Modelling Micro-Doppler Signature of Drone

Propellers in Distributed ISAC

Heraldo Cesar Alves Costa∗ , Saw James Myint∗ , Carsten Andrich∗† , Sebastian W. Giehl∗ ,
Christian Schneider∗† , Reiner S. Thomä∗
∗ Technische Universität Ilmenau, Institute for Information Technology, Ilmenau, Germany
† Fraunhofer Institute for Integrated Circuits IIS, Ilmenau, Germany
[email protected]

Abstract—Integrated Sensing and Communication (ISAC) can be critical for accurate target tracking and classification,
arXiv:2401.14287v1 [eess.SP] 25 Jan 2024

comprises detection and analysis of non-cooperative targets by ex- especially in complex environments where multiple targets
ploiting the resources of the mobile radio system. In this context, may be present. Therefore, the development of advanced signal
micro-Doppler is of great importance for target classification, in
order to distinguish objects with local movements. For developing processing techniques for bistatic micro-Doppler data is of
algorithms for target classification, it is necessary to have a large great interest to the radar community. Nevertheless, despite
amount of target signatures. Aiming to generate these data, this the importance of bistatic micro-Doppler measurements in
paper proposes a mathematical model for the micro-Doppler of ISAC systems, there is currently a limited amount of data and
drone rotating propellers, and validate the proposed model by research on this topic compared to monostatic micro-Doppler.
comparing it to measured micro-Doppler. Results show that the
proposed mathematical model can generate micro-Doppler data In order to fill this gap and develop adequate techniques,
very similar to those from measurement data. it is important to have access to a large amount of target
Index Terms—ISAC, BiRa, dynamic reflectivity, drone, micro- data. Bistatic Radar measurement facility (BiRa), installed at
Doppler, bistatic, OFDM. the Ilmenau University of Technology, can play an interest-
ing role in the way to solve this problem, since it is able
I. I NTRODUCTION to automatically perform multiple multiaspect micro-Doppler
Integrated Sensing and Communication (ISAC) is consid- measurements along a defined set of bistatic geometry posi-
ered to be one of the key features of future 6G mobile radio. tions, with different parameters, and without user intervention
ISAC stands for the detection and analysis of non-cooperative [1]. This allows collecting numerous measurement data in a
objects (“targets”) by exploiting the inherent resources of the single unmanned measurement run.
mobile radio system on both the radio access and network However, training some classification algorithms such as
level, using the signals transmitted simultaneously for both machine learning and deep learning layouts requires a huge
communication and target illumination purposes. Therefore, amount of training data. In this scenario, a simple model is of
ISAC performs target detection in configurations typical for great help for generating large amounts of micro-Doppler data,
communication systems, which comprise the waveform — since a model can be more general than measurements. It is
usually Orthogonal Frequency Division Multiplexing (OFDM) able to generate data for all distances, which is not always fea-
and derivatives —, its numerology, multiuser access (OFDMA, sible with measurements. Moreover, a model can be scalable,
TDMA), pilot schemes, multi-link (multistatic) propagation, in effort and accuracy, in order to avoid overcomplexity.
channel state estimation, and synchronization [1]. The present work proposes a new OFDM bistatic micro-
One important aspect of multi-link propagation characteri- Doppler mathematical model for drone rotating propellers,
zation is bistatic target dynamic reflectivity and, in this context, developed based on the classic model for monostatic radars
micro-Doppler, which refers to the small fluctuations in the presented in [3], intended to become an interesting tool for
Doppler shift caused by the motion of a target’s internal parts, future development of ISAC target classification algorithms.
such as rotating blades or moving limbs. Bistatic case is the Section II introduces the mathematical formulation of the
native geometry in mobile radio, and micro-Doppler provides proposed model. Section III presents the BiRa measurement
valuable information about the target’s moving components, system and the measurement setup used in the current paper.
which can be used to identify the type of target and distinguish Section IV validates micro-Doppler signatures generated with
it from other objects. Due to these reasons, micro-Doppler is the proposed model by comparing them to those obtained from
a phenomenon that is becoming increasingly important in the the measurements. The conclusion and outlook can be found
context of ISAC systems. in Section V.
Bistatic micro-Doppler measurement and analysis are im-
portant tools for characterizing targets, since bistatic geometry II. P ROPELLER M ICRO -D OPPLER M ODEL
provides diversity advantage over monostatic case, giving
additional information about the target’s motion and structure A. Classic Propeller Micro-Doppler Model
[2]. Correct interpretation of bistatic micro-Doppler signatures Suppose that a radar transmits a signal described as:
 x(t) = γ0 exp(jω0 t) (1)
where γ0 is the amplitude of the transmitted signal, ω0 =
2πf0 = 2π λc0 , f0 is the transmitter central frequency, c is the
light speed, and λ0 is the wavelength corresponding to f0 .
(m)
If a scatter point target P has a range RP (t) with respect
to a monostatic radar, its returns to this radar can be expressed
as [4]:
" #
 (m)
2RP (t)
yP (t) = γP (t) exp jω0 t− + z(t) (2)
c

