Page 258 - Integrated Wireless Propagation Models

236 C h a p t e r F o u r

For all the matrixes for each case or scenario we list above, we can simply follow the
JLO procedure to fine-tune the model. In this approach, the values of the model param­
eters are obtained through minimization of the cost function given by

(4.4.2.2.5)

where

(4.4.2.2.6)
The JLO method solves for all parameters of the model simultaneously and gener­
ally provides the best fit to the measured data. However, although numerically optimal,
the values for the parameters may not be physically intuitive and may vary substan­
tially from cell to cell.

4.4.3 Verification of the Lee Model
The work of fine-tuning the Lee model was published by Kostanic et al.9 as described in
Sec. 4.4.2 and the result is discussed here. Kostanic used a different approach to integrate
the measured data with the Lee microcell model. The Lee rnicrocell model and tuning
techniques have been implemented by the WIZARD RF propagation software tool of
1
Agilent technologies. 9
Applying the Lee microcell model and the JFO tuning method, the range of stan­
dard deviations was shown to be between 5.3 and 6.6 dB. These low values from the
measurement errors rank better than the typical standard deviations observed in dense
urban environments, which are often as high as 8 to 10 dB.
Another illustration of the accuracy of the model is presented in Figs. 4.4.3.1 and 4.4.3.2.
The predictions are calculated using only initial default rnicrocell propagation parameters
and no building losses. These predictions are used for comparison with the predictions
from the tuned model. Figure 4.4.3.1 shows a scatter plot of predicted versus measured
points for the default (no building losses added) and the tuned models. The ideal zero­
mean, no-scatter line is shown as reference. The default model underpredicts by 18 dB
when compared with the measured data. As shown in Fig. 4.2.1.2.2, the building loss L8
remains constant as 18 dB after a building block exceeds 250 ft. The basic Lee rnicrocell
prediction model in Sec. 4.2 describes the calculation of prediction in microcell environ­
ments by adding the building block loss to the default LOS loss. Therefore, by adding 18 dB
to the prediction of the default model if the building block data are not available, the results
are the same as the prediction from the tuned model, as shown in Fig. 4.4.3.1. The tuned
model introduced by the JLO tuning technique also shows the tight fit to the reference line.
Both predictions are fairly accurate, but the basic Lee rnicrocell model is much simpler to
use than the tuned Lee model by using JLO procedures. Although the prediction from the
default model plus the loss due to the building block curve is just as accurate as from the
tuned model, it may not have as sound a calculation as the tuned model does. However,
the tuned model proves that the default model plus the building block curve, which is the
basic Lee rnicrocell model, is the right approach to making the prediction.
A similar comparison can be observed from the histogram of measurement errors
shown in Fig. 4.4.3.2. Histograms for both the default model and the tuned Lee micro­
cell model are shown. The difference of 20 dB between two prediction results is the loss
due to building blocks, which the default model does not include. Removal of the bias