IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO.

3, THIRD QUARTER 2019 2361

A Survey of Air-to-Ground Propagation Channel

Modeling for Unmanned Aerial Vehicles
Wahab Khawaja , Ismail Guvenc , David W. Matolak , Uwe-Carsten Fiebig, and Nicolas Schneckenburger

Abstract—In recent years, there has been a dramatic increase $100 billion market opportunity exists for UAVs in commer-
in the use of unmanned aerial vehicles (UAVs), particularly cial, civil government, and military sectors combined between
for small UAVs, due to their affordable prices, wide avail- now and 2020 [17].
ability, and relative ease of operability. Existing and future
applications of UAVs include remote surveillance and monitoring, Wireless connectivity with UAVs is a major enabler for
relief operations, package delivery, and communication back- integration of UAVs into a national air space and for facil-
haul infrastructure. Additionally, UAVs are envisioned as an itating new use cases. However, UAV operating environments
important component of 5G wireless technology and beyond. The and scenarios introduce unique technical challenges, and these
unique application scenarios for UAVs necessitate accurate air- are recently being investigated by telecommunications com-
to-ground (AG) propagation channel models for designing and
evaluating UAV communication links for control/non-payload as
panies such as AT&T [18], Vodafone [19], Ericsson [20],
well as payload data transmissions. These AG propagation mod- Nokia [21], [22] and Qualcomm [23], among others. One of
els have not been investigated in detail, relative to terrestrial the major challenges is to have realistic air-to-ground (AG)
propagation models. In this paper, a comprehensive survey is propagation models for various UAV operating environments
provided on available AG channel measurement campaigns, large and scenarios, in order to fulfill the ever increasing demands
and small scale fading channel models, their limitations, and
of high rate data transfer for emerging UAV applications.
future research directions for UAV communication scenarios.
In particular, having accurate characterization of the AG
Index Terms—Air-to-ground (AG), channel measurement, channel is of paramount importance for designing robust
channel modeling, drone, large and small scale fading, sounding,
and effective waveforms, modulation techniques, resource
unmanned aerial vehicle (UAV).
allocation and link adaptation approaches, and multiple
antenna techniques.
The AG channel for UAVs has not been studied exten-
I. I NTRODUCTION sively when compared with the existing literature on the
NMANNED aerial vehicles (UAVs), also commonly
U referred as “drones”, have long been used for military
and specialized applications [1]–[5]. Owing to recent tech-
terrestrial propagation channels. The available AG propaga-
tion channel models used for higher altitude aeronautical
communications generally cannot be employed directly for
nological advancements, they have attracted major attention low-altitude UAV communications due to differences in chan-
from industry for uses (“use cases”) from delivery to commu- nel scattering environment. Small UAVs may also possess
nications, surveillance, inspection, transportation, search and distinct structural and flight characteristics such as different
rescue, among others [6]–[15]. These UAVs can vary in size airframe shadowing features due to unique body shapes and
from small toys that fit in the palm of a human hand (where materials, and potentially sharper pitch, roll, and yaw rates of
the “unmanned” designation obviously becomes unnecessary) change during flight. In this survey, we will provide a com-
to large military aircraft such as the General Atomics MQ-9 prehensive, unified review of the existing work on UAV AG
Reaper (commonly termed Predator) [16], with a wingspan propagation channels. We will discuss recent channel measure-
over 15 m. The small battery powered toys can typically ment campaigns and modeling efforts to characterize the AG
fly for up to 15 minutes, whereas the larger UAVs can be channel for UAVs. We will also describe future research chal-
designed for long-endurance (30 hours) and high-altitude oper- lenges and possible enhancements relating to UAV propagation
ations (higher than 15 km). According to Goldman Sachs, a channels.
Manuscript received January 1, 2018; revised June 30, 2018 and December
The rest of the paper is organized as follows. Section II
24, 2018; accepted March 21, 2019. Date of publication May 8, 2019; date provides a review of various use cases of interest for UAVs
of current version August 20, 2019. This work was supported in part by NSF to provide some context, and gives a brief literature review
under Grant CNS-1453678, and in part by National Aeronautics and Space on existing efforts for modeling propagation characteristics of
Administration (NASA) through Federal Award under Grant NNX17AJ94A.
(Corresponding author: Wahab Khawaja.) aerial links. Subsequently, Section III explains some of the
W. Khawaja and I. Guvenc are with the Department of Electrical and unique propagation channel characteristics for the UAV AG
Computer Engineering, North Carolina State University, Raleigh, NC 27606 channels such as operating frequencies, scattering, antenna
USA (e-mail: [email protected]; [email protected]).
D. W. Matolak is with the Department of Electrical Engineering, University effects, and Doppler, and compares the differences with terres-
of South Carolina, Columbia, SC 29208 USA (e-mail: [email protected]). trial channels. A review of some key considerations for AG
U.-C. Fiebig and N. Schneckenburger are with the German Aerospace channel measurements is provided in Section IV, including
Center (DLR), Institute of Communications and Navigation, 82234 Weßling,
Germany (e-mail: [email protected]; [email protected]). measurement frequencies, configurations, and environments,
Digital Object Identifier 10.1109/COMST.2019.2915069 unique challenges for AG measurements, different sounding
1553-877X  c 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2362 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

waveform types for AG measurements, and the effect of which they communicate. These GSs are also usually the UAV
elevation angle on measurement results. control stations.
Section V continues to review the existing UAV AG prop-
agation measurements and some representative simulation
studies in the literature. In particular, path loss and shad- A. UAV Use Cases
owing, delay dispersion, narrowband fading, Doppler spread, Use of commercial UAVs has recently seen exceptional
throughput and bit error rate characteristics, and the effects growth that is forecast to continue in the forseeable future.
of different measurement environment types are reviewed The easy operability, availability of multiple flight controls,
for some existing studies in the literature. Section VI dis- high maneuverability, and increasing payload weight have led
cusses AG propagation channel models, including models to the use of UAVs for many real time civilian applications.
based on deterministic and stochastic models, their combi- This includes for examples remote surveillance, filming, disas-
nation, and ray tracing simulations. After classification and ter relief, transport of goods, and communication relaying, as
review of different channel model types, we provide a survey well as recreation. According to statistics provided by the mar-
of path loss and large scale fading models, airframe shad- ket research company Tractica, the shipment of commercial
owing, small scale fading models, modeling of intermittent UAV units is expected to reach 2.7 million in 2025 with the
multipath components (MPCs), effect of frequency bands, and services offered rising to $8.7 billion in the next decade [25].
multiple-input-multiple-output (MIMO) propagation models. Another promising application of UAVs is in supporting
We also review the recent 3GPP AG UAV channel models, broadband cellular communications, and wireless local area
and provide a comparison of existing AG propagation mod- networks (WLANS) such as IEEE 802.11. Cellular use may
els with each other and with traditional cellular and satellite be in hot spot areas during peak demand events, and in
channel models. cases of a natural calamity where the existing communication
Future challenges and research directions for UAV chan- infrastructure is damaged. It is expected that future 5G imple-
nel measurements and modeling are provided in Section VII, mentation will include UAVs as autonomous communicating
and concluding remarks follow in Section VIII. All acronyms nodes, possibly for providing low latency and highly reli-
and variables used throughout this survey paper are given in able communications. Qualcomm is testing the operability of
Table I and Table II, respectively. Multiple tables and figures UAVs for current LTE and future 5G cellular applications [26].
are provided to tabulate, classify, and review existing aerial Facebook and Google are also exploring the possibility of
channel modeling studies in the literature in a unified man- using UAVs for Internet connectivity to remote areas [27].
ner. Table III provides a review of AG channel measurements
in the literature along with their measurement configurations,
Table IV classifies related measurement studies with respect B. Literature Review on Aerial Propagation
to five different measurement environments, Fig. 2 provides a
The available UAV based AG wireless propagation channel
taxonomy of measurement scenarios and related literature in
research can be largely categorized into two major portions.
terms of aerial vehicle type and measurement environment,
The first one is payload communications, where the payload
Table V reviews the existing literature on large scale AG
can be narrow-band or wide-band and is mostly application
propagation and summarizes key path loss parameters, Fig. 5
dependent. The second one is control and non-payload com-
classifies the literature on UAV AG channel models in terms
munications (CNPC) for telemetric control of UAVs. Most of
of deterministic and stochastic models, and finally Table VI
the CNPC employs the unlicensed bands, e.g., 2.4 GHz, and
summarizes small scale AG model parameters documented in
5.8 GHz; this is not preferred by the aviation community as
the literature.
these bands can be congested and may be easily jammed. In
the USA, CNPC is potentially planned for a portion of L-band
II. UAV U SE C ASES AND L ITERATURE R EVIEW (0.9 GHz - 1.2 GHz) and C-band (5.03 GHz - 5.091 GHz),
Various organizations have developed classifications for although as is common in spectrum allocation, use of these
UAVs according to size, with designations large, medium, bands is still being negotiated [28], [29]. Channel measure-
and small being typical. In the U.S., the Federal Aviation ments and modeling for UAVs are (other than bandwidth and
Administration (FAA) has issued rules for small UAVs weigh- carrier frequency) largely independent of whether signaling is
ing less than 55 pounds (25 kg) [24]. Highlights of these for payload or CNPC.
rules include the requirement for a visual line-of-sight (LOS) AG communications can be traced back to 1920 [30], with
from pilot to aircraft, flight under daylight or during twilight manually operated radio telegraphs. Lower frequency bands
(within 30 minutes of official sunrise/sunset) with appropri- were used in the early 1930s but did not support simultaneous
ate lighting for collision avoidance, a maximum flight ceiling voice communications in both directions (AG and ground-to-
of 400 feet (122 m) above the ground (higher if the UAV is air (GA)). From the early 1940s, double sideband amplitude
within 122 m of a construction site), and a maximum speed modulation (DSB-AM) in the very high frequency (VHF) band
of 100 mph (87 knots, or 161 km/h). Restrictions also apply (118 MHz - 137 MHz) was adopted for voice communications
regarding proximity to airports, and generally, a licensed pilot between pilots and ground controllers. This system supported
must operate or supervise UAV operation. In this survey paper a maximum of 140 channels until 1979. Multiplexing and
our focus is on the smaller UAVs, and specifically on the air- multiple access were frequency division with manual chan-
ground (AG) channel between these UAVs and the GSs with nel assignment by air traffic control. In more dense air traffic

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2363

spaces, to enable larger numbers of simultaneous transmis- channel was provided. A general finding was that no accu-
sions, 25 kHz DSB-AM channels were subdivided into three rate and comprehensive wideband AG propagation channel
channels of width 8.33 kHz. model based on empirical data existed for L-band and C-band.
The civilian aeronautical AG communications continues to However, the literature review content in the survey is now
use the reliable analog DSB-AM system today, although since dated and new research has since appeared in the literature. A
1990 some small segments of the VHF band in some geo- similar short description of overall propagation channel char-
graphic locations are being upgraded to a digital VHF data link acteristics for UAVs was provided in [40] with analysis of two
that can in principle support 2280 channels [31], [32]. This ray geometrical model and its applications in different scenar-
system employs time-division as well as frequency-division, ios. Limitations of existing AG channel models for UAVs were
with single-carrier phase-shift keying modulation. Military also discussed.
AG communications uses different frequency bands (ultra- A recent survey on AG propagation channel modeling
high frequency) and modulation schemes for short and long was provided in [41]. The survey discussed measurement
ranges [33]. Due to very low data rates, the civil aviation and analytical channel models available in the literature. The
systems cannot support modern AG communication require- AG propagation channel measurements were divided into
ments. In 2007, use of portion of the L-band was suggested three parts. The first part covered narrowband, and wideband
for new civil aviation systems, and two such systems known channel measurements, the second part discussed channel mea-
as L-band Digital Aeronautical Communications Systems, or surements using 802.11 radios, whereas the third part covered
LDACS, were developed [32]. Due to compatibility with the cellular infrastructure based AG propagation channel mea-
numerous existing systems that operate in the L-band, the surements. A table of AG propagation channel measurements
LDACS system is still being refined. LDACS is currently being available in the literature was also provided. Additionally, large
standardized by the International Civil Aviation Organization. scale and small scale fading statistics and their respective mod-
There are numerous studies available in the literature on els available in the literature were provided. The survey also
the characteristics of aeronautical channels [31], [34]–[37]. covered analytical channel models from the literature on AG
Aeronautical communications can be broadly classified into propagation. The analytical channel models were divided into
communications between the pilot or crew with the ground three categories, namely, deterministic, stochastic (based on
controller and wireless data communication for passengers. time delay line (TDL)), and geometric. Additionally, the sur-
Both of these types of communication are dependent on the vey provided air-to-air (AA) channel characterization studies
flight route characteristics. In [34] the propagation channel is from the literature with a common result of small path loss
divided into three main phases of flight, termed as parking exponent (PLE) as compared to terrestrial or AG propagation.
and taxiing, en-route, and take off and landing. Each phase of Towards the end of the survey, important issues relating to
flight was described by different channel characteristics (type AG propagation were discussed including airframe shadowing,
of fading, Doppler spread, and delay), but this relatively stationary interval and diversity gain.
early paper was not comprehensive nor fully supported by
measurements. III. UAV AG P ROPAGATION C HANNEL C HARACTERISTICS
There are also long distance AG propagation channel studies In this section, salient characteristics of UAV AG prop-
available for satellites and high altitude platforms (HAPs). The agation channel are described. A common AG propagation
AG propagation channel in these studies can be considered as scenario is shown in Fig. 1 in the presence of terrestrial obsta-
a UAV communication channel, but due to long distances from cles which are also commonly referred as scatterers. In the
the earth surface, normally greater than 17 km, modeling of figure, hG , hS , hU represents the height of the ground sta-
these links may also need to take into account upper atmo- tion (GS), scatterers, and UAV above the ground, respectively,
spheric effects. Depending on frequency and UAV altitude, d is the slant range between the UAV antennas and the GS, and
they may also be much more susceptible to lower tropospheric θ is the elevation angle between GS and UAV antennas. (We
effects such as fading from hydrometeors [38]. For most of note that airborne scatterers may be present as well, but for this
these longer distance platforms, a LOS component is required paper, for the AG link, we neglect this secondary condition.)
because of power limitations, hence the AG channel amplitude
fading is typically modeled as Ricean [39]. As the deploy- A. Comparison of UAV AG and Terrestrial Propagation
ment of UAVs as communication nodes in the near future is
The AG channel exhibits distinctly different characteristics
expected to be at much lower altitudes compared to that of
from those of other well studied terrestrial communication
HAPs and satellites, in this survey we focus only on lower
channels, e.g., the urban channel. There is the inherent advan-
altitude UAV AG propagation channels.
tage over terrestrial communications in terms of a higher
likelihood of LOS propagation. This reduces transmit power
C. Existing Surveys on UAV AG Channel Propagation requirements and can translate to higher link reliability as well.
There are some survey studies available for UAV AG prop- In cases where only non-LOS (NLOS) paths exist, when the
agation channel measurements and modeling [37], [40], [41]. elevation angle to the UAV is large enough, the AG channel
In [37], a survey on wideband AG propagation channel may incur smaller diffraction and shadowing losses than near
was provided with focus on the L-band and C-band envi- ground terrestrial links.
sioned as possible candidates for future non-payload AG On the other hand, the AG channel can exhibit significantly
communications. A tapped delay line model for time varying higher rates of change than typical terrestrial communication

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2364 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

Fig. 1. A typical air-to-ground propagation scenario with a UAV.

channels because of UAV velocities. When the channel is mod- C-bands envisioned for CNPC, and for the currently popular
eled statistically, this can mean that the channel’s statistics unlicensed bands for payload communications, tropospheric
are approximately constant (the channel is wide-sense sta- attenuations from atmospheric gases and hydrometeors are
tionary (WSS)) for only a small spatial extent. This is often mostly negligible. This will not be true for operation at higher
loosely termed “non-stationarity.” If the UAV is not in the frequency bands, e.g., at Ku, Ka, and other so-called millime-
direct vicinity of scattering objects or the GS, the characteris- ter wave (mmWave) bands, which may be as high as 100 GHz.
tics of the channel could instead actually change very slowly, These higher frequency bands can hence suffer both larger
especially for hovering UAVs. In such a case, adverse propa- free-space path loss (FSPL) as well as tropospheric attenua-
gation conditions, e.g., deep fades of the received signal, may tions. Because of this, these frequency bands will generally be
last several seconds or even minutes, hence common commu- used for short-range AG links.
nication techniques of interleaving or averaging would not be In contrast to the attenuation characteristics compared with
affective. In many cases, when UAV altitudes are well above lower frequency bands, mmWave bands offer a large amount
scattering objects, the AG channel’s “non-stationarity” will be of bandwidth, which is their primary appeal for 5G cellu-
attributable to the direct surroundings of the GS, e.g., the close lar systems. Large bandwidths can be more robust to the
by buildings or the ground surface composition around the GS. larger values of Doppler shift and Doppler spread encoun-
Additionally, AG communications with UAVs face many tered with UAVs moving at high velocity. The mmWave bands
other challenges, due to arbitrary mobility patterns and diverse may be supported by UAVs in the future for high speed data
types of communication applications [42]–[45]. As an aerial applications as communication payload; however, for CNPC,
node, some of the UAV specifics that need to be taken into mmWave bands may not be a good option due to higher atten-
account include airframe shadowing, mechanical and elec- uation and smaller diffractions, making these bands unreliable
tronic noise from UAV electronics and motors, and finally in case of even small blockages.
antenna characteristics, including size, orientation, polariza-
tion, and array operation (e.g., beam steering) for MIMO
systems. For UAVs in motion, the effect of Doppler shifts C. Scattering Characteristics for UAV AG Propagation
and spread must also be considered for specific communica- In an AG propagation channel using UAVs, the MPCs
tion applications [46], [47]. For a given setting, an optimum appear due to reflections from the earth surface, from terrestrial
UAV height may need to be considered, e.g., for maintaining objects (ground scatterers), and sometimes from the airframe
LOS in that environment [48]. of the UAV itself. The characteristics of the channel will be
dependent on the material, shape, and size of the scattering
objects. The strongest MPC apart from the LOS component
B. Frequency Bands for UAV AG Propagation in an AG propagation scenario is often the single reflection
As with all communication channels, a fundamental consid- from the earth surface. This gives rise to the well known two
eration is the frequency band, since propagation characteristics ray model.
can vary significantly with frequency. There are typically two For high enough frequencies, the scatterers on the ground
popular bands, 2.4 GHz and 5.8 GHz, used by commercial and around the UAV can be modeled as points scatterers on
UAVs for CNPC operations during flight. However, other the surface of two respective cylinders or spheres [49], [50]
frequency bands may be used for additional features, e.g., for or ellipsoids, and these can be bounded (truncated) by
transferring videos from the UAV to the GS at 3.4 GHz. The intersection of the elliptical planes on the ground [51], [52].
5.8 GHz band is a better choice than the 2.4 GHz for major- These topologies can help in deriving geometrical charac-
ity of scenarios, because of lower interference. For the L and teristics of the AG propagation scenario. The distribution of

