JOURNAL OF SPACECRAFT AND ROCKETS

Analysis of Satellite Constellations for the Continuous Coverage

of Ground Regions

Guangming Dai,∗ Xiaoyu Chen,† and Maocai Wang‡

China University of Geosciences, 430074 Wuhan, People’s Republic of China
Elena Fernández§
Technical University of Catalonia, 08034 Barcelona, Spain
and
Tuan Nam Nguyen¶ and Gerhard Reinelt**
Heidelberg University, 69120 Heidelberg, Germany
DOI: 10.2514/1.A33826
This paper studies the problem of analyzing multisatellite constellations with respect to their coverage capacity of
Downloaded by UNIVERSITY OF ADELAIDE on August 21, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.A33826

areas on Earth’s surface. The geometric configuration of constellation projection points on Earth’s surface is
investigated. A geometric subdivision approach is described, and the coverage target area belonging to each satellite
and its maximum circle radius are defined and calculated. Accordingly, the target area can be decomposed into
subregions, and thus the multisatellite coverage problem is decomposed into a one-satellite coverage problem. An
accurate and effective solution method is proposed that solves both continuous and discontinuous coverage problems
for any type of ground area. In addition, a procedure for calculating satellite orbital parameters is also proposed. The
performance of our approach is analyzed using the Globalstar system as an example, and it is shown that it compares
favorably with the classical grid-point technique and the longitude method.

Nomenclature ∂Ωi = the boundary of coverage target area Ωi

C = the satellites constellation Ω = target area
DT
O = the Delaunay triangulation of O Ωi = the coverage target area of Si
d = coverage spherical radius Θ
P; α = spherical circle with the center point P
θ; φ and
Fi = the coverage feature point set of Si radius α
f = mapping an orientation angle to a point on the θ, φ = the longitude and latitude angle
spherical circle
g
β = one-to-one mapping from β to a unique point
H = orbital altitude I. Introduction
h
Θ 0  = projection function from a spherical Θ 0 to the

M
P =
yz-plane
coordination rotation matrix in R3×3
S ATELLITES find numerous applications in communication,
navigation, imaging, and remote sensing [1,2]. The requirements
for data accuracy and real-time observations become more and more
O = the finite set in some metric space M with distance
extensive as the need for Earth’s observation data continuously
function d
increases [3]. To satisfy these demands, systems of multiple satellites
Oe = Earth’s center
have to be designed and their performance has to be analyzed.
P, P 0 = point on Earth’s surface
Earth observation from remote satellites and data transmission
Re = Earth’s radius
between satellites and facilities are both related to satellite coverage.
Ru
α = the rotation with angle α around the u axis
In general, satellite coverage problems arise when a target area on the
r = spherical circumcircle radius
Earth’s surface must be visible to one or more satellites. In principle,
rmax
i = maximum circle radius of Si
S = the satellite there are two main types of coverage problems. One class assumes
VD
O = the Voronoi diagram of O that the parameters of the satellites are fixed and studies their
VR
Oi  = the Voronoi region associated with the point Oi coverage capacity measured in terms of the percentage of the target
α = coverage angle area that is covered (visible). The second class of problems refers
β = the orientation angle to the determination of the positions of the satellites (i.e., their
γ min = minimum elevation angle constellation or configuration), including satellites orbital
parameters, in order to achieve a maximum coverage. Special
variants arise in both settings if full coverage of the target area is
requested and an accurate determination of the achieved coverage is
Received 6 December 2016; revision received 10 May 2017; accepted for
necessary with respect to a variety of coverage areas. Furthermore,
publication 26 June 2017; published online 11 August 2017. Copyright ©
2017 by the American Institute of Aeronautics and Astronautics, Inc. All because satellites are moving and covering areas are subject to
rights reserved. All requests for copying and permission to reprint should be change, in practice these problems must be extended to account for
submitted to CCC at www.copyright.com; employ the ISSN 0022-4650 varying constellations within given time intervals, also denoted as
(print) or 1533-6794 (online) to initiate your request. See also AIAA Rights reconstruction periods.
and Permissions www.aiaa.org/randp. Most methods in coverage analysis are based on the visibility of a
*Professor Doctor, School of Computer Science; [email protected]. subset of points of the given target area [4]. Hence, an accurate
†
Ph.D., School of Computer Science; [email protected] coverage of such points becomes a prerequisite, given that the
berg.de (Corresponding Author). performance of satellite constellations significantly depends on the
‡
Assistant Professor Doctor, School of Computer Science; cugwangmc@
126.com.
precision with which such points are covered. Recent research on
§
Professor Doctor, Statistics and Operations Research Department; coverage analysis often resorts to simulation or analytical methods.
[email protected]. Such methods exhibit a low computational efficiency and little
¶
Doctor, Institute of Computer Science; [email protected]. reliability of the results. Thus alternative methods are needed for large
**Professor Doctor, Institute of Computer Science; [email protected]. target areas when high accuracy is required.
Article in Advance / 1
 2 Article in Advance / DAI ET AL.

In this paper we present a method to obtain the coverage target area modeled the Earth’s surface as a series of longitudinal strips and
and coverage capacity for each satellite of a constellation. The investigated the cumulative coverage problem by determining the
method is based on a geometric subdivision of the target area exact latitude intervals that must be observed at each time instant. The
obtained from the spherical Voronoi diagram and its associated accuracy of the method only depends on the choice of the time step
Delaunay triangulation. Our geometric subdivision method also and the number of strips that are used for instantaneous coverage
allows to compute directly the satellites orbital parameters when full computations. More recently, Xu and Huang [19] proposed a new
coverage is imposed. From the coverage areas obtained for a fixed algorithm for the design of revisited orbitals in satellite coverage. In
time point, the full coverage and the continuous coverage capacity particular, they analyzed the global coverage performance and the
can be efficiently and accurately computed in the desired relationship between the coverage percentage and the orbital altitude
reconstruction period. Overall, the constellation design process by mapping every orbit pass into its ascending and descending nodes
improves significantly. Furthermore, our method is very versatile as it on different latitude circles, and by evaluating the performance at
solves discontinuous and continuous coverage problems for any type different altitudes. For the constraint on the elevation angle,
of target areas. Ulybyshev [20] and Seyedi and Safavi [21] proposed statistical
This paper is organized as follows. Section II reviews some related methods for the computation of satellite coverage times and service
literature and Sec. III introduces the main concepts related to satellite areas, based on the analysis of satellites coverage capacity to each
coverage of a ground region. In Sec. IV we review the properties of latitude. Such procedures eliminate the errors of traditional
two classical geometric structures and define the orientation angle. techniques, due to points with discrete values in the latitude direction.
Section V defines the coverage target area belonging to a single The Rosette Constellation [22] was proposed by Ballard in the
satellite and presents a computational approach for the analysis of the 1980s based on a concept different from numerical simulation. This
Downloaded by UNIVERSITY OF ADELAIDE on August 21, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.A33826

