Tradespace Exploration of the Next Generation
Alexa Aguilar∗, Patrick Butler*, Jennifer Collins*, Markus Guerster*, Bjarni Kristinsson*, Patrick McKeen*, Kerri
Massachusetts Institute of Technology, Cambridge, MA 02139
With this paper, we describe a tradespace exploration analysis for the next generation con-
stellation of communication satellites resulting in a recommendation for a future system. In
particular, we compare our proposal with ViaSat-3 and SpaceX’s Starlink constellation. In
order to arrive at a recommendation for an optimal constellation design, we ﬁrst identify the
design space by creating a morphological matrix and applying necessary constraints (see Table
1 for the architectural decision). The morphological matrix decisions are selected based on
variability in heritage versus state-of-the-art designs, and include options with diﬀerent Tech-
nology Readiness Level (TRL). The resulting 3,120 feasible architectures are evaluated using
both cost and performance estimates. Costs are determined from component costs, TRL, and
heritage. Performance scoring is based on a modiﬁed Signal-to-Noise Ratio (SNR) calculation,
which includes technical factors such as the downlink budget and latency, as well as system
factors such as crosslinks, architectures, and coverage. The ﬁnal design recommendation is a
Radio Frequency (RF) crosslink, bent pipe, Ka/Ku-band satellite with an electronically steered
antenna and projected mass of 125 kg. The system is a constellation of 312 satellites, spread
across 6 orbital planes at 444 km of altitude with global coverage and an estimated system
capacity of about 2 Tbps. Estimates place the cost at $8.9 billion with a NPV of $1.4 billion
over a total lifetime of ten years. Latency is expected to be around 25 ms. As with many space
systems, our proposed design comes with a number of risks. Outside of typical regulatory,
technological, and programmatic risks, providing satellite communications, particularly data
services, comes with a unique risk: the price of user terminals. In order to provide public
consumer broadband, in addition to other attractive markets such as 5G, the price of user
terminals must decrease to an aﬀordable price of $100 per terminal.
BW = Bandwidth
DoD = Depth of Discharge
DR = Data rate
FBOL = Beginning of Life Power Flux Density
FEOL = End of Life Power Flux Density
GRX = Receiver Gain
GT X = Transmitter Gain
k= Boltzmann’s Constant
Lat m = Atmospheric Losses
LID = Lithium Ion Density
Lpat h = Path Losses
LRX = Receiver Losses
LT X = Transmitter Losses
ηtr a ns f e r = Power transfer eﬃciency from solar cell to load
ηsol ar cell = Solar cell eﬃciency
Pgen = Orbit average power generated
∗Graduate Student, Aeronautics and Astronautics Engineering, MIT, 77 Massachusetts Ave, Cambridge, MA 02139, AIAA Student
†Associate Professor, Aeronautics and Astronautics Engineering, MIT, 77 Massachusetts Ave, Cambridge, MA 02139, AIAA Fellow
Ford Professor of Engineering, Aeronautics and Astronautics Engineering, MIT, 77 Massachusetts Ave, Cambridge, MA 02139, AIAA Fellow
PTX = Transmit Power
PRX = Received Power
SN R = Signal to Noise Ratio
Tecli ps e = Eclipse period per orbit
Tsys = System Temperature
by Morgan Stanley show the market for satellite communications is expected to grow exponentially
from $24B in 2018 to $128B by 2028 (see Figure 1)[
]. Driving growth in this market are upgrades to 5G networks,
where satellite communications can play an important role in backhauling, tower feed, mobility, and hybrid multiplay, as
well as consumer broadband [2–4].
Fig. 1 Predicted growth in the satellite services market over the next 20 years.
Demand for data continues to increase for consumers, with data traﬃc from mobile devices increasing at over 50%
per year. Consumer broadband is also expected to be an area for growth, with nearly three billion people that do not have
internet access or are underserved. In the US alone, over 24 million people do not have access to terrestrial broadband
and about 14 million lack LTE broadband access [
]. Only two million people globally are currently being served by
satellite internet, according to the 2017 Satellite State of the Industry Report [
]. Satellite-based consumer broadband is
well-suited to provide internet to these individuals, given the diﬃculties and marginal proﬁt of terrestrial networks in
rural areas or developing nations.
In order for the satellite industry to capture the additional demand, new, revolutionary and cost-eﬀective constellations
are necessary. Many new constellation are being proposed or developed, but which of those provide the best combination
of performance and cost? And are there additional constellation architectures that dominate those designs? With this
paper we aim to address those questions.
Previous works have shown processes for analyzing complex constellation systems, and methods for modeling
key performance metrics and cost inﬂuences. [
] developed a model similar to the one presented in this work that
incorporates both performance and cost metrics to analyze a large number of satellite constellation architectures for
the SCaN program. While [
] emphasized the communication link to evaluate constellation performance, crosslink
functionality was not considered, non-dominated architectures within the generated tradespace were not identiﬁed,
and a method for system selection using the tool was not shown. Elbert systematically formulated the key objectives
of a satellite based communication constellation from the business standpoint. The enumeration of the constellation
architectures was constrained to sun synchronous orbits. The trade space was visualized in the mapping of datarate and
constellation cost to identify the Pareto front and downselected designs [
]. This paper greatly expands the design space,
relaxes the orbital constraint and includes key emerging technologies of tomorrows communication constellations.
We aim to approach the above questions more systematically by using a tradespace exploration technique to compare
many alternative architectures with respect to performance and cost.
First, in Section III we identiﬁed the design space by creating a morphological matrix and applying constraints.
After narrowing down the tradespace from 15,552 concepts to 3,120 constrained systems. Systems were evaluated
for both cost and performance estimates. Costs were determined from components, TRL, and heritage. Performance
scoring was based on a modiﬁed link budget equation which encapsulated crosslinks, architectures, coverage, and other
relevant factors. Using normalized cost and performance metrics, the Pareto Front was identiﬁed. Then, in Section
IV, we downselected into four promising systems ranging signiﬁcantly in performance and costs. Section V describes
ﬁnancial, regulatory, marketing, and risk considerations during the implementation of those systems. Finally, Section
VI concludes the paper.
III. Conceptual Design
A. Tradespace Overview
The design space of a Lower Earth Orbit (LEO) constellation was broken down into a morphological matrix, as
shown in Table 1. In this table, the rows are potential decision to make when deﬁning the system, and the columns
represent one of four (or fewer) options for a corresponding decision. The decisions and their options are based on a
review of other systems, and are discussed further below.
Table 1 Morphological Matrix
Decision Option 1 Option 2 Option 3 Option 4 # Options
Crosslink Type None RF Optical - 3
Parabolic or Horn
Steered Laser - 3
Coverage Global (pole to
Eﬀective global (+/-
Relay Type Bent Pipe Regenerative - - 2
None Ring Mesh - 3
Yes No - - 2
User Terminal Type Parabolic Optical Electronically
Steered - 3
Mass <10 10 −100 100 −500 500+4
Frequency Band X band Ku/Ka band V band Optical 4
The decisions in the morphological matrix are based on a review of existing and proposed satellite telecommunications
]. Given the goal of designing an advanced LEO system that competes with Viasat-3 and Starlink, many
of the systems under consideration have not yet been deployed (e.g., Telesat, Starlink, Boeing). Only a handful of LEO
telecommunications constellations (Iridium, Globalstar, Orbcomm) have been implemented, and none oﬀer a data rate
comparable to the FCC minimum download speed of 25 Mbps .
One of the ﬁrst decisions we arrived at was whether the design should transmit data by crosslinking between satellites
and, if so, how it should be accomplished. Many proposed designs such as Starlink feature crosslinks between satellites
, however others, such as O3b, only transmit and receive data during ground station passes . This led to the decision
regarding crosslink type, with the options of none (no crosslink), an RF crosslink, or an optical/laser-based crosslink.
In the event of a crosslink decision, a conditional decision known as constellation architecture must be selected.
The ﬁrst option, none, accounts for no crosslink implementation with "none". Another option is to have each satellite
only link to the satellite leading and lagging within the orbital plane, known as a ring architecture. Finally, the mesh
architecture captures the scenario in which satellites can communicate across planes, creating a mesh architecture.
Communication satellites can be classiﬁed into two types of data relays: "bent pipe" or "regenerative". A bent pipe
relay ampliﬁes and reroutes the received signal, while a regenerative relay demodulates, decodes, encodes, modulates,
and ampliﬁes the received data, which raises the SNR relative to the alternative [
]. This applies to constellations with
and without crosslink capability.
The technology used for uplinks and downlinks is another key trade. Optical communications oﬀer higher data rates
and comparable SWaP to their RF counterparts, but are not robust to cloud-cover or weather. RF communications have
less stringent pointing requirements and may be used in a range of weather conditions, though higher frequencies are
susceptible to loss due to weather eﬀects. Starlink is expected to use V-band, whereas ViaSat uses Ka-band  .
The decision regarding antenna type was driven by the current evolution of satellite antenna technology. There is the
traditional option for a parabolic or horn antenna, but there is also the opportunity to use electronically steered antennas,
e.g. phased arrays, which is planned for use in designs such as O3b’s mPower system [
]. Finally, in the case of optical
communications, the antenna is replaced by a laser. Analogous decisions must also be made for the user terminal.
