,y)=100m by 1:200.000 (50m by
1:100.000). The altimetric accuracy (z) is from 20m up to 40m by
1:200.000 (10 to 20m by 1:100.000), depending on the relief. City
packages or urban data sets should be used for the coverage
analysis in towns and cities. The typical resolution is 20mx20m.
The DTM created by digitization of 1:50.000 scale topographic al
maps. The clutter information based on 10m resolution Spot
satellite imagery. Typical is a 15-class land usage map. The
planimetric accuracy is (x,y): 20m. The altimetric accuracy (z) is
from 7m up to 12m.
y
Keep in mind, that: y y y y y 100-meter clutter databases are
created from topographical paper maps that are older than 3 years
(depending on the country) usually. Thus, the land usage
information is not up-to-date. Satellite images guarantee
up-to-date land usage information. Clutter databases created by a
satellite image do not guarantee reliable clutter information. The
clutter database is the interpretation of a person, who analyzes
groups or cluster of gray -pattern in the image and try to assign
to this group the most likely suitable clutter category. The
resolution of the terrain database is not a guarantee for the
quality of the database Terrain databases with more than 20 clutter
classes (recommended are 15 -clutter classes) can provide extra geo
information for the RF planer (e.g. traffic distribution) and
perhaps improve the coverage analysis. However, it requires more
survey data (see RFGuidelineCWMeasurements.doc) and complicates the
model tuning process. In city packages, the data supplier offers
extra clutter categories like MainRoads or OpenInUrban. These
classes are created by line data information. All pixels are
covered by the line data represents Roads or Streets will be
assigned to the clutter category OpenInUrban. In case of tuning
empirical models like COST231-Hata, do not insist on this extra geo
information, if possible. Experience has shown that it complicates
the tuning process and it didnt improve noticeable the coverage
analysis. Furthermore, clutter categories like urban or suburban
already characterize the local probability of building density and
roads in a certain area.
y
To avoid problems performing the model tuning process: y Check
the real planimetric accuracy by paper maps or by the survey
routes. If the shift is greater than the declared planimetric
accuracy, then in worst case the survey routes cover the wrong
clutter areas.
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 21 of 53
RF Guideline Propagation Model Tuning
y
Check the clutter assignment by high-resolution paper maps or by
the local knowledge of the country project teams.
5.2
CW Survey Data
CW field strength measurements are necessary in order to tune an
empirical propagation model. Increasing the number of survey data
improves the reliability of the tuned propagation model. However, a
tuned propagation model is only good as the quality of the input CW
survey data used to calibrate it. To collect a statistically
significant amount of accurate survey data take into account the
subjects as follows: y y GPS, D-GPS or any other kind of
positioning system should geographically reference the data points
along the survey route. To fulfil the sampling theorem with respect
to the Doppler shift (i.e. record at least two impulse responses
per wavelength), the field strength should be measured every half
wavelength ( /2) by triggering by the survey vehicle (e.g. wheel
triggering). To get of the long term fading, fast fluctuations have
to be filtered out. This is done by averaging the survey data over
a gliding window with a minimum length of 40 (Lee-criteria). The
maximum length is 200 . To achieve a representative set of data
collection the number of test site locations should be at least of
10. The test sites locations should be representative of the sites
in the planned or running mobile network (antenna heights and
environment surrounding the test site). The test site antenna
height depends on the average operating antenna heights of the RF
network in the future. For Microcells (or UMTS), the test site
antenna height should be 12,5m and 25m. For Macrocells the test
site antenna height should be 25m and 40m. For the different test
site antenna heights, the same measurement routes have to be
driven. For measurements within Macrocells the minimum length for
one survey route should be between 200 km and 300 km. For CW
measurements within Microcells the minimum length for one survey
route should be 100km The survey data should be evenly distributed
with respect to distance from the test site and distributed with
respect to the clutter categories that are used in the
topographical database. After averaging the survey data a minimum
of 300 data points per 1000m-distance interval and per clutter
category is recommended. Example: For the calibration of a
prediction model for suburban application and a planned prediction
radius of 10km, the minimum number of overall measured (and
averaged points) for the clutter category suburban should be 300
*(10000/1000) = 3000 points. Therefore, it is important to plan the
survey routes with help of digitized terrain database, especially
the land usage database. The survey routes should be zigzag or
stair routes to avoid street direction propagation relative to the
test site especially for urban (build up) environments and dense
vegetation areas. Indeed, guiding effects in dense urban street
canyons with LOS may lead to path loss values which are up to 4 dB
lower than under normal inner-city propagation conditions. Avoid
surveys on short-term conditions that could impact on wave
propagation like peak ours or the weather (thunderstorm or snow
fall) and the wetness of surroundings after rainfall. Furthermore,
keep in mind the impact on wave propagation on the seasons
(deciduous forest in winter).
y
y
y
y
y
y
y
A further guidance is given in the word document
RFGuidelineCWMeasurements.doc.Author: Doc-ID: Date: H.H. Rhrig
Lucent Technologies Proprietary RFET-QA-REP-00-010-V01.00 12
September 00 Use Pursuant to Company Instructions Revision: Page:
1.1 22 of 53
RF Guideline Propagation Model Tuning
5.35.3.1
Start Parameter Values of Propagation ModelAircom ASSET Standard
Macrocell Model
Aircom International ASSET supports a Standard Macrocell model
based on the Okumura-Hata and/or the COST-231-Hata model. These
classical field-strength prediction models were developed for large
radio cells, i.e. radio paths larger than 1km. Furthermore, the
models applied to large base station antenna heights, i.e. the base
antenna is installed considerable over the rooftops of the
surrounding buildings. The Aircom ASSET Standard Macrocell Model is
defined as follows: L(d) = k 1 + k 2lg(d) + k 3(Hms) + k 4lg(Hms) +
k 5lg(Heff) + k 6lg(Heff)lg(d) + k 7Ldiffn + C_Loss (5.3.1)
Where: d Hms Heff Ldiffn k1 k2 k3 , k 4 k5 k6 k7 C_Loss distance
from the base station to the mobile station [km] height of the
mobile station above ground [m]. effective base station antenna
height [m] diffraction loss calculated using either Epstein
Peterson or Deygout intercept, corresponds to a constant offset
slope adjustment coefficient correction factor used to take into
account the effective mobile antenna height Effective antenna
height gain. This is the multiplying factor for the log of the
effective antenna height LOG10(Heff) is the multiplying factor for
LOG 10(Heff)log(d) multiplying factor for the determined
diffraction loss clutter adjustment coefficient
Aircom Asset supports the knife-edge diffraction methods y y y y
Epstein-Peterson Bullington Deygout Japanese-Atlas
Aircom Asset supports the effective antenna height calculation
methods y y y y Absolute Average Relative Slope
Author: Doc-ID: Date:
H.H. Rhrig Lucent Technologies Proprietary
RFET-QA-REP-00-010-V01.00 12 September 00 Use Pursuant to Company
Instructions
Revision: Page:
1.1 23 of 53
RF Guideline Propagation Model Tuning
5.3.2
Classification of Hata Adjustment Coefficients to ASSET k
-parameter
Table 2 shows the classification of the
Okumura-Hata/COST-231-Hata adjustment coefficients to the
kparameters of the Aircom ASSET Standard Macrocell Model. The
Okumura-Hata and COST-231-Hata adjustment coefficients obtained
from the equations (3.1.1) and (3.2.1).Aircom ASSET k-parameter
Okumura-Hata Model150 MHz < f