Content uploaded by Minseok Kim
Author content
All content in this area was uploaded by Minseok Kim on Mar 07, 2025
Content may be subject to copyright.
Available via license: CC BY 4.0
Content may be subject to copyright.
3598 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 74, NO. 3, MARCH 2025
Millimeter-Wave Urban Cellular Channel
Characterization and Recipe for High-Precision
Site-Specific Channel Simulation
Hibiki Tsukada , Member, IEEE, Naoya Suzuki , Banibrata Bag, Member, IEEE, Riku Takahashi,
and Minseok Kim , Senior Member, IEEE
Abstract—To design a reliable communication system leveraging
millimeter-wave (mm-wave) technology-gaining popularity for its
capacity to deliver multi-gigabit-per-second data rates-it is essen-
tial to consider the site-specific nature of mm-wave propagation.
Conventional site-general stochastic channel models often fail to
reproduce channel responses in specific usage scenarios or environ-
ments accurately. To enable high-precision channel simulation that
captures site-specific characteristics, this paper proposes a channel
modeling framework leveraging the widely accepted 3GPP map-
based hybrid channel modeling approach and provides a detailed
methodology for applying it to real-world scenarios with practical
examples. First, an extensive measurement campaign was con-
ducted in typical urban macro and micro cellular environments us-
ing an in-house dual-band (24/60 GHz) double-directional channel
sounder. The mm-wave channel behavior was then characterized,
with a focus on differences between the two frequencies. Following
this, site-specific large-scale and small-scale channel properties
were parameterized. As a critical component for enhancing predic-
tion accuracy, this paper introduces an exponential decay model for
the power-delay characteristics of non-line-of-sight clusters, which
are often significantly overestimated by deterministic prediction
tools. Finally, using the in-house channel model simulator (CP-
SQDSIM) developed for grid-wise channel data (PathGridData)
generation, a marked improvement in prediction accuracy over
the existing 3GPP map-based channel model was demonstrated.
Index Terms—5G mobile communication, channel model
simulator, hybrid channel model, millimeter wave (mm-wave),
outdoor urban radio channels, wireless emulator.
I. INTRODUCTION
WITH a projected 50 billion devices worldwide by 2030,
communication between them is more prevalent than
Received 10 February 2024; revised27 July 2024; accepted 2 November 2024.
Date of publication 6 November 2024; date of current version 5 March 2025. This
work was supported by the Ministry of Internal Affairs and Communications
of Japan through “R&D for the Realization of High-Precision Radio Wave
Emulator in Cyberspace ” under Grant JPJ000254. The review of this article
was coordinated by Prof. Mugen Peng. (corresponding author: Minseok Kim.)
Hibiki Tsukada is with NTT Access Network Service Systems Laboratories,
Kanagawa 239-0847, Japan.
Naoya Suzuki is with NEC Corporation, Kanagawa 211-8666, Japan.
Banibrata Bag is with the Department of Electronics and Communication
Engineering, Haldia Institute of Technology, Haldia 721657, India.
Riku Takahashiis with Mitsubishi Electric Corporation, Kanagawa 247-8501,
Japan.
Minseok Kim is with the Graduate School of Science and Technology, Niigata
University, Niigata 950-2181, Japan (e-mail: mskim@eng.niigata-u.ac.jp).
Digital Object Identifier 10.1109/TVT.2024.3492719
ever. This surge in traffic volume leads to spectral efficiency
degradation and frequency spectrum congestion, and mutual
interference is predicted to become even more severe. Therefore,
to address these challenges and meet the ever-increasing demand
for improving data rates as well as personalized experiences con-
cerning quality-of-services (QoS), utilizing the millimeter-wave
(mm-wave) band is indispensable. The new radio access tech-
nology (RAT) using mm-wave bands (24–71 GHz has been iden-
tified for IMT frequencies in WRC-19) can provide enormous
bandwidth, low latency, and ultra-high data rate in the range of
Gbps, which makes mm-wave technology suitable for a variety
of advanced services and future applications. However, while
mm-wave bands offer various advantages mentioned above, the
current fifth-generation (5G) technology is still struggling to
fully utilize them due to their unique propagation characteristics,
which differ significantly from those of low-frequency bands [1].
Developing accurate channel models that reflect the diverse
effects of specific usage scenarios or environments requires
extensive measurement data collection across various scenar-
ios and frequencies. To date, numerous research groups and
organizations worldwide have undertaken campaigns to measure
radio characteristics across a wide range of mm-wave bands
and in different scenarios. For instance, extensive propagation
measurements in both outdoor and indoor urban environments
at various frequencies, including 28 GHz, 38 GHz, 60 GHz,
and 73 GHz, have been conducted [1],[2],[3], and mm-wave
propagation channel characteristics in urban scenarios at 28 GHz
have been investigated [4],[5]. Many other groups [6],[7],[8]
are similarly involved in measurement campaigns investigating
mm-wave propagation channel characteristics in various sce-
narios and channel parameters for new wireless communication
technologies such as beamforming and massive MIMO; these
efforts contribute to developing a suitable channel model for
5G or beyond [9],[10],[11],[12],[13]. However, further study,
such as site-specific behavior and frequency-dependent charac-
teristics, should be explored to make full use of the mm-wave
bands.
Again, the utilization of mm-wave bands can realize a sig-
nificantly higher data rate due to their abundant unexploited
bandwidth. However, the coverage area of these bands is typ-
ically constrained by extremely large propagation loss [14],
[15]. Furthermore, the propagation characteristics in mm-wave
© 2024 The Authors. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see
https://creativecommons.org/licenses/by/4.0/
TSUKADA et al.: MILLIMETER-WAVE URBAN CELLULAR CHANNEL CHARACTERIZATION AND RECIPE FOR HIGH-PRECISION 3599
bands are highly environment-dependent, with communication
links easily obstructed by small obstacles such as trees, cars,
and signboards along streets. In addition, the mm-wave signal
gets diffused through scattering by a very tiny object or rough
surfaces of a large object [14],[15],[16]. Consequently, existing
channel models designed for lower-frequency bands cannot sat-
isfactorily predict the propagation channel’s behavior with high
precision for the mm-wave bands. Therefore, developing a more
suitable channel model for mm-wave new RATs is essential [17],
striking a balance between computational complexity and pre-
diction accuracy. Moreover, a simulation tool is necessary for
system verification and evaluation.
In recent years, various organizations and research communi-
ties have developed different approaches for designing and sim-
ulating channel models for cellular networks and local area net-
works (LANs). Some widely accepted channel models include
3GPP [18],[19], WINNER [20], COST2100 [21], METIS [22],
QuaDRiGa [23], NYUSIM [12], and MiWEBA [24],[25]. While
each of these channel modeling approaches adheres to its own
standards, highlighting unique priorities and strengths, they
sometimes align with other methods. Channel models can gen-
erally be categorized into stochastic, deterministic, and hybrid
models. The stochastic approach, also known as the geometry-
based stochastic channel model (GSCM), such as WINNER,
3GPP, COST2100, and so forth, has been widely adopted for
conventional wireless communication systems due to its ease
of implementation. However, it cannot provide high-precision
reproducibility capturing the effects of specific usage scenarios
or environments in mm-wave new RATs [3]. Meanwhile, the
deterministic channel model approach, such as a ray tracing (RT)
based modeling, utilizes a delicate or simplified 3D geometric
description of the propagation environment to produce channel
responses in a specific environment [26],[27],[28],[29]. While
these approaches are promising for accurate site-specific predic-
tions, they are not always convenient due to their computational
complexity and limited accuracy when obtaining complex mi-
croscopic interactions for a large-scale propagation area [11],
[30],[31],[32].