continuous coverage area associated with multisatellites. In Sec. VI method presents better constellation coverage properties, which are
we give a method for the computation of orbital parameters when full analyzed in terms of the largest coverage circle range between
coverage is imposed. The results assessing the correctness and the anywhere on the Earth’s surface and the nearest subpoint of the
efficiency of the proposed methods using the Globalstar system as an satellite. In [23,24] authors introduce specific methods implemented
example are presented in Sec. VII [5], where we also compare the for particular cases of discontinuous coverage analysis. They are still
performance of our approach with two popular coverage methods. associated with the calculating procedure with a priori selected
The paper ends in Sec. VIII with a summary and some conclusions. classes of orbital structures or the particular character of the Earth
surface coverage type. Mozhaev [25] analyzed the kinematics of the
вставить в общ обзор specific intersatellites constellation and studied a single continuous
II. Literature Review global coverage problem using symmetry group theory. Ulybyshev
In 1970, Walker [6] studied circular orbital patterns and particular [4] defined a function for the (full or partial) coverage capacity of a
constellation configurations for the continuous coverage of global satellite for a geographical region, which uses as parameters the right
regions, ensuring that every point on the Earth’s surface is always ascension of ascending node (RAAN) and the latitude. In [26,27] the
visible from at least one satellite, under the constraint of the minimum same author introduced a geometric pattern, named coverage belt,
elevation angle. Crossley and Williams [7] proposed simulated and developed a new method for the analysis of maximum revisiting
annealing and genetic algorithms for the same problem. Asvial et al. times to a specified latitude in discontinuous coverage satellite
[8] addressed the design of constellation configurations, taking into constellations. Razoumny [28] analyzed and investigated the satellite
account the total number of satellites and their altitudes, the angle constellation design for Earth discontinuous coverage based on the
shift between satellites, the angle between planes, and the inclination geometry analytic solutions for latitude coverage by single satellite.
angle. Multiobjective evolutionary algorithms, dealing with the Mohammadi [29] introduced the concept of the best coverage region
discrete and nonlinear characteristics of the metrics, have been for the receiving stations from a low-Earth-orbit (LEO) remote
developed for the design of satellite constellations for regional sensing satellite. He proposed a construction method guaranteeing
coverage by Ferringer and Spencer [9] and by Wang et al. [10]. the requirements of the receiving stations, based on the analysis of the
Ferringer et al. [11] proposed two parallel multiobjective variation of the satellite’s altitude for the coverage areas obtained
evolutionary algorithmic frameworks (master–slave and island under different minimum elevation angle constraints. In this
approach) for the same problem, which approximate the true Pareto approach, the best coverage region is defined as the coverage area of
frontier. the satellite. Sengupta et al. [30] proposed a semianalytical technique
One challenge when applying evolutionary algorithms to for the study of the coverage of LEO satellites. They expressed the
constellation design is how to incorporate suitably satellite coverage area of a satellite as a function of orbital elements and
kinematical features and constellation coverage characteristics obtained accurate results using numerical integration methods. The
relevant to the optimization process. Given that satellite orbitals are above-mentioned simulation methods are usually time-consuming
determined by six parameters that need to be repeatedly coded and and of low accuracy. Analytical methods are used only for the
updated, large computing times and memory space are usually continuous coverage analysis of a single satellite and regular target
required. Thus, when addressing constellation design, it seems more areas (e.g., a single point, a latitude or longitude line, or the global
suitable to consider only a few satellites together with coverage region). They are used for analyzing the coverage performance of a
requirements for a series of point targets. single satellite and are exact over time only if the orbital parameters
From a methodological point of view, recent research on Earth are analyzable, but they are not suitable for calculating the coverage
coverage analysis can be classified as numerical simulation or capacity of satellite constellations for any specified ground areas.
analytical methods. Currently, the grid-point technique is the most
commonly used method for satellite coverage analysis in several
applications [12–14]. This technique was first introduced in early III. Basic Coverage Definitions and Properties
1973 by Morrison [15] for the percentage satellite coverage on the For our purposes Earth is assumed to be a perfect sphere with
Earth’s surface, and for the statistical analysis of the full coverage of a center Oe and radius Re . For a satellite S and a point P on Earth’s
region for a system with 16 synchronous satellites. For the evaluation surface visible from S, the elevation angle at P is the angle between
of the satellite coverage capacity Jiang et al. [16] computed the the line connecting S and P and the tangent plane to the Earth’s
percentage coverage of a ground region by a constellation at any time surface at P. The subpoint O of S is the projection of S on the surface
by 1) dividing the target area into a number of discrete grids according and is the unique point on the Earth’s surface with elevation angle
to certain rules, and 2) taking the satellite’s coverage capacity to the π∕2. It is obtained by intersecting the line connecting S and Oe with
center of each grid as the coverage capacity for the entire grid area. the Earth’s surface. The coverage area of S contains all the points on
Mortari et al. [17] applied the Flower constellation theory to the the Earth’s surface that can be observed from S with elevation angle
maximization of the global coverage and the network connectivity of at least γ min and is determined by a spherical circle (shaded part of
via intersatellite links. On the other hand, Casten and Gross [18] Fig. 1) associated with the coverage angle α whose spherical radius is
 Article in Advance / DAI ET AL. 3

d ≥ maxfr1 ; r2 ; : : : ; rN g (2)

