514 IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 6, NO.

4, AUGUST 2017

Radio Channel Modeling for UAV Communication Over Cellular Networks

Rafhael Amorim, Huan Nguyen, Preben Mogensen, István Z. Kovács, Jeroen Wigard, and Troels B. Sørensen

Abstract—The main goal of this letter is to obtain models for links [6]. To study the feasibility of cellular-based commu-
path loss exponents and shadowing for the radio channel between nication for drones, a good understanding of the propagation
airborne unmanned aerial vehicles (UAVs) and cellular networks. channel between UAVs and ground stations is required. It is
In this pursuit, field measurements were conducted in live LTE
networks at the 800 MHz frequency band, using a commercial reasonable to assume the channel will present different behav-
UAV. Our results show that path loss exponent decreases as the iors for an aerial user when compared to a regular ground user.
UAV moves up, approximating freespace propagation for hori- UAVs flying above rooftops, vegetation and terrain elevations,
zontal ranges up to tens of kilometers at UAV heights around 100 are more likely to observe radio path clearance to the base
m. Our findings support the need of height-dependent parame- stations in the surrounding areas and therefore more likely to
ters for describing the propagation channel for UAVs at different
heights. experience line-of-sight (LOS) propagations [6] for larger dis-
tances resulting in higher level of interference from a larger
Index Terms—UAV, drone, path loss, propagation, channel number of surrounding BSs [7].
modelling, drone communication, field measurement, air-to-
ground, radio channel measurements. Some efforts to characterize the aerial channel were
presented by Matolak and Sun [8], [9], where measurements
were performed using single dedicated links at 900 MHz and
I. I NTRODUCTION 5 GHz bands, with large drones flying at heights between
NMANNED aerial vehicles (UAVs), also known as 500 m and 2 km, but the effect of height dependency is
U drones, have been used for military applications for more
than 20 years. More recently, technological developments
not directly assessed, neither heights below 150m, which are
expected to be heavily used by commercial drones in the near
regarding batteries, electronics and lightweight materials have future.
made UAVs more accessible to the public, creating a boom in Some previous studies have suggested it is impor-
the market of small and medium scale UAVs. However, due to tant to obtain a model that accounts for the dependency
concerns with public safety most of their applications are still observed in the propagation channel to UE heights [3], [4].
limited by countries regulations to visual-line-of-sight (VLOS) Goddemeier et al. [10] present a modification to the two-ray
ranges and maximum heights between 100 and 150 m [1]. model which introduces variation in the path loss exponent
Emerging UAV applications present potential to reduce risk according to the UE height, based on GSM and UMTS mea-