To address the limitations of purely stochastic or deterministic
channel models, hybrid models have been developed to combine
both approaches, simplifying the modeling of complex scattered
multipaths while preserving accuracy in environment-specific
paths generated by specular reflections. This approach enables
the inclusion of deterministic components from prominent en-
vironmental structures, such as the ground, trees, and building
walls, alongside non-deterministic components from smaller,
unpredictable indoor or outdoor objects whose locations are dif-
ficult to anticipate. Some well-known hybrid channel models are
QuaDRiGa [33], 3GPP map-based model [19], and Q-D channel
model [34],[35]. In these models, obtaining deterministic com-
ponents using ray tracing is straightforward, but incorporating
attenuation caused by diffuse scattering in mm-wave propaga-
tion is challenging. To address this issue, the ray tracing calibra-
tion method is proposed to incorporate various types of diffuse
scattering models in ray tracing calculation [36]. However, this
kind of approach can not be very practical due to the difficulty of
the complicated setup of the model. In [36], ray tracing results
are compensated with object-specific diffuse scattering models
against individual rays by distinguishing surface materials of the
3D model and setting scattering parameters for each material.
However, such a detailed setup can be especially challenging
in macro environments. This paper, therefore, proposes a cali-
bration method, utilizing an exponential decay model for power
delay characteristics of non-line-of-sight (NLoS) clusters, which
is a novel and feasible idea to incorporate the non-negligible
effects of diffuse scattering into the deterministic components
to improve the accuracy.
Before implementing a new wireless communication system,
it is essential to thoroughly analyze site-specific propagation be-
haviors and assess its link-level and system-level performances.
However, large-scale system verification for advanced radio
equipment and wireless technology by deploying hundreds of
wireless devices in the target area can be time-consuming and
costly. Furthermore, this approach may not ensure reproducibil-
ity due to possible changes in environment and condition, render-
ing it impractical. Thus, it is crucial to develop a high-precision
radio wave simulator/emulator that can accurately estimate and
verify new radio system behaviors, which significantly reduces
the time and costs associated with the process while allowing
for variations in scenarios, network layouts, and other required
specifications [37]. Therefore, this study considers a hybrid
channel model with site-specific channel representation (SSCR)
exploiting geometric data, such as 2D maps and 3D CAD models
of specific environments for high-precision channel simulation.
A. Contributions
As described above, this paper proposes a Q-D channel
model framework for high-precision channel simulation reflect-
ing site-specific characteristics, leveraging a widely accepted
3GPP map-based hybrid channel modeling approach. Specif-
ically, to improve prediction accuracy, a calibration method of
deterministic components is proposed. It also provides a detailed
recipe for applying it to an actual scenario. The paper’s primary
contributions are as follows:
rAn extensive double-directional (D-D) channel measure-
ment campaign was conducted in typical urban cellu-
lar environments, including urban macrocell (UMa) and
microcell (UMi) scenarios, at two different carrier fre-
quencies (24 and 60 GHz) using an in-house chan-
nel sounder. Most of the existing multi-frequency mea-
surements were sequentially conducted at a single fre-
quency, and dual-frequency simultaneous measurements
are scarce. Joint clustering with composite datasets of 24
and 60 GHz, which can extract common clusters existing
at both frequencies, enables the characterization of the
cluster-level frequency-dependentbehavior. From the mea-
surement data, the site-specific channel parameters, such as
large-scale parameters (LSPs) and small-scale parameters
(SSPs), are extracted.
rA measurement-based exponential decay model for the
power delay characteristics of non-line-of-sight (NLoS)
clusters is proposed, which is essential for improving
prediction accuracy. In ray tracing simulation, the powers
3600 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 74, NO. 3, MARCH 2025
of the NLoS clusters are significantly overestimated due
to a lack of consideration of actually existing additional
interaction losses caused by shadowing by small objects
or diffuse scattering, which is impractical to account for
deterministically. The proposed model can provide an effi-
cient calibration tool for ray-tracing results, which is incor-
porated into the channel model to reproduce site-specific
mm-wave channel characteristics more accurately.
rA high-precision, site-specific channel model framework
has been developed. In this framework, calibrated deter-
ministic paths-derived from ray tracing results using the
proposed method-are combined with random paths gen-
erated from measured site-specific statistical parameters
of LSPs and SSPs to produce accurate channel responses.
Additionally, the spatial consistency procedure in the 3GPP
model is also applied using the measured site-specific
correlation distances [38]. An in-house channel model
simulator, cyber-physical system Q-D wireless channel
simulator (CPSQDSIM), was also developed to generate
grid-wise channel data (so-called PathGridData). It can
efficiently and accurately produce radio wave behaviors
tailored to any specific environment.
rThe proposed channel model was validated against mea-
sured data in two distinct environments at carrier frequen-
cies of 24 and 60 GHz. A comparison of the statistical
characteristics between the proposed channel model and
the existing 3GPP map-based channel model demonstrates
a substantial improvement in accuracy with the proposed
model.
B. Organisation
The subsequent parts of this paper are organized as follows:
Section II provides a comprehensive description of the
site-specific channel modeling concept and the methodology
involved. This section also details the channel sounder
used in the measurement campaigns. Section III covers the
post-processing and channel characterization results. Section IV
describes the extraction of channel parameters and introduces
the recipe parameters to capture site-specific characteristics.
Section Vdescribes the extension of the 3GPP map-based
model. Section VI introduces the channel model simulator,
CPSQDSIM, along with the implementation of the spatial
consistency procedure. Additionally, this section presents
some simulation results that illustrate the proposed model’s
performance. Finally, Section VII concludes the paper.
II. METHODOLOGY
A. Site-Specific Channel Modeling Concept
This paper considers a Q-D channel model for the SSCR as
illustrated in Fig. 1. In this model, dominant paths are generated
deterministically by simplified ray tracing in the first step. Then,
the second step creates the centroids of the random clusters
stochastically for the resulting channel responses to exhibit the
measured site-specific inter-cluster properties (LSPs). In the
third step, the complete shape of the clusters is determined by
Fig. 1. Quasi-deterministic (Q-D) channel model concept.
Fig. 2. Evaluation models: Area1 (UMa) and Area2 (UMi).
applying the power spread stochastically, where some multipath
components (MPCs) are added around each path or cluster
centroid in delay and angle domains to capture the measured
site-specific intra-cluster properties (SSPs).
In this approach, the site-specific propagation characteristics
of a given environment are accurately represented. Advanta-
geously, the deterministic clusters obtained by ray tracing are
inherently spatially consistent because the evolution of angles
and delays are calculated based on the geometry of the environ-
ment. Further, the channel model also generates spatially con-
sistent random clusters following the 3GPP spatial consistency
procedure with measured site-specific correlation distance [38].
B. Evaluation Model Environments and Measurement
Campaign
Our goal is to develop a radio channel simulation technique
for a CPS wireless emulator that can replicate a communication
environment within a virtual space and simulate site-specific
radio propagation channels with high precision, significantly
reducing the need for extensive field testing of wireless devices
in real physical space [37]. In this study, the channel models
are developed for two evaluation model environments in typical
urban areas of Yokohama, Kanagawa, Japan: a UMa scenario
around a railway station, JR Kannai (Area1), and a UMi sce-
nario near a shopping mall, Yokohama World Porters (Area2)
as shown in Fig. 2. The channel measurement campaigns were
TSUKADA et al.: MILLIMETER-WAVE URBAN CELLULAR CHANNEL CHARACTERIZATION AND RECIPE FOR HIGH-PRECISION 3601
Fig. 3. Measurement setup of each evaluation area. (a) Area1 (UMa).
(b) Area2 (UMi).
conducted in the same environments. Aerial maps of these areas,
indicating the measurement points, are shown in Fig. 3.
In Area1 (UMa), measurement data were taken at 49 distinct
points along the sidewalk as marked in Fig. 3(a) [16],[39].The
transmitter (Tx) as a base station (BS) was installed on the roof of
an eight-story building, with a BS antenna height of 31.0 m. The
receiver (Rx), representing a mobile station (MS), was placed
at each marked point, with an MS antenna height of 1.5 m. The
distance between the BS and MS antennas varied from 40 to
350 m. In Area2 (UMi), measurement data were taken at 25
distinct points along the sidewalk, as shown in Fig. 3(b) [39].
