2022 IEEE International Workshop on Metrology for AeroSpace, 27-29 June 2022, Pisa, Italy.

DOI: 10.1109/MetroAeroSpace54187.2022.9856394

RCS Measurements of UAVs and Their Statistical

Analysis
Massimo Rosamilia† , Augusto Aubry† , Alessio Balleri# , Vincenzo Carotenuto† , and Antonio De Maio†
† Department of Electrical Engineering and Information Technology,

University of Naples “Federico II”, Naples, Italy, I-80125

# Centre for Electronic Warfare, Information and Cyber,

Cranfield University, Defence Academy of the United Kingdom, Shrivenham, SN6 8LA

Abstract—This paper deals with Radar Cross Section (RCS)

measurements of five small Unmanned Aerial Vehicles (UAVs) in RAM
a semi-controlled environment as a function of azimuth aspect
angle, polarization, and frequency in the range 8.2-18 GHz. The V
N R Drone
experimental setup and the data pre-processing, which include Tx
coherent background subtraction, range gating procedures, and A A RAM
Rx
calibration, are presented. Then, a thorough statistical analysis M
of the measured RCSs is provided by means of the Cramér–von Turntable
Mises (CVM) distance and the Kolmogorov–Smirnov (KS) test. PC

I. I NTRODUCTION
Collecting drone data and analyzing their Radar Cross
Section (RCS) is a critical step towards the design of ap- Fig. 1. A notional representation of the experimental setup.
propriate system architectures capable of dealing with these
types of targets as well as for the development of an accurate
performance prediction of existing algorithms. 100. The raw RCS data are collected in a semi-controlled
In this regard, in [1], the RCSs of small UAVs have been environment as a function of frequency (in the interval 8.2-
measured for different aspect angles in the frequency interval 18 GHz), target azimuth aspect angle and polarization. In this
8-12 GHz and in VV polarization. The measurements have respect, a description of the experimental setup and the data
also been examined using the Inverse Synthetic Aperture pre-processing is provided. Specifically, the pre-processing
Radar (ISAR) method, which provides useful information operations, which include Coherent Background Subtraction
regarding the components that mostly contribute to the UAV (CBS) and range gating procedures, are illustrated with a
signatures. Besides, in [2], six commercial UAVs have been discussion on the RCS calibration relying on the substitution
measured at 15 GHz and 25 GHz for both HH and VV po- method. The results are analyzed considering sliding frequency
larizations. The RCS measurement of two off-the-shelf drones intervals of 200 MHz corresponding to a range resolution
in the frequency band 5.8-8.2 GHz has been addressed in [3], of 0.75 m, which allows to model the drones as point-like
and in [4] RCS data of several drones have been collected targets. Finally, a detailed first-order statistical analysis of
in the frequency range 26-40 GHz. Some measurements in the measured RCSs is performed by fitting the data with
the Ku radar band have been conducted in [5], whereas, (one- and two- parameters) distributions typically employed
unlike aforementioned references, [6] has presented three- to model RCS fluctuations [23], via the minimization of the
dimensional RCS measurements of a nano-drone from 23 GHz Cramér–von Mises (CVM) distance between the empirical and
to 25 GHz. A highly accurate UAV RCS simulation has been the theoretical Cumulative Distribution Functions (CDFs). The
developed in [7] and the corresponding results compared with Kolmogorov–Smirnov (KS) test is also employed to further
those of measurements and simpler simulation approaches. assess the goodness-of-fit of the selected distribution.
In [8] the RCSs of some nano and micro drones have been The rest of the paper is organized as follows. Section II
collected in the X-band for several elevation angles, and some presents the experimental setup. In Section III, the statistical
statistics related to measured RCS data have been provided. behavior of the measured RCSs is analyzed. Finally, conclu-
To the best of the authors’ knowledge, in the open literature, sions are drawn in Section IV.
the statistical analysis of the RCS of commercial drones
in a frequency range including the X and Ku radar bands II. E XPERIMENTAL S ETUP
has not been yet addressed. Aimed at filling this gap, this
paper presents the statistical analysis of the RCS signatures In this section, a description of the experimental setup
of five drones: an AscTec Firefly, an AscTec Pelican, a involved in the measurement campaign is provided along with
Venom VN10, a Parrot AR.DRONE, and a DJI Matrice details on data pre-processing and calibration. The RCS data