where z(t) is the noise, and γP (t) is the amplitude of the

returned signal, given by [5]
r
cσP
γP (t) = γ0 (3)
(4π 3 )RP (t)4 f02 Fig. 1. Bistatic rotating point geometry

with σP being the RCS of the scatter point P .

Since the wavelengths involved in the proposed ISAC sys- low-resolution CW waveform. This model can be used with
tems are smaller than the dimensions of most drone structures, good level of success for cases with both low-resolution and
except for some micro and nano drones which are out of the narrowband.
scope of this study, the electromagnetic scattering of these
B. Bistatic Geometry
targets behave similarly to what happens in the optical area.
As the interaction among different scattering centers is very In order to be able to derive a similar model for the
weak, the scattering can be assumed to be a local linear broadband as well as bistatic case, it is first necessary to have
process, i.e., the target returns in a given range resolution a geometrical model of the bistatic range of a single rotating
cell is the superposition (coherent sum) of each independent point.
scattering center in that resolution cell at that instant [3]. Such a math model was presented in [8], where the author
In this case, the target’s total scattering echo is given by develops the mathematical formulation of the bistatic range
the coherent sum of all independent scattering centers at that of a scatter point P, in a bistatic radar environment, rotating
instant. Therefore, considering a blade i as a line of infinite around a point O, according to Figure 1. This bistatic range
point scatters going from the center O until a point Ptip , is given by
distant l = LB from the center O, [6] derives a mathematical (b)
model for the monostatic slow-time returns of a rotor with NB RP (t) ≈ RO − AB l cos(ωt + φB ) (6)
rotating blades of length LB each: where RO = RT +RR is the total bistatic range of the rotation
NB 
ω0
center O, and
γ̃i (t)e[j c (−2RO (t)+LB cos(ωt+φB (i)) cos(β))] ·
X
h̃i (t) = p
i=1
  AB = 4 cos(β/2)2 − (cos βT + cos βR )2 (7)
ω0 LB 
sin βR sin(φT − φR )

· L sinc cos(ωt + φB (i)) cos(β) + z̃(t) φB = φT − arctan (8)
c sin βT + sin βR cos(φT − φR )
(4)
Based on (6), and using the same method as in subsection
where β is the elevation angle from the radar to the rotation II-A for the mono-static case, it is now possible to derive a
center with respect to the rotation plane, and model for bistatic returns of a rotor with NB rotating blades
i−1 of length LB each.
φB (i) = φ0 + 2π (5)
NB
C. OFDM Waveform
(m)
It is important to note that changes in RP (t) generate The model from equation (1) assumes a simple CW wave-
changes in the phase of the signal. If the target is moving, form with center frequency f0 , where ω0 = 2πf0 . This
(m)
RP (t) corresponding to each symbol echo is changing. As formulation can be considered a good approximation for radars
a result of this, the phase term becomes a changing function with narrow bandwidth. However, for broadband systems this
(m)
on RP (t), and the Doppler frequency is given by the derivate formulation might not properly reproduce the signal features,
of this phase with respect to the slow-time [7]. since the same target scatter point can produce different
Therefore, this model represents the monostatic slow-time Doppler contributions due to the different frequency compo-
response of a rotor with Nb blades, when our radar/sensor uses nents of the transmitted signal.
 Once again, considering a blade i as a line of infinite point
scatters, each of its returns can be calculated, by using