Here, the BS antenna was mounted at a height of 3.0 m on the
sidewalk, while the MS antenna height was 1.5 m at each marked
location. In Fig. 3(b), the measurement points from WP9 to WP23
are closely spaced for spatial consistency measurement [38].
The distances between the BS and MS antennas ranged from
27 to 105 m. Note that the measurement data may be subject to
some influence from nearby vehicles and pedestrians at the MS
locations.
C. Channel Sounding
Fig. 4shows the in-house 24/60 GHz dual-band D-D channel
sounder used in this study [39],[40]. The measurement system
comprises dual-band radio frequency (RF) heads and baseband
(BB) processing units. Here, the dual-band RF heads include
commercial off-the-shelf (COTS)-phased array antenna beam-
forming transceivers (EVK02001 for 24 GHz and EVK06002
for 60 GHz, SIVERS IMA [41]). An RF head consists of four
phased array antennas with 90◦azimuth coverage directing
toward −135◦,−45◦,+45◦, and +135◦for a full azimuth angle
sweep. By stacking the 24 GHz RF heads on the 60 GHz
RF heads and employing a dual 4 ×4 MIMO time division
multiplexing (TDM) scheme, the system enables simultaneous
measurement of 32 channels (16 for 60 GHz and 16 for 24 GHz,
respectively) [39],[40].
The 60 GHz transceiver circuit consists of a 16-element uni-
form linear array (ULA), whereas the 24 GHz transceiver circuit
consists of a 2-by-8-element uniform planar array (UPA). Both
transceivers synthesize narrow beam patterns with a half-power
TAB L E I
SYSTEM PARAMETERS
beamwidth (HPBW) of approximately 6◦and 15◦within an
azimuth range of ±45◦. Considering the beam’s HPBW and
patterns, a 90◦azimuth angle sweep was achieved by using five
beams for each Tx and Rx array at 24 GHz, whereas 11 beams
for each Tx array and 12 beams for each Rx array at 60 GHz.
Using four antenna arrays can achieve a 360◦full azimuth angle
sweep. The Tx power is approximately 32 and 41 dBm in terms
of equivalent isotropic radiated power(EIRP) f or 24 and 60 GHz,
respectively. The system parameters in each frequency band are
presented in Table I. Further details on the channel-sounding
system are provided in the authors’ previous works [39],[40].
III. MM-WAV E DUAL-BAND CHANNEL CHARACTERIZATION
This section presents the mm-wave channel characteristics
at two different frequencies obtained in the measurement cam-
paigns conducted in the evaluation model environments [16],
[38],[39] described above.
A. Post-Processing
1) Multipath Component (MPC) and Cluster Extraction: In
the measurements, band-limited, D-D angle resolved channel
transfer functions (DDCTF) denoted by Hk,nT,nRwere ob-
tained. The D-D angle resolved channel impulse response (DD-
CIR) can be obtained by applying the Inverse Fourier Transform
of the DDCTF as follows:
h(ˇτ, ˇ
φT,ˇ
φR)=F−1{H(ˇ
f, ˇ
φT,ˇ
φR)},(1)
where the delay tap is represented by ˇτ∈{nΔτ|n=
0,...,N −1}, where Δτ=1/W, with Nand Wbeing the
number of frequency tones and the bandwidth of the sounding
signal (multitone), respectively. The frequency bin (tone or
sub-carrier) is denoted by ˇ
f∈{kΔf|k=0,...,N −1}, where
Δf=W/N. The transmitting and receiving pointing angles
are represented by ˇ
φT∈{nTΔφT|nT=0,...,N
T−1}and
ˇ
φR∈{nRΔφR|nR=0,...,N
R−1}, respectively. Here, ΔφT
and ΔφRindicate the scanning intervals of the Tx and Rx beams.
Then, the D-D angular delay power spectrum (DDADPS) is
3602 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 74, NO. 3, MARCH 2025
Fig. 4. Mm-wave dual-band (24/60 GHz) double-directional channel sounder configuration. (a) Transmitter circuit. (b) Receiver circuit.
obtained as
P(ˇτ, ˇ
φT,ˇ
φR)=E|h(ˇτ, ˇ
φT,ˇ
φR)|2,(2)
where Erepresents the expectation operator. After noise-
filtering the DDADPS, the omnidirectional power delay profile
(PDP), and the angular power spectrum (APS) are synthesized to
visually represent the power distribution of the delay and angular
domains as
PDP(ˇτ)=
ˇ
φT,ˇ
φR
P(ˇτ, ˇ
φT,ˇ
φR),(3)
APS(ˇ
φx)=
ˇτ,ˇ
φy
P(ˇτ, ˇ
φT,ˇ
φR),(4)
where x∈{T,R}and y∈{T,R}\x.
In addition, the MPCs were extracted using the in-house
Sub-grid CLEAN algorithm [16],[40],[42], which is a succes-
sive interference cancellation (SIC) method to obtain multipath
parameters by sequentially subtracting an image of an MPC
from the DDADPS in the order of the power magnitude. The
replica is created from the continuous function of the beam
pattern and the signal autocorrelation function [43].Inthis
study, the angular and delay resolutions of the Sub-grid CLEAN
algorithm were set to 0.1◦and 0.01 ns, respectively. Clustering
was then applied to the extracted MPCs to group them based
on similar angle and delay parameters. Grouping MPCs in this
way helps to identify clusters of scattered waves that likely
originate from the same physical objects or interactions in
the environment. This approach provides a more realistic and
physically meaningful model of the wireless channel. Clustering
was performed on the composite datasets of the MPCs obtained
at each frequency, as shown in Fig. 5. To compensate for the
increased propagation loss at 60 GHz due to the frequency
difference, the power of each 60 GHz MPC was scaled up by
20 log10(58.32/24.15)=7.66 [dB]. After clustering, the MPCs
were separated back into their respective frequency datasets, and
the power of each 60 GHz MPC was down-scaled to its original
value. This process enabled the identification of common clus-
ters present at both frequencies, as well as uncommon (unique)
clusters found only at one frequency. Here, the K-PowerMeans
Fig. 5. Extraction of common and uncommon clusters at 24 and 60 GHz.
algorithm was employed for clustering, with the number of
clusters, K, manually set through visual inspection to ensure
the results remained relevant to the physical scattering objects.
B. Observation of Frequency Dependence in Scattering
Processes
As described above, the common clusters obtained by using
two different datasets simultaneously measured at two different
carrier frequencies (24 and 60 GHz) enabled some interesting
observations on the cluster-level frequency-dependent behavior.
Fig. 6shows the omnidirectional PDPs obtained at NW7 of
Area1 and WP17 of Area2 in as examples. Here, the cutoff
level for noise-filtering for the 60 GHz band was set by the value
of 7.66 dB (as much as the additional propagation loss) lower
than that for the 24 GHz band to maintain the same detection
range. The results of NW7 show 13 and 5 clusters in the 24
and 60 GHz bands, respectively, with five common clusters, and
those of WP17 show 27 and 21 clusters in the 24 and 60 GHz
bands, respectively, with 18 common clusters. Note that all the
TSUKADA et al.: MILLIMETER-WAVE URBAN CELLULAR CHANNEL CHARACTERIZATION AND RECIPE FOR HIGH-PRECISION 3603
Fig. 6. Comparison of PDPs obtained at two different frequencies in two different environments. FSPG denotes free space path gain. (a) NW7@24 GHz (Area1).
(b) NW7@60 GHz (Area1). (c) WP17@24 GHz (Area2). (d) WP17@60 GHz (Area2).
clusters in the 60 GHz band also exist in the 24 GHz band in
NW7, but this is not always the case like in WP17. It is seen in
NW7 that several clusters with long delays exist in the 24 GHz
band, whereas in the 60 GHz band, most of the clusters have
short delays. That is because the clusters with long propagation
distances are not observed in the 60 GHz band due to significant
propagation loss. As can be seen from Fig. 6(b), there is only one
cluster of #7 in the 60 GHz band with a longer delay than #3
due to significant attenuation, which is a significant difference
from the 24 GHz band shown in Fig. 6(a).InWP17, two PDPs in
Fig. 6(c) and (d) show similar delay distribution, but the powers
of the NLoS clusters at 60 GHz significantly decreased due to
additional interaction loss as in NW7.
