504 IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 7, NO. 5, SEPTEMBER 2017
Fig. 2. (a) Three possible Kiosk downloading scenarios and (b) measurement
campaign: VNA with automatic stage surrounded by wave absorber [17].
Fig. 3. Antenna pattern: (a) 2-D view of vertical cut and (b) 3-D view.
by measurement of the S-parameters (S21) for varying distances
between Tx and Rx using an automatic translation stage.
In practice, different users may place their devices in different
positions, which not only results in various distances between
Tx and Rx, but also influences the angles of departure (AoDs)
arrival (AoAs) at Tx and Rx. However, due to constraints on the
experimental equipment, measurements at different angles have
not been realized. To extract the channel characteristics of the
scenarios s uch as multipath composition, angular profile and po-
larization i mpact for varying Tx and Rx positions, a 3-D RT sim-
ulator is calibrated based on the measurement results. Initially,
the highest reflection order considered by the simulator is set to
6. The antenna patterns of Tx and Rx are considered as well.
The metal plate is assumed to be a perfect dielectric conductor.
The parameters to be calibrated are the PET absorption loss A
pet
and its dielectric parameters
r
and δ. The initial PET absorption
loss is set to A
pet
= 2 dB and the initial dielectric parameters
are
r
=6.4,δ =0.1171. Fig. 4 compares the uncalibrated RT
result with the measurement for a snapshot of Scenario (i). The
delays of rays match well with the measurement, indicating the
accuracy of the geometric environment model. However, the
errors of the simulated amplitudes are larger than 5 dB. A ge-
Fig. 4. Comparison between uncalibrated RT result and measurement of a
snapshot of Scenario (i).
netic algorithm is applied to find the best set of A
pet
,
r
and δ for
the PET plate that minimize the mean absolute error of the power
delay profile. The calibration progress is shown in Fig. 5(a), the
algorithm converges after 40 runs, and the mean absolute error
of the power delay profile between calibrated results and mea-
surements is reduced from 5 to 0.99 dB. The calibrated material
parameters of the PET plate are A
pet
=1.97 dB,
r
=2.1899,
and δ =0.1187. The calibrated results of Scenario (i)–(iii) are
shown in Fig. 5(b) and 5(c), respectively. As can be seen from
the figures, the number of reflection rays varies among different
scenarios, due to the different alignments of the metal plates
and PET plate. The average RMS delay spread decreases from
Scenario (i) to (iii) (0.39, 0.22, and 0.12 ns, respectively), and
the average Rician K-factor in terms of rays increases from Sce-
nario (i) to (iii) (14.91, 22.13, and 26.07 dB, respectively). Thus,
the overall contribution of multipath components of Scenario (i)
is the highest and that of Scenario (iii) is the smallest among
the three scenarios. The average errors of RMS delay spreads
and Rician K-factors of the RT results are 0.02 ns and 0.05 dB,
respectively, indicating that the calibrated RT results match well
with the measurements in terms of the amplitudes and delays
of significant multipath components. Therefore, the calibrated
RT simulator is sufficiently accurate in describing the three po-
tential Kiosk downloading scenarios. In the following section,
RT simulation is used to overcome the limits of measurement
to analyze more characteristics of the Kiosk downloading prop-
agation channel, and RT results perform as reference data for
channel modeling.
III. C
HANNEL MODELING AND RESULTS
In this section, the calibrated RT simulator is used to run
intensive simulations for different scenario configurations. For
each of the three scenarios, the Tx is fixed, while the Rx position
varies within a 3-D space near the front cover, as depicted in
Fig. 6. The width and the height of the metal plate of Tx and
Rx are 0.2 m, respectively. The 3-D grid distance between each
neighboring Rx position is 0.04 m. In this paper, 576 snapshots
(Rx positions) are simulated for Scenario (i), 540 snapshots are
simulated for Scenario (ii), and 340 snapshots are simulated for
Scenario (iii). The total number of snapshots vary due to the con-
straint on the valid communication space in different scenarios.