© 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media,
including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists,
or reuse of any copyrighted component of this work in other works
 TABLE I coherently in the frequency domain from all the data acquired
S ETUP AND ACQUISITION PARAMETERS . in the presence of the target. Range-gating has been then
applied to the high range resolution background-free profile
Parameter Setup 1 Setup 2
Analyzed Frequency Bandwidth 8.2-12.4 GHz 12.4-18 GHz to further isolate the target response in range from residual
Azimuth Rotation Step 0.1 degrees 0.1 degrees multipath reflections which could not be eliminated with the
Target-antennas Distance ≈ 7.2 m ≈ 3.4 m CBS [10], [11]. To this end, a tailored rectangular window,
Distance from Ceiling 2.71 m 2.71 m with parameters matched to the drone size and the target-
Height above Floor 1.28 m 1.28 m antennas distance (see Table I), has been used. The frequency
Range Gating 6.5-7.8 m 1.5-4.6 m
Number of FFT/IFFT Points 400100 400100 spectrum of the clean signatures has been then used to extract
the point-like target response over a moving bandwidth of 200
MHz, corresponding to a range resolution of 0.75 m. Precisely,
has been collected in a laboratory environment using the the frequency bandwidth is discretized in several overlap-
measurement setup depicted in Fig. 1, which is composed of ping frequency sub-bandwidths of 200 MHz having central
• Radar Absorbing Material (RAM) panels to mitigate frequencies {8.3 GHz + (i × 100) MHz, i = 0, . . . , 96} and
multipath reflections from the ceiling, walls and floor; the data are processed separately in each of them. Therein,
• a 2-port MS46322A Anritsu Vector Network Analyzer the target can be approximated as a point-like reflector (i.e.,
(VNA), which measures the frequency response of the target scatterers within the range resolution cell) whose power
illuminated area over a pre-defined bandwidth; response (i.e., non-calibrated RCS) is extracted as the squared
• a LinearX precision turntable with an angular step res- magnitude peak in the time domain. However, in the process
olution of (up to) 0.1 degrees, fully controlled remotely of measuring the absolute RCS of a particular target, it is
from a PC; essential to include an accurate RCS calibration step. The
• a standard PC to control and synchronize the turntable substitution method [10] is the most often used calibration
and the VNA via the Laboratory Virtual Instrument procedure for RCS measurements, which involves measuring
Engineering Workbench (LabVIEW) as well as to store a calibrating target (with a known RCS) with the same test-
and process the raw data; bed used to collect data from the target under test [10]. As a
• a pair of identical standard horn antennas, one for trans- result, each measurement related to the RCS of the calibrating
mission and the other for reception; they are connected to target at a given frequency is compared with the theoretical
the two ports of the VNA by means of low-loss coaxial RCS, and the resulting dB difference is utilized to calibrate the
cables and co-located on a tripod. The positions of the an- target measurements, provided that the test-bed, as well as the
tennas have been adjusted with a cross-laser level to steer system parameters, are stationary. In the performed campaign,
the antenna boresights at the target. Rotating the antennas a conductive 10 cm diameter sphere has been used to calibrate
allowed data collection for different polarizations. the drone measurement data.
Before proceeding with the measurements, the VNA has been
calibrated using the standard “thru” calibration procedure to
III. D RONE RCS S TATISTICAL B EHAVIOR
provide a measurement setup with a flat frequency response
up to the antennas. The LabVIEW scripts have been designed
to trigger a turntable step rotation after the data acquisition The classic approach for evaluating radar detection per-
by the VNA at a specific aspect angle. This has guaranteed formance is based on the assumption that the target’s RCS
collections of frequency responses with a stationary target. In fluctuation follows one of the Swerling models I-V [12].
particular, a short time delay has been also included before However, as confirmed by some practical cases, amplitude
collecting a new measurement to ensure that the setup was fluctuations do not always comply with the aforementioned
actually stationary after each step rotation. models, resulting in a mismatch between the actual and
Although the VNA measured all S-parameters at each the theoretical radar performance. Indeed, several alternative
frequency [9] of interest, for the considered experiments only fluctuation models (e.g., Weibull, Log-normal, shadowed Rice,
S21 has been recorded and analyzed. The HH-pol and VV- two-state Rayleigh-chi, just to mention a few) have been
pol returns from five drones have been measured versus proposed in the open literature to cope with this problem [12]–
frequency and azimuth aspect angle in the interval 8.2-18 [14]. Using a suitable statistical description for the target RCS
GHz. A summary of the experimental parameters used to behavior enables the accurate prediction of radar detection per-
collect and analyze the data is reported in Table I, whereas formance as well as the design of appropriate signal processing
the specifications of the drones are listed in Table II. Note architectures. Toward this goal, in this section, the measured
that the employed measurement setup falls in the so-called RCS signatures of several drones are statistically analyzed
near-field non-anechoic range scenario [10]. by fitting the data with well-known and commonly used
distributions (at most bi-parametric), over different frequencies
A. Data Pre-processing and Calibration and polarizations. Then, the most appropriate statistical model
For each acquisition, a background measurement (obtained for each drone RCS collection (in the aspect angle domain) is
in the absence of the drone) has been collected and subtracted selected resorting to the CVM distance and the KS test.
 TABLE II
M EASURED D RONES S PECIFICATIONS .