We further investigated scattering processes from the clus-
tering results to understand the frequency dependence of the
two frequency bands. Table II lists some selected common
clusters with the path gains, the power difference between the
two frequencies, and the excess loss. The excess loss is obtained
by subtracting 7.66 [dB] from the power difference. Regard-
ing NW7, the power differences are greater than the additional
propagation loss of 7.66 dB except #3. From the results, the
interaction loss in the 60 GHz band is significantly greater than
in the 24 GHz band. A similar trend is observed in WP17.It
TAB L E I I
CHARACTERISTICS OF COMMON CLUSTERS (UNIT:DB)
was also seen that the powers of NLoS clusters are significantly
attenuated due to significant interaction loss by various shapes of
small objects and various surface conditions of large scatterers.
C. Channel Characterization
This section provides a statistical analysis to investigate gen-
eral trends underlying the measurement data.
1) Cluster Power: Fig. 7(a) depicts the cumulative distribu-
tion function (CDF) of the total cluster power, excluding the LoS
power at each Rx point. Here, the values of Area1 (UMa) were
3604 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 74, NO. 3, MARCH 2025
Fig. 7. Statistical properties of clusters. (a) Total cluster power (LoS power excluded). (b) Cluster power difference. (c) Cluster delay.
adjusted by the beam pattern in the vertical plane relative to the
Rx position. The average value of each frequency in the UMa
scenario is −124.34 dB at 24 GHz and −115.06 dB at 60 GHz.
This results in a power difference of 9.28 dB, which is 1.62 dB
higher than the additional free space propagation loss (FSPL)
of 7.66 dB, on average. In contrast, Area2 (UMi) exhibits an
average values of −87.18 dB and −99.27 dB at 24 and 60 GHz,
respectively. This results in a power difference of 12.09 dB,
which is 4.43 dB greater than the additional FSPL. This might
be because of the more significant interaction loss and oxygen
absorption loss in the 60 GHz band.
2) Cluster Power Difference: The CDF of the power differ-
ence between two frequencies for each common cluster across
all Rx points is presented in Fig. 7(b). The results indicate that
the power difference of the LoS cluster is approximately equal to
the additional FSPL, as indicated by the dashed line. However,
the power of the LoS cluster fluctuates due to fading, which is
caused by interference from ground reflections and potentially
other local scattering sources. In Area2 (UMi), specific points,
particularly near the intersection, may partially block the first
Fresnel zone by pedestrians or traffic signs and lights. The
power difference distribution of the NLoS clusters is widely
spread, ranging from negative to large positive values. These
negative values indicate that the power is more significant in the
60 GHz band, possibly due to some interacting objects, such
as curved walls and small objects around the MS. However,
further investigation is necessary to elucidate the underlying
mechanisms. One observation is that the power difference in
common clusters varies considerably with the geometry of the
scatterer and the position of Tx and Rx.
3) Cluster Relative Delay: Fig. 7(c) illustrates the relative
delay of clusters with respect to the LoS delay. In Area1
(UMa), the common clusters exhibit a mean relative delay of
approximately 66 ns, indicating shorter delays. The uncommon
clusters at 24 GHz show delays similar to those at 60 GHz within
a shorter delay range, around 370 ns, with the probability rising
rapidly as the delay increases. However, the uncommon clusters
at 24 GHz are more widely distributed, with delays extending
beyond 370 ns. This can be attributed to the lower attenuation in
the 24 GHz band compared to the 60 GHz band, which permits
longer propagation paths involving multiple bounce reflections
from distant buildings relative to the Rx. The trend in the distri-
bution of common and uncommon clusters is similar in Area2
(UMi) as well. Additionally, in Area2, the antenna heights at
both the Tx and Rx are lower than the surrounding buildings.
This configuration results in more arrival paths with significant
power originating from scatterers located farther from the Rx,
leading to a greater relative delay compared to Area1 (UMa).
4) Cluster Number: Fig. 8presents the number of clusters
at each Rx point as a function of the Tx-Rx separation dis-
tance, along with a linear regression model obtained by the
least squares method for each frequency. Both Fig. 8(a) and
(b) indicate that the number of clusters is higher in the 24 GHz
band compared to the 60 GHz band. Again, this is because the
60 GHz band experiences significantly higher interaction losses.
The slope of the regression line is negative for all environments
and frequencies, indicating that the number of clusters decreases
as distance increases.
IV. SITE-SPECIFIC MODEL PARAMETER EXTRACTION
This section details the calculation method for extracting
channel parameters and presents the results. In the site-specific
channel model developed in this study, the channel parame-
ters are categorized into LSPs and SSPs. LSPs represent the
spatiotemporal power distribution across clusters, while SSPs
describe the spatiotemporal power distribution of MPCs within
each cluster. In the 3GPP map-based model, these parameters are
defined as site-general parameters for categorized typical sce-
narios, such as urban and suburban areas. However, these gen-
eralized parameters are often insufficient for accurately repro-
ducing channel responses in specific usage scenarios or unique
environments. Therefore, this study replaces several important
parameters among those defined as LSPs and SSPs in the 3GPP
map-based model with the measured values, thereby deriving
site-specific channel parameters. Furthermore, this section also
introduces a specific set of parameters called Recipe Parameters
(RPs), which are proposed to enhance the accuracy of the model.
A. Large-Scale Parameters (LSPs)
According to the calculation method defined in the 3GPP
model [19], the delay spread (DS) is calculated as
DS =ˇτ(ˇτ−μτ)PDP(ˇτ)
ˇτPDP(ˇτ)(5)
TSUKADA et al.: MILLIMETER-WAVE URBAN CELLULAR CHANNEL CHARACTERIZATION AND RECIPE FOR HIGH-PRECISION 3605
Fig. 8. Cluster numbers vs. distance. (a) Area1 (UMa). (b) Area2 (UMi).
where PDP(ˇτ)denotes the omnidirectional PDP obtained in (3),
and the delay mean μτis calculated as
μτ=ˇτˇτPDP(ˇτ)
ˇτPDP(ˇτ).(6)
Next, the azimuth spread of departure (ASD) is calculated as
ASD =
−2ln ˇ
φTexp (j ˇ
φT)APST(ˇ
φT)
ˇ
φTAPST(ˇ
φT)(7)
where APSTdenotes the azimuth power spectrum (APS) at Tx,
and is obtained in (4). The azimuth spread of arrival (ASA) can
be calculated in the same way. The K-factor (K) is derived from
the power ratio between the LoS path and the NLoS clusters. It
is calculated as
K[dB] = PLoS −10 log10 ⎛
⎝
NNLoS
j=1
10(PNLoS
j/10)⎞
⎠(8)
where PLoS and PNLoS
jdenote the path gains of the LoS path and
NLoS clusters in dB. The delay scaling factor, rτ, is a parameter
in the 3GPP model that characterizes the rate of power decay as
a function of cluster delay. It is defined as the ratio of the average
delay of NLoS clusters to DS as
rτ=τ
DS (9)
where τis the relative delay mean of NLoS clusters obtained as
τ=1
NNLoS
NNLoS
j=1
(τj−min (τj)) ,(10)
and τjdenotes the absolute delay of the jth NLoS cluster. This
factor is used to model the attenuation of power within each
cluster over time, accounting for the gradual decrease in power
with increasing delay. By adjusting rτ, the model can repre-
sent different decay rates for various environments, enabling a
more accurate representation of the PDP and its spatiotemporal
characteristics in site-specific scenarios.
