5G

1. Received December 7, 2016, accepted December 17, 2016, date of publication January 9, 2017, date of current version March 8, 2017.
Digital Object Identifier 10.1109/ACCESS.2017.2650261
Channel Measurements, Modeling, Simulation
and Validation at 32 GHz in Outdoor
Microcells for 5G Radio Systems
XIONGWEN ZHAO1, (Senior Member, IEEE), SHU LI1, QI WANG1, MENGJUN WANG2,
SHAOHUI SUN2, AND WEI HONG3, (Fellow, IEEE)
1School of Electrical and Electronic Engineering, North China Electric Power University, Beijing 102206, China
2China Academy of Telecommunication Technology, Beijing 100191, China
3State Key Laboratory of Millimeter Waves, Nanjing 210096, China
Corresponding author: S. Li (lishu@ncepu.edu.cn)
This work was supported in part by the China Academy of Telecommunication Technology, in part by the State Key Laboratory of
Millimeter Waves, Southeast University, China, under Grant K201517, and in part by the Fundamental Research Funds for
the Central Universities under Grant 2015 XS09.
ABSTRACT In this paper, based on outdoor microcellular channel measurements at 32 GHz for 5G radio
systems, a comprehensive channel modeling, simulation, and validation are performed. The directional-
scan-sounding measurements using a horn antenna rotated with an angular step at the receiver are carried
out, which constitutes a virtual array to form a single-input multiple-output radio channel. The directional-
and omni-directional path-loss models are developed by using close-in and floating-intercept methods. Non-
parametric and parametric methods are applied to extract large-scale channel parameters (LSPs). The non-
parametric method is based on the definition of a channel parameter, whereas the parametric method is
derived by the space-alternating generalized expectation–maximization (SAGE) algorithm, which can de-
embed an antenna pattern. It is found that the LSPs in the angular domain are significantly different by using
the two methods; however, the LSPs in the delay domain almost stay the same. By comparing the LSPs
with the parameter table at 32 GHz with 3GPP standard, it is found that 3GPP LSPs should be corrected at
the International Telecommunications Union-assigned millimeter wave (mmWave) frequencies for 5G. In
this paper, the channel simulation is implemented by using the quasi-deterministic radio channel generator
(QuaDRiGa) platform recommended by 3GPP. By comparing the LSPs with the simulated and measured
results, it is found that QuaDRiGa is a good platform at the mmWave band, even if it is originally developed
for channel simulation below 6 GHz. The results of this paper are important and useful in the simulations
and design of future 5G radio systems at 32 GHz.
INDEX TERMS mmWave, 32 GHz, channel measurement, direction-scan-sounding, path-loss, SAGE,
QuaDRiGa, simulation, validation.
I. INTRODUCTION
The demand in high-data-rate transmission is expected to
grow explosively with the advent of the fifth generation (5G)
radio systems in the next few years as stated in European
METIS (Mobile and wireless communications Enablers
for the Twenty-twenty Information Society) project [1].
Millimeter Waves (mmWave) are regarded as the key fre-
quency candidates for 5G, which can offer very high data rate
in broadband mobile and backhaul services [2], [3]. There-
fore, accurate channel models and parameters at mmWave are
urgently needed in the link and system level simulations for
5G radio systems.
In recent years, the characteristics of indoor radio channels
have been studied in higher frequency bands (HFB), e.g.
10-11 GHz [4]–[6], 28 GHz [1]–[12], 60 GHz [13], [14],
and 70-73 GHz [11], [14], [15]. The measurement campaigns
for outdoor urban cellular network have been carried out in
HFB of 10, 18, 28, 38, 60, 72 and 81-86 GHz [16]–[20].
In worldwide, there are some research projects related to 5G
mmWave channel measurements and modeling work such as
METIS [1], NYU WIRELESS [11], [16], mmMAGIC [22],
MiWEBA [23], and 3GPP [24] and so on. Based on the
available work, 3GPP finalized and published its 5G channel
models in this year [24], and its main target was to get
1062
2169-3536
2017 IEEE. Translations and content mining are permitted for academic research only.
Personal use is also permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
VOLUME 5, 2017

