This paper has been accepted for presentation at the 20th Mediterranean Communication and Computer Networking Conference.

©2022 IEEE.
Please cite it as: P. Testolina, M. Lecci, A. Traspadini, and M. Zorzi, “An Open Framework to Model Diffraction by Dynamic Blockers in Millimeter
Wave Simulations,” in IEEE 20th Mediterranean Communication and Computer Networking Conference (MedComNet), Paphos, Cyprus, Jun. 2022.

An Open Framework to Model Diffraction by

Dynamic Blockers in Millimeter Wave Simulations
Paolo Testolina, Mattia Lecci, Alessandro Traspadini, and Michele Zorzi
Department of Information Engineering, University of Padova, Italy
E-mail: {testolina,leccimat,traspadini,zorzi}@dei.unipd.it

(RTs) exploit a digital reconstruction of the environment to

Abstract—The millimeter wave (mmWave) band will be ex- achieve a greater degree of accuracy, at the cost of additional
ploited to address the growing demand for high data rates computational load [6]. Generally, the digital model includes
and low latency. The higher frequencies, however, are prone to
limitations on the propagation of the signal in the environment. the static elements of the scenario, e.g., the floor, ceiling and
Thus, highly directional beamforming is needed to increase the walls, and the tables, screens and other objects, depending
antenna gain. Another crucial problem of the mmWave fre- on its level of detail. Its design is not trivial, as it requires
quencies is their vulnerability to blockage by physical obstacles. the use of a Computer-Aided Design (CAD) software, a
To this aim, we studied the problem of modeling the impact complex and time-consuming task. Furthermore, dynamic
of second-order effects on mmWave channels, specifically the
arXiv:2206.05000v1 [cs.NI] 10 Jun 2022

susceptibility of the mmWave signals to physical blockers. With elements, e.g., moving humans and vehicles, are not generally
respect to existing works on this topic, our project focuses considered, as their movements can not be included in a
on scenarios where mmWaves interact with multiple, dynamic single CAD file. Thus, these elements are often overlooked
blockers. Our open source software includes diffraction-based in RT simulations, despite the significant role they play in the
blockage models and interfaces directly with an open source propagation of the signal [7], [8]. Furthermore, the study of
Radio Frequency (RF) ray-tracing software.
Index Terms—5G, millimeter wave networks, simulation, the impact of blockers on the network performance has been
channel model, diffraction, propagation. limited to ad hoc scenarios, often considering a single blocker
and only at the PHY layer. On the contrary, their effect
on large-scale, high-level network simulations has not yet
I. I NTRODUCTION been fully characterized, due to the complexity of designing
The millimeter wave (mmWave) frequencies feature large the scenario and modeling the diffraction when considering
chunks of untapped bandwidth that can increase the data moving obstacles.
rate provided to the end users, and the small wavelength In this work, we present the Blockage Manager1 , a novel
enables the design of antenna arrays with tens of elements in open-source software to model dynamic blockers in RT
a small form factor to support beamformed transmissions. simulations. The application was designed to post-process
While these promising characteristics make the mmWave information typically obtained from RT software, offering the
technology able to meet the requirements of 5th generation user a simple yet powerful interface to introduce blockage
(5G) cellular systems and Wi-Fi networks [1], [2], there are models on top of them, making their simulations more real-
several concerns regarding the propagation characteristics at istic and dynamic. Starting from a pre-processed simulation
these frequencies that justify a more accurate study and the allows the user to later add as many obstacles as desired,
need for new channel models. First, the high propagation with custom mobility and settings, without the need to run
loss limits the coverage region of mmWave base stations, an entirely new ray-tracing simulation from scratch, which
although large antenna arrays and complex beamforming can take a significant amount of time. The software has been
techniques can mitigate the problem. Second, at mmWaves, designed to present as simple an API as possible, so that
the increased diffraction loss results in deep shadow regions, the user only has to create a high-level description of the
thus further degrading the communications performance [3], desired obstacles, while still having some control on fine-
[4]. Furthermore, mmWave signals can be easily blocked by tuning parameters if desired, and leaving all the complexity
obstacles (e.g., vehicles, buildings, vegetation, human bod- to the simulator.
ies), which may prevent direct Line-of-Sight (LoS) communi- The aim of this work is to describe an open-source software
cations. These (often unpredictable) propagation components able to (i) interface directly with qd-realization [9], an open-
make it imperative to accurately model the dynamics of source RF ray-tracing software, importing and exporting
the surrounding environment, and to design communication channel traces with a single line of code, (ii) process pre-
protocols, and especially beam-tracking algorithms, taking computed RT simulations, adding multiple mobile obstacles
into account these disruptive events. on top of them, (iii) support already implemented blockage
In this context, ray tracing has emerged as an essential models, with varying degrees of complexity and accuracy.
tool to model the mmWave channel [5], especially, but not Furthermore, we showcase some basic network scenarios
limited to, when detailed link-level protocols need to be sim- processed with our Blockage Manager, describing the ob-
ulated. With respect to stochastic channel models, ray tracers tained results and discussing future works that will enable it
to closely mimic real channel behaviors in the presence of
This work was partially supported by the National Institute of Standards
and Technology (NIST) under award no. 60NANB20D082. The work of M.
Lecci and P. Testolina was supported by Fondazione CaRiPaRo under grants 1 The Blockage Manager is available at https://github.com/signetlabdei/rt-
“Dottorati di Ricerca” 2018 and 2019. blockage-manager.

