See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/363737876

Satellite Visibility Window Estimation using Doppler Measurement for IoT

Applications

Preprint · September 2022

DOI: 10.36227/techrxiv.21173032

CITATIONS READS
0 153

2 authors:

Iza Shafinaz Mohamad Hashim Akram Al-Hourani

RMIT University RMIT University
10 PUBLICATIONS 42 CITATIONS 156 PUBLICATIONS 7,201 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Iza Shafinaz Mohamad Hashim on 22 September 2022.

The user has requested enhancement of the downloaded file.

 1

Satellite Visibility Window Estimation using

Doppler Measurement for IoT Applications
Iza S. Mohamad Hashim, Akram Al-Hourani

Abstract—Many Internet-of-Things (IoT)-over-satellite appli- that can be made; (i) optical observations such as visual,
cations rely on affordable location-aware but energy-constrained photographic, and laser ranging measurements, and (ii) radio
IoT sensors. In this paper, we propose a novel method to estimate measurements such as radio ranging, radar, Doppler observa-
the satellite visibility window in IoT devices based on simple
Doppler measurements. We present two scenarios where the tions, and interferometry [4]. Assuming a perfect Keplerian
orbital information of the serving satellite’s is initially unknown motion, an orbit can be fully described using six fundamental
to the IoT device: (i) we assume that the geographic coordinates elements: the semi-major axis, eccentricity, inclination angle,
are known to the IoT device, and (ii) we assume that the argument of periapsis, true anomaly, and the right ascension
coordinates are completely unknown. Accordingly, we derive the of ascending node. The three established orbit determination
Doppler measurement likelihood function, and simplify it to a
root mean square error (RMSE) minimization problem. From methods include Gaussian, Gibbs’ method, and Laplacian [4].
a sequence of Doppler measurements, we estimate the orbital The Gaussian orbit determination only requires the knowledge
parameters of the serving satellite using a stochastic optimizer of two orbital positions and the observation interval. Then,
to minimize the RMSE. From the orbital estimation, we then the velocity can be computed from a distance between two
predict the satellite visibility window (satellite pass). To gauge points on an ellipse and the difference in the observation times.
the accuracy of the window estimation, we apply the intersection-
over-union metric to compute the overlapping visibility window However, this method has limited use cases in satellites as
between the ground truth and the estimation, and consequently it only allows small spread between the true anomalies an-
present the results based on extensive Monte Carlo simulations. gles [5]. The Gibbs’ method takes three position observations,
Index Terms—Localization, orbital determination, satellite and thus provide better estimation since only one ellipse can
pass, visibility window, Internet-of-Things, IoT-over-Satellite. pass through three points. Time is not essential in this method
except when the satellite’s position on the ellipse is desired.
Laplacian utilizes three of the right ascension, declination, and
I. I NTRODUCTION time observations unlike the other methods that use Cartesian
Internet-of-Things (IoT)-over-satellite applications have positions. By using three pointing angles and their observation
been proliferating in recent years due to the highly accessible times, the orbit can be fully characterized [6]. These traditional
satellite platforms, especially in the low Earth orbit (LEO). methods have been used over the last centuries for planetary
The moderate costs of development and deployment of LEO motion and other celestial bodies; however, they are not
satellites have enabled the expansion of IoT-over-satellite suitable to be applied in IoT use cases since either the position
applications [1]. However, LEO satellites are known to have or angular measurements are not available in simple IoT
a short visibility window due to their fast relative velocity. devices.
Typically, the orbital period of LEO satellites is around 90 Modern methods such as [7] has introduced a least-squares
to 120 minutes; thus, LEO satellites are only visible within a (LS) technique for orbit determination, where a computed
few minutes window, also called the satellite pass duration [2]. curve is fitted to the Doppler observations using LS. The curve
On the other hand, many IoT access technologies are based on is obtained from the six Keplerian orbital parameters. By using
simple protocols with short and random transmissions. There- differential corrections, the estimation of the parameters is
fore, to increase the probability of a successful transmission improved iteratively until they converge to the ground truth. In
between the IoT device and receiving satellite, it is crucial addition, two orbit determination methods are proposed in [8];
to align the IoT transmission time with the satellite visibility the first method utilizes a perturbed circular motion predic-
window. By doing so, the lifetime of IoT devices can be tion model which can estimate the orbit from the collected
prolonged as well as enhancing the transmissions (frames) telemetry radio measurements with no NORAD database
success rate. Besides obtaining sensor data from the ground information, while for the second method, the SGP4 model
device, it is also beneficial to gain its location information [3]. and NORAD two-line element (TLE) information are used.
Therefore, ensuring a window alignment between the ground Moreover, a maximum likelihood estimation (MLE) method is
device and satellite is one of the major requirements for more proposed by [9] to solve the orbit determination problem. The
reliable IoT services. estimator can produce the optimal parameter (state) estimates
Estimating the visibility window requires some knowledge from the Doppler observations. The parameter estimates are
of satellites orbit. To do so, there are two types of observations determined by maximizing the conditional density function of
the state given the observations.
The authors are with School of Engineering, RMIT University, Melbourne, The work in [9] demonstrates how standalone Doppler shift
Australia.
Emails: {iza.mohamad.hashim, akram.hourani}@rmit.edu.au measurements can be used to estimate an orbit. However, [9]
Manuscript received xx xxxx 2022; revised xx xxxx 202x. attempts to solve the estimation analytically; thus, requiring
 2