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2365

TABLE I
ACRONYMS U SED IN T HIS PAPER

scattering objects, on land or water, can be modeled stochas- The majority of AG channel measurements employ stand
tically, and this concept can be used to create so-called alone (single) antennas, whereas in [57], an antenna array
geometrically-based stochastic channel models (GBSCMs). is used. There are some single-input-multiple-output (SIMO)
For aircraft moving through an area above such a distribu- and MIMO antenna configurations available in the literature
tion, this gives rise to intermittent MPCs [53], as also seen in for AG propagation measurements [58], [59]. Omni-directional
vehicle-to-vehicle channels. antennas are most popular for vehicular communications due
In case of propagation over water the path loss (PL) is to their superior performance during motion, whereas direc-
similar to that of free space [54], with a strong surface reflec- tional antennas (having better range via directional gain) can
tion. The other MPCs from the water surface are weaker, perform poorly during motion due to mis-alignment losses.
and of approximately equal power and time-of-arrival (TOA), With high maneuverability of UAVs during flight, omni-
whereas MPCs from obstacles on the water surface, e.g., large directional antennas are generally better suited than directional
ships, can be stronger. antennas. A potential major drawback of any antenna on-board
UAVs is the shadowing from the body of the UAV. Similarly,
D. Antenna Configurations for UAV AG Propagation orientation of antennas on-board UAVs can affect the commu-
The antenna is one of the critical components for AG nication performance [60], [61]. The omni-directional antenna
communications due to limited space, and limitations of orientation is found to affect the received signal strength (RSS)
the aerodynamic structure [55], [56]. Factors that affect AG and the system throughput. A better throughput performance
link performance are the number, type and orientation of was reported with horizontal-horizontal orientation as com-
the antennas used, as well as the UAV shape and material pared to vertical-vertical orientation in [60], whereas in [61],
properties. it was observed that horizontal antenna orientation can

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2366 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

TABLE II
VARIABLES U SED IN T HIS PAPER The use of multiple antennas to enable diversity can yield
spatial diversity gains even in sparse multipath environ-
ments [62], [63]. It was demonstrated that using measurements
that employ MIMO and SIMO in AG propagations can yield
spatial diversity dependent on the antenna geometry and sur-
rounding environment. Similarly, multiple antennas can be
used for spatial selectivity such as beam forming/steering.
However, due to limited space on-board UAVs, space diversity
using multiple antennas is difficult to achieve, especially for
lower carrier frequencies. Beamforming using antenna arrays
operating at mmWave frequencies, for example, can be used
to overcome fading and improve coverage, but array process-
ing will require high computational resources on-board. The
employment of MIMO systems for enhancing the channel
capacity of the AG propagation channel has been suggested
in [64], [65]. By changing the diameter of a circular antenna
array and the UAV flying altitude, different values of MIMO
channel capacity were obtained [64]. On the other hand,
in [65], optimizing the distance between the antenna elements
using linear adaptive antenna arrays was proposed to increase
MIMO channel capacity.

E. Doppler Effects
Doppler shifts can introduce carrier frequency offset (CFO)
and inter-carrier interference (ICI), especially for orthogo-
nal frequency division multiplexing (OFDM) implementations.
There are several studies that consider modeling of Doppler
spread [34], [46], [47], [66]–[70] for AG scenarios. Some
channel access algorithms, e.g., multi carrier code division
multiple access, have been shown to be robust against Doppler
spread in AG propagation [71], while Doppler spread is
controlled by adjusting the carrier spacing of OFDM.
Due to UAV motion, there are Doppler frequency shifts
that depend on the velocity of the UAV and the geometry.
Higher Doppler frequency presents a problem if the differ-
ent signal paths are associated with largely different Doppler
frequencies, yielding large Doppler spread. This can happen
if the aircraft is relatively close to the GS. If the aircraft is
further away from the GS, and at sufficient altitude, the paths
should all have a very similar Doppler frequency as the objects
in the close surroundings of the GS causing MPCs are seen
all under similar angles from the aircraft. The effect of a large
Doppler frequency that is constant for all MPCs should be
well mitigated by frequency synchronization.
In order to describe the statistical characteristics of a fading
channel, typically first and second order fading statistics are
used. The majority of the AG propagation literature discusses
first order fading statistics. The second order statistics of
envelope level crossing rate and average fade duration are dis-
cussed in [49], [68], but many authors address other second
order properties, primarily correlation functions in the time or
frequency domains.

IV. AG C HANNEL M EASUREMENTS : C ONFIGURATIONS ,

C HALLENGES , S CENARIOS , WAVEFORMS
help to overcome the difference in the yaw, similarly, Several AG channel measurement campaigns using piloted
vertical orientation was found to perform better during aircraft and UAVs have been recently reported in the lit-
tilting of UAV. erature. These measurements were conducted in different

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2367

environments and with different measurement parameters. In channel characteristics. For a given environmental scenario,
this section, we provide a brief classification of these mea- there may be an optimal height of the GS [72], e.g., this might
surements based on environmental scenario, sounding signal, be a balancing of attenuation and multipath diversity.
carrier frequency, bandwidth, and antenna specifications and Example propagation measurements using rotorcraft and air
placement. As available, we also provide UAV type and speed, balloons during flight and hovering are available in [61], [68],
heights of UAV and GS from terrain surface, link distance [72], [85]. These AG propagation measurements were obtained
between transmitter (TX) and receiver (RX), elevation angle, at different UAV heights ranging from 16 m to 11 km, and
and the channel statistics provided by the cited authors. These link distances 16.5 m to 142 km. The UAV latitude, longitude,
channel measurement parameters are given in Table III. yaw, pitch, and roll readings are typically obtained from global
In the reported AG propagation measurements, either TX positioning system (GPS) RXs and often stored on-board.
or RX on UAV/GS is stationary. Measurements with both Apart from conventional AG channel sounding, there are
TX and RX moving for AG propagation are rare. A notable some indirect UAV AG channel measurements available from
contribution of wide-band AG propagation measurements is use of radios employing different versions of protocols of the
available in the form of multiple campaigns conducted in the IEEE 802.11 standards [60], [61], [84], [85]. The IEEE 802.11
L and C bands using SIMO antenna configuration for differ- supported devices offer a very flexible platform and may pro-
ent terrain types and over water/sea [53], [54], [58], [73]–[79]. vide insight for UAV deployments in different topologies and
The rest of the cited channel measurements are conducted applications, e.g., UAV swarms. Yet because of the specific
in different frequency bands ranging from narrow-band to features of 802.11, the resulting measurements are applicable
ultra-wideband (UWB) with various types of sounding signals. to particular protocol setup and radio configuration, and rarely
provide detailed propagation channel characteristics.
A. Channel Measurement Configurations AA communications with UAVs has not been studied exten-
sively in the literature [94]. The AA communications is
These channel measurements used different types and con-
particularly important for scenarios where multiple drones
figurations of antennas. The most commonly used antenna type
communicate among a swarm. This swarm then usually com-
is omni-directional and the most commonly used configura-
municates with one or more GS via a back-haul link from one
tion is single-input-single-output (SISO). The positioning of an
or several of the UAVs. The AA communications is similar
antenna on the UAV is important to avoid both shadowing from
to free space with a strong LOS and often a weak ground
the airframe and disruption of the aircraft’s aerodynamics. In
reflection, but this is dependent on the flight altitude and envi-
the majority of measurements the antennas were mounted on
ronment. The communication channel is mostly non-dispersive
the bottom of the aircraft’s fuselage or wings. The orientation
for higher altitudes but can be rapidly time-varying, depen-
of antennas on UAV and ground can also affect the signal char-
dent on the relative velocities of the UAVs and the scattering
acteristics [60], [61], [84], [85]. This characteristic is most
environment [95].
important during banking turns, and when the aircraft pitch
angle deviates from horizontal. The elevation angle between
the TX and the RX antennas is dependent on the height of B. Challenges in AG Channel Measurements
UAV and GS and often continuously varies during the flight. There are many challenges in AG channel measurement
In the majority of the communication applications envi- campaigns as compared to terrestrial measurements. The
sioned for UAVs, the aerial node is expected to be stationary biggest challenges are the payload limitation of the UAVs, and
(or mostly so) in space for a given time. As noted, for com- the operating range and height of UAVs, which in the USA
munications with a mobile UAV, the velocity will affect the is set by the FAA [96]. Larger UAVs also incur larger test
channel statistics. For UAVs operating at higher velocities, costs. Due to restrictions on the height of UAVs above ground,
the coherence time of the channel decreases, and this trans- UAVs at lower altitudes have lower LOS probability and are
lates into a larger Doppler spread. For connections to multiple hence more susceptible to shadowing, especially in suburban
UAVs, where hand-overs are required, this means that the num- and urban areas. Due to limitations on payload, higher trans-
ber of handovers will also generally increase with velocity, and mit power measurements on-board the UAVs are difficult to
this will require additional processing. Additionally, higher achieve, and similarly, complex RX processing on-board UAVs
velocities will result in increased air friction and mechani- can consume a prohibitive amount of power.
cal turbulence that generally result in increased noise levels. Other challenges include wireless precise frequency syn-
Many of the AG channel measurements in the literature have chronization for channel measurements, varying conditions of
been conducted with fixed wing aircraft with maximum speeds the terrain during flight, meteorological conditions (winds and
varying from 17 m/s to 293 m/s. The speed of rotorcraft and rain), antenna positioning on the UAV, precise location mea-
air balloons is much less than that of fixed wing aircraft, and surement of UAVs in space over time, diverse telemetry control
ranges from 8 m/s to 20 m/s. for different types of UAVs having specific latencies, band-
The height of the UAV above ground is an important chan- width and reliability issues, and limited flight time for most
nel parameter and will also affect the channel characteristics. small UAVs due to limited battery life [42]–[44]. Due to the
For example, increasing the height of the UAV usually results motion of UAVs in three dimensional space, it is challenging
in reduced affect of MPCs [93] from surrounding scatterers. to precisely measure the distance between the UAV and the
Another benefit of higher UAV altitude is larger coverage area GS. Momentary wind gusts that cause sudden shifts in UAV
on the ground. Similarly, the height of GS will also affect the position can make it difficult to accurately track the UAV path.

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2368 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

TABLE III
R EVIEW OF I MPORTANT E MPIRICAL AG C HANNEL M EASUREMENT S TUDIES IN THE L ITERATURE AND T HEIR M EASUREMENT C ONFIGURATIONS ON
S OUNDING WAVEFORM , F REQUENCY, BANDWIDTH , T RANSMIT P OWER , UAV T YPE , T RANSMIT /R ECEIVE H EIGHT

The most common technique of measuring the instantaneous C. AG Propagation Scenarios

distance is by using GPS traces on both the UAV and GS, but A typical type of terrestrial channel sounding equipment, a
of course GPS devices have accuracy limitations and naviga- vector network analyzer, cannot be used for UAV based AG
tion signals may also be susceptible to interference in different channel sounding due to payload constraints, physical synchro-
flying zones. nization link requirements, and UAV mobility [97]. Therefore,

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2369

TABLE IV
L ITERATURE ON UAV AG P ROPAGATION C HARACTERISTICS FOR F IVE D IFFERENT F LIGHT E NVIRONMENTS

Fig. 2. Measurement scenarios for UAV AG propagation channel.

channel sounding for both narrow-band and wide-band chan- but do not have robust movement characteristics. The non-
nels using impulse, correlative, or chirp sounding techniques balloon UAVs can be broadly classified as fixed wing and
are employed, where the RX is typically on the ground due rotorcraft. The fixed wing UAVs can glide and attain higher
to payload and processing constraints. air speeds and generally travel farther than the rotorcraft, but
Proper selection of channel measurement parameters in a rotorcraft are more agile, e.g., most can move straight ver-
given environment is critical for obtaining accurate channel tically. Rotorcraft also have the ability to hover, which is
statistics for a given application. The AG propagation envi- not possible for nearly all fixed wing UAVs. The UAV AG
ronment is generally classified on the basis of the terrain propagation scenarios in different environments with particu-
type, namely flat, hilly, mountainous, and over water. A par- lar characteristics are described in Table IV. In the rest of this
ticular terrain can have a given cover, e.g., grass, forest, or subsection, we review the different AG measurement scenarios
buildings. The most widely accepted terrain cover classifi- depicted in Fig. 2.
cation is provided by the International Telecommunication 1) Open Space: A major part of the literature on AG prop-
Union (ITU) [98]. In this survey we classify the cited mea- agation covers open (flat) terrain. This open terrain can have
surement scenarios as open (flat), hilly/mountainous, and over different terrain covers that affect the channel characteristics.
water. Each scenario can be subdivided on the basis of the One of the major terrain cover types is buildings. The distri-
terrain cover as shown in Fig. 2. bution of building sizes, heights, and their area-wise densities
For any environment, different types of radio controlled allows sub-classification into urban, suburban and rural areas
UAVs can be used. Balloons or dirigibles are simple to operate as depicted in Fig 2. In case of urban and suburban areas,

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2370 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

(a) (b)

(c) (d)

Fig. 3. Sounding signals (a) Chirp signal, (b) Short duration Gaussian pulse sounding signal at center frequency of 1 MHz and fractional bandwidth of 60%,
(c) PRN sequence of polynomial degree 10 shown half of the overall period, (d) OFDM sounding signal resource mapping with 64 sub-carriers, 16 symbols
and 6 pilots.

there is a higher concentration of man made structures in a satellite channels, e.g., [100]–[102]. In these studies, propa-
given space, e.g., buildings, roads, bridges, large signs, etc. gation effects–typically attenuation–from particular volumes
The distribution (and composition) of these complex scatterer of trees, along with temporal fade statistics are analyzed for
structures can strongly influence the channel characteristics. In long range AG communications. Generally for AG propaga-
rural areas, typically buildings are sparse, and of lower height tion with a GS within a forest, the channel characteristics are
than in urban settings, although large warehouses and other dominated by the type and density of trees. Small UAVs within
structures could yield strong MPCs. a forest experience different scattering characteristics depend-
2) Hilly/Mountainous: The hilly/mountainous terrain is ing upon height, e.g., the scattering near the tree trunk will
characterized by uneven ground heights; equivalently, a large be different from that near the tree crown [86]. The scat-
standard deviation of terrain height. The propagation PL in tering is also dependent on the type and density of leaves
hilly and mountainous areas will mostly follow the two ray and branches of the trees, and hence for deciduous trees, can
model with adjustments due to surface roughness, and poten- vary seasonally.
tially reflections from smooth sections of mountain slopes or 4) Water/Sea: The AG propagation channel for over water
an occasional large building. The PL over or beyond terrain settings is similar to that for open settings, with different
obstructions can employ established models for diffraction, surface reflectivity and roughness than ground. The PL can
e.g., [99] but with first Fresnel zone clearance between the be represented using a two ray PL model, with variations
TX and the RX, PL is close to free space [75], [76]. Channel attributable to surface roughness (see small-scale fading in
dispersion, typically quantified by the root mean square-delay the following section). The RMS-DS in this case is generally
spread (RMS-DS), is generally smaller than in urban/suburban smaller than in environments with a large number of obsta-
environments [76] but can be large if a strong reflection occurs cles (urban, suburban), although if large objects are on or just
from a large and distant mountain slope. Generally, hilly and off shore, these may produce significant reflections and large
mountainous settings present fewer reflections than more pop- delay spreads if geometry permits.
ulated regions because of the absence of large numbers of In case of propagation over sea, the height of waves in a
nearby scatterers. rough sea can introduce additional scattering and even diffrac-
3) Forest: There are few comprehensive studies covering tion for very low height stations on the sea. An interesting
AG propagation in forests, especially with UAVs, although propagation phenomenon that can also occur over sea is
there are numerous publications for roadside shadowing for ducting, where anomalous index of refraction variation with

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2371

height results in propagation loss less than that of free (2) NLOS communication for a substantial fraction of users.
space [91]. This phenomenon is dependent on frequency and The effect of elevation angle can also vary with the type of
meteorological conditions, and is thus typically addressed antennas used. If the communication is directional, and the
statistically [103]. UAV and GS beams are aligned, then the effect of the ele-
vation angle is negligible. However, if the communication is
D. AG Channel Sounding Waveforms omni-directional, then the effect of the antenna gain at higher
As noted in [67], [92], common channel sounding sig- elevation angles will be significant.
nals include short pulses (approximately impulses), direct In [106], [107], the effect of elevation angle at different UAV
sequence spread spectrum (DS-SS) signals for correlative pro- altitudes for different antenna orientations was discussed. The
cessing, linearly varying frequency modulated (chirp) signals, gain of the omni-directional dipole antenna in the elevation
and multi-tone signals. Different example sounding signals are plane was modeled as a trigonometric function of physical
shown in Fig. 3 representing a chirp signal, RF Gaussian pulse, elevation angle between the UAV and the GS. It was shown
pseudo-random number (PRN) sequence, and OFDM sound- that the effect of the elevation angle on the received power was
ing signals. These sounding signals have been used in different essentially deterministic when the UAV is hovering, but was
measurement campaigns summarized in Table III in differ- not as easily modeled when the UAV was in circular motion
ent AG channel measurement scenarios given in Fig. 2. Short around the ground RX.
duration pulses are direct approximations of input impulses The effect of elevation angle on UAV AG omni-directional
and MPCs can be directly measured in the time domain communications at 28 GHz was discussed in [108]. It was
(e.g., via a sampling oscilloscope). The primary drawback observed that received power was mainly dependent on the ele-
is generation of sufficient pulse energies to reach long dis- vation angles of the LOS and ground reflected components (as
tances, and large peak-to-average power ratios (PAPR). The they were the strongest). The received power was a function of
DS-SS signaling uses pseudo-random (PR) sequences to gen- link distance and elevation angle, with the latter a strong func-
erate a wideband noise-like signal that is demodulated with tion of antenna gain: for some distances and angles, antenna
a sliding (or sometimes a stepped) correlator; this correla- gain effects dominated, but at larger distances the antenna gain
tion processing yields an estimate of the channel impulse was less significant than the effect of attenuation (FSPL) with
response (CIR). The DS-SS technique can use binary phase distance.
shift keying transmission and with modest filtering this yields Similarly, in [93], the received power was shown to fol-
a low PAPR. Chirp sounding has the advantage of high low a two ray model. It was also shown that the received
frequency resolution and the potential to sweep over large power and RMS-DS dependence on UAV height is a func-
frequency ranges; PAPR can be the ideal value of unity. The tion of the specific propagation environment. For example, in
chirp technique yields the channel transfer function, from a rural area, the effect of ground scatterers was negligible at
which the CIR is obtained via inverse Fourier transformation. higher UAV altitudes, whereas for an urban area, the effect
Another popular technique is the use of a multitone signal, of scatterers was observed at higher UAV altitudes. Similar
with the idea of sampling the channel transfer function. This is results were reported for TOA, angle of arrival, and angle of
in essence an OFDM based channel sounding. One advantage departure of MPCs in [108], where larger temporal and angu-
of using OFDM sounding is that known data can be used for lar spreads were observed at UAV heights comparable to the
sounding, hence allowing some data transmission along with scatterer heights.
channel sounding [104]. The OFDM signals have the advan-
tage of a flat spectrum but of course a sinc (sin(x)/x) delay V. UAV AG P ROPAGATION M EASUREMENT AND
domain response and a large PAPR. Details on these various S IMULATION R ESULTS IN THE L ITERATURE
sounding signals can be found in the literature, e.g., [105]. Several types of channel statistics are useful for characteriz-
Different carrier frequencies can be used to sound the AG ing the channel for different applications. For AG propagation,
channel and in principle this is completely arbitrary, but most the channel statistics are similar to those gathered for terres-
measurements aim at frequency bands in which UAV use is at trial channels. In general, propagation channels are linear and
least possible. Measurements have ranged from 100 MHz to time varying, but can sometimes be approximated or mod-
18 GHz with perhaps most of the measurements carried out eled as time-invariant. For linearly time-varying channels, the
in the 5 GHz band (5.06 GHz - 5.8 GHz). Similarly, sound- CIR or its Fourier transform, the time varying channel trans-
ing signal bandwidth varies, from very narrow-band to several fer function (CTF), completely characterizes the channel [53],
tens of MHz or more. In [72], UWB channel sounding with [54], [57], [58], [62], [68], [73]–[79], [92]. As noted, due to
a bandwidth of 2.2 GHz was used, yielding sub-nanosecond relative motion of the UAV, the AG channel may be station-
time resolution. ary only for small distances [58]. Thus, stationary distance
needs to be taken into account when estimating the channel
E. Elevation Angle Effects statistics [54], [109], [110].
It is important to consider the effect of elevation angle Another higher-level parameter that has been used by some
for UAV AG communications, as it is for satellite com- researchers to characterize the quality of the AG propagation
munications. This is in contrast to the case of terrestrial channel is throughput, but of course this is highly dependent
communications, where the effect of the elevation angle is upon the TX and the RX implementation, and parameters of
less significant due to (1) smaller heights of TX and RX, and the air interface, such as the number of antennas and the

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2372 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

transmit power. Hence this measure is of limited use for assess-

ing the AG channel itself. Similarly, for MIMO channels,
beam-forming gain, diversity, and capacity of the channel are
often estimated. Some commonly reported channel character-
istics for AG propagation channels are given in the following
subsections.