Fig. 9shows the CDFs of the LSPs obtained from the mea-
sured data in Area1 (UMa) and Area2 (UMi) scenarios, where
Fig. 9(a)–(c) are the CDFs for the 24 GHz band, and Fig. 9(d)–(f)
are those for the 60 GHz band. Each plot in Fig. 9also includes
reference CDFs from the 3GPP model for UMa and UMi sce-
narios. The distributions for DS in UMi (Fig. 9(d)) and ASD in
UMa (Fig. 9(e)) show close alignment with the 3GPP reference
distributions. However, other parameters differ significantly
from the reference, highlighting the substantial influence of the
surrounding environment on propagation characteristics. These
discrepancies underscore the need for site-specific parameters
to adapt the channel model effectively to specific environmental
conditions.
Table III provides the extracted LSPs. The table reveals that,
while the 3GPP model suggests identical mean values for the
Rician K factor across all scenarios and frequencies, the mea-
sured Rician K factor values can differ considerably depending
on the scenario and frequency. Specifically, the mean values
of the Rician K factor at 60 GHz are higher in both envi-
ronments, suggesting greater attenuation of the NLoS clusters.
This variation highlights the need to account for frequency-
and scenario-specific conditions to enhance the accuracy of the
channel model.
B. Small-Scale Parameters (SSPs)
The intra-cluster DS (cDS) and intra-cluster ASD/ASA
(cASD/ASA) are calculated as small-scale parameters, which
were derived from the individual MPCs within each cluster as
cDS =Mn
m=1(τm−μτ)2Pm
Mn
m=1Pm
,(11)
cASD/ASA =
−2ln Mn
m=1exp (jφm)Pm
Mn
m=1Pm,(12)
where τm,φm, and Pmrepresent the delay time, depar-
ture/arrival angle (AoD or AoA), and path gain of the mth MPC,
respectively, while Mnindicates the total number of MPCs
belonging to the nth cluster.
The calculated SSPs are also presented in Table III.Itis
noteworthy that the 60 GHz band yields smaller SSP values
compared to the 24 GHz band due to increased attenuation and
decreased power dispersion at higher frequencies. Additionally,
the measured data for cASD and cASA in both scenarios and
frequencies exhibit larger values than those proposed in the
3606 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 74, NO. 3, MARCH 2025
Fig. 9. Comparison between the CDFs of the LSPs obtained from the 3GPP model and the actual measured data, where ‘Meas UMa’ and ‘Meas UMi’ in the
legend denote the results obtained in Area1 and Area2, respectively, and the models obtained by the measurement data are also plotted in the dotted lines.
(a) DS@24 GHz. (b) ASD@24 GHz. (c) ASA@24 GHz. (d) DS@60 GHz. (e) ASD@60 GHz. (f) ASA@60 GHz.
TABLE III
MODEL PARAMETERS OBTAINED FROM MEASUREMENTS (LOS)
3GPP model, implying that some environmental factors may
have contributed to the increased variation.
C. Recipe Parameters
In the 3GPP map-based model, deterministic clusters are
generated by ray tracing. For simulation in urban environments
that extend from hundreds of meters to several kilometers, a
3D model-predominantly consisting of homogeneous materials
such as concrete and glass, with a simplified structure-is used.
Section IV-A highlighted a significant point, namely, that the
surrounding environment heavily influences the characteristics
of NLoS clusters. The deterministic clusters in the channel
model, which are calculated on the simplified 3D model, of-
ten lead to significant differences from the measured channel
responses.
TSUKADA et al.: MILLIMETER-WAVE URBAN CELLULAR CHANNEL CHARACTERIZATION AND RECIPE FOR HIGH-PRECISION 3607
To achieve a more realistic and accurate representation of
deterministic clusters, detailed information on the orientation,
dimension, and dielectric properties of every building object
within the evaluation area is essential. Yet, this is usually un-
realistic in complex urban cellular environments with a signif-
icant separation distance between Tx and Rx. To address the
mentioned challenges, this study proposes an exponential decay
model for cluster power delay characteristics for the calibration
of deterministic clusters as a recipe to refine the 3GPP map-based
model. This model approximates the power attenuation with
respect to the cluster arrival as in [44].
The exponential decay model for the power delay character-
isticsofNLoSclustersisexpressedas
P(τ)[dB]=Pm(τ)+ζ(13)
for 0 <τ <τ
c, where
Pm(τ)=P0+10 log10 exp −τ
β0 (14)
where P0and β0denote the initial path gain in dB and power
decay coefficient, respectively. ζ∼N(0,σ
2
SF)is a log-normal
random variable where σSF denotes a standard deviation of
cluster shadow fading. The following procedure calculates the
model parameters. First, for each scenario and frequency, all
datasets of NLoS clusters observed at every measurement point
are merged. In practice, the system’s noise floor usually limits
the measurement’s dynamic range, significantly truncating some
clusters. This leads to a biased estimation of the decay model
parameters. The model used only the clusters in which the
delay is less than the delay threshold τc=1.5μs for both
frequencies. Here, the threshold indicates the value from which
the probability that the cluster power becomes equal to or less
than the noise level increases.
In the model parameter estimation procedure, the likelihood
function is used to obtain the probability of the lth cluster being
above the noise level, denoted by pl, which is crucial because
only a subset of clusters are observable in measurements. The
likelihood function is formulated as follows:
LF =
L
l=1
1
plσSF√2πexp −(Pl−Pm(τl))2
2σ2
SF (15)
where τland Pldenote the lth cluster’s delay and power, respec-
tively. Here, the noise path gains, Pn, are set by −140 dB at 24
GHz and −147.66 dB at 60 GHz. To determine the probability pl
of the lth cluster being above the noise level, the complementary
error function erfc(·)is utilized as
pl=1
2erfc Pn−Pl
√2σSF .(16)
Using (15), the corresponding log-likelihood function
LLF =
L
l=1ln (plσSF)+ (Pl−Pm(τl))2
2σ2
SF ,(17)
and the parameters are determined by minimizing LLF using
the maximum likelihood estimation method as
(P∗
0,β∗
0,σ
∗
SF)=arg min
P0,β0,σSF
LLF (18)
Fig. 10 presents the parameter estimation results for each
scenario and frequency. The grey markers indicate the path gains
of the NLoS clusters, while the red line shows the power decay
line drawn by using the estimated parameters. The fitting results
obtained without using a truncated normal distribution are also
displayed in the black line for comparison. Upon comparing the
results across frequencies, it is observed that β0at 60 GHz is
smaller by 12 ns and 29 ns for UMa and UMi, respectively,
compared to that at 24 GHz, indicating more significant attenu-
ation. Furthermore, σSF is about 1.5 dB larger in both scenarios,
suggesting that the propagation at the 60 GHz band is more
influenced by diffused scattering by walls or small objects, and
oxygen absorption than in the 24 GHz band. The variation in
β0between scenarios also suggests that this parameter reflects
specific environmental characteristics in modeling, such as the
building density causing propagation interactions and variations
in wall materials. As a result, applying the exponential decay
model to the calibration of the deterministic components ob-
tained by ray tracing can efficiently represent the effects of
diffuse scattering and cluster fading in the deterministic compo-
nents.
V. E XTENSION OF 3GPP MAP-BASED MODEL
As described above, stochastic site-general channel mod-
els, developed based on extensive field measurement data, are
typically employed for modeling medium- to long-range radio
propagation. However, relying on these site-general stochastic
models restricts prediction accuracy, as channel characteristics
are heavily influenced by the specific features of the individual
environment, as discussed in Section IV and in [43]. Conse-
quently, a hybrid channel model that combines deterministic
and stochastic descriptions, as mentioned above, emerges as an
effective solution.
The 3GPP model has evolved to enable performance eval-
uation of new techniques through the three-dimensional (3D)
extension, frequency extension above 6 GHz, and spatial con-
sistency procedures. Further, a hybrid channel modeling frame-
work, the 3GPP map-based model [19], is also incorporated.
Since the 3GPP model is currently the most widely accepted
standard channel model for evaluating 5G RATs, ensuring com-
patibility with the 3GPP framework is essential when developing
a new channel model. The following subsections overview the
existing 3GPP map-based model and describe the proposed
extension.