the derived likelihood function to be differentiable. In this TABLE I: Notations and Symbols
work, Doppler measurements are performed at the ground Symbol Definition Value [Unit]
IoT device, which are consequently used to estimate the orbit a Orbit’s semi-major axis 7571 [km]
and the corresponding visibility windows. By acquiring the e Orbit’s eccentricity 0
i Orbit’s inclination 87.9◦
knowledge of the satellite pass, the power can be efficiently h Satellite’s altitude 1200 [km]
conserved where the IoT device needs only to turn on its T Orbital period 109.26 [min]
transceiver when the satellite is visible, and it can then be µG Geocentric gravitational constant 3.986 × 1014 [m3 /s2 ]
Ω Right ascension of ascending node 240◦
switched off during the rest of the time [10]. Many literature θo Initial mean anomaly −60◦
works [8], [9] propose the utilization of Doppler shift mea- v Relative velocity - [m/s]
surements obtained from telemetry signals. Thus, we further ρ Slant distance between satellite - [m]
and ground device
leverage these Doppler measurements from communication t Time variable - [s]
signals for estimating the visibility window. ∆t Sampling time 15 [s]
Accordingly, this letter presents a framework to determine fd Doppler shift frequency measurement - [Hz]
µd True Doppler shift frequency - [Hz]
the satellite visibility window in two scenarios; (i) when σd Standard deviation of 10 [Hz]
geographic location is available to the ground IoT device by Doppler shift measurement
an external input, e.g., using global navigation satellite system Θ Latitude of IoT ground device −23.5◦
Φ Longitude of IoT ground device 136◦
(GNSS), and (ii) when IoT device has no knowledge of its
location. In the first scenario, the prediction can be performed
even when limited information regarding the satellite is avail- a given reference time. Accordingly, the shape of the elliptical
able; namely only the satellite’s altitude is assumed to be orbit can be expressed as follows in a polar coordinate system:
known. Therefore, we need to estimate three relative orbital
parameters: the orbit’s inclination, right ascension of ascending a(1 − e2 )
r= (1)
node (RAAN), and initial mean anomaly with respect to the 1 + e cos θ
IoT device timing. In the second scenario, the IoT device needs where a is the semi-major axis, e ∈ [0, 1] is the eccentricity,
to estimate its own latitude. Thus, we need to estimate an and θ is the true anomaly. The orbit’s shape is determined by a
additional parameter to the first scenario on top of the other and e parameters, whereas the satellite’s position in the orbit
three orbital parameters. Following is a summary of the main is defined by the polar coordinates: distance to the origin r
contributions of this paper: and angle θ. Since we assume a circular orbit, the eccentricity
• It proposes a novel framework to predict the satellite is e = 0, thus reducing the ellipse equation to r = a. Also,
visibility window based solely on Doppler measurements the true anomaly is equivalent to both the eccentric and mean
obtained from the communication signal. anomaly when e = 0.
• It compares the performance of different stochastic opti- Without loss of generality, we focus in this paper on the
mizer methods in minimizing the objective function. polar-orbiting LEO satellite where the orbital inclination can
• It analyzes the predicted visibility window for several be in the range of i ∈ [60◦ , 90◦ ] and the altitude h ∈
passes using a newly introduced intersection-over-union [200 km, 2000 km], owing to the popularity of such orbit in
metric. the current satellite constellation, such as OneWeb, Starlink,
and Iridium. Also, LEO satellite orbits are preferable for IoT
applications because of the reduced power budget.
II. S YSTEM M ODEL Furthermore, we follow Kepler’s second and third laws to
To estimate the satellite’s visibility window, we outline the simulate the satellite’s motion around the orbit. In satellite
(i) geometric model of the satellite orbit, and (ii) Doppler application, Kepler’s second law states that a satellite which
measurements error model. For convenience, we list relevant orbits the Earth will cover the same area for the same time
notations and symbols in Table I. interval. As for Kepler’s third law, the square of the orbit’s
period, T is proportional to the cube of the orbit’s semi-major
axis, a,
A. Satellite orbit geometric model s
4π 2