There is also a decision regarding the mass of the satellite. This indicates the size of the satellite, what could be put
onboard, and directly correlates to the satellite’s performance (size of solar panels, antenna size, drag management needs,
thermal inertia, etc.). We split this decision into four categories: less than 10 kg, 10-100 kg, 100-500 kg, and 500+ kg.
Finally, there is the decision of coverage. The goal is to compete with Viasat-3 and the proposed constellation
Starlink, therefore the system needs to cover a large part of the globe - continental and regional systems would not
compete in the same market. A LEO constellation cannot be limited by longitude anyway, unless the satellites are not
used for large portions of their orbit. Thus, the coverage decision is based on the size of band we oﬀer around the
equator. The options are an equatorial band for
latitude, and eﬀectively global band within
access to nearly all inhabited areas , and global, pole-to-pole, coverage.
Clearly, some of these decisions interfere with one another, creating constraints. For example, if the frequency band
is optical, the antenna and user terminal types must also be optical. These constraints are:
C1: If the Frequency Band is Optical, then the Antenna Type and User Terminal Type must also be Optical (and
vice versa). These three decisions must either all be Optical, or none of them are Optical.
C2: If there are crosslinks, the type of crosslink architecture must be considered. Therefore, if the Crosslink Type
is RF or Optical, the Constellation Architecture must be either Ring or Mesh. If the Crosslink Type is None, the
Constellation Architecture must be None.
These constraints, and their eﬀects, are also illustrated in the pair of decision trees in the Appendix G.
With the nine system architecture design decisions determined and their inter-relating constraints identiﬁed, the next
step was to determine which screening metrics to use in order to compare system designs. Communication satellites
are limited by regulatory bodies, namely the FCC, on the allocation of frequency, bandwidth, and power ﬂux density
(PFD). Historically, communication satellite manufacturers struggle with proﬁts, and many ﬁle for bankruptcy before
the delivery of the system [
]; for this analysis, closing the downlink budget to deliver value to stakeholders
and generating a positive NPV were required; consequently, the most important metrics were determined to be cost
and performance. An explanation of the quantitative analysis of the cost metric is described in Section III.E and
an explanation of the performance metric approach can be found in Section III.F. Speciﬁc entries for each cost and
performance metric are shown in Appendix A.
The screening metrics that were chosen highlight a trade-oﬀ in development and operation between cost and
performance. In order to quantify the relationship between the two, all constrained combinations were modeled. The
result of this algorithm graphically represents the entire design space, from which all the dominated solutions were
identiﬁed to create the Pareto front.
E. Cost Metric Approach
The cost metric used in this analysis is a scalar from zero to ten, with zero being no cost added and ten being the
most expensive option. The cost for each decision option was determined using several diﬀerent methods, such as
market research of commercially available components, historical FCC spectrum availability, technology heritage, and
qualitative assessments of expected costs. An in-depth cost metric discussion can be found in the Appendix B.
F. Performance Metric Approach
In order to eﬀectively assess the impact of each metric on the constellation performance, the design variables were
incorporated into an equation similar to a non-normalized, multi-attribute utility function that focuses on communication
performance. The algorithm begins with the downlink budget to baseline performance. It then incorporates losses
from the system (e.g. path loss) and data rate to arrive at energy per bit over noise. The algorithm then includes all the
decision metric factors that will either penalize or increase the performance of the link. For example, the relay loss can
be thought of as a loss in SNR, since either decision aﬀects the probability of making an error at the user terminal - thus
it is a loss factor in the performance equation. The pseudo-link budget equation used to assess constellation performance
is shown in Eq.(1),
Psys =Pt x +Gtx +Gr x −Lp+10 log(lat ) − k−Lr el ay +gf−Lat m +Rcl (1)
and a detailed description of the equation and its variables can be found in Appendix C.
G. Pareto Front
As can be seen in Fig. 2, the Pareto front shows a non-linear positive correlation between performance and cost.
The low cost portion of the frontier experiences steep performance improvements with minimal cost increases. As the
relative costs move greater than 2.5, the performance improves less compared to the increase in relative cost, indicating
diminishing returns. The Pareto front oﬀers 27 optimal non-dominated design solutions for a LEO based satellite
H. Reduction of Non-Dominated Solutions
A detailed analysis of the Pareto front led to a downselection of four system designs. One method used to downselect
was to qualitatively analyze the Pareto front and ﬁnd the points with the highest gain in performance for the lowest
change in cost (e.g. the highest slope). Speciﬁc designs were also chosen to represent the full spectrum of decision
options that were available along the Pareto front. Table 2 shows the details of four designs that provide the best tradeoﬀ
between cost and performance.
Fig. 2 Downselected Designs
NanoSat is the design with the most standardized technology (e.g. X-band frequency), but the compact form factor
and lack of ﬂight heritage results in an overall moderate TRL level. In other words, it is expected the material cost of an
individual satellite will be relatively low, however developing and miniaturizing sophisticated communications payloads
will result in development risks, integration problems and risks. Overall NanoSat will blur the boundaries between
subsystems due to the constrained form factor.
MidRange is the transition to the next mass (and volume) category, where many of the integration and miniaturization
problems should be alleviated. Additionally, the increased surface area allows for higher power generation and
utilizing electronically steered antennas introduces greater capability in delivering value, such as adding beamforming
Table 2 Selected Designs
Decision/Name NanoSat MidRange NextGen Tech Heavy
Mass [kg] 10 10 −100 100 −500 500+
Frequency X band X band Ku/Ka band Ku/Ka band
Antenna Type Parabolic Electronically
Crosslink Type None None RF Optical
Relay Type Bent Pipe Bent Pipe Bent Pipe Regenerative
Architecture None None Ring Ring
Coverage Equatorial Equatorial Eﬀective Global Eﬀective Global
Resource Allocation Yes Yes Yes Yes
User Terminal Electronically
Cost Score [-] 0.00 0.05 0.23 0.59
Performance Score [-] 0.08 0.28 0.68 0.96
In contrast to NanoSat and MidRange, the NextGen design has eﬀective global coverage with Ku/Ka Band frequency,
and RF crosslink between satellites in a ring architecture. Similar to the transition between NanoSat and MidRange,
shifting to the next mass category makes more power available per satellite. Combined with crosslink capabilities, the
system can service customers in areas diﬃcult to service, such as the maritime and aviation industries. This results in a
substantially higher performance score.
Tech Heavy is similar to SpaceX’s Starlink design and has low TRL level due to regenerative relays and the optical
crosslinks. We expect signiﬁcant developmental risk and technological challenges that may result in both cost and
schedule overruns. The additional optical crosslink capability with regenerative payloads allows for a performance
increase when compared to NextGen, but we also expect a signiﬁcant cost increase.
IV. Detailed Design
With the design space reduced to a handful of unique systems, a detailed analysis is required to further evaluate the
cost vs performance trade. We investigate speciﬁc LEO altitude for optimal operation, and the inﬂuence of coverage on
system performance. The foundation of this analysis is comprised of three major technical capability estimations: a
downlink budget, a power budget, and system coverage (see Fig. 3). The outputs of those functions are fed into an orbital
lifetime estimator and a cost estimator, which then determine the value of the three performance metrics: $/Megabyte,
system capacity, and total cost. A summary of the detailed design functions are found in subsequent sections.
A. Link Budget
A traditional link budget is comprised of two to three parts: a downlink, an uplink, and a crosslink (if applicable).
In the detailed design, the link budget is analyzed from a downlink-only perspective because the emphasis is put on
delivering data to customers. The link budget starts by determining the power received at the user terminal with Eq.(2).
Prx =Pt x +Gtx +Gr x −Ltx −Lr x −Latm −Lp ath [dB](2)
In this equation,
is the power of the transmitter,
are the gains of the transmitter and receiver,
are the losses at the transmitter and receiver,
is atmospheric losses, and
is the path
loss introduced by the separation between the transmitter and receiver.
The communication system is subject to noise, with Johnson noise and ampliﬁer noise contributing the most power
to the bands of interest. Bit Error Rate (BER), the number of expected errors arising from noise in a link, is a function
of the modulation scheme and
(energy-per-bit). Many BER curves have been published capturing this relation for
diﬀerent modulation schemes, and can be used to determine the required SNR ratio for a system [
]. For a system
using QPSK with 10−5BER can be achieved with a SNR of 9.2 dB.
Fig. 3 Analysis Foundation
To determine the SNR of the system, denoted
, Eq.(3) can be used, where
is the power received calculated
from Eq.(2) in decibels, DR is the data rate in decibels,
is Boltzmann’s constant in decibels, and
is the system
temperature in decibels. Note that SNR is equivalent to Eb
N0in this equation.
SNRsys =Prx −DR −kdB −Tsys [dB](3)
According to the Shannon-Hartley theorem, the maximum achievable data rate in a system is band-limited and noise
limited, and in the case can be found as
with Eq.(4), where BW is the bandwidth of the link in Hertz,
power received from Eq.(2) in Watts, kis Boltzmann’s constant, and Tsys is the system temperature in Kelvin.
DRlinear =BW 1+Prx
kTsy sBW [bps](4)
By implementing a 3 dB margin, the beam-speciﬁc data rate can be found by decrementing the maximum achievable
data rate in the SNR equation (Eq.(3)) until the margin criteria is met. The total system throughput then is found by
multiplying the data rate per beam by the number of beams per satellite.