then this proves that the constellation covers the surface of the Earth
completely.
Note that the coverage radius d is a technical parameter of a
satellite. This parameter is denoted as coverage angle θ introduced in
[20]. However, the Earth-central angle ϑ that is defined in the first
covering function of [20] focuses on a single point of a given latitude,
whereas the maximum circle radius r of a satellite that is defined and
computed based on two geometric structures (see Sec. V) is the
maximum spherical distance between its subpoint and any point in its
coverage target area. This kind of analysis will be extended to any
type of coverage areas on Earth’s surface, provided that it can be
divided into several adjacent spherical convex polygons (according
to certain rules). Also in such cases it will be easy to determine
whether or not the constellation can completely cover the ground
Fig. 1 Coverage area of a satellite on Earth. area. Moreover, if condition (2) holds for a given constellation at any
time of a given reconstruction period, then the constellation fulfills
the single continuous coverage property for the given ground area.
Downloaded by UNIVERSITY OF ADELAIDE on August 21, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.A33826

d  Re α. Figure 1 illustrates the coverage area for a typical

observation activity of a satellite S located at an orbital altitude H for
a minimum elevation angle γ min . The well-known relation of the IV. Spherical Geometric Subdivision
coverage angle α with the orbital altitude H and the minimum A. Voronoi Diagrams and Delaunay Triangulations
elevation angle γ min is given by The Voronoi diagram and the Delaunay triangulation are two
classical geometric structures in computational geometry [31,32],

 which produce partitions of a given region or set of points in a metric
Re cos
γ min 
α  arccos − γ min (1) space. Because of their interesting properties they are widely used in
Re  H various areas such as multispectral sensor image processing.
Moreover, the closest point property of Delaunay triangulations is
Relation (1) indicates that, for the case of one single satellite, when iteratively used in several graph matching and image registration
the minimum elevation angle γ min is fixed, the size of the coverage methods, such as the ones proposed by Zhao et al. [33].
circle depends only on the orbital altitude H of the satellite. The general definition of a Voronoi diagram is as follows. Let
Consider now three satellites at the same orbital altitude. Let A, B, O  fO1 ; O2 ; : : : ; On g be a finite set of points (set of generators) in
and C denote their respective subpoints. Because all satellites are at some metric space M with distance function d. The Voronoi region
the same orbital altitude, their coverage areas have the same spherical associated with the point Oi is the set VR
Oi   fO ∈
radius d. Let ΔABC be the spherical triangle on the Earth’s surface Mjd
O; Oi  ≤ d
O; Oj ; for j  1; : : : ; N; j ≠ ig, that is, the set
defined by these points and let P and r be, respectively, the center and of points for which Oi is closest among the generators. The set of all
radius of the spherical circumcircle of ΔABC. Furthermore let P 0 N Voronoi regions is called the Voronoi diagram VD
O of O. It
denote the point in ΔABC that minimizes the maximum of the gives a partition of M into polygons with mutually disjoint interiors.
distances to A, B, and C. Note that when P 0 ∈ ΔABC, then P 0 For a given Voronoi diagram of O the corresponding Delaunay
triangulation DT
O of O is obtained if two points in O are connected
coincides with P. Otherwise P 0 is a point in the perimeter of ΔABC.
by a shortest line if and only if their corresponding Voronoi regions
The two cases are illustrated in Fig. 2, where the coverage areas of
have a nonempty intersection. Under appropriate conditions that are
satellites A, B, and C intersecting ΔABC are delimited by the blue,
usually satisfied, two Voronoi regions do not intersect at all or intersect
green, and red arcs, respectively. The exact definition and generation
in more than one point and DT
O has 3N − 6 edges defining 2N − 4
approach will be given below. triangles covering the space M (where triangles can only intersect in
Thus, in order to have full coverage of ΔABC with the three lines) [22]. The Delaunay triangulation thus also provides a partition of
satellites it is enough to ensure that the point P 0 is covered, that M (into triangles with mutually disjoint interiors).
is, d ≥ r. Both concepts are dual to each other. It is also possible to first
This condition can be extended to the case of a constellation compute the Voronoi diagrams directly and then obtain the Delaunay
consisting of multiple satellites in circular orbit at the same orbital triangulation. There are many efficient algorithms for computing
altitude. Suppose that the Earth’s surface can be divided into several either of the two and we do not elaborate on them here [34–37].
adjacent spherical triangles (with pairwise disjoint interiors) by In our application we have a constellation of N satellites C 
connecting the subpoints of the satellites according to certain rules fS1 ; S2 ; : : : ; SN g and we consider Voronoi diagram and Delaunay
(see Sec. IV). Then, only the points P 0 i and the maximum distance ri triangulation for the set O  fO1 ; O2 ; : : : ; ON g of their respective
of each spherical triangle must be computed, and if the coverage subpoints. So the metric space is the surface of the Earth and distances
radius d of the satellites satisfies are spherical distances. Accordingly, we speak about the spherical
Voronoi diagram and the spherical Delaunay triangulation of O.
Note that Voronoi regions are (spherical) convex polygons. Now,
every point on Earth’s surface belongs to a unique Voronoi region, a
unique Delaunay triangle, respectively (except for the case that a
point is on a line of the Voronoi diagram or the Delaunay
triangulation). So both VD
O and DT
O can be used to partition
Earth’s surface. For an example see Fig. 3.
It should be noted that on the sphere there are no infinite Voronoi
regions and that VD
O can be obtained from DT
O by connecting
the circumcircle centers of its triangles following certain rules. And
no point of O is in the interior of the circumcircle of any spherical
triangle of DT
O.
In this paper, we compute the Delaunay triangulation basically
Fig. 2 Spherical triangle defined by three satellites. using the approach of [34] and then generate the Voronoi diagram
 4 Article in Advance / DAI ET AL.

Fig. 3 Construction of coverage area for a constellation.