Drone # Rotors Weight Width Depth Height Primary Use

AscTec Firefly 6 1600 g 470 mm 430 mm 165 mm Mapping/Surveying
AscTec Pelican 4 1650 g 360 mm 360 mm 188 mm Film & Photo/Mapping/Surveying
Venom VN10 4 148 g 290 mm 210 mm 38 mm Film & Photo
Parrot AR.DRONE 2.0 4 420 g 517 mm 517 mm 127 mm Film & Photo
DJI Matrice 100 4 2355 g 759 mm 755 mm 205 mm Film & Photo/Mapping/Surveying

The RCS analysis of the AscTec Firefly is discussed in

Subsection III-A, then the other drones are studied in Sub-
section III-B.

A. Statistical Analysis of AscTec Firefly RCS

A detailed analysis of the AscTec Firefly’s RCS is provided
in this subsection. Before proceeding further, it is worth
mentioning that the worst case noise power (averaged over
the two polarizations) observed in the measurements, after pre-
processing, subband analysis, and calibration, is NF ≈ −34.5
dB for the frequency band 8.2-12.4 GHz, and NF ≈ −44.5 Fig. 2. Mean values of AscTec Firefly’s RCS versus frequency for HH and VV
dB for the interval 12.4-18 GHz. Given the noise level NF , an polarizations (solid and dashed lines, respectively). The top (bottom) border
of the shaded area represents the mean value plus (minus) standard deviation.
estimate of the measurements SNR in a given frequency bin,
which provides insights on the measurement accuracy [15], is 90°
given by the ratio between the target RCS and NF , i.e., it can 120° 60°
be practically computed as SN R(f, p) = σ̂(f, p)/NF , with -10
σ̂(f, p) the measured RCS of the target at frequency f and 150° 30°
-20
polarization p. For the case at hand, apart for a few outliers,
the SNR is in the order of 20 dB. -30

As to the RCS, Fig. 2 displays both the mean and standard -40 9.5 GHz
180° 0°
10.7 GHz
deviation values (with respect to aspect angle) versus fre-
quency, with reference to both HH and VV RCS acquisitions.
Remarkably, similar average RCS values are obtained in
210° 330°
both the HH and VV polarizations, in agreement with [1].
The plots also reveal that, for a given frequency bin, the
standard deviation of the measured RCS is about 15 dB, which 240° 300°
270°
might be attributed to the presence of a few major scatterers
(a)
whose interaction significantly changes with aspect angles.
This behavior is supported by Fig. 3, which illustrates the 90°
120° 60°
RCS of the drone in polar coordinates (at varying azimuth
-10
angle) for HH polarization and different central frequencies,
i.e., 9.5 GHz, 10.7 GHz, 14.1 GHz, and 15.9 GHz. The figure 150° -20 30°