and cost for many commercial activities [2], but they would surements collected by a stationary balloon located at 1900 m
require larger operational ranges. The research community of the serving base station. Measurements in LTE using a fly-
is putting efforts into creating solutions for a safe integra- ing UAV were reported in [7] and results suggest there is a
tion of drones in the airspace for beyond-VLOS flight ranges. clearance of the radio path, obtained with higher UE heights,
An essential element in this is the development of a reliable reduces the shadowing variation while it increases the received
communication link between the pilot/controller and the UAV. signal power from the interfering cells and the number of vis-
The cellular networks are natural candidates to provide not ible neighboring cells, but no propagation model is presented.
only this link, known as CNPC (control and non-payload com- The present work differs from the previous studies, as it
munication) [1], [3], [4] or C2 (communication and control directly assesses the effects of the LTE UAV-UE heights in
link) [5], but also to serve data traffic for applications such the path loss exponent and shadowing variation, and proposes
as live streaming or sensor readings. Mobile operators already a height dependent modeling for both. A wider range of dis-
have ground infrastructures implemented and a ubiquitous cov- tances and diverse surrounding base stations are assessed using
erage that can be adapted to serve such air-to-infrastructure a flying LTE UAV-UE, connected to two real LTE networks
at 800 MHz in Denmark.
Manuscript received April 13, 2017; accepted May 19, 2017. Date of pub- This letter is organized as follows. The setup used in the
lication May 31, 2017; date of current version August 21, 2017. The associate trials and the data processing methodology are introduced in
editor coordinating the review of this paper and approving it for publication Section II. Section III present the measurements results, while
was D. Tarchi. (Corresponding author: Rafhael Amorim.)
R. Amorim, H. Nguyen, and T. B. Sørensen are with the Department the modeling of the height-dependent radio propagation chan-
of Electronic Systems, Aalborg University, 9220 Aalborg, Denmark (e-mail: nel is presented in Section IV. This letter follows with the
rma@es.aau.dk; hcn@es.aau.dk; tbs@es.aau.dk). conclusion in Section V.
P. Mogensen is with the Department of Electronic Systems, Aalborg
University, 9220 Aalborg, Denmark, and also with Nokia Bell Labs,
9220 Aalborg, Denmark (e-mail: pm@es.aau.dk).
I. Z. Kovács and J. Wigard are with Nokia Bell Labs, II. M EASUREMENT S ETUP AND DATA P ROCESSING
9220 Aalborg, Denmark (e-mail: istvan.kovacs@nokia-bell-labs.com;
jeroen.wigard@nokia-bell-labs.com). A measurement campaign was performed in October 2016,
Digital Object Identifier 10.1109/LWC.2017.2710045 using the setup reported in Table I. The scanner was mounted
2162-2345  c 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/
redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
AMORIM et al.: RADIO CHANNEL MODELING FOR UAV COMMUNICATION OVER CELLULAR NETWORKS 515