2. X. Zhao et al.: Channel Measurements, Modeling, Simulation and Validation at 32 GHz in Outdoor Microcells
universal path-loss models and the large scale channel param-
eters (LSPs) within 6 – 100GHz to meet 5G link and system
level simulations at different carrier frequencies. However,
3GPP outputs were based on only a few of measured car-
rier frequencies with different measurement environments in
the world, the frequency dependent path-loss models and
LSPs are for sure required to be validated by more coming
measurements. This kind of work could be left to Interna-
tional Telecommunications Union (ITU) which has started its
5G channel standardization right now. ITU assigned different
spectrum segments between 24 and 86 GHz for 5G radio
systems in World Radio Communications Conference (WRC)
last year [25], in which 24.25-27.5 GHz and 31.8-33.4 GHz
might be the primary choice for 5G wireless access [26].
So far there are some channel measurements and modeling
work available at 24.25-27.5 GHz, however, there are no
results at 31.8-33.4 GHz by us knowledge, which is why we
focus on 32 GHz channel in this work.
Channel measurements in mmWave are now carried out
with either a network analyzer or a channel sounder, where at
least one horn antenna is required to be used at the transceiver
to increase the measurement distance. In a practical measure-
ment, because it’s very difficult to align two horn antennas,
therefore an omni-directional antenna is commonly used at
the transmitter (Tx) and a horn is equipped at the receiver (Rx)
to perform directional-scan-sounding (DSS) measurements.
To extract channel LSPs, e.g. rms delay- and angular-spread,
number of clusters etc. from measurement data, the non-
parametric [16]–[19] and parametric methods [4], [12] are
applied, the former one is based on the definitions of
channel parameters, while the later one is based on high
resolution method, e.g. using SAGE (Space-Alternating Gen-
eralized Expectation-maximization). While applying SAGE
algorithm to estimate the channel parameters, the concept of
multipath component (MPC) is defined as a propagation path
over the environment and the signal model which reflects
the MPCs and antenna array configuration is required. The
advantage of parametric method over non-parametric one is
that it can de-embed antenna pattern when extracting LSPs,
then more accurate LSPs can be extracted from measured
channel impulse responses (CIRs), especially in the angular
domain [12], [27].
A geometric based stochastic channel model (GSCM)
was proposed in Wireless World Initiative New Radio
Project (WINNER) [28] and ITU-R [29], which is very popu-
lar platform for 4G channel simulation. To extend WINNER
and ITU-R model to three-dimensional (3D) MIMO model,
quasi deterministic radio channel generator (QuaDRiGa) has
been developed first [30]. Most recently, QuaDRiGa was
extended to mmWave and recommended for 5G channel
simulations by 3GPP [24]. However, the accuracy of the
platform is yet to be verified by measurements, especially at
ITU assigned mmWave frequency bands.
In this paper, the DSS channel measurements for out-
door microcells at 32 GHz with 1 GHz bandwidth are car-
ried out. The horn antenna is rotated in both azimuth and
elevation planes, respectively with specific angular steps.
Virtual single-input and multiple-output (SIMO) system can
be formed by horn rotating positions. Moreover, the horn
antenna pattern is measured in an anechoic chamber, then
the LSPs and the parameters inside clusters can be extract
by SAGE. Two kinds of path-loss models are developed,
namely directional and omni-directional path loss models.
The former one is to find the maximum received power
while the horn is rotating for beamforming and tracking, and
the later one is to integrate the received powers in all dif-
ferent horn orientations at a specific measurement location.
Finally, a parameter table is summarized as in WINNER [28]
to implement channel simulation and validation by using
QuaDRiGa platform, the simulation results are compared
with the modeling results in [17], mmMAGIC [22], and
3GPP TR 38.900 [24] for outdoor microcells, and also val-
idated by the measurements at 32 GHz in this work.
The novelties of this paper include: (1) the first out-
door microcell measurements are performed at 32 GHz ITU
assigned frequency spectrum by us knowledge; (2) Channel
simulation and validation are done using 3GPP recommended
QuaDRiGa platform, which has not been done so far at
mmWave. (3) A comprehensive parameter table and path-loss
models are given first time at 32 GHz in the link and system
level simulations for 5G radio systems, where the LSPs are
extracted by SAGE to de-embed antenna pattern.
The rest of the paper is organized as follows. The mea-
surement campaign and system are introduced in Section II.
In Section III, the SAGE algorithm based on the DSS signal
model is described and an example of estimation is pre-
sented. In Section IV, the model parameters are extracted
and the comparison among this work with mmMAGIC, NYU
WIRELESS, and 3GPP are summarized. Section V presents
the channel simulation and validation by using QuaDRiGa
simulation platform. The conclusion is drawn in Section VI.
II. MEASUREMENT CAMPAIGN AND SYSTEM
The outdoor microcellular measurements were conducted
in the campus of North China Electric Power Univer-
sity (NCEPU) in Beijing, China in two different scenarios
as shown in Fig. 1(a) and (b), respectively. Less building
windows at the two sides of the road with sparser but taller
trees in Scenario #1 than Scenario #2. In each scenario, a
line-of-sight (LoS) and a non-line-of-sight (NLoS) routes are
measured as shown in Fig. 2(a) and (b), respectively. The
BS and MS are located in a crane and a trolley with heights
of 6.1 meter and 1.8 meter, respectively, and equipped an
omni-directional antenna and a horn antenna. The horn has
10◦ half-power-beamwidth (HPBW) and is able to rotate in
both of the azimuth and elevation planes by using a stepper
motor. To get the complete 3D angle-of-arrival (AoA) infor-
mation, the horn is rotated from 0◦ to 360◦ with 5◦ angular
step in the azimuth plane, and several co-elevation angles are
selected to repeat such rotating measurement with respect to
the distance between the BS and MS. For example, at the
first MS measurement location (closest to the BS) in route
VOLUME 5, 2017 1063

3. X. Zhao et al.: Channel Measurements, Modeling, Simulation and Validation at 32 GHz in Outdoor Microcells
FIGURE 1. Measurement environments. (a) Scenario #1 and
(b) Scenario #2.
FIGURE 2. Layout of measurement campaigns. (a) Scenario #1.
(b) Scenario #2. Legend of (a) is suitable for (b) as well.
R1-1 shown in Fig. 2(a), the co-elevation angle is set from
−10◦ to 50◦ with step of 10◦, thus 72∗7 directions are mea-
sured in 3D space. As the distance between the BS and MS
is larger, the co-elevation angles can be reduced because of
FIGURE 3. Block diagram of measurement system.
TABLE 1. Measurement system parameters.
very time consuming measurement. The co-elevation angles
at each MS location are shown in the legend of Fig. 2(a).
In both of the scenarios, there are 20 and 14 MS locations are
measured for the LoS and NLoS routes, respectively. At each
MS location, about 21,600 CIRs are collected when only 2D
measurements are considered, thus more than 430,000 and
300,000 CIRs are recorded in scenarios #1 and #2, respec-
tively. The measurements were mainly conducted at midnight
to avoid people’s movement. The measurement system is
developed by Keysight with block diagram shown in Fig. 3
with system level parameters listed in Table 1 in this work.
The detailed information of the sliding correlated sounder and
the calibration method are described in [31].
III. PARAMETER ESTIMATION FOR
PROPAGATION PATHS
A. SIGNAL MODEL FOR THE DSS SYSTEM
SAGE algorithm was widely used in WINNER to extract
multipath parameters for MIMO channel measurements [28],
SAGE implementation can be found in [32]. In this mea-
surements, an biconical horn is used at the Tx while the
horn antennas is rotated in different directions, therefore the
measurement system can be regarded as a virtual SIMO with
number of M elements or directions in the Rx, the receive
signal of a single impinging wave can be written as
s (t; ρl) = [s1 (t; ρl) , . . . , sM (t; ρl)]T
= c (θl, ϕl) αl exp (j2πυlt) u (t − τl) (1)
where ρl = [τl, θl, ϕl, υl, αl] is the parameter set of path l
to be estimated, where τ is delay, θ is the elevation-angle-of-
arrivals (EAoA), ϕ is the azimuth-angle-of-arrivals (AAoA),
1064 VOLUME 5, 2017