highlights that the RCS is characterized by a fast fluctuation -30

in angle. As already said, this behavior is mainly determined
-40
by the composition of the dominant scatterers returns. 180° 0°
14.1 GHz
15.9 GHz
Let us now focus on the first-order statistical analysis. Since
the drone RCS strongly changes with aspect angle, it appears
reasonable the exploitation of a statistical model to describe 210° 330°
the target fluctuation and accurately predict radar detection
performance. Inspired by previous studies on target RCS 240° 300°
fluctuation statistics, in this paper the distributions (at most 270°
bi-parametric) reported in Table III are considered to model (b)
the RCS of the AscTec Firefly. In the table, Γ(x), γ(a, b), and Fig. 3. Polar plot of AscTec Firefly RCS [dBsqm] for HH polarization in
erf(x) indicate the Gamma, the incomplete Gamma, and the the frequency bin with central frequency: (a) 9.5 GHz (blue curve) and 10.7
error functions, respectively. GHz (red curve), (b) 14.1 GHz (blue curve) and 15.9 GHz (red curve).
 TABLE III
CDF S OF THE CONSIDERED MODELS .

Distribution CDF Parameter 1 Parameter 2 Mean

Exponential F (x; λ) = 1 − e−λx, x ≥ 0 λ > 0, rate - µ = 1/λ
Gamma F (x; α, β) = Γ(α) γ α, βx , x ≥ 0
1
α > 0, shape β > 0, scale µ = αβ
σ2
h  i
LogNormal F (x; χ, σ) = 12 1 + erf lnσx−χ
√
2
,x > 0 χ ∈ (−∞, +∞) σ>0 µ = eχ+ 2

b
Weibull F (x; a, b) = 1 − e−(x/a) , x ≥ 0 a ∈ (0, +∞), scale b ∈ (0, +∞), shape µ = aΓ(1 + 1/b)

