Professional Documents
Culture Documents
1 Introduction
This document has been completely orientated on the previous Model Tuning Concepts for
the former Tornado (Planet – MSI). It has been found that Model Calibration process can
exactly be interpreted as the same, where main efforts were seen in the transformation of the
already found standard model parameters. Since their values and algorithms are correlated to
initial rich experience with Tornado, further new algorithms and parameters in TornadoN
have to be carefully and consciously applied. However enough information is provided for a
successful empirical propagation model evaluation with the right/sufficient set of
usable/practicable parameters and algorithms.
The propagation models are the main aim to adapt radio attenuation in the real field. The
tuning of its certain parameters can classify different environments like urban or rural areas.
Therefore, different statistical formulas are existing in order to investigate appropriate
scenarios with a high reliability. In this document the tuning process shall be analysed on
macro- (Tx-Antenna height above roof) build-up scenarios. Finally detailed concepts for
model tuning, resulting in reliable radio predictions, are given.
The following pages are explaining the most known and practicable radio propagation
formulas for build-up area, Okumura-Hata formula for macro environments. Additionally
their implementations within TornadoN (Aircom Enterprise – ASSET) and the parameters
correspondence are clarified. Chapter three concerns with concepts for tuning the mentioned
models. The last chapter four indicates the main factors of experienced models and therefore
can recommend appropriate models to certain scenarios, including its starting values.
It should be noticed that the used terrain databases have the recommended typical raster
resolutions for macro cell, 20m -for urban areas-, 50m, 100m (depending on the roughness of
the regional surface) -for regional areas-, and for micro cells 5m. Since the Okumura-Hata
formula is a pure statistical approach for predictions of radio propagation behaviour, the usage
of finer (macro scenario) raster resolutions for their certain application areas are not justified
by the propagation models.
The Hata radio propagation formula is based on the Okumura measurements in the urban area
of Tokyo at frequencies of up to a maximum of 1500 MHz. This formula applies to flat urban
areas. Correction terms are specified by Hata for suburban and open areas and are represented
by L(Clutter).
f hBS h d
L pathloss 69.55 26.16 log 13.82 log a MS s * log L(clutter)
MHz m m km
h
s 44.9 6.55 log BS
m
The formula had to be modified slightly to remain applicable in the frequency range of
1800MHz. The result of this modification was the COST-231 Hata model.
f hBS h d
L pathloss 46.3 33.9 log 13.82 log a MS s * log L(clutter)
MHz m m km
h
s 44.9 6.55 log BS
m
These two approaches form a usable basis for most 900 and 1800 applications on level terrain.
The pathloss coefficients of the Hata model have been converted for a flexible use within the
TornadoN “Standard Macrocell” model. Changes to the basic parameters for these equations
should be kept to a minimum.
where :
The losses due to obstruction by terrain obstacles are determined by extension of the model
for knife-edge diffraction. They can be weighed with a selectable factor K7. Multiple
diffraction caused by adjacent obstacles would produce high overall attenuation. This effect
can be limited by merging several obstacles within a selectable distance.
K(clutter) is called clutter gain and represents a fixed power correction for environments
which differ from the one assumed by Hata situated in Tokyo (urban). The clutter gain has to
be defined for each clutter that exists in the terrain database. This value is treated as the GAIN
of the individual clutter class with respect to the reference of Tokyo-City. Obviously, the
model can be referred to a specific clutter, whose clutter gain will be put at zero, while each
other clutter correction factor will be estimated with respect to it.
If the original Hata-formula -COST231 respectively- is compared with the in TORNADO
implemented standard model, the following terms can be corresponded to the different K-
factors (without K4, K7).
f d h h d h
L pathloss 69.55 26.16 log 44.9 log 13.82 log BS 6.55 log BS * log a MS L(clutter)
MHz km m m km m
K1 K2 K5 K6 K3 K(Clutter)
NOTE: With respect to the changed distance-unit within TorndaoN which is kilometer (km),
the formula has NOT to be converted. Previously in the old Tornado, additional terms
regarding the K1 and K5 factor had to be included.
The TornadoN tools offer several approaches to building prediction models for the signal
level and for the field strength distribution to be expected in a planning area. This guide is
limited to the Standard Macrocell Models which are the most manageable methods for the
planner, and which can be used in practice in all occurring scenarios. This description does
not cover all of the options for model calibration, especially offered in the Standard Model.