TABLE I
M EASUREMENTS S ETUP I NFORMATION also performed on the nearby roads around the flying paths.
During the drive test, the antenna is mounted on top of a car at
1.5 m height. With distances around 2 km from the closest BS,
the propagation path is most of the time blocked by surround-
ing trees, buildings and hills, and therefore non line-of-sight
is dominant condition in the drive test.
Each RSRP sample recorded by the scanner, Ri , recorded
from a site at a distance di in meters, was translated into a
path loss sample PLi , according to the following equation
PLi = PTx + Ga (θ, φ) − Ri [dB], (1)
where PTx represents the average transmitted power per ref-
erence symbol in the network, and Ga is the antenna gain
for the azimuthal θ and elevation φ angles measured between
the base station and the UAV. The antenna gain is calculated
through the horizontal interpolation algorithm (HPI) applied
over the horizontal and vertical antenna diagrams, obtained
from manufacturers. Example of antennas used in the networks
include: Kathrein 80010699, Kathrein 80010647V01, among
others. The calculation of elevation considers BS’s and UAV’s
heights altogether with terrain topography. At the highest flight
level, 120m, the UAV is flying above cellular base stations,
Fig. 1. UAV-Scanner Mounting used for the measurements. which are usually downtilted for optimized ground coverage.
However, distances ranges in this letter are limited to 1-22km
and elevation angles were in the range of 0.25 to 2.9 degrees.
When the geometrical elevation angle is added to antennas tilt,
the maximum angle to the main beam of base station antennas
is around 10 degrees, with more than 95% of samples below
7.5 degrees. In order to avoid the roll off region of the antenna
patterns and miscompensation of the antenna gains in Eq. (1),
samples lying outside the -6dB vertical and horizontal lobes
of the BS antenna pattern were filtered out from the analysis.
The effect of fast fading components in the measurements
are mitigated by obtaining the local mean of samples for PLi
using windows of length equal to 40λ [11], where λ represents
the radio wavelength at 800 MHz. The pair of averaged path
loss samples and distances, (PLj , dj ) were then used to obtain
a regression, in the least square sense, to fit a log-distance
Fig. 2. Site Locations and Terrain Profile for the measurement campaign in alpha-beta (AB) model, widely used in [12]:
Fyn, Denmark.
PLest (d) = α10 log10 (d) + β + Xσ [dB]. (2)
In Eq. (2), PLest represents the estimated path loss for a
underneath a commercial UAV connected to a dipole antenna, receiver located at a 3-D distance d (in meters) from the
whose gain is small and assumed negligible for the purpose transmitter; α represents the path loss exponent and β is the
of this analysis, vertically placed as depicted in Fig. 1. The intercept point with the line d = 1 m. Finally, Xσ is a random
scanner is capable of reporting radio measurements from up variable that accounts for shadowing variation modeled with
to 32 cells per recorded sample. The reports include the UAV normal distribution and standard deviation σ , assumed equal
GPS locations and reference signal received power (RSRP) to the standard deviation of the regression residuals [12].
and physical cell ID (PCI) from each detected cell. The mea- At very large distance, some path loss samples might be
surements were repeated for two different Danish operators cropped, as the received power is not high enough to over-
with independent networks and their results were combined come the noise plus interference level so that the broadcast
to produce the outcome presented in Section III. The terrain channel can be successfully decoded. The sensitivity level
profile and the location of sites in a radius of 35 km around (PLsens ), i.e., path loss value when cropping occurs, is height-
the flight zone are showed in Fig. 2. The UAV was set to fly dependent as the interference increases with the flight height
over two circular paths of 500m diameter, set 7 km apart from (it will become more evident in Section III). This cropping
each other (see Fig. 2). The UAV heights, measured from the negatively affects the path loss analysis: it causes the path
take-off spot, according to the maximum limits allowed by loss slope to be skewed downward, thus underestimating path
local regulations. On ground, a reference drive-test (DT) was loss exponent. Therefore, a threshold distance (dmax (hu)) is
516 IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 6, NO. 4, AUGUST 2017

