Content uploaded by Jose-Maria Molina-Garcia-Pardo
Author content
All content in this area was uploaded by Jose-Maria Molina-Garcia-Pardo on Jun 29, 2015
Content may be subject to copyright.
IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS, VOL. 13, 2014 1047
Deterministic and Experimental Indoor mmW
Channel Modeling
Maria-Teresa Martinez-Ingles, Davy P. Gaillot, Juan Pascual-Garcia, Jose-Maria Molina-Garcia-Pardo,
Martine Lienard, and José-Víctor Rodríguez
Abstract—This letter presents an extensive multidimensional
analysis of line-of-sight (LOS) experimental data and simula-
tions at 60 GHz over a 9-GHz bandwidth. Numerical versions of
the measured multiple-input–multiple-output (MIMO) channel
transfer functions were obtained with a ray-tracing engine that
includes single-order diffuse scattering. The received power, RMS
delay spread (DS), and maximum excess delay (MED) computed
from both measured and simulated data indicate that diffuse
scattering improves ray-tracing-based modeling. Moreover, the
multipath components (MPCs) extracted from both sets of data
using the high-resolution estimator RiMAX were statistically com-
pared. The analysis of the results shows that even a raw description
of the environment can be used to predict millimeter-wave (mmW)
propagation with ray tracing.
Index Terms—Channel modeling, millimeter-wave (mmW), ray
tracing, RiMAX.
I. INTRODUCTION
FUTURE wireless communications systems have envi-
sioned the millimeter-wave (mmW) frequency band as a
promising response to overcome the Gbps barrier [1]. To this
end, two wireless medium access controls and physical layers
have been proposed by IEEE, one for personal area networks
(802.15.3c [2]) and another for wireless LAN (802.11ad [3]).
In [2], it is mentioned that significant efforts were carried out to
develop models as realistic as possible. However, the number
of available measurements and related data in the 57–64-GHz
range, from which the model was based, was insufficient to
fully characterize the underlying environments. Some recent
papers, such as [4], present clustering results for a double-di-
rectional 60-GHz multiple-input–multiple-output (MIMO)
channel model. The authors of [5] provide a deep review of
frequency-domain measurement results selected from major
research teams.
Manuscript received March 07, 2014; revised April 24, 2014 and April
24, 2014; accepted May 24, 2014. Date of publication May 29, 2014; date
of current version June 11, 2014. This work was supported by MINECO,
Spain, under Grant TEC2010-20841-C04-03, the European FEDER funds,
and STSM under Grant COST IC-1004. (Corresponding author: Jose-Maria
Molina-Garcia-Pardo.)
M.-T. Martinez-Inglés, J. Pascual-Garcia, J.-M. Molina-Garcia-Pardo,
and J.-V. Rodríguez are with the Departamento Tecnologías de la Informa-
ción y las Comunicaciones, Universidad Politécnica de Cartagena, Murcia
30202, Spain (e-mail: mteresa.martinez@upct.es; juan.pascual@upct.es;
josemaria.molina@upct.es; jvictor.rodriguez@upct.es).
D. P. Gaillot and M. Lienard are with the IEMN/TELICE, Electronics De-
partment, University of Lille 1, 59655 Villeneuve d’Ascq, France (e-mail: davy.
gaillot@univ-lille1.fr; martine.lienard@univ-lille1.fr).
Color versions of one or more of the figures in this letter are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/LAWP.2014.2327054
On the other hand, deterministic field prediction methods are
widely used for estimating essential radio channel characteris-
tics [6], [7]. Nonetheless, as far as the authors are concerned, a
comprehensive comparison of ray tracing simulations including
diffuse scattering with measurements in the mmW frequency
band is clearly missing. Only in [8] can such an approach be
found, where the authors developed a point cloud-based full dif-
fuse propagation prediction method. However, the overall field
is only described as fully diffuse backscattered from the point
cloud measured by a laser device.