©2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including
reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists,
or reuse of any copyrighted component of this work in other works.
obstacles, allowing the community to better study and design existing models in the literature that deal with this problem.
link-level protocols for mmWave communications. Namely, we first report the main works on blockage at the
In the remainder of the paper, we first report the state channel modeling level, and then its modeling at the higher
of the art on diffraction modeling (Sec. II). Then, Sec. III layers of the network stack.
offers a brief overview of the framework architecture, giving The authors of [19] analyzed the scattering of a vehicle
the essential details to understand its main components. In and showed that RT simulations of complex objects allow
Sec. IV, we present the blockage models that are currently the derivation of simplified scattering models. This is fun-
implemented in the framework. Finally, the results in Sec. V damental, since only by reducing the complexity of large
serve the double purpose of showcasing the usage of the objects does it become possible to simulate realistic scenarios.
Blockage Manager and of highlighting the impact of dynamic Their results show that, rather than introducing a complex
blockage in mmWave networks, and Sec. VI presents some model with a high polygon count, it is possible to reduce the
concluding remarks and future work directions. complexity of simulations involving cars by only considering
a few key points producing scattering, e.g., wing mirrors.
II. S TATE OF THE A RT Another simplification was described also for human
Signals at mmWave frequencies are prone to limitations on blockage in [20], where the authors analyzed their mea-
their propagation in the environment. Therefore, understand- surements taken at 10.5 GHz and concluded that a metal
ing and precisely characterizing the interaction between radio cylinder affects the radio channel similarly to a human body.
waves and the surroundings is fundamental to characterize This result justifies the usage of such a simple model in
communications in different scenarios. practical simulations rather than modeling a detailed person,
This reason has led to different modeling approaches for a significantly harder task to perform with a correspondingly
the mmWave channel, which have various degrees of com- much higher computational cost.
plexity and accuracy, and can be applied to different contexts The 3GPP also proposes a map-based hybrid channel
and evaluations [6]. Thus, channel models can be summarized model [13] using [21] for the deterministic component of
in: analytical, stochastic, and Quasi Deterministic (QD). the channel, and adding a random component to model
Analytical channel models: These models generally clusters of MPCs. It also proposes the usage of two blockage
offer a simplified representation of the channel, based on models, to better represent a dynamic and realistic channel. In
propagation loss and a random variable representing fading. the stochastic blockage model (Blockage model A), random
This kind of model has a limited accuracy and can be used to rectangular regions are blocked, considering the temporal and
characterize communications in a generic environment [10]– spatial correlation for completeness. The geometric blockage
[12], without accounting for the features of specific scenarios model (Blockage model B) deploys rectangular screens, spec-
and their interaction with the elements typical of mmWave ifying their relative dimensions and mobility patterns. In both
propagation (e.g., realistic antenna arrays and beamforming). cases, a Knife-Edge Diffraction (KED) at four edges is de-
Stochastic channel models: They derive the entries of fined, using the arctan approximation, described in Sec. IV,
the channel matrix from a set of random distributions, whose for the diffraction loss.
parameters are determined by statistical fits on channel mea- A low-complexity enhancement of this approximation was
surements for a generic scenario (e.g., a common rural or studied in [22] using a Fourier-based model for human ob-
urban environment). Their speed and their stochastic nature stacles. Its authors compared their simulations with measure-
allow to easily generate the radio channels for generic, large- ments at 2.44 GHz in an indoor Non-Line-of-Sight (NLoS)
scale scenarios. Models of this type, e.g., the one proposed scenario and showed that their approximation produces shad-
in 3GPP TR 38.901 [13], have been used in the performance owing loss closer to the Fresnel formulas for Double Knife-
evaluations of mmWave networks [14], [15]. Edge Diffraction (DKED) than the simpler arctan model.
Quasi-Deterministic channel models: QD channels, in- Besides, the validity of the KED model was proved for
stead, can accurately model the interactions of the mmWave narrow obstacles also using complex specific solvers for
signal with a specific environment [16]–[18]. However, since Maxwell’s equations as described in [23].
the Multi Path Components (MPCs) are generated by the A series of modifications and criteria to improve the
interaction between the transmitted signal and the elements accuracy of the 3GPP model are provided in [24]. It proposed,
of a given environment, they can offer improved accuracy at in case of multiple blockers in close proximity, to consider
the cost of a higher processing load. them as a single blocker with increased size. The authors also
Including the effect of blockage on the channel model suggest a way to increase the accuracy when the transmitted
is a non-trivial task. The high-level, generic nature of the beam does not cover the entire obstacle surface, for instance,
analytical models makes it difficult to model the blockage by when highly directional antennas at a relative short distance
multiple, possibly moving obstacles (e.g., people, vehicles, are employed. In such cases, the idea consists of taking
foliage, etc.). Similarly, since stochastic channel models do into consideration the transmitted beam pattern to choose
not require a model of the propagation environment, it is the position of the diffraction points on the surface of the
difficult to define a temporally and spatially consistent behav- obstacles. A measurement campaign was carried out, and
ior of the blockers. Furthermore, modeling specific mobility its results showed that with these modifications the loss
patterns does not agree well with the stochastic nature of the prediction provided by the 3GPP model can be improved.
channel, and can decrease the advantages of these kinds of The literature also includes some studies analyzing block-
models (i.e., their being lightweight and easy to use). Finally, age experimentally [7], [8], [25]. The authors in [7] analyzed
as mentioned in Sec. I, placing (possibly dynamic) blockers accurate measurements of human body blockage at 60 GHz
in an RT simulation can also be challenging, thus motivating with a person that traversed a linear trajectory perpendicular
this work. In the remainder of this section, we present the to the direct path between the Transmitter (TX) and the
Receiver (RX). They derived a diffraction model and com- for instance the diffraction effect of obstacles and considering
pared it with their measurements. Eventually, they showed appropriately the presence of multiple blockers, can provide
that when a human obstacle is close to the path of the direct more precise results. For example, even in a simple human
ray, it also creates a strong reflected ray, thus acting as a 2-ray blockage scenario where an obstacle moves perpendicularly
propagation model. to the line between the TX and the RX, the loss is higher
A similar scenario was studied in [8] at 73 GHz. Its anal- when the obstacle crosses the LoS closer to one of the
ysis showed that the use of directive antennas has a crucial nodes. This phenomenon, shown in our simulations, can be
impact on human blockage measurement results, especially described only by taking into account the diffraction effect
when the blocker was close either to the TX or to the RX. of the obstacles. Thus, it can be easily understood that in
Furthermore, [26] explores the impact of human block- more complex scenarios with many obstacles this effect has
age on the channel model specifically proposed for a critical impact on results. This is the reason why we
IEEE 802.11ad [27], based on measurements at 60 GHz. Its implemented an open source software able to (i) include
authors analyzed ray tracing simulations in a human blockage diffraction-based blockage models, (ii) interface directly with
scenario considering also diffraction effects. They found that an open source Radio Frequency (RF) ray-tracing software,
the impact of human blockage is twofold. First of all, the (iii) support multiple mobile blockers.
Signal-to-Noise Ratio (SNR) at the RX decreases due to the
additional attenuation, even if smart antennas are employed. III. S OFTWARE A RCHITECTURE
On the other hand, the delay spread and the frequency
selectivity in general increase, because the channel conditions The Blockage Manager was designed as a self-contained