 
Downloaded by UNIVERSITY OF ADELAIDE on August 21, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.A33826

from it. By subdividing the target area into several adjacent convex z
polygons, the constellation coverage problem can be decomposed β  arctan (4)
y
into a set of single-satellite-coverage problems.
There is a natural mapping, which we denote as h, that projects the
B. Orientation Angle spherical circle Θ 0 onto the yz plane. Since there exists a one-to-one
Consider a spherical circle Θ
P; α with center point P
θ; φ and mapping g that assigns to each orientation angle β ∈ 0; 2π a unique
radius α, where θ and φ are the longitude and latitude of P, point in h
Θ 0 , it is clear that the projection h
Θ 0  is a circle on the yz
respectively. Let us rotate the coordinates such that in the new plane. Hence, g maps an angle β ∈ 0; 2π to g
β 
0; cos β; sin β.
coordinate system, the x axis passes through the center P 0 of the The inverse projection f mapping an orientation angle to a point on
rotated spherical circle. The coordinate rotation matrix is the original spherical circle Θ maps β ∈ 0; 2π to
M−1
cos α; sin α cos β; sin α sin β, where M−1 denotes the inverse
of the rotation mapping (3). Note that f can also be expressed
P 0  M
P  Ry
−φRz
θ (3) as f  M−1 ∘ h−1 ∘ g.

where M ∈ R3×3 with

V. Constellation Coverage Analysis
2 3 A key issue for the design of satellite orbits and for the evaluation of
cos
θ − sin
θ 0 the performance of constellations is the accurate computation and
6 7 analysis of the coverage for a given target area Ω. This topic is
Rz
θ  4 sin
θ cos
θ 05 and
addressed in this section where we give some definitions relating the
0 0 1 Voronoi diagram computed for a constellation C  fS1 ; S2 ; : : : ; SN g
2 3
cos
φ 0 − sin
φ of N satellites to its coverage properties. First we focus on one satellite
6 7 Si and later extend the concepts to the full constellation. As before, we
Ry
−φ  4 0 1 0 5 denote by O  fO1 ; O2 ; : : : ; ON g the set of subpoints of the
sin
φ 0 cos
φ constellation.
1) The coverage target area Ωi of Si is the intersection of the
Voronoi region VR
Oi  and the target area Ω. If Ω is a polygon, then
Figure 4 gives an illustration of the spherical circle Θ 0 after the Ωi is also a polygon. Its vertices are all the vertices of VR
Oi  in the
rotation M
Θ  fM
QjQ ∈ Θg, where now the x axis passes target area plus some “new” vertices generated by the intersection of
through the center P 0 
0; 0 of the rotated spherical circle. the boundary of VR
Oi  with the spherical arcs that define Ω.
The orientation angle with respect to P 0 of any point in the rotated 2) The maximum circle radius of Si is the maximum spherical
spherical circle is denoted by β. It is clear that the orientation angle of distance between Oi and any point in its coverage target area Ωi .
any point Q 0 
x; y; z, x ≠ 0, in the rotated spherical circle is 3) A coverage feature point of Si is any point on the boundary of Ωi
that potentially could be of maximum distance to Oi . The set of
feature points of Si is denoted by Fi. Indeed Fi contains all the
vertices of Ωi . Furthermore, when some of the lines that define Ω are
small arcs, then Fi may contain additional points, corresponding to
the properties and intersection characteristics of small arcs.
From the above definitions it follows that for any point in Ωi ,
1 ≤ i ≤ N, the subpoint Oi is a nearest subpoint among all subpoints
of the satellites. Furthermore, all the coverage target areas only
intersect in boundary points, as they are defined as subsets of Voronoi
regions.

A. Construction of the Coverage Target Area

The coverage area of one satellite can be studied directly according
to Eq. (1). Suppose now that the target area Ω must be covered
by a constellation C  fS1 ; S2 ; : : : ; SN g. To obtain the coverage area
of C we compute the Voronoi diagram for the subpoints O 
fO1 ; O2 ; : : : ; ON g and the coverage target areas Ωi together with the
feature points Fi . As an example see Fig. 3, showing the Voronoi
Fig. 4 Orientation angle computation using projection. diagram and Delaunay triangulation for the Walker constellation [6]
 Article in Advance / DAI ET AL. 5

with the configuration parameters
T∕P∕F 
48∕8∕1 and orbital and r2 < Re . Let us further assume that
O1 ; r1  covers points A and
inclination Incl  52°. B, and
O2 ; r2  covers points C and D. Because the curvature of the
Full coverage of a target area Ω is attained by a constellation if each spherical great circle is smaller than that of the spherical small circle,
⌢
satellite has full coverage of its coverage target area Ωi . When some we have
1∕Re  <
1∕r1  and
1∕Re  <
1∕r2 . Thus, arcs AB and
Si does not have full coverage of Ωi , then no other satellite can cover ⌢
any point in the uncovered part of Ωi . In this case, the constellation CD must be inside the circles
O1 ; r1  and
O2 ; r2 , respectively. It is
cannot totally cover Ω. obvious that the point of maximum distance to center point O1 ,
⌢
As a consequence of the above analysis, after the Voronoi diagram among all the points on arc AB, is either A or B. The same holds with
of O has been obtained, the constellation full-coverage problem for a ⌢
target area Ω is equivalent to a series of N independent coverage C and D for arc CD relative to O2 (see Fig. 6a).
problems, each of them restricted to its own target area Ωi . The region 3) If the boundary ∂Ωi of the coverage area belonging to satellite Si
⌢
Ωi  VR
Oi  ∩ Ω is a spherical polygon. The boundary ∂Ωi is contains a small arc segment Θ, then, in addition to the points defined
composed of several spherical great and small arc segments, which in the above item, we add to Fi the intersection points of the adjacent
are generated according to the intersection of VR
Oi  with Ω, and the small arc segments, as depicted in Fig. 7.
⌢ ⌢ ⌢
orientation angle of the feature point. The spherical arc is denoted as Suppose now that small arcs AB, CD, and EF on a spherical small
great arc segments if the center of the corresponding spherical
circle
O; r belong to ∂Ωi . Let us also assume that there is another
circumcircle is Oe , and as small arc segments otherwise.
spherical small circle
O 0 ; r 0 , where O 0 is the subpoint and r 0 < Re
Therefore, the question that arises now is how to determine
(see Fig. 6b). When the distance between O 0 and O is greater than or
whether or not a given satellite Si has full coverage of its target area ⌢
Downloaded by UNIVERSITY OF ADELAIDE on August 21, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.A33826

Ωi . The answer is given in terms of the coverage of feature points. Let equal to r, then all of the points on the small arc AB are inside
Fi  ffi1 ; fi2 ; : : : ; fim g be the set of feature points of Si and rij be
O 0 ; r 0 , provided that the spherical small circle
O 0 ; r 0  covers both
the spherical distance from subpoint Oi to feature point fij . The A and B (see
O1 ; r1  in Fig. 6b). Otherwise, two different cases may
maximum circle radius of Si is arise. In the first case we have that
1∕r ≤
1∕r 0  (see spherical
small circle
O2 ; r2  in Fig. 6b). Then, full coverage of the small arc
⌢
rmax
i  maxfri1 ; ri2 ; : : : ; rim g (5) CD is implied by coverage of both C and D. In the second case,