B. Power Budget
The power budget determines the amount of power required for full system operation in eclipse and sizes solar
panels to meet the power requirement for end of life operation. According to SMAD Table 10-2, the payload consumes
40% of the spacecraft power [
], which means by estimating payload power consumed, satellite power consumed can
be determined. For example, the ClydeSpace CPUT X-Band transmitter advertises less than 10 W power consumption
with a single patch antenna. For NanoSat, we have assumed a gimbaled parabolic antenna conﬁguration and an X-band
radio capable of using multiple antennas simultaneously. Assuming double radio power is required to accommodate
additional antennas, each antenna requires roughly 1 W for mechanical control, and ADCS can compensate for antenna
movement, then roughly 26 W is required for payload operation only. This implies 65 W is required for full operation,
and to add an additional margin for battery safety, the required power is rounded up to 70 W. This power estimate
accounts for both bus power and payload power duty cycled over the duration of an orbit.
Using Kepler’s 3rd Law, the orbital period of a single satellite can be estimated (Eq.5), and through means of
geometric analysis and assuming a circular orbit (6), the percentage of the period in eclipse can be determined (7).
Tecli ps e =Torb it ∗(π−α)
Where Reis the Earth’s radius, his the orbital altitude, and µis the gravitational parameter for Earth.
The objective of this function is to size solar panels to meet power requirements. This is done iteratively by
initializing the solar panels to an optimistic value (e.g. smaller surface area than expected), and stepping up the area
incrementally until the orbit average power generated is greater than the power required.
The power generated by the solar cells is found with 8
Pre q Tecl i p s e
+Pre q (Tor bi t −Tecl i p s e )
Torb −Tecl ipse [W](8)
Where a ηeand ηdare the energetic eﬃciencies between the solar array and load.
Solar array sizing must be done with respect to End of Life (EOL) performance. Assuming triple junction Gallium
Arsenide cells, the EOL power ﬂux density is calculated with 9 - 11
Ld=(1−d)satellite life (9)
PBOL =Psun ∗ηS A ∗Idcos(θinc ) [ W
PEO L =PBO L Ld[W
is the expected yearly degradation rate due to radiation,
is the solar constant with units
solar cell eﬃciency,
is 0.77 for GaAs cells, and
is the angle of incidence of the solar ray vector onto the array.
Solar array size is found using Eq.12
PEO L [m2](12)
Battery size is based on the system required power such that full operation is possible during eclipse periods. Healthy
battery depletion practice involves setting a depth-of-discharge (DoD) threshold such that nominal operations avoid
draining the battery past its DoD. This allows for longer battery lifetime and more charge cycles [
]. Eq.(13) shows a
method for estimating the required battery capacity:
Battery Capacity =
DoD ∗ηtr a ns f e r [kWh](13)
is orbit average power generated,
is from Eq.(7),
is the percent depth of discharge as a
fraction of total battery capacity, ηtransfer is the power transfer eﬃciency of the battery regulators.
To estimate the mass of the system, four assumptions were made: ﬁrst, each system has propulsion; second, batteries
are 32% of the satellite’s mass; third, the batteries of the system are lithium-ion; and fourth, solar panels are roughly 2
]. Combining these assumptions provides us with a system for estimating the dry mass of a satellite, as
shown in Eq.(14):
mdr y =Battery Capacity
0.32 ∗LID +2∗SA [kg](14)
is found from Eq.(13),
is the lithium-ion energy density, and
is the surface area of
the solar panels in m
. To ﬁnd the wet mass of the satellite it is assumed the rocket launches from Cape Canaveral and
thereby satellites are released in a 28.5
inclined orbit. To achieve equatorial coverage the satellites can remain in the
dropped oﬀ orbital plane while for the eﬀective global and global coverages the orbital planes have to be changed, the
required velocity change is found according to Eq.(15).
is the orbital velocity and
is the desired change in
orbital inclination. The preferred propulsion system for the orbital change is ionic Electrospray Propulsion System
(iEPS) with Is p of 1200s. The wet mass of the satellite is found with Eq.16, with g=9.81m/s2.
∆V=2Vorb it sin(∆i) [m
Is p g[kg](16)
C. Coverage Area
To estimate the required size of each downselected design, a geometric relationship was derived between satellite
antenna ﬁeld of view, crossover between satellites’ equatorial ﬁeld of view (to account for user and ground station
handover) and the satellites’ altitude. After analyzing constellation dynamics on a Mercator projection of Earth, it
was concluded that the assumption of constellation size grows linearly as latitude coverage increases as ﬁrst order
D. Orbital Lifetime Estimation
The orbital lifetime estimation is a spline extrapolation from the results produced in [
]. Based on the surface area
and mass of the satellite, the orbital decay can be found for a range of altitudes. Although all orbits are placed below the
Van Allen belts, radiation damage and other hardware failures are likely to become limiting factors in lifetime before
orbital decay, and a maximum lifetime of ten years is assumed for each system.
E. Cost Estimation
Detailed cost estimation was implemented using the ﬁnancial analysis described in Section V.A.
F. Downselected Designs
The driving performance metric for these systems is the $/MB vs. altitude function, which captures the total cost of
the system, the lifetime of the system, and the total system capacity. The objective is to optimize from this metric across
systems to get the best price per capacity, making it competitive from both a business and performance perspective.
Diﬀerent coverages produce diﬀerent $/MB curves as the cost and system capacity ratio ﬂuctuates. These curves are
shown in Fig. 4. In each case, the optimum coverage value (where $/MB is minimized) was selected to represent that
design for system comparisons in Section IV.G. As can be seen in Fig. 4a, the absolute minimum of $/MB occurs on the
Global curve at 487 km altitude. Figure 4b shows the lowest $/MB for the MidRange constellation occurs at 505 km with
Global coverage and Fig. 4c indicates that the optimal (in terms of lowest $/MB) altitude for the NextGen constellation
with Global coverage occurs at 444 km. Finally, Fig. 4d shows that the Tech Heavy constellation is optimized at 411 km
with Equatorial coverage. The comparison of $/MB versus altitude for the selected coverage option in each design can be
found in Fig. 4e. The key characteristics of each of these systems at their optimized altitude are summarized in Table 3.
G. Detailed System Trades
Once the individual systems are vetted for optimal performance, the overall highest-performing system must be
determined. Table 3 gives a side-by-side comparison of all of the systems, and Fig. 5 shows each of the system’s
performance metrics against one another.
Here, the intention of this study are revisited and ﬁnal downselection is aligned with mission requirements and
objectives. The selection process deviates slightly from simply selecting the design with the lowest $/MB; although
NanoSat has the overall smallest $/MB, it cannot compete with reference cases Starlink and Viasat in terms of capacity
or throughput. Looking at altitude 444 km, NextGen exceeds ViaSat capacity and latency, has the highest NPV, and
it can provide the subscribers with 10 Mbps of throughput during peak operation. Section VI gives the ﬁnal system
recommendation and rationale.
H. AI Implementation
As with many ﬁelds and industries, there are a number of applications for artiﬁcial intelligence (AI) and machine
learning (ML) in space and, speciﬁcally, satellite telecommunications. This could include onboard AI on the satellite
itself, or ground-based systems that instruct the satellite but require more mass or power than is practical in orbit. There
are a number of ways it could be included in the system, but we focused on three: health monitoring, dynamic resource
allocation, and smart routing.
(a) NanoSat (b) MidRange
(c) NextGen (d) Tech Heavy
(e) Comparison of Designs for Selected Coverage
Fig. 4 Cost per Megabyte versus Altitude for the Downselected Designs
The ﬁrst of these uses is health monitoring. There are massive amounts of data that can be used to understand a
satellite’s performance, providing information on status, health indicators, and overall behavior and environment. These
data can be used to improve eﬃciency, diagnose problems and even predict problems before they occur. However,
the amount of information is immense, and it is easy to overlook patterns or small anomalies. AI could be useful, by
better interpreting the data and ﬁnding new patterns. Most of this AI will likely be ground based, given the power and
complexity required to run the system, where it could provide decision support for operators. The system could also
synthesize data from satellites across the constellation to more eﬃciently address problems.
A second use of AI is dynamic resource allocation. Eﬃciently distributing the satellites’ power and bandwidth, as
well as optimally pointing the beams, can be complex, computationally intensive questions. AI could be used to reach
(a) Total Cost Over All Systems as a Function of Altitude (b) Total System Capacity as a Function of Altitude
Fig. 5 Performance metrics compared for downselected designs
Table 3 Downselected Designs Comparison
Decission/Name NanoSat MidRange NextGen Tech Heavy
Mass [kg] 12.7 38.3 125 480
Frequency X Band X Band Ku/Ka Band Ku/Ka Band
Antenna Type Parabolic Electronically Steered Electronically Steered Electronically Steered
Crosslink Type None None RF Optical
Relay Type Bent Pipe Bent Pipe Bent Pipe Regenerative
Architecture None None Ring Ring
Coverage Global Global Global Equatorial
Altitude [km] 487 505 444 411
Est. Constellation Size 264 253 312 203
System Capacity [Gbps] 102.5 370.5 1832.5 3203
Development Cost $33M $33M $48M $64M
AI Software Cost $5M $5M $5M $5M
Total Satellite Cost $38M $1,453M $8,112M $12,434M
Total Launch Cost $270M $682M $744M $1,302M
Regulatory Fees $10M $10M $20M $20M
Total Constellation Cost $356M $2,183M $8,929M $13,825M
NPV -$328M $501M $1,401M -$1,748M
better solutions to this problem. Neural networks and machine learning could be used to increase eﬃciency. Another
option would be to use genetic algorithms. The performance could be improved even further by using machine learning
to determine patterns in usage and need, and allow the satellite to start acting predictively. Due to power and mass of the
computing system, this AI would likely need to be ground-based.