change from LoS to NLoS. Both effects deteriorate the link, software to model the effect of blockage in RT traces. Doing
causing a higher bit error probability during communications. it in post-processing allows running the computationally-
The modern literature not only provides examples of works demanding RT only once, including in the simulation all the
studying blockage at mmWaves from a physical measurement static elements with a single CAD model. Any distribution of
and modeling perspective, but also recognizes it as an issue the dynamic elements, i.e., blockers, can then be introduced
that mmWave communication protocols need to address. in the static scenario using the Blockage Manager, that can
In fact, while high frequencies are in principle capable of use the same RT output as a static baseline for any blocker
delivering the exceptional performance that they promise, configuration.
they are also heavily affected by bad channel conditions, The Blockage Manager is organized in modules, each
making the system much less stable thus worsening the providing a set of classes and functions to handle the elements
Quality of Experience (QoE). required for the simulation. Here, we report a brief overview
An analysis of blockage was performed in [25] to evaluate of the different modules, to offer some insights on the
the effect of blockers on an End-to-End (E2E) application. implementation and to clarify how the results presented in
The authors used a phased array system that measured the Sec. V were obtained.
channel in multiple directions, in order to fully understand Although geometry libraries exist in Python [37], [38], in
the path diversity. order to meet all the simulation requirements (3D geometry
Different analyses on Transmission Control Protocol (TCP) with fast computation time), and to build a self-contained
[28]–[31], studying the behavior of this protocol at mmWave package, we decided to implement all the basic geometrical
frequencies, employed obstacles to simulate NLoS scenarios. objects and operations from scratch. Thus, we developed a
Indeed, the extreme variability of the signal quality over custom Geometry module to handle the geometry of the
mmWave links yields either degraded TCP goodput and very ray and of the obstacles themselves, and to define consistent
low utilization of the resources at mmWave frequencies, or, in mobility patterns of the latter. This module underlies most of
the presence of link-layer retransmissions, high goodput at the the operations carried out during the simulation and can be
price of high latency. Therefore, especially at mmWaves, the optimized independently in the future, thanks to the common
effect of obstacles is crucial to understand the actual condition object interface.
of the channel. The RT traces can be imported in the Blockage Man-
As blockage is so impactful, especially on TCP-based ager using the Scenario interface, that defines common
communications, some works have proposed and tested ways methods to import/export traces in the target format, as
of relying on multiple simultaneous connections, on both well as to access and update sets of rays between nodes.
high and low frequencies, to balance resilience and raw Currently, the Scenario interface supports traces from the
throughput. Protocols such as Multipath TCP (MPTCP) have qd-realization software, which is specifically able to handle
been analyzed [31]–[36], showing their advantages when, for channel traces for multiple users and timesteps. However, the
example, buildings block the direct mmWave path or fast interface allows for the possibility to support different ray-
mobility cannot be fully supported by the mmWave network. tracing formats, that could be implemented in the future. The
In [29], the authors assessed the goodput of a scenario information for each ray (delay, path gain, phase, path) is
in which obstacles of different sizes are placed in the area stored in a Ray object, that also offers a simple interface to
between the TX and the RX. The application layer simulated consistently compute Angles of Departure (AoDs) and Angles
a file transfer with full buffer and the authors noticed that of Arrival (AoAs).
the number of obstacles plays a major role in the achievable Then, blockers can be introduced in the scenario with
goodput. With fewer obstacles, there is a higher probability the common Obstacle interface, that handles obstructions,
of having a LoS channel and this has a crucial impact on the diffraction, and other effects that a generic obstacle may
data rate available at the physical layer. impose over the Rays of the imported Scenario. Currently,
Thus, the use of more accurate channel models, modeling a sphere, a rectangular and an orthogonal-rectangular screens
 edge The Fresnel integral can be computed by numerical calcu-
lation, however [39] provides a simple way to compute an
approximate result.
TX RX
Obstruction
When an obstacle cuts through a transmission path, the
simplest obstruction model consists in applying a constant
loss L during the shadowing window, resulting in a sharp
transition between an obstructed and an unobstructed path.
Fig. 1: DKED geometry.
The obstruction loss is generally computed based on the
absorption properties of the obstacle. Depending on the model
are implemented. We defined the orthogonal-rectangular ob- assumptions, an obstructed path may be completely removed
stacle as an ideal rectangular screen that behaves as if it from the scenario. Its simplicity makes this model desirable
were orthogonal to any considered ray, when computing the in terms of computational and implementation complexity,
interaction between the two. As detailed in Sec. IV, this so that it has often been chosen for end-to-end network
artificial obstacle was introduced to meet the hypotheses of simulations [40]–[42]. Unfortunately, the unrealistic sharp
several diffraction models, that could thus be included in drop not only yields imprecise results, but can also affect
the software. On the contrary, the rectangular screen can some beam-tracking algorithms, which are of the utmost
be tilted in both the azimuth and elevation directions, thus importance when it comes to mmWave communication.
providing a more general obstacle mobility at the cost of a METIS
limited set of available diffraction models. The movement
of an obstacle during the simulation is described by a The Mobile and wireless communications Enablers for
MobilityModel. At each time step, the position of each the Twenty-twenty Information Society (METIS) channel
obstacle is updated based on such model, thereby providing model [21] provides a simplified method to account for the
accurate and temporally-correlated mobility and making the diffraction contribution at mmWave frequencies. The screen
channel temporally consistent. The whole simulation, i.e., is assumed to be vertical and perpendicularly oriented with
computing the interactions of the rays with the obstacles, and respect to the considered ray segment in the projection from
updating the positions of the obstacles at each time step, is above. The diffraction loss is modeled using a KED model
handled by the Environment, that constitutes the core of for the four edges of the screen as:
the software, with just a few lines of code. L = −20 log10 (1 − (lh1 + lh2 )(lw1 + lw2 )) [dB] (4)
IV. B LOCKAGE M ODELS where lhi and lwi are the Single Knife-Edge Diffraction
When considering the interaction between an object and a (SKED) at edge i, corresponding to the height h and the
signal propagating in the free space from point T to point width w of the obstacle. The loss of each single edge is
R, the Fresnel Zones offer a useful model to analyze the provided by:
intensity of the diffraction. The Fresnel zones are concentric  q 
π π(dT ,i +dR,i −r)
ellipses with focal points at T and R, and radius atan ± 2 λ
r li = (5)
dT dR π
rn = nλ (1)
dT + dR where, as shown in Fig. 1, dT,i and dR,i are the distances
where λ is the wavelength of the transmitted signal, dT between the nodes and edge i of the screen, and r is the
(dR ) is the distance between the diffraction point and the distance between the TX and the RX. If the link is NLoS,
transmitter (receiver), and n is the order of the Fresnel Zone. then each contribution is taken as positive, whereas for LoS
If the cross section of the First Fresnel Zone is obstructed, condition, only the farthest edge from the link provides a
diffraction becomes the dominant term in the interaction with positive contribution.
the object, with significant impact on the communication
Double Knife-Edge Diffraction (DKED)
performance. In the Blockage Manager, several models to
describe the diffraction loss are available. In this section, we This model [43] computes the SKED from the screen edges
provide a brief overview of those that were considered for evaluating both amplitudes and phases. Since only the lateral
the simulations described in Sec. V. edges are considered for the KED, the interaction of the
As the majority of the diffraction models is based on the obstacle with the propagating waves is that of a vertical
Fresnel formulation, we define the complex Fresnel integral stripe of infinite vertical extent. The SKED for edge i can
as: Z ν be computed as:
πs2
ej 2 ds
   