A. Path Loss/Shadowing
Most of the AG propagation campaigns address PL and
if present, shadowing, in different scenarios. For AG chan-
nels with a LOS component, PL modeling begins with FSPL;
when the earth surface reflection is present (not blocked or
suppressed via directional antennas), PL can be described by
the well-known two-ray model. Parallel to the developments
in terrestrial settings, most of the measurements employ the
log-distance PL model where the loss increase with distance
is indicated by the PLE. In [72], PL is calculated for open
field and suburban areas for different UAV and GS heights for
a small hovering UAV. Comprehensive PL measurements in L
and C bands were carried out in different propagation scenar-
ios in [53], [54], [58], [73]–[79] as summarized in Table III.
The values of PLE were found to be slightly different for
urban, suburban, hilly, and over water scenarios, but are gen-
erally close to the free-space value of 2 with standard deviation
around the linear fit typically less than 3 dB.
In [61], it was observed that the PLEs for IEEE 802.11 com-
munications were different during UAV hovering and moving
due to different orientations of the on-board UAV antennas.
Therefore, antenna patterns can distort the true channel PL
characteristics and removing their effect is not always easy or
possible. On the other hand, for the specific UAV configuration
Fig. 4. (a) The LOS signal power variation due to ground multipath
used, the resulting PL model is still useful. Typically, PL for propagation. The power is normalized to FSPL, (b) Measurement scenario
LOS and NLOS conditions are provided separately, e.g., [111], environment in [115].
where for the NLOS case, there is an additional small-scale
(often modeled as Rayleigh) fading term, and a constant reflec-
tion term in addition to the LOS PL. Analogously, the LOS was found to be a function of the elevation angle, where
models for L- and C-bands can incorporate Ricean small scale the shadowing magnitude was estimated by using the uniform
effects [58]. In [112], the reported PL is described as a func- theory of diffraction.
tion of the elevation angle between the low altitude platform In Fig. 4(a) we show an example for the variation of the
and GS, θ given as follows: LOS signal power due to ground reflected MPCs versus the
  link distance d. Specifically, this is the combined effect of
Δh
PL = 20 log + 20 log(fMHz ) − 27.55, (1) the LOS component and the unresolved ground reflection.
sin θ
The measurements were taken in a rural environment using
where Δh = hU − hG is the difference between the height of a 10 MHz signal bandwidth. The GS height hG was 23 m.
the low altitude platform (UAV) and the GS, fMHz is the oper- The UAV trajectory is shown in Fig. 4(b). During the mea-
ating frequency expressed in MHz. The argument Δh/ sin θ surements, the specular reflection point first passed over the
is simply the link distance expressed as a function of the roof of a building and then over open grassy fields [115].
elevation angle. From Fig. 4(a) we observe a periodic variation of the received
PL including shadowing is reported in [57], [68], [71], [72], power: an attenuation of the signal by more than 10 dB is not
[113], [114], where we note that in LOS cases without actual uncommon. These signal fades will of course generally nega-
obstruction of the first Fresnel zone, the physical mechanism tively impact the performance of any communication system.
causing PL variation is not actually shadowing but often small- For an increasing link distance the frequency of the variation
scale effects. In [68], PL and its associated shadowing was decreases–a direct manifestation of the two-ray model. Thus
attributed to buildings only when the UAV was flying near the in such a channel, even for a UAV flying at a high speed a
ground whereas when flying higher, actual shadowing was not fade can easily last several seconds. It is essential to note that
present but variation from small-scale fading still occurred. a ground MPC may not always be present, e.g., for the case
One can also estimate losses due to “partial” shadowing by when the ground is a poorly reflecting ground surface, or the
conventional methods. For example the shadowing in [113] surface is very rough relative to the signal wavelength.

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2373

The PL provides complete information on link attenuation, characterize the AG channel amplitude fading. In [57], as gen-
but another indirect parameter often used for channel atten- erally expected, the authors found that the K-factor increased
uation estimation is RSS. In [60], [61], [84], RSS indicator with increasing elevation angle. The Ricean K-factor as a
data for an AG propagation channel based on IEEE 802.11a function of link distance was given in [71], during multiple
transmissions with different antenna orientations was provided. phases of flight (parking and taxiing, take off and landing,
Data on fluctuations in RSS due to multipath fading from tall and en-route). The en-route phase showed the largest K-factor,
building reflections was provided in [81], where the RSS was followed by take off and landing, and parking and taxiing.
found to decrease due to polarization mismatch between the In [86], it was observed that the K-factor will differ with dif-
TX and the RX antennas when the aerial vehicle made a bank- ferent types of scattering trees: values of K ranging from 2 dB
ing turn. The accuracy of RSS values in commercial products to 10 dB were reported.
can vary considerably, so when these are used, care should be The K-factor was measured for both L-band and C-band AG
taken in calibration. propagation in [53], [58], [76], [79] for urban, suburban, hilly
and mountainous settings, and also for over fresh water and sea
scenarios. The mean values of K-factor for urban areas were
B. Delay Dispersion
reported to be 12 dB and 27.4 dB for L-band and C-band
The power delay profile (PDP) is the “power version” of the respectively. The mean K-factor values for hilly and moun-
CIR. This can be computed “instantaneously,” or more tradi- tainous terrain were reported to be 12.8 dB and 29.4 dB for
tionally, as an average over a given spatial volume (where L-band and C-band respectively, whereas for over sea settings,
the channel can be considered WSS). Various AG propaga- K-factor mean values for L-band and C-band were found to
tion studies in different environments have measured PDPs, be 12.5 dB and 31.3 dB, respectively. Worth pointing out is
and via the PDP the most common estimate of the delay- that in these “strong LOS” channels, the K-factor does not
domain dispersion is estimated: the RMS-DS. Other dispersion strongly depend on the GS environment. Also observed was
measures such as the delay window or delay interval are also that the C-band K-factor was larger than the L-band K-factor
sometimes reported. Statistics for the RMS-DS itself are often in all environments. This is attributable to two causes: first, the
computed, e.g., in [57], mean RMS-DS values for different C-band measurement signal bandwidth was larger than that of
elevation angles were reported. As generally expected from L-band, ameliorating fading, and second, for any given inci-
geometry, the RMS-DS was found to decrease as elevation dent angle and surface roughness (e.g., ground, or ocean), as
angle increases. In [72] PDPs were measured for open areas, carrier frequency increases, the surface roughness with respect
suburban areas, and areas covered with foliage. to the wavelength also increases, and hence incident signals are
The Saleh-Valenzuela model, originally developed for scattered in multiple directions rather than being reflected in
indoor channels, is sometimes used to model the PDP when a single direction (toward the RX). With fewer and/or weaker
MPCs appear grouped or “clustered” in delay. This model MPCs at the higher frequency, the K-factor is larger.
specifies the MPCs by such clusters, and the number of clus-
ters is different for different environmental scenarios. PDPs
D. Doppler Spread
were measured for different environments in [54], [74]–[79],
and resulting RMS-DS statistics were provided. As expected, The Doppler effect is a well-known phenomenon for wire-
the delay spread was found to be dependent on the terrain less mobile communications. Considering AG propagation
cover with maximum delay spread values of 4 μs for urban with UAVs in a multipath environment, if we let φi repre-
and suburban settings. The largest RMS-DS values generally sent the angle between the aircraft velocity vector and the
occur when there are large buildings that can provide strong direction from which the i th MPC is received, the Doppler
φi
MPC reflections. For hilly and mountainous terrain, maximum frequency shift of this i th MPC is fdi = v cos λ , where v is
RMS-DS values of 1 μs for hilly regions and 180 ns for the UAV velocity, and λ is the wavelength of the radio wave.
the mountainous terrain were reported. In over water settings, (We assume here that the GS is motionless, else a more general
the maximum RMS-DS value reported was 350 ns. Again, in formulation for the Doppler shift must be used.) If MPCs are
all these settings cited here, a LOS component was present received with different Doppler frequencies this phenomenon
between GS and UAV, hence for the majority of the time, produces spectral broadening, called Doppler spread.
RMS-DS was small, on the order of a few tens of nanosec- In [34], [71], simulations were used to find the Doppler
onds. In [116], a finite-difference time domain model for the shift and its effect on the channel at different phases of
electric field propagating at very low heights over sea was flight (parking and taxiing, en-route, and take off and land-
developed. An RMS delay spread model for VHF to 3 GHz ing). Doppler spread in a multipath environment implementing
was presented, with RMS-DS a function of wave height. OFDM systems was considered in [67], where arriving MPCs
were observed to have different frequency offsets. In such a
case, if the RX CFO synchronizer cannot mitigate the effect
C. Narrowband Fading and Ricean K-Factor of these different frequency offsets, this results in ICI. In [47],
Small scale amplitude fading in AG propagation channels a mitigation technique for Doppler shift was proposed for
usually follows a Ricean distribution due to the presence of a the case where the UAV is relaying between two communi-
LOS component. The Ricean K-factor is defined as the ratio of cation nodes. The UAV acts as a repeater that provides the
dominant channel component power to the power in the sum required frequency shift to mitigate the Doppler effect. A three
of all other received components. The K-factor is often used to dimensional AG Doppler delay spread model was provided

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2374 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

in [46] for high scattering scenarios. Doppler spread for AG level crossing rate and average fade duration were analyzed for
propagation is also discussed in [62], [66], [72], [73], [92]. different heights and horizontal distances of the aerial platform
from the GS. This work is unique as the second order chan-
E. Throughput and Bit Error Rate Statistics nel fading statistics for AG propagation using UAVs are rarely
available in the literature. However, the velocity of the aerial
Other than the channel characteristics reviewed earlier, there
platform considered was very less. Additionally, it would have
are other performance indicators that can be measured. Two
been more interesting if second order statistics were compared
of these are throughput and bit error ratio (BER) with particu-
at different UAV velocities in the same environment.
lar communication technologies. As with RSSI measurements,
In [53], [78], wideband AG propagation channel measure-
these are useful for the particular technology and environ-
ments were reported for suburban and urban areas in the
ment in question, but may offer very little that is directly
L-band and C-band. It was observed that reflections from
relevant to modeling the AG channel. The throughput of an
high rise buildings resulted in increase in the RMS-DS. The
AG propagation channel was investigated in several studies,
large scale and small scale fading at two different frequency
most commonly using the IEEE 802.11 protocol. Throughput
bands in the similar environment were found to be different.
analysis using different versions of the IEEE 802.11 pro-
In [57], wideband CW sounding signal at a center frequency
tocol were carried out in [60], [61], for different antenna
of 2.05 GHz was used in an urban area like environment of
orientations, propagation distances, and UAV elevations. A
a university campus. There were 4 to 6 story tall buildings
throughput analysis of IEEE 802.11n was carried out in [42],
and the terrain was rolling. It was observed that the RMS-
where–as expected–it was found that throughput is directly
DS increased with decreasing the elevation angle, whereas the
dependent on the modulation and coding scheme. Throughput
number of MPCs remain the same, suggesting that at lower
analysis for data relaying and ferrying for an AG propaga-
elevation angles the MPCs have larger power. In both [57]
tion channel was carried out in [43]. It was observed here that
and [68], measurements were performed in the approximately
mobile relaying can achieve more than twice the throughput
similar environments and at same center frequency, only the
of static relaying for a given delay tolerant system.
bandwidths were different. However, the multipath fading dis-
Some results for BER as a function of signal-to-noise-
tribution in [57] was found to be Rayleigh/Ricean, whereas
ratio (SNR) for AG propagation channels are available in the
for [68], it was found to be Loo.
literature to compare the performance of different implemen-
2) Rural/Open Field: In [80], a rural environment similar
tation schemes. In [117], BER was measured against SNR
to an airport with large and small buildings and open grassy
for different modes of LDACS1, as a function of distance
fields was considered for channel measurements. Wideband
and for different phases of flight. A similar study was con-
OFDM channel sounding was performed in the L-band at
ducted in [118], where BER was measured against SNR for
a center frequency of 970 MHz. These channel measure-
an over sea AG propagation channel with distance measuring
ments were preformed at an aerial height up to 11 km. Due
equipment (DME) co-channel interference present. In [119],
to higher altitude, the tropospheric effects were also con-
BER versus SNR analysis was performed for different flight
sidered. In addition, interference effects such as from DME
route phases for different values of Ricean K-factor. BER ver-
were also taken into account. A link distance of 350 km was
sus SNR analysis was performed in [70] for comparing the
considered that is much larger than available in the litera-
effect of presence and absence of ICI for an IEEE 802.11a
ture. In [71], channel measurements were carried out near
OFDM system in the presence of additive white Gaussian
airports using a wideband CW signal centered at 5.75 GHz.
noise (AWGN).
Channel measurements were obtained for different flight sce-
narios of parking and taxiing, en-route, and take off and
F. Effects of UAV AG Measurement Environment landing. Received power and small scale fading statistics at
The different AG propagation channel measurement cam- different flight scenarios were analyzed, where during taxing
paigns can be broadly classified based on terrain, terrain cover and take-off higher RMS-DS, K-factor and Doppler shift were
and sounding signal characteristics. In this subsection, we observed. In [90], measurements were conducted in order to
will provide a brief overview and comparison of different explore the feasibility of fixed cellular network for teleme-
approaches. try and control of UAVs, particularly, in the perspective of
1) Urban/Suburban: UWB AG propagation channel mea- radio propagation at shorter distances in the sky as com-
surements were provided in [72], using PRN sounding pulses. pared to terrestrial. The center frequency of operation was
These measurements are unique as large bandwidth AG chan- 0.915 GHz. Comparisons of AG measurement results with the
nel measurements are not available in the literature. However, COST-231-WI were provided showing that received power
the link distance and UAV heights considered are small due from measurements was overestimated by COST-231-WI
to small transmit power allowed by the FCC. Buildings over model.
a flat terrain in the suburban area with an average height of In [87], channel measurements were carried out in an open
12 m, resulted in additional reflections and respective shape field using a modulated signal centered at 5.8 GHz, with
of the PDP as compared to the open area. In [68], a contin- two TXs operating at slightly different frequencies. A MIMO
uous wave (CW) narrowband sounding signal with a center configuration was used for channel measurements with direc-
frequency of 2 GHz was used in an urban scenario. There tional and omni-directional antennas. It was observed that
were uniformly built buildings in the urban area with an aver- multipath interference can be reduced using directional anten-
age height of 22 m. Second order channel fading statistics of nas. Additionally, using multiple antennas on-board UAVs can
Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2375

