Each - Draper Laboratory
Each - Draper Laboratory
Each - Draper Laboratory
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
North (m)<br />
Node 1 Node 2<br />
30<br />
25<br />
20<br />
15<br />
10<br />
5<br />
0<br />
x1t1spr<br />
0 10 20 30<br />
East (m)<br />
Figure 6. Conditions after particle filter initialization.<br />
Available measurements were processed every 0.5 s. Tag<br />
speed was held constant at 0.5 m/s. No dead reckoning<br />
sensors were employed, so that the geolocation estimates<br />
calculated by both filters were not propagated between<br />
measurements; however, the error covariance matrices<br />
were increased within both filters using (21). The process<br />
noise covariance matrix Q(i) = v(i)I was calculated using<br />
sequential differencing of the position estimates to estimate<br />
the variance v(i).<br />
The simulation was run for 94 s at a time step of 0.5 s.<br />
The time delay (in meters) for the direct and indirect paths<br />
are plotted in Figure 7. The two indirect paths from Node<br />
1 have a single crossover point at 20 s. The two indirect<br />
paths from Node 2 have a single crossover point at 70 s,<br />
with a near-crossover at 17 s. The data association algorithm<br />
given in the previous section was employed using<br />
quadratic regression models and produced no data association<br />
errors.<br />
The true and estimated paths over time for both filters<br />
are shown in Figure 8. True tag location is shown by the<br />
solid black line. The estimated path for the multipath filter<br />
(MP) is shown by the solid colored line, while the estimated<br />
path for the conventional filter (CV) is shown by the<br />
dotted colored line. While both direct paths are detected<br />
(for the first 55 s), the MP filter and the CV filter produce<br />
identical geolocation estimates (blue line). After the direct<br />
path from Node 2 is lost at 55.5 s, the CV filter is able to<br />
navigate off the direct path from Node 1 only, while the MP<br />
filter, in addition, is able to navigate off the indirect path<br />
from Node 1 reflected off the bottom wall and the indirect<br />
path from Node 2 reflected off the West wall. The MP<br />
filter estimate (solid red line) produces very small tracking<br />
errors, while the CV filter errors (dotted silver line) start to<br />
grow. When both direct paths become undetected at 73.5<br />
s, the CV filter can no longer track at all; its geolocation<br />
estimate remains constant for the remainder of the simulation.<br />
In comparison, the MP filter is able to navigate off the<br />
10 Innovative Indoor Geolocation Using RF Multipath Diversity<br />
m<br />
m<br />
North (m)<br />
70<br />
60<br />
50<br />
40<br />
30<br />
20<br />
Node 1<br />
10<br />
0 20 40 60 80 100<br />
60<br />
50<br />
40<br />
30<br />
20<br />
Time (s)<br />
x1dy<br />
10<br />
0 20 40 60 80 100<br />
30<br />
25<br />
20<br />
15<br />
10<br />
5<br />
0<br />
Node 2<br />
Time (s)<br />
Figure 7. Measurement delay vs. time.<br />
Node 1 Node 2<br />
x1tag<br />
0 10 20 30<br />
East (m)<br />
Figure 8. Comparison of true and estimated paths.<br />
detected indirect paths. Between 73.5 and 77.5 s, the MP<br />
filter navigates off the indirect path from Node 1 reflected<br />
off the bottom wall and both indirect paths from Node 2.<br />
At 78 s, the indirect path from Node 2 reflected from the<br />
West wall becomes undetected, and the MP filter is reduced<br />
to using both indirect path measurements off the bottom<br />
wall. At 84 s, all four indirect paths become detectable and<br />
are used by the MP filter until the end of the simulation.