1∕r >
1∕r 0  (see spherical small circle
O3 ; r3  in Fig. 6b). Then,
Thus, the re-statement of condition (2) for full coverage of the ⌢
target area Ω by constellation C is neither E nor F is of longest distance to O 0 among the points on EF.
Now, to obtain such a maximum distance point, we connect O3 with
d ≥ rmax  maxfrmax
1; rmax
2; : : : ; rmax
Ng (6) O and extend this arc to a spherical great circle, which intersects with
the spherical small circle
O; r in points P and Q. Indeed it can be
where, as before, d denotes the spherical radius of the coverage areas shown that the point of maximum distance to O3 is either P or Q,
⌢
of the satellites for the coverage angle α associated with a minimum provided that they are on arc EF. Such a maximum distance point
elevation angle γ min , that is, d  Re α. also becomes one of the feature points of subpoint O3 .

B. Calculation of the Coverage Feature Points C. Continuous Coverage Analysis of Constellations

Feature points play a crucial role in the coverage problem of a The analysis of full coverage of a target area Ω by a constellation C
constellation for a target area. Each coverage target area is given by at a fixed time point performed above can be easily extended to study
the set of feature points on the Earth’s surface with shortest distance to the continuous coverage of Ω by C over a given reconstruction period.
the subpoint of a satellite. In some cases, which we list below, it is Indeed, the necessary and sufficient condition that guarantees that C
possible to easily compute the set of feature points of a given satellite. offers continuous full coverage for Ω throughout the given
1) If the coverage target area Ωi of Si is empty, then Si has no reconstruction period is that Eq. (6) holds at any time of the
coverage feature point and Fi  ∅. reconstruction period. Hence, for this analysis we will discretize the
2) If the boundary ∂Ωi of the coverage area of satellite Si contains reconstruction period and check the condition at each time point of
arc segments of spherical great circles and does not contain any small the discretized reconstruction period.
arc segment, then the set of feature points Fi is given by the Observe that the position of each satellite changes as time varies, as
intersection points of adjacent arc segments of ∂Ωi , as depicted in it moves through different points of its orbit. Hence the subpoint of
⌢ ⌢
Fig. 5. Suppose that great arcs AB and CD on the spherical great each satellite also depends on the time t and accordingly on the set of
circle with center point O and radius Re belong to ∂Ωi and that we subpoints Ot  fOt1 ; Ot2 ; : : : ; OtN g. That is, for a given minimum
also have the spherical small circles
O1 ; r1  and
O2 ; r2  with elevation angle γ min , the coverage angle αt obtained with expression
subpoints O1 and O2 and respective radii r1 and r2 satisfying r1 < Re (1), together with the associated spherical radius dt  Re αt , depends

Fig. 5 Subdivision of target area only with great arcs (like global target area and spherical polygon area).
 6 Article in Advance / DAI ET AL.

Fig. 6 Spherical arcs intersections.

Downloaded by UNIVERSITY OF ADELAIDE on August 21, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.A33826

on t. This means that, to check whether or not condition (6) holds at the Voronoi diagram of the satellites constellation. Then, for each
any time of the reconstruction period, the construction must be satellite Si ∈ S, we compute its set of feature points Fti , the spherical
repeated for each time point of the reconstruction period, including distances rtij from the subpoints to their feature points, and its circle
the procedure for obtaining the Voronoi diagram, the target areas of radius value rtmax
i. Finally we compute the maximum circle radius
the different satellites, and their feature points. Hence, the condition for the constellation rtmax  maxfrtmax
ij1 ≤ i ≤ Ng.
that guarantees full coverage for Ω throughout the reconstruction Since condition (6) holds when dt  Re αt ≥ rtmax for all t, we
period is dt  Re αt ≥ rtmax , for all t in the discretized reconstruction compute the maximum radius rmax  maxfrtmax jt ∈ Tg. Then, we
period. take as spherical radius d  maxfdt jt ∈ Tg, associated with the
coverage angle d  maxfαt jt ∈ Tg  rmax ∕Re .
From Eq. (1) we obtain
VI. Orbital Parameters Design for the Continuous
Coverage Problem Re cos
γ min 
h − Re (7)
In the previous sections we have seen that the spherical geometric cos
rmax ∕Re   γ min 
subdivision based on Voronoi diagrams allows to address the
continuous full-coverage problem, that is, the problem of
determining whether or not a given configuration for a satellite as altitude for the satellites in the constellation.
constellation provides continuous full coverage for a given
target area. B. Calculation of Orbital Inclination
As we see below, the proposed geometric subdivision can also be It is well-known that the optimal orbital inclination corresponds to
used to address the problem of finding a constellation of N satellites the minimum value of the maximum coverage circle radius in the
that provides full coverage for a given target area. As it is known, constellation reconstruction period.
there are several orbital parameters that define the properties of a Suppose that the coverage target area is located in the latitude
constellation of satellites. Eccentricity and perigee are typically fixed region ϕl ; ϕu . The maximum latitude value of a point in the target
in the design of circular orbital parameters, and so they need not be area is ϕmax  maxfjϕl j; jϕu jg. Therefore, for the full coverage of the
computed. Moreover, the initial values of the RAAN and Mean target area, the satellite must cover the point with maximum latitude
Anomaly are not considered to be important in the continuous ϕmax . Using Eq. (1), the minimum orbital inclination for a given
coverage problem, and so they can be distributed evenly. On the other orbital altitude ht and minimum observation elevation angle γ min is
hand, it is well-known that Walker’s configuration [6] produces given by
constellations with good symmetry and stability properties. Hence
for the design of a constellation for the continuous coverage problem,
we will follow Walker’s configuration and assume that satellites are Incltmin  maxfϕmax − αt ; 0g (8)
distributed in a uniform way. As a result, only the length of the
semimajor axis (altitude h) and the inclination angle Incl offer some Figure 8 illustrates the satellite coverage of the Earth’s surface in
freedom of choice for orbital design. Below, we show how to obtain the Earth-centered inertial (ECI) reference frame. It reflects the
these parameters. For our analysis we recall that, for any two satellites coverage status when the satellite passes through the highest and
in the constellation, the relative motion trajectory in space changes lowest points in the ECI frame.
periodically. Therefore, the values of orbital inclination and
maximum coverage circle radius, satisfying continuous coverage to
target, change continuously in the constellation reconstruction
period.