A Keplerian orbit can be fully described using six funda- T = a3 (2)
µG
mental parameters; (i) semi-major axis a, (ii) eccentricity e,
which defines the oblateness of the orbit, (iii) inclination angle where 4π 2 /µG is the proportionality constant and µG is
i, which is the angle between the equatorial plane and orbital the geocentric gravitational constant which is equivalent to
plane, (iv) right ascension of the ascending node Ω which is 3.986 × 1014 m3 /s2 . Fig 1 is an example of a circular orbit
the rotation of the orbital plane from the reference axis, (v) with three orbital parameters; RAAN Ω, initial mean anomaly
argument of periapsis ω which is the angle from the ascending θo , and inclination i.
node to the perigee point, however, in a circular orbit scenario, Accordingly, for generating the ground truth orbit, we apply
there is no perigee point; thus this angle can be set to zero the Simplified General Pertubations 4 (SGP4) propagator using
without loss of generality, and (vi) true anomaly which is the the supplied Keplerian orbital elements. SGP4 takes into
angle between the perigee point and the satellite’s position at account the secular and periodic variations that are caused by
 3

As an inverse problem, we want to use the readily available

Doppler measurements to estimate the orbit and hence future
visibility windows.

A. Orbit determination
In this framework, we first estimate the satellite’s orbit using
the Doppler shift frequency measurements. To sufficiently
define an orbit, all aforementioned six parameters need to be
estimated in addition to the observer location. To reduce the
complexity of computation, we reduce the number of orbital
parameters needed to be estimated. In IoT applications, we
investigate the two following cases:
• Case 1: to estimate three of the satellite’s orbital parame-
ters, namely the inclination, RAAN, and the initial mean
anomaly. This is because we assume a typical case of a
Fig. 1 Illustration of a satellite orbital element for circular orbit circular orbit, where we can assume e = 0 and argument
of periapsis at zero without loss of generality. Also, we
the atmospheric drag and Earth’s geometry. The propagator is assume the altitude is known. In addition, the position of
suitable for satellites that have an orbital period of less than the IoT ground device is assumed to be known.
• Case 2: to estimate the inclination, RAAN, the initial
225 minutes.
mean anomaly of the satellite’s orbit, in addition to the
latitude of the ground IoT device. Note that we do not
B. Doppler measurement model need the longitude of the device as the orbit is referenced
Doppler shift is caused by the change in signal’s frequency to the device’s timing and assuming its longitude as the
due to the relative velocity between the transmitter (satellite) meridian. The only known parameter is the satellite’s
and the receiver (IoT device). The model assumes that the altitude.
satellite transmitter is orbiting the Earth in a certain orbit while
the ground IoT device receiver is stationary with respect to the B. Likelihood function
Earth-centered Earth-fixed coordinates (ECEF). The relative To estimate the orbital parameters, we derive the likelihood
velocity can be computed as follows, function that represents the Doppler measurements given a
ρ(t + ∆t) − ρ(t) set of orbital parameters and location. Then, we maximize the
v(t) = lim (3) function by utilizing a stochastic optimizer to find the parame-
∆t→0 ∆t
ters. Accordingly, the state vector is given by, x = [i, Ω, θo ] for
where ρ is the slant distance between the satellite and the
case 1 and x = [i, Ω, θo , Θ] for case 2, where Θ is the latitude
ground device, t is the time variable, and ∆t is the simulation
of the IoT device. Since, Doppler shift measurements are
time step.
normally distributed, the likelihood function can be expressed
To model the Doppler shift frequency measurement as
as follows:
realistically as possible, we measure the Doppler and obtain  !2 
(k) (k)
the ground truth of the measurements with the help of a 
(k)
 1 1 f d − µ d (x)
GPS disciplined oscillator based on NOAA 15 satellite signal p fd |x = √ exp −  (4)
σd 2π 2 σd
as explained in our previous work in [11]. We perform a
Kolmogorov-Smirnov normality test on the measurements, where fd is the measured Doppler shift frequency at time index
where Doppler error distribution passes the test with 95% k, µd is the true Doppler shift frequency which depends on
confidence, indicating a suitable resemblance to the Normal the state vector x, and σd is the error standard deviation.
distribution. Therefore, the Doppler shift measurement is rep- We can simplify the likelihood maximization problem,