TABLE II
M EASUREMENTS S UMMARY

affected by UAV’s transmissions. This claim is also sup-

ported by the average number of detected cells per sample
that increased from 5.1 (DT) to 16.9 (120m). This height
dependent behavior in the distance range and number of sig-
Fig. 3. Path loss vs Distance - Measurement Results and regression model nificant interference sources complies with previous results
for UAV heights hu = 1.5 and 120 m. reported in [7]. It is also worth mentioning that the measure-
ments suggest the signal power threshold increased at higher
heights in all measured routes. This is exemplified by the
applied, where we removed samples greater than this distance value of PLsens in Table II. This behavior might be attributed
to avoid the bias due to saturated samples. The choice of the to the higher interference levels, and it indicates the number
threshold distance is important, as if it is set too high, the of significant interfering sites could be even higher, as some
slope will experience the effect of cropping; if too low, a sig- might not be identified due to falling short of the required
nificant number of points will be removed from the analysis, signal-to-interference plus noise ratio (SINR) level.
and this might compromise the statistical significance of the Another finding that goes in line with the radio path clear-
regression values. ance with height regards the observed values for the shadowing
In this letter the threshold distance is selected as follows: variation. For DT measurements it is approximately 7.7, which
First Psens was defined as 99%-percentile of all measured PLj is aligned with reference values in [13] for ground level mea-
for a given height. The 99%-ile was chosen in order not to surements. As the UAV moves up this value decreases up to
make the assumed sensitivity value too low due to outliers. 3.4 dB, indicating a significant reduction in the shadowing
Then, dmax (hu ) was iteratively increased until the following variation. Part of the remaining variation might be attributed
stopping criteria is reached: to the non-omni directional pattern of the receiver antenna and
self-shadowing components.
PLest (dmax ) ≤ PLsens − σ, (3)
where PLest (dmax ) represents the estimated value for the path IV. PATH L OSS M ODELLING AND D ISCUSSION
loss at dmax , using the regression presented in eq (2), using
The results in Section III made clear the propagation
all points that satisfy dj ≤ dmax (hu ).. Assuming a Gaussian
environment is significantly different for airborne UAVs and
shadowing distribution, ≈ 15% of the samples at dmax (hu ) are
ground level users. Based on such observations, and in the
expected to be above PLest (dmax (hu )) + σ (in the cropping
work in [3], [4], [7], and [10], it is proposed here an extension
region). The expected down-bias in the path loss slope using
of the model in Eq. (2) using height-dependent parameters.
this criteria is within 0.1, and therefore, negligible for the later
Path loss exponents should decay with increases in UAV
remarks presented in this letter.
heights. In this letter, a logarithmic regression was used to
obtain a group of height-dependent parameters to be used in
III. M EASUREMENT R ESULTS eq. (2). The logarithmic function was chosen assuming height-
The results obtained through the methodology described in related radio path clearance, i.e., the path loss exponents
Section IV are presented in Fig. 3, where it is possible to see reduction, is more prominent to small increments in elevation
that there is a clear reduction in path loss exponents as hu at low heights, where there are more concentration of build-
increases, from 3.7 at ground level to 2.0 at 120 m. It results ings, vegetations and other obstacles. The height-dependent
in significant differences in the path loss attenuation, specially models are found in equations (4)–(6).
for larger distances: for 3D distances close to 10 km the signal  
attenuation is 20 dB higher on ground level compared to the α(hu ) = max pα1 + pα2 log10 (hu ), 2 , (4)
measurements at 120m. β(hu ) = pβ1 + pβ2 log10 (min(hu , hFSPL )) [dB] (5)
The summary of the results for the other flight tests can σ (hu ) = pσ1 + pσ2 log10 (min(hu , hFSPL )) [dB], (6)
be appreciated in Table II that supports the expectations of
better radio clearance at higher heights, with the path loss where hFSPL is the height where free space propagation is
exponent approaching free space propagation at higher flight assumed (α = 2.0). The values of p1 and p2 obtained based
levels [7]. In practical terms, such observation implies an on the reported measurements are exposed in Table III. Such
expected increase in the interference level observed by UAVs, parameters modelling serves as a reference for rural scenarios,
as well as a higher number of neighbor base stations being and are valid for ranges limited to 1.5 m ≤ hu ≤ 120 m and
AMORIM et al.: RADIO CHANNEL MODELING FOR UAV COMMUNICATION OVER CELLULAR NETWORKS 517

TABLE III
H EIGHT-D EPENDENT M ODEL PARAMETERS cells at higher levels. A practical consequence of these obser-
vations is an expected SINR degradation at higher elevations,
to be evaluated in future works.
In order to investigate the interference problem and evalu-
ate mechanisms to deal with it, system level simulations are
required. The main contribution of this letter is presented in
Section IV. It proposes that path loss and shadowing param-
eters for airbone UAVs connected to cellular networks must
follow height-dependent models, as a more efficient way of
performing spatial prediction, as the radio path becomes more
unobstructed with increases in height.

ACKNOWLEDGMENT
The authors would like to thank the team from DroneFyn
Denmark which assisted in this research by conducting the
UAV Flights. The authors would also like to acknowledge
the important contribution made by Rohde & Schwarz, which
provided the TSMA scanner used in the measurements.