F (ν) = (2) 1+j 1 1
0
li = − C(νi ) − j − S(νi ) (6)
2 2 2
where ν is the Fresnel-Kirchhoff diffraction parameter:
r and then, this model computes the superposition of each
2 dT + dR contribution to obtain the diffraction loss:
ν(h) = h (3)
λ dT dR
L = −20 log10 (|l1 + l2 |) [dB] (7)
and depends on the obstruction depth h. The real and imag-
inary parts of F (ν) can be recalled as follows: Besides, as the Semi-empirical ITU model, it provides a valid
solution even for non-orthogonal obstacles, and thus can be
F (ν) = C(ν) + jS(ν) exploited for a larger variety of obstacles.
 0 0

Diffraction Loss [dB]

Diffraction Loss [dB]

Baseline
−10 Obstruction
−10
METIS
DKED+PC
−20 DKED −20
ITU SE
−0.4 −0.2 0 0.2 0.4 0 2 4 6 8
Obstacle Location [m] Obstacle Distance from TX [m]

(a) Models at 60 GHz (b) Different obstacle’s distance

Fig. 2: Comparison between the implemented models using a carrier frequency of 60 GHz.

10 GHz 30 GHz 60 GHz 100 GHz

Diffraction Loss [dB]

0 0 0 0

−10 −10 −10 −10

−20 −20 −20 −20

−30 −30 −30 −30

−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
Obstacle Location [m] Obstacle Location [m] Obstacle Location [m] Obstacle Location [m]

(a) METIS (b) DKED (c) DKED+PC (d) Semi-Empirical ITU

Fig. 3: Comparison of the implemented models at different frequencies.

Double Knife-Edge Diffraction with Phase Correction V. R ESULTS

(DKED+PC) In this section, we present a set of results obtained using
In [43], the authors present also a modified version of the Blockage Manager, to showcase the framework and offer
the DKED. Namely, the SKED is computed for the two some insight on the impact of blockage modeling on network
contributions (front and back of the body) as described simulations, and to compare the different diffraction models.
by Eq. (6) for the DKED model. Additionally, this method First, in Sec. V-A, we will show their general behavior as
takes into account the phase shift of the diffracted rays path directly described by the equations, thus considering the
when combining the two contributions. effect that the obstacle has when passing through the direct
  ray between transmitter and receiver.
2π∆d1 2π∆d2 
L = −20 log10 l1 · e−j λ + l2 · e−j λ 

