IWSCFF 19-1983

A REVIEW OF SATELLITE CONSTELLATION

RECONFIGURATION AND ITS APPLICATIONS

Olivier L. de Weck*

This paper provides a summary of key concepts in satellite constellation reconfiguration. This
includes a review of the literature over the last 15 years from the author’s perspective and its
applicability to the next generation of mega constellations such as OneWeb, Starlink, and the
next generation of Earth Observation constellations (such as TROPICS). We provide a new h-
T diagram to visualize satellite constellation reconfigurability at the system level. Particular
emphasis in the future will be placed on the potential of electric propulsion and actively using
perturbations such as atmospheric drag and J2-effects in a future proliferated Low Earth Orbit
(LEO) environment. Amongst different methods for designing for reconfigurability we discuss
real options, we review optimization algorithms such as Simulated Annealing (SA) as well as
the auction algorithm as a means for optimizing constellation reconfigurability. A particularly
useful way to reconfigure is to switch between drifting and repeating ground track (RGT)
orbits to improve revisit statistics over key points of interest, a concept known as ReCon.

STATIC CONSTELLATION DESIGN

The design of satellite constellations goes back as far as the 1970s and 1980s with seminal
papers by Adams and Rider (polar constellations, also known as Street of Coverage SOC [1])
and Walker constellations (in inclined orbits [2]), amongst others. An important early satellite
constellation was the Global Positioning System (GPS) which was operationally deployed in
the mid 1980s. The definition of a satellite constellation is:
A satellite constellation is an on-orbit ensemble of satellites that is coordinated such that
coverage and functionality emerge in a way that could not be provided by individual sat-
ellites alone.
The minimum number of satellites in a particular
orbital plane can be estimated through the follow-
ing geometry (see Fig. 1) to ensure continuous
coverage (at least n=1 fold):

Rsat = Re + h

sin(η) = Re/Rsat η=sin-1(Re/Rsat)

β=π/2−η=
π/2−sin-1(Re/Rsat)
=cos-1(Re/Rsat) Smin=ceil [π/β]

where Smin is the minimum number of satellites per

plane assuming a zero elevation angle. Figure 1: Single circular plane geometry

*
Professor of Aeronautics and Astronautics and Engineering Systems, Department of Aeronautics and Astronautics,
Room 33-410, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA
Figure 2 shows the theoretical minimum number of satellites Smin required per orbital plane,
along with the actual number for a selected number of constellations.

Figure 2: Minimum number of satellites per orbital plane (zero elevation angle)
The total number of satellites in the constellation then depends additionally on the number of
orbital planes. A reference satellite constellation design module was developed and bench-
marked [3,4], see Figure 3, for circular and hybrid constellations. Classic satellite constellations
are static, that is the number of satellites does not change over the life of the constellation and
the orbital elements of each satellite - their two line elements TLE - are frozen (except for
natural perturbations such as atmospheric drag). In Fig. 3 the arrow A à B indicates the possi-
bility of reconfiguring a constellation from a configuration A to B.
RECONFIGURABILITY Benchmarking of Constellation Module
4
There are a number of reasons why 10
satellite constellations should be re- Legend
configurable: o Walker
x SOC
Globalstar
Staged deployment over
Orbital Altitude [km]

•
time to phase a satellite Iridium
constellation during devel- 3
10 n=4
opment or as a function of A
n=3
evolving demand [3] B n=2
• Preferential revisit statistics Walker dominates
by favoring certain points n=1
SOC dominates
of interest (POI) on Earth in Curves: ε=0
terms of coverage [17,18]
2
• Compensating for defunct 10
10 20 30 40 50 60 70 80 90 100
satellites to self-repair the
Number of Satellites
constellation (this could
be accomplished by a Figure 3: Optimization of static satellite constellations
combination of the first
two aspects).
This paper provides and overview of past and ongoing research in satellite constellation
reconfiguration and sets out a few directions for future work. Figure 4 shows the original GPS

2
deployment schedule which was executed by the U.S. Air Force in three phases between 1975
and 1984. The constellation was placed in 12-hr orbits (h=20,200 km) with initially 6 satellites
(phase 1) then 11 satellites (phase 2) and finally 24 satellites for full operational capability after
1984. In this case the deployment of the constellation in phases was pre-planned.