4. X. Zhao et al.: Channel Measurements, Modeling, Simulation and Validation at 32 GHz in Outdoor Microcells
υ is Doppler frequency and α is the amplitude. u (t) is the
reference signal. The steering vector c (θ, ϕ) can be expressed
as
c (θ, ϕ) =



c1 (θ, ϕ)
.
.
.
cM (θ, ϕ)



=







f1 (θ, ϕ) exp
j2π
λ
he (θ, ϕ) , r1i
.
.
.
fM (θ, ϕ) exp
j2π
λ
he (θ, ϕ) , rM i







(2)
where r1, r2, . . . rM are the positions of the M elements in
Cartesian coordinate system, f is the complex antenna pat-
tern, e (θ, ϕ) is the unit direction vector and hi is inner prod-
uct. Then the receiving signal Y (t) = [Y1 (t) , . . . , YM (t)]T
is given by
Y (t) =
L
X
l=1
s (t; ρl) +
r
N0
2
N (t) (3)
where N0 is a positive constant, N (t) = [N1 (t) , . . . , NM (t)]T
is M- dimensional complex white Gaussian noise.
FIGURE 4. Directional-scan-sounding (DSS) system with the biconical
horn and horn antennas at the Tx and Rx, respectively.
DSS is the most widely used method in millimeter wave
channel measurement as shown in Fig. 4. Assuming that the
channel is time-invariant and the feed point remains in the
center when the horn is rotating, then the steering vector
c (θ, ϕ) in (2) can be written as
c (θ, ϕ) =



c1 (θ, ϕ)
.
.
.
cM (θ, ϕ)


 =



f θ − θ1, ϕ − ϕ1
.
.
.
f θ − θM , ϕ − ϕM


 (4)
where e θi, ϕi
is the unit direction vector of i-th orientation.
As the position of the transceiver keeps unchanged with no
moving scatterers during the horn rotating, s (t; ρl) in (1) can
be written as
s (t; ρl) = c (θl, ϕl) αlu (τ − τl) . (5)
FIGURE 5. MPCs at location 1 in route R3-4 with θ = 0◦. (a) MPCs
estimated by SAGE algorithm. (b) Concatenated PDPs derived from
72 directional CIRs.
B. EXTRACTION OF CHANNEL PARAMETERS USING SAGE
By applying the signal model (4)-(5) in the DSS system,
channel parameters for the propagation paths related to delay
and angular domains can be extracted by SAGE. Since the
full spatial scanning in the azimuth and elevation planes is too
time-consuming, for most the measured locations especially
those locations far away from the Tx, only 3∼5 elevation
angles are scanned. In this case, the EAoAs extracted by
SAGE can have estimation error according to [33]. As an
alternative, we estimate the delay and AAoAs of the MPCs
for each co-elevation angle. The SAGE estimation parameters
are set as follows: the number of the MPCs is 200, the
maximum iteration number is 10, the MPCs with power at
least 3 dB larger than the noise level are picked up from
200 estimated MPCs.
As an example, a scatter plot of the delays, AoAs, and gains
of the estimated MPCs in the NLoS case measured in route
R3-4 at location 1 with co-elevation θ = 0◦ is illustrated in
Fig. 5(a) by using SAGE. For sake of comparison, the cor-
responding plot of the delays, AoAs and gains from original
CIRs are displayed in Fig. 5(b). From Fig. 5(a) and (b), it is
seen that all distinguishable MPCs are extracted. The perfor-
mance of the MPCs extraction is evaluated by calculating the
percentage of reconstruction energy with respect to the raw
VOLUME 5, 2017 1065

5. X. Zhao et al.: Channel Measurements, Modeling, Simulation and Validation at 32 GHz in Outdoor Microcells
measured CIRs. The mean reconstruction energy in this work
is about 88%.
IV. CHANNEL MODELS AND PARAMETERS
A. PATH-LOSS MODEL
In this work, two kinds of path-loss models are developed
based on the DSS measurements, namely directional and
omni-directional path-loss models. The former one is to find
the maximum receiver power or minimum path-loss in each
measurement location with different horn rotation angles,
which could be useful for beamforming or tracking. The
path-loss in the later one can be calculated as [11] to inte-
grate power in all the horn rotating directions at a specific
location
PLi [dB] = Pt [dBm] − 10 log10
×
X
az
X
el
Pri (θel, ϕaz) [mW]
#
(6)
where Pri (θel, ϕaz) is the received power after removing the
antenna gain for azimuth angle ϕaz and elevation angel θel in
measured location i, Pt is the transmit power.
Two well-known models are used to develop path-loss
models in this work. The first one is called close-in (CI)
model, in which the path-loss intercept is calculated by
assuming 1 meter reference distance in free space [16] and
is given by
PLCI = 20 log10
4πf
c
+ 10n log10
d
d0
(7)
where n is the path-loss exponent, f is the carrier frequency,
c is the speed of light, d0 = 1 m is the reference distance
and d is the separation of the Tx and Rx in 3D space. The
second one is called floating intercept (FI) model [27] , which
is calculated as
PLFI = β + 10α log10
d
d0
(8)
where α is the distance dependence coefficient similar with n
in CI model and β is the floating path-loss intercept.
Figure 6(a) and 6(b) shows the path-loss models as well as
shadow fading for Scenarios #1 and #2, respectively. As a ref-
erence, free space path-loss models are also included. From
Fig. 6(a) it is seen that omni-directional path-loss models are
slightly below the directional path-loss models both in the
LoS and NLoS routes since the directional path-loss model
only contains the path-loss for strongest path, while omni-
directional path-loss model also includes the other propaga-
tion paths. The directional and omni-directional path losses
for the LoS routes are roughly equal to the free space path-
loss. The shadow fading (SF) of the CI model is slightly larger
than that in FI model, and the difference is within 1 dB. The
path-loss models and shadow fading by the CI method are
summarized in Table 2.
FIGURE 6. Directional and omni-directional path-loss models and
shadow fading using CI and FI models. (a) Scenario #1. (b) Scenario #2.
B. RMS DELAY SPREAD AND ANGULAR SPREAD
Root mean square (RMS) delay spread (RDS) is calculated
by
DS =
v
u
u
t
PL
l=1 Plτ2
l
PL
l=1 Pl
− τ2
0 (9)
1066 VOLUME 5, 2017

