An Experimental Security Analysis of Two Satphone Standards

Abstract

General-purpose communication systems such as GSM and UMTS have been in the focus of security researchers for over a decade now. Recently also technologies that are only used under more specific circumstances have come into the spotlight of academic research and the hacker scene alike. A striking example of this is recent work [Driessen et al. 2012] that analyzed the security of the over-the-air encryption in the two existing ETSI satphone standards GMR-1 and GMR-2. The firmware of handheld devices was reverse-engineered and the previously unknown stream ciphers A5-GMR-1 and A5-GMR-2 were recovered. In a second step, both ciphers were cryptanalized, resulting in a ciphertext-only attack on A5-GMR-1 and a known-plaintext attack on A5-GMR-2.
In this work, we extend the aforementioned results in the following ways: First, we improve the proposed attack on A5-GMR-1 and reduce its average-case complexity from 2³² to 2²¹ steps. Second, we implement a practical attack to successfully record communications in the Thuraya network and show that it can be done with moderate effort for approximately $5,000. We describe the implementation of our modified attack and the crucial aspects to make it practical. Using our eavesdropping setup, we recorded 30 seconds of our own satellite-to-satphone communication and show that we are able to recover Thuraya session keys in half an hour (on average). We supplement these results with experiments designed to highlight the feasibility of also eavesdropping on the satphone's emanations.
The purpose of this article is threefold: Develop and demonstrate more practical attacks on A5-GMR-1, summarize current research results in the field of GMR-1 and GMR-2 security, and shed light on the amount of work and expertise it takes from setting out to analyze a complex system to actually break it in the real world.

Full text

10
An Experimental Security Analysis of Two Satphone Standards
BENEDIKT DRIESSEN, Ruhr-University Bochum
RALF HUND, Ruhr-University Bochum
CARSTEN WILLEMS, Ruhr-University Bochum
CHRISTOF PAAR, Ruhr-University Bochum
THORSTEN HOLZ, Ruhr-University Bochum
General purpose communication systems such as GSM and UMTS have been in the focus of security re-
searchers for over a decade now. Recently also technologies that are only used under more specific circum-
stances have come into the spotlight of academic research and the hacker scene alike. A striking example
of this is recent work [Driessen et al. 2012] that analyzed the security of the over-the-air encryption in the
two existing ETSI satphone standards GMR-1 and GMR-2. The firmware of handheld devices was reverse-
engineered and the previously unknown stream ciphers A5-GMR-1 and A5-GMR-2 were recovered. In a
second step, both ciphers were cryptanalized, resulting in a ciphertext-only attack on A5-GMR-1 and a
known-plaintext attack on A5-GMR-2.
In this work, we extend the afore-mentioned results in the following ways: First, we improve the proposed
attack on A5-GMR-1 and reduce its average case complexity from 232 to 221 steps. Second, we implement
a practical attack to successfully record communications in the Thuraya network and show that it can be
done with moderate effort for approx. $5 000. We describe the implementation of our modified attack and
the crucial aspects to make it practical. Using our eavesdropping setup, we recorded 30 seconds of our own
satellite-to-satphone communication and show that we are able to recover Thuraya session keys in half an
hour (on average). We supplement these results with experiments designed to highlight the feasibility of
also eavesdropping on the satphone’s emanations.
The purpose of this paper is threefold: Develop and demonstrate more practical attacks on A5-GMR-1,
summarize current research results in the field of GMR-1 and GMR-2 security, and shed light on the amount
of work and expertise it takes from setting out to analyze a complex system to actually break it in the real
world.
Additional Key Words and Phrases: Mobile Security; Satellite Phone Systems; Cryptanalysis; Binary Anal-
ysis; Real-world Attack; Stream Cipher
ACM Reference Format:
Driessen, B., Hund, R., Willems, R., Paar, C., Holz, T.. 2012. An Experimental Security Analysis of Two
Satphone Standards. ACM Trans. Info. Syst. Sec. 16, 3, Article 10 (November 2013), 30 pages.
DOI =10.1145/0000000.0000000 http://doi.acm.org/10.1145/0000000.0000000
1. INTRODUCTION
Mobile communication systems have revolutionized the way we interact with each
other. Instead of depending on landline connections with fixed locations, we can talk to
other people wherever we are and also transmit data from (almost) arbitrary locations.
Especially the Global System for Mobile Communications (GSM) has evolved into an
extremely large-scale system; with more than four billion subscribers in 2011, it is the
This work has been supported in part by the Ministry of Economic Affairs and Energy of the State of North
Rhine-Westphalia (Grant IV.5-43-02/2-005-WFBO-009).
Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted
without fee provided that copies are not made or distributed for profit or commercial advantage and that
copies show this notice on the first page or initial screen of a display along with the full citation. Copyrights
for components of this work owned by others than ACM must be honored. Abstracting with credit is per-
mitted. To copy otherwise, to republish, to post on servers, to redistribute to lists, or to use any component
of this work in other works requires prior specific permission and/or a fee. Permissions may be requested
from Publications Dept., ACM, Inc., 2 Penn Plaza, Suite 701, New York, NY 10121-0701 USA, fax +1 (212)
869-0481, or permissions@acm.org.
c
2013 ACM 1094-9224/2013/11-ART10 $15.00
DOI 10.1145/0000000.0000000 http://doi.acm.org/10.1145/0000000.0000000
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:2 B. Driessen et al.
most widely deployed standard for cellular networks. Many other cellular network
standards like Universal Mobile Telecommunications System (UMTS), CDMA2000
(also known as IMT Multi-Carrier (IMT-MC)), or 3GPP Long Term Evolution (LTE)
exist and are continuously enhanced to meet growing customer demands.
Cellular mobile networks require a so-called cell site to create a cell within the net-
work. The cell site provides all the infrastructure necessary for exchanging radio sig-
nals between mobile handsets and the provider network. For example, a typical cell
site contains one or more sets of transmitters/receivers, antennas, digital signal pro-
cessors to perform all computations, a GPS receiver for timing and other control elec-
tronics. The cells within a network have only a limited operating distance and, thus,
a certain proximity to a cell site is always necessary to establish a connection to the
mobile network.
In practice, however, it is not always possible to be close to a cell site and there are
many use cases in which no coverage is provided. Workers on an oil rig or on board of
a ship, researchers on a field trip in a desert or near the poles, aid workers in remote
areas or areas that are affected by a natural disaster, journalists working in politically
unstable areas, or certain military and governmental undertakings are a few of many
uses cases where terrestrial cellular networks are not available. To overcome this lim-
itation, satellite systems were introduced that provide telephony and data services
based on telecommunications satellites. In such systems, the mobile handset (typically
called satphone) communicates directly with satellites in orbit. Thus, coverage can be
provided without the need of a highly interconnected infrastructure on the Earth’s
surface.
There are two major satphone protocol families, standardized by the European
Telecommunications Standards Institute (ETSI), that were both developed in the past
few years:
—Geostationary Earth Orbit (GEO) Mobile Radio Interface (better known as GMR-1)
is a family of ETSI standards that were derived from the terrestrial cellular stan-
dard GSM. In fact, the specifications of GMR are an extension of the GSM stan-
dard, where certain aspects of the specification are adjusted for satphone settings.
This protocol family is supported by several providers (e.g., Thuraya, SkyTerra,
TerreStar) and has continuously undergone several revisions to support a broader
range of services.
— The GMR-2 family is even closer to GSM. It deviates from the GMR-1 specifications
in numerous ways, most notably the network architecture is different.
The specification documents of GMR-1 and GMR-2 are available online, but do not
provide any information about implementation details of security aspects. More pre-
cisely, it was not publicly known which encryption algorithms are actually used to
secure the communication channels between a satphone and a satellite. Since an at-
tacker can easily eavesdrop on the radio signals between satphone and satellite, even
at some distance, it is obvious that weak encryption would be a serious threat to confi-
dentiality. At this point, it was thus unclear what effort would be needed by an attacker
to actually intercept telephony and data services for common satphone systems.
In this paper, we build on our previous work [Driessen et al. 2012], which reverse-
engineered the stream ciphers A5-GMR-1 and A5-GMR-2, used in the respective stan-
dards. In contrast to the original publication, we focus less on the process of reverse-
engineering itself, but rather collect all relevant information regarding security and
configuration aspects of the system: We describe the network architecture of satellite
telecommunication systems and, to some degree, how they operate on the physical and
protocol level. We describe the architecture of the satphones themselves, the ciphers
we found for GMR-1 and GMR-2 and our results of cryptanalyzing them. We improve
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:3
the complexity of attacking A5-GMR-1 by a factor of 211 due to targeting a different
channel and exploiting the fact that frame numbers and initial states in A5-GMR-1
are linearly related. Latter property allows us to mount an attack that uses multi-
ple speech data frames (instead of only one control channel frame), which leads to
less guessing when the attack is performed. This considerable improvement enables
us to mount an attack on the Thuraya network, for which we proceed to describe the
hardware and software requirements for an actual attack. We discuss crucial aspects
when implementing and executing an eavesdropping attack, experimentally establish
the feasibility of direct uplink interception and ultimately show that GMR-1 privacy is
practically non-existent.
Effectively, we thus demonstrate that current satphone systems are vulnerable to
eavesdropping attacks; the results of this paper can be used to build an interceptor for
satellite telecommunication systems.
2. BACKGROUND AND RELATED WORK
We now introduce the background information necessary to understand the basics of
satellite telephone systems (with a focus on GMR-1), their security mechanisms and
the architecture of the mobile handsets. More information about these topics can be
found in the literature [Wright 1995; Matolak et al. 2002; ETSI 2001a; Maral and
Bousquet 2009; Jim Geovedi and Raoul Chiesa 2011]. Furthermore, we discuss related
work in this area.