provide robustness against received power fluctuations due to including woods, where the shadowing from the woods was
varying antenna orientations on the aerial platform. However, found to be significantly different than obtained from the
no propagation model was provided. In [62], wideband PRN, buildings.
binary phase shift keying (BPSK) sounding signal was used at
a center frequency of 0.915 GHz. Analysis of spatial diversity G. Simulations for Channel Characterization
for MIMO signaling was carried out. Additionally, near field
Apart from measurement campaigns for AG propagation
scattering region analysis around the UAV and the GS were
channel modeling, some simulation based channel character-
carried out. It was observed that additional spatial diversity
izations are also available in the literature, where the real
could be obtained from objects near the GS. The MPCs were
time environmental scenarios are imitated using computer
reported to be sparse similar as observed in [108], though at
simulations. Simulations in urban/suburban areas were per-
a lower frequency.
formed in [69], [112], [114], [119]. The antenna considered
3) Mountains/Hilly, Over Sea, Forest: In [92], AG prop-
in these environments was omni-directional. Different carrier
agation channel measurements were performed in a near
frequencies 200 MHz, 700 MHz, 1 GHz, 2 GHz, 2.5 GHz,
mountainous area using a wideband chirp signal at a cen-
5 GHz, and 5.8 GHz were covered for AG channel charac-
ter frequency of 5.12 GHz. The overall terrain was flat with
terization in the urban/suburban environments, and different
some nearby mountains resulting in moderate MPCs. The
heights of UAVs, ranging from 200 m to 2000 m were con-
channel characteristics were largely dictated by the LOS com-
sidered. The PL (from simulated RSS) was estimated. Over
ponent. The effect of airframe shadowing were also observed.
sea based channel simulations were carried out in [116],
In [75], [76] channel measurements were performed in the
where a channel simulator imitating the sea environment was
L-band and C-band in an urban hilly area. Curved earth
developed. Carrier frequencies from 3 kHz - 3 GHz were used,
two ray (CE2R) model with weaker reflections and scatter-
with the TX and the RX placed 3.75 m above the sea surface.
ing from the hilly terrain and terrain cover were observed.
The main goal of the study was to quantify sea surface shad-
Additionally, multiple clusters of MPCs due to reflections
owing for the marine communication channel using UAVs.
from hills and buildings on hills were observed. In [81],
The channel characteristics of PL and RMS-DS were modeled
a wideband frequency modulated continuous wave (FMCW)
based on the sea surface height.
signal centered at 5.06 GHz was used for channel measure-
In [118], simulations were conducted in environmental sce-
ments in a hilly area. RSS fluctuations due to MPCs from
narios consisting of over sea, hilly, and mountainous terrain.
nearby buildings were reported. In the measurement cam-
Performance of AG communications using filter bank multi-
paigns of [75], [76], [81], the frequency bands were close in
carrier (FBMC) modulation systems and LDACS were com-
approximately similar environments. However, different results
pared. The results showed that FBMC has better performance
were observed, e.g., in [76], a better fitting with CE2R model
than LDACs, especially in the presence of interference from
was observed in the C-band from the received power however,
DME signals. In the presence of the AG channel, the FBMC
a free space attenuation fitting was observed in [81] for one
and LDACS performance is comparable. Other simulations of
of the flight tracks.
communication systems employed over AG propagation chan-
In [91], wideband channel measurements were performed
nels, for particular simulation scenarios, are also available in
using PRN sequence at a center frequency of 5.7 GHz over
the literature [47], [67], [120].
the sea. Multipath channel statistics were analyzed at dif-
In [121], the effect of the UAV height for optimal coverage
ferent heights of the aerial platform. It was observed that
radius was considered. It is observed that by adjusting UAV
the CIR can be represented using a 3 ray model. Elevation
altitude, outage probability can be minimized: a larger “foot-
and evaporation ducting effects were observed that resulted in
print” is produced with a higher UAV altitude, but of course
reduction of the attenuation. Over sea measurements were also
increased altitude can increase PL. An optimum UAV height
carried out in [54], [58], [74], for over sea and near harbor
is evaluated that maximizes the coverage area for a given
areas. A dominant two ray model was observed for the all the
SNR threshold. The Ricean K-factor was found to increase
cases. No ducting effect was observed over the sea for these
exponentially with elevation angle between UAV and GS,
measurements as opposed to [91].
given as K = c1 expc2 θ , where c1 and c2 are constants
In [86], an AG propagation scenario through a forest was
dependent on the environment and system parameters. The
imitated in an anechoic chamber using different heights of
relation between minimizing outage probability or maximiz-
the TX and the RX and with different species of trees.
ing coverage area for a given SNR threshold is solved only
Channel measurements were carried out in the X-band and
based on PL without considering the effect of scatterers in
Ku-bands. Different diffuse scattering regions from differ-
the environment. The consideration of geometry of scatter-
ent parts and respective species of the trees were observed,
ers in the analysis would of course make it more robust and
resulting in corresponding small scale fading statistics. Similar
realistic.
results were observed in [72], where foliage in the form
of a medium sized tree obstructed the direct LOS path
between the UAV and the GS resulting in peculiar small VI. UAV AG P ROPAGATION M ODELS
scale fading due to diffraction and scattering from different The UAV AG propagation measurements discussed in the
parts of the tree. In [89], channel measurements were carried previous section are useful for developing models for different
out using CW sounding signal centered at 2 GHz. Received environments. In the literature, UAV AG propagation channel
power was measured in different propagation environments models have been developed using deterministic or statistical
Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2376 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

Fig. 5. AG channel model characterization.

approaches, or their combination. These channel models can amplitude, delay, and duration) for each environment based on
be for narrow-band, wide-band, or even UWB communica- a large set of measurement data.
tions. Complete channel models include both large scale and Purely stochastic channel models can be obtained either
small scale effects. In this section, we categorize AG propaga- from geometric and numerical analysis without using measure-
tion channel models in the literature as shown in Fig. 5, and ments or they can be wholly empirical. Early cellular radio
review some of the important channel models. channel models, e.g., the COST 207 models, are examples
of the latter. These types of models are becoming less and
less common over time though, as incorporation of known
A. AG Propagation Channel Model Types physical information is shown to improve accuracy, and the
Time-variant channel models can be obtained via deter- greater model complexity is no longer prohibitive because of
ministic or stochastic methods or by their combination. The continuing advances in computer memory capacity and com-
deterministic methods often use ray tracing (or, geometry) putational power. Geometric based channel models for AG
to estimate the CIR in a given environment. These deter- propagation generally require three spatial dimensions to be
ministic channel models can have very high accuracy but accurate. The associated velocity vector for UAV motion in
require extensive data to characterize the real environment. space also requires three dimensions, although 2D approxi-
This includes the sizes, shapes, and locations of all obstacles mations can often be very accurate. In order to model the
in the environment, along with the electrical properties (per- scatterers around the GS, two elliptical planes intersecting a
mittivity, conductivity) of all materials. Hence such models are main ellipsoid were considered in [51], [52], [111], [119],
inherently site-specific. They also tend to require adjustment where the MPCs are defined by the ellipsoid and the two
of parameters when comparing with measurement data. Since elliptical planes. Scatterers are considered to be randomly
ray tracing based techniques employ high-frequency approxi- distributed on two spheres surrounding the TX and the RX
mations, they are not always accurate. They are not as accurate in [50]. In [49], [64], the distribution of scatterers around the
as full wave electromagnetic solutions, e.g., the method of GS is modeled using a three dimensional cylinder.
moments and finite difference time domain methods for solv- The geometry-based stochastic channel models (GBSCMs)
ing Maxwell equations [122], but ray tracing methods are of can be further classified into regular shaped GBSCMs (RS-
course far less complex than these full-wave solutions. Such GBSCMs) or irregular shaped GBSCMs (IS-GBSCMs). For
deterministic simulators are also very complex when they are RS-GBSCMs, the scatterers are assumed to be distributed on
used to model time varying channels. Ray tracing was used regular shapes, e.g., ellipsoids, cylinders, or spheres. These
in [46], [83], [112], [114], [123], [124] for different fully models often result in closed form solutions, but are of course
deterministic AG propagation scenarios. generally unrealistic. In contrast, the IS-GBSCM distributes
The models in [54], [76], [78] are a mix of deterministic the scatterers at random locations through some statistical dis-
and stochastic models (sometimes termed quasi-deterministic). tribution. The properties of the scatterers in both cases are
Specifically, the LOS and earth surface reflection are modeled generally defined beforehand. In some cases, authors assume
deterministically via geometry, and the remaining MPCs are a large number of scatterers a priori, and via the Central Limit
modeled stochastically, with parameter distributions (for MPC Theorem, obtain a Ricean amplitude distribution to obtain

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2377

TABLE V
R EVIEW OF E XISTING L ITERATURE ON L ARGE S CALE AG P ROPAGATION AND THE PATH L OSS PARAMETERS IN T HOSE PAPERS

estimates of the CIR based upon some geometry. Alternatively, scale fading models in the literature cover both the PL
signal interaction from randomly distributed scatterers can and shadowing.
be estimated directly, or with the help of ray tracing soft- 1) Free Space Path Loss Model: In the majority of the liter-
ware [112], [114], [123]. A non-geometric stochastic channel ature, the well-known terrestrial based log-distance PL model
model (NGSCM) based on a Markov process is provided with FSPL reference (“close-in,” CI) is used:
in [66]. The ground to air fading channel was described by
a Markov process that switches between the Ricean and Loo PLCI (d ) = PL0 + 10γ log10 (d /d0 ) + XFS , (2)
models, depending on the flight altitude.
where LCI (d ) is the model PL as a function of distance,
PL0 is the PL at reference distance d0 in free space given
B. Path Loss and Large Scale Fading Models by 10 log [( 4πd
λ ) ], γ is the PLE obtained using minimum
0 2

As noted, in mostly-LOS AG channels, large scale fading mean square error best fit, and XFS is a random variable to
only occurs when the LOS path between UAV and GS gets account for shadowing, or in the case of LOS channels, the
obstructed by an object that is large relative to the wave- variation about the linear fit. In free space the value of PLE
length. Some models for this attenuation mechanism exist is 2, but as seen from Table V, measured values of PLE vary
(e.g., terrain diffraction, tree shadowing), but not much mea- from approximately 1.5 to 4. One might conceptually divide
surement data for UAV channels obstructed by buildings has the path between the UAV and the GS into two components:
been reported. When the LOS path does not get obstructed, the free space component above the ground and the remain-
the only other truly large-scale effect is the two-ray vari- ing terrestrial influenced components. When the GS antenna
ation from the earth surface MPC. There are numerous height is well above surrounding obstacles, we expect the ter-
measurement campaigns in the literature for PL estimation restrial components to have smaller effect and the PLE is near
in different environments, as summarized in Table V. Large to that of free space.

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2378 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

2) Floating Intercept Path Loss Model: Another PL model in [108], [112], [114], [126], [133]. In [108], effect of three
used in the literature for large scale fading is floating inter- dimensional antenna radiation pattern is included in the
cept (FI) [128]. This model is similar to (2), but the FSPL at received power calculation. At higher elevation angles with
reference distance is eliminated and the model is dependent omni-directional antennas at both the TX and the RX sides,
on two parameters represented as α and β [22], where α is the low gain antenna radiation regions in the elevation plane
the slope and β represents the intercept given as were exposed at the TX and the RX sides resulting in reduction
of the received power. This loss, due to small antenna radi-
PLFI (d ) = α10 log10 (d ) + β + XFI , (3)
ation gain in the elevation plane at higher UAV heights was
where XFI is a random variable representing the variation of reduced as the UAV moved away from the GS. Similar obser-
the PL. vations with antenna radiation pattern were made in [126].
3) Dual Slope Path Loss Model: The two PL models dis- In [112], FSPL was provided, taking into account the height
cussed above are based on single slope. These models hold in of the aerial platform, therefore the distance d obtained from
areas where the characteristics of the channel do not change the geometry of the setup in the FSPL expression was modi-
Δh . In [114], PL considering the elevation angle
fied as d = sin
drastically. The two PL models in (2) and (3) have different θ
interpretations. The model in (2) uses a reference distance of was modeled for θ > 10. The PL was modeled for both LOS
typically 1 m from the transmit antenna and find the slope and NLOS paths using intrinsic coefficients and claimed to
using minimum mean square error, whereas no reference dis- be independent of the antenna heights at the TX and the RX.
tance is used for calculating the PL in (3) and linear least Similar three dimensional PL model was provided in [133].
square error regression is used to obtain the fitting for the 6) Two-Ray PL Model: The two ray PL model described
data. Additionally, β in (3) is not PLE, as it is not obtained earlier in Section III-C is provided in [37], [54], [73]–[75],
through a free space reference distance near the antenna [129]. [77]–[79], [91]. In case of two ray PL modeling, the variation
At short distances, the model in (3) provides an underesti- of the PL with distance has distinctive peaks due to destructive
mated PL, whereas it provides an overestimated PL at larger summation of the dominant and surface-reflected component.
distances [130], therefore, use of CI model in (2) is recom- In the majority of PL models, PL variation is approximated
mended for all the scenarios. However, in some settings with as a log-normal random variable. This variation can be either
NLOS paths and complex geometries resulting in higher order due to shadowing from the UAV body (see next subsection) or
reflections and diffractions, these single-slope models can have from MPCs attributable to terrestrial scatterers such as build-
large regression errors. In such cases, a dual slope (DS) PL ings [37], [53], [54], [57], [71], [72], [75], [76], [78], [85],
model is sometimes used [127], [131]. This model is similar [91], [111], [113].
to the FI model, but has two different slopes for different link 7) Log-Distance FI Model: In [22], log-distance FI mod-
distance ranges, and can be represented as els for the PLE and shadowing for the AG radio channel
between airborne UAVs and cellular networks are presented
PLDS (d)
 for 800 MHz and UAV heights from 1.5 m to 120 m above
αd1 10 log10 (d) + βd1 + XDS , d ≤ d1 ground. In [127], the low altitude AG UAV wireless channel
=
αd1 10 log10 (d1 ) + βd1 + αd2 10 log10 (d/d1 ) + XDS , d > d1 has been investigated for a scenario where a UAV was fly-
(4) ing above an ensemble of containers at 5.76 GHz. Narrow-
and wideband measurements have been carried out. The paper
where αd1 , αd2 , are the slopes of the fits for at two link
presents a modified PL model and PDPs. Most interesting
distance ranges separated by threshold d1 , βd1 is the inter-
is that in this particular environment, delay dispersion actu-
cept, and XDS is a random variable representing the variation
ally increases with altitude as the UAV rises above metallic
in the fit.
structures.
4) Log-Distance Path Loss Model: PL estimates using log-
distance models (2) are given in [53], [60], [61], [71], [72], 8) LOS/NLOS Mixture Path Loss Model: Another com-
[76]–[78], [83]–[85], [91], [111], [116], [126], [132]. There mon model used in the literature [48], [134]–[139] averages
are other PL models that consider shadowing for NLOS the PL over the probabilities of LOS and NLOS PL as
paths, and additional losses incurred from other obstacles [82], follows [48], [140]:
[113], [125]. In [113], shadowing loss was considered in the
PLavg = P (LOS) × PLLOS + [1 − P (LOS)] × PLNLOS ,
modeling and evaluated as a function of the elevation angle
for NLOS paths. The shadowing loss was calculated based (5)
on the uniform theory of diffraction. The distribution of the
where PLLOS and PLNLOS are the PL in LOS and NLOS
shadowing was found to be normal. Strong shadowing was
conditions, respectively, P (LOS) denotes the probability of
observed in [125], in an urban area, mostly due to knife edge
having a LOS link between the UAV and the ground node,
diffractions from the surrounding buildings. On the other hand,
given by [48], [140]:
in [82], additional losses were considered at the GS and the
UAV for the overall PL modeling. ⎡ ⎛  ⎞⎤
(n+1/2)(hU −hG ) 2
⎢ ⎜ hU −
m
5) Modified FSPL Model: Due to the potential three m+1 ⎟⎥
P (LOS) = ⎣1 − exp⎝− ⎠⎦,
dimensional motion of UAVs, modified FSPL models 2Ω2
n=0
accounting for UAV altitude can also be developed; sev-
eral that are a function of elevation angle are considered (6)

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2379

√
where we have m = floor(r ςξ − 1), r is the horizontal dis- For complex geometrical environments with large number of
tance between the UAV and the ground node, hU and hG are scatterers in a LOS path between the GS and the UAV, a DS
as shown in Fig. 1 of this survey, ς is the ratio of built-up land PL (4) model may be best. This is because the received power
area to the total land area, ξ is the mean number of buildings can be differentiated into two different categories [108]. One
per unit area (in km2 ), and Ω characterizes the height (denoted is due to persistent components arising from LOS and ground
by H) distribution of buildings, which is based on a Rayleigh reflection paths, whereas, others from the surrounding scatter-
distribution (P (H ) = (H /Ω)2 exp(−H /2Ω2 )). In [48], for a ers. Therefore, DS may be a better option for these scenarios.
specific value of θ in [48, Fig. 2], a sigmoid function is also The FI model in (3) may be preferred in certain specialized
fitted to (6) for different environments (urban, suburban, dense environments, e.g., [127], due to their ease in applicability
urban, and highrise urban) to enable analytical tractability of and PL model predictions only for a given area. In Table V,
UAV height optimization. Since (5) averages the PL over large the model types denoted log-distance refer to the general log-
number of potential LOS/NLOS link possibilities, it should be distance equation for PL with different reference distances and
used carefully if used with system-level analysis while calcu- additional parameters.
lating end metrics such as throughput and outage. Similarly,
PL variability should be added to the model of (5). C. Airframe Shadowing
Overall, a comparison of different PL models in Table V,
Airframe shadowing occurs when the body of the aircraft
shows that two ray PL model is a better choice for open field,
itself obstructs the LOS to the GS. This impairment is some-
rural areas and over sea, whereas higher ray model, e.g., 3
what unique to AG communications, and not much exists
ray or 4 ray model can be employed for environments with
in the literature on this effect. One reason for this is that
larger number of scatterers whose heights are comparable to
such shadowing can be largely (but not always completely)
the height of the UAV. This has been validated by channel
alleviated by using multiple spatially separated antennas: air-
measurements carried out in [54], [76], [78], where CE2R
frame shadowing on one antenna can be made unlikely to
model was found to provide better fitting for over water, har-
occur at the same time as shadowing on the other(s). In addi-
bor and mountainous settings, however, the log distance PL
tion to frequency and antenna placement, shadowing results
model was found to have better fit for the environments with
also depend on the exact shape, size, and material of the
higher scatterers, e.g., urban and suburban. Similarly, two ray
aircraft. For small rotorcraft, depending on frequency and
PL fitting with L-band was better as compared to C-band.
antenna placement, airframe shadowing could be minimal.
However, in [91], the measured data obtained over sea was
Example measurement results, as well as models for air-
compared with FSPL and two ray model, and it was observed
frame shadowing, for a fixed wing medium sized aircraft, were
that both models overestimated the PL. This was mainly due
provided in [141].
to ducting effect over sea surface causing reduction in the PL.
For these results, at frequencies of 970 and 5060 MHz, wing
In [81], the RSS obtained from wideband measurements at
shadowing attenuations were generally proportional to aircraft
5.06 GHz were found to closely fit the FSPL for one location
roll angle, with maximum shadowing depths exceeding 35 dB
whereas providing upper bound for another location. However,
at both frequencies. Shadowing durations depend upon flight
in [76], with similar propagation environment at C-band, it was
maneuvers, but for long, slow banking turns, can exceed tens
observed that CE2R model provided a better fit.
of seconds.
Selection of a suitable PL model for a given AG propagation
An illustration of airframe shadowing is shown in Fig. 6
scenario is pivotal. In most of the literature, the PL model for
where received power is plotted against time for a wide-
of (2) is used due to its simplicity and provision of a standard
band (50 MHz) signal in C-band before, during, and after
platform based on reference distance FSPL for comparison of
the medium-sized aircraft made a banking turn. The received
measurements in different environments. A reference distance
power on two aircraft antennas (denoted C1, C2), bottom
of 1 m is often taken as a standard for short-range systems, but
mounted and separated by approximately 1.2 m, is shown.
larger values are also used. However, in some scenarios, where
Attenuations due to airframe shadowing, along with the polar-
the reference FSPL is not available, the FI model (3) may
ization mismatch that occurs during the aircraft maneuver,
be used. Yet due to lack of any standard physical reference,
exceed approximately 30 dB in this case.
the FI slope cannot be deemed PLE and will be dependent
on the environment. Additionally, the variability of the PL
is generally a zero mean Gaussian random variable that has D. Small Scale Fading Models
approximately similar values for both the CI and FI model Small-scale fading models apply to narrow-band chan-
types. nels or to individual MPCs, or taps in tapped delay line
A general recommendation for selection of AG PL model wide-band models, with bandwidth up to some maximum
for a given measurement scenario from Table V is as follows: value (i.e., small scale fading may not pertain to MPCs
for an open flat or hilly area with light suburban, rural or no in a UWB channel). The depth of small scale amplitude
terrain cover, and for over water, the two ray PL model or free fades on a given signal also generally varies inversely
space reference log-distance model (2) may be preferred. This with signal bandwidth [142]. Stochastic fading models are
is due to small number of MPCs reaching the aerial plat- obtained through analysis, empirical data, or through geo-
form and high probability of presence of a dominant LOS metric analysis and simulations [49]–[51], [64], [111], [119].
and ground reflected component only in these environments. As noted in Section VI-A, the GBSCMs can be subdivided

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2380 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

TABLE VI
R EVIEW OF THE E XISTING L ITERATURE ON S MALL S CALE AG P ROPAGATION C HANNEL FADING C HARACTERISTICS

distribution typically provides a better fit [57], [65], [68],

[69], [71], [120], [143], and of course, other distributions
such as the Nakagami-m and Weibull distributions might also
be employed. Small scale fading rates depend upon velocity,
and these rates are proportional to the Doppler spreads of the
MPCs [70], [71], [74], [81].