In this letter, an extensive measurement campaign has been
carried out to measure the MIMO channel transfer functions
at 60 GHz in an office. The transmitting array was moved
over 20 line-of-sight (LOS) positions, whereas the receiving
array stayed at the same position. Additionally, all MIMO
channels were simulated by using a ray-tracing engine that
implements single-order diffuse scattering [9] from which the
received power, RMS delay spread (DS), and excess delay were
compared to the measured channels. Also, and for the sake of
comparison, the high-resolution algorithm RiMAX [10] was
applied to both data (experiment and simulation) to extract the
geometrical parameters of the multipath components (MPCs).
Finally, the arrival and departure angular spreads were com-
puted to compare the measured and simulated MPCs.
II. SCENARIO AND MEASUREMENTS/SIMULATIONS
A. Scenario
The measurement scenario is a laboratory located on the
first floor of the Universidad Politécnica de Cartagena research
building (Spain). The 4.5 72.5-m laboratory is furnished
with several closets, shelves, desktops, and chairs. In addition,
the laboratory is equipped with several computers and elec-
tronic devices. The walls are typical plasterboard walls, and
the floor and ceiling are made of concrete. In Fig. 1, a top view
of the measured scenario is depicted, as well as the measured
positions. Twenty separate transmitter (Tx) locations and one
receiver position (Rx) were considered for this study.
For all positions, a 0.5-m and 1-m distance was selected be-
tween each Tx row and column, respectively. All distances were
measured with a laser to obtain the most accurate precision pos-
sible. It is noteworthy that an LOS existed for all positions.
B. Measurements
The measurements were conducted using a Rohde &
Schwartz ZVA67 vector network analyzer (VNA). The mea-
sured frequency range was 57–66 GHz using 4096 frequency
points. A 10-Hz intermediate frequency was selected, and a dy-
namic range of more than 100 dB was obtained. Two amplifiers
1536-1225 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1048 IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS, VOL. 13, 2014
Fig. 1. Indoor scenario.
were used in the transmission to compensate for the attenuation
of the cables. The system is THROUGH calibrated to eliminate
the effect of cables and amplifiers. Both Tx and Rx antennas
are vertically polarized antennas (Q-par QOM55–65 VRA)
with 4.5 dBi gain. The antennas have omnidirectional patterns
in the H-plane and 40 ,28 , and 21 beamwidth at half-power
centered in the E-plane at 55, 60, and 65 GHz, respectively.
The height of the transmitting antenna was 1.44 and 1.54 m for
the receiving antenna. A virtual uniform linear array (ULA)
with five positions is used for Rx ( -axis orientation), whereas
a6 6 virtual uniform rectangular array (URA) parallel to
Rx was used for Tx, resulting in 180 possible channels. The
interelement distance was set to 2 mm for both arrays.
C. Simulations
Asimplified, yet faithful, numerical model of the scenario
has been developed with the main furniture. Suitable permit-
tivity and conductivity values, constant through the whole band-
width, were assigned to the scenario elements [11]. The 3-D ray
tracing (3D RT) technique employed in this work is fully written
in MATLAB and provides the computation of the usual reflected
and diffracted components. Furthermore, single-order diffuse
components have been simulated with the directive model in
order to increase the accuracy of the model [9]. Following the
approach of [8], the best parameters for the directive model
(and ) were found by comparing the mea-
suredpowerdelayprofile (PDP) to the simulated PDP obtained
with different combinations of the model parameters. The max-
imum number of reflections was set to two. These values pro-
vide a natural convergence of the algorithm for all simulations.
Finally, we note that the radiating pattern of the antennas was
included to adjust the complex gain of each path. However, the
precision stage guide and mounting bracket were not included
in the model.
For each Tx–Rx pair, 3D RT simulations were performed
over the same frequency points as the VNA to obtain numerical
versions of the measured MIMO channels. This is simply done
by summing the contribution of all waves for each frequency
value to reconstruct the transfer functions in the frequency
domain. This simplifies the comparison between measurements
and simulations. The classical MPC data model used in the lit-
erature and this work is frequency-dependent in the time-delay
domain. However, it is narrowband for the spatial domain (i.e.,
single frequency). Indeed, a wideband description of the spatial
domain would imply measuring the radiating pattern for each
frequency point, which is simply not feasible. Hence, a sub-band
could be selected from both measured and simulated channels
to perform a parametric estimation of the MPCs.