2.1. Network Layout
Thuraya implements the GMR-1 standard and provides satellite telephony for most
of Europe, the Middle East, North, Central and East Africa, Asia and Australia. To
achieve this coverage, the network consists of two overlapping regions, each handled by
a different satellite. Thuraya satellites are operating in Geosynchronous Orbit (GSO),
where they do not stay on a position but follow a fixed movement pattern, typically
an analemma. Currently, there are two operational1satellites named Thuraya-2 and
Thuraya-3. The former is relevant here, since it is centered on the Middle East and
supplies most of Europe as well as a large portion of the African continent with con-
nectivity, see Figure 1.
Thuraya offers a diverse range of products for fixed installations, handhelds (i.e.,
satphones) and even solutions for the maritime environment. With the help of Thu-
raya, voice, fax and IP-based data can be transmitted where “traditional” infrastruc-
tures (e.g., GSM, UMTS, WLAN, etc.) are not available. In addition to the satellites, a
set of terrestrial gateways and one primary gateway (located in Sharjah, United Arab
Emirates) handle the entire network as depicted in Figure 2. Gateway stations pro-
vide the connectivity to tethered networks, e.g., telephone calls to a landline are for-
warded to the Public Switched Telephone Network (PSTN) and enable maintenance
and configuration purposes. For this so-called ground segment, conventional wave-
length (3.400 −3.625 GHz and 6.425 −6.725 GHz) signals are used. The user segment
operates on L-band carriers assigned to spotbeams, which are Thuraya’s equivalent to
cells in GSM (albeit covering far more area). In the L-band, the frequency band from
1.525 to 1.559 GHz is assigned for downlink (space-to-earth) communication while the
uplink (earth-to-space transmissions) operates between 1.6265 and 1.6605 GHz. Up-
link and downlink are divided into 1087 paired carrier frequencies with a spacing of
31.25 KHz.
1Thuraya-1 has ceased to operate in May 2007 and has been moved to “junk orbit” [TBS 2012].
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:4 B. Driessen et al.
Fig. 1. Network coverage of the Thuraya-2 satellite, [Peter 2013]
2.2. Channels
Just like in GSM, the Time Division Multiple Access (TDMA) time slot architecture
is employed which partitions a carrier frequency into disjunct timeslots of a fixed
length. Figure 3 shows how a TDMA frame (middle) is split into 24 timeslots (bot-
tom) of 5
3ms each. 16 TDMA frames are grouped together into one multiframe (top),
which is 640 ms long. Furthermore, multiframes are consolidated into a superframe,
of which 4 496 comprise a hyperframe. It should be noted that each TDMA frame has
a19-bit TDMA frame number; numbering starts at 0, the number is incremented with
each new frame.
Several logical channels (called channels from now on) can share a carrier frequency
by being mapped on different timeslots. Due to this architecture, a channel is uniquely
determined by a frequency and a sequence of Timeslot Numbers (TN). There are dif-
ferent types of channels, but all are either Traffic Channels (TCH) for voice, fax or IP-
based data, or Control Channels (CCH). Data is sent over these channels in the form
of frames (i.e., blocks of consecutive bits), that are encoded (cf. Section 2.3) by adding
redundancy to protect against transmission failures. Frames are enumerated by their
respective TDMA frame numbers, which we simply call frame numbers from now on.
For some channels, the encoded data is subsequently encrypted, see Section 2.3. The
encoded (and encrypted) data is finally modulated before it is transmitted via the
phone’s antenna. The encoding scheme differs from channel to channel and is depen-
dent on the respective reliability requirements as defined in the various documents of
the standard.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:5
Ground Segment
PSTN
User Segment
C-Band
C-Band
C-Band
C-Band
L-Band
L-Band
Fig. 2. Layout of the Thuraya network
...
Fig. 3. TDMA architecture of GMR-1 networks
Specific channels relevant for our attack are the Frequency Correction Channel
(FCCH), the Common Control Channel (CCCH) and the Traffic Channel-3 (TCH3).
The FCCH is initially (e.g., after power up) used by the satphone to determine its rel-
ative time and frequency error in order to synchronize with the satellite. The CCCH
is used to send information to the phone when a new channel (e.g., TCH3) needs to be
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:6 B. Driessen et al.
established2. These assignment messages contain an Absolute Radio-Frequency Chan-
nel Number (ARFCN) and a TN, which is, as explained above, all that is required to
use the channel. After TCH3 has been set up on the uplink and downlink, it can be
used to transmit speech data.
In Section 5.1 we will go more into the process of translating ARFCNs into frequen-
cies and how we can actually tune to the TCH3 channel.
2.3. Encoding and Encryption
As mentioned before, all data has to be encoded before it is sent to travel the distance
of around 36 000 Km between ground and satellite.
modulationencryption
intraburst
multiplexing
scrambling
channel
interleaving
convolutional
encoding
block
encoding
Fig. 4. Generic encoding (and encryption) scheme for information in the GMR-1 system
Encoding always increases the size of the encoded data, thus adding redundancy
which allows error detection and possibly correction. Figure 4 shows the multiple en-
coding steps of which not all are always applied to data, depending on the channel it
is sent on. The general encoding procedure is as follows:
Each channel uses the following sequence and order of operations:
— The information bits are encoded with a systematic block code, i.e.,
CRC, building words of information and parity bits;
— these information and parity bits are encoded with a convolutional
code, building the coded bits;
— the coded bits are reordered and potentially interleaved over multi-
ple bursts;
— the interleaved bits are scrambled and, in some cases, multiplexed
with other bits (before or after encryption);
—[ETSI 2002, p. 11]
To protect against eavesdropping of data sent over the air, the encoded bits are en-
crypted with a proprietary cipher. However, doing it the way it is done in GMR-1 leads
to a property we exploit for our real-world attack (cf. Section 4.5).
Encryption in GMR-1 is enabled on a per-session basis, i.e., for the duration of one
call a session key Kc is established (see Figure 5 for a sketch of the respective proto-
col). This key is derived from a challenge RAND sent by the network and a long-term
key Ki, known only to the satphone and network. In the specifications, the key deriva-
tion algorithm is denoted as A8, which serves the same3role as the A8 function in
GSM. On the handheld side it is implemented on the phone’s SIM card, where also the
corresponding long-term key is stored.
With the help of the session key, data can be encrypted with an algorithm denoted
as A5 (shorthand for A5-GMR-1). This algorithm is a stream cipher that encrypts data
2TCH3 is typically established at the beginning of a call.
3It is possible to use GSM SIM cards for Thuraya’s network.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:7
Satphone Network
(has Ki) (has Ki)
Authentication request RAND
←−−−−−−−−−−−−−−−−−−−−−−−−−
SRES =A3(Ki;RAND)Authentication response SRE S
−−−−−−−−−−−−−−−−−−−−−−−−−→
Kc =A8(K i;RAN D)Cipher mode “ON”
←−−−−−−−−−−−−−−−−−−−−−−−−− Kc =A8(Ki;RAN D)
Cipher mode complete
−−−−−−−−−−−−−−−−−−−−−−−−−→
.
.
.
d0=A5(Kc; 0, N0,· · · )Data N0,d0
−−−−−−−−−−−−−−−−−−−−−−−−−→
Data N1,d1
←−−−−−−−−−−−−−−−−−−−−−−−−− d1=A5(Kc; 1, N1,· · · )
Data N2,d2
←−−−−−−−−−−−−−−−−−−−−−−−−− d2=A5(Kc; 1, N2,· · · )
.
.
.
Fig. 5. Protocol for establishing a session key Kc between satphone and provider network
based on the session key and its TDMA frame number and direction (i.e., whether it is
received or sent by a satphone). A second property of the protocol is that it simultane-
ously authenticates the phone against the network—with the help of the A3 algorithm.
Overall, this protocol is strikingly similar to what is implemented in GSM.
2.4. Satellite Telephone Architecture
We now briefly introduce the general architectural structure of satellite phones and the
hardware typically found in such devices. In a later section, we provide more details on
the specific phones we studied during our analysis, including a discussion of the actual
hardware.
In general, the architecture of satellite phones is similar to the architecture of cel-
lular phones [Welte 2010]. Both types of handsets have to perform a lot of processing
of speech and signal data, thus they typically ship with a dedicated digital signal pro-
cessor. Consequently, the afore mentioned, computationally intensive operations are
done by the DSP. More relevant for our purpose are the facts that DSPs are also suit-
able for executing cryptographic algorithms and that encryption is part of the encoding
process, which makes DSP code a prime candidate for holding GMR cipher code.
Nevertheless, the core of a satphone is a standard microprocessor (usually an ARM-
based CPU) that serves as the central control unit within the system. This CPU ini-
tializes the DSP during the boot process. Furthermore, both processors share parts of
the main memory or other peripheral devices to implement inter-processor communi-
cation. To understand the flow of code and data on a phone, we thus needed to analyze
the communication between the two processors.
The operating system running on the main CPU is typically a highly specialized
system that is designed with respect to the special requirements of a phone system
(e.g., limited resources, reliability, real-time constraints, etc.). All of the software is de-
ployed as one large, statically linked firmware binary which typically contains ARM
code mixed with DSP code. For our analysis, we were especially interested in the inter-
processor communication functionality provided by the operating system as well as the
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:8 B. Driessen et al.
DSP initialization routine. This is due to the fact that we needed to extract and trans-
form the DSP code, as it is found in the firmware, into the format actually executed by
the DSP.
2.5. Related Work
Satellite telecommunication systems are related to terrestrial cellular systems since
the GMR standards are derived from the GSM standard. We can thus leverage work
on the analysis of cellular systems for our security analysis as discussed in the fol-
lowing. Briceno et al. published in 1999 an implementation of the GSM A5/1 and
A5/2 algorithms, which they apparently obtained by reverse engineering a GSM hand-
set [Briceno et al. 1999]. However, no details about the analysis process were ever
published and it remains unclear how the algorithms were actually derived. Our anal-
ysis is also based on actual implementations of the ciphers; we discuss the general
approach in Section 3 and provide analysis details in later sections.