N
(
Z LB X 
(b)
yi (t, τ ) = γn D(n, m) exp jωn ·
0 n=1
 (b)
)
RP (t)
!
(b)
RP (t) τ− c 1
· t− rect  −  dl + z(t)
c T 2
(13)
If l1 and l2 correspond to the limits of the blade length
µ
comprised by the pulse length in the delay τ = N T , then
Fig. 2. BiRa measurement system: the two gantries can scan the object at (
the center of a turntable more than the dimension of the hemisphere for both N
(b)
X
incident and scattering. yi (t, τ ) = γn D(n, m) exp (jωn t) ·
n=1
 ) (14)
Z l2 
In ISAC, it is common to use OFDM waveforms and, for ωn (b)
· exp − j (RP (t)) dl + z(t)
some applications, these waveforms can occupy a considerably l1 c
broad bandwidth. This case deserves a special attention and
The returns of all NB blades, with φB given, similarly to
will be analyzed in the next subsection.
(5), by φB (i) = φB + 2π i−1
NB , can be summed with
D. Bistatic OFDM Mathematical Formulation ( )
NB
The mth transmitted symbol of a system with N OFDM (b)
X
(b)
y (t, τ ) = yi (t, τ ) (15)
subcarriers can be written as i=1
N
X  µ  Solving the integral, calculating its limits, and eliminating
x(µ, m) = D(n, m) exp jωn (m + )T (9)
N the carrier frequency component, the proposed model is con-
n=1
cluded by describing the baseband signal as
where T is the time duration of the OFDM symbol, and µ (
N
and m are discrete fast-time and slow-time indexes, respec- (b)
X  nµ 
µ
tively, so that the total discrete time t = mT + N T. y (µ, m) = D(n, m) exp j2π ·
n=1
N
Additionally, D(n, m) is the complex amplitude for each
NB 
subcarrier, also called complex modulation symbol, given by X ωn
· γni (m)ej c (−RO (t)+∆R+ (µ,i)) · (16)
the modulation technique, e.g. phase-shift keying (PSK).
i=1
Furthermore, in order to have orthogonality, ωn is given by )
n l2 (µ) − l1 (µ) ω
n

ωn = 2πfn = 2π(f0 + ) (10) · sinc ∆R− (µ, i) + z(t)
T 2 c
with fn denoting the central frequency of each subcarrier. where
So, as exp(j2πnm) = 1, and using t in the expression as l2 (µ) + l1 (µ)
a matter of simplicity, (9) can be rewritten as ∆R+ (µ, i) = AB cos(ωt + φB (i)) (17)
2
N
X nµ
x(µ, m) = exp(jω0 t) D(n, m) exp(j2π ) (11)
N l2 (µ) − l1 (µ)
n=1 ∆R− (µ, i) = AB cos(ωt + φB (i)) (18)
2
Then, using the same logic as in [7], the HRR baseband
bistatic returns from a scatter point P , with respect to the with l1 (µ) and l2 (µ) being the medians, respectively, of the
mth symbol, can be written as following sets:
N
( !
 (b) µ µ
(b)
X RP (t) 
RO (t) − ( N − 1)cT RO (t) − N cT

yP (t, τ ) = γn D(n, m) exp jωn t − . 0, , (19)
n=1
c AB cos(ωt + φB (i)) AB cos(ωt + φB (i))
 (b)
)
RP (t)
τ− c 1 µ µ
RO (t) − ( N − 1)cT RO (t) − N
 
.rect  −  + z(t) cT
T 2 LB , , (20)
AB cos(ωt + φB (i)) AB cos(ωt + φB (i))
(12) Micro-Doppler signatures, generated by the proposed
Now, if P is a rotating scatter point, the bistatic range model, are presented and compared to measurement in Section
(b)
RP (t) is given by (6). IV.
 TABLE I
MICRO-DOPPLER MEASUREMENT SETUP

Waveform OFDM-based
System Central frequency 3.7 GHz
Bandwidth 200 MHz
Total number of carriers 1600
OFDM
Carriers with energy 1280
Settings
Symbol duration 8µs
Number of rotating propellers 1
Blades per propeller 2
Drone Propeller radius 16.55 cm
Rotation speed 1500 rpm
Propeller material carbon fiber Fig. 4. Micro-Doppler measurement in BiRa.