6. X. Zhao et al.: Channel Measurements, Modeling, Simulation and Validation at 32 GHz in Outdoor Microcells
TABLE 2. Comparison of the GSCM parameters in this work and open literature.
where L is the total number of paths estimated by SAGE and
τ0 is the mean excess delay given by
τ0 =
v
u
u
t
PL
l=1 Plτl
PL
l=1 Pl
. (10)
The rms angle spread (RAS) is calculated as in [34].
By defining the circular angle spread, differences caused by
the zero degree selection could be avoided.
RAS = min
1
σAS (1) =
v
u
u
t
PL
l=1 ψl,µ (1)
2
Pl
PL
l=1 Pl
(11)
where
ψl,µ (1) =





ϕ + 2$, if ϕ −$
ϕ, if |ϕ| ≤ $
ϕ − 2$, if ϕ $
(12)
where $ = π for the AAoDs (Azimuth Angle-of-Departure)
and AAoAs and $ = π/2 for the EAoDs (Elevation Angle-
of-Departure ) and EAoAs, ϕ = ψl (1) − µψ (1) and
µψ (1) =
PL
l=1 ψl (1) Pl
PL
l=1 Pl
(13)
where 1 designates angle shift from −$ to $ and
ψl (1) =





(ψl + 1) + 2$, if (ψl + 1) −$
(ψl + 1) , if |(ψl + 1)| ≤ $
(ψl + 1) − 2$, if (ψl + 1) $
(14)
For the two typical microcell scenarios measured in this
work, parametric method based on SAGE as well as non-
parametric method are used to get the RDS and RAS.
By using non-parametric method , they are derived by the
PDPs (Power-Delay-Profiles) and PAPs (Power-Angular-
Profiles) obtained from original CIRs. The noise floor is esti-
mated by the last two hundred delay samples where no signal
received and the noise cut threshold is set as 5dB above noise
level. The cumulative probability functions (CDFs) of the
RDS for Scenarios #1 and #2 are shown in Fig. 7(a) and (b),
respectively. The CDFs of the azimuth RAS (ARAS) for
Scenario #1 and #2 are shown in Fig. 8(a) and (b), respec-
tively. It is seen from Fig. 7 that the RDS distributions by non-
parametric method and SAGE are very close. However, the
ARAS distributions have relative big difference as seen from
Fig. 8. Non-parametric method is over-estimate the ARAS.
The same phenomenon is also observed in [27], which is
because the non-parametric method is based on the rms angu-
lar spread definition in (11) to extract it from measured CIRs,
the effect of virtual array pattern is included, while SAGE
algorithm can de-embed the array pattern from the measured
CIRs. The RDS and ARAS for the NLoS scenarios are larger
than those for the LoS scenarios, which is consistent with the
results below 6 GHz [28]. It is also found from Fig. 8 that RDS
in Scenario #2 is much larger than that in Scenario #1 which
might be due to the dense building windows in Scenario #2
with specular reflection from windows every now and then.
The statistical results of the RDS and ARAS are summarized
in Table 2.
VOLUME 5, 2017 1067

7. X. Zhao et al.: Channel Measurements, Modeling, Simulation and Validation at 32 GHz in Outdoor Microcells
FIGURE 7. Rms delay spread derived by non-parametric method and
SAGE. (a) Scenario #1. (b) Scenario #2.
FIGURE 8. Azimuth rms angular spread derived by non-parametric
method and SAGE. (a) Scenario #1. (b) Scenario #2.
C. CLUSTER PARAMETERS AND RICEAN K-FACTOR
Cluster is a widely used concept in the GSCMs such
as 3GPP SCM and WINNER models as well as in
Saleh-Valenzula (SV) model for ultra-wideband [35].
To extract the MPCs by SAGE, the cluster should be dis-
tinguished in delay-angular domain. In order to implement
channel simulation, the cluster level parameters, e.g. number
of clusters, cluster RDS and RAS etc. are essential. A mul-
tipath component distance (MCD)-based method is adopted
to group the MPCs [36]. The MCD between the i-th and j-th
MPCs (i 6= j) is given by
MCDi,j =
q
MCD2
AoA,ij + MCD2
τ,ij (15)
where MCDAoA,ij and MCDτ,ij are the MCD in angular
domain and delay domain, respectively, calculated as
MCDAoA,ij =
j − i
,
MCDτ,ij = ζ

9. τj − τi

11. στ
1τ2
max
(16)
where i and j are the directions of arrival given
by  = [cos (ϕ) cos (θ) , sin (ϕ) cos (θ) , sin (θ)]T
for
the i-th and j-th MPCs, respectively, τi and τj are the
delays of the i-th and j-th MPCs, respectively, 1τmax =
max

