#### Geostationary orbit height calculation A geostationary orbi...
Here is an illustration of the footprint of a single geostationary ...
### Signal power degradation with distance in passive and active sa...
### Satellite communication frequency bands The first satellite...
### SATNET Also known as the Atlantic Packet Satellite Network, wa...
Today you can get a Satellite Data Hotspot for $250. E.g. [Global...
### Teledesic project Teledesic was an American company founded in...
When using a **Carrier-sense multiple access** MAC protocol the use...
#### Frequency-division multiple access (FDMA) FDMA involves divid...
To better visualize the problem being described, here is an animati...
### Autonomous systems (ASes) An AS is simply a network owned by a...
### BGP The Border gateway protocol, operates among ASes. It is co...
IEEE Communications Magazine • March 2001
Satellite-Based Internet: A Tutorial
0163-6804/01/$10.00 © 2001 IEEE
In a satellite-based Internet system, satellites
are used to interconnect heterogeneous network
segments and to provide ubiquitous direct Inter-
net access to homes and businesses. This article
presents satellite-based Internet architectures
and discusses multiple access control, routing,
satellite transport, and integrating satellite net-
works into the global Internet.
The Internet has enjoyed explosive growth in the
past few years. At the same time, the prolifera-
tion of new applications, and expansion in the
number of hosts (computers connected to the
Internet) and of users impose new technical
challenges to Internet development. New Inter-
net infrastructure and technologies capable of
providing high-speed and high-quality services
are needed to accommodate multimedia applica-
tions with diverse quality of service (QoS)
requirements. Furthermore, in order to provide
ubiquitous Internet access, appropriate mobility
support is required.
A satellite communication system, distin-
guished by its global coverage, inherent broad-
cast capability, bandwidth-on-demand flexibility,
and the ability to support mobility, is an excel-
lent candidate to provide broadband integrated
Internet services to globally scattered users. A
satellite system, if properly designed, can cover
the entire surface of the Earth, making it
extremely appealing to aeronautical and mar-
itime users, and to those in remote areas lack-
ing terrestrial communication infrastructure.
Even for the densely wired parts of the world, it
offers an alternative to the increasingly congest-
ed terrestrial links. A satellite network is inher-
ently a broadcast system. It is particularly
attractive to point-to-multipoint and multi-
point-to-multipoint communications which are
experiencing rapid development, especially in
broadband multimedia applications. Satellite
networks can serve as broadband access net-
works, high-speed backbone networks connect-
ing heterogeneous networks, or simply as
communication links between users with fixed
or mobile terminals.
However, the interoperation between a satel-
lite system and the existing terrestrial Internet
infrastructure introduces new challenges. This
article attempts to survey ongoing research
efforts on integrating satellite systems into the
global Internet. It clarifies the crucial technical
difficulties and provides insights for further
research. The rest of the article is organized as
follows. In the next section we present the basic
background on satellite systems. The article then
describes proposed satellite-based Internet archi-
tectures. Several technical issues in constructing
satellite-based Internet and some suggested solu-
tions are discussed. The final section provides a
summary and identifies some future research
A satellite system consists of a space segment
and a ground segment. The ground segment
consists of gateway stations (GSs), a network
control center (NCC), and operation control
centers (OCCs). The NCC and OCCs handle
overall network resource management, satellite
operation, and orbiting control. The GSs act as
network interfaces between various external net-
works and the satellite network. They also per-
form protocol, address, and format conversions.
The space segment is composed of satellites,
which may be classified into geostationary orbit
(GSO) and nongeostationary orbit (NGSO)
satellite, including medium earth orbit (MEO)
and low earth orbit (LEO) satellite, according to
the orbit altitude above the Earth’s surface.
GSO — The majority of satellites in operation
nowadays are placed in GSO orbit. The GSO
satellite is 35,786 km above the equator, and its
revolution around the Earth is synchronized with
the Earth’s rotation. Therefore, it appears fixed
to an observer on the Earth’s surface, and may
serve as a repeater in the sky. Its high altitude
allows each GSO satellite to cover approximately
one third of the Earth’s surface, excluding the
high latitude areas. The area of coverage of a
satellite is called its footprint. Three GSO satel-
lites are sufficient for global coverage. However,
the cost of launching GSO satellites is high. Due
to its high altitude and the inherent signal degra-
Yurong Hu and Victor O. K. Li, The University of Hong Kong
IEEE Communications Magazine • March 2001
dation with distance, large antennas and trans-
mission power are required for both the GSO
satellite and ground terminals. The most signifi-
cant problem is the large propagation delay for
GSO satellite links. The typical value of round-
trip delay is 250–280 ms, which is undesirable for
real-time traffic.
MEO and LEO — MEO’s distance from the
Earth’s surface is from 3000 km up to the GSO
orbit with a typical round-trip propagation delay
of 110–130 ms. LEOs are located 200–3000 km
above the Earth’s surface. For a LEO satellite the
round-trip delay is 20–25 ms, which is comparable
to that of a terrestrial link. Since LEO/
MEO satellites are closer to the Earth’s surface,
the necessary antenna size and transmission
power level are much smaller; but their footprints
are also much smaller. A constellation of a large
number of satellites is necessary for global cover-
age. The lower the orbit altitude, the greater the
number of satellites required. In addition, since
satellites travel at high speeds relative to the
Earth’s surface, a user may need to be handed off
from satellite to satellite as they pass rapidly over-
head. Therefore, steerable antennas are crucial to
maintain continuous service.
Satellite Payload — The satellite payload is
responsible for the satellite communication func-
tions. Once the satellite is launched, it is very
expensive and almost impossible to upgrade or
repair. The space environment, with radiation,
rain, and space debris, is harsh for satellites.
Therefore, the satellite payload is required to be
simple and robust. Traditional satellites, espe-
cially GSOs, serve as bent pipes. They act as
repeaters between two communication points on
the ground. There is no onboard processing
(OBP). It is simple and easy to implement. Some
satellite systems allow OBP, including demodu-
lation/remodulation, decoding/recoding,
transponder/beam switching, and routing to pro-
vide more efficient channel utilization. OBP can
support high-capacity intersatellite links (ISLs)
connecting two satellites within line of sight. By
using a sophisticated constellation with ISLs,
connectivity in space without any terrestrial
resource is possible.