III. MPC EXTRACTION THROUGH RIMAX
Both measured and simulated MIMO radio channels were
processed with the RiMAX maximum-likelihood algorithm.
This estimator was developed to extract the propagation paths
parameters (time delays, azimuth/elevation angles, and com-
plex gains) and dense multipath components (DMC) [10].
The DMC is stochastic by nature and includes both diffuse
scattering and paths that cannot be resolved. The incorporation
of the DMCs into the data model was shown to improve the
accuracy and validity of the estimated propagation paths [10].
Estimated parameters are typically subject to data model and
calibration errors. Hence, the purpose of estimating the pa-
rameters from the simulated channels instead of actually using
those predicted by the 3D RT is to compare fairly the measured
and simulated data. The variance of the measured noise was
computed from the measured channels for each position and
added to the simulated channels since it is used in the estimator
data model.
The following set of parameters is obtained for each multi-
path component :
(1)
where are the complex amplitude, el-
evation angle of departure, azimuth angle of arrival and depar-
ture, and time delay, respectively.
To comply with the narrowband spatial model hypothesis
discussed earlier, a 1.12-GHz bandwidth around 57.56 GHz
(512 frequency points) was selected out of the 9 GHz available
bandwidth. In addition, the whole MIMO array was selected
to perform the estimation (6 6URAforTxandfive-element
ULA for Rx). The five-element ULA restricts the angular
estimation between 90 and 90 , but is here sufficient to
grasp the physics of the propagation mechanisms. Also, note
that elevation estimation is limited to the Tx array in this work.
Finally, we note that the radiation pattern of the antennas was
not included into the estimator data model. Preliminary estima-
tions from 3D RT simulated transfer functions have shown that
the azimuth and elevation angles are correctly grasped with an
omnidirectional approach. Also, the error between the RT and
estimated transfer functions was in the order of 0.05 dB. Due to
the configuration of the room, the most energetic paths arrive
with low elevation angles where the antenna gain is almost
maximal.
IV. RESULTS
A. 9-GHz Bandwidth
The PDP is computed as the inverse Fourier transform of
obtained from both measurements and simulations, and aver-
aged over all 180 channel transfer functions. As an example,
Fig. 2 presents the PDP for position 3 with 9-GHz bandwidth.
For the sake of clarity, no measurement noise was added to the
simulations. It is visually shown that the shape of the simulated
PDP matches well that of the measured one between 10 and
30 ns. However, the baseline of the simulated PDP diverges be-
yond 30 ns.
MARTINEZ-INGLES et al.: DETERMINISTIC AND EXPERIMENTAL INDOOR mmW CHANNEL MODELING 1049
TAB L E I
RELATIVE RECEIVED POWER,DS,AND MED IN 9GHZFROM MEASURED AND SIMULATED (WITH AND WITHOUT DIFFUSE SCATTERING)CHANNELS
Fig. 2. Power delay profile for measurements and simulations.
This proves that second-order diffusion should be added to
fit correctly the shape of the PDP. However, those components
were not considered in this work to reduce the computation
time, but also because its contribution to the total energy of the
channel is very weak. For instance, a maximum difference of
1–2 dB was obtained between the simulated (without second-
order scattering) and measured transfer functions. Both results
indicate the correctness of the3DRTdatamodel.Thewide-
band relative received power in decibels can be computed from
the PDP as the sum of all PDP components
(2)
The RMS DS (second-order moment) that gives us a measure
of the maximum data throughput without equalization can also
be computed from the PDP by [12]
(3)
where are the components of the PDP within a
threshold and their corresponding delay. Finally, the max-
imum excess delay (MED) can be computed as the delay
relative to the first arriving path
(4)
Table I presents the wideband received power, DS, and MED
for simulations (with and without diffuse scattering) and mea-
surements with 9-GHz bandwidth for the first five positions.