Smart routing is a subset of resource allocation. In this case, the resource to be allocated is bandwidth across nodes
in the network. At each point, a user’s data could be transmitted a number of ways. The satellite could function as a
bent-pipe, and simply take the user’s uplink and transmit to the destination. It could take the uplink, then transmit to a
ground station, send the data over cables to the destination (or to another ground station, uplink to a diﬀerent satellite,
and downlink to the destination). It could also take the uplink, then pass the data through many diﬀerent options of
crosslinks, before being beamed down to the destination, or to a ground station, etc.. At any stage, the data could go
between crosslinks, to the ground and over cables, or back up to satellites. For instance, the data for a user in rural New
Mexico (headed to a company in Moscow) might go to the nearest satellite, passed by crosslink to a neighboring satellite,
then beamed down to a ground station in Texas. The data might be downlinked here due to the satellites over the Eastern
Seaboard being at capacity. The data could travel by cable to , before being re-uplinked and transmitted over the Atlantic
by crosslinks, then beamed down to Moscow. Such a route is very complicated, but could, under speciﬁc situations, be
the optimum solution. The availability of such complicated routes could improve performance of the network overall by
allowing each connection in the network to operate closer to its maximum, preventing unused bandwidth in connections
that are under-utilized. This would result in the entire system operating closer to peak eﬃciency. This is an incredibly
complex computational problem, but machine learning as well as other algorithms (such as genetic algorithms) could
help arrive at eﬃcient, more optimal solutions, and thus increase throughput. The second advantage of such is a system
would appear in the event of a breakdown. When a connection is blocked by weather, or if a satellite fails, an AI could
identify the damaged satellite, ﬁnd new paths for the information and route the data around the broken connection or
satellite. This would result in an adaptive network that responds to failures, reducing downtime and increasing reliability.
Due to the mass and power of the computers required for this sort of system (in addition to the needs of input from
satellites across the constellation), it is recommended to be ground-based.
The only AI option incorporated in the design space was resource allocation, which was selected for each of the
four downselected designs. The other two uses, health monitoring and smart routing, could be implemented for any
relevant design. Hence, recommending further study to explore the exact details of how they could be implemented,
the magnitude of the beneﬁts achievable, and the possible increase in eﬃciency. None of these systems are necessary
for the designs to work, but they have the opportunity to improve all of them. Therefore, suggesting developing these
technologies and implementing them for any of the ﬁnal designs.
V. Implementation Plan
Costs were broken down by development, satellite, launch, and regulatory costs for each of the detailed designs.
Development costs were estimated based on the anticipated design time, number of full-time employees, and a salary
and beneﬁts package of $200,000 annually. AI software cost estimation was developed in a similar way, with a annual
salary of $500,000 with ten employees for two years. The satellite costs were derived from estimated cost based on bus
size, technology readiness level, and anticipated high-volume manufacturing eﬃciencies. Finally, the launch cost was
determined by examining the satellite weight, orbital altitude, and the number of orbital planes. For “NanoSat”, the
Electron Rocket from Rocket Labs was considered, whereas Falcon 9 was assumed to estimate launch costs for the other
An estimate was also used for the licensing fees and includes fees associated with consultants, lawyers, International
Telecommunications Union (ITU), and FCC. It was expected for Ka/Ku band to have higher fees due to their high use in
The ﬁnancial costs in Table 3 represent the costs from project initiation through full constellation launch completion.
To conduct an analysis of the ﬁnancial viability of each satellite design, a ﬁnancial pro forma was developed from
project initiation through the 10-year proposed lifecycle of the satellite constellation. The operating cost of the satellite
constellation was assumed to be 12% of the total development cost across the lifetime of the constellation, based on
operations of other satellites[
]. The assumed salary and beneﬁt package of $200,000 per year per employee continued
through the lifecycle of the constellation. The revenue model is based oﬀ $50 per month per subscriber. Due to the
presumed vertical integration of the design and manufacturing, initial launches are assumed to begin within two years in
2020. Subscribers will begin to use the satellite connectivity service once the constellation has been fully deployed.
Initially, it is assumed that the subscriber base will be at 500,000 users (13% market share of the 2021 satellite internet
users). The subscribers are assumed to grow at a 50% compound annual growth rate (CAGR) through 2029. However,
this growth is limited. The user base is not allowed to grow past the point where the constellation could support half of
the users at 10Mbps, to ensure the system can still cope with peak demand. The weighted average cost of capital for the
pro forma is calculated at 10%, which is in line with current technology and space industry estimates. The resultant
cumulative net present value (NPV) of each downselected design is shown in Fig. 6.
Fig. 6 NPV of Downselected Designs
For any of the designs, the implementation necessary to meet regulatory standards should not diﬀer greatly from
past constellations and telecommunications platforms.
To comply with regulations, broadcast licenses will need to be obtained, orbit debris analysis will need to be
performed, and appropriate documents (such as end-of-life orbit plans) must be ﬁled with the FCC (and equivalent
organizations around the world), ITU, FAA/AST, State Department, and so on. These will need to be completed, like
any other constellation.
The design should not have any additional regulatory hurdles. The only technologies that could be considered
new or diﬀerent on these satellites are electronically steered antennas, the use of AI (onboard or ground-based), and
optical links. For electronically steered antennas, the regulations are likely identical to more traditional antennas. Any
organization utilizing it will need to verify the regulations are equivalent. Like any other system, they will also need to
verify that these regulations are met through analysis and testing. For AI-based systems on board, there will need to be
software/hardware limits in case the algorithm acts unexpectedly. For example, if the AI attempts to allocate too much
power to a downlink, a hardware or software limit should prevent this. Such an approach should satisfy any regulatory
questions regarding AI, and is probably a smart move even if not demanded by regulations. Optical links are the one
possible option that require additional regulatory actions. For optical crosslinks, the system will need to be cleared with
the Joint Space Operations Center Laser Clearinghouse, in addition to the FCC and ITU.
As a new player in the satellite internet market, any company will need a marketing strategy to recruit customers and
build a reputation in this ﬁeld. Of course, the ability to building a marketing strategy on the aspects being promoted are
highly dependent on the design selected.
Emphasizing the ubiquity of coverage, low-latency connection, and high-throughput connectivity are all possible
factors to use to attract individual clients. These could also be used to attract businesses, along with the emphasis on the
Some other key potential markets for the system are mobility (maritime and aviation, particularly), trucking, oil rigs,
and network backhauling. These systems would not be marketed directly to consumers, but would require deals and
agreements with airlines, shipping companies, trucking companies, and network providers. The best approach for this is
having a system that is competitive in performance, reliability and price.
There is also the potential to use the network to enter other markets, besides satellite internet. One option would be
partnering with cell phone companies to improve coverage and reliability in countries without developed ﬁber networks.
Partnerships in this market could help gain access to new markets, particularly if we partner with data-intensive industries
that typically rely on terrestrial networks. Using the system, they could expand into regions that were previously
unavailable, due to lack of infrastructure, safety issues, or other problems.
Finally, rural areas are another key potential market for the system. This is a user base where large sections of the
population either use satellite communications or could be well-served by satellite communications, given the lack of
reliable alternatives. A network such as ours, that is high-throughput with extensive coverage, could appeal to these
customers. In particular, it could compete with low-reliability or low-throughput broadband or cellular options. It is
true the idea of satellite internet and cutting edge technology may pique some people’s interests, but recruiting and
retaining users will be most heavily dependent on customer experience. Therefore, any implementation will need a
strategic focus on the customer experience. The customer cares mostly for the ways in which they interact with the
system, including user terminals. In the internet service market, we can also aim to provide the service end-to-end. This
would allow us full control over the network and the customer experience. Using this to focus on customer service and
provide a top-notch delivery, we could reduce the frustration of the customers, provide a great experience when using
our product, and command the highest possible margins.
Any enterprise carries risks, and this is especially true in the space industry, with its high upfront costs and complex
technologies. A cutting-edge system and aggressive business case, as described here, carries an added share of risks as
The ﬁrst group of risks are technological. This project would be at the forefront of satellite telecommunications and,
therefore, relies on technologies that have not been used on this scale before. One key risk is the integration of these
technologies. AI, electronically steered antennas, satellite crosslinks, and X-band satellite links have all been separately
demonstrated, however they have not previously been combined in a consumer telecommunications network located in
the harsh environment of space. There could also be unforeseen issues with integrating these technologies, causing cost
and budget overruns, or requiring design alterations.
Though it is not a speciﬁcally new technology, there are a similar potential issues regarding a constellation of this
scale. Without any previous systems of this size, it is diﬃcult to predict the problems that will be encountered. Issues
that do not occur in a system of 20 satellites might appear in a system of 200. Similarly, electronically steered antennas
have been demonstrated, but not used as user terminal antennas on a large scale. These antennas may also experience
new problems when in the ﬁeld and subjected to the environmental and wear conditions of a user station. Finally, there
may be an adoption hurdle for the users, as an electronically steered terminal could be quite diﬀerent from those with
which they are more familiar.