E. Intermittent MPCs
Another AG characteristic that may be of interest in high-
fidelity and long-term channel models is the intermittent nature
of MPCs. From geometry, it is easy to deduce that for a given
vehicle trajectory in some environment, individual MPCs will
persist only for some finite span of time [54]. This has been
noted in V2V channels as well, but with UAVs and their
potentially larger velocities, the intermittent MPC (IMPC)
dynamics can be greater. These IMPCs arise (are “born”)
Fig. 6. Received power vs. time for illustration of shadowing before, during and disappear (“die”) naturally in GBSCMs. They may also
and after the banking turn of medium sized aircraft at C-band. be modeled using discrete time Markov chains. The IMPCs
can significantly change the CIR for some short time span,
hence yielding wide variation in RMS-DS. (Another mani-
into RS-GBSCM and IS-GBSCM. In [119], a time-variant festation of so-called “non-stationarity.”) Example models for
IS-GBSCM was provided with a Ricean distribution for the IMPCs–their probability of occurrence, duration, delay,
small scale fading. Time-variant RS-GBSCM were provided and amplitude–appear in [54], [58], [79], [108]. In these stud-
in [50], [64], and these also illustrated Ricean small scale ies it was found that IMPCs follow a random process which
fading. is highly dependent on the geometry, and distribution of the
A NGSCM was provided in [66], where GA fading was scatterers around a given UAV trajectory.
described using Ricean and Loo models. The Loo model was In Fig. 7, from [37] the fading of MPCs as a function of time
derived based on the assumption that the amplitude attenuation and delay are shown. The amplitude of MPCs generally decay
of the LOS component due to foliage in a land mobile satel- with excess delay at a given time instant. Additionally, there
lite link follows a log-normal distribution, and that the fading is a continuous birth and death process of MPCs at different
due to MPCs follows a Rayleigh distribution. The switching instants of time. This can be represented using CIR as [37]:
between Ricean and Loo models was controlled by a Markov
M (t)−1
process dependent on flight height. In [51], a GBSCM for 
MPCs was provided in the form of shape factors describing h(t, τ ) = pi (t)ai (t) exp(j φi (t))δ(τ − τi (t)), (7)
angular spread, angular compression, and direction of maxi- i=0
mum fading using the probability density function (PDF) of where h(t, τ ) is the time variant CIR, M(t) is the
angle of arrival. total number of MPCs at time instant t, pi (t) repre-
Table VI provides measured small scale AG fading charac- sents the multipath persistence process coefficient and can
teristics reported in the literature for various environments. take binary values [0, 1]. The amplitude, phase and delay
As previously noted, the most common small scale fading of i th MPC at time instant t are represented as ai (t),
distribution for AG propagation is the Ricean. As in ter- φi (t) and τi (t), respectively. The phase term is given as
restrial channels, for the NLOS case, the Rayleigh fading φi (t) = 2πfdi (t)(t − τi (t)) − fc (t)τi (t), where fdi (t) =

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2381

Fig. 7. Fading and birth and death process of intermittent MPCs from [37].

v (t)fc (t) cos(Θi (t)/c is the Doppler frequency of the i th

MPC, Θi (t) is the aggregate phase angle in the i th delay bin,
c is the speed of light and fc represents the carrier frequency.
The channel transfer function H(f, t) from (7) is then given as
follows:
M (t)−1
  
H (f , t) = pi (t)ai (t) exp j 2πfdi (t)(t − τi (t))
i=0
× exp(−j 2πfc τi (t)) exp(−j 2πf τi (t)), (8)
The effect of the Doppler spread is typically negligible
compared to carrier frequency at lower velocities. Therefore
the carrier frequency term will dominate the variation of the
transfer function.
Fig. 8(a) shows a sequence of PDPs versus link distance for
a near-urban AG link near Cleveland, OH, USA. Flight param-
eters can be found in [78]. In this figure, the IMPCs are clearly Fig. 8. (a) Sequence of PDPs versus link distance for a near-urban AG
visible, here caused by reflections from obstacles near the Lake link near Cleveland, OH, USA. (b) RMS-DS vs. link distance for a hilly
environment in Palmdale, California.
Erie shoreline. In Fig. 8(b) RMS-DS vs. link distance for a
hilly environment in Palmdale, California is shown. The inter-
mittent nature of the MPCs produces “spikes” and “bumps”
and suburban areas in [72]. A narrow sub-band frequency anal-
in the RMS-DS values, illustrating the potential rapid time
ysis was provided, where higher mean attenuation and larger
variation of AG channels.
variance of received power was observed at higher frequency
sub-bands. Narrowband frequency measurement campaign for
F. Effect of Frequency Bands on Channel Models three different frequencies of 2 GHz, 3.5 GHz, and 5.5 GHz
The selection of frequency bands for CNPC and payload was carried out in an urban area in [113]. Higher attenuation
communications using AG links for UAVs are discussed in was observed at respective higher frequency bands, whereas
Section III-B. In the literature, most of the AG propagation the standard deviation did not show much difference for three
channel measurements were carried out using wideband spec- different frequency bands. Wideband channel measurements at
trum, whereas some narrowband channel measurements are L-band and C-band were provided in [54], [76], [78] for differ-
also available. The current and future communications are ent propagation environments. Different attenuation, and small
expected to deploy OFDM technology. Therefore, narrowband scale effects were observed at the two different frequency
employed by OFDM needs to be studied extensively in differ- bands in a given propagation environment, e.g., in [76], chan-
ent AG propagation environments. The propagation channel nel measurements were performed in a hilly/mountainous area.
characteristics using narrowband are expected to be affected Similarly, higher K-factor and PLE were observed for C-band
more as compared to wideband in case of UAV in motion as compared to L-band, whereas, larger shadowing variable
especially in a scatterer rich environment, e.g., dense subur- was reported for L-band as compared to C-band.
ban. This is due to fading, affecting the narrowband more as The mmWave AG propagation channel characteristics using
compared to the wideband when the UAV is in motion. UAVs at 28 GHz and 60 GHz in different environments
UWB AG propagation channel measurements in the were provided in [93] using ray tracing simulations. The RSS
frequency band 3.1 GHz - 5.3 GHz were performed in open generally followed a two ray model at both frequencies in

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2382 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

different propagation environments, however, higher rate of reflections, whereas in case of forests, the reflection and shad-
maxima/minima at 60 GHz was observed as compared to owing from the trees is a major contributor to the propagation
28 GHz. In [82], two ray model was used for PL represen- channel characteristics. In rural and urban cases, the reflec-
tation at 2 GHz in an urban environment, whereas, channel tions from the walls and surfaces of building structures play
measurements at 2 GHz obtained in [57] in a similar envi- an important role.
ronment were found to be better fitted with a log distance PL Time-variant GBSCMs for MIMO systems provided in [50],
model. [64], [119], [146] were explored through simulations. Where,
different propagation geometries and scatterer distributions
were used in order to analyze the capacity for MIMO AG
G. MIMO AG Propagation Channel Models channels. A common observation was that higher capacity is
achievable with MIMO AG systems. A simulation based AG
The use of MIMO systems for AG UAV communications
MIMO channel propagation model was provided in [111] for
has been gaining popularity. The rationale, increased through-
a hilly area. The results indicate increased throughput from
put and reliability, is the same one driving mmWave and future
spatial multiplexing and higher SNR from the MIMO system
5G research. In [144], it was shown that it is possible to attain
in comparison to SISO, as expected. A stochastic model for
higher spatial multiplexing gains in LOS channels by properly
a mobile to mobile AG MIMO propagation channel was
selecting the antenna separation and orientation as a function
presented in [50]. These results show that there was consid-
of carrier wavelength and link distance. This careful alignment
erable capacity increase and reduction in outage probabilities
is not always practical or possible with UAVs, especially when
using MIMO systems if perfect instantaneous CSI is avail-
mobile.
able. In [146], geometry-based simulations were conducted
The advantages of spatial diversity and multiplexing gains
for a massive MIMO implementation for a UAV AG propa-
in MIMO are often only moderate due to limited scattering
gation channel. The simulation results illustrate the expected
available near UAVs or GSs. In [145], it was demonstrated
result of a significant capacity increase when a large number
that due to limited spatial diversity in the AG channel, only
of antennas is used at the GS.
moderate capacity gains are possible. In order to obtain bet-
ter spatial multiplexing gains, larger antenna separations are
required, and this requires large antenna arrays that are not fea- H. Analysis and Comparison of Different AG Propagation
sible on-board small UAVs. Use of higher carrier frequencies Channel Models
makes it possible to use electrically-large antenna arrays, but In this subsection, we will briefly analyze and compare
higher frequencies yield higher PL (this can be mitigated different AG propagation channel models for UAVs in the
somewhat by beamforming, at the expense of the complexity literature.
required for beam steering). Moreover, accurate channel state 1) Large Scale Fading Models: Large scale fading models
information (CSI) is important for MIMO systems for higher for AG propagation available in the literature, generally can be
performance, but in a rapidly varying AG propagation channel, fitted with a modified FSPL model. In [72], PL was measured
it can be difficult to provide accurate CSI and hence MIMO for open and suburban scenarios in the presence and absence
gains can be limited. The use of MIMO on airborne plat- of foliage. The PL was found to be highest for the foliage due
forms also incurs additional cost, computational complexity, to obstruction. In addition, the PL was found to be dependent
and power consumption. on the height of the GS, apart from the height of the UAV.
There is a limited number of studies available in the lit- The PLE reported was above 2.5 for all the propagation scenar-
erature for MIMO AG propagation channel measurements. ios involving open and suburban areas. Additionally, the PLE
In [62], a detailed measurement analysis of the AG MIMO for suburban scenario was slightly higher than the open area
propagation channel was provided. It was observed that a con- scenario. In [86], PL due to diffraction and scattering from
siderable spatial de-correlation of the received signal at the different species and parts of the tree are provided. It was
GS is achieved due to the interaction of non-planar wave- observed that loss from the trunk of the tree is due to diffuse
fronts. These wavefronts are generated due to near field effects scattering, whereas, it was mostly due to diffraction on the
from the measurement vehicle, on which the GS antennas were edges of the crown of the tree. In [53], [54], [58], [75], [76],
mounted. Spatial diversity from antennas located on the UAV [78], PL for channel measurements in the urban, suburban,
was also observed, interestingly at higher elevation angles. hilly, mountainous, and over sea scenarios were provided. A
The authors suggest that having scatterers near the GS can comparison of PL obtained from free space, analytical CE2R
yield larger spatial diversity. The received signal in [87] was model and measurements was provided. The PLE for C-band
analyzed for multiple-input-single-output (MISO) and MIMO and L-band for different measurement scenarios was found to
systems, and it was observed that the use of MIMO systems be approximately equal or less than the FSPL. For over sea
enables a more robust channel for changes in antenna orienta- scenario, ducting was not observed whereas in [91], ducting
tions arising from UAV maneuvering. In [89], MIMO system was observed that resulted in reduction of the PL as compared
performance was tested in different scenarios of the outdoor to the FSPL. A study for PL during different flight scenarios of
environment, including urban, rural, open field, and forest. The take off, en-route, landing, taxing and parking were provided
effect of terrain cover on the received power was analyzed for in [71]. Higher PLE was observed for take off and en-route
these different scenarios with the result that the propagation as compared to other flight scenarios. In [125], excess PL was
channel in the open field is mostly influenced by the ground provided for an urban scenario. It was found that excess loss

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2383

was dependent on different diffractions from the edges of the on specific environment, center frequency of operation, band-
surrounding buildings. width, antenna type and configuration, among other factors.
Antenna orientation effects on the RSS were provided Here we provide some remarks for context.
in [61], where the PLE was found to be close to FSPL in dif- Among other applications, UAVs are likely to be used in
ferent antenna orientations in urban and open field. The PLE future cellular communication networks either as a base sta-
was found to be higher in the urban area as compared to open tion [148], [149] or a user equipment node [83]. When a UAV
field. A similar study taking into account the antenna radia- is used as a base station, the corresponding propagation chan-
tion effects is available in [108], [126], where it was found nel may exhibit cellular base station characteristics if they are
that minimum received power was observed when the UAV hovering (with no mobility) at similar heights/environments
was on the top of the base station due to minimum antenna as in terrestrial base stations, and assuming they communicate
gain in the elevation plane at that point. with similar user equipment as in terrestrial cellular networks.
2) Small Scale Fading Models: In the literature, there are In such cases, terrestrial cellular channel models might be
limited small scale models for AG propagation using UAVs. applicable for AG channels [147].
However, there is a comprehensive wideband measurement On the other hand, many times the UAVs may be mobile,
and modeling campaign at the L-band and C-band available in might operate at significantly higher altitudes than terrestrial
the literature [54], [76], [78]. This campaign was conducted in base stations, and the operational environment may also be
different propagation environments over sea, hilly, mountain- very different, all of which should be taken into account for the
ous, suburban and urban areas. This campaign provides the channel model. A major difference in the AG channel would
bulk of the small scale modeling information for AG propa- be that for typical situations, the vast majority of scatterers
gation channels in the L-band and C-band. A TDL model was are located around the GS and not the UAV. This of course
used to represent the channel response in all the scenarios. has a direct influence on the MPC characteristics, mainly the
A two ray model in addition to varying number of intermit- Doppler frequencies of the MPCs. In particular, if the UAV
tent MPCs in different environments. K-factor obtained from is moving and while the GS is stationary, all MPCs usually
measurements provided a measure of the strength of MPCs have a very similar Doppler frequency, in strong contrast to
when compared with the LOS component. In all the scenar- the terrestrial channel.
ios, the K-factor for C-band was found to be higher than the When the UAV is used as a user equipment node (commu-
L-band. The highest K-factor was observed for the C-band for nicating with a terrestrial base station), the terrestrial cellular
the over sea scenario, followed by hilly/mountainous, and sub- channel models cannot be directly applied [107]. This is true
urban/urban scenarios. The K-factor in the L-band was found because when aloft, hovering or moving, the UAV experiences
to vary less in different propagation environments as compared large and small scale fading characteristics different from
to the C-band. ground based users because of the different spatial distribu-
A small scale propagation channel model for UWB was tion of nearby obstacles. Channel models for a user equipment
provided in [72], for suburban and open field scenarios. The node as a UAV in a cellular communication network are
small scale fading amplitudes were found to be Nakagami provided in the recent version of the 3GPP documentation
distributed. A modified Saleh-Valenzuela model was used covering UAV communications [150]. Details are discussed in
to model the CIR. Different number of MPC clusters were Section VI-K.
observed in the open field as compared to the suburban envi- One can also compare UAV channel models with satellite
ronment. It was observed that the RMS-DS changed with channel models [151], [152]. Overall, both satellite and UAV
the height of the UAV in the suburban scenario, whereas it communication links have higher probability of LOS than ter-
was approximately flat for the open field scenario. Similarly, restrial cellular links [153]. One of the main differences is
higher TOA of MPCs was observed for suburban scenario as that for most satellite applications, we are not interested in
compared to open field for different UAV heights. In [68], small elevation angles (e.g., smaller than 5◦ ). This is due to
second order channel statistics of average fade duration and the negative effects of the tropo-/ionosphere on the propaga-
level crossing rate were provided for the narrowband AG sig- tion. However, in the AG channel we often end up with very
nal propagation. The amplitude of the MPCs was found to be low elevation angles and long link distances: this implies that
log-normally distributed. A time series generator was used for the effects of the troposphere can be significantly larger than
emulation of RSS based on the analytical model and compared for the typically high elevation angles used in satellite appli-
with the measurement results. cations. Many satellite communication links are directional
and point to point, whereas UAV communications need not
be directional.
I. Comparison of Traditional Channel Models With UAV AG
Propagation Channel Models J. Ray Tracing Simulations
The UAV AG propagation channel naturally has similari- In the literature, in addition to measurements, channel char-
ties with outdoor terrestrial propagation channels, at least for acterization for AG propagation is also carried out using
elevated base stations. We do not provide a comprehensive simulations. These simulators are either based on customized
comparison between UAV AG channels and terrestrial chan- channel environments on a given software platform or can
nels here, as there are numerous terrestrial channel models be realized using ray tracing simulations. There are PL mod-
for a variety of terrestrial environments [147] that differ based els available for these simulated environments [48], [112],

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2384 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

Fig. 9. Ray tracing simulation scenario for over sea scenario, where the
UAV flies over a straight line.

(a)

(a)

(b)

Fig. 10. Ray tracing PL results for over sea water settings, (a) C-
band (5.03 GHz - 5.091 GHz), (b) L-band (0.9 GHz - 1.2 GHz).

(b)

Fig. 12. Path loss versus distance with/without scatterers, and without the
sea surface: (a) 100 m to 2 km range, and (b) 1300 m to 1350 m range.

there are no specific experimental studies available in the liter-

ature that experimentally validate the channel models proposed
using geometrical analysis and simulations in [48], [112],
[114], [116], [133], [143].
Ray tracing was used for mmWave channel characterization
for 28 GHz and 60 GHz frequency bands for UAV AG prop-
agation in [93]. Different environments were realized, namely
urban, suburban, rural and over sea. It was observed that the
RSS followed the two ray model and is further affected by the
presence of scatterers in the surroundings. The RMS-DS was
Fig. 11. Zoomed in results of PL for over sea water simulations of Fig. 10
also affected by the presence of scatterers in the surrounding
for C-band at link distances of 13 km - 14 km. environment and the UAV height in the given environment.
If the height of the scatterers were comparable with the UAV
height, higher RMS-DS was observed due to multiple reflec-
[114], [116], [133], [143]. Urban environmental scenarios for tions from the randomly distributed scatterers. In contrast, if
LOS and NLOS paths were considered in [48], [112], [114] the height of the scatterers is small, we have smaller RMS-DS
where log-distance and modified FSPL models were sug- at higher UAV heights due to fewer significant MPCs reaching
gested. In [116] a log-distance path model was provided for the UAV. This phenomena is verified at 28 GHz and 60 GHz,
LOS and NLOS paths for over sea settings in a simulated where at 60 GHz, we have smaller RMS-DS than at 28 GHz
environment. However, to the best knowledge of the authors, due to higher attenuation of MPCs.