A. Existing Framework of 3GPP Map-Based Model
The 3GPP map-based model is a hybrid channel model that
supports frequencies from 0.5 to 100 GHz and covers eight
typical scenarios, including UMi and UMa. The channel model
methodology is detailed step by step in Chapter 8 of [19],
comprising 13 steps in total. Step 1) and 2) describe the setting
3608 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 74, NO. 3, MARCH 2025
Fig. 10. Power delay characteristics of NLoS clusters and the exponential decay models. (a) Area1 (UMa) @ 24 GHz. (b) Area1 (UMa) @ 60 GHz. (c) Area2
(UMi) @ 24 GHz. (d) Area2 (UMi) @ 60 GHz.
of the environment and network layout. Step 3) explains the
ray tracing method for generating deterministic clusters and
the information required for output. Step 4) to 12) describe
the method for generating random clusters from channel model
parameters. Step 13) describes the calculation of the channel
transfer function from both deterministic and random clusters.
While the developed model mainly follows the aforementioned
steps for compatibility, it modifies several points. The 3GPP
map-based model eventually generates channel coefficients, but
the developed model aims to generate a clustered MPC dataset
(PathGridData) for further processing within the CPS wireless
channel emulator [37].
B. Recipe for Reflecting Site-Specific Characteristics
1) Deterministic Cluster Calibration: In Section IV-C,we
outlined the methodology for applying the power delay decay
model of NLoS clusters as recipe parameters. This subsection
provides further details on integrating the power delay decay
model into the 3GPP model, which is crucial for accurately
reflecting the characteristics of the measured environment. It
details a statistical calibration method to adjust the power of
deterministic clusters generated by ray tracing on the simplified
3D model. An additional step is incorporated into Step 3) of the
3GPP model to facilitate this calibration process.
1) Step 3+) Power compensation for NLoS deterministic
clusters: The NLoS deterministic cluster power is obtained
by the developed decay model as (14) for each BS-MS link.
The compensation is made by using the z-score of the power
deviation from the model obtained by the standardization as
zlRT =lRT −μ
σ
(19)
for lRT =2,...,L
RT (NLoS clusters). The power difference
lRT =PlRT −Pm(τlRT )(20)
where PlRT and Pm(τlRT )denote the lRT -th deterministic cluster’s
power and the value obtained from (14), respectively. μand σ
denote the mean and standard deviation of a random variable of
(20), respectively. Finally, the compensated cluster’s power is
obtained by
P
lRT [dB] = P0+10 log10 exp −τlRT
β+zlRT ζ. (21)
TSUKADA et al.: MILLIMETER-WAVE URBAN CELLULAR CHANNEL CHARACTERIZATION AND RECIPE FOR HIGH-PRECISION 3609
Fig. 11. Block diagram of CPSQDSIM.
Fig. 12. PDP characteristics with varying Rx position from WP1 to WP25, where the PDPs generated by the models were synthesized from the angle-resolved
channel impulse responses reconstructed by using the measurement system’s characteristics. (a) PDPs. (b) Deterministic cluster generation by RT at 7 m (upper)
and 30 m (lower).
2) Site-Specific Random Cluster Generation: Random clus-
ters are generated according to the procedures outlined in [19],
utilizing site-specific LSPs and SSPs derived from measurement
data.
VI. DEVELOPMENT AND VALIDATION OF CHANNEL MODEL
GENERATOR
A. CPSQDSIM
The channel model simulator, CPSQDSIM, has been de-
veloped based on our proposed channel modeling framework,
which produces a grid-wise clustered MPC dataset called Path-
GridData. As illustrated in Fig. 11, the simulator takes deter-
ministic cluster centroids obtained via ray tracing, along with
site-specific statistical parameters, such as LSPs and SSPs,
to generate random clusters. It then creates the PathGridData
according to the proposed channel model recipe detailed in
Section V. CPSQDSIM can accept any site-specific measured
LSPs/SSPs dataset as user input, enabling accurate reproduc-
tion of radio wave propagation characteristics for the targeted
environment.
Fig. 13. PDP snapshot at the Rx position of WP3, where the PDPs generated by
the models were synthesized from the angle-resolved channel impulse responses
reconstructed by using the measurement system’s characteristics.
B. Spatial Consistency
A common problem with previous drop-based stochastic
channel models of microwave and mm-wave channels is that
3610 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 74, NO. 3, MARCH 2025
Fig. 14. Comparison of LSP (CDFs) for the 3GPP model, proposed model, and measured data at different scenarios and frequencies. (a) Area1 (UMa)@24
GHz. (b) Area1 (UMa)@60 GHz. (c) Area2 (UMi)@24 GHz. (d) Area2 (UMi)@60 GHz.
they do not allow for the temporally or spatially consistent
simulation that is required in beamforming evaluations. The
3GPP model addresses this by introducing a spatial consistency
procedure that updates each cluster’s angles and delays as the
user moves along a route, using a linear approximation process.
The spatial consistency has been implemented in CPSQDSIM
using the delay and angle generation procedure described in
3GPP Spatial Consistency Procedure A, following Step 5) and
Step 7), respectively [19].Att0=0, when the MS/BS is initially
dropped into the network, the power, delay, and angle of a
spatially consistent cluster are generated. In the subsequent time
step, designated as tk=tk−1+Δt, the power, delay, and angle
of the clusters are recalculated, considering the MS’s position in
the previous time step at tk−1, as well as those of the cluster at
tk, in conjunction with the MS’s velocity and moving direction.
The spatial consistency procedure is applied to the random
clusters by updating their delay and angle based on the move-
ment direction of the Rx. The birth and death of each ran-
dom cluster are governed by a cluster visible region, which
is determined by an exponentially distributed random variable
derived from measurement data. Each cluster is updated in the
angle and delay domains according to 3GPP Spatial Consistency
Procedure A, up to the limit of its visible region. When a random
cluster dies, a new random cluster is generated dynamically at
runtime. This process is repeated for all grid points traversed
by the Rx, resulting in spatially consistent channel responses.
As described in Section II, spatial consistency was measured
using a setup with a stationary Tx and a moving Rx, advancing
approximately 1 m along the direction indicated by the arrow.
Our previous investigation shows that, in Area2, the average
cluster birth and death distances are 5.41 m and 4.16 m in the 24
and 60 GHz bands, respectively. Additional analysis for Area1
is provided in [38].
C. Validation
To validate the developed channel model, the PDP along the
measurement route is compared with observed data. Addition-
ally, the statistical properties of DS, ASD, and ASA produced
by the proposed model and the 3GPP map-based model are eval-
uated against measurement results. Ray tracing calculations for
deterministic clusters were conducted using Wireless InSite [45]
on simplified 3D models of Area1 (UMa) and Area2 (UMi),
with the number of allowed reflections and diffractions set to
one. The channel parameters for random clusters are drawn from
Table III, and the RPs were applied exclusively to the developed
model.
Fig. 12 shows the transition of the PDP along the Rx route
in Area2, with results from the existing 3GPP map-based
model (left), the proposed model (center), and the measured
data (right). In Fig. 12(a), a snapshot of the ray tracing simu-
lation used to generate deterministic clusters is also provided
(Fig. 12(b)) for an intuitive understanding of the environment. It
should be noted that the PDPs generated by the models were
synthesized from angle-resolved channel impulse responses,
reconstructed based on the measurement system’s character-
istics. Fig. 12(a) shows that the existing 3GPP model dis-
plays unusually high power levels in the NLoS determinis-
tic clusters, highlighted by the white circles. In contrast, the
TSUKADA et al.: MILLIMETER-WAVE URBAN CELLULAR CHANNEL CHARACTERIZATION AND RECIPE FOR HIGH-PRECISION 3611
proposed model more closely matches the trends observed in
the measurement data. Specifically, Fig. 13 presents a single
PDP measured at WP3, with its location indicated in Fig. 3(b).
This comparison demonstrates that the 3GPP map-based model
significantly overestimates the power of NLoS clusters, whereas
the proposed model, through calibrated NLoS cluster power,
aligns well with the measured data. This result indicates that
the proposed model effectively captures the characteristics of
both deterministic and random clusters in agreement with the
measurements.