Figure 4: Initial Global Positioning System (GPS) Deployment Schedule (1975-1984)

Following the bankruptcy of Iridium and Glob-
alstar it was proposed that satellite constella-
tions should contain flexibility in order to
change the number of satellites as a function of
demand (rather than being pre-planned). In this
case a flexible staged deployment strategy can
be evaluated using a real options approach [3].
Figure 5 shows an optimal staged deployment
strategy for satellite constellations under uncer-
tainty. An optimal deployment path here in-
cludes a stage 1 (h=2000 km, 21 satellites, 3
planes), followed by a stage 2 (h=800 km with
50 satellites in 3 planes), followed by a stage 3
(h=400 km with 112 satellites in 8 planes). The
lifecycle cost for such a telecommunications
constellation is between $1-2 billion. The num-
ber of subscribers in this case can vary between
about 5,000 and about 300,000. All satellites
across stage 1, 2 and 3 are of identical design. It
is interesting to observe that as the constellation
grows it becomes more suboptimal compared to
a static (fixed) constellation that is not reconfig-
urable. Figure 5: Staged Deployment of Satellite Con-
stellations as a form of Reconfigurability [3]

3
 The main difference between Fig. 4 (GPS) and Fig. 5 (on-demand stage deployed configu-
ration) is that in the former case the stages of deployment are pre-programmed and are known
at the start, while in the second case the reconfiguration may or may not be triggered depend-
ing on an uncertain variable (in this case subscriber demand), which is set of real options.
The second major reason why satellite constellations may want to reconfigure is to im-
prove the revisit statistics (guaranteed maximum time between adjacent ground passes)
[17,18]. One way to do this is to switch between a so-called regional observation mode
(ROM) that uses repeating ground-track orbits and a global observation mode (GOM). Given
the large number of variables involved, which are a mix of discrete and continuous variables
the design space for reconfigurable constellations tends to be large and complex. Figure 6
shows the application of simulated annealing (SA) to reconfigurable satellite constellation op-
timization. In order to apply SA it is necessary to define an appropriate fitness function.

Figure 6: Simulated Annealing of optimal satellite constellation reconfiguration [17]

4
AUCTION ALGORITHM
As the number of satellites grows the optimal reconfiguration of a constellation from a config-
uration A to a configuration B (see Fig. 7 left), becomes challenging. One way to optimally
solve the problem is through the so called auction algorithm [5] (see Fig. 7 right). In this algo-
rithm we have, on the left side, existing satellites TA in constellation A (each with their own
Two Line Elements TLE) and new to-be launched satellites TB- TA. On the right side we have
TB slots to be populated in constellation B. Each assignment from the left side to one of the
open slots on the right side has a “cost” in terms of ΔV. For example, ΔV1j is the cost of moving
the first satellite from A to the j-th slot in constellation B.

Figure 7: left - reconfiguration from A to B, right - posing the problem as an auction

