Page 82 - Autonomous Mobile Robots

Millimeter Wave RADAR Power-Range Spectra Interpretation 65

range bin as shown in Equation (2.10) [27].

M

Z = x i (2.10)
i=1
This sum is multiplied by a threshold, τ (in this case τ = 0.033), for
later comparison with a test sample power value. The value for τ is chosen
for achieving the desired value of P fa , the design false alarm probability, in the
absenceoftargets[28]. Thescalar τ isafunctionofthenumberofreferencecells
M (here M = 70) and P fa is (1×10 −6 ) for the RADAR used here [10]. The test
sampleY iseitheranoise-plus-clutterobservationoratargetreturn. Thevariable
threshold τZ is compared with Y. A target is declared to be present if

Y >τZ (2.11)

The range bin in Figure 2.15 was obtained from an environment contain-
ing a concrete wall at approximately 18 m. The detected features are indicated
along with the adaptive threshold. The moving average will set the threshold
above which targets are considered detected. Due to the phase noise, the power
returned from the target is widened along the range axis, resulting in more
feature detections at approximately 18 m. In Figure 2.15a and b, CFAR “picks
out” features which lie at closest range. Features at a longer range, however,
will not be detected as the noise power variance estimate by the CFAR pro-
cessor becomes incorrect due to the range bin distortion caused by the high pass
filter.

2.5.1 The Effect of the High Pass Filter on CFAR
In general, since the gain of the high pass filter is not linear (Figure 2.6a) the sum
of the noisy received power values in Equation (2.10) is inaccurate at higher
ranges, which ultimately results in the missed detection of targets at these range
values. This is evident from Figure 2.15b where CFAR detects a feature (corner
reflector) at 10.25 m while it misses a feature (building) at 138 m. The second
reflection is due to the beam-width of the RADAR, as part of the transmitted
signal passes the corner reflector. It would therefore be useful to reduce the
power–range bias before applying the CFAR method. Therefore, to correctly
implement the CA-CFAR method here, first, the average of two noise only
6
range bins can be obtained, the result of which should be subtracted from the
range bin under consideration. This is carried out to obtain a range independent,
high pass filter gain for the resultant bin.
The CFAR method has been applied to the range bin of Figure 2.11b, the
full 200 m bin of which is shown in Figure 2.16a, after subtracting the high
6 The noise only range bins are obtained by pointing the RADAR toward open space.

© 2006 by Taylor & Francis Group, LLC

FRANKL: “dk6033_c002” — 2006/3/31 — 17:29 — page 65 — #25