There has been much work on the security analysis of the ciphers used within GSM,
e.g., [Golic 1997; Biham and Dunkelman 2000; Biryukov et al. 2000; Ekdahl and Jo-
hansson 2003; Bogdanov et al. 2007; Nohl and Paget 2009; Dunkelman et al. 2010].
The cipher used in GMR-1 is related to the A5/2 algorithm, but the configuration
of the cipher is different. Our attack on this algorithm builds on existing ideas for
A5/2 [Petrovic and Fuster-Sabater 2000; Barkan et al. 2008], which we extended to
enable a time-ciphertext trade-off.
3. GENERAL APPROACH
In this section, we outline the general methodology we used for identifying and ex-
tracting encryption algorithms from satellite phones. Furthermore, we also discuss
the assumptions that helped us during the reverse engineering phase.
We analyzed two representative phones that operate according the two different
standards we are interested in. More precisely, we analyzed the firmwares of the fol-
lowing two phones:
— the Thuraya SO-2510 satphone that implements the GMR-1 specification
— the Inmarsat IsatPhone Pro satphone that implements the GMR-2 specification
The starting point of our analysis was the publicly available firmware upgrade of
each of these two devices. The entire analysis was performed purely statically since we
initially did not have a real satellite phone at our disposal that we could instrument
to perform a dynamic analysis. Furthermore, we did not have access to a whole device
simulator that enables debugging of arbitrary firmware image, thus we had to develop
our own set of analysis tools. However, the ARM code for the main microprocessor
(used by both phones) can be partially executed and debugged in a CPU emulator such
as QEMU.
The approach we followed to analyze both satphones can be separated into the fol-
lowing five phases:
(1) Obtain the firmware update package and the respective update program (usually a
Windows executable).
(2) Extract the firmware image from the package.
(3) Reconstruct the correct memory mappings of the code and data sections in the
firmware image.
(4) Identify the DSP initialization procedure in order to extract the DSP code/mapping.
(5) Search for the encryption algorithms in the DSP code using specific heuristics as
well as control and data flow analysis techniques.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:9
Several steps can be automated, but some manual analysis is nevertheless required.
We successfully applied this method to the two phones we analyzed. In addition, we
speculate that also other kinds of satphones can be analyzed in this way.
These basic assumptions helped us to find the relevant pieces of code in a shorter
amount of time:
(1) The length of the session key is known.
(2) The length of the keystream, as generated by the cipher, is equal to the length of
the encrypted frame.
(3) Since the GMR standards are derived from GSM, the ciphers bear at least some
resemblance to the well-known, LFSR-based A5 algorithms.
Actual lengths for first two assumptions can be derived from the publicly available
parts of the GMR specifications [ETSI 2001b; 2001c]. These assumptions enabled us to
decrease the search space of potential code. The last assumption is rather speculative,
but helped us in finding one of the two algorithms (but did not hold for the second
cipher).
4. SECURITY ANALYSIS OF GMR-1
We used the Thuraya SO-2510 phone as an example for a handset that operates ac-
cording to the GMR-1 standard. This decision was solely driven by the fact that the
firmware of this satphone is publicly available from the vendor’s website. In fact, we
did not analyze any other GMR-1 satellite phone, but since the protocol is standardized
we are confident that our analysis results apply to all other GMR-1 phones as well.
4.1. Hardware Architecture
The Thuraya SO-2510 runs on a Texas Instruments OMAP1510 platform. The core
of the platform is an ARM-925 CPU along with a TI TMS320C5000 signal processor.
This information can be deduced from corresponding strings in the binary and from
pictures of the actual components soldered on the circuit board [OsmocomGMR 2012].
Figure 6 provides a high-level overview of the architecture.
Both processors can communicate with each other using a special shared peripherals
bus. Furthermore, they share the same RAM and can access additional memory (e.g.,
SRAM or Flash) on equal terms. Initially, DSP code or data has to be loaded by the
ARM CPU into the specific memory regions of the DSP. The DSP code can be located
in either the on-chip SARAM (which holds 96 Kb of memory) or in the SRAM, which
is accessed through the memory interface controller (MIC). Writes to the SARAM re-
gion of the DSP are especially interesting for extracting the corresponding DSP code.
The official OMAP1510 documents suggest pre-defined memory regions to be used by
the ARM-MMU for mapping this memory area [Texas Instruments 2012]. During our
analysis, we could confirm that the firmware uses exactly the same mappings.
4.2. Finding the Cipher
We were able to find the cipher A5-GMR-1 in the firmware of a Thuraya SO-2510
phone with the help of IDA Pro by ranking functions in the DSP code according to
their percentage of XOR and SHIFT operations. The four topmost functions in this
ranking turned out to implement the different linear feedback shift registers (LFSR)
of the cipher . The interested reader is referred to the original publication [Driessen
et al. 2012] for more details.
4.3. Structure of the Cipher
The cipher used in GMR-1 is a typical stream cipher. Its design is a modification of
the A5/2 cipher [Petrovic and Fuster-Sabater 2000; Barkan et al. 2008], which is used
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:10 B. Driessen et al.
Memory Interface
Controller (MIC)
ARM Core
MMU
Flash /
SRAM
SDRAM
DSP
MMU C55x DSP
SARAM
ARM / DSP
Shared Peripherals
Fig. 6. The OMAP1510 platform [Texas Instruments 2012]
in GSM networks. Similar to A5/2, the cipher uses four LFSRs which are clocked ir-
regularly. We call these LFSRs R1, R2, R3and R4, see Figure 7 for a schematic of the
structure.
Fig. 7. The A5-GMR-1 cipher
Comparing A5/2 and A5-GMR-1, we see that for most registers the feedback poly-
nomials and also the selection of input taps for the non-linear majority-function M
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:11
Table I. Configuration of LFSRs in A5-GMR-1 and A5/2
A5-GMR-1 A5/2
Size Feedback polynomial Taps Final Feedback polynomial Taps Final
R119 x19 +x18 +x17 +x14 + 1 1,6,15 11 x19 +x5+x2+x+ 1 12,14,15 18
R222 x22 +x21 +x17 +x13 + 1 3,8,14 1 x22 +x+ 1 9,13,16 21
R323 x23 +x22 +x19 +x18 + 1 4,15,19 0 x23 +x15 +x2+x+ 1 13,16,18 22
R417 x17 +x14 +x13 +x9+ 1 1,6,15 - x17 +x5+ 1 3,7,10 -
with
M:{0,1}37→ {0,1}
(x2, x1, x0)27→ x2x1⊕x2x0⊕x0x1
were changed. Also, the positions of the bits that are XORed with the respective out-
puts of the majority functions are different. All feedback-polynomials have five mono-
mials, which is not the case for A5/2, as shown in Table I.
4.4. Mode of Operation
Next we focus on the mode of operation. Clocking a single LFSR means evaluating its
respective feedback polynomial and using the resulting bit to overwrite the leftmost
position of the LFSR, after shifting its current state by one bit to the right. When the
cipher is clocked for the l-th time with irregular clocking active, the following happens:
(1) The irregular clocking component Cevaluates all taps of R4, the remaining registers
are clocked accordingly, i.e.,
(a) Iff M(R4,1, R4,6, R4,15 ) = R4,15, register R1is clocked.
(b) Iff M(R4,1, R4,6, R4,15 ) = R4,6, register R2is clocked.
(c) Iff M(R4,1, R4,6, R4,15 ) = R4,1, register R3is clocked.
(2) The taps of R1, R2and R3are evaluated and one bit of keystream is output accord-
ingly, i.e.,
zl=M(R1,1, R1,6, R1,15)⊕ M(R2,3, R2,8, R2,14 )⊕
M(R3,4, R4,15, R3,19 )⊕R1,11 ⊕R2,1⊕R3,0
is generated.
(3) R4is clocked.
The cipher is operated in two modes, initialization and generation mode. Running
the cipher in former mode includes setting the initial state of the cipher, which is done
in the following way:
(1) All four registers are set to zero.
(2) A 64-bit initialization vector α= (α0, ..., α63 )is computed by XORing the bits of the
19-bit frame number Nand 64-bit session key K, i.e.,
α=F(K, N ) = (K0, K1, K2, K3⊕N6, K4⊕N7,
K5⊕N8, K6⊕N9, K7⊕N10,
K8⊕N11, K9⊕N12 , K10 ⊕N13,
K11 ⊕N14, K12 ⊕N15 ,
K13 ⊕N16, K14 ⊕N17 , K15 ⊕N18,
K16, K17 , ..., K21, K22 ⊕N4,
K23 ⊕N5, K24, ..., K59 , K60 ⊕N0,
K61 ⊕N1, K62 ⊕N2, K63 ⊕N3)
(3) The bits of αare re-ordered to α0with
α0= (α15, α14 , ..., α0, α31, α30 , ..., α16, α47, ..., α32, α63, ..., α48 )2
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:12 B. Driessen et al.
and clocked into all four registers in this order. To clock one bit of α0into R1, its
feedback polynomial is evaluated and the resulting bit then clocked into R1,after
XORing it with the α0bit. The same bit of α0is also clocked into R2,R3and R4.
Then, the second bit of α0is clocked into all four registers in this manner and so on.
While doing this, irregular clocking is deactivated, i.e., all registers are clocked for
each bit of α0.
(4) The least-significant bits of all four registers are set to 1, i.e., R1,0=R2,0=R3,0=
R4,0= 1.
We denote the whole initialization process by
(β0, ..., β18
| {z }
R1
, β19, ..., β40
| {z }
R2
, β41, ..., β63
| {z }
R3
, β64, ..., β80
| {z }
R4
) = G(K, N ),
where βis a 81-bit string, comprised of the consecutive bits of the four initialized reg-
isters. After all registers are initialized, irregular clocking is activated and the cipher
is clocked 250 times. The resulting output bits are discarded.
Now the cipher is switched into generation mode and clocked for 2·mtimes, gen-
erating one bit of keystream at a time. Here, mis the length of an encrypted frame.