The auction algorithm has been used to create a Constellation Reconfiguration Map, see Fig.
8. Depending on the type of reconfiguration, the average quantity of ΔV (per satellite) necessary
to achieve the maneuvers of the on orbit satellites can be very different. Different types
reconfigurations were considered: reconfigurations in altitude, reconfigurations in inclination,
and reconfigurations in both altitude and inclination. Reconfiguration in RAAN was not
explored explicitly, but it is an implicit function of the number of orbital planes in A and B. A
reconfiguration in altitude only conserves the type of constellation (polar SOC or Walker). An
example of reconfiguration in altitude is the reconfiguration from a GEO polar constellation
into a MEO polar constellation. Inversely, a reconfiguration in inclination conserving the
altitude can change the constellation type. See Figure 8 for such as constellation reconfiguration
map.
The problem of determining the optimal assignment for transferring on-orbit satellites in con-
stellation A to constellation B is such that the total ΔV for the reconfiguration is minimized. It
has been shown that the auction algorithm, used for solving general network flow problems,
can efficiently and reliably determine the optimum assignment of satellites of A to slots of B.
Based on this methodology, reconfiguration maps can be created, which show the energy re-
quired for transforming one constellation into another as a function of type (Street-of-Cover-
age, Walker, etc..), altitude, ground elevation angle and n-fold of coverage. Suggested exten-
sions of this past work include quantification of the tradeoff between reconfiguration time and
ΔV (also listed in this paper as future work), multiple successive reconfigurations, balancing
propellant consumption within the constellation during reconfiguration as well as using recon-
figurability as an objective during initial constellation design.
Fig. 8 suggests that reconfiguring constellations in LEO is most expensive, > 3 km/s per satel-
lite, particularly when plane changes are involved. For example, reconfiguring a 16/8/5
Walker constellation at 60-degree inclination with a minimum elevation angle of 5 degrees
into a 24/4/0 satellite constellation in polar orbit is prohibitive. However, reconfiguring a

5
18/6/2 Walker constellation in MEO to a GEO Walker constellation at 21/7/3 requires less
than 2 km/s and may be feasible.

Figure 8: Satellite Constellation Reconfiguration Map (units of ΔV are [km/s])

REGIONAL OBSERVATION MODEL (ROM) versus Global Mode (GOM)

Remote sensing missions for disaster monitoring or reconnaissance are usually target-agnostic
in that target locations are unknown a priori. Therefore, reconfigurability in satellite orbits can
increase responsiveness and quality of imagery data. An optimization tool for reconfigurable
constellations (ReCon) was developed by Legge, Paek, Smith, Miller, de Weck [17, 18] and
others. It is still currently under development, e.g. by Lowey and Steighner et al. The tool
concurrently optimizes the individual satellite design and the constellation geometry.
The proposed concept of operations of ReCon has two operational modes: global observation
mode (GOM) and regional observation mode (ROM). Satellites in GOM have drifting ground
tracks that provide a global coverage, and ROM features repeating ground tracks (RGT) that
increase the access frequency to a particular target.
A weighted-sum genetic algorithm (GA) is used to identify non-dominated optimal solutions
along a Pareto front in the multi-objective tradespace. GA is an alternative, albeit less effi-
cient algorithm than the SA shown in Figure 6. Figure 9 compares satellites in ROM (top) and
GOM (bottom) modes. The difference is clear. In ROM the satellites in the constellation have
(nearly) overlapping ground tracks which favor certain nadir points (defined by their latitude
and longitude) at the expense of other locations on Earth’s surface that are not visited often.

6
 Figure 9: top - Regional Observation Mode, bottom - Global Observation Mode
Figure 10 describes a concept of operations (ConOps) over the life cycle of ReCon. First, sat-
ellites are launched into a GOM orbit which is a default mode when there is no specific target
to observe. When a target of interest is identified on the ground, the satellites switch to ROM
and observe the target with increased access. In this mode the satellite changes altitude to
move to a repeating ground track (RGT) orbit. Such orbits meet the following condition.
Ground tracks repeat themselves when an Earth nodal day and one period of a satellite orbit
make a ratio of two integers. This synchronization ratio, or the RGT ratio, is defined by:

where a satellite makes NS revolutions around the Earth while the Earth rotates around its axis
ND times, i.e., in ND days. Similarly, TS is the orbital period of a satellite and TD is the Earth
nodal day. For orbital elements, n is the mean motion of a satellite; dM/dt (=n+Δn ) is the per-

7
turbed mean motion; ωE is the rotation rate of the Earth; dω/dt is the drift rate of the argu-
ment of perigee; and dΩ/dt is the nodal regression rate. When perturbations up to J2 terms are
considered calculations are more complex, see [17,18] for details.
A number of possible ROM (RGT) orbits that are sun-synchronous are shown in Table 1.
Table 1: Sun-synchronous RGT Orbits (altitude shown is hRGT)