The fitting of the above-mentioned distributions with the with F̂ (σ̂i (f, p)) the empirical CDF at a given frequency and
data is performed considering the RCS measurements for polarization of the measured RCS values (for different aspect
different aspect angles at a given frequency f and in a angles) collected in the vector σ̂(f, p), whereas γ(αKS ) is
polarization p = {HH, V V }. Formally, the parameter vector the decision threshold (which does not depend on the tested
of the distributions is determined as a solution to the following distribution) set to ensure the desired significance level αKS .
optimization problem Besides, the implementation of (4) is tantamount to comparing
b p) = arg min CV M (σ̂(f, p), F (x; θ))
θ(f, (1) the p-value of Dn (f, p), under the null hypothesis, with
θ αKS [19]. Considering a significance level αKS = 0.01, the
where F (x; θ) is the CDF of the distribution under outcomes of the KS tests unveils that, regardless of frequency
test, θ denotes the distributional parameters, σ̂(f, p) = and polarization, the H0 hypothesis cannot be rejected.
[σ̂1 (f, p), σ̂2 (f, p), . . . , σ̂n (f, p)]T ∈ Rn , with n = 3600, is the Finally, Fig. 4 concludes this subsection with the analysis
vector of the measured/observed RCS, and [16] of the Gamma parameters obtained with the fitting procedure.
CV M (σ̂(f, p), F (x; θ)) = In particular, Fig. 4(a) compares the sample mean of the
v measured RCS with the first moment of the fitted Gamma
u n  2
u 1 X 2i − 1 (2) distribution. The results show that Gamma expectation is very
t + − F (σ̂i (f, p); θ)
12n i=1 2n close to the mean RCS σ̃p , p = {HH, V V } of the drone. The
shape parameter of the optimally fitted Gamma is examined
is the CVM distance. The optimization problem in (1) is tack- in Fig. 4(b) versus the frequency. A close examination of the
led by means of the iterative algorithm proposed in [17] which figure reveals that the shape parameter is relatively close to
is implemented in MATLAB with the function fminsearch, 1, indicating that the RCSs can be well represented using a
using as initial estimates of the distributional parameters statistical distribution close to an Exponential (Swerling I-II)
those obtained via the MATLAB function fitdist. The average model. However, based on the results shown in Table IV, a
CVM distances between the optimally fitted (according to (1)) plain Exponential distribution is unable to fully predict the be-
distributions and the empirical CDF for different frequencies havior of the measured data, and a bi-parametric distribution,
are reported in the first column of Table IV. Inspection of i.e., the Gamma, appears necessary to better capture the data
the table reveals that the Gamma CDF in general achieves statistical properties.
the lowest average CVM distance from the measured data,
whereas the Weibull distribution ranks second. Hence, to B. Statistical Analysis of Others Drones RCS
confirm the quality of the Gamma model to faithfully describe In this subsection, the statistical analysis is conducted on the
the collected data, a KS test is performed [16], [18]. This collected RCSs of the other tested drones (see Table II). Fig. 5
is a non parametric statistical procedure which can be used shows the mean and standard deviation values of the RCS
to assess the goodness-of-fit between the empirical and the versus frequency for HH and VV polarizations. Like the results
theoretical RCS distributions, i.e., it tests the simple hypothesis obtained for the AscTec Firefly (see Fig. 2), the considered
b p)), p = {HH, V V }. drones achieve similar RCS values in both HH and VV
H0 : σ̂i (f, p) has the CDF F (x; θ(f,
polarizations. Moreover, large fluctuations in the RCS values
(3)
can be observed, with a standard deviation in the order of 10
In a nutshell, the KS test tackles the hypothesis testing
dB for the AscTec Pelican and DJI Matrice 100 (Figs. 5(a)
problem (3) by comparing a threshold with the largest absolute
and 5(d)) and 20 dB for the Venom VN10 and the Parrot
difference between the empirical CDF of the data and the
AR.DRONE 2.0 (Figs. 5(b) and 5(c)). As to the statistical
theoretical one. Formally,
analysis, Table IV reports the mean values (over the frequency)
H1
Dn (f, p) ≷ γ(αKS ) (4) of the CVM distances computed between the empirical and
H0 theoretical CDFs of the measured RCSs in both HH and VV
where polarizations. Interestingly, the Gamma model is still able to
b p)) achieve the lowest average CVM distance in almost all the
Dn (f, p) = sup F̂ (σ̂i (f, p)) − F (σ̂i (f, p); θ(f,
σ̂i (f,p)∈σ̂(f,p) scenarios, with some exceptions where Weibull distribution
(5) prevails over the others (see for instance the case of Asctec
 0.08 0.08
been performed checking the adequacy of some distribu-
0.07 0.07
tions (at most bi-parametric) to describe the first-order RCS
0.06 0.06
statistics. The results have highlighted that, in the considered
RCS [m2 ]

0.05 0.05
frequency bands, the RCSs of the drones assume quite small
0.04 0.04
values and are characterized by strong fluctuations in angle.
0.03 0.03
Besides, from a statistical standpoint, the Gamma distribution
0.02 0.02
proved capable of modeling such measurement variability,
0.01 0.01
8 9 10 11 12 13 14 15 16 17 18 characterized, in the majority of cases, by Exponential-like
Frequency [GHz]

(a)
fluctuations. Precisely, the RCS variability can be usually
described using Gamma shape parameter values close to 1.
1.4
Therefore, the developed analysis can provide useful insights
1.3 toward an accurate prediction of the radar detection perfor-
1.2 mance for the considered targets.
Shape