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2385

the geometry of the scenario and will cause the PL to vary

along the trajectory of the UAV. A similar effect at the larger
link distance range of 13 km - 13.5 km in Fig. 12(a) can be
observed in Fig. 12(b).
Fig. 13 shows measured and model PL results from [54]
for over-sea settings. There is a good match between the
ray tracing simulation results and analytical results for this
over sea scenario in Fig. 10, but when comparing measure-
ment data with simulation data, we observe more fluctuations
in measurements due to several factors: measurement equip-
ment variation, ambient noise, and in particular scattering
from the rough sea surface, which is not as easily mod-
eled with the basic ray tracing. Also, plotted along with
the measurement data in Fig. 13 are analytical results for
free space and curved- and flat-earth two ray models. The
curved- and flat-earth two-ray models are obtained using
the specific geometry and conditions of the measurement
environment.

K. Channel Models for 3GPP Cellular-Connected UAVs

The latest version of 3GPP model [150] provides channel
modeling details for UAV AG communications, considering
the user equipment on the UAV communicating with the fixed
base station. These details include LOS probability, PL mod-
els, and fast fading models. The height of the user equipment
on-board UAV in the air can be smaller or greater than the
base station height.
LOS probability for rural macro (RMa), urban
macro (UMa), and urban micro (UMi) are provided for
different aerial user heights. The LOS probability is smaller
Fig. 13. Measurement results for PL over sea scenario from [54]: for all the scenarios when the UAV height is small due to
(a) C-band (5.03 GHz - 5.091 GHz), (b) L-band (0.9 GHz - 1.2 GHz). obstruction from scatterers on the ground. As the height of
the aerial user increases, the LOS probability also increases,
e.g., for RMa scenario, we have 100% LOS probability after
50 m of aerial user height, whereas it is 100 m for UMa
Ray tracing simulations using Wireless InSite software were scenario. The LOS probability expression is dependent on the
carried out to estimate PL for an over-sea scenario as shown in height of the UAV. For example, for a RMa scenario, with a
Fig. 9. The channel measurement parameters were set accord- UAV height in the range of 10 m - 40 m, and for a UMa and
ing to [54], and the simulated PL results were compared with UMi scenario with a UAV height range of 22.5 m - 100 m,
the measured values. Fig. 10 shows the simulated PL results. the LOS probability expression is given as follows [150]:
In this simulated environment, we have buildings as scatterers 
near the TX. Due to reflections and diffractions from these 1,    for (dh ≤ d1 )
PLOS = d1 dh d1 (9)
d + exp − p1 1 − d , for (dh > d1 )
scatterers we observe additional fluctuations on top of the
h h
two ray propagation model. The deviations are due to MPCs
reflected and diffracted from the different-shaped scatterers where dh is the horizontal distance of the UAV from the base
at different angles. These weak MPCs reach the UAV RX at station. The value of variables d1 and p1 are dependent on the
different link distances resulting in variations from the two UAV height and the scenario considered.
ray model as shown in Fig. 11 at a link distance between The PL models are modified FSPL models taking into
13 km - 14 km. account the height of the user with respective constants. For
Similarly in Fig. 12(a), the effect of MPCs from scatterers the three scenario of RMa, UMa, and UMi, the lower bound
around the TX for link distances 100 m - 2 km are shown. height of user is 1.5 m and goes up to 300 m. The PL models
It can be observed that without the scatterers and seawa- using UAVs are divided into two categories based on the height
ter (with ground only), we have a perfect two ray PL model. of the aerial user. For a RMa scenario, with height of aerial
Yet in the presence of the scatterers around the TX, super- user in the range 1.5 m - 10 m, PL model from [154, Sec. 7.4]
imposed upon this effect is variation from additional MPCs were recommended for both LOS and NLOS paths, whereas
from the scatterers; this yields what can be modeled as a ran- for aerial user heights of above 10 m, additional PL mod-
dom PL component on top of the two ray model, or in effect els are provided for LOS and NLOS paths [150]. Similarly,
a small scale fading. This effect is of course dependent on for the UMa and UMi scenarios, instead of 10 m, 22.5 m

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2386 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

upper bound height is used for separation of two PL model rich multipath, and it is in these cases where communica-
categories. The distribution of the shadow fading in all the tion must work very reliably. Moreover, we think that the
scenarios is found to be log-normal. The standard deviation propagation conditions for flights that bring UAVs intention-
of the shadow fading for LOS scenarios for aerial user height ally close to building facades, power lines, containers, and
of above 10 m for RMa and 22.5 m for UMa and UMi is a other objects (e.g., for inspection) should also be investi-
function of height of the aerial user. On the other hand, for gated as such propagation may exhibit special or atypical
aerial user height of less than 10 m for RMa, and 22.5 m for features.
UMa and UMi, and NLOS scenarios it is assigned a constant In addition of using UAVs as a node in different propaga-
value. tion environments for data source and relaying, they can be
Fast fading models are also provided in [150] for aerial employed for data ferrying operations, similar to aerial com-
users. The models are provided for aerial user height of 10 m - munication postal service. A possible scenario can be sending
300 m for RMa and 22.5 m - 300 m for UMa and UMi. Three data to a remote location not accessible through a direct radio
different alternatives are provided for evaluation of fast fading link. This task would be accomplished by first loading data
models. In each alternative specific parameters are provided onto a UAV and then taking it to the desired location for
or parameters from [154] are used for fast fading modeling transmission.
for RMa, UMa, and UMi scenarios. 1) Cognitive Communications Using UAVs: The CNPC
links for UAVs are critical for safe operation of the UAVs.
However, only specific frequency bands are suited for these
VII. F UTURE R ESEARCH A REAS FOR AG UAV
links. With the introduction of new communication devices
C HANNEL M EASUREMENTS AND M ODELS
into the already congested unlicensed bands, it is imperative
In this section we discuss limitations of currently avail- to find alternate frequency bands that are robust and reliable
able AG channel measurements and models and their possible for CNPC links. Cognitive radio technology (CRT) can offer a
enhancements. We also identify some representative consider- possible solution to this problem [155]–[157]. In these studies,
ations for future research. Our aim is to incite development advantages of using CRT technology with UAVs are discussed,
of more comprehensive, realistic, and accurate propagation e.g., dynamic spectrum access, reduced energy consumption
channel models for future UAV communication applications. and smaller delays as compared to current UAV communica-
tion approaches. Similarly, future applications of using CRT
A. Future Small UAV Scenarios with UAVs are discussed in the field of traffic and border
In future scenarios small UAVs will fly in cities, across surveillance, disaster relief operations and military applica-
suburban areas, and over rural terrain. There are two con- tions in the battlefield. Some of the future challenges include
ceptually very different communication approaches for small design of antenna and specific medium access, routing and
UAVs: the first approach is based on centralized communi- transport layer protocols that are optimally efficient during
cations, i.e., UAVs communicate with base stations similar mobility. Overall, CRT can help in accessing licensed and unli-
to the concept of 3G and 4G cellular mobile radio. These censed frequency bands for robust operations of telemetry in
base stations would preferably be located on elevated posi- different propagation environments. In addition, CRT can be
tions such as towers or roof tops and have antennas whose used for different payload AG communications. However, a
radiation patterns are optimized for serving these UAVs. The major hurdle in implementing the CRT technology for CNPC
second approach foresees direct communications among all operations is the safety of flight operations due to dynamic
UAVs, similar to vehicular communications such as ITS-G5 spectrum access.
(intelligent transportation systems communications standard at 2) Artificial Intelligence Integration Into UAVs: Artificial
5.9 GHz). Both approaches have their pros and cons in terms intelligence (AI) is expected to affect every aspect of our future
of robustness, latency, and capacity; as implied, no decision lives. Integration of AI into self driving vehicles is now a real-
has been made so far on which approach to use and only a ity [158], [159]. Similarly, many autonomous flying operations
few channel measurements have been carried out so far for are integrated into UAVs, e.g., autopilot, collision avoidance.
both approaches. However, integrating a wider spectrum of AI into UAVs
The scenarios that have to be considered for future prop- for different future applications can result in revolutionized
agation measurements should encompass urban, suburban, performance. These integrations in the communication domain
industrial, rural, and even indoor or “quasi-confined” areas on-board UAVs include optimum three dimensional path plan-
such as large arenas or stadiums. Attention should be directed ning, frequency selection for CNPC operations for different
not only to en-route situations; even though these might be propagation environments, beam steering, and data load bal-
less demanding in terms of propagation conditions, strong ancing. However, higher standards of safety are required for
MPCs are likely to occur due to reflections from smooth wet the autonomous working of UAVs.
ground or bodies of water, and from large buildings with 3) Marine Applications: Apart from exciting future appli-
metallized window fronts. In addition we also recommend cations on the ground, UAVs can be used for different future
investigating the channel for take-off and landing scenarios, marine applications. These applications include monitoring of
be it on roof tops, in gardens, or in other specifically assigned the sea surfaces using radar, using sonar to study underwater
areas. In these take-off and landing conditions, propagation marine life and seabed characteristics, surveying of ship pas-
may be unfavorable due to shadowing, strong diffraction, and sage lanes in constantly shifting ice sheets around the poles.

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2387

Similar to sea based applications, UAVs can be used for differ- which are known to exhibit interference in various frequency
ent fresh water applications, e.g., monitoring the water flows bands. For the development of countermeasures the relevant
at different points on a river or barrage. The real time data parameters of the interfering signals shall be assessed such as
collected using sensors on-board UAVs can be sent to remote typical locations of interfering TXs, transmit power, antenna
monitoring stations either through satellite or radio links. patterns, bandwidth, duty cycle, and optionally modulation
schemes.
In addition to the UAV settings, there are several other fac-
B. UAV AG Propagation Measurements tors that need to be taken into account for comprehensive
Existing AG propagation channel measurements and models AG propagation measurements using UAVs. One of these is
mostly apply to aeronautical communications at higher flight the placement and orientation of antennas. The placement of
altitudes than envisioned for small UAVs. These smaller struc- antennas should be such that there is minimum shadowing
tures have limited on-board computation capabilities, strict and noise effect from the air-frame and motors while flying.
power limitations, and can only fly at much lower altitudes, Achieving this is not always easy, and will usually be UAV-
and at present, only for short durations. There is a growing specific. The antenna orientation has been shown to result
demand for higher data rates, low latency, and high reliability in different throughputs and RSS values [60], [61], [84], [85]
for future communications, and this will be challenging for for different flight maneuvers. In order to provide better cov-
current civilian UAV architectures. erage during flight, omni-directional antennas on both the
Additionally, as noted in Section I, there are usually TX and the RX are commonly used, especially for CNPC
two types of communications maintained simultaneously for communications. An important factor during UAV flight with
UAVs: payload and CNPC. However, currently there are no omni-directional antennas is the three dimensional pattern of
standards adopted worldwide for these two types of commu- the antenna radiation [108]. This factor is important when the
nications for UAVs. Both can have their own operating bands elevation angle between the GS and the UAV is high. Use
that may or may not overlap. The CNPC communication links of directional antennas is dependent on the specific applica-
are pivotal for maintaining safety of flight and any interference tion and coverage. When selecting UAV antenna options, the
can be catastrophic. Standards organizations are thus working mechanical viability for a given UAV type should also be taken
on robust loss of link procedures. Moreover, the CNPC needs into account, e.g., a long helical antenna or yagi uda structure
to be secure and resistant to jamming and hacking attacks. The may be difficult to mount on a fixed wing aircraft compared
USA has developed a standard, primarily for medium and large to a horn or patch antenna.
aircraft [160], with standards envisioned for smaller UAVs in There is no fixed number of antennas recommended for opti-
the future. mum performance, and the number of antennas will depend
Future measurement campaigns should take into account on the operating frequency, UAV size, and operational envi-
not only a great variety of buildings - small and large ones, ronments. In many experiments multiple antennas are used on
rectangular and irregularly shaped ones, industrial facilities, UAVs, and these may be helpful for improved coverage and
halls, and towers - but also reflecting areas like bodies of diversity gains, but at the expense of increased computation,
water, streets, and squares, and demanding situations when space, and power requirements.
a UAV lands on a terrace or the like. Especially for modeling The ambient conditions on-board UAVs must also be taken
the UAV-to-UAV channel, different velocities and flight situ- into account for precise measurement of any communication
ations should be investigated, e.g., two UAVs flying toward link characteristics (for CNPC or otherwise). These ambient
each other, with one UAV near ground and the other up in the conditions include noise from the motors, noise from aircraft
air, and swarms of UAVs flying with the same velocity. For electronics, air friction while moving, sudden air gusts, tem-
cellular-like deployments, interference is likely to be a signif- perature variations, and outside-system interference. The latter
icant issue that influences network planning. Thus, it would may be particularly severe for unlicensed bands. Another con-
be useful to have measurements up to far distances (and over sideration with the use of unlicensed bands and commodity
different terrain). We envisage that the UAV-to-UAV channel radios is that the adaptive modulation and coding algorithms
for small UAVs in urban areas is as diverse as the car-to-car employed for terrestrial networks (which often assume quasi-
channel, the latter being modeled as a 2.5 dimensional chan- stationary conditions) may not work so well when directly
nel whereas the UAV-to-UAV channel will often need to be applied to highly dynamic UAV AG propagation channels.
modeled as a 3 dimensional channel. Nearly all current day channel measurements take advantage
UAV Communications are likely to operate in protected of positioning information, typically from global navigation
frequency bands. The frequency assignments can be exclu- satellite systems, with GPS being the most widely used. In
sive or on a primary basis, but these regulations may not addition to position information, GPS signals also provide an
prevent interfering signals which can drastically degrade the accurate time reference. Depending on measurement require-
performance of the UAV communication system. Interfering ments and the envisioned application, the accuracy of GPS
signals are unintentional or intentional. Both cases have to may or may not be sufficient, and this should be considered
be considered although the first case might be more frequent. before beginning measurement campaigns.
We recommend assessing the expected amount of interference When using UAVs in swarms, the location and mobility
and measuring interference in various environments. The focus aware routing methods that are used for terrestrial networks
should be on highly populated as well as industrial areas may need to be adapted to account for the three dimensional

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2388 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

movement of UAVs. Similarly, route selection algorithms for [3] D. H. Shim, H. J. Kim, and S. Sastry, “Control system design for
mobility aware networks will need to consider the fast varying rotorcraft-based unmanned aerial vehicles using time-domain system
identification,” in Proc. IEEE Int. Conf. Control Appl., Anchorage, AK,
channel conditions during UAV flight. USA, 2000, pp. 808–813.
[4] H. Shim, “Hierarchical flight control system synthesis for rotorcraft-
based unmanned aerial vehicles,” Ph.D. dissertation, Eng. Mech. Eng.,
C. UAV AG Propagation Channel Models Univ. California at Berkeley, Berkeley, CA, USA, 2000.
The UAV AG propagation literature mostly covers the [5] J. D. Blom, Unmanned Aerial Systems: A Historical Perspective,
vol. 45. Fort Leavenworth, KS, USA: Combat Stud. Inst. Press, 2010.
modeling of PL, as described in Section VI-B. As noted, and
[6] M. McFarland, UPS Drivers May Tag Team Deliveries With
as is common for terrestrial channels, the PL models are typ- Drones, CNN News, Atlanta, GA, USA, Feb. 2017. [Online].
ically provided as a function of link distance. For UAVs there Available: https://money.cnn.com/2017/02/21/technology/ups-drone-
might be other models appropriate for certain cases, for exam- delivery/index.html
[7] J. Desjardins, Amazon and UPS Are Betting Big on Drone Delivery,
ple a PL model as a function of UAV altitude in a given setting, Bus. Insider, New York, NY, USA, Mar. 2018. [Online]. Available:
or even indoor UAV PL models for certain settings (e.g., large https://www.businessinsider.com/amazon-and-ups-are-betting-big-on-
arenas). drone-delivery-2018-3
[8] Amazon Prime Air Drone Delivery. Accessed: Dec. 10, 2018.
The most accurate UAV AG propagation channel mod- [Online]. Available: https://www.amazon.com/Amazon-Prime-Air/
els are of course time varying, but in some cases these b?ie=UTF8&node=8037720011
can be specialized to time-invariant approximations, e.g., [9] Google Project Wing. Accessed: Dec. 10, 2018. [Online]. Available:
when a UAV is hovering above an area of static objects. https://x.company/projects/wing/
[10] Google Project Loon. Accessed: Dec. 10, 2018. [Online]. Available:
In [54], [58], [76], [79], the channel is considered to be https://loon.co/
quasi stationary only for short distances, and small scale [11] Uber Elevate. Accessed: Dec. 10, 2018. [Online]. Available:
fading parameters are evaluated over that stationarity interval. https://www.uber.com/us/en/elevate/
Additional studies of the stationarity distance should be con- [12] L. Bandoim, Uber Plans to Launch Food-Delivery Drones,
Forbes, Jersey City, NJ, USA, Oct. 2018. [Online]. Available:
ducted for other UAV propagation scenarios, using multiple https://www.forbes.com/sites/lanabandoim/2018/10/23/uber-plans-to-
metrics: the PDP correlation coefficient, correlation matrix launch-food-delivery-drones/#71c10d17e147
collinearity, spectral divergence, and evolutionary spectrum [13] J. Johnsson and A. Levin, Boeing Is Getting Ready to Sell
Flying Taxis, Bloomberg, New York, NY, USA, Mar. 2018.
have all been used, but each metric has its own advantages and [Online]. Available: https://www.bloomberg.com/news/articles/2018-
disadvantages. Depending on environments, additional UAV 03-01/boeing-is-getting-ready-to-sell-flying-taxis-within-a-decade
measurement campaigns will likely result in more elaborate [14] Zipline Medical Supply Drone Delivery. Accessed: Dec. 12, 2018.
[Online]. Available: http://www.flyzipline.com/
UAV AG propagation channel models, that may make use of
[15] Flytrex on Demand Drone Delivery. Accessed: Dec. 12, 2018. [Online].
MPC clusters, spatial (angular) information, and correlations Available: http://www.flytrex.com/
among model parameters. Ultimately, deterministic and hybrid [16] Wikipedia. General Atomics MQ-9 Reaper. Accessed: Jul. 3, 2017.
channel models using GBSCM principles will likely evolve to [Online]. Available: https://en.wikipedia.org/wiki/General_Atomics_
MQ-9_Reaper
be the most widely used to characterize UAV AG propagation. [17] Goldman Sachs. Drones: Reporting for Work. Accessed:
Dec. 12, 2018. [Online]. Available: https://www.goldmansachs.com/
insights/technology-driving-innovation/drones/
VIII. C ONCLUDING R EMARKS [18] Art Pregler and AT&T. (Feb. 2017). When COWs Fly: AT&T
In this paper, we have provided a comprehensive survey for Sending LTE Signals From Drones. [Online]. Available:
https://about.att.com/innovationblog/cows_fly
AG propagation channels for UAVs. The measurement cam-
[19] Vodafone. (Nov. 2018). Beyond Visual Line of Sight Drone Trial Report.
paigns in the literature for AG propagation were summarized, [Online]. Available: https://www.vodafone.com/content/dam/vodafone-
with information provided on the type of channel sounding images/media/Downloads/Vodafone_BVLOS_drone_trial_report.pdf
signal, its center frequency, bandwidth, transmit power, UAV [20] X. Lin et al., “The sky is not the limit: LTE for unmanned aerial
vehicles,” IEEE Commun. Mag., vol. 56, no. 4, pp. 204–210, Apr. 2018.
speed, height of UAV and GS, link distance, elevation angle, [21] I. Kovacs, R. Amorim, H. C. Nguyen, J. Wigard, and P. Mogensen,
and local GS environment characteristics. Air-ground chan- “Interference analysis for UAV connectivity over LTE using aerial radio
nel statistics from the literature were also provided. Various measurements,” in Proc. IEEE Veh. Technol. Conf. (VTC), Toronto, ON,
Canada, 2017, pp. 1–6.
UAV propagation scenarios and important implementation fac-
[22] R. Amorim, H. Nguyen, P. Mogensen, I. Z. Kovács, J. Wigard, and
tors for these measurements were also discussed. Large scale T. B. Sørensen, “Radio channel modeling for UAV communication
fading, small scale fading, MIMO channel characteristics and over cellular networks,” IEEE Wireless Commun. Lett., vol. 6, no. 4,
models, and channel simulations were all described. Finally, pp. 514–517, Aug. 2017.
[23] Qualcomm. (May 2017). LTE Unmanned Aircraft Systems. [Online].
future research directions and challenges were discussed. We Available: https://www.qualcomm.com/media/documents/files/lte-
expect that more elaborate and accurate AG propagation mea- unmanned-aircraft-systems-trial-report.pdf
surement campaigns and channel models will be developed in [24] FAA Small Unmanned Aircraft Regulations, Federal Aviation Admin.,
Washington, DC, USA, 2018.
the future, and we hope this study will be of use in that regard.
[25] Commercial Drone Shipments to Surpass 2.6 Million Units Annually
by 2025, Tractica, Boulder, CO, USA. Accessed: Nov. 28, 2017.
R EFERENCES [26] Qualcomm. Leading the World to 5G: Evolving Cellular
Technologies for Safer Drone Operation. Accessed: May 17, 2017.
[1] C. S. Sharp, O. Shakernia, and S. S. Sastry, “A vision system for [Online]. Available: https://www.qualcomm.com/invention/
landing an unmanned aerial vehicle,” in Proc. IEEE Int. Conf. Robot. technologies/lte/advanced-pro/cellular-drone-communication
Autom. (ICRA), vol. 2. Seoul, South Korea, 2001, pp. 1720–1727. [27] T. Patterson. Google, Facebook, SpaceX, OneWeb Plan to Beam
[2] G. P. Jones, IV, L. G. Pearlstine, and H. F. Percival, “An assessment Internet Everywhere. Accessed: May 17, 2017. [Online]. Available:
of small unmanned aerial vehicles for wildlife research,” Wildlife Soc. http://www.cnn.com/2015/10/30/tech/pioneers-google-facebook-
Bull., vol. 34, no. 3, pp. 750–758, 2006. spacex-oneweb-satellite-drone-balloon-internet/