A 30-dB threshold was used to eliminate noise and low-power
contributions for the computation of the received power, DS,
and MED. A mean RMS DS of 4.1 ns was computed from the
Fig. 3. Simulated and measured relative received power.
measured and simulated data using all positions. In comparison
to the literature, a 4.8-ns RMS DS was reported in [5] for a
28.2-m room (3.7 ns for 32.9 m in this work). A maximum
excess delay of 29.8 and 27.2 ns was obtained using (4) from
the measurements and simulations, respectively.
Furthermore, the results show that including diffuse scat-
tering, which accounts for 10% of the total energy for all
simulated positions, improves the accuracy of the simulations.
Evidently, this energetic contribution would strongly increase
for obstructed or non-line-of sight (OLOS or NLOS) scenarios
since the LOS component is found to be 10–15 dB more
energetic than secondary paths.
The relative received power computed using (2) is displayed
in Fig. 3 for both measurements and simulations. From the av-
eraged received power, a typical one-slope model was fitted to
the data
(5)
where is the received power for a 1-m reference distance,
the decay factor, and the distance between Tx and Rx.
The one-slope models obtained from the measured and simu-
lated data are in good agreement with a computed decay factor
of 1.52 and 1.51, respectively. These values are in the range
of 1.2 and 2.0 reported in [2] and [5] for mmW LOS measure-
ments. Indeed, the authors in [2] explained that can be smaller
than 2 when wave-guiding and reverberation effects are present,
resulting in power levels increase by multipath aggregation. In
addition, a distribution of the measured and simulated results
around the average received power can be modeled by a zero
mean Gaussian distribution . A 2.17- and 0.23-dB standard
deviation for was found for the measurements and simula-
tions, respectively.
1050 IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS, VOL. 13, 2014
TAB L E II
EXTRACTED PARAMETERS FOR SIMULATED AND MEASURED DATA
B. 1.12 GHz Bandwidth
The MPCs were extracted from both measured and simulated
channel transfer functions over a 1.12-GHz bandwidth (relative
bandwidth less than 2% and thus satisfying the narrowband as-
sumption). Once all MPCs are estimated, the total number of
rays (TNR) is obtained after applying a 30-dB threshold to re-
move the weaker paths. The angular spread (AS) in azimuth and
elevation for direction of arrival and departure angles was also
computed as [13]
(6)
where for the three possible angular
spreads. Note that angle ambiguities were removed to correctly
compute the azimuth AS for the direction of departure.
Table II summarizes in details the parameters for the posi-
tions 1–5 extracted from both simulated and measured data, as
well as the averaged values and standard deviation for the 20 po-
sitions. The average number of extracted MPCs (total number
of rays, TNR) using RiMAX is 15 for the simulations and 60
for measurements. Here, the difference is attributed to the sim-
plified modeling of the propagation scenario.
Finally, an AS value of 41.5 (63.6 )and55.1 (65.5 )was
computed for () from the simulations and measure-
ments, respectively. Similarly, an AS value of 18.4 and 16.6
was computed for from the simulations and measurements,
respectively. In general, a larger spread is obtained from the
measurements than the simulations due to the modeling of the
room, as discussed previously. Nevertheless, the results can be
considered quite satisfactory.
Authors from [4] measured , , ,
values of 40 ,17.2 ,11.4 , and 17.2 , while we have
measured 55.1 ,65.6 , 24.3 and 16.6 .Alargedifferenceis
observed between both .Thisisattributedtothefact
that the authors in [4] used a URA for Rx, whereas a ULA
was chosen in this work. Hence, all rays are folded back into
, which results into increasing its angular spread.