R EFERENCES
[1] “Roadmap for the integration of civil remotely-piloted aircraft
systems into the European aviation system,” EUROCONTROL,
Brussels, Belgium, Tech. Rep., 2013. [Online]. Available:
https://publicintelligence.net/eu-rpa-roadmap/
[2] “The future of drones according to the AT&T foundry,” AT&T
Fig. 4. Height dependent models for the regression parameters of eq. (2). Foundry, Palo Alto, CA, USA, Ericsson, Stockholm, Sweden,
and Rocket Space, San Francisco, CA, USA, Tech. Rep.,
Oct. 2016. [Online]. Available: https://www.qualcomm.com/invention/
technologies/lte/advanced-pro/cellular-drone-communication
distance ranges similar to those in Table II, in a lightly hilly [3] “Consideration on the channel model for LTE-based aerial
rural environment, with base stations height between 20 and vehicles,” ZTE Microelectron., Tongji Univ., Shanghai, China,
50 meters. Tech. Rep. R1-1705163, Apr. 2017.
[4] “Evaluation scenarios and channel models for drones,” Alcatel-Lucent
Using the slope and intercept estimation given by the height Shanghai Bell Labs, Shanghai, China, Nokia, Espoo, Finland,
dependent model, and applying it to all measurements col- Tech. Rep. R1-1704430, Apr. 2017.
lected, the average offset between the measured samples and [5] “Integration of civil unmanned aircraft systems (UAS) in the national
airspace system (NAS) roadmap,” U.S. Dept. Transp., FAA, Washington,
the estimated ones is equal to −0.3 dB, and the standard devi- DC, USA, Tech. Rep. 2012-AJG-502, 2013.
ation for all heights was kept under the same values presented [6] K. Welch, “Evolving cellular technologies for safer drone operation,”
in Table II. It means the proposed height-dependent model is Qualcomm 5G White Paper Present., San Diego, CA, USA, Tech. Rep.,
Oct. 2016.
capable of providing a good model for the measured data, at [7] B. V. D. Bergh, A. Chiumento, and S. Pollin, “LTE in the sky: Trading
expense of just two optimization variables per parameter. off propagation benefits with interference costs for aerial nodes,” IEEE
A visual example of the height dependent model using the Commun. Mag., vol. 54, no. 5, pp. 44–50, May 2016.
[8] D. W. Matolak and R. Sun, “Air-ground channel characterization for
parameters in Table III, is shown in Fig. 4. The 95% confi- unmanned aircraft systems: The hilly suburban environment,” in Proc.
dence interval for the value of α estimated on the measured IEEE 80th Veh. Technol. Conf. (VTC Fall), Vancouver, BC, Canada,
data is also shown, assuming Xσ to be Gaussian distributed (in Sep. 2014, pp. 1–5.
[9] D. W. Matolak and R. Sun, “Air-ground channel characterization for
which case the estimate is Student t-distributed [14]). These unmanned aircraft systems: The near-urban environment,” in Proc.
values suggest the difference among the exponents at higher IEEE Mil. Commun. Conf. (MILCOM), Tampa, FL, USA, Oct. 2015,
levels (hu = 60 or 120 m) compared to those at lower levels pp. 1656–1660.
[10] N. Goddemeier, K. Daniel, and C. Wietfeld, “Role-based connectiv-
(hu = 1.5 or 15) is statistically significant. ity management with realistic air-to-ground channels for cooperative
UAVs,” IEEE J. Sel. Areas Commun., vol. 30, no. 5, pp. 951–963,
Jun. 2012.
V. C ONCLUSION [11] W. C. Y. Lee, “Estimate of local average power of a mobile radio signal,”
IEEE Trans. Veh. Technol., vol. VT-34, no. 1, pp. 22–27, Feb. 1985.
This letter analyzed a set of live network measurements [12] T. Rappaport, Wireless Communications: Principles and Practice
conducted with a radio network LTE scanner attached to an (Prentice Hall Communications Engineering and Emerging Technologies
Series). Upper Saddle River, NJ, USA: Prentice-Hall, 2002.
airborne UAV. Flights were conducted at heights compliant [13] V. Erceg et al., “An empirically based path loss model for wireless chan-
with current regulations, up to 120m. The results for the path nels in suburban environments,” IEEE J. Sel. Areas Commun., vol. 17,
loss exponents and the shadowing standard deviations imply no. 7, pp. 1205–1211, Jul. 1999.
[14] D. G. Altman, D. Machin, T. N. Bryant, and M. J. Gardner, Statistics
better radio clearance as the UAV moves up. This finding is With Confidence: Confidence Intervals and Statistical Guidelines.
corroborated by the increase in the average number of detected London, U.K.: BMJ Books, Feb. 2000.