Depending on the direction4bit, either the first half or the second half of the 2·m
keystream bits is used for encryption/decryption. We denote the l-th keystream bit by
z(N)
l, where 0≤l < 2·mis the number of irregular clockings (after warm-up) and
Nthe frame number that was used for initialization. Since our cryptanalysis will fo-
cus on the downlink, we denote the continuous keystream for frames N, N + 1, ... (as
decrypted by the phone) by z, where
z=z(N)
0, z(N)
1, ..., z(N)
m−1, z(N+1)
0, ..., z(N+1)
m−1, z(N+2)
0, ...2
is the concatenation of the first halves of z(N), z(N+1) , ... respectively. The choice of m
depends on the type of channel for which data is encrypted or decrypted. For the TCH3
channel, each frame has a length of m= 208 bits.
4.5. Cryptanalysis
The attack we present in the following is a variant of the ciphertext-only attack orig-
inally presented in [Driessen et al. 2012], which itself was inspired by previous at-
tacks [Petrovic and Fuster-Sabater 2000; Barkan et al. 2003] on A5/2. Please note that
we treat bit strings as column vectors and vice versa. We now briefly review the pro-
posed attack that exploits several weaknesses which are either due to the design of the
cipher or due to the use of the cipher in GMR-1:
(1) Given R4, the clocking behavior of A5-GMR-1 is uniquely determined.
(2) Since the inputs to each majority-component are only from one register, one bit of
keystream can always be expressed as an easy to linearize quadratic equation over
GF (2).
(3) In GMR-1, encryption is applied after encoding (which is entirely linear in GF (2))
and scrambling5.
(4) For each two keystreams generated by the same session key but different frame
numbers, the respective initial states are linearly related by the XOR-differences of
the frame numbers.
Due to the first and second observation and given enough keystream bits for a partic-
ular frame Nwe can guess R4, clock the entire cipher for several times and generate
4The first mbits are used on the handset’s side for decryption, on the provider network side for encryption
5While encoding adds redundancy, scrambling is used to “[...] randomize the number of 0s and 1s in the
output bit stream.” [ETSI 2002].
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:13
a linearized system of equations over GF(2), i.e.,
Ax=z(N),
describing keystream bits as linear combinations of terms which either are individual
bits or products of two bits from the initial state of R1, R2and R3. If we guess R4
correctly and Ahas full rank, solving the equation system gives the correct initial state
which can easily be used to obtain the session key. Please note that, even if the session
key is fixed, for different frame numbers not only the keystream but also the initial
state and the matrix describing its relation to the keystream will be different. The
required number of linearly independent equations, and hence the minimum6number
of known keystream bits, is denoted as vwith
v=18 −k1
2+21 −k2
2+22 −k3
2+ (18 −k1) + (21 −k2) + (22 −k3).
Here, the numbers 18,21 and 22 are due to the sizes of registers R1, R2and R3re-
spectively (one bit is subtracted due to the fixed 1 per LFSR, cf. Subsection 4.4). By
k1, k2and k3we denote the number of bits we may additionally guess for each of these
LFSRs. Fixing variables helps to decrease the size of the equation systems and the
number of required keystream bits, but also increases the average amount of bits to
guess for the whole attack to 215+k1+k2+k3.
We now use the principle we have outlined above (and the fact that encryption is
applied to encoded data) for a ciphertext-only attack which explicitly targets the TCH3
channel in GMR-1. Encoding, scrambling and encrypting a 160-bit speech-frame d(N)
with frame number Ncan be expressed as
c(N)=Gtd(N)⊕s⊕z(N),
where Gtis the transpose7of the 160×208 generator matrix Gof the code, sis a 208-bit
pseudo-random scrambling sequence, z(N)the keystream generated for this frame and
c(N)the resulting 208-bit codeword. Gand scan be derived from the specification of
TCH3 (cf. Subsection 5.2), additionally a parity-check matrix Hcan be derived from G
with Hc= 0 iff c=Gtd. Due to this property, if we invert scrambling for a codeword
c(N), we get
Hc(N)⊕s=HGtd(N)⊕z(N)=Hz(N).
Given a syndrome (i.e., a bit vector indicating whether decoding was completed without
errors, potentially enabling error correction) vector r(N)=Hc(N)⊕s, we can, again,
set up an equation system in variables x0, x1, ..., xv−1of the initial state by guessing
R4, clocking the cipher 250 times (to account for the warm-up phase) and another 208
times, i.e.,
H(Ax) = Sx=r(N).
Here, Ais the 208 ×vmatrix that describes the linear relation between xand the bits
z(N)
0, z(N)
1..., z(N)
207 generated by the cipher. Please note that His a 48 ×208 matrix and
subsequently Sis a 48 ×vmatrix which implies that for v > 48 this system is not
uniquely solvable. In order to obtain an equation system where Shas full rank, we
6We need at least as many keystream bits as we have variables and thus equations. However, since not all
equations we obtain by clocking the cipher based on R4are necessarily linearly independent, we may need
even more keystream bits.
7We deviate from the traditional notation of encoding via c=dGand decoding via s=Hctfor the sake of
clarity, since we consider column vectors instead of row vectors.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:14 B. Driessen et al.
need to generate and collect equations from several encrypted frames for consecutive
frame numbers N, N +1, .... For a fixed session key, the initial states for different frame
numbers are linearly related by the XOR-differences of the frame numbers. Taking
these differences into account when generating equations allows to build a uniquely
solvable equation systems and solving this equation system gives a potential initial
state which could have generated z(N).
Now we describe the actual steps of our attack for which we assume that we are in
possession of n48-bit syndromes
r(N0)=Hc(N0)⊕s, ..., r(Nn−1)=Hc(Nn−1)⊕s
which correspond to TCH3 downlink data encrypted under the same session key.
Our attack is parameterized by n, k1, k2, k3and N0and recovers the initial state
β=G(K, N0). Before we proceed, we need to introduce the helper-function V(·)which
can be applied to extract certain bits of the 81-bit state of A5-GMR-1. Depending on
the configuration of the attack, V(·)will extract a bitstring which corresponds to these
positions of the overall state whose bits we have guessed (i.e., R4and parts of the other
registers).
(1) Systematically guess the bitstring γwhich has 20 + k1+k2+k3bits (also incorpo-
rating the fixed bit per LFSR). For each syndrome 0≤i<ndo the following:
(a) Compute the 81-bit difference δ=G(0, N0)⊕ G(0, Ni)in the initialization state
for frame number N0and Ni.
(b) Modify γby XORing it with the corresponding positions of δ, i.e., γ0=γ⊕ V(δ).
(c) Based on γ0and δgenerate a linearized 458×vmatrix B(and vector yfor the one
constant per equation) which describes the linear relation between the initial
state for N0and the 458 keystream bits generated for r(Ni).
(d) Take the warm-up phase into account by discarding the first 250 rows of Bto
obtain a 208 ×vmatrix B0and also discarding the first 250 elements of yto
obtain y0.
(e) Compute the 48 ×vmatrix S0and vector r0such that
S0=HB0and
r0=Hy0⊕r(Ni)=Hy0⊕c(Ni)
and add those rows of S0(and the corresponding bits from r0) to the equation
system Sx=r, which are linearly independent from all previously existing
rows of S.
(f) Abort if Shas full rank.
(2) Solve the equation system by computing x=S−1rand combine the guessed bits and
xappropriately to obtain the 81-bit initialization state candidate β.
(3) Initialize A5-GMR-1 with βand clock it to obtain 208 bits of keystream z0(N0)for
frame number N0and test whether
Hc(N0)⊕s⊕z0(N0)= 0.
If this equation holds, applying the obtained keystream produced a valid codeword.
This implies we have produced the correct keystream and therefore (most likely)
the correct initial state.
Once we have β=G(K, N0), we can set up another equation system
Lα=β⊕with α=L−1(β⊕) = F(K, N0)
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:15
where Ldescribes the process of clocking αinto all four LFSRs (and setting the lowest
bit per LFSR to 1 which is expressed by ). Solving the equation system resolves the
initialization vector αfrom which we can easily derive the session key K=F(α, N0).
5. A REAL-WORLD ATTACK ON THURAYA
In this section we describe the details of our real-world attack on the TCH3 channel in
the Thuraya network.
5.1. Recording TCH3 Data
Executing the attack described in Section 4.5 requires acquiring and setting up ap-
propriate hardware to generate and receive real-world data in the Thuraya network.
Here, we describe our hard- and software setup that allows us to record speech data
frames.
L-Band
C-Band
Satphone AntennaSoftware RadioLaptop
Fig. 8. A schematic of the attack setup
Figure 8 provides a schematic overview of our attack setup: We use a satphone to
establish a call in the Thuraya network and place an antenna nearby, thus receiving all
downlink transmissions. Attached to the antenna is a Software Defined Radio (SDR)
system. With the help of the SDR hardware and some software running on the laptop,
we can demodulate and decode received transmissions. It is important to note here
that we only receive the downlink and not the uplink. We focus on this part of the
communication for two reasons:
(1) Demodulation of downlink transmissions is (mostly) readily available as part of
OsmocomGMR, while this is not true for the uplink.
(2) The downlink can be received (at least) in the entire area which is assigned to one
spotbeam (see below).
Furthermore, if we can decrypt the downlink, we can also decrypt the uplink – both
share the same session key for encryption.
The software we use here is based on the OsmocomGMR project [Munaut 2012],
which is a subproject of Osmocom and is maintained by Sylvain Munaut. The aim
of the Osmocom project family is to establish open-source implementations of a wide
range of communication standards, e.g., GSM, TETRA and even GMR-1. Although the
implementation of OsmocomGMR is still in its infancy it is evolved enough for our
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:16 B. Driessen et al.
purposes; we were able to use it with only a few tweaks—although not in a completely
automated fashion.