13. τj − τi

15. ; ∀i, j ∈ [i, . . . , L]
, and ζ is a delay scaling
factor to balance the weights of the MCD in angular and delay
domains. The clustering algorithm consists of the following
three steps [12]:
1) Choose a reference MPC which has the largest power
among all the MPCs in a set eligible for extracting clusters.
2) Calculate the MCD between the reference MPC and all
other MPCs in the set, select those MPCs with the MCDs less
than a predefined threshold denoted with MCDth, and group
them together with the reference MPC as one cluster.
3) Remove the MPCs already allocated to a cluster from the
MPC set, and re-execute step (1) to find the next cluster. This
procedure stops until all the MPCs are assigned to certain
clusters. The parameter ζ and MCDth are determined by a
visual inspection evaluated by whether the clustering results
can map to the physical environment. In this work, we found
that ζ = 5 and MCDth = 0.25 are appropriate values to
take. The 3D scatter plots (delay-AAoA-relative power) are
shown in Fig. 9 in which Fig. 9(a) is by clustering result, and
Fig. 9(b) is got by the measured CIRs in delay and angular
domains. Clusters grouped by the MPCs are distinguished
with different colors in Fig. 9(a) where the Circles represent
the scatter locations. It is seen from Fig. 9 that clustering
result agrees well with the physical environment. Based on
the clustering result, the cluster RDS and RAS can be cal-
culated by (9)-(14) from the propagation paths within each
cluster. The cluster level parameters are also summarized
in Table 2.
Ricean K-factor (KF) is the power ratio between the
LoS component and the sum of other propagation compo-
nents [37]. The CDF plots of K-factor are shown in Fig. 10. It
is seen that the K-factor fits well into norm distribution. The
K-factor in Scenario #2 is less than that in Scenario #1 due
to more windows in the buildings at the two sides of the road
are scattered the power in Scenario #2. Figure 11 shows the
CDF plots of number of clusters for the four measured routes,
they fit well into normal distribution as well. The mean values
of number of clusters for the LoS routes are about 8, which
is the same as in WINNER microcell scenario [28]. For the
NLoS routes, the mean values of number of clusters here are
less than 10, which is less than the WINNER value of 16.
This is because in mmWave propagation, blockage, e.g. by a
1068 VOLUME 5, 2017

16. X. Zhao et al.: Channel Measurements, Modeling, Simulation and Validation at 32 GHz in Outdoor Microcells
FIGURE 9. Scatter plots for the cluster locations in the NLOS route R3-4.
(a) Clustering result. (b) Results by measured CIRs in delay and angular
domains.
FIGURE 10. K-factor for the LoS routes in Scenarios #1 and #2.
building corner, can cause big loss, which may resulting in
fewer number of clusters in the NLoS cases. The statistical
values of K-factor and number of clusters are summarized
in Table 2.
D. COMPARISON AND SUMMARY OF THE CHANNEL
MODELS AND PARAMETERS
In order to implement channel simulations and validations
in next Section, Table 2 summarizes the statistical values of
FIGURE 11. Number of cluster for the four measurement routes in
Scenarios #1 and #2.
the LSPs, the parameters in clusters and the path-loss mod-
els. For comparison purpose, the corresponding parameters
in [17] at 28 GHz, in mmMAGIC [22] and 3GPP [24] at
32 GHz are also listed in the same table. For the path-loss
models and shadow fading, it is found in Table 2 that the path-
loss exponents (PLEs) expressed by A for the LoS cases are
in range of 1.8 to 2.2, which are close to that in free space
of 2.0 in this and other work. For the NLoS cases, the shadow
fading in this work is smaller than other work, and the PLE
for Scenario #1 is slightly larger than other work.
It is also observed from Table 2 that cluster parameters
such as number of clusters, cluster ASA and ESA in this
work agree well with [17] and [22] except the 3GPP [24].
It is also found that although the scenarios in Table 2 are all
for outdoor microcells, there are relatively big differences for
the LSPs, e.g. mean values of the RDS and RAS measured
in NYC Campus are larger than Daojeon in both of the LoS
and NLoS cases, and those values given by 3GPP [24] are
larger than mmMAGIC [22]. In this work, these differences
are also observed in Scenarios #1 and #2, which is caused by
different outdoor environments such as window density on
building surfaces, the trees and cars in the road etc. Therefore,
despite the differences among these parameters derived by
the measurements with different environments, from Table 2
it’s seen that the channel models and parameters in this
work have relatively good agreements with [17] and [22].
The 3GPP parameter table should be corrected at the ITU
assigned carrier frequencies. It’s also recommend to use the
parameter table derived at a specific carrier to implement
channel simulation other than using 3GPP frequency depen-
dent parameters because, as mentioned in the introduction,
3GPP has developed frequency dependent channel models
and parameters from 6 to 100 GHz, but its parameter table is
based on the measurements performed only in a few discrete
carrier frequencies with different measurement environments
in the world.
V. CHANNEL SIMULATION AND VALIDATION
We use 3GPP recommended channel simulation platform
called QuaDRiGa [17] to implement channel simulation
and validation, where our proposed parameters in Table 2
at 32 GHz are used as the inputs to generate chan-
nel coefficient.The implementation of QuaDRiGa is avail-
able as an open source in [38]. Because of the usage of
VOLUME 5, 2017 1069

17. X. Zhao et al.: Channel Measurements, Modeling, Simulation and Validation at 32 GHz in Outdoor Microcells
FIGURE 12. CDF plots of the rms delay spread derived by the simulated
channel coefficient and measurements for the LoS and NLoS cases.
(a) Scenario #1. (b) Scenario #2.
FIGURE 13. CDF plots of the azimuth angular spread of arrival (ARAS) by
the simulated channel coefficient and measurements for the LoS and
NLoS cases. (a) Scenario #1. (b) Scenario #2.
omni-directional antenna at the transmitter in our 32 GHz
measurements, the AAoDs (Azimuth-Angle-of-Departures)
and EAoDs (Elevation-Angle-of-Departures) are not avail-
able, therefore the AAoDs and EAoDs derived from NYC
Campus in [17] are adopted in our simulations. The reasons
for selecting [17] are that the measurements were also con-
ducted in university campus, and the other LSPs in Table 2
have relative good agreement with our measurement results.
The settings of simulation scenario are as follows: the
Tx is equipped with an omni-directional antenna, its height
FIGURE 14. CDF plots of the elevation angular spread of arrival (ERAS) by
the simulated channel coefficient and measurements for the LoS and
NLoS cases. (a) Scenario #1. (b) Scenario #2.
FIGURE 15. CDF plots of K-factor by simulated channel coefficient and
measurements for the LoS cases in Scenarios #1 and #2.
is 6 meters and is located at the middle of map and sur-
rounded by 250 receivers with height of 1.8 meters, they
are equipped with dipole antenna distributed uniformly
in a circle with radius of 200 meters. Both Scenarios
#1 and #2 are simulated with the LoS and NLoS cases.
Figures 12-15 show the comparison between measurement
results and simulation outputs of the RDS, ARAS and ERAS
as well as the K-factor, respectively. Figures 12(a)–15(a) are
for Scenario #1, while Figs. 12(b)–15(b) are for Scenario #2,
respectively.
In the simulations, the RDSs are calculated by the sim-
ulated channel coefficients where the delays are generated
by the simulator. The ARASs and ERASs are calculated
using the power of each path based on simulated channel
coefficient, the AoAs and EoAs are generated by the simu-
lator, respectively. K-factor is calculated by (17). It is seen
from Figs. 12–15 that the simulation results agree very well
with measured results for both of the LoS and NLoS cases
in Scenario #1 and #2. At the same time it can be found
that QuaDRiGa is good platform for channel simulation in
mmWave band at 32 GHz.
1070 VOLUME 5, 2017