[dB] Then, Sec. V-B will consider a more realistic static sce-
(8) nario, where transmitter and receiver are still fixed but placed
where ∆di is the length of the i-th diffracted ray. The model in a room, and reflections, computed using a ray-tracer, also
is presented and calibrated for ultra-wideband measurements interact with the moving obstacle.
between 4 and 10 GHz. However, its analytical derivation is Finally, in Sec. V-C a more complex dynamic scenario is
based on the Fresnel diffraction theory, that holds also for evaluated, where we consider an access point placed on the
higher frequencies, as reported in [7]. ceiling of a room and a user moving away from it. In this
case, multiple obstacles move in the scenario, to make the
simulation even more realistic.
ITU-R P.526-15
All the scenarios but one consider the obstacle(s) passing
The International Telecommunication Union (ITU) pro- through the LoS ray to better highlight the effect of the
vides guidelines on the modeling of the diffraction in [39]. diffraction on the received power. As detailed in the follow-
Based on the Uniform Theory of Diffraction (UTD) and on ing, we consider also the case where the obstacles do not
high-precision measurements, two approximations are given intercept the main ray, but only the reflected ones, to show
for estimating the minimum and the average diffraction that diffraction is relevant also in this situation. For this work,
loss. Furthermore, a Semi-Empirical method (ITU SE) is we consider an (orthogonal-rectangular) thin screen, which
derived for thin rectangular screens, that exhibits the rapid provides a simplified yet realistic representation of the human
fluctuations of the field strength due to the constructive and body, according to measurements [7], [43] and standards [21],
destructive interference of the diffraction from the edges. [39]. The obstruction loss is set to 10 dB, obtained by
This method can be used also for non-orthogonal obstacles. averaging the mean loss of the considered diffraction models.
Moreover, with reasonable accuracy, it does not require the First, the qd-realization traces are imported in the Blockage
Fresnel integral to be solved. Both models can be applied Manager, that applies the diffraction loss to the rays ac-
when the wavelength is fairly small in relation to the size of cording to the obstacle configuration. Link-layer simulations
the obstacles, which is the case when considering human-size are then run on the traces using the same custom MATLAB
blockers at mmWave frequencies. The mathematical details simulator as in [6], [44], to observe the blockage effect on the
of this method are reported in [39]. SNR. However, given that the format of the output traces of
 40

35

14 m

SNR [dB]
30 METIS
Baseline
TX RX

7m
{
DKED+PC
DKED
ITU SE
3m

25
Obstruction
{

1m 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Time [s]

(a) Static scenario. (b) Max reflection order: 2

Fig. 4: Static scenario including second-order reflections.

In particular, the difference between the simple obstruction

and the more complex diffraction models can be clearly
observed. Complex physics-based models, in fact, show an
oscillatory behavior before and after the obstacle even inter-
cepts the LoS, due to its effect on the surrounding propaga-
3 RXref
tion environment, with peaks of almost 2 dB. Furthermore,
z

0 TXref
within the blockage region, very sharp deep fades affect the
19
0 channel with over 10 dB extra losses, due to destructive
interference among the rays curving around the obstacle. This
5 phenomenon, together with the high obstruction loss, can
y
make channel estimation and adaptation harder, thus further
x
reducing the communication efficiency in the presence of
10 0
obstacles. It is also possible to observe that different models
Fig. 5: Dynamic scenario have different average obstruction losses, making us question
which one, if any, is close to the real-world measurements.
Instead, Fig. 2b shows what happens for the different
the Blockage Manager is the same as that of qd-realization, models when the same obstacle moves starting close to the
they are compatible with other simulators such as Network TX and moving towards the RX, always obstructing the direct
Simulator 3 (ns-3) [45]. The main parameters used for link- path between the two. For all diffraction models, the distance
level simulations are listed in Table I. Beamforming at both between the obstacle and the nodes is taken into account
the transmitter and the receiver is achieved using an antenna when modeling its effect on the channel. The figure shows
array with omnidirectional antenna elements and λ/2 element a symmetric behavior, justified by the symmetry of wireless
spacing. propagation, and all curves follow approximately the same
A. Model Comparison trend. Specifically, the loss tends to be higher when the
obstacle is close to one of the nodes, and is lowest when
All implemented models have their own distinct charac-
the obstacle is exactly in the middle, a trend that is observed
teristics derived by their formulations. To give the reader an
irrespective of the distance between TX and RX. Intuitively,
idea of their behavior, we show a set of key comparisons for
in fact, rays that need a sharper turn to surpass the obstacle
these models in scenarios of interest.
(i.e., when the obstacle is close to one of the nodes) lose
We first consider a TX at 1.6 m height, and a RX at
more energy than rays that need a shallower turn (i.e., when
the same height placed 8 m away from the transmitter. An
the obstacle is farthest). A different explanation can be given
obstacle of size 0.2×1.7 m passes perpendicularly halfway
by thinking about the apparent size of the obstacle, as seen
between TX and RX, blocking the LoS between the two
from one of the nodes. The same obstacle has a different
nodes. To highlight the effect of the models, we only process
apparent size depending on how far it is from its observer,
the direct ray, normalizing the received power to obtain the
where closer objects appear larger than farther objects and
diffraction loss. Fig. 2a shows the losses of the different
thus result in a deeper shadow.
models at a 60 GHz carrier frequency.
Finally, it is important to remember that these models
heavily depend on the carrier frequency. In fact, in general
TABLE I: Simulation parameters
higher frequencies will present sharper and larger losses, due
TX Antenna 8×8 RX Antenna 4×4 to the larger electrical size of the obstacle with respect to the
λ
Antenna Spacing 2
Antenna Element omni-directional wavelength. The short wavelength will also be more prone
Carrier Frequency 60 GHz Bandwidth 2.16 GHz to creating constructive and destructive interference between
TX Power 20 dBm Noise Figure 10 dB
the rays bending around the obstacle, creating deeper and
 METIS Baseline DKED+PC
DKED ITU SE Obstruction

38
50

SNR [dB]
} 40 36
} }

30 34
20
32
0 5 10 15 5 5.2 5.4 5.6 5.8 6
Time [s] Time [s]

(a) Obstacle locations (static). (b) Full scenario (c) Zoom

Fig. 6: Dynamic scenario with 4 static obstacles.