A. Calculation of Orbital Altitude

As we have seen, to obtain continuous full coverage for the target
area in the reconstruction period, condition (6) must hold at any time.
That is, for a given minimum elevation angle γ min , at each time instant
t of the reconstruction period, the spherical radius dt  Re αt
associated with the coverage angle αt obtained with Eq. (1) must be
such that condition (6) holds.
Next we explain how to analytically compute a value for the
altitude of the satellites in the constellation that guarantees the above Fig. 7 Subdivision of target area with small arcs (like latitude target
condition. For each time point t we obtain the set of subpoints Ot and area and spherical circle target area).
 Article in Advance / DAI ET AL. 7

Fig. 8 Satellite coverage in the ECI reference frame.

VII. Computational Experiments

In this section we analyze the Globalstar mobile satellite system,
designed by Loral and Qualcomm of America, with the geometric Fig. 10 Maximum coverage circle radius with orbital inclination angle
subdivision method proposed in this paper. This system is a Walker over time.
constellation with configuration parameters
T∕P∕F 
48∕8∕1.
Downloaded by UNIVERSITY OF ADELAIDE on August 21, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.A33826

Note that in this notation T denotes the total number of satellites and
not the reconstruction time period. The satellite orbital inclination is the coverage demand are depicted in Fig. 11. Figure 11a illustrates the
Incl  52° and the altitude is h  1414 km. It provides single variation of the minimum elevation angle with respect to the orbital
continuous full coverage of the latitude region S70°; N70° with a inclination angle, at different times under the full-coverage
minimum coverage elevation angle γ min  10° [5]. In our study we constraint. Figure 11b shows the variation of the minimum elevation
use a discretized reconstruction period of 142.6 s. angle with respect to the orbital inclination under the constraint of
continuous full coverage. From the above results we can conclude the
A. Subdivision following.
1) To satisfy the continuous full coverage of the latitude region
Because of the symmetry property of Walker constellations, we
under the constraint of minimum elevation angle γ min  10°, the
need to analyze only the coverage capacity of the Globalstar system
satellite orbital inclination is in the range 52.224°; 55.094°. So with
for the latitude region 0°; N70°. The construction of the Voronoi
the original system configuration parameter Incl  52° the latitude
diagram at t  2015–01–01 00:00:00 is shown in Fig. 9, where every region S70°; N70° cannot be covered completely.
convex polygon in the graph corresponds to the coverage area of a 2) Figures 10 and 11 give optimal inclination values. It can be seen
satellite. that the maximum coverage circle radius is rmax  25.586° and the
minimum elevation angle is γ  10.935°, while the orbital
B. Coverage Calculation inclination is Incl  53.24°.
1. Analysis of System Configuration Parameters Therefore, the configuration
T∕P∕F 
48∕8∕1 with inclina-
We re-optimize the orbital inclination according to Eq. (8) under tion Incl  53.24° can be regarded as the best constellation
the assumption that all other parameters are given and fixed. Then, we configuration parameters given the continuous full-coverage
use Eq. (6) to calculate the value of the coverage circle radius, requirements for the latitude region. In this context the minimum
satisfying the continuous full coverage of the latitude region. The orbital altitude value is h  1347.679 km.
results are given in Fig. 10. It illustrates the variation of the maximum Suppose now that all the constellation parameters are constant.
coverage circle radius relative to the orbital inclination angle at Figure 12 gives the maximum coverage gap of the constellation for
different times of the reconstruction period when full coverage is different latitude regions. The x axis represents time in the
imposed. constellation reconstruction period and the y axis gives the upper
The results of the computations of the minimum elevation angle bound values of the latitude regions 0°; y° in the graph. The shaded
corresponding to the maximum coverage circle radius that satisfies area indicates coverage gap intervals in different latitude regions. For

Fig. 9 Spherical subdivision for the Globalstar system.

 8 Article in Advance / DAI ET AL.

Fig. 11 Variation of the minimum elevation angle relative to orbital inclination.

Downloaded by UNIVERSITY OF ADELAIDE on August 21, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.A33826

Fig. 12 Maximum coverage gap of constellation in different latitude

regions. Fig. 14 Coverage computation times in relation to the partition
precision.

with the classical grid-point technique and the longitude strip

method. For the analyzed data, the computing time needed by our
method is only 20 ms and does not change with the constellation and
the type and size of coverage area. The computational results for the
grid-point technique and the longitude strip method are depicted in
Fig. 14. The x axis indicates the partition precision, given as the
number of longitude strips or grid points in 1 km on Earth’s surface.
The y axis shows the instantaneous coverage computing time for the
latitude region with the Globalstar system. Given that the grid-point
technique and the longitude strip method are both numerical
simulation methods, their major drawback is that results with high
accuracy can be obtained only with a considerable increase in
computing times and memory requirements.
Fig. 13 Minimum elevation angle with different latitude regions The computations show that the optimal system parameters
imposing continuous full coverage. obtained with our method are basically the same as the original
parameters produced by the Globalstar system.
example, the coverage gap interval during the reconstruction period
for the latitude region 0°; N70.3° is t1 ; t2 .
Further results are presented in Fig. 13 on the former basis giving VIII. Conclusions
the variation of the minimum elevation angle for different latitude This paper has proposed a new method based on spherical
regions under the constraint of continuous full coverage. The x axis geometric subdivision for the analysis of the continuous coverage of
represents the upper bound value 0°; x° of the latitude regions. The y satellites and the calculation of satellite orbital parameters for the full-
axis represents the minimum elevation angle under the constraint of coverage problem. The proposed method is exact for fixed time t and
continuous full coverage. It is clear that the original constellation works on any type of coverage target areas. Although only
configuration parameters can offer only single continuous full approximate in the constellation reconstruction period, it is also very
coverage for the latitude region S69.81°; N69.81° or less. effective for evaluating a series of further meaningful performance
indices for the continuous and the discontinuous coverage problem.
2. Efficiency Analysis (Note the exception that when computing the spherical Delaunay
For the same target area 0°; N70° of the latitude region we subdivision not all subpoints must be located in the same
compared the computational efficiency of our new coverage method semispherical surface since, otherwise, the subdivision is invalid.)
 Article in Advance / DAI ET AL. 9