resented as fd ∼ N µd , σd2 , where µd is the true Doppler based on K measurements, into a root mean squared error
shift given by µd = vfc /c, and the standard deviation σd , (RMSE) minimization problem, as follows,
where c is the propagation speed of light in vacuum, and fc   
(k) (k)
2  
is the carrier frequency.
K fd − µd (x)
(k)
Y
− ln p(fd |x) = − ln C  exp −  ,
  
2σd2
k=1
III. S ATELLITE VISIBILITY WINDOW ESTIMATION   2 
FRAMEWORK (k) (k)
K
X fd − µd (x)
Satellite passes and the corresponding visibility window = −  + C, accordingly,
 
2σd2
depend on the time, the location of the observer, and the k=1
orbit of the satellite. Accordingly, the visibility prediction is v
u K f (k) − µ(k) (x) 2
u  
calculated by propagating the satellite orbit and iteratively uX d d
sampling it using a fine time grid [12]. However, this is only ϵ(x) = t (5)
possible if we have a prior knowledge of the satellite’s orbit. K
k=1
 4

which represents the RMSE between the measured Doppler and observer locations. Accordingly, the satellite is deemed to
vector and √
calculated mean based on the orbital geometry. be visible when its elevation angle exceeds a certain minimum
C = 1/(σd 2π) is a constant, which does not contribute to threshold ζh , where the visibility window(s) in a given time
the minimization problem. Accordingly, we apply a stochastic interval [a, b] can be expressed as,
optimizer to minimize the RMSE, which is equivalent to
W = {t ∈ [a, b] : ζ > ζh } . (7)
maximizing the likelihood function.
IV. S IMULATION AND R ESULTS
C. Stochastic Optimizer
In the simulation, we utilize a single transmitting satellite
There are several stochastic optimization algorithms to and a receiving ground IoT device. Firstly, we propagate
choose from, such as pattern search (PS), simulated an- the orbital scenario and based on the simulated geometry
nealing (SA), genetic algorithm (GA), multi-objective genetic we obtain Doppler measurements as explained in Sec. II-B.
algorithm (MOGA), particle swarm (PSO), and surrogate Accordingly, we use the first two passes for the orbital
optimization (SO). As such, we investigate the performance estimation, and the consecutive passes for testing the goodness
using a range of optimizers to find the most suitable for this of the estimation. The obtained results are based on 300 Monte
problem. Thus, we run the optimizers under Case 1 condition Carlo runs, where the simulation steps are further explained
using the same satellite orbit for 300 runs. The goal of the as follows,
optimizer is to minimize the RMSE cost function in (5), by
finding the most suitable state vector,
A. Orbital estimation results
x̂ = argmin [ϵ(x)] (6) The statistics of the estimated parameters errors for both
x
Case 1 and Case 2 are shown in Fig. 3. The median error is
where x̂ represents the estimated state vector. represented as a dot marker. It can be seen that for Case 1, all
of the three orbital parameters errors have a narrow spread, and
10 4 their error medians are approximately close to 0◦ . In contrast,
Objective function