50
50

SNR [dB]
40 40

30 30
20
20
0 5 10 15 2.2 2.4 2.6 2.8 3 3.2
(a) The 15 moving obstacles inter- Time [s] Time [s]
cept the LoS ray as the RX moves
through the room. (b) Full scenario (c) Zoom
Fig. 7: Dynamic scenario with 15 dynamic obstacles.

more frequent loss peaks. Longer wavelength, instead, will be C. Dynamic Scenario
able to more easily bend around an obstacle, greatly reducing
For this scenario, a different room is considered: the length
the obstruction losses. These results can also be observed in
and width are respectively equal to 19 m and 10 m, whereas
Fig. 3.
the height is 3 m as described in Fig. 5.
B. Static Scenario Although the environment, a rectangular room, is similar
We tested our framework in a scenario inspired by [7]. to the previous one, in this case the receiver is not fixed
Specifically, two static nodes are placed in positions p1 = and the two nodes are not at the same height. Indeed the
(1, 3, 1.6) and p2 = (9, 3, 1.6) within a 14×7×3 m3 transmitter has a fixed position in p1 = (5, 0.1, 2.9) and the
room, as shown in Fig. 4a. We simulated the channel with receiver starts from a position right below the transmitter,
qd-realization [9], an open-source ray-tracing software for pstart = (5, 0.1, 1.5), and moves away from it at 1.2 m/s for
mmWave propagation, considering up to second-order reflec- about 15.7 s, reaching the position pfinal = (5, 18.9, 1.5) next
tions. to the farthest wall from the TX position. The total number of
We then imported the channel trace into the Blockage Man- samples is 3133, achieved using a sampling period of 5 ms.
ager software, configuring an orthogonal-rectangular screen Following the results from [6], we simulated up to second-
moving from pstart = (5, 0, 0) upwards at 1.2 m/s, and order reflections, and excluded rays less powerful than the
sampled the channel every 3.4 ms for 1500 samples, for a most prominent one by more than 40 dB. These parameters
total of about 5 s of simulation time. We considered a thin were shown to yield a good balance between computational
screen with a width of 20 cm and a height of 1.7 m, in an effort and accuracy of results. We then imported the results
attempt to emulate the size of the human body. into the Blockage Manager software with two different ob-
Fig. 4b reports the SNR observed by the receiving node. stacle settings.
As the obstacle crosses through the LoS, the SNR decreases In the first scenario, reported in Fig. 6a, 4 static obstacles
rapidly, showing a behavior similar to that described in of size 0.4×1.7 m2 are considered, none of them blocking the
Sec. V-A both in shape and in amplitude. Specifically, ob- direct ray. The obstacles are placed at decreasing distances
serving a diffraction loss so close to the one presented in from the LoS, specifically in p1 = (5.2, 15.14, 0), p2 =
Fig. 2 highlights how the LoS ray carries the great majority (5.8, 7.62, 0), p3 = (5.4, 11.38, 0), and p4 = (6.2, 3.86, 0).
of the signal power to the receiver, while the reflected ray can As shown in Fig. 6, although the obstacles do not create
not compensate for the diffraction loss caused by the blocker. strong or sudden effects on the received power, they still
On the other hand, while the average loss is similar to the play an important role in the total received power, showing
one described in Sec. V-A, the deepest fades are mitigated differences with respect to the baseline of up to 8.2 dB. On the
by secondary reflected rays carrying far less power. Namely, other hand, as the RX passes by the first three obstacles, all
the troughs observed in most of the diffraction models often models behave almost indistinguishably from each other, with
overestimate the fade depth [7]. However, a complete channel an absolute error of at most 1.6 dB with respect to the simple
simulation, where the transceivers and the obstacles are obstruction model. On the contrary, the diffraction introduced
immersed in a realistic three-dimensional environment, tends by the last and closest screen is significant, with the SNR of
to average them out, reducing the overall overshooting. the diffraction models presenting a difference of about 5 dB
from that of the obstruction model. This justifies our design [4] I. A. Hemadeh, K. Satyanarayana, M. El-Hajjar, and L. Hanzo,
choice of setting a distance threshold on the diffraction model, “Millimeter-wave communications: Physical channel models, design
considerations, antenna constructions, and link-budget,” IEEE Commu-
so that obstacles far enough from the ray will not affect it, nications Surveys and Tutorials, vol. 20, no. 2, pp. 870–913, Second
making it sufficient to only model the most significant effects Quarter 2018.
of the diffraction. Similarly, we also allow the user to fall back [5] M. Lecci, P. Testolina, M. Giordani, M. Polese, T. Ropitault, C. Gentile,
N. Varshney, A. Bodi, and M. Zorzi, “Simplified Ray Tracing for the
to a simpler obstruction model for secondary or less powerful Millimeter Wave Channel: A Performance Evaluation,” in Information
rays, which have a much smaller effect on the total received Theory and Applications Workshop (ITA), San Diego, CA, USA, Feb.
power. 2020.
[6] M. Lecci, P. Testolina, M. Polese, M. Giordani, and M. Zorzi, “Ac-
Finally, in the second scenario we consider 15 obstacles curacy Versus Complexity for mmWave Ray-Tracing: A Full Stack
traversing the room at regular intervals, as represented in Perspective,” IEEE Transactions on Wireless Communications, vol. 20,
Fig. 7a. The results in Fig. 7 show all the effects discussed so no. 12, pp. 7826–7841, Dec. 2021.