A second group of risks are more operational. Any satellite telecommunications network is complex to operate
and requires many levels of bureaucracy and regulatory approval. As this project might be applied to new markets
and use cases, there could be struggles in adapting the existing frameworks, on top of the issues inherent in all
satellite communications projects. Establishing ground stations in multiple countries will require meeting their national
regulations, which vary. This is not a huge risk, but should be taken into consideration, particularly given the new
technologies and ideas that may be utilized. The regulations in many countries may not have caught up with new
technologies, or may be punitively conservative regarding new concepts. General regulatory approval will also be
needed from the FCC, ITU, and other organizations. This is a known process, but there may be new hurdles for new
technologies (see Section V.B). Some of the markets and use cases under consideration may be more diﬃcult. Using
satellites to compete with 5G or as part of the 5G structure is diﬀerent than previous cases. It is possible that this system,
in the juncture of two previously separate domains, could face added restrictions, regulations, and requirements.
Finally, as with any business endeavor, this project carries market risks. We are making predictions of the market
based on current trends and expert opinions. If the market grows less than predicted, or changes in a way that makes
the system ill-suited, then the system may have fewer paying customers and generate less revenue, until an alternate
market and use case can be found. On the other hand, if the market grows faster than predicted, there is the possibility of
opportunity cost for having a system that is too limited or small-scale. One speciﬁc risk is, with the expanding satellite
internet market, the price of satellite internet capacity decreases signiﬁcantly. In that case, our revenue estimates, based
on expected prices, would be invalid, and lower prices might be necessary to remain competitive. Similarly, growth in
terrestrial networks could make the system less able to compete in certain markets. High-speed ﬁber networks, networks
run by the community (rather than ISPs), and networks in areas that are currently underserved (such as rural areas) are
particular examples which could cut into the predicted market. As mentioned previously, another risk is that the product
(the user terminals or another aspect) is not well-received by consumers, reducing market penetration. If the cost of user
terminals is prohibitively high, then consumers cannot aﬀord to start services. The price of user terminals must be
driven down to gain traction with a wider customer base.
Political and regulatory changes could also introduce market risk. One key example is net neutrality. With the
overturn of net neutrality, many consumers could see their ISP increase their prices based on the content they access. In
locations where there are few ISPs to choose from, a satellite internet product has the potential to beneﬁt by oﬀering a
neutral product. If net neutrality is reinstated (or companies act neutrally on their own), then this market opportunity will
not be present. On the opposite side of this, without net neutrality, there is the opportunity to partner with companies to
distribute their content for free or reduced prices in markets that would otherwise be diﬃcult to proﬁt in (similar to
Facebook’s internet.org and Free Basics program). The reinstatement of Net Neutrality would remove this business case.
Many changes to the market could reduce the possibility for market share and thus revenue, but the changes could also
oﬀer opportunity. The risks and possibilities are hard to determine ahead of time, and could be caused by unpredictable
A. Summary Selection
Based on the detailed analysis comparing performance vs. cost and NPV of each system, the ﬁnal recommendation
is the “NextGen” design. The key characteristics of this selected design can be found in Table 4.
Table 4 Design Choice Characteristics
Designation Mass [kg] Frequency Band Antenna
NextGen 125 Ka/Ku Electronically
Steered RF Bent Pipe Ring
Cost 2029 Sub-
Global 444 312 1832.5 $8.9B 8M $1.4B
The system’s high NPV is the primary reason for our recommendation. Satellite communication companies have
historically struggled with generating revenue; by optimizing over NPV, potential cash ﬂow is the driving metric in
determining next generation satellite communication systems. The constellation is able to provide nearly 2 Tbps across
the system, which competes with Viasat-3. Furthermore, the high system capacity places this design as the ideal system
to provide mobile broadband datarate (10 Mbps) at peak demand [
]. NextGen can service over 4 million users with 10
Mbps during peak demand.
NextGen provides high performance at an acceptable price. The low price per megabyte is an advantage, but it is
also worth noting that it does not have excessive development costs. The NPV for NextGen is at $1.4B with the project’s
nominal market analysis. Compare this with the “Tech Heavy” design where its very high development cost results in a
negative NPV of -$1.75B.
NextGen is a truly global constellation; global coverage is desirable because it oﬀers customers internet anytime,
anywhere; for comparison, the MidRange design technically provides global coverage, but the satellite system does not
have crosslinks, which greatly limits connectivity. This causes issues over oceans (where there is a high mobility market
demand) since it is diﬃcult to install ground stations. Having a constellation with crosslinks, like NextGen, also reduces
the total number of ground stations required, reducing the cost of the whole project. Furthermore, crosslinks allow
for live health monitoring of the system and make the system more robust by enabling alternative routing in the case
of system degradation. Finally, the crosslinks allow the system to tap the potentially lucrative maritime and aviation
markets, meaning that the system could bring in more revenue than simply by providing satellite internet, meaning the
revenue, value, and NPV may be even higher than we have predicted.
B. Comparison to Reference Cases
The selected design, NextGen, outperforms some of the reference cases. ViaSat-3 has already been deployed and
both NextGen and Starlink are under development, making it diﬃcult to perform more than a high-level analysis. A
more thorough analysis on announced speciﬁcations and market estimations is likely to be inaccurate. This is especially
true in the satellite industry, as many projects take much longer and cost much more than advertised. More speciﬁc
design and performance comparisons are show in Table 5. It should also be noted that the reported cost of ViaSat is low,
as it only includes manufacturing costs, and does not include launch or development costs.
Table 5 Design and Performance Comparisons
Feature NextGen ViaSat-3 Starlink
Orbit LEO GEO LEO
Coverage Global Regional Global
Constellation Size 312 3 4,425+
Mass [kg] 125 6,400 400
Latency [ms] 25 638 25
System Capacity [Gbps] 1833 3,000 88,500
Cost [$B] $8.9 $1.9 $10
Compared to ViaSat-3, “NextGen” is superior in latency and disaggregated risks. As ViaSat-3 is in GEO, the
round-trip time for a signal is, at best (with the satellite directly overhead) nearly half of a second. This is too long for
voice communication or video conferencing, and is even long enough to make webpages load more slowly. Any sort of
interactive information, or communication that requires a back-and-forth, is going to be dramatically slower and feel
very “laggy” to users. NextGen, which is only at 1/70th of the altitude, will have much less of an issue. NextGen has a
capacity of nearly 2 Tbps, putting it in the same class as ViaSat-3. In addition, with dynamic resources allocation, the
system should be able to better utilize this capacity. The ﬁnal advantage to “NextGen” over ViaSat-3 is the number of
satellites. By deploying 312 satellites instead of 3, the impact of a single failure is greatly reduced. Satellites can be
replaced, if needed, for signiﬁcantly less cost, and each satellite handles a smaller fraction of the data, making any
problems less catastrophic. This ties in with the use of crosslinks, as well. Data can be routed around problem satellites,
and nearby satellites can help transmit to various locations. Data can be routed a multitude of diﬀerent ways, through
the crosslinks or on the ground, which can allow for more eﬃcient usage of the network and better throughput overall.
tNextGen has a number of advantages over ViaSat-3 including lower latency, higher system capacity, and disaggregated
risks. For more in-depth information on ViaSat-3, see Appendix D.
For Starlink, we found it impossible to outperform the current cost, schedule, performance, and technology claims
made by SpaceX. Since it was not possible to beat this system, a feasibility study was conducted to see if the system will
be a potentially outperforming competitor. Our analysis indicated that Starlink will not be a commercially successful
venture due to cost sensitivities, lack of historical precedent for necessary funding, and competing internal R&D funding
priorities within SpaceX. For more in-depth analysis, see Appendix F.
Furthermore, the “Tech Heavy” solution shares many similar attributes to the Starlink, with optical crosslinks, a
regenerative payload, slightly heavier mass (480 vs. 400 kg), and LEO altitude. Though “Tech Heavy” has signiﬁcantly
fewer satellites (203 vs. 4,425) and less technology on board, NPV analysis indicates that it will lose nearly $2 billion
(though this may be connected to it not being a global system). It is worth noting that SpaceX has made grandiose
claims (such as the Falcon Heavy) in the past, and eventually delivered, so it is possible that some form of Starlink may
eventually succeed. However, though it was eventually successful, the Falcon Heavy was delivered years later than
originally anticipated. The most likely case for Starlink to succeed is with a modiﬁed schedule, costs, and performance
estimates. Without changes to current claims, it is unlikely to have a positive NPV or make enough money from the
incurred costs. For more information on current Starlink plans, see Appendix E.
C. Sample Strategic Plan
Demand for data and connectivity is forecasted to grow exponentially in the foreseeable future. Terrestrial networks
will attempt to expand with the growing demand, but could struggle to meet all needs [
]. Satellite internet capacity
needs to to be expanded in order to meet this demand, and there is a clear market opportunity in this fast-growing market.
Satellite internet could also play a key role in the 5G rollout. Hybrid networks will prove to be essential in providing
internet connectivity to everyone in the world. An additional strong component in the favor of satellite internet is that
current Net Neutrality rules may make satellite internet more appealing for customers .