Frequency Bands — The most commonly used
satellite frequency bands are C band (4–8 GHz),
Ku band (10–18 GHz), and Ka band (18–31
GHz). With a higher frequency band and a cor-
responding shorter wavelength, smaller antennas
can be used to receive the signal. Some satellite
systems use C band and thus employ large anten-
nas with minimum diameter of 2–3 m. The
majority of direct broadcast satellites use Ku
band for broadcasting as well as for Internet
connections from the server to the users, with a
terrestrial return link. A Ku band antenna can
be as small as 18 inches in diameter. There are
proposals to provide a Ka band return link for
these systems. Ka band potentially offers much
higher bandwidth than Ku band, and can use
very small antennas, but it suffers from environ-
mental impairments such as fading and rain
attenuation. There are also plans to use frequen-
cies beyond Ka band, but the technologies for
using those frequencies are immature and fur-
ther investigation is needed.
The satellite-based Internet has several architec-
tural options due to the diverse designs of satel-
lite systems, in orbit types (GSO, MEO, LEO),
payload choice (OBP or bent pipe), and ISL
designs. There are suggestions that multiple satel-
lite types (i.e., GSOs, MEOs, and LEOs) be
included in a hybrid GSO/NGSO network to fully
utilize the best characteristics of each orbit type.
A satellite network can serve as part of the
Internet backbone, a high-speed access network,
or both. Using a satellite system as part of the
Internet backbone has a long history that dates
back to the Atlantic SATNET interconnecting
ARPANET with European research networks
[1]. However, the idea of using satellites as a
solution of the last mile problem (i.e., connect-
ing users to network access points), inspired by
the usage of cost-effective very small aperture
terminals (VSAT) and improvements in satellite
technologies, is relatively new.
A typical satellite-based Internet scenario
with bent-pipe satellites is depicted in Fig. 1.
The idea of using
satellites as a
solution of the
last mile problem
(i.e., connecting
users to the
network access
points), inspired
by the usage of
VSATs and
in satellite
technologies, is
relatively new.
Figure 1. The satellite-based Internet with the bent-pipe architecture.
User A
Direct access users
User B
Gateway Gateway
IEEE Communications Magazine • March 2001
The satellites adopted can be GSO, MEO, or
LEO. It provides Internet access as well as data
trunking service. The satellite network interfaces
with the ground Internet infrastructure via GSs
on the Earth. It may be the only access method
for some users (e.g., user A) when no other
communication method is available, or a backup
connection in addition to an existing terrestrial
access network (e.g., user B).
However, the bent-pipe architecture’s lack of
direct communication paths in space results in
low spectrum efficiency and long latency. OBP
and ISLs may be used to help construct a net-
work in the sky (Fig. 2). This architecture is
again a combination access and backbone net-
work. Teledesic is one such system using a con-
stellation of 288 LEO satellites with ISLs [2].
The rich connectivity in space will provide more
flexibility but also bring complex routing issues,
which will be discussed later.
In Table 1 we summarize some proposed
worldwide broadband satellite systems that aim
to provide high-speed Internet service. We have
not included Iridium [3], which started operation
in 1998, since it is primarily designed for voice
and paging services.
In the two aforementioned general architec-
tures, user terminals are assumed to be interac-
tive, which means they can directly transmit data
up to the satellite and receive data from the satel-
lite. Although rapid advancements in satellite
technology have spawned small user terminals,
such as ultra small aperture terminals (USAT)
with 60 cm antennas, the interactive terminal is
still expensive and thus frustrates direct-to-home
implementation. Enlightened by Internet traffic
asymmetry where considerably more data is trans-
mitted from the server to the end user than in the
reverse direction (e.g., Web browsing), there is a
trend to offer Internet access via direct broadcast
satellites (DBSs) used for television broadcasting
[1]. Each home has a receive-only satellite dish to
collect data delivered in the high-speed satellite
broadcast channel. The reverse path to the server
is provided by a terrestrial link (Fig. 3). Hughes’s
DirecPC system is an example. In order to make
full use of the wide bandwidth of satellite broad-
cast links, DBS is also extended by using the
receive-only terminals as gateways to interconnect
remote networks.
The above architecture contradicts the tradi-
tional symmetric network assumption of two-way
balanced load and identical link characteristics,
and causes the so-called unidirectional routing
problem elaborated on in the next section.
In this section we summarize the technical chal-
lenges in designing and implementing the satel-
lite-based Internet. We focus on special
requirements unique to satellite systems, and
leave out general considerations common to ter-
restrial networks. Multiple access control
schemes, essential for satellite systems, are
described first. Then we investigate transmitting
IP packets in satellite networks. Finally, trans-
port issues based on TCP modification and satel-
lite-specific transport protocols are presented.
In interactive satellite systems, a large number of
user terminals widely scattered within the satellite
footprint contend for the satellite uplink channel.
Multiple access control (MAC), defined as a set
of rules for controlling access to a shared channel
Figure 2. The satellite-based Internet with the OBP and ISL architecture.
Direct access users
Figure 3. Internet access via DBS.
Reverse link
GW: Gateway
High-speed forward link
Multiple access
control, defined
as a set of rules
for controlling
access to a
shared channel
contending users,
plays an
important role in
efficiently and
fairly utilizing the
limited satellite
system resources.
IEEE Communications Magazine • March 2001
among contending users, plays an important role
in efficiently and fairly utilizing the limited satel-
lite system resources. MAC protocol perfor-
mance can significantly affect higher-layer
protocols and the QoS provided by the system.
The performance of MAC protocols depends
on the characteristics of both the shared commu-
nication media and the traffic. The long latency
in satellite channels (especially the GSO links)
excludes some MAC schemes used in terrestrial
local area networks (LANs) such as carrier sense
multiple access (CSMA), and the limited power
resource in satellites constrains the transponder
and computational capacity on board the satel-
lite. Internet traffic turns out to be bursty in
nature. Besides the current best-effort service,
the Internet is expected to provide diverse QoS
guarantees (e.g., on delay, delay jitter, packet
loss ratio) for a wide range of traffic types. Thus,
a candidate MAC protocol must implement pri-
orities. Real-time traffic with transmission dead-
lines is usually given higher priority than
non-real-time traffic.