far in the previous sections, combined in a more complex and [7] A. Bhardwaj, D. Caudill, C. Gentile, J. Chuang, J. Senic, and D. G.
Michelson, “Geometrical-empirical channel propagation model for
realistic scenario. In particular, it is possible to observe that (i) human presence at 60 GHz,” IEEE Access, vol. 9, pp. 38 467–38 478,
all models behave as expected during obstruction (highlighted 2021.
in gray), (ii) obstacles affect the channel also when not di- [8] G. R. MacCartney, S. Deng, S. Sun, and T. S. Rappaport, “Millimeter-
wave human blockage at 73 GHz with a simple double knife-edge
rectly blocking the LoS, (iii) constructive interference before diffraction model and extension for directional antennas,” in IEEE 84th
and after obstruction can actually be very significant, even Vehicular Technology Conference (VTC-Fall), Sep. 2016, pp. 1–6.
overshooting the baseline, (iv) using a constant obstruction [9] wigig-tools, “qd-realization,” https://github.com/signetlabdei/qd-
realization/tree/feature/power-threshold, open-source ray-tracer
loss does not represent well the complex interaction between implementation.
transceivers and obstacles, (v) when the receiver reaches the [10] T. Bai and R. W. Heath, “Coverage and rate analysis for millimeter-
other side of the room, and with multiple rays having similar wave cellular networks,” IEEE Transactions on Wireless Communica-
tions, vol. 14, no. 2, pp. 1100–1114, Feb. 2015.
power to the direct path, obstruction is less severe and small [11] J. G. Andrews, T. Bai, M. N. Kulkarni, A. Alkhateeb, A. K. Gupta,
scale fading is actually the main concern. and R. W. Heath, “Modeling and analyzing millimeter wave cellular
systems,” IEEE Transactions on Communications, vol. 65, no. 1, pp.
VI. C ONCLUSIONS 403–430, Jan. 2017.
In this paper we presented a novel open source tool, the [12] P. Ferrand, M. Amara, S. Valentin, and M. Guillaud, “Trends and
challenges in wireless channel modeling for evolving radio access,”
Blockage Manager, to model the diffraction by dynamic IEEE Communications Magazine, vol. 54, no. 7, pp. 93–99, Jul. 2016.
obstacles in RT traces. Namely, the software post-processes [13] 3GPP, “Study on channel model for frequencies from 0.5 to 100 GHz,”
the output from a ray-tracer, allowing the user to introduce an 3rd Generation Partnership Project (3GPP), Technical Report (TR)
38.901, Jun. 2018, version 15.0.0.
arbitrary number of obstacles in the simulation, and modeling [14] M. Gapeyenko, V. Petrov, D. Moltchanov, S. Andreev, Y. Koucheryavy,
their interactions with the rays choosing from a number M. Valkama, M. R. Akdeniz, and N. Himayat, “An Analytical
of diffraction models. The potential of the software was Representation of the 3GPP 3D Channel Model Parameters for
MmWave Bands,” in Proceedings of the 2nd ACM Workshop on
showcased with simple yet insightful network simulations, Millimeter Wave Networks and Sensing Systems, ser. mmNets ’18.
which already provided interesting results. We hope that New Delhi, India: Association for Computing Machinery, 2018, p.
this tool lays the foundations for more accurate, large-scale 33–38. [Online]. Available: https://doi.org/10.1145/3264492.3264498
[15] T. Zugno, M. Polese, N. Patriciello, B. Bojović, S. Lagen, and M. Zorzi,
studies on the effects of blockage on mmWave networks. “Implementation of a Spatial Channel Model for Ns-3,” in ACM
Comparison with real-world data is an essential step to Workshop on Ns-3 (WNS3), Gaithersburg, MD, USA, Jun. 2020.
validate the framework. A more precise calibration of the [16] M. Lecci, M. Polese, C. Lai, J. Wang, C. Gentile, N. Golmie, and
diffraction models against measurements is currently being M. Zorzi, “Quasi-Deterministic Channel Model for mmWaves: Math-
ematical Formalization and Validation,” in IEEE Global Communica-
considered, and we plan to provide a calibration toolbox as tions Conference (GLOBECOM), Taipei, Taiwan, Dec. 2020.
an additional module for the framework. [17] V. Degli-Esposti, F. Fuschini, E. M. Vitucci, M. Barbiroli, M. Zoli,
Future steps will also focus on making the Blockage Man- L. Tian, X. Yin, D. A. Dupleich, R. Müller, C. Schneider, and
R. S. Thomä, “Ray-tracing-based mm-Wave beamforming assessment,”
ager able to describe a much wider variety of scenarios, intro- IEEE Access, vol. 2, pp. 1314–1325, 2014.
ducing new obstacles and diffraction models. For instance, a [18] J. McKown and R. Hamilton, “Ray tracing as a design tool for radio
foliage model could be implemented to increase the accuracy networks,” IEEE Network, vol. 5, no. 6, pp. 27–30, Nov. 1991.
[19] K. Schuler, D. Becker, and W. Wiesbeck, “Extraction of virtual scatter-
in outdoor environments where the signal propagates through ing centers of vehicles by ray-tracing simulations,” IEEE Transactions
trees and vegetation. Furthermore, UTD may be considered as on Antennas and Propagation, vol. 56, no. 11, pp. 3543–3551, Nov.
a more general approach to model the diffraction, including 2008.
obstacles of arbitrary shape and dimension. [20] M. Ghaddar, L. Talbi, T. A. Denidni, and A. Sebak, “A conducting
cylinder for modeling human body presence in indoor propagation
Finally, a full stack analysis of a real scenario with accurate channel,” IEEE Transactions on Antennas and Propagation, vol. 55,
obstacle models is essential to understand more precisely the no. 11, pp. 3099–3103, Nov. 2007.
actual effect of multiple, dynamic obstacles on the final user [21] “METIS Channel Models,” Deliverable D1.4, Feb. 2015.
[22] J. Medbo and F. Harrysson, “Channel modeling for the stationary UE