Fig. 9. Thuraya spotbeams (with numeric IDs) over central Europe, [Munaut 2012]
In addition to software, the OsmocomGMR project also provides8information re-
garding the configuration of the Thuraya network. As shown in Figure 9, there are
two spotbeams assigned to Germany, they have the IDs 289 and 291. Since our experi-
ments were performed in Bochum, we picked the spotbeam with the former ID, which
is assigned the ARFCN 1007. Given the fact that the downlink frequency band is di-
vided into 1087 physical channels (starting at 1.525 GHz) with a spacing of 31.25 KHz,
ARFCN 1007 translates into a radio frequency of
fD= 1.525 GHz +31.25
2KHz + 1007 ·31.25 KHz = 1.556484375 GHz
for the downlink of the CCH channel.
To obtain real-world data for our attack, we used our Thuraya satphone and estab-
lished some calls to a landline. By simultaneously tuning the SDR to fD, we were able
to intercept TCH3 assignments which were sent to our phone via the CCH (cf. Sec-
tion 2.2). After some experimentation we found that TCH3 is typically assigned to one
of these three ARFCNs: 1008,1009 and 1011, a fact that is now also documented on
the website of OsmocomGMR. Upon observing the ARFCN assignment, we could tune
to the newly assigned frequency and capture most of the encrypted downlink speech
data (missing only a fraction at the beginning of the call). All subsequent data (in-
cluding frame numbers) was stored on a harddisk and could directly be used in our
cryptanalysis.
At this point, it should be noted that we only target the downlink, which can be
received easily and does not require immediate proximity to an eavesdropping target.
However, downlink data only gives half of a communication, which is why we have also
investigated the possibility of receiving uplink transmissions. The Thuraya SO-2510
has a small, helical antenna which not only radiates where the satphone is pointed
at, but also to the sides (although with lower power). We have determined this level
of power by establishing a call on the roof of the university while holding the phone
strictly vertical and measuring the side radiation from a fixed distance. Using a signal
analyzer and a horizontally polarized antenna with 5.85 dB gain, the uplink signal was
detected at 1.65 GHz and determined to have a power of −33 dBm at a distance of 15 m.
This implies that, assuming a free path loss of 110 dB and further propagation losses
8See http://gmr.osmocom.org/trac/wiki/Thuraya_Beams.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:17
of 20–30 dB, it is entirely possible to directly receive the uplink signal at distances of
5Km and more9, given a direct line of sight.
We speculate that a more indirect approach might exploit the fact that Thuraya-2
and Thuraya-3 operate according to the “bent pipe” principle. In this setup, the satel-
lite just acts as a redirector of incoming data, i.e., uplink data sent by a satphone is
simply redirected to the ground segment (although shifted to a different frequency
band). This implies that uplink data of mobile devices in the user segment can be
intercepted by placing a satellite dish “closely” to the central gateway. However, imple-
menting this method of interception is hindered by the lack of public specifications for
the C-Band, which is used to transmit data between ground segment and satellite. An-
other difficulty stems from the fact that several concurrent communications are sent
over this link to the gateway in parallel, therefore L-Band downlink and C-Band up-
link data need to be matched. To us, it seems nonetheless reasonable to assume that
this can be done when considering that Thuraya handles “only” 13 750 calls simulta-
neously, up- and downlink share frame numbers (and we can get frame numbers from
the downlink) and timing of uplink data in the C-Band can probably be predicted quite
accurately.
5.2. Encoding and Parity-Check Matrix
As stated in the previous sections, a key step to move from a known-plaintext to a
ciphertext-only scenario is collapsing all linear encoding steps into a single matrix G
and deriving the respective parity-check matrix H. Obtaining Gis straightforward: All
relevant encoding steps for the TCH3 channel can be found in the respective document
of the specification [ETSI 2001d]. These steps include:
(1) Block encoding
(2) Convolutional encoding
(3) Interleaving
(4) (Scrambling)
(5) Multiplexing
Each step—except for scrambling—can modeled as multiplication of an information
vector with an appropriately constructed matrix Miwith 0≤i≤3. Given these ma-
trices, their product is the 160 ×208 encoding matrix
G=
3
Y
i=0
Mi.
The corresponding parity-check matrix Hwith
H(Gtd)=0 for all d∈ {0,1}160
can be obtained with these steps:
(1) Use Gaussian elimination to find a permutation matrix Pwith
LGP =G0= (I160|T)
for some L, where the left hand side of G0is the 160 ×160 identity matrix. Here, G0
is the systematic form of the encoding matrix.
(2) From the systematic form of G,H0can be obtained easily, i.e.,
H0=Tt|I48.
The result is a 48 ×208 matrix, which is appropriate for code words encoded with
G0.
9Using a high gain antenna and/or a more sensitive receiver should significantly boost reception levels.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:18 B. Driessen et al.
(3) From H0the parity-check matrix Hfor the actual form of Gcan be obtained via
another matrix multiplication, i.e.,
H=H0P−1.
The resulting matrix is given (rows encoded hexadecimally) in Figure 10.
2008020200802000028000a0202020a080800000800000000000
0410014010001000004050144450140044104440400000000000
080200808020080000a000280808082820200000200000000000
4415044000400405005000044454441400401000100000000000
8020080020080200022000888080808808080000080000000000
0004010050440100001050544000101454541040040000000000
4014040040500005000000104050400010501000000000000000
8028080080a0000a0000002080a0800020a02000000000000000
a0280802002020020080a0a02080008080800000008000000000
4400054010401004004050140410144004104400004000000000
280a020080080800802028280820002020200000002000000000
4414040010400404001014005454501004541400001000000000
8822008020800208020088888008000808080000000800000000
0405054050000104000014544044505454541040000400000000
0000080000800008802088000880808808000008000000000000
000002000020000a0200a000802020a080000080000000000000
880a0a82800808088280a00088a8a82000080000000080000000
0000000000000400004000504050004010105040000040000000
a0220a80a080020a00a02800a0a8a80800800000000020000000
0411000100400105005014540454000454405400000010000000
28280a82202020028220880028a8a88000200000000008000000
4405014140441000001014140400141054541440000004000000
0411044100040000001044400004444444444004000000000000
200a020020280002800000082028200008280800000000000000
a802088000a82800828020008828282080880000000000800000
0410004050001401004040145440544014001440000000400000
a8200a0000a80a0800a00800a088880820a00000000000200000
4004044110040501005004444414500454004400000000100000
a808028000a822020220800028a0a08008280000000000080000
0004050150441104001010540040141010145440000000040000
0411004100440004400000400444040040444000000000000000
0000008000080002808028002008082820000020000000000000
282a00828088000a800020008828a82000080000000000008000
4404054110000001000010141450041014401400000000004000
882a0800a0a0000a80000800a088a80800800000000000002000
0010004110440004000004501450100450545000000000001000
a02a02022028000a8000800028a0a88000200000000000000800
4010000100000004000040040444444004040400000000000400
5400054150000000000000545400540054005400000000000000
0411000100400005004014401440001450444010000000000000
000800028080280002002020a080000080000000000000000080
4010050100001400000040005000505010501040000000000040
00020000a0200a00008008082820000020000000000000000020
4015040010400505004004145440400000501000000000000010
0020000220082200002080808808000008000000000000000008
0010000110041100001040404404000004000000000000000004
4014050040400000010050100040505050501040000000000000
0822008200880008800000800888080080888000000000000000
Fig. 10. The (encoded) Hmatrix of the TCH3 channel in Thuraya
5.3. Parameterization
As described in Section 4.5, our ciphertext-only attack on TCH3 is parameterized by
the tuple (n, k1, k2, k3). To actually execute the attack, we have experimentally estab-
lished a working set of parameters.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:19
0 10 20 30 40 50 60 70
0
10
20
30
40
50
60
70
80
90
100
Length of consecutive frame number streak
Percentage of streak length
Fig. 11. Percentage of observed frame number streak lengths obtained from several short calls
First of all, we have determined how many streaks of TCH3 frames with consecutive
frame numbers of a certain length we can expect to obtain with our eavesdropping
setup. By streak we denote a set of received frames with numbers N0, N1, N2, ..., Nn−1
with Ni=Ni−1+ 1 for all 0<i<n, where ndenotes the length of a streak. We
have analyzed the TCH3 data of several 10 second calls and plotted the percentage of
observed streak lengths in Figure 11. The longest streak we have observed consists of
56 TCH3 frames, which serves as an upper bound for the next step.
Table II. Guessed bits of the LFSRs R1,R2and R3
k1k2k3#Variables #Frames (avg.) #Frames (max.)
0 2 4 532 12.42 24
0 3 3 532 12.73 25
1 2 3 533 13.01 25
In order to determine how many bits of the LFSRs R1, R2and R3we need to guess10
(in addition to completely guessing R4), we have performed experiments: We have
systematically evaluated all combinations of k1, k2and k3with the aim to minimize
k1+k2+k3. We found that we need to guess at least 6bits of R1to R3, in order to
always achieve full rank from 56 TCH3 frames. Of the 28 possibilities, only three were
found to always guarantee full rank of the obtained matrices, see Table II. Looking at
the results, we picked k1= 0, k2= 2, k3= 4 because the average and maximum number
of frames required to achieve full rank is lowest. Also, we can be certain that 33% of
the frame number streaks (cf. Figure 11) are at least 24 frames long. This is helpful,
because now we can pick multiple, different subsets of 24 TCH3 frames for our attack,
which allows to validate any session key we may find.
In contrast to the original attack [Driessen et al. 2012], which targets the FACCH3,
uses only one frame and has an average case complexity of 232, we now achieve a
complexity of guessing and solving
215+k1+k2+k3= 215+0+2+4 = 221
10Please note that, for simplicity, we always guess the LSBs of each LFSR. It is certainly an interesting
question which bits should be guessed for optimal performance: guessing different positions might lead to a
higher percentage of linearly independent equations, which in turn could to reduce the number of guessed
bits and thus improve the performance of the attack.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:20 B. Driessen et al.
equation systems. This is due to the fact that we use multiple TCH3 frames to collect
linearly independent equations. In this way, by gaining more equations we gain more
information about the initial state of the cipher and thus have to guess less bits; it was
this considerable improvement that made the attack practical, as shall be described in
the following.