18. X. Zhao et al.: Channel Measurements, Modeling, Simulation and Validation at 32 GHz in Outdoor Microcells
VI. CONCLUSION
In this paper, the channel measurements are carried out for
outdoor microcells at 32 GHz ITU assigned mmWave band
for 5G radio systems with bandwidth of 1 GHz. SAGE algo-
rithm is used to extract the MPCs based on the measure-
ments by rotating a horn antenna in azimuth and elevation
planes to form a virtual array. The directional- and omni-
directional path-loss models are developed by using both
of the CI and FI models. Based on the MPC parameters
extracted by SAGE, the LSPs are calculated and compared
with the results by the non-parametric method. The chan-
nel parameters in delay domain, e.g. the RDSs derived by
non-parametric method and SAGE algorithm are very close,
however, the angular domain parameters, such as the RASs
derived by non-parametric method are much larger than those
by SAGE which can de-embed array pattern from the MPCs.
By using SAGE estimated MPCs, clusters are identified using
MCD based clustering approach, and then the number of
cluster and cluster level parameters are calculated. Parameter
table at 32 GHz is summarized and compared with open
literature, project report and 3GPP standard, which shows
that our results are more close to the NYU WIRELESS and
mmMAGIC outputs, however, a relative big difference from
the 3GPP results existed. 3GPP frequency dependent param-
eter table is based only on the measurements performed in a
few discrete carrier frequencies with different environments
in the world, its parameter table should be validated and cor-
rected at the ITU already assigned carrier frequencies. In this
work, 3GPP recommended QuaDRiGa platform is applied to
implement channel simulation, it’s found that QuaDRiGa is
a good platform at 32 GHz mmWave band even if it’s orig-
inally a platform for channel simulation below 6 GHz. The
results in this work are important in 5G link and system level
simulations at 32 GHz. The comprehensive studies here in
channel measurements, modeling, simulation and validation
at 32 GHz could be very useful in 5G channel research, the
methodologies can be extended to the other ITU assigned
spectra as well.
REFERENCES
[1] METIS. (2015). Mobile and Wireless Communications Enablers for
the Twenty-Twenty Information Society, D1.4, METIS Channel Models.
[Online]. Available: http://www.metis2020.com
[2] T. S. Rappaport et al., ‘‘Millimeter wave mobile communications
for 5G cellular: It will work!’’ IEEE Access, vol. 1, pp. 335–349,
May 2013.
[3] W. Roh et al., ‘‘Millimeter-wave beamforming as an enabling technol-
ogy for 5G cellular communications: Theoretical feasibility and pro-
totype results,’’ IEEE Commun. Mag., vol. 52, no. 2, pp. 106–113,
Feb. 2014.
[4] A. Roivainen, C. F. Dias, N. Tervo, V. Hovinen, M. Sonkki, and
M. Latva-Aho, ‘‘Geometry-based stochastic channel model for two-story
lobby environment at 10 GHz,’’ IEEE Trans. Antennas Propag., vol. 64,
no. 9, pp. 3990–4003, Sep. 2016.
[5] M. Kim, J.-I. Takada, Y. Chang, J. Shen, and Y. Oda, ‘‘Large scale
characteristics of urban cellular wideband channels at 11 GHz,’’ in Proc.
Eur. Conf. Antennas Propag. (EuCAP), Lisbon, Portugal, Apr. 2015,
pp. 1–4.
[6] K. Belbase, M. Kim, and J.-I. Takada, ‘‘Study of propagation mecha-
nisms and identification of scattering objects in indoor multipath channels
at 11 GHz,’’ in Proc. Eur. Conf. Antennas Propag. (EuCAP), Lisbon,
Portugal, Apr. 2015, pp. 1–4.
[7] X. Wu, Y. Zhang, C.-X. Wang, G. Goussetis, E.-H. M. Aggoune, and
M. M. Alwakeel, ‘‘28 GHz indoor channel measurements and modelling
in laboratory environment using directional antennas,’’ in Proc. Eur. Conf.
Antennas Propag. (EuCAP), Lisbon, Portugal, Apr. 2015, pp. 1–5.
[8] F. S. Huang, L. Tian, Y. Zheng, and J. H. Zhang, ‘‘Propagation characteris-
tics of indoor radio channel from 3.5 GHz to 28 GHz,’’ in Proc. 84th Veh.
Technol. Conf. (VTC-Fall), Montreal, QC, Canada, Sep. 2016.
[9] X. Gao, L. Tian, P. Tang, T. Jiang, B. Liu, and J. H. Zhang, ‘‘Channel char-