Fig. 14 presents the CDFs of DS, ASA, and ASD obtained
from Area1 (UMa) and Area2 (UMi) at 24 and 60 GHz. Upon
examining Fig. 14(a) and (b), it is apparent that both the DS and
ASD/ASA of the proposed model at both frequencies align more
closely with the measured data than the existing 3GPP model
in the UMa scenario. On the other hand, regarding the UMi
scenario, both the DS and ASD/ASA of the proposed model
at both frequencies are closer to the measured value than the
3GPP model, as shown in Fig. 14(c) and (d), respectively. Since
the proposed model is not a fully deterministic model but a
Q-D model, it produces channel responses that are statistically
similar to real-world measurements rather than exact replicas.
However, even so, there is a noticeable gap between the proposed
and measured distribution in the ASD of the UMi scenario at
both frequencies. This is primarily attributed to the fact that the
simplified 3D model could not perfectly include all dominant
scattering objects that were present during the measurement.
However, our future research will address this issue to improve
outcomes.
VII. CONCLUSION
To achieve high-precision channel simulations that capture
site-specific characteristics, this paper proposed a Q-D channel
model framework. Then, a detailed recipe to apply that to two
actual scenarios of Area1 (UMa) and Area2 (UMi) at two
mm-wave frequencies was presented. For channel modeling,
an extensive measurement campaign was conducted, and the
mm-wave channel behavior was characterized, focusing on the
difference between the two frequencies. The measurement re-
sults revealed a significant power decrease in the NLoS clusters.
This trend was more significant in the 60 GHz band than in
the 24 GHz band. From the results, a measurement-based ex-
ponential decay model for power delay characteristics of NLoS
clusters was proposed as an essential component for improving
prediction accuracy.
In the developed channel simulation framework, deterministic
paths obtained through ray tracing with calibration by the devel-
oped exponential decay model and measured site-specific sta-
tistical parameters of LSPs and SSPs are combined to generate
accurate channel responses. The spatial consistency procedure
in the 3GPP model was also incorporated using the measured
site-specific correlation distances [38]. Furthermore, an in-house
channel model simulator, CPSQDSIM, was developed to gener-
ate a grid-wise channel dataset, PathGridData. It can efficiently
and accurately produce radio wave behaviors for any specific
targeted environment.
The proposed channel model was validated against measured
data from two distinct environments at carrier frequencies of
24 and 60 GHz. Statistical comparisons with the existing 3GPP
map-based channel model demonstrated a substantial improve-
ment in accuracy, affirming the effectiveness of the proposed
model.
REFERENCES
[1] T. S. Rappaport et al., “Millimeter wave mobile communications for 5G
cellular: It will work!,” IEEE Access, vol. 1, pp. 335–349, 2013.
[2] 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.
[3] T. S. Rappaport, Y. Xing, G. R. MacCartney, A. F. Molisch, E. Mellios,
and J. Zhang, “Overview of millimeter wave communications for fifth-
generation (5G) wireless networks—With a focus on propagation models,”
IEEE Trans. Antennas Propag., vol. 65, no. 12, pp. 6213–6230, Dec. 2017.
[4] J. Ko et al., “Millimeter-wave channel measurements and analysis for
statistical spatial channel model in in-building and urban environments at
28 GHz,” IEEE Trans. Wireless Commun., vol. 16, no. 9, pp. 5853–5868,
Sep. 2017.
[5] S. Hur et al., “Proposal on millimeter-wave channel modeling for 5G
cellular system,” IEEE J. Sel. Topics Signal Process., vol. 10, no. 3,
pp. 454–469, Apr. 2016.
[6] M. J. Kazemi, A. Abdipur, and A. Mohammadi, “Indoor propagation
MIMO channel modeling in 60 GHz using SBR based 3D ray tracing
technique,” in Proc. 2nd Conf. Millimeter-Wave Terahertz Technol., 2012,
pp. 25–28.
[7] X. Gao, O. Edfors, F. Rusek, and F. Tufvesson, “Massive MIMO per-
formance evaluation based on measured propagation data,” IEEE Trans.
Wireless Commun., vol. 14, no. 7, pp. 3899–3911, Jul. 2015.
[8] X. Wu et al., “60-GHz millimeter-wave channel measurements and mod-
eling for indoor office environments,” IEEE Trans. Antennas Propag.,
vol. 65, no. 4, pp. 1912–1924, Apr. 2017.
[9] V. Degli-Esposti et al., “Ray-tracing-Based mm-Wave beamforming as-
sessment,” IEEE Access, vol. 2, pp. 1314–1325, 2014.
[10] P. Ferrand, M. Amara, S. Valentin, and M. Guillaud, “Trends and chal-
lenges in wireless channel modeling for evolving radio access,” IEEE
Commun. Mag., vol. 54, no. 7, pp. 93–99, Jul. 2016.
[11] K. Haneda et al., “5G 3GPP-Like channel models for outdoor urban
microcellular and macrocellular environments,” in Proc. IEEE 83rd Veh.
Technol. Conf., 2016, pp. 1–7.
[12] S. Ju and T. S. Rappaport, “Millimeter-wave extended NYUSIM channel
model for spatial consistency,” inProc. IEEE Glob. Commun. Conf., 2018,
pp. 1–6.
[13] S. Sun, T. S. Rappaport, M. Shafi, P. Tang, J. Zhang, and P.J. Smith, “Propa-
gation models and performance evaluation for 5G millimeter-wave bands,”
IEEE Trans. Veh. Technol., vol. 67, no. 9, pp. 8422–8439, Sep. 2018.
[14] S. Deng, G. Maccartney, and T. Rappaport, “Indoor and outdoor 5G
diffraction measurements and models at 10, 20, and 26 GHz,” in Proc.
IEEE Glob. Commun. Conf., Dec. 2016, pp. 1–7.
[15] A. Maltsev et al., “Characteristics of indoor millimeter-wave channel at
60 GHz in application to perspective WLAN system,” in Proc. 4th Eur.
Conf. Antennas Propag., 2010, pp. 1–5.
[16] M. Kim, N. Suzuki, H. Tsukada, and R. Takahashi, “Frequencydependence
of millimeter-wave urban macrocell multipath cluster channels,” in Proc.
Eur. Conf. Antennas Propag., Florence, Italy, Mar. 2023, pp. 1–5.
[17] O. O. Erunkulu, A. M. Zungeru, C. K. Lebekwe, M. Mosalaosi, and
J. M. Chuma, “5G mobile communication applications: A survey and
comparison of use cases,” IEEE Access, vol. 9, pp. 97251–97295, 2021.
[18] 3GPP, “Spatial channel model for multiple input multiple output (MIMO)
simulations, release 11,” 3rd Generation Partnership Project, Tech. Rep.
25.996, Sep. 2012.
[19] “Study on channel model for frequencies from 0.5 to 100 GHz (3GPP TR
38.901 version 16.1.0 Release 16),” Accessed: Nov. 2024. [Online]. Avail-
able: https://www.etsi.org/deliver/etsi_tr/138900_138999/138901/ 16.01.
00_60
[20] P.Kyösti et al., “WINNER II channel models,” D1.1.2 V1.2, IST-4-027756
WINNER II, Tech. Rep., 2007.
[21] L. Liu et al., “The COST 2100 MIMO channel model,” IEEE Wireless
Commun., vol. 19, no. 6, pp. 92–99, Dec. 2012.
[22] L. Raschkowski et al., “METIS channel models,”ICT-317669-METIS/D1.
4, Jul. 2015.
3612 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 74, NO. 3, MARCH 2025
[23] S. Jaeckel, L. Raschkowski, K. Börner, 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.
[24] R. J. Weiler et al., “Quasi-deterministic millimeter-wave channel models
in MiWEBA,” EURASIP J. Wireless Commun. Netw., vol. 84, Mar. 2016,
Art. no. 84, doi: 10.1186/s13638-016-0568-6.
