88 ◾ Advances in Communications-Based Train Control Systems
In Equation 5.1, the path loss exponent
has been obtained, and
is an
inherent parameter of one kind of leaky waveguide. Next, we will discuss the dis-
tribution model of small-scale fading.
5.4.3 Determination of the Small-Scale Fading
e statistical model of the small-scale fading needs to be determined to show how
much the signal level can vary on the basis of the large-scale fading. A number
of dierent distributions have been proposed for small-scale amplitude fading in
indoor and outdoor environments, including Rice, Rayleigh, Nakagami, Weibull,
and log-normal distributions. In order to determine the distribution of the small-
scale fading of the channel with leaky waveguide, we rst need to remove the eects
of the large-scale fading. And the length of small-scale area (SSA) should be deter-
mined as
to get enough data to estimate the model precisely for one SSA[8].
As the empirical decorrelation distance is
, we select one sample from the mea-
surement results in each
section. And the samples in dierent decorrelation
distance areas are spatial independent (uncorrelated).
We select ve kinds of classic distributions mentioned in the preceding text as
the candidate models of the small-scale fading for the channel with leaky wave-
guide. ese models are also used in [8], where the AICc is used. AICc is also called
the second-order AIC, which is dened as follows [9]:
AIC
AICc AIC
e
ij
n
N
i
ij
in j
ij ij
j
lx k
kk
,
=1
,
,
,,
2((|)) 2
2(
=− +
=+
∑
log
θ
jj
Nk
+
−−
1)
1
(5.6)
where:
means the
th SSA
j
means the
j
th candidate model
,
means the estimated parameters of the
th candidate model for the
th SSA
using the maximum likelihood estimator (MLE)
,
is the
th sample of the
th SSA
is the number of parameters of the
j
th candidate model
i
is the total number of samples of the
th SSA
For each SSA, we calculate the value of AICc of each candidate model and select the
best-t model with the lowest value of AICc. Figure 5.5 shows that the log-normal
distribution provides the best t in a majority of the cases, about
%. is is valid
for 20 measurements in Beijing Subway Yizhuang Line. erefore, we can see that the
log-normal model is the best parametric t to the distribution of the small-scale fading.
In addition, we show the plots of the empirical cumulative distribution func-
tions(CDFs) of measurement data and the candidate models in Figure5.6, where