The various possible presentation methods may have effects which are to a certain extent
contrary or partially counteractive or accumulative. These dependencies are described here.
When modelling is practised on a multilevel basis involving the simultaneous use of several
parameters with different effects on the prediction (e.g. simultaneous use of "clutter factor"
and "clutter height"), the overall behaviour of the model quickly becomes difficult to assess.
All terrain have particular features with regard to radio attenuation. It is recommended, that an
area visit/survey is proceeded by the engineer responsible for adjustments of the prediction
model, even though radio measurements are available. While tuning, the engineer will also
remember the peculiarities of the area and be in a better position to explain any deviations in
the measured results. Radio planners should avoid creating models which are so complex that
their behaviour is difficult or impossible to follow. It is a fallacy to assume that the quality of
a prediction increases with the number of tuning parameters. Although it is certainly possible
to emulate the measured result of a specific survey with great precision by activating several
variables, it should not be overlooked that the aim of the model calibration is not to reproduce
individual measurement routes as exactly as possible, but to elaborate the overall radio
characteristics of an area. The resulting model should retain its informative value in other,
comparable areas for which no measurements are available. This goal is best achieved by
evaluating as many surveys which are typical of the area as possible and hence can be re-used
with constant high accuracy.
One extension of the Hata-formula concerns heavy signal fluctuations near the site, caused by
LOS and NLOS paths, make the average field strength rise over the values predicted by
models tuned for higher distances, where a lower occurrence of LOS paths makes signal
variability be more restrained. “Two-piece” models were defined to take into account this
effect; the global curve is composed by two pieces with different slopes (signal VS distance),
joined at a certain distance (generally 1 km, that is the lower limit for Hata model validity).
A plausible approach to modeling the near section would be to accept the free space loss at a
distance of 20m and the attenuation calculated with the Hata formula at a distance of 1 km.
This results in the following slope within the near field up to a transition break point
dBP=1km.:
NOTE: When designing a 2-piece model, the planner must make sure that there is no
attenuation step at the transition point. The K1n factor must therefore be adapted accordingly.
Path-
Loss
K2far
K1far
K2near
K1near
dBP log d
Another approach to describe the propagation area is, to use for each clutter class the height
and separation parameter INSTEAD of its correction factor. This could be very accurate
especially in urban areas where diffraction is the dominant attenuation path. As clutter heights
extends the terrain height and is determining the obstruction loss, the clutter separation factor
separates the mobile from the surrounding clutters and hence prevents from high losses by
adjusting the heights for use in the diffraction calculation (K7*D). In this context a lower K1-
factor is needed to decrease the propagation curve with the higher diffraction loss which is
especially excessive in the near field.
Note: With the introduction of clutter heights, the propagation model becomes a more
deterministic method in comparison to the pure empirical Okumura-Hata formula. Originally
no diffraction losses were involved and only two statistical clutter classes (suburban and
open) were evaluated.
A decision between these two clutter tuning-concepts shall be made in accordance to the
following aspect. The height and separation concept have to be seen as an alternative, since
the most experience is existing on tuning the correction factors which can be applied on
similar environments. The main background is that the tuning of heights and separation
parameters requires more detailed and accurate/reliable terrain data (height and clutter) to
guaranty slight deviations between the average parameters of each clutter and the realistic
measures of the dominant obstacles within a clutter area. A high resolution terrain database
like 20m for urban areas, results in a lot of morphological clutter information. Hence if
models of these two concepts are compared, the model tuned by the height and separation
clutter parameters results in a less prediction error of course, but corresponds only to the
certain clutter area and finally can not probably be applied on similar clutter scenarios.
However in case the coverage area is affected by morphological obstacles situated in a clutter
class between transmitter and receiver, the tuning of the clutter factors can not take into
account such propagation characteristic. Even if the path clutter algorithm is enabled, the
investigated parameters of path clutter would only be reliable if a lot of measurement data
have been evaluated. The diffraction algorithm on the other hand also includes obstruction
losses from each morphological obstacle encountered on the propagation path, provided that
the height factor is specified on every clutter class.