Generally speaking, a good MAC scheme for
a satellite-based network should be simple to
implement, robust, and flexible to accommodate
network reconfiguration. The MAC should be
able to achieve high throughput, maintain chan-
nel stability, and enjoy low protocol overhead
and small access delay.
Depending on how bandwidth is allocated
among all contenders, candidate MAC schemes
for satellite systems can be categorized into
three groups: fixed assignment, random access,
and demand assignment.
Fixed Assignment — Fixed assignment may be
made on a frequency, time, or code basis. Major
techniques include frequency-division multiple
access (FDMA), time-division multiple access
(TDMA), and code-division multiple access
(CDMA). In FDMA and TDMA systems, each
station utilizes its own dedicated channel. They
are contention-free, and can provide QoS guar-
antees. However, this is at the expense of ineffi-
cient utilization of resources. Their lack of flexi-
bility and scalability makes them only suitable
for small-scale networks with stable traffic pat-
terns. FDMA was the first fixed assignment mul-
tiple access method used in satellite systems.
TDMA is popular mainly because of its compati-
bility with the nonlinear nature of transponders
and is used in the majority of current satellite
systems. In a CDMA system, each user is
assigned a unique code sequence which is used
to spread the data signal over a wider bandwidth
than that required to transmit the data. If code
sequences are guaranteed to be orthogonal, all
other simultaneous transmissions in the same
channel act as additive interference to the
desired signal and can be removed completely at
the receiver side, where a reverse procedure,
despread, is taken to recover the original data.
Thus, in a CDMA system, the whole bandwidth
is used by all users, making it more flexible for
system expansion.
Random Access — Due to technological
advances, small and inexpensive terminals (i.e.,
VSATs and USATs) with lower data rates are
now widely available, thus stimulating home or
personal use of satellite access service. The num-
ber of stations within a satellite footprint increas-
es from a few to several hundreds or thousands.
In addition, the traffic generated by each user is
very bursty. Fixed assignment schemes are
replaced by contention-based random access (i.e.,
Aloha and its variations). In random access
schemes, each station transmits data regardless of
the transmission status of others. Retransmissions
after collision increase the average packet delay,
and frequent collisions may cause low throughput.
Demand Assignment — Although random
access may better accommodate a large number
of terminals with bursty traffic, it provides no
QoS guarantees. Demand assignment multiple
access (DAMA) protocols attempt to solve this
Table 1. A summary of proposed worldwide broadband satellite systems.
System Major sponsors Constellation Satellite payload Frequency band Data rate Service date
Astrolink Lockheed Martin Up to 9 GSO OBP and ISLs Ka Up to 200 Mb/s 2003
satellites downlink
Up to 20 Mb/s uplink
Skybridge Alcatel Espace, 80 LEOsatellites Bent-pipe Ku 16 kb/s–20 Mb/s downlink 2002
Loral Space at 1469 km 16 kb/s–2 Mb/s uplink
Spaceway Hughes 4 GSO satellites OBP and ISLs Ka Up to 92 Mb/s downlink 2002
(ultimately 21 16 kb/s–6 Mb/s uplink
Teledesic Motorola, 288 LEO satellites OBP and ISLs Ka 16 kb/s–64 Mb/s downlink 2004
Lockheed Martin at 1375 km 16 kb/s–2 Mb/s uplink
This section describes protocols used by those interactive
satellite systems which use a satellite link for the reverse
link from the users to the Internet servers, and are not
applicable to those systems, such as DirecPC, which use a
terrestrial reverse link.
In order to efficiently utilize the power of transponders,
we must drive them into a saturation area where the
amplifiers operate as nonlinear devices. In an FDMA sys-
tem, users’ signals may be received simultaneously, and
the nonlinear amplifier generates undesired interference.
In TDMA, only one user accesses the transponder at any
given time interval, and the problem is avoided.
IEEE Communications Magazine • March 2001
problem by dynamically allocating system band-
width in response to user requests. A resource
request must be granted before actual data trans-
mission. The transmission of requests is itself a
multiple access problem. However, since the
request message is typically much shorter than
actual data transmission, we can afford to have
reservation requests collide and be retransmitted.
After a successful reservation, bandwidth is allo-
cated on an overall FDMA or TDMA architec-
ture, and data transmission is guaranteed to be
collision-free. This article focuses on the TDMA
architecture in which equal-sized time slots are
grouped into frames, repeated periodically.
The reservation may be made under central-
ized or distributed control. The central con-
troller can be located at an Earth station or at a
satellite with OBP. For a ground-based con-
troller, the minimum request delay is two round-
trip times before the reservation request is
granted. The minimum request delay can be
halved for a space-based controller, but the
satellite payload capacity limits this implementa-
tion. Distributed control, in which each station
receives all request information from the satel-
lite broadcast channel and makes a decision on
its own, is more robust and reliable. The channel
overhead associated with reservation announce-
ments is reduced greatly, and the minimum
reservation delay is as small as one round-trip
time. Although it may put the processing burden
on the stations, distributed control is still pre-
ferred, considering its overall advantages.
Resource reservation can be made either
explicitly or implicitly. Explicit reservation is on
a per-transmission basis, and usually a dedicated
reservation channel is shared among all stations.
Each station sends a short request via the reser-
vation channel specifying the number of time
slots needed. Stations access the reservation
channel in fixed assignment mode, such as
TDMA reservation, or random access mode,
such as Aloha reservation. Data is transmitted in
the data channel after successful reservation.
In implicit reservation, there is no explicit
reservation message, and a successful data trans-
mission in a slot serves as an indication of reser-
vation for the corresponding time slot in
subsequent frames. Packets belonging to a long
transmission can repeatedly occupy the same slot
in consecutive frames. An empty slot in a frame
indicates the end of the transmission, and other
users can then contend for this slot starting in the
next frame. This scheme is attractive for relatively
steady traffic patterns such as voice and video
connections. An example of this scheme is Reser-
vation Aloha, which uses the first data packet as
an implicit request unit and accesses the available
time slots via the slotted Aloha protocol.