After being in ROM and completing their

mission all satellites return to GOM again af-
ter target observation, and this reconfigura-
tion occurs until their fuel runs out. At the
end of life, the ReCon satellites perform de-
commissioning re-entry to altitude hD.
The simplifying assumptions are as follows:
• Observation mode changes always occur in
all satellites. Partial moves are not considered
• Both GOM and ROM are circular orbits
with same inclinations.
• Transitions between two modes are per-
formed by a two impulse Hohmann transfer. Figure 10: CONOPS for ReCon
Optimal ReCon results are shown in Figure
11 in terms of trading off overall constellation mass (y-axis, objective J2) versus revisit time
(x-axis), also allowing for staged deployment. Revisit times can be reduced below 1 hour.

Figure 11: Sample results of ReCon optimization for GOM and ROM

8
The constellations corresponding to the configurations in Paths A (solid line) and B (dashed
line) are shown in Figure 12. It can be seen that for Path A going from 3 satellites (3/3/1) to
eventually a (18/3/1) constellation improves revisit time to about 0.75 hrs (#9). In path B the
endpoint is constellation #12 with a 24/6/1 constellations with a 30 min (0.5 hr revisit time).

Figure 12: Walker constellations shown for ReCon (i: T/P/R in Walker notation)
This kind of reconfiguration from ROM to GOM mode is currently actively being considered
for both reconfigurable civilian and military Earth Observation missions.
CONCLUSIONS (incl. h-T diagram)
Rather than designing satellite constellations only for a static mission design point, the idea of
reconfigurability has been proposed and intensely researched over the last 15 years. Reconfig-
uration can be done for several reasons such as staged deployment (phasing a constellation
into existence either in a pre-planned or dynamic fashion), to give preferential coverage to
certain points on Earth and also to compensate for on-orbit satellite failures or to manage pro-
pellant consummation at the constellation level. We have developed a better understanding of
the different kinds of satellite constellation reconfigurations and can now visualize those on
an h-T chart, see Figure 13. Reconfigurations in the h-T chart are horizontal (same orbit, but
different number of satellites), or vertical (same number of satellites but different orbits).

Satellite Constellations
100000

(3) GPS
Orbital Altitude [km]

10000

Staged Deployment
(2)
SD1 (1) Globalstar (4) OneWeb
1000 (1) Iridium
#1 #9 #12 (4) Starlink
GOM Path A
ROM ROM (2)
Path B GOM SD2 (2)
(5) ReCon SD3

100
1 10 100 1000 10000
Number of Satellites in the Constellation T [-]

Figure 13: Types of satellite constellation reconfigurations (h-T diagram)

9
A diagonal reconfiguration path on the h-T diagram indicates that both the orbit and the num-
ber of satellites is changing. This is the most aggressive type of reconfiguration. The numbers
in parentheses in Fig. 13 indicate the following types of constellations.
1 – Classic static constellation (Iridium, Globalstar). These constellations were launched in
the 1990s and early 2000s and have a few dozen satellites. They are not reconfigurable.
2 - Flexible staged deployment strategy (real options) with altitude change. This type of re-
configuration allows to both change the altitude and number of satellites at the same time.
SD1, SD2 and SD3 shown in Fig. 13 correspond to Stages 1, 2 and 3 in Fig. 5.
3 – Pre-planned staged deployment (see GPS phase 1, 2 and 3). This is a pre-planned staged
deployment (see Fig. 4 for GPS) with no inherent reconfigurability after orbital insertion.
4 – New generation of Mega-constellations (OneWeb, Starlink, etc..). These constellations
will have hundreds or thousands of satellites and will need to consider multiple levels of re-
configuration, including repositioning within a plane to compensate for failed satellites.
5 – Regional to Global Constellation Reconfiguration ReCon (ROM ßReConàGOM type
reconfigurability). Path A is shown from #1 to #9, Path B from #1 to #12, see also Fig. 12.