1.1
R EFERENCES
1
[1] M. Pieraccini, L. Miccinesi, and N. Rojhani, “RCS measurements and
0.9 ISAR images of small UAVs,” IEEE Aerosp. Electron. Syst. Magazine,
vol. 32, no. 9, pp. 28–32, 2017.
0.8
8 9 10 11 12 13 14 15 16 17 18 [2] M. Ezuma, C. K. Anjinappa, M. Funderburk, and I. Guvenc, “Radar
Frequency [GHz] cross section based statistical recognition of UAVs at microwave fre-
(b) quencies,” IEEE Trans. Aerosp. Electron. Syst., vol. 58, no. 1, pp. 27–46,
2022.
Fig. 4. Analysis of the Gamma distribution’s parameters versus frequency, [3] A. Herschfelt, C. R. Birtcher, R. M. Gutierrez, Y. Rong, H. Yu, C. A.
fitted with the RCS measurements of AscTec Firefly. (a) Comparison between Balanis, and D. W. Bliss, “Consumer-grade drone radar cross-section
the mean RCS values and the expectation of the fitted Gamma distribution and micro-doppler phenomenology,” in 2017 IEEE Radar Conference
versus frequency for HH and VV polarization; (b) Gamma shape parameter (RadarConf), 2017, pp. 0981–0985.
values versus frequency. [4] V. Semkin, J. Haarla, T. Pairon, C. Slezak, S. Rangan, V. Viikari, and
C. Oestges, “Analyzing radar cross section signatures of diverse drone
models at mmWave frequencies,” IEEE Access, vol. 8, pp. 48958–48969,
2020.
Pelican in HH or Venom VN10 in VV). However, under these [5] J. Ochodnický, Z. Matousek, M. Babjak, and J. Kurty, “Drone detection
specific instances, the mean CVM distances achieved by the by Ku-band battlefield radar,” in 2017 International Conference on
Gamma and the Weibull model are relatively close. Moreover, Military Technologies (ICMT), 2017, pp. 613–616.
[6] A. Balleri, “Measurements of the radar cross section of a nano-drone at
unlike the Weibull, the Gamma fluctuation law enables a quite K-band,” in 2021 IEEE 8th International Workshop on Metrology for
simple and closed-form analytical evaluation of the detection AeroSpace (MetroAeroSpace), 2021, pp. 283–287.
performance [14]. [7] P. J. Speirs, A. Murk, M. Renker, P. Wellig, and U. Aulenbacher,
“High-detail simulations of consumer-grade UAV RCS signatures, and
Fig. 6 shows the shape parameter values of the fitted comparisons against measurements,” STO-MP-MSG-SET-183.
Gamma versus frequency for both polarizations. For most [8] P. Sedivy and O. Nemec, “Drone RCS statistical behaviour,” STO-MP-
cases, the Gamma shape parameter is close to 1, underlining MSG-SET-183, 2021.
[9] Calibration and Measurement Guide, ShockLine MS46122A/B,
that the measured RCS first-order statistics are not far from an MS46131A, ME7868A, and MS46322A/B Series Vector Network Ana-
Exponential-like behavior. The only exception which is worth lyzer, Number 10410-00336. Anritsu Company, 2021.
of a further investigation is the AscTec Pelican. As a matter [10] IEEE, “Recommended practice for radar cross-section test procedures,”
IEEE Std 1502-2020 (Revision of IEEE Std 1502-2007), pp. 1–78, 2020.
of fact, Fig. 6(a) reveals that for the frequency band 10.2- [11] M. F. Sundermeier and D. Fischer, “Compact radar cross-section
10.8 GHz and VV polarization, the Gamma shape parameter measurement setup and performance evaluation,” Advances in Radio
is close to 2, meaning that the fluctuation follows a chi-squared Science, vol. 19, pp. 147–152, 2021.
[12] M. A. Richards, J. A. Scheer, and W. A. Holm, Principles of Modern
distribution with 4 degrees of freedom. Remarkably, this latter Radar: Basic Principles, Volume 1, Institution of Engineering and
distribution (used in the Swerling 3 and 4 models) is typically Technology, 2010.
employed to model targets composed of scatterers of similar [13] D. A. Shnidman, “Expanded swerling target models,” IEEE Trans.
Aerosp. Electron. Syst., vol. 39, no. 3, pp. 1059–1069, 2003.
strength plus
√ one dominant scatterer, with the latter having [14] A. De Maio, A. Farina, and G. Foglia, “Target fluctuation models and
RCS 1 + 2σo , where σo is the sum of RCS of the randomly their application to radar performance prediction,” IEE Proceedings-
distributed equal-strength scatterers [12]. Radar, Sonar and Navigation, vol. 151, no. 5, pp. 261–269, 2004.
[15] “IEEE recommended practice for radar cross-section test procedures,”
IEEE Std 1502-2020 (Revision of IEEE Std 1502-2007), pp. 1–78, 2020.
IV. C ONCLUSION [16] R. D’Agostino and M. Stephens, Goodness-of-Fit Techniques, New
York: Marcel Dekker, 1986.
This paper has considered the statistical analysis of the RCS [17] J. A. Nelder and R. Mead, “A simplex method for function minimiza-
of small UAVs collected in a semi-controlled environment as tion,” The computer journal, vol. 7, no. 4, pp. 308–313, 1965.
a function of frequency, angle, and polarization. Specifically, [18] W. W. Daniel, Applied Nonparametric Statistics, Duxbury advanced
series in statistics and decision sciences. PWS-KENT Pub., 1990.
five drones of different sizes and characteristics have been [19] R. Simard and P. L’Ecuyer, “Computing the two-sided kolmogorov-
measured in the frequency range 8.2-18 GHz, and their RCSs smirnov distribution,” Journal of Statistical Software, vol. 39, pp. 1–18,
have been analyzed considering first order statistics over a 2011.
moving bandwidth of 200 MHz. The statistical analysis has
 TABLE IV
M EAN CVM DISTANCES BETWEEN EMPIRICAL AND THEORETICAL CDF.