The terrain profile (topography) can also be specified as itself on a model for a certain area. It
is considered on the path loss calculation by evaluating the effective antenna height between
transmitter and receiver. There are several algorithms in the standard model for determining
the effective antenna height. In each case the best-suited method must be determined from the
prevailing terrain profile. If only one specific road needs to be covered in hilly terrain, the
choice may be different to that of a whole area.
ATTENTION: The model to be tuned in TornadoN is that one named standard macrocell
model 3: user reference guide strongly recommends to use it, because it performs more
Harsha Nath Jha SaMax Nepal Pvt. Ltd. Samax.nepal@gmail.com
accurate antenna masking, whereas models 1 and 2 are included only for cases when
upgrading and re-tuning are not desired.
The receive signal level reveals a long term fading characteristic with normal distribution,
even in areas with a very homogeneous building character. The typical standard deviation for
this spread is approx. 7dB in urban areas. The level predicted with TornadoN is the mean
level of the signal. The statistic evaluation of the prediction error with TornadoN generates
the mean deviation (mean error = predicted path loss - measured pathloss)) and the quadratic
mean of the error (RMS value). There is no point in attempting to use all available means to
force the RMS value under the value of the typical standard deviation. In practical terms, a
RMS error of around 8dB can be considered as a very good tuning result. For the mean error
you should aim for a slightly positive value (approx. +3dB...+2dB). This would provide for a
slight safety margin in the predicted “pessimistic” value.
A very important parameter is the error regression curve slope, because the closer to zero it is
the more faithfully it simulates the actual field strength decay with distance from site. The
improve of coherence between the model slope and the actual power attenuation with distance
from radio base station let it be possible to deduce cell dimensions with little error margin
and, as a consequence, the number of predicted cells necessary to cover an area will be more
reliable. A slight positive sign of the error regression curve slope (max +10 dB) also refers to
a healthy pessimistic approach in order to give a higher probability in overlapping cells than
being below the predicted signal (at a lower probability) resulting in coverage gaps.
Model performance analysis, when applying 2-piece model feature, shall consequently be
seen before (near) and after (far) transition point. Their application area, nearfield as the
coverage representative and farfield as the interferer representative, then can individually be
evaluated.
The following picture of model and measurement data analysis has been taken from the visual
feature in former Tornado Survey Tool. Since here, received signal strengths are compared
(predicted – measured) inverted signs on the model performance indicators (regression slope,
mean value of error, both with negative values) were main target for a slight pessimistic
propagation model.
+RMS
Mean Error
-RMS
Standard
Deviation
Signal Regression
= Slope
- Remove deletion of LOS- and NLOS points: Check the analysis result for the entire
measurement route (except for the dubious route sections or where the recorded level is
drowned in the receiver noise)
Calibration procedure for clutter -heights and -separation concept slightly changes in Model
tuning because of extending terrain height with clutter height and its inclusion together with
the separation parameter to calculate the diffraction loss. In this context it is really
recommended to accurately investigate the environment for average clutter height
assignments. The model tuning of an area with several clutter classes have to be seen as an
iterative process, since changes of each clutter parameters on the propagation path influence
the diffraction calculation. As clutter height parameter is representing the real coverage area,
the clutter separation is the tuning parameter which can characterise the height distribution
within a clutter class. The option in the survey tool to distinguish between LOS and NLOS-
measurement data now involves the clutter height. However when evaluating LOS-data
(delete obstructed) fine tuning of K2-factor shall be done at the end, since the K1-factor
definitely changes when tuning the parameters for diffraction loss of the obstructed data.
The important steps of model calibration using clutter height and separation concept can be
seen as follows. (Replaces the certain steps (italic script) of the model calibration with clutter
factors)
- In the start model, average clutter heights and clutter separation have to be set instead of
clutter factors (set to zero!).
- Consider NLOS data (delete non-obstructed)
K7 -Tuning and K1-Alignment; changes of several clutter separation in accordance to the
median error of the certain clutter class. This will result in adapting its impact on
separation parameters of following (away from Tx) clutter classes and the K1-factor.
Afterwards K4-Tuning (K1-Adaption) with respect to error regression curve.
- Consider LOS data (delete obstructed)
Slight tuning of K1/K2-factor
- For fine-tuning the mentioned process (bold steps) perhaps have to be repeated
NOTE:
Due to the fact that K1 definitely raises/lowers predicted path loss, K1-key values shall be
found for certain building environments. Evaluated clutter diffraction characteristics (height
and separation) shall be analysed to achieve experience for its application on similar building
structures or clutter classes or even build up scenarios.