III. B I R A M EASUREMENTS broadband receiver, allows the dynamic choice of complex

A. BiRa Measurement System and Drones software-generated waveforms and live signal processing, be-
ing very effective for micro-Doppler measurements.
In order to validate the model from Section II, it is necessary Moreover, BiRa allows efficiently setting up sets of Tx/Rx
to perform measurements combining both bistatic geometry positions, with joint control of gantries positions, turntable
and broadband OFDM waveform. A solution for this kind of azimuth, and device under test (DUT) settings. Therefore,
measurement is the recently inaugurated BiRa measurement it is possible to perform long fully automated measurement
system, situated at the Thuringian Innovation Center for Mo- runs, including multiple bistatic angles, without requiring any
bility (ThIMo) of the Ilmenau University of Technology. operator interaction.
This advanced system comprises two pivoting gantries, each Another important aspect for this measurement infrastruc-
equipped with a transmitting (Tx) and receiving (Rx) antenna, ture, is that for micro-Doppler analysis purposes, in order to
respectively [1]. The target object of interest is positioned at provide reference speed, it is desirable to have full control
the center of the turntable as in Figure 2. Therefore, the target over the rotors’ rotation, which is not feasible with most
can be illuminated and observed from any desired bistatic standard commercial drones. Therefore, a custom-built system
angle in azimuth, from any elevation bistatic angle in the upper was assembled, allowing us to remotely set the propeller’s
hemisphere, and from 25° elevation in the lower hemisphere. angular velocities. This system, presented on Figure 4, is
A particularly noteworthy feature is that its mechanical composed by a Tarot IRON MAN 650 mechanical structure,
system is independent and can be integrated with a vast with motors, sensors, and an Arduino on a PCB board for
variety of Tx/Rx measurement systems, thus allowing multiple setting up motors thrust values and measuring rotation speeds,
combinations of applications, waveforms, bandwidths, carrier as well as an Ethernet shield for remote commanding.
frequencies, etc. Furthermore, BiRa has a wideband channel
sounder, offering an instantaneous bandwidth of 4 GHz, in
B. Bistatic Micro-Doppler Measurement Setup
addition to a standard vector network analyzer (VNA), which
can be used for a vast set of static reflectivity measurement A setup was implemented, using a wideband OFDM-based
scenarios, from narrowband to ultra-wideband (UWB). An- transmit signal called Newman sequence with constant spectral
other very useful system available is the universal software magnitude and minimal crest-factor [9]. Table I details the
radio peripheral (USRP), which in combination with BiRa’s measurement system setup.

Fig. 3. Rotating propeller Micro-Doppler measurement signature in frequency Fig. 5. Comparison between simulation and measurement micro-Doppler
domain (16384 symbols, subsampling = 8). signatures of a rotating propeller (16384 symbols, subsampling = 8).
Fig. 6. Comparison of frequency-domain micro-Doppler signatures of a Fig. 7. Comparison of frequency-domain micro-Doppler signatures of a
rotating propeller from measurement data, with different bistatic angles (16384 rotating propeller from simulated data, with different bistatic angles (16384
symbols, subsampling = 8). The legend indicates the bistatic angle in each symbols, subsampling = 8). The legend indicates the bistatic angle in each
measurement. simulation.