The new proposed method also produces analytical solutions as [13] Ma, D. M., Hong, Z. C., Lee, T. H., and Chang, B. J., “Design of a Micro-
well as the range for each parameter value so as to satisfy the Satellite Constellation for Communication,” Acta Astronautica, Vol. 82,
coverage requirements. The satellite orbital parameters can be No. 1, Jan. 2013, pp. 54–59.
obtained directly from the maximum coverage circle radii. doi:10.1016/j.actaastro.2012.04.037
[14] Lluch, I., and Golkar, A., “Satellite-to-Satellite Coverage Optimization
Moreover, optimal satellite orbital inclinations and altitudes can be Approach for Opportunistic Inter-Satellite Links,” Proceedings of the IEEE
obtained according to the original design objectives. The results Aerospace Conference, IEEE Publ., Piscataway, NJ, March 2014, pp. 1–13.
also indicate that traditional methods based on grid-point division doi:10.1109/AERO.2014.6836307
or longitude strip partition are usually unsuitable for very large [15] Morrison, J. J., “A System of Sixteen Synchronous Satellites for
regions under high accuracy requirements. Worldwide Navigation and Surveillance,” Technical Department of
Summarizing, the decision space for an optimal constellation Transportation Federal Aviation Administration Systems Research and
design is reduced considerably, and the constellation design process Development Service, Washington, D.C., March 1973.
improves significantly. [16] Jiang, Y., Yang, S., Zhang, G., and Li, G., “Coverage Performances
Analysis on Combined-GEO-IGSO Satellite Constellation,” Journal of
Electronics (China), Vol. 28, No. 2, March 2011, pp. 228–234.
doi:10.1007/s11767-011-0581-1
Acknowledgment [17] Mortari, D., Sanctis, M. D., and Lucente, M., “Design of Flower
This work was supported by National Natural Science Foundation Constellations for Telecommunication Services,” Proceedings of the
of China under Grant No. 41571403 and No. 61472375. IEEE, Vol. 99, No. 11, July 2011, pp. 2008–2019.
doi:10.1109/JPROC.2011.2158766
[18] Casten, R. G., and Gross, R. P., “Satellite Cumulative Earth Coverage,”
Downloaded by UNIVERSITY OF ADELAIDE on August 21, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.A33826

