Digital Signal Processing Reference
In-Depth Information
Fig. 4.3
RSSI measurements for an outdoor 802.11 antenna located on top of Cory Hall at UC
Berkeley. The
left image
shows the resulting contours for a given RSSI threshold, as we will com-
pute in this chapter. The two
right images
show measurement points and denoised RSSI signals
with 95% confidence intervals for two different slices. The horizontal
dashed line
corresponds to
the RSSI contour threshold on the
left image
. Note the high noise levels, the absence of a clear
trend and the very different results along the two directions
figure show pathloss measurements along the slices indicated on the left overview
image. Each of the continuous pathloss curves in Fig.
4.3
is obtained by computing
for each point along the horizontal axis a local moving average (details are discussed
in Sect.
4.3.2
). Along with the moving average the variance is computed, and hence
95% confidence intervals (assuming the error is normally distributed), as indicated
in the figures.
4.1.3.2 Observations from Measurements
From these measurements, the following observations can be made.
No Clear Trend:
As can be seen in the slices drawn from the measured signal
strength in Fig.
4.3
, it is not possible to find a clear trend in the pathloss as function
of distance. Reception can degrade significantly behind a building, and then improve
again at larger distances when the line-of-sight is established again.
Anisotropic Due to Shadowing:
As shown in Fig.
4.3
, the propagation trend
looks very different along the two different paths. It can hence be concluded that
it is not possible to fit a single trend to the measurements. Instead we need to fit
local trends at run-time, not relying on the assumption that a single model fits the
entire propagation surface properly.
Noisy Measurements Due to Fast Fading:
The RSSI sample points in Fig.
4.3
appear very noisy. A technique is needed to locally smooth out the noise and obtain
the solid curve along with associated confidence intervals.
The main conclusions drawn from the measurements are that:
(1) there is no clear global trend,
(2) propagation conditions vary significantly depending on the direction (since ob-
stacles vary depending on the exact path followed) and
(3) measurements are very noisy.
Clearly, it is very difficult to propose a single pathloss or shadowing model at design
time that would fit any run time condition. In this chapter a local channel estimation