acteristics analysis of angle and clustering in indoor office environment
at 28 GHz,’’ in Proc. 84th Veh. Technol. Conf. (VTC-Fall), Montreal, QC,
Canada, Sep. 2016.
[10] M. Lei, J. Zhang, T. Lei, and D. Du, ‘‘28-GHz indoor channel mea-
surements and analysis of propagation characteristics,’’ in Proc. IEEE
25th Annu. Int. Symp. Pers., Indoor, Mobile Radio Commun. (PIMRC),
Sep. 2014, pp. 208–212.
[11] G. R. Maccartney, T. S. Rappaport, S. Sun, and S. Deng, ‘‘Indoor office
wideband millimeter-wave propagation measurements and channel models
at 28 and 73 GHz for ultra-dense 5G wireless networks,’’ IEEE Access,
vol. 3, pp. 2388–2424, 2015.
[12] X. Yin, C. Ling, and M.-D. Kim, ‘‘Experimental multipath-cluster
characteristics of 28-GHz propagation channel,’’ IEEE Access, vol. 3,
pp. 3138–3150, 2015.
[13] R. C. Daniels and R. W. Heath, ‘‘60 GHz wireless communica-
tions: Emerging requirements and design recommendations,’’ IEEE Veh.
Technol. Mag., vol. 2, no. 3, pp. 41–50, Sep. 2007.
[14] P. F. M. Smulders, ‘‘Statistical characterization of 60-GHz indoor
radio channels,’’ IEEE Trans. Antennas Propag., vol. 57, no. 10,
pp. 2820–2829, Oct. 2009.
[15] N. Zhang, X. Yin, S. X. Lu, M. Du, and X. Cai, ‘‘Measurement-based
angular characterization for 72 GHz propagation channels in indoor envi-
ronments,’’ in Proc. Globecom Workshops (GC Wkshps), Austin, TX,
USA, Dec. 2014, pp. 370–376.
[16] T. S. Rappaport, G. R. Maccartney, M. K. Samimi, and S. Sun, ‘‘Wideband
millimeter-wave propagation measurements and channel models for future
wireless communication system design,’’ IEEE Trans. Commun., vol. 63,
no. 9, pp. 3029–3056, Sep. 2015.
[17] S. Hur et al., ‘‘Proposal on millimeter-wave channel modeling for 5G cel-
lular system,’’ IEEE J. Sel. Topics Signal Process., vol. 10, no. 3,
pp. 454–469, Apr. 2016.
[18] S. Hur et al., ‘‘Wideband spatial channel model in an urban cellular
environments at 28 GHz,’’ in Proc. Eur. Conf. Antennas Propag. (EuCAP),
Apr. 2015, pp. 1–5.
[19] Y. Azar et al., ‘‘28 GHz propagation measurements for outdoor cellular
communications using steerable beam antennas in New York City,’’ in
Proc. IEEE Int. Conf. Commun. (ICC), Budapest, Hungary, Jun. 2013,
pp. 5143–5147.
[20] V. Semkin, U. Virk, A. Karttunen, K. Haneda, and A. V. Räisänen,
‘‘E-band propagation channel measurements in an urban street canyon,’’ in
Proc. Eur. Conf. Antenna Propag. (EuCAP), Lisbon, Portugal, Apr. 2015,
pp. 1–4.
[21] H. C. Nguyen, I. Rodriguez, T. B. Sorensen, L. L. Sanchez, I. Kovacs, and
P. Mogensen, ‘‘An empirical study of urban macro propagation at 10,
18 and 28 GHz,’’ in Proc. IEEE 83rd Veh. Technol. Conf. (VTC-Spring),
Nanjing, China, May 2016, pp. 1–5.
[22] mmMagic. (2016). mm-Wave Based Mobile Radio Access Network for
5G Integrated Communications, W2.1, 6–100 GHz Channel Modeling for
5G: Measurement and Modeling Plans in mmMAGIC. [Online]. Available:
https://5g-mmmagic.eu
[23] MiWEBA. (2014). Millimetre-Wave Evolution for Backhaul and Access,
D5.1, Channel Modeling and Characterization. [Online]. Available:
http://www.miweba.eu
[24] Study on Channel Model for Frequency Spectrum Above 6 GHz,
TR 38.900, 3GPP. (Sep. 2016). [Online]. Available: http://www.3gpp.org
[25] World Radio Communication Conference (WRC-15), ITU 500-E, Geneva,
Switzerland, Nov. 2015.
[26] RSPG. (Sep. 2016). Responses to RSPG Public Consultation on 5G.
[Online]. Available: http://rspg-spectrum.eu
[27] Y. Ji, X. Yin, H. Wang, X. Lu, and C. Cao, ‘‘Antenna-de-embedded
characterization for 13-17 GHz wave propagation in indoor envi-
ronments,’’ IEEE Antennas Wireless Propag. Lett., to be published,
doi: 10.1109/AWPL.2016.2553455.
[28] IST-4-027756 WINNER II, D1.1.2 ver 1.2, Final Report on Link Level and
System Level Channel Models, (WINNER), Feb. 2008.
[29] ‘‘Report ITU-R M.2135 guidelines for evaluation of radio interface tech-
nologies for IMT-advanced,’’ International Telecommunication Union-
Radio Communication Sector, Tech. Rep. ITU-R M.2135-1, 2009.
VOLUME 5, 2017 1071