The goal for implementing this technology going forward should be to compete on performance and cost, with
a customer-oriented solution. To achieve this, a new satellite internet provider should partner with consumer-facing
companies to ﬁnd user-friendly solutions that easily interface with people’s lives. In order for this satellite constellation
to be eﬀective, investments must be made in consumer-based hardware, such as aﬀordable user terminals and cell phone
plug-ins for satellite connectivity. There is also the opportunity to expand the role of satellite internet through operating
services, such as WiFi cafés for rural villages.
With this consumer-oriented focus, a satellite internet provider would be able to develop a strong retail presence.
Emphasis needs to be placed on sales channels, industry partnerships, and product architecture. By creating a vertically
integrated system from design to manufacturing and through to service-based operations, the company will have greater
control over the entire ecosystem. This will allow the provider to capture the larger margins that are available closer to
A. Speciﬁcation of Metrics
Table 6 Summary of Cost and Performance Estimates
Decision Option Cost Performance Inputs
None 0 cl f =1
RF 3 cl f =3
Optical 9 cl f =6
Parabolic/Horn 8 see Appendix C
Electronically Steered 10 see Appendix C
Laser 9 see Appendix C
Global (pole-to-pole) 10 nsats =213
Eﬀective Global (±60◦) 7 nsats =183
Equatorial (±45◦) 6 nsats =160
Relay Type Bent Pipe 2 Lrelay ∝nsats
Regenerative 10 Lrelay ∝nsats
None 0 ar ch =1
Ring 3 arch =3
Mesh 10 arc h =7
Dynamic Resource Allocation Yes 2 ηr a =1
No 3 ηr a =1
User Terminal Type
Parabolic 1 grt =0.3dB
Optical 10 gr t =5.24 dB
Electronically Steered 10 gr t =10.8dB
<10 1 m=10
10-100 2 m=45
100-500 5 m=200
>500 10 m=500
Latm =0.61 dB
reuse = 4
Latm =0.97 dB
reuse = 4
Latm =7.42 dB
reuse = 7
Latm =10 dB
reuse = 1
B. Cost Metrics
A. Antenna Type
Options: Parabolic, Phased Array, or Laser. See Table 7
Table 7 Summary of Cost Estimates for Antenna Type
Phased Array 10
The costs for parabolic and phased arrays are based oﬀ satellite communication systems for yachts. The cost of a
laser system is based of MIT STARlab’s internally developed system. Intellian’s 65cm Ku-band gimballed parabolic
dish can be acquired for approximately $20,000. STARlab estimates the raw material cost of their laser system to
be approximately $20,000. Commercially, phased arrays for satellite ground stations can be acquired for as low as
$1,000 each from ThinKom Solutions for example. However these arrays include only single transmitter and receiver.
Experts estimate the cost of multi-transmitter and multi-receiver phased arrays being substantially higher than of their
single channel counterparts. There have been no commercial satellites deployed with laser based system for ground
communication. Comparably there is limited heritage of phased arrays on commercial satellites.
We took a conservative approach towards placing cost estimates on the antenna types since it is the main payload.
Because we do not have a price point for a multichannel phased array antenna it gets a maximum cost rating of 10. Due
the low heritage of lasers and their beamforming limitation they receive a cost estimate of 9. The parabolic dishes have
extensive heritage, known design cycles but their limitation and high cost stems from required gimbal system, high mass
and volume. Parabolic antennas receive a cost estimate of 8.
B. Crosslink Type
Options: Radio Frequency, Optical, or None. See Table 8
Table 8 Summary of Cost Estimates for Crosslink Type
The cost estimate for crosslink type is based of the Antenna type discussion. However the cost of the RF system is
assumed to be much lower than for the optical system because of lower pointing requirements and allows for multiple
satellite communication in parallel. The optical system will require an individual laser for each link. Furthermore,
optical crosslinks do not have extensive heritage, which poses additional development cost and program risk. Therefore
the optical system is given the same rating as of laser antenna type. Meanwhile due to lower risk, more heritage and
more lenient pointing requirements RF gets the cost rating of 3 compared with optical’s 9. The None crosslink type has
been given a cost score of zero.
Options: Global, Eﬀective Global, or Equatorial. See Table 9
Table 9 Summary of Cost Estimates for Coverage
Eﬀective Global 7
The cost metric of coverage is based of the estimated number of satellites, number of orbital planes and relative
inclination to the launch latitude. For a simple communications payload (ﬁeld of view 60
overlap between satellites
at 400 km altitude) then global coverage requires 4140 satellites, eﬀective global (
) requires 2944, while equatorial
requires 2024. Considering only the number of required satellites, Global gets a cost rating of 10, Eﬀective global 8 and
Equatorial 5. Due to the size of Global constellation we can assume we will get lower launch cost per satellite, hence we
increase the cost of Equatorial to 6. However we reduce the cost of Eﬀective Global by one point as the launch maneuver
to deliver a satellite to polar orbit is higher than for regular inclined orbits, hence Eﬀective Global has a rating of 7.
D. Relay Type
Options: Bent Pipe or Regenerative. See Table 10
Table 10 Summary of Cost Estimates for Relay Type
Relay Type Bent Pipe 2
Regenerative relays have not been commercially ﬂown, therefore it is assumed they have limited heritage and have a
steep price point. Meanwhile bent pipe relays have been ﬂown for decades, as a result bent pipe gets a cost score of 2
and regenerative receives maximum score of 10.
Options: None, Ring, or Mesh. See Table 11
Table 11 Summary of Cost Estimates for Architecture
Mesh architecture has been deﬁned as the capability to form a crosslink with any of its neighboring satellites. This
will materialize as requiring communications capabilities on four additional faces of the satellite, increasing costs,
pointing, data handling, and weight. Therefore Mesh gets a cost rating of 10. Ring architecture is deﬁned as the
capability to communicate to the satellite in your orbital plane, resulting in crosslinks located on two additional faces of
the satellite. Ring was given a cost estimate of 3 based on it is substantially simpler crosslink interface and less surface
area constraint. Having no architecture does not incur any costs.
F. Dynamic Resource Allocation
Options:Yes or No. See Table 12
Table 12 Summary of Cost Estimates for Dynamic Resource Allocation
Dynamic Resource Allocation Yes 2
It was assumed that the developmental cost of making a dynamic resource allocation software would lower the
lifetime cost of the constellation, by lowering operator costs relative to development cost. Therefore having dynamic
resource allocation gets a cost rating of 2 while not having one receives a score of 3.
G. User Terminal Type
Options: Parabolic, Electronically Steered, or Optical. See Table 13
Table 13 Summary of Cost Estimates for User Terminal Type
User Terminal Type
Electronically Steered 1
Given that we are a LEO constellation, the user terminals need to be able to track the satellite as it traces over the sky.
Therefore the common parabolic VSAT terminal pointing towards a GEO satellite will not suﬃce. This brings us back
000 yacht gimballed parabolic antenna. On MIT campus there is ongoing research into making an optical
ground station, with an estimated cost of the custom system in the tens of thousands of dollars. Meanwhile, electronically
steered user terminals can passively track the satellites and we have already mentioned single transmitter/receiver
phased array which is priced at
000. Other options for electronically steered or ﬂat panel antennas are Isotropic
Solutions Limited (ISL) and Kymeta Corp.’s antennas which are projected to radically lower component costs. As a
result electronically steered antennas get a cost estimate of 1, meanwhile parabolic and optical receive a cost score of 10.
H. Frequency Band
Options: X-Band, Ku/Ka-band, V-band, or Optical. See Table 14
Table 14 Summary of Cost Estimates for Frequency Band
There are several factors that went into the frequency band cost estimate, most notably the available frequency from
FCC and ITU to get spectrum allocated and the estimated cost, risk and heritage of pre-antenna electronics. As a result
of the congested spectrum of X-band receives a cost score of 8, the Ku/Ka-bands are not yet congested but is rumored
to be allocated to terrestrial 5G receives a cost score of 5. However V-band has neither of the previously mentioned
problems, but it is limiting by its physical frequency and sophisticated electronics to operate with the spectrum, therefore
a score of 4 has been assigned. The optical spectrum is not regulated by the FCC but by FDA, technically it is not
regulated of speciﬁc frequency bands but rather by maximum power ﬂux hence optical receives a cost of 2.
C. Performance Algorithm
Our performance algorithm is deﬁned by Eq. (17). The purpose of the performance algorithm was to creating a
ranking systems for system down-selection.
Psys =PT x +GT x +GRT −LP+10 log10 (cPS )+10 log10 (lat) − k−Lrelay +gf−Latm +Rcl (17)
These variables are broken down and explained below.
•Transmit power (PT X )
–The power generated on the satellite is proportional to the available transmit power
–PTX =10 log10(1360 ∗0.3∗ηR A ∗S A)
–30% is the solar array eﬃciency
–ηR A varies with dynamic resource allocation and S A varies with satellite mass
•Transmitter gain (GTX )
–Equation being employed is dependent on the antenna technology.