Priority-Oriented Demand Assignment
(PODA) and first-in first-out (FIFO) Ordered
Demand Assignment (FODA) [4] combine
implicit and explicit requests. Each PODA
TDMA frame consists of a control part and a
data part with an adjustable boundary. Explicit
requests contend in the control part by slotted
Aloha, while implicit requests are piggybacked
on data packets. FODA further divides the data
part into stream and datagram subframes. One
FIFO stream queue and two FIFO datagram
queues, for short interactive traffic and bulky
traffic, are maintained with decreasing priorities.
The latter two types of traffic are transmitted in
the datagram subframes.
There are proposals to make use of the unre-
served resource after the demand assignment.
Combined free/demand assignment multiple
access (CFDAMA) [5] freely assigns remaining
channels according to some strategy (e.g., round-
robin). In combined random access and TDMA-
reservation multiple access (CRRMA) [6],
remaining resources are open for random access.
By randomly accessing the unreserved channel,
some bursty interactive traffic may be transmitted
immediately without waiting for the two-hop
reservation delay. A hybrid scheme called Round-
Robin Reservation (RRR) is based on fixed
TDMA. The number of stations is required to be
less than or equal to the number of time slots.
Each station obtains a dedicated channel, and
extra or unused slots are accessed in a round-
robin manner or via slotted Aloha. In satellite sys-
tems, large GSs interfacing with terrestrial
networks function as multiplexers for traffic from
those directly connected networks. Gateway sta-
tions are usually much more heavily loaded than
small terminals, and the number of GSs is much
smaller than that of small terminals. The RRR
mechanism may be suitable for this scenario. For
example, GSs may obtain dedicated time slots,
while small terminals contend for the remaining
slots in each frame. Similar hybrid methods may
combine the advantages of different schemes, but
further performance analyses and simulations are
needed to demonstrate their feasibility.
Routing Issues in a LEO Constellation
The significant advantages of LEO with OBP
and ISLs, such as small delay and full connectivi-
ty, make it a very attractive approach to the
Internet in the sky. In such networks, the major
technical issue is the complex dynamic routing
issue due to satellite movements.
Dynamic Topology — Due to the relative
movement between the LEO satellite and the
Earth, a satellite has a very short visible period
to motionless users on the ground. To maintain
24-hr continuous coverage, a carefully designed
satellite constellation is crucial. At any time
there should be at least one satellite within line
of sight of a user. When a satellite moves out of
a user’s visual field and another satellite moves
in, intersatellite handover happens. For a satel-
lite with multiple antennas and transponders, the
satellite footprint is divided into a number of
spotbeams, each covered by an antenna beam.
Thus, frequent interbeam handover from spot-
beam to spotbeam occurs within a single satel-
lite’s visible period.
The ISLs in the constellation form a mesh net-
work topology. Each satellite is typically able to
set up 4–8 ISLs. There are two types of ISLs:
intraplane ISLs connecting adjacent satellites in
the same orbit, and interplane ISLs connecting
neighboring satellites in adjoining orbits. Intra-
plane ISLs are maintained permanently, but some
interplane ISLs may be temporarily switched off
A hybrid scheme
Reservation is
based on fixed
number of
stations is
required to be
less than or equal
to the number of
time slots. Each
station obtains a
channel, and
extra or unused
slots are accessed
in a round-robin
manner or via
slotted Aloha.
IEEE Communications Magazine • March 2001
when the viewing angle or distance between two
satellites changes too fast for the steerable
antennas to follow. This may occur between two
counter-rotating orbits or when two orbits cross.
The routing scheme should be able to handle
topological variations. Fortunately, although the
constellation topology changes frequently, it is
highly periodic and predictable because of the
strict orbital movements of the satellites.
Some dynamic routing mechanisms popular
in the Internet, such as distance vector (DV)
and link state algorithm (LSA), are not directly
applicable in satellite constellation routing,
because frequent topological changes in satellite
constellation will cause large overhead and oscil-
lation if these schemes are used. Two new con-
cepts tailored to dynamic satellite constellation
are worth mentioning: discrete-time dynamic vir-
tual topology routing (DT-DVTR) [7] and the
virtual node (VN) [8].
DT-DVTR — DT-DVTR makes full use of
the periodic nature of satellite constellation
and works completely offline. It divides the
system period
into a set of time intervals
so that the topology changes only at the
beginning of each time interval and remains
constant until the next time interval. In
each interval, the routing problem is a static
topology routing problem that can be solved
easily. A number of consecutive routing
tables are then stored onboard and
retrieved when the topology changes. With
this strategy, online computational com-
plexity is transformed into a large storage
requirement on the satellites. In order to
minimize the storage needed and the inter-
satellite handover attempts when topology
changes, an optimization procedure can be
used to choose the best path or a small set
of paths from the series of instantaneous
routes. Although it can significantly reduce
the storage size, some links may become
congested while others are underutilized.
VN — The objective of this scheme is to hide
the topology changes from the routing pro-
tocols. A virtual topology is set up with VNs
superimposed on the physical topology of
the satellite constellation. Even as satellites
are moving across the sky, the virtual topolo-
gy remains unchanged. Each VN keeps state
information, including routing tables and
information of users within the VN’s cover-
age area. In a certain period, a VN is repre-
sented by a certain physical satellite. As this
satellite disappears over the horizon, the VN
is represented by the next satellite passing
overhead. The state information is also
transferred from the first satellite to the sec-
ond. A routing decision is made on the virtu-
al topology, and the protocols are not aware
of the dynamic satellite constellation con-
cealed in state transfers.
Based on these two concepts, some routing
schemes are proposed for carrying IP packets
through the satellite constellation. Some com-
mercial satellite systems (e.g., Teledesic) use
proprietary routing techniques that are highly
dependent on explicit orbital and constellation
knowledge and optimized for specific designs.
Due to their lack of generality, they are not cov-
ered in this article.
IP Routing at the Satellites — To route IP
packets through a satellite constellation, it seems
straightforward to adopt IP routing at the satel-
lites. This strategy is addressed in [9], and is based
on the VN concept. It can seamlessly integrate the
space network with the terrestrial Internet, and
permits direct support for IP multicast and IP
QoS (integrated and differentiated service mod-
els). However, how to deal with variable-length IP
packets, the scalability problem of onboard rout-
ing tables, and computational and processing
capacity limitations in space devices are challeng-
ing problems. The scheme is still in its infancy,
and some practical problems are unsolved in the
implementation of the VN concept.