19. X. Zhao et al.: Channel Measurements, Modeling, Simulation and Validation at 32 GHz in Outdoor Microcells
[30] S. Jaeckel, L. Raschkowski, K. Borner, and L. Thiele, ‘‘QuaDRiGa:
A 3-D multi-cell channel model with time evolution for enabling virtual
field trials,’’ IEEE Trans. Antennas Propag., vol. 62, no. 6, pp. 3242–3256,
Jun. 2014.
[31] Z. Wen and H. Kong, ‘‘mmWave MIMO channel sounding for 5G,’’ in
Proc. 1st Int. Conf. 5G Ubiquitous Connectivity (5GU), Äkäslompolo,
Finland, 2014, pp. 192–197.
[32] A. Richter and R. S. Thoma, ‘‘Joint maximum likelihood estimation
of specular paths and distributed diffuse scattering,’’ in Proc. IEEE
61st Veh. Technol. Conf. (VTC-Spring), vol. 1. Stockholm, Sweden,
May/Jun. 2005, pp. 11–15.
[33] M. Landmann, M. Kaske, and R. S. Thoma, ‘‘Impact of incomplete and
inaccurate data models on high resolution parameter estimation in multi-
dimensional channel sounding,’’ IEEE Trans. Antennas Propag., vol. 60,
no. 2, pp. 557–573, Feb. 2012.
[34] ‘‘Spatial channel model for multiple input multiple output (MIMO) simu-
lations, v11.0.0,’’ 3GPP, Tech. Rep. 3GPP TR 25.996, 2012.
[35] A. A. M. Saleh and R. A. Valenzuela, ‘‘A statistical model for indoor multi-
path propagation,’’ IEEE J. Sel. Areas Commun., vol. 5, no. 2, pp. 128–137,
Feb. 1987.
[36] M. Steinbauer, H. Ozcelik, H. Hofstetter, C. F. Mecklenbrauker, and
E. Bonek, ‘‘How to quantify multipath separation,’’ IEICE Trans.
Electron., vol. E85-C, no. 3, pp. 552–557, Mar. 2002.
[37] S. R. Saunders and A. Aragón-Zavala, Antennas and Propagation for
Wireless Communication Systems. Hoboken, NJ, USA: Wiley, 2007.
[38] Fraunhofer Heinrich Hertz Institute. (2016). The Implementation of
Quasi Deterministic Radio Channel Generator (QuaDRiGa), v1.4.1-551.
[Online]. Available: http://www.quadriga-channel-model.de
XIONGWEN ZHAO (SM’06) received the Ph.D.
degree (Hons.) from Helsinki University of Tech-
nology (TKK), Finland, in 2002.
From 1992 to 1998, he was a Director and a
Senior Engineer with the Laboratory of Communi-
cations System Engineering, China Research Insti-
tute of Radiowave Propagation, Beijing, China.
From 1999 to 2004, he was a Senior Researcher
and a Project Manager with the Radio Laboratory,
TKK, in the areas of MIMO channel modeling and
measurements at 2, 5, and 60GHz and UWB. From 2004 to 2011, he was with
Elektrobit Corporation (EB), Espoo, Finland, as a Senior Specialist at EB
Wireless Solutions. From 2004 to 2007, he was with European WINNER
Project as a Senior Researcher in MIMO channel modeling for 4G radio
systems. From 2006 to 2008, he also with the field of wireless network
technologies, such as WiMAX and wireless mesh networks (WMNs). From
2008 to 2009, he was in satellite mobile communications for GMR-1 3G,
DVB-SH RF link budget, and antenna performance evaluations. From 2010
to 2011, he was involved in the spectrum-sharing and interference manage-
ment between satellite and terrestrial LTE networks. He is currently a Full
Professor of Wireless Communications with North China Electric Power
University, Beijing, China, and chairs several projects by the National Sci-
ence Foundation of China (NSFC), the State Key Laboratories and Industries
on channel measurements, modeling, and simulations.
Prof. Zhao is a Reviewer of the IEEE transactions, journals, letters, and
conferences papers. He received the IEEE Vehicular Technology Society
Neal Shepherd Best Propagation Paper Award in 2014. He has served as
the TPC Member, Session Chair, and a Keynote Speaker for numerous
international and national conferences.
SHU LI received the B.Sc. degree in electronic
information technology from North China Elec-
tric Power University, Beijing, China, in 2012,
where he is currently pursuing the Ph.D. degree in
electrical engineering and information technology.
His current research interests include 3D-MIMO
channel modeling, mmWave channel model-
ing, high-resolution parameter extraction algo-
rithm, time-evolution channels, and D2D
communications.
QI WANG received the B.S. degree from North
China Electric Power University, Beijing, China,
in 2012, where she is currently pursuing the
Ph.D. degree. Her recent research interests include
mmWave communication, massive MIMO chan-
nel modeling, and human blocking modeling.
MENGJUN WANG received the M.S. degree in
communication and information systems from the
China Academy of Telecommunication Technol-
ogy (CATT). He is currently a Senior Engineer
with CATT. His research fields include mmWave
mobile communications, MIMO technology, and
heterogeneous wireless networks.
SHAOHUI SUN was born in Shaoguan, China,
in 1972. He received the M.S. degree in computer
engineering and the Ph.D. degree in communica-
tion and information systems from Xidian Univer-
sity, Xi’an, China, in 1999 and 2003, respectively.
From 2003 to 2006, he was a Post-Doctoral
Fellow with the China Academy of Telecommuni-
cation Technology (CATT), Beijing, China. From
2006 to 2010, he was with the Datang Mobile
Communications Equipment Ltd., Beijing, where
he has been deeply involved in the development and standardization of the
Third-Generation Partnership Project Long-Term Evolution (3GPP LTE).
Since 2011, he has been the Chief Technical Officer with the Datang Wire-
less Mobile Innovation Center, CATT. His current research area of interest
includes multiple antenna technology, heterogeneous wireless networks, and
relay.
WEI HONG (M’92–SM’07–F’12) received the
B.S. degree from the University of Information
Engineering, Zhengzhou, China, in 1982, and the
M.S. and Ph.D. degrees from Southeast University,
Nanjing, China, in 1985 and 1988, respectively, all
in radio engineering.
Since 1988, he has been with the State Key
Laboratory of Millimeter Waves, serving as the
Director of the Laboratory since 2003, and is cur-
rently a Professor and the Dean of the School of
Information Science and Engineering, Southeast University. In 1993, and
from 1995 to 1998, he was a short-term Visiting Scholar with the University
of California at Berkeley and at Santa Cruz, CA, USA, respectively. He has
been engaged in numerical methods for electromagnetic problems, mmWave
and THz theory and technology, antennas, and RF technology for wireless
communications. He has authored and co-authored more than 300 technical
publications and two books.
Dr. Hong is a Fellow of CIE, MTT-S AdCom Member (2014-2016),
Vice-President of the Microwave Society and Antenna Society of CIE,
Chairperson of the IEEE MTT-S/AP-S/EMC-S Joint Nanjing Chapter. He
has served as the Associate Editor of the IEEE TRANSACTIONS ON MICROWAVE
THEORY and TECHNIQUES from 2007 to 2010, and an Editor board member
for IJAP, China Communications, and the Chinese Science Bulletin. He was
twice awarded the National Natural Prizes (second and fourth class), thrice
awarded the First-Class Science and Technology Progress Prizes from the
Ministry of Education of China and Jiangsu Province Government. He also
received the Foundations for China Distinguished Young Investigators and
for Innovation Group issued by NSF of China.
1072 VOLUME 5, 2017