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2389

[28] B. Kerczewski, “Spectrum for UAS control and non-payload commu- [52] S. M. Gulfam, J. Syed, M. N. Patwary, and M. Abdel-Maguid, “On
nications,” in Proc. IEEE Integr. Commun. Navig. Surveillance Conf. the spatial characterization of 3-D air-to-ground radio communication
(ICNS), 2013, pp. 1–21. channels,” in Proc. IEEE Int. Conf. Commun. (ICC), London, U.K.,
[29] B. R. Jackson, “Telemetry, command and control of UAS in the national 2015, pp. 2924–2930.
airspace,” in Proc. Eur. Telemetry Test Conf., Nuremberg, Germany, [53] D. W. Matolak and R. Sun, “Air–ground channel characterization for
2016, pp. 145–155. unmanned aircraft systems: The near-urban environment,” in Proc.
[30] F. White, “Air–ground communications: History and expectations,” IEEE Mil. Commun. Conf. (MILCOM), Tampa, FL, USA, 2015,
IEEE Trans. Commun., vol. 21, no. 5, pp. 398–407, May 1973. pp. 1656–1660.
[31] M. S. B. Mahmoud et al., Aeronautical Air–Ground Data Link [54] D. W. Matolak and R. Sun, “Air–ground channel characterization for
Communications. New York, NY, USA: Wiley, 2014. unmanned aircraft systems—Part I: Methods, measurements, and mod-
[32] M. Schnell, U. Epple, D. Shutin, and N. Schneckenburger, “LDACS: els for over-water settings,” IEEE Trans. Veh. Technol., vol. 66, no. 1,
Future aeronautical communications for air-traffic management,” IEEE pp. 26–44, Jan. 2017.
Commun. Mag., vol. 52, no. 5, pp. 104–110, May 2014. [55] S. Kaul et al., “Effect of antenna placement and diversity on vehic-
[33] VHF/UHF Military Monitoring. Accessed: May 31, 2017. ular network communications,” in Proc. IEEE Sensor Mesh Ad Hoc
[Online]. Available: http://wiki.radioreference.com/index.php/VHF/ Commun. Netw., San Diego, CA, USA, Jun. 2007, pp. 112–121.
UHF_Military_Monitoring [56] A. R. Ruddle, “Simulation of far-field characteristics and measurement
techniques for vehicle-mounted antennas,” in Proc. IEE Colloquium
[34] E. Haas, “Aeronautical channel modeling,” IEEE Trans. Veh. Technol.,
Antennas Automotives, 2000, pp. 1–8.
vol. 51, no. 2, pp. 254–264, Mar. 2002.
[57] W. G. Newhall et al., “Wideband air-to-ground radio channel measure-
[35] B. G. Gates, “Aeronautical communications,” Elect. Eng. III A ments using an antenna array at 2 GHz for low-altitude operations,”
Radiocommun. J., vol. 94, no. 11, pp. 74–81, Mar. 1947. in Proc. IEEE Mil. Commun. Conf. (MILCOM), vol. 2. Boston, MA,
[36] D. F. Lamiano, K.-H. Leung, L. C. Monticone, W. J. Wilson, and USA, 2003, pp. 1422–1427.
B. Phillips, “Digital broadband VHF aeronautical communications for [58] D. W. Matolak and R. Sun, “Antenna and frequency diversity
air traffic control,” in Proc. Integr. Commun. Navig. Surveillance Conf., in the unmanned aircraft systems bands for the over-sea setting,”
Arlington, VA, USA, May 2009, pp. 1–12. in Proc. IEEE Digit. Avionics Syst. Conf. (DASC), Oct. 2014,
[37] D. W. Matolak, “Air–ground channels & models: Comprehensive pp. 6A4-1–6A4-10.
review and considerations for unmanned aircraft systems,” in Proc. [59] J. Chen, B. Daneshrad, and W. Zhu, “MIMO performance evaluation
IEEE Aerosp. Conf., Mar. 2012, pp. 1–17. for airborne wireless communication systems,” in Proc. Mil. Commun.
[38] C. Levis, J. T. Johnson, and F. L. Teixeira, Radiowave Propagation: Conf. (MILCOM), Baltimore, MD, USA, Nov. 2011, pp. 1827–1832.
Physics and Applications. Hoboken, NJ, USA: Wiley, 2010. [60] C.-M. Cheng, P.-H. Hsiao, H. T. Kung, and D. Vlah, “Performance
[39] G. M. Djuknic, J. Freidenfelds, and Y. Okunev, “Establishing wireless measurement of 802.11a wireless links from UAV to ground nodes with
communications services via high-altitude aeronautical platforms: A various antenna orientations,” in Proc. Int. Conf. Comput. Commun.
concept whose time has come?” IEEE Commun. Mag., vol. 35, no. 9, Netw. (ICCCN), Arlington, VA, USA, 2006, pp. 303–308.
pp. 128–135, Sep. 1997. [61] E. Yanmaz, R. Kuschnig, and C. Bettstetter, “Channel measure-
[40] D. W. Matolak and R. Sun, “Unmanned aircraft systems: Air–ground ments over 802.11a-based UAV-to-ground links,” in Proc. IEEE Glob.
channel characterization for future applications,” IEEE Veh. Technol. Commun. Conf. (GLOBECOM) Workshops, Houston, TX, USA, 2011,
Mag., vol. 10, no. 2, pp. 79–85, Jun. 2015. pp. 1280–1284.
[41] A. A. Khuwaja, Y. Chen, N. Zhao, M.-S. Alouini, and P. Dobbins, “A [62] T. J. Willink, C. C. Squires, G. W. K. Colman, and M. T. Muccio,
survey of channel modeling for UAV communications,” IEEE Commun. “Measurement and characterization of low-altitude air-to-ground
Surveys Tuts., vol. 20, no. 4, pp. 2804–2821, 2018. MIMO channels,” IEEE Trans. Veh. Technol., vol. 65, no. 4,
[42] M. Asadpour, B. V. den Bergh, D. Giustiniano, K. A. Hummel, pp. 2637–2648, Apr. 2016.
S. Pollin, and B. Plattner, “Micro aerial vehicle networks: An experi- [63] D. W. Matolak, H. Jamal, and R. Sun, “Spatial and frequency correla-
mental analysis of challenges and opportunities,” IEEE Commun. Mag., tions in two-ray air–ground SIMO channels,” in Proc. IEEE Int. Conf.
vol. 52, no. 7, pp. 141–149, Jul. 2014. Commun. (ICC), Paris, France, May, 2017, pp. 1–6.
[43] Y. Zeng, R. Zhang, and T. J. Lim, “Wireless communications [64] X. Gao, Z. Chen, and Y. Hu, “Analysis of unmanned aerial vehicle
with unmanned aerial vehicles: Opportunities and challenges,” IEEE MIMO channel capacity based on aircraft attitude,” WSEAS Trans. Inf.
Commun. Mag., vol. 54, no. 5, pp. 36–42, May 2016. Sci. Appl., vol. 10, no. 2, pp. 58–67, 2013.
[44] L. Afonso, N. Souto, P. Sebastiao, M. Ribeiro, T. Tavares, [65] C. Zhang and Y. Hui, “Broadband air-to-ground communications with
and R. Marinheiro, “Cellular for the skies: Exploiting mobile adaptive MIMO datalinks,” in Proc. IEEE Digit. Avionics Syst. Conf.
network infrastructure for low altitude air-to-ground communica- (DASC), Seattle, WA, USA, 2011, pp. 4D4-1–4D4-10.
tions,” IEEE Aerosp. Electron. Syst. Mag., vol. 31, no. 8, pp. 4–11, [66] J. Yang, P. Liu, and H. Mao, “Model and simulation of narrow-
Aug. 2016. band ground-to-air fading channel based on Markov process,” in Proc.
Netw. Comput. Inf. Security Conf. (NCIS), vol. 1. Guilin, China, 2011,
[45] Z. Xiao, P. Xia, and X.-G. Xia, “Enabling UAV cellular with millimeter-
pp. 142–146.
wave communication: Potentials and approaches,” IEEE Commun.
[67] C. Bluemm, C. Heller, B. Fourestie, and R. Weigel, “Air-to-ground
Mag., vol. 54, no. 5, pp. 66–73, May 2016.
channel characterization for OFDM communication in C-band,” in
[46] M. Ibrahim and H. Arslan, “Air–ground doppler-delay spread spectrum Proc. Int. Conf. Signal Process. Commun. Syst. (ICSPCS), 2013,
for dense scattering environments,” in Proc. IEEE Mil. Commun. Conf. pp. 1–8.
(MILCOM), Tampa, FL, USA, Oct. 2015, pp. 1661–1666. [68] M. Simunek, F. P. Fontán, and P. Pechac, “The UAV low elevation
[47] R. Essaadali and A. Kouki, “A new simple unmanned aerial vehicle propagation channel in urban areas: Statistical analysis and time-
doppler effect RF reducing technique,” in Proc. IEEE Mil. Commun. series generator,” IEEE Trans. Antennas Propag., vol. 61, no. 7,
Conf., Baltimore, MD, USA, Nov. 2016, pp. 1179–1183. pp. 3850–3858, Jul. 2013.
[48] A. Al-Hourani, S. Kandeepan, and S. Lardner, “Optimal LAP altitude [69] V. Vahidi and E. Saberinia, “Orthogonal frequency division
for maximum coverage,” IEEE Wireless Commun. Lett., vol. 3, no. 6, multiplexing and channel models for payload communications of
pp. 569–572, Dec. 2014. unmanned aerial systems,” in Proc. Int. Conf. Unmanned Aircraft Syst.
[49] X. Cheng, C.-X. Wang, D. I. Laurenson, and A. V. Vasilakos, “Second (ICUAS), Arlington, VA, USA, 2016, pp. 1156–1161.
order statistics of non-isotropic mobile-to-mobile Ricean fading chan- [70] Z. Wu, H. Kumar, and A. Davari, “Performance evaluation of OFDM
nels,” in Proc. IEEE Int. Conf. Commun. (ICC), Dresden, Germany, transmission in UAV wireless communication,” in Proc. Southeastern
2009, pp. 1–5. Symp. Syst. Theory (SSST), Tuskegee, AL, USA, 2005, pp. 6–10.
[50] A. Ksendzov, “A geometrical 3D multi-cluster mobile-to-mobile [71] H. D. Tu and S. Shimamoto, “A proposal of wide-band air-to-ground
MIMO channel model with Rician correlated fading,” in Proc. IEEE communication at airports employing 5-GHz band,” in Proc. IEEE
Int. Congr. Ultra Modern Telecommun. (ICUMT) Conf., Lisbon, Wireless Commun. Netw. Conf. (WCNC), Budapest, Hungary, 2009,
Portugal, 2016, pp. 191–195. pp. 1–6.
[51] S. M. Gulfam, S. J. Nawaz, A. Ahmed, and M. N. Patwary, “Analysis [72] W. Khawaja, I. Guvenc, and D. W. Matolak, “UWB channel sounding
on multipath shape factors of air-to-ground radio communication chan- and modeling for UAV air-to-ground propagation channels,” in Proc.
nels,” in Proc. IEEE Wireless Telecommun. Symp. (WTS), London, IEEE Glob. Commun. Conf. (GLOBECOM), Washington, DC, USA,
U.K., 2016, pp. 1–5. Dec. 2016, pp. 1–7.

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
2390 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 21, NO. 3, THIRD QUARTER 2019