ATM Switching at the Satellites — Many pro-
posed systems use ATM as the network protocol
for the constellation (i.e., Cyberstar, Astrolink,
Spaceway, and Skyway) with a satellite-specific
signaling protocol and link layer protocol [10]. An
ATM version of DT-DVTR is investigated in [7],
where all the virtual channel connections between
the same pair of ingress and egress satellites are
grouped into a virtual path connection (VPC),
and onboard switching is done according to the
VPC labels. A modified S-ATM packet is suggest-
ed to reduce the overhead without changing the
cell size [10]. If such a system is adopted to pro-
vide Internet service, IP over ATM or other simi-
lar technologies will be used.
External Routing Issues — It is reasonable to
assume that the internal routing schemes for
satellite constellation will continue to be hetero-
geneous. Satellite manufacturers and operators
will probably select routing methods best suited
to their own system designs. The internal proto-
col should be kept simple. Details of the satellite
network should be hidden from the terrestrial
Internet as well. Today’s Internet achieves this
kind of isolation by using the autonomous sys-
tem (AS) concept. Typically, some external rout-
ing schemes are used for inter-AS routing, while
internal routing is handled by the AS’s own
internal routing protocol.
A satellite system can be considered an AS in
the Internet (Fig. 4) [9]. A number of border
gateways (BGs) running exterior routing proto-
cols (e.g., Border Gateway Protocol, BGP, used
by terrestrial ASs) will communicate with terres-
trial ASs. Only BGs on the constellation periph-
ery need be aware of the outside addresses and
topological information. All packets going
through a satellite constellation enter the satel-
lite AS from one entry BG, which is responsible
for determining the exit BG of each packet. If
necessary, the entry/exit BGs perform encapsula-
tion/decapsulation and address resolution. The
BGs can be implemented either onboard the
satellites or in ground GSs. If space-based BGs
are used, computational and storage require-
ments may be too much for the satellites. On the
other hand, if terrestrial gateways are used,
How to deal with
variable length IP
packets, the
problem of
on-board routing
tables, and
and processing
limitations in
space devices are
problems. The
scheme is still in
its infancy, and
some practical
problems are
The system period is the least common multiple of the
orbit period and the Earth’s rotation period.
IEEE Communications Magazine • March 2001
packets must be bounced back to the ground for
IP routing, introducing an extra round-trip delay.
However, ground BGs are more realistic. Fur-
thermore, external routing protocols popular in
terrestrial networks cannot simply be reused in
satellite constellations. In terrestrial networks,
any internal link within an AS always has smaller
cost than an inter-AS link. It is generally true
because in terrestrial networks an AS is usually
limited to a small geographical area, while an
inter-AS link travels a much longer way. But a
satellite system extends globally, and routing
within a satellite constellation may be as expen-
sive as traversing several ASs. Thus, multiple
pairs of BGs should be used from a satellite con-
stellation to a destination in some AS.
Unidirectional Routing — As described earlier,
Internet access via DBS poses the unidirectional
routing problem which cannot be handled by tra-
ditional dynamic routing schemes where bidirec-
tional links are assumed. For example, in distance
vector routing, a router receiving the distance vec-
tor tuple {destination, cost} from its neighbor
deduces that it can reach the destination via this
neighbor. It is no longer true in the satellite
broadcast scenario where the direct reverse link
to the satellite does not exist. Multicast routing
protocols (e.g., DVMRP) based on the reverse
shortest path tree also face such a problem [11].
There are three ways to handle this problem.
Instead of dynamic routing, static routing may
be an option; but with thousands of users served
by a DBS, it is impossible to manually configure
all of the routing entries. The other two methods
are routing protocol modification and tunneling,
proposed in the Internet Engineering Task Force
(IETF) Unidirectional Link Routing (UDLR)
working group. They are discussed next.
Routing Protocol Modification — In unidi-
rectional routing, the router at one end of a uni-
directional link with a send-only interface is
referred to as a feeder, while the router at the
other end of the unidirectional link with a
receive-only interface is called a receiver. The
key idea of the modification is twofold. First, the
modified protocol should enable a receiver to
identify the potential feeders whenever it
receives routing updates from them, and to
ignore the unusable routing information in those
packets while keeping the useful reports to
maintain the neighboring connectivity [11]. Sec-
ond, the receiver periodically delivers its own
routing message to all feeders through the ter-
restrial reverse channel. Thus, when a feeder
gets the routing information, it can update the
related routing entries for reachable destinations
through the unidirectional link passing the
receiver. The idea is used by the UDLR working
group in the proposals to modify some popular
protocols (i.e., RIP, OSPF, and DVMRP).
Tunneling — Tunneling offers a link layer
approach to hide the network asymmetry from
the routing process. A virtual bidirectional link is
set up between a DBS and a user by encapsula-
tion and decapsulation. This virtual link is called
a tunnel. Packets destined for the DBS from the
user are delivered via the tunnel. The tunnel end-
point at the user side first encapsulates the pack-
et, and then passes it to the routing protocol
where it is delivered through the actual terrestrial
reverse channel. When the packet arrives at the
satellite, the tunnel endpoint captures it, decapsu-
lates it, and forwards it to the routing protocol to
which it seems to come from a bidirectional link.
The above two approaches are simple, and
since tunneling is transparent to all upper layer
protocols, it may be quickly implemented in
DBS Internet access architectures. However, the
two schemes are designed based on point-to-
point unidirectional links, although satellites are
point-to-multipoint broadcast systems. Thus, fur-
ther study is needed to design new approaches
optimized for this architecture. The two
approaches also focus only on the routing issue
within a single AS and fail to address interdo-
main routing. New interdomain routing schemes
that can handle unidirectional links are needed.
The TCP/IP and UDP/IP protocol suites form the
basis of the Internet. Due to their tremendous
legacy, it is unlikely they will be totally discarded
in the near future. Therefore, the satellite-based
Internet is expected to continue to serve applica-
tions based on TCP and UDP. However, the per-
formance of both protocols will be affected by the
long latency and error-prone characteristics of
satellite links. The impacts on TCP will be much
greater, and heated debates have been spawned
regarding the feasibility of TCP in a satellite envi-
ronment. Researchers working with NASA’s
ACTS satellites are performing research regard-
ing TCP/IP over satellite connections. The IETF
TCP over Satellite working group is also dedicat-
ed to improvement of TCP performance in satel-
lite systems. In this section we first present the
main limitations of TCP over satellite links and
then summarize the ongoing research efforts on
the satellite transport problem.