V. C ONCLUSION
In this letter, we have presented an extensive multidimen-
sional analysis of LOS experimental data and ray tracing
simulations including single-order diffuse scattering in the
millimeter-wave frequency band for 20 transmitting positions
in an 80-m office. The results support the idea that diffuse
scattering, which accounts for 10% of the total energy, must
be taken into account in simulations to faithfully reconstruct
channel transfer functions. However, second-order scattering
might not be necessary to assess the propagation parameters
of the mmW channels. The RiMAX algorithm was used to
extract the MPC parameters from both measured and simulated
datasets. From this analysis, a good agreement is reached
between the time delays and power angular spreads computed
from the RT and measured channels. Those values are also
found to be similar to other results published in the scien-
tific literature. In summary, the results show that even a raw
description of the environment can be used to predict mmW
propagation with ray tracing.
REFERENCES
[1] P. Smulders, “Exploiting the 60 GHz band for local wireless multi-
media access: Prospects and future directions,” IEEE Commun. Mag.,
vol. 40, no. 1, pp. 140–147, Jan. 2002.
[2] Part 15.3: WirelessM edium Access Control (MAC) and Physical Layer
(PHY) Specifications for High Rate Wireless Personal Area Networks
(WPANs), Amendment 2: Millimeter-Wavebased Alternative Physical
Layer Extension, IEEE802.15.3c-2009, Oct. 2009.
[3] Part 11: Wireless LAN Medium Access Control (MAC) and Physical
Layer (PHY) Specifications Amendment 3: Enhancements for Very
High Throughput in the 60 GHz Band, IEEE802.11ad draft, Dec. 2011.
[4] C. Gustafson, K. Haneda, S. Wyne, and F. Tufvesson, “On mm-wave
multi-path clustering and channel modeling,” IEEE Trans. Antennas
Propag., vol. 62, no. 3, pp. 1445–1455, Mar. 2014.
[5] W. Fu, J. Hu, and S. Zhang, “Frequency-domain measurement of 60
GHz indoor channels: A measurement setup, literature data, and anal-
ysis,” IEEE Instrum. Meas. Mag., vol. 16, no. 2, pp. 34–40, Apr. 2013.
[6] W . Peter, W. Keusgen, and R. Felbecker, “Measurement and raytracing
simulation of the 60 GHz indoor broadband channel: Model accuracy
and parameterization,” in Proc. 2nd EuCAP, Edinburgh, U.K., Nov.
2007, pp. 1–8.
[7] M. Kazemi, A. Abdipur, and A. Mohammadi, “Indoor propagation
MIMO channel modelingin60GHzusingSBRbased3Draytracing
technique,” in Proc. 2nd MMWaTT, Tehran, Iran, Dec. 24–25, 2012,
pp. 25–28.
[8] J. Jarvelainen et al., “60 GHz radio wave propagation prediction in
a hospital environment using an accurate room structural model,” in
Proc. LAPC, Loughborough, U.K., Nov. 12–13, 2012, pp. 1–4.
[9] V. Degli-Esposti, F. Fuschini, E. M. Vitucci, and G. Falciasecca, “Mea-
surement and modelling of scattering from buildings,” IEEE Trans. An-
tennas Propag., vol. . 55, no. 1, pp. 143–153, Jan. 2007.
[10] A. Richter, “Estimation of radio channel parameters: Models and algo-
rithms,” Dr.-Ing. dissertation, TU Ilmenau, Ilmenau, Germany, 2005.
[11] L. M. Correia and P. O. Françês, “Estimation of materials charac-
teristics from power measurements at 60 GHz,” in Proc. 5th IEEE
Int. Symp. Pers., Indoor Mobile Radio Commun. Wireless Netw.,The
Hague, Netherlands, Sep. 18–23, 1994, vol. 2, pp. 510–513.
[12]T. S. Rappaport, Wireless Communications: Principle and Practice,
2nd ed. Upper Saddle River, NJ, USA: Prentice-Hall, 2002.
[13] B. H. Fleury, “First- and second-order characterization of direction dis-
persion and space selectivity in the radio channel,” IEEE Trans. Inf.
Theory, vol. 46, no. 6, pp. 2027–44, Sep. 2000.