[25] A. Maltsev, R. Maslennikov, A. Sevastyanov, A. Lomayev, and A. Kho-
ryaev,“Statistical channel model for 60 GHz WLAN systems in conference
room environment,” in Proc. 4th Eur. Conf. Antennas Propag., 2010,
pp. 1–5.
[26] H. C. Nguyen, G. R. MacCartney, T. Thomas, T. S. Rappaport, B. Vejl-
gaard, and P. Mogensen, “Evaluation of empirical ray-tracing model for
an urban outdoor scenario at 73 GHz E-band,” in Proc. IEEE 80th Veh.
Technol. Conf., 2014, pp. 1–6.
[27] S. Hur et al., “28 GHz channel modeling using 3D ray-tracing in urban
environments,” in Proc. 9th Eur. Conf. Antennas Propag., 2015, pp. 1–5.
[28] A. Schiavoni, A. Leoni, D. Arena, and R. Lanzo, “Ray tracing simulations
at millimeter waves in different indoor and outdoor scenarios,” in Proc.
10th Eur. Conf. Antennas Propag., 2016, pp. 1–5.
[29] F. Fuschini, E. M. Vitucci, M. Barbiroli, G. Falciasecca, and V. Degli-
Esposti, “Ray tracing propagation modeling for future small-cell and
indoor applications: A review of current techniques,” Radio Sci., vol. 50,
no. 6, pp. 469–485, 2015.
[30] Y.-G.Lim, Y. J. Cho, M. S. Sim, Y. Kim, C.-B. Chae, and R. A. Valenzuela,
“Map-based millimeter-wave channel models: An overview, data for B5G
evaluation and machine learning,” IEEE Wireless Commun., vol. 27, no. 4,
pp. 54–62, Aug. 2020.
[31] M. Lecci, P. Testolina, M. Polese, M. Giordani, and M. Zorzi, “Accuracy
versus complexity for mmWave ray-tracing: A full stack perspective,”
IEEE Trans.Wireless Commun., vol. 20, no. 12, pp. 7826–7841, Dec. 2021.
[32] C.-X. Wang, J. Bian, J. Sun, W. Zhang, and M. Zhang, “A survey of 5G
channel measurements and models,”IEEE Commun. Surveys Tuts., vol. 20,
no. 4, pp. 3142–3168, Fourthquarter 2018.
[33] S. Jaeckel, L. Raschkowski, K. Börner, 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. 808–812,
Jun. 2014.
[34] A. Bodi et al., “NIST quasi-deterministic (Q-D) channel realization soft-
ware documentation,” Nat. Inst. Standards Technol., Gaithersburg, MD,
USA, Tech. Rep., Jan. 2021.
[35] A. Maltsev et al., “Quasi-deterministic approach to mmWave channel
modeling in the FP7 MiWEBA project,” in Proc. WWRF’33, 2014, pp. 1–6.
[36] R. Charbonnier et al., “Calibration of ray-tracing with diffuse scattering
against 28-GHz directional urban channel measurements,” IEEE Trans.
Veh. Technol., vol. 69, no. 12, pp. 14264–14276, Dec. 2020.
[37] “Advancement of radio wave simulation system technology in virtual
space, ministry of internal affairs and communications, Japan,” Accessed:
Aug. 12, 2023. [Online]. Available: https://www.tele.soumu.go.jp/j/sys/
fees/purpose/emulator/ index.htm
[38] N. Suzuki, H. Tsukada, R. Takahashi, B. Bag, and M. Kim, “Character-
ization of spatial consistency of cluster channels in urban environments
at 24 and 60 GHz,” IEEE Antennas Wireless Propag. Lett., vol. 23, no. 5,
pp. 1583–1587, May 2024.
[39] M. Kim et al., “A 24/60-GHz dual-band double-directional channel
sounder using COTS phased arrays,” in Proc. IEEE Int. Conf. Commun.,
Seoul, Korea, May 2022, pp. 1113–1117.
[40] M. Kim, S. Tang, and K. Kumakura, “Fast double-directional full az-
imuth sweep channel sounder using low-cost COTS beamforming RF
transceivers,” IEEE Access, vol. 9, pp. 80288–80299, 2021.
[41] “EVK06002/00, SIVERS IMA,” Accessed: Aug. 12, 2023. [Online].
Available: https://www.siversima.com/product/evk-06002- 00/
[42] M. Kim, T. Iwata, S. Sasaki, and J. Takada, “Millimeter-waveradio channel
characterization using multi-dimensional sub-grid CLEAN algorithm,”
IEICE Trans. Commun., vol. E103-B, no. 7, pp. 767–779, Feb. 2020.
[43] H. Tsukada, K. Kumakura, S. Tang, and M. Kim, “Millimeter-wave
channel model parameters for various office environments,” IEEE Access,
vol. 10, pp. 60387–60396, 2022.
[44] K. Haneda, J. Järveläinen, A. Karttunen, M. Kyrö, and J. Putkonen, “A
statistical spatio-temporal radio channel model for large indoor environ-
ments at 60 and 70 GHz,” IEEE Trans. Antennas Propag., vol. 63, no. 6,
pp. 2694–2704, Jun. 2015.
[45] “Wireless EM propagation software: Wireless InSite remcom,” Accessed:
Feb. 04, 2024. [Online]. Available: https://www.remcom.com/wireless-
insite-em- propagation-software
Hibiki Tsukada (Member, IEEE) received the B.E.
and M.E. degrees in electrical and electronic engi-
neering from Niigata University, Niigata, Japan, in
2021 and 2023, respectively. He is currently with
NTT Access Network Service Systems Laboratories.
His research interests include millimeter-wave radio
channel sounding and modeling. He is also a Member
of the IEICE.
Naoya Suzuki received the B.E. and M.E. degrees
in electrical and electronic engineering from Niigata
University, Niigata, Japan, in 2022 and 2024, re-
spectively. He is currently with NEC Corporation.
His research interests include millimeter-wave radio
channel sounding and modeling. He is also a Student
Member of the IEICE.
Banibrata Bag (Member, IEEE) received the B.E.
degree in computer science & engineering from Dr.
B.C. Roy Engineering College, Durgapur, India, in
2004, the M.Tech. degree in electronics and commu-
nication engineering from Techno Main Salt Lake,
Kolkata, India, in 2009, and the Ph.D. degree from
Jadavpur University, Kolkata, India, in 2022. In 2010,
he joined the Department of Electronics and Com-
munication Engineering, Haldia Institute of Technol-
ogy, Haldia, India, as an Assistant Professor. He is
currently a Specially Appointed Assistant Professor
with the Graduate School of Science and Technology, Niigata University,
Niigata, Japan. His research interests include millimeter-wave, terahertz radio
propagation channel modeling, and optical wireless communications.
Riku Takahashi received the B.E. and M.E. degrees
in electrical and electronic engineering from Niigata
University, Niigata, Japan, in 2022 and 2024, respec-
tively. He is currently with Mitsubishi Electric Corpo-
ration. His research interests include millimeter-wave
and terahertz radio channel sounding and modeling.
He is also a Student Member of the IEICE.
Minseok Kim (Senior Member, IEEE) was born in
Seoul, Korea. He receivedthe B.S. degree in electrical
engineering from Hanyang University, Seoul, Korea,
the M.E. and D.E. degrees in division of electri-
cal and computer engineering, Yokohama National
University (YNU), Yokohama, Japan, in 1999, 2002,
and 2005, respectively. In 2007, he was an Assistant
Professor with the Tokyo Institute of Technology,
Tokyo, Japan, and a Visiting Scholar with the Geor-
gia Institute of Technology, Atlanta, GA, USA, in
2010. In 2014, he joined the Graduate School of
Engineering, Niigata University, Niigata, Japan, as an Associate Professor.
His research interests include radio propagation channel measurement and
modeling, millimeter-wave radar, radio tomographic imaging techniques, and
MIMO/antenna array signal processing. He is also an Associate Editor for IEEE
ACCESS and IEEE ANTENNAS WIRELESS PROPAGATION LETTERS.HeisaSenior
Member of the IEICE.