TCP Performance over Satellite — TCP uses
a positive feedback mechanism to achieve rate
control and reliable delivery. The long latency of
Figure 4. A LEO constellation's autonomous system.
LEO constellation
IP router
External AS
IEEE Communications Magazine • March 2001
satellite links (especially GSO links) increases
the TCP end-to-end delay and results in sluggish
acknowledgments. The slow feedback will weak-
en the functionality of rate control and conges-
tion avoidance, and thus affect the throughput.
In addition, a potential problem is the large fluc-
tuation of measured round-trip time (RTT) that
may be caused by dynamic topology in LEO con-
stellation networks. Large variations in RTT
measurements may result in false timeouts and
In the initial slow start stage of TCP transmis-
sion, although the sending rate increases exponen-
tially, it is still too slow for the high-bandwidth
satellite links. One proposed solution is to increase
the initial value of the window. TCP originally
allows a window size of 64 kbytes, which also lim-
its the maximum sending rate to 64 kbytes/RTT.
The satellite link will be underutilized, and a large
window scaling up to the bandwidth-delay product
of the satellite link is required for higher through-
put. A set of window scaling options to TCP imple-
mentation are defined in IETF Request for
Comments (RFC) 1323.
Satellite links are subject to various impair-
ments (i.e., interference, fading, shadowing, and
rain attenuation). Therefore, a high bit error rate
(BER) is expected. Although advanced modula-
tion, coding schemes, and forward error correction
(FEC) techniques are used to reduce the BER, in
some environments high BER persists. But TCP
does not distinguish between corrupted data caused
by transmission error and packet loss due to con-
gestion; both are unacknowledged and interpreted
as a notification of network congestion. When
there is a corrupted packet, the window size is
halved even though there is no congestion. Fur-
thermore, transmission errors on a satellite link are
bursty in nature, especially under bad weather con-
ditions. Bursty errors in one RTT will dramatically
reduce the throughput. The Space Communica-
tions Standards-Transport Protocol (SCPS-TP)
[12] defined for the general space environment
provides two mechanisms to distinguish the sources
of loss and responds differently. In addition, net-
work asymmetry can also impair TCP perfor-
mance. Satellite network asymmetry occurs in two
situations. One is in the asymmetric DBS Internet
access architecture described earlier. The other is
due to bandwidth asymmetry in some interactive
satellite terminals. These terminals may be capable
of downloading at tens of megabits per second, but
with uplink speed of only several hundred kilobits
per second. The limited reverse link capacity may
cause the acknowledgment starvation problem.
The backlogged feedback will slow down the win-
dow refresh. In addition, acknowledgment loss due
to reverse link congestion may trigger unnecessary
Another problem inherent in TCP is the fair-
ness issue between different TCP connections
with various RTTs. When those TCP connec-
tions share a bottlenecked link, the TCP connec-
tions with longer RTTs will suffer unfair
bandwidth allocation.
Performance Enhancements — The IETF
TCP over Satellite working group has recently
made a number of recommendations to enhance
the performance of TCP over satellite links in its
RFCs. The last two schemes listed below are
non-TCP techniques:
TCP selective acknowledgment-(SACK)
options (RFC2018) allow the receiver to
specify the correctly received segments.
Thus, the sender needs to retransmit only
the lost packets. TCP SACK can recover
multiple losses in a transmission window
within one RTT.
TCP for transaction (T/TCP) (RFC1644)
attempts to reduce the connection hand-
shaking latency from two RTTs to one
RTT, which is a significant improvement
for short transmissions.
Persistent TCP connection, supported in
HTTP1.1 (RFC2068), allows multiple small
transfers to download in a single persistent
TCP connection. It is more efficient.
The Path maximum transfer unit (MTU)
discovery mechanism allows TCP to use the
largest possible packet size, thus avoiding
IP segmentation. It reduces the overhead,
and eliminates fragmentation and defrag-
• FEC is employed in link layer protocols to
improve the quality of satellite links, but it
should not be expected to fix all problems
associated with manmade noise, such as
military jamming, and some natural noise,
such as that caused by rain attenuation.
Besides FEC, some other link layer
approaches (e.g., bit interleaving, link layer
automatic repeat request schemes) can also
be used to improve packet error rate in
transmissions over satellite links.
TCP extensions can solve some of the limita-
tions of standard TCP over satellite links, but
other problems such as long end-to-end latency
and asymmetry are not effectively addressed.
One way to alleviate the effects of large end-to-
end latency is to split the TCP connection into
two or more parts at the GSs connecting the
satellite network and terrestrial networks. There
are three approaches to splitting TCP connec-
tions over satellite links:
TCP spoofing: The divided connections are
isolated by the GSs, which prematurely
send spoofing acknowledgments upon
receiving packets. The GSs at split points
are also responsible for retransmitting any
missing data. The performance of TCP
spoofing is examined in [13].
TCP splitting: Instead of spoofing, the con-
nection is fully split. A proprietary transport
protocol can be used in a satellite network
without interference to standard TCP in ter-
restrial networks [14]. It is more flexible, and
some kind of protocol converter should be
implemented at the splitting points.
Web caching: In contrast to the above two
schemes, the TCP connection is split by a
Web cache in the satellite network. Users in
the satellite network connected to this Web
cache need not set up TCP connections all
the way to servers outside if the required
contents are available from the cache. Web
caching effectively reduces connection laten-
cy and bandwidth consumption.
In [14], a Satellite Transport Protocol (STP)
is designed and used in the TCP splitting
The satellite-
based Internet is
expected to
continue to serve
based on TCP
and UDP.
However, the
performance of
both protocols
will be affected
by the long
latency and
characteristics of
satellite links.