Proceedings of the AAS/AIAA Astrodynamics Specialist Conference,

References Univelt, San Diego, CA, Aug. 1981.
[1] Rainer, S., Klaus, B., and Marco, D. E., “Small Satellites for Global [19] Xu, M., and Huang, L., “An Analytic Algorithm for Global Coverage
Coverage: Potential and Limits,” ISPRS Journal of Photogrammetry of the Revisiting Orbit and Its Application to the CFOSAT Satellite,”
and Remote Sensing, Vol. 65, No. 6, Sept. 2010, pp. 492–504. Astrophysics and Space Science, Vol. 352, No. 2, Aug. 2014,
doi:10.1016/j.isprsjprs.2010.09.003 pp. 497–502.
[2] Sandau, R., “Status and Trends of Small Satellite Missions for Earth doi:10.1007/s10509-014-1939-2
Observation,” Acta Astronautica, Vol. 66, No. 1, Jan. 2010, pp. 1–12. [20] Ulybyshev, Y., “Geometric Analysis of Low-Earth-Orbit Satellite
doi:10.1016/j.actaastro.2009.06.008 Communication Systems: Covering Functions,” AIAA Journal of
[3] Zhao, Z. Y., Cai, Y. W., Zhao, Y. B., and Li, Y., “Research on Small Spacecraft and Rockets, Vol. 37, No. 3, May. 2000, pp. 385–391.
Satellite Formation Scheme for Earth Observation Task,” Proceedings doi:10.2514/2.3572
of the 2015 International Conference on Automation, Mechanical [21] Seyedi, Y., and Safavi, S. M., “On the Analysis of Random Coverage
Control and Computational Engineering, Atlantis Press, Paris, France, Time in Mobile LEO Satellite Communications,” IEEE Communica-
April 2015, pp. 1678–1683. tions Letters, Vol. 16, No. 5, March 2012, pp. 612–615.
doi:10.2991/amcce-15.2015.299 doi:10.1109/LCOMM.2012.031912.112323
[4] Ulybyshev, Y., “Satellite Constellation Design for Complex Coverage,” [22] Ballard, A. H., “Rosette Constellation of Earth Satellite,” IEEE
AIAA Journal of Spacecraft and Rockets, Vol. 45, No. 4, July 2008, Transaction on Aerospace and Electronic Systems, Vol. AES-16, No. 5,
pp. 843–849. Sept. 1980, pp. 656–673.
doi:10.2514/1.35369 doi:10.1109/TAES.1980.308932
[5] Crowe, K. E., and Raines, R. A., “A Model to Describe the Distribution [23] Sarno, S., Graziano, M. D., and DÉrrico, M., “Polar Constellations
of Transmission Path Elevation Angles to the Iridium and Globalstar Design for Discontinuous Coverage,” Acta Astronautica, Vol. 127,
Satellite Systems,” IEEE Communications Letters, Vol. 3, No. 8, Oct. 2016, pp. 367–374.
Aug. 1999, pp. 242–244. doi:10.1016/j.actaastro.2016.06.001
doi:10.1109/4234.781008 [24] Li, T., Xiang, J., Wang, Z., and Zhang, Y., “Circular Revisit Orbits
[6] Walker, J. G., “Circular Orbit Patterns Providing Continuous Whole Design for Responsive Mission over a Single Target,” Acta Astronaut,
Earth Coverage,” Royal Aircraft Establishment, Technical Rept. 70211 Vol. 127, Oct. 2016, pp. 219–225.
(UDC 629.195:521.6), Farnborough, U.K., Nov. 1970. doi:10.1016/j.actaastro.2016.05.037
[7] Crossley, W. A., and Williams, E. A., “Simulated Annealing and Genetic [25] Mozhaev, G. V., “Capabilities of Kinematically Regular Satellite
Algorithm Approaches for Discontinuous Coverage Satellite Systems with Symmetry Groups of the Second Type in the Problem of
Constellation Design,” Engineering Optimization, Vol. 32, No. 3, Continuous Single Coverage of the Earth,” Cosmic Research, Vol. 43,
2000, pp. 353–371. No. 3, May 2005, pp. 205–212.
doi:10.1080/03052150008941304 doi:10.1007/s10604-005-0036-2
[8] Asvial, M., Tafazolli, R., and Evans, B. G., “Satellite Constellation [26] Ulybyshev, Y., “Geometric Analysis and Design Method for Discontinuous
Design and Radio Resource Management Using Genetic Algorithm,” Coverage Satellite Constellations,” AIAA Journal of Guidance, Control,
IEE Proceedings-Communications, Vol. 151, No. 3, July 2004, and Dynamics, Vol. 37, No. 2, Feb. 2014, pp. 549–557.
pp. 204–209. doi:10.2514/1.60756
doi:10.1049/ip-com:20040291 [27] Ulybyshev, Y. P., “A General Analysis Method for Discontinuous
[9] Ferringer, M. P., and Spencer, D. B., “Satellite Constellation Design Coverage Satellite Constellations,” AIAA Journal of Guidance, Control,
Tradeoffs Using Multiple-Objective Evolutionary Computation,” and Dynamics, Vol. 38, No. 12, Aug. 2015, pp. 2475–2483.
AIAA Journal of Spacecraft and Rockets, Vol. 43, No. 6, Nov. 2006, doi:10.2514/1.G001254
pp. 1404–1411. [28] Razoumny, Y. N., “Fundamentals of the Route Theory for Satellite
doi:10.2514/1.18788 Constellation Design for Earth Discontinuous Coverage. Part 1:
[10] Wang, L., Wang, Y., Chen, K., and Zhang, H., “Optimization of Analytic Emulation of the Earth Coverage,” Acta Astronautica,
Regional Coverage Reconnaissance Satellite Constellation by NSGA-II Vol. 128, July 2016, pp. 722–740.
Algorithm,” Proceedings of the 2008 IEEE International Conference on doi:10.1016/j.actaastro.2016.07.013
Information and Automation, IEEE Publ., Piscataway, NJ, June 2008, [29] Mohammadi, L., “Determination of the Best Coverage Area for
pp. 1111–1116. Receiver Stations of LEO Remote Sensing Satellites,” Proceedings of
doi:10.1109/ICINFA.2008.4608165 the 2008 3rd International Conference on Information and
[11] Ferringer, M. P., Clifton, R. S., and Thompson, T. G., “Efficient and Communication Technologies: From Theory to Applications, IEEE
Accurate Evolutionary Multi-Objective Optimization Paradigms for Publ., Piscataway, NJ, April 2008, pp. 1–4.
Satellite Constellation Design,” AIAA Journal of Spacecraft and doi:10.1109/ICTTA.2008.4530157
Rockets, Vol. 44, No. 3, May 2007, pp. 682–691. [30] Sengupta, P., Vadali, R. S., and Alfriend, T. K., “Satellite Orbit Design
doi:10.2514/1.26747 and Maintenance for Terrestrial Coverage,” AIAA Journal of Spacecraft
[12] Jiang, Y., Zhang, G., Li, G., Xie, Z., and Yang, S., “Study on Orthogonal and Rockets, Vol. 47, No. 1, Jan. 2010, pp. 177–187.
IGSO Global Communication Satellite Constellation,” 2011 6th doi:10.2514/1.44120
International ICST Conference on Communications and Networking in [31] Devillers, O., “Delaunay Triangulation and Randomized Construc-
China (CHINACOM), Aug. 2011, pp. 1064–1068. tions,” Encyclopedia of Algorithms, Springer, Oct. 2014, pp. 1–7.
doi:10.1109/ChinaCom.2011.6158313 doi:10.1007/978-3-642-27848-8 711-1
 10 Article in Advance / DAI ET AL.

[32] Okabe, A., Boots, B., Sugihara, K., and Chiu, S. N., Spatial Model Development, Vol. 6, No. 4, Dec. 2013, pp. 1353–1365.
Tessellations: Concepts and Applications of Voronoi Diagrams, doi:10.5194/gmd-6-1353-2013
2nd ed., Wiley, Hoboken, NJ, May 2008, pp. 46–105. [36] Yan, D. M., Wang, W., Levy, B., and Liu, Y., “Efficient Computation of
doi:10.1002/9780470317013 Clipped Voronoi Diagram for Mesh Generation,” Computer-Aided
[33] Zhao, M., An, B., Wu, Y., Chen, B., and Sun, S., “A Robust Delaunay Design, Vol. 45, No. 4, April 2013, pp. 843–852.
Triangulation Matching for Multispectral/Multidate Remote Sensing doi:10.1016/j.cad.2011.09.004
Image Registration,” IEEE Geoscience and Remote Sensing Letters, [37] Peterka, T., Morozov, D., and Phillips, C., “High-Performance
Vol. 12, No. 4, April 2015, pp. 711–715. Computation of Distributed-Memory Parallel 3D Voronoi and Delaunay
doi:10.1109/LGRS.2014.2359518 Tessellation,” Proceedings of the International Conference for High
[34] Zhou, L., Huang, D., Li, C., and Zhou, D., “Algorithm for GPS Network Performance Computing, Networking, Storage and Analysis, IEEE
Construction Based on Spherical Delaunay Triangulated Irregular Publ., Piscataway, NJ, Nov. 2014, pp. 997–1007.
Network,” Journal of Southwest Jiaotong University, Vol. 42, No. 3, doi:10.1109/SC.2014.86
June 2007, pp. 380–383.
[35] Jacobsen, D. W., Gunzburger, M., Ringler, T., Burkardt, J., and Peterson, E. G. Lightsey
J., “Parallel Algorithms for Planar and Spherical Delaunay Construction Associate Editor
with an Application to Centroidal Voronoi Tessellations,” Geoscientific
Downloaded by UNIVERSITY OF ADELAIDE on August 21, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.A33826