The lower table summarises pros and cons of the two model tuning concepts. Additionally
their recommended application fields are also mentioned.
The “Through-Clutter loss” in TornadoN option allows to take account of clutter not only in
the predicted pixel, but also in those pixels interposed between the transmitter and the
receiver. The clutter loss for a predicted pixel is obtained by adding to the clutter offset (offset
loss) the loss due to the clutter lying between the base station and the mobile station
(Through-Clutter loss in dB/km), included inside a certain distance starting from the predicted
point (this distance is named dthrough in TornadoN): a variable weight is associated, in
TornadoN, to the clutter lying along dthrough, linearly distributed from 1 (the predicted point) to
0 (the point at dthrough distance from the predicted one), whereas Tornado old allows to define
more weight distribution laws.
TCLossi i
Total Loss Offset Loss
i 1 K N Loss in dtthrough
The offset loss, in Tornado old approach, is conceived as the loss due to a certain clutter when
no other clutter is interposed between the transmitter and the receiver: path clutter option is
used in order to take account of effect of interposed different environment classes, but this
effect is null if the interposed clutter is the same as in the prediction pixel.
On the contrary, in TornadoN, the through-clutter loss has effect even if the same clutter is
present along dthrough; so in order to maintain a behaviour similar to that one implemented in
Tornado old (more logical), if the through-clutter option is enabled, it is recommended to set
the through-clutter loss equal to the offset loss, and than to modify the offset loss by
subtracting from it the through-clutter loss calculated along dthrough (the part of the
previous equation inside the frame).
Obviously, when through-clutter option is disabled, the original configuration must be
resumed (offset loss must be set equal to its original calibrated value, and dthrough and through-
clutter loss must be set to zero).
Example:
1 2 3 4
Tx Rx
dthrough (4 pixels)
Recommended operations:
Harsha Nath Jha SaMax Nepal Pvt. Ltd. Samax.nepal@gmail.com
1) through-clutter loss = 10 dB (equal to the offset loss)
2) new offset loss =
10 – [(10/K)*(1/N) + (10/K)*(2/N) + (10/K)*(3/N) + (10/K)*(4/N)] = 8.75 dB
Why this? Because in this way, if path clutter is enabled both in Tornado and
in TornadoN, the same final clutter loss (offset + through-clutter) is obtained
when one single clutter is found along dthrough.
The following pages concern with experienced models which contain evaluated and even
practicable values for each parameter of the macro propagation models described before.
These should help the planner to have first startup values for the initial process of tuning a
model and therefore definitely shorten the model tuning progress. Again it is strictly
recommended that the practicable value ranges shall be kept in order to gain reliable model in
context with radio planning aspects and finally not to reach the lowest prediction error by
mathematical sophisticated methods.
The K1-factor for 1800 MHz is extended by additional 3 dB with respect to COST231
recommendation for metropolitan areas. The K5 and K6 parameters, which represent the
weighting of the BS antenna height and its effect on the slope in the Hata formula, should be
changed as little as possible, since it is otherwise difficult to apply the model to other areas.
K7-factor is only expected a percentage of the diffraction loss, since diffraction algorithm is
Epstein-Peterson which calculates high losses. Initial test results indicate K3, K4 factor for
pathloss influence of the mobile antenna height are advisably set to zero.
The near field parameter K2n (and K1n respectively) is height dependent where the value
changes with different BTS height. It is not practicable to have a different model for each Site
and therefore these parameters correspond to a reference height of 35m.