∗Parabolic: GT X =20 log10 ηaπd
∗Phased Array: GT X =20 log10 4πS Aηa
∗Laser: GT X =10 log10 4π
Ωwhere Ωis the solid angle of the HPBW
–Assumed constant orbital altitude of 400 km throughout
–Varies with frequency and mass (S A)
•Receiver Gain on Noise (GRT )
–Based oﬀ O3b, Starlink, and PorTel ground systems
∗Parabolic, O3b: GRT =5.24
∗Phased Array, Starlink: GRT =10.8
∗Optical, PorTel from MIT STAR Lab: GRT =20 log10 πDr x
•Path Loss (LP)
–Assumed constant orbital altitude of 400 km
–Lp=10 log10 λ
•Capacity Per Satellite (cPS )
–Function of architecture, bandwidth, reuse factor, and crosslink factor
–The variable cl is an integer that scales with architecture decision and crosslink factor
The architecture choice is a vector of 1, 3, and 7, where each number represents the theoretical
number of connections a satellite in each architecture may have. One indicates no architecture, where
only downlink and uplink with the ground are an option. Three indicates a ring architecture where a
satellite in an orbital plane can transmit and receive from the ground, the lagging satellite, and the
leading satellite. Seven indicates a mesh architecture that can transmit and receive from the same
satellites as the ring, but also satellites in neighboring orbital planes.
The crosslink factor,
acts as a performance multiplier vector with values 1, 3, and 6 corresponding to
decisions: no crosslink, RF crosslink, and optical, respectively.
–cl =cl f ∗arch
Crosslinks add to the capacity of the system. It’s assumed at any given time, the satellite is free to use its
allocated bandwidth to perform crosslinks. The reuse factor further increases the usable bandwidth of the
–cps =r euse ∗BW +cl ∗BWC L .
–Varies by constellation architecture and crosslinks
Latency, a metric with units of time, was based on the system’s ability to get the information to ground once
it was received.
Assuming no supporting ground infrastructure to help route data to users, systems without crosslinks would
take an orbital period to deliver data to users.
With crosslinks, this latency was reduced by sending data to neighbors; within the crosslink design decision
are two architectures: ring and mesh. Ring architectures were limited to sending data to spacecraft directly
in front of or behind itself in the same orbital plane, making the latency factor smaller, but not as fast as it
could be. Mesh architecture allows for systems to communicate cross-plane and with no regard to where
the target spacecraft lies in formation as long as it’s in line-of-sight. This allows for the least amount of
•Relay Loss (LRelay) (adapted from )
–Bent Pipe: 10 log10(nsats )
Bent pipe systems are subject to noise and have no on-board processing to help preserve the received data.
Received signals are ﬁltered and ampliﬁed, but this leaves room for amplifying erroneous signals. This
implies each time a signal is passed through a bent-pipe satellite, the noise power increases. To account for
the worst-case of signal to noise ratio, it is assumed the signal will need to travel through every satellite in
–Regenerative: 10 log10 √nsats
Regenerative systems have the ability to demodulate and decode the received signal to better preserve the
data and mitigate the eﬀects of noise. This process greatly reduces the probability of making an error,
however it is not a perfect process and the SNR is degraded by a square-root factor of the worst-case
(traveling through all satellites).
•Global Factor (gf)
Assumes at any given time at any given location there are a number of satellites with 3dB extra power beam
down to user. Varies by coverage. The less coverage, the less assistance a satellite gets from a neighbor
•Atmospheric Losses (Latm )
Determined for each frequency band Assuming 50 km travelled in sea level atmospheric conditions,
Matlab’s internal gaspl function was used.
•Crosslink Rate (Rcl )
The rate at which satellites can transmit and receive to one another in space, directly related to the satellite
D. Reference Case: ViaSat-3
The new ViaSat-3 systems consists of three geostationary satellites which service the Americas, EMEA, and Asia
providing service by 2019, 2020, and 2021, respectively. The most recently published cost estimates have indicated a
price of $625M per satellite, although news has hinted towards price increase. The system is being built by Boeing and
is expected to have a lifespan of 15 years. The systems has circular polarization with a four frequency reuse pattern with
operators able to adapt the system where demand is unexpectedly higher or lower. The capacity of each satellite is
estimated to be >1 Tbps, with residential broadband expected to provide 100+ Mbps data rates. The system operates in
the Ka-band. The beam and channel characteristics can be seen in Tables 15 and 16.
From a business perspective, there is a strong focus on changing current residential broadband business plan. ViaSat
has removed hard cap Internet plans, where your service is restricted after a hard cap has been met and has replaced
it with a “soft cap”. In this system when the daily limit has been reached, the customer is able to use any available
bandwidth on the network, with speeds likely slower. While the heaviest streamers and gamers are not a good ﬁt for
satellite internet, ViaSat is targeting markets where customers currently only able to get 10 Mbps. The US Market
is relatively untapped with 10-15 million households being underserved or unserved by ISPs. Currently, ViaSat has
690,000 residential broadband subscribers in the United States. With ViaSat-1, the company is hoping to expand to
international markets. The targeted markets can be seen in Table 17. Trials have started in Northern Mexico for this
service. The the global demand for consumer broadband is viewed as “pent-up” and “just a question of bringing the
capacity online” by recent ViaSat Statements. The ground system for ViaSat-3 is expected to have 250 gateways with <2
meter antennas costing around $1M each. This is a paradigm shift from ViaSat-1 and ViaSat-2, which had 20 and 25
gateways, respectively and cost millions of dollars each.
Table 15 ViaSat-3 beam formulation, G/T, and EIRP
Characteristic A-Type Beam B-Type Beam
(Operation Functions) (Service to End Users)
Beams 20 72
G/T (dB/K) 17.1 - 21.9 18.2 - 22.7
Peak Downlink EIRP (dBW) 56.9 - 62.2 62.7 - 67.0
Table 16 ViaSat-3 frequency, modulation type, and channel access method for forward and return links
Characteristic Forward Return
28.1-29.1 GHz and 29.5-30.0
18.3-19.3 GHz and 19.7-20.2
16-APSK, 8PSK, and QPSK
Channel Access Method TDM 500 MHz wide carrier
MF-TDMA multiple band-
Table 17 ViaSat-3 targeted markets by location
Americas EMEA Asia∗
Consumer Broadband Residential Broadband Commercial Aeronautics
Commercial Aeronautics Commercial Aeronautics Government (US)
E. Reference Case: StarLink
In November 2016, SpaceX ﬁled for a Ka/Ku Band frequency allocation with the FCC. In the application, SpaceX
proposes using the spectrum to provide global broadband service via a Low-Earth Orbiting (LEO) satellite constellation
known as “Starlink”. In total, Starlink will be comprised of 4425 satellites in 5 altitudes using Ka/Ku band, with 1600 in
1 altitude, and 2825 in 4 diﬀerent altitudes. It should be noted SpaceX also ﬁled for a V Band frequency allocation for
7518 satellites in Very Low Earth Orbit (VLEO). Team Charlie did not consider the VLEO part of the system because
of the role deﬁnition (i.e. designing a state-of-the-art communication system in LEO), and to simplify the Starlink case
in the design space.
Each spacecraft has a projected average capacity of 20 Gbps, with an aggregate system capacity of 88.5 Tbps (full
system deployment). SpaceX advertises Starlink can provide end-users with up to 1 Gbps using space terminal phased
arrays, user terminal phased arrays, beamforming, optical crosslinks, and intelligent digital processing technology.
Beam and channel characteristics can be found in Tables 18 and 19. The use of space terminal phased arrays enables
electronically steered spot beams and the ability to employ beamforming. The use of user terminal phased arrays enables
precision spacecraft tracking and relaxes the mechanical requirements needed to slew.
SpaceX plans to incrementally deploy LEO spacecraft for operations. The “beta” deployment consists of 800,
single-altitude satellites that will provide service the US; the “initial” deployment will be the beta deployment and 800
more satellites (i.e. a 1600 total spacecraft) that provide global coverage to +/- 60
latitude; the “ﬁnal” deployment
consists of the remaining 2825 satellites that provide complete global coverage (pole-to-pole service).
Table 18 Starlink Beam Formation, G/T, and EIRP Schedule S Values
Characteristic Gateway Beam User Beam
(Operation Functions) (Service to End Users)
Beams 4 16
G/T (dB/K) 13.7 - 8.7 9.8 - 8.7
Peak Downlink EIRP (dBW) 39.44 - 0 39.44 - 0
Table 19 StarLink Frequency Allocation and Modulation Type
Characteristic Uplink Downlink
14.0 -14.5 10.7 - 12.7
27.5 - 29.1 17.8 - 18.6
29.5 - 30.0 18.8 - 19.3
47.2 - 52.4 37.5 - 42.5
Modulation Type BPSK, QAM OQPSK, QAM
F. Starlink Feasibility Study
Since the systems being considered are all direct competitors to Starlink, a review of publicly available information
was completed in order to determine the feasibility of Starlink from a business perspective. With the amount of
technology Starlink has reported to include on their satellites, including optical crosslinks, regenerative payloads,
V-band hardware, and electronically steered arrays, a separate study would need to be considered for the technical
and programmatic risks. It is unclear how much of this technology was included on the recent test ﬂights, where the
satellites weighed 400 kg [
] . The most recently released information anticipates an operational constellation of 800
satellites by 2020 [
]. There will be a total of 4,425 satellites by 2027, with a possible total of 7,725 satellites
beyond 2027 [
]. Estimates from industry sources indicate the cost of a Falcon 9 to be $37M [
]. The number of
employees is estimated to be at least 60 considering the current number of open positions (there are 33 open positions
]) and is expected to grow to 1,000 by 2020 [
]. While the current satellite cost remains unknown, the development,
launch, and satellite constellation costs are estimated in Table 20.