[73] D. W. Matolak and R. Sun, “Air–ground channel measurements & [95] N. Goddemeier and C. Wietfeld, “Investigation of air-to-air channel
modeling for UAS,” in Proc. Integr. Commun. Navig. Surveillance characteristics and a UAV specific extension to the Rice model,” in
Conf. (ICNS), 2013, pp. 1–9. Proc. IEEE Glob. Commun. (Globecom) Workshops, San Diego, CA,
[74] R. Sun and D. W. Matolak, “Over-harbor channel modeling with direc- USA, Dec. 2015, pp. 1–5.
tional ground station antennas for the air–ground channel,” in Proc. [96] Federal Aviation Administration. FAA Rules for UAVs. Accessed:
IEEE Mil. Commun. Conf. (MILCOM), Baltimore, MD, USA, 2014, Feb. 25, 2017. [Online]. Available: https://www.faa.gov/uas/
pp. 382–387. beyond_the_basics/
[75] D. W. Matolak and R. Sun, “Air–ground channel characterization for [97] Universitat Politäcnica de Catalunya. Vector Network Analyzer
unmanned aircraft systems: The hilly suburban environment,” in Proc. Specifications. Accessed: May 18, 2017. [Online]. Available:
Veh. Technol. Conf. (VTC), Vancouver, BC, Canada, 2014, pp. 1–5. http://www.upc.edu/sct/en/documents_equipament/d_160_id-655-2.pdf
[76] R. Sun and D. W. Matolak, “Air–ground channel characterization for [98] International Telecommunication Union. Terrain Cover Types.
unmanned aircraft systems—Part II: Hilly and mountainous settings,” Accessed: Jul. 5, 2017. [Online]. Available: https://www.itu.int/
IEEE Trans. Veh. Technol., vol. 66, no. 3, pp. 1913–1925, Mar. 2017. oth/R0A04000031/en
[77] D. W. Matolak and R. Sun, “Air–ground channels for UAS: Summary [99] International Telecommunication Union. Propagation by Diffraction.
of measurements and models for L-and C-bands,” in Proc. Integr. Accessed: Jul. 5, 2017. [Online]. Available: http://www.itu.int/
Commun. Navig. Surveillance (ICNS), 2016, pp. 8B2-1–8B2-11. dms_pubrec/itu-r/rec/p/R-REC-P.526-13-201311-I!!PDF-E.pdf
[78] D. W. Matolak and R. Sun, “Air–ground channel characterization [100] M. Kvicera, F. P. Fontán, J. Israel, and P. Pechac, “A new model for
for unmanned aircraft systems—Part III: The suburban and near- scattering from tree canopies based on physical optics and multiple
urban environments,” IEEE Trans. Veh. Technol., vol. 66, no. 8, scattering theory,” IEEE Trans. Antennas Propag., vol. 65, no. 4,
pp. 6607–6618, Aug. 2017. pp. 1925–1933, Apr. 2017.
[79] D. W. Matolak, “Channel characterization for unmanned aircraft [101] F. Kawamata, “Optimum frame size for land mobile satellite communi-
systems,” in Proc. Eur. Conf. Antennas Propag. (EuCAP), Lisbon, cation channels,” in Proc. IEEE Glob. Commun. (GLOBECOM) Conf.,
Portugal, 2015, pp. 1–5. vol. 1. Houston, TX, USA, Nov. 1993, pp. 583–587.
[80] N. Schneckenburger et al., “Measurement of the L-band air-to-ground [102] A. A. Aboudebra, K. Tanaka, T. Wakabayashi, S. Yamamoto, and
channel for positioning applications,” IEEE Trans. Aerosp. Electron. H. Wakana, “Signal fading in land-mobile satellite communication
Syst., vol. 52, no. 5, pp. 2281–2297, Oct. 2016. systems: Statistical characteristics of data measured in Japan using
[81] K. Takizawa, T. Kagawa, S. Lin, F. Ono, H. Tsuji, and R. Miura, ETS-VI,” IEE Proc. Microw. Antennas Propag., vol. 146, no. 5,
“C-band aircraft-to-ground (A2G) radio channel measurement for pp. 349–354, Oct. 1999.
unmanned aircraft systems,” in Proc. Conf. Wireless Pers. Multimedia [103] International Telecommunication Union. Ducting Over Sea
Commun. (WPMC), Sydney, NSW, Australia, 2014, pp. 754–758. Calculation. Accessed: Jul. 5, 2017. [Online]. Available: http://
[82] F. Ono, K. Takizawa, H. Tsuji, and R. Miura, “S-band radio propagation www.itu.int/md/dologin_md.asp?id=R03-WRC03-C-0025!A27-L188!
characteristics in urban environment for unmanned aircraft systems,” MSW-E
in Proc. Antennas Propag. (ISAP) Conf., Hobart, TAS, Australia, 2015, [104] Q. Chen, “Wideband channel sounding techniques for dynamic spec-
pp. 1–4. trum access networks,” Ph.D. dissertation, Elect. Eng. Comput. Sci.,
Univ. Kansas, Lawrence, KS, USA, 2009.
[83] N. Goddemeier, K. Daniel, and C. Wietfeld, “Coverage evaluation of
[105] J. Parsons, The Mobile Radio Propagation Channel, 2nd ed. West
wireless networks for unmanned aerial systems,” in Proc. IEEE Glob.
Sussex, U.K.: Wiley, 2000.
Commun. Conf. (GLOBECOM) Workshops, Miami, FL, USA, 2010,
[106] J. Chen, D. Raye, W. Khawaja, P. Sinha, and I. Guvenc, “Impact of 3D
pp. 1760–1765.
UWB antenna radiation pattern on air-to-ground drone connectivity,”
[84] E. Yanmaz, R. Kuschnig, and C. Bettstetter, “Achieving air-ground
in Proc. Veh. Technol. Conf. (VTC), Chicago, IL, USA, Aug. 2018,
communications in 802.11 networks with three-dimensional aerial
pp. 1–5.
mobility,” in Proc. IEEE INFOCOM, 2013, pp. 120–124.
[107] W. Khawaja, O. Ozdemir, F. Erden, I. Guvenc, and D. Matolak, “UWB
[85] N. Ahmed, S. S. Kanhere, and S. Jha, “On the importance of link air-to-ground propagation channel measurements and modeling using
characterization for aerial wireless sensor networks,” IEEE Commun. UAVs,” in Proc. IEEE Aerosp. Conf., 2019.
Mag., vol. 54, no. 5, pp. 52–57, May 2016. [108] W. Khawaja, O. Ozdemir, and I. Guvenc, “Temporal and spatial char-
[86] E. L. Cid, A. V. Alejos, and M. G. Sanchez, “Signaling through acteristics of mmWave propagation channels for UAVs,” in Proc. Glob.
scattered vegetation: Empirical loss modeling for low elevation angle Symp. Millimeter Waves (GSMM), Boulder, CO, USA, May 2018,
satellite paths obstructed by isolated thin trees,” IEEE Veh. Technol. pp. 1–6.
Mag., vol. 11, no. 3, pp. 22–28, Sep. 2016. [109] A. Paier et al., “Non-WSSUS vehicular channel characterization in
[87] J. Romeu, A. Aguasca, J. Alonso, S. Blanch, and R. R. Martins, “Small highway and urban scenarios at 5.2GHz using the local scattering
UAV radiocommunication channel characterization,” in Proc. Eur. function,” in Proc. Int. Workshop Smart Antennas, Vienna, Austria,
Conf. Antennas Propag. (EuCAP), Barcelona, Spain, 2010, pp. 1–5. Feb. 2008, pp. 9–15.
[88] H. T. Kung, C.-K. Lin, T.-H. Lin, S. J. Tarsa, and D. Vlah, “Measuring [110] O. Renaudin, V. M. Kolmonen, P. Vainikainen, and C. Oestges, “Non-
diversity on a low-altitude UAV in a ground-to-air wireless 802.11 stationary narrowband MIMO inter-vehicle channel characterization
mesh network,” in Proc. IEEE Glob. Commun. Conf. (GLOBECOM) in the 5-GHz band,” IEEE Trans. Veh. Technol., vol. 59, no. 4,
Workshops, 2010, pp. 1799–1804. pp. 2007–2015, May 2010.
[89] J. Zelenỳ, F. Pérez-Fontán, and P. Pechač, “Initial results from a [111] M. Wentz and M. Stojanovic, “A MIMO radio channel model for low-
measurement campaign for low elevation angle links in different envi- altitude air-to-ground communication systems,” in Proc. IEEE Veh.
ronments,” in Proc. Eur. Conf. Antennas Propag. (EuCAP), 2015, Technol. Conf. (VTC), Boston, MA, USA, 2015, pp. 1–6.
pp. 1–4. [112] A. Al-Hourani, S. Kandeepan, and A. Jamalipour, “Modeling air-to-
[90] E. Teng, J. Falcao, C. Dominguez, F. Mokaya, P. Zhang, and ground path loss for low altitude platforms in urban environments,” in
B. Iannucci, Aerial Sensing and Characterization of Three-Dimensional Proc. IEEE Glob. Commun. (GLOBECOM) Conf., Austin, TX, USA,
RF Fields, Univ. Buffalo, Buffalo, NY, USA, Sep. 2016. 2014, pp. 2898–2904.
[91] Y. S. Meng and Y. H. Lee, “Measurements and characterizations of [113] J. Holis and P. Pechac, “Elevation dependent shadowing model
air-to-ground channel over sea surface at C-band with low airborne for mobile communications via high altitude platforms in built-up
altitudes,” IEEE Trans. Veh. Technol., vol. 60, no. 4, pp. 1943–1948, areas,” IEEE Trans. Antennas Propag., vol. 56, no. 4, pp. 1078–1084,
May 2011. Apr. 2008.
[92] J. Kunisch, I. De La Torre, A. Winkelmann, M. Eube, and T. Fuss, [114] Q. Feng, J. McGeehan, E. K. Tameh, and A. R. Nix, “Path loss models
“Wideband time-variant air-to-ground radio channel measurements at for air-to-ground radio channels in urban environments,” in Proc. IEEE
5 GHz,” in Proc. Eur. Conf. Antennas Propag. (EUCAP), Rome, Italy, Veh. Technol. Conf. (VTC), vol. 6. Melbourne, VIC, Australia, 2006,
2011, pp. 1386–1390. pp. 2901–2905.
[93] W. Khawaja, O. Ozdemir, and I. Guvenc, “UAV air-to-ground channel [115] N. Schneckenburger, T. Jost, D. Shutin, and U. C. Fiebig, “Line of
characterization for mmWave systems,” in Proc. IEEE Veh. Technol. sight power variation in the air to ground channel,” in Proc. EuCAP,
Conf. (VTC), Toronto, ON, Canada, Sep. 2017, pp. 1–5. Davos, Switzerland, 2016, pp. 1–5.
[94] M. Walter, S. Gligorević, T. Detert, and M. Schnell, “UHF/VHF air- [116] I. J. Timmins and S. O’Young, “Marine communications channel
to-air propagation measurements,” in Proc. IEEE Eur. Conf. Antennas modeling using the finite-difference time domain method,” IEEE Trans.
Propag. (EuCAP), Barcelona, Spain, 2010, pp. 1–5. Veh. Technol., vol. 58, no. 6, pp. 2626–2637, Jul. 2009.

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.
KHAWAJA et al.: SURVEY OF AG PROPAGATION CHANNEL MODELING FOR UAVs 2391

[117] H. Jamal, D. W. Matolak, and R. Sun, “Comparison of L-DACS and [139] M. Alzenad, A. El-Keyi, F. Lagum, and H. Yanikomeroglu, “3D
FBMC performance in over-water air–ground channels,” in Proc. IEEE placement of an unmanned aerial vehicle base station (UAV-BS) for
Digit. Avionics Syst. Conf. (DASC), Prague, Czech Republic, 2015, energy-efficient maximal coverage,” IEEE Wireless Commun. Lett.,
pp. 2D6-1–2D6-9. vol. 6, no. 4, pp. 434–437, Aug. 2017.
[118] H. Jamal and D. W. Matolak, “FBMC and L-DACS performance [140] International Telecommunication Union. (2003). Propagation
for future air-to-ground communication systems,” IEEE Trans. Veh. Data and Prediction Methods Required for the Design of
Technol., vol. 66, no. 6, pp. 5043–5055, Jun. 2017. Terrestrial Broadband Millimetric Radio Access Systems. Accessed:
[119] S. Blandino, F. Kaltenberger, and M. Feilen, “Wireless channel sim- Nov. 27, 2017. [Online]. Available: http://www.catr.cn/catr/catr/
ulator testbed for airborne receivers,” in Proc. IEEE Glob. Commun. itu/itur/iturlist.jsp?docplace=P&vchar1=P.1410-2
(Globecom) Workshops, 2015, pp. 1–6. [141] R. Sun, D. W. Matolak, and W. Rayess, “Air-ground channel character-
[120] F. Jiang and A. L. Swindlehurst, “Optimization of UAV heading for ization for unmanned aircraft systems—Part IV: Airframe shadowing,”
the ground-to-air uplink,” IEEE J. Sel. Areas Commun., vol. 30, no. 5, IEEE Trans. Veh. Technol., vol. 66, no. 9, pp. 7643–7652, Sep. 2017.
pp. 993–1005, Jun. 2012. [142] W. Q. Malik, B. Allen, and D. J. Edwards, “Impact of bandwidth
[121] M. M. Azari, F. Rosas, K. C. Chen, and S. Pollin, “Optimal UAV on small-scale fade depth,” in Proc. IEEE Glob. Commun. Conf.
positioning for terrestrial-aerial communication in presence of fading,” (GLOBECOM), Washington, DC, USA, Nov. 2007, pp. 3837–3841.
in Proc. IEEE Glob. Commun. (GLOBECOM) Conf., Washington, DC, [143] Y. Zheng, Y. Wang, and F. Meng, “Modeling and simulation of pathloss
USA, Dec. 2016, pp. 1–7. and fading for air-ground link of HAPs within a network simulator,” in
[122] Z. Yun and M. F. Iskander, “Ray tracing for radio propaga- Proc. Int. Conf. Cyber Enabled Distrib. Comput. Knowl. Disc., Beijing,
tion modeling: Principles and applications,” IEEE Access, vol. 3, China, 2013, pp. 421–426.
pp. 1089–1100, 2015. [144] F. Bohagen, P. Orten, and G. E. Oien, “Design of optimal high-rank
[123] K. Daniel, M. Putzke, B. Dusza, and C. Wietfeld, “Three dimen- line-of-sight MIMO channels,” IEEE Trans. Wireless Commun., vol. 6,
sional channel characterization for low altitude aerial vehicles,” in no. 4, pp. 1420–1425, Apr. 2007.
Proc. Int. Symp. Wireless Commun. Syst. (ISWCS), York, U.K., 2010, [145] D. Rieth, C. Heller, D. Blaschke, and G. Ascheid, “On the practicability
pp. 756–760. of airborne MIMO communication,” in Proc. IEEE Digit. Avionics Syst.
[124] Y. Wu et al., “Ray tracing based wireless channel modeling over the Conf. (DASC), Prague, Czech Republic, 2015, pp. 1–10.
sea surface near Diaoyu Islands,” in Proc. Int. Conf. Comput. Intell. [146] P. Chandhar, D. Danev, and E. G. Larsson, “Massive MIMO as enabler
Theory Syst. Appl. (CCITSA), 2015, pp. 124–128. for communications with drone swarms,” in Proc. Int. Conf. Unmanned
[125] M. Simunek, P. Pechac, and F. P. Fontan, “Excess loss model for low Aircraft Syst. (ICUAS), Arlington, VA, USA, Jun. 2016, pp. 347–354.
elevation links in urban areas for UAVs,” Radioengineering, vol. 20, [147] 3GPP. (2018). The Mobile Broadband Standard. [Online]. Available:
no. 3, pp. 561–568, 2011. http://www.3gpp.org/specifications
[126] T. Tavares et al., “Generalized LUI propagation model for UAVs com- [148] J. Lyu, Y. Zeng, R. Zhang, and T. J. Lim, “Placement optimization
munications using terrestrial cellular networks,” in Proc. IEEE Veh. of UAV-mounted mobile base stations,” IEEE Commun. Lett., vol. 21,
Technol. Conf. (VTC), Boston, MA, USA, 2015, pp. 1–6. no. 3, pp. 604–607, Mar. 2017.
[127] X. Cai et al., “Low altitude UAV propagation channel modelling,” in [149] V. V. C. Ravi and H. S. Dhillon, “Downlink coverage probability in
Proc. IEEE 11th Eur. Conf. Antennas Propag. (EUCAP), Paris, France, a finite network of unmanned aerial vehicle (UAV) base stations,” in
2017, pp. 1443–1447. Proc. IEEE Conf. Signal Process. Adv. Wireless Commun. (SPAWC),
[128] IST-4-027756 WINNER II. (2003). D1.1.2 V1.0 WINNER II Channel Edinburgh, U.K., 2016, pp. 1–5.
Models. Accessed: Nov. 29, 2017. [Online]. Available: http://www2.tu- [150] “Enhanced LTE support for aerial vehicles specifications, RAN 78,”
ilmenau.de/nt/generic/paper_pdfs/Part%20II%20of%20D1.1.2.pdf 3GPP, Sophia Antipolis, France, Rep. 36.777. Accessed: May 17, 2018.
[129] T. S. Rappaport, G. R. MacCartney, M. K. Samimi, and S. Sun, [Online]. Available: https://portal.3gpp.org/desktopmodules/
“Wideband millimeter-wave propagation measurements and channel Specifications/SpecificationDetails.aspx?specificationId=3231
models for future wireless communication system design,” IEEE Trans. [151] A. Matese et al., “Intercomparison of UAV, aircraft and satellite remote
Commun., vol. 63, no. 9, pp. 3029–3056, Sep. 2015. sensing platforms for precision viticulture,” Remote Sens., vol. 7, no. 3,
[130] T. S. Rappaport. Investigation of Prediction Accuracy, Sensitivity, and pp. 2971–2990, 2015.
Parameter Stability of Large-Scale Propagation Path Loss Models [152] L. Point. Satellite Versus UAV Mapping: How Are They Different.
From 500 MHz to 100 GHz. Accessed: May 7, 2018. [Online]. Accessed: Nov. 24, 2018. [Online]. Available: http://www.landpoint.
Available: http://wireless.engineering.nyu.edu/presentations/NTIA- net/satellite-versus-uav-mapping-how-are-they-different/
propagation-presentation-JUNE-15-2016_v1%203.pdf [153] T. C. Tozer and D. Grace, “High-altitude platforms for wireless com-
[131] T. S. Rappaport, Wireless Communications: Principles and Practice, munications,” Electron. Commun. Eng. J., vol. 13, no. 3, pp. 127–137,
vol. 2. Upper Saddle River, NJ, USA: Prentice-Hall, 1996. Jun. 2001.
[132] J. R. Child, “Air-to-ground propagation at 900 MHz,” in Proc. [154] “Study on channel model for frequencies from 0.5 to 100 GHz,”
IEEE Veh. Technol. Conf. (VTC), vol. 35. Boulder, CO, USA, 1985, 3GPP, Sophia Antipolis, France, Rep. ETSI TR 138 901. Accessed:
pp. 73–80. May 17, 2018. [Online]. Available: http://www.etsi.org/deliver/
[133] P. J. Park, S.-M. Choi, D. H. Lee, and B.-S. Lee, “Performance of UAV etsi_tr/138900_138999/138901/14.00.00_60/tr_138901v140000p.pdf
(unmanned aerial vehicle) communication system adapting WiBro with [155] Y. Saleem, M. H. Rehmani, and S. Zeadally, “Integration of cognitive
array antenna,” in Proc. Int. Conf. Adv. Commun. Technol. (ICACT), radio technology with unmanned aerial vehicles: Issues, opportuni-
vol. 2, 2009, pp. 1233–1237. ties, and future research challenges,” J. Netw. Comput. Appl., vol. 50,
[134] M. Mozaffari, W. Saad, M. Bennis, and M. Debbah, “Mobile unmanned pp. 15–31, Apr. 2015.
aerial vehicles (UAVs) for energy-efficient Internet of Things com- [156] P. Jacob, R. P. Sirigina, A. S. Madhukumar, and V. A. Prasad,
munications,” IEEE Trans. Wireless Commun., vol. 16, no. 11, “Cognitive radio for aeronautical communications: A survey,” IEEE
pp. 7574–7589, Nov. 2017. Access, vol. 4, pp. 3417–3443, 2016.
[135] D. Athukoralage, I. Guvenc, W. Saad, and M. Bennis, “Regret based [157] G. Stamatescu, D. Popescu, and R. Dobrescu, “Cognitive radio as
learning for UAV assisted LTE-U/WiFi public safety networks,” in solution for ground-aerial surveillance through WSN and UAV infras-
Proc. IEEE Conf. Glob. Commun. (GLOBECOM), Washington, DC, tructure,” in Proc. Conf. IEEE Electron. Comput. Artif. Intell. (ECAI),
USA, 2016, pp. 1–7. 2014, pp. 51–56.
[136] M. Alzenad, A. El-Keyi, and H. Yanikomeroglu, “3D placement of an [158] T. Wong. Artificial Intelligence (AI) and Self Driving
unmanned aerial vehicle base station for maximum coverage of users Cars. Accessed: May 12, 2018. [Online]. Available:
with different QoS requirements,” IEEE Wireless Commun. Lett., to be http://www.semiconwest.org/programs-catalog/smart-automotive-
published. future-smart-connected-self-driving-cars/artificial-intelligence-ai-and-
[137] E. Kalantari, I. Bor-Yaliniz, A. Yongacoglu, and H. Yanikomeroglu, self-driving-cars
“User association and bandwidth allocation for terrestrial and aerial [159] Tesla. Full Self Driving Cars. Accessed: May 12, 2018. [Online].
base stations with backhaul considerations,” in Proc. Pers. Indoor Available: https://www.tesla.com/autopilot
Mobile Radio Commun. (PIMRC), Montreal, QC, Canada, Oct. 2017, [160] Radio Technical Commission for Aeronautics. Command and Control
pp. 1–6. (C2) Data Link Minimum Operational Performance Standards.
[138] R. I. Bor-Yaliniz, A. El-Keyi, and H. Yanikomeroglu, “Efficient 3- Accessed: Jul. 7, 2017. [Online]. Available: https://global.ihs.com/
D placement of an aerial base station in next generation cellular doc_detail.cfm?document_name=RTCA%20DO-362&item_s_key=00
networks,” in Proc. IEEE Int. Conf. Commun. (ICC), 2016, pp. 1–5. 694348

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on August 11,2023 at 04:15:15 UTC from IEEE Xplore. Restrictions apply.