2 Case 2 has a wider spread and the error medians are higher
1.5 than Case 1.
1
1 15 15
0.5 2
10 10
0 0.5

Error of o [deg]

[deg]
[deg]
Error of i [deg]

SA PS GA MOGA SO PSO 1 5 5
Stochastic optimizers
0 0 0 0
Error of

Error of
Fig. 2 Boxplot of the objective function (RMSE) using a range -5 -5
-1
of stochastic optimizers, where the dot marker represents the -0.5 -10
-10
median error. -2
-15 -15
-1
1 2 1 2 1 2 2
It can be observed from Fig. 2 that the particle swarm Case
optimizer has the best performance in minimizing the objective
function. PSO is one of the bio-inspired algorithms that is Fig. 3 Boxplot of the estimated parameters errors (inclination,
based on the movement of organisms in a group such as RAAN, initial mean anomaly and latitude) for Case 1 and
school of fish, and flock of bird [13]. It is categorized as a Case 2.
metaheuristic optimization algorithm where it combines the
diversification and intensification components. The organisms, In Fig. 4 we further depict an example of the true and
referred to as particles in the algorithm are spread out in estimated orbit in a three dimensional (3D) geometry. The
the solution space, and each particle tries to find the optimal estimated orbit is computed from the estimated parameters,
solution, which refers to diversification. Once the particles and has a very good match to the true orbit for both Case 1
have found a good solution within the space, they will move and Case 2.
towards the good solution region to conduct a focused search,
which is the intensification component of the algorithm [14]. B. Evaluating the visibility window
Finding a good balance between these two components can To fairly evaluate the estimation fitness of the visibility
ensure the algorithm to converge to the global solution. Also, window, we adopt the metric of intersection-over-union (IoU)
it is a simple algorithm which requires only basic mathematical which is commonly used in evaluating bounding box predic-
operations, thus making it computationally inexpensive [13]. tion in machine vision applications [15]. IoU is computed
by finding the percentage of the overlapping bounding boxes,
where in our case, the bounding box is the satellite visibility
D. Visibility window
window. The following shows the IoU computation,
Satellite’s elevation angle ζ with respect to an observer is
measured between the satellite point and the local horizon of A∩ Ŵ ∩ W
IoU = = , (8)
the observer. The angle can be calculated based on the satellite A∪ Ŵ ∪ W
 5

IoU (2 Meas)
0.8

0.6
1

IoU (4 Meas)
Pass index
0.8

Mean
0.6
0 5 10 15 20 25 30 35
Pass index

Fig. 6 Statistics of the intersection-over-union (IoU) for satel-

Fig. 4 An example 3D orbit comparison between the ground lite visibility window (Case 1) after several passes using two
truth and Case 1 / Case 2 orbital estimation. and four passes of Doppler measurements with confidence
interval of 95%.