The following table contains of all K-factors applying on 900 and 1800 startup models. Find
before the parameters conversion rule from former Tornado to TornadoN :
Tornado TornadoN
K1T K1TN = - (K1T + 3*K2T)
K2T K2TN = - K2T
K3T K3TN = - K6T
K4T K4TN = 0
K5T K5TN = - (K3T + 3*K5T)
K6T K6TN = - K5T
K7T K7TN = K4T
Hata specifies correction functions for the clutter types “suburban” and “open area”. Based on
the basic equations listed in 3.1, these have the following values in dB:
Because of the very approximate graduation and the limitation to 2 categories, these can only
be typical values. Even the basic value Urban = 0dB can apply only to a very specific building
development type (in this case, Tokyo). In principle, all intermediate values are possible,
depending on the surface and building structures. In Central Europe values of 0 to around
12dB are feasible for urban areas. The prediction results in clutter types “open area” should
undergo a critical verification. The related level supplement would in regions closer than
about 500m from the base station predict a path loss lower than free space attenuation
The following table provides the experienced values/ranges for typical clutter classifications
for 900MHz and 1800MHz. The clutter coefficients conversion rule for TornadoN is the
following: KcTN = - KcT
TornadoN:
Example 900 MHz 1800MHz
Dense urban >10 floors, narrow streets 0 0
Harsha Nath Jha SaMax Nepal Pvt. Ltd. Samax.nepal@gmail.com
(type Tokyo)
Urban 6...7 floors, streets app. 20m -3 -3
(type Munich centre)
Suburban 4...5 floors -(8..9) -(11..15)
(medium wide streets)
3...4 floors -(9..11) -(13..17)
(open suburbs)
Industrial (heavy buildings) -(8..10) -(12..15)
(open type, light buildings) -(13..15) -(17..20)
Village open type, Bavarian -(14..15) -(18..20)
Forest Tropical rain forest, 30m -(5..8) -(6..10)
Dense deciduous forest -(10..12) -(13..15)
Light pine forest -(15..17) -(18..20)
Quasi open Rural, single houses -(18..22) -(20..25)
Open (Heath) -(23..25) -(25..30)
(Sahara) -28 -35
Water Sea, Lake -28 -35
In case clutter heights are used, “clutter factors” have to be set to zero. Regarding clutter
separation, the value shall be taken equal to half the average street width in the corresponding
clutter class; in this way, the nearest obstacle (the first building encountered looking from the
receiver towards the transmitting site) is that located on the side of the street itself. So, no
obstructing edge is discarded, and an accurate description of diffraction can be obtained.
Please notice the following list of differences between through-clutter feature in TornadoN
and path clutter feature in Tornado:
For disabling set: Through Clutter Loss Distance = Through Clutter Loss = 0
Diffraction Algorithm
The following table shows the algorithms for diffraction calculation given in TornadoN, a
comparison with former Tornado is useful. The Epstein-Peterson method shall be the standard
set algorithm for diffraction calculation.
ATTENTION:
There is a discrepancy between the diffraction loss calculated by the Profile feature and the
value mapped in the coverage prediction: the Profile feature shows the correct diffraction loss
computed by the application of the algorithm reported at page B-10 of ASSET User
Reference Guide, whereas for the mapped value the model doesn’t consider diffraction losses
unless obstacles block the line of sight. In other words, in the coverage prediction the
diffraction loss is taken into account only if at least half Fresnel zone is obstructed, otherwise
the diffraction loss is set at zero !
Furthermore, concentric rings due to Earth curvature are created: this effect takes place when
large prediction is calculated on totally flat (or with very low slope) terrain, because the
terrain profile algorithm rounds heights to the nearest meter, producing a profile that falls
away from the Tx (because of Earth curvature) in little steps rather than smoothly. A knife
edge is created whenever the mobile is behind one of these little steps.
1 Tornado does not apply the algorithm Spot described in its user reference guide
(which neglects mobile antenna height), but the algorithm Relative described in
TornadoN manual (which considers also mobile antenna height). On the contrary,
Enterprise does not apply the algorithm Relative described in its reference guide, but
the algorithm Spot described in Tornado manual.
3 The different prediction and the effect of concentric rings is due to the same reason
described for diffraction.
ATTENTION: for large prediction, also Earth curvature is considered as increasing Heff.
If the “Slope height” method is selected, it should be noted that the terrain profile has a strong
effect on the effective height determined for each area element. Extreme values must be
avoided by choosing suitable limits (e.g. 30m...300m). The choice of the correct “slope
distance ds” is also very critical (approx. 1000m ). Great care is particularly important if the
model is to be used in other areas.
On wide (and even inclined) plains the “Absolute” in TornadoN method is the most suitable.
“Relative” should preferably be used for slightly uneven terrain with a higher base station
location. The angle of incidence near the mobile should not be too small. “Slope height” is
suited to locations in a geographic basin or on a central mountain midst a plain, even if the
incidence is glancing