IEEE Communications Magazine • March 2001
approach as well as for traffic management in a
satellite network. STP is based on the basic
operation of Service-Specific Connection-Orient-
ed Protocol (SSCOP). The sender periodically
requests the receiver to report successful recep-
tions, and retransmission is triggered by explicit
selective negative acknowledgment. It uses a
hybrid window and rate congestion control
mechanism. STP performs well in asymmetric
networks since the reverse traffic is significantly
reduced, but it does not distinguish between dif-
ferent sources of packet loss and also leaves the
fairness problem unsolved.
In this article we present an introduction to the
satellite-based Internet. Some possible architec-
tures based on bent-pipe and OBP satellites are
discussed. Several technical challenges, including
multiple access control, IP routing in LEO con-
stellations, unidirectional routing, and satellite
transport issues are investigated. In addition to
what we elaborate on in this article, some impor-
tant research issues are identified as follows:
IP QoS support. There is no lack of
research regarding QoS support in satellite
systems. However, most of them are based
on ATM QoS classes [8, 15], and the map-
ping of ATM service classes to IP QoS
requirements is a nontrivial problem.
Moreover, the implementation of TCP/IP
over ATM brings much overhead, extra
processing time, and protocol complexity.
Direct support of the Internet integrated
or differentiated service model is desired.
In [9], multiprotocol label switching
(MPLS) is proposed to support Internet
QoS (integrated or differentiated service)
in a satellite-based network. The integra-
tion of space and terrestrial communica-
tion systems, internetworking different
satellite networks, and the advent of hybrid
satellite systems will bring more redundan-
cy and routing choices. QoS routing in
satellite systems will be a very important
research problem.
Traffic and congestion control. To ensure
that the satellite network achieves desired
performance and fulfills the IP QoS require-
ments, a set of mechanisms to control traffic
and avoid congestion is required. A well-
designed MAC protocol will not by itself
prevent congestion in the network. Traffic
management, traffic shaping, policing, and
scheduling are also required. Some preven-
tive congestion control schemes, such as
admission control, and efficient congestion
notification schemes are important to main-
tain the specified QoS guarantees.
[1] H. D. Clausen, H. Linder, and B. Collini-Nocker, Internet
over Direct Broadcast Satellites, IEEE Commun. Mag.,
June 1999, pp. 14651.
[2] D. J. Bem, T. W. Wieckowski, and R. J. Zielinski, Broad-
band Satellite Systems, IEEE Commun. Surveys, vol. 3,
no. 1, 2000.
[3] S. R. Pratt et al., An Operational and Performance
Overview of the IRIDIUM Low Earth Orbit Satellite Sys-
tem, IEEE Commun. Surveys, vol. 2, no. 2, 1999.
[4] N. Celandroni, and E. Ferro, The FODA-TDMA Satellite
Access Scheme: Presentation, Study of the System, and
Results, IEEE. Trans. Commun., vol. 39, no. 12, Dec.
1991, pp. 182331.
[5] Le-Ngoc and S. V. Krishnamurthy, Performance of
Combined Free/Demand Assignment Multiple-Access
Schemes in Satellite Communications, Int’l. J. Satellite
Commun., vol. 14, no. 1, Jan./Feb. 1996, pp. 1121.
[6] H. W. Lee and J. W. Mark, Combined Random/Reserva-
tion Access for Packet Switched Transmission over a
Satellite with Onboard Processing: Part I Global
Beam Satellite, IEEE Trans. Commun., vol. COM-31,
Oct. 1983, pp. 116171.
[7] Markus Werner, A Dynamic Routing Concept for ATM-
Based Satellite Personal Communication Networks,
IEEE JSAC, vol. 15, no. 8, Oct. 1997, pp. 163648.
[8] R. Manger and C. Rosenberg, QoS Guarantees for Mul-
timedia Services on a TDMA-Based Satellite Network,
IEEE Commun. Mag., July 1997, pp. 5665.
[9] L. Wood et al., IP Routing Issues in Satellite Constella-
tion Networks, Int’l. J. Satellite Commun., to appear.
[10] I. Mertzanis et al., Protocol Architectures for Satellite
ATM Broadband Networks, IEEE Commun. Mag., Mar.
1999, pp. 4654.
[11] W. Dabbous, E. Duros, and T. Ernst, Dynamic Routing
in Networks with Unidirectional Links, Proc. 2nd Int’l.
Wksp. Satellite-Based Information Services (WOSBIS
’97), Budapest, Hungary, Oct. 1997.
[12] R. C. Durst, G. J. Miller, and E. J. Travis, TCP Exten-
sions for Space Communications, ACM MobiComm
’96, Nov. 1996.
[13] M. Allman, H. Kruse, and S. Ostermann, TCP Perfor-
mance over Satellite Links, WOSBIS ’96), Nov. 1996.
[14] T. R. Henderson and R. H. Hatz, Transport Protocols
for Internet-Compatible Satellite Networks, IEEE JSAC,
vol. 72, no. 2, Feb. 1999, pp. 32644.
[15] R. Goyal et al., Traffic Management for TCP/IP over
Satellite ATM networks, IEEE Commun. Mag., Mar.
1999, pp. 5661.
YURONG HU (yrhu@eee.hku.hk) received her B.E. from
Tsinghua University, China, in 1999. She is currently an
M.Phil. student in the Department of Electrical and Elec-
tronic Engineering at the University of Hong Kong, China.
Her research interests are in the areas of satellite networks,
multimedia communications, and Internet QoS.
ICTOR O.K. LI [F] (vli@eee.hku.hk) received his S.B., S.M.,
E.E., and Sc.D. degrees in electrical engineering and com-
puter science from the Massachusetts Institute of Technol-
ogy in 1977, 1979, 1980, and 1981, respectively. He is
Chair Professor of Information Engineering at the Universi-
ty of Hong Kong, and Managing Director of Versitech Ltd.,
the technology transfer and commercial arm of the univer-
sity. Previously, he was professor of electrical engineering
at the University of Southern California (USC), Los Angeles,
and director of the USC Communication Sciences Institute.
He has published over 200 technical articles, and has lec-
tured and consulted extensively around the world. His
research interest is in information technologies, focusing
on the Internet and wireless networks.