mean CVM distance HH / VV

Distribution AscTec Firefly AscTec Pelican Venom VN10 Parrot AR.DRONE 2.0 DJI Matrice 100
Exponential 0.93 / 0.73 1.21 / 1.43 1.17 / 1.08 0.93 / 0.99 0.85 / 0.77
Gamma 0.59 / 0.49 0.73 / 0.66 0.62 / 0.69 0.66 / 0.57 0.62 / 0.55
LogNormal 1.11 / 1.37 1.09 / 1.15 1.26 / 1.32 1.06 / 1.21 0.98 / 1.04
Weibull 0.63 / 0.50 0.67 / 0.68 0.65 / 0.66 0.67 / 0.58 0.64 / 0.58

(a) (b)

(c) (d)
Fig. 5. RCS mean and standard deviation values versus frequency for a) AscTec Pelican, b) Venom VN10, c) Parrot AR.DRONE 2.0, d) DJI Matrice 100.

2.5 1.6

2 1.4
Shape

Shape

1.5 1.2

1 1

0.5 0.8
8 9 10 11 12 13 14 15 16 17 18 8 9 10 11 12 13 14 15 16 17 18
Frequency [GHz] Frequency [GHz]

(a) (b)
1.4 1.4

1.3 1.3

1.2
1.2
1.1
Shape

Shape

1.1
1
1
0.9

0.8 0.9

0.7 0.8
8 9 10 11 12 13 14 15 16 17 18 8 9 10 11 12 13 14 15 16 17 18
Frequency [GHz] Frequency [GHz]

(c) (d)
Fig. 6. Shape parameter values of the Gamma distribution fitted with the RCS data of a) AscTec Pelican, b) Venom VN10, c) Parrot AR.DRONE 2.0, d)
DJI Matrice 100, versus frequency in HH and VV polarization.
 Cranfield University
CERES Research Repository https://dspace.lib.cranfield.ac.uk/
Cranfield Defence and Security Staff publications (CDS)

RCS measurements of UAVs and their

statistical analysis

Rosamilia, Massimo
2022-06-29
Attribution-NonCommercial 4.0 International

Rosamilia M, Aubry A, Balleri A, et al., (2022) RCS measurements of UAVs and their statistical
analysis. In: 2022 IEEE International Workshop on Metrology for AeroSpace, 27-29 June 2022,
Pisa, Italy. DOI: 10.1109/MetroAeroSpace54187.2022.9856394
http://www.metroaerospace.org/index-metroaerospace
Downloaded from CERES Research Repository, Cranfield University