Table 20 Cost build up of Starlink
2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026
Employees 50 240 430 620 810 1000 1000 1000 1000 1000 1000 1000
0 0 0 100 450 800 1153 1506 1859 2213 2950 3688
Sats/Year 0 0 0 100 350 350 353 353 353 353 738 738
9 43 77 112 146 180 180 180 180 180 180 180
0 0 0 111 333 333 333 333 333 333 629 629
0 0 0 100 350 350 353 353 353 353 738 738
0 0 0 0.5 1.75 1.75 1.8 1.8 1.8 1.8 3.7 3.7
0 0 0 1 3.5 3.5 3.5 3.5 3.5 3.5 7.4 7.4
681 683 683 683 683 1,220 1,220
9.5 9.5 9.5 9,571 9.5 19.4 19.4
With the large number of satellites, the cost of the system is unsurprisingly very sensitive to the cost of the satellite.
Three diﬀerent satellite costs were estimated: $1M (low), $5M (med), and $10M (high), resulting in three diﬀerent total
constellation costs at $10B, $28, and $50B. SpaceX has publicly estimated the constellation cost to be $10B, as well,
suggesting a satellite cost of $1M [
]. Considering the amount of unproven technology on the satellites, this number
seems remarkably low. Likely, Starlink is relying signiﬁcantly on savings due to mass production. However, if parallels
may be made to Elon Musk’s other company, Tesla, mass production eﬃciency takes years to achieve and can be elusive.
From the NYTimes, the company aimed to produce “20,000 Model 3s a month by December. More recently, Tesla had
aimed to produce at least 2,500 Model 3s per week by the end of the ﬁrst quarter. But that is more than it managed in
the entire fourth quarter, and it has produced fewer than 10,000 in the ﬁrst quarter, the company said Tuesday,” [
The likelihood of making a satellite for <$1M without production expertise is very unlikely.
As can be seen from the results of the cost build up, Starlink quickly moves from a “feasible” $10B cost to multiple
tens of billions, still using satellite prices that are unheard of for the technologies involved, including optical crosslinks,
regenerative payloads, electronically steered arrays, and untested V-band ﬂight hardware. Historically, satellite endeavors
have only raised single-digit billions of dollars considering Iridium NexGen ($2.9B), OneWeb ($1.7B), and O3B ($1.2B)
[30–32]. Over 16 years, SpaceX itself has only raised $1.58B .
Considering the customer base that would generate revenue for this constellation, Starlink would require 1M
subscribers in the low case and 20M in the high case. This assumes that subscribers cannot start to use the service until
it is operational in 2020, and that the average price is $50 / subscriber (used in high case) or $150 / subscriber used in
low case), It also assumes a 40% margin. These margins are likely necessary in order for Starlink to generate a return
on investment and help fund BFR, which has been speculated. This margin was pulled from IntelSat’s 2017 10k for
reference. The current consumer broadband service market had 1.9 million subscribers in 2016, according to 2017:
State of the Satellite Industry report. This means that Starlink will need to at least double the current market in order to
make money in the best possible scenario report. In the worse case, Starlink would need to exponentially grow the
consumer base. Without investing in aﬀordable user terminals, which can cost up to $1,000, it will be diﬃcult to grow
the user base.
Additionally, Starlink appears to be a competing priority for BFR, which is consuming a large portion of SpaceX’s
R&D expenditure. It is unclear how SpaceX can focus on two capital-intensive long term investments at the same time.
SpaceX has had a single priority in the past of building rockets and moving to multiple competing priorities may prove
Finally, due to the unfeasibly low satellite cost of $1M, relatively low fundraising for space ventures of $2B, large
required number of subscribers, and SpaceX’s internal competing priorities for R&D funding, Starlink should not been
seen as a viable competitor with the currently stated constellation timelines, cost estimates, and conﬁguration.
This paper originated as a project from a graduate class taught by MIT Aeronautics and Astronautics Professors
Kerri Cahoy and Ed Crawley during Spring 2018.
“Space: Investment Implications of the Final Frontier,” URL
“The role of satellites in 5G ,” NetWorld 2020, Vol. 322, No. 10, pp. 891–921. doi:http://dx.doi.org/10.1002/andp.19053221004,
URL https://www.networld2020.eu/wp-content/uploads/2014/02/SatCom- in-5G_v5.pdf.
“ROLE OF SATELLITE SYSTEMS IN THE FUTURE 5G ECOSYSTEM,” European Space Agency. URL
“2018 Broadband Deployment Report,” URL
progress-reports/2018- broadband-deployment- report.
“2017 State of the Industry Report,” Satellite Industry Association. URL
Marc Sanchez, e. a., “Exploring the architectural trade space of NASAs Space Communication and Navigation Program,” IEEE
Aerospace. URL https://ieeexplore.ieee.org/document/6497173.
Gianluca Palermoa, P. G., Alessandro Golkarbc, “Earth Orbiting Support Systems for commercial low Earth orbit data relay:
Assessing architectures through tradespace exploration,” Acta Astronautica 111, pg 48-60. URL
“SpaceX Testing its Own Satellite Broadband Internet Network,” CNBC. URL
spacex-testing- its-own- satellite-broadband- internet-network.html.
Team, B., “FCC Broadband Deﬁnition Has Changed Before and Will Change Again,” URL
 Proakis, J. G., and Salehi, M., Communication Systems Engineering, 2nd ed., Prentice Hall, 2002.
“FCC approves SpaceX’s plan to provide broadband services with Starlink satellites,” GeekWire. URL
geekwire.com/2018/fcc-approves- spacexs-plan- provide-broadband- services-starlink-satellites/.
Inc, V., “Viasat, Boeing Enter Next Phase of ViaSat-3 Satellite Integration (Press Release),” URL
//investors.viasat.com/news-releases/news- release-details/viasat- boeing-enter- next-phase-
Inc, V., “Viasat Phased Array Flat Panel Antenna Selected by SES Networks for the O3b mPOWER System,”
https://www.viasat.com/news/viasat-phased- array-flat- panel-antenna- selected-ses-networks-
Matti Kummu, O. V., “The world by latitudes: A global analysis of human population, development level and environment
across the north-south axis over the past half century,” Applied Geography, 31(2):495-507. URL
Feder, B., “Globalstar Bankrupt Satellite Company to Be Sold for 55 Million,” URL
16/business/globalstar-bankrupt- satellite-company- to-be- sold-for-55-million.html.
Mellow, C., “The Rise and Fall and Rise of Iridium,” URL
and-rise- of-iridium- 5615034/.
 Wertz, J., Everett, D., and Puschell, J., Space Mission Engineering: The New SMAD, 1st ed., Microcosm Press, 2011.
 Wertz, J., and Larson, W., Space Mission Analysis and Design, 3rd ed., Springer, 1999.
David J. Hoﬀman, e. a., “THIN-FILM SOLAR ARRAY EARTH ORBIT MISSION APPLICABILITY ASSESSMENT ,” 17th
Space Photovoltaic research and Technology Conference. URL
David Krejci, e. a., “Emission Characteristics of Passively Fed Electrospray Microthrusters with Propellant Reservoirs,”
Spacecraft and Rockets, Vol. 54, No. 2 (2017), pp. 447-458. URL https://arc.aiaa.org/doi/10.2514/1.A33531.
“Satellite Orbital Decay Calculations,” The Australian Space Weather Agency. URL
“NASA FY2018 Budget,” URL
Clarke, R., “Expanding mobile wireless capacity: The challenges presented by technology and economics,” Telecommunications
Policy, Vol 38, Issue 8-9, pg. 693-708. URL https://doi.org/10.1016/j.telpol.2013.11.006.
Boyle, A., “Why net neutrality’s peril raises the stakes for future satellite broadband options,” URL
com/2017/net-neutralitys- peril-boost- prospects-global- satellite-broadband/.
de Selding, P. B., “SpaceX’s reusable Falcon 9: What are the real cost savings for customers?” SpaceNews. URL
http://spacenews.com/spacexs-reusable- falcon-9- what-are- the-real-cost-savings-for-customers/.
 SpaceX, “SpaceX Open Positions ,” URL https://www.spacex.com/careers/list?location%5B%5D=1276.
Boyle, A., “SpaceX ﬁles FCC application for internet access network with 4,425 satellites,” GeekWire. URL
//www.geekwire.com/2016/spacex-fcc- application-internet- 4425-satellites/.
Emre, K., “SpaceX’s Shotwell: Starlink internet will cost about $10 billion and ‘change the world’,” Florida Today.
internet-constellation- cost-10- billion-and- change-world/554028002/.
Boudette, N., “For Tesla, ‘Production Hell’ Looks Like the Reality of the Car Business,” New York Times. URL
“Iridium secures funding for $2.9bn satellite project,” Financial Times. URL
Henry, C., “OneWeb gets $1.2 billion in SoftBank-led investment,” SpaceNews. URL
gets-1- 2-billion- in-softbank- led-investment/.
McDermid, R., “Google-backed satellite provider O3b raises $1.2B to bring the world online,” Venture Beat. URL
Foust, J., “New funding round values SpaceX at $21.2 billion,” SpaceNews. URL
round-values- spacex-at- 21-2- billion/.