5.4. Implementation
With the parameters we have established in the previous section, we need to generate
and solve on average 221 equation systems with matrices of dimension 532 ×532 and
consequentially test as many state candidates (by decoding via another matrix opera-
tion) to obtain one session key. To speed up the actual attack, we exploit the fact that
the matrices which describe the linear relation between internal state of the cipher
and keystream depend only on the bits we guess. These matrices are simply multi-
plied with H, which is fixed and static (cf. Section 2.3 and Section 5.2). Thus, once we
have fixed a set of parameters we are going to use in the attack, we need to generate
and store the resulting Smatrices only once. This effectively splits the execution of the
attack into a pre-computation and recovery phase:
— In the pre-computation phase, a parameter set is chosen; all matrices, as result of
the bits we guess, are generated and stored. This step has to be done only once.
— In the recovery phase, the generated matrices are read from disk and subsequently
used to build and solve equation systems for actual TCH3 channel data. This step
has to be repeated for every new GMR-1 TCH3 session.
To further optimize the execution time of the recovery phase, we apply two more tricks:
(1) In the pre-computation phase, we already test the matrices for linear dependencies
and bring them in upper triangular form. Since we have to apply the resulting row
operations also to the syndromes (i.e., the results of multiplying ciphertexts with
H) in the recovery phase, we have track and store them, too. While this requires
some more storage space, not having to do linearity testing in the recovery phase
and knowing that the lower tridiagonal part of each matrix is zero is a considerable
improvement in terms of computation time and storage space.
(2) We use a very fast implementation of the Lempel-Ziv (LZ) compression algorithm
[Ziv and Lempel 1977] in order to minimize the required storage space and the
performance impact of reading and writing matrices from to a standard hard-
disk. Although compression introduces some computational overhead in the pre-
computation phase, decompression is fast enough to significantly speed-up our at-
tack (when compared to an attack using uncompressed files).
Although our attack only uses TCH3 data, our implementation is also able to gen-
erate and handle mixtures of FACCH3, TCH3 and keystream frames for an attack,
which makes it considerably more complex. However, more details on the implemen-
tation are not relevant here, which is why the results of our attack will be presented
next.
5.5. Results
In this section we shortly subsume the results of executing our proposed attack and
the hardware we have used. Figure 12 shows the components we have used to perform
the attack:
(1) An accessory11 antenna (typically used for installing satphones in cars) was used to
receive Thuraya traffic.
11Earlier attempts at building such an antenna failed; assembling helical antennas is a very delicate process
and requires experience and very high precision.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:21
Fig. 12. The attack setup: (1) antenna, (2) software radio, (3) laptop, (4) satphone
(2) An Ettus USRP-2 device was used to digitize received transmissions (coming from
the antenna) and send them to an attached laptop.
(3) A laptop is used to control the USRP-2, apply demodulation steps and execute the
implemented attack.
(4) A Thuraya SO-2510 satphone was used as handheld device for communicating over
the Thuraya network.
In the pre-computation phase, we have generated approx. 400 GB of system matrices
in a negligible amount of time. We have executed a 30 second call between satphone
and landline, of which more than 27 seconds of TCH3 data could be saved to disk.
Given the eavesdropped data and pre-computed matrices, we were able to find the
session key for multiple subsets of 24 frames in 32.1minutes (on average).
It must be stressed here, that—since the speech codecs of Thuraya still have not
been reverse-engineered—actually listening to a conversation is not possible for us.
However, since the codecs can be reverse-engineered too (potentially even by applying
similar heuristics as used by us) this is no real obstacle.
6. SECURITY ANALYSIS OF GMR-2
To obtain the code responsible for implementing the cipher according to the GMR-2
standard, we analyzed the latest publicly available firmware image of the Inmarsat
IsatPhone Pro, which was released in June 2010. Only Inmarsat handsets support the
GMR-2 standard at this point and we are confident that our analysis results apply to
all of these satphones.
6.1. Hardware Architecture
The Inmarsat IsatPhone Pro runs on an Analog Devices LeMans AD6900 platform.
The core of the platform is an ARM-926EJ-S CPU, which is supplemented by a Black-
fin DSP (see Figure 13 for a schematic overview). This architecture can be deduced
from plain text strings within the firmware image. We identified an operating system
function that returns information on the underlying hardware of the system and this
function returns the platform name as a static string.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:22 B. Driessen et al.
Blackfin
DSP Core
ARM Core
System
RAM
Bus Controller
Bus Controller External
Memory
Boot
ROM
Shared
Peripherals
Fig. 13. The LeMans AD6900 platform [Jose Fridman, Analog Devices ]
Both processors connect to the same bus interface, which is attached to the system
RAM, any external memory that might be present as well as the shared peripherals
(e.g., SIM card, keypad, SD/MMC slots, etc.). The system is initialized by the boot
ROM code of the ARM CPU. The ARM CPU then has the task to initialize the DSP for
further operations.
6.2. Finding the Cipher
We were able to reverse-engineer the A5-GMR-2 cipher from a firmware update for the
Inmarsat IsatPhone Pro. We used IDA to analyze the ARM portion of the firmware and
a custom disassembler to disassemble the code running on the Blackfin DSP. Starting
from a function which XORs keystream and plaintext data, we traced the usage of the
keystream buffer back to its origin and from there to the cipher which fills it. This
process was mostly done manually, although it only became feasible after applying
a trick to narrow down the search space. For more details, please read the original
publication [Driessen et al. 2012].
6.3. Structure of the Cipher
After having obtained the cipher’s assembler code, we had to find a more abstract
description in order to enhance intuitive understanding of its way of functioning. We
arbitrarily chose to split the cipher into several distinct components which emerged
after examining its functionality. Note that, for the sake of symmetry, we denote the
cipher as A5-GMR-2, although it shows no resemblance to any of the A5-type ciphers
and is called GMR-2-A5 in the respective specification [ETSI 2001c].
The cipher uses a 64-bit encryption-key and operates on bytes. When the cipher
is clocked, it generates one byte of keystream, which we denote by Zl, where lrep-
resents the number of clockings. The cipher exhibits an eight byte state register
S= (S0, S1, ..., S7)28and three major components we call F,Gand H. Additionally,
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:23
there is a 1-bit register Tthat outputs the so-called “toggle-bit” (which alternates be-
tween 1 and 0 for each clock of the cipher) and a 3-bit register Cthat implements a
counter (counting from zero to seven) which is incremented for each clock of the cipher.
Figure 14 provides a schematic overview of the cipher structure. In the following, we
detail the inner workings of each of the three major components.
3 8
4
8
1
6
6
64
Fig. 14. The A5-GMR-2 cipher
Fig. 15.F-component of A5-GMR-2
We begin with the F-component, which is certainly the most interesting part of this
cipher—Figure 15 shows its internal structure. On the left we see another 64-bit reg-
ister split into eight bytes (K0, K1, ..., K7)28. The register is read from two sides, on the
lower side one byte is extracted according to the value of c, i.e., the output of the lower
multiplexer is Kc. The upper multiplexer outputs another byte, but this one is deter-
mined by a 4-bit value we will call α. On the right side, two smaller sub-components
T1:{0,1}47→{0,1}3
T2:{0,1}37→{0,1}3
are implemented via table-lookups (see Table III).
The input of T1is determined by p, Kcand the toggle-bit t. Note that we use p=Zl−1
as a shorthand for the byte of keystream that was generated in the preceding clock.
We model the behavior of the small vertical multiplexer by N(·), which we define as
N:{0,1}×{0,1}87→{0,1}4
(t, x7, x6, ..., x0)7→ ((x3, x2, x1, x0)2if t= 0,
(x7, x6, x5, x4)2if t= 1.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:24 B. Driessen et al.
Table III. T1and T2as lookup-table
xT1(x)T2(x)T2(T1(x))
(0,0,0,0)22 4 6
(0,0,0,1)25 5 3
(0,0,1,0)20 6 4 *
(0,0,1,1)26 7 2
(0,1,0,0)23 4 7
(0,1,0,1)27 3 1
(0,1,1,0)24 2 4 *
(0,1,1,1)21 1 5
(1,0,0,0)23 - 7
(1,0,0,1)20 - 4 *
(1,0,1,0)26 - 2
(1,0,1,1)21 - 5
(1,1,0,0)25 - 3
(1,1,0,1)27 - 1
(1,1,1,0)24 - 4 *
(1,1,1,1)22 - 6
With the help of N, which returns either the higher or lower nibble of its second input,
the following holds for the output of the mentioned multiplexer
α=N(t, Kc⊕p) = N(cmod 2, Kc⊕p).
The output of the upper multiplexer is rotated to the right by as many positions as
indicated by the output of T2, therefore the 8-bit output O0and the 4-bit value O1are
of the following form,
O0= (KT1(α)≫T2(T1(α)))28
O1= (Kc,7⊕p7⊕Kc,3⊕p3,
Kc,6⊕p6⊕Kc,2⊕p2,
Kc,5⊕p5⊕Kc,1⊕p1,
Kc,4⊕p4⊕Kc,0⊕p0)2.
The G-component gets the outputs of the F-component as inputs, i.e., I0=O0, I1=
O1. Additionally, the one byte S7of the state is used as input. As can be seen in Fig-
4
4
4
4
4
4
8
8
8
82
2
6
6
4
4
4
4
4
4
Fig. 16.G-component of A5-GMR-2
ure 16, three sub-components, denoted as B1,B2,B3, are employed—again, they are
implemented in the form of lookup-tables. Each of these components works on 4-bit
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:25
inputs and equally returns 4-bit. After analyzing the tables, we found that all three
simply implement linear boolean arithmetic, i.e.,
B1:{0,1}47→ {0,1}4
(x3, x2, x1, x0)27→ (x3⊕x0, x3⊕x2⊕x0, x3, x1)2,
B2:{0,1}47→ {0,1}4
(x3, x2, x1, x0)7→ (x1, x3, x0, x2)2,
B3:{0,1}47→ {0,1}4
(x3, x2, x1, x0)7→ (x2, x0, x3⊕x1⊕x0, x3⊕x0)2.