For micro-Doppler analysis, the signal was processed in angle β. Figure 6 shows how the Doppler spread due to the
slow-time, i.e., a vector of the returns of the target in a propeller rotation decreases when the bistatic angle increases.
single range bin, along different symbols. Some subsampling As expected, the maximum Doppler spread is observed in
can also be applied in order to reduce the amount of data the monostatic case, while in forward configuration almost no
and the slow-time sampling frequency, and thus, to achieve a Doppler spread is present. We can also notice, that distance
better visualization of the data, focusing the analysis on the among impulses does not change with respect to bistatic
bandwidth of interest. In this work, a subsampling factor of 8 angles, since they depend only on the rotation frequency and
means that we use only every eighth symbol available. the number of blades.
C. Measurements Analysis IV. C OMPARISON B ETWEEN S IMULATION AND
A Doppler frequency signature of the custom drone with M EASUREMENT
only one propeller rotating, obtained from the measurement, is
A. Low Resolution Case
presented in Figure 3. There, we can see, above the noisy basis,
a strong impulse centered in f = 0 Hz, and many impulses Figure 5 presents a comparison between two micro-Doppler
on both sides of this stronger one. The central impulse occurs signatures: one simulated using the model developed in Sec-
due to reflections of the drone’s static parts, while the other tion II, and one obtained from the measurement described in
impulses come from the rotating parts. These impulses are Section III. Both use the parameters described in Table I, and
separated by ∆f = N frot , where frot is the propeller rotation geometrical configuration with bistatic angles of 30°.
frequency and N is the number of blades in the propeller. In Additionally, to generate the simulated data, the OFDM
this case, the rotation frequency was frot = 1500 rpm = 25 coefficients D(n, m) were obtained from the reference data
Hz and N = 2, so ∆f = 50 Hz, which is consistent with used for the measurement.
what can be seen in the image. Figure 5 shows that the simulated signature presents the
We can also see that this sequence of impulses ends in impulses mentioned in Subsection III-C in the same positions
the Doppler frequencies equivalent to the target’s maximum as in the measurement data target signature.
positive and negative relative speeds. In the current measure- The simulated signature well matches the response obtained
ment, the maximum velocity is given by v = 2πfrot l = from measured data, presenting the same Doppler spread and
2π × 25 × 0.1655 = 641.25 m/s, where l = 0.1655 m is the same distance between impulses, which are the most
the radius of the propeller. Bistatic Doppler is given by [10]: important features for classification.
Comparison between Figures 7 and 6 shows that the pro-
2v cos(β/2) cos(δ) posed model is also capable of emulating the dependence on
fD = (21)
λ the bistatic angle seen in the measurement data.
where β is the bistatic angle and δ the angle between the Finally, the Pearson correlation coefficient, i.e. the ratio
bistatic bisector, and v the direction of target motion. In this between the covariance of the simulated and the measured
case, we have δ = 0 and β = 30°. So fD max = 555.3 Hz. micro-Doppler signatures, and the product of their standard
An important effect that can be inferred from equation (21) deviations, returns a cross-correlation coefficient of approxi-
is the dependency that the Doppler spread has on the bistatic mately 0.98 across the bistatic angles from 30° to 180°.
 TABLE II
MICRO-DOPPLER HRR MEASUREMENT SETUP

Waveform OFDM-based
System Central frequency 7 GHz
Bandwidth 2.46 GHz
Total number of carriers 2500
OFDM
Carriers with energy 2048
Settings
Symbol duration 1.02µs
Number of rotating propellers 2
Blades per propeller 2
Drone Propeller radius 16.55 cm
Rotation speeds 1500 and 2000 rpm
Propeller material carbon fiber Fig. 8. Comparison of simulation (left) and measurement (right) range-
Doppler signatures of a drone with two rotating propellers.