experience in different practical situations. scenario,” in 7th European Conference on Antennas and Propagation
(EuCAP), Gothenburg, Sweden, Apr. 2013.
R EFERENCES [23] M. D. Casciato, “Radio wave diffraction and scattering models for
[1] T. S. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y. Azar, K. Wang, G. N. wireless channel simulation,” Ph.D. dissertation, University of Michi-
Wong, J. K. Schulz, M. Samimi, and F. Gutierrez, “Millimeter wave gan, 2001.
mobile communications for 5G cellular: It will work!” IEEE Access, [24] D. Prado-Alvarez, S. Inca, D. Martı́n-Sacristán, and J. F. Monserrat,
vol. 1, pp. 335–349, 2013. “Millimeter-wave human blockage model enhancements for direc-
[2] S. Rangan, T. S. Rappaport, and E. Erkip, “Millimeter-wave cellular tional antennas and multiple blockers,” IEEE Communications Letters,
wireless networks: Potentials and challenges,” Proceedings of the IEEE, vol. 25, no. 9, pp. 2776–2780, Sep. 2021.
vol. 102, no. 3, pp. 366–385, Mar. 2014. [25] C. Slezak, M. Zhang, M. Mezzavilla, and S. Rangan, “Understanding
[3] S. Deng, G. R. MacCartney, and T. S. Rappaport, “Indoor and outdoor End-to-End Effects of Channel Dynamics in Millimeter Wave 5G New
5G diffraction measurements and models at 10, 20, and 26 GHz,” in Radio,” in IEEE 19th International Workshop on Signal Processing
IEEE Global Communications Conference (GLOBECOM), Dec. 2016. Advances in Wireless Communications (SPAWC), 2018, pp. 1–5.
[26] M. Jacob, S. Priebe, A. Maltsev, A. Lomayev, V. Erceg, and T. Kurner,
“A ray tracing based stochastic human blockage model for the IEEE
802.11ad 60 GHz channel model,” in 5th European Conference on
Antennas and Propagation (EuCAP), Rome, Italy, Apr. 2011.
[27] A. Maltsev et al., “Channel models for 60 GHz WLAN systems,”
09/0334r8, IEEE Task Group ad (TGad), Tech. Rep., May 2010.
[28] M. Zhang, M. Polese, M. Mezzavilla, J. Zhu, S. Rangan, S. Panwar,
and M. Zorzi, “Will TCP Work in mmWave 5G Cellular Networks?”
IEEE Communications Magazine, vol. 57, no. 1, pp. 65–71, January
2019.
[29] M. Polese, M. Mezzavilla, S. Rangan, and M. Zorzi, “Mobility Man-
agement for TCP in MmWave Networks,” in 1st ACM Workshop on
Millimeter-Wave Networks and Sensing Systems (mmNets), ser. mmNets
’17, Snowbird, Utah, USA, Oct. 2017.
[30] M. Polese, M. Mezzavilla, M. Zhang, J. Zhu, S. Rangan, S. Panwar,
and M. Zorzi, “milliProxy: A TCP proxy architecture for 5G mmWave
cellular systems,” in 51st Asilomar Conference on Signals, Systems,
and Computers, 2017, pp. 951–957.
[31] M. Polese, R. Jana, and M. Zorzi, “TCP in 5G mmWave networks: Link
level retransmissions and MP-TCP,” in IEEE Conference on Computer
Communications Workshops (INFOCOM WKSHPS), 2017, pp. 343–
348.
[32] ——, “TCP and MP-TCP in 5G mmWave Networks,” IEEE Internet
Computing, vol. 21, no. 5, pp. 12–19, Sep. 2017.
[33] M. Mezzavilla, M. Zhang, M. Polese, R. Ford, S. Dutta, S. Rangan,
and M. Zorzi, “End-to-end simulation of 5G mmWave networks,” IEEE
Communications Surveys and Tutorials, vol. 20, no. 3, pp. 2237–2263,
Third Quarter 2018.
[34] Y. Ren, W. Yang, X. Zhou, H. Chen, and B. Liu, “A survey on TCP
over mmWave,” Computer Communications, vol. 171, pp. 80–88, Apr.
2021.
[35] I. Khan, M. Ghoshal, S. Aggarwal, D. Koutsonikolas, and J. Widmer,
“Multipath TCP in Smartphones Equipped with Millimeter Wave
Radios,” in 15th ACM Workshop on Wireless Network Testbeds, Exper-
imental Evaluation and Characterization (WiNTECH), New Orleans,
LA, USA, Jan. 2022.
[36] C. Lee, S. Song, H. Cho, G. Lim, and J.-M. Chung, “Optimal Multipath
TCP Offloading Over 5G NR and LTE Networks,” IEEE Wireless
Communications Letters, vol. 8, no. 1, pp. 293–296, Feb. 2019.
[37] A. Meurer, C. P. Smith, M. Paprocki, O. Čertı́k, S. B. Kirpichev,
M. Rocklin, A. Kumar, S. Ivanov, J. K. Moore, S. Singh,
T. Rathnayake, S. Vig, B. E. Granger, R. P. Muller, F. Bonazzi,
H. Gupta, S. Vats, F. Johansson, F. Pedregosa, M. J. Curry, A. R.
Terrel, v. Roučka, A. Saboo, I. Fernando, S. Kulal, R. Cimrman,
and A. Scopatz, “Sympy: symbolic computing in python,” PeerJ
Computer Science, vol. 3, p. e103, Jan. 2017. [Online]. Available:
https://doi.org/10.7717/peerj-cs.103
[38] S. Gillies et al., “Shapely: manipulation and analysis of geometric ob-
jects,” 2007–. [Online]. Available: https://github.com/shapely/shapely
[39] ITU-R, “Propagation by diffraction,” Recommendation ITU-R P.526-
15, Oct. 2019.
[40] R. Ford, M. Zhang, S. Dutta, M. Mezzavilla, S. Rangan, and M. Zorzi,
“A Framework for End-to-End Evaluation of 5G MmWave Cellular
Networks in Ns-3,” in Proceedings of the ACM Workshop on Ns-3
(WNS3), Seattle, WA, USA, Jun. 2016.
[41] H. Assasa and J. Widmer, “Implementation and Evaluation of a WLAN
IEEE 802.11ad Model in Ns-3,” in Proceedings of the Workshop on
Ns-3 (WNS3), Seattle, WA, USA, Jun. 2016.
[42] ——, “Extending the IEEE 802.11ad Model: Scheduled Access, Spatial
Reuse, Clustering, and Relaying,” in Proceedings of the Workshop on
Ns-3 (WNS3), Porto, Portugal, Jun. 2017.
[43] J. Kunisch and J. Pamp, “Ultra-wideband double vertical knife-edge
model for obstruction of a ray by a person,” in IEEE International
Conference on Ultra-Wideband, Hannover, Germany, Sep. 2008.
[44] P. Testolina, M. Lecci, M. Polese, M. Giordani, and M. Zorzi, “Scalable
and accurate modeling of the millimeter wave channel,” in International
Conference on Computing, Networking and Communications (ICNC),
Feb. 2020, pp. 969–974.
[45] T. R. Henderson, M. Lacage, G. F. Riley, C. Dowell, and J. Kopena,
“Network simulations with the ns-3 simulator,” vol. 14, no. 14, 2008,
p. 527.