One way to
alleviate the
effects of large
latency is to split
the TCP
connection into
two or more
parts at the
gateway stations
connecting the
satellite network
and terrestrial


### Signal power degradation with distance in passive and active satellites For a **passive** satellite (effectivel just a reflector, no amplification) the following expression can be used to calculate the received signal power: $$ P_{r} = \frac{P_{t} \cdot G_{t} \cdot G_{r} \cdot \sigma}{(4\pi)^{2} \cdot d_{t}^{2} \cdot d_{r}^{2}}$$ Where: * $P_{t}, P_{r}$: Transmitter and receiver power * $G_{t}, G_{r}$: Gain of transmitter and receiver antennas * $d_{t}, d_{r}$: Distance of ground stations to satellite * $\sigma$: satellite gain/loss factor Assuming $d_{t} \approx d_{r}$ we get: $$ P_{r} = K \cdot \frac{\sigma}{d^{4}} \cdot P_{t} $$ As we can see the received power signal is inversely proportional to the fourth power of the distance. For an **active** satellite we get: $$ P_{r} = \frac{P_{t} \cdot G_{t} \cdot G_{r} \cdot g_{s} \cdot A_{s}}{(4\pi)^{2} \cdot d_{t}^{2} \cdot d_{r}^{2}}$$ Where $g_{s}$ is the transponder amplification coefficient and $A_{s}$ is the effective area of the active satellite antenna. Assuming $g_{s}$ is proportional to $4\pi d_{r}^{2}$ we get: $$ P_{r} = K^{'}\cdot \frac{A_{s}}{d^{2}} \cdot P_{t} $$ Here is an illustration of the footprint of a single geostationary satellite. ![](https://i.imgur.com/DATWhiX.gif) Today you can get a Satellite Data Hotspot for $250. E.g. [Globalstar Sat-Fi2](https://www.magnumelectronics.com/Globalstar-Satellite-Hotspot-p/SATFI-2-US.htm) ![](https://i.imgur.com/2xJgnMn.png) ### SATNET Also known as the Atlantic Packet Satellite Network, was an early satellite network that provided connectivity between the US and Europe. The SATNET link was made available in 1973. Due to its different characteristics than the rest of ARPANET, it ended up having a prominent role in pushing the design of the Internet protocol suite towards being able to support very heterogeneous networks. ![](https://i.imgur.com/cg1s0YF.png) *ARPANET diagram (SATNET link highlighted)* When using a **Carrier-sense multiple access** MAC protocol the user verifies the absence of other ongoing data traffic before transmitting on a shared transmission medium. ### Teledesic project Teledesic was an American company founded in the 1990s which made plans to use low-earth orbiting satellites to provide wireless Internet service almost everywhere on Earth. Some plans envisioned a system that could provide uplinks of as much as 100 Mbit/s and downlinks of up to 720 Mbit/s. The system never ended up being fully deployed due to financing problems. The network of 840 satellites would have cost over $9 billion to build. ![](https://i.imgur.com/yogRjpT.jpg) *Conceptual drawing of a Teledesic satellite* ### BGP The Border gateway protocol, operates among ASes. It is concerned with exchanging reachability information amongst ASes in a scalable way. To better visualize the problem being described, here is an animation of SpaceX's Starlink system when fully deployed: ![](https://media1.giphy.com/media/XHvz5dwJTjmMIqynv9/giphy.gif) ### Autonomous systems (ASes) An AS is simply a network owned by a commercial (or government) entity which is somehow connected to the broader Internet. ASes are identified by a unique 16bit number. #### Frequency-division multiple access (FDMA) FDMA involves dividing the frequency bandwidth of the channel you are using into many non-overlapping sub-channels and allocating each sub-channel to a separate user. #### Time-division multiple access (TDMA) In TDMA you allocate users to time slots when they can transmit. #### Code Division Multiple Access (CDMA) CDMA is an example of multiple access, where several users can send information simultaneously over a single communication channel. ### Satellite communication frequency bands The first satellite to orbit the Earth was Sputnik 1, carried two radio beacons on frequencies of 20.005 and 40.01 MHz. Over time satellite communications have been shifting towards higher and higher frequencies. Sputnik 1 (1957 - the first satellite to orbit the earth) carried two radio beacons on frequencies of 20MHz and 40MHz. At these frequencies, amateur radio enthusiasts could easily tune into its transmissions as it passed overhead (you can listen to samples [here](https://www.youtube.com/watch?v=-YSm2qFwRpI) ). The Explorer earth orbiter missions (1958) used VHF (~100MHz). The first Lunar probe (Pioneer III) used UHF (960MHz). Over time the tendency was to increase the frequency, first to S-band (2.3 GHz), then to X-Band (8.4GHz), Ku-Band (10-18GHz) and so on. The two main reasons for this was competition for frequency allocations at lower frequencies (the frequencies you can use to communicate with are tightly regulated) and the need for more spacecraft antenna directivity. #### Geostationary orbit height calculation A geostationary orbit is a an orbit where the satellite remains always in the same position from the point of view of a static observer on earth. ![](https://i.imgur.com/MIIj1jG.png) The satellite is kept in orbit by the gravitational force of the earth exerted on the satellite ($F_g$). Therefore: $$ F_g = m_{sat} \cdot a_{sat} $$ We can find $F_g$ using Newton's Universal law of gravitation. $$ F_g = G \frac{m_{earth} \cdot m_{sat}}{r^{2}} $$ The acceleration of an object moving around a circle can be expressed using the angular velocity (\omega) of that object (which in this case is the same angular velocity as earth’s angular velocity around its axis): $$ a_{sat} = r \cdot \omega^{2} $$ Earth spins around it’s axis approximately every 24h (86400s). Therefore its angular velocity around its axis is: $$ \omega = \frac{2\pi}{86400}$$ From this we get $$ m_{sat} \cdot r \cdot \omega^{2} = G \frac{m_{earth} \cdot m_{sat}}{r^{2}} $$ $$ \frac{G \cdot m_{earth}}{r^{2}} = r \cdot \omega^{2} $$ $$ \frac{G \cdot m_{earth}}{r^{2}} = r \cdot \omega^{2} $$ $$ r = \sqrt[3]{\frac{G \cdot m_{earth}}{\omega^{2}}} $$ $$ r \approx 4.2232 \cdot 10^7 m $$ Now we just have to subtract the earth’s radius from $r$ to get the altitude of the satellite. $$ 4.2232 \cdot 10^7 - 6.371 \cdot 10^6 \approx 35,861km $$