FUTURE DIRECTIONS
Recent developments and announcements suggest that there is a new wave of satellite constel-
lations coming and renewed interest reaching well into the early 2020s [11-14]. For example,
OneWeb is launching a constellation with 648 satellites (not counting spares), Starlink (Space
X) plans to deploy initially 1,600 satellites and amazon has recently announced a constella-
tion with over 3,000 satellites (Project Kuiper). At this large number of satellites, embedded
autonomy and reconfigurability will be essential features.
We see the following topics as the next frontier for satellite constellation reconfiguration re-
search. This research will require both modelling and simulation, but also online experimenta-
tion and validation with existing satellite constellations:
• Large satellite launches per orbital plane with large launchers (e.g. 36 satellites per
launch as in OneWeb), versus on demand replacement of one or two satellites with
dedicated small launchers. What is the optimal launch strategy?

• What is the impact of electric propulsion on capabilities of reconfigurable constella-

tions? What are optimal spiral transfer scenarios (e.g. from ROM to GOM) and how
competitive are they compared to impulsive Hohmann transfers? How can we best
trade off energy consumption (power levels) and reconfiguration time?

• Peer-to-peer tasking in satellite constellations where satellites in the same and across
planes can auto-plan observations and uplinks and downlinks and send each other
tasks dynamically, potentially bypassing ground stations. Bidding mechanisms (such
as the auction algorithm) in satellite constellations, eventually moving to federated
satellite systems (FSS). Application of reconfigurability to constellations of nanosat-
ellites for Earth Observation [6-10].

• High performance computing (HPC) for mega-constellations. How can we use surro-
gate models to handle reconfigurability for hundreds or thousands of satellites? How
can we visualize reconfigurations in mega-constellations? How to handle the recon-
figuration of hybrid constellations that may contain a mix of LEO and GEO or circu-
lar and elliptical orbits [4]?

10
 • Autonomous collision avoidance due to space debris [11]. This is related to the
emerging field of space traffic management and what some call proliferated LEO.

• Economic aspects of satellite constellation reconfiguration. When is it worth it?

When does it add too much complexity and cost? Tradespace Analysis for Constella-
tions (TAT-C) with reconfigurability [10]

• Technology Roadmapping for reconfigurable future satellite constellations [12]. Key

technologies such as electric propulsion, optical laser communications, digital beam-
forming (phased array antennas), smart materials and software defined radio SDR),
among others, have the potential to make important contributions to satellite reconfig-
urability at the individual satellite, but also at the constellation level.

• Multidisciplinary Design Optimization (MDO) for concurrent satellite design, launch

strategy and on-orbit operations for reconfigurable constellations [15].

• Reconfigurability in satellite constellations that are designed to track fast-moving

phenomena (e.g. such as cyclones, diurnal cycles, seasonal phenomena like snow-
melt) and disasters (earthquakes, floods, conflict zones) [16].

REFERENCES

[1] W.S. Adams, L. Rider, “Circular polar constellations providing single or multiple cover-
age above a specified latitude”, The Journal of the Astronautical Sciences, 35 (2) (1987) 155–
192

[2] J.G. Walker, “Some circular orbit patterns providing continuous whole Earth coverage”,
Journal of the British Interplanetary Society, 24 (1971) 369–384

[3] de Weck, O.L., de Neufville R. and Chaize M., “Staged Deployment of Communications
Satellite Constellations in Low Earth Orbit”, Journal of Aerospace Computing, Information,
and Communication, 1 (3), 119-136, March 2004

[4] Underwood J., de Weck O.L., Chan S., Samuels A., Shah N. “Optimization of Hybrid Sat-
ellite Constellations using Multiple Layers and Mixed Circular-Elliptical Orbits”, AIAA-
2004-3205, 22nd AIAA International Communications Satellite Systems Conference and Ex-
hibit, Monterey, California, May 9-12, 2004

[5] de Weck, O.L., Scialom, U. and Siddiqi A., “Optimal reconfiguration of satellite constel-
lations with the auction algorithm”, Acta Astronautica, 62 (2-3), 112-130, January/February
2008

