Digital Signal Processing Reference
In-Depth Information
Ta b l e 4 . 1
Performance of the algorithm for the examples of Fig.
4.7
Number of
updates
Percentage
misclassified
Example 1 (710 nodes)
primary sender
724
0
.
28
secondary sender
721
0
.
14
secondary sender update 1
722
0
.
98
secondary sender update 2
714
1
.
12
Example 2 (723 nodes)
primary sender
732
0
.
83
secondary sender
730
0
.
14
secondary sender update 1
732
1
.
24
secondary sender update 2
732
2
.
21
Example 3 (712 nodes)
primary sender
721
0
.
70
secondary sender
723
0
secondary sender update 1
722
0
.
14
secondary sender update 2
719
0
.
98
the middle of the building, the whole zone is obstructed, leading to larger losses. In
this simulation study, the model is slightly adapted to allow received signals inside
buildings. The exact form of the model is not that important since our technique
should work for any possible situation. The resulting shadowing loss is
max
0
,
5
.
125
2
,
−
x
L
−
x
O
D/
2
L
S
=
×
(4.8)
where
x
O
is the position of the obstacle,
D
its width and
x
L
the orthogonal pro-
jection of
x
O
on the ray. The fraction
x
L
−
x
O
D/
2
tells us how close to the center of
the building the line-of-sight ray crosses. When the line-of-sight ray crosses the
building through the center, we have a maximal
L
S
of 10.25 dB per obstacle which
is a value chosen from [51]. From the center of the building, the shadowing loss
decreases linearly until 0 dB.
4.3.5.2 Results and Discussion
The resulting iterative power adjustment is illustrated for 3 examples in each column
of Fig.
4.7
. The primary sender's contour approximation is drawn using a solid line.
The real contour is drawn as the dashed line. The computed minimal contour-to-
contour distance is indicated by the straight line connecting the two corresponding
nodes. The buildings are drawn as gray boxes. The simulation model as described
above was used, with a noise power of 4 corresponding to the real-world measure-
ments. We used quadratic MLS with a kernel width of 95 m in all examples.
As can be seen in all images, the computed contours approximate the exact ones
very well: Only 0.7% on average of all nodes was misclassified. Note that these