B. High Range Resolution Case

V. C ONCLUSIONS AND O UTLOOK
The classical model for rotating blades micro-Doppler pre- This paper presents a model for simulating the bistatic
sented in [3] assumes that the target response is completely micro-Doppler signature of drone propellers in distributed
contained in one range resolution cell. ISAC. The simulated target signatures are compared to sig-
However, when the range resolution is smaller than the natures obtained from real BiRa measurement.
target dimensions, the latter has an extended response in fast- As a conclusion, it is shown that the proposed model
range dimension, which stands for the high range resolution produces the same basic features of the bistatic micro-Doppler
(HRR) case. This is achieved by using higher signal bandwidth signature achieved in the signatures obtained from measure-
c
B, since the range resolution is given by ∆R = 2B . [4] ment data. In addition, the model reproduces the relation
An additional advantage of the model proposed in the between the bistatic angle and the micro-Doppler spread in
current work is the ability to generate HRR data. the same way it appears in the measurement data signature.
It is also possible to simulate a drone with multiple pro- Future work can comprise the use of simulated data using
pellers by setting the coordinates of the rotation center point this model to train target classification algorithms, and the
O (Fig. 1) of each rotor according to the drone geometry, validation of this training method using measurement data.
generating each propeller returns using the proposed model,
and then summing the returns from each propeller and from ACKNOWLEDGMENT
the static parts of the drone. Propellers and static parts return This research is funded by the BMBF project 6G-
intensity can be adjusted by changing the RCS parameter σP ICAS4Mobility, Project No. 16KISK229K.
in (3). Future work will present a more detailed analysis of
R EFERENCES
multi-propeller drone simulation using this model.
Figure 8 shows a comparison between the range-Doppler [1] R. S. Thomä, C. Andrich, S. J. Myint, C. Schneider, and G. Sommerkorn,
“Characterization of multi-link propagation and bistatic target reflectivity
signatures of a drone with two propellers from two sources: for distributed ISAC,” arXiv preprint arXiv:2210.11840, 2022.
simulated from the proposed model and retrieved from BiRa [2] G. E. Smith, “Radar target micro-doppler signature classification,” PhD
measurements. Both cases use the parameters from Table II, thesis, University College London, 2008, available at https://discovery.
ucl.ac.uk/id/eprint/18688/1/18688.pdf.
thus represent a high range resolution scenario. The main [3] D. Jianjiang, Y. Zhiqiang, Y. Dazhi, and R. Chongji, “Modeling and
properties of the drone micro-Doppler signatures, such as validation of modulated characteristics for aircraft rotating structure
range of the propellers, Doppler spread, and distance between in the air surveillance radars,” in IEEE National Radar Conference -
Proceedings, vol. 2005-January, 2005.
impulses in frequency direction are the same on both sides. [4] M. A. Richards, Fundamentals of Radar Signal Processing, 2nd ed.
McGraw-Hill, 2014, ch. 2.
[5] K. M. Braun, “OFDM radar algorithms in mobile communication
C. Advantages and Disadvantages networks,” Ph.D. dissertation, Karlsruher Institut für Technologie (KIT),
2014. [Online]. Available: https://d-nb.info/104838490X/34
Many other methods can be used for modeling drone [6] J. Misiurewicz, K. Kulpa, and Z. Czekala, “Analysis of recorded
propellers, such as Finite Element Method (FEM), ray tracing, helicopter echo,” in Radar 97 (Conf. Publ. No. 449). IET, 1997, pp.
primitive-based modeling from finite discrete scatter points, 449–453.
[7] X. Qu, K. Li, Q. Zhang, and B. Liang, “Micro-doppler feature extraction
and classic primitive continuous modeling. Compared to other under passive radar based on orthogonal frequency division multiplexing
methods, the most important advantages of the proposed communication signal,” The Journal of Engineering, vol. 2019, pp.
method are its simplicity, low computational cost, and ability 6889–6893, 2019.
[8] X. Ai, X. Zou, J. Yang, J. Liu, and Y. Li, “Feature extraction of rotating
to represent the HRR case, while taking the waveform into target based on bistatic micro-Doppler analysis,” in Proceedings of 2011
account. Its main disadvantage is that it doesn’t take into IEEE CIE International Conference on Radar, RADAR 2011, vol. 1,
account some electromagnetic phenomena, such as diffraction 2011.
[9] S. Boyd, “Multitone signals with low crest factor,” IEEE Transactions
and multipath, which can introduce inaccuracies. However, on Circuits and Systems, vol. 33, no. 10, pp. 1018–1022, 1986.
its simplicity makes it suitable for applications where a large [10] N. J. Willis, Bistatic radar. SciTech Publishing, 2005, vol. 2, ch. 6.
amount of simulated data with various parameters is required.