[6] Nag S., de Weck O., "Tradespace Exploration of Distributed Nanosatellite Formations for
BRDF Estimation", 7th International Workshop on Satellite Constellations and Formation
Flying (IWSCFF), Lisbon, Portugal, March 15 2013 - best paper award

[7] Nag S., LeMoigne J., de Weck, “Cost and Risk Analysis of Small Satellite Constellation
for Earth Observation”, DOI: 10.1109/AERO.2014.6836396, 2014 IEEE Aerospace Confer-
ence, Big Sky, Montana, March 1-8, 2014

[8] Nag S., Gatebe C. K., Hilker T., Hall F.G. de Weck O.L., “Measuring Bi-Directional Re-

11
flectance for Gross Primary Productivity with a Constellation of SmallSats”, American Geo-
physical Union (AGU) Fall Meeting, December 15-19, 2014

[9] Paek S., Kronig L., de Weck O., “Satellite Constellation Design for the SOLVE Mission
Investigating Diurnal Cycles of Vegetation Phenomena”, 26th AAS/AIAA Space Flight Me-
chanics Meeting, AAS 16-360, CA, February 14-18, 2016

[10] Le Moigne J., Dabney P., de Weck O., Foreman V., Grogan P., Holland M., Hughes S.,
Nag S., “Tradespace Analysis Tool for Designing Constellations (TAT-C)”, Paper
#3108, 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), Fort
Worth, Texas, July 23–28, 2017

[11] Foreman V.; Siddiqi A.; de Weck O., “Large Satellite Constellation Orbital Debris Im-
pacts: Case Studies of OneWeb and SpaceX Proposals”, AIAA-2017-5200, AIAA Space 2017,
Orlando, Florida, 12-14 September 2017

[12] Foreman V., Siddiqi A., de Weck O.L., “Enabling Technologies for Modern Low Earth
Orbit Constellations”, Paper DY3-2, Council of Engineering Systems Universities Sympo-
sium, CESUN 2018, Tokyo, Japan, 20-22 June 2018

[13] Foreman V., Siddiqi A., de Weck O., “Advantages and Limitations of Small Satellites in
Low Earth Orbit Constellations: A Prospective Review”, 32nd Annual Small Satellite Confer-
ence, Logan Utah, 4-9 August, 2018

[14] Foreman V., Siddiqi A., de Weck O., “From International Space Station to International
Constellations: A new paradigm for cooperation for Earth Observation?”, IAC-
18.B1.1.2, 69th International Astronautical Conference IAC, Bremen, Germany, 1-5 October
2018

[15] Paek S.W., de Weck O., Smith M.W., “Concurrent Design Optimization of Earth Obser-
vation Satellites and Reconfigurable Constellations”, Journal of the British Interplanetary So-
ciety (JBIS), 70(1), 19-35, 2017

[16] Paek, S.W., Kronig, L.G., Ivanov, A.B. and de Weck, O.L., “Satellite constellation de-
sign algorithm for remote sensing of diurnal cycles phenomena”, Advances in Space Re-
search, 62 (9), pp. 2529-2550, November 2018

[17] Paek, S.W.; Kim, S.; de Weck, O., “Optimization of Reconfigurable Satellite Constella-
tions Using Simulated Annealing and Genetic Algorithm”, Sen-
sors, 19(4), 765, https://doi.org/10.3390/s19040765 , February 2019

[18] S.W. Paek, R.S. Legge and M.W. Smith, “Reconfigurable Satellite Constellations
for Geospatially Adaptive Earth Observation Missions”, 7th International Workshop on Satel-
lite Constellations and Formation Flying (IWSCFF), Lisbon, Portugal, March 15 2013

Selected Nomenclature

h – orbital altitude [km]

n – number-fold of satellite coverage from a spot on Earth’s surface [-]
s – slant range [km]
T – number of satellites in the constellation [-], orbital period [s]
ΔV – velocity change [m/s]
β – central angle [-]

12