where A∩ is the area of the intersection, and A∪ is the area

of union of the estimated and the ground truth windows. Ŵ into a root mean square (RMSE) minimization problem. Then,
represents the estimated visibility window, whereas W denotes we applied a stochastic optimizer to minimize the RMSE.
the ground truth which was previously defined in (7). From the estimated orbit, we investigated the accuracy of
We illustrate in Fig. 5 further example of the satellite visi- the visibility window using the intersection-over-union metric.
bility window simulation, showing both the estimated window Based on the results, it can be concluded that the visibility
and the ground truth window. In this example, the true and window can be adequately estimated based on the Doppler
estimated window are fully overlapping. The figure also shows measurements.
the estimated and the ground truth Doppler profiles. R EFERENCES
[1] A. Nascetti, E. Pittella, P. Teofilatto, and S. Pisa, “High-gain S-band
patch antenna system for Earth-observation CubeSat satellites,” IEEE
Antennas and Wireless Propagation Letters, vol. 14, pp. 434–437, 2015.
[2] A. Al-Hourani, “A tractable approach for predicting pass duration in
dense satellite networks,” IEEE Communications Letters, vol. 25, no. 8,
pp. 2698–2702, 2021.
[3] I. S. M. Hashim, A. Al-Hourani, and B. Ristic, “Satellite localization
of IoT devices using signal strength and Doppler measurements,” IEEE
Wireless Communications Letters, pp. 1–5, 2022.
[4] R. Hart, “Single-station tracking for orbit determination of small satel-
lites,” in Proceedings of the Small Satellite Conference, 1987.
[5] V. Arroyo, A. Cordero, and J. R. Torregrosa, “Approximation of artificial
satellites’ preliminary orbits: The efficiency challenge,” Mathematical
and Computer Modelling, vol. 54, no. 7, pp. 1802–1807, 2011.
Fig. 5 Satellite visibility window (Case 1) with the bounding [6] R. R. Bate, D. D. Muller, and J. E. White, Fundamentals of astrody-
boxes denoted as the green shaded area for the estimated namics. New York: Dover Publications, 1971.
[7] R. B. Patton, “Orbit determination from single pass Doppler observa-
window, and red for the true window. tions,” IRE Transactions on Military Electronics, vol. MIL-4, no. 2/3,
pp. 336–344, 1960.
To further investigate the performance of the proposed [8] A. A. Spiridonov, V. A. Saetchnikov, D. V. Ushakov, V. E. Cherny, and
A. G. Kezik, “Small satellite orbit determination methods based on the
framework in estimating the visibility window, we present the Doppler measurements by Belarusian State University ground station,”
IoU statistics comparing two cases: (i) when the measurements IEEE Journal on Miniaturization for Air and Space Systems, vol. 2,
from two passes are used for the estimation, and (ii) when no. 2, pp. 59–66, 2021.
[9] B. D. Tapley, B. E. Schutz, and G. H. Born, Chapter 4 - Fundamentals of
using the measurements from four passes. The comparison is Orbit Determination. Burlington: Academic Press, 2004, pp. 159–284.
shown in Fig. 6, where it is imperative to note that by adding [10] I. Ali, N. Al-Dhahir, and J. E. Hershey, “Doppler characterization for
more measurements in the estimation, we are able to improve LEO satellites,” IEEE Transactions on Communications, vol. 46, no. 3,
pp. 309–313, 1998.
the bounds of the prediction. However, in practical scenarios, [11] I. S. M. Hashim and Al-Hourani, “Satellite-based localization of IoT
there is an energy cost for each Doppler measurement. Conse- devices using joint Doppler and Angle-of-Arrival estimation,” Submitted
quently, it is a designer choice to balance the trade-off between to IEEE Transactions of Aerospace and Electronics, pp. 1–11, 2022.
[12] I. Ali, P. G. Bonanni, N. Al-Dhahir, and J. E. Hershey, “Satellite visibility
accuracy and battery life of the device. prediction,” Doppler Applications in LEO Satellite Communication
Systems: CD-Rom Included, pp. 53–66, 2002.
[13] J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proceed-
V. C ONCLUSION ings of ICNN’95 - International Conference on Neural Networks, vol. 4,
1995, pp. 1942–1948 vol.4.
In this letter, we presented a framework for predicting [14] C. Blum and A. Roli, “Metaheuristics in combinatorial optimization:
satellite’s visibility window using a series of Doppler mea- Overview and conceptual comparison,” ACM Comput. Surv., vol. 35,
surements. We estimated the satellite’s orbit by deriving the no. 3, p. 268–308, Sep 2003.
[15] H. Rezatofighi, N. Tsoi, J. Gwak, A. Sadeghian, I. Reid, and S. Savarese,
likelihood function of the Doppler measurements given a set “Generalized intersection over union: a metric and a loss for bounding
of orbital parameters; inclination, RAAN, and initial mean box regression,” in Proceedings - 2019 IEEE/CVF Conference on
anomaly. We simplified the likelihood function maximization Computer Vision and Pattern Recognition, 2019, pp. 658–666.

View publication stats