Since these sub-components and the XORs are linear and all other operations on single
bits just amount to permutations, the G-component is entirely linear. Therefore, we can
write the 6-bit outputs O0
0, O0
1as linear functions of the inputs I0, I1and S7, i.e.,
O0
0= (I0,7⊕I0,4⊕S7,5, I0,7⊕I0,6⊕I0,4⊕S7,7, I0,7⊕S7,4, I0,5⊕S7,6,
I1,3⊕I1,1⊕I1,0, I1,3⊕I1,0)2,
O0
1= (I0,3⊕I0,0⊕S7,1, I0,3⊕I0,2⊕I0,0⊕S7,3, I0,3⊕S7,0,
I0,1⊕S7,2, I1,2, I1,0)2.
Finally, the H-component gets I0
0=O0
0and I0
1=O0
1as input and constitutes the
non-linear “filter” of the cipher (see Figure 17). Here, two new sub-components
S2:{0,1}67→ {0,1}4
S6:{0,1}67→ {0,1}4
are used and implemented via lookup-tables. Interestingly, these tables were taken
from the DES, i.e., S2is the second S-box and S6represents the sixth S-box of DES.
However, in this cipher, the S-boxes have been reordered to account for the different
addressing, i.e., the four most-significant bits of the inputs to S2and S6select the S-
box-column, the two least-significant bits select the row. Note that this is crucial for
the security of the cipher. The inputs to the S-boxes are swapped with the help of two
Fig. 17.H-component of A5-GMR-2
multiplexers, depending on the value of t. Given the inputs I0
0, I0
1and twe can express
the l-th byte of keystream as
Zl=((S2(I0
1),S6(I0
0))24if t= 0,
(S2(I0
0),S6(I0
1))24if t= 1.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:26 B. Driessen et al.
6.4. Mode of Operation
Next we describe the mode of operation. When the cipher is clocked for the l-th time,
the following happens:
(1) Based on the current state of the S-, C- and T-register, the cipher generates one
byte Zlof keystream.
(2) The T-register is toggled, i.e., if it was 1previously, it is set to 0and vice versa.
(3) The C-register is incremented by one, when 8is reached the register is reset to 0.
(4) The S-register is shifted by a byte to the right, i.e., S7:= S6, S6:= S5etc. The
previous value of S7is fed into the G-component, the subsequent output Zlof His
written back to S0, i.e., S0:= Zl. This value is also passed to the F-component as
input for the next iteration.
The cipher is operated in two modes, initialization and generation. In the initializa-
tion phase, the following steps are performed:
(1) The T- and C-register are set to 0.
(2) The 64-bit encryption-key is written into the K-register in the F-component.
(3) The state-register Sis initialized with the 22-bit frame number N, this procedure
is dependent on the “direction bit” but not detailed here as it is irrelevant for the
remainder of this paper.
After C, T and Shave been initialized, the cipher is clocked eight times, but the result-
ing keystream is discarded.
After initialization is done, the cipher is clocked to generate and output actual
keystream bytes. By Z(N)
lwe denote the l-th (0≤l≤14) byte of keystream generated
after initialization (and warm-up) with frame number N. In GMR-2, the frame number
is always incremented after 15 bytes of keystream, which forces a re-initialization of
the cipher. Therefore, the keystream that is actually used is the concatenation of these
15-byte blocks. We denote continuous keystream for frames N, N + 1, ... by Z, with
Z=Z(N)
0, ..., Z(N)
14 , Z(N+1)
0, ..., Z(N+1)
14 , Z(N+2)
0, ...28.
6.5. Cryptanalysis
In this section, we present a known-plaintext attack that is based on several observa-
tions that can be made when carefully examining the F-component (and the starred
rows in Table III):
(1) If α∈ {(0,0,1,0)2,(1,0,0,1)2}then T1(α) = 0 and T2(T1(α)) = 4, thus O0=
(N(0, K0),N(1, K0))24.
(2) If α∈ {(0,1,1,0)2,(1,1,1,0)2}then T1(α) = 4 and T2(T1(α)) = 4, thus O0=
(N(0, K4),N(1, K4))24.
(3) If T1(α) = c, both multiplexers select the same key-byte. We call this a read-collision
in Kc.
In the following, we describe how to obtain K0and K4with high probability, which
is then leveraged in a second step in order to guess the remaining 48 bits of Kin an
efficient way.
The key idea to derive K0is to examine keystream bytes (Zi, Zi−1, Zi−8)28with i∈
{8,23,38, ...}in order to detect when a read-collision in K0has happened during the
generation of Zi. Please note that due to our choice of ithis
Z8=Z(N)
8, Z23 =Z(N+1)
8, Z38 =Z(N+2)
8, ...
holds, i.e., for each iwe already know that the lower multiplexer has selected K0. In
general, if the desired read-collision has happened in the F-component, the outputs of
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:27
the F-component are
O0= (p3⊕α3, p2⊕α2, p1⊕α1, p0⊕α0, K0,7, K0,6, K0,5, K0,4)2,
O1= (K0,7⊕p7⊕α3, K0,6⊕p6⊕α2, K0,5⊕p5⊕α1, K0,4⊕p4⊕α0)2,
and the subsequent outputs of Gare
O0
0= (p3⊕α3⊕p0⊕α0⊕S7,5, p3⊕α3⊕p2⊕α2⊕p0⊕α0⊕S7,7,
p3⊕α3⊕S7,4, p1⊕α1⊕S7,6,
K0,7⊕p7⊕α3⊕K0,5⊕p5⊕α1⊕K0,4⊕p4⊕α0,
K0,7⊕p7⊕α3⊕K0,4⊕p4⊕α0)2,
O0
1= (K0,7⊕K0,4⊕S7,1, K0,7⊕K0,6⊕K0,4⊕S7,3,
K0,7⊕S7,0, K0,5⊕S7,2,
K0,6⊕p6⊕α2,
K0,4⊕p4⊕α0)2.
Considering the H-component, we also know that
Zi= (S2(O0
1),S6(O0
0))24
holds.
In order to determine K0, we examine the inputs and outputs of S6and S2in the
H-component, starting with S6. Due to the reordering of the DES S-boxes, the column
of S6is selected by the four most-significant bits of O0
0. If we assume a collision in K0
has happened while generating Zi, we can compute these most-significant bits due to
the fact that
S7=Zi−8and p=Zi−1
are also known for all of our choices of i. If, for α∈ {(0,0,1,0)2,(1,0,0,1)2}the lower
nibble of Ziis found in the row with index β, a collision may indeed have happened
and the lower two bits of O0
0must be (β1, β0)2, which implies
K0,7⊕K0,5⊕K0,4=β1⊕p7⊕α3⊕p5⊕α1⊕p4⊕α0,
K0,7⊕K0,4=β0⊕p7⊕α3⊕p4⊕α0.
Here we gain “some” information about the bits of K0,K0,5can even be computed.
We can then use the output of S2to verify whether a collision has happened for the
particular αwe used above. Due to the structure of the S-box, there are only four 6-bit
inputs γwith
S2(γ) = (Zi,7, Zi,6, Zi,5, Zi,4)2.
Due to our partial knowledge about (K0,4, K0,5, K0,7)2we can test for each γwhether
the following relations hold:
γ5
?
=β0⊕p7⊕α3⊕p4⊕α0⊕S7,1,
γ4⊕γ1
?
=β0⊕p7⊕α3⊕p4⊕α0⊕S7,3⊕p6⊕α2,
γ3⊕γ0
?
=β0⊕p7⊕α3⊕S7,0,
γ2⊕γ5
?
=β1⊕p7⊕α3⊕p5⊕α1⊕p4⊕α0⊕S7,1⊕S7,2.
If all of these relations hold for one γ, we can be sure with sufficiently high probability
that a read-collision has indeed happened. A probable hypothesis for K0is now given
by
(γ3⊕S7,0, γ1⊕p6⊕α2, γ2⊕S7,2, γ0⊕p4⊕α0, p3⊕α3, p2⊕α2, p1⊕α1, p0⊕α0)2.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:28 B. Driessen et al.
Our method detects all read-collisions, but there may also be false positives, therefore
the process described above must be iterated for a few times for different portions
of the keystream. Typically, over time, one or two hypotheses occur more often than
others and distinguish themselves quite fast from the rest. Experiments show that
about a dozen key-frames are usually enough so that the correct key-byte is among
the first two hypotheses. The principle we outlined above not only works for K0, it also
allows to recover the value of K4when α∈ {(0,1,1,0)2,(1,1,1,0)2},i∈ {12,27,42, ...}
are chosen appropriately.
In the following we assume that we have obtained a set of hypotheses for K0—we
might also have K4, but this improves the efficiency of the remainder of the attack
only slightly. Based on these hypotheses, starting with the most plausible one, we can
brute-force the remaining key-bytes separately. Please note that the following pro-
cess will only produce the correct key, if our hypothesis for K0was correct. To ob-
tain K1, ..., K7we examine a few keystream-bytes for a second time, while focusing on
the F-component. For each Kjwith j∈ {0,1, ..., 7}for which we already have a hy-
pothesis, we can use the corresponding key-stream bytes (Zi+j, Zi+j−1, Zi+j−8)28with
i∈ {8,23,38, ...}to compute
α=N(jmod 2, Kj⊕Zi+j−1).
If we do not already have a plausible hypothesis for Kkwith k=T1(α), we can simply
try out all possible values δ∈ {0,1, ..., 255}and compute the output of the cipher. If we
find for one value that the output equals Zi+jwe keep δas hypothesis for Kk. This can
be repeated for a few different iuntil a hypothesis for the full key has been recovered.
Since the validity of the full hypothesis solely depends on the correctness of K0, we
must verify each key candidate by generating and comparing keystream.
The overall complexity of this attack depends on how many hypotheses for K0are
used to derive the remaining key. Given 15 −20 key-frames, the correct byte for K0is
usually ranked as best hypothesis so deriving the complete key means testing
(7 ·28)/2≈210
single byte hypotheses for the missing bytes (on average). Clearly, a keystream/time
trade-off is possible: The more key-frames are available to test hypotheses for K0, the
more the right hypothesis distinguishes itself from all others. As a matter of fact, the
most extreme trade-off is simply trying all 28possible values for K0(without even
ranking them), which reduces the required amount of known keystream to about 400–
500 bits but increases the computational complexity to
(7 ·28·28)/2≈218
guesses on average.
7. CONCLUSION AND IMPLICATIONS
In this paper, we merge prior work [Driessen et al. 2012] on the analysis of ETSI’s
satellite communication standards GMR-1 and GMR-2 with a practical attack on the
Thuraya network. Reverse-engineering the firmware of two satphones revealed that
the ciphers used in these standards are surprisingly weak. While A5-GMR-1 is a
modification of the A5/2 cipher (used in GSM), the second stream cipher we found
is an entirely proprietary design. Both ciphers could be broken by us; A5-GMR-1 in a
ciphertext-only setting, while we require a handful of keystreams for A5-GMR-2. The
recovered algorithms as well as the presented attacks were validated experimentally.
Compared to our prior work, we improve the ciphertext-only attack on GMR-1 by a
factor of 211. We do this by targeting a different channel and adapting the attack in or-
der to obtain more equations (thus guessing less bits of R1, R2and R3) from multiple,
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

An Experimental Security Analysis of Two Satphone Standards 10:29
consecutive TCH3 frames. We describe hard- and software requirements, as well as
the necessary information regarding the configuration of the Thuraya network, which
allowed us to execute the proposed attack. We demonstrate the practicability of our
attack by obtaining session keys from Thuraya downlink data within half an hour (on
average). Thus, for GMR-1, we have documented all steps from performing black-box
analysis of the system to finally executing a very efficient real-world attack. The en-
tire process took approx. 6months, the cost of the required equipment is $5 000, while
storage requirements (400 GB for precomputed data) and computational costs are neg-
ligible. These facts, together with our discussion of uplink interception at a distance
of 5Km (or by indirect means in the C-Band), show that interception of GMR-1-based
communication is entirely within reach of any attacker with only modest financial
means but sufficient dedication.
This work does not improve our attack on GMR-2, which uses (in the configuration
optimized for keystream) a handful of keystream frames and requires 218 guesses. Very
recently, Li et al. presented a different attack [Li et al. 2013], trading computation time
for keystream. As a result, only one frame of keystream is required and the complexity
increased by a factor of 210. This variant is still a known-plaintext attack and vali-
dating the feasibility of obtaining keystream from a real network (such as Inmarsat’s
network) is the next logical step. To perform this task, access to (and understanding
of) live data from such a network is required. A presentation [Ortega and Muniz 2012]
indicates that a framework to obtain this data exists. The authors were so kind to pro-
vide us with sample data, but the setup used to acquire it proved to introduce some
errors on its own, which hampers analysis. However, given some time to improve the
analysis framework as well as understand the structure of occuring plaintexts, real-
world attacks building on these tools are very likely to occur in the future.
ACKNOWLEDGMENTS
Understanding existing, closed-source communication infrastructures, their security aspects and implica-
tions is inherently important and only made possible by dedicated individuals. Therefore, we gratefully
acknowledge that implementing and carrying out our attack would not have been possible (in the given
timeframe) without the availability of the OsmocomGMR [Munaut 2012] project.
Furthermore, we would like to thank Rainer Kronberger and Stephan Werker of the High Frequency
Laboratory of Cologne University of Applied Science for supporting our work with measuring and approxi-
mating the uplink reception distance, thus greatly helping to establish a better understanding of the risk of
direct uplink interception.
Finally, we want to thank the anonymous reviewers for their insightful comments, which helped to im-
prove this work.
REFERENCES
BARKAN, E., BIHA M, E., AND KELLER, N. 2003. Instant Ciphertext-Only Cryptanalysis of GSM encrypted
communication. In International Crytology Conference (CRYPTO). 600–616.
BARKAN, E., BIHAM, E., AND KELLER, N. 2008. Instant Ciphertext-Only Cryptanalysis of GSM Encrypted
Communication. Journal of Cryptology 21.
BIHAM, E. AND DUNK ELM AN, O. 2000. Cryptanalysis of the A5/1 GSM Stream Cipher. In Indocrypt.
BIRYUKOV, A., S HAMIR, A., AND WAGNER, D. 2000. Real Time Cryptanalysis of A5/1 on a PC. In Fast
Software Encryption (FSE).
BOGDANOV, A., E ISENBARTH, T., AND RUPP, A. 2007. A Hardware-Assisted Realtime Attack on A5/2 With-
out Precomputations. In Cryptographic Hardware and Embedded Systems (CHES).
BRICENO, M., GOLDBERG, I., AND WAGNER, D. 1999. A pedagogical implementation of the GSM A5/1
and A5/2 “voice privacy” encryption algorithms. Originally published at http://www.scard.org, mirror at
http://cryptome.org/gsm-a512.htm.
DRIESSEN, B., HU ND, R., WILLEMS, C., PA AR, C., AND HOLZ, T. 2012. Don’t Trust Satellite Phones: A
Security Analysis of Two Satphone Standards. In IEEE Symposium on Security and Privacy. 128–142.
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.

10:30 B. Driessen et al.
DUNKEL MAN, O., K ELLER, N., AND SHAMIR, A. 2010. A Practical-Time Related-Key Attack on the KA-
SUMI Cryptosystem Used in GSM and 3G Telephony. In International Crytology Conference (CRYPTO).
EKDAHL, P. AND JOHANSSON, T. 2003. Another Attack on A5/1. IEEE Transactions on Information The-
ory 49, 1.
ETSI. 2001a. ETSI TS 101 376-3-2 V1.1.1 (2001-03); GEO-Mobile Radio Interface Specifications; Part 3:
Network specifications; Sub-part 2: Network Architecture; GMR-1 03.002. Tech. rep.
ETSI. 2001b. ETSI TS 101 376-3-9 V1.1.1 (2001-03); GEO-Mobile Radio Interface Specifications; Part 3:
Network specifications; Sub-part 9: Security related Network Functions; GMR-1 03.020. Tech. rep.
ETSI. 2001c. ETSI TS 101 377-3-10 V1.1.1 (2001-03); GEO-Mobile Radio Interface Specifications; Part 3:
Network specifications; Sub-part 9: Security related Network Functions; GMR-2 03.020. Tech. rep.
ETSI. 2001d. ETSI TS 101 377-5-3 V1.1.1 (2001-03); GEO-Mobile Radio Interface Specifications; Part 5:
Radio interface physical layer specifications; Sub-part 3: Channel Coding; GMR-2 05.003. Tech. rep.
ETSI. 2002. ETSI TS 101 376-5-3 V1.2.1 (2002-04); GEO-Mobile Radio Interface Specifications; Part 5:
Radio interface physical layer specifications; Sub-part 3: Channel Coding; GMR-1 05.003. Tech. rep.
GOLIC, J. D. 1997. Cryptanalysis of alleged A5 stream cipher. In Proceedings of the 16th annual interna-
tional conference on Theory and application of cryptographic techniques. EUROCRYPT’97. Springer-
Verlag, 239–255.
JIM GEOVEDI AND RAOUL CHI ESA. 2011. Hacking a Bird in the Sky. In HITBSecConf, Amsterdam.
JOSE FRID MAN, ANALOG DEVICES. How to optimize H.264 video decode on a digital baseband processor.
LI, R., LI, H., LI, C., AND SU N, B. 2013. A Low Data Complexity Attack on the GMR-2 Cipher Used in the
Satellite Phones. In International Workshop on Fast Software Encryption (FSE).
MARAL, G. AND BOUSQUET, M. 2009. Satellite Communications Systems: Systems, Techniques and Tech-
nology 5 Ed. John Wiley & Sons.
MATO LA K, D., N OERPEL, A., G OODINGS, R., STAAY, D., AND BA LDA SAN O, J. 2002. Recent progress in
deployment and standardization of geostationary mobile satellite systems. In Military Communications
Conference (MILCOM).
MUNAUT, S. 2012. OsmocomGMR.
NOHL, K. AND PAGET, C. 2009. GSM: SRSLY? 26th Chaos Communication Congress.
ORTEGA, A. AND MUNIZ , S. 2012. Satellite baseband mods: Taking control of the InmarSat
GMR-2 phone terminal. ekoparty Security Conference. http://www.groundworkstech.com/blog/
ekoparty2012satellitebasebandmods.
OSMOCO MGMR. 2012. Thuraya SO-2510.
PETER. 2013. Airborne satellite weather data.
PETROVIC, S. AND FUS TER-SABAT ER, A. 2000. Cryptanalysis of the A5/2 Algorithm. Tech. rep. http://
eprint.iacr.org/2000/052.
TBS. 2012. The Satellite Encyclopedia.
TEXAS INSTRUMENTS. 2012. The OMAP 5910 Platform.
WELTE , H. 2010. Anatomy of contemporary GSM cellphone hardware.
WRIGHT, D. 1995. Reaching out to remote and rural areas: Mobile satellite services and the role of Inmarsat.
Telecommunications Policy 19, 2, 105 – 116.
ZIV, J. AND LEMPEL, A. 1977. A universal algorithm for sequential data compression. IEEE Transactions
on Information Theory 23, 3, 337 – 343.
Received December 2012; revised May 2013; accepted November 2013
ACM Transactions on Information and System Security, Vol. 16, No. 3